chore: bump version to 0.0.2-beta.3

* [AGC] Decode VOP3P and emit packed f16 arithmetic (first slice)

On gfx10 the VOP3P family lives under its own 0b110011000 prefix (word0
top byte 0xCC), which the major-opcode switch currently routes to the
SMEM branch, so packed instructions were decoded as scalar memory ops
and emitted as silent no-ops. Intercept the exact 9-bit prefix ahead of
the switch, decode the five packed-f16 arithmetic opcodes with their
op_sel/op_sel_hi/neg_lo/neg_hi/clamp modifiers, and emit them as
UnpackHalf2x16 -> component-wise f32 vec2 ops -> PackHalf2x16 so no
Float16 capability is needed.

Bit layout and opcode numbers pinned to LLVM MC test encodings
(vop3p.s, gfx10_vop3p_literalv216.txt) and VOP3PInstructions.td.
Unsupported modifiers, packed constants and out-of-scope packed opcodes
fail with a clear error instead of emitting wrong results.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* [AGC] Make packed f16 exact and drop v_pk_fma_f16 (review response)

Address the FP16 correctness review on the VOP3P slice.

Replace GLSL UnpackHalf2x16/PackHalf2x16 with explicit integer f16<->f32
conversions (EmitHalfToFloat/EmitFloatToHalf): exact widening with subnormal
normalisation, and narrowing with round-to-nearest-even, overflow-to-Inf and
NaN/Inf handling. Their subnormal and rounding behaviour no longer depends on
implementation-defined float-controls modes.

With exact conversions, v_pk_add_f16 and v_pk_mul_f16 are bit-exact to a true
f16 op (f32 result rounds losslessly to f16; a f16 product fits in f32). Emit
v_pk_min_f16/v_pk_max_f16 as fminnum_like/fmaxnum_like (NaN operand returns the
other; ordered numeric compare) instead of GLSL FMin/FMax.

v_pk_fma_f16 now fails emission loudly: a fused f16 FMA rounds once, an f32
multiply-add then pack double-rounds (fma(0x4100,0x7522,0x04EA) is 0x7A6B fused
vs 0x7A6A via f32). Exact fused emulation is a planned follow-up slice.

ShaderDump gains an Expect model (Translates/DecodeFails/EmitFails) and packed
regressions: arith, non-default modifiers, and loud-failure pins for the fma
case above and for clamp.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: tensorcrush <tensorcrush@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Antigravity AI <antigravity@gemini.com>

.github/pull_request_template.md

@@ -0,0 +1,27 @@
+## Before submitting
+
+Please read our contribution guidelines before opening a pull request:
+
+➡️ [**CONTRIBUTING.md**](https://github.com/sharpemu/sharpemu/blob/main/CONTRIBUTING.md)
+
+By opening this pull request, you confirm that you have read and agree to follow the contribution guidelines.
+
+## Testing
+
+If applicable, list the game(s) you tested and briefly describe the results.
+
+Example:
+
+- Demon's Souls (PPSA01341) – Boots to splash screen.
+- Dreaming Sarah – Save/load works correctly.
+
+If your changes do not affect runtime behavior (e.g. documentation, tooling, CI), write `N/A`.
+
+## Checklist
+
+By submitting this pull request, you confirm that:
+
+- [ ] I have read and followed `CONTRIBUTING.md`.
+- [ ] I tested my changes or marked the testing section as `N/A`.
+- [ ] I wrote this pull request description myself and did not paste AI-generated explanations.
+- [ ] I listed the game(s) I tested (or marked the testing section as `N/A`).

.github/workflows/notify-site.yml

@@ -0,0 +1,31 @@
+# Copyright (C) 2026 SharpEmu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+# Tells the website repo to rebuild when a release is published, so
+# sharpemu.app/downloads lists the new build within a minute.
+name: Notify website
+
+on:
+  release:
+    types: [published, released, edited, deleted]
+  workflow_dispatch:
+
+jobs:
+  dispatch:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Trigger sharpemu-site rebuild
+        env:
+          TOKEN: ${{ secrets.SITE_DISPATCH_TOKEN }}
+        run: |
+          if [ -z "$TOKEN" ]; then
+            echo "SITE_DISPATCH_TOKEN is not set — skipping website rebuild."
+            exit 0
+          fi
+          curl -fsS -X POST \
+            -H "Authorization: Bearer $TOKEN" \
+            -H "Accept: application/vnd.github+json" \
+            -H "X-GitHub-Api-Version: 2022-11-28" \
+            https://api.github.com/repos/sharpemu/sharpemu-site/dispatches \
+            -d '{"event_type":"release-published"}'
+          echo "Website rebuild requested."

.github/workflows/pr-build-links.yml

@@ -0,0 +1,120 @@
+# Copyright (C) 2026 SharpEmu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+# Posts (and keeps updated) a single PR comment linking the per-platform build
+# artifacts once "Build and Release" finishes.
+#
+# This is a workflow_run workflow on purpose: PRs from forks run "Build and
+# Release" with a read-only GITHUB_TOKEN and cannot comment. workflow_run runs
+# afterwards in the base-repo context with a read-write token and does not check
+# out untrusted fork code, so it can comment safely. Because of that, GitHub only
+# triggers it from the copy on the default branch — it does nothing until merged
+# to main.
+name: PR Build Links
+
+on:
+  workflow_run:
+    workflows: ["Build and Release"]
+    types:
+      - completed
+
+permissions:
+  contents: read
+  actions: read
+  pull-requests: write
+
+jobs:
+  comment:
+    name: Post artifact links
+    runs-on: ubuntu-latest
+    # Only for successful PR builds — artifacts exist only when the build passed.
+    if: >-
+      github.event.workflow_run.event == 'pull_request' &&
+      github.event.workflow_run.conclusion == 'success'
+    steps:
+      - name: Post or update the artifact-links comment
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const run = context.payload.workflow_run;
+            const { owner, repo } = context.repo;
+
+            // Resolve the PR. Same-repo PRs are in workflow_run.pull_requests, but
+            // fork PRs leave it empty AND their head commit is not on a base-repo
+            // branch, so listPullRequestsAssociatedWithCommit misses them too. Match
+            // the open PR by its head "owner:branch" instead, which works for forks.
+            let prNumber;
+            if (run.pull_requests && run.pull_requests.length > 0) {
+              prNumber = run.pull_requests[0].number;
+            } else {
+              let candidates = [];
+              const headOwner = run.head_repository && run.head_repository.owner
+                ? run.head_repository.owner.login
+                : null;
+              if (headOwner && run.head_branch) {
+                const byHead = await github.rest.pulls.list({
+                  owner, repo, state: 'open',
+                  head: `${headOwner}:${run.head_branch}`, per_page: 10,
+                });
+                candidates = byHead.data;
+              }
+              if (candidates.length === 0) {
+                const byCommit = await github.rest.repos.listPullRequestsAssociatedWithCommit({
+                  owner, repo, commit_sha: run.head_sha,
+                });
+                candidates = byCommit.data.filter(pr => pr.state === 'open');
+              }
+              if (candidates.length === 0) {
+                core.info('No open PR for this build; nothing to comment.');
+                return;
+              }
+              prNumber = candidates[0].number;
+            }
+
+            // Collect the per-platform artifacts the build produced.
+            const artifacts = await github.paginate(
+              github.rest.actions.listWorkflowRunArtifacts,
+              { owner, repo, run_id: run.id, per_page: 100 },
+            );
+            const platforms = [
+              { key: 'win-x64', label: 'Windows (win-x64)' },
+              { key: 'linux-x64', label: 'Linux (linux-x64)' },
+              { key: 'osx-x64', label: 'macOS (osx-x64)' },
+            ];
+            const rows = [];
+            for (const platform of platforms) {
+              const artifact = artifacts.find(a => a.name.includes(platform.key));
+              if (!artifact) {
+                continue;
+              }
+              const url = `https://github.com/${owner}/${repo}/actions/runs/${run.id}/artifacts/${artifact.id}`;
+              rows.push(`| ${platform.label} | [\`${artifact.name}\`](${url}) |`);
+            }
+            if (rows.length === 0) {
+              core.info('No platform artifacts on this run; nothing to comment.');
+              return;
+            }
+
+            const marker = '<!-- pr-build-links -->';
+            const body = [
+              marker,
+              `### 📦 Build artifacts — \`${run.head_sha.substring(0, 7)}\``,
+              '',
+              '| Platform | Download |',
+              '| --- | --- |',
+              ...rows,
+              '',
+              `From [build run #${run.run_number}](${run.html_url}). ` +
+                'Downloads require a GitHub login and expire after 90 days.',
+            ].join('\n');
+
+            // Upsert one comment so repeated builds refresh it in place.
+            const comments = await github.paginate(github.rest.issues.listComments, {
+              owner, repo, issue_number: prNumber, per_page: 100,
+            });
+            const existing = comments.find(c => c.body && c.body.includes(marker));
+            if (existing) {
+              await github.rest.issues.updateComment({ owner, repo, comment_id: existing.id, body });
+            } else {
+              await github.rest.issues.createComment({ owner, repo, issue_number: prNumber, body });
+            }

.github/workflows/workflow.yml

@@ -7,6 +7,8 @@ on:
   push:
     branches:
       - "**"
+    tags:
+      - "v*"
     paths-ignore:
       - "**/*.md"
       - "**/*.png"
@@ -34,29 +36,36 @@ jobs:
     name: Init
     runs-on: ubuntu-latest
     outputs:
-      archive-name: ${{ steps.vars.outputs.archive-name }}
-      artifact-name: ${{ steps.vars.outputs.artifact-name }}
       release-name: ${{ steps.vars.outputs.release-name }}
       release-tag: ${{ steps.vars.outputs.release-tag }}
+      safe-ref: ${{ steps.vars.outputs.safe-ref }}
       short-sha: ${{ steps.vars.outputs.short-sha }}
+      version: ${{ steps.vars.outputs.version }}
     steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
       - name: Compute workflow variables
         id: vars
         shell: bash
         run: |
           short_sha="${GITHUB_SHA::7}"
           safe_ref="$(echo "${GITHUB_REF_NAME}" | tr '[:upper:]' '[:lower:]' | sed 's#[^a-z0-9._-]#-#g')"
-          archive_name="sharpemu-win64-${short_sha}.zip"
-          artifact_name="sharpemu-win64-${short_sha}"
-          release_tag="win64-${safe_ref}-${short_sha}"
-          release_name="SharpEmu win64 ${short_sha}"
+          version="$(python3 -c 'import sys, xml.etree.ElementTree as ET; version = ET.parse("Directory.Build.props").getroot().findtext(".//SharpEmuVersion"); sys.exit("Directory.Build.props is missing SharpEmuVersion") if not version or not version.strip() else print(version.strip())')"
+          release_tag="v${version}"
+          release_name="SharpEmu v${version}"
+
+          if [ "${GITHUB_REF_TYPE}" = "tag" ] && [ "${GITHUB_REF_NAME}" != "${release_tag}" ]; then
+            echo "Release tag ${GITHUB_REF_NAME} does not match project version ${release_tag}." >&2
+            exit 1
+          fi
 
           {
             echo "short-sha=${short_sha}"
-            echo "archive-name=${archive_name}"
-            echo "artifact-name=${artifact_name}"
+            echo "safe-ref=${safe_ref}"
             echo "release-tag=${release_tag}"
             echo "release-name=${release_name}"
+            echo "version=${version}"
           } >> "$GITHUB_OUTPUT"
 
   reuse:
@@ -92,14 +101,20 @@ jobs:
           cache: true
           cache-dependency-path: |
             Directory.Packages.props
-            src/**/packages.lock.json
+            Directory.Build.props
 
       - name: Restore solution
-        run: dotnet restore SharpEmu.slnx --locked-mode
+        run: dotnet restore SharpEmu.slnx
 
       - name: Build solution
         run: dotnet build SharpEmu.slnx -c Release --no-restore
 
+      # Runs every test project in the solution; a test failure fails the build, as
+      # does an aerolib.bin generation failure in the build step above (the MSBuild
+      # task logs an error and returns false).
+      - name: Run tests
+        run: dotnet test SharpEmu.slnx -c Release --no-build --verbosity normal
+
       - name: Validate synthetic shaders
         run: dotnet run --project tools/SharpEmu.Tools.ShaderDump/SharpEmu.Tools.ShaderDump.csproj -c Release -- artifacts/shader-dump
 
@@ -110,7 +125,8 @@ jobs:
         run: |
           New-Item -ItemType Directory -Path $env:RELEASE_DIR -Force | Out-Null
 
-          $archivePath = Join-Path $env:RELEASE_DIR "${{ needs.init.outputs.archive-name }}"
+          $archiveName = "sharpemu-${{ needs.init.outputs.version }}-win-x64.zip"
+          $archivePath = Join-Path $env:RELEASE_DIR $archiveName
           if (Test-Path $archivePath) {
             Remove-Item $archivePath -Force
           }
@@ -120,8 +136,72 @@ jobs:
       - name: Upload build artifact
         uses: actions/upload-artifact@v7
         with:
-          name: ${{ needs.init.outputs.artifact-name }}
-          path: ${{ env.RELEASE_DIR }}\${{ needs.init.outputs.archive-name }}
+          name: sharpemu-win-x64-${{ needs.init.outputs.short-sha }}
+          path: ${{ env.RELEASE_DIR }}\sharpemu-${{ needs.init.outputs.version }}-win-x64.zip
+          if-no-files-found: error
+
+  build-posix:
+    name: Build ${{ matrix.rid }}
+    needs:
+      - init
+      - reuse
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - os: ubuntu-latest
+            rid: linux-x64
+          - os: macos-latest
+            rid: osx-x64
+    env:
+      DOTNET_NOLOGO: true
+      NUGET_PACKAGES: ${{ github.workspace }}/.nuget/packages
+      PUBLISH_DIR: ${{ github.workspace }}/artifacts/publish/${{ matrix.rid }}
+      RELEASE_DIR: ${{ github.workspace }}/artifacts/release
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Setup .NET SDK
+        uses: actions/setup-dotnet@v5
+        with:
+          dotnet-version: 10.0.103
+          cache: true
+          cache-dependency-path: |
+            Directory.Packages.props
+            Directory.Build.props
+
+      - name: Restore solution
+        run: dotnet restore SharpEmu.slnx
+
+      - name: Build solution
+        run: dotnet build SharpEmu.slnx -c Release --no-restore
+
+      # Also runs on Linux/macOS: the aerolib.bin MSBuild task does platform-sensitive
+      # path handling, so a cross-platform generation regression surfaces here.
+      - name: Run tests
+        run: dotnet test SharpEmu.slnx -c Release --no-build --verbosity normal
+
+      - name: Publish ${{ matrix.rid }} CLI
+        run: dotnet publish src/SharpEmu.CLI/SharpEmu.CLI.csproj -c Release -r ${{ matrix.rid }} --self-contained true --no-restore -p:PublishDir="$PUBLISH_DIR"
+
+      - name: Stage MoltenVK next to the build
+        if: matrix.rid == 'osx-x64'
+        run: scripts/fetch-macos-moltenvk.sh "$PUBLISH_DIR"
+
+      - name: Create release archive
+        run: |
+          mkdir -p "$RELEASE_DIR"
+          # tar keeps the executable bit, which zip would drop.
+          tar -czf "$RELEASE_DIR/sharpemu-${{ needs.init.outputs.version }}-${{ matrix.rid }}.tar.gz" \
+            -C "$PUBLISH_DIR" .
+
+      - name: Upload build artifact
+        uses: actions/upload-artifact@v7
+        with:
+          name: sharpemu-${{ matrix.rid }}-${{ needs.init.outputs.short-sha }}
+          path: ${{ env.RELEASE_DIR }}/sharpemu-${{ needs.init.outputs.version }}-${{ matrix.rid }}.tar.gz
           if-no-files-found: error
 
   release:
@@ -129,35 +209,42 @@ jobs:
     needs:
       - init
       - build
-    if: github.event_name == 'workflow_dispatch' || (github.event_name == 'push' && github.ref == 'refs/heads/main')
+      - build-posix
+    # Versioned releases are immutable and tag-driven. Branch and manual runs
+    # still produce Actions artifacts without modifying a published release.
+    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
     runs-on: ubuntu-latest
     permissions:
       contents: write
     steps:
-      - name: Download build artifact
+      - name: Download build artifacts
         uses: actions/download-artifact@v8
         with:
-          name: ${{ needs.init.outputs.artifact-name }}
           path: release
 
-      - name: Create or update release
+      - name: Create release
         shell: bash
         env:
-          ARCHIVE_NAME: ${{ needs.init.outputs.archive-name }}
           GH_REPO: ${{ github.repository }}
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           RELEASE_NAME: ${{ needs.init.outputs.release-name }}
           RELEASE_TAG: ${{ needs.init.outputs.release-tag }}
+          VERSION: ${{ needs.init.outputs.version }}
         run: |
-          asset_path="release/${ARCHIVE_NAME}"
-          notes="Automated Windows build for commit ${GITHUB_SHA}."
+          mapfile -t assets < <(find release -type f \( -name '*.zip' -o -name '*.tar.gz' \) | sort)
+          if [ "${#assets[@]}" -eq 0 ]; then
+            echo "No release assets found." >&2
+            exit 1
+          fi
+
+          notes="Automated SharpEmu v${VERSION} build for commit ${GITHUB_SHA}."
 
           if gh release view "${RELEASE_TAG}" >/dev/null 2>&1; then
-            gh release upload "${RELEASE_TAG}" "${asset_path}" --clobber
-            gh release edit "${RELEASE_TAG}" --title "${RELEASE_NAME}" --notes "${notes}"
-          else
-            gh release create "${RELEASE_TAG}" "${asset_path}" \
-              --title "${RELEASE_NAME}" \
-              --notes "${notes}" \
-              --target "${GITHUB_SHA}"
+            echo "Release ${RELEASE_TAG} already exists and will not be modified." >&2
+            exit 1
           fi
+
+          gh release create "${RELEASE_TAG}" "${assets[@]}" \
+            --verify-tag \
+            --title "${RELEASE_NAME}" \
+            --notes "${notes}"

CONTRIBUTING.md

@@ -5,6 +5,11 @@ SPDX-License-Identifier: GPL-2.0-or-later
 
 # Contributing
 
+> [!IMPORTANT]
+> The pull request template is mandatory.
+>
+> Pull requests that do not follow the template or leave the required checklist incomplete will be closed without review, even if the proposed code is technically correct or beneficial. Please review these contribution guidelines before submitting a pull request.
+
 Contributions are always welcome!
 
 Before opening a pull request, please keep the following in mind:

Directory.Build.props

@@ -9,7 +9,8 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <ImplicitUsings>enable</ImplicitUsings>
     <Nullable>enable</Nullable>
     <GenerateDocumentationFile>true</GenerateDocumentationFile>
-    <RestorePackagesWithLockFile>true</RestorePackagesWithLockFile>
+    <SharpEmuVersion>0.0.2-beta.3</SharpEmuVersion>
+    <Version>$(SharpEmuVersion)</Version>
 
     <RepoRoot>$([MSBuild]::NormalizeDirectory('$(MSBuildThisFileDirectory)'))</RepoRoot>
 

Directory.Packages.props

@@ -12,7 +12,11 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <PackageVersion Include="Avalonia.Fonts.Inter" Version="11.3.18" />
     <PackageVersion Include="Avalonia.Themes.Fluent" Version="11.3.18" />
     <PackageVersion Include="Iced" Version="1.21.0" />
+    <PackageVersion Include="Microsoft.Build.Framework" Version="17.14.8" />
+    <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.11.0" />
+    <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.12.0" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.14.1" />
+    <PackageVersion Include="Silk.NET.Input" Version="2.23.0" />
     <PackageVersion Include="Silk.NET.Vulkan" Version="2.23.0" />
     <PackageVersion Include="Silk.NET.Vulkan.Extensions.EXT" Version="2.23.0" />
     <PackageVersion Include="Silk.NET.Vulkan.Extensions.KHR" Version="2.23.0" />
@@ -22,4 +26,4 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <PackageVersion Include="xunit" Version="2.9.3" />
     <PackageVersion Include="xunit.runner.visualstudio" Version="3.1.1" />
   </ItemGroup>
-</Project>
+</Project>

README.md

@@ -25,8 +25,9 @@ SPDX-License-Identifier: GPL-2.0-or-later
 
 ---
 
-> [!WARNING]  
-> Currently the primary development target is Windows.
+> [!NOTE]  
+> SharpEmu supports Windows x64, Linux x64, and macOS x64. Apple Silicon Macs
+> can run the macOS x64 build through Rosetta 2.
 
 > [!WARNING]  
 > SharpEmu is an experimental PS5 emulator developed from scratch in C#. The current focus is on accuracy and infrastructure setup rather than game-specific compatibility.
@@ -40,6 +41,16 @@ This project is developed purely for research and educational purposes. There ar
 SharpEmu focuses exclusively on the PlayStation 5.  
 Our goal is **not** to emulate PS4 games, as there is already an excellent emulator dedicated to that platform: **ShadPS4**.
 
+## Games Tested
+
+|               Demons Souls Remake                   |                     Dreaming Sarah                         |
+| :-----------------------------------------------------------: | :--------------------------------------------------------------------------------------------: |
+| ![Bloodborne screenshot](./.github/images/demons-souls.jpg) | ![Dreaming Sarah](./.github/images/dreaming-sarah.jpg) |
+
+|                  Void Terrarium                     |                 Dead Cells                    |
+| :------------------------------------------------------------------------: | :------------------------------------------------------------------: |
+| ![Void Terrarium](./.github/images/void-terrarium.jpg) | ![Dead Cells](./.github/images/dead-cells.jpg) |
+
 ## Status
 
 The emulator can currently load the `eboot.bin` of real games, execute native CPU instructions, and partially handle kernel-related functionality. However, several critical components are still missing.
@@ -59,33 +70,33 @@ Current capabilities include:
 
 Some games have reached like `sceVideoOut` and AGC stages.
 
-Currently the project primarily targets Windows. Cross-platform support (Linux and macOS) is planned, but development is currently focused on Windows to simplify early-stage debugging and iteration.
+SharpEmu supports Windows, Linux, and macOS hosts. Video output uses Vulkan on
+Windows and Linux, and MoltenVK on macOS. Platform support is still experimental,
+so compatibility and performance vary by game, operating system, and GPU driver.
 
 ## Using
 
-* Build or Publish project or download in release tab.
-* Open Powershell.
-  * Run Emulator GUI.
-  * Or command: `.\SharpEmu "eboot.bin" 2>&1 | Tee-Object -FilePath "log.txt"`
+Download the release archive for your operating system, extract it, and launch
+SharpEmu with the path to a legally obtained game's `eboot.bin`.
 
-## Games Tested
+Windows PowerShell:
 
-* **Demon's Souls Remake**
-  * [Demon's Souls [PPSA01341]](https://github.com/par274/sharpemu/issues/2)
-  * Demon's Souls is now video loop. Shaders are ready to be converted to SPIR-V/Vulkan. We are continuing our work on this.
-  ![DeS videoOut submit first frame](./.github/images/des-videoout-shaders.jpg)
+```powershell
+.\SharpEmu.exe "C:\path\to\game\eboot.bin" 2>&1 |
+  Tee-Object -FilePath "SharpEmu.log"
+```
 
-* **Poppy Playtime Chapter 1**
-  * [Poppy Playtime Chapter 1 [PPSA20591]](https://github.com/par274/sharpemu/issues/3)
+Linux and macOS:
 
-* **SILENT HILL: The Short Message**
-  * [SILENT HILL: The Short Message [PPSA10112]](https://github.com/par274/sharpemu/issues/4)
+```bash
+chmod +x ./SharpEmu
 
-* **Dreaming Sarah**
-  * [Dreaming Sarah [PPSA02929]](https://github.com/par274/sharpemu/issues/9)
-  * Real texture rendering for this game;
-  ![Splash texture](./.github/images/dreaming-sarah.jpg)
+./SharpEmu "/path/to/game/eboot.bin" 2>&1 |
+  tee SharpEmu.log
+```
 
+A Vulkan-capable GPU and current graphics driver are required. The macOS
+release includes the MoltenVK Vulkan implementation.
 
 > [!IMPORTANT]  
 > This project does **not** support or condone piracy.  
@@ -94,8 +105,8 @@ Currently the project primarily targets Windows. Cross-platform support (Linux a
 
 ## Build
 
-1. Install the **.NET SDK**.
-2. Clone the repository: `git clone https://github.com/par274/sharpemu.git`
+1. Install the .NET SDK version specified in [`global.json`](./global.json).
+2. Clone the repository: `git clone https://github.com/sharpemu/sharpemu.git`
 3. Open the solution file (`SharpEmu.slnx`) in **VSCode**.
 4. Build the project: `dotnet build` or `dotnet publish`
 5. Build artifacts will be located in the `artifacts` directory.
@@ -121,7 +132,7 @@ Provided valuable references for filesystem handling and low-level C# implementa
 
 # License
 
-- [**GPL-2.0 license**](https://github.com/par274/sharpemu/blob/main/LICENSE)
+- [**GPL-2.0 license**](https://github.com/sharpemu/sharpemu/blob/main/LICENSE)
 
 ## Contributing
 

REUSE.toml

@@ -7,10 +7,10 @@ path = [
     "global.json",
     "**/packages.lock.json",
     "scripts/ps5_names.txt",
-    "src/SharpEmu.HLE/Aerolib/aerolib.bin",
     "src/SharpEmu.GUI/Languages/**",
     "_logs/**",
     ".github/images/**",
+    ".github/pull_request_template.md",
     "assets/images/**"
 ]
 precedence = "aggregate"

SharpEmu.slnx

@@ -11,8 +11,12 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <Project Path="src/SharpEmu.HLE/SharpEmu.HLE.csproj" />
     <Project Path="src/SharpEmu.Libs/SharpEmu.Libs.csproj" />
     <Project Path="src/SharpEmu.Logging/SharpEmu.Logging.csproj" />
+    <Project Path="src/SharpEmu.ShaderCompiler/SharpEmu.ShaderCompiler.csproj" />
+    <Project Path="src/SharpEmu.ShaderCompiler.Vulkan/SharpEmu.ShaderCompiler.Vulkan.csproj" />
+    <Project Path="src/SharpEmu.SourceGenerators/SharpEmu.SourceGenerators.csproj" />
   </Folder>
   <Folder Name="/tests/">
     <Project Path="tests/SharpEmu.Libs.Tests/SharpEmu.Libs.Tests.csproj" />
+    <Project Path="tests/SharpEmu.SourceGenerators.Tests/SharpEmu.SourceGenerators.Tests.csproj" />
   </Folder>
 </Solution>

docs/bink2-bridge.md

@@ -0,0 +1,46 @@
+<!--
+Copyright (C) 2026 SharpEmu Emulator Project
+SPDX-License-Identifier: GPL-2.0-or-later
+-->
+
+# Bink 2 bridge
+
+Demon's Souls plays Bink 2 (.bk2) files through a Bink implementation linked
+directly into eboot.bin. It does not use libSceVideodec, therefore an HLE video
+decoder cannot observe or replace those frames.
+
+SharpEmu observes successful guest .bk2 opens and, when a Bink bridge is
+available, presents its decoded BGRA frames at the normal guest-flip boundary.
+This preserves the game's own timing and lets the host Vulkan presenter display
+the movie without trying to execute the PS5-specific Bink GPU decode path.
+
+Without an adapter, Bink movies are skipped by default: their open call returns
+not-found so games that mark cinematics as optional progress to their next
+state instead of waiting on an empty Bink GPU texture.
+
+Set SHARPEMU_BINK_MODE=dummy to retain the open and show a built-in,
+non-decoded placeholder frame. This requires no SDK, but is a visual diagnostic
+only; it does not decode the movie or alter its game logic. Set
+SHARPEMU_BINK_MODE=native to force native bridge mode.
+
+## Supplying the adapter
+
+Bink 2 is proprietary. Obtain a compatible Mac Bink 2 SDK from RAD Game Tools,
+then compile sharpemu_bink2_bridge.c against the SDK's bink.h and Mac library.
+The adapter deliberately contains only a three-function C ABI so the managed
+emulator never depends on RAD's private binary ABI.
+
+Place the resulting libsharpemu_bink2_bridge.dylib next to the SharpEmu
+executable, or point to it explicitly:
+
+    SHARPEMU_BINK2_BRIDGE=/absolute/path/libsharpemu_bink2_bridge.dylib \
+      ./SharpEmu /path/to/eboot.bin
+
+The expected exports are sharpemu_bink2_open_utf8,
+sharpemu_bink2_decode_next_bgra, and sharpemu_bink2_close. The supplied
+adapter opens one movie, exposes BGRA pixels, and advances after each decoded
+frame. The managed side validates dimensions and retains ownership of the
+destination buffer.
+
+If the bridge is absent in native mode, SharpEmu logs one informational line
+and retains the existing guest rendering path.

native/bink2-bridge/sharpemu_bink2_bridge.c

@@ -0,0 +1,51 @@
+/*
+ * Copyright (C) 2026 SharpEmu Emulator Project
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ *
+ * Build this small adapter with a licensed RAD Bink 2 SDK. The SDK and its
+ * headers are not distributed by SharpEmu. See docs/bink2-bridge.md.
+ */
+
+#include <stdint.h>
+#include "bink.h"
+
+typedef struct sharpemu_bink2_info {
+  uint32_t width;
+  uint32_t height;
+  uint32_t frames_per_second_numerator;
+  uint32_t frames_per_second_denominator;
+} sharpemu_bink2_info;
+
+int sharpemu_bink2_open_utf8(const char *path, HBINK *movie, sharpemu_bink2_info *info) {
+  HBINK bink;
+  if (!path || !movie || !info) return 0;
+
+  bink = BinkOpen(path, 0);
+  if (!bink) return 0;
+
+  *movie = bink;
+  info->width = bink->Width;
+  info->height = bink->Height;
+  info->frames_per_second_numerator = bink->FrameRate;
+  info->frames_per_second_denominator = bink->FrameRateDiv;
+  return 1;
+}
+
+int sharpemu_bink2_decode_next_bgra(HBINK movie, uint8_t *destination,
+                                    uint32_t stride, uint32_t destination_bytes) {
+  uint64_t needed;
+  if (!movie || !destination || stride < movie->Width * 4) return 0;
+  needed = (uint64_t)stride * movie->Height;
+  if (needed > destination_bytes) return 0;
+
+  /* Async Bink I/O has not filled the next frame yet; retry on the next host present. */
+  if (BinkWait(movie)) return 0;
+  BinkDoFrame(movie);
+  BinkCopyToBuffer(movie, destination, stride, movie->Height, 0, 0, BINKSURFACE32RA);
+  BinkNextFrame(movie);
+  return 1;
+}
+
+void sharpemu_bink2_close(HBINK movie) {
+  if (movie) BinkClose(movie);
+}

scripts/RELEASE-USE.md

@@ -0,0 +1,90 @@
+<!--
+Copyright (C) 2026 SharpEmu Emulator Project
+SPDX-License-Identifier: GPL-2.0-or-later
+-->
+
+## Release Script
+
+SharpEmu releases are prepared and published using `scripts/release.py`.
+
+The release process consists of two steps:
+
+1. Prepare the version bump through a pull request.
+2. Create and push the release tag after the pull request has been merged.
+
+### Preparing a Release
+
+Run:
+
+```bash
+python scripts/release.py prepare 0.0.2-beta.2
+```
+
+This command will:
+
+- Verify that the working tree is clean.
+- Update the local `main` branch.
+- Create a new branch named `release/0.0.2-beta.2`.
+- Update `SharpEmuVersion` in `Directory.Build.props`.
+- Create a version bump commit.
+- Push the release branch to the remote repository.
+
+Afterwards, open a pull request from:
+
+```text
+release/0.0.2-beta.2
+```
+
+into:
+
+```text
+main
+```
+
+### Publishing a Release
+
+Once the pull request has been merged, update your local repository:
+
+```bash
+git switch main
+git pull --ff-only
+```
+
+Then create and push the release tag:
+
+```bash
+python scripts/release.py tag 0.0.2-beta.2
+```
+
+This command will:
+
+- Verify that the working tree is clean.
+- Confirm that `Directory.Build.props` contains the requested version.
+- Create an annotated Git tag (`v0.0.2-beta.2`).
+- Push the tag to the remote repository.
+
+Pushing the tag automatically triggers the GitHub Release workflow.
+
+### Version Format
+
+Specify the version **without** the `v` prefix.
+
+Examples:
+
+```text
+0.0.2
+0.0.2-alpha.1
+0.0.2-beta.1
+0.0.2-beta.2
+0.0.2-rc.1
+```
+
+The script automatically prefixes the Git tag with `v`.
+
+### Notes
+
+- Run `prepare` only from the `main` branch.
+- Run `tag` only after the version bump pull request has been merged.
+- Do not create release tags manually before merging the version bump.
+- Both commands require a clean working tree.
+- The version in `Directory.Build.props` must exactly match the version passed to the `tag` command.

scripts/check_sysabi_aerolib.py

@@ -1,172 +0,0 @@
-# Copyright (C) 2026 SharpEmu Emulator Project
-# SPDX-License-Identifier: GPL-2.0-or-later
-
-#!/usr/bin/env python3
-
-"""Offline check: SysAbiExport ExportName must hash to its Nid (name2nid).
-
-NIDs absent from aerolib.bin are skipped (unknown/unresolved symbols).
-Known historic mislabels may be allowlisted with a one-line reason.
-
-Run from the repository root:
-  python scripts/check_sysabi_aerolib.py
-  python scripts/check_sysabi_aerolib.py --strict
-"""
-
-from __future__ import annotations
-
-import argparse
-import hashlib
-import re
-import struct
-import sys
-from base64 import b64encode as base64enc
-from binascii import unhexlify as uhx
-from pathlib import Path
-
-SRC_ROOT = Path("src")
-AEROLIB_BIN = Path("src/SharpEmu.HLE/Aerolib/aerolib.bin")
-SYSABI_EXPORT_RE = re.compile(r"\[SysAbiExport\((.*?)\)\]", re.DOTALL)
-NID_RE = re.compile(r'Nid\s*=\s*"([^"]+)"')
-EXPORT_NAME_RE = re.compile(r'ExportName\s*=\s*"([^"]+)"')
-
-# NID -> reason. Keep minimal; fix ExportName when safe instead of growing this list.
-ALLOWLISTED_NIDS: dict[str, str] = {
-    "KMcEa+rHsIo": "Historic kernel MapMemory stub bound to sceAvPlayerAddSource NID; API rewrite deferred.",
-    "WV1GwM32NgY": "Historic WebApi2 init alias for PushEventCreateHandle NID; ABI rewrite deferred.",
-}
-
-
-def name2nid(name: str) -> str:
-    symbol = hashlib.sha1(name.encode() + uhx("518D64A635DED8C1E6B039B1C3E55230")).digest()
-    id_val = struct.unpack("<Q", symbol[:8])[0]
-    nid = base64enc(uhx("%016x" % id_val), b"+-").rstrip(b"=")
-    return nid.decode("utf-8")
-
-
-def find_repo_root() -> Path:
-    cwd = Path.cwd()
-    if (cwd / SRC_ROOT).is_dir() and (cwd / "scripts").is_dir():
-        return cwd
-    script_root = Path(__file__).resolve().parent.parent
-    if (script_root / SRC_ROOT).is_dir():
-        return script_root
-    raise SystemExit("Run from the repository root (src/ and scripts/ expected).")
-
-
-def load_aerolib_nids(aerolib_path: Path) -> set[str]:
-    data = aerolib_path.read_bytes()
-    if len(data) < 4:
-        raise SystemExit(f"Aerolib binary too small: {aerolib_path}")
-
-    count = struct.unpack_from("<I", data, 0)[0]
-    offset = 4
-    nids: set[str] = set()
-    for _ in range(count):
-        if offset >= len(data):
-            raise SystemExit(f"Truncated aerolib.bin while reading NIDs: {aerolib_path}")
-        nid_len = data[offset]
-        offset += 1
-        nid = data[offset : offset + nid_len].decode("utf-8")
-        offset += nid_len
-        if offset + 2 > len(data):
-            raise SystemExit(f"Truncated aerolib.bin name length: {aerolib_path}")
-        name_len = struct.unpack_from("<H", data, offset)[0]
-        offset += 2 + name_len
-        nids.add(nid)
-    return nids
-
-
-def iter_sysabi_exports(cs_path: Path, text: str):
-    for match in SYSABI_EXPORT_RE.finditer(text):
-        block = match.group(1)
-        nid_match = NID_RE.search(block)
-        export_match = EXPORT_NAME_RE.search(block)
-        if nid_match is None or export_match is None:
-            continue
-
-        nid = nid_match.group(1)
-        export_name = export_match.group(1)
-        nid_attr = f'Nid = "{nid}"'
-        abs_pos = text.find(nid_attr, match.start(), match.end())
-        if abs_pos < 0:
-            abs_pos = match.start()
-        line = text.count("\n", 0, abs_pos) + 1
-        yield cs_path, line, nid, export_name
-
-
-def scan(src_root: Path, catalog_nids: set[str]):
-    checked = 0
-    mismatches = []
-    skipped_no_catalog = 0
-    allowlisted = 0
-
-    for cs_path in sorted(src_root.rglob("*.cs")):
-        text = cs_path.read_text(encoding="utf-8")
-        for path, line, nid, export_name in iter_sysabi_exports(cs_path, text):
-            checked += 1
-            computed = name2nid(export_name)
-            if computed == nid:
-                continue
-
-            if nid not in catalog_nids:
-                skipped_no_catalog += 1
-                continue
-
-            if nid in ALLOWLISTED_NIDS:
-                allowlisted += 1
-                continue
-
-            mismatches.append((path, line, nid, export_name, computed))
-
-    return checked, mismatches, skipped_no_catalog, allowlisted
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(
-        description="Check that SysAbiExport ExportName values hash to their Nid via name2nid."
-    )
-    parser.add_argument(
-        "--strict",
-        action="store_true",
-        help="Exit 1 when any non-skipped/non-allowlisted ExportName does not hash to its Nid.",
-    )
-    parser.add_argument(
-        "--quiet",
-        action="store_true",
-        help="Print only the summary line.",
-    )
-    args = parser.parse_args()
-
-    repo_root = find_repo_root()
-    aerolib_path = repo_root / AEROLIB_BIN
-    if not aerolib_path.is_file():
-        raise SystemExit(f"Missing Aerolib catalog: {aerolib_path.as_posix()}")
-
-    catalog_nids = load_aerolib_nids(aerolib_path)
-    checked, mismatches, skipped_no_catalog, allowlisted = scan(
-        repo_root / SRC_ROOT, catalog_nids
-    )
-    ok = checked - len(mismatches) - skipped_no_catalog - allowlisted
-
-    if not args.quiet:
-        for path, line, nid, export_name, computed in mismatches:
-            rel = path.relative_to(repo_root).as_posix()
-            print(
-                f"{rel}:{line}: NID={nid} ExportName={export_name!r} "
-                f"computed={computed}"
-            )
-
-    print(
-        f"checked={checked} ok={ok} fail={len(mismatches)} "
-        f"skipped_no_catalog={skipped_no_catalog} allowlisted={allowlisted} "
-        f"allowlist_size={len(ALLOWLISTED_NIDS)}"
-    )
-
-    if args.strict and mismatches:
-        return 1
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())

scripts/fetch-macos-moltenvk.sh

@@ -0,0 +1,37 @@
+#!/usr/bin/env bash
+# Copyright (C) 2026 SharpEmu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+#
+# Downloads the official (universal x86_64+arm64) MoltenVK dylib and stages
+# it next to a SharpEmu build as libvulkan.1.dylib. The macOS build runs as
+# an x86-64 process under Rosetta 2, so Homebrew's arm64-only Vulkan
+# libraries cannot be used; the presenter looks for this app-local copy.
+#
+# Usage: scripts/fetch-macos-moltenvk.sh [output-dir]
+#        (default output: artifacts/bin/Debug/net10.0/osx-x64)
+set -euo pipefail
+
+MVK_VERSION="${MVK_VERSION:-v1.4.0}"
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+OUT_DIR="${1:-$REPO_ROOT/artifacts/bin/Debug/net10.0/osx-x64}"
+
+if [[ ! -d "$OUT_DIR" ]]; then
+  echo "output directory does not exist: $OUT_DIR (build first?)" >&2
+  exit 2
+fi
+
+WORK_DIR="$(mktemp -d)"
+trap 'rm -rf "$WORK_DIR"' EXIT
+
+echo ">> Downloading MoltenVK $MVK_VERSION..."
+curl -sL -o "$WORK_DIR/mvk.tar" \
+  "https://github.com/KhronosGroup/MoltenVK/releases/download/$MVK_VERSION/MoltenVK-macos.tar"
+tar -xf "$WORK_DIR/mvk.tar" -C "$WORK_DIR" \
+  MoltenVK/MoltenVK/dynamic/dylib/macOS/libMoltenVK.dylib
+
+DYLIB="$WORK_DIR/MoltenVK/MoltenVK/dynamic/dylib/macOS/libMoltenVK.dylib"
+file "$DYLIB" | grep -q x86_64 || { echo "downloaded dylib lacks x86_64 slice" >&2; exit 3; }
+
+cp "$DYLIB" "$OUT_DIR/libMoltenVK.dylib"
+cp "$DYLIB" "$OUT_DIR/libvulkan.1.dylib"
+echo ">> Staged libMoltenVK.dylib + libvulkan.1.dylib in $OUT_DIR"

scripts/generate_aerolib_binary.py

@@ -1,54 +0,0 @@
-# Copyright (C) 2026 SharpEmu Emulator Project
-# SPDX-License-Identifier: GPL-2.0-or-later
-
-#!/usr/bin/env python3
-
-import hashlib
-import struct
-from base64 import b64encode as base64enc
-from binascii import unhexlify as uhx
-from pathlib import Path
-
-NAMES = 'scripts/ps5_names.txt'
-OUTPUT = 'src/SharpEmu.HLE/Aerolib/aerolib.bin'
-
-def name2nid(name):
-    symbol = hashlib.sha1(name.encode() + uhx('518D64A635DED8C1E6B039B1C3E55230')).digest()
-    id_val = struct.unpack('<Q', symbol[:8])[0]
-    nid = base64enc(uhx('%016x' % id_val), b'+-').rstrip(b'=')
-    return nid.decode('utf-8')
-
-def generate():
-    names_path = Path(NAMES)
-    output_path = Path(OUTPUT)
-
-    entries = []
-    with open(names_path, 'r', encoding='utf-8') as f:
-        for line in f:
-            name = line.strip()
-            if name:
-                nid = name2nid(name)
-                entries.append((nid, name))
-
-    print(f"Found {len(entries)} entries")
-
-    data = bytearray()
-    data.extend(struct.pack('<I', len(entries)))
-
-    for nid, name in entries:
-        nid_bytes = nid.encode('utf-8')
-        name_bytes = name.encode('utf-8')
-        data.append(len(nid_bytes))
-        data.extend(nid_bytes)
-        data.extend(struct.pack('<H', len(name_bytes)))
-        data.extend(name_bytes)
-
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    with open(output_path, 'wb') as f:
-        f.write(data)
-
-    print(f"Generated: {output_path} ({len(data):,} bytes)")
-    print(f"Total entries: {len(entries)}")
-
-if __name__ == "__main__":
-    generate()

scripts/release.py

@@ -0,0 +1,408 @@
+#!/usr/bin/env python3
+
+# Copyright (C) 2026 SharpEmu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+from __future__ import annotations
+
+import argparse
+import re
+import subprocess
+import sys
+from pathlib import Path
+
+
+VERSION_PATTERN = re.compile(r"^\d+\.\d+\.\d+(?:-[0-9A-Za-z.-]+)?$")
+VERSION_ELEMENT_PATTERN = re.compile(
+    r"(<SharpEmuVersion>)([^<]+)(</SharpEmuVersion>)"
+)
+
+
+class ReleaseError(RuntimeError):
+    pass
+
+
+def run_git(
+    *args: str,
+    cwd: Path,
+    capture_output: bool = False,
+) -> str:
+    command = ["git", *args]
+
+    try:
+        result = subprocess.run(
+            command,
+            cwd=cwd,
+            check=True,
+            text=True,
+            capture_output=capture_output,
+        )
+    except FileNotFoundError:
+        raise ReleaseError("Git was not found in PATH.") from None
+    except subprocess.CalledProcessError as error:
+        stderr = error.stderr.strip() if error.stderr else ""
+        detail = f"\n{stderr}" if stderr else ""
+
+        raise ReleaseError(
+            f"Git command failed: {' '.join(command)}{detail}"
+        ) from error
+
+    return result.stdout.strip() if capture_output else ""
+
+
+def find_repository_root(script_path: Path) -> Path:
+    root = run_git(
+        "rev-parse",
+        "--show-toplevel",
+        cwd=script_path.resolve().parent,
+        capture_output=True,
+    )
+
+    return Path(root)
+
+
+def get_status(repository_root: Path) -> str:
+    return run_git(
+        "status",
+        "--porcelain",
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+
+def ensure_clean_worktree(repository_root: Path) -> None:
+    status = get_status(repository_root)
+
+    if status:
+        raise ReleaseError(
+            "The working tree is not clean.\n\n"
+            f"{status}\n\n"
+            "Commit, stash, or remove these changes first."
+        )
+
+
+def get_current_branch(repository_root: Path) -> str:
+    branch = run_git(
+        "branch",
+        "--show-current",
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+    if not branch:
+        raise ReleaseError(
+            "HEAD is detached. Switch to a branch before continuing."
+        )
+
+    return branch
+
+
+def ensure_branch_does_not_exist(
+    repository_root: Path,
+    branch: str,
+    remote: str,
+) -> None:
+    local_branch = run_git(
+        "branch",
+        "--list",
+        branch,
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+    if local_branch:
+        raise ReleaseError(f"Branch {branch} already exists locally.")
+
+    remote_branch = run_git(
+        "ls-remote",
+        "--heads",
+        remote,
+        branch,
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+    if remote_branch:
+        raise ReleaseError(
+            f"Branch {branch} already exists on {remote}."
+        )
+
+
+def ensure_tag_does_not_exist(
+    repository_root: Path,
+    tag: str,
+    remote: str,
+) -> None:
+    local_tag = run_git(
+        "tag",
+        "--list",
+        tag,
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+    if local_tag:
+        raise ReleaseError(f"Tag {tag} already exists locally.")
+
+    remote_tag = run_git(
+        "ls-remote",
+        "--tags",
+        remote,
+        f"refs/tags/{tag}",
+        cwd=repository_root,
+        capture_output=True,
+    )
+
+    if remote_tag:
+        raise ReleaseError(f"Tag {tag} already exists on {remote}.")
+
+
+def read_version(props_path: Path) -> str:
+    if not props_path.exists():
+        raise ReleaseError(f"Version file not found: {props_path}")
+
+    content = props_path.read_text(encoding="utf-8")
+    match = VERSION_ELEMENT_PATTERN.search(content)
+
+    if match is None:
+        raise ReleaseError(
+            f"SharpEmuVersion was not found in {props_path.name}."
+        )
+
+    return match.group(2).strip()
+
+
+def update_version(props_path: Path, version: str) -> str:
+    content = props_path.read_text(encoding="utf-8")
+    current_version = read_version(props_path)
+
+    if current_version == version:
+        raise ReleaseError(
+            f"SharpEmuVersion is already set to {version}."
+        )
+
+    updated_content, replacement_count = VERSION_ELEMENT_PATTERN.subn(
+        rf"\g<1>{version}\g<3>",
+        content,
+        count=1,
+    )
+
+    if replacement_count != 1:
+        raise ReleaseError(
+            "Expected exactly one SharpEmuVersion element."
+        )
+
+    props_path.write_text(
+        updated_content,
+        encoding="utf-8",
+        newline="\n",
+    )
+
+    return current_version
+
+
+def prepare_release(
+    repository_root: Path,
+    props_path: Path,
+    version: str,
+    remote: str,
+) -> None:
+    ensure_clean_worktree(repository_root)
+
+    current_branch = get_current_branch(repository_root)
+
+    if current_branch != "main":
+        raise ReleaseError(
+            f"Prepare must be run from main, not {current_branch}."
+        )
+
+    run_git(
+        "pull",
+        "--ff-only",
+        remote,
+        "main",
+        cwd=repository_root,
+    )
+
+    branch = f"release/{version}"
+    ensure_branch_does_not_exist(
+        repository_root,
+        branch,
+        remote,
+    )
+
+    previous_version = read_version(props_path)
+
+    run_git(
+        "switch",
+        "-c",
+        branch,
+        cwd=repository_root,
+    )
+
+    try:
+        update_version(props_path, version)
+
+        relative_props_path = props_path.relative_to(repository_root)
+
+        run_git(
+            "add",
+            relative_props_path.as_posix(),
+            cwd=repository_root,
+        )
+        run_git(
+            "commit",
+            "-m",
+            f"chore: bump version to {version}",
+            cwd=repository_root,
+        )
+        run_git(
+            "push",
+            "-u",
+            remote,
+            branch,
+            cwd=repository_root,
+        )
+    except Exception:
+        print(
+            "\nPrepare failed. The release branch may still exist locally.",
+            file=sys.stderr,
+        )
+        raise
+
+    print()
+    print(f"Prepared release {previous_version} -> {version}")
+    print(f"Branch pushed: {branch}")
+    print()
+    print("Open a pull request from:")
+    print(f"  {branch}")
+    print("into:")
+    print("  main")
+    print()
+    print("After merging the PR, run:")
+    print(f"  python scripts/release.py tag {version}")
+
+
+def create_release_tag(
+    repository_root: Path,
+    props_path: Path,
+    version: str,
+    remote: str,
+) -> None:
+    ensure_clean_worktree(repository_root)
+
+    current_branch = get_current_branch(repository_root)
+
+    if current_branch != "main":
+        raise ReleaseError(
+            f"Tagging must be run from main, not {current_branch}."
+        )
+
+    run_git(
+        "pull",
+        "--ff-only",
+        remote,
+        "main",
+        cwd=repository_root,
+    )
+
+    current_version = read_version(props_path)
+
+    if current_version != version:
+        raise ReleaseError(
+            "Version mismatch:\n"
+            f"  Directory.Build.props: {current_version}\n"
+            f"  Requested tag:         {version}"
+        )
+
+    tag = f"v{version}"
+
+    ensure_tag_does_not_exist(
+        repository_root,
+        tag,
+        remote,
+    )
+
+    run_git(
+        "tag",
+        "-a",
+        tag,
+        "-m",
+        f"SharpEmu {version}",
+        cwd=repository_root,
+    )
+    run_git(
+        "push",
+        remote,
+        tag,
+        cwd=repository_root,
+    )
+
+    print()
+    print(f"Successfully pushed tag {tag}.")
+    print("The release workflow should start automatically.")
+
+
+def parse_arguments() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="Prepare or tag a SharpEmu release."
+    )
+
+    subparsers = parser.add_subparsers(
+        dest="command",
+        required=True,
+    )
+
+    for command in ("prepare", "tag"):
+        subparser = subparsers.add_parser(command)
+        subparser.add_argument(
+            "version",
+            help="Version without the v prefix, e.g. 0.0.2-beta.2.",
+        )
+        subparser.add_argument(
+            "--remote",
+            default="origin",
+            help="Git remote. Default: origin.",
+        )
+
+    arguments = parser.parse_args()
+
+    if not VERSION_PATTERN.fullmatch(arguments.version):
+        parser.error(
+            "Version must look like 0.0.2, "
+            "0.0.2-beta.2, or 0.0.2-rc.1."
+        )
+
+    return arguments
+
+
+def main() -> int:
+    arguments = parse_arguments()
+
+    try:
+        repository_root = find_repository_root(Path(__file__))
+        props_path = repository_root / "Directory.Build.props"
+
+        if arguments.command == "prepare":
+            prepare_release(
+                repository_root,
+                props_path,
+                arguments.version,
+                arguments.remote,
+            )
+        else:
+            create_release_tag(
+                repository_root,
+                props_path,
+                arguments.version,
+                arguments.remote,
+            )
+    except ReleaseError as error:
+        print(f"Error: {error}", file=sys.stderr)
+        return 1
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/test-linux-docker.sh

@@ -0,0 +1,32 @@
+#!/usr/bin/env bash
+# Copyright (C) 2026 SharpEmu Emulator Project
+# SPDX-License-Identifier: GPL-2.0-or-later
+#
+# Smoke-tests the linux-x64 build inside an amd64 container. Useful from any
+# host (including Apple Silicon, where Docker runs the amd64 image under
+# emulation) to confirm the cross-platform layer keeps working on Linux.
+#
+# Usage: scripts/test-linux-docker.sh /path/to/eboot.bin
+set -euo pipefail
+
+GAME_PATH="${1:-}"
+if [[ -z "$GAME_PATH" || ! -f "$GAME_PATH" ]]; then
+  echo "usage: $0 <path-to-eboot.bin>" >&2
+  exit 2
+fi
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+GAME_DIR="$(cd "$(dirname "$GAME_PATH")" && pwd)"
+GAME_FILE="$(basename "$GAME_PATH")"
+PUBLISH_DIR="$REPO_ROOT/artifacts/publish/SharpEmu.CLI/Debug/net10.0/linux-x64"
+
+echo ">> Publishing linux-x64 self-contained build..."
+dotnet publish "$REPO_ROOT/src/SharpEmu.CLI" \
+  -c Debug -r linux-x64 --self-contained -p:PublishSingleFile=false
+
+echo ">> Running inside linux/amd64 container..."
+docker run --rm --platform linux/amd64 \
+  -v "$PUBLISH_DIR":/app:ro \
+  -v "$GAME_DIR":/game:ro \
+  mcr.microsoft.com/dotnet/runtime-deps:10.0 \
+  /app/SharpEmu --log-level=info "/game/$GAME_FILE"

src/SharpEmu.CLI/Program.cs

@@ -45,6 +45,11 @@ internal static partial class Program
     [STAThread]
     private static int Main(string[] args)
     {
+        // Avoid blocking full collections while guest and render threads are
+        // running, and establish the GC mode before the runtime reserves the
+        // fixed guest address-space window.
+        System.Runtime.GCSettings.LatencyMode = System.Runtime.GCLatencyMode.SustainedLowLatency;
+
         try
         {
             return Run(args);
@@ -58,11 +63,15 @@ internal static partial class Program
 
     private static int Run(string[] args)
     {
-        args = NormalizeInternalArguments(args, out var isMitigatedChild);
-        if (args.Length == 0 && !isMitigatedChild)
+        if (Updater.TryApply(args, out var updateExitCode))
+        {
+            return updateExitCode;
+        }
+
+        args = NormalizeInternalArguments(args, out var isMitigatedChild);
+
+        if (args.Length == 0)
         {
-            // No arguments: open the desktop frontend. Any argument selects
-            // the classic CLI behavior below.
             return GuiLauncher.Run();
         }
 
@@ -75,6 +84,142 @@ internal static partial class Program
             TryEnableConsoleFileMirror(earlyLogFilePath);
         }
 
+        if (!CheckHostArchitecture())
+        {
+            return 5;
+        }
+
+        if (OperatingSystem.IsMacOS() || OperatingSystem.IsLinux())
+        {
+            if (OperatingSystem.IsMacOS())
+            {
+                ConfigureMoltenVkDefaults();
+                PreloadMacVulkanLoader();
+            }
+
+            // GLFW requires window creation and event processing on the
+            // process main thread: AppKit demands it on macOS, and X11 has a
+            // single event queue that must be serviced from the main thread
+            // (a window created and polled off it may never map, which showed
+            // as a running game with no visible window on Linux). Emulation
+            // moves to a worker thread and the main thread services the window
+            // work the video presenter posts. Windows keeps a per-thread event
+            // queue, so its window stays on the presenter's own thread.
+            var exitCode = 0;
+            HostMainThread.Enable();
+            var emulation = new Thread(() =>
+            {
+                try
+                {
+                    exitCode = RunEmulator(args, isMitigatedChild);
+                }
+                finally
+                {
+                    HostMainThread.Shutdown();
+                }
+            }, 32 * 1024 * 1024)
+            {
+                Name = "SharpEmu Emulation",
+            };
+            emulation.Start();
+            HostMainThread.Pump();
+            emulation.Join();
+            return exitCode;
+        }
+
+        return RunEmulator(args, isMitigatedChild);
+    }
+
+    /// <summary>
+    /// The supported host execution model, checked before any emulation
+    /// starts: the CPU backend executes guest x86-64 code natively, so the
+    /// host process must be x86-64 — win-x64/linux-x64 on x64 hardware, or
+    /// osx-x64 under Rosetta 2 on Apple Silicon (Rosetta translates the
+    /// whole process, so it still reports as X64 here). An arm64 process
+    /// (e.g. the osx-arm64 build) can browse the GUI but cannot run games;
+    /// failing up front distinguishes that from MoltenVK, signal-handler,
+    /// or guest-memory startup problems.
+    /// </summary>
+    private static bool CheckHostArchitecture()
+    {
+        if (RuntimeInformation.ProcessArchitecture == Architecture.X64)
+        {
+            return true;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][ERROR] Unsupported process architecture " +
+            $"{RuntimeInformation.ProcessArchitecture}: guest code executes " +
+            "natively, so SharpEmu must run as an x86-64 process.");
+        if (OperatingSystem.IsMacOS())
+        {
+            Console.Error.WriteLine(
+                "[LOADER][ERROR] On Apple Silicon, use the osx-x64 build under " +
+                "Rosetta 2 (install with: softwareupdate --install-rosetta).");
+        }
+
+        return false;
+    }
+
+    /// <summary>
+    /// Applies MoltenVK performance defaults before the Vulkan loader is
+    /// loaded. Existing user-provided values always take precedence.
+    /// </summary>
+    private static void ConfigureMoltenVkDefaults()
+    {
+        try
+        {
+            _ = MacSetEnv("MVK_CONFIG_SYNCHRONOUS_QUEUE_SUBMITS", "0", 0);
+            _ = MacSetEnv("MVK_CONFIG_SHOULD_MAXIMIZE_CONCURRENT_COMPILATION", "1", 0);
+            _ = MacSetEnv("MVK_CONFIG_USE_METAL_ARGUMENT_BUFFERS", "1", 0);
+            _ = MacSetEnv("MVK_CONFIG_RESUME_LOST_DEVICE", "1", 0);
+        }
+        catch (Exception exception)
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][WARN] Failed to set MoltenVK defaults: {exception.Message}");
+        }
+    }
+
+    /// <summary>
+    /// Makes a Vulkan loader visible to GLFW's dlopen("libvulkan.1.dylib").
+    /// Homebrew's Vulkan libraries are arm64-only and cannot load into this
+    /// x86-64 (Rosetta 2) process, so a universal libMoltenVK.dylib placed
+    /// next to the executable (named libvulkan.1.dylib) is preloaded here;
+    /// dyld then resolves GLFW's bare-name dlopen to the loaded image.
+    /// </summary>
+    private static void PreloadMacVulkanLoader()
+    {
+        var candidates = new[]
+        {
+            Path.Combine(AppContext.BaseDirectory, "libvulkan.1.dylib"),
+            Path.Combine(AppContext.BaseDirectory, "libMoltenVK.dylib"),
+            Path.Combine(
+                Environment.GetFolderPath(Environment.SpecialFolder.UserProfile),
+                ".sharpemu", "x64lib", "libvulkan.1.dylib"),
+        };
+        foreach (var candidate in candidates)
+        {
+            if (File.Exists(candidate) && NativeLibrary.TryLoad(candidate, out _))
+            {
+                Console.Error.WriteLine($"[LOADER][INFO] Vulkan loader preloaded: {candidate}");
+                return;
+            }
+        }
+
+        if (NativeLibrary.TryLoad("libvulkan.1.dylib", out _))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            "[LOADER][WARN] No x86-64 Vulkan loader found; video output will be unavailable. " +
+            "Place a universal libMoltenVK.dylib (from the MoltenVK releases) next to SharpEmu " +
+            "as libvulkan.1.dylib.");
+    }
+
+    private static int RunEmulator(string[] args, bool isMitigatedChild)
+    {
         Console.Error.WriteLine($"[DEBUG] SharpEmu starting with {args.Length} args");
 
         if (!isMitigatedChild && TryRunMitigatedChild(args, out var childExitCode))
@@ -82,7 +227,17 @@ internal static partial class Program
             return childExitCode;
         }
 
-        if (!TryParseArguments(args, out var ebootPath, out var runtimeOptions, out var logLevel, out var logFilePath))
+        if (!TryExtractHostSurfaceArgument(args, out var emulatorArgs, out var hostSurface, out var hostSurfaceError))
+        {
+            Console.Error.WriteLine($"[LOADER][ERROR] {hostSurfaceError}");
+            return 1;
+        }
+
+        HostSessionControl.SetEmbeddedHostSurface(
+            hostSurface?.WindowHandle ?? 0,
+            hostSurface?.DisplayHandle ?? 0);
+
+        if (!TryParseArguments(emulatorArgs, out var ebootPath, out var runtimeOptions, out var logLevel, out var logFilePath))
         {
             PrintUsage();
             return 1;
@@ -109,66 +264,122 @@ internal static partial class Program
 
         Console.Error.WriteLine("[DEBUG] Creating runtime...");
 
-        using var runtime = SharpEmuRuntime.CreateDefault(runtimeOptions);
-
-        OrbisGen2Result result;
-        ConsoleCancelEventHandler? cancelHandler = null;
         try
         {
-            cancelHandler = (_, eventArgs) =>
+            if (hostSurface is not null && !VulkanVideoHost.TryAttachSurface(hostSurface))
             {
-                eventArgs.Cancel = true;
-                VideoOutExports.NotifyHostInterrupt();
-            };
-            Console.CancelKeyPress += cancelHandler;
+                Console.Error.WriteLine("[LOADER][ERROR] The requested GUI host surface is already active.");
+                return 3;
+            }
 
-            Console.Error.WriteLine($"[DEBUG] Running: {ebootPath}");
-            result = runtime.Run(ebootPath);
-            Console.Error.WriteLine($"[DEBUG] Result: {result}");
-        }
-        catch (Exception ex)
-        {
-            Console.Error.WriteLine($"[DEBUG] Exception: {ex}");
-            Log.Error("SharpEmu failed to run.", ex);
-            return 3;
+            using var runtime = SharpEmuRuntime.CreateDefault(runtimeOptions);
+
+            OrbisGen2Result result;
+            ConsoleCancelEventHandler? cancelHandler = null;
+            try
+            {
+                cancelHandler = (_, eventArgs) =>
+                {
+                    eventArgs.Cancel = true;
+                    VideoOutExports.NotifyHostInterrupt();
+                };
+                Console.CancelKeyPress += cancelHandler;
+
+                Console.Error.WriteLine($"[DEBUG] Running: {ebootPath}");
+                result = runtime.Run(ebootPath);
+                Console.Error.WriteLine($"[DEBUG] Result: {result}");
+            }
+            catch (Exception ex)
+            {
+                Console.Error.WriteLine($"[DEBUG] Exception: {ex}");
+                Log.Error("SharpEmu failed to run.", ex);
+                return 3;
+            }
+            finally
+            {
+                if (cancelHandler is not null)
+                {
+                    Console.CancelKeyPress -= cancelHandler;
+                }
+            }
+
+            Log.Info($"SharpEmu execution completed. Result={result} (0x{(int)result:X8})");
+            if (!string.IsNullOrWhiteSpace(runtime.LastSessionSummary))
+            {
+                Log.Info(runtime.LastSessionSummary);
+            }
+
+            if (!string.IsNullOrWhiteSpace(runtime.LastBasicBlockTrace))
+            {
+                Log.Info("BB trace:");
+                Log.Info(runtime.LastBasicBlockTrace);
+            }
+
+            if (!string.IsNullOrWhiteSpace(runtime.LastMilestoneLog))
+            {
+                Log.Info(runtime.LastMilestoneLog);
+            }
+
+            if (result != OrbisGen2Result.ORBIS_GEN2_OK && !string.IsNullOrWhiteSpace(runtime.LastExecutionDiagnostics))
+            {
+                Log.Warn(runtime.LastExecutionDiagnostics);
+            }
+
+            if (runtimeOptions.ImportTraceLimit > 0 && !string.IsNullOrWhiteSpace(runtime.LastExecutionTrace))
+            {
+                Log.Info("Import trace:");
+                Log.Info(runtime.LastExecutionTrace);
+            }
+
+            return result == OrbisGen2Result.ORBIS_GEN2_OK ? 0 : 4;
         }
         finally
         {
-            if (cancelHandler is not null)
+            HostSessionControl.SetEmbeddedHostSurface(0);
+            if (hostSurface is not null)
             {
-                Console.CancelKeyPress -= cancelHandler;
+                VulkanVideoHost.RequestClose();
+                VulkanVideoHost.DetachSurface(hostSurface);
+                hostSurface.Dispose();
+            }
+        }
+    }
+
+    private static bool TryExtractHostSurfaceArgument(
+        IReadOnlyList<string> args,
+        out string[] emulatorArgs,
+        out VulkanHostSurface? hostSurface,
+        out string? error)
+    {
+        const string hostSurfacePrefix = "--host-surface=";
+        var remaining = new List<string>(args.Count);
+        hostSurface = null;
+        error = null;
+        foreach (var argument in args)
+        {
+            if (!argument.StartsWith(hostSurfacePrefix, StringComparison.OrdinalIgnoreCase))
+            {
+                remaining.Add(argument);
+                continue;
+            }
+
+            if (hostSurface is not null)
+            {
+                emulatorArgs = [];
+                error = "more than one GUI host surface was specified";
+                return false;
+            }
+
+            var descriptor = argument[hostSurfacePrefix.Length..];
+            if (!VulkanHostSurface.TryCreateChildProcessSurface(descriptor, out hostSurface, out error))
+            {
+                emulatorArgs = [];
+                return false;
             }
         }
 
-        Log.Info($"SharpEmu execution completed. Result={result} (0x{(int)result:X8})");
-        if (!string.IsNullOrWhiteSpace(runtime.LastSessionSummary))
-        {
-            Log.Info(runtime.LastSessionSummary);
-        }
-
-        if (!string.IsNullOrWhiteSpace(runtime.LastBasicBlockTrace))
-        {
-            Log.Info("BB trace:");
-            Log.Info(runtime.LastBasicBlockTrace);
-        }
-
-        if (!string.IsNullOrWhiteSpace(runtime.LastMilestoneLog))
-        {
-            Log.Info(runtime.LastMilestoneLog);
-        }
-
-        if (result != OrbisGen2Result.ORBIS_GEN2_OK && !string.IsNullOrWhiteSpace(runtime.LastExecutionDiagnostics))
-        {
-            Log.Warn(runtime.LastExecutionDiagnostics);
-        }
-
-        if (runtimeOptions.ImportTraceLimit > 0 && !string.IsNullOrWhiteSpace(runtime.LastExecutionTrace))
-        {
-            Log.Info("Import trace:");
-            Log.Info(runtime.LastExecutionTrace);
-        }
-
-        return result == OrbisGen2Result.ORBIS_GEN2_OK ? 0 : 4;
+        emulatorArgs = remaining.ToArray();
+        return true;
     }
 
     private static void EnsureCliConsole()
@@ -265,7 +476,9 @@ internal static partial class Program
         return handle != 0 && handle != -1;
     }
 
-    private static string[] NormalizeInternalArguments(string[] args, out bool isMitigatedChild)
+    private static string[] NormalizeInternalArguments(
+        string[] args,
+        out bool isMitigatedChild)
     {
         isMitigatedChild = false;
         var trustedMitigatedChild = string.Equals(
@@ -1204,4 +1417,7 @@ internal static partial class Program
         uint creationDisposition,
         uint flagsAndAttributes,
         nint templateFile);
+
+    [DllImport("libSystem", EntryPoint = "setenv")]
+    private static extern int MacSetEnv(string name, string value, int overwrite);
 }

src/SharpEmu.CLI/SharpEmu.CLI.csproj

@@ -16,19 +16,34 @@ SPDX-License-Identifier: GPL-2.0-or-later
          console window; CLI mode re-attaches to the parent terminal's console. -->
     <OutputType>WinExe</OutputType>
     <AssemblyName>SharpEmu</AssemblyName>
-    <RuntimeIdentifiers>win-x64;linux-x64;osx-arm64</RuntimeIdentifiers>
+    <!-- osx-x64 is the macOS target: the CPU backend executes guest x86-64
+         natively, so on Apple Silicon it runs under Rosetta 2. -->
+    <RuntimeIdentifiers>win-x64;linux-x64;osx-x64;osx-arm64</RuntimeIdentifiers>
     <SelfContained>true</SelfContained>
     <PublishSingleFile>true</PublishSingleFile>
     <IncludeNativeLibrariesForSelfExtract>true</IncludeNativeLibrariesForSelfExtract>
     <EnableCompressionInSingleFile>true</EnableCompressionInSingleFile>
     <ImplicitUsings>enable</ImplicitUsings>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
-    <Version>0.0.1</Version>
-    <ServerGarbageCollection>true</ServerGarbageCollection>
+    <!-- Server GC reserves one large heap range per logical processor. On
+         high-core-count hosts those ranges can overlap fixed PS5 image bases
+         before the guest address space is established. Workstation GC avoids
+         that collision and keeps the emulator's required mappings available. -->
+    <ServerGarbageCollection>false</ServerGarbageCollection>
     <ConcurrentGarbageCollection>true</ConcurrentGarbageCollection>
     <TieredPGO>true</TieredPGO>
   </PropertyGroup>
 
+  <!-- Background GC's write-watch revisit calls FlushProcessWriteBuffers,
+       which on macOS uses thread_get_register_pointer_values; under Rosetta 2
+       that Mach call can stall indefinitely on threads executing translated
+       guest code, wedging the whole runtime (every allocating thread then
+       blocks behind the never-finishing GC). Non-concurrent GC never takes
+       that path. Windows and Linux keep concurrent GC. -->
+  <PropertyGroup Condition="$([System.String]::Copy('$(RuntimeIdentifier)').StartsWith('osx'))">
+    <ConcurrentGarbageCollection>false</ConcurrentGarbageCollection>
+  </PropertyGroup>
+
   <PropertyGroup Condition="'$(Configuration)' == 'Release'">
     <GenerateDocumentationFile>false</GenerateDocumentationFile>
     <DebugType>none</DebugType>
@@ -39,6 +54,7 @@ SPDX-License-Identifier: GPL-2.0-or-later
   <PropertyGroup Condition="'$(RuntimeIdentifier)' == 'win-x64' Or '$(RuntimeIdentifier)' == ''">
     <ApplicationIcon>..\..\assets\images\SharpEmu.ico</ApplicationIcon>
     <Win32Icon>..\..\assets\images\SharpEmu.ico</Win32Icon>
+    <ApplicationManifest>app.manifest</ApplicationManifest>
   </PropertyGroup>
 
   <PropertyGroup>
@@ -63,7 +79,7 @@ SPDX-License-Identifier: GPL-2.0-or-later
   <Target Name="KeepGlfwOutsideSingleFile" AfterTargets="ComputeResolvedFilesToPublishList">
     <ItemGroup>
       <_GlfwPublishFiles Include="@(ResolvedFileToPublish)"
-                         Condition="$([System.String]::Copy('%(ResolvedFileToPublish.Filename)').StartsWith('glfw'))" />
+                         Condition="$([System.String]::Copy('%(ResolvedFileToPublish.Filename)').StartsWith('glfw')) Or $([System.String]::Copy('%(ResolvedFileToPublish.Filename)').StartsWith('libglfw'))" />
       <ResolvedFileToPublish Remove="@(_GlfwPublishFiles)" />
       <ResolvedFileToPublish Include="@(_GlfwPublishFiles)">
         <ExcludeFromSingleFile>true</ExcludeFromSingleFile>

src/SharpEmu.CLI/app.manifest

@@ -0,0 +1,16 @@
+<?xml version="1.0" encoding="utf-8"?>
+
+<!--
+Copyright (C) 2026 SharpEmu Emulator Project
+SPDX-License-Identifier: GPL-2.0-or-later
+-->
+
+<assembly manifestVersion="1.0" xmlns="urn:schemas-microsoft-com:asm.v1">
+  <assemblyIdentity version="1.0.0.0" name="SharpEmu" />
+  <compatibility xmlns="urn:schemas-microsoft-com:compatibility.v1">
+    <application>
+      <!-- Required by Avalonia NativeControlHost on Windows 10 and 11. -->
+      <supportedOS Id="{8e0f7a12-bfb3-4fe8-b9a5-48fd50a15a9a}" />
+    </application>
+  </compatibility>
+</assembly>

src/SharpEmu.CLI/packages.lock.json

@@ -135,6 +135,23 @@
           "Ultz.Native.GLFW": "3.4.0"
         }
       },
+      "Silk.NET.Input.Common": {
+        "type": "Transitive",
+        "resolved": "2.23.0",
+        "contentHash": "QbJVV7kFBHEByayXCYdJtXXI9Sp4a+QAf0IdGV6uCWkFYcEmqBYW3aaNGvFOdSwTBDbHL5T/OtOCrGh4qYhk7A==",
+        "dependencies": {
+          "Silk.NET.Windowing.Common": "2.23.0"
+        }
+      },
+      "Silk.NET.Input.Glfw": {
+        "type": "Transitive",
+        "resolved": "2.23.0",
+        "contentHash": "KGHYqsv/IQRJtD6dloYh2tN4CkaM40vxM2kj0cGKBoCQiBDYHHhJiyDTyMPx0W7Fz5IgnhnG42ELmIAa0DH69A==",
+        "dependencies": {
+          "Silk.NET.Input.Common": "2.23.0",
+          "Silk.NET.Windowing.Glfw": "2.23.0"
+        }
+      },
       "Silk.NET.Maths": {
         "type": "Transitive",
         "resolved": "2.23.0",
@@ -199,9 +216,9 @@
         "type": "Project",
         "dependencies": {
           "Iced": "[1.21.0, )",
-          "SharpEmu.HLE": "[1.0.0, )",
-          "SharpEmu.Libs": "[1.0.0, )",
-          "SharpEmu.Logging": "[1.0.0, )"
+          "SharpEmu.HLE": "[0.0.2-beta.2, )",
+          "SharpEmu.Libs": "[0.0.2-beta.2, )",
+          "SharpEmu.Logging": "[0.0.2-beta.2, )"
         }
       },
       "sharpemu.gui": {
@@ -211,20 +228,25 @@
           "Avalonia.Desktop": "[11.3.18, )",
           "Avalonia.Fonts.Inter": "[11.3.18, )",
           "Avalonia.Themes.Fluent": "[11.3.18, )",
-          "SharpEmu.Logging": "[1.0.0, )",
+          "SharpEmu.Core": "[0.0.2-beta.2, )",
+          "SharpEmu.Libs": "[0.0.2-beta.2, )",
+          "SharpEmu.Logging": "[0.0.2-beta.2, )",
           "Tmds.DBus.Protocol": "[0.21.3, )"
         }
       },
       "sharpemu.hle": {
         "type": "Project",
         "dependencies": {
-          "SharpEmu.Logging": "[1.0.0, )"
+          "SharpEmu.Logging": "[0.0.2-beta.2, )"
         }
       },
       "sharpemu.libs": {
         "type": "Project",
         "dependencies": {
-          "SharpEmu.HLE": "[1.0.0, )",
+          "SharpEmu.HLE": "[0.0.2-beta.2, )",
+          "SharpEmu.ShaderCompiler": "[0.0.2-beta.2, )",
+          "SharpEmu.ShaderCompiler.Vulkan": "[0.0.2-beta.2, )",
+          "Silk.NET.Input": "[2.23.0, )",
           "Silk.NET.Vulkan": "[2.23.0, )",
           "Silk.NET.Vulkan.Extensions.EXT": "[2.23.0, )",
           "Silk.NET.Vulkan.Extensions.KHR": "[2.23.0, )",
@@ -234,6 +256,18 @@
       "sharpemu.logging": {
         "type": "Project"
       },
+      "sharpemu.shadercompiler": {
+        "type": "Project",
+        "dependencies": {
+          "SharpEmu.HLE": "[0.0.2-beta.2, )"
+        }
+      },
+      "sharpemu.shadercompiler.vulkan": {
+        "type": "Project",
+        "dependencies": {
+          "SharpEmu.ShaderCompiler": "[0.0.2-beta.2, )"
+        }
+      },
       "Avalonia": {
         "type": "CentralTransitive",
         "requested": "[11.3.18, )",
@@ -282,6 +316,16 @@
         "resolved": "1.21.0",
         "contentHash": "dv5+81Q1TBQvVMSOOOmRcjJmvWcX3BZPZsIq31+RLc5cNft0IHAyNlkdb7ZarOWG913PyBoFDsDXoCIlKmLclg=="
       },
+      "Silk.NET.Input": {
+        "type": "CentralTransitive",
+        "requested": "[2.23.0, )",
+        "resolved": "2.23.0",
+        "contentHash": "Xzl+tVwAp2eEd8blmGQjmJrsZPnp3PWG0KJjiAQHaY2Zr/ELVWeAROKXmZdCAvexzmte2JVGEy/dxnMycbxlpg==",
+        "dependencies": {
+          "Silk.NET.Input.Common": "2.23.0",
+          "Silk.NET.Input.Glfw": "2.23.0"
+        }
+      },
       "Silk.NET.Vulkan": {
         "type": "CentralTransitive",
         "requested": "[2.23.0, )",
@@ -434,6 +478,59 @@
         "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
       }
     },
+    "net10.0/osx-x64": {
+      "Avalonia.Angle.Windows.Natives": {
+        "type": "Transitive",
+        "resolved": "2.1.25547.20250602",
+        "contentHash": "ZL0VLc4s9rvNNFt19Pxm5UNAkmKNylugAwJPX9ulXZ6JWs/l6XZihPWWTyezaoNOVyEPU8YbURtW7XMAtqXH5A=="
+      },
+      "Avalonia.Native": {
+        "type": "Transitive",
+        "resolved": "11.3.18",
+        "contentHash": "8g53DROFW6wVJAnTsE1Iu5bdZO5r0oqxbdbMQODs1QDYCK2IU/y/x/vQVKCYOvqxl4tXkzU9tExv8XqSGPWthQ==",
+        "dependencies": {
+          "Avalonia": "11.3.18"
+        }
+      },
+      "HarfBuzzSharp.NativeAssets.Linux": {
+        "type": "Transitive",
+        "resolved": "8.3.1.1",
+        "contentHash": "3EZ1mpIiKWRLL5hUYA82ZHteeDIVaEA/Z0rA/wU6tjx6crcAkJnBPwDXZugBSfo8+J3EznvRJf49uMsqYfKrHg=="
+      },
+      "HarfBuzzSharp.NativeAssets.macOS": {
+        "type": "Transitive",
+        "resolved": "8.3.1.1",
+        "contentHash": "jbtCsgftcaFLCA13tVKo5iWdElJScrulLTKJre36O4YQTIlwDtPPqhRZNk+Y0vv4D1gxbscasGRucUDfS44ofQ=="
+      },
+      "HarfBuzzSharp.NativeAssets.Win32": {
+        "type": "Transitive",
+        "resolved": "8.3.1.1",
+        "contentHash": "UsJtQsfAJoFDZrXc4hCUfRPMqccfKZ0iumJ/upcUjz/cmsTgVFGNEL5yaJWmkqsuFYdMWbj/En5/kS4PFl9hBA=="
+      },
+      "SkiaSharp.NativeAssets.Linux": {
+        "type": "Transitive",
+        "resolved": "2.88.9",
+        "contentHash": "cWSaJKVPWAaT/WIn9c8T5uT/l4ETwHxNJTkEOtNKjphNo8AW6TF9O32aRkxqw3l8GUdUo66Bu7EiqtFh/XG0Zg==",
+        "dependencies": {
+          "SkiaSharp": "2.88.9"
+        }
+      },
+      "SkiaSharp.NativeAssets.macOS": {
+        "type": "Transitive",
+        "resolved": "2.88.9",
+        "contentHash": "Nv5spmKc4505Ep7oUoJ5vp3KweFpeNqxpyGDWyeEPTX2uR6S6syXIm3gj75dM0YJz7NPvcix48mR5laqs8dPuA=="
+      },
+      "SkiaSharp.NativeAssets.Win32": {
+        "type": "Transitive",
+        "resolved": "2.88.9",
+        "contentHash": "wb2kYgU7iy84nQLYZwMeJXixvK++GoIuECjU4ECaUKNuflyRlJKyiRhN1MAHswvlvzuvkrjRWlK0Za6+kYQK7w=="
+      },
+      "Ultz.Native.GLFW": {
+        "type": "Transitive",
+        "resolved": "3.4.0",
+        "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
+      }
+    },
     "net10.0/win-x64": {
       "Avalonia.Angle.Windows.Natives": {
         "type": "Transitive",

src/SharpEmu.Core/Cpu/CpuDispatcher.cs

@@ -7,30 +7,33 @@ using SharpEmu.Core.Cpu.Native;
 using SharpEmu.Core.Loader;
 using SharpEmu.Core.Memory;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
-using SharpEmu.Logging;
 
 namespace SharpEmu.Core.Cpu;
 
 public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
 {
-    private static readonly SharpEmuLogger Log = SharpEmuLog.For("Dispatcher");
-
     private enum EntryFrameKind
     {
         ProcessEntry,
         ModuleInitializer,
     }
 
-    private const ulong StackBaseAddress = 0x7FFF_F000_0000UL;
+    // The top of the x86-64 user address space (0x7FFD..0x7FFF) is only
+    // freely mappable on Windows; on macOS/Linux it hosts the dyld shared
+    // cache / vdso and (under Rosetta 2) the translator runtime, so POSIX
+    // hosts use the equivalent layout one slot lower at 0x6FFx.
+    private static readonly ulong StackBaseAddress = OperatingSystem.IsWindows() ? 0x7FFF_F000_0000UL : 0x6FFF_F000_0000UL;
     private const ulong StackSize = 0x0020_0000UL;
-    private const ulong TlsBaseAddress = 0x7FFE_0000_0000UL;
+    private static readonly ulong TlsBaseAddress = OperatingSystem.IsWindows() ? 0x7FFE_0000_0000UL : 0x6FFE_0000_0000UL;
     private const ulong TlsSize = 0x0001_0000UL;
-    private const ulong TlsPrefixSize = 0x0000_1000UL;
-    private const ulong BootstrapStubBaseAddress = 0x7FFD_F000_0000UL;
-    private const ulong BootstrapPayloadBaseAddress = 0x7FFD_E000_0000UL;
-    private const ulong DynlibFallbackStubBaseAddress = 0x7FFD_D000_0000UL;
-    private const ulong ReturnToHostStubBaseAddress = 0x7FFD_C000_0000UL;
+    // The static TLS blocks live at negative offsets from the TCB (FreeBSD
+    // amd64 variant II). Keep every host in sync with GuestTlsTemplate's
+    // startup reservation; PS5 modules routinely reach beyond one host page.
+    private const ulong TlsPrefixSize = GuestTlsTemplate.StartupStaticTlsReservation;
+    private static readonly ulong BootstrapStubBaseAddress = OperatingSystem.IsWindows() ? 0x7FFD_F000_0000UL : 0x6FFD_F000_0000UL;
+    private static readonly ulong BootstrapPayloadBaseAddress = OperatingSystem.IsWindows() ? 0x7FFD_E000_0000UL : 0x6FFD_E000_0000UL;
+    private static readonly ulong DynlibFallbackStubBaseAddress = OperatingSystem.IsWindows() ? 0x7FFD_D000_0000UL : 0x6FFD_D000_0000UL;
+    private static readonly ulong ReturnToHostStubBaseAddress = OperatingSystem.IsWindows() ? 0x7FFD_C000_0000UL : 0x6FFD_C000_0000UL;
     private const ulong BootstrapRegionSize = 0x0000_1000UL;
     private const ulong ReturnToHostStubStride = 0x0100_0000UL;
     private const ulong BootstrapPayloadResultOffset = 0x28UL;
@@ -42,19 +45,16 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
     ];
     private readonly IVirtualMemory _virtualMemory;
     private readonly IModuleManager _moduleManager;
-    private readonly IHostPlatform? _hostPlatform;
     private INativeCpuBackend? _nativeCpuBackend;
 
     public CpuDispatcher(
         IVirtualMemory virtualMemory,
         IModuleManager moduleManager,
-        INativeCpuBackend? nativeCpuBackend = null,
-        IHostPlatform? hostPlatform = null)
+        INativeCpuBackend? nativeCpuBackend = null)
     {
         _virtualMemory = virtualMemory ?? throw new ArgumentNullException(nameof(virtualMemory));
         _moduleManager = moduleManager ?? throw new ArgumentNullException(nameof(moduleManager));
         _nativeCpuBackend = nativeCpuBackend;
-        _hostPlatform = hostPlatform;
     }
 
     public ulong? LastEntryPoint { get; private set; }
@@ -87,8 +87,8 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         string processImageName = "eboot.bin",
         CpuExecutionOptions executionOptions = default)
     {
-        Log.Debug("=== DispatchEntry START ===");
-        Log.Debug($"entryPoint=0x{entryPoint:X16}, generation={generation}");
+        Console.Error.WriteLine("[DISPATCHER] === DispatchEntry START ===");
+        Console.Error.WriteLine($"[DISPATCHER] entryPoint=0x{entryPoint:X16}, generation={generation}");
 
         try
         {
@@ -96,7 +96,8 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         }
         catch (Exception ex)
         {
-            Log.Critical($"FATAL EXCEPTION in DispatchEntry: {ex.GetType().Name}: {ex.Message}", ex);
+            Console.Error.WriteLine($"[DISPATCHER] FATAL EXCEPTION in DispatchEntry: {ex.GetType().Name}: {ex.Message}");
+            Console.Error.WriteLine($"[DISPATCHER] Stack trace: {ex.StackTrace}");
             throw;
         }
     }
@@ -109,8 +110,8 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         string moduleName = "module",
         CpuExecutionOptions executionOptions = default)
     {
-        Log.Debug("=== DispatchModuleInitializer START ===");
-        Log.Debug($"moduleInit=0x{entryPoint:X16}, generation={generation}, module={moduleName}");
+        Console.Error.WriteLine("[DISPATCHER] === DispatchModuleInitializer START ===");
+        Console.Error.WriteLine($"[DISPATCHER] moduleInit=0x{entryPoint:X16}, generation={generation}, module={moduleName}");
 
         try
         {
@@ -125,7 +126,8 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         }
         catch (Exception ex)
         {
-            Log.Critical($"FATAL EXCEPTION in DispatchModuleInitializer: {ex.GetType().Name}: {ex.Message}", ex);
+            Console.Error.WriteLine($"[DISPATCHER] FATAL EXCEPTION in DispatchModuleInitializer: {ex.GetType().Name}: {ex.Message}");
+            Console.Error.WriteLine($"[DISPATCHER] Stack trace: {ex.StackTrace}");
             throw;
         }
     }
@@ -139,7 +141,7 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         CpuExecutionOptions executionOptions = default,
         EntryFrameKind frameKind = EntryFrameKind.ProcessEntry)
     {
-        Log.Debug("DispatchEntryCore STARTING...");
+        Console.Error.WriteLine("[DISPATCHER] DispatchEntryCore STARTING...");
 
         LastEntryPoint = entryPoint;
         LastTrapInfo = null;
@@ -270,7 +272,7 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
             entryFrameDiagnostic,
             Environment.NewLine,
             "CpuEngine: native-only");
-        _nativeCpuBackend ??= new DirectExecutionBackend(_moduleManager, _hostPlatform);
+        _nativeCpuBackend ??= new DirectExecutionBackend(_moduleManager);
         if (_nativeCpuBackend.TryExecute(
                 context,
                 entryPoint,
@@ -311,7 +313,7 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
             LastMilestoneLog,
             Environment.NewLine,
             $"CpuEngine native-only failed: {backendError}");
-        Log.Error($"Native backend FAILED: {backendError}");
+        Console.Error.WriteLine($"[DISPATCHER] Native backend FAILED: {backendError}");
         return FailEarly(
             OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_IMPLEMENTED,
             CpuExitReason.NativeBackendUnavailable);
@@ -370,11 +372,19 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
 
     private static bool InitializeTls(CpuContext context, ulong tlsBase)
     {
-        return context.TryWriteUInt64(tlsBase - 0xF0, 0) &&
-               context.TryWriteUInt64(tlsBase + 0x00, tlsBase) &&
-               context.TryWriteUInt64(tlsBase + 0x10, tlsBase) &&
-               context.TryWriteUInt64(tlsBase + 0x28, 0xC0DEC0DECAFEBABEUL) &&
-               context.TryWriteUInt64(tlsBase + 0x60, tlsBase);
+        if (!context.TryWriteUInt64(tlsBase - 0xF0, 0) ||
+            !context.TryWriteUInt64(tlsBase + 0x00, tlsBase) ||
+            !context.TryWriteUInt64(tlsBase + 0x10, tlsBase) ||
+            !context.TryWriteUInt64(tlsBase + 0x28, 0xC0DEC0DECAFEBA00UL) ||
+            !context.TryWriteUInt64(tlsBase + 0x60, tlsBase))
+        {
+            return false;
+        }
+
+        // Seed the static TLS block below the thread pointer with the main
+        // module's initialized thread-locals (variant II layout).
+        SharpEmu.HLE.GuestTlsTemplate.SeedThreadBlock(context, tlsBase);
+        return true;
     }
 
     private static bool InitializeGuestFrameChainSentinel(CpuContext context)
@@ -400,35 +410,54 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         ulong programExitHandlerAddress)
     {
         var imageName = string.IsNullOrWhiteSpace(processImageName) ? "eboot.bin" : processImageName;
-        var encodedNameLength = Encoding.UTF8.GetByteCount(imageName);
-        Span<byte> argv0Buffer = encodedNameLength + 1 <= 512
-            ? stackalloc byte[encodedNameLength + 1]
-            : new byte[encodedNameLength + 1];
-        if (Encoding.UTF8.GetBytes(imageName.AsSpan(), argv0Buffer) != encodedNameLength)
+        var arguments = new List<string>(3) { imageName };
+        var configuredArguments = Environment.GetEnvironmentVariable("SHARPEMU_GUEST_ARGS");
+        if (!string.IsNullOrWhiteSpace(configuredArguments))
         {
-            return false;
+            // The PS5 entry-parameter ABI exposes three inline argv pointers.
+            // Two compatibility arguments are therefore safe without changing
+            // the fixed 0x20-byte structure expected by existing titles.
+            var compatibilityArguments = configuredArguments.Split(
+                (char[]?)null,
+                StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries);
+            arguments.AddRange(compatibilityArguments.Take(2));
         }
 
-        argv0Buffer[encodedNameLength] = 0;
         var cursor = context[CpuRegister.Rsp];
-
-        var argv0Address = AlignDown(cursor - (ulong)argv0Buffer.Length, 16);
-        if (!context.Memory.TryWrite(argv0Address, argv0Buffer))
+        var argumentAddresses = new ulong[arguments.Count];
+        for (var index = arguments.Count - 1; index >= 0; index--)
         {
-            return false;
+            var encoded = Encoding.UTF8.GetBytes(arguments[index] + '\0');
+            cursor = AlignDown(cursor - (ulong)encoded.Length, 16);
+            if (!context.Memory.TryWrite(cursor, encoded))
+            {
+                return false;
+            }
+
+            argumentAddresses[index] = cursor;
         }
 
         const ulong entryParamsSize = 0x20;
-        var entryParamsAddress = AlignDown(argv0Address - entryParamsSize, 16);
-        if (!context.TryWriteUInt32(entryParamsAddress + 0x00, 1) ||
-            !context.TryWriteUInt32(entryParamsAddress + 0x04, 0) ||
-            !context.TryWriteUInt64(entryParamsAddress + 0x08, argv0Address) ||
-            !context.TryWriteUInt64(entryParamsAddress + 0x10, 0) ||
-            !context.TryWriteUInt64(entryParamsAddress + 0x18, 0))
+        var entryParamsAddress = AlignDown(cursor - entryParamsSize, 16);
+        if (!TryWriteUInt32(context, entryParamsAddress + 0x00, (uint)arguments.Count) ||
+            !TryWriteUInt32(context, entryParamsAddress + 0x04, 0) ||
+            !context.TryWriteUInt64(entryParamsAddress + 0x08, argumentAddresses[0]) ||
+            !context.TryWriteUInt64(
+                entryParamsAddress + 0x10,
+                argumentAddresses.Length > 1 ? argumentAddresses[1] : 0) ||
+            !context.TryWriteUInt64(
+                entryParamsAddress + 0x18,
+                argumentAddresses.Length > 2 ? argumentAddresses[2] : 0))
         {
             return false;
         }
 
+        if (arguments.Count > 1)
+        {
+            Console.Error.WriteLine(
+                $"[DISPATCHER] Guest arguments: {string.Join(' ', arguments.Skip(1))}");
+        }
+
         var entryStackPointer = entryParamsAddress - sizeof(ulong);
         if (!context.TryWriteUInt64(entryStackPointer, 0))
         {
@@ -461,6 +490,13 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         return value & ~(alignment - 1);
     }
 
+    private static bool TryWriteUInt32(CpuContext context, ulong address, uint value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(buffer, value);
+        return context.Memory.TryWrite(address, buffer);
+    }
+
     private static string BuildEntryFrameDiagnostic(
         ulong entryPoint,
         CpuContext context,
@@ -657,6 +693,13 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         return _virtualMemory.TryWrite(address, buffer);
     }
 
+    /// <summary>
+    /// True when the disposed native backend left its session state alive
+    /// because guest workers were still executing guest code. The guest
+    /// address space must then stay mapped as well.
+    /// </summary>
+    internal bool NativeSessionLeaked { get; private set; }
+
     public void Dispose()
     {
         if (_nativeCpuBackend is IDisposable disposableBackend)
@@ -664,6 +707,7 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
             disposableBackend.Dispose();
         }
 
+        NativeSessionLeaked = _nativeCpuBackend is DirectExecutionBackend { GuestSessionLeaked: true };
         _nativeCpuBackend = null;
     }
 }

src/SharpEmu.Core/Cpu/Emulation/BmiInstructionEmulator.cs

@@ -0,0 +1,305 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Numerics;
+
+namespace SharpEmu.Core.Cpu.Emulation;
+
+/// <summary>
+/// Pure software implementations of the BMI1, BMI2 and ABM general-purpose-register
+/// bit-manipulation instructions.
+///
+/// The direct-execution backend runs guest PS5 code natively on the host CPU. The PS5's
+/// Zen 2 cores implement BMI1/BMI2/ABM, but a host CPU that predates those extensions raises
+/// #UD (STATUS_ILLEGAL_INSTRUCTION) when it meets one of these opcodes. This class provides the
+/// register-only arithmetic so the exception handler can finish the instruction in software and
+/// resume, instead of aborting the title.
+///
+/// The methods deliberately operate on plain integers rather than on the OS CONTEXT record so the
+/// semantics can be unit-tested in isolation; the unsafe register/memory plumbing lives in the
+/// backend adapter. Each method returns the result already masked to the operand width and, where
+/// the instruction is defined to touch flags, updates <paramref name="eflags"/> in place. Flags
+/// documented as "undefined" by the vendor manuals are left untouched so behaviour stays
+/// deterministic across hosts.
+/// </summary>
+public static class BmiInstructionEmulator
+{
+    private const uint FlagCarry = 1u << 0;
+    private const uint FlagZero = 1u << 6;
+    private const uint FlagSign = 1u << 7;
+    private const uint FlagOverflow = 1u << 11;
+
+    private static ulong WidthMask(GprOperandSize size) =>
+        size == GprOperandSize.Bits64 ? ulong.MaxValue : 0xFFFF_FFFFUL;
+
+    private static int WidthBits(GprOperandSize size) => (int)size;
+
+    private static bool SignSet(ulong value, GprOperandSize size) =>
+        size == GprOperandSize.Bits64 ? (value >> 63) != 0 : (value & 0x8000_0000UL) != 0;
+
+    private static uint WithFlag(uint eflags, uint flag, bool set) =>
+        set ? eflags | flag : eflags & ~flag;
+
+    // Applies the CF/ZF/SF/OF set shared by ANDN/BLS*/BZHI: OF is always cleared, ZF and SF follow
+    // the result, and the caller supplies CF because each instruction defines it differently.
+    private static uint ApplyLogicFlags(uint eflags, ulong result, GprOperandSize size, bool carry)
+    {
+        eflags = WithFlag(eflags, FlagCarry, carry);
+        eflags = WithFlag(eflags, FlagZero, result == 0);
+        eflags = WithFlag(eflags, FlagSign, SignSet(result, size));
+        eflags = WithFlag(eflags, FlagOverflow, false);
+        return eflags;
+    }
+
+    /// <summary>ANDN: <c>dest = (~src1) &amp; src2</c>. CF and OF are cleared.</summary>
+    public static ulong Andn(ulong src1, ulong src2, GprOperandSize size, ref uint eflags)
+    {
+        var result = (~src1 & src2) & WidthMask(size);
+        eflags = ApplyLogicFlags(eflags, result, size, carry: false);
+        return result;
+    }
+
+    /// <summary>BLSI: isolate the lowest set bit, <c>dest = (-src) &amp; src</c>. CF = (src != 0).</summary>
+    public static ulong Blsi(ulong src, GprOperandSize size, ref uint eflags)
+    {
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var result = ((0UL - s) & s) & mask;
+        eflags = ApplyLogicFlags(eflags, result, size, carry: s != 0);
+        return result;
+    }
+
+    /// <summary>BLSMSK: mask up to and including the lowest set bit, <c>dest = (src - 1) ^ src</c>. CF = (src == 0).</summary>
+    public static ulong Blsmsk(ulong src, GprOperandSize size, ref uint eflags)
+    {
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var result = ((s - 1) ^ s) & mask;
+        eflags = ApplyLogicFlags(eflags, result, size, carry: s == 0);
+        return result;
+    }
+
+    /// <summary>BLSR: reset the lowest set bit, <c>dest = (src - 1) &amp; src</c>. CF = (src == 0).</summary>
+    public static ulong Blsr(ulong src, GprOperandSize size, ref uint eflags)
+    {
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var result = ((s - 1) & s) & mask;
+        eflags = ApplyLogicFlags(eflags, result, size, carry: s == 0);
+        return result;
+    }
+
+    /// <summary>
+    /// BEXTR: extract <c>len</c> bits of <paramref name="src"/> starting at bit <c>start</c>, where
+    /// start = control[7:0] and len = control[15:8]. Only ZF (per result) and cleared CF/OF are defined.
+    /// </summary>
+    public static ulong Bextr(ulong src, ulong control, GprOperandSize size, ref uint eflags)
+    {
+        var bits = WidthBits(size);
+        var start = (int)(control & 0xFF);
+        var length = (int)((control >> 8) & 0xFF);
+
+        ulong result;
+        if (start >= bits)
+        {
+            result = 0;
+        }
+        else
+        {
+            var shifted = (src & WidthMask(size)) >> start;
+            if (length == 0)
+            {
+                result = 0;
+            }
+            else if (length >= bits)
+            {
+                result = shifted;
+            }
+            else
+            {
+                result = shifted & ((1UL << length) - 1);
+            }
+        }
+
+        result &= WidthMask(size);
+        eflags = WithFlag(eflags, FlagZero, result == 0);
+        eflags = WithFlag(eflags, FlagCarry, false);
+        eflags = WithFlag(eflags, FlagOverflow, false);
+        return result;
+    }
+
+    /// <summary>
+    /// BZHI: zero the bits of <paramref name="src"/> from bit position <c>index[7:0]</c> upward.
+    /// CF is set when the requested position is at or beyond the operand width.
+    /// </summary>
+    public static ulong Bzhi(ulong src, ulong index, GprOperandSize size, ref uint eflags)
+    {
+        var bits = WidthBits(size);
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var n = (int)(index & 0xFF);
+
+        ulong result;
+        bool carry;
+        if (n >= bits)
+        {
+            result = s;
+            carry = true;
+        }
+        else
+        {
+            result = s & ((1UL << n) - 1);
+            carry = false;
+        }
+
+        eflags = ApplyLogicFlags(eflags, result, size, carry);
+        return result;
+    }
+
+    /// <summary>TZCNT: count trailing zero bits. If src == 0 the result is the operand width and CF is set.</summary>
+    public static ulong Tzcnt(ulong src, GprOperandSize size, ref uint eflags)
+    {
+        var bits = WidthBits(size);
+        var s = src & WidthMask(size);
+
+        ulong result;
+        bool carry;
+        if (s == 0)
+        {
+            result = (ulong)bits;
+            carry = true;
+        }
+        else
+        {
+            result = (ulong)(size == GprOperandSize.Bits64
+                ? BitOperations.TrailingZeroCount(s)
+                : BitOperations.TrailingZeroCount((uint)s));
+            carry = false;
+        }
+
+        eflags = WithFlag(eflags, FlagCarry, carry);
+        eflags = WithFlag(eflags, FlagZero, result == 0);
+        return result;
+    }
+
+    /// <summary>LZCNT: count leading zero bits. If src == 0 the result is the operand width and CF is set.</summary>
+    public static ulong Lzcnt(ulong src, GprOperandSize size, ref uint eflags)
+    {
+        var bits = WidthBits(size);
+        var s = src & WidthMask(size);
+
+        ulong result;
+        bool carry;
+        if (s == 0)
+        {
+            result = (ulong)bits;
+            carry = true;
+        }
+        else
+        {
+            result = (ulong)(size == GprOperandSize.Bits64
+                ? BitOperations.LeadingZeroCount(s)
+                : BitOperations.LeadingZeroCount((uint)s));
+            carry = false;
+        }
+
+        eflags = WithFlag(eflags, FlagCarry, carry);
+        eflags = WithFlag(eflags, FlagZero, result == 0);
+        return result;
+    }
+
+    /// <summary>RORX: rotate <paramref name="src"/> right by <paramref name="count"/> (masked to the operand width). No flags.</summary>
+    public static ulong Rorx(ulong src, int count, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var rotate = count & (bits - 1);
+        if (rotate == 0)
+        {
+            return s;
+        }
+
+        return ((s >> rotate) | (s << (bits - rotate))) & mask;
+    }
+
+    /// <summary>SARX: arithmetic shift right by <paramref name="count"/> (masked to the operand width). No flags.</summary>
+    public static ulong Sarx(ulong src, int count, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var mask = WidthMask(size);
+        var shift = count & (bits - 1);
+        if (size == GprOperandSize.Bits64)
+        {
+            return (ulong)((long)src >> shift);
+        }
+
+        return (ulong)(uint)((int)(uint)src >> shift) & mask;
+    }
+
+    /// <summary>SHLX: logical shift left by <paramref name="count"/> (masked to the operand width). No flags.</summary>
+    public static ulong Shlx(ulong src, int count, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var mask = WidthMask(size);
+        var shift = count & (bits - 1);
+        return (src << shift) & mask;
+    }
+
+    /// <summary>SHRX: logical shift right by <paramref name="count"/> (masked to the operand width). No flags.</summary>
+    public static ulong Shrx(ulong src, int count, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var mask = WidthMask(size);
+        var s = src & mask;
+        var shift = count & (bits - 1);
+        return s >> shift;
+    }
+
+    /// <summary>PDEP: deposit contiguous low bits of <paramref name="src"/> into the positions selected by <paramref name="mask"/>. No flags.</summary>
+    public static ulong Pdep(ulong src, ulong mask, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var selector = mask & WidthMask(size);
+        ulong result = 0;
+        var bit = 0;
+        for (var i = 0; i < bits; i++)
+        {
+            var position = 1UL << i;
+            if ((selector & position) != 0)
+            {
+                if (((src >> bit) & 1UL) != 0)
+                {
+                    result |= position;
+                }
+
+                bit++;
+            }
+        }
+
+        return result & WidthMask(size);
+    }
+
+    /// <summary>PEXT: gather the bits of <paramref name="src"/> selected by <paramref name="mask"/> into contiguous low bits. No flags.</summary>
+    public static ulong Pext(ulong src, ulong mask, GprOperandSize size)
+    {
+        var bits = WidthBits(size);
+        var selector = mask & WidthMask(size);
+        ulong result = 0;
+        var bit = 0;
+        for (var i = 0; i < bits; i++)
+        {
+            if ((selector & (1UL << i)) != 0)
+            {
+                if (((src >> i) & 1UL) != 0)
+                {
+                    result |= 1UL << bit;
+                }
+
+                bit++;
+            }
+        }
+
+        return result & WidthMask(size);
+    }
+}

src/SharpEmu.Core/Cpu/Emulation/GprOperandSize.cs

@@ -0,0 +1,14 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.Core.Cpu.Emulation;
+
+/// <summary>
+/// Operand width for the emulated general-purpose-register instructions. The numeric value is the
+/// bit width, so it can double as the "count leading/trailing zeros of an all-zero source" result.
+/// </summary>
+public enum GprOperandSize
+{
+    Bits32 = 32,
+    Bits64 = 64,
+}

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.Diagnostics.cs

@@ -7,8 +7,8 @@ using System.Collections.Generic;
 using System.Linq;
 using System.Runtime.InteropServices;
 using System.Threading;
+using SharpEmu.Core.Cpu.Disasm;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
 using SharpEmu.Logging;
 
 namespace SharpEmu.Core.Cpu.Native;
@@ -17,6 +17,55 @@ public sealed partial class DirectExecutionBackend
 {
 	private static readonly ConcurrentDictionary<ulong, byte> _knownExecutablePages = new();
 
+	private static readonly bool _perfHleHistogram =
+		string.Equals(System.Environment.GetEnvironmentVariable("SHARPEMU_PERF_HLE"), "1", System.StringComparison.Ordinal);
+	private static readonly System.Collections.Concurrent.ConcurrentDictionary<string, long> _perfHleCounts = new();
+	private static long _perfHleTotal;
+	private static long _perfHleDispatchTicks;
+
+	private static void RecordPerfHleDispatchTime(long ticks)
+	{
+		var total = System.Threading.Interlocked.Add(ref _perfHleDispatchTicks, ticks);
+		var calls = System.Threading.Interlocked.Read(ref _perfHleTotal);
+		if (calls > 0 && calls % 500000 == 0)
+		{
+			var avgUs = (double)total / System.Diagnostics.Stopwatch.Frequency * 1_000_000.0 / calls;
+			System.Console.Error.WriteLine($"[PERF][HLE] managed_dispatch_avg={avgUs:F3}us total_managed_s={(double)total / System.Diagnostics.Stopwatch.Frequency:F2}");
+		}
+	}
+
+	private static readonly bool _perfHleNoDict =
+		string.Equals(System.Environment.GetEnvironmentVariable("SHARPEMU_PERF_HLE_NODICT"), "1", System.StringComparison.Ordinal);
+
+	private static void RecordPerfHleCall(string name)
+	{
+		var total = System.Threading.Interlocked.Increment(ref _perfHleTotal);
+		if (!_perfHleNoDict)
+		{
+			_perfHleCounts.AddOrUpdate(name, 1, static (_, v) => v + 1);
+		}
+
+		if (total % 500000 == 0 && !_perfHleNoDict)
+		{
+			// Snapshot via foreach (a safe moving enumerator) before sorting.
+			// LINQ over a ConcurrentDictionary uses ICollection.CopyTo, which
+			// throws ArgumentException if another thread adds a key between the
+			// Count read and the copy — that exception was being swallowed into
+			// a CPU_TRAP return and crashing the guest.
+			var snapshot = new System.Collections.Generic.List<System.Collections.Generic.KeyValuePair<string, long>>(_perfHleCounts.Count + 16);
+			foreach (var kvp in _perfHleCounts)
+			{
+				snapshot.Add(kvp);
+			}
+
+			var top = snapshot
+				.OrderByDescending(kvp => kvp.Value)
+				.Take(20)
+				.Select(kvp => $"{kvp.Key}={kvp.Value}");
+			System.Console.Error.WriteLine($"[PERF][HLE] total={total} top: {string.Join(", ", top)}");
+		}
+	}
+
 	private void RecordRecentImportTrace(
 		long dispatchIndex,
 		string nid,
@@ -25,101 +74,40 @@ public sealed partial class DirectExecutionBackend
 		ulong arg1,
 		ulong arg2)
 	{
-		_recentImportTrace[_recentImportTraceWriteIndex] = new RecentImportTraceEntry(
+		var trace = _recentImportTrace;
+		trace[_recentImportTraceWriteIndex] = new RecentImportTraceEntry(
 			dispatchIndex,
 			nid,
 			returnRip,
 			arg0,
 			arg1,
-			arg2);
-		_recentImportTraceWriteIndex = (_recentImportTraceWriteIndex + 1) % _recentImportTrace.Length;
-		if (_recentImportTraceCount < _recentImportTrace.Length)
+			arg2,
+			GuestThreadExecution.CurrentGuestThreadHandle,
+			Environment.CurrentManagedThreadId);
+		_recentImportTraceWriteIndex = (_recentImportTraceWriteIndex + 1) % trace.Length;
+		if (_recentImportTraceCount < trace.Length)
 		{
 			_recentImportTraceCount++;
 		}
 	}
 
-	private void RecordDeferredBootstrapTrace(
-		long dispatchIndex,
-		ulong op,
-		ulong symbolPointer,
-		ulong outputPointer,
-		ulong returnRip)
-	{
-		lock (_deferredBootstrapTraceGate)
-		{
-			_deferredBootstrapTrace[_deferredBootstrapTraceWriteIndex] = new DeferredBootstrapTraceEntry(
-				dispatchIndex,
-				op,
-				symbolPointer,
-				outputPointer,
-				returnRip);
-			_deferredBootstrapTraceWriteIndex =
-				(_deferredBootstrapTraceWriteIndex + 1) % _deferredBootstrapTrace.Length;
-			if (_deferredBootstrapTraceCount < _deferredBootstrapTrace.Length)
-			{
-				_deferredBootstrapTraceCount++;
-			}
-		}
-	}
-
-	private void DrainDeferredBootstrapTraces()
-	{
-		if (!_logBootstrap)
-		{
-			return;
-		}
-
-		DeferredBootstrapTraceEntry[] pending;
-		lock (_deferredBootstrapTraceGate)
-		{
-			if (_deferredBootstrapTraceCount == 0)
-			{
-				return;
-			}
-
-			pending = new DeferredBootstrapTraceEntry[_deferredBootstrapTraceCount];
-			var readIndex = (_deferredBootstrapTraceWriteIndex - _deferredBootstrapTraceCount +
-				_deferredBootstrapTrace.Length) % _deferredBootstrapTrace.Length;
-			for (var i = 0; i < _deferredBootstrapTraceCount; i++)
-			{
-				pending[i] = _deferredBootstrapTrace[(readIndex + i) % _deferredBootstrapTrace.Length];
-			}
-
-			_deferredBootstrapTraceCount = 0;
-		}
-
-		foreach (var entry in pending)
-		{
-			var symbolText = "<unreadable>";
-			if (TryReadAsciiZ(entry.SymbolPointer, 256, out var sym))
-			{
-				symbolText = sym;
-			}
-
-			Console.Error.WriteLine(
-				$"[LOADER][TRACE] bootstrap_call#{entry.DispatchIndex}: op=0x{entry.Op:X16} " +
-				$"sym_ptr=0x{entry.SymbolPointer:X16} sym='{symbolText}' " +
-				$"out_ptr=0x{entry.OutputPointer:X16} ret=0x{entry.ReturnRip:X16}");
-		}
-	}
-
 	private void DumpRecentImportTrace()
 	{
-		if (_recentImportTraceCount == 0)
+		var trace = _recentImportTrace;
+		if (trace is null || _recentImportTraceCount == 0)
 		{
 			return;
 		}
-		Log.Info($"   Recent import calls ({_recentImportTraceCount}):");
-		int num = (_recentImportTraceWriteIndex - _recentImportTraceCount + _recentImportTrace.Length) % _recentImportTrace.Length;
+		Log.Info($"   Recent import calls for managed={Environment.CurrentManagedThreadId} guest=0x{GuestThreadExecution.CurrentGuestThreadHandle:X16} ({_recentImportTraceCount}):");
+		int num = (_recentImportTraceWriteIndex - _recentImportTraceCount + trace.Length) % trace.Length;
 		for (int i = 0; i < _recentImportTraceCount; i++)
 		{
-			int num2 = (num + i) % _recentImportTrace.Length;
-			var entry = _recentImportTrace[num2];
+			int num2 = (num + i) % trace.Length;
+			var entry = trace[num2];
 			if (!string.IsNullOrEmpty(entry.Nid))
 			{
 				Log.Info(
-					$"     #{entry.DispatchIndex} nid={entry.Nid} ret=0x{entry.ReturnRip:X16} " +
+					$"     #{entry.DispatchIndex} managed={entry.ManagedThreadId} guest=0x{entry.GuestThreadHandle:X16} nid={entry.Nid} ret=0x{entry.ReturnRip:X16} " +
 					$"rdi=0x{entry.Arg0:X16} rsi=0x{entry.Arg1:X16} rdx=0x{entry.Arg2:X16}");
 			}
 		}
@@ -135,9 +123,8 @@ public sealed partial class DirectExecutionBackend
 		int num2 = 0;
 		List<ulong> list = new List<ulong>(16);
 		ulong num3 = scanStart;
-		var hostMemory = ResolveDiagnosticsHostMemory();
-		HostRegionInfo lpBuffer;
-		while (num3 < scanEnd && hostMemory.Query(num3, out lpBuffer))
+		MEMORY_BASIC_INFORMATION64 lpBuffer;
+		while (num3 < scanEnd && VirtualQuery((void*)num3, out lpBuffer, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) != 0)
 		{
 			ulong baseAddress = lpBuffer.BaseAddress;
 			ulong num4 = baseAddress + lpBuffer.RegionSize;
@@ -147,7 +134,7 @@ public sealed partial class DirectExecutionBackend
 			}
 			ulong value = Math.Max(num3, baseAddress);
 			ulong num5 = Math.Min(num4, scanEnd);
-			if (lpBuffer.State == HostRegionState.Committed && IsReadableProtection(lpBuffer.RawProtection) && !IsExecutableProtection(lpBuffer.RawProtection))
+			if (lpBuffer.State == 4096 && IsReadableProtection(lpBuffer.Protect) && !IsExecutableProtection(lpBuffer.Protect))
 			{
 				ulong num6 = AlignUp(value, 8uL);
 				for (ulong num7 = num6; num7 + 8 <= num5; num7 += 8)
@@ -186,6 +173,50 @@ public sealed partial class DirectExecutionBackend
 		{
 			return;
 		}
+		const int preludeSize = 192;
+		Span<byte> prelude = stackalloc byte[preludeSize];
+		if (returnRip >= preludeSize && cpuContext.Memory.TryRead(returnRip - preludeSize, prelude))
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][TRACE] Import#{dispatchIndex} pre-return bytes @0x{returnRip - preludeSize:X16}: " +
+				BitConverter.ToString(prelude.ToArray()).Replace("-", " "));
+
+			List<DecodedInst>? bestCallChain = null;
+			var preludeAddress = returnRip - preludeSize;
+			for (var startOffset = 0; startOffset < preludeSize; startOffset++)
+			{
+				var cursor = preludeAddress + (ulong)startOffset;
+				var candidate = new List<DecodedInst>();
+				while (cursor < returnRip && candidate.Count < 96 &&
+					IcedDecoder.TryReadGuestBytes(cpuContext.Memory, cursor, 15, out var instructionBytes) &&
+					IcedDecoder.TryDecode(cursor, instructionBytes, out var instruction) &&
+					instruction.Length > 0 &&
+					cursor + (ulong)instruction.Length <= returnRip)
+				{
+					candidate.Add(instruction);
+					cursor += (ulong)instruction.Length;
+				}
+
+				if (cursor == returnRip &&
+					candidate.Count > 0 &&
+					string.Equals(candidate[^1].Mnemonic, "Call", StringComparison.OrdinalIgnoreCase) &&
+					(bestCallChain is null || candidate.Count > bestCallChain.Count))
+				{
+					bestCallChain = candidate;
+				}
+			}
+
+			if (bestCallChain is not null)
+			{
+				Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} pre-return disassembly:");
+				foreach (var instruction in bestCallChain.TakeLast(32))
+				{
+					Console.Error.WriteLine(
+						$"[LOADER][TRACE]   0x{instruction.Rip:X16}: {instruction.Text} " +
+						$"bytes={IcedDecoder.FormatBytes(instruction.Bytes)}");
+				}
+			}
+		}
 		Span<byte> destination = stackalloc byte[128];
 		if (!cpuContext.Memory.TryRead(returnRip, destination))
 		{
@@ -237,15 +268,14 @@ public sealed partial class DirectExecutionBackend
 			ulong callRip = returnRip + (ulong)i;
 			ulong target = unchecked((ulong)((long)(callRip + 5) + rel32));
 			Log.Debug($"Import#{dispatchIndex} near-call @{callRip:X16}: target=0x{target:X16}");
-			var importEntries = _importEntries;
-			for (int importIndex = 0; importIndex < importEntries.Length; importIndex++)
+			for (int importIndex = 0; importIndex < _importEntries.Length; importIndex++)
 			{
-				if (importEntries[importIndex].Address != target)
+				if (_importEntries[importIndex].Address != target)
 				{
 					continue;
 				}
 
-				string nid = importEntries[importIndex].Nid;
+				string nid = _importEntries[importIndex].Nid;
 				if (_moduleManager.TryGetExport(nid, out var export))
 				{
 					Log.Debug(
@@ -272,14 +302,14 @@ public sealed partial class DirectExecutionBackend
 					{
 						Log.Debug(
 							$"Import#{dispatchIndex} near-call PLT slot: [0x{slot:X16}] = 0x{slotTarget:X16}");
-						for (int importIndex = 0; importIndex < importEntries.Length; importIndex++)
+						for (int importIndex = 0; importIndex < _importEntries.Length; importIndex++)
 						{
-							if (importEntries[importIndex].Address != slotTarget)
+							if (_importEntries[importIndex].Address != slotTarget)
 							{
 								continue;
 							}
 
-							string nid = importEntries[importIndex].Nid;
+							string nid = _importEntries[importIndex].Nid;
 							if (_moduleManager.TryGetExport(nid, out var export))
 							{
 								Log.Debug(
@@ -303,6 +333,28 @@ public sealed partial class DirectExecutionBackend
 		return value == 65534 || value == 4294967294u || value == 18446744073709551614uL;
 	}
 
+	private static ulong ParseOptionalHexAddress(string? value)
+	{
+		if (string.IsNullOrWhiteSpace(value))
+		{
+			return 0;
+		}
+
+		var text = value.Trim();
+		if (text.StartsWith("0x", StringComparison.OrdinalIgnoreCase))
+		{
+			text = text[2..];
+		}
+
+		return ulong.TryParse(
+			text,
+			System.Globalization.NumberStyles.HexNumber,
+			System.Globalization.CultureInfo.InvariantCulture,
+			out var address)
+			? address
+			: 0;
+	}
+
 	private static bool IsPlausibleReturnAddress(ulong address)
 	{
 		return address >= 12884901888L && address < 17592186044416L && !IsUnresolvedSentinel(address);
@@ -352,7 +404,7 @@ public sealed partial class DirectExecutionBackend
 		{
 			return false;
 		}
-		if (!ResolveDiagnosticsHostMemory().Query(address, out var lpBuffer))
+		if (VirtualQuery((void*)address, out var lpBuffer, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
 		}
@@ -361,7 +413,7 @@ public sealed partial class DirectExecutionBackend
 		{
 			return false;
 		}
-		if (lpBuffer.State != HostRegionState.Committed || !IsReadableProtection(lpBuffer.RawProtection))
+		if (lpBuffer.State != 4096 || !IsReadableProtection(lpBuffer.Protect))
 		{
 			return false;
 		}
@@ -393,12 +445,12 @@ public sealed partial class DirectExecutionBackend
 			return true;
 		}
 
-		if (!ResolveDiagnosticsHostMemory().Query(address, out var lpBuffer))
+		if (VirtualQuery((void*)address, out var lpBuffer, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
 		}
 
-		var executable = lpBuffer.State == HostRegionState.Committed && IsExecutableProtection(lpBuffer.RawProtection);
+		var executable = lpBuffer.State == 4096 && IsExecutableProtection(lpBuffer.Protect);
 		if (executable)
 		{
 			_knownExecutablePages.TryAdd(pageAddress, 0);
@@ -417,14 +469,6 @@ public sealed partial class DirectExecutionBackend
 		return (value + num) & ~num;
 	}
 
-	// Diagnostics helpers are static (reachable from static handler paths), so
-	// they use the platform injected into the backend active on this thread and
-	// fall back to the process-wide singleton only when no run is bound.
-	private static IHostMemory ResolveDiagnosticsHostMemory()
-	{
-		return _activeExecutionBackend?._hostMemory ?? HostPlatform.Current.Memory;
-	}
-
 	private static bool IsReadableProtection(uint protect)
 	{
 		if ((protect & 0x100) != 0 || (protect & 1) != 0)

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.Exceptions.cs

@@ -9,9 +9,7 @@ using System.Reflection;
 using System.Runtime.InteropServices;
 using System.Threading;
 using SharpEmu.Core.Cpu.Disasm;
-using SharpEmu.Core.Cpu.Native.Windows;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
 
 namespace SharpEmu.Core.Cpu.Native;
 
@@ -19,17 +17,25 @@ public sealed partial class DirectExecutionBackend
 {
 	private const ulong LazyCommitWindowBytes = 0x0200_0000UL;
 	private static int _lazyCommitTraceCount;
+	private static int _guestAllocatorHoleRecoveries;
+	private static int _auxiliaryThreadExecuteFaultRecoveries;
 
 	private unsafe void SetupExceptionHandler()
 	{
+		if (!OperatingSystem.IsWindows())
+		{
+			SetupPosixExceptionHandler();
+			return;
+		}
+
 		if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_DISABLE_RAW_HANDLER"), "1", StringComparison.Ordinal))
 		{
-			_rawExceptionHandlerStub = _faultHandling.CreateHandlerThunk(RawVectoredHandlerPtrManaged, _hostRspSlotTlsIndex, _tlsGetValueAddress);
+			_rawExceptionHandlerStub = CreateExceptionHandlerTrampoline(RawVectoredHandlerPtrManaged);
 			if (_rawExceptionHandlerStub == 0)
 			{
 				throw new InvalidOperationException("Failed to create raw exception handler trampoline");
 			}
-			_rawExceptionHandler = _faultHandling.AddFirstChanceHandler(_rawExceptionHandlerStub);
+			_rawExceptionHandler = (nint)AddVectoredExceptionHandler(1u, _rawExceptionHandlerStub);
 			Console.Error.WriteLine($"[LOADER][INFO] Raw exception handler installed: 0x{_rawExceptionHandler:X16}");
 		}
 		else
@@ -39,22 +45,22 @@ public sealed partial class DirectExecutionBackend
 
 		_handlerDelegate = VectoredHandler;
 		_handlerHandle = GCHandle.Alloc(_handlerDelegate);
-		_exceptionHandlerStub = _faultHandling.CreateHandlerThunk(Marshal.GetFunctionPointerForDelegate(_handlerDelegate), _hostRspSlotTlsIndex, _tlsGetValueAddress);
+		_exceptionHandlerStub = CreateExceptionHandlerTrampoline(Marshal.GetFunctionPointerForDelegate(_handlerDelegate));
 		if (_exceptionHandlerStub == 0)
 		{
 			throw new InvalidOperationException("Failed to create exception handler trampoline");
 		}
-		_exceptionHandler = _faultHandling.AddFirstChanceHandler(_exceptionHandlerStub);
+		_exceptionHandler = (nint)AddVectoredExceptionHandler(1u, _exceptionHandlerStub);
 		Console.Error.WriteLine($"[LOADER][INFO] Exception handler installed: 0x{_exceptionHandler:X16}");
 
 		_unhandledFilterDelegate = UnhandledExceptionFilter;
 		_unhandledFilterHandle = GCHandle.Alloc(_unhandledFilterDelegate);
-		_unhandledFilterStub = _faultHandling.CreateHandlerThunk(Marshal.GetFunctionPointerForDelegate(_unhandledFilterDelegate), _hostRspSlotTlsIndex, _tlsGetValueAddress);
+		_unhandledFilterStub = CreateExceptionHandlerTrampoline(Marshal.GetFunctionPointerForDelegate(_unhandledFilterDelegate));
 		if (_unhandledFilterStub == 0)
 		{
 			throw new InvalidOperationException("Failed to create unhandled exception filter trampoline");
 		}
-		_faultHandling.SetUnhandledFilter(_unhandledFilterStub);
+		SetUnhandledExceptionFilter(_unhandledFilterStub);
 	}
 
 	private unsafe int UnhandledExceptionFilter(void* exceptionInfo)
@@ -62,8 +68,8 @@ public sealed partial class DirectExecutionBackend
 		try
 		{
 			EXCEPTION_RECORD* exceptionRecord = ((EXCEPTION_POINTERS*)exceptionInfo)->ExceptionRecord;
-			ulong rip = ReadCtxU64(((EXCEPTION_POINTERS*)exceptionInfo)->ContextRecord, CTX_RIP);
-			ulong rsp = ReadCtxU64(((EXCEPTION_POINTERS*)exceptionInfo)->ContextRecord, CTX_RSP);
+			ulong rip = ReadCtxU64(((EXCEPTION_POINTERS*)exceptionInfo)->ContextRecord, 248);
+			ulong rsp = ReadCtxU64(((EXCEPTION_POINTERS*)exceptionInfo)->ContextRecord, 152);
 			Console.Error.WriteLine("[LOADER][FATAL] Unhandled exception filter fired.");
 			Console.Error.WriteLine($"[LOADER][FATAL]   Code: 0x{exceptionRecord->ExceptionCode:X8}");
 			Console.Error.WriteLine($"[LOADER][FATAL]   Exception Address: 0x{(ulong)(nint)exceptionRecord->ExceptionAddress:X16}");
@@ -102,19 +108,28 @@ public sealed partial class DirectExecutionBackend
 				return 0;
 			}
 
-			ulong rip = ReadCtxU64(contextRecord, CTX_RIP);
-			ulong rsp = ReadCtxU64(contextRecord, CTX_RSP);
-
-			// Thread-mode probe: a hardware exception raised while this thread is inside
-			// the managed import gateway means the VEH->managed reentry happened from
-			// cooperative GC mode — a ReversePInvokeBadTransition candidate.
-			if (LogThreadMode && _threadModeGatewayDepth > 0)
+			ulong rip = ReadCtxU64(contextRecord, 248);
+			ulong rsp = ReadCtxU64(contextRecord, 152);
+			if (TryRecoverGuestInt41(exceptionCode, contextRecord, rip))
 			{
-				TraceThreadMode(
-					$"veh_in_gateway code=0x{exceptionCode:X8} rip=0x{rip:X16} gateway_depth={_threadModeGatewayDepth}");
+				return -1;
+			}
+			if (TryRecoverAuxiliaryThreadExecuteFault(exceptionRecord, contextRecord, rip))
+			{
+				return -1;
 			}
 
-			if (exceptionCode == WindowsFaultCodes.AccessViolation && TryHandleLazyCommittedPage(exceptionRecord, rip, rsp))
+			if (exceptionCode == 3221225477u && TryHandleLazyCommittedPage(exceptionRecord, rip, rsp))
+			{
+				return -1;
+			}
+			if (exceptionCode == 3221225477u &&
+				TryRecoverGuestAllocatorHole(exceptionRecord, contextRecord, rip))
+			{
+				return -1;
+			}
+			if (exceptionCode == StatusIllegalInstruction &&
+				TryRecoverIllegalInstruction(contextRecord, rip))
 			{
 				return -1;
 			}
@@ -129,10 +144,10 @@ public sealed partial class DirectExecutionBackend
 
 			switch (exceptionCode)
 			{
-				case WindowsFaultCodes.AccessViolation:
+				case 3221225477u:
 					LogAccessViolationTrace(exceptionAddress, exceptionRecord);
 					break;
-				case WindowsFaultCodes.FastFail:
+				case 3221226505u:
 					{
 						ulong p0 = exceptionRecord->NumberParameters >= 1 ? (*exceptionRecord->ExceptionInformation) : 0;
 						ulong p1 = exceptionRecord->NumberParameters >= 2 ? exceptionRecord->ExceptionInformation[1] : 0;
@@ -142,27 +157,56 @@ public sealed partial class DirectExecutionBackend
 					}
 			}
 
-			ulong rax = ReadCtxU64(contextRecord, CTX_RAX);
-			ulong rbx = ReadCtxU64(contextRecord, CTX_RBX);
-			ulong rcx = ReadCtxU64(contextRecord, CTX_RCX);
-			ulong rdx = ReadCtxU64(contextRecord, CTX_RDX);
-			ulong rsi = ReadCtxU64(contextRecord, CTX_RSI);
-			ulong rdi = ReadCtxU64(contextRecord, CTX_RDI);
-			ulong rbp = ReadCtxU64(contextRecord, CTX_RBP);
-			ulong r8 = ReadCtxU64(contextRecord, CTX_R8);
-			ulong r9 = ReadCtxU64(contextRecord, CTX_R9);
-			ulong r10 = ReadCtxU64(contextRecord, CTX_R10);
-			ulong r11 = ReadCtxU64(contextRecord, CTX_R11);
-			ulong r12 = ReadCtxU64(contextRecord, CTX_R12);
-			ulong r13 = ReadCtxU64(contextRecord, CTX_R13);
-			ulong r14 = ReadCtxU64(contextRecord, CTX_R14);
-			ulong r15 = ReadCtxU64(contextRecord, CTX_R15);
+			ulong rax = ReadCtxU64(contextRecord, 120);
+			ulong rbx = ReadCtxU64(contextRecord, 144);
+			ulong rcx = ReadCtxU64(contextRecord, 128);
+			ulong rdx = ReadCtxU64(contextRecord, 136);
+			ulong rsi = ReadCtxU64(contextRecord, 168);
+			ulong rdi = ReadCtxU64(contextRecord, 176);
+			ulong rbp = ReadCtxU64(contextRecord, 160);
+			ulong r8 = ReadCtxU64(contextRecord, 184);
+			ulong r9 = ReadCtxU64(contextRecord, 192);
+			ulong r10 = ReadCtxU64(contextRecord, 200);
+			ulong r11 = ReadCtxU64(contextRecord, 208);
+			ulong r12 = ReadCtxU64(contextRecord, 216);
+			ulong r13 = ReadCtxU64(contextRecord, 224);
+			ulong r14 = ReadCtxU64(contextRecord, 232);
+			ulong r15 = ReadCtxU64(contextRecord, 240);
 
 			Console.Error.WriteLine("[LOADER][INFO] =========================================");
 			Console.Error.WriteLine("[LOADER][INFO] NATIVE EXCEPTION CAUGHT!");
 			Console.Error.WriteLine($"[LOADER][INFO]   Code: 0x{exceptionCode:X8}");
 			Console.Error.WriteLine($"[LOADER][INFO]   Exception Address: 0x{exceptionAddress:X16}");
 			Console.Error.WriteLine($"[LOADER][INFO]   RIP: 0x{rip:X16}");
+			Console.Error.WriteLine(
+				$"[LOADER][INFO]   Host thread: managed={Environment.CurrentManagedThreadId} " +
+				$"name='{Thread.CurrentThread.Name ?? "<unnamed>"}'");
+			if (_activeGuestThreadState is { } activeGuestThread)
+			{
+				Console.Error.WriteLine(
+					$"[LOADER][INFO]   Guest thread: handle=0x{activeGuestThread.ThreadHandle:X16} " +
+					$"name='{activeGuestThread.Name}' state={activeGuestThread.State} " +
+					$"last_import={activeGuestThread.LastImportNid ?? "<none>"} " +
+					$"last_ret=0x{activeGuestThread.LastReturnRip:X16}");
+				Console.Error.WriteLine(
+					$"[LOADER][INFO]   Last import registers: " +
+					$"rax=0x{Volatile.Read(ref activeGuestThread.LastImportRax):X16} " +
+					$"result_valid={Volatile.Read(ref activeGuestThread.LastImportResultValid) != 0} " +
+					$"rdi=0x{activeGuestThread.LastImportRdi:X16} " +
+					$"rsi=0x{activeGuestThread.LastImportRsi:X16} " +
+					$"rdx=0x{activeGuestThread.LastImportRdx:X16} " +
+					$"rcx=0x{activeGuestThread.LastImportRcx:X16} " +
+					$"r8=0x{activeGuestThread.LastImportR8:X16} " +
+					$"r9=0x{activeGuestThread.LastImportR9:X16}");
+				Console.Error.WriteLine(
+					$"[LOADER][INFO]   Last import stack args: " +
+					$"0=0x{activeGuestThread.LastImportStack0:X16} " +
+					$"1=0x{activeGuestThread.LastImportStack1:X16} " +
+					$"2=0x{activeGuestThread.LastImportStack2:X16} " +
+					$"3=0x{activeGuestThread.LastImportStack3:X16} " +
+					$"4=0x{activeGuestThread.LastImportStack4:X16} " +
+					$"5=0x{activeGuestThread.LastImportStack5:X16}");
+			}
 			if (TryFormatNearestRuntimeSymbol(rip, out string symbol))
 			{
 				Console.Error.WriteLine("[LOADER][INFO]   RIP symbol: " + symbol);
@@ -187,7 +231,7 @@ public sealed partial class DirectExecutionBackend
 
 			ulong accessType = 0;
 			ulong target = 0;
-			if (exceptionCode == WindowsFaultCodes.AccessViolation && exceptionRecord->NumberParameters >= 2)
+			if (exceptionCode == 3221225477u && exceptionRecord->NumberParameters >= 2)
 			{
 				accessType = *exceptionRecord->ExceptionInformation;
 				target = exceptionRecord->ExceptionInformation[1];
@@ -200,27 +244,57 @@ public sealed partial class DirectExecutionBackend
 				};
 				Console.Error.WriteLine("[LOADER][INFO]   AV access: " + accessText);
 				Console.Error.WriteLine($"[LOADER][INFO]   AV target: 0x{target:X16}");
-				if (_hostMemory.Query(target, out var mbi))
+				if (VirtualQuery((void*)target, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) != 0)
 				{
-					Console.Error.WriteLine($"[LOADER][INFO]   AV target region: base=0x{mbi.BaseAddress:X16} size=0x{mbi.RegionSize:X16} state=0x{mbi.RawState:X08} protect=0x{mbi.RawProtection:X08}");
+					Console.Error.WriteLine($"[LOADER][INFO]   AV target region: base=0x{mbi.BaseAddress:X16} size=0x{mbi.RegionSize:X16} state=0x{mbi.State:X08} protect=0x{mbi.Protect:X08}");
 				}
 
 			}
 
-			try
+			Console.Error.WriteLine("[LOADER][INFO]   Stack qwords (RSP..):");
+			for (int i = 0; i < 16; i++)
 			{
-				Console.Error.WriteLine("[LOADER][INFO]   Stack qwords (RSP..):");
-				for (int i = 0; i < 16; i++)
+				ulong stackAddr = rsp + (ulong)(i * 8);
+				if (!TryReadHostQword(stackAddr, out ulong value))
 				{
-					ulong stackAddr = rsp + (ulong)(i * 8);
-					ulong value = (ulong)Marshal.ReadInt64((nint)stackAddr);
-					Console.Error.WriteLine($"[LOADER][INFO]     [rsp+0x{i * 8:X2}] @0x{stackAddr:X16} = 0x{value:X16}");
+					Console.Error.WriteLine("[LOADER][WARNING]   Could not read stack qwords.");
+					break;
+				}
+				Console.Error.WriteLine($"[LOADER][INFO]     [rsp+0x{i * 8:X2}] @0x{stackAddr:X16} = 0x{value:X16}");
+			}
+
+			if (string.Equals(
+					Environment.GetEnvironmentVariable("SHARPEMU_DUMP_FAULT_STACK_WINDOW"),
+					"1",
+					StringComparison.Ordinal))
+			{
+				Console.Error.WriteLine("[LOADER][INFO]   Full fault stack window (RSP-0x300..RSP+0x100):");
+				var windowStart = rsp >= 0x300 ? rsp - 0x300 : 0;
+				for (var stackAddr = windowStart; stackAddr < rsp + 0x100; stackAddr += 8)
+				{
+					if (!TryReadHostQword(stackAddr, out var value))
+					{
+						continue;
+					}
+
+					var relative = unchecked((long)(stackAddr - rsp));
+					var relativeText = relative >= 0
+						? $"+0x{relative:X}"
+						: $"-0x{-relative:X}";
+					var symbolText = TryFormatNearestRuntimeSymbol(value, out var stackSymbol)
+						? $" [{stackSymbol}]"
+						: string.Empty;
+					Console.Error.WriteLine(
+						$"[LOADER][INFO]     [rsp{relativeText}] " +
+						$"@0x{stackAddr:X16} = 0x{value:X16}{symbolText}");
 				}
 			}
-			catch
-			{
-				Console.Error.WriteLine("[LOADER][WARNING]   Could not read stack qwords.");
-			}
+
+			DumpPointerWindow("fault-register-rbx", rbx, 0x60);
+			DumpPointerWindow("fault-register-rsi", rsi, 0x60);
+			DumpPointerWindow("fault-register-rdi", rdi, 0x60);
+			DumpPointerWindow("fault-register-r13", r13, 0x60);
+			DumpPointerWindow("fault-register-r14", r14, 0x60);
 
 			try
 			{
@@ -228,12 +302,15 @@ public sealed partial class DirectExecutionBackend
 				ulong frame = rbp;
 				for (int i = 0; i < 12; i++)
 				{
-					if (frame < 140733193388032L || frame > 140737488355327L)
+					if (frame < 0x10000)
 					{
 						break;
 					}
-					ulong next = (ulong)Marshal.ReadInt64((nint)frame);
-					ulong ret = (ulong)Marshal.ReadInt64((nint)(frame + 8));
+					if (!TryReadHostQword(frame, out ulong next) || !TryReadHostQword(frame + 8, out ulong ret))
+					{
+						Console.Error.WriteLine("[LOADER][WARNING]   Could not walk RBP frame chain.");
+						break;
+					}
 					string extra = TryFormatNearestRuntimeSymbol(ret, out string retSym) ? $" [{retSym}]" : string.Empty;
 					Console.Error.WriteLine($"[LOADER][INFO]     frame#{i}: rbp=0x{frame:X16} ret=0x{ret:X16}{extra} next=0x{next:X16}");
 					if (next <= frame)
@@ -250,16 +327,15 @@ public sealed partial class DirectExecutionBackend
 
 			switch (exceptionCode)
 			{
-				case WindowsFaultCodes.AccessViolation:
+				case 3221225477u:
 					Console.Error.WriteLine("[LOADER][ERROR]   Type: Access Violation");
 					Console.Error.WriteLine("[LOADER][ERROR]   This usually means:");
 					Console.Error.WriteLine("[LOADER][ERROR]     - Guest code called an unmapped import");
 					Console.Error.WriteLine("[LOADER][ERROR]     - Guest code accessed unmapped memory");
 					Console.Error.WriteLine("[LOADER][ERROR]     - Need to implement HLE for this NID");
-					try
+					byte[] code = new byte[16];
+					if (TryReadHostBytes(rip, code))
 					{
-						byte[] code = new byte[16];
-						Marshal.Copy((nint)rip, code, 0, code.Length);
 						Console.Error.WriteLine("[LOADER][INFO]   Code at RIP: " + BitConverter.ToString(code).Replace("-", " "));
 						if (code[0] == 100)
 						{
@@ -275,34 +351,67 @@ public sealed partial class DirectExecutionBackend
 							Console.Error.WriteLine($"[LOADER][INFO]   RBP: 0x{rbp:X16} (mod 16 = {rbp % 16})");
 							Console.Error.WriteLine($"[LOADER][INFO]   RSP: 0x{rsp:X16} (mod 16 = {rsp % 16})");
 						}
-						if (rip > 16)
+						byte[] before = new byte[16];
+						if (rip > 16 && TryReadHostBytes(rip - 16, before))
 						{
-							byte[] before = new byte[16];
-							Marshal.Copy((nint)(rip - 16), before, 0, before.Length);
 							Console.Error.WriteLine("[LOADER][INFO]   Code before RIP: " + BitConverter.ToString(before).Replace("-", " "));
 						}
-						if (rip > 32)
+						byte[] window = new byte[64];
+						if (rip > 32 && TryReadHostBytes(rip - 32, window))
 						{
-							byte[] window = new byte[64];
-							Marshal.Copy((nint)(rip - 32), window, 0, window.Length);
 							Console.Error.WriteLine("[LOADER][INFO]   Code window [RIP-0x20..]: " + BitConverter.ToString(window).Replace("-", " "));
 						}
+						for (var stackIndex = 0; stackIndex < 16; stackIndex++)
+						{
+							byte[] stackSlot = new byte[8];
+							if (!TryReadHostBytes(rsp + (ulong)(stackIndex * 8), stackSlot))
+							{
+								continue;
+							}
+							var candidate = BitConverter.ToUInt64(stackSlot);
+							if (candidate < _entryPoint || candidate >= _entryPoint + 0x10000000 || candidate < 24)
+							{
+								continue;
+							}
+							byte[] callSiteWindow = new byte[48];
+							if (TryReadHostBytes(candidate - 24, callSiteWindow))
+							{
+								Console.Error.WriteLine(
+									$"[LOADER][INFO]   Stack guest-code candidate [rsp+0x{stackIndex * 8:X2}]=0x{candidate:X16}, bytes [-0x18..]: " +
+									BitConverter.ToString(callSiteWindow).Replace("-", " "));
+							}
+						}
 					}
-					catch
+					else
 					{
 						Console.Error.WriteLine("[LOADER][ERROR]   Could not read code at RIP");
 					}
 					DumpRecentImportTrace();
-					DumpGuestDisasmDiagnostics(rip, rbp);
+					DumpGuestDisasmDiagnostics(rip, rbp, rsp);
+					DumpGuestRegisterWindowDiagnostics(
+						rax, rbx, rcx, rdx, rsi, rdi, rbp, rsp,
+						r8, r9, r10, r11, r12, r13, r14, r15);
 					DumpGuestReferenceDiagnostics();
 					DumpGuestPointerWindowDiagnostics();
 					break;
-				case WindowsFaultCodes.Breakpoint:
+				case 2147483651u:
 					Console.Error.WriteLine("[LOADER][WARNING]   Type: Breakpoint (int3)");
 					Console.Error.WriteLine("[LOADER][WARNING]   Unexpected breakpoint in direct-bridge mode");
 					break;
-				case WindowsFaultCodes.IllegalInstruction:
+				case 1073741845u:
+					Console.Error.WriteLine("[LOADER][ERROR]   Type: Abort (SIGABRT)");
+					DumpRecentImportTrace();
+					DumpGuestDisasmDiagnostics(rip, rbp, rsp);
+					break;
+				case 3221225501u:
 					Console.Error.WriteLine("[LOADER][INFO]   Type: Illegal Instruction");
+					byte[] illegalCode = new byte[16];
+					if (TryReadHostBytes(rip, illegalCode))
+					{
+						Console.Error.WriteLine("[LOADER][INFO]   Code at RIP: " + BitConverter.ToString(illegalCode).Replace("-", " "));
+					}
+					DumpRecentImportTrace();
+					DumpGuestDisasmDiagnostics(rip, rbp, rsp);
 					break;
 			}
 
@@ -316,6 +425,108 @@ public sealed partial class DirectExecutionBackend
 		}
 	}
 
+	private unsafe bool TryRecoverAuxiliaryThreadExecuteFault(
+		EXCEPTION_RECORD* exceptionRecord,
+		void* contextRecord,
+		ulong rip)
+	{
+		if (exceptionRecord->ExceptionCode != 3221225477u ||
+			rip >= 0x0000000800000000UL ||
+			_activeGuestThreadState is not { Name: "tbb_thead" } activeThread)
+		{
+			return false;
+		}
+
+		var hostExit = ActiveEntryReturnSentinelRip;
+		if (hostExit < 0x10000)
+		{
+			hostExit = unchecked((ulong)_guestReturnStub);
+		}
+		if (hostExit < 0x10000)
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][WARN] Could not recover auxiliary TBB execute fault: target=0x{rip:X16} " +
+				$"active_exit=0x{ActiveEntryReturnSentinelRip:X16} guest_return_stub=0x{unchecked((ulong)_guestReturnStub):X16}");
+			return false;
+		}
+
+		_ = TryPatchActiveGuestReturnSlot(hostExit);
+		WriteCtxU64(contextRecord, 120, 0);
+		WriteCtxU64(contextRecord, 248, hostExit);
+		var recovery = Interlocked.Increment(ref _auxiliaryThreadExecuteFaultRecoveries);
+		Console.Error.WriteLine(
+			$"[LOADER][WARN] Recovered auxiliary TBB execute fault #{recovery}: " +
+			$"thread=0x{activeThread.ThreadHandle:X16} target=0x{rip:X16} -> host_exit=0x{hostExit:X16}");
+		return true;
+	}
+
+	private unsafe bool TryRecoverGuestInt41(uint exceptionCode, void* contextRecord, ulong rip)
+	{
+		if (!_ignoreGuestInt41 || exceptionCode != 3221225477u || rip < 0x10000)
+		{
+			return false;
+		}
+
+		byte[] opcode = new byte[2];
+		if (!TryReadHostBytes(rip, opcode) || opcode[0] != 0xCD || opcode[1] != 0x41)
+		{
+			return false;
+		}
+
+		var count = Interlocked.Increment(ref _ignoredGuestInt41Count);
+		WriteCtxU64(contextRecord, 248, rip + 2);
+		if (count <= 16 || count % 65536 == 0)
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][WARN] Ignored guest int 0x41 trap #{count} at 0x{rip:X16} (SHARPEMU_IGNORE_INT41=1)");
+			Console.Error.Flush();
+		}
+		return true;
+	}
+
+	private unsafe static bool TryRecoverGuestAllocatorHole(
+		EXCEPTION_RECORD* exceptionRecord,
+		void* contextRecord,
+		ulong rip)
+	{
+		if (string.Equals(
+				Environment.GetEnvironmentVariable("SHARPEMU_DISABLE_GUEST_ALLOCATOR_HOLE_RECOVERY"),
+				"1",
+				StringComparison.Ordinal) ||
+			exceptionRecord->NumberParameters < 2 ||
+			exceptionRecord->ExceptionInformation[0] != 0 ||
+			exceptionRecord->ExceptionInformation[1] != 8 ||
+			ReadCtxU64(contextRecord, CTX_RDI) != 0 ||
+			rip < 0x10000)
+		{
+			return false;
+		}
+
+		// Demon's Souls occasionally leaves an empty payload in a locked pool
+		// tree node. The allocator dereferences payload+8 before reaching its
+		// existing empty-pool fallback. Match the instruction stream instead of
+		// a title-specific absolute address, then resume at that fallback so the
+		// lock is released and the allocator can try its next backing pool.
+		const ulong allocatorHoleSignature = 0x634CFF568D08778BUL;
+		if (*(ulong*)rip != allocatorHoleSignature || *((byte*)rip + 8) != 0xF2)
+		{
+			return false;
+		}
+
+		const ulong emptyPoolFallbackDelta = 0x8E;
+		WriteCtxU64(contextRecord, CTX_RIP, rip + emptyPoolFallbackDelta);
+		var recovery = Interlocked.Increment(ref _guestAllocatorHoleRecoveries);
+		if (recovery <= 16 || (recovery & (recovery - 1)) == 0)
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][WARN] Guest allocator empty-node adapter recovery #{recovery}: " +
+				$"rip=0x{rip:X16} -> 0x{rip + emptyPoolFallbackDelta:X16}");
+			Console.Error.Flush();
+		}
+
+		return true;
+	}
+
 	private static bool IsBenignHostDebugException(uint exceptionCode)
 	{
 		return exceptionCode is DBG_PRINTEXCEPTION_C or DBG_PRINTEXCEPTION_WIDE_C or MS_VC_THREADNAME_EXCEPTION;
@@ -334,8 +545,8 @@ public sealed partial class DirectExecutionBackend
 			EXCEPTION_POINTERS* pointers = (EXCEPTION_POINTERS*)exceptionInfo;
 			EXCEPTION_RECORD* record = pointers->ExceptionRecord;
 			void* contextRecord = pointers->ContextRecord;
-			ulong rip = contextRecord != null ? ReadCtxU64(contextRecord, CTX_RIP) : 0;
-			ulong rsp = contextRecord != null ? ReadCtxU64(contextRecord, CTX_RSP) : 0;
+			ulong rip = contextRecord != null ? ReadCtxU64(contextRecord, 248) : 0;
+			ulong rsp = contextRecord != null ? ReadCtxU64(contextRecord, 152) : 0;
 			ulong accessType = record->NumberParameters >= 1 ? *record->ExceptionInformation : 0;
 			ulong target = record->NumberParameters >= 2 ? record->ExceptionInformation[1] : 0;
 			Console.Error.WriteLine(
@@ -395,7 +606,7 @@ public sealed partial class DirectExecutionBackend
 		}
 	}
 
-	private void DumpGuestDisasmDiagnostics(ulong rip, ulong rbp)
+	private void DumpGuestDisasmDiagnostics(ulong rip, ulong rbp, ulong rsp)
 	{
 		if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_DISASM"), "1", StringComparison.Ordinal))
 		{
@@ -407,12 +618,20 @@ public sealed partial class DirectExecutionBackend
 			DumpGuestInstructionStream("fault-prelude", rip - 0x20, 24);
 		}
 
+		// Optimized guest code frequently omits frame pointers. The return
+		// address at RSP is then more useful than an RBP walk and identifies the
+		// exact call site that supplied the faulting arguments.
+		if (TryReadHostQword(rsp, out var stackReturn) && stackReturn >= 0x60)
+		{
+			DumpGuestInstructionStream("stack-return-prelude", stackReturn - 0x60, 40);
+		}
+
 		try
 		{
 			ulong frame = rbp;
 			for (int i = 0; i < 3; i++)
 			{
-				if (frame < 140733193388032L || frame > 140737488355327L)
+				if (frame < 0x10000)
 				{
 					break;
 				}
@@ -481,7 +700,7 @@ public sealed partial class DirectExecutionBackend
 		ulong address = scanBase;
 		while (address < scanEnd)
 		{
-			if (!_hostMemory.Query(address, out var mbi))
+			if (VirtualQuery((void*)address, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 			{
 				break;
 			}
@@ -493,9 +712,9 @@ public sealed partial class DirectExecutionBackend
 				break;
 			}
 
-			if (mbi.State == HostRegionState.Committed &&
-				IsReadableProtection(mbi.RawProtection) &&
-				IsExecutableProtection(mbi.RawProtection))
+			if (mbi.State == MEM_COMMIT &&
+				IsReadableProtection(mbi.Protect) &&
+				IsExecutableProtection(mbi.Protect))
 			{
 				ScanExecutableRegionForTargetReferences(regionBase, regionEnd, targetList, hitCounts, maxHitsPerTarget);
 			}
@@ -556,6 +775,55 @@ public sealed partial class DirectExecutionBackend
 		}
 	}
 
+	private void DumpGuestRegisterWindowDiagnostics(
+		ulong rax,
+		ulong rbx,
+		ulong rcx,
+		ulong rdx,
+		ulong rsi,
+		ulong rdi,
+		ulong rbp,
+		ulong rsp,
+		ulong r8,
+		ulong r9,
+		ulong r10,
+		ulong r11,
+		ulong r12,
+		ulong r13,
+		ulong r14,
+		ulong r15)
+	{
+		if (!string.Equals(
+				Environment.GetEnvironmentVariable("SHARPEMU_LOG_REGISTER_WINDOWS"),
+				"1",
+				StringComparison.Ordinal))
+		{
+			return;
+		}
+
+		// A register can be the only surviving reference to the object or
+		// argument array that caused a native guest fault. Capture a compact
+		// window while the process is alive so the post-mortem log can
+		// distinguish an absent object from a partially initialized one.
+		var registers = new (string Name, ulong Value)[]
+		{
+			("rax", rax), ("rbx", rbx), ("rcx", rcx), ("rdx", rdx),
+			("rsi", rsi), ("rdi", rdi), ("rbp", rbp), ("rsp", rsp),
+			("r8", r8), ("r9", r9), ("r10", r10), ("r11", r11),
+			("r12", r12), ("r13", r13), ("r14", r14), ("r15", r15),
+		};
+		var seen = new HashSet<ulong>();
+		foreach (var (name, value) in registers)
+		{
+			if (value < 0x10000 || !seen.Add(value))
+			{
+				continue;
+			}
+
+			DumpPointerWindow($"register-{name}", value, 0x80);
+		}
+	}
+
 	private void ScanExecutableRegionForTargetReferences(
 		ulong regionBase,
 		ulong regionEnd,
@@ -800,13 +1068,13 @@ public sealed partial class DirectExecutionBackend
 			return false;
 		}
 
-		if (!_hostMemory.Query(address, out var mbi))
+		if (VirtualQuery((void*)address, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
 		}
 
 		ulong regionEnd = mbi.BaseAddress + mbi.RegionSize;
-		if (mbi.State != HostRegionState.Committed || !IsReadableProtection(mbi.RawProtection) || regionEnd <= address || address > regionEnd - 8)
+		if (mbi.State != MEM_COMMIT || !IsReadableProtection(mbi.Protect) || regionEnd <= address || address > regionEnd - 8)
 		{
 			return false;
 		}
@@ -823,6 +1091,61 @@ public sealed partial class DirectExecutionBackend
 		}
 	}
 
+	private static bool TryReadHostQword(ulong address, out ulong value)
+	{
+		if (!OperatingSystem.IsWindows())
+		{
+			// A stray read inside the signal handler would raise a nested
+			// SIGSEGV and kill the process before diagnostics finish, so
+			// probe the region table instead of relying on try/catch.
+			return TryReadStackU64(address, out value);
+		}
+
+		value = 0;
+		try
+		{
+			value = (ulong)Marshal.ReadInt64((nint)address);
+			return true;
+		}
+		catch
+		{
+			return false;
+		}
+	}
+
+	private unsafe static bool TryReadHostBytes(ulong address, byte[] buffer)
+	{
+		if (address < 65536)
+		{
+			return false;
+		}
+
+		if (!OperatingSystem.IsWindows())
+		{
+			// See TryReadHostQword: probe every touched page before reading.
+			ulong end = address + (ulong)buffer.Length;
+			for (ulong page = address & 0xFFFFFFFFFFFFF000uL; page < end; page += 4096)
+			{
+				if (VirtualQuery((void*)page, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0 ||
+					mbi.State != MEM_COMMIT ||
+					!IsReadableProtection(mbi.Protect))
+				{
+					return false;
+				}
+			}
+		}
+
+		try
+		{
+			Marshal.Copy((nint)address, buffer, 0, buffer.Length);
+			return true;
+		}
+		catch
+		{
+			return false;
+		}
+	}
+
 	private string FormatPointerWithNearestSymbol(ulong value)
 	{
 		string text = $"0x{value:X16}";
@@ -918,25 +1241,25 @@ public sealed partial class DirectExecutionBackend
 		{
 			return false;
 		}
-		if (!_hostMemory.Query(faultAddress, out var mbi))
+		if (VirtualQuery((void*)faultAddress, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
 		}
 
 		ulong pageBase = faultAddress & 0xFFFFFFFFFFFFF000uL;
-		uint commitProtect = ResolveLazyCommitProtection(accessType, mbi.RawAllocationProtection);
+		uint commitProtect = ResolveLazyCommitProtection(accessType, mbi.AllocationProtect);
 		int traceIndex = Interlocked.Increment(ref _lazyCommitTraceCount);
 		bool traceLazyCommit = ShouldTraceLazyCommit(traceIndex);
 		if (traceLazyCommit)
 		{
-			Console.Error.WriteLine($"[LOADER][TRACE] lazy-query#{traceIndex}: fault=0x{faultAddress:X16} owner={owner} rip=0x{rip:X16} rsp=0x{rsp:X16} state=0x{mbi.RawState:X08} base=0x{mbi.BaseAddress:X16} size=0x{mbi.RegionSize:X16} alloc=0x{mbi.RawAllocationProtection:X08} prot=0x{mbi.RawProtection:X08}");
+			Console.Error.WriteLine($"[LOADER][TRACE] lazy-query#{traceIndex}: fault=0x{faultAddress:X16} owner={owner} rip=0x{rip:X16} rsp=0x{rsp:X16} state=0x{mbi.State:X08} base=0x{mbi.BaseAddress:X16} size=0x{mbi.RegionSize:X16} alloc=0x{mbi.AllocationProtect:X08} prot=0x{mbi.Protect:X08}");
 		}
 
-		if (mbi.State == HostRegionState.Committed && IsAccessCompatible(accessType, mbi.RawProtection))
+		if (mbi.State == 4096 && IsAccessCompatible(accessType, mbi.Protect))
 		{
 			if (traceLazyCommit)
 			{
-				Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit-race#{traceIndex}: fault=0x{faultAddress:X16} protect=0x{mbi.RawProtection:X08}");
+				Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit-race#{traceIndex}: fault=0x{faultAddress:X16} protect=0x{mbi.Protect:X08}");
 			}
 			return true;
 		}
@@ -945,10 +1268,10 @@ public sealed partial class DirectExecutionBackend
 		ulong committedBase = 0;
 		ulong committedSize = 0;
 
-		if (mbi.State == HostRegionState.Free)
+		if (mbi.State == 65536)
 		{
 			if (TryGetLazyCommitWindow(faultAddress, mbi.BaseAddress, mbi.RegionSize, out var windowBase, out var windowSize) &&
-				TryReserveThenCommit(_hostMemory, windowBase, windowSize, windowBase, windowSize, commitProtect))
+				TryReserveThenCommit(windowBase, windowSize, windowBase, windowSize, commitProtect))
 			{
 				committed = true;
 				committedBase = windowBase;
@@ -957,7 +1280,7 @@ public sealed partial class DirectExecutionBackend
 			else
 			{
 				ulong largeBase = faultAddress & 0xFFFFFFFFFFE00000uL;
-				if (TryReserveThenCommit(_hostMemory, largeBase, 2097152uL, largeBase, 2097152uL, commitProtect))
+				if (TryReserveThenCommit(largeBase, 2097152uL, largeBase, 2097152uL, commitProtect))
 				{
 					committed = true;
 					committedBase = largeBase;
@@ -968,13 +1291,13 @@ public sealed partial class DirectExecutionBackend
 			if (!committed)
 			{
 				ulong region64kBase = faultAddress & 0xFFFFFFFFFFFF0000uL;
-				if (TryReserveThenCommit(_hostMemory, region64kBase, 65536uL, region64kBase, 65536uL, commitProtect))
+				if (TryReserveThenCommit(region64kBase, 65536uL, region64kBase, 65536uL, commitProtect))
 				{
 					committed = true;
 					committedBase = region64kBase;
 					committedSize = 65536uL;
 				}
-				else if (TryReserveThenCommit(_hostMemory, pageBase, 4096uL, pageBase, 4096uL, commitProtect))
+				else if (TryReserveThenCommit(pageBase, 4096uL, pageBase, 4096uL, commitProtect))
 				{
 					committed = true;
 					committedBase = pageBase;
@@ -987,7 +1310,7 @@ public sealed partial class DirectExecutionBackend
 				return false;
 			}
 
-			TryCommitRange(_hostMemory, pageBase + 4096, 4096uL, commitProtect);
+			TryCommitRange(pageBase + 4096, 4096uL, commitProtect);
 			if (traceLazyCommit)
 			{
 				Console.Error.WriteLine($"[LOADER][TRACE] lazy-reserve-commit#{traceIndex}: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
@@ -995,13 +1318,13 @@ public sealed partial class DirectExecutionBackend
 			return true;
 		}
 
-		if (mbi.State != HostRegionState.Reserved)
+		if (mbi.State != 8192)
 		{
 			return false;
 		}
 
 		if (TryGetLazyCommitWindow(faultAddress, mbi.BaseAddress, mbi.RegionSize, out var commitWindowBase, out var commitWindowSize) &&
-			TryCommitRange(_hostMemory, commitWindowBase, commitWindowSize, commitProtect))
+			TryCommitRange(commitWindowBase, commitWindowSize, commitProtect))
 		{
 			committed = true;
 			committedBase = commitWindowBase;
@@ -1010,7 +1333,7 @@ public sealed partial class DirectExecutionBackend
 		else
 		{
 			ulong largeCommitBase = faultAddress & 0xFFFFFFFFFFE00000uL;
-			if (TryCommitRange(_hostMemory, largeCommitBase, 2097152uL, commitProtect))
+			if (TryCommitRange(largeCommitBase, 2097152uL, commitProtect))
 			{
 				committed = true;
 				committedBase = largeCommitBase;
@@ -1021,19 +1344,19 @@ public sealed partial class DirectExecutionBackend
 		if (!committed)
 		{
 			ulong region64kBase = faultAddress & 0xFFFFFFFFFFFF0000uL;
-			if (TryCommitRange(_hostMemory, region64kBase, 65536uL, commitProtect))
+			if (TryCommitRange(region64kBase, 65536uL, commitProtect))
 			{
 				committed = true;
 				committedBase = region64kBase;
 				committedSize = 65536uL;
 			}
-			else if (TryCommitRange(_hostMemory, pageBase, 8192uL, commitProtect))
+			else if (TryCommitRange(pageBase, 8192uL, commitProtect))
 			{
 				committed = true;
 				committedBase = pageBase;
 				committedSize = 8192uL;
 			}
-			else if (TryCommitRange(_hostMemory, pageBase, 4096uL, commitProtect))
+			else if (TryCommitRange(pageBase, 4096uL, commitProtect))
 			{
 				committed = true;
 				committedBase = pageBase;
@@ -1046,7 +1369,7 @@ public sealed partial class DirectExecutionBackend
 			return false;
 		}
 
-		TryCommitRange(_hostMemory, pageBase + 4096, 4096uL, commitProtect);
+		TryCommitRange(pageBase + 4096, 4096uL, commitProtect);
 		if (traceLazyCommit)
 		{
 			Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit#{traceIndex}: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
@@ -1087,33 +1410,31 @@ public sealed partial class DirectExecutionBackend
 			return true;
 		}
 
-		// The commit protection is one of the two raw values ResolveLazyCommitProtection
-		// produces (0x40 RWX / 0x04 RW); the enum mapping reproduces those exactly.
-		static bool TryCommitRange(IHostMemory hostMemory, ulong baseAddress, ulong length, uint protection)
+		static unsafe bool TryCommitRange(ulong baseAddress, ulong length, uint protection)
 		{
 			if (length == 0)
 			{
 				return false;
 			}
-			return hostMemory.Commit(baseAddress, length, protection == 64u ? HostPageProtection.ReadWriteExecute : HostPageProtection.ReadWrite);
+			return VirtualAlloc((void*)baseAddress, (nuint)length, 4096u, protection) != null;
 		}
 
-		static bool TryReserveRange(IHostMemory hostMemory, ulong baseAddress, ulong length)
+		static unsafe bool TryReserveRange(ulong baseAddress, ulong length)
 		{
 			if (length == 0)
 			{
 				return false;
 			}
-			return hostMemory.Reserve(baseAddress, length, HostPageProtection.ReadWrite) != 0;
+			return VirtualAlloc((void*)baseAddress, (nuint)length, 8192u, 4u) != null;
 		}
 
-		static bool TryReserveThenCommit(IHostMemory hostMemory, ulong reserveAddress, ulong reserveSize, ulong commitAddress, ulong commitSize, uint protection)
+		static bool TryReserveThenCommit(ulong reserveAddress, ulong reserveSize, ulong commitAddress, ulong commitSize, uint protection)
 		{
-			if (!TryReserveRange(hostMemory, reserveAddress, reserveSize))
+			if (!TryReserveRange(reserveAddress, reserveSize))
 			{
 				return false;
 			}
-			return TryCommitRange(hostMemory, commitAddress, commitSize, protection);
+			return TryCommitRange(commitAddress, commitSize, protection);
 		}
 
 		static bool IsAccessCompatible(ulong accessType, uint protection)

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.IllegalInstruction.cs

@@ -0,0 +1,362 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Buffers.Binary;
+using System.Threading;
+using Iced.Intel;
+using SharpEmu.Core.Cpu.Emulation;
+
+namespace SharpEmu.Core.Cpu.Native;
+
+// Software fallback for the BMI1/BMI2/ABM general-purpose-register instructions.
+//
+// Guest code runs natively, so the guest and host share the same virtual address space and the
+// same registers (the OS delivers them in the CONTEXT record on a fault). When the host CPU lacks
+// one of these extensions it raises #UD instead of executing the opcode; without this the title
+// simply aborts. Here we decode the faulting instruction, evaluate it against the trapped register
+// and memory state, write the result back into the CONTEXT, step RIP past the instruction and ask
+// the OS to continue. Only the register-only BMI/ABM forms are handled; anything else returns false
+// and falls through to the existing diagnostics unchanged, so this can never mis-handle an opcode it
+// does not fully model.
+public sealed partial class DirectExecutionBackend
+{
+    // Windows x64 CONTEXT.EFlags lives just past the segment selectors. The GPR offsets it shares
+    // with the rest of the backend are the CTX_* constants declared in DirectExecutionBackend.cs.
+    private const int CTX_EFLAGS = 68;
+
+    // STATUS_ILLEGAL_INSTRUCTION (#UD surfaced by the Windows vectored handler).
+    private const uint StatusIllegalInstruction = 0xC000001Du;
+
+    private const int MaxInstructionBytes = 15;
+
+    // Instruction-window sizes tried in turn so a fault near a page boundary still decodes.
+    private static readonly int[] DecodeWindowSizes = { MaxInstructionBytes, 11, 8, 4, 2 };
+
+    private static int _bmiSoftwareFallbackAnnounced;
+    private static long _bmiInstructionsEmulated;
+
+    private unsafe bool TryRecoverIllegalInstruction(void* contextRecord, ulong rip)
+    {
+        if (!TryReadFaultingInstruction(rip, out var instruction))
+        {
+            return false;
+        }
+
+        if (instruction.Op0Kind != OpKind.Register ||
+            !TryGetGprSlot(instruction.Op0Register, out var destOffset, out var size))
+        {
+            return false;
+        }
+
+        if (!TryEvaluate(contextRecord, in instruction, size, out var result, out var flagsChanged, out var eflags))
+        {
+            return false;
+        }
+
+        WriteCtxU64(contextRecord, destOffset, result);
+        if (flagsChanged)
+        {
+            WriteCtxU32(contextRecord, CTX_EFLAGS, eflags);
+        }
+
+        WriteCtxU64(contextRecord, CTX_RIP, rip + (ulong)instruction.Length);
+
+        Interlocked.Increment(ref _bmiInstructionsEmulated);
+        if (Interlocked.Exchange(ref _bmiSoftwareFallbackAnnounced, 1) == 0)
+        {
+            Console.Error.WriteLine(
+                "[LOADER][INFO] Host lacks a BMI/ABM extension used by the guest; " +
+                "emulating those instructions in software.");
+        }
+
+        return true;
+    }
+
+    private unsafe bool TryEvaluate(
+        void* contextRecord,
+        in Instruction instruction,
+        GprOperandSize size,
+        out ulong result,
+        out bool flagsChanged,
+        out uint eflags)
+    {
+        result = 0;
+        flagsChanged = false;
+        eflags = ReadCtxU32(contextRecord, CTX_EFLAGS);
+
+        switch (instruction.Mnemonic)
+        {
+            case Mnemonic.Andn:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var andnSrc1) ||
+                    !TryReadOperand(contextRecord, in instruction, 2, size, out var andnSrc2))
+                {
+                    return false;
+                }
+
+                result = BmiInstructionEmulator.Andn(andnSrc1, andnSrc2, size, ref eflags);
+                flagsChanged = true;
+                return true;
+
+            case Mnemonic.Blsi:
+            case Mnemonic.Blsmsk:
+            case Mnemonic.Blsr:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var blsSrc))
+                {
+                    return false;
+                }
+
+                result = instruction.Mnemonic switch
+                {
+                    Mnemonic.Blsi => BmiInstructionEmulator.Blsi(blsSrc, size, ref eflags),
+                    Mnemonic.Blsmsk => BmiInstructionEmulator.Blsmsk(blsSrc, size, ref eflags),
+                    _ => BmiInstructionEmulator.Blsr(blsSrc, size, ref eflags),
+                };
+                flagsChanged = true;
+                return true;
+
+            case Mnemonic.Bextr:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var bextrSrc) ||
+                    !TryReadOperand(contextRecord, in instruction, 2, size, out var bextrControl))
+                {
+                    return false;
+                }
+
+                result = BmiInstructionEmulator.Bextr(bextrSrc, bextrControl, size, ref eflags);
+                flagsChanged = true;
+                return true;
+
+            case Mnemonic.Bzhi:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var bzhiSrc) ||
+                    !TryReadOperand(contextRecord, in instruction, 2, size, out var bzhiIndex))
+                {
+                    return false;
+                }
+
+                result = BmiInstructionEmulator.Bzhi(bzhiSrc, bzhiIndex, size, ref eflags);
+                flagsChanged = true;
+                return true;
+
+            case Mnemonic.Tzcnt:
+            case Mnemonic.Lzcnt:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var cntSrc))
+                {
+                    return false;
+                }
+
+                result = instruction.Mnemonic == Mnemonic.Tzcnt
+                    ? BmiInstructionEmulator.Tzcnt(cntSrc, size, ref eflags)
+                    : BmiInstructionEmulator.Lzcnt(cntSrc, size, ref eflags);
+                flagsChanged = true;
+                return true;
+
+            case Mnemonic.Rorx:
+                if (instruction.Op2Kind != OpKind.Immediate8 ||
+                    !TryReadOperand(contextRecord, in instruction, 1, size, out var rorxSrc))
+                {
+                    return false;
+                }
+
+                result = BmiInstructionEmulator.Rorx(rorxSrc, instruction.Immediate8, size);
+                return true;
+
+            case Mnemonic.Sarx:
+            case Mnemonic.Shlx:
+            case Mnemonic.Shrx:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var shiftSrc) ||
+                    !TryReadOperand(contextRecord, in instruction, 2, size, out var shiftCount))
+                {
+                    return false;
+                }
+
+                result = instruction.Mnemonic switch
+                {
+                    Mnemonic.Sarx => BmiInstructionEmulator.Sarx(shiftSrc, (int)shiftCount, size),
+                    Mnemonic.Shlx => BmiInstructionEmulator.Shlx(shiftSrc, (int)shiftCount, size),
+                    _ => BmiInstructionEmulator.Shrx(shiftSrc, (int)shiftCount, size),
+                };
+                return true;
+
+            case Mnemonic.Pdep:
+            case Mnemonic.Pext:
+                if (!TryReadOperand(contextRecord, in instruction, 1, size, out var packSrc) ||
+                    !TryReadOperand(contextRecord, in instruction, 2, size, out var packMask))
+                {
+                    return false;
+                }
+
+                result = instruction.Mnemonic == Mnemonic.Pdep
+                    ? BmiInstructionEmulator.Pdep(packSrc, packMask, size)
+                    : BmiInstructionEmulator.Pext(packSrc, packMask, size);
+                return true;
+
+            default:
+                return false;
+        }
+    }
+
+    private unsafe bool TryReadFaultingInstruction(ulong rip, out Instruction instruction)
+    {
+        // Try the full instruction window first, then shrink so a fault near the end of a mapped
+        // page (where fewer than 15 bytes are readable) still decodes.
+        foreach (var attempt in DecodeWindowSizes)
+        {
+            var buffer = new byte[attempt];
+            if (!TryReadHostBytes(rip, buffer))
+            {
+                continue;
+            }
+
+            var decoder = Decoder.Create(64, new ByteArrayCodeReader(buffer));
+            decoder.IP = rip;
+            decoder.Decode(out instruction);
+            if (instruction.Code != Code.INVALID && instruction.Length > 0 && instruction.Length <= attempt)
+            {
+                return true;
+            }
+        }
+
+        instruction = default;
+        return false;
+    }
+
+    private unsafe bool TryReadOperand(
+        void* contextRecord,
+        in Instruction instruction,
+        int operandIndex,
+        GprOperandSize size,
+        out ulong value)
+    {
+        value = 0;
+        switch (instruction.GetOpKind(operandIndex))
+        {
+            case OpKind.Register:
+                if (!TryGetGprSlot(instruction.GetOpRegister(operandIndex), out var offset, out _))
+                {
+                    return false;
+                }
+
+                var raw = ReadCtxU64(contextRecord, offset);
+                value = size == GprOperandSize.Bits64 ? raw : raw & 0xFFFF_FFFFUL;
+                return true;
+
+            case OpKind.Memory:
+                if (!TryComputeMemoryAddress(contextRecord, in instruction, out var address))
+                {
+                    return false;
+                }
+
+                var byteCount = size == GprOperandSize.Bits64 ? 8 : 4;
+                var buffer = new byte[byteCount];
+                if (!TryReadHostBytes(address, buffer))
+                {
+                    return false;
+                }
+
+                value = byteCount == 8
+                    ? BinaryPrimitives.ReadUInt64LittleEndian(buffer)
+                    : BinaryPrimitives.ReadUInt32LittleEndian(buffer);
+                return true;
+
+            default:
+                return false;
+        }
+    }
+
+    private unsafe bool TryComputeMemoryAddress(void* contextRecord, in Instruction instruction, out ulong address)
+    {
+        address = 0;
+
+        // FS/GS-relative operands need the guest segment base, which is not modelled here.
+        if (instruction.SegmentPrefix != Register.None)
+        {
+            return false;
+        }
+
+        if (instruction.IsIPRelativeMemoryOperand)
+        {
+            address = instruction.IPRelativeMemoryAddress;
+            return true;
+        }
+
+        var effective = instruction.MemoryDisplacement64;
+        if (instruction.MemoryBase != Register.None)
+        {
+            if (!TryGetGpr64Offset(instruction.MemoryBase, out var baseOffset))
+            {
+                return false;
+            }
+
+            effective += ReadCtxU64(contextRecord, baseOffset);
+        }
+
+        if (instruction.MemoryIndex != Register.None)
+        {
+            if (!TryGetGpr64Offset(instruction.MemoryIndex, out var indexOffset))
+            {
+                return false;
+            }
+
+            effective += ReadCtxU64(contextRecord, indexOffset) * (ulong)instruction.MemoryIndexScale;
+        }
+
+        address = effective;
+        return true;
+    }
+
+    // Maps a 32- or 64-bit GPR to its CONTEXT offset and reports the operand width it implies.
+    private static bool TryGetGprSlot(Register register, out int offset, out GprOperandSize size)
+    {
+        switch (register)
+        {
+            case Register.EAX: offset = CTX_RAX; size = GprOperandSize.Bits32; return true;
+            case Register.ECX: offset = CTX_RCX; size = GprOperandSize.Bits32; return true;
+            case Register.EDX: offset = CTX_RDX; size = GprOperandSize.Bits32; return true;
+            case Register.EBX: offset = CTX_RBX; size = GprOperandSize.Bits32; return true;
+            case Register.ESP: offset = CTX_RSP; size = GprOperandSize.Bits32; return true;
+            case Register.EBP: offset = CTX_RBP; size = GprOperandSize.Bits32; return true;
+            case Register.ESI: offset = CTX_RSI; size = GprOperandSize.Bits32; return true;
+            case Register.EDI: offset = CTX_RDI; size = GprOperandSize.Bits32; return true;
+            case Register.R8D: offset = CTX_R8; size = GprOperandSize.Bits32; return true;
+            case Register.R9D: offset = CTX_R9; size = GprOperandSize.Bits32; return true;
+            case Register.R10D: offset = CTX_R10; size = GprOperandSize.Bits32; return true;
+            case Register.R11D: offset = CTX_R11; size = GprOperandSize.Bits32; return true;
+            case Register.R12D: offset = CTX_R12; size = GprOperandSize.Bits32; return true;
+            case Register.R13D: offset = CTX_R13; size = GprOperandSize.Bits32; return true;
+            case Register.R14D: offset = CTX_R14; size = GprOperandSize.Bits32; return true;
+            case Register.R15D: offset = CTX_R15; size = GprOperandSize.Bits32; return true;
+            default:
+                if (TryGetGpr64Offset(register, out offset))
+                {
+                    size = GprOperandSize.Bits64;
+                    return true;
+                }
+
+                size = GprOperandSize.Bits32;
+                return false;
+        }
+    }
+
+    private static bool TryGetGpr64Offset(Register register, out int offset)
+    {
+        switch (register)
+        {
+            case Register.RAX: offset = CTX_RAX; return true;
+            case Register.RCX: offset = CTX_RCX; return true;
+            case Register.RDX: offset = CTX_RDX; return true;
+            case Register.RBX: offset = CTX_RBX; return true;
+            case Register.RSP: offset = CTX_RSP; return true;
+            case Register.RBP: offset = CTX_RBP; return true;
+            case Register.RSI: offset = CTX_RSI; return true;
+            case Register.RDI: offset = CTX_RDI; return true;
+            case Register.R8: offset = CTX_R8; return true;
+            case Register.R9: offset = CTX_R9; return true;
+            case Register.R10: offset = CTX_R10; return true;
+            case Register.R11: offset = CTX_R11; return true;
+            case Register.R12: offset = CTX_R12; return true;
+            case Register.R13: offset = CTX_R13; return true;
+            case Register.R14: offset = CTX_R14; return true;
+            case Register.R15: offset = CTX_R15; return true;
+            default: offset = 0; return false;
+        }
+    }
+}

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.NativeWorker.cs

@@ -6,7 +6,6 @@ using System.Collections.Generic;
 using System.Runtime.InteropServices;
 using System.Threading;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
 
 namespace SharpEmu.Core.Cpu.Native;
 
@@ -36,6 +35,20 @@ public sealed partial class DirectExecutionBackend
 	private bool _nativeWorkersDisposed;
 	private int _nativeWorkerCreationFailedLogged;
 
+	private const uint StackSizeParamIsAReservation = 0x00010000u;
+
+	[DllImport("kernel32.dll", SetLastError = true)]
+	private static extern nint CreateThread(
+		nint lpThreadAttributes,
+		nuint dwStackSize,
+		nint lpStartAddress,
+		nint lpParameter,
+		uint dwCreationFlags,
+		out uint lpThreadId);
+
+	[DllImport("kernel32.dll", SetLastError = true)]
+	private static extern uint WaitForSingleObject(nint hHandle, uint dwMilliseconds);
+
 	// Runs an emitted guest entry stub. Preferred path is a pooled native worker
 	// thread; falls back to the historical inline calli (guest frames above this
 	// thread's managed frames) when workers are disabled or unavailable.
@@ -48,7 +61,7 @@ public sealed partial class DirectExecutionBackend
 		var worker = RentNativeGuestExecutor();
 		if (worker is null)
 		{
-			_hostThreading.SetTlsValue(_hostRspSlotTlsIndex, (nint)hostRspSlot);
+			TlsSetValue(_hostRspSlotTlsIndex, (nint)hostRspSlot);
 			return CallNativeEntry(entryStub);
 		}
 		try
@@ -79,7 +92,10 @@ public sealed partial class DirectExecutionBackend
 
 	private NativeGuestExecutor? RentNativeGuestExecutor()
 	{
-		if (NativeGuestWorkersDisabled)
+		// NativeGuestExecutor emits a Win32 wait loop and creates it with
+		// kernel32!CreateThread. POSIX hosts use the established inline entry
+		// path until the worker loop has a pthread/eventfd implementation.
+		if (!OperatingSystem.IsWindows() || NativeGuestWorkersDisabled)
 		{
 			return null;
 		}
@@ -172,8 +188,20 @@ public sealed partial class DirectExecutionBackend
 		private static nint _exitThreadAddress;
 
 		private readonly DirectExecutionBackend _backend;
-		private readonly AutoResetEvent _workAvailable = new(false);
-		private readonly AutoResetEvent _workCompleted = new(false);
+		// Windows uses AutoResetEvent (its SafeWaitHandle is a real kernel
+		// event the emitted loop can wait on); POSIX uses worker-event
+		// semaphores shared the same way via PosixHostStubs.
+		private readonly AutoResetEvent? _workAvailable;
+		private readonly AutoResetEvent? _workCompleted;
+		private nint _workSemaphore;
+		private nint _doneSemaphore;
+
+		// RunPrologue/RunEpilogue compile to the host ABI (SysV on POSIX); the
+		// emitted loop calls them with Win64 registers, so POSIX routes the
+		// calls through register-shuffling thunks (shared by all workers).
+		private static nint _posixPrologueThunk;
+		private static nint _posixEpilogueThunk;
+		private static readonly object PosixThunkGate = new();
 		private GCHandle _selfHandle;
 		private void* _controlBlock;
 		private void* _loopStub;
@@ -212,11 +240,16 @@ public sealed partial class DirectExecutionBackend
 		private NativeGuestExecutor(DirectExecutionBackend backend)
 		{
 			_backend = backend;
+			if (OperatingSystem.IsWindows())
+			{
+				_workAvailable = new AutoResetEvent(false);
+				_workCompleted = new AutoResetEvent(false);
+			}
 		}
 
 		public static NativeGuestExecutor? TryCreate(DirectExecutionBackend backend)
 		{
-			if (!EnsureHostRuntimeExports(backend._hostSymbols))
+			if (!EnsureKernel32Exports())
 			{
 				return null;
 			}
@@ -229,27 +262,32 @@ public sealed partial class DirectExecutionBackend
 			return executor;
 		}
 
-		private static bool EnsureHostRuntimeExports(IHostSymbolResolver symbols)
+		private static bool EnsureKernel32Exports()
 		{
 			if (_exitThreadAddress != 0)
 			{
 				return _waitForSingleObjectAddress != 0 && _setEventAddress != 0;
 			}
-			_waitForSingleObjectAddress = symbols.GetAddress(HostRuntimeFunction.WaitForSingleObject);
-			_setEventAddress = symbols.GetAddress(HostRuntimeFunction.SetEvent);
-			_exitThreadAddress = symbols.GetAddress(HostRuntimeFunction.ExitThread);
+			nint kernel32 = GetModuleHandle("kernel32.dll");
+			if (kernel32 == 0)
+			{
+				return false;
+			}
+			_waitForSingleObjectAddress = GetProcAddress(kernel32, "WaitForSingleObject");
+			_setEventAddress = GetProcAddress(kernel32, "SetEvent");
+			_exitThreadAddress = GetProcAddress(kernel32, "ExitThread");
 			return _waitForSingleObjectAddress != 0 && _setEventAddress != 0 && _exitThreadAddress != 0;
 		}
 
 		private bool Initialize()
 		{
 			_selfHandle = GCHandle.Alloc(this);
-			_controlBlock = (void*)_backend._hostMemory.Allocate(0, 4096u, HostPageProtection.ReadWrite);
+			_controlBlock = VirtualAlloc(null, 4096u, 12288u, 4u);
 			if (_controlBlock == null)
 			{
 				return false;
 			}
-			_loopStub = (void*)_backend._hostMemory.Allocate(0, LoopStubSize, HostPageProtection.ReadWriteExecute);
+			_loopStub = VirtualAlloc(null, LoopStubSize, 12288u, 64u);
 			if (_loopStub == null)
 			{
 				return false;
@@ -258,8 +296,34 @@ public sealed partial class DirectExecutionBackend
 			var prologuePtr = (nint)(delegate* unmanaged<nint, nint>)&RunPrologue;
 			var epiloguePtr = (nint)(delegate* unmanaged<nint, int, void>)&RunEpilogue;
 			var executorHandle = GCHandle.ToIntPtr(_selfHandle);
-			var workHandle = _workAvailable.SafeWaitHandle.DangerousGetHandle();
-			var doneHandle = _workCompleted.SafeWaitHandle.DangerousGetHandle();
+			nint workHandle;
+			nint doneHandle;
+			if (OperatingSystem.IsWindows())
+			{
+				workHandle = _workAvailable!.SafeWaitHandle.DangerousGetHandle();
+				doneHandle = _workCompleted!.SafeWaitHandle.DangerousGetHandle();
+			}
+			else
+			{
+				lock (PosixThunkGate)
+				{
+					if (_posixPrologueThunk == 0)
+					{
+						_posixPrologueThunk = PosixHostStubs.CreateWin64ToSysVThunk(prologuePtr);
+						_posixEpilogueThunk = PosixHostStubs.CreateWin64ToSysVThunk(epiloguePtr);
+					}
+				}
+				prologuePtr = _posixPrologueThunk;
+				epiloguePtr = _posixEpilogueThunk;
+				_workSemaphore = PosixHostStubs.CreateWorkerEvent();
+				_doneSemaphore = PosixHostStubs.CreateWorkerEvent();
+				if (_workSemaphore == 0 || _doneSemaphore == 0)
+				{
+					return false;
+				}
+				workHandle = _workSemaphore;
+				doneHandle = _doneSemaphore;
+			}
 
 			byte* code = (byte*)_loopStub;
 			int offset = 0;
@@ -331,15 +395,17 @@ public sealed partial class DirectExecutionBackend
 			*(int*)(code + skipJump) = skipEntryOffset - (skipJump + sizeof(int));
 
 			uint oldProtect = 0;
-			if (!_backend._hostMemory.Protect((ulong)_loopStub, LoopStubSize, HostPageProtection.ReadExecute, out oldProtect))
+			if (!VirtualProtect(_loopStub, LoopStubSize, 32u, &oldProtect))
 			{
 				return false;
 			}
-			_backend._hostMemory.FlushInstructionCache((ulong)_loopStub, LoopStubSize);
-			_threadHandle = _backend._hostThreading.CreateNativeThread(
-				(nint)_loopStub,
+			FlushInstructionCache(GetCurrentProcess(), _loopStub, LoopStubSize);
+			_threadHandle = CreateThread(
 				0,
 				WorkerStackReservation,
+				(nint)_loopStub,
+				0,
+				StackSizeParamIsAReservation,
 				out _nativeThreadId);
 			if (_threadHandle == 0)
 			{
@@ -377,8 +443,8 @@ public sealed partial class DirectExecutionBackend
 			_runYieldRequested = false;
 			_runYieldReason = null;
 			_runForcedExit = false;
-			_workAvailable.Set();
-			_workCompleted.WaitOne();
+			SignalWorkAvailable();
+			WaitWorkCompleted();
 			_runContext = null;
 			_runState = null;
 			yieldRequested = _runYieldRequested;
@@ -391,6 +457,28 @@ public sealed partial class DirectExecutionBackend
 			return _runNativeResult;
 		}
 
+		private void SignalWorkAvailable()
+		{
+			if (_workAvailable is not null)
+			{
+				_workAvailable.Set();
+				return;
+			}
+
+			_ = PosixHostStubs.SignalWorkerEvent(_workSemaphore);
+		}
+
+		private void WaitWorkCompleted()
+		{
+			if (_workCompleted is not null)
+			{
+				_workCompleted.WaitOne();
+				return;
+			}
+
+			_ = PosixHostStubs.WaitWorkerEvent(_doneSemaphore, -1);
+		}
+
 		[UnmanagedCallersOnly]
 		private static nint RunPrologue(nint executorHandle)
 		{
@@ -445,7 +533,7 @@ public sealed partial class DirectExecutionBackend
 			_prevYieldRequested = _activeGuestThreadYieldRequested;
 			_prevYieldReason = _activeGuestThreadYieldReason;
 			_prevState = _activeGuestThreadState;
-			_prevHostRspSlot = backend._hostThreading.GetTlsValue(backend._hostRspSlotTlsIndex);
+			_prevHostRspSlot = TlsGetValue(backend._hostRspSlotTlsIndex);
 			_prevGuestThreadHandle = GuestThreadExecution.EnterGuestThread(_runGuestThreadHandle);
 			_entered = true;
 			_activeExecutionBackend = backend;
@@ -457,11 +545,11 @@ public sealed partial class DirectExecutionBackend
 			_activeGuestThreadYieldReason = null;
 			_activeGuestThreadState = _runState;
 			backend.BindTlsBase(_runContext!);
-			backend._hostThreading.SetTlsValue(backend._hostRspSlotTlsIndex, _runHostRspSlot);
+			TlsSetValue(backend._hostRspSlotTlsIndex, _runHostRspSlot);
 			if (_runState is { } state)
 			{
 				_prevHostThreadId = Volatile.Read(ref state.HostThreadId);
-				Volatile.Write(ref state.HostThreadId, unchecked((int)backend._hostThreading.CurrentThreadId));
+				Volatile.Write(ref state.HostThreadId, unchecked((int)GetCurrentThreadId()));
 			}
 			if (_runAffinityMask != 0)
 			{
@@ -491,7 +579,7 @@ public sealed partial class DirectExecutionBackend
 			{
 				Volatile.Write(ref state.HostThreadId, _prevHostThreadId);
 			}
-			_backend._hostThreading.SetTlsValue(_backend._hostRspSlotTlsIndex, _prevHostRspSlot);
+			TlsSetValue(_backend._hostRspSlotTlsIndex, _prevHostRspSlot);
 			GuestThreadExecution.RestoreGuestThread(_prevGuestThreadHandle);
 			_activeExecutionBackend = _prevBackend;
 			_activeCpuContext = _prevContext;
@@ -520,7 +608,7 @@ public sealed partial class DirectExecutionBackend
 			}
 			try
 			{
-				_workAvailable.Set();
+				SignalWorkAvailable();
 			}
 			catch (ObjectDisposedException)
 			{
@@ -528,8 +616,8 @@ public sealed partial class DirectExecutionBackend
 			var exited = _threadHandle == 0;
 			if (_threadHandle != 0)
 			{
-				exited = _backend._hostThreading.WaitForThreadExit(_threadHandle, 1000u);
-				_backend._hostThreading.CloseThreadHandle(_threadHandle);
+				exited = WaitForSingleObject(_threadHandle, 1000u) == 0u;
+				CloseHandle(_threadHandle);
 				_threadHandle = 0;
 			}
 			if (!exited)
@@ -543,20 +631,30 @@ public sealed partial class DirectExecutionBackend
 			}
 			if (_loopStub != null)
 			{
-				_backend._hostMemory.Free((ulong)_loopStub);
+				VirtualFree(_loopStub, 0u, 32768u);
 				_loopStub = null;
 			}
 			if (_controlBlock != null)
 			{
-				_backend._hostMemory.Free((ulong)_controlBlock);
+				VirtualFree(_controlBlock, 0u, 32768u);
 				_controlBlock = null;
 			}
 			if (_selfHandle.IsAllocated)
 			{
 				_selfHandle.Free();
 			}
-			_workAvailable.Dispose();
-			_workCompleted.Dispose();
+			_workAvailable?.Dispose();
+			_workCompleted?.Dispose();
+			if (_workSemaphore != 0)
+			{
+				PosixHostStubs.DestroyWorkerEvent(_workSemaphore);
+				_workSemaphore = 0;
+			}
+			if (_doneSemaphore != 0)
+			{
+				PosixHostStubs.DestroyWorkerEvent(_doneSemaphore);
+				_doneSemaphore = 0;
+			}
 		}
 	}
 }

src/SharpEmu.Core/Cpu/Native/DirectExecutionBackend.PosixSignals.cs

@@ -0,0 +1,409 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using System.Runtime.InteropServices;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Core.Cpu.Native;
+
+public sealed unsafe partial class DirectExecutionBackend
+{
+	// POSIX bridge for the Windows vectored-exception-handler logic. A
+	// sigaction(SIGSEGV/SIGBUS/SIGILL) handler rebuilds the EXCEPTION_POINTERS
+	// view the shared handlers expect (Win64 CONTEXT register offsets) from
+	// the signal's mcontext, runs the same recovery chain the VEH path uses
+	// (unresolved-import trap sentinels, demand-paging of lazily-committed
+	// guest pages, fault diagnostics), and writes register changes back into
+	// the mcontext so sigreturn resumes the repaired guest. Unrecovered
+	// faults are forwarded to the previously installed handler so the .NET
+	// runtime keeps turning its own faults into managed exceptions.
+
+	private const int PosixSigIll = 4;
+	private const int PosixSigTrap = 5;
+	private const int PosixSigAbort = 6;
+	private const int PosixSigSegv = 11;
+	private static readonly int PosixSigBus = OperatingSystem.IsMacOS() ? 10 : 7;
+
+	// struct sigaction: the handler pointer leads on both platforms; Darwin
+	// packs { handler(8), mask(4), flags(4) }, Linux glibc/musl packs
+	// { handler(8), mask(128), flags(4), restorer(8) }.
+	private static readonly int PosixSigactionSize = OperatingSystem.IsMacOS() ? 16 : 152;
+	private static readonly int PosixSigactionFlagsOffset = OperatingSystem.IsMacOS() ? 12 : 136;
+
+	private static readonly int PosixSaSigInfo = OperatingSystem.IsMacOS() ? 0x0040 : 0x0004;
+	private static readonly int PosixSaNoDefer = OperatingSystem.IsMacOS() ? 0x0010 : 0x40000000;
+
+	// siginfo_t.si_addr: Darwin { signo, errno, code, pid, uid, status, addr },
+	// Linux { signo, errno, code, pad32, addr }.
+	private static readonly int PosixSigInfoAddressOffset = OperatingSystem.IsMacOS() ? 24 : 16;
+
+	// Darwin ucontext_t stores a pointer to __darwin_mcontext64 at +48; the
+	// general registers live in its __ss thread state after the 16-byte
+	// exception state. Linux glibc embeds mcontext_t inline at +40 with the
+	// registers in gregs[23]. Rosetta 2 delivers the regular x86-64 layout
+	// to translated processes.
+	private const int DarwinUcontextMcontextOffset = 48;
+	private const int DarwinMcontextErrOffset = 4;
+	private const int DarwinMcontextFaultAddressOffset = 8;
+	private const int LinuxUcontextGregsOffset = 40;
+	private const int LinuxGregsErrOffset = 19 * 8;
+
+	// Byte offsets of the general registers relative to GetPosixRegisterBase,
+	// ordered to match the contiguous Win64 CONTEXT block CTX_RAX..CTX_RIP
+	// (rax, rcx, rdx, rbx, rsp, rbp, rsi, rdi, r8..r15, rip). Verified
+	// against the x86-64 platform headers.
+	private static readonly int[] PosixRegisterOffsets = OperatingSystem.IsMacOS()
+		? new[] { 16, 32, 40, 24, 72, 64, 56, 48, 80, 88, 96, 104, 112, 120, 128, 136, 144 }
+		: new[] { 104, 112, 96, 88, 120, 80, 72, 64, 0, 8, 16, 24, 32, 40, 48, 56, 128 };
+
+	private static DirectExecutionBackend? _posixSignalBackend;
+	private static bool _posixSignalHandlersInstalled;
+	private static bool _posixRawRecoveryEnabled;
+	private static bool _posixSignalWarmup;
+	private static readonly nint[] _posixPreviousActions = new nint[32];
+	private static int _posixSignalTraceCount;
+	private static long _perfSignalCount;
+	private static readonly bool _perfSignalCounter =
+		string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_PERF_MEM"), "1", StringComparison.Ordinal);
+
+	[ThreadStatic]
+	private static int _posixSignalHandlerDepth;
+
+	private void SetupPosixExceptionHandler()
+	{
+		if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_DISABLE_POSIX_SIGNALS"), "1", StringComparison.Ordinal))
+		{
+			Console.Error.WriteLine("[LOADER][WARN] POSIX signal exception bridge disabled by SHARPEMU_DISABLE_POSIX_SIGNALS=1; guest faults will not be recovered.");
+			return;
+		}
+
+		_posixSignalBackend = this;
+		if (_posixSignalHandlersInstalled)
+		{
+			return;
+		}
+
+		_posixRawRecoveryEnabled = !string.Equals(
+			Environment.GetEnvironmentVariable("SHARPEMU_DISABLE_RAW_HANDLER"), "1", StringComparison.Ordinal);
+		if (!_posixRawRecoveryEnabled)
+		{
+			Console.Error.WriteLine("[LOADER][INFO] Raw sentinel recovery disabled by SHARPEMU_DISABLE_RAW_HANDLER=1");
+		}
+
+		WarmUpPosixSignalPath();
+		SharpEmu.HLE.GuestImageWriteTracker.WarmUp();
+
+		if (!InstallPosixSignalHandler(PosixSigSegv) ||
+			!InstallPosixSignalHandler(PosixSigBus) ||
+			!InstallPosixSignalHandler(PosixSigIll) ||
+			!InstallPosixSignalHandler(PosixSigTrap) ||
+			!InstallPosixSignalHandler(PosixSigAbort))
+		{
+			throw new InvalidOperationException("Failed to install POSIX fault signal handlers");
+		}
+
+		_posixSignalHandlersInstalled = true;
+		Console.Error.WriteLine("[LOADER][INFO] POSIX signal exception bridge installed (SIGSEGV/SIGBUS/SIGILL)");
+	}
+
+	/// <summary>
+	/// Runs the signal-recovery path once with fabricated inputs before the
+	/// handlers are installed. The first entry into the handler must not
+	/// require JIT compilation (a fault can interrupt arbitrary runtime
+	/// states), and under Rosetta 2 the signal trampoline cannot enter x86
+	/// code that has never been executed (and therefore never translated): a
+	/// cold handler is silently never invoked and the faulting instruction
+	/// retries forever.
+	/// </summary>
+	private void WarmUpPosixSignalPath()
+	{
+		byte* fakeUcontext = stackalloc byte[512];
+		new Span<byte>(fakeUcontext, 512).Clear();
+		byte* fakeMcontext = stackalloc byte[512];
+		new Span<byte>(fakeMcontext, 512).Clear();
+		if (OperatingSystem.IsMacOS())
+		{
+			*(byte**)(fakeUcontext + DarwinUcontextMcontextOffset) = fakeMcontext;
+		}
+
+		_posixSignalWarmup = true;
+		try
+		{
+			((delegate* unmanaged<int, nint, nint, void>)&HandlePosixSignal)(PosixSigSegv, 0, (nint)fakeUcontext);
+
+			// Warm the branches the fabricated fault above skips without
+			// spamming diagnostics: the benign-exception path through
+			// VectoredHandler, the lazy-commit probe (fault address 0 bails
+			// out immediately), and the chain helper (signal 0 has no saved
+			// action and sigaction(0, ...) fails with EINVAL).
+			EXCEPTION_RECORD record = default;
+			record.ExceptionCode = DBG_PRINTEXCEPTION_C;
+			byte* contextRecord = stackalloc byte[Win64ContextSize];
+			new Span<byte>(contextRecord, Win64ContextSize).Clear();
+			EXCEPTION_POINTERS pointers;
+			pointers.ExceptionRecord = &record;
+			pointers.ContextRecord = contextRecord;
+			_ = VectoredHandler(&pointers);
+
+			record.ExceptionCode = 3221225477u;
+			record.NumberParameters = 2;
+			// 0x70000 is never guest-owned, so this walks the vmem region
+			// scan and the PRT range check, then bails out silently.
+			record.ExceptionInformation[1] = 0x70000;
+			_ = TryHandleLazyCommittedPage(&record, 0, 0);
+			ChainPreviousPosixAction(0, 0, 0);
+		}
+		finally
+		{
+			_posixSignalWarmup = false;
+		}
+	}
+
+	private static bool InstallPosixSignalHandler(int signal)
+	{
+		byte* action = stackalloc byte[PosixSigactionSize];
+		new Span<byte>(action, PosixSigactionSize).Clear();
+		*(nint*)action = (nint)(delegate* unmanaged<int, nint, nint, void>)&HandlePosixSignal;
+		// No SA_ONSTACK: the runtime's alternate stacks are far too small for
+		// the recovery/diagnostic path (JIT compilation of cold handler code
+		// can run inside the signal frame). Guest faults deliver onto the 2MB
+		// guest stack, host faults onto the regular thread stack — the same
+		// stacks Windows dispatches exceptions on.
+		*(int*)(action + PosixSigactionFlagsOffset) = PosixSaSigInfo | PosixSaNoDefer;
+
+		var previous = (byte*)NativeMemory.AllocZeroed((nuint)PosixSigactionSize);
+		if (sigaction(signal, action, previous) != 0)
+		{
+			NativeMemory.Free(previous);
+			Console.Error.WriteLine($"[LOADER][ERROR] sigaction({signal}) failed: errno={Marshal.GetLastPInvokeError()}");
+			return false;
+		}
+
+		_posixPreviousActions[signal] = (nint)previous;
+		return true;
+	}
+
+	[UnmanagedCallersOnly]
+	private static void HandlePosixSignal(int signal, nint siginfo, nint ucontext)
+	{
+		if (_posixSignalHandlerDepth > 0)
+		{
+			// A fault inside our own fault handler (diagnostics touched an
+			// unmapped address): restore the default action and return so the
+			// re-executed instruction terminates the process.
+			RestoreDefaultPosixAction(signal);
+			return;
+		}
+
+		_posixSignalHandlerDepth++;
+		if (_perfSignalCounter)
+		{
+			var n = Interlocked.Increment(ref _perfSignalCount);
+			if (n % 100000 == 0)
+			{
+				Console.Error.WriteLine($"[PERF][MEM] posix_faults={n}");
+			}
+		}
+		try
+		{
+			// Guest-image write tracking runs first: it only needs the fault
+			// address (safe for host and guest threads alike) and must resume
+			// the faulting write immediately after restoring write access.
+			if (signal != PosixSigIll &&
+				siginfo != 0 &&
+				SharpEmu.HLE.GuestImageWriteTracker.TryHandleWriteFault(
+					*(ulong*)((byte*)siginfo + PosixSigInfoAddressOffset)))
+			{
+				return;
+			}
+
+			if (TryHandlePosixFault(signal, siginfo, ucontext))
+			{
+				return;
+			}
+		}
+		catch
+		{
+			// A managed exception must never unwind out of a signal frame.
+		}
+		finally
+		{
+			_posixSignalHandlerDepth--;
+		}
+
+		ChainPreviousPosixAction(signal, siginfo, ucontext);
+	}
+
+	private static bool TryHandlePosixFault(int signal, nint siginfo, nint ucontext)
+	{
+		byte* registers = GetPosixRegisterBase(ucontext);
+		if (registers == null)
+		{
+			return false;
+		}
+
+		byte* contextRecord = stackalloc byte[Win64ContextSize];
+		new Span<byte>(contextRecord, Win64ContextSize).Clear();
+		int[] offsets = PosixRegisterOffsets;
+		for (int i = 0; i < offsets.Length; i++)
+		{
+			WriteCtxU64(contextRecord, CTX_RAX + i * 8, *(ulong*)(registers + offsets[i]));
+		}
+
+		EXCEPTION_RECORD record = default;
+		record.ExceptionAddress = (void*)ReadCtxU64(contextRecord, CTX_RIP);
+		if (signal == PosixSigIll)
+		{
+			record.ExceptionCode = 3221225501u;
+		}
+		else if (signal == PosixSigTrap)
+		{
+			record.ExceptionCode = 2147483651u;
+		}
+		else if (signal == PosixSigAbort)
+		{
+			record.ExceptionCode = 1073741845u;
+		}
+		else
+		{
+			ulong faultAddress = GetPosixFaultAddress(siginfo, registers);
+			record.ExceptionCode = 3221225477u;
+			record.NumberParameters = 2;
+			record.ExceptionInformation[0] = GetPosixAccessType(registers, faultAddress, ReadCtxU64(contextRecord, CTX_RIP));
+			record.ExceptionInformation[1] = faultAddress;
+		}
+
+		EXCEPTION_POINTERS pointers;
+		pointers.ExceptionRecord = &record;
+		pointers.ContextRecord = contextRecord;
+
+		int traceIndex = _posixSignalWarmup ? 0 : Interlocked.Increment(ref _posixSignalTraceCount);
+		bool traceSignal = traceIndex > 0 && (traceIndex <= 16 || traceIndex % 1024 == 0 ||
+			string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_POSIX_SIGNALS"), "1", StringComparison.Ordinal));
+		if (traceSignal)
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][TRACE] posix-signal#{traceIndex}: sig={signal} rip=0x{ReadCtxU64(contextRecord, CTX_RIP):X16} " +
+				$"fault=0x{record.ExceptionInformation[1]:X16} access={record.ExceptionInformation[0]} rsp=0x{ReadCtxU64(contextRecord, CTX_RSP):X16}");
+			Console.Error.Flush();
+		}
+
+		// Sentinel recovery runs first: on Windows both vectored handlers see
+		// every fault anyway, and recovering here avoids dumping the full
+		// VectoredHandler diagnostics for each recoverable trap.
+		int disposition = 0;
+		if (_posixRawRecoveryEnabled)
+		{
+			disposition = TryRecoverUnresolvedSentinel(&pointers);
+		}
+		if (disposition != -1 && !_posixSignalWarmup && _posixSignalBackend is { } backend)
+		{
+			disposition = backend.VectoredHandler(&pointers);
+		}
+		if (traceSignal)
+		{
+			Console.Error.WriteLine(
+				$"[LOADER][TRACE] posix-signal#{traceIndex}: recovered={disposition == -1} new_rip=0x{ReadCtxU64(contextRecord, CTX_RIP):X16}");
+			Console.Error.Flush();
+		}
+		if (disposition != -1 && !_posixSignalWarmup)
+		{
+			return false;
+		}
+
+		for (int i = 0; i < offsets.Length; i++)
+		{
+			*(ulong*)(registers + offsets[i]) = ReadCtxU64(contextRecord, CTX_RAX + i * 8);
+		}
+		return true;
+	}
+
+	private static byte* GetPosixRegisterBase(nint ucontext)
+	{
+		if (ucontext == 0)
+		{
+			return null;
+		}
+
+		if (OperatingSystem.IsMacOS())
+		{
+			return *(byte**)((byte*)ucontext + DarwinUcontextMcontextOffset);
+		}
+
+		return (byte*)ucontext + LinuxUcontextGregsOffset;
+	}
+
+	private static ulong GetPosixFaultAddress(nint siginfo, byte* registers)
+	{
+		ulong address = siginfo != 0 ? *(ulong*)((byte*)siginfo + PosixSigInfoAddressOffset) : 0;
+		if (address == 0 && OperatingSystem.IsMacOS())
+		{
+			address = *(ulong*)(registers + DarwinMcontextFaultAddressOffset);
+		}
+
+		return address;
+	}
+
+	private static ulong GetPosixAccessType(byte* registers, ulong faultAddress, ulong rip)
+	{
+		// x86 page-fault error code: bit 1 = write access, bit 4 = instruction
+		// fetch. Fall back to comparing the fault address against RIP when
+		// the error code is not populated (e.g. under Rosetta 2 translation).
+		ulong error = OperatingSystem.IsMacOS()
+			? *(uint*)(registers + DarwinMcontextErrOffset)
+			: *(ulong*)(registers + LinuxGregsErrOffset);
+		if ((error & 0x10) != 0)
+		{
+			return 8;
+		}
+		if ((error & 0x2) != 0)
+		{
+			return 1;
+		}
+
+		return faultAddress != 0 && faultAddress == rip ? 8u : 0u;
+	}
+
+	private static void RestoreDefaultPosixAction(int signal)
+	{
+		byte* action = stackalloc byte[PosixSigactionSize];
+		new Span<byte>(action, PosixSigactionSize).Clear();
+		_ = sigaction(signal, action, null);
+	}
+
+	private static void ChainPreviousPosixAction(int signal, nint siginfo, nint ucontext)
+	{
+		byte* previous = (uint)signal < (uint)_posixPreviousActions.Length
+			? (byte*)_posixPreviousActions[signal]
+			: null;
+		nint handler = previous != null ? *(nint*)previous : 0;
+		if (handler == 0)
+		{
+			// SIG_DFL (or nothing saved): reinstate the default action and
+			// return, so re-executing the faulting instruction terminates the
+			// process with the original fault context intact.
+			RestoreDefaultPosixAction(signal);
+			return;
+		}
+		if (handler == 1)
+		{
+			// SIG_IGN
+			return;
+		}
+
+		int flags = *(int*)(previous + PosixSigactionFlagsOffset);
+		if ((flags & PosixSaSigInfo) != 0)
+		{
+			((delegate* unmanaged<int, nint, nint, void>)handler)(signal, siginfo, ucontext);
+		}
+		else
+		{
+			((delegate* unmanaged<int, void>)handler)(signal);
+		}
+	}
+
+	[DllImport("libc", SetLastError = true)]
+	private static extern int sigaction(int signum, void* act, void* oldact);
+}

src/SharpEmu.Core/Cpu/Native/NullHostFaultHandling.cs

@@ -0,0 +1,49 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE.Host;
+
+namespace SharpEmu.Core.Cpu.Native;
+
+/// <summary>
+/// Placeholder for hosts whose fault bridge is installed directly by the
+/// execution backend. POSIX uses its sigaction bridge and never calls these
+/// Windows-shaped registration methods.
+/// </summary>
+internal sealed class NullHostFaultHandling : IHostFaultHandling
+{
+    public static NullHostFaultHandling Instance { get; } = new();
+
+    private NullHostFaultHandling()
+    {
+    }
+
+    public nint CreateHandlerThunk(nint managedCallback, uint hostRspSwitchTlsSlot, nint tlsGetValueAddress)
+    {
+        _ = managedCallback;
+        _ = hostRspSwitchTlsSlot;
+        _ = tlsGetValueAddress;
+        return 0;
+    }
+
+    public void FreeThunk(nint thunk)
+    {
+        _ = thunk;
+    }
+
+    public nint AddFirstChanceHandler(nint thunk)
+    {
+        _ = thunk;
+        return 0;
+    }
+
+    public void RemoveHandler(nint handle)
+    {
+        _ = handle;
+    }
+
+    public void SetUnhandledFilter(nint thunk)
+    {
+        _ = thunk;
+    }
+}

src/SharpEmu.Core/Cpu/Native/PosixHostStubs.cs

@@ -0,0 +1,425 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Core.Cpu.Native;
+
+/// <summary>
+/// POSIX replacements for the kernel32 helpers the native backend embeds in
+/// emitted x86-64 code. Every stub exposed here follows the Win64 calling
+/// convention the emitted call sites were written for (first argument in
+/// ECX, result in RAX, Win64 non-volatile registers preserved), so the
+/// emission code stays identical across platforms.
+/// </summary>
+internal static unsafe class PosixHostStubs
+{
+    private static readonly object Gate = new();
+    private static bool _initialized;
+    private static nint _tlsGetValueStub;
+    private static nint _queryPerformanceCounterStub;
+    private static nint _switchToThreadStub;
+    private static nint _sleepStub;
+
+    public static nint TlsGetValueStubAddress
+    {
+        get { EnsureInitialized(); return _tlsGetValueStub; }
+    }
+
+    public static nint QueryPerformanceCounterStubAddress
+    {
+        get { EnsureInitialized(); return _queryPerformanceCounterStub; }
+    }
+
+    public static nint SwitchToThreadStubAddress
+    {
+        get { EnsureInitialized(); return _switchToThreadStub; }
+    }
+
+    public static nint SleepStubAddress
+    {
+        get { EnsureInitialized(); return _sleepStub; }
+    }
+
+    public static nint CreateWorkerEvent()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            return dispatch_semaphore_create(0);
+        }
+
+        var semaphore = Marshal.AllocHGlobal(64);
+        if (sem_init(semaphore, 0, 0) != 0)
+        {
+            Marshal.FreeHGlobal(semaphore);
+            return 0;
+        }
+
+        return semaphore;
+    }
+
+    public static bool SignalWorkerEvent(nint handle)
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            _ = dispatch_semaphore_signal(handle);
+            return true;
+        }
+
+        return sem_post(handle) == 0;
+    }
+
+    public static bool WaitWorkerEvent(nint handle, int timeoutMilliseconds)
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            if (timeoutMilliseconds < 0)
+            {
+                return dispatch_semaphore_wait(handle, ulong.MaxValue) == 0;
+            }
+
+            var deadline = dispatch_time(0, timeoutMilliseconds * 1_000_000L);
+            return dispatch_semaphore_wait(handle, deadline) == 0;
+        }
+
+        if (timeoutMilliseconds < 0)
+        {
+            while (sem_wait(handle) != 0)
+            {
+            }
+
+            return true;
+        }
+
+        var deadlineTicks = Environment.TickCount64 + timeoutMilliseconds;
+        while (sem_trywait(handle) != 0)
+        {
+            if (Environment.TickCount64 >= deadlineTicks)
+            {
+                return false;
+            }
+
+            Thread.Sleep(1);
+        }
+
+        return true;
+    }
+
+    public static void DestroyWorkerEvent(nint handle)
+    {
+        if (handle == 0)
+        {
+            return;
+        }
+
+        if (OperatingSystem.IsMacOS())
+        {
+            dispatch_release(handle);
+            return;
+        }
+
+        _ = sem_destroy(handle);
+        Marshal.FreeHGlobal(handle);
+    }
+
+    /// <summary>Allocates a pthread TLS key, mirroring kernel32!TlsAlloc.</summary>
+    public static uint TlsAlloc()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            nuint key;
+            return pthread_key_create_mac(&key, 0) == 0 ? (uint)key : uint.MaxValue;
+        }
+
+        uint key32;
+        return pthread_key_create_linux(&key32, 0) == 0 ? key32 : uint.MaxValue;
+    }
+
+    public static bool TlsFree(uint key)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_key_delete_mac((nuint)key) == 0
+            : pthread_key_delete_linux(key) == 0;
+    }
+
+    public static bool TlsSetValue(uint key, nint value)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_setspecific_mac((nuint)key, value) == 0
+            : pthread_setspecific_linux(key, value) == 0;
+    }
+
+    public static nint TlsGetValue(uint key)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_getspecific_mac((nuint)key)
+            : pthread_getspecific_linux(key);
+    }
+
+    /// <summary>Stable numeric id of the calling thread (kernel32!GetCurrentThreadId).</summary>
+    public static uint GetCurrentThreadId()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            ulong tid;
+            return pthread_threadid_np(0, &tid) == 0 ? unchecked((uint)tid) : 0u;
+        }
+
+        return unchecked((uint)gettid());
+    }
+
+    /// <summary>
+    /// Wraps a managed callback (compiled for the SysV ABI on POSIX .NET) in a
+    /// thunk that accepts up to four integer arguments in the Win64 ABI the
+    /// emitted x86-64 call sites use. Win64 passes args in rcx/rdx/r8/r9 and
+    /// treats rdi/rsi as non-volatile; SysV expects rdi/rsi/rdx/rcx and
+    /// clobbers them, so the thunk saves rdi/rsi, shuffles the registers, keeps
+    /// the stack 16-byte aligned for the call, and forwards the rax result.
+    /// </summary>
+    public static nint CreateWin64ToSysVThunk(nint sysvTarget)
+    {
+        var page = (byte*)HostMemory.Alloc(
+            null,
+            4096,
+            HostMemory.MEM_COMMIT | HostMemory.MEM_RESERVE,
+            HostMemory.PAGE_EXECUTE_READWRITE);
+        if (page == null)
+        {
+            throw new OutOfMemoryException("Failed to allocate Win64->SysV thunk page");
+        }
+
+        var offset = 0;
+        Emit(page, ref offset, 0x57);                   // push rdi
+        Emit(page, ref offset, 0x56);                   // push rsi
+        Emit(page, ref offset, 0x48, 0x89, 0xCF);       // mov rdi, rcx
+        Emit(page, ref offset, 0x48, 0x89, 0xD6);       // mov rsi, rdx
+        Emit(page, ref offset, 0x4C, 0x89, 0xC2);       // mov rdx, r8
+        Emit(page, ref offset, 0x4C, 0x89, 0xC9);       // mov rcx, r9
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08); // sub rsp, 8 (realign to 16)
+        EmitMovRaxImm64(page, ref offset, sysvTarget);  // mov rax, target
+        Emit(page, ref offset, 0xFF, 0xD0);             // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08); // add rsp, 8
+        Emit(page, ref offset, 0x5E);                   // pop rsi
+        Emit(page, ref offset, 0x5F);                   // pop rdi
+        Emit(page, ref offset, 0xC3);                   // ret
+
+        if (!HostMemory.Protect(page, 4096, HostMemory.PAGE_EXECUTE_READ, out _))
+        {
+            throw new InvalidOperationException("Failed to protect Win64->SysV thunk page");
+        }
+
+        HostMemory.FlushInstructionCache(page, (nuint)offset);
+        return (nint)page;
+    }
+
+    private static void EnsureInitialized()
+    {
+        if (_initialized)
+        {
+            return;
+        }
+
+        lock (Gate)
+        {
+            if (_initialized)
+            {
+                return;
+            }
+
+            BuildStubs();
+            _initialized = true;
+        }
+    }
+
+    private static void BuildStubs()
+    {
+        var page = (byte*)HostMemory.Alloc(
+            null,
+            4096,
+            HostMemory.MEM_COMMIT | HostMemory.MEM_RESERVE,
+            HostMemory.PAGE_EXECUTE_READWRITE);
+        if (page == null)
+        {
+            throw new OutOfMemoryException("Failed to allocate POSIX host helper stub page");
+        }
+
+        var offset = 0;
+        _tlsGetValueStub = EmitTlsGetValue(page, ref offset);
+        _queryPerformanceCounterStub = EmitQueryPerformanceCounter(page, ref offset);
+        _switchToThreadStub = EmitSwitchToThread(page, ref offset);
+        _sleepStub = EmitSleep(page, ref offset);
+
+        if (!HostMemory.Protect(page, 4096, HostMemory.PAGE_EXECUTE_READ, out _))
+        {
+            throw new InvalidOperationException("Failed to protect POSIX host helper stub page");
+        }
+
+        HostMemory.FlushInstructionCache(page, (nuint)offset);
+    }
+
+    private static nint EmitTlsGetValue(byte* page, ref int offset)
+    {
+        var start = (nint)(page + offset);
+        if (OperatingSystem.IsMacOS())
+        {
+            // On macOS x86-64 pthread keys index the gs-based thread specific
+            // data array directly, so TlsGetValue(index in ecx) collapses to a
+            // single load that clobbers nothing but RAX.
+            Emit(page, ref offset, 0x89, 0xC8);                                     // mov eax, ecx
+            Emit(page, ref offset, 0x65, 0x48, 0x8B, 0x04, 0xC5, 0, 0, 0, 0);       // mov rax, gs:[rax*8]
+            Emit(page, ref offset, 0xC3);                                           // ret
+            return start;
+        }
+
+        // Linux: call pthread_getspecific, preserving the registers that are
+        // volatile in SysV but non-volatile in Win64 (rsi, rdi).
+        var pthreadGetSpecific = ResolveLibcExport("pthread_getspecific");
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        Emit(page, ref offset, 0x89, 0xCF);                                         // mov edi, ecx
+        EmitMovRaxImm64(page, ref offset, pthreadGetSpecific);                      // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitQueryPerformanceCounter(byte* page, ref int offset)
+    {
+        // BOOL QueryPerformanceCounter(LARGE_INTEGER* out in rcx): the emitted
+        // consumers only need a monotonically increasing counter, which rdtsc
+        // provides without leaving Win64-safe registers.
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x0F, 0x31);                                         // rdtsc
+        Emit(page, ref offset, 0x48, 0xC1, 0xE2, 0x20);                             // shl rdx, 32
+        Emit(page, ref offset, 0x48, 0x09, 0xD0);                                   // or rax, rdx
+        Emit(page, ref offset, 0x48, 0x89, 0x01);                                   // mov [rcx], rax
+        Emit(page, ref offset, 0xB8, 0x01, 0x00, 0x00, 0x00);                       // mov eax, 1
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitSwitchToThread(byte* page, ref int offset)
+    {
+        var schedYield = ResolveLibcExport("sched_yield");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        EmitMovRaxImm64(page, ref offset, schedYield);                              // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xB8, 0x01, 0x00, 0x00, 0x00);                       // mov eax, 1
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitSleep(byte* page, ref int offset)
+    {
+        // void Sleep(DWORD milliseconds in ecx) -> usleep(microseconds in edi).
+        var usleep = ResolveLibcExport("usleep");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        Emit(page, ref offset, 0x89, 0xCF);                                         // mov edi, ecx
+        Emit(page, ref offset, 0x81, 0xFF, 0xFF, 0x0F, 0x00, 0x00);                 // cmp edi, 0xFFF
+        Emit(page, ref offset, 0x76, 0x05);                                         // jbe +5
+        Emit(page, ref offset, 0xBF, 0xFF, 0x0F, 0x00, 0x00);                       // mov edi, 0xFFF (cap at ~4s)
+        Emit(page, ref offset, 0x69, 0xFF, 0xE8, 0x03, 0x00, 0x00);                 // imul edi, edi, 1000
+        EmitMovRaxImm64(page, ref offset, usleep);                                  // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint ResolveLibcExport(string name)
+    {
+        var libc = NativeLibrary.Load(OperatingSystem.IsMacOS() ? "libSystem.dylib" : "libc.so.6");
+        return NativeLibrary.GetExport(libc, name);
+    }
+
+    private static void Emit(byte* page, ref int offset, params byte[] bytes)
+    {
+        foreach (var value in bytes)
+        {
+            page[offset++] = value;
+        }
+    }
+
+    private static void EmitMovRaxImm64(byte* page, ref int offset, nint value)
+    {
+        Emit(page, ref offset, 0x48, 0xB8);
+        *(long*)(page + offset) = value;
+        offset += sizeof(long);
+    }
+
+    [DllImport("libc", EntryPoint = "pthread_key_create", SetLastError = true)]
+    private static extern int pthread_key_create_mac(nuint* key, nint destructor);
+
+    [DllImport("libc", EntryPoint = "pthread_key_create", SetLastError = true)]
+    private static extern int pthread_key_create_linux(uint* key, nint destructor);
+
+    [DllImport("libc", EntryPoint = "pthread_key_delete")]
+    private static extern int pthread_key_delete_mac(nuint key);
+
+    [DllImport("libc", EntryPoint = "pthread_key_delete")]
+    private static extern int pthread_key_delete_linux(uint key);
+
+    [DllImport("libc", EntryPoint = "pthread_setspecific")]
+    private static extern int pthread_setspecific_mac(nuint key, nint value);
+
+    [DllImport("libc", EntryPoint = "pthread_setspecific")]
+    private static extern int pthread_setspecific_linux(uint key, nint value);
+
+    [DllImport("libc", EntryPoint = "pthread_getspecific")]
+    private static extern nint pthread_getspecific_mac(nuint key);
+
+    [DllImport("libc", EntryPoint = "pthread_getspecific")]
+    private static extern nint pthread_getspecific_linux(uint key);
+
+    [DllImport("libc")]
+    private static extern int pthread_threadid_np(nint thread, ulong* threadId);
+
+    [DllImport("libc")]
+    private static extern int gettid();
+
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_create(long value);
+
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_signal(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_wait(nint semaphore, ulong timeout);
+
+    [DllImport("libc")]
+    private static extern ulong dispatch_time(ulong when, long deltaNanoseconds);
+
+    [DllImport("libc")]
+    private static extern void dispatch_release(nint handle);
+
+    [DllImport("libc")]
+    private static extern int sem_init(nint semaphore, int shared, uint value);
+
+    [DllImport("libc")]
+    private static extern int sem_post(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_wait(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_trywait(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_destroy(nint semaphore);
+}

src/SharpEmu.Core/Cpu/Native/StubManager.cs

@@ -3,7 +3,6 @@
 
 using System.Runtime.InteropServices;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
 
 namespace SharpEmu.Core.Cpu.Native;
 
@@ -12,15 +11,17 @@ public sealed unsafe class StubManager : IDisposable
     private readonly List<nint> _allocatedStubs = new();
     private readonly Dictionary<string, nint> _importHandlers = new();
     private readonly Dictionary<ulong, nint> _stubAddresses = new();
-    private readonly IHostMemory _hostMemory;
     private byte* _pltMemory;
     private int _pltOffset;
     private const int PltMemorySize = 1024 * 1024; // 1MB for stubs
 
-    public StubManager(IHostMemory? hostMemory = null)
+    public StubManager()
     {
-        _hostMemory = hostMemory ?? HostPlatform.Current.Memory;
-        _pltMemory = (byte*)_hostMemory.Allocate(0, PltMemorySize, HostPageProtection.ReadWriteExecute);
+        _pltMemory = (byte*)VirtualAlloc(
+            null,
+            (nuint)PltMemorySize,
+            AllocationType.Reserve | AllocationType.Commit,
+            MemoryProtection.ExecuteReadWrite);
 
         if (_pltMemory == null)
         {
@@ -184,7 +185,7 @@ public sealed unsafe class StubManager : IDisposable
     {
         if (_pltMemory != null)
         {
-            _hostMemory.Free((ulong)_pltMemory);
+            VirtualFree(_pltMemory, 0, FreeType.Release);
             _pltMemory = null;
         }
 
@@ -193,5 +194,29 @@ public sealed unsafe class StubManager : IDisposable
         _stubAddresses.Clear();
     }
 
+    private static void* VirtualAlloc(void* lpAddress, nuint dwSize, AllocationType flAllocationType, MemoryProtection flProtect) =>
+        HostMemory.Alloc(lpAddress, dwSize, (uint)flAllocationType, (uint)flProtect);
+
+    private static bool VirtualFree(void* lpAddress, nuint dwSize, FreeType dwFreeType) =>
+        HostMemory.Free(lpAddress, dwSize, (uint)dwFreeType);
+
+    [Flags]
+    private enum AllocationType : uint
+    {
+        Commit = 0x1000,
+        Reserve = 0x2000,
+    }
+
+    [Flags]
+    private enum MemoryProtection : uint
+    {
+        ExecuteReadWrite = 0x40,
+    }
+
+    private enum FreeType : uint
+    {
+        Release = 0x8000,
+    }
+
     public delegate void ImportHandler(CpuContext context);
 }

src/SharpEmu.Core/Cpu/Native/Windows/Win64ContextOffsets.cs

@@ -11,6 +11,7 @@ namespace SharpEmu.Core.Cpu.Native.Windows;
 /// </summary>
 internal static class Win64ContextOffsets
 {
+    public const int Size = 0x4D0;
     public const int Mxcsr = 52;
     public const int Rax = 120;
     public const int Rcx = 128;

src/SharpEmu.Core/Cpu/TrackedCpuMemory.cs

@@ -50,4 +50,9 @@ public sealed class TrackedCpuMemory : ICpuMemory, ITrackedCpuMemory, IGuestMemo
         address = 0;
         return false;
     }
+
+    public bool TryFreeGuestMemory(ulong address)
+    {
+        return _inner is IGuestMemoryAllocator allocator && allocator.TryFreeGuestMemory(address);
+    }
 }

src/SharpEmu.Core/Loader/ProgramHeader.cs

@@ -48,6 +48,8 @@ public enum ProgramHeaderType : uint
 
     Tls = 7,
 
+    GnuEhFrame = 0x6474E550,
+
     SceRela = 0x60000000,
 
     SceProcParam = 0x61000001,

src/SharpEmu.Core/Loader/SelfImage.cs

@@ -24,7 +24,10 @@ public sealed class SelfImage
         ulong procParamAddress = 0,
         string? title = null,
         string? titleId = null,
-        string? version = null)
+        string? version = null,
+        uint tlsModuleId = 0,
+        ulong tlsMemorySize = 0,
+        ulong tlsStaticOffset = 0)
     {
         ArgumentNullException.ThrowIfNull(programHeaders);
         ArgumentNullException.ThrowIfNull(mappedRegions);
@@ -44,6 +47,9 @@ public sealed class SelfImage
         Title = title;
         TitleId = titleId;
         Version = version;
+        TlsModuleId = tlsModuleId;
+        TlsMemorySize = tlsMemorySize;
+        TlsStaticOffset = tlsStaticOffset;
     }
 
     public bool IsSelf { get; }
@@ -75,4 +81,11 @@ public sealed class SelfImage
     public string? TitleId { get; }
 
     public string? Version { get; }
+
+    public uint TlsModuleId { get; }
+
+    public ulong TlsMemorySize { get; }
+
+    /// <summary>Variant II distance from the thread pointer to this module's static TLS base.</summary>
+    public ulong TlsStaticOffset { get; }
 }

src/SharpEmu.Core/Loader/SelfLoader.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using System.Buffers.Binary;
+using System.Diagnostics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Text;
@@ -10,16 +11,14 @@ using SharpEmu.Core;
 using SharpEmu.Core.Cpu;
 using SharpEmu.Core.Memory;
 using SharpEmu.HLE;
-using SharpEmu.Logging;
 
 namespace SharpEmu.Core.Loader;
 
 public sealed class SelfLoader : ISelfLoader
 {
-    private static readonly SharpEmuLogger Log = SharpEmuLog.For("Loader");
     private const uint ElfMagic = 0x7F454C46;
-    private const uint Ps4SelfMagic  = 0x4F153D1D;
-    private const uint Ps5SelfMagic  = 0x5414F5EE;
+    private const uint Ps4SelfMagic = 0x4F153D1D;
+    private const uint Ps5SelfMagic = 0x5414F5EE;
     private const ulong SelfSegmentFlag = 0x800;
     private const int PageSize = 0x1000;
     private const ulong ImportStubBaseAddress = 0x0000_7000_0000_0000UL;
@@ -58,11 +57,22 @@ public sealed class SelfLoader : ISelfLoader
     private const long DtSceSymTab = 0x61000039;
     private const long DtSceSymTabSize = 0x6100003F;
 
+    private const uint RelocationTypeNone = 0;
     private const uint RelocationTypeAbsolute64 = 1;
+    private const uint RelocationTypePc32 = 2;
+    private const uint RelocationTypePlt32 = 4;
     private const uint RelocationTypeGlobalData = 6;
     private const uint RelocationTypeJumpSlot = 7;
     private const uint RelocationTypeRelative = 8;
-    private const uint RelocationTypeTlsModuleId = 16;
+    private const uint RelocationTypeUnsigned32 = 10;
+    private const uint RelocationTypeSigned32 = 11;
+    private const uint RelocationTypeTlsModuleId = 16;   // R_X86_64_DTPMOD64
+    private const uint RelocationTypeTlsDtpOff64 = 17;    // R_X86_64_DTPOFF64
+    private const uint RelocationTypeTlsTpOff64 = 18;     // R_X86_64_TPOFF64
+    private const uint RelocationTypePc64 = 24;
+    private const uint RelocationTypeSize32 = 32;
+    private const uint RelocationTypeSize64 = 33;
+    private const uint RelocationTypeRelative64 = 38;
     private const ulong Ps5MainImageBase = 0x0000000800000000UL;
     private const ulong Ps4MainImageBase = 0x0000000000400000UL;
     private const ulong Ps5ModuleSearchStart = 0x0000000804000000UL;
@@ -88,6 +98,11 @@ public sealed class SelfLoader : ISelfLoader
     private static readonly int SelfSegmentSize = Unsafe.SizeOf<SelfSegment>();
     private static readonly int ProgramHeaderSize = Unsafe.SizeOf<ProgramHeader>();
 
+    static SelfLoader()
+    {
+        RunRelocationSelfChecks();
+    }
+
     public SelfImage Load(ReadOnlySpan<byte> imageData, IVirtualMemory virtualMemory)
     {
         return Load(imageData, virtualMemory, fs: null, mountRoot: null);
@@ -156,16 +171,22 @@ public sealed class SelfLoader : ISelfLoader
         {
             virtualMemory.Clear();
             _nextTlsModuleId = 1;
+            GuestTlsTemplate.Reset();
         }
 
-        var tlsModuleId = _nextTlsModuleId == 0 ? 1u : _nextTlsModuleId;
-        Log.Debug($"TLS load_start clear={clearVirtualMemory} next={_nextTlsModuleId} assigned={tlsModuleId}");
-
         var loadContext = ParseLayout(imageData);
         var elfHeader = ReadUnmanaged<ElfHeader>(imageData, loadContext.ElfOffset);
         ValidateElfHeader(elfHeader);
 
         var programHeaders = ParseProgramHeaders(imageData, loadContext, elfHeader);
+        var hasTlsSegment = TryGetProgramHeader(programHeaders, ProgramHeaderType.Tls, out var processTlsHeader, out _) &&
+            processTlsHeader.MemorySize != 0;
+        var tlsModuleId = hasTlsSegment
+            ? (_nextTlsModuleId == 0 ? 1u : _nextTlsModuleId)
+            : 0u;
+        Console.Error.WriteLine(
+            $"[LOADER][TLS] load_start clear={clearVirtualMemory} next={_nextTlsModuleId} " +
+            $"assigned={tlsModuleId} has_pt_tls={hasTlsSegment}");
 
         var totalImageSize = CalculateTotalImageSize(programHeaders);
         Console.WriteLine($"Total image size needed: 0x{totalImageSize:X} ({totalImageSize} bytes)");
@@ -178,8 +199,9 @@ public sealed class SelfLoader : ISelfLoader
             {
                 if (!physicalVm.TryAllocateAtExact(imageBase, totalImageSize, executable: true, out var allocatedBase))
                 {
+                    var reason = physicalVm.DescribeAddressForDiagnostics(imageBase);
                     throw new InvalidOperationException(
-                        $"Could not allocate main image at required base 0x{imageBase:X16} (size=0x{totalImageSize:X}).");
+                        $"Could not allocate main image at required base 0x{imageBase:X16} (size=0x{totalImageSize:X}): {reason}.");
                 }
 
                 imageBase = allocatedBase;
@@ -198,6 +220,13 @@ public sealed class SelfLoader : ISelfLoader
         }
 
         MapLoadSegments(imageData, loadContext, programHeaders, virtualMemory, imageBase);
+        // Register every module before relocations so DTPMOD/DTPOFF/TPOFF use
+        // the module's real PT_TLS identity and Variant II static offset.
+        var tlsInfo = RegisterModuleTlsTemplate(
+            programHeaders,
+            virtualMemory,
+            imageBase,
+            tlsModuleId);
         var importStubs = ResolveAndPatchImportStubs(
             imageData,
             loadContext,
@@ -255,11 +284,11 @@ public sealed class SelfLoader : ISelfLoader
             Console.WriteLine($"[LOADER] PH[{i}]: type={ph.HeaderType}, vaddr=0x{ph.VirtualAddress:X16} -> 0x{ph.VirtualAddress + imageBase:X16}, memsz=0x{ph.MemorySize:X}");
         }
 
-        if (_nextTlsModuleId == tlsModuleId && _nextTlsModuleId < uint.MaxValue)
+        if (tlsModuleId != 0 && _nextTlsModuleId == tlsModuleId && _nextTlsModuleId < uint.MaxValue)
         {
             _nextTlsModuleId++;
         }
-        Log.Debug($"TLS load_done assigned={tlsModuleId} next={_nextTlsModuleId}");
+        Console.Error.WriteLine($"[LOADER][TLS] load_done assigned={tlsModuleId} next={_nextTlsModuleId}");
 
         return new SelfImage(
             loadContext.IsSelf,
@@ -276,7 +305,10 @@ public sealed class SelfLoader : ISelfLoader
             procParamAddress,
             applicationInfo.Title,
             applicationInfo.TitleId,
-            applicationInfo.Version);
+            applicationInfo.Version,
+            tlsModuleId,
+            tlsInfo.MemorySize,
+            tlsInfo.StaticOffset);
     }
 
     private static (string? Title, string? TitleId, string? Version) TryLoadParamJson(
@@ -329,7 +361,7 @@ public sealed class SelfLoader : ISelfLoader
         if (magic is Ps4SelfMagic or Ps5SelfMagic)
         {
             var selfHeader = ReadUnmanaged<SelfHeader>(imageData, 0);
-            if (!selfHeader.HasKnownLayout || selfHeader.Unknown != 0x22)
+            if (!selfHeader.HasKnownLayout)
             {
                 throw new InvalidDataException("SELF header signature is not recognized.");
             }
@@ -348,10 +380,17 @@ public sealed class SelfLoader : ISelfLoader
             return new LoadContext(IsSelf: true, elfOffset, selfHeader.FileSize, segments);
         }
 
+        // Not a recognized (fake-signed) SELF. Only a bare, decrypted ELF is
+        // acceptable here; anything else — most commonly a still-encrypted
+        // retail eboot — must be reported clearly rather than failing later
+        // with an opaque "not a valid ELF header" message.
         if (magic != ElfMagic)
         {
             throw new InvalidDataException(
-                $"Unsupported executable signature 0x{magic:X8}");
+                $"Image is neither a decrypted ELF nor a recognized fake-signed SELF " +
+                $"(leading bytes 0x{magic:X8}). This is almost certainly a still-encrypted " +
+                $"retail eboot — SharpEmu has no decryption keys and requires a decrypted / " +
+                $"fake-signed (fSELF) image.");
         }
 
         return new LoadContext(IsSelf: false, ElfOffset: 0, SelfFileSize: 0, Array.Empty<SelfSegment>());
@@ -414,15 +453,15 @@ public sealed class SelfLoader : ISelfLoader
 
             var virtualAddress = header.VirtualAddress + imageBase;
 
-            Log.Debug($"Segment {index}: VAddr=0x{virtualAddress:X16}, FileSize=0x{header.FileSize:X}, MemSize=0x{header.MemorySize:X}, Align=0x{header.Alignment:X}");
+            Console.Error.WriteLine($"[LOADER] Segment {index}: VAddr=0x{virtualAddress:X16}, FileSize=0x{header.FileSize:X}, MemSize=0x{header.MemorySize:X}, Align=0x{header.Alignment:X}");
             if (header.Alignment > 1)
             {
                 var vaddrMod = virtualAddress % header.Alignment;
                 var offsetMod = header.Offset % header.Alignment;
                 if (vaddrMod != offsetMod)
                 {
-                    Log.Warning(
-                        $"Segment {index} ELF alignment mismatch! " +
+                    Console.Error.WriteLine(
+                        $"[LOADER] WARNING: Segment {index} ELF alignment mismatch! " +
                         $"VAddr=0x{virtualAddress:X}, Offset=0x{header.Offset:X}, Align=0x{header.Alignment:X}, " +
                         $"VAddr%Align=0x{vaddrMod:X}, Offset%Align=0x{offsetMod:X}");
                 }
@@ -465,6 +504,43 @@ public sealed class SelfLoader : ISelfLoader
         return 0;
     }
 
+    private static ModuleTlsInfo RegisterModuleTlsTemplate(
+        IReadOnlyList<ProgramHeader> programHeaders,
+        IVirtualMemory virtualMemory,
+        ulong imageBase,
+        uint tlsModuleId)
+    {
+        if (!TryGetProgramHeader(programHeaders, ProgramHeaderType.Tls, out var tlsHeader, out _) ||
+            tlsHeader.MemorySize == 0)
+        {
+            return default;
+        }
+
+        // tdata (initialized) bytes come from the mapped segment; tbss is the
+        // implicitly-zero remainder up to MemorySize.
+        var fileSize = (int)Math.Min(tlsHeader.FileSize, tlsHeader.MemorySize);
+        var initImage = fileSize > 0 ? new byte[fileSize] : [];
+        if (fileSize > 0 &&
+            !virtualMemory.TryRead(imageBase + tlsHeader.VirtualAddress, initImage))
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][TLS] Failed to read TLS init image at 0x{imageBase + tlsHeader.VirtualAddress:X}; seeding zeros.");
+            initImage = [];
+        }
+
+        var staticOffset = GuestTlsTemplate.RegisterModule(
+            tlsModuleId,
+            initImage,
+            tlsHeader.MemorySize,
+            tlsHeader.Alignment,
+            tlsHeader.VirtualAddress);
+        Console.Error.WriteLine(
+            $"[LOADER][TLS] Module {tlsModuleId} TLS template: memsz=0x{tlsHeader.MemorySize:X} " +
+            $"filesz=0x{tlsHeader.FileSize:X} align=0x{tlsHeader.Alignment:X} " +
+            $"static_offset=0x{staticOffset:X} total_static=0x{GuestTlsTemplate.StaticTlsSize:X}");
+        return new ModuleTlsInfo(tlsHeader.MemorySize, staticOffset);
+    }
+
     private static IReadOnlyDictionary<ulong, string> ResolveAndPatchImportStubs(
         ReadOnlySpan<byte> imageData,
         LoadContext loadContext,
@@ -544,12 +620,17 @@ public sealed class SelfLoader : ISelfLoader
         uint maxSymbolIndex = 0;
         foreach (var relocation in relocations)
         {
+            if (relocation.Type == RelocationTypeNone)
+            {
+                continue;
+            }
+
             if (!IsSupportedRelocationType(relocation.Type))
             {
                 continue;
             }
 
-            if (relocation.Type is RelocationTypeRelative or RelocationTypeTlsModuleId)
+            if (relocation.Type is RelocationTypeNone or RelocationTypeRelative or RelocationTypeRelative64 or RelocationTypeTlsModuleId)
             {
                 continue;
             }
@@ -633,13 +714,16 @@ public sealed class SelfLoader : ISelfLoader
 
         importedRelocations = BuildImportedRelocations(descriptors);
 
-        var stubsByAddress = CreateImportStubMapping(virtualMemory, orderedImportNids);
+        var stubImportNids = orderedImportNids
+            .Where(nid => ShouldCreateImportStub(nid, descriptors, moduleManager))
+            .ToArray();
+        var stubsByAddress = CreateImportStubMapping(virtualMemory, stubImportNids);
         Console.WriteLine($"[LOADER] Created {stubsByAddress.Count} import stubs");
 
-        int printCount = Math.Min(10, orderedImportNids.Count);
+        int printCount = Math.Min(10, stubImportNids.Length);
         for (int i = 0; i < printCount; i++)
         {
-            var nid = orderedImportNids[i];
+            var nid = stubImportNids[i];
             var addr = stubsByAddress.First(x => x.Value == nid).Key;
         }
 
@@ -663,45 +747,65 @@ public sealed class SelfLoader : ISelfLoader
 
         foreach (var descriptor in descriptors)
         {
-            ulong targetValue;
+            ulong symbolValue;
             if (descriptor.ImportNid is null)
             {
-                targetValue = AddSigned(descriptor.SymbolValue, descriptor.Addend);
+                symbolValue = descriptor.SymbolValue;
             }
             else
             {
-                if (!addressesByNid.TryGetValue(descriptor.ImportNid, out var stubAddress))
+                if (addressesByNid.TryGetValue(descriptor.ImportNid, out var stubAddress))
                 {
-                    throw new InvalidOperationException($"Import stub not found for NID '{descriptor.ImportNid}'.");
+                    symbolValue = stubAddress;
                 }
-
-                targetValue = AddSigned(stubAddress, descriptor.Addend);
-            }
-
-            if (targetValue < 0x1000)
-            {
-                if (descriptor.ValueKind == RelocationValueKind.TlsModuleId)
+                else if (descriptor.IsWeak)
                 {
-                    Log.Debug(
-                        $"Patching DTPMOD64 at 0x{descriptor.TargetAddress:X} with module id 0x{targetValue:X}");
+                    // ELF unresolved weak definitions use S=0. They must not
+                    // receive a trap import stub, which would turn a permitted
+                    // null test into a call to an unresolved-import handler.
+                    symbolValue = 0;
                 }
                 else
                 {
-                    Log.Warning($"!!! CRITICAL !!! Patching address 0x{descriptor.TargetAddress:X} with INVALID value 0x{targetValue:X} for NID {descriptor.ImportNid ?? "(null)"}");
-                    Log.Warning($"  SymbolValue=0x{descriptor.SymbolValue:X}, Addend=0x{descriptor.Addend:X}, StubAddress=0x{(addressesByNid.TryGetValue(descriptor.ImportNid ?? "", out var sa) ? sa : 0):X}");
+                    throw new InvalidOperationException($"Import stub not found for NID '{descriptor.ImportNid}'.");
                 }
             }
 
-            if (!TryWriteUInt64(virtualMemory, descriptor.TargetAddress, targetValue))
+            var targetValue = ComputeRelocationValue(descriptor, symbolValue);
+
+            if (targetValue < 0x1000 && descriptor.ValueKind is
+                RelocationValueKind.TlsOffset or
+                RelocationValueKind.PcRelative or
+                RelocationValueKind.SymbolSize)
             {
-                throw new InvalidDataException($"Failed to patch relocation at 0x{descriptor.TargetAddress:X16}.");
+                // A TLS offset (TPOFF64/DTPOFF64) is a signed displacement, not a
+                // mapped address, so a small or negative value here is expected.
+            }
+            else if (targetValue < 0x1000 && !descriptor.IsWeak)
+            {
+                if (descriptor.ValueKind == RelocationValueKind.TlsModuleId)
+                {
+                    Console.Error.WriteLine(
+                        $"[LOADER][TLS] Patching DTPMOD64 at 0x{descriptor.TargetAddress:X} with module id 0x{targetValue:X}");
+                }
+                else
+                {
+                    Console.Error.WriteLine($"[LOADER] !!! CRITICAL !!! Patching address 0x{descriptor.TargetAddress:X} with INVALID value 0x{targetValue:X} for NID {descriptor.ImportNid ?? "(null)"}");
+                    Console.Error.WriteLine($"[LOADER]   SymbolValue=0x{descriptor.SymbolValue:X}, Addend=0x{descriptor.Addend:X}, StubAddress=0x{(addressesByNid.TryGetValue(descriptor.ImportNid ?? "", out var sa) ? sa : 0):X}");
+                }
+            }
+
+            if (!TryWriteRelocationValue(virtualMemory, descriptor, targetValue, out var writeError))
+            {
+                throw new InvalidDataException(
+                    $"Failed to patch relocation at 0x{descriptor.TargetAddress:X16}: {writeError}");
             }
 
             if (descriptor.TargetAddress >= 0x00000008030FC300UL &&
                 descriptor.TargetAddress <= 0x00000008030FC3F0UL)
             {
-                Log.Debug(
-                    $"[RELOC] target=0x{descriptor.TargetAddress:X16} value=0x{targetValue:X16} addend=0x{descriptor.Addend:X} nid={(descriptor.ImportNid ?? "<sym>")}");
+                Console.Error.WriteLine(
+                    $"[LOADER][RELOC] target=0x{descriptor.TargetAddress:X16} value=0x{targetValue:X16} addend=0x{descriptor.Addend:X} nid={(descriptor.ImportNid ?? "<sym>")}");
             }
         }
 
@@ -794,29 +898,36 @@ public sealed class SelfLoader : ISelfLoader
         {
             if (IsFocusRelocationOffset(relocation.Offset, imageBase))
             {
-                Log.Debug(
-                    $"[FOCUS][SCAN] off=0x{relocation.Offset:X16} type={relocation.Type} sym={relocation.SymbolIndex} addend=0x{relocation.Addend:X}");
+                Console.Error.WriteLine(
+                    $"[LOADER][FOCUS][SCAN] off=0x{relocation.Offset:X16} type={relocation.Type} sym={relocation.SymbolIndex} addend=0x{relocation.Addend:X}");
             }
 
             if (!IsSupportedRelocationType(relocation.Type))
             {
-                if (IsFocusRelocationOffset(relocation.Offset, imageBase))
+                // Surface unsupported relocation types loudly instead of
+                // skipping silently: TLS-relative (17/18), COPY (5), and
+                // IRELATIVE/ifunc (37) leave their targets unrelocated, which
+                // manifests later as reads of zero or calls into address 0.
+                ReportUnsupportedRelocation(relocation.Type, relocation.Offset, imageBase);
+                if (relocation.Type is 5 or 37)
                 {
-                    Log.Debug($"[FOCUS][SKIP] unsupported type={relocation.Type}");
+                    throw new NotSupportedException(
+                        $"Relocation type {relocation.Type} requires deferred runtime-linker processing.");
                 }
                 continue;
             }
 
-            if (!TryResolveMappedAddress(virtualMemory, relocation.Offset, imageBase, sizeof(ulong), out var targetAddress))
+            var relocationWriteSize = GetRelocationWriteSize(relocation.Type);
+            if (!TryResolveMappedAddress(virtualMemory, relocation.Offset, imageBase, relocationWriteSize, out var targetAddress))
             {
                 if (IsFocusRelocationOffset(relocation.Offset, imageBase))
                 {
-                    Log.Debug("[FOCUS][SKIP] target address not mapped");
+                    Console.Error.WriteLine("[LOADER][FOCUS][SKIP] target address not mapped");
                 }
                 continue;
             }
 
-            if (relocation.Type == RelocationTypeRelative)
+            if (relocation.Type is RelocationTypeRelative or RelocationTypeRelative64)
             {
                 descriptors.Add(new RelocationDescriptor(
                     targetAddress,
@@ -830,7 +941,12 @@ public sealed class SelfLoader : ISelfLoader
 
             if (relocation.Type == RelocationTypeTlsModuleId)
             {
-                var dtpmodValue = tlsModuleId == 0 ? 1u : tlsModuleId;
+                if (tlsModuleId == 0)
+                {
+                    throw new InvalidDataException(
+                        $"R_X86_64_DTPMOD64 at 0x{targetAddress:X16} references a module without PT_TLS.");
+                }
+                var dtpmodValue = tlsModuleId;
                 descriptors.Add(new RelocationDescriptor(
                     targetAddress,
                     0,
@@ -851,50 +967,100 @@ public sealed class SelfLoader : ISelfLoader
             {
                 if (targetAddress >= FocusRelocGuestStart && targetAddress <= FocusRelocGuestEnd)
                 {
-                    Log.Debug($"[FOCUS][SKIP] symbol read failed index={symbolIndex}");
+                    Console.Error.WriteLine($"[LOADER][FOCUS][SKIP] symbol read failed index={symbolIndex}");
                 }
                 continue;
             }
 
-            var addend = relocation.Type is RelocationTypeGlobalData or RelocationTypeJumpSlot ? 0 : relocation.Addend;
-            var symbolBind = GetSymbolBind(symbol.Info);
-            if (symbolBind == SymbolBindLocal)
+            if (relocation.Type is RelocationTypeTlsDtpOff64 or RelocationTypeTlsTpOff64)
             {
-                var symbolAddress = ResolveMappedAddressOrFallback(virtualMemory, symbol.Value, imageBase);
-                if (symbolAddress == 0)
+                // Variant II static TLS: the module block sits at
+                // [tp - blockSize, tp). DTPOFF64 is the module-relative offset
+                // (st_value + addend); TPOFF64 is that offset expressed relative
+                // to the thread pointer, i.e. minus the aligned block size.
+                var tlsSymbolOffset = AddSigned(symbol.Value, relocation.Addend);
+                if (!GuestTlsTemplate.TryGetStaticOffset(tlsModuleId, out var moduleStaticOffset))
                 {
-                    Log.Warning(
-                        $"Skipping local relocation with invalid symbol value 0x{symbol.Value:X} " +
-                        $"at target 0x{targetAddress:X16}, type={relocation.Type}, sym={symbolIndex}");
+                    Console.Error.WriteLine(
+                        $"[LOADER][TLS] Missing PT_TLS registration for module {tlsModuleId}; " +
+                        $"cannot apply relocation {relocation.Type} at 0x{targetAddress:X16}.");
+                    continue;
                 }
-
+                var tlsValue = relocation.Type == RelocationTypeTlsTpOff64
+                    ? unchecked(tlsSymbolOffset - moduleStaticOffset)
+                    : tlsSymbolOffset;
                 descriptors.Add(new RelocationDescriptor(
                     targetAddress,
-                    addend,
+                    0,
                     null,
-                    symbolAddress,
-                    RelocationValueKind.Pointer,
+                    tlsValue,
+                    RelocationValueKind.TlsOffset,
                     IsDataImport: false));
                 continue;
             }
 
+            var symbolBind = GetSymbolBind(symbol.Info);
+            if (symbolIndex == 0)
+            {
+                descriptors.Add(CreateSymbolRelocationDescriptor(
+                    relocation,
+                    targetAddress,
+                    symbol,
+                    symbolAddress: 0,
+                    importNid: null,
+                    isWeak: false));
+                continue;
+            }
+
+            if (symbolBind == SymbolBindLocal)
+            {
+                var symbolAddress = relocation.Type is RelocationTypeSize32 or RelocationTypeSize64
+                    ? 0
+                    : ResolveMappedAddressOrFallback(virtualMemory, symbol.Value, imageBase);
+                if (symbolAddress == 0)
+                {
+                    if (relocation.Type is not (RelocationTypeSize32 or RelocationTypeSize64))
+                    {
+                        Console.Error.WriteLine(
+                            $"[LOADER] Skipping local relocation with invalid symbol value 0x{symbol.Value:X} " +
+                            $"at target 0x{targetAddress:X16}, type={relocation.Type}, sym={symbolIndex}");
+                        continue;
+                    }
+                }
+
+                descriptors.Add(CreateSymbolRelocationDescriptor(
+                    relocation,
+                    targetAddress,
+                    symbol,
+                    symbolAddress,
+                    importNid: null,
+                    isWeak: false));
+                continue;
+            }
+
             if (symbol.Value != 0)
             {
-                var symbolAddress = ResolveMappedAddressOrFallback(virtualMemory, symbol.Value, imageBase);
+                var symbolAddress = relocation.Type is RelocationTypeSize32 or RelocationTypeSize64
+                    ? 0
+                    : ResolveMappedAddressOrFallback(virtualMemory, symbol.Value, imageBase);
                 if (symbolAddress == 0)
                 {
-                    Log.Warning(
-                        $"Skipping relocation with invalid symbol value 0x{symbol.Value:X} " +
-                        $"at target 0x{targetAddress:X16}, type={relocation.Type}, sym={symbolIndex}");
+                    if (relocation.Type is not (RelocationTypeSize32 or RelocationTypeSize64))
+                    {
+                        Console.Error.WriteLine(
+                            $"[LOADER] Skipping relocation with invalid symbol value 0x{symbol.Value:X} " +
+                            $"at target 0x{targetAddress:X16}, type={relocation.Type}, sym={symbolIndex}");
+                        continue;
+                    }
                 }
 
-                descriptors.Add(new RelocationDescriptor(
+                descriptors.Add(CreateSymbolRelocationDescriptor(
+                    relocation,
                     targetAddress,
-                    addend,
-                    null,
+                    symbol,
                     symbolAddress,
-                    RelocationValueKind.Pointer,
-                    IsDataImport: false));
+                    importNid: null,
+                    isWeak: false));
                 continue;
             }
 
@@ -902,16 +1068,26 @@ public sealed class SelfLoader : ISelfLoader
             {
                 if (targetAddress >= FocusRelocGuestStart && targetAddress <= FocusRelocGuestEnd)
                 {
-                    Log.Debug($"[FOCUS][SKIP] bind={symbolBind} not importable");
+                    Console.Error.WriteLine($"[LOADER][FOCUS][SKIP] bind={symbolBind} not importable");
                 }
                 continue;
             }
 
             if (!TryReadNullTerminatedAscii(stringTable, symbol.NameOffset, out var symbolName))
             {
+                if (symbolBind == SymbolBindWeak)
+                {
+                    descriptors.Add(CreateSymbolRelocationDescriptor(
+                        relocation,
+                        targetAddress,
+                        symbol,
+                        symbolAddress: 0,
+                        importNid: null,
+                        isWeak: true));
+                }
                 if (targetAddress >= FocusRelocGuestStart && targetAddress <= FocusRelocGuestEnd)
                 {
-                    Log.Debug($"[FOCUS][SKIP] symbol name read failed offset={symbol.NameOffset}");
+                    Console.Error.WriteLine($"[LOADER][FOCUS][SKIP] symbol name read failed offset={symbol.NameOffset}");
                 }
                 continue;
             }
@@ -919,6 +1095,16 @@ public sealed class SelfLoader : ISelfLoader
             var nid = ExtractNid(symbolName);
             if (string.IsNullOrWhiteSpace(nid))
             {
+                if (symbolBind == SymbolBindWeak)
+                {
+                    descriptors.Add(CreateSymbolRelocationDescriptor(
+                        relocation,
+                        targetAddress,
+                        symbol,
+                        symbolAddress: 0,
+                        importNid: null,
+                        isWeak: true));
+                }
                 continue;
             }
 
@@ -927,16 +1113,74 @@ public sealed class SelfLoader : ISelfLoader
                 orderedImportNids.Add(nid);
             }
 
-            descriptors.Add(new RelocationDescriptor(
+            descriptors.Add(CreateSymbolRelocationDescriptor(
+                relocation,
                 targetAddress,
-                addend,
-                nid,
-                0,
-                RelocationValueKind.Pointer,
-                IsDataImport: GetSymbolType(symbol.Info) == SymbolTypeObject));
+                symbol,
+                symbolAddress: 0,
+                importNid: nid,
+                isWeak: symbolBind == SymbolBindWeak));
         }
     }
 
+    private static RelocationDescriptor CreateSymbolRelocationDescriptor(
+        ElfRelocation relocation,
+        ulong targetAddress,
+        ElfSymbol symbol,
+        ulong symbolAddress,
+        string? importNid,
+        bool isWeak)
+    {
+        var valueKind = relocation.Type switch
+        {
+            RelocationTypePc32 or RelocationTypePlt32 or RelocationTypePc64 => RelocationValueKind.PcRelative,
+            RelocationTypeSize32 or RelocationTypeSize64 => RelocationValueKind.SymbolSize,
+            _ => RelocationValueKind.Pointer,
+        };
+        var writeKind = relocation.Type switch
+        {
+            RelocationTypePc32 or RelocationTypePlt32 or RelocationTypeSigned32 => RelocationWriteKind.Int32,
+            RelocationTypeUnsigned32 or RelocationTypeSize32 => RelocationWriteKind.UInt32,
+            _ => RelocationWriteKind.UInt64,
+        };
+        var symbolValue = valueKind == RelocationValueKind.SymbolSize
+            ? symbol.Size
+            : symbolAddress;
+        var addend = relocation.Type is RelocationTypeGlobalData or RelocationTypeJumpSlot
+            ? 0
+            : relocation.Addend;
+        return new RelocationDescriptor(
+            targetAddress,
+            addend,
+            importNid,
+            symbolValue,
+            valueKind,
+            IsDataImport: GetSymbolType(symbol.Info) == SymbolTypeObject,
+            writeKind,
+            isWeak);
+    }
+
+    private static bool ShouldCreateImportStub(
+        string nid,
+        IReadOnlyList<RelocationDescriptor> descriptors,
+        IModuleManager? moduleManager)
+    {
+        for (var i = 0; i < descriptors.Count; i++)
+        {
+            var descriptor = descriptors[i];
+            if (!string.Equals(descriptor.ImportNid, nid, StringComparison.Ordinal))
+            {
+                continue;
+            }
+            if (!descriptor.IsWeak || moduleManager?.TryGetExport(nid, out _) == true)
+            {
+                return true;
+            }
+        }
+
+        return false;
+    }
+
     private static void RegisterRuntimeSymbolsAndHooks(
         ReadOnlySpan<byte> imageData,
         LoadContext loadContext,
@@ -959,8 +1203,8 @@ public sealed class SelfLoader : ISelfLoader
 
         if (sectionSymbols > 0 || dynamicSymbols > 0)
         {
-            Log.Info(
-                $"Runtime symbol index populated: section={sectionSymbols}, dynamic={dynamicSymbols}, total={runtimeSymbols.Count}");
+            Console.Error.WriteLine(
+                $"[LOADER] Runtime symbol index populated: section={sectionSymbols}, dynamic={dynamicSymbols}, total={runtimeSymbols.Count}");
         }
     }
 
@@ -1034,8 +1278,8 @@ public sealed class SelfLoader : ISelfLoader
 
         if (preInitializers.Count != 0 || initializers.Count != 0)
         {
-            Log.Info(
-                $"Initializers discovered: preinit={preInitializers.Count}, init={initializers.Count}");
+            Console.Error.WriteLine(
+                $"[LOADER] Initializers discovered: preinit={preInitializers.Count}, init={initializers.Count}");
         }
     }
 
@@ -1418,8 +1662,8 @@ public sealed class SelfLoader : ISelfLoader
 
         var requiredBytes = checked((ulong)orderedImportNids.Count * ImportStubSlotSize);
         var mapSize = AlignUp(Math.Max(requiredBytes, (ulong)PageSize), PageSize);
-        Log.Debug(
-            $"CreateImportStubMapping: nids={orderedImportNids.Count}, required=0x{requiredBytes:X}, map_size=0x{mapSize:X}");
+        Console.Error.WriteLine(
+            $"[LOADER] CreateImportStubMapping: nids={orderedImportNids.Count}, required=0x{requiredBytes:X}, map_size=0x{mapSize:X}");
         if (mapSize > int.MaxValue)
         {
             throw new NotSupportedException("Import stub mapping exceeds 2 GB and is not supported.");
@@ -1691,11 +1935,51 @@ public sealed class SelfLoader : ISelfLoader
     private static bool IsSupportedRelocationType(uint relocationType)
     {
         return relocationType is
+            RelocationTypeNone or
             RelocationTypeAbsolute64 or
+            RelocationTypePc32 or
+            RelocationTypePlt32 or
             RelocationTypeGlobalData or
             RelocationTypeJumpSlot or
             RelocationTypeRelative or
-            RelocationTypeTlsModuleId;
+            RelocationTypeUnsigned32 or
+            RelocationTypeSigned32 or
+            RelocationTypeTlsModuleId or
+            RelocationTypeTlsDtpOff64 or
+            RelocationTypeTlsTpOff64 or
+            RelocationTypePc64 or
+            RelocationTypeSize32 or
+            RelocationTypeSize64 or
+            RelocationTypeRelative64;
+    }
+
+    private static readonly HashSet<uint> _reportedUnsupportedRelocationTypes = new();
+
+    private static void ReportUnsupportedRelocation(uint relocationType, ulong offset, ulong imageBase)
+    {
+        // Report each distinct unsupported type once to keep the log useful.
+        lock (_reportedUnsupportedRelocationTypes)
+        {
+            if (!_reportedUnsupportedRelocationTypes.Add(relocationType))
+            {
+                if (IsFocusRelocationOffset(offset, imageBase))
+                {
+                    Console.Error.WriteLine($"[LOADER][FOCUS][SKIP] unsupported type={relocationType}");
+                }
+
+                return;
+            }
+        }
+
+        var name = relocationType switch
+        {
+            5 => "R_X86_64_COPY",
+            37 => "R_X86_64_IRELATIVE (ifunc)",
+            _ => "unknown",
+        };
+        Console.Error.WriteLine(
+            $"[LOADER][ERROR] Unsupported relocation type {relocationType} ({name}) rejected " +
+            $"(first at off=0x{offset:X16}); COPY requires dependency symbol storage and IRELATIVE requires resolver execution.");
     }
 
     private static ulong DetermineRequestedImageBase(
@@ -1851,16 +2135,16 @@ public sealed class SelfLoader : ISelfLoader
         tableBytes = GC.AllocateUninitializedArray<byte>((int)size);
 
         var guestAddr = location + imageBase;
-        Log.Debug($"TryLoadTableBytes: trying guest address 0x{guestAddr:X}");
+        Console.Error.WriteLine($"[LOADER] TryLoadTableBytes: trying guest address 0x{guestAddr:X}");
         if (virtualMemory.TryRead(guestAddr, tableBytes))
         {
-            Log.Debug($"TryLoadTableBytes: loaded from guest memory at 0x{guestAddr:X}");
+            Console.Error.WriteLine($"[LOADER] TryLoadTableBytes: loaded from guest memory at 0x{guestAddr:X}");
             return true;
         }
 
         if (virtualMemory.TryRead(location, tableBytes))
         {
-            Log.Debug($"TryLoadTableBytes: loaded from absolute guest address 0x{location:X}");
+            Console.Error.WriteLine($"[LOADER] TryLoadTableBytes: loaded from absolute guest address 0x{location:X}");
             return true;
         }
 
@@ -1868,11 +2152,11 @@ public sealed class SelfLoader : ISelfLoader
         {
             var slice = elfData.Slice((int)location, (int)size);
             tableBytes = slice.ToArray();
-            Log.Debug($"TryLoadTableBytes: loaded from elfData as file offset at 0x{location:X}");
+            Console.Error.WriteLine($"[LOADER] TryLoadTableBytes: loaded from elfData as file offset at 0x{location:X}");
             return true;
         }
 
-        Log.Warning($"TryLoadTableBytes: FAILED for location 0x{location:X}");
+        Console.Error.WriteLine($"[LOADER] TryLoadTableBytes: FAILED for location 0x{location:X}");
         tableBytes = Array.Empty<byte>();
         return false;
     }
@@ -1929,17 +2213,17 @@ public sealed class SelfLoader : ISelfLoader
 
     private static ulong ResolveMappedAddressOrFallback(IVirtualMemory virtualMemory, ulong address, ulong imageBase)
     {
-        Log.Debug($"ResolveMappedAddressOrFallback addr=0x{address:X} imageBase=0x{imageBase:X16}");
+        Console.Error.WriteLine($"[LOADER][TEST] ResolveMappedAddressOrFallback addr=0x{address:X} imageBase=0x{imageBase:X16}");
 
         if (address == 0)
         {
-            Log.Debug("-> return 0 (null)");
+            Console.Error.WriteLine("[LOADER][TEST] -> return 0 (null)");
             return 0;
         }
 
         if (TryResolveMappedAddress(virtualMemory, address, imageBase, 1, out var resolved))
         {
-            Log.Debug($"-> resolved raw 0x{resolved:X16}");
+            Console.Error.WriteLine($"[LOADER][TEST] -> resolved raw 0x{resolved:X16}");
             return resolved;
         }
 
@@ -1948,18 +2232,18 @@ public sealed class SelfLoader : ISelfLoader
             var rebased = address + imageBase;
             if (TryResolveMappedAddress(virtualMemory, rebased, imageBase, 1, out var resolvedRebased))
             {
-                Log.Debug($"-> resolved rebased 0x{resolvedRebased:X16}");
+                Console.Error.WriteLine($"[LOADER][TEST] -> resolved rebased 0x{resolvedRebased:X16}");
                 return resolvedRebased;
             }
         }
 
         if (address < 0x10000)
         {
-            Log.Debug($"-> reject small 0x{address:X}");
+            Console.Error.WriteLine($"[LOADER][TEST] -> reject small 0x{address:X}");
             return 0;
         }
 
-        Log.Debug($"-> fallback raw 0x{address:X}");
+        Console.Error.WriteLine($"[LOADER][TEST] -> fallback raw 0x{address:X}");
         return address;
     }
 
@@ -2049,6 +2333,106 @@ public sealed class SelfLoader : ISelfLoader
         return virtualMemory.TryWrite(address, buffer);
     }
 
+    private static ulong ComputeRelocationValue(RelocationDescriptor descriptor, ulong resolvedSymbolValue)
+    {
+        var baseValue = descriptor.ValueKind == RelocationValueKind.SymbolSize
+            ? descriptor.SymbolValue
+            : resolvedSymbolValue;
+        var value = AddSigned(baseValue, descriptor.Addend);
+        return descriptor.ValueKind == RelocationValueKind.PcRelative
+            ? unchecked(value - descriptor.TargetAddress)
+            : value;
+    }
+
+    private static bool TryWriteRelocationValue(
+        IVirtualMemory virtualMemory,
+        RelocationDescriptor descriptor,
+        ulong value,
+        out string? error)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(ulong)];
+        var length = sizeof(ulong);
+        switch (descriptor.WriteKind)
+        {
+            case RelocationWriteKind.UInt64:
+                BinaryPrimitives.WriteUInt64LittleEndian(buffer, value);
+                break;
+
+            case RelocationWriteKind.UInt32:
+                if (value > uint.MaxValue)
+                {
+                    error = $"value 0x{value:X16} overflows an unsigned 32-bit relocation";
+                    return false;
+                }
+                BinaryPrimitives.WriteUInt32LittleEndian(buffer, (uint)value);
+                length = sizeof(uint);
+                break;
+
+            case RelocationWriteKind.Int32:
+                var signedValue = unchecked((long)value);
+                if (signedValue is < int.MinValue or > int.MaxValue)
+                {
+                    error = $"value {signedValue} overflows a signed 32-bit relocation";
+                    return false;
+                }
+                BinaryPrimitives.WriteInt32LittleEndian(buffer, (int)signedValue);
+                length = sizeof(int);
+                break;
+
+            default:
+                error = $"unknown relocation write kind {descriptor.WriteKind}";
+                return false;
+        }
+
+        error = virtualMemory.TryWrite(descriptor.TargetAddress, buffer[..length])
+            ? null
+            : "target memory is not writable";
+        return error is null;
+    }
+
+    private static int GetRelocationWriteSize(uint relocationType)
+    {
+        return relocationType is
+            RelocationTypePc32 or
+            RelocationTypePlt32 or
+            RelocationTypeUnsigned32 or
+            RelocationTypeSigned32 or
+            RelocationTypeSize32
+                ? sizeof(uint)
+                : sizeof(ulong);
+    }
+
+    [Conditional("DEBUG")]
+    private static void RunRelocationSelfChecks()
+    {
+        var pc32 = new RelocationDescriptor(
+            TargetAddress: 0x1000,
+            Addend: -4,
+            ImportNid: null,
+            SymbolValue: 0x1800,
+            RelocationValueKind.PcRelative,
+            IsDataImport: false,
+            RelocationWriteKind.Int32);
+        Debug.Assert(
+            unchecked((long)ComputeRelocationValue(pc32, pc32.SymbolValue)) == 0x7FC,
+            "R_X86_64_PC32 did not apply S + A - P.");
+
+        var weak = new RelocationDescriptor(
+            TargetAddress: 0x2000,
+            Addend: 7,
+            ImportNid: "weak",
+            SymbolValue: 0,
+            RelocationValueKind.Pointer,
+            IsDataImport: false,
+            IsWeak: true);
+        Debug.Assert(
+            ComputeRelocationValue(weak, resolvedSymbolValue: 0) == 7,
+            "An unresolved weak relocation did not use S=0.");
+        Debug.Assert(
+            !ShouldCreateImportStub("weak", [weak], moduleManager: null),
+            "An unresolved weak symbol incorrectly received a trap import stub.");
+    }
+
     private static ulong AlignUp(ulong value, ulong alignment)
     {
         var mask = alignment - 1;
@@ -2257,6 +2641,16 @@ public sealed class SelfLoader : ISelfLoader
         {
             throw new InvalidDataException($"Unsupported ELF program header entry size: {header.ProgramHeaderEntrySize}.");
         }
+
+        // The CPU backend executes guest instructions natively, so a non
+        // x86-64 image can only fail deep in execution with an opaque SIGILL.
+        // Catch it here with an actionable message. (EM_X86_64 == 62.)
+        const ushort ElfMachineX86_64 = 62;
+        if (header.Machine != ElfMachineX86_64)
+        {
+            throw new InvalidDataException(
+                $"Unsupported ELF machine type 0x{header.Machine:X4}; SharpEmu only runs x86-64 (EM_X86_64) images.");
+        }
     }
 
     private static unsafe T ReadUnmanaged<T>(ReadOnlySpan<byte> source, int offset)
@@ -2337,10 +2731,25 @@ public sealed class SelfLoader : ISelfLoader
         public uint Type => (uint)(Info & uint.MaxValue);
     }
 
+    private readonly record struct ModuleTlsInfo(ulong MemorySize, ulong StaticOffset);
+
     private enum RelocationValueKind : byte
     {
         Pointer = 0,
-        TlsModuleId = 1
+        TlsModuleId = 1,
+        // A pre-computed TLS offset written verbatim (TPOFF64/DTPOFF64). Unlike
+        // Pointer it is a signed displacement, not a mapped address, so it is
+        // patched as-is without the low-address validity warning.
+        TlsOffset = 2,
+        PcRelative = 3,
+        SymbolSize = 4,
+    }
+
+    private enum RelocationWriteKind : byte
+    {
+        UInt64 = 0,
+        UInt32 = 1,
+        Int32 = 2,
     }
 
     private readonly record struct RelocationDescriptor(
@@ -2349,7 +2758,9 @@ public sealed class SelfLoader : ISelfLoader
         string? ImportNid,
         ulong SymbolValue,
         RelocationValueKind ValueKind,
-        bool IsDataImport);
+        bool IsDataImport,
+        RelocationWriteKind WriteKind = RelocationWriteKind.UInt64,
+        bool IsWeak = false);
 
     private enum SelfSegmentResolveStatus
     {
@@ -2379,15 +2790,15 @@ public sealed class SelfLoader : ISelfLoader
         private readonly ushort _size2;
         private readonly ulong _fileSize;
         private readonly ushort _segmentCount;
-        private readonly ushort _unknown;
+        private readonly ushort _flags;
         private readonly uint _padding;
 
         public ushort SegmentCount => _segmentCount;
 
-        public ushort Unknown => _unknown;
-
         public ulong FileSize => _fileSize;
 
+        // Version, key type, and flags are signing metadata and vary across
+        // valid SELF images. They do not change the fixed header layout.
         public bool HasKnownLayout =>
             ((_ident0 == 0x4F &&
               _ident1 == 0x15 &&
@@ -2397,14 +2808,9 @@ public sealed class SelfLoader : ISelfLoader
               _ident1 == 0x14 &&
               _ident2 == 0xF5 &&
               _ident3 == 0xEE)) &&
-            _ident4 == 0x00 &&
             _ident5 == 0x01 &&
             _ident6 == 0x01 &&
-            _ident7 == 0x12 &&
-            _ident8 == 0x01 &&
-            _ident9 == 0x01 &&
-            _ident10 == 0x00 &&
-            _ident11 == 0x00;
+            _ident7 == 0x12;
     }
 
     [StructLayout(LayoutKind.Sequential, Pack = 1)]

src/SharpEmu.Core/Memory/PhysicalVirtualMemory.cs

@@ -42,12 +42,113 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
 
     private readonly IHostMemory _hostMemory;
     private ulong _guestAllocationArenaBase;
-    private ulong _guestAllocationOffset;
+    private readonly SortedDictionary<ulong, ulong> _guestAllocationFreeRanges = new();
+    private readonly Dictionary<ulong, (ulong Offset, ulong Size)> _guestAllocations = new();
     private static readonly ulong LazyReservePrimeBytes = ResolveLazyReservePrimeBytes();
 
     public PhysicalVirtualMemory(IHostMemory? hostMemory = null)
     {
-        _hostMemory = hostMemory ?? HostPlatform.Current.Memory;
+        _hostMemory = hostMemory ?? CrossPlatformHostMemory.Instance;
+    }
+
+    private sealed class CrossPlatformHostMemory : IHostMemory
+    {
+        public static readonly CrossPlatformHostMemory Instance = new();
+
+        public ulong Allocate(ulong desiredAddress, ulong size, HostPageProtection protection) =>
+            unchecked((ulong)HostMemory.Alloc(
+                (void*)desiredAddress,
+                (nuint)size,
+                HostMemory.MEM_RESERVE | HostMemory.MEM_COMMIT,
+                ToRawProtection(protection)));
+
+        public ulong Reserve(ulong desiredAddress, ulong size, HostPageProtection protection) =>
+            unchecked((ulong)HostMemory.Alloc(
+                (void*)desiredAddress,
+                (nuint)size,
+                HostMemory.MEM_RESERVE,
+                ToRawProtection(protection)));
+
+        public bool Commit(ulong address, ulong size, HostPageProtection protection) =>
+            HostMemory.Alloc(
+                (void*)address,
+                (nuint)size,
+                HostMemory.MEM_COMMIT,
+                ToRawProtection(protection)) != null;
+
+        public bool Free(ulong address) =>
+            HostMemory.Free((void*)address, 0, HostMemory.MEM_RELEASE);
+
+        public bool Protect(
+            ulong address,
+            ulong size,
+            HostPageProtection protection,
+            out uint rawOldProtection) =>
+            HostMemory.Protect(
+                (void*)address,
+                (nuint)size,
+                ToRawProtection(protection),
+                out rawOldProtection);
+
+        public bool ProtectRaw(
+            ulong address,
+            ulong size,
+            uint rawProtection,
+            out uint rawOldProtection) =>
+            HostMemory.Protect((void*)address, (nuint)size, rawProtection, out rawOldProtection);
+
+        public bool Query(ulong address, out HostRegionInfo info)
+        {
+            if (HostMemory.Query((void*)address, out var raw) == 0)
+            {
+                info = default;
+                return false;
+            }
+
+            var state = raw.State switch
+            {
+                HostMemory.MEM_FREE_STATE => HostRegionState.Free,
+                HostMemory.MEM_RESERVE => HostRegionState.Reserved,
+                _ => HostRegionState.Committed,
+            };
+
+            info = new HostRegionInfo(
+                raw.BaseAddress,
+                raw.AllocationBase,
+                raw.RegionSize,
+                state,
+                raw.State,
+                FromRawProtection(raw.Protect),
+                raw.Protect,
+                raw.AllocationProtect);
+            return true;
+        }
+
+        public void FlushInstructionCache(ulong address, ulong size) =>
+            HostMemory.FlushInstructionCache((void*)address, (nuint)size);
+
+        private static uint ToRawProtection(HostPageProtection protection) => protection switch
+        {
+            HostPageProtection.NoAccess => HostMemory.PAGE_NOACCESS,
+            HostPageProtection.ReadOnly => HostMemory.PAGE_READONLY,
+            HostPageProtection.ReadWrite => HostMemory.PAGE_READWRITE,
+            HostPageProtection.Execute => HostMemory.PAGE_EXECUTE,
+            HostPageProtection.ReadExecute => HostMemory.PAGE_EXECUTE_READ,
+            HostPageProtection.ReadWriteExecute => HostMemory.PAGE_EXECUTE_READWRITE,
+            HostPageProtection.ExecuteWriteCopy => 0x80,
+            _ => HostMemory.PAGE_NOACCESS,
+        };
+
+        private static HostPageProtection FromRawProtection(uint protection) => protection switch
+        {
+            HostMemory.PAGE_READONLY => HostPageProtection.ReadOnly,
+            HostMemory.PAGE_READWRITE => HostPageProtection.ReadWrite,
+            HostMemory.PAGE_EXECUTE => HostPageProtection.Execute,
+            HostMemory.PAGE_EXECUTE_READ => HostPageProtection.ReadExecute,
+            HostMemory.PAGE_EXECUTE_READWRITE => HostPageProtection.ReadWriteExecute,
+            0x80 => HostPageProtection.ExecuteWriteCopy,
+            _ => HostPageProtection.NoAccess,
+        };
     }
 
     public bool TryAllocateAtExact(ulong desiredAddress, ulong size, bool executable, out ulong actualAddress)
@@ -97,6 +198,24 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
         return true;
     }
 
+    public string DescribeAddressForDiagnostics(ulong address)
+    {
+        if (!_hostMemory.Query(address, out var info))
+        {
+            return "unable to query host memory at this address";
+        }
+
+        return info.State switch
+        {
+            HostRegionState.Free => "address reports free, but the exact-address reservation still failed",
+            HostRegionState.Reserved =>
+                $"already reserved by another host allocation (base=0x{info.AllocationBase:X16}, size=0x{info.RegionSize:X})",
+            HostRegionState.Committed =>
+                $"already committed by another host allocation (base=0x{info.AllocationBase:X16}, size=0x{info.RegionSize:X}, protect=0x{info.RawProtection:X})",
+            _ => $"in an unexpected host state (raw=0x{info.RawState:X})",
+        };
+    }
+
     public ulong AllocateAt(ulong desiredAddress, ulong size, bool executable = true, bool allowAlternative = true)
     {
         if (size == 0)
@@ -248,6 +367,35 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
         var requestedCursor = AlignUp(desiredAddress, effectiveAlignment);
         var cursor = GetAllocationSearchCursor(desiredAddress, requestedCursor, effectiveAlignment, executable);
 
+        // macOS needs alignment over-allocation; Linux uses exact-address search.
+        if (OperatingSystem.IsMacOS())
+        {
+            var reserveSize = effectiveAlignment > PageSize
+                ? alignedSize + effectiveAlignment
+                : alignedSize;
+            try
+            {
+                var posixAddress = AllocateAt(cursor, reserveSize, executable, allowAlternative: true);
+                if (posixAddress != 0)
+                {
+                    var alignedBase = AlignUp(posixAddress, effectiveAlignment);
+                    if (alignedBase + alignedSize <= posixAddress + reserveSize)
+                    {
+                        actualAddress = alignedBase;
+                        UpdateAllocationSearchCursor(desiredAddress, effectiveAlignment, executable, alignedBase + alignedSize);
+                        return true;
+                    }
+
+                    ReleaseUntrackedAllocation(posixAddress);
+                }
+            }
+            catch
+            {
+            }
+
+            return false;
+        }
+
         for (var attempt = 0; attempt < 0x10000; attempt++)
         {
             if (cursor == 0 || ulong.MaxValue - cursor < alignedSize)
@@ -261,19 +409,10 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
                 continue;
             }
 
-            try
-            {
-                actualAddress = AllocateAt(cursor, alignedSize, executable, allowAlternative: false);
-                if (actualAddress == cursor)
-                {
-                    UpdateAllocationSearchCursor(desiredAddress, effectiveAlignment, executable, actualAddress + alignedSize);
-                    return true;
-                }
-
-                actualAddress = 0;
-            }
-            catch
+            if (TryAllocateAtExact(cursor, alignedSize, executable, out actualAddress))
             {
+                UpdateAllocationSearchCursor(desiredAddress, effectiveAlignment, executable, actualAddress + alignedSize);
+                return true;
             }
 
             cursor = AlignUp(cursor + effectiveAlignment, effectiveAlignment);
@@ -282,6 +421,28 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
         return false;
     }
 
+    private void ReleaseUntrackedAllocation(ulong address)
+    {
+        _gate.EnterWriteLock();
+        try
+        {
+            for (var i = 0; i < _regions.Count; i++)
+            {
+                if (_regions[i].VirtualAddress == address)
+                {
+                    _regions.RemoveAt(i);
+                    break;
+                }
+            }
+        }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
+
+        _hostMemory.Free(address);
+    }
+
     public bool TryAllocateGuestMemory(ulong size, ulong alignment, out ulong address)
     {
         address = 0;
@@ -301,7 +462,9 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
                         GuestAllocationArenaSize,
                         executable: false,
                         allowAlternative: true);
-                    _guestAllocationOffset = GuestAllocationArenaStartOffset;
+                    _guestAllocationFreeRanges.Add(
+                        GuestAllocationArenaStartOffset,
+                        GuestAllocationArenaSize - GuestAllocationArenaStartOffset);
                 }
                 catch (Exception)
                 {
@@ -309,14 +472,89 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
                 }
             }
 
-            var alignedOffset = AlignUp(_guestAllocationOffset, alignment);
-            if (alignedOffset > GuestAllocationArenaSize || size > GuestAllocationArenaSize - alignedOffset)
+            ulong rangeOffset = 0;
+            ulong rangeSize = 0;
+            ulong alignedOffset = 0;
+            var found = false;
+            foreach (var range in _guestAllocationFreeRanges)
+            {
+                alignedOffset = AlignUp(range.Key, alignment);
+                if (alignedOffset >= range.Key &&
+                    alignedOffset - range.Key <= range.Value &&
+                    size <= range.Value - (alignedOffset - range.Key))
+                {
+                    rangeOffset = range.Key;
+                    rangeSize = range.Value;
+                    found = true;
+                    break;
+                }
+            }
+
+            if (!found)
             {
                 return false;
             }
 
+            _guestAllocationFreeRanges.Remove(rangeOffset);
+            if (alignedOffset > rangeOffset)
+            {
+                _guestAllocationFreeRanges.Add(rangeOffset, alignedOffset - rangeOffset);
+            }
+
+            var allocationEnd = alignedOffset + size;
+            var rangeEnd = rangeOffset + rangeSize;
+            if (allocationEnd < rangeEnd)
+            {
+                _guestAllocationFreeRanges.Add(allocationEnd, rangeEnd - allocationEnd);
+            }
+
             address = _guestAllocationArenaBase + alignedOffset;
-            _guestAllocationOffset = alignedOffset + size;
+            _guestAllocations.Add(address, (alignedOffset, size));
+            return true;
+        }
+    }
+
+    public bool TryFreeGuestMemory(ulong address)
+    {
+        lock (_guestAllocationGate)
+        {
+            if (!_guestAllocations.Remove(address, out var allocation))
+            {
+                return false;
+            }
+
+            var freeOffset = allocation.Offset;
+            var freeSize = allocation.Size;
+            ulong? previousOffset = null;
+            ulong? nextOffset = null;
+
+            foreach (var range in _guestAllocationFreeRanges)
+            {
+                if (range.Key < freeOffset)
+                {
+                    previousOffset = range.Key;
+                    continue;
+                }
+
+                nextOffset = range.Key;
+                break;
+            }
+
+            if (previousOffset is { } previous &&
+                previous + _guestAllocationFreeRanges[previous] == freeOffset)
+            {
+                freeOffset = previous;
+                freeSize += _guestAllocationFreeRanges[previous];
+                _guestAllocationFreeRanges.Remove(previous);
+            }
+
+            if (nextOffset is { } next && freeOffset + freeSize == next)
+            {
+                freeSize += _guestAllocationFreeRanges[next];
+                _guestAllocationFreeRanges.Remove(next);
+            }
+
+            _guestAllocationFreeRanges.Add(freeOffset, freeSize);
             return true;
         }
     }
@@ -380,7 +618,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
             }
 
             _guestAllocationArenaBase = 0;
-            _guestAllocationOffset = 0;
+            _guestAllocationFreeRanges.Clear();
+            _guestAllocations.Clear();
         }
     }
 
@@ -439,12 +678,33 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
 
     private void ApplySegmentProtection(ulong mapStart, ulong mapEnd, ProgramHeaderFlags flags)
     {
+        var runStart = mapStart;
+        var runFlags = ProgramHeaderFlags.None;
+        var hasRun = false;
+
         for (var pageAddress = mapStart; pageAddress < mapEnd; pageAddress += PageSize)
         {
             _pageProtections.TryGetValue(pageAddress, out var existingFlags);
             var mergedFlags = existingFlags | flags;
             _pageProtections[pageAddress] = mergedFlags;
-            SetProtection(pageAddress, PageSize, mergedFlags);
+
+            if (!hasRun)
+            {
+                runStart = pageAddress;
+                runFlags = mergedFlags;
+                hasRun = true;
+            }
+            else if (mergedFlags != runFlags)
+            {
+                SetProtection(runStart, pageAddress - runStart, runFlags);
+                runStart = pageAddress;
+                runFlags = mergedFlags;
+            }
+        }
+
+        if (hasRun)
+        {
+            SetProtection(runStart, mapEnd - runStart, runFlags);
         }
     }
 
@@ -789,9 +1049,14 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryA
         _gate.EnterReadLock();
         try
         {
-            return FindRegion(virtualAddress, 1) is not null
-                ? (void*)virtualAddress
-                : null;
+            var region = FindRegion(virtualAddress, 1);
+            if (region is null ||
+                (region.IsReservedOnly && !EnsureRangeCommitted(virtualAddress, 1, region)))
+            {
+                return null;
+            }
+
+            return (void*)virtualAddress;
         }
         finally
         {

src/SharpEmu.Core/Memory/VirtualMemory.cs

@@ -41,15 +41,14 @@ public sealed class VirtualMemory : IVirtualMemory
 
         lock (_gate)
         {
-            foreach (var existing in _regions)
+            var insertionIndex = FindInsertionIndex(virtualAddress);
+            if ((insertionIndex > 0 && virtualAddress < _regions[insertionIndex - 1].EndAddress) ||
+                (insertionIndex < _regions.Count && endAddress > _regions[insertionIndex].Region.VirtualAddress))
             {
-                if (virtualAddress < existing.EndAddress && endAddress > existing.Region.VirtualAddress)
-                {
-                    throw new InvalidOperationException("Attempted to map an overlapping virtual memory region.");
-                }
+                throw new InvalidOperationException("Attempted to map an overlapping virtual memory region.");
             }
 
-            _regions.Add(new MappedRegion(
+            _regions.Insert(insertionIndex, new MappedRegion(
                 new VirtualMemoryRegion(virtualAddress, memorySize, fileOffset, (ulong)fileData.Length, protection),
                 endAddress,
                 backingMemory));
@@ -74,12 +73,12 @@ public sealed class VirtualMemory : IVirtualMemory
     {
         lock (_gate)
         {
-            if (!TryResolveRegion(virtualAddress, destination.Length, out var region, out var offset))
+            if (!TryValidateRange(virtualAddress, destination.Length, ProgramHeaderFlags.Read, out var regionIndex))
             {
                 return false;
             }
 
-            region.BackingMemory.AsSpan(offset, destination.Length).CopyTo(destination);
+            CopyFromRegions(virtualAddress, destination, regionIndex);
             return true;
         }
     }
@@ -88,39 +87,127 @@ public sealed class VirtualMemory : IVirtualMemory
     {
         lock (_gate)
         {
-            if (!TryResolveRegion(virtualAddress, source.Length, out var region, out var offset))
+            if (!TryValidateRange(virtualAddress, source.Length, ProgramHeaderFlags.Write, out var regionIndex))
             {
                 return false;
             }
 
-            source.CopyTo(region.BackingMemory.AsSpan(offset, source.Length));
+            CopyToRegions(virtualAddress, source, regionIndex);
             return true;
         }
     }
 
-    private bool TryResolveRegion(ulong virtualAddress, int length, out MappedRegion region, out int offset)
+    private bool TryValidateRange(
+        ulong virtualAddress,
+        int length,
+        ProgramHeaderFlags requiredProtection,
+        out int regionIndex)
     {
-        foreach (var candidate in _regions)
+        regionIndex = FindContainingRegionIndex(virtualAddress);
+        if (regionIndex < 0)
         {
-            if (virtualAddress < candidate.Region.VirtualAddress || virtualAddress >= candidate.EndAddress)
-            {
-                continue;
-            }
-
-            var candidateOffset = checked((int)(virtualAddress - candidate.Region.VirtualAddress));
-            if (candidateOffset + length > candidate.BackingMemory.Length)
-            {
-                break;
-            }
-
-            region = candidate;
-            offset = candidateOffset;
-            return true;
+            return false;
         }
 
-        region = default;
-        offset = 0;
-        return false;
+        var currentAddress = virtualAddress;
+        var remaining = length;
+        var currentIndex = regionIndex;
+        while (true)
+        {
+            if (currentIndex >= _regions.Count)
+            {
+                return false;
+            }
+
+            var region = _regions[currentIndex];
+            if (currentAddress < region.Region.VirtualAddress ||
+                currentAddress >= region.EndAddress ||
+                (region.Region.Protection & requiredProtection) == 0)
+            {
+                return false;
+            }
+
+            if (remaining == 0)
+            {
+                return true;
+            }
+
+            var available = region.EndAddress - currentAddress;
+            var chunkLength = (int)Math.Min((ulong)remaining, available);
+            remaining -= chunkLength;
+            if (remaining == 0)
+            {
+                return true;
+            }
+
+            currentAddress += (ulong)chunkLength;
+            currentIndex++;
+        }
+    }
+
+    private int FindContainingRegionIndex(ulong virtualAddress)
+    {
+        var insertionIndex = FindInsertionIndex(virtualAddress);
+        if (insertionIndex < _regions.Count &&
+            _regions[insertionIndex].Region.VirtualAddress == virtualAddress)
+        {
+            return insertionIndex;
+        }
+
+        var candidateIndex = insertionIndex - 1;
+        return candidateIndex >= 0 && virtualAddress < _regions[candidateIndex].EndAddress
+            ? candidateIndex
+            : -1;
+    }
+
+    private void CopyFromRegions(ulong virtualAddress, Span<byte> destination, int regionIndex)
+    {
+        var copied = 0;
+        var currentAddress = virtualAddress;
+        while (copied < destination.Length)
+        {
+            var region = _regions[regionIndex++];
+            var regionOffset = checked((int)(currentAddress - region.Region.VirtualAddress));
+            var chunkLength = Math.Min(destination.Length - copied, region.BackingMemory.Length - regionOffset);
+            region.BackingMemory.AsSpan(regionOffset, chunkLength).CopyTo(destination[copied..]);
+            copied += chunkLength;
+            currentAddress += (ulong)chunkLength;
+        }
+    }
+
+    private void CopyToRegions(ulong virtualAddress, ReadOnlySpan<byte> source, int regionIndex)
+    {
+        var copied = 0;
+        var currentAddress = virtualAddress;
+        while (copied < source.Length)
+        {
+            var region = _regions[regionIndex++];
+            var regionOffset = checked((int)(currentAddress - region.Region.VirtualAddress));
+            var chunkLength = Math.Min(source.Length - copied, region.BackingMemory.Length - regionOffset);
+            source.Slice(copied, chunkLength).CopyTo(region.BackingMemory.AsSpan(regionOffset, chunkLength));
+            copied += chunkLength;
+            currentAddress += (ulong)chunkLength;
+        }
+    }
+
+    private int FindInsertionIndex(ulong virtualAddress)
+    {
+        var lower = 0;
+        var upper = _regions.Count;
+        while (lower < upper)
+        {
+            var middle = lower + ((upper - lower) / 2);
+            if (_regions[middle].Region.VirtualAddress < virtualAddress)
+            {
+                lower = middle + 1;
+            }
+            else
+            {
+                upper = middle;
+            }
+        }
+
+        return lower;
     }
 
     private readonly record struct MappedRegion(VirtualMemoryRegion Region, ulong EndAddress, byte[] BackingMemory);

src/SharpEmu.Core/Runtime/SharpEmuRuntime.cs

@@ -7,25 +7,22 @@ using SharpEmu.Core.Cpu.Disasm;
 using SharpEmu.Core.Loader;
 using SharpEmu.Core.Memory;
 using SharpEmu.HLE;
-using SharpEmu.HLE.Host;
 using SharpEmu.Libs.VideoOut;
 using SharpEmu.Libs.Kernel;
 using SharpEmu.Libs.AppContent;
 using SharpEmu.Libs.SaveData;
-using SharpEmu.Logging;
+using SharpEmu.Libs.Fiber;
+using SharpEmu.Libs.SystemService;
 using System.Buffers.Binary;
 using System.Collections.Generic;
 using System.Linq;
 using System.Text;
-using System.IO;
 
 namespace SharpEmu.Core.Runtime;
 
 public sealed class SharpEmuRuntime : ISharpEmuRuntime
 {
-    private static readonly SharpEmuLogger Log = SharpEmuLog.For("Runtime");
-
-    private readonly record struct LoadedModuleImage(string Path, SelfImage Image);
+    private readonly record struct LoadedModuleImage(string Path, SelfImage Image, int Handle, bool StartAtBoot);
 
     private static readonly HashSet<string> PreloadSkipModules = new(StringComparer.OrdinalIgnoreCase)
     {
@@ -84,22 +81,19 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             ImportTraceLimit = Math.Max(0, options.ImportTraceLimit),
         };
         var moduleManager = new ModuleManager();
-        moduleManager.RegisterFromAssembly(typeof(KernelExports).Assembly, Generation.Gen4 | Generation.Gen5, Aerolib.Instance);
+        // The compile-time generated registry (SharpEmu.SourceGenerators) is the sole
+        // registration source; content tests in SharpEmu.Libs.Tests pin its invariants.
+        moduleManager.RegisterExports(SharpEmu.Generated.SysAbiExportRegistry.CreateExports(Generation.Gen4 | Generation.Gen5));
         moduleManager.Freeze();
 
-        // Resolve the host platform once at the composition root; on unsupported
-        // OSes this throws PlatformNotSupportedException with a clear message
-        // instead of failing on the first native call.
-        var hostPlatform = HostPlatform.Current;
-
-        var virtualMemory = new PhysicalVirtualMemory(hostPlatform.Memory);
+        var virtualMemory = new PhysicalVirtualMemory();
 
         var fileSystem = new PhysicalFileSystem();
 
         return new SharpEmuRuntime(
             new SelfLoader(),
             virtualMemory,
-            new CpuDispatcher(virtualMemory, moduleManager, hostPlatform: hostPlatform),
+            new CpuDispatcher(virtualMemory, moduleManager),
             moduleManager,
             Aerolib.Instance,
             cpuExecutionOptions,
@@ -137,20 +131,21 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
     {
         var normalizedEbootPath = Path.GetFullPath(ebootPath);
         using var app0Binding = BindApp0Root(normalizedEbootPath);
-        Log.Info($"Loading: {ebootPath}");
+        Console.Error.WriteLine($"[RUNTIME] Loading: {ebootPath}");
         LastExecutionDiagnostics = null;
         LastExecutionTrace = null;
         LastSessionSummary = null;
         LastBasicBlockTrace = null;
         LastMilestoneLog = null;
+        FiberExports.ResetRuntimeState();
         KernelModuleRegistry.Reset();
         var image = LoadImage(normalizedEbootPath);
-        VideoOutExports.ConfigureApplicationInfo(image.Title, image.TitleId, image.Version, BuildInfo.CommitSha);
+        VideoOutExports.ConfigureApplicationInfo(image.Title, image.TitleId, image.Version);
         SaveDataExports.ConfigureApplicationInfo(image.TitleId);
-        LogAppBundleInfo(normalizedEbootPath, image);
-        RegisterLoadedModule(normalizedEbootPath, image, isMain: true, isSystemModule: false);
+        SystemServiceExports.ConfigureApplicationInfo(image.TitleId);
+        _ = RegisterLoadedModule(normalizedEbootPath, image, isMain: true, isSystemModule: false);
         KernelRuntimeCompatExports.ConfigureProcessProcParamAddress(image.ProcParamAddress);
-        Log.Info($"Entry: 0x{image.EntryPoint:X16}");
+        Console.Error.WriteLine($"[RUNTIME] Entry: 0x{image.EntryPoint:X16}");
         var generation = image.ElfHeader.AbiVersion == 2 ? Generation.Gen5 : Generation.Gen4;
         var activeImportStubs = new Dictionary<ulong, string>(image.ImportStubs);
         var activeRuntimeSymbols = new Dictionary<string, ulong>(image.RuntimeSymbols, StringComparer.Ordinal);
@@ -173,7 +168,7 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             processImageName);
         if (initializerResult is { } failedInitializerResult)
         {
-            Log.Error($"Initializer dispatch failed: {failedInitializerResult}");
+            Console.Error.WriteLine($"[RUNTIME] Initializer dispatch failed: {failedInitializerResult}");
             LastExecutionTrace = _cpuDispatcher.LastImportResolutionTrace;
             LastMilestoneLog = _cpuDispatcher.LastMilestoneLog;
             LastSessionSummary = BuildSessionSummary(_cpuDispatcher.LastSessionSummary);
@@ -181,8 +176,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             return failedInitializerResult;
         }
 
-        Log.Info($"Dispatching, gen: {generation}");
-        Log.Debug($"About to call DispatchEntry with entryPoint=0x{image.EntryPoint:X16}");
+        Console.Error.WriteLine($"[RUNTIME] Dispatching, gen: {generation}");
+        Console.Error.WriteLine($"[RUNTIME] About to call DispatchEntry with entryPoint=0x{image.EntryPoint:X16}");
 
         var result = _cpuDispatcher.DispatchEntry(
             image.EntryPoint,
@@ -192,8 +187,18 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             processImageName,
             _cpuExecutionOptions);
 
-        Log.Info($"DispatchEntry returned: {result}");
-        Log.Info($"Dispatch result: {result}");
+        Console.Error.WriteLine($"[RUNTIME] DispatchEntry returned: {result}");
+        Console.Error.WriteLine($"[RUNTIME] Dispatch result: {result}");
+
+        // Stop is a host operation, not an emulation failure. The detailed
+        // trace and session-summary builders can traverse a partially torn
+        // down native backend, delaying the GUI exit callback indefinitely.
+        if (HostSessionControl.IsShutdownRequested)
+        {
+            Console.Error.WriteLine("[RUNTIME] Skipping post-exit diagnostics for host shutdown.");
+            return result;
+        }
+
         LastExecutionTrace = _cpuDispatcher.LastImportResolutionTrace;
         LastMilestoneLog = _cpuDispatcher.LastMilestoneLog;
         LastSessionSummary = BuildSessionSummary(_cpuDispatcher.LastSessionSummary);
@@ -364,66 +369,6 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
         return result;
     }
 
-    private static void LogAppBundleInfo(string ebootPath, SelfImage image)
-    {
-        var executableName = Path.GetFileName(ebootPath);
-        if (string.IsNullOrWhiteSpace(executableName))
-        {
-            executableName = "eboot.bin";
-        }
-
-        var displayName = string.IsNullOrWhiteSpace(image.Title) ? "(unknown)" : image.Title!.Trim();
-        var titleId = string.IsNullOrWhiteSpace(image.TitleId) ? "(unknown)" : image.TitleId!.Trim();
-        var version = string.IsNullOrWhiteSpace(image.Version) ? "(unknown)" : image.Version!.Trim();
-        var contentId = ResolveContentId(Path.GetDirectoryName(ebootPath)) ?? "(unknown)";
-
-        var builder = new StringBuilder();
-        builder.AppendLine("App bundle info:");
-        builder.AppendLine($"- Display name: {displayName}");
-        builder.AppendLine($"- Version: {version}");
-        builder.AppendLine($"- Title ID: {titleId}");
-        builder.AppendLine($"- Content ID: {contentId}");
-        builder.AppendLine($"- Executable: {executableName}");
-        builder.Append("- Platform: PlayStation 5");
-        Log.Info(builder.ToString());
-    }
-
-    private static string? ResolveContentId(string? bundleRoot)
-    {
-        if (string.IsNullOrEmpty(bundleRoot))
-        {
-            return null;
-        }
-
-        foreach (var candidate in new[]
-        {
-            Path.Combine(bundleRoot, "sce_sys", "param.json"),
-            Path.Combine(bundleRoot, "param.json"),
-        })
-        {
-            if (!File.Exists(candidate))
-            {
-                continue;
-            }
-
-            try
-            {
-                using var doc = System.Text.Json.JsonDocument.Parse(File.ReadAllBytes(candidate));
-                if (doc.RootElement.TryGetProperty("contentId", out var contentId))
-                {
-                    return contentId.GetString();
-                }
-            }
-            catch (Exception ex) when (ex is IOException or System.Text.Json.JsonException)
-            {
-            }
-
-            break;
-        }
-
-        return null;
-    }
-
     private static App0BindingScope? BindApp0Root(string normalizedEbootPath)
     {
         const string app0VariableName = "SHARPEMU_APP0_DIR";
@@ -438,6 +383,20 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             return null;
         }
 
+        // Dump tools commonly place decrypted executables in an app0/decrypted
+        // sidecar while leaving Unity data in the parent app0 directory. Keep
+        // loading code/modules beside the decrypted eboot, but mount /app0 at
+        // the content-bearing parent so boot.config, metadata and assets resolve.
+        if (string.Equals(Path.GetFileName(app0Root), "decrypted", StringComparison.OrdinalIgnoreCase))
+        {
+            var parentRoot = Path.GetDirectoryName(app0Root);
+            if (!string.IsNullOrWhiteSpace(parentRoot) &&
+                Directory.Exists(Path.Combine(parentRoot, "Media")))
+            {
+                app0Root = parentRoot;
+            }
+        }
+
         Environment.SetEnvironmentVariable(app0VariableName, app0Root);
         return new App0BindingScope(app0VariableName);
     }
@@ -483,6 +442,56 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
         return null;
     }
 
+    private bool TryGetEhFrameInfo(
+        SelfImage image,
+        ulong imageSize,
+        out ulong ehFrameHeaderAddress,
+        out ulong ehFrameAddress,
+        out ulong ehFrameSize)
+    {
+        ehFrameHeaderAddress = 0;
+        ehFrameAddress = 0;
+        ehFrameSize = 0;
+        var imageBase = image.EntryPoint >= image.ElfHeader.EntryPoint
+            ? image.EntryPoint - image.ElfHeader.EntryPoint
+            : 0UL;
+        for (var i = 0; i < image.ProgramHeaders.Count; i++)
+        {
+            var header = image.ProgramHeaders[i];
+            if (header.HeaderType != ProgramHeaderType.GnuEhFrame || header.MemorySize < 8)
+            {
+                continue;
+            }
+
+            var headerAddress = imageBase + header.VirtualAddress;
+            Span<byte> ehHeader = stackalloc byte[8];
+            if (!_virtualMemory.TryRead(headerAddress, ehHeader) ||
+                ehHeader[0] != 1 ||
+                ehHeader[1] != 0x1B)
+            {
+                continue;
+            }
+
+            var relativeOffset = BinaryPrimitives.ReadInt32LittleEndian(ehHeader[4..]);
+            ehFrameHeaderAddress = headerAddress;
+            ehFrameAddress = unchecked((ulong)((long)headerAddress + 4 + relativeOffset));
+            if (ehFrameAddress < 0x10000)
+            {
+                continue;
+            }
+
+            var relativeEhFrameAddress = ehFrameAddress - imageBase;
+            ehFrameSize = header.VirtualAddress > relativeEhFrameAddress
+                ? header.VirtualAddress - relativeEhFrameAddress
+                : imageSize > header.VirtualAddress
+                    ? imageSize - header.VirtualAddress
+                    : 0;
+            return true;
+        }
+
+        return false;
+    }
+
     private OrbisGen2Result? RunPreloadedModuleInitializers(
         IReadOnlyList<LoadedModuleImage> loadedModuleImages,
         Generation generation,
@@ -492,20 +501,30 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
         for (var i = 0; i < loadedModuleImages.Count; i++)
         {
             var loadedModule = loadedModuleImages[i];
+            if (!loadedModule.StartAtBoot)
+            {
+                continue;
+            }
+
             var initEntryPoint = loadedModule.Image.InitFunctionEntryPoint;
             if (initEntryPoint < 0x10000)
             {
                 continue;
             }
 
+            if (!KernelModuleRegistry.TryBeginModuleStart(loadedModule.Handle, out _))
+            {
+                continue;
+            }
+
             var moduleName = Path.GetFileName(loadedModule.Path);
             if (string.IsNullOrWhiteSpace(moduleName))
             {
                 moduleName = $"module#{i}";
             }
 
-            Log.Info(
-                $"Starting module {moduleName}: dt_init=0x{initEntryPoint:X16}");
+            Console.Error.WriteLine(
+                $"[RUNTIME] Starting module {moduleName}: dt_init=0x{initEntryPoint:X16}");
 
             var result = _cpuDispatcher.DispatchModuleInitializer(
                 initEntryPoint,
@@ -514,10 +533,13 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
                 activeRuntimeSymbols,
                 moduleName,
                 _cpuExecutionOptions);
+            KernelModuleRegistry.CompleteModuleStart(
+                loadedModule.Handle,
+                result == OrbisGen2Result.ORBIS_GEN2_OK);
             if (result != OrbisGen2Result.ORBIS_GEN2_OK)
             {
-                Log.Error(
-                    $"Module start failed: {moduleName} -> {result}");
+                Console.Error.WriteLine(
+                    $"[RUNTIME] Module start failed: {moduleName} -> {result}");
                 return result;
             }
         }
@@ -538,8 +560,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             return null;
         }
 
-        Log.Info(
-            $"Running initializers for {label}: preinit={image.PreInitializerFunctions.Count}, init={image.InitializerFunctions.Count}");
+        Console.Error.WriteLine(
+            $"[RUNTIME] Running initializers for {label}: preinit={image.PreInitializerFunctions.Count}, init={image.InitializerFunctions.Count}");
 
         var result = RunInitializerList(
             $"{label}:preinit",
@@ -578,8 +600,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
                 continue;
             }
 
-            Log.Debug(
-                $"  Initializer {label}[{i}] -> 0x{initializerAddress:X16}");
+            Console.Error.WriteLine(
+                $"[RUNTIME]   Initializer {label}[{i}] -> 0x{initializerAddress:X16}");
 
             var result = _cpuDispatcher.DispatchEntry(
                 initializerAddress,
@@ -611,12 +633,17 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
 
         var moduleDirectories = new[]
         {
-            Path.Combine(ebootDirectory, "sce_module"),
-            Path.Combine(ebootDirectory, "sce_modules"),
-            Path.Combine(ebootDirectory, "Media", "Modules"),
+            (Path: Path.Combine(ebootDirectory, "sce_module"), StartAtBoot: true),
+            (Path: Path.Combine(ebootDirectory, "sce_modules"), StartAtBoot: true),
+            (Path: Path.Combine(ebootDirectory, "Media", "Modules"), StartAtBoot: true),
+            // Unity native plugins are loaded later through sceKernelLoadStartModule. Map
+            // them up front so the HLE loader can return a real module handle and dlsym
+            // can resolve their exports, but defer DT_INIT until the guest requests them.
+            (Path: Path.Combine(ebootDirectory, "Media", "Plugins"), StartAtBoot: false),
         }
-        .Distinct(StringComparer.OrdinalIgnoreCase)
-        .Where(Directory.Exists)
+        .GroupBy(entry => entry.Path, StringComparer.OrdinalIgnoreCase)
+        .Select(group => group.First())
+        .Where(entry => Directory.Exists(entry.Path))
         .ToArray();
 
         if (moduleDirectories.Length == 0)
@@ -626,28 +653,30 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
 
         var allModulePaths = moduleDirectories
             .SelectMany(directory => Directory
-                .EnumerateFiles(directory)
-                .OrderBy(path => path, StringComparer.OrdinalIgnoreCase))
-            .Where(path =>
+                .EnumerateFiles(directory.Path)
+                .OrderBy(path => path, StringComparer.OrdinalIgnoreCase)
+                .Select(path => (Path: path, directory.StartAtBoot)))
+            .Where(entry =>
             {
-                var extension = Path.GetExtension(path);
+                var extension = Path.GetExtension(entry.Path);
                 return string.Equals(extension, ".prx", StringComparison.OrdinalIgnoreCase) ||
                        string.Equals(extension, ".sprx", StringComparison.OrdinalIgnoreCase);
             })
-            .Distinct(StringComparer.OrdinalIgnoreCase)
+            .GroupBy(entry => entry.Path, StringComparer.OrdinalIgnoreCase)
+            .Select(group => group.First())
             .ToArray();
 
         var modulePaths = allModulePaths
-            .Where(ShouldPreloadModule)
+            .Where(entry => ShouldPreloadModule(entry.Path))
             .ToArray();
         var skippedModules = allModulePaths
-            .Where(path => !ShouldPreloadModule(path))
-            .Select(Path.GetFileName)
+            .Where(entry => !ShouldPreloadModule(entry.Path))
+            .Select(entry => Path.GetFileName(entry.Path))
             .Where(name => !string.IsNullOrWhiteSpace(name))
             .ToArray();
         if (skippedModules.Length > 0)
         {
-            Log.Info($"Skipping {skippedModules.Length} core module(s): {string.Join(", ", skippedModules)}");
+            Console.Error.WriteLine($"[RUNTIME] Skipping {skippedModules.Length} core module(s): {string.Join(", ", skippedModules)}");
         }
 
         if (modulePaths.Length == 0)
@@ -655,14 +684,15 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             return loadedImages;
         }
 
-        Log.Debug($"Module search directories: {string.Join(", ", moduleDirectories)}");
-        Log.Info($"Loading {modulePaths.Length} module(s)...");
+        Console.Error.WriteLine($"[RUNTIME] Module search directories: {string.Join(", ", moduleDirectories.Select(entry => entry.Path))}");
+        Console.Error.WriteLine($"[RUNTIME] Loading {modulePaths.Length} module(s)...");
         var loadedModules = 0;
         var failedModules = 0;
         var mergedImportCount = 0;
         var mergedSymbolCount = 0;
-        foreach (var modulePath in modulePaths)
+        foreach (var moduleEntry in modulePaths)
         {
+            var modulePath = moduleEntry.Path;
             try
             {
                 var fileInfo = new FileInfo(modulePath);
@@ -687,25 +717,52 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
 
                 mergedImportCount += MergeImportStubs(importStubs, moduleImage.ImportStubs, modulePath);
                 mergedSymbolCount += MergeRuntimeSymbols(runtimeSymbols, moduleImage.RuntimeSymbols);
-                RegisterLoadedModule(modulePath, moduleImage, isMain: false, isSystemModule: false);
-                loadedImages.Add(new LoadedModuleImage(modulePath, moduleImage));
+                InstallNativePluginCompatibilityHooks(importStubs, moduleImage, modulePath);
+                var moduleHandle = RegisterLoadedModule(modulePath, moduleImage, isMain: false, isSystemModule: false);
+                var moduleName = Path.GetFileName(modulePath);
+                var startAtBoot = moduleEntry.StartAtBoot ||
+                    string.Equals(moduleName, "libfmod.prx", StringComparison.OrdinalIgnoreCase) ||
+                    string.Equals(moduleName, "libfmodstudio.prx", StringComparison.OrdinalIgnoreCase);
+                loadedImages.Add(new LoadedModuleImage(modulePath, moduleImage, moduleHandle, startAtBoot));
                 loadedModules++;
 
-                Log.Info(
-                    $"Loaded module {Path.GetFileName(modulePath)}: entry=0x{moduleImage.EntryPoint:X16}, imports={moduleImage.ImportStubs.Count}, symbols={moduleImage.RuntimeSymbols.Count}");
+                Console.Error.WriteLine(
+                    $"[RUNTIME] Loaded module {Path.GetFileName(modulePath)}: entry=0x{moduleImage.EntryPoint:X16}, imports={moduleImage.ImportStubs.Count}, symbols={moduleImage.RuntimeSymbols.Count}");
             }
             catch (Exception ex)
             {
                 failedModules++;
-                Log.Error($"Module load failed: {modulePath} ({ex.GetType().Name}: {ex.Message})", ex);
+                Console.Error.WriteLine($"[RUNTIME] Module load failed: {modulePath} ({ex.GetType().Name}: {ex.Message})");
             }
         }
 
-        Log.Info(
-            $"Module preload summary: loaded={loadedModules}, failed={failedModules}, merged_imports={mergedImportCount}, merged_symbols={mergedSymbolCount}");
+        Console.Error.WriteLine(
+            $"[RUNTIME] Module preload summary: loaded={loadedModules}, failed={failedModules}, merged_imports={mergedImportCount}, merged_symbols={mergedSymbolCount}");
         return loadedImages;
     }
 
+    private static void InstallNativePluginCompatibilityHooks(
+        IDictionary<ulong, string> importStubs,
+        SelfImage moduleImage,
+        string modulePath)
+    {
+        if (!string.Equals(Path.GetFileName(modulePath), "libfmod.prx", StringComparison.OrdinalIgnoreCase))
+        {
+            return;
+        }
+
+        ReadOnlySpan<string> compatibilityNids = ["uPLTdl3psGk"];
+        foreach (var nid in compatibilityNids)
+        {
+            if (moduleImage.RuntimeSymbols.TryGetValue(nid, out var address) && address >= 0x10000)
+            {
+                importStubs[address] = nid;
+                Console.Error.WriteLine(
+                    $"[RUNTIME] Installed FMOD compatibility hook: {nid} -> 0x{address:X16}");
+            }
+        }
+    }
+
     private void RebindImportedDataSymbols(
         SelfImage mainImage,
         IReadOnlyList<LoadedModuleImage> loadedModuleImages,
@@ -722,8 +779,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
 
         if (rebound != 0 || unresolved != 0)
         {
-            Log.Info(
-                $"Imported data rebind: rebound={rebound}, unresolved={unresolved}");
+            Console.Error.WriteLine(
+                $"[RUNTIME] Imported data rebind: rebound={rebound}, unresolved={unresolved}");
         }
     }
 
@@ -755,8 +812,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             {
                 if (logRebind)
                 {
-                    Log.Warning(
-                        $"Imported data unresolved: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} addend=0x{unchecked((ulong)relocation.Addend):X16}");
+                    Console.Error.WriteLine(
+                        $"[RUNTIME] Imported data unresolved: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} addend=0x{unchecked((ulong)relocation.Addend):X16}");
                 }
 
                 unresolved++;
@@ -768,8 +825,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             {
                 if (logRebind)
                 {
-                    Log.Error(
-                        $"Imported data write-failed: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} value=0x{reboundValue:X16}");
+                    Console.Error.WriteLine(
+                        $"[RUNTIME] Imported data write-failed: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} value=0x{reboundValue:X16}");
                 }
 
                 unresolved++;
@@ -778,8 +835,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
 
             if (logRebind)
             {
-                Log.Debug(
-                    $"Imported data rebound: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} value=0x{reboundValue:X16}");
+                Console.Error.WriteLine(
+                    $"[RUNTIME] Imported data rebound: nid={relocation.Nid} target=0x{relocation.TargetAddress:X16} value=0x{reboundValue:X16}");
             }
 
             rebound++;
@@ -815,8 +872,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             {
                 if (!string.Equals(existingNid, nid, StringComparison.Ordinal))
                 {
-                    Log.Warning(
-                        $"Import stub conflict at 0x{address:X16}: keep={existingNid}, skip={nid} ({Path.GetFileName(modulePath)})");
+                    Console.Error.WriteLine(
+                        $"[RUNTIME] Import stub conflict at 0x{address:X16}: keep={existingNid}, skip={nid} ({Path.GetFileName(modulePath)})");
                 }
 
                 continue;
@@ -910,7 +967,7 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
         return !PreloadSkipModules.Contains(fileName);
     }
 
-    private static void RegisterLoadedModule(string modulePath, SelfImage image, bool isMain, bool isSystemModule)
+    private int RegisterLoadedModule(string modulePath, SelfImage image, bool isMain, bool isSystemModule)
     {
         if (!TryComputeImageRange(image, out var baseAddress, out var size))
         {
@@ -918,15 +975,27 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             size = 0;
         }
 
+        _ = TryGetEhFrameInfo(
+            image,
+            size,
+            out var ehFrameHeaderAddress,
+            out var ehFrameAddress,
+            out var ehFrameSize);
         var handle = KernelModuleRegistry.RegisterModule(
             modulePath,
             baseAddress,
             size,
             image.EntryPoint,
+            image.InitFunctionEntryPoint,
+            ehFrameHeaderAddress,
+            ehFrameAddress,
+            ehFrameSize,
             isMain,
             isSystemModule);
-        Log.Info(
-            $"Registered module handle={handle} name={Path.GetFileName(modulePath)} base=0x{baseAddress:X16} size=0x{size:X16}");
+        KernelModuleRegistry.RegisterModuleSymbols(handle, image.RuntimeSymbols);
+        Console.Error.WriteLine(
+            $"[RUNTIME] Registered module handle={handle} name={Path.GetFileName(modulePath)} base=0x{baseAddress:X16} size=0x{size:X16}");
+        return handle;
     }
 
     private static bool TryComputeImageRange(SelfImage image, out ulong baseAddress, out ulong size)
@@ -1093,6 +1162,16 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
             disposableDispatcher.Dispose();
         }
 
+        if (_cpuDispatcher is CpuDispatcher { NativeSessionLeaked: true })
+        {
+            // A guest worker is still inside guest code; unmapping the guest
+            // address space under it would fault the whole process, which
+            // hosts the GUI launcher in embedded sessions.
+            Console.Error.WriteLine(
+                "[RUNTIME] Guest workers were still active at teardown; keeping the guest address space mapped.");
+            return;
+        }
+
         if (_virtualMemory is IDisposable disposableMemory)
         {
             disposableMemory.Dispose();

src/SharpEmu.Core/SharpEmu.Core.csproj

@@ -14,6 +14,10 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <PackageReference Include="Iced" />
   </ItemGroup>
 
+  <ItemGroup>
+    <InternalsVisibleTo Include="SharpEmu.Libs.Tests" />
+  </ItemGroup>
+
   <PropertyGroup>
     <NoWarn>$(NoWarn);1591</NoWarn>
   </PropertyGroup>

src/SharpEmu.Core/packages.lock.json

@@ -1,127 +0,0 @@
-{
-  "version": 2,
-  "dependencies": {
-    "net10.0": {
-      "Iced": {
-        "type": "Direct",
-        "requested": "[1.21.0, )",
-        "resolved": "1.21.0",
-        "contentHash": "dv5+81Q1TBQvVMSOOOmRcjJmvWcX3BZPZsIq31+RLc5cNft0IHAyNlkdb7ZarOWG913PyBoFDsDXoCIlKmLclg=="
-      },
-      "Microsoft.DotNet.PlatformAbstractions": {
-        "type": "Transitive",
-        "resolved": "3.1.6",
-        "contentHash": "jek4XYaQ/PGUwDKKhwR8K47Uh1189PFzMeLqO83mXrXQVIpARZCcfuDedH50YDTepBkfijCZN5U/vZi++erxtg=="
-      },
-      "Microsoft.Extensions.DependencyModel": {
-        "type": "Transitive",
-        "resolved": "9.0.9",
-        "contentHash": "fNGvKct2De8ghm0Bpfq0iWthtzIWabgOTi+gJhNOPhNJIowXNEUE2eZNW/zNCzrHVA3PXg2yZ+3cWZndC2IqYA=="
-      },
-      "Silk.NET.Core": {
-        "type": "Transitive",
-        "resolved": "2.23.0",
-        "contentHash": "D7AT/nnwlB+4RZ84XY8QNGBZMJI5z9l4CSSETIJ1wCfRJzRt/341y3MRZ4HbnFz4r/IGaWOEZr86iE+0/65yyQ==",
-        "dependencies": {
-          "Microsoft.DotNet.PlatformAbstractions": "3.1.6",
-          "Microsoft.Extensions.DependencyModel": "9.0.9"
-        }
-      },
-      "Silk.NET.GLFW": {
-        "type": "Transitive",
-        "resolved": "2.23.0",
-        "contentHash": "UIs4sH57xlPUNHQ/1bt9rymPWlGy8IMDCNv86h0iM4TOA1CkIx0XM/n/tA4AReh1zQkNrvkxPEdZ3Blvy1dyXg==",
-        "dependencies": {
-          "Silk.NET.Core": "2.23.0",
-          "Ultz.Native.GLFW": "3.4.0"
-        }
-      },
-      "Silk.NET.Maths": {
-        "type": "Transitive",
-        "resolved": "2.23.0",
-        "contentHash": "r8PdIVzME8EH0qAgbmRPO87I4GfgR2j8TofT7EMuRJDf1QluoQwnVypDoFJjQ2ZBSRsGYk5unYxxogI05Ogsmw=="
-      },
-      "Silk.NET.Windowing.Common": {
-        "type": "Transitive",
-        "resolved": "2.23.0",
-        "contentHash": "ThStSinmY9KQI8DGiF5XEhkLJVnBcgRTBTzL9ijg1wMZAYuckz7ykrNw04fjRm2Gryh6tCNGbvz2XaY0efeFzg==",
-        "dependencies": {
-          "Silk.NET.Core": "2.23.0",
-          "Silk.NET.Maths": "2.23.0"
-        }
-      },
-      "Silk.NET.Windowing.Glfw": {
-        "type": "Transitive",
-        "resolved": "2.23.0",
-        "contentHash": "aYBudKmENmvLRn9p15HbdvlQTnnXskcDfTfbYwSb/4fr263rGLwYuDw/txUEc2jihHJiWCp5+75Y7z5wTJWl7g==",
-        "dependencies": {
-          "Silk.NET.GLFW": "2.23.0",
-          "Silk.NET.Windowing.Common": "2.23.0"
-        }
-      },
-      "Ultz.Native.GLFW": {
-        "type": "Transitive",
-        "resolved": "3.4.0",
-        "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
-      },
-      "sharpemu.hle": {
-        "type": "Project",
-        "dependencies": {
-          "SharpEmu.Logging": "[1.0.0, )"
-        }
-      },
-      "sharpemu.libs": {
-        "type": "Project",
-        "dependencies": {
-          "SharpEmu.HLE": "[1.0.0, )",
-          "Silk.NET.Vulkan": "[2.23.0, )",
-          "Silk.NET.Vulkan.Extensions.EXT": "[2.23.0, )",
-          "Silk.NET.Vulkan.Extensions.KHR": "[2.23.0, )",
-          "Silk.NET.Windowing": "[2.23.0, )"
-        }
-      },
-      "sharpemu.logging": {
-        "type": "Project"
-      },
-      "Silk.NET.Vulkan": {
-        "type": "CentralTransitive",
-        "requested": "[2.23.0, )",
-        "resolved": "2.23.0",
-        "contentHash": "3/irtlSWXZ3eTi8N6nelI6L34NTB8ZJHpqVMNzZx2aX7Ek9YEQ34NoQW8/Tljrtmkg8KRhHW8hKTEzZaKV8PgA==",
-        "dependencies": {
-          "Silk.NET.Core": "2.23.0"
-        }
-      },
-      "Silk.NET.Vulkan.Extensions.EXT": {
-        "type": "CentralTransitive",
-        "requested": "[2.23.0, )",
-        "resolved": "2.23.0",
-        "contentHash": "+Oth189ksRiL6HvGCwIdnsYHawqrbO8y49u1H61z3wsfcHhQZeVDYe/wF5LD7fk3NcdgDvwFD3mLm1QWhdZySw==",
-        "dependencies": {
-          "Silk.NET.Core": "2.23.0",
-          "Silk.NET.Vulkan": "2.23.0"
-        }
-      },
-      "Silk.NET.Vulkan.Extensions.KHR": {
-        "type": "CentralTransitive",
-        "requested": "[2.23.0, )",
-        "resolved": "2.23.0",
-        "contentHash": "uRaf4j+SmH3DumjSSSUbFg33BnsGZUyXGj93O9NgGKZSJN3OTmNmQDxRew+/KiVLcgH6qzbto8aNGZ++j9GFWg==",
-        "dependencies": {
-          "Silk.NET.Core": "2.23.0",
-          "Silk.NET.Vulkan": "2.23.0"
-        }
-      },
-      "Silk.NET.Windowing": {
-        "type": "CentralTransitive",
-        "requested": "[2.23.0, )",
-        "resolved": "2.23.0",
-        "contentHash": "OPNPmt/lRyUKVYrFLQXVxyATqD3MKLc1iY1oKx1/2GppgmZxVZPwN12tekrQ4C7408kgB1L5JD1Wnirqqeb2kg==",
-        "dependencies": {
-          "Silk.NET.Windowing.Common": "2.23.0",
-          "Silk.NET.Windowing.Glfw": "2.23.0"
-        }
-      }
-    }
-  }
-}

src/SharpEmu.GUI/App.axaml

@@ -29,6 +29,7 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <SolidColorBrush x:Key="DangerBrush" Color="#E5484D" />
     <SolidColorBrush x:Key="DangerHoverBrush" Color="#F2555A" />
     <SolidColorBrush x:Key="SuccessBrush" Color="#46C46B" />
+    <SolidColorBrush x:Key="InfoBrush" Color="#58A6FF" />
     <SolidColorBrush x:Key="TileHoverBrush" Color="#1A2130" />
     <SolidColorBrush x:Key="TileSelectedBrush" Color="#212A3F" />
   </Application.Resources>
@@ -55,6 +56,22 @@ SPDX-License-Identifier: GPL-2.0-or-later
       <Setter Property="Padding" Value="10,3" />
     </Style>
 
+    <!-- Session status/hotkey badges: the title-id pill geometry with a
+         tinted fill so state (RUNNING) and keys (F11) read at a glance. -->
+    <Style Selector="Border.badge">
+      <Setter Property="CornerRadius" Value="999" />
+      <Setter Property="Padding" Value="8,2" />
+      <Setter Property="BorderThickness" Value="1" />
+    </Style>
+    <Style Selector="Border.badge.running">
+      <Setter Property="Background" Value="#1E46C46B" />
+      <Setter Property="BorderBrush" Value="#5546C46B" />
+    </Style>
+    <Style Selector="Border.badge.key">
+      <Setter Property="Background" Value="#1E58A6FF" />
+      <Setter Property="BorderBrush" Value="#5558A6FF" />
+    </Style>
+
     <Style Selector="TextBlock.sectionTitle">
       <Setter Property="FontSize" Value="11" />
       <Setter Property="FontWeight" Value="SemiBold" />

src/SharpEmu.GUI/EmulatorProcess.cs

@@ -1,7 +1,6 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-using System.ComponentModel;
 using System.Diagnostics;
 using System.Runtime.InteropServices;
 using System.Text;
@@ -10,32 +9,36 @@ using Microsoft.Win32.SafeHandles;
 namespace SharpEmu.GUI;
 
 /// <summary>
-/// Launches the SharpEmu CLI as a child process with the same CET/CFG mitigation
-/// opt-outs the CLI would apply to its own relaunched child, while capturing
-/// stdout/stderr through pipes. The CLI's internal relaunch is suppressed via
-/// SHARPEMU_DISABLE_MITIGATION_RELAUNCH so output is not lost to a detached
-/// console. A kill-on-close job object ties the emulator's lifetime to the GUI.
+/// Owns an isolated emulator process. Guest virtual memory is fixed-address and
+/// cannot be reliably reused while guest-created host threads are still alive,
+/// so the GUI must never execute a game in its own process.
 /// </summary>
 internal sealed class EmulatorProcess : IDisposable
 {
-    private const uint EXTENDED_STARTUPINFO_PRESENT = 0x00080000;
-    private const uint CREATE_NO_WINDOW = 0x08000000;
-    private const int STARTF_USESTDHANDLES = 0x00000100;
-    private const uint HANDLE_FLAG_INHERIT = 0x00000001;
-    private const uint INFINITE = 0xFFFFFFFF;
-    private const int PROC_THREAD_ATTRIBUTE_MITIGATION_POLICY = 0x00020007;
-    private const uint JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE = 0x00002000;
-    private const int JobObjectExtendedLimitInformation = 9;
-    private const ulong PROCESS_CREATION_MITIGATION_POLICY_CONTROL_FLOW_GUARD_ALWAYS_OFF = 0x00000002UL << 40;
-    private const ulong PROCESS_CREATION_MITIGATION_POLICY2_CET_USER_SHADOW_STACKS_ALWAYS_OFF = 0x00000002UL << 28;
-    private const ulong PROCESS_CREATION_MITIGATION_POLICY2_USER_CET_SET_CONTEXT_IP_VALIDATION_ALWAYS_OFF = 0x00000002UL << 32;
-    private const ulong PROCESS_CREATION_MITIGATION_POLICY2_XTENDED_CONTROL_FLOW_GUARD_ALWAYS_OFF = 0x00000002UL << 40;
+    public const int HostStopExitCode = -2;
+
+    private const uint ExtendedStartupInfoPresent = 0x00080000;
+    private const uint CreateNoWindow = 0x08000000;
+    private const int StartfUseStdHandles = 0x00000100;
+    private const uint HandleFlagInherit = 0x00000001;
+    private const uint Infinite = 0xFFFFFFFF;
+    private const int ProcThreadAttributeMitigationPolicy = 0x00020007;
+    private const uint JobObjectLimitKillOnJobClose = 0x00002000;
+    private const int JobObjectExtendedLimitInformationClass = 9;
+    private const string MitigatedChildFlag = "--sharpemu-mitigated-child";
+    private const string MitigatedChildEnvironment = "SHARPEMU_MITIGATED_CHILD";
+    private const ulong ControlFlowGuardAlwaysOff = 0x00000002UL << 40;
+    private const ulong CetUserShadowStacksAlwaysOff = 0x00000002UL << 28;
+    private const ulong UserCetSetContextIpValidationAlwaysOff = 0x00000002UL << 32;
+
+    private static readonly object EnvironmentGate = new();
 
     private readonly object _sync = new();
     private nint _processHandle;
     private nint _jobHandle;
     private Process? _fallbackProcess;
     private bool _running;
+    private bool _stopRequested;
     private bool _disposed;
 
     public event Action<string, bool>? OutputReceived;
@@ -55,29 +58,34 @@ internal sealed class EmulatorProcess : IDisposable
 
     public void Start(string exePath, IReadOnlyList<string> arguments, string? workingDirectory)
     {
+        ArgumentException.ThrowIfNullOrWhiteSpace(exePath);
+        ArgumentNullException.ThrowIfNull(arguments);
+
         lock (_sync)
         {
-            ObjectDisposedException.ThrowIf(_disposed, this);
+            ThrowIfDisposed();
             if (_running)
             {
                 throw new InvalidOperationException("The emulator process is already running.");
             }
 
-            if (OperatingSystem.IsWindows())
-            {
-                StartWindows(exePath, arguments, workingDirectory);
-            }
-            else
-            {
-                StartFallback(exePath, arguments, workingDirectory);
-            }
-
-            _running = true;
+            _stopRequested = false;
         }
+
+        if (OperatingSystem.IsWindows())
+        {
+            StartWindows(exePath, arguments, workingDirectory);
+            return;
+        }
+
+        StartFallback(exePath, arguments, workingDirectory);
     }
 
     public void Stop()
     {
+        nint processHandle;
+        nint jobHandle;
+        Process? fallbackProcess;
         lock (_sync)
         {
             if (!_running)
@@ -85,28 +93,35 @@ internal sealed class EmulatorProcess : IDisposable
                 return;
             }
 
-            // Prefer terminating the job: it kills the whole tree, including
-            // any children the emulator spawned, even when the main process
-            // is wedged in a GPU driver call.
-            if (_jobHandle != 0)
-            {
-                _ = TerminateJobObject(_jobHandle, 1);
-            }
+            _stopRequested = true;
+            processHandle = _processHandle;
+            jobHandle = _jobHandle;
+            fallbackProcess = _fallbackProcess;
+        }
 
-            if (_processHandle != 0)
-            {
-                _ = TerminateProcess(_processHandle, 1);
-            }
+        if (jobHandle != 0)
+        {
+            _ = TerminateJobObject(jobHandle, unchecked((uint)HostStopExitCode));
+            return;
+        }
 
-            try
+        if (processHandle != 0)
+        {
+            _ = TerminateProcess(processHandle, unchecked((uint)HostStopExitCode));
+            return;
+        }
+
+        try
+        {
+            if (fallbackProcess is { HasExited: false })
             {
-                _fallbackProcess?.Kill(entireProcessTree: true);
-            }
-            catch (InvalidOperationException)
-            {
-                // Already exited.
+                fallbackProcess.Kill(entireProcessTree: true);
             }
         }
+        catch (InvalidOperationException)
+        {
+            // The process exited while Stop was handling the request.
+        }
     }
 
     public void Dispose()
@@ -124,121 +139,206 @@ internal sealed class EmulatorProcess : IDisposable
         Stop();
     }
 
-    private void StartWindows(string exePath, IReadOnlyList<string> arguments, string? workingDirectory)
+    private void StartFallback(string exePath, IReadOnlyList<string> arguments, string? workingDirectory)
     {
-        // The CLI would otherwise relaunch itself into a mitigated child whose
-        // console output cannot flow through our pipes.
-        Environment.SetEnvironmentVariable("SHARPEMU_DISABLE_MITIGATION_RELAUNCH", "1");
-
-        var securityAttributes = new SECURITY_ATTRIBUTES
+        var startInfo = new ProcessStartInfo(exePath)
         {
-            nLength = Marshal.SizeOf<SECURITY_ATTRIBUTES>(),
-            bInheritHandle = 1,
+            WorkingDirectory = string.IsNullOrWhiteSpace(workingDirectory)
+                ? Path.GetDirectoryName(exePath) ?? Environment.CurrentDirectory
+                : workingDirectory,
+            UseShellExecute = false,
+            CreateNoWindow = true,
+            RedirectStandardOutput = true,
+            RedirectStandardError = true,
+            StandardOutputEncoding = Encoding.UTF8,
+            StandardErrorEncoding = Encoding.UTF8,
         };
-
-        if (!CreatePipe(out var stdoutRead, out var stdoutWrite, ref securityAttributes, 0) ||
-            !CreatePipe(out var stderrRead, out var stderrWrite, ref securityAttributes, 0))
+        foreach (var argument in arguments)
         {
-            throw new Win32Exception(Marshal.GetLastWin32Error(), "Failed to create output pipes.");
+            startInfo.ArgumentList.Add(argument);
         }
 
-        _ = SetHandleInformation(stdoutRead, HANDLE_FLAG_INHERIT, 0);
-        _ = SetHandleInformation(stderrRead, HANDLE_FLAG_INHERIT, 0);
+        var process = Process.Start(startInfo)
+            ?? throw new InvalidOperationException("Could not start the emulator process.");
+        process.EnableRaisingEvents = true;
+        process.OutputDataReceived += (_, eventArgs) => ForwardOutput(eventArgs.Data, isError: false);
+        process.ErrorDataReceived += (_, eventArgs) => ForwardOutput(eventArgs.Data, isError: true);
+        process.Exited += (_, _) => OnExited(process.ExitCode);
 
-        var startupInfoEx = new STARTUPINFOEX();
-        startupInfoEx.StartupInfo.cb = Marshal.SizeOf<STARTUPINFOEX>();
-        startupInfoEx.StartupInfo.dwFlags = STARTF_USESTDHANDLES;
-        startupInfoEx.StartupInfo.hStdOutput = stdoutWrite;
-        startupInfoEx.StartupInfo.hStdError = stderrWrite;
+        lock (_sync)
+        {
+            _fallbackProcess = process;
+            _running = true;
+        }
 
+        process.BeginOutputReadLine();
+        process.BeginErrorReadLine();
+    }
+
+    private unsafe void StartWindows(string exePath, IReadOnlyList<string> arguments, string? workingDirectory)
+    {
+        nint stdoutRead = 0;
+        nint stdoutWrite = 0;
+        nint stderrRead = 0;
+        nint stderrWrite = 0;
         nint attributeList = 0;
         nint mitigationPolicies = 0;
+        nint processHandle = 0;
+        nint threadHandle = 0;
+
         try
         {
+            var security = new SecurityAttributes
+            {
+                Size = Marshal.SizeOf<SecurityAttributes>(),
+                InheritHandle = 1,
+            };
+            if (!CreatePipe(out stdoutRead, out stdoutWrite, ref security, 0) ||
+                !CreatePipe(out stderrRead, out stderrWrite, ref security, 0))
+            {
+                throw new InvalidOperationException($"Could not create emulator output pipes (Win32 error {Marshal.GetLastWin32Error()}).");
+            }
+
+            if (!SetHandleInformation(stdoutRead, HandleFlagInherit, 0) ||
+                !SetHandleInformation(stderrRead, HandleFlagInherit, 0))
+            {
+                throw new InvalidOperationException($"Could not configure emulator output pipes (Win32 error {Marshal.GetLastWin32Error()}).");
+            }
+
             nuint attributeListSize = 0;
             _ = InitializeProcThreadAttributeList(0, 1, 0, ref attributeListSize);
             attributeList = Marshal.AllocHGlobal((nint)attributeListSize);
             if (!InitializeProcThreadAttributeList(attributeList, 1, 0, ref attributeListSize))
             {
-                throw new Win32Exception(Marshal.GetLastWin32Error(), "Failed to initialize the process attribute list.");
+                throw new InvalidOperationException($"Could not initialize process mitigation attributes (Win32 error {Marshal.GetLastWin32Error()}).");
             }
 
-            startupInfoEx.lpAttributeList = attributeList;
-
-            var policy1 = PROCESS_CREATION_MITIGATION_POLICY_CONTROL_FLOW_GUARD_ALWAYS_OFF;
-            var policy2 =
-                PROCESS_CREATION_MITIGATION_POLICY2_CET_USER_SHADOW_STACKS_ALWAYS_OFF |
-                PROCESS_CREATION_MITIGATION_POLICY2_USER_CET_SET_CONTEXT_IP_VALIDATION_ALWAYS_OFF;
-
             mitigationPolicies = Marshal.AllocHGlobal(sizeof(ulong) * 2);
-            Marshal.WriteInt64(mitigationPolicies, unchecked((long)policy1));
-            Marshal.WriteInt64(nint.Add(mitigationPolicies, sizeof(long)), unchecked((long)policy2));
-
+            Marshal.WriteInt64(mitigationPolicies, unchecked((long)ControlFlowGuardAlwaysOff));
+            Marshal.WriteInt64(
+                nint.Add(mitigationPolicies, sizeof(long)),
+                unchecked((long)(CetUserShadowStacksAlwaysOff | UserCetSetContextIpValidationAlwaysOff)));
             if (!UpdateProcThreadAttribute(
-                attributeList,
-                0,
-                PROC_THREAD_ATTRIBUTE_MITIGATION_POLICY,
-                mitigationPolicies,
-                (nuint)(sizeof(ulong) * 2),
-                0,
-                0))
+                    attributeList,
+                    0,
+                    (nint)ProcThreadAttributeMitigationPolicy,
+                    mitigationPolicies,
+                    (nuint)(sizeof(ulong) * 2),
+                    0,
+                    0))
             {
-                throw new Win32Exception(Marshal.GetLastWin32Error(), "Failed to apply the mitigation policy.");
+                throw new InvalidOperationException($"Could not apply process mitigations (Win32 error {Marshal.GetLastWin32Error()}).");
             }
 
-            var currentDirectory = workingDirectory ?? Environment.CurrentDirectory;
-            var created = CreateProcessW(
-                exePath,
-                new StringBuilder(BuildCommandLine(exePath, arguments)),
-                0,
-                0,
-                true,
-                EXTENDED_STARTUPINFO_PRESENT | CREATE_NO_WINDOW,
-                0,
-                currentDirectory,
-                ref startupInfoEx,
-                out var processInfo);
+            var startup = new StartupInfoEx();
+            startup.StartupInfo.Size = Marshal.SizeOf<StartupInfoEx>();
+            startup.StartupInfo.Flags = StartfUseStdHandles;
+            startup.StartupInfo.StdOutput = stdoutWrite;
+            startup.StartupInfo.StdError = stderrWrite;
+            startup.AttributeList = attributeList;
 
-            if (!created)
+            var childArguments = new List<string>(arguments.Count + 1) { MitigatedChildFlag };
+            childArguments.AddRange(arguments);
+            var commandLine = new StringBuilder(BuildCommandLine(exePath, childArguments));
+            ProcessInformation processInfo;
+            lock (EnvironmentGate)
             {
-                var error = Marshal.GetLastWin32Error();
-                throw new Win32Exception(error, $"Failed to start '{exePath}' with CET/CFG mitigation disabled (Win32 error {error}: {new Win32Exception(error).Message}).");
+                var previousValue = Environment.GetEnvironmentVariable(MitigatedChildEnvironment);
+                try
+                {
+                    Environment.SetEnvironmentVariable(MitigatedChildEnvironment, "1");
+                    if (!CreateProcessW(
+                            exePath,
+                            commandLine,
+                            0,
+                            0,
+                            true,
+                            ExtendedStartupInfoPresent | CreateNoWindow,
+                            0,
+                            string.IsNullOrWhiteSpace(workingDirectory)
+                                ? Path.GetDirectoryName(exePath) ?? Environment.CurrentDirectory
+                                : workingDirectory,
+                            ref startup,
+                            out processInfo))
+                    {
+                        throw new InvalidOperationException($"Could not start the emulator process (Win32 error {Marshal.GetLastWin32Error()}).");
+                    }
+                }
+                finally
+                {
+                    Environment.SetEnvironmentVariable(MitigatedChildEnvironment, previousValue);
+                }
             }
 
-            CloseHandle(processInfo.hThread);
-            _processHandle = processInfo.hProcess;
+            processHandle = processInfo.Process;
+            threadHandle = processInfo.Thread;
+            CloseHandle(stdoutWrite);
+            stdoutWrite = 0;
+            CloseHandle(stderrWrite);
+            stderrWrite = 0;
 
-            _jobHandle = CreateJobObjectW(0, null);
-            if (_jobHandle != 0 &&
-                (!TryEnableKillOnJobClose(_jobHandle) || !AssignProcessToJobObject(_jobHandle, processInfo.hProcess)))
+            var jobHandle = CreateJobObjectW(0, null);
+            if (jobHandle != 0 &&
+                (!TryEnableKillOnJobClose(jobHandle) || !AssignProcessToJobObject(jobHandle, processHandle)))
             {
-                CloseHandle(_jobHandle);
-                _jobHandle = 0;
+                CloseHandle(jobHandle);
+                jobHandle = 0;
             }
 
-            StartReaderThread(stdoutRead, isError: false);
-            StartReaderThread(stderrRead, isError: true);
-            StartExitWatcherThread();
+            lock (_sync)
+            {
+                _processHandle = processHandle;
+                _jobHandle = jobHandle;
+                _running = true;
+            }
+            processHandle = 0;
+
+            StartPipeReader(stdoutRead, isError: false);
+            stdoutRead = 0;
+            StartPipeReader(stderrRead, isError: true);
+            stderrRead = 0;
+            StartWindowsExitWatcher(_processHandle);
         }
         catch
         {
-            CloseHandle(stdoutRead);
-            CloseHandle(stderrRead);
+            if (processHandle != 0)
+            {
+                _ = TerminateProcess(processHandle, 1);
+            }
+
             throw;
         }
         finally
         {
-            // The child owns duplicated pipe write ends; closing ours lets the
-            // readers observe EOF when the child exits.
-            CloseHandle(stdoutWrite);
-            CloseHandle(stderrWrite);
-
+            if (threadHandle != 0)
+            {
+                CloseHandle(threadHandle);
+            }
+            if (processHandle != 0)
+            {
+                CloseHandle(processHandle);
+            }
+            if (stdoutRead != 0)
+            {
+                CloseHandle(stdoutRead);
+            }
+            if (stdoutWrite != 0)
+            {
+                CloseHandle(stdoutWrite);
+            }
+            if (stderrRead != 0)
+            {
+                CloseHandle(stderrRead);
+            }
+            if (stderrWrite != 0)
+            {
+                CloseHandle(stderrWrite);
+            }
             if (attributeList != 0)
             {
                 DeleteProcThreadAttributeList(attributeList);
                 Marshal.FreeHGlobal(attributeList);
             }
-
             if (mitigationPolicies != 0)
             {
                 Marshal.FreeHGlobal(mitigationPolicies);
@@ -246,103 +346,56 @@ internal sealed class EmulatorProcess : IDisposable
         }
     }
 
-    private void StartFallback(string exePath, IReadOnlyList<string> arguments, string? workingDirectory)
+    private void StartPipeReader(nint handle, bool isError)
     {
-        var startInfo = new ProcessStartInfo
+        var readerThread = new Thread(() =>
         {
-            FileName = exePath,
-            WorkingDirectory = workingDirectory ?? Environment.CurrentDirectory,
-            UseShellExecute = false,
-            CreateNoWindow = true,
-            RedirectStandardOutput = true,
-            RedirectStandardError = true,
-        };
-        foreach (var argument in arguments)
-        {
-            startInfo.ArgumentList.Add(argument);
-        }
-
-        var process = new Process { StartInfo = startInfo, EnableRaisingEvents = true };
-        process.OutputDataReceived += (_, e) =>
-        {
-            if (e.Data is not null)
+            using var safeHandle = new SafeFileHandle(handle, ownsHandle: true);
+            using var stream = new FileStream(safeHandle, FileAccess.Read, 4096, isAsync: false);
+            using var reader = new StreamReader(stream, Encoding.UTF8, detectEncodingFromByteOrderMarks: true);
+            while (reader.ReadLine() is { } line)
             {
-                OutputReceived?.Invoke(e.Data, false);
+                ForwardOutput(line, isError);
             }
-        };
-        process.ErrorDataReceived += (_, e) =>
-        {
-            if (e.Data is not null)
-            {
-                OutputReceived?.Invoke(e.Data, true);
-            }
-        };
-        process.Exited += (_, _) =>
-        {
-            int exitCode;
-            try
-            {
-                exitCode = process.ExitCode;
-            }
-            catch (InvalidOperationException)
-            {
-                exitCode = -1;
-            }
-
-            OnExited(exitCode);
-        };
-
-        process.Start();
-        process.BeginOutputReadLine();
-        process.BeginErrorReadLine();
-        _fallbackProcess = process;
-    }
-
-    private void StartReaderThread(nint readHandle, bool isError)
-    {
-        var thread = new Thread(() =>
-        {
-            using var stream = new FileStream(new SafeFileHandle(readHandle, ownsHandle: true), FileAccess.Read);
-            using var reader = new StreamReader(stream, Encoding.UTF8);
-            try
-            {
-                while (reader.ReadLine() is { } line)
-                {
-                    OutputReceived?.Invoke(line, isError);
-                }
-            }
-            catch (IOException)
-            {
-                // Pipe broken on process teardown.
-            }
-        })
+        }, 256 * 1024)
         {
             IsBackground = true,
             Name = isError ? "SharpEmu stderr reader" : "SharpEmu stdout reader",
         };
-        thread.Start();
+        readerThread.Start();
     }
 
-    private void StartExitWatcherThread()
+    private void StartWindowsExitWatcher(nint processHandle)
     {
-        var processHandle = _processHandle;
-        var thread = new Thread(() =>
+        var watcher = new Thread(() =>
         {
-            _ = WaitForSingleObject(processHandle, INFINITE);
-            var exitCode = GetExitCodeProcess(processHandle, out var rawExitCode)
-                ? unchecked((int)rawExitCode)
-                : -1;
+            _ = WaitForSingleObject(processHandle, Infinite);
+            var exitCode = 1;
+            if (GetExitCodeProcess(processHandle, out var nativeExitCode))
+            {
+                exitCode = unchecked((int)nativeExitCode);
+            }
+
             OnExited(exitCode);
-        })
+        }, 128 * 1024)
         {
             IsBackground = true,
             Name = "SharpEmu exit watcher",
         };
-        thread.Start();
+        watcher.Start();
     }
 
-    private void OnExited(int exitCode)
+    private void ForwardOutput(string? line, bool isError)
     {
+        if (!string.IsNullOrEmpty(line))
+        {
+            OutputReceived?.Invoke(line, isError);
+        }
+    }
+
+    private void OnExited(int nativeExitCode)
+    {
+        int exitCode;
         lock (_sync)
         {
             if (!_running)
@@ -350,13 +403,13 @@ internal sealed class EmulatorProcess : IDisposable
                 return;
             }
 
+            exitCode = _stopRequested ? HostStopExitCode : nativeExitCode;
             _running = false;
             if (_processHandle != 0)
             {
                 CloseHandle(_processHandle);
                 _processHandle = 0;
             }
-
             if (_jobHandle != 0)
             {
                 CloseHandle(_jobHandle);
@@ -372,24 +425,19 @@ internal sealed class EmulatorProcess : IDisposable
 
     private static bool TryEnableKillOnJobClose(nint jobHandle)
     {
-        var extendedLimitInfo = new JOBOBJECT_EXTENDED_LIMIT_INFORMATION
+        var info = new JobObjectExtendedLimitInformation
         {
-            BasicLimitInformation = new JOBOBJECT_BASIC_LIMIT_INFORMATION
+            BasicLimitInformation = new JobObjectBasicLimitInformation
             {
-                LimitFlags = JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE,
+                LimitFlags = JobObjectLimitKillOnJobClose,
             },
         };
-
-        var size = Marshal.SizeOf<JOBOBJECT_EXTENDED_LIMIT_INFORMATION>();
+        var size = Marshal.SizeOf<JobObjectExtendedLimitInformation>();
         var memory = Marshal.AllocHGlobal(size);
         try
         {
-            Marshal.StructureToPtr(extendedLimitInfo, memory, false);
-            return SetInformationJobObject(
-                jobHandle,
-                JobObjectExtendedLimitInformation,
-                memory,
-                unchecked((uint)size));
+            Marshal.StructureToPtr(info, memory, false);
+            return SetInformationJobObject(jobHandle, JobObjectExtendedLimitInformationClass, memory, unchecked((uint)size));
         }
         finally
         {
@@ -397,128 +445,124 @@ internal sealed class EmulatorProcess : IDisposable
         }
     }
 
-    private static string BuildCommandLine(string processPath, IReadOnlyList<string> args)
+    private static string BuildCommandLine(string processPath, IReadOnlyList<string> arguments)
     {
-        var builder = new StringBuilder();
-        builder.Append(QuoteArgument(processPath));
-        for (var i = 0; i < args.Count; i++)
+        var builder = new StringBuilder(QuoteArgument(processPath));
+        foreach (var argument in arguments)
         {
             builder.Append(' ');
-            builder.Append(QuoteArgument(args[i]));
+            builder.Append(QuoteArgument(argument));
         }
 
         return builder.ToString();
     }
 
-    private static string QuoteArgument(string argument)
+    private static string QuoteArgument(string value)
     {
-        if (argument.Length == 0)
+        if (value.Length == 0)
         {
             return "\"\"";
         }
 
-        var needsQuotes = false;
-        foreach (var c in argument)
+        if (!value.Any(static c => char.IsWhiteSpace(c) || c == '"'))
         {
-            if (char.IsWhiteSpace(c) || c == '"')
-            {
-                needsQuotes = true;
-                break;
-            }
+            return value;
         }
 
-        if (!needsQuotes)
-        {
-            return argument;
-        }
-
-        var builder = new StringBuilder(argument.Length + 2);
+        var builder = new StringBuilder(value.Length + 2);
         builder.Append('"');
-
-        var backslashCount = 0;
-        foreach (var c in argument)
+        var slashCount = 0;
+        foreach (var character in value)
         {
-            if (c == '\\')
+            if (character == '\\')
             {
-                backslashCount++;
+                slashCount++;
                 continue;
             }
 
-            if (c == '"')
+            if (character == '"')
             {
-                builder.Append('\\', (backslashCount * 2) + 1);
-                builder.Append('"');
-                backslashCount = 0;
+                builder.Append('\\', (slashCount * 2) + 1);
+                builder.Append(character);
+                slashCount = 0;
                 continue;
             }
 
-            if (backslashCount > 0)
+            if (slashCount > 0)
             {
-                builder.Append('\\', backslashCount);
-                backslashCount = 0;
+                builder.Append('\\', slashCount);
+                slashCount = 0;
             }
 
-            builder.Append(c);
+            builder.Append(character);
         }
 
-        if (backslashCount > 0)
+        if (slashCount > 0)
         {
-            builder.Append('\\', backslashCount * 2);
+            builder.Append('\\', slashCount * 2);
         }
 
         builder.Append('"');
         return builder.ToString();
     }
 
-    [StructLayout(LayoutKind.Sequential)]
-    private struct SECURITY_ATTRIBUTES
+    private void ThrowIfDisposed()
     {
-        public int nLength;
-        public nint lpSecurityDescriptor;
-        public int bInheritHandle;
+        if (_disposed)
+        {
+            throw new ObjectDisposedException(nameof(EmulatorProcess));
+        }
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    private struct SecurityAttributes
+    {
+        public int Size;
+        public nint SecurityDescriptor;
+        public int InheritHandle;
     }
 
     [StructLayout(LayoutKind.Sequential, CharSet = CharSet.Unicode)]
-    private struct STARTUPINFO
+    private struct StartupInfo
     {
-        public int cb;
-        public nint lpReserved;
-        public nint lpDesktop;
-        public nint lpTitle;
-        public int dwX;
-        public int dwY;
-        public int dwXSize;
-        public int dwYSize;
-        public int dwXCountChars;
-        public int dwYCountChars;
-        public int dwFillAttribute;
-        public int dwFlags;
-        public short wShowWindow;
-        public short cbReserved2;
-        public nint lpReserved2;
-        public nint hStdInput;
-        public nint hStdOutput;
-        public nint hStdError;
+        public int Size;
+        public nint Reserved;
+        public nint Desktop;
+        public nint Title;
+        public int X;
+        public int Y;
+        public int XSize;
+        public int YSize;
+        public int XCountChars;
+        public int YCountChars;
+        public int FillAttribute;
+        public int Flags;
+        public short ShowWindow;
+        public short Reserved2Count;
+        public nint Reserved2;
+        public nint StdInput;
+        public nint StdOutput;
+        public nint StdError;
     }
 
     [StructLayout(LayoutKind.Sequential)]
-    private struct STARTUPINFOEX
+    private struct StartupInfoEx
     {
-        public STARTUPINFO StartupInfo;
-        public nint lpAttributeList;
+        public StartupInfo StartupInfo;
+        public nint AttributeList;
     }
 
     [StructLayout(LayoutKind.Sequential)]
-    private struct PROCESS_INFORMATION
+    private struct ProcessInformation
     {
-        public nint hProcess;
-        public nint hThread;
-        public int dwProcessId;
-        public int dwThreadId;
+        public nint Process;
+        public nint Thread;
+        public int ProcessId;
+        public int ThreadId;
     }
 
     [StructLayout(LayoutKind.Sequential)]
-    private struct JOBOBJECT_BASIC_LIMIT_INFORMATION
+    private struct JobObjectBasicLimitInformation
     {
         public long PerProcessUserTimeLimit;
         public long PerJobUserTimeLimit;
@@ -532,7 +576,7 @@ internal sealed class EmulatorProcess : IDisposable
     }
 
     [StructLayout(LayoutKind.Sequential)]
-    private struct IO_COUNTERS
+    private struct IoCounters
     {
         public ulong ReadOperationCount;
         public ulong WriteOperationCount;
@@ -543,10 +587,10 @@ internal sealed class EmulatorProcess : IDisposable
     }
 
     [StructLayout(LayoutKind.Sequential)]
-    private struct JOBOBJECT_EXTENDED_LIMIT_INFORMATION
+    private struct JobObjectExtendedLimitInformation
     {
-        public JOBOBJECT_BASIC_LIMIT_INFORMATION BasicLimitInformation;
-        public IO_COUNTERS IoInfo;
+        public JobObjectBasicLimitInformation BasicLimitInformation;
+        public IoCounters IoInfo;
         public nuint ProcessMemoryLimit;
         public nuint JobMemoryLimit;
         public nuint PeakProcessMemoryUsed;
@@ -555,66 +599,37 @@ internal sealed class EmulatorProcess : IDisposable
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool CreatePipe(
-        out nint hReadPipe,
-        out nint hWritePipe,
-        ref SECURITY_ATTRIBUTES lpPipeAttributes,
-        uint nSize);
+    private static extern bool CreatePipe(out nint readPipe, out nint writePipe, ref SecurityAttributes attributes, uint size);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool SetHandleInformation(nint hObject, uint dwMask, uint dwFlags);
+    private static extern bool SetHandleInformation(nint handle, uint mask, uint flags);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool InitializeProcThreadAttributeList(
-        nint lpAttributeList,
-        int dwAttributeCount,
-        int dwFlags,
-        ref nuint lpSize);
+    private static extern bool InitializeProcThreadAttributeList(nint list, int count, int flags, ref nuint size);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool UpdateProcThreadAttribute(
-        nint lpAttributeList,
-        uint dwFlags,
-        nint attribute,
-        nint lpValue,
-        nuint cbSize,
-        nint lpPreviousValue,
-        nint lpReturnSize);
+    private static extern bool UpdateProcThreadAttribute(nint list, uint flags, nint attribute, nint value, nuint size, nint previousValue, nint returnSize);
 
     [DllImport("kernel32.dll")]
-    private static extern void DeleteProcThreadAttributeList(nint lpAttributeList);
+    private static extern void DeleteProcThreadAttributeList(nint list);
 
     [DllImport("kernel32.dll", EntryPoint = "CreateJobObjectW", SetLastError = true, CharSet = CharSet.Unicode)]
-    private static extern nint CreateJobObjectW(nint lpJobAttributes, string? lpName);
+    private static extern nint CreateJobObjectW(nint attributes, string? name);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool SetInformationJobObject(
-        nint hJob,
-        int jobObjectInfoClass,
-        nint lpJobObjectInfo,
-        uint cbJobObjectInfoLength);
+    private static extern bool SetInformationJobObject(nint job, int infoClass, nint info, uint size);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool AssignProcessToJobObject(nint hJob, nint hProcess);
+    private static extern bool AssignProcessToJobObject(nint job, nint process);
 
     [DllImport("kernel32.dll", EntryPoint = "CreateProcessW", SetLastError = true, CharSet = CharSet.Unicode)]
     [return: MarshalAs(UnmanagedType.Bool)]
-    private static extern bool CreateProcessW(
-        string applicationName,
-        StringBuilder commandLine,
-        nint processAttributes,
-        nint threadAttributes,
-        [MarshalAs(UnmanagedType.Bool)] bool inheritHandles,
-        uint creationFlags,
-        nint environment,
-        string currentDirectory,
-        ref STARTUPINFOEX startupInfo,
-        out PROCESS_INFORMATION processInformation);
+    private static extern bool CreateProcessW(string applicationName, StringBuilder commandLine, nint processAttributes, nint threadAttributes, [MarshalAs(UnmanagedType.Bool)] bool inheritHandles, uint flags, nint environment, string currentDirectory, ref StartupInfoEx startupInfo, out ProcessInformation processInformation);
 
     [DllImport("kernel32.dll", SetLastError = true)]
     private static extern uint WaitForSingleObject(nint handle, uint milliseconds);

src/SharpEmu.GUI/GameSurfaceHost.cs

@@ -0,0 +1,613 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using Avalonia;
+using Avalonia.Controls;
+using Avalonia.Platform;
+using Avalonia.Threading;
+using SharpEmu.Libs.VideoOut;
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.GUI;
+
+/// <summary>
+/// Native child surface owned by Avalonia. The isolated emulator process uses
+/// its platform handle to create the Vulkan presentation surface, keeping the
+/// guest address space out of the GUI process.
+/// </summary>
+public sealed class GameSurfaceHost : NativeControlHost
+{
+    private const uint SwpNoSize = 0x0001;
+    private const uint SwpNoMove = 0x0002;
+    private const uint SwpNoZOrder = 0x0004;
+    private const uint SwpNoActivate = 0x0010;
+    private const uint SwpShowWindow = 0x0040;
+    private const uint SwpHideWindow = 0x0080;
+    private const uint WsChild = 0x40000000;
+    private const uint WsVisible = 0x10000000;
+    private const uint WsClipSiblings = 0x04000000;
+    private const uint WsClipChildren = 0x02000000;
+    private const uint CsOwnDc = 0x0020;
+    private const uint WmSetCursor = 0x0020;
+    private const uint WmMouseMove = 0x0200;
+    private const int IdcArrow = 32512;
+    private const int CursorHideDelayMs = 2500;
+
+    private VulkanHostSurface? _surface;
+    private nint _windowHandle;
+    private nint _x11Display;
+    private string? _win32ClassName;
+    private WindowProcedure? _windowProcedure;
+    private nint _metalLayer;
+    private bool _presentationVisible = true;
+    private DispatcherTimer? _cursorIdleTimer;
+    private bool _cursorAutoHide;
+    private bool _cursorHidden;
+    private long _lastPointerActivity;
+
+    public GameSurfaceHost()
+    {
+        PropertyChanged += (_, change) =>
+        {
+            if (change.Property == BoundsProperty)
+            {
+                UpdateSurfaceSize();
+            }
+        };
+        LayoutUpdated += (_, _) =>
+        {
+            // Fullscreen can change a monitor's DPI scale without changing
+            // the logical Bounds. Refresh the native child from physical size.
+            UpdateSurfaceSize();
+
+            // NativeControlHost may make its HWND visible again as part of a
+            // later arrange pass. Keep a loading surface hidden until its
+            // child process reports a real first frame.
+            if (!_presentationVisible)
+            {
+                ApplyPresentationVisibility();
+            }
+        };
+    }
+
+    public event EventHandler<VulkanHostSurface>? SurfaceAvailable;
+
+    public event EventHandler<VulkanHostSurface>? SurfaceDestroyed;
+
+    public VulkanHostSurface? Surface => _surface;
+
+    public void RefreshSurfaceSize() => UpdateSurfaceSize();
+
+    /// <summary>
+    /// Hides the platform child without detaching the Vulkan surface. This
+    /// allows the launcher to return to its library while guest teardown is
+    /// still finishing on the render thread.
+    /// </summary>
+    public void SetPresentationVisible(bool visible)
+    {
+        _presentationVisible = visible;
+        ApplyPresentationVisibility();
+    }
+
+    /// <summary>
+    /// Auto-hides the mouse cursor over the game surface after a short idle
+    /// period; any pointer movement brings it back. Enabling (again) restarts
+    /// the idle countdown, so both "first frame presented" and "entered
+    /// fullscreen" can arm it. Windows-only; a no-op elsewhere.
+    /// </summary>
+    public void SetCursorAutoHide(bool enabled)
+    {
+        if (!OperatingSystem.IsWindows())
+        {
+            return;
+        }
+
+        _cursorAutoHide = enabled;
+        _lastPointerActivity = System.Diagnostics.Stopwatch.GetTimestamp();
+        if (enabled)
+        {
+            _cursorIdleTimer ??= CreateCursorIdleTimer();
+            _cursorIdleTimer.Start();
+            return;
+        }
+
+        _cursorIdleTimer?.Stop();
+        ShowCursorNow();
+    }
+
+    private DispatcherTimer CreateCursorIdleTimer()
+    {
+        var timer = new DispatcherTimer
+        {
+            Interval = TimeSpan.FromMilliseconds(250),
+        };
+        timer.Tick += (_, _) => HideCursorWhenIdle();
+        return timer;
+    }
+
+    private void HideCursorWhenIdle()
+    {
+        if (!_cursorAutoHide || _cursorHidden || _windowHandle == 0)
+        {
+            return;
+        }
+
+        var idleMs = (System.Diagnostics.Stopwatch.GetTimestamp() - _lastPointerActivity) *
+            1000 / System.Diagnostics.Stopwatch.Frequency;
+        if (idleMs < CursorHideDelayMs)
+        {
+            return;
+        }
+
+        // Only swallow the cursor while it is actually over the game surface;
+        // hovering launcher chrome (console, toolbar) must keep the arrow.
+        if (!GetCursorPos(out var point) || WindowFromPoint(point) != _windowHandle)
+        {
+            return;
+        }
+
+        _cursorHidden = true;
+        _ = SetCursor(0);
+    }
+
+    private void ShowCursorNow()
+    {
+        if (!_cursorHidden)
+        {
+            return;
+        }
+
+        _cursorHidden = false;
+        _ = SetCursor(LoadCursorW(0, IdcArrow));
+    }
+
+    private void ApplyPresentationVisibility()
+    {
+        if (_windowHandle == 0)
+        {
+            return;
+        }
+
+        var visible = _presentationVisible;
+        if (OperatingSystem.IsWindows())
+        {
+            // SW_HIDE can be ignored for a window's initial show state. Force
+            // the state through SetWindowPos so an old child swapchain cannot
+            // remain composed while the next game is loading.
+            var flags = SwpNoSize | SwpNoMove | SwpNoZOrder | SwpNoActivate |
+                (visible ? SwpShowWindow : SwpHideWindow);
+            _ = SetWindowPos(_windowHandle, 0, 0, 0, 0, 0, flags);
+        }
+        else if (OperatingSystem.IsLinux() && _x11Display != 0)
+        {
+            _ = visible
+                ? XMapWindow(_x11Display, _windowHandle)
+                : XUnmapWindow(_x11Display, _windowHandle);
+            _ = XFlush(_x11Display);
+        }
+        else if (OperatingSystem.IsMacOS())
+        {
+            SendBool(_windowHandle, "setHidden:", !visible);
+        }
+    }
+
+    protected override IPlatformHandle CreateNativeControlCore(IPlatformHandle control)
+    {
+        PlatformHandle handle;
+        if (OperatingSystem.IsWindows())
+        {
+            handle = CreateWin32(control);
+        }
+        else if (OperatingSystem.IsLinux())
+        {
+            handle = CreateX11(control);
+        }
+        else if (OperatingSystem.IsMacOS())
+        {
+            handle = CreateMacOS();
+        }
+        else
+        {
+            throw new PlatformNotSupportedException("SharpEmu's embedded Vulkan surface is unsupported on this platform.");
+        }
+
+        UpdateSurfaceSize();
+        if (_surface is { } surface)
+        {
+            SurfaceAvailable?.Invoke(this, surface);
+        }
+
+        return handle;
+    }
+
+    protected override void DestroyNativeControlCore(IPlatformHandle control)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            SetCursorAutoHide(false);
+        }
+
+        var surface = _surface;
+        _surface = null;
+
+        if (OperatingSystem.IsWindows())
+        {
+            DestroyWin32();
+        }
+        else if (OperatingSystem.IsLinux())
+        {
+            DestroyX11();
+        }
+        else if (OperatingSystem.IsMacOS())
+        {
+            DestroyMacOS();
+        }
+
+        if (surface is not null)
+        {
+            SurfaceDestroyed?.Invoke(this, surface);
+        }
+    }
+
+    private PlatformHandle CreateWin32(IPlatformHandle control)
+    {
+        _win32ClassName = $"SharpEmuGameSurface-{Guid.NewGuid():N}";
+        _windowProcedure = WindowProcedureImpl;
+        var classInfo = new WndClassEx
+        {
+            Size = (uint)Marshal.SizeOf<WndClassEx>(),
+            Style = CsOwnDc,
+            WindowProcedure = Marshal.GetFunctionPointerForDelegate(_windowProcedure),
+            Instance = GetModuleHandleW(null),
+            ClassName = _win32ClassName,
+        };
+
+        if (RegisterClassExW(ref classInfo) == 0)
+        {
+            throw new InvalidOperationException($"Could not register the embedded game window class (Win32 error {Marshal.GetLastWin32Error()}).");
+        }
+
+        _windowHandle = CreateWindowExW(
+            0,
+            _win32ClassName,
+            "SharpEmu Game Surface",
+            WsChild | (_presentationVisible ? WsVisible : 0) | WsClipSiblings | WsClipChildren,
+            0,
+            0,
+            1,
+            1,
+            control.Handle,
+            0,
+            classInfo.Instance,
+            0);
+        if (_windowHandle == 0)
+        {
+            var error = Marshal.GetLastWin32Error();
+            _ = UnregisterClassW(_win32ClassName, classInfo.Instance);
+            throw new InvalidOperationException($"Could not create the embedded game window (Win32 error {error}).");
+        }
+
+        _surface = new VulkanHostSurface(
+            VulkanHostSurfaceKind.Win32,
+            _windowHandle,
+            classInfo.Instance);
+        return new PlatformHandle(_windowHandle, "HWND");
+    }
+
+    private PlatformHandle CreateX11(IPlatformHandle control)
+    {
+        _x11Display = XOpenDisplay(0);
+        if (_x11Display == 0)
+        {
+            throw new InvalidOperationException("Could not connect to the X11 server for the embedded game surface.");
+        }
+
+        _windowHandle = XCreateSimpleWindow(
+            _x11Display,
+            control.Handle,
+            0,
+            0,
+            1,
+            1,
+            0,
+            0,
+            0);
+        if (_windowHandle == 0)
+        {
+            XCloseDisplay(_x11Display);
+            _x11Display = 0;
+            throw new InvalidOperationException("Could not create the X11 embedded game surface.");
+        }
+
+        if (_presentationVisible)
+        {
+            _ = XMapWindow(_x11Display, _windowHandle);
+        }
+        _ = XFlush(_x11Display);
+        _surface = new VulkanHostSurface(VulkanHostSurfaceKind.Xlib, _windowHandle, _x11Display);
+        return new PlatformHandle(_windowHandle, "X11");
+    }
+
+    private PlatformHandle CreateMacOS()
+    {
+        _metalLayer = CreateObjectiveCObject("CAMetalLayer");
+        _windowHandle = CreateObjectiveCObject("NSView");
+        SendBool(_windowHandle, "setWantsLayer:", true);
+        SendPointer(_windowHandle, "setLayer:", _metalLayer);
+        SendBool(_windowHandle, "setHidden:", !_presentationVisible);
+
+        _surface = new VulkanHostSurface(VulkanHostSurfaceKind.Metal, _windowHandle, metalLayerHandle: _metalLayer);
+        return new PlatformHandle(_windowHandle, "NSView");
+    }
+
+    private void UpdateSurfaceSize()
+    {
+        if (_surface is null)
+        {
+            return;
+        }
+
+        var renderScale = (VisualRoot as TopLevel)?.RenderScaling ?? 1.0;
+        var width = Math.Max(1, (int)Math.Round(Bounds.Width * renderScale));
+        var height = Math.Max(1, (int)Math.Round(Bounds.Height * renderScale));
+        var sizeChanged = _surface.PixelWidth != width || _surface.PixelHeight != height;
+        _surface.UpdatePixelSize(width, height);
+
+        if (!sizeChanged)
+        {
+            return;
+        }
+
+        if (OperatingSystem.IsWindows() && _windowHandle != 0)
+        {
+            _ = SetWindowPos(
+                _windowHandle,
+                0,
+                0,
+                0,
+                width,
+                height,
+                SwpNoMove | SwpNoZOrder | SwpNoActivate);
+        }
+        else if (OperatingSystem.IsLinux() && _x11Display != 0 && _windowHandle != 0)
+        {
+            _ = XResizeWindow(_x11Display, _windowHandle, (uint)width, (uint)height);
+            _ = XFlush(_x11Display);
+        }
+        else if (OperatingSystem.IsMacOS() && _metalLayer != 0)
+        {
+            SendDouble(_metalLayer, "setContentsScale:", renderScale);
+        }
+    }
+
+    private void DestroyWin32()
+    {
+        if (_windowHandle != 0)
+        {
+            _ = DestroyWindow(_windowHandle);
+            _windowHandle = 0;
+        }
+
+        if (!string.IsNullOrWhiteSpace(_win32ClassName))
+        {
+            _ = UnregisterClassW(_win32ClassName, GetModuleHandleW(null));
+            _win32ClassName = null;
+        }
+
+        _windowProcedure = null;
+    }
+
+    private void DestroyX11()
+    {
+        if (_x11Display != 0 && _windowHandle != 0)
+        {
+            _ = XDestroyWindow(_x11Display, _windowHandle);
+        }
+        if (_x11Display != 0)
+        {
+            _ = XCloseDisplay(_x11Display);
+        }
+
+        _windowHandle = 0;
+        _x11Display = 0;
+    }
+
+    private void DestroyMacOS()
+    {
+        if (_windowHandle != 0)
+        {
+            SendVoid(_windowHandle, "release");
+        }
+        if (_metalLayer != 0)
+        {
+            SendVoid(_metalLayer, "release");
+        }
+
+        _windowHandle = 0;
+        _metalLayer = 0;
+    }
+
+    private nint WindowProcedureImpl(nint window, uint message, nint wParam, nint lParam)
+    {
+        if (message == WmMouseMove)
+        {
+            _lastPointerActivity = System.Diagnostics.Stopwatch.GetTimestamp();
+            ShowCursorNow();
+        }
+        else if (message == WmSetCursor && _cursorHidden)
+        {
+            // Win32 re-resolves the cursor on every mouse message; returning
+            // TRUE here keeps the parent chain from restoring the arrow.
+            _ = SetCursor(0);
+            return 1;
+        }
+
+        return DefWindowProcW(window, message, wParam, lParam);
+    }
+
+    private static nint CreateObjectiveCObject(string className)
+    {
+        var classHandle = objc_getClass(className);
+        if (classHandle == 0)
+        {
+            throw new InvalidOperationException($"Objective-C class '{className}' is unavailable.");
+        }
+
+        var instance = objc_msgSend_id(classHandle, sel_registerName("alloc"));
+        instance = objc_msgSend_id(instance, sel_registerName("init"));
+        if (instance == 0)
+        {
+            throw new InvalidOperationException($"Could not create Objective-C '{className}'.");
+        }
+
+        return instance;
+    }
+
+    private static void SendVoid(nint receiver, string selector) =>
+        objc_msgSend_void(receiver, sel_registerName(selector));
+
+    private static void SendBool(nint receiver, string selector, bool value) =>
+        objc_msgSend_bool(receiver, sel_registerName(selector), value ? (byte)1 : (byte)0);
+
+    private static void SendPointer(nint receiver, string selector, nint value) =>
+        objc_msgSend_pointer(receiver, sel_registerName(selector), value);
+
+    private static void SendDouble(nint receiver, string selector, double value) =>
+        objc_msgSend_double(receiver, sel_registerName(selector), value);
+
+    private delegate nint WindowProcedure(nint window, uint message, nint wParam, nint lParam);
+
+    [StructLayout(LayoutKind.Sequential, CharSet = CharSet.Unicode)]
+    private struct WndClassEx
+    {
+        public uint Size;
+        public uint Style;
+        public nint WindowProcedure;
+        public int ClassExtra;
+        public int WindowExtra;
+        public nint Instance;
+        public nint Icon;
+        public nint Cursor;
+        public nint Background;
+        public string? MenuName;
+        public string? ClassName;
+        public nint IconSmall;
+    }
+
+    [DllImport("kernel32.dll", EntryPoint = "GetModuleHandleW", CharSet = CharSet.Unicode)]
+    private static extern nint GetModuleHandleW(string? moduleName);
+
+    [DllImport("user32.dll", EntryPoint = "RegisterClassExW", SetLastError = true, CharSet = CharSet.Unicode)]
+    private static extern ushort RegisterClassExW(ref WndClassEx classInfo);
+
+    [DllImport("user32.dll", EntryPoint = "UnregisterClassW", SetLastError = true, CharSet = CharSet.Unicode)]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool UnregisterClassW(string className, nint instance);
+
+    [DllImport("user32.dll", EntryPoint = "CreateWindowExW", SetLastError = true, CharSet = CharSet.Unicode)]
+    private static extern nint CreateWindowExW(
+        uint extendedStyle,
+        string className,
+        string windowName,
+        uint style,
+        int x,
+        int y,
+        int width,
+        int height,
+        nint parent,
+        nint menu,
+        nint instance,
+        nint parameter);
+
+    [DllImport("user32.dll", EntryPoint = "DestroyWindow", SetLastError = true)]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool DestroyWindow(nint window);
+
+    [DllImport("user32.dll", EntryPoint = "SetWindowPos", SetLastError = true)]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool SetWindowPos(
+        nint window,
+        nint insertAfter,
+        int x,
+        int y,
+        int width,
+        int height,
+        uint flags);
+
+    [DllImport("user32.dll", EntryPoint = "DefWindowProcW", CharSet = CharSet.Unicode)]
+    private static extern nint DefWindowProcW(nint window, uint message, nint wParam, nint lParam);
+
+    [DllImport("user32.dll", EntryPoint = "SetCursor")]
+    private static extern nint SetCursor(nint cursor);
+
+    [DllImport("user32.dll", EntryPoint = "LoadCursorW", CharSet = CharSet.Unicode)]
+    private static extern nint LoadCursorW(nint instance, nint cursorName);
+
+    [DllImport("user32.dll", EntryPoint = "GetCursorPos")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool GetCursorPos(out NativePoint point);
+
+    [DllImport("user32.dll", EntryPoint = "WindowFromPoint")]
+    private static extern nint WindowFromPoint(NativePoint point);
+
+    [StructLayout(LayoutKind.Sequential)]
+    private struct NativePoint
+    {
+        public int X;
+        public int Y;
+    }
+
+    [DllImport("libX11.so.6", EntryPoint = "XOpenDisplay")]
+    private static extern nint XOpenDisplay(nint displayName);
+
+    [DllImport("libX11.so.6", EntryPoint = "XCreateSimpleWindow")]
+    private static extern nint XCreateSimpleWindow(
+        nint display,
+        nint parent,
+        int x,
+        int y,
+        uint width,
+        uint height,
+        uint borderWidth,
+        ulong border,
+        ulong background);
+
+    [DllImport("libX11.so.6", EntryPoint = "XMapWindow")]
+    private static extern int XMapWindow(nint display, nint window);
+
+    [DllImport("libX11.so.6", EntryPoint = "XUnmapWindow")]
+    private static extern int XUnmapWindow(nint display, nint window);
+
+    [DllImport("libX11.so.6", EntryPoint = "XResizeWindow")]
+    private static extern int XResizeWindow(nint display, nint window, uint width, uint height);
+
+    [DllImport("libX11.so.6", EntryPoint = "XDestroyWindow")]
+    private static extern int XDestroyWindow(nint display, nint window);
+
+    [DllImport("libX11.so.6", EntryPoint = "XCloseDisplay")]
+    private static extern int XCloseDisplay(nint display);
+
+    [DllImport("libX11.so.6", EntryPoint = "XFlush")]
+    private static extern int XFlush(nint display);
+
+    [DllImport("/usr/lib/libobjc.A.dylib")]
+    private static extern nint objc_getClass(string name);
+
+    [DllImport("/usr/lib/libobjc.A.dylib")]
+    private static extern nint sel_registerName(string name);
+
+    [DllImport("/usr/lib/libobjc.A.dylib", EntryPoint = "objc_msgSend")]
+    private static extern nint objc_msgSend_id(nint receiver, nint selector);
+
+    [DllImport("/usr/lib/libobjc.A.dylib", EntryPoint = "objc_msgSend")]
+    private static extern void objc_msgSend_void(nint receiver, nint selector);
+
+    [DllImport("/usr/lib/libobjc.A.dylib", EntryPoint = "objc_msgSend")]
+    private static extern void objc_msgSend_bool(nint receiver, nint selector, byte value);
+
+    [DllImport("/usr/lib/libobjc.A.dylib", EntryPoint = "objc_msgSend")]
+    private static extern void objc_msgSend_pointer(nint receiver, nint selector, nint value);
+
+    [DllImport("/usr/lib/libobjc.A.dylib", EntryPoint = "objc_msgSend")]
+    private static extern void objc_msgSend_double(nint receiver, nint selector, double value);
+}

src/SharpEmu.GUI/GuiConsoleMirror.cs

@@ -0,0 +1,116 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Text;
+
+namespace SharpEmu.GUI;
+
+/// <summary>
+/// Mirrors process-wide console output into the launcher console while
+/// retaining the original streams for shell users and file logging.
+/// </summary>
+internal sealed class GuiConsoleMirror : IDisposable
+{
+    private readonly TextWriter _originalOut;
+    private readonly TextWriter _originalError;
+    private int _disposed;
+
+    private GuiConsoleMirror(Action<string, bool> writeLine)
+    {
+        _originalOut = Console.Out;
+        _originalError = Console.Error;
+        Console.SetOut(new MirroringWriter(_originalOut, line => writeLine(line, false)));
+        Console.SetError(new MirroringWriter(_originalError, line => writeLine(line, true)));
+    }
+
+    public static GuiConsoleMirror Install(Action<string, bool> writeLine) => new(writeLine);
+
+    public void Dispose()
+    {
+        if (Interlocked.Exchange(ref _disposed, 1) != 0)
+        {
+            return;
+        }
+
+        Console.SetOut(_originalOut);
+        Console.SetError(_originalError);
+    }
+
+    private sealed class MirroringWriter : TextWriter
+    {
+        private readonly TextWriter _inner;
+        private readonly Action<string> _writeLine;
+        private readonly StringBuilder _line = new();
+        private readonly object _gate = new();
+
+        public MirroringWriter(TextWriter inner, Action<string> writeLine)
+        {
+            _inner = inner;
+            _writeLine = writeLine;
+        }
+
+        public override Encoding Encoding => _inner.Encoding;
+
+        public override void Write(char value)
+        {
+            lock (_gate)
+            {
+                _inner.Write(value);
+                Append(value);
+            }
+        }
+
+        public override void Write(string? value)
+        {
+            if (value is null)
+            {
+                return;
+            }
+
+            lock (_gate)
+            {
+                _inner.Write(value);
+                foreach (var character in value)
+                {
+                    Append(character);
+                }
+            }
+        }
+
+        public override void WriteLine(string? value)
+        {
+            lock (_gate)
+            {
+                _inner.WriteLine(value);
+                if (!string.IsNullOrEmpty(value))
+                {
+                    _line.Append(value);
+                }
+
+                FlushLine();
+            }
+        }
+
+        private void Append(char value)
+        {
+            if (value == '\r')
+            {
+                return;
+            }
+
+            if (value == '\n')
+            {
+                FlushLine();
+                return;
+            }
+
+            _line.Append(value);
+        }
+
+        private void FlushLine()
+        {
+            _writeLine(_line.ToString());
+            _line.Clear();
+        }
+    }
+}

src/SharpEmu.GUI/GuiSettings.cs

@@ -48,6 +48,8 @@ public sealed class GuiSettings
     /// <summary>Publish launcher/game status to Discord Rich Presence.</summary>
     public bool DiscordRichPresence { get; set; } = true;
 
+    public bool CheckForUpdatesOnStartup { get; set; } = true;
+
     /// <summary>Names of SHARPEMU_* switches set to "1" in the emulator's environment at launch.</summary>
     public List<string> EnvironmentToggles { get; set; } = new();
 

src/SharpEmu.GUI/Languages/br.json

@@ -26,6 +26,15 @@
   "Library.Loading": "Carregando biblioteca…",
 
   "Options.General": "Opções Gerais",
+  "Options.Env.Tab": "Ambiente",
+  "Options.Section.Environment": "VARIÁVEIS DE AMBIENTE",
+  "Options.Env.Desc": "Variáveis de ambiente passadas ao emulador durante a inicialização.",
+  "Options.Env.Bthid.Desc": "Informa o Bluetooth HID como indisponível para títulos cujo middleware de volante/FFB fica verificando indefinidamente.\nNormalmente, deixe desativado. Alguns títulos travam quando a inicialização falha.",
+  "Options.Env.LoopGuard.Desc": "Não força o encerramento de títulos que repetem a mesma chamada por tempo excessivo.\nExperimente esta opção se um jogo fechar sozinho durante o carregamento.",
+  "Options.Env.VkValidation.Desc": "Ativa as camadas de validação do Vulkan para depuração de GPU.\nLento. Requer que o Vulkan SDK esteja instalado.",
+  "Options.Env.DumpSpirv.Desc": "Extrai os shaders AGC e suas traduções SPIR-V para a pasta shader-dumps.\nUse ao reportar bugs de shader ou renderização.",
+  "Options.Env.LogDirectMemory.Desc": "Registra alocações de memória direta e falhas no console.\nUse quando um jogo aborta ou fecha durante a inicialização (boot).",
+  "Options.Env.LogNp.Desc": "Registra chamadas da biblioteca NP (PlayStation Network) no console.",
   "Options.Section.Emulation": "EMULAÇÃO",
   "Options.Section.Logging": "LOGS",
   "Options.Section.Launcher": "INICIALIZADOR",
@@ -74,7 +83,7 @@
   "Console.Title": "CONSOLE",
   "Console.SearchWatermark": "Pesquisar...",
   "Console.AutoScroll": "Rolagem automática",
-  "Console.Split": "Recortar",
+  "Console.Split": "Dividir",
   "Console.Copy": "Copiar",
   "Console.Clear": "Limpar",
   "Console.WindowTitle": "Console do SharpEmu",
@@ -125,5 +134,13 @@
   "Dialog.PsExecutables": "Executáveis de PS",
   "Dialog.SaveLogFile": "Selecione onde salvar o arquivo de log",
   "Dialog.PlainTextFiles": "Arquivos de texto simples",
-  "Dialog.LogFiles": "Arquivos de log"
-}
+  "Dialog.LogFiles": "Arquivos de log",
+
+  "Options.About" : "Sobre",
+  "About.Github.Label": "GitHub",
+  "About.Github.Desc": "Código-fonte, problemas e desenvolvimento do projeto.",
+  "About.Discord.Label": "Discord",
+  "About.Discord.Desc": "Participe da comunidade, obtenha suporte e acompanhe o desenvolvimento.",
+  "About.GithubButton": "Contribua no GitHub!",
+  "About.DiscordButton": "Entre no nosso Discord!"
+}

src/SharpEmu.GUI/Languages/en.json

@@ -31,9 +31,11 @@
   "Options.Env.Desc": "Switches passed to the emulator as environment variables at launch.",
   "Options.Env.Bthid.Desc": "Report Bluetooth HID as unavailable for titles whose wheel/FFB middleware polls forever.\nLeave off normally. Some titles freeze when init fails.",
   "Options.Env.LoopGuard.Desc": "Do not force quit titles that repeat the same call for too long.\nTry this when a game exits on its own while loading.",
+  "Options.Env.WritableApp0.Desc": "Allow titles to create and write files inside their install folder.\nNeeded by unpackaged dumps that write their save or config data under /app0.",
   "Options.Env.VkValidation.Desc": "Enable Vulkan validation layers for GPU debugging.\nSlow. Requires the Vulkan SDK to be installed.",
   "Options.Env.DumpSpirv.Desc": "Dump AGC shaders and their SPIR-V translations to the shader-dumps folder.\nUse when reporting shader or rendering bugs.",
   "Options.Env.LogDirectMemory.Desc": "Log direct memory allocations and failures to the console.\nUse when a game aborts or exits during boot.",
+  "Options.Env.LogIo.Desc": "Log file open, read, and path-resolve activity to the console.\nUse when a game cannot find its data files during boot.",
   "Options.Env.LogNp.Desc": "Log NP (PlayStation Network) library calls to the console.",
   "Options.Section.Emulation": "EMULATION",
   "Options.Section.Logging": "LOGGING",
@@ -139,8 +141,25 @@
   "Options.About" : "About",
   "About.Github.Label": "GitHub",
   "About.Github.Desc": "Source code, issues and project development.",
+  "About.Github.LatestCommitLabel": "Latest Commit",
+  "About.Github.LatestCommitDescription": "Latest commit on the main branch",
   "About.Discord.Label": "Discord",
   "About.Discord.Desc": "Join the community, get support and follow development.",
   "About.GithubButton": "Contribute in GitHub!",
-  "About.DiscordButton": "Join our Discord!"
+  "About.DiscordButton": "Join our Discord!",
+
+  "Updater.Auto.Label": "Check for updates on startup",
+  "Updater.Auto.Desc": "Checks GitHub without delaying startup.",
+  "Updater.Label": "Updates",
+  "Updater.Check": "Check for updates",
+  "Updater.DownloadRestart": "Download and restart",
+  "Updater.Status.Ready": "Current build: {0}",
+  "Updater.Status.Checking": "Checking for updates…",
+  "Updater.Status.Current": "You are up to date ({0}).",
+  "Updater.Status.Available": "A new build is available: {0}",
+  "Updater.Status.Downloading": "Downloading update… {0}%",
+  "Updater.Status.Installing": "Installing update…",
+  "Updater.Status.Timeout": "Update check timed out after 10 seconds.",
+  "Updater.Status.Failed": "Could not check for updates.",
+  "Updater.Status.Unsupported": "Automatic updating requires a Windows, Linux or macOS x64 build."
 }

src/SharpEmu.GUI/Languages/es.json

@@ -131,6 +131,8 @@
   "About.Github.Label": "GitHub",
   "About.Github.Desc": "Código fuente, issues y desarrollo del proyecto.",
   "About.Discord.Label": "Discord",
+  "About.Github.LatestCommitLabel": "Último Commit",
+  "About.Github.LatestCommitDescription": "Último commit en la rama main",
   "About.Discord.Desc": "Únete a la comunidad, recibe soporte y sigue el desarrollo.",
   "About.GithubButton": "Contribuye en GitHub!",
   "About.DiscordButton": "Únete a nuestro Discord!"

src/SharpEmu.GUI/Languages/fr.json

@@ -26,6 +26,15 @@
   "Library.Loading": "Chargement de la bibliothèque…",
 
   "Options.General": "Général",
+  "Options.Env.Tab": "Environnement",
+  "Options.Section.Environment": "VARIABLES D’ENVIRONNEMENT",
+  "Options.Env.Desc": "Paramètres passés à l’émulateur comme variables d’environnement au lancement.",
+  "Options.Env.Bthid.Desc": "Signaler le HID Bluetooth comme indisponible pour les titres dont le middleware volant/FFB interroge indéfiniment.\nLaissez désactivé normalement. Certains titres se bloquent si l’initialisation échoue.",
+  "Options.Env.LoopGuard.Desc": "Ne pas forcer la fermeture des titres qui répètent le même appel trop longtemps.\nEssayez ceci quand un jeu se ferme tout seul pendant le chargement.",
+  "Options.Env.VkValidation.Desc": "Activer les couches de validation Vulkan pour le débogage GPU.\nRalentit. Nécessite l’installation du Vulkan SDK.",
+  "Options.Env.DumpSpirv.Desc": "Exporter les shaders AGC et leurs traductions SPIR-V dans le dossier shader-dumps.\nÀ utiliser pour signaler des bugs de shader ou de rendu.",
+  "Options.Env.LogDirectMemory.Desc": "Journaliser les allocations de mémoire directe et les échecs dans la console.\nÀ utiliser quand un jeu plante ou se ferme pendant le démarrage.",
+  "Options.Env.LogNp.Desc": "Journaliser les appels de la bibliothèque NP (PlayStation Network) dans la console.",
   "Options.Section.Emulation": "ÉMULATION",
   "Options.Section.Logging": "JOURNALISATION",
   "Options.Section.Launcher": "LANCEUR",
@@ -125,5 +134,13 @@
   "Dialog.PsExecutables": "Exécutables PlayStation",
   "Dialog.SaveLogFile": "Choisir l’emplacement du fichier journal",
   "Dialog.PlainTextFiles": "Fichiers texte brut",
-  "Dialog.LogFiles": "Fichiers journaux"
+  "Dialog.LogFiles": "Fichiers journaux",
+
+  "Options.About": "À propos",
+  "About.Github.Label": "GitHub",
+  "About.Github.Desc": "Code source, issues et développement du projet.",
+  "About.Discord.Label": "Discord",
+  "About.Discord.Desc": "Rejoignez la communauté, obtenez de l’aide et suivez le développement.",
+  "About.GithubButton": "Contribuer sur GitHub !",
+  "About.DiscordButton": "Rejoindre notre Discord !"
 }

src/SharpEmu.GUI/Languages/hu.json

@@ -0,0 +1,146 @@
+{
+  "_languageName": "Hungarian",
+
+  "Page.Library": "Könyvtár",
+  "Page.Options": "Beállítások",
+  "Page.GameCount.One": "1 játék",
+  "Page.GameCount.Other": "{0} játékok",
+
+  "Library.SearchWatermark": "Keresés a könyvtárban",
+  "Library.AddFolder": "＋ Mappa hozzáadása",
+  "Library.Rescan": "⟳ Újrakeresés",
+  "Library.OpenFile": "Fájl megnyitása…",
+
+  "Library.Context.Launch": "Inditás",
+  "Library.Context.OpenFolder": "Játékmappa megnyitása",
+  "Library.Context.CopyPath": "Elérési út másolása",
+  "Library.Context.CopyTitleId": "Cím ID másolása",
+  "Library.Context.Remove": "Eltávolítás a Könyvtárból",
+
+  "Library.Empty.Title": "A könyvtárad üres",
+  "Library.Empty.Hint": "Add meg a játékaidat tartalmazó mappát a kezdáshez.",
+  "Library.Empty.SearchTitle": "Nincs találat a elemre",
+  "Library.Empty.SearchHint": "A könyvtárban nincs olyan elem, amely egyezne a „{0}” kifejezéssel.",
+  "Library.Empty.AddFolder": "＋ Játékmappa hozzáadása",
+
+  "Library.Loading": "Könyvtár betöltése",
+
+  "Options.General": "Általános",
+  "Options.Env.Tab": "Környezet",
+  "Options.Section.Environment": "KÖRNYEZETI VÁLTOZÓK",
+  "Options.Env.Desc": "Indításkor környezeti változóként az emulátorhoz átadott kapcsolók.",
+  "Options.Env.Bthid.Desc": "Jelenti, amely címeknél a Bluetooth HID nem elérhető, amelyeknél a kormány/FFB-közbenső szoftver végtelenül lekérdezi az adatokat.\nNormál esetben hagyja ki. Egyes címek lefagyanak, ha az inicializálás sikertelen.",
+  "Options.Env.LoopGuard.Desc": "Ne erőltesse a kilépést azoknál a címeknél, amelyek túl sokáig ismételnek ugyanazt a hívást.\nPróbálja ki ezt, ha egy játék betöltés közben magától kilép.",
+  "Options.Env.VkValidation.Desc": "Engedélyezze a Vulkan-érvényesítési rétegeket a GPU hibakereséshez.\nLassú. A Vulkan SDK telepítését igényli.",
+  "Options.Env.DumpSpirv.Desc": "Az AGC-shaderek és azok SPIR-V-fordításainak mentése a shader-dumps mappába.\nHasználd shader- vagy renderelési hibák jelentésekor.",
+  "Options.Env.LogDirectMemory.Desc": "A közvetlen memóriaallokációk és hibák naplózása a konzolra.\nHasználd, ha egy játék a rendszerindítás során megszakad vagy kilép.",
+  "Options.Env.LogNp.Desc": "Az NP (PlayStation Network) könyvtárhívásokat naplózza a konzolra.",
+  "Options.Section.Emulation": "EMULÁCIÓ",
+  "Options.Section.Logging": "LOGOLÁS",
+  "Options.Section.Launcher": "INDITÓ",
+
+  "Options.CpuEngine.Label": "CPU motor",
+  "Options.CpuEngine.Desc": "A játék kódjának futtatásához használt végrehajtó motor.",
+  "Options.CpuEngine.Native": "Natív",
+
+  "Options.Strict.Label": "Szigorú dynlib felbontás",
+  "Options.Strict.Desc": "Indítás megszakítása, ha egy importált szimbólum nem oldható fel.",
+
+  "Options.LogLevel.Label": "Naplózási szint",
+  "Options.LogLevel.Desc": "Az emulátor konzol kimenetének részletessége.",
+  "Options.LogLevel.Trace": "Trace",
+  "Options.LogLevel.Debug": "Debug",
+  "Options.LogLevel.Info": "Információ",
+  "Options.LogLevel.Warning": "Figyelmeztetés",
+  "Options.LogLevel.Error": "Hiba",
+  "Options.LogLevel.Critical": "Kritikus",
+
+  "Options.TraceImports.Label": "Import trace limit",
+  "Options.TraceImports.Desc": "Az első N darab import nyomon követése modulonként (0 = ki).",
+
+  "Options.LogToFile.Label": "Naplozás fájlba",
+  "Options.LogToFile.Desc": "Az emulátor kimenetének tükrözése egy log fájlba.",
+
+  "Options.LogFilePath.Label": "Naplófájl elérési útja",
+  "Options.LogFilePath.Default": "Nincs egyéni út — a logok az emulátor melletti user/logs mappába kerülnek.",
+  "Options.LogFilePath.Select": "Kiválasztás…",
+
+  "Options.OverrideLogFile.Label": "Naplófájl felülírása",
+  "Options.OverrideLogFile.Desc": "A pontos fájlútvonal használata a cím ID és időbélyeg hozzáfűzése helyett.",
+
+  "Options.TitleMusic.Label": "Címzene",
+  "Options.TitleMusic.Desc": "A kiválasztott játék előnézeti zenéjének ismétlése a könyvtárban.",
+
+  "Options.Discord.Label": "Discord jelenlét",
+  "Options.Discord.Desc": "A futó játék megjelenítése a Discord profilodon.",
+
+  "Options.Language.Label": "Emulátor nyelve",
+  "Options.Language.Desc": "Az indítóban használt nyelv. Azonnal érvénybe lép.",
+
+  "Common.On": "Be",
+  "Common.Off": "Ki",
+
+  "Console.Title": "KONZOL",
+  "Console.SearchWatermark": "Keresés...",
+  "Console.AutoScroll": "Automatikus görgetés",
+  "Console.Split": "Felosztás",
+  "Console.Copy": "Másolás",
+  "Console.Clear": "Törlés",
+  "Console.WindowTitle": "SharpEmu Konzol",
+
+  "Launch.NoGameSelected": "Nincs játék kiválasztva",
+  "Launch.NoGameHint": "Válassz egy játékot a könyvtárból, vagy nyiss meg közvetlenül egy eboot.bin fájlt.",
+  "Launch.Idle": "Tétlen",
+  "Launch.Console": "≡ Konzol",
+  "Launch.Launch": "▶  Inditás",
+  "Launch.Stop": "■  Leállítás",
+  "Launch.Running": "Fut — {0}",
+  "Launch.Stopping": "Leállítás…",
+  "Launch.Exited": "Kilépett a következő kóddal: {0} ({1})",
+  "Launch.ExeNotFound": "A SharpEmu futtatható fájl nem található. Előbb építsd fel a SharpEmu.CLI projektet (dotnet build).",
+  "Launch.LogFile": "Naplófájl: {0}",
+  "Launch.Command": "$ SharpEmu {0}",
+  "Launch.StartFailed": "Nem sikerült elindítani az emulátort: {0}",
+  "Launch.ProcessExited": "A folyamat kilépett a következő kóddal: {0} ({1}).",
+
+  "Exit.Ok": "OK",
+  "Exit.InvalidArguments": "nemérvényes argumentumok",
+  "Exit.EbootNotFound": "eboot nem található",
+  "Exit.RuntimeException": "runtime exception",
+  "Exit.EmulationError": "emulációs hiba",
+  "Exit.Unknown": "ismeretlen",
+
+  "Status.EmulatorLocating": "Emulátor: keresés…",
+  "Status.EmulatorPath": "Emulátor: {0}",
+  "Status.EmulatorNotFound": "Emulátor: a SharpEmu futtatható fájl nem található — előbb építsd fel a SharpEmu.CLI-t.",
+  "Status.ScanningLibrary": "Könyvtár beolvasása…",
+  "Status.AddFolderPrompt": "Adj hozzá egy játékmappát a könyvtár feltöltéséhez.",
+  "Status.LibraryScanned": "Könyvtár beolvasva: {0} játék {1} mappában.",
+  "Status.CouldNotOpenFolder": "Nem sikerült megnyitni a mappát: {0}",
+  "Status.CopiedToClipboard": "{0} másolva a vágólapra.",
+  "Status.RemovedFromLibrary": "„{0}” eltávolítva a könyvtárból. A visszaállításához add hozzá újra a mappáját.",
+  "Status.Running": "Fut {0}",
+  "Status.Stopping": "Leállítás…",
+  "Status.Idle": "Nyugodt",
+
+  "Clipboard.Path": "Út",
+  "Clipboard.TitleId": "Cím ID",
+
+  "Discord.Playing": "Játékban {0}",
+  "Discord.Browsing": "Böngéssz a könyvtárban",
+
+  "Dialog.ChooseGameFolder": "Válassz egy mappát ami a játékaidat tartalmazza",
+  "Dialog.OpenExecutable": "Futtatható fájl megnyitása az indításhoz",
+  "Dialog.PsExecutables": "PS futtatható fájlok",
+  "Dialog.SaveLogFile": "Válaszd ki, hogy hova szeretnéd menteni a napló fájlokat",
+  "Dialog.PlainTextFiles": "Egyszerű szöveges fájlok",
+  "Dialog.LogFiles": "Naplózási fájlok",
+
+  "Options.About" : "Erről",
+  "About.Github.Label": "GitHub",
+  "About.Github.Desc": "Forrás kód, hibajelentések és a projekt fejlesztése.",
+  "About.Discord.Label": "Discord",
+  "About.Discord.Desc": "Csatlakozz a közösséghe, kérj segítéget és kövesd nyomon a fejlesztést.",
+  "About.GithubButton": "Járulj hozzá GitHubon!",
+  "About.DiscordButton": "Csatlakozz a Discordunhoz!"
+}

src/SharpEmu.GUI/Languages/pt.json

@@ -0,0 +1,146 @@
+{
+  "_languageName": "Português (Portugal)",
+
+  "Page.Library": "Biblioteca",
+  "Page.Options": "Opções",
+  "Page.GameCount.One": "1 jogo",
+  "Page.GameCount.Other": "{0} jogos",
+
+  "Library.SearchWatermark": "Pesquisar biblioteca…",
+  "Library.AddFolder": "＋ Adicionar pasta",
+  "Library.Rescan": "⟳ Reanalisar",
+  "Library.OpenFile": "Abrir ficheiro…",
+
+  "Library.Context.Launch": "Iniciar",
+  "Library.Context.OpenFolder": "Abrir pasta do jogo",
+  "Library.Context.CopyPath": "Copiar caminho",
+  "Library.Context.CopyTitleId": "Copiar ID do título",
+  "Library.Context.Remove": "Remover da biblioteca",
+
+  "Library.Empty.Title": "A sua biblioteca está vazia",
+  "Library.Empty.Hint": "Adicione uma pasta com os seus jogos para começar.",
+  "Library.Empty.SearchTitle": "Nenhum jogo corresponde à sua pesquisa",
+  "Library.Empty.SearchHint": "Nada na biblioteca corresponde a “{0}”.",
+  "Library.Empty.AddFolder": "＋ Adicionar pasta de jogos",
+
+  "Library.Loading": "A carregar biblioteca…",
+
+  "Options.General": "Geral",
+  "Options.Env.Tab": "Ambiente",
+  "Options.Section.Environment": "VARIÁVEIS DE AMBIENTE",
+  "Options.Env.Desc": "Switches passados ao emulador como variáveis de ambiente no arranque.",
+  "Options.Env.Bthid.Desc": "Reporta o Bluetooth HID como indisponível para títulos cujo middleware de volante/FFB fica à espera indefinidamente.\nDeixe desativado normalmente. Alguns títulos bloqueiam quando a inicialização falha.",
+  "Options.Env.LoopGuard.Desc": "Não force o encerramento de títulos que repetem a mesma chamada durante demasiado tempo.\nExperimente isto quando um jogo fecha sozinho durante o carregamento.",
+  "Options.Env.VkValidation.Desc": "Ativa as camadas de validação do Vulkan para depuração da GPU.\nLento. Requer que o Vulkan SDK esteja instalado.",
+  "Options.Env.DumpSpirv.Desc": "Extrai os shaders AGC e as respetivas traduções SPIR-V para a pasta shader-dumps.\nUtilize ao reportar problemas de shaders ou renderização.",
+  "Options.Env.LogDirectMemory.Desc": "Regista alocações de memória direta e falhas na consola.\nUtilize quando um jogo aborta ou fecha durante o arranque.",
+  "Options.Env.LogNp.Desc": "Regista chamadas da biblioteca NP (PlayStation Network) na consola.",
+  "Options.Section.Emulation": "EMULAÇÃO",
+  "Options.Section.Logging": "REGISTOS",
+  "Options.Section.Launcher": "LANÇADOR",
+
+  "Options.CpuEngine.Label": "Motor de CPU",
+  "Options.CpuEngine.Desc": "Motor de execução utilizado para correr o código do jogo.",
+  "Options.CpuEngine.Native": "Nativo",
+
+  "Options.Strict.Label": "Resolução estrita de dynlib",
+  "Options.Strict.Desc": "Falha o arranque quando um símbolo importado não pode ser resolvido.",
+
+  "Options.LogLevel.Label": "Nível de registo",
+  "Options.LogLevel.Desc": "Nível de detalhe da saída da consola do emulador.",
+  "Options.LogLevel.Trace": "Rastreio",
+  "Options.LogLevel.Debug": "Depuração",
+  "Options.LogLevel.Info": "Informação",
+  "Options.LogLevel.Warning": "Aviso",
+  "Options.LogLevel.Error": "Erro",
+  "Options.LogLevel.Critical": "Crítico",
+
+  "Options.TraceImports.Label": "Limite de rastreio de importações",
+  "Options.TraceImports.Desc": "Rastreia as primeiras N importações por módulo (0 = desativado).",
+
+  "Options.LogToFile.Label": "Registar para ficheiro",
+  "Options.LogToFile.Desc": "Duplicar a saída do emulador para um ficheiro de registo.",
+
+  "Options.LogFilePath.Label": "Caminho do ficheiro de registo",
+  "Options.LogFilePath.Default": "Sem caminho personalizado — os registos vão para user/logs junto ao emulador.",
+  "Options.LogFilePath.Select": "Selecionar…",
+
+  "Options.OverrideLogFile.Label": "Substituir ficheiro de registo",
+  "Options.OverrideLogFile.Desc": "Utilizar o caminho de ficheiro exato em vez de acrescentar o ID do título e a hora.",
+
+  "Options.TitleMusic.Label": "Música do título",
+  "Options.TitleMusic.Desc": "Repetir em loop a música de pré-visualização do jogo selecionado na biblioteca.",
+
+  "Options.Discord.Label": "Presença no Discord",
+  "Options.Discord.Desc": "Mostrar o jogo em execução no seu perfil do Discord.",
+
+  "Options.Language.Label": "Idioma do emulador",
+  "Options.Language.Desc": "Idioma utilizado em todo o lançador. Aplica-se de imediato.",
+
+  "Common.On": "Ativado",
+  "Common.Off": "Desativado",
+
+  "Console.Title": "CONSOLA",
+  "Console.SearchWatermark": "Pesquisar...",
+  "Console.AutoScroll": "Deslocamento automático",
+  "Console.Split": "Dividir",
+  "Console.Copy": "Copiar",
+  "Console.Clear": "Limpar",
+  "Console.WindowTitle": "Consola do SharpEmu",
+
+  "Launch.NoGameSelected": "Nenhum jogo selecionado",
+  "Launch.NoGameHint": "Escolha um jogo da biblioteca ou abra um eboot.bin diretamente.",
+  "Launch.Idle": "Inativo",
+  "Launch.Console": "≡ Consola",
+  "Launch.Launch": "▶  Iniciar",
+  "Launch.Stop": "■  Parar",
+  "Launch.Running": "Em execução — {0}",
+  "Launch.Stopping": "A parar…",
+  "Launch.Exited": "Terminou com o código {0} ({1})",
+  "Launch.ExeNotFound": "Executável do SharpEmu não encontrado. Compile primeiro o projeto SharpEmu.CLI (dotnet build).",
+  "Launch.LogFile": "Ficheiro de registo: {0}",
+  "Launch.Command": "$ SharpEmu {0}",
+  "Launch.StartFailed": "Falha ao iniciar o emulador: {0}",
+  "Launch.ProcessExited": "O processo terminou com o código {0} ({1}).",
+
+  "Exit.Ok": "OK",
+  "Exit.InvalidArguments": "argumentos inválidos",
+  "Exit.EbootNotFound": "eboot não encontrado",
+  "Exit.RuntimeException": "exceção em tempo de execução",
+  "Exit.EmulationError": "erro de emulação",
+  "Exit.Unknown": "desconhecido",
+
+  "Status.EmulatorLocating": "Emulador: a localizar…",
+  "Status.EmulatorPath": "Emulador: {0}",
+  "Status.EmulatorNotFound": "Emulador: executável do SharpEmu não encontrado — compile primeiro o SharpEmu.CLI.",
+  "Status.ScanningLibrary": "A analisar biblioteca…",
+  "Status.AddFolderPrompt": "Adicione uma pasta de jogos para preencher a biblioteca.",
+  "Status.LibraryScanned": "Biblioteca analisada: {0} jogo(s) em {1} pasta(s).",
+  "Status.CouldNotOpenFolder": "Não foi possível abrir a pasta: {0}",
+  "Status.CopiedToClipboard": "{0} copiado para a área de transferência.",
+  "Status.RemovedFromLibrary": "“{0}” removido da biblioteca. Adicione novamente a pasta para o restaurar.",
+  "Status.Running": "A executar {0}",
+  "Status.Stopping": "A parar…",
+  "Status.Idle": "Inativo",
+
+  "Clipboard.Path": "Caminho",
+  "Clipboard.TitleId": "ID do título",
+
+  "Discord.Playing": "A jogar {0}",
+  "Discord.Browsing": "A navegar na biblioteca",
+
+  "Dialog.ChooseGameFolder": "Escolha uma pasta com jogos",
+  "Dialog.OpenExecutable": "Abrir um executável para iniciar",
+  "Dialog.PsExecutables": "Executáveis PS",
+  "Dialog.SaveLogFile": "Selecione onde guardar o ficheiro de registo",
+  "Dialog.PlainTextFiles": "Ficheiros de Texto Simples",
+  "Dialog.LogFiles": "Ficheiros de Registo",
+
+  "Options.About" : "Sobre",
+  "About.Github.Label": "GitHub",
+  "About.Github.Desc": "Código-fonte, problemas e desenvolvimento do projeto.",
+  "About.Discord.Label": "Discord",
+  "About.Discord.Desc": "Junte-se à comunidade, obtenha suporte e acompanhe o desenvolvimento.",
+  "About.GithubButton": "Contribua no GitHub!",
+  "About.DiscordButton": "Junte-se ao nosso Discord!"
+}

src/SharpEmu.GUI/Languages/tr.json

@@ -125,5 +125,20 @@
   "Dialog.PsExecutables": "PS çalıştırılabilirleri",
   "Dialog.SaveLogFile": "Günlük dosyasının kaydedileceği yeri seçin",
   "Dialog.PlainTextFiles": "Düz Metin Dosyaları",
-  "Dialog.LogFiles": "Günlük Dosyaları"
+  "Dialog.LogFiles": "Günlük Dosyaları",
+
+  "Updater.Auto.Label": "Açılışta güncellemeleri denetle",
+  "Updater.Auto.Desc": "Açılışı geciktirmeden GitHub'ı denetler.",
+  "Updater.Label": "Güncellemeler",
+  "Updater.Check": "Güncellemeleri denetle",
+  "Updater.DownloadRestart": "İndir ve yeniden başlat",
+  "Updater.Status.Ready": "Mevcut build: {0}",
+  "Updater.Status.Checking": "Güncellemeler denetleniyor…",
+  "Updater.Status.Current": "Güncelsiniz ({0}).",
+  "Updater.Status.Available": "Yeni build mevcut: {0}",
+  "Updater.Status.Downloading": "Güncelleme indiriliyor… %{0}",
+  "Updater.Status.Installing": "Güncelleme kuruluyor…",
+  "Updater.Status.Timeout": "Güncelleme denetimi 10 saniye sonra zaman aşımına uğradı.",
+  "Updater.Status.Failed": "Güncellemeler denetlenemedi.",
+  "Updater.Status.Unsupported": "Otomatik güncelleme Windows, Linux veya macOS x64 build'i gerektirir."
 }

src/SharpEmu.GUI/MainWindow.axaml

@@ -5,6 +5,8 @@ SPDX-License-Identifier: GPL-2.0-or-later
 
 <Window xmlns="https://github.com/avaloniaui"
         xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
+        xmlns:local="clr-namespace:SharpEmu.GUI"
+        xmlns:primitives="clr-namespace:Avalonia.Controls.Primitives;assembly=Avalonia.Controls"
         x:Class="SharpEmu.GUI.MainWindow"
         Title="SharpEmu"
         Width="1280" Height="820"
@@ -18,7 +20,7 @@ SPDX-License-Identifier: GPL-2.0-or-later
         Icon="avares://SharpEmu.GUI/Assets/SharpEmu.ico"
         KeyDown="OnKeyDown">
 
-  <Grid RowDefinitions="Auto,*,Auto">
+  <Grid x:Name="RootLayout" RowDefinitions="Auto,*,Auto">
 
     <!-- Selected-game backdrop: key art behind the main content, dimmed by
          a scrim so tiles and text stay readable. Fades on selection. -->
@@ -56,13 +58,19 @@ SPDX-License-Identifier: GPL-2.0-or-later
     </Grid>
 
     <!-- Main content -->
-    <Grid Grid.Row="1" Margin="32,24,32,20" RowDefinitions="Auto,*,Auto,Auto">
+    <Grid x:Name="MainContent" Grid.Row="1" Margin="32,24,32,20" RowDefinitions="Auto,*,Auto,Auto">
+
+      <!-- The game owns the full client area while running. Session controls
+           use a native popup so they can stay above this native child surface. -->
+      <Border x:Name="GameView" Grid.Row="0" Grid.RowSpan="4" IsVisible="False" Background="#000000" ClipToBounds="True">
+        <Grid x:Name="GameSurfaceContainer" />
+      </Border>
 
       <!-- Library / Options page switcher, with the library toolbar sharing
            the same row on the right. Plain buttons (not TabItem) so there is
            no underline; LB/RB hint chips flank the pair and the gamepad's
            shoulder buttons actually switch pages from anywhere. -->
-      <Grid Grid.Row="0" ColumnDefinitions="Auto,*,Auto" Margin="0,0,0,20">
+      <Grid x:Name="ContentToolbar" Grid.Row="0" ColumnDefinitions="Auto,*,Auto" Margin="0,0,0,20">
         <StackPanel Grid.Column="0" Orientation="Horizontal" Spacing="14" VerticalAlignment="Center">
           <Border Classes="padHint" VerticalAlignment="Center">
             <TextBlock Text="LB" FontSize="11" FontWeight="Bold" Foreground="{StaticResource MutedBrush}" />
@@ -315,6 +323,16 @@ SPDX-License-Identifier: GPL-2.0-or-later
                         <ToggleSwitch Grid.Column="1" x:Name="DiscordToggle" OnContent="On" OffContent="Off"
                                       VerticalAlignment="Center" />
                       </Grid>
+
+                      <Grid ColumnDefinitions="*,Auto">
+                        <StackPanel VerticalAlignment="Center" Spacing="2" Margin="0,0,16,0">
+                          <TextBlock x:Name="AutoUpdateLabel" Text="Check for updates on startup" FontSize="13" />
+                          <TextBlock x:Name="AutoUpdateDesc" Text="Checks GitHub without delaying startup."
+                                     FontSize="11" Foreground="{StaticResource MutedBrush}" TextWrapping="Wrap" />
+                        </StackPanel>
+                        <ToggleSwitch Grid.Column="1" x:Name="AutoUpdateToggle" OnContent="On" OffContent="Off"
+                                      IsChecked="True" VerticalAlignment="Center" />
+                      </Grid>
                     </StackPanel>
                   </Border>
                   <Border Classes="card">
@@ -322,6 +340,40 @@ SPDX-License-Identifier: GPL-2.0-or-later
                       <TextBlock x:Name="AboutSectionTitle"
                                 Classes="sectionTitle"
                                 Text="ABOUT" />
+
+                      <!--Latest commit info-->
+                      <Grid ColumnDefinitions="*,Auto">
+                        <StackPanel Grid.Column="0" Orientation="Horizontal" Spacing="8" VerticalAlignment="Center">
+                          <Image Source="avares://SharpEmu.GUI/Assets/commit-icon.png"
+                                    Width="24" Height="24" VerticalAlignment="Center" />
+                          <StackPanel Spacing="2" VerticalAlignment="Center">
+                            <TextBlock x:Name="LatestCommitLabel" Text="Latest commit" FontSize="13"/>
+                            <TextBlock x:Name="LatestCommitDescription" Text="Latest commit on the main branch" FontSize="11" Foreground="{StaticResource MutedBrush}"/>
+                          </StackPanel>
+                        </StackPanel>
+
+                        <Button Grid.Column="1" x:Name="LatestCommitHashText" Classes="ghost"
+                          Content="Loading…" FontSize="12" FontFamily="Consolas,monospace"
+                          VerticalAlignment="Center" IsEnabled="False" />
+
+                      </Grid>
+
+                      <!--Update-->
+                      <Grid ColumnDefinitions="*,Auto">
+                        <StackPanel Grid.Column="0" Orientation="Horizontal" Spacing="8" VerticalAlignment="Center">
+                          <Image Source="avares://SharpEmu.GUI/Assets/update-icon.png"
+                                    Width="24"
+                                    Height="24"
+                                    VerticalAlignment="Center" />
+                          <StackPanel Spacing="2" VerticalAlignment="Center">
+                            <TextBlock x:Name="UpdateLabel" Text="Updates" FontSize="13" />
+                            <TextBlock x:Name="UpdateStatusText" Text="Current build: dev" FontSize="11"
+                                      Foreground="{StaticResource MutedBrush}" TextWrapping="Wrap" />
+                          </StackPanel>
+                        </StackPanel>
+                        <Button Grid.Column="1" x:Name="UpdateButton" Classes="ghost"
+                                Content="Check for updates" VerticalAlignment="Center" />
+                      </Grid>
                       
                       <!--Github-->
                       <Grid ColumnDefinitions="*,Auto">
@@ -420,6 +472,17 @@ SPDX-License-Identifier: GPL-2.0-or-later
                                       VerticalAlignment="Center" />
                       </Grid>
 
+                      <Grid ColumnDefinitions="*,Auto">
+                        <StackPanel VerticalAlignment="Center" Spacing="2" Margin="0,0,16,0">
+                          <TextBlock Text="SHARPEMU_WRITABLE_APP0" FontSize="13" FontFamily="Consolas,monospace" />
+                          <TextBlock x:Name="EnvWritableApp0Desc"
+                                     Text="Allow titles to create and write files inside their install folder.&#10;Needed by unpackaged dumps that write their save or config data under /app0."
+                                     FontSize="11" Foreground="{StaticResource MutedBrush}" TextWrapping="Wrap" />
+                        </StackPanel>
+                        <ToggleSwitch Grid.Column="1" x:Name="EnvWritableApp0Toggle" OnContent="On" OffContent="Off"
+                                      VerticalAlignment="Center" />
+                      </Grid>
+
                       <Grid ColumnDefinitions="*,Auto">
                         <StackPanel VerticalAlignment="Center" Spacing="2" Margin="0,0,16,0">
                           <TextBlock Text="SHARPEMU_VK_VALIDATION" FontSize="13" FontFamily="Consolas,monospace" />
@@ -453,6 +516,17 @@ SPDX-License-Identifier: GPL-2.0-or-later
                                       VerticalAlignment="Center" />
                       </Grid>
 
+                      <Grid ColumnDefinitions="*,Auto">
+                        <StackPanel VerticalAlignment="Center" Spacing="2" Margin="0,0,16,0">
+                          <TextBlock Text="SHARPEMU_LOG_IO" FontSize="13" FontFamily="Consolas,monospace" />
+                          <TextBlock x:Name="EnvLogIoDesc"
+                                     Text="Log file open, read, and path-resolve activity to the console.&#10;Use when a game cannot find its data files during boot."
+                                     FontSize="11" Foreground="{StaticResource MutedBrush}" TextWrapping="Wrap" />
+                        </StackPanel>
+                        <ToggleSwitch Grid.Column="1" x:Name="EnvLogIoToggle" OnContent="On" OffContent="Off"
+                                      VerticalAlignment="Center" />
+                      </Grid>
+
                       <Grid ColumnDefinitions="*,Auto">
                         <StackPanel VerticalAlignment="Center" Spacing="2" Margin="0,0,16,0">
                           <TextBlock Text="SHARPEMU_LOG_NP" FontSize="13" FontFamily="Consolas,monospace" />
@@ -502,7 +576,7 @@ SPDX-License-Identifier: GPL-2.0-or-later
       </Border>
 
       <!-- Launch bar; capped so it does not sprawl on a maximized window. -->
-      <Border Grid.Row="3" Classes="card" Margin="0,12,0,0" Padding="14" MaxWidth="1280">
+      <Border Grid.Row="3" x:Name="LaunchBar" Classes="card" Margin="0,12,0,0" Padding="14" MaxWidth="1280">
         <StackPanel Spacing="14">
           <Grid ColumnDefinitions="Auto,*,Auto">
 
@@ -567,6 +641,66 @@ SPDX-License-Identifier: GPL-2.0-or-later
       </Border>
     </Grid>
 
+    <!-- Avalonia's regular overlay layer cannot appear over a native child
+         HWND/X11/Metal surface. Keep the running-session controls in a native
+         popup so the game reaches the bottom status bar without losing Stop. -->
+    <primitives:Popup x:Name="SessionBarPopup"
+                      IsOpen="False"
+                      PlacementTarget="{Binding #GameView}"
+                      Placement="Bottom"
+                      VerticalOffset="-66"
+                      Topmost="True"
+                      ShouldUseOverlayLayer="False"
+                      TakesFocusFromNativeControl="False"
+                      IsLightDismissEnabled="False">
+      <Border Classes="card" Width="598" Height="58" CornerRadius="16" Padding="14,8">
+        <Grid ColumnDefinitions="*,Auto">
+          <StackPanel Spacing="3" VerticalAlignment="Center">
+            <StackPanel Orientation="Horizontal" Spacing="8">
+              <TextBlock x:Name="SessionGameTitle" Text="GAME RUNNING" FontSize="13" FontWeight="SemiBold"
+                         MaxWidth="240" TextTrimming="CharacterEllipsis" VerticalAlignment="Center" />
+              <Border Classes="badge running" VerticalAlignment="Center">
+                <TextBlock Text="RUNNING" FontSize="9" FontWeight="Bold" LetterSpacing="1"
+                           Foreground="{StaticResource SuccessBrush}" />
+              </Border>
+            </StackPanel>
+            <StackPanel Orientation="Horizontal" Spacing="7">
+              <Border x:Name="SessionF11Badge" Classes="badge key" VerticalAlignment="Center">
+                <TextBlock Text="F11" FontSize="9" FontWeight="Bold"
+                           Foreground="{StaticResource InfoBrush}" />
+              </Border>
+              <TextBlock x:Name="SessionHintText" Text="Fullscreen" FontSize="11"
+                         Foreground="{StaticResource MutedBrush}" VerticalAlignment="Center" />
+            </StackPanel>
+          </StackPanel>
+          <StackPanel Grid.Column="1" Orientation="Horizontal" Spacing="8" VerticalAlignment="Center">
+            <Button x:Name="SessionConsoleButton" Classes="ghost" Content="≡ Console" />
+            <Button x:Name="SessionStopButton" Classes="danger" Content="■  Stop" IsEnabled="False" />
+          </StackPanel>
+        </Grid>
+      </Border>
+    </primitives:Popup>
+
+    <!-- This is a native popup rather than an Avalonia overlay because the
+         emulated Vulkan surface is a native child window. -->
+    <primitives:Popup x:Name="SessionLoadingPopup"
+                      IsOpen="False"
+                      PlacementTarget="{Binding #GameView}"
+                      Placement="Center"
+                      Topmost="True"
+                      ShouldUseOverlayLayer="False"
+                      TakesFocusFromNativeControl="False"
+                      IsLightDismissEnabled="False">
+      <Border Classes="card" Width="380" CornerRadius="18" Padding="22,18">
+        <StackPanel Spacing="12">
+          <TextBlock x:Name="SessionLoadingTitle" Text="Loading game" FontSize="16" FontWeight="SemiBold" />
+          <TextBlock x:Name="SessionLoadingDetail" Text="Preparing the emulation session..." FontSize="12"
+                     Foreground="{StaticResource MutedBrush}" TextTrimming="CharacterEllipsis" />
+          <ProgressBar IsIndeterminate="True" Height="5" />
+        </StackPanel>
+      </Border>
+    </primitives:Popup>
+
     <!-- Status bar -->
     <Grid x:Name="StatusBar" Grid.Row="2" Height="32" Background="{StaticResource ChromeBrush}"
           ColumnDefinitions="*,Auto">

src/SharpEmu.GUI/SharpEmu.GUI.csproj

@@ -9,12 +9,18 @@ SPDX-License-Identifier: GPL-2.0-or-later
        the executable is started without arguments. -->
   <PropertyGroup>
     <GenerateDocumentationFile>false</GenerateDocumentationFile>
-    <Version>0.0.1</Version>
+    <!-- Required by the source-generated LibraryImport stubs in the linked
+         controller readers below. -->
+    <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
   </PropertyGroup>
 
   <!-- Dependency-free; provides the BuildInfo provenance shown in the
        title bar. -->
   <ItemGroup>
+    <!-- The GUI owns the native presentation control while each game runs in
+         an isolated emulator process. -->
+    <ProjectReference Include="..\SharpEmu.Core\SharpEmu.Core.csproj" />
+    <ProjectReference Include="..\SharpEmu.Libs\SharpEmu.Libs.csproj" />
     <ProjectReference Include="..\SharpEmu.Logging\SharpEmu.Logging.csproj" />
   </ItemGroup>
 
@@ -30,6 +36,8 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <AvaloniaResource Include="..\..\assets\images\SharpEmu.ico" Link="Assets/SharpEmu.ico" />
     <AvaloniaResource Include="..\..\assets\images\github.png" Link="Assets/github.png" />
     <AvaloniaResource Include="..\..\assets\images\discord.png" Link="Assets/discord.png" />
+    <AvaloniaResource Include="..\..\assets\images\update-icon.png" Link="Assets/update-icon.png" />
+    <AvaloniaResource Include="..\..\assets\images\commit-icon.png" Link="Assets/commit-icon.png" />
   </ItemGroup>
 
   <ItemGroup>
@@ -37,18 +45,4 @@ SPDX-License-Identifier: GPL-2.0-or-later
       <LogicalName>Languages.%(Filename)%(Extension)</LogicalName>
     </EmbeddedResource>
   </ItemGroup>
-
-
-
-  <!-- The controller readers (DualSense raw HID + Xbox XInput) are shared
-       with the emulator's pad HLE. They are dependency-free, so they are
-       compiled in directly rather than pulling a reference to all of
-       SharpEmu.Libs into the launcher. -->
-  <ItemGroup>
-    <Compile Include="..\SharpEmu.Libs\Pad\PadState.cs" Link="Input/PadState.cs" />
-    <Compile Include="..\SharpEmu.Libs\Pad\HidNative.cs" Link="Input/HidNative.cs" />
-    <Compile Include="..\SharpEmu.Libs\Pad\DualSenseReader.cs" Link="Input/DualSenseReader.cs" />
-    <Compile Include="..\SharpEmu.Libs\Pad\XInputReader.cs" Link="Input/XInputReader.cs" />
-  </ItemGroup>
-
 </Project>

src/SharpEmu.GUI/Updater.cs

@@ -0,0 +1,253 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Diagnostics;
+using System.Formats.Tar;
+using System.IO.Compression;
+using System.Net.Http.Headers;
+using System.Runtime.InteropServices;
+using System.Text.Json;
+
+namespace SharpEmu.GUI;
+
+/// <summary>Self-contained Windows updater; the emulator layers do not depend on it.</summary>
+public static class Updater
+{
+    private const string ApplyArgument = "--sharpemu-apply-update";
+    private const string LatestReleaseUrl = "https://api.github.com/repos/sharpemu/sharpemu/releases/latest";
+    private static readonly TimeSpan CheckTimeout = TimeSpan.FromSeconds(10);
+    private static readonly HttpClient Http = CreateHttpClient();
+
+    public sealed record UpdateInfo(string Sha, string Name, string DownloadUrl, long Size);
+
+    public static async Task<UpdateInfo?> CheckAsync(string? currentSha, CancellationToken cancellationToken = default)
+    {
+        var platform = CurrentPlatform();
+        using var timeout = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken);
+        timeout.CancelAfter(CheckTimeout);
+
+        using var response = await Http.GetAsync(LatestReleaseUrl, timeout.Token);
+        response.EnsureSuccessStatusCode();
+        return ParseRelease(
+            await response.Content.ReadAsStringAsync(timeout.Token),
+            currentSha,
+            platform.Rid,
+            platform.Extension);
+    }
+
+    public static async Task DownloadAndRestartAsync(
+        UpdateInfo update,
+        IProgress<int>? progress = null,
+        CancellationToken cancellationToken = default)
+    {
+        var root = Path.Combine(Path.GetTempPath(), "SharpEmu.Update");
+        var payload = Path.Combine(root, "payload");
+        if (Directory.Exists(root))
+        {
+            Directory.Delete(root, recursive: true);
+        }
+
+        Directory.CreateDirectory(root);
+        var archive = Path.Combine(root, update.Name);
+        using (var response = await Http.GetAsync(update.DownloadUrl, HttpCompletionOption.ResponseHeadersRead, cancellationToken))
+        {
+            response.EnsureSuccessStatusCode();
+            await using var input = await response.Content.ReadAsStreamAsync(cancellationToken);
+            await using var output = File.Create(archive);
+            var buffer = new byte[81920];
+            long written = 0;
+            int read;
+            while ((read = await input.ReadAsync(buffer, cancellationToken)) > 0)
+            {
+                await output.WriteAsync(buffer.AsMemory(0, read), cancellationToken);
+                written += read;
+                progress?.Report(update.Size == 0 ? 0 : (int)(written * 100 / update.Size));
+            }
+
+            if (written != update.Size)
+            {
+                throw new InvalidDataException($"Downloaded {written} bytes; expected {update.Size}.");
+            }
+        }
+
+        var platform = CurrentPlatform();
+        var stagedExe = ExtractArchive(archive, payload, platform.Extension, platform.ExecutableName);
+
+        var start = new ProcessStartInfo(stagedExe)
+        {
+            UseShellExecute = false,
+            WorkingDirectory = payload,
+        };
+        start.ArgumentList.Add(ApplyArgument);
+        start.ArgumentList.Add(Environment.ProcessId.ToString());
+        start.ArgumentList.Add(AppContext.BaseDirectory);
+        using var helper = Process.Start(start)
+            ?? throw new InvalidOperationException("The update installer could not be started.");
+    }
+
+    /// <summary>Runs from the downloaded executable after the old GUI exits.</summary>
+    public static bool TryApply(string[] args, out int exitCode)
+    {
+        exitCode = 0;
+        if (args.Length != 3 || args[0] != ApplyArgument)
+        {
+            return false;
+        }
+
+        try
+        {
+            if (int.TryParse(args[1], out var oldPid))
+            {
+                try
+                {
+                    if (!Process.GetProcessById(oldPid).WaitForExit(30_000))
+                    {
+                        throw new TimeoutException("SharpEmu did not close within 30 seconds.");
+                    }
+                }
+                catch (ArgumentException)
+                {
+                    // The old process has already exited.
+                }
+            }
+
+            var source = AppContext.BaseDirectory;
+            var target = Path.GetFullPath(args[2]);
+            foreach (var file in Directory.EnumerateFiles(source, "*", SearchOption.AllDirectories))
+            {
+                var relative = Path.GetRelativePath(source, file);
+                if (relative.Equals("gui-settings.json", StringComparison.OrdinalIgnoreCase) ||
+                    relative.StartsWith("user" + Path.DirectorySeparatorChar, StringComparison.OrdinalIgnoreCase) ||
+                    relative.StartsWith("logs" + Path.DirectorySeparatorChar, StringComparison.OrdinalIgnoreCase) ||
+                    relative.StartsWith("Languages" + Path.DirectorySeparatorChar, StringComparison.OrdinalIgnoreCase))
+                {
+                    continue;
+                }
+
+                var destination = Path.Combine(target, relative);
+                Directory.CreateDirectory(Path.GetDirectoryName(destination)!);
+                File.Copy(file, destination, overwrite: true);
+                if (!OperatingSystem.IsWindows())
+                {
+                    File.SetUnixFileMode(destination, File.GetUnixFileMode(file));
+                }
+            }
+
+            using var restarted = Process.Start(new ProcessStartInfo(
+                Path.Combine(target, CurrentPlatform().ExecutableName))
+            {
+                UseShellExecute = false,
+                WorkingDirectory = target,
+            }) ?? throw new InvalidOperationException("The updated SharpEmu could not be started.");
+        }
+        catch (Exception ex)
+        {
+            exitCode = 1;
+            try
+            {
+                File.WriteAllText(Path.Combine(args[2], "update-error.log"), ex.ToString());
+            }
+            catch
+            {
+                // Best-effort diagnostics only.
+            }
+        }
+
+        return true;
+    }
+
+    private static UpdateInfo? ParseRelease(
+        string json,
+        string? currentSha,
+        string rid,
+        string extension)
+    {
+        using var document = JsonDocument.Parse(json);
+        var candidates = new List<(DateTimeOffset Created, UpdateInfo Update)>();
+        foreach (var asset in document.RootElement.GetProperty("assets").EnumerateArray())
+        {
+            var name = asset.GetProperty("name").GetString() ?? "";
+            var marker = $"-{rid}-";
+            var markerIndex = name.LastIndexOf(marker, StringComparison.OrdinalIgnoreCase);
+            if (!name.EndsWith(extension, StringComparison.OrdinalIgnoreCase) ||
+                markerIndex < 0)
+            {
+                continue;
+            }
+
+            var sha = name[(markerIndex + marker.Length)..^extension.Length];
+            if (sha.Length < 7 || !sha.All(Uri.IsHexDigit))
+            {
+                continue;
+            }
+
+            candidates.Add((
+                asset.GetProperty("created_at").GetDateTimeOffset(),
+                new UpdateInfo(
+                    sha,
+                    name,
+                    asset.GetProperty("browser_download_url").GetString()!,
+                    asset.GetProperty("size").GetInt64())));
+        }
+
+        var latest = candidates.OrderByDescending(candidate => candidate.Created).FirstOrDefault().Update;
+        return latest is null || string.Equals(latest.Sha, currentSha, StringComparison.OrdinalIgnoreCase)
+            ? null
+            : latest;
+    }
+
+    private static string ExtractArchive(
+        string archive,
+        string payload,
+        string extension,
+        string executableName)
+    {
+        if (extension == ".zip")
+        {
+            ZipFile.ExtractToDirectory(archive, payload);
+        }
+        else
+        {
+            Directory.CreateDirectory(payload);
+            using var compressed = File.OpenRead(archive);
+            using var gzip = new GZipStream(compressed, CompressionMode.Decompress);
+            TarFile.ExtractToDirectory(gzip, payload, overwriteFiles: false);
+        }
+
+        var executable = Path.Combine(payload, executableName);
+        if (!File.Exists(executable))
+        {
+            throw new InvalidDataException($"The update archive does not contain {executableName}.");
+        }
+
+        if (!OperatingSystem.IsWindows())
+        {
+            File.SetUnixFileMode(executable, File.GetUnixFileMode(executable) | UnixFileMode.UserExecute);
+        }
+
+        return executable;
+    }
+
+    private static HttpClient CreateHttpClient()
+    {
+        var client = new HttpClient { Timeout = Timeout.InfiniteTimeSpan };
+        client.DefaultRequestHeaders.UserAgent.Add(new ProductInfoHeaderValue("SharpEmu", "0.0.1"));
+        client.DefaultRequestHeaders.Accept.Add(new MediaTypeWithQualityHeaderValue("application/vnd.github+json"));
+        return client;
+    }
+
+    private static PlatformInfo CurrentPlatform()
+    {
+        if (RuntimeInformation.ProcessArchitecture != Architecture.X64)
+        {
+            throw new PlatformNotSupportedException("SharpEmu releases require an x64 process.");
+        }
+
+        if (OperatingSystem.IsWindows()) return new("win-x64", ".zip", "SharpEmu.exe");
+        if (OperatingSystem.IsLinux()) return new("linux-x64", ".tar.gz", "SharpEmu");
+        if (OperatingSystem.IsMacOS()) return new("osx-x64", ".tar.gz", "SharpEmu");
+        throw new PlatformNotSupportedException();
+    }
+
+    private sealed record PlatformInfo(string Rid, string Extension, string ExecutableName);
+}

src/SharpEmu.GUI/packages.lock.json

@@ -1,189 +0,0 @@
-{
-  "version": 2,
-  "dependencies": {
-    "net10.0": {
-      "Avalonia": {
-        "type": "Direct",
-        "requested": "[11.3.18, )",
-        "resolved": "11.3.18",
-        "contentHash": "2C4UxhWUObWGgYKWic1x5BMMWGJP6SElb91WeOxs+X/iR26rtkqpxFFwwo50FXS9AyYnHfk8QKXDEfe7oT/kZA==",
-        "dependencies": {
-          "Avalonia.BuildServices": "11.3.2",
-          "Avalonia.Remote.Protocol": "11.3.18",
-          "MicroCom.Runtime": "0.11.0"
-        }
-      },
-      "Avalonia.Desktop": {
-        "type": "Direct",
-        "requested": "[11.3.18, )",
-        "resolved": "11.3.18",
-        "contentHash": "bilMPa5vYiis6fbNovb6esKytBnOCEGojBa1XFegLCRHCP6g6PvZwS0XF/YOAGkENRlHG8dI7lohOpQ9bIkq1g==",
-        "dependencies": {
-          "Avalonia": "11.3.18",
-          "Avalonia.Native": "11.3.18",
-          "Avalonia.Skia": "11.3.18",
-          "Avalonia.Win32": "11.3.18",
-          "Avalonia.X11": "11.3.18"
-        }
-      },
-      "Avalonia.Fonts.Inter": {
-        "type": "Direct",
-        "requested": "[11.3.18, )",
-        "resolved": "11.3.18",
-        "contentHash": "27u6hB3Y2Ue586yjfeVakberY73VNQXtuKwe/P927XG1QPlhsfmOyifLHDDpSHG85Zl1x/Xv9IZ3+tk9FnjcZQ==",
-        "dependencies": {
-          "Avalonia": "11.3.18"
-        }
-      },
-      "Avalonia.Themes.Fluent": {
-        "type": "Direct",
-        "requested": "[11.3.18, )",
-        "resolved": "11.3.18",
-        "contentHash": "+Q/TJoynD0zNuu5w2gD+xcTl7GNKJFxlPYAndRLs/mTDrNbbsvv/271WyIysbMPsXSjCyBDp7RCZzQkpD6x5Bg==",
-        "dependencies": {
-          "Avalonia": "11.3.18"
-        }
-      },
-      "Tmds.DBus.Protocol": {
-        "type": "Direct",
-        "requested": "[0.21.3, )",
-        "resolved": "0.21.3",
-        "contentHash": "hDwB8WsQoyALQKqIbwzS68UKdlnafDm4T/DkO/JrA/YIneP/rKv96SxYPVXeh3FP4i/SXfShrYftKLtciJAIlw=="
-      },
-      "Avalonia.Angle.Windows.Natives": {
-        "type": "Transitive",
-        "resolved": "2.1.25547.20250602",
-        "contentHash": "ZL0VLc4s9rvNNFt19Pxm5UNAkmKNylugAwJPX9ulXZ6JWs/l6XZihPWWTyezaoNOVyEPU8YbURtW7XMAtqXH5A=="
-      },
-      "Avalonia.BuildServices": {
-        "type": "Transitive",
-        "resolved": "11.3.2",
-        "contentHash": "qHDToxto1e3hci5YqbG9n0Ty8mlp3zBUN5wT66wKqaDVzXyQ0do3EnRILd4Ke9jpvsktaPpgE0YjEk7hornryQ=="
-      },
-      "Avalonia.FreeDesktop": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "aUwv8BNruRUOaUfMu4U3uibIUS60/rSHgGOhd8zBkLkpxY3JFJvgRbeq5ZzHIyKXCuKi18PO00YHAgCarp3wdw==",
-        "dependencies": {
-          "Avalonia": "11.3.18",
-          "Tmds.DBus.Protocol": "0.21.3"
-        }
-      },
-      "Avalonia.Native": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "8g53DROFW6wVJAnTsE1Iu5bdZO5r0oqxbdbMQODs1QDYCK2IU/y/x/vQVKCYOvqxl4tXkzU9tExv8XqSGPWthQ==",
-        "dependencies": {
-          "Avalonia": "11.3.18"
-        }
-      },
-      "Avalonia.Remote.Protocol": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "vw+6ZfgTuu72dA9aVWn6u56t2nrBd5MoMU0wo/qI9XJAl/c0oYYphIvwLvJP1JorubQY4UE3d0ac8ULBhrGBiA=="
-      },
-      "Avalonia.Skia": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "/B4aXmNRNjG8I5U/a1xJI+bIi0XO6DDzS3mBrIKlVnJRY2CyZiUeESRQXLnIU77Z9TvqkUROs+D47s085YjFtA==",
-        "dependencies": {
-          "Avalonia": "11.3.18",
-          "HarfBuzzSharp": "8.3.1.1",
-          "HarfBuzzSharp.NativeAssets.Linux": "8.3.1.1",
-          "HarfBuzzSharp.NativeAssets.WebAssembly": "8.3.1.1",
-          "SkiaSharp": "2.88.9",
-          "SkiaSharp.NativeAssets.Linux": "2.88.9",
-          "SkiaSharp.NativeAssets.WebAssembly": "2.88.9"
-        }
-      },
-      "Avalonia.Win32": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "eioUHkM2PeLPETd1aEks3rvb9plbba6buIrNdrqCpwE/qgHKUjvRNBd5mUQfAbGgTLiAes524gB8uUMDhrsJVQ==",
-        "dependencies": {
-          "Avalonia": "11.3.18",
-          "Avalonia.Angle.Windows.Natives": "2.1.25547.20250602"
-        }
-      },
-      "Avalonia.X11": {
-        "type": "Transitive",
-        "resolved": "11.3.18",
-        "contentHash": "m4Ki/G5Dovnq+6QzfS0iGbK8V77Q6oTjToMLOB0CxPCCrl3Oxywh6kIjuGJDPaN6kopMmjxlNShyQf+vPYL+JA==",
-        "dependencies": {
-          "Avalonia": "11.3.18",
-          "Avalonia.FreeDesktop": "11.3.18",
-          "Avalonia.Skia": "11.3.18"
-        }
-      },
-      "HarfBuzzSharp": {
-        "type": "Transitive",
-        "resolved": "8.3.1.1",
-        "contentHash": "tLZN66oe/uiRPTZfrCU4i8ScVGwqHNh5MHrXj0yVf4l7Mz0FhTGnQ71RGySROTmdognAs0JtluHkL41pIabWuQ==",
-        "dependencies": {
-          "HarfBuzzSharp.NativeAssets.Win32": "8.3.1.1",
-          "HarfBuzzSharp.NativeAssets.macOS": "8.3.1.1"
-        }
-      },
-      "HarfBuzzSharp.NativeAssets.Linux": {
-        "type": "Transitive",
-        "resolved": "8.3.1.1",
-        "contentHash": "3EZ1mpIiKWRLL5hUYA82ZHteeDIVaEA/Z0rA/wU6tjx6crcAkJnBPwDXZugBSfo8+J3EznvRJf49uMsqYfKrHg=="
-      },
-      "HarfBuzzSharp.NativeAssets.macOS": {
-        "type": "Transitive",
-        "resolved": "8.3.1.1",
-        "contentHash": "jbtCsgftcaFLCA13tVKo5iWdElJScrulLTKJre36O4YQTIlwDtPPqhRZNk+Y0vv4D1gxbscasGRucUDfS44ofQ=="
-      },
-      "HarfBuzzSharp.NativeAssets.WebAssembly": {
-        "type": "Transitive",
-        "resolved": "8.3.1.1",
-        "contentHash": "loJweK2u/mH/3C2zBa0ggJlITIszOkK64HLAZB7FUT670dTg965whLFYHDQo69NmC4+d9UN0icLC9VHidXaVCA=="
-      },
-      "HarfBuzzSharp.NativeAssets.Win32": {
-        "type": "Transitive",
-        "resolved": "8.3.1.1",
-        "contentHash": "UsJtQsfAJoFDZrXc4hCUfRPMqccfKZ0iumJ/upcUjz/cmsTgVFGNEL5yaJWmkqsuFYdMWbj/En5/kS4PFl9hBA=="
-      },
-      "MicroCom.Runtime": {
-        "type": "Transitive",
-        "resolved": "0.11.0",
-        "contentHash": "MEnrZ3UIiH40hjzMDsxrTyi8dtqB5ziv3iBeeU4bXsL/7NLSal9F1lZKpK+tfBRnUoDSdtcW3KufE4yhATOMCA=="
-      },
-      "SkiaSharp": {
-        "type": "Transitive",
-        "resolved": "2.88.9",
-        "contentHash": "3MD5VHjXXieSHCleRLuaTXmL2pD0mB7CcOB1x2kA1I4bhptf4e3R27iM93264ZYuAq6mkUyX5XbcxnZvMJYc1Q==",
-        "dependencies": {
-          "SkiaSharp.NativeAssets.Win32": "2.88.9",
-          "SkiaSharp.NativeAssets.macOS": "2.88.9"
-        }
-      },
-      "SkiaSharp.NativeAssets.Linux": {
-        "type": "Transitive",
-        "resolved": "2.88.9",
-        "contentHash": "cWSaJKVPWAaT/WIn9c8T5uT/l4ETwHxNJTkEOtNKjphNo8AW6TF9O32aRkxqw3l8GUdUo66Bu7EiqtFh/XG0Zg==",
-        "dependencies": {
-          "SkiaSharp": "2.88.9"
-        }
-      },
-      "SkiaSharp.NativeAssets.macOS": {
-        "type": "Transitive",
-        "resolved": "2.88.9",
-        "contentHash": "Nv5spmKc4505Ep7oUoJ5vp3KweFpeNqxpyGDWyeEPTX2uR6S6syXIm3gj75dM0YJz7NPvcix48mR5laqs8dPuA=="
-      },
-      "SkiaSharp.NativeAssets.WebAssembly": {
-        "type": "Transitive",
-        "resolved": "2.88.9",
-        "contentHash": "kt06RccBHSnAs2wDYdBSfsjIDbY3EpsOVqnlDgKdgvyuRA8ZFDaHRdWNx1VHjGgYzmnFCGiTJBnXFl5BqGwGnA=="
-      },
-      "SkiaSharp.NativeAssets.Win32": {
-        "type": "Transitive",
-        "resolved": "2.88.9",
-        "contentHash": "wb2kYgU7iy84nQLYZwMeJXixvK++GoIuECjU4ECaUKNuflyRlJKyiRhN1MAHswvlvzuvkrjRWlK0Za6+kYQK7w=="
-      },
-      "sharpemu.logging": {
-        "type": "Project"
-      }
-    }
-  }
-}

src/SharpEmu.HLE/CpuContext.cs

@@ -1,6 +1,7 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+using System.Buffers;
 using System.Buffers.Binary;
 using System.Text;
 
@@ -25,6 +26,12 @@ public sealed class CpuContext(ICpuMemory memory, Generation generation)
 
     public ulong GsBase { get; set; }
 
+    /// <summary>x87 control word observed at the current guest boundary.</summary>
+    public ushort FpuControlWord { get; set; } = 0x037F;
+
+    /// <summary>MXCSR observed at the current guest boundary.</summary>
+    public uint Mxcsr { get; set; } = 0x1F80;
+
     public ulong this[CpuRegister register]
     {
         get => _registers[(int)register];
@@ -238,23 +245,63 @@ public sealed class CpuContext(ICpuMemory memory, Generation generation)
             return false;
         }
 
-        var bytes = new byte[capacity];
-        for (var index = 0; index < bytes.Length; index++)
+        const int StackBufferLength = 512;
+        const int ReadChunkLength = 128;
+        var rented = capacity > StackBufferLength ? ArrayPool<byte>.Shared.Rent(capacity) : null;
+        Span<byte> bytes = rented is null ? stackalloc byte[StackBufferLength] : rented;
+        try
         {
-            if (!Memory.TryRead(address + (ulong)index, bytes.AsSpan(index, 1)))
+            var length = 0;
+            while (length < capacity)
             {
-                return false;
+                // Bulk-read in bounded chunks rather than the full capacity: the string
+                // may end just before unmapped memory, and overreading past the
+                // terminator by more than a chunk could fault where the old
+                // byte-by-byte loop succeeded.
+                var chunk = Math.Min(ReadChunkLength, capacity - length);
+                var span = bytes.Slice(length, chunk);
+                if (Memory.TryRead(address + (ulong)length, span))
+                {
+                    var terminator = span.IndexOf((byte)0);
+                    if (terminator >= 0)
+                    {
+                        value = Encoding.UTF8.GetString(bytes[..(length + terminator)]);
+                        return true;
+                    }
+
+                    length += chunk;
+                    continue;
+                }
+
+                // The chunk touches an unreadable range; fall back to per-byte reads so a
+                // terminator sitting before the bad byte still yields the string.
+                for (var i = 0; i < chunk; i++)
+                {
+                    if (!Memory.TryRead(address + (ulong)(length + i), bytes.Slice(length + i, 1)))
+                    {
+                        return false;
+                    }
+
+                    if (bytes[length + i] == 0)
+                    {
+                        value = Encoding.UTF8.GetString(bytes[..(length + i)]);
+                        return true;
+                    }
+                }
+
+                length += chunk;
             }
 
-            if (bytes[index] == 0)
+            value = Encoding.UTF8.GetString(bytes[..capacity]);
+            return true;
+        }
+        finally
+        {
+            if (rented is not null)
             {
-                value = Encoding.UTF8.GetString(bytes, 0, index);
-                return true;
+                ArrayPool<byte>.Shared.Return(rented);
             }
         }
-
-        value = Encoding.UTF8.GetString(bytes);
-        return true;
     }
 
     public bool PushUInt64(ulong value)

src/SharpEmu.HLE/GuestCStringAttribute.cs

@@ -0,0 +1,20 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE;
+
+/// <summary>
+/// Marks a string parameter of a [SysAbiExport] handler as a guest null-terminated
+/// UTF-8 pointer. The generated thunk reads the string from the argument register's
+/// address (bounded by <see cref="MaxLength"/>) before invoking the handler, and
+/// returns ORBIS_GEN2_ERROR_MEMORY_FAULT to the guest when the read fails — including
+/// a null pointer, matching TryReadNullTerminatedUtf8's contract.
+/// </summary>
+[AttributeUsage(AttributeTargets.Parameter, Inherited = false, AllowMultiple = false)]
+public sealed class GuestCStringAttribute : Attribute
+{
+    public GuestCStringAttribute(int maxLength) => MaxLength = maxLength;
+
+    /// <summary>Upper bound in bytes for the guest read, terminator included.</summary>
+    public int MaxLength { get; }
+}

src/SharpEmu.HLE/GuestImageWriteTracker.cs

@@ -0,0 +1,613 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Globalization;
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE;
+
+/// <summary>
+/// Detects guest CPU writes into memory that backs a host GPU image. On PS5
+/// render targets alias unified memory, so games freely mix CPU writes and GPU
+/// draws on the same surface (Chowdren titles memset their fog layers every
+/// frame). Host GPU images are separate storage, so the video backend needs to
+/// know when the guest CPU rewrote a surface to re-upload it. Ranges are
+/// write-protected; the first write faults, the fault handler restores write
+/// access and marks the range dirty, and the video backend consumes the dirty
+/// flag once per flip and re-arms protection after re-uploading.
+/// </summary>
+public static unsafe class GuestImageWriteTracker
+{
+    private const int ProtRead = 0x1;
+    private const int ProtWrite = 0x2;
+    private const int ClockMonotonicRaw = 4;
+
+    private sealed class TrackedRange
+    {
+        public ulong Address;
+        public ulong ByteCount;
+        public ulong Start;
+        public ulong End;
+        public int Dirty;
+        public int Armed;
+        public int FirstCpuWriteSeen;
+        public int PendingFirstCpuWrite;
+        public bool TraceLifetime;
+        public long SourceSequence;
+        public long FirstCpuWriteTraceSequence;
+        public long FirstCpuWriteTimestampNanoseconds;
+        public ulong FirstCpuWriteAddress;
+        public ulong FirstCpuWritePage;
+        public string Source = "unspecified";
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    private struct Timespec
+    {
+        public long Seconds;
+        public long Nanoseconds;
+    }
+
+    private static readonly object _gate = new();
+    private static readonly Dictionary<ulong, TrackedRange> _rangesByAddress = new();
+
+    // Snapshot array read lock-free from the signal handler; rebuilt on every
+    // mutation under the gate. Signal handlers must not take managed locks.
+    private static TrackedRange[] _rangeSnapshot = [];
+
+    private static readonly bool _enabled = !OperatingSystem.IsWindows() &&
+        Environment.GetEnvironmentVariable("SHARPEMU_GUEST_IMAGE_CPU_SYNC") != "0";
+    private static readonly (bool Wildcard, ulong[] Addresses) _lifetimeTraceFilter =
+        ParseAddressList(Environment.GetEnvironmentVariable("SHARPEMU_TRACE_GUEST_IMAGE_ADDRS"));
+    private static readonly (bool Wildcard, string[] Sources) _lifetimeSourceTraceFilter =
+        ParseSourceList(Environment.GetEnvironmentVariable("SHARPEMU_TRACE_GUEST_MEMORY_LIFETIME"));
+    private static readonly bool _lifetimeTraceEnabled =
+        _lifetimeTraceFilter.Wildcard ||
+        _lifetimeTraceFilter.Addresses.Length != 0 ||
+        _lifetimeSourceTraceFilter.Wildcard ||
+        _lifetimeSourceTraceFilter.Sources.Length != 0;
+    private static readonly long _lifetimeTraceEpochNanoseconds =
+        _enabled && _lifetimeTraceEnabled ? GetMonotonicNanoseconds() : 0;
+    private static long _lifetimeTraceSequence;
+
+    [DllImport("libc", EntryPoint = "mprotect", SetLastError = true)]
+    private static extern int Mprotect(nint address, nuint length, int protection);
+
+    [DllImport("libc", EntryPoint = "clock_gettime", SetLastError = false)]
+    private static extern int ClockGetTime(int clockId, Timespec* time);
+
+    public static bool Enabled => _enabled;
+
+    /// <summary>
+    /// Exercises the fault-handling path once outside signal context so every
+    /// branch is JIT-compiled (and, under Rosetta 2, translated) before a real
+    /// fault arrives — a cold signal path is silently never entered there.
+    /// </summary>
+    public static void WarmUp()
+    {
+        if (!_enabled)
+        {
+            return;
+        }
+
+        var scratch = NativeMemory.AllocZeroed(4096);
+        try
+        {
+            // Warm the timestamp P/Invoke used by the signal-safe scalar
+            // capture path before a real protected-page write reaches it.
+            _ = GetMonotonicNanoseconds();
+            var address = (ulong)scratch;
+            Track(address, 4096);
+            _ = TryHandleWriteFault(address);
+            _ = ConsumeDirty(address);
+            Untrack(address);
+        }
+        finally
+        {
+            NativeMemory.Free(scratch);
+        }
+    }
+
+    /// <summary>Registers a range and arms write protection on it.</summary>
+    public static void Track(
+        ulong address,
+        ulong byteCount,
+        long sourceSequence = 0,
+        string source = "unspecified")
+    {
+        if (!_enabled || address == 0 || byteCount == 0)
+        {
+            return;
+        }
+
+        var (start, length) = PageAlign(address, byteCount);
+        lock (_gate)
+        {
+            _rangesByAddress.TryGetValue(address, out var range);
+            if (range is not null &&
+                (range.Start != start ||
+                 range.End != start + length ||
+                 range.ByteCount != byteCount))
+            {
+                // Never resize an object that is still reachable from the
+                // signal handler's lock-free snapshot. Retire it and publish
+                // a fresh immutable range.
+                DisarmLocked(range, "replace-range");
+                _rangesByAddress.Remove(address);
+                range = null;
+            }
+
+            if (range is null)
+            {
+                range = new TrackedRange
+                {
+                    Address = address,
+                    ByteCount = byteCount,
+                    Start = start,
+                    End = start + length,
+                    TraceLifetime =
+                        ShouldTraceRange(start, start + length) || ShouldTraceSource(source),
+                    SourceSequence = sourceSequence,
+                    Source = source,
+                };
+                _rangesByAddress[address] = range;
+                RebuildSnapshotLocked();
+            }
+            else
+            {
+                FlushPendingFirstCpuWrite(range);
+            }
+
+            range.SourceSequence = sourceSequence;
+            range.Source = source;
+            range.TraceLifetime =
+                ShouldTraceRange(range.Start, range.End) || ShouldTraceSource(source);
+            ArmLocked(range, "arm");
+        }
+    }
+
+    public static void Untrack(ulong address)
+    {
+        if (!_enabled)
+        {
+            return;
+        }
+
+        lock (_gate)
+        {
+            if (_rangesByAddress.TryGetValue(address, out var range))
+            {
+                DisarmLocked(range, "untrack");
+                _rangesByAddress.Remove(address);
+                RebuildSnapshotLocked();
+            }
+        }
+    }
+
+    /// <summary>
+    /// Returns true when the guest CPU wrote the range since the last call,
+    /// clearing the flag. The caller re-arms via <see cref="Rearm"/> after it
+    /// finished reading the guest bytes.
+    /// </summary>
+    public static bool ConsumeDirty(ulong address)
+    {
+        if (!_enabled)
+        {
+            return false;
+        }
+
+        lock (_gate)
+        {
+            if (!_rangesByAddress.TryGetValue(address, out var range))
+            {
+                return false;
+            }
+
+            FlushPendingFirstCpuWrite(range);
+            return Interlocked.Exchange(ref range.Dirty, 0) != 0;
+        }
+    }
+
+    /// <summary>
+    /// Non-consuming variant of <see cref="ConsumeDirty"/>: reports whether
+    /// the range has been written since it was last re-armed, leaving the
+    /// flag for the owner that evicts and re-uploads.
+    /// </summary>
+    public static bool PeekDirty(ulong address)
+    {
+        if (!_enabled)
+        {
+            return false;
+        }
+
+        lock (_gate)
+        {
+            if (!_rangesByAddress.TryGetValue(address, out var range))
+            {
+                return false;
+            }
+
+            FlushPendingFirstCpuWrite(range);
+            return Volatile.Read(ref range.Dirty) != 0;
+        }
+    }
+
+    public static void Rearm(ulong address)
+    {
+        if (!_enabled)
+        {
+            return;
+        }
+
+        lock (_gate)
+        {
+            if (_rangesByAddress.TryGetValue(address, out var range))
+            {
+                ArmLocked(range, "rearm");
+            }
+        }
+    }
+
+    /// <summary>
+    /// Prepares pages touched by a managed HLE memory write. Native guest
+    /// stores fault and enter <see cref="TryHandleWriteFault"/> through the
+    /// POSIX signal bridge, but a managed Buffer.MemoryCopy into a protected
+    /// page is surfaced by the runtime as a fatal AccessViolation instead of
+    /// a resumable guest fault. Visit every page in the write span up front so
+    /// all overlapping texture owners are dirtied and made writable.
+    /// </summary>
+    public static void NotifyManagedWrite(ulong address, ulong byteCount)
+    {
+        if (!_enabled || address == 0 || byteCount == 0)
+        {
+            return;
+        }
+
+        var end = address > ulong.MaxValue - byteCount
+            ? ulong.MaxValue
+            : address + byteCount;
+        var candidate = address;
+        while (candidate < end)
+        {
+            _ = TryHandleWriteFault(candidate);
+            var nextPage = (candidate & ~0xFFFUL) + 0x1000UL;
+            if (nextPage <= candidate)
+            {
+                break;
+            }
+            candidate = nextPage;
+        }
+    }
+
+    /// <summary>
+    /// Flushes scalar first-write records captured by the POSIX signal handler.
+    /// Call only from ordinary managed execution, never from signal context.
+    /// </summary>
+    public static void FlushPendingDiagnostics()
+    {
+        if (!_enabled || !_lifetimeTraceEnabled)
+        {
+            return;
+        }
+
+        lock (_gate)
+        {
+            foreach (var range in _rangesByAddress.Values)
+            {
+                FlushPendingFirstCpuWrite(range);
+            }
+        }
+    }
+
+    /// <summary>
+    /// Signal-handler entry: if the fault address lies in a tracked, armed
+    /// range, restore write access, mark the range dirty, and return true so
+    /// the faulting write can be retried. Must not allocate or lock.
+    /// </summary>
+    public static bool TryHandleWriteFault(ulong faultAddress)
+    {
+        if (!_enabled || faultAddress == 0)
+        {
+            return false;
+        }
+
+        var ranges = Volatile.Read(ref _rangeSnapshot);
+        var writableStart = ulong.MaxValue;
+        var writableEnd = 0UL;
+        for (var index = 0; index < ranges.Length; index++)
+        {
+            var range = ranges[index];
+            if (faultAddress < range.Start || faultAddress >= range.End)
+            {
+                continue;
+            }
+
+            writableStart = Math.Min(writableStart, range.Start);
+            writableEnd = Math.Max(writableEnd, range.End);
+        }
+
+        if (writableStart == ulong.MaxValue)
+        {
+            return false;
+        }
+
+        // Ranges are page-aligned and may overlap (font atlases and other
+        // suballocations commonly share pages). Unprotecting one range also
+        // makes every overlapping tracked page writable. Expand to the full
+        // transitive overlap and dirty/disarm every owner, otherwise only the
+        // first dictionary entry observes the write and the others retain a
+        // stale cached texture indefinitely.
+        var expanded = true;
+        while (expanded)
+        {
+            expanded = false;
+            for (var index = 0; index < ranges.Length; index++)
+            {
+                var range = ranges[index];
+                if (range.Start >= writableEnd || range.End <= writableStart)
+                {
+                    continue;
+                }
+
+                var start = Math.Min(writableStart, range.Start);
+                var end = Math.Max(writableEnd, range.End);
+                if (start != writableStart || end != writableEnd)
+                {
+                    writableStart = start;
+                    writableEnd = end;
+                    expanded = true;
+                }
+            }
+        }
+
+        var needsUnprotect = false;
+        for (var index = 0; index < ranges.Length; index++)
+        {
+            var range = ranges[index];
+            if (range.Start < writableEnd && range.End > writableStart &&
+                Volatile.Read(ref range.Armed) != 0)
+            {
+                needsUnprotect = true;
+                break;
+            }
+        }
+
+        if (needsUnprotect &&
+            Mprotect(
+                (nint)writableStart,
+                (nuint)(writableEnd - writableStart),
+                ProtRead | ProtWrite) != 0)
+        {
+            return false;
+        }
+
+        for (var index = 0; index < ranges.Length; index++)
+        {
+            var range = ranges[index];
+            if (range.Start >= writableEnd || range.End <= writableStart)
+            {
+                continue;
+            }
+
+            var wasArmed = Interlocked.Exchange(ref range.Armed, 0) != 0;
+            if (wasArmed &&
+                range.TraceLifetime &&
+                Interlocked.CompareExchange(ref range.FirstCpuWriteSeen, 1, 0) == 0)
+            {
+                // Signal context: capture preallocated scalar fields only.
+                // Formatting and I/O are deferred to a locked safe path.
+                range.FirstCpuWriteTraceSequence =
+                    Interlocked.Increment(ref _lifetimeTraceSequence);
+                range.FirstCpuWriteTimestampNanoseconds = GetMonotonicNanoseconds();
+                range.FirstCpuWriteAddress = faultAddress;
+                range.FirstCpuWritePage = faultAddress & ~0xFFFUL;
+                Volatile.Write(ref range.PendingFirstCpuWrite, 1);
+                Volatile.Write(ref range.FirstCpuWriteSeen, 2);
+            }
+
+            Volatile.Write(ref range.Dirty, 1);
+        }
+
+        return true;
+    }
+
+    private static void ArmLocked(TrackedRange range, string operation)
+    {
+        FlushPendingFirstCpuWrite(range);
+        if (Interlocked.Exchange(ref range.Armed, 1) == 1)
+        {
+            return;
+        }
+
+        // A new publication/rearm starts a new first-write lifetime.
+        Volatile.Write(ref range.FirstCpuWriteSeen, 0);
+        var failed = Mprotect(
+            (nint)range.Start,
+            (nuint)(range.End - range.Start),
+            ProtRead) != 0;
+        if (failed)
+        {
+            Volatile.Write(ref range.Armed, 0);
+        }
+
+        if (range.TraceLifetime)
+        {
+            TraceLifetime(
+                range,
+                failed ? $"{operation}-failed-errno-{Marshal.GetLastPInvokeError()}" : operation);
+        }
+    }
+
+    private static void DisarmLocked(TrackedRange range, string operation)
+    {
+        FlushPendingFirstCpuWrite(range);
+        var wasArmed = Interlocked.Exchange(ref range.Armed, 0) == 1;
+        if (wasArmed)
+        {
+            _ = Mprotect(
+                (nint)range.Start,
+                (nuint)(range.End - range.Start),
+                ProtRead | ProtWrite);
+        }
+
+        if (range.TraceLifetime)
+        {
+            TraceLifetime(range, wasArmed ? operation : $"{operation}-already-disarmed");
+        }
+    }
+
+    private static void RebuildSnapshotLocked()
+    {
+        _rangeSnapshot = _rangesByAddress.Values.ToArray();
+    }
+
+    private static (ulong Start, ulong Length) PageAlign(ulong address, ulong byteCount)
+    {
+        const ulong pageMask = 0xFFFUL;
+        var start = address & ~pageMask;
+        var end = (address + byteCount + pageMask) & ~pageMask;
+        return (start, end - start);
+    }
+
+    private static bool ShouldTraceRange(ulong start, ulong end)
+    {
+        if (_lifetimeTraceFilter.Wildcard)
+        {
+            return true;
+        }
+
+        var addresses = _lifetimeTraceFilter.Addresses;
+        for (var index = 0; index < addresses.Length; index++)
+        {
+            if (addresses[index] >= start && addresses[index] < end)
+            {
+                return true;
+            }
+        }
+
+        return false;
+    }
+
+    private static (bool Wildcard, ulong[] Addresses) ParseAddressList(string? addresses)
+    {
+        if (string.IsNullOrWhiteSpace(addresses))
+        {
+            return (false, []);
+        }
+
+        var parsedAddresses = new List<ulong>();
+        foreach (var token in addresses.Split(
+                     [',', ';', ' ', '\t'],
+                     StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))
+        {
+            if (token == "*")
+            {
+                return (true, []);
+            }
+
+            var span = token.AsSpan();
+            if (span.StartsWith("0x", StringComparison.OrdinalIgnoreCase))
+            {
+                span = span[2..];
+            }
+
+            if (ulong.TryParse(span, NumberStyles.HexNumber, CultureInfo.InvariantCulture, out var parsed))
+            {
+                parsedAddresses.Add(parsed);
+            }
+        }
+
+        return (false, parsedAddresses.ToArray());
+    }
+
+    private static bool ShouldTraceSource(string source)
+    {
+        if (_lifetimeSourceTraceFilter.Wildcard)
+        {
+            return true;
+        }
+
+        return Array.IndexOf(_lifetimeSourceTraceFilter.Sources, source) >= 0;
+    }
+
+    private static (bool Wildcard, string[] Sources) ParseSourceList(string? sources)
+    {
+        if (string.IsNullOrWhiteSpace(sources))
+        {
+            return (false, []);
+        }
+
+        var parsedSources = new List<string>();
+        foreach (var token in sources.Split(
+                     [',', ';'],
+                     StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))
+        {
+            if (token == "*")
+            {
+                return (true, []);
+            }
+
+            parsedSources.Add(token);
+        }
+
+        return (false, parsedSources.ToArray());
+    }
+
+    private static void FlushPendingFirstCpuWrite(TrackedRange range)
+    {
+        var spin = new SpinWait();
+        while (Volatile.Read(ref range.FirstCpuWriteSeen) == 1)
+        {
+            spin.SpinOnce();
+        }
+
+        if (!range.TraceLifetime || Interlocked.Exchange(ref range.PendingFirstCpuWrite, 0) == 0)
+        {
+            return;
+        }
+
+        TraceLifetime(
+            range,
+            "first-cpu-write-disarm",
+            range.FirstCpuWriteAddress,
+            range.FirstCpuWritePage,
+            range.FirstCpuWriteTraceSequence,
+            range.FirstCpuWriteTimestampNanoseconds);
+    }
+
+    private static void TraceLifetime(
+        TrackedRange range,
+        string operation,
+        ulong faultAddress = 0,
+        ulong faultPage = 0,
+        long traceSequence = 0,
+        long timestampNanoseconds = 0)
+    {
+        if (traceSequence == 0)
+        {
+            traceSequence = Interlocked.Increment(ref _lifetimeTraceSequence);
+        }
+
+        if (timestampNanoseconds == 0)
+        {
+            timestampNanoseconds = GetMonotonicNanoseconds();
+        }
+
+        var elapsedMilliseconds =
+            (timestampNanoseconds - _lifetimeTraceEpochNanoseconds) / 1_000_000.0;
+        Console.Error.WriteLine(
+            $"[WT][LIFETIME] seq={traceSequence} t_ms={elapsedMilliseconds:F3} " +
+            $"event={operation} source_seq={range.SourceSequence} source='{range.Source}' " +
+            $"requested=0x{range.Address:X16}+0x{range.ByteCount:X} " +
+            $"range=0x{range.Start:X16}..0x{range.End:X16} " +
+            $"fault=0x{faultAddress:X16} page=0x{faultPage:X16}");
+    }
+
+    private static long GetMonotonicNanoseconds()
+    {
+        Timespec time;
+        return ClockGetTime(ClockMonotonicRaw, &time) == 0
+            ? unchecked((time.Seconds * 1_000_000_000L) + time.Nanoseconds)
+            : 0;
+    }
+}

src/SharpEmu.HLE/GuestThreadExecution.cs

@@ -23,10 +23,32 @@ public readonly record struct GuestThreadSnapshot(
     ulong LastReturnRip,
     string? BlockReason);
 
+/// <summary>
+/// Continuation state for a blocked guest thread, replacing the closure pair a blocking
+/// wait used to allocate. TryWake runs under the scheduler's guest-thread gate and
+/// returns true when the waiter has a final result and the thread should be re-readied;
+/// false leaves it parked. Resume runs later on the woken thread outside that gate, and
+/// its return value becomes the guest's RAX for the resumed call.
+/// </summary>
+public interface IGuestThreadBlockWaiter
+{
+    int Resume();
+
+    bool TryWake();
+}
+
 public interface IGuestThreadScheduler
 {
     bool SupportsGuestContextTransfer { get; }
 
+    /// <summary>
+    /// Associates a pthread identity created on the primary guest executor
+    /// with its live CPU context. Primary execution does not pass through
+    /// TryStartThread, but kernel exception delivery must still be able to
+    /// target it.
+    /// </summary>
+    void RegisterGuestThreadContext(ulong threadHandle, CpuContext context);
+
     bool TryStartThread(CpuContext creatorContext, GuestThreadStartRequest request, out string? error);
 
     bool TryJoinThread(
@@ -39,6 +61,19 @@ public interface IGuestThreadScheduler
 
     int WakeBlockedThreads(string wakeKey, int maxCount = int.MaxValue);
 
+    /// <summary>
+    /// Applies a new guest scheduling priority to a live thread, mapping it
+    /// onto the host thread if one is running. Returns false when the thread
+    /// handle is unknown.
+    /// </summary>
+    bool TrySetGuestThreadPriority(ulong guestThreadHandle, int guestPriority);
+
+    /// <summary>
+    /// Records a new affinity mask for a guest thread and re-applies it to
+    /// the host thread where the platform supports it.
+    /// </summary>
+    bool TrySetGuestThreadAffinity(ulong guestThreadHandle, ulong affinityMask);
+
     IReadOnlyList<GuestThreadSnapshot> SnapshotThreads();
 
     bool TryCallGuestFunction(
@@ -51,11 +86,36 @@ public interface IGuestThreadScheduler
         string reason,
         out string? error);
 
+    bool TryCallGuestFunction(
+        CpuContext callerContext,
+        ulong entryPoint,
+        ulong arg0,
+        ulong arg1,
+        ulong arg2,
+        ulong stackAddress,
+        ulong stackSize,
+        string reason,
+        out ulong returnValue,
+        out string? error);
+
     bool TryCallGuestContinuation(
         CpuContext callerContext,
         GuestCpuContinuation continuation,
         string reason,
         out string? error);
+
+    /// <summary>
+    /// Asynchronously invokes an installed kernel exception handler as the
+    /// target guest thread. This is used by IL2CPP's stop-the-world collector:
+    /// the handler acknowledges suspension and may remain blocked until the
+    /// collecting thread resumes it.
+    /// </summary>
+    bool TryRaiseGuestException(
+        CpuContext callerContext,
+        ulong threadHandle,
+        ulong handler,
+        int exceptionType,
+        out string? error);
 }
 
 public readonly record struct GuestImportCallFrame(
@@ -80,13 +140,34 @@ public readonly record struct GuestCpuContinuation(
     ulong Rdi,
     ulong R8,
     ulong R9,
+    ulong R10,
+    ulong R11,
     ulong R12,
     ulong R13,
     ulong R14,
-    ulong R15);
+    ulong R15,
+    ushort FpuControlWord,
+    uint Mxcsr,
+    bool RestoreFullFpuState);
 
 public static class GuestThreadExecution
 {
+    private sealed class DelegateGuestThreadBlockWaiter : IGuestThreadBlockWaiter
+    {
+        private readonly Func<int> _resume;
+        private readonly Func<bool> _tryWake;
+
+        public DelegateGuestThreadBlockWaiter(Func<int> resume, Func<bool> tryWake)
+        {
+            _resume = resume;
+            _tryWake = tryWake;
+        }
+
+        public int Resume() => _resume();
+
+        public bool TryWake() => _tryWake();
+    }
+
     [ThreadStatic]
     private static ulong _currentGuestThreadHandle;
 
@@ -106,10 +187,7 @@ public static class GuestThreadExecution
     private static string? _pendingBlockWakeKey;
 
     [ThreadStatic]
-    private static Func<int>? _pendingBlockResumeHandler;
-
-    [ThreadStatic]
-    private static Func<bool>? _pendingBlockWakeHandler;
+    private static IGuestThreadBlockWaiter? _pendingBlockWaiter;
 
     [ThreadStatic]
     private static long _pendingBlockDeadlineTimestamp;
@@ -157,8 +235,7 @@ public static class GuestThreadExecution
         _pendingBlockContinuationValid = false;
         _pendingBlockContinuation = default;
         _pendingBlockWakeKey = null;
-        _pendingBlockResumeHandler = null;
-        _pendingBlockWakeHandler = null;
+        _pendingBlockWaiter = null;
         _pendingBlockDeadlineTimestamp = 0;
         _pendingEntryExit = false;
         _pendingEntryExitValue = 0;
@@ -179,8 +256,7 @@ public static class GuestThreadExecution
         _pendingBlockContinuationValid = false;
         _pendingBlockContinuation = default;
         _pendingBlockWakeKey = null;
-        _pendingBlockResumeHandler = null;
-        _pendingBlockWakeHandler = null;
+        _pendingBlockWaiter = null;
         _pendingBlockDeadlineTimestamp = 0;
         _pendingEntryExit = false;
         _pendingEntryExitValue = 0;
@@ -211,8 +287,7 @@ public static class GuestThreadExecution
         CpuContext? context,
         string reason,
         string? wakeKey = null,
-        Func<int>? resumeHandler = null,
-        Func<bool>? wakeHandler = null,
+        IGuestThreadBlockWaiter? waiter = null,
         long blockDeadlineTimestamp = 0)
     {
         if (!IsGuestThread)
@@ -222,8 +297,7 @@ public static class GuestThreadExecution
 
         _pendingBlockReason = string.IsNullOrWhiteSpace(reason) ? "guest_thread_blocked" : reason;
         _pendingBlockWakeKey = string.IsNullOrWhiteSpace(wakeKey) ? _pendingBlockReason : wakeKey;
-        _pendingBlockResumeHandler = resumeHandler;
-        _pendingBlockWakeHandler = wakeHandler;
+        _pendingBlockWaiter = waiter;
         _pendingBlockDeadlineTimestamp = blockDeadlineTimestamp;
         if (context is not null && TryCaptureCurrentBlockContinuation(context, out var continuation))
         {
@@ -239,6 +313,23 @@ public static class GuestThreadExecution
         return true;
     }
 
+    // Compatibility bridge for exports that still describe blocked work as a
+    // resume/wake delegate pair. New hot paths should provide an
+    // IGuestThreadBlockWaiter directly to avoid allocating closures.
+    public static bool RequestCurrentThreadBlock(
+        CpuContext? context,
+        string reason,
+        string? wakeKey,
+        Func<int> resumeHandler,
+        Func<bool> wakeHandler,
+        long blockDeadlineTimestamp = 0) =>
+        RequestCurrentThreadBlock(
+            context,
+            reason,
+            wakeKey,
+            new DelegateGuestThreadBlockWaiter(resumeHandler, wakeHandler),
+            blockDeadlineTimestamp);
+
     public static bool TryConsumeCurrentThreadBlock(out string reason)
     {
         return TryConsumeCurrentThreadBlock(out reason, out _, out _);
@@ -255,7 +346,6 @@ public static class GuestThreadExecution
             out hasContinuation,
             out _,
             out _,
-            out _,
             out _);
     }
 
@@ -264,19 +354,50 @@ public static class GuestThreadExecution
         out GuestCpuContinuation continuation,
         out bool hasContinuation,
         out string wakeKey,
-        out Func<int>? resumeHandler,
-        out Func<bool>? wakeHandler)
+        out IGuestThreadBlockWaiter? waiter)
     {
         return TryConsumeCurrentThreadBlock(
             out reason,
             out continuation,
             out hasContinuation,
             out wakeKey,
-            out resumeHandler,
-            out wakeHandler,
+            out waiter,
             out _);
     }
 
+    public static bool TryConsumeCurrentThreadBlock(
+        out string reason,
+        out GuestCpuContinuation continuation,
+        out bool hasContinuation,
+        out string wakeKey,
+        out IGuestThreadBlockWaiter? waiter,
+        out long blockDeadlineTimestamp)
+    {
+        reason = _pendingBlockReason ?? string.Empty;
+        if (string.IsNullOrEmpty(reason))
+        {
+            continuation = default;
+            hasContinuation = false;
+            wakeKey = string.Empty;
+            waiter = null;
+            blockDeadlineTimestamp = 0;
+            return false;
+        }
+
+        continuation = _pendingBlockContinuation;
+        hasContinuation = _pendingBlockContinuationValid;
+        wakeKey = _pendingBlockWakeKey ?? reason;
+        waiter = _pendingBlockWaiter;
+        blockDeadlineTimestamp = _pendingBlockDeadlineTimestamp;
+        _pendingBlockReason = null;
+        _pendingBlockContinuation = default;
+        _pendingBlockContinuationValid = false;
+        _pendingBlockWakeKey = null;
+        _pendingBlockWaiter = null;
+        _pendingBlockDeadlineTimestamp = 0;
+        return true;
+    }
+
     public static bool TryConsumeCurrentThreadBlock(
         out string reason,
         out GuestCpuContinuation continuation,
@@ -286,32 +407,16 @@ public static class GuestThreadExecution
         out Func<bool>? wakeHandler,
         out long blockDeadlineTimestamp)
     {
-        reason = _pendingBlockReason ?? string.Empty;
-        if (string.IsNullOrEmpty(reason))
-        {
-            continuation = default;
-            hasContinuation = false;
-            wakeKey = string.Empty;
-            resumeHandler = null;
-            wakeHandler = null;
-            blockDeadlineTimestamp = 0;
-            return false;
-        }
-
-        continuation = _pendingBlockContinuation;
-        hasContinuation = _pendingBlockContinuationValid;
-        wakeKey = _pendingBlockWakeKey ?? reason;
-        resumeHandler = _pendingBlockResumeHandler;
-        wakeHandler = _pendingBlockWakeHandler;
-        blockDeadlineTimestamp = _pendingBlockDeadlineTimestamp;
-        _pendingBlockReason = null;
-        _pendingBlockContinuation = default;
-        _pendingBlockContinuationValid = false;
-        _pendingBlockWakeKey = null;
-        _pendingBlockResumeHandler = null;
-        _pendingBlockWakeHandler = null;
-        _pendingBlockDeadlineTimestamp = 0;
-        return true;
+        var consumed = TryConsumeCurrentThreadBlock(
+            out reason,
+            out continuation,
+            out hasContinuation,
+            out wakeKey,
+            out var waiter,
+            out blockDeadlineTimestamp);
+        resumeHandler = waiter is null ? null : waiter.Resume;
+        wakeHandler = waiter is null ? null : waiter.TryWake;
+        return consumed;
     }
 
     public static long ComputeDeadlineTimestamp(TimeSpan timeout)
@@ -360,10 +465,15 @@ public static class GuestThreadExecution
             context[CpuRegister.Rdi],
             context[CpuRegister.R8],
             context[CpuRegister.R9],
+            context[CpuRegister.R10],
+            context[CpuRegister.R11],
             context[CpuRegister.R12],
             context[CpuRegister.R13],
             context[CpuRegister.R14],
-            context[CpuRegister.R15]);
+            context[CpuRegister.R15],
+            context.FpuControlWord,
+            context.Mxcsr,
+            RestoreFullFpuState: false);
         return true;
     }
 

src/SharpEmu.HLE/GuestTlsTemplate.cs

@@ -0,0 +1,442 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+using System.Diagnostics;
+
+namespace SharpEmu.HLE;
+
+/// <summary>
+/// Process-wide registry of ELF PT_TLS templates and per-thread dynamic thread
+/// vectors. AMD64 uses TLS Variant II: startup/static module blocks precede the
+/// thread pointer, while modules introduced after a thread was initialized are
+/// represented by separately allocated DTV entries.
+/// </summary>
+public static class GuestTlsTemplate
+{
+    // Must match CpuDispatcher/DirectExecutionBackend's mapped prefix. PS5
+    // modules can require more than one host page of Variant II static TLS;
+    // Dreaming Sarah's startup image, for example, reaches 0x1870 bytes.
+    public const ulong StartupStaticTlsReservation = 0x10000UL;
+    private static readonly object _gate = new();
+    private static readonly SortedDictionary<ulong, ModuleTemplate> _modules = new();
+    private static readonly Dictionary<ulong, ThreadDtv> _threadDtvs = new();
+    private static ulong _staticTlsSize;
+    private static ulong _maximumAlignment = 1;
+    private static ulong _generation;
+
+    static GuestTlsTemplate()
+    {
+        RunTlsLayoutSelfChecks();
+    }
+
+    private sealed class ModuleTemplate
+    {
+        public required ulong ModuleId { get; init; }
+        public required byte[] InitImage { get; init; }
+        public required ulong MemorySize { get; init; }
+        public required ulong Alignment { get; init; }
+        public required ulong AlignmentBias { get; init; }
+        public required ulong StaticOffset { get; init; }
+    }
+
+    private sealed class ThreadDtv
+    {
+        public ulong Generation { get; set; }
+        public Dictionary<ulong, DtvEntry> Entries { get; } = new();
+        public nint DtvAllocationBase { get; set; }
+        public ulong DtvAddress { get; set; }
+    }
+
+    private sealed class DtvEntry
+    {
+        public required ulong Address { get; init; }
+        public nint AllocationBase { get; init; }
+    }
+
+    /// <summary>Main executable's initialized PT_TLS bytes.</summary>
+    public static byte[] InitImage
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return _modules.TryGetValue(1, out var module)
+                    ? (byte[])module.InitImage.Clone()
+                    : [];
+            }
+        }
+    }
+
+    /// <summary>Aligned size of the main executable's static TLS block.</summary>
+    public static ulong BlockSize
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return _modules.TryGetValue(1, out var module)
+                    ? module.StaticOffset
+                    : 0;
+            }
+        }
+    }
+
+    /// <summary>Main executable's PT_TLS alignment.</summary>
+    public static ulong Alignment
+    {
+        get
+        {
+            lock (_gate)
+            {
+                return _modules.TryGetValue(1, out var module)
+                    ? module.Alignment
+                    : 1;
+            }
+        }
+    }
+
+    /// <summary>Total Variant II static TLS span required below the TCB.</summary>
+    public static ulong StaticTlsSize
+    {
+        get { lock (_gate) { return _staticTlsSize; } }
+    }
+
+    /// <summary>Largest PT_TLS alignment required by the registered modules.</summary>
+    public static ulong MaximumAlignment
+    {
+        get { lock (_gate) { return _maximumAlignment; } }
+    }
+
+    /// <summary>Current DTV generation, incremented for every registered module.</summary>
+    public static ulong Generation
+    {
+        get { lock (_gate) { return _generation; } }
+    }
+
+    /// <summary>
+    /// Backwards-compatible main-module registration entry point.
+    /// </summary>
+    public static void Set(ReadOnlySpan<byte> initImage, ulong memorySize, ulong alignment)
+    {
+        Reset();
+        RegisterModule(1, initImage, memorySize, alignment, alignmentBias: 0);
+    }
+
+    /// <summary>
+    /// Registers one module's PT_TLS template and allocates its Variant II
+    /// static offset. Modules must be registered in loader module-ID order.
+    /// </summary>
+    public static ulong RegisterModule(
+        ulong moduleId,
+        ReadOnlySpan<byte> initImage,
+        ulong memorySize,
+        ulong alignment,
+        ulong alignmentBias = 0)
+    {
+        if (moduleId == 0)
+        {
+            throw new ArgumentOutOfRangeException(nameof(moduleId));
+        }
+
+        var normalizedAlignment = alignment == 0 ? 1 : alignment;
+        if ((normalizedAlignment & (normalizedAlignment - 1)) != 0)
+        {
+            throw new InvalidDataException($"PT_TLS alignment 0x{alignment:X} is not a power of two.");
+        }
+        if (memorySize > int.MaxValue || (ulong)initImage.Length > memorySize)
+        {
+            throw new InvalidDataException("PT_TLS template size is invalid or exceeds the supported process limit.");
+        }
+
+        lock (_gate)
+        {
+            if (_modules.TryGetValue(moduleId, out var existing))
+            {
+                if (existing.MemorySize != memorySize ||
+                    existing.Alignment != normalizedAlignment ||
+                    existing.AlignmentBias != alignmentBias ||
+                    !existing.InitImage.AsSpan().SequenceEqual(initImage))
+                {
+                    throw new InvalidOperationException($"TLS module {moduleId} was registered with a different template.");
+                }
+
+                return existing.StaticOffset;
+            }
+
+            // FreeBSD/AMD64 Variant II: choose the smallest offset satisfying
+            // offset - previous >= size and (-offset) % align == p_vaddr % align.
+            var staticOffset = CalculateStaticOffset(
+                _staticTlsSize,
+                memorySize,
+                normalizedAlignment,
+                alignmentBias);
+            if (staticOffset > StartupStaticTlsReservation)
+            {
+                throw new InvalidOperationException(
+                    $"Static TLS requires 0x{staticOffset:X} bytes, but startup maps only " +
+                    $"0x{StartupStaticTlsReservation:X} bytes below the thread pointer.");
+            }
+            _modules.Add(moduleId, new ModuleTemplate
+            {
+                ModuleId = moduleId,
+                InitImage = initImage.ToArray(),
+                MemorySize = memorySize,
+                Alignment = normalizedAlignment,
+                AlignmentBias = alignmentBias,
+                StaticOffset = staticOffset,
+            });
+            _staticTlsSize = staticOffset;
+            _maximumAlignment = Math.Max(_maximumAlignment, normalizedAlignment);
+            _generation++;
+            return staticOffset;
+        }
+    }
+
+    /// <summary>Returns the static TP-relative block offset for a module.</summary>
+    public static bool TryGetStaticOffset(ulong moduleId, out ulong staticOffset)
+    {
+        lock (_gate)
+        {
+            if (_modules.TryGetValue(moduleId, out var module))
+            {
+                staticOffset = module.StaticOffset;
+                return true;
+            }
+        }
+
+        staticOffset = 0;
+        return false;
+    }
+
+    /// <summary>Clears all module templates and frees dynamic DTV blocks.</summary>
+    public static void Reset()
+    {
+        lock (_gate)
+        {
+            foreach (var dtv in _threadDtvs.Values)
+            {
+                FreeDynamicEntries(dtv);
+            }
+
+            _threadDtvs.Clear();
+            _modules.Clear();
+            _staticTlsSize = 0;
+            _maximumAlignment = 1;
+            _generation++;
+        }
+    }
+
+    /// <summary>
+    /// Initializes every currently registered startup TLS block for a thread
+    /// and records their addresses in that thread's DTV.
+    /// </summary>
+    public static void SeedThreadBlock(CpuContext context, ulong threadPointer)
+    {
+        ArgumentNullException.ThrowIfNull(context);
+        if (threadPointer == 0)
+        {
+            return;
+        }
+
+        lock (_gate)
+        {
+            if (_threadDtvs.Remove(threadPointer, out var oldDtv))
+            {
+                FreeDynamicEntries(oldDtv);
+            }
+
+            var dtv = new ThreadDtv();
+            _threadDtvs.Add(threadPointer, dtv);
+            foreach (var module in _modules.Values)
+            {
+                dtv.Entries[module.ModuleId] = CreateStaticOrDynamicEntry(context, threadPointer, module);
+            }
+            RebuildGuestDtv(context, threadPointer, dtv);
+        }
+    }
+
+    /// <summary>
+    /// Implements the DTV lookup used by <c>__tls_get_addr</c>. Unknown module
+    /// IDs and offsets outside the module's TLS image are rejected with zero.
+    /// </summary>
+    public static ulong ResolveAddress(CpuContext context, ulong moduleId, ulong offset)
+    {
+        ArgumentNullException.ThrowIfNull(context);
+        if (context.FsBase == 0 || moduleId == 0)
+        {
+            return 0;
+        }
+
+        lock (_gate)
+        {
+            if (!_modules.TryGetValue(moduleId, out var module) || offset >= module.MemorySize)
+            {
+                return 0;
+            }
+
+            if (!_threadDtvs.TryGetValue(context.FsBase, out var dtv))
+            {
+                dtv = new ThreadDtv();
+                _threadDtvs.Add(context.FsBase, dtv);
+                foreach (var startupModule in _modules.Values)
+                {
+                    dtv.Entries[startupModule.ModuleId] = CreateStaticOrDynamicEntry(context, context.FsBase, startupModule);
+                }
+                RebuildGuestDtv(context, context.FsBase, dtv);
+            }
+
+            var entryAdded = false;
+            if (!dtv.Entries.TryGetValue(moduleId, out var entry))
+            {
+                // This module appeared after the thread's startup TLS layout was
+                // seeded. Model rtld's lazy DTV allocation instead of assuming
+                // unused space exists below the already-live thread pointer.
+                entry = CreateDynamicEntry(module);
+                dtv.Entries.Add(moduleId, entry);
+                entryAdded = true;
+            }
+
+            if (entryAdded || dtv.Generation != _generation)
+            {
+                RebuildGuestDtv(context, context.FsBase, dtv);
+            }
+
+            return checked(entry.Address + offset);
+        }
+    }
+
+    private static DtvEntry CreateStaticOrDynamicEntry(
+        CpuContext context,
+        ulong threadPointer,
+        ModuleTemplate module)
+    {
+        if (threadPointer >= module.StaticOffset)
+        {
+            var address = threadPointer - module.StaticOffset;
+            var zeroImage = module.MemorySize == 0 ? [] : new byte[(int)module.MemorySize];
+            if ((zeroImage.Length == 0 || context.Memory.TryWrite(address, zeroImage)) &&
+                (module.InitImage.Length == 0 || context.Memory.TryWrite(address, module.InitImage)))
+            {
+                return new DtvEntry { Address = address };
+            }
+        }
+
+        return CreateDynamicEntry(module);
+    }
+
+    private static DtvEntry CreateDynamicEntry(ModuleTemplate module)
+    {
+        var size = Math.Max(1UL, module.MemorySize);
+        var allocationSize = checked(size + module.Alignment - 1);
+        if (allocationSize > int.MaxValue)
+        {
+            throw new OutOfMemoryException("TLS module allocation exceeds the supported host allocation size.");
+        }
+
+        var allocationBase = Marshal.AllocHGlobal((int)allocationSize);
+        var alignedAddress = AlignUp(unchecked((ulong)allocationBase), module.Alignment);
+        Marshal.Copy(new byte[(int)size], 0, unchecked((nint)alignedAddress), (int)size);
+        if (module.InitImage.Length != 0)
+        {
+            Marshal.Copy(module.InitImage, 0, unchecked((nint)alignedAddress), module.InitImage.Length);
+        }
+
+        return new DtvEntry
+        {
+            Address = alignedAddress,
+            AllocationBase = allocationBase,
+        };
+    }
+
+    private static void FreeDynamicEntries(ThreadDtv dtv)
+    {
+        foreach (var entry in dtv.Entries.Values)
+        {
+            if (entry.AllocationBase != 0)
+            {
+                Marshal.FreeHGlobal(entry.AllocationBase);
+            }
+        }
+        if (dtv.DtvAllocationBase != 0)
+        {
+            Marshal.FreeHGlobal(dtv.DtvAllocationBase);
+            dtv.DtvAllocationBase = 0;
+            dtv.DtvAddress = 0;
+        }
+    }
+
+    private static void RebuildGuestDtv(CpuContext context, ulong threadPointer, ThreadDtv dtv)
+    {
+        var maximumModuleId = _modules.Count == 0 ? 0UL : _modules.Keys.Max();
+        var byteSize = checked(2UL * sizeof(ulong) + maximumModuleId * sizeof(ulong));
+        if (byteSize > int.MaxValue)
+        {
+            throw new OutOfMemoryException("Guest DTV exceeds the supported host allocation size.");
+        }
+
+        var newAllocation = Marshal.AllocHGlobal((int)Math.Max((ulong)sizeof(ulong) * 2, byteSize));
+        var newAddress = unchecked((ulong)newAllocation);
+        Marshal.Copy(new byte[(int)Math.Max((ulong)sizeof(ulong) * 2, byteSize)], 0, newAllocation, (int)Math.Max((ulong)sizeof(ulong) * 2, byteSize));
+        Marshal.WriteInt64(newAllocation, 0, unchecked((long)_generation));
+        Marshal.WriteInt64(newAllocation, sizeof(ulong), unchecked((long)maximumModuleId));
+        foreach (var (moduleId, entry) in dtv.Entries)
+        {
+            var slotOffset = checked((int)(2UL * sizeof(ulong) + (moduleId - 1) * sizeof(ulong)));
+            Marshal.WriteInt64(newAllocation, slotOffset, unchecked((long)entry.Address));
+        }
+
+        if (!context.TryWriteUInt64(threadPointer + sizeof(ulong), newAddress))
+        {
+            Marshal.FreeHGlobal(newAllocation);
+            throw new InvalidOperationException("Failed to install the guest DTV pointer in the thread control block.");
+        }
+
+        if (dtv.DtvAllocationBase != 0)
+        {
+            Marshal.FreeHGlobal(dtv.DtvAllocationBase);
+        }
+        dtv.DtvAllocationBase = newAllocation;
+        dtv.DtvAddress = newAddress;
+        dtv.Generation = _generation;
+    }
+
+    private static ulong AlignUp(ulong value, ulong alignment)
+    {
+        var mask = alignment - 1;
+        return checked((value + mask) & ~mask);
+    }
+
+    private static ulong CalculateStaticOffset(
+        ulong previousOffset,
+        ulong size,
+        ulong alignment,
+        ulong alignmentBias)
+    {
+        var result = checked(previousOffset + size + alignment - 1);
+        return result - ((result + alignmentBias) & (alignment - 1));
+    }
+
+    [Conditional("DEBUG")]
+    private static void RunTlsLayoutSelfChecks()
+    {
+        var firstOffset = RegisterModule(1, [0x11, 0x22], 0x20, 0x10, 0);
+        var secondOffset = RegisterModule(2, [0x7A], 0x18, 0x20, 8);
+        Debug.Assert(firstOffset == 0x20, "First Variant II TLS offset is incorrect.");
+        Debug.Assert(secondOffset >= firstOffset + 0x18, "TLS modules overlap in the static layout.");
+        Debug.Assert((unchecked(0UL - secondOffset) & 0x1F) == 8, "PT_TLS virtual-address congruence was not preserved.");
+
+        var lateEntry = CreateDynamicEntry(_modules[2]);
+        try
+        {
+            Debug.Assert(lateEntry.AllocationBase != 0, "A late TLS module did not receive a dynamic DTV block.");
+            Debug.Assert(Marshal.ReadByte(unchecked((nint)lateEntry.Address)) == 0x7A, "Late-module tdata was not initialized.");
+            Debug.Assert(Marshal.ReadByte(unchecked((nint)lateEntry.Address + 1)) == 0, "Late-module tbss was not zero initialized.");
+        }
+        finally
+        {
+            Marshal.FreeHGlobal(lateEntry.AllocationBase);
+            Reset();
+        }
+    }
+}

src/SharpEmu.HLE/HleDataSymbols.cs

@@ -13,7 +13,9 @@ public static class HleDataSymbols
     private const string LibcNeedFlagNid = "P330P3dFF68";
     private const string LibcInternalNeedFlagNid = "ZT4ODD2Ts9o";
     private const int ProgNameMaxBytes = 511;
-    private const ulong StackChkGuardValue = 0xC0DEC0DECAFEBABEUL;
+    // Terminator canaries reserve the low byte as NUL. Keep the process data
+    // symbol and every per-thread TLS copy byte-for-byte identical.
+    private const ulong StackChkGuardValue = 0xC0DEC0DECAFEBA00UL;
 
     private static readonly object _gate = new();
     private static readonly nint _stackChkGuardAddress = Allocate(sizeof(ulong) * 2);

src/SharpEmu.HLE/Host/HostGamepadState.cs

@@ -0,0 +1,46 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host;
+
+/// <summary>
+/// Host-neutral gamepad button flags. Named after the PlayStation layout the guest API
+/// exposes, but the numeric values are the seam's own — the HLE pad exports translate
+/// them to SCE_PAD_BUTTON bits, so guest ABI values never leak into host backends.
+/// </summary>
+[Flags]
+public enum HostGamepadButtons : uint
+{
+    None = 0,
+    Up = 1 << 0,
+    Down = 1 << 1,
+    Left = 1 << 2,
+    Right = 1 << 3,
+    Cross = 1 << 4,
+    Circle = 1 << 5,
+    Square = 1 << 6,
+    Triangle = 1 << 7,
+    L1 = 1 << 8,
+    R1 = 1 << 9,
+    L2 = 1 << 10,
+    R2 = 1 << 11,
+    L3 = 1 << 12,
+    R3 = 1 << 13,
+    Options = 1 << 14,
+    TouchPad = 1 << 15,
+}
+
+/// <summary>
+/// Snapshot of one host gamepad: sticks are 0..255 with 128 centered and Y growing
+/// downward; triggers 0..255. Unmanaged on purpose so per-frame polls can stackalloc
+/// snapshot buffers.
+/// </summary>
+public readonly record struct HostGamepadState(
+    bool Connected,
+    HostGamepadButtons Buttons,
+    byte LeftX,
+    byte LeftY,
+    byte RightX,
+    byte RightY,
+    byte LeftTrigger,
+    byte RightTrigger);

src/SharpEmu.HLE/Host/HostPlatform.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using System.Runtime.InteropServices;
+using SharpEmu.HLE.Host.Posix;
 using SharpEmu.HLE.Host.Windows;
 
 namespace SharpEmu.HLE.Host;
@@ -28,7 +29,14 @@ public static class HostPlatform
             return new WindowsHostPlatform();
         }
 
+        if ((OperatingSystem.IsLinux() || OperatingSystem.IsMacOS()) &&
+            RuntimeInformation.ProcessArchitecture == Architecture.X64)
+        {
+            return new PosixHostPlatform();
+        }
+
         throw new PlatformNotSupportedException(
-            "SharpEmu native guest execution requires a host platform backend and none exists for this OS/architecture yet (currently Windows x64 only).");
+            "SharpEmu native guest execution requires an x86-64 process on Windows, Linux, or macOS. " +
+            "On Apple Silicon, use the osx-x64 build under Rosetta 2.");
     }
 }

src/SharpEmu.HLE/Host/IHostAudioOutput.cs

@@ -0,0 +1,23 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host;
+
+/// <summary>
+/// Host audio-output device access. The HLE audio exports convert guest submissions to
+/// interleaved stereo 16-bit PCM (the format every backend accepts) and feed them through
+/// streams opened here; everything device-specific — queueing, backpressure, native
+/// buffer lifetime — lives behind <see cref="IHostAudioStream"/>.
+/// </summary>
+public interface IHostAudioOutput
+{
+    /// <summary>Backend identifier for diagnostics (e.g. "winmm").</summary>
+    string BackendName { get; }
+
+    /// <summary>
+    /// Opens an interleaved stereo 16-bit PCM output stream at the given sample rate.
+    /// Throws when the host has no usable output device; callers degrade to a silent
+    /// port and pace the guest instead.
+    /// </summary>
+    IHostAudioStream OpenStereoPcm16Stream(uint sampleRate);
+}

src/SharpEmu.HLE/Host/IHostAudioStream.cs

@@ -0,0 +1,18 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host;
+
+/// <summary>
+/// One open host audio output stream. Submissions are interleaved stereo 16-bit PCM at
+/// the sample rate the stream was opened with.
+/// </summary>
+public interface IHostAudioStream : IDisposable
+{
+    /// <summary>
+    /// Submits one buffer. May block briefly while the device drains its queue (this is
+    /// what paces the guest's audio loop); returns false when the stream cannot accept
+    /// audio, in which case the caller paces the guest itself.
+    /// </summary>
+    bool Submit(ReadOnlySpan<byte> stereoPcm16);
+}

src/SharpEmu.HLE/Host/IHostInput.cs

@@ -0,0 +1,44 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host;
+
+/// <summary>
+/// Host input devices: gamepad state snapshots, force-feedback/lightbar sinks, and the
+/// keyboard-fallback queries. Which physical readers exist (DualSense over raw HID,
+/// XInput, evdev, ...) is a backend detail; merge policy between devices and the
+/// keyboard lives in the HLE pad exports.
+/// </summary>
+public interface IHostInput
+{
+    /// <summary>Starts the background device readers once; safe to call repeatedly.</summary>
+    void EnsureStarted();
+
+    /// <summary>
+    /// Fills <paramref name="destination"/> with snapshots of currently connected
+    /// gamepads and returns how many were written (0 when none are connected).
+    /// </summary>
+    int GetGamepadStates(Span<HostGamepadState> destination);
+
+    /// <summary>Human-readable name of the first connected gamepad, or null.</summary>
+    string? DescribeConnectedGamepad();
+
+    /// <summary>Sets rumble on all connected gamepads; large = strong/left motor.</summary>
+    void SetRumble(byte largeMotor, byte smallMotor);
+
+    /// <summary>
+    /// Approximates per-trigger vibration on gamepads without independent trigger
+    /// actuators; null leaves that trigger's current value unchanged.
+    /// </summary>
+    void SetTriggerRumble(byte? leftTrigger, byte? rightTrigger);
+
+    void SetLightbar(byte red, byte green, byte blue);
+
+    void ResetLightbar();
+
+    /// <summary>True when a window of this process has keyboard focus.</summary>
+    bool IsHostWindowFocused();
+
+    /// <summary>Windows virtual-key code semantics; other backends translate.</summary>
+    bool IsKeyDown(int virtualKey);
+}

src/SharpEmu.HLE/Host/IHostPlatform.cs

@@ -16,4 +16,8 @@ public interface IHostPlatform
     IHostThreading Threading { get; }
 
     IHostSymbolResolver Symbols { get; }
+
+    IHostAudioOutput Audio { get; }
+
+    IHostInput Input { get; }
 }

src/SharpEmu.HLE/Host/IHostThreading.cs

@@ -24,6 +24,12 @@ public interface IHostThreading
 
     bool TrySetCurrentThreadAffinity(nuint affinityMask);
 
+    /// <summary>
+    /// Asks the OS for ~1 ms timed-wait granularity for the life of the process
+    /// (idempotent; best-effort). No-op on platforms whose default is already fine.
+    /// </summary>
+    void RequestTimerResolution();
+
     /// <summary>
     /// Creates a raw OS thread executing native code at <paramref name="entry"/> with
     /// <paramref name="stackReserveBytes"/> of reserved (not committed) stack.

src/SharpEmu.HLE/Host/Posix/PosixAlsaAudioStream.cs

@@ -0,0 +1,173 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// ALSA-based playback for Linux. The PCM device is opened in blocking mode
+/// with a device buffer sized to match the 32KB queue the other backends
+/// keep, so snd_pcm_writei itself provides the backpressure pacing. The
+/// "default" device routes through PulseAudio/PipeWire on desktops and to
+/// the hardware on bare ALSA setups; SHARPEMU_ALSA_DEVICE overrides it.
+/// </summary>
+internal sealed unsafe class PosixAlsaAudioStream : IHostAudioStream
+{
+    // 32KB of stereo PCM16 at 48kHz is ~170ms; keep the same device-side
+    // queue depth the WinMM/CoreAudio ports enforce in managed code.
+    private const uint DeviceLatencyMicroseconds = 170_000;
+    private const int StreamPlayback = 0;
+    private const int FormatS16LittleEndian = 2;
+    private const int AccessReadWriteInterleaved = 3;
+    private const int ErrorPipe = -32; // -EPIPE, underrun
+    private const int ErrorStreamPipe = -86; // -ESTRPIPE, suspended
+
+    private readonly object _gate = new();
+    private nint _pcm;
+    private bool _disposed;
+
+    public PosixAlsaAudioStream(uint sampleRate)
+    {
+        if (!OperatingSystem.IsLinux())
+        {
+            throw new PlatformNotSupportedException("ALSA audio is only available on Linux.");
+        }
+
+        var device = Environment.GetEnvironmentVariable("SHARPEMU_ALSA_DEVICE");
+        if (string.IsNullOrWhiteSpace(device))
+        {
+            device = "default";
+        }
+
+        var status = snd_pcm_open(out _pcm, device, StreamPlayback, 0);
+        if (status != 0)
+        {
+            throw new InvalidOperationException(
+                $"snd_pcm_open(\"{device}\") failed: {DescribeError(status)}.");
+        }
+
+        status = snd_pcm_set_params(
+            _pcm,
+            FormatS16LittleEndian,
+            AccessReadWriteInterleaved,
+            2,
+            sampleRate,
+            1,
+            DeviceLatencyMicroseconds);
+        if (status != 0)
+        {
+            _ = snd_pcm_close(_pcm);
+            _pcm = 0;
+            throw new InvalidOperationException(
+                $"snd_pcm_set_params({sampleRate} Hz) failed: {DescribeError(status)}.");
+        }
+    }
+
+    public bool Submit(ReadOnlySpan<byte> stereoPcm16)
+    {
+        lock (_gate)
+        {
+            if (_disposed)
+            {
+                return false;
+            }
+
+            return WritePcm(stereoPcm16, (uint)(stereoPcm16.Length / 4));
+        }
+    }
+
+    public void Dispose()
+    {
+        lock (_gate)
+        {
+            if (_disposed)
+            {
+                return;
+            }
+
+            _disposed = true;
+            if (_pcm != 0)
+            {
+                _ = snd_pcm_drop(_pcm);
+                _ = snd_pcm_close(_pcm);
+                _pcm = 0;
+            }
+        }
+    }
+
+    private bool WritePcm(ReadOnlySpan<byte> pcm, uint frames)
+    {
+        var recovered = false;
+        fixed (byte* data = pcm)
+        {
+            var offset = 0L;
+            while (offset < frames)
+            {
+                var written = snd_pcm_writei(
+                    _pcm,
+                    data + (offset * 4),
+                    (nuint)(frames - offset));
+                if (written >= 0)
+                {
+                    offset += written;
+                    continue;
+                }
+
+                // One recovery attempt per submit covers underruns (-EPIPE)
+                // and suspend/resume (-ESTRPIPE); anything else, or a second
+                // failure, drops the buffer rather than stalling the guest.
+                if (recovered ||
+                    (written != ErrorPipe && written != ErrorStreamPipe) ||
+                    snd_pcm_recover(_pcm, (int)written, 1) != 0)
+                {
+                    return false;
+                }
+
+                recovered = true;
+            }
+        }
+
+        return true;
+    }
+
+    private static string DescribeError(long status)
+    {
+        var message = Marshal.PtrToStringUTF8(snd_strerror((int)status));
+        return $"{message ?? "unknown error"} ({status})";
+    }
+
+    private const string Alsa = "libasound.so.2";
+
+    [DllImport(Alsa)]
+    private static extern int snd_pcm_open(
+        out nint pcm,
+        [MarshalAs(UnmanagedType.LPUTF8Str)] string name,
+        int stream,
+        int mode);
+
+    [DllImport(Alsa)]
+    private static extern int snd_pcm_set_params(
+        nint pcm,
+        int format,
+        int access,
+        uint channels,
+        uint rate,
+        int softResample,
+        uint latencyUs);
+
+    [DllImport(Alsa)]
+    private static extern long snd_pcm_writei(nint pcm, byte* buffer, nuint frames);
+
+    [DllImport(Alsa)]
+    private static extern int snd_pcm_recover(nint pcm, int error, int silent);
+
+    [DllImport(Alsa)]
+    private static extern int snd_pcm_drop(nint pcm);
+
+    [DllImport(Alsa)]
+    private static extern int snd_pcm_close(nint pcm);
+
+    [DllImport(Alsa)]
+    private static extern nint snd_strerror(int error);
+}

src/SharpEmu.HLE/Host/Posix/PosixCoreAudioStream.cs

@@ -0,0 +1,261 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// AudioQueue-based playback for macOS. Buffers are enqueued as stereo PCM16
+/// and returned by the queue's internal thread through the output callback;
+/// Submit applies the same 32KB backpressure the WinMM backend uses so guest
+/// pacing works identically.
+/// </summary>
+internal sealed unsafe class PosixCoreAudioStream : IHostAudioStream
+{
+    private const int MaximumQueuedPcmBytes = 32 * 1024;
+    private const uint FormatLinearPcm = 0x6C70636D; // 'lpcm'
+    private const uint FlagIsSignedInteger = 0x4;
+    private const uint FlagIsPacked = 0x8;
+
+    private readonly object _gate = new();
+    private readonly AutoResetEvent _completion = new(false);
+    private readonly Queue<nint> _freeBuffers = new();
+    private GCHandle _selfHandle;
+    private nint _queue;
+    private int _queuedPcmBytes;
+    private bool _started;
+    private bool _disposed;
+
+    public PosixCoreAudioStream(uint sampleRate)
+    {
+        if (!OperatingSystem.IsMacOS())
+        {
+            throw new PlatformNotSupportedException("CoreAudio is only available on macOS.");
+        }
+
+        var format = new AudioStreamBasicDescription
+        {
+            SampleRate = sampleRate,
+            FormatId = FormatLinearPcm,
+            FormatFlags = FlagIsSignedInteger | FlagIsPacked,
+            BytesPerPacket = 4,
+            FramesPerPacket = 1,
+            BytesPerFrame = 4,
+            ChannelsPerFrame = 2,
+            BitsPerChannel = 16,
+        };
+
+        _selfHandle = GCHandle.Alloc(this);
+        var status = AudioQueueNewOutput(
+            &format,
+            &OutputCallback,
+            GCHandle.ToIntPtr(_selfHandle),
+            0,
+            0,
+            0,
+            out _queue);
+        if (status != 0)
+        {
+            _selfHandle.Free();
+            throw new InvalidOperationException($"AudioQueueNewOutput failed with OSStatus {status}.");
+        }
+    }
+
+    public bool Submit(ReadOnlySpan<byte> stereoPcm16)
+    {
+        lock (_gate)
+        {
+            if (_disposed || _queue == 0)
+            {
+                return false;
+            }
+
+            var outputLength = stereoPcm16.Length;
+            while (_queuedPcmBytes != 0 &&
+                   _queuedPcmBytes + outputLength > MaximumQueuedPcmBytes)
+            {
+                Monitor.Exit(_gate);
+                try
+                {
+                    // Dispose can free the event while this thread waits
+                    // outside the gate; treat that like a timed-out wait.
+                    if (!_completion.WaitOne(TimeSpan.FromSeconds(1)))
+                    {
+                        return false;
+                    }
+                }
+                catch (ObjectDisposedException)
+                {
+                    return false;
+                }
+                finally
+                {
+                    Monitor.Enter(_gate);
+                }
+
+                if (_disposed)
+                {
+                    return false;
+                }
+            }
+
+            if (!TryTakeBuffer(outputLength, out var buffer))
+            {
+                return false;
+            }
+
+            var audioData = ((AudioQueueBuffer*)buffer)->AudioData;
+            stereoPcm16.CopyTo(new Span<byte>(audioData, outputLength));
+
+            ((AudioQueueBuffer*)buffer)->AudioDataByteSize = (uint)outputLength;
+            if (AudioQueueEnqueueBuffer(_queue, buffer, 0, 0) != 0)
+            {
+                _freeBuffers.Enqueue(buffer);
+                return false;
+            }
+
+            _queuedPcmBytes += outputLength;
+            if (!_started)
+            {
+                if (AudioQueueStart(_queue, 0) != 0)
+                {
+                    // A queue that never starts never drains, so later
+                    // submits would block on backpressure until their
+                    // timeout. Tear the queue down and fail fast instead.
+                    _ = AudioQueueDispose(_queue, true);
+                    _queue = 0;
+                    _queuedPcmBytes = 0;
+                    _freeBuffers.Clear();
+                    return false;
+                }
+
+                _started = true;
+            }
+
+            return true;
+        }
+    }
+
+    public void Dispose()
+    {
+        lock (_gate)
+        {
+            if (_disposed)
+            {
+                return;
+            }
+
+            _disposed = true;
+            if (_queue != 0)
+            {
+                // Synchronous dispose stops the queue, frees its buffers, and
+                // guarantees no further callbacks reference this instance.
+                _ = AudioQueueDispose(_queue, true);
+                _queue = 0;
+            }
+
+            _freeBuffers.Clear();
+            // Wake any submitter waiting on backpressure before the event
+            // goes away; a late waiter observes ObjectDisposedException and
+            // bails out in Submit.
+            _completion.Set();
+            _completion.Dispose();
+            if (_selfHandle.IsAllocated)
+            {
+                _selfHandle.Free();
+            }
+        }
+    }
+
+    private bool TryTakeBuffer(int length, out nint buffer)
+    {
+        while (_freeBuffers.TryDequeue(out buffer))
+        {
+            if (((AudioQueueBuffer*)buffer)->AudioDataBytesCapacity >= (uint)length)
+            {
+                return true;
+            }
+
+            _ = AudioQueueFreeBuffer(_queue, buffer);
+        }
+
+        return AudioQueueAllocateBuffer(_queue, (uint)length, out buffer) == 0;
+    }
+
+    [UnmanagedCallersOnly]
+    private static void OutputCallback(nint userData, nint queue, nint buffer)
+    {
+        if (GCHandle.FromIntPtr(userData).Target is not PosixCoreAudioStream port)
+        {
+            return;
+        }
+
+        lock (port._gate)
+        {
+            if (port._disposed)
+            {
+                return;
+            }
+
+            port._queuedPcmBytes -= checked((int)((AudioQueueBuffer*)buffer)->AudioDataByteSize);
+            port._freeBuffers.Enqueue(buffer);
+        }
+
+        port._completion.Set();
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    private struct AudioStreamBasicDescription
+    {
+        public double SampleRate;
+        public uint FormatId;
+        public uint FormatFlags;
+        public uint BytesPerPacket;
+        public uint FramesPerPacket;
+        public uint BytesPerFrame;
+        public uint ChannelsPerFrame;
+        public uint BitsPerChannel;
+        public uint Reserved;
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    private struct AudioQueueBuffer
+    {
+        public uint AudioDataBytesCapacity;
+        public void* AudioData;
+        public uint AudioDataByteSize;
+        public nint UserData;
+        public uint PacketDescriptionCapacity;
+        public nint PacketDescriptions;
+        public uint PacketDescriptionCount;
+    }
+
+    private const string AudioToolbox =
+        "/System/Library/Frameworks/AudioToolbox.framework/AudioToolbox";
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueNewOutput(
+        AudioStreamBasicDescription* format,
+        delegate* unmanaged<nint, nint, nint, void> callback,
+        nint userData,
+        nint callbackRunLoop,
+        nint runLoopMode,
+        uint flags,
+        out nint queue);
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueAllocateBuffer(nint queue, uint bufferByteSize, out nint buffer);
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueFreeBuffer(nint queue, nint buffer);
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueEnqueueBuffer(nint queue, nint buffer, uint packetDescriptionCount, nint packetDescriptions);
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueStart(nint queue, nint startTime);
+
+    [DllImport(AudioToolbox)]
+    private static extern int AudioQueueDispose(nint queue, [MarshalAs(UnmanagedType.I1)] bool immediate);
+}

src/SharpEmu.HLE/Host/Posix/PosixHostAudio.cs

@@ -0,0 +1,21 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// POSIX audio output: CoreAudio (AudioQueue) on macOS, ALSA on Linux. Both
+/// streams accept the seam's interleaved stereo PCM16 and pace the guest via
+/// device-queue backpressure.
+/// </summary>
+internal sealed class PosixHostAudio : IHostAudioOutput
+{
+    public string BackendName => OperatingSystem.IsMacOS() ? "coreaudio" : "alsa";
+
+    public IHostAudioStream OpenStereoPcm16Stream(uint sampleRate)
+    {
+        return OperatingSystem.IsMacOS()
+            ? new PosixCoreAudioStream(sampleRate)
+            : new PosixAlsaAudioStream(sampleRate);
+    }
+}

src/SharpEmu.HLE/Host/Posix/PosixHostInput.cs

@@ -0,0 +1,186 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// Bridges a window-provided input source into the host input seam. POSIX
+/// hosts have no user32/XInput/raw-HID readers; keyboard and gamepad state
+/// come from the presenter's GLFW window instead, which registers itself via
+/// <see cref="SetSource"/> once the window exists. Until then (and with no
+/// window at all, e.g. headless runs) every query reports neutral input.
+/// Rumble and lightbar are unsupported by the GLFW input layer and no-op.
+/// </summary>
+public interface IPosixWindowInputSource
+{
+    /// <summary>True while the window's keyboard is delivering events.</summary>
+    bool HasKeyboardFocus { get; }
+
+    /// <summary>Windows virtual-key semantics; the source translates.</summary>
+    bool IsKeyDown(int virtualKey);
+
+    /// <summary>Same contract as <see cref="IHostInput.GetGamepadStates"/>.</summary>
+    int GetGamepadStates(Span<HostGamepadState> destination);
+
+    string? DescribeConnectedGamepad();
+}
+
+// Public so the presenter's window layer (SharpEmu.Libs) can register its
+// input source; the platform still constructs the singleton itself.
+public sealed class PosixHostInput : IHostInput
+{
+    private static volatile IPosixWindowInputSource? _source;
+
+    /// <summary>Called by the presenter's window layer when input is ready.</summary>
+    public static void SetSource(IPosixWindowInputSource source)
+    {
+        _source = source;
+    }
+
+    public void EnsureStarted()
+    {
+        // Device readers are event-driven off the window thread; nothing to start.
+    }
+
+    public int GetGamepadStates(Span<HostGamepadState> destination)
+    {
+        return _source?.GetGamepadStates(destination) ?? 0;
+    }
+
+    public string? DescribeConnectedGamepad() => _source?.DescribeConnectedGamepad();
+
+    public void SetRumble(byte largeMotor, byte smallMotor)
+    {
+    }
+
+    public void SetTriggerRumble(byte? leftTrigger, byte? rightTrigger)
+    {
+    }
+
+    public void SetLightbar(byte red, byte green, byte blue)
+    {
+    }
+
+    public void ResetLightbar()
+    {
+    }
+
+    public bool IsHostWindowFocused()
+    {
+        // GLFW only delivers key events to the focused window, so a
+        // delivering keyboard implies focus.
+        return _source?.HasKeyboardFocus ?? IsEmbeddedX11WindowFocused();
+    }
+
+    public bool IsKeyDown(int virtualKey)
+    {
+        var source = _source;
+        if (source is not null)
+        {
+            return source.IsKeyDown(virtualKey);
+        }
+
+        return IsEmbeddedX11WindowFocused() && IsEmbeddedX11KeyDown(virtualKey);
+    }
+
+    private static bool IsEmbeddedX11WindowFocused()
+    {
+        if (!OperatingSystem.IsLinux())
+        {
+            return false;
+        }
+
+        var display = HostSessionControl.EmbeddedHostDisplay;
+        var window = HostSessionControl.EmbeddedHostWindow;
+        if (display == 0 || window == 0 || XGetInputFocus(display, out var focusedWindow, out _) == 0 || focusedWindow == 0)
+        {
+            return false;
+        }
+
+        return GetTopLevelWindow(display, focusedWindow) == GetTopLevelWindow(display, window);
+    }
+
+    private static bool IsEmbeddedX11KeyDown(int virtualKey)
+    {
+        var display = HostSessionControl.EmbeddedHostDisplay;
+        var keysym = ToX11Keysym(virtualKey);
+        if (display == 0 || keysym == 0)
+        {
+            return false;
+        }
+
+        var keycode = XKeysymToKeycode(display, keysym);
+        if (keycode == 0)
+        {
+            return false;
+        }
+
+        var keymap = new byte[32];
+        XQueryKeymap(display, keymap);
+        return (keymap[keycode >> 3] & (1 << (keycode & 7))) != 0;
+    }
+
+    private static nint GetTopLevelWindow(nint display, nint window)
+    {
+        var current = window;
+        for (var depth = 0; depth < 16; depth++)
+        {
+            if (XQueryTree(display, current, out var root, out var parent, out var children, out _) == 0)
+            {
+                return 0;
+            }
+
+            if (children != 0)
+            {
+                XFree(children);
+            }
+
+            if (parent == 0 || parent == root)
+            {
+                return current;
+            }
+
+            current = parent;
+        }
+
+        return 0;
+    }
+
+    private static nuint ToX11Keysym(int virtualKey)
+    {
+        return virtualKey switch
+        {
+            0x08 => 0xFF08, // Backspace
+            0x09 => 0xFF09, // Tab
+            0x0D => 0xFF0D, // Return
+            0x1B => 0xFF1B, // Escape
+            0x25 => 0xFF51, // Left
+            0x26 => 0xFF52, // Up
+            0x27 => 0xFF53, // Right
+            0x28 => 0xFF54, // Down
+            >= 0x41 and <= 0x5A => (nuint)virtualKey,
+            _ => 0,
+        };
+    }
+
+    [System.Runtime.InteropServices.DllImport("libX11.so.6")]
+    private static extern int XGetInputFocus(nint display, out nint focus, out int revertTo);
+
+    [System.Runtime.InteropServices.DllImport("libX11.so.6")]
+    private static extern int XQueryKeymap(nint display, [System.Runtime.InteropServices.Out] byte[] keysReturn);
+
+    [System.Runtime.InteropServices.DllImport("libX11.so.6")]
+    private static extern byte XKeysymToKeycode(nint display, nuint keysym);
+
+    [System.Runtime.InteropServices.DllImport("libX11.so.6")]
+    private static extern int XQueryTree(
+        nint display,
+        nint window,
+        out nint root,
+        out nint parent,
+        out nint children,
+        out uint childCount);
+
+    [System.Runtime.InteropServices.DllImport("libX11.so.6")]
+    private static extern int XFree(nint data);
+}

src/SharpEmu.HLE/Host/Posix/PosixHostMemory.cs

@@ -0,0 +1,588 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// POSIX virtual memory backend implemented over mmap/mprotect/munmap with a
+/// shadow region table that answers VirtualQuery-style questions and tracks
+/// page protections.
+/// POSIX anonymous mappings are demand-paged by the kernel, so Win32
+/// "reserve-only" regions are mapped as committed memory directly and
+/// commit requests become protection changes.
+/// </summary>
+internal sealed unsafe class PosixHostMemory : IHostMemory
+{
+    private const uint MEM_COMMIT = 0x1000;
+    private const uint MEM_RESERVE = 0x2000;
+    private const uint MEM_RELEASE = 0x8000;
+    private const uint MEM_FREE_STATE = 0x10000;
+    private const uint MEM_PRIVATE = 0x20000;
+
+    private const uint PAGE_NOACCESS = 0x01;
+    private const uint PAGE_READONLY = 0x02;
+    private const uint PAGE_READWRITE = 0x04;
+    private const uint PAGE_EXECUTE = 0x10;
+    private const uint PAGE_EXECUTE_READ = 0x20;
+    private const uint PAGE_EXECUTE_READWRITE = 0x40;
+
+    private const ulong PageSize = 0x1000;
+
+    private struct BasicInfo
+    {
+        public ulong BaseAddress;
+        public ulong AllocationBase;
+        public uint AllocationProtect;
+        public ulong RegionSize;
+        public uint State;
+        public uint Protect;
+        public uint Type;
+    }
+
+    public ulong Allocate(ulong desiredAddress, ulong size, HostPageProtection protection)
+    {
+        return (ulong)Posix.Alloc(
+            (void*)desiredAddress,
+            (nuint)size,
+            MEM_COMMIT | MEM_RESERVE,
+            ToNativeProtection(protection));
+    }
+
+    public ulong Reserve(ulong desiredAddress, ulong size, HostPageProtection protection)
+    {
+        return (ulong)Posix.Alloc(
+            (void*)desiredAddress,
+            (nuint)size,
+            MEM_RESERVE,
+            ToNativeProtection(protection));
+    }
+
+    public bool Commit(ulong address, ulong size, HostPageProtection protection)
+    {
+        return Posix.Alloc(
+            (void*)address,
+            (nuint)size,
+            MEM_COMMIT,
+            ToNativeProtection(protection)) != null;
+    }
+
+    public bool Free(ulong address)
+    {
+        return Posix.Free((void*)address, 0, MEM_RELEASE);
+    }
+
+    public bool Protect(
+        ulong address,
+        ulong size,
+        HostPageProtection protection,
+        out uint rawOldProtection)
+    {
+        return Posix.Protect(
+            (void*)address,
+            (nuint)size,
+            ToNativeProtection(protection),
+            out rawOldProtection);
+    }
+
+    public bool ProtectRaw(
+        ulong address,
+        ulong size,
+        uint rawProtection,
+        out uint rawOldProtection)
+    {
+        return Posix.Protect(
+            (void*)address,
+            (nuint)size,
+            rawProtection,
+            out rawOldProtection);
+    }
+
+    public bool Query(ulong address, out HostRegionInfo info)
+    {
+        if (Posix.Query((void*)address, out var nativeInfo) == 0)
+        {
+            info = default;
+            return false;
+        }
+
+        info = new HostRegionInfo(
+            nativeInfo.BaseAddress,
+            nativeInfo.AllocationBase,
+            nativeInfo.RegionSize,
+            nativeInfo.State switch
+            {
+                MEM_COMMIT => HostRegionState.Committed,
+                MEM_RESERVE => HostRegionState.Reserved,
+                _ => HostRegionState.Free,
+            },
+            nativeInfo.State,
+            ToHostProtection(nativeInfo.Protect),
+            nativeInfo.Protect,
+            nativeInfo.AllocationProtect);
+        return true;
+    }
+
+    public void FlushInstructionCache(ulong address, ulong size)
+    {
+        _ = address;
+        _ = size;
+        // The supported POSIX process is x86-64 (including Rosetta 2), whose
+        // instruction cache is coherent. A future arm64 backend must call the
+        // platform instruction-cache invalidation API here.
+    }
+
+    private static uint ToNativeProtection(HostPageProtection protection) => protection switch
+    {
+        HostPageProtection.NoAccess => PAGE_NOACCESS,
+        HostPageProtection.ReadOnly => PAGE_READONLY,
+        HostPageProtection.ReadWrite => PAGE_READWRITE,
+        HostPageProtection.Execute => PAGE_EXECUTE,
+        HostPageProtection.ReadExecute => PAGE_EXECUTE_READ,
+        HostPageProtection.ReadWriteExecute => PAGE_EXECUTE_READWRITE,
+        HostPageProtection.ExecuteWriteCopy => PAGE_EXECUTE_READWRITE,
+        _ => throw new ArgumentOutOfRangeException(nameof(protection), protection, null),
+    };
+
+    private static HostPageProtection ToHostProtection(uint protection) => protection switch
+    {
+        PAGE_READONLY => HostPageProtection.ReadOnly,
+        PAGE_READWRITE => HostPageProtection.ReadWrite,
+        PAGE_EXECUTE => HostPageProtection.Execute,
+        PAGE_EXECUTE_READ => HostPageProtection.ReadExecute,
+        PAGE_EXECUTE_READWRITE => HostPageProtection.ReadWriteExecute,
+        _ => HostPageProtection.NoAccess,
+    };
+
+    private static class Posix
+    {
+        private const int PROT_NONE = 0x0;
+        private const int PROT_READ = 0x1;
+        private const int PROT_WRITE = 0x2;
+        private const int PROT_EXEC = 0x4;
+
+        private const int MAP_PRIVATE = 0x02;
+        private static readonly int MAP_ANON = OperatingSystem.IsMacOS() ? 0x1000 : 0x20;
+        private static readonly int MAP_NORESERVE = OperatingSystem.IsMacOS() ? 0 : 0x4000;
+
+        // Linux-only: fail instead of clobbering an existing mapping.
+        private const int MAP_FIXED_NOREPLACE = 0x100000;
+
+        private const int KERN_SUCCESS = 0;
+
+        // On Darwin fixed placement is represented by the absence of
+        // VM_FLAGS_ANYWHERE. Do not add VM_FLAGS_OVERWRITE: overlap must fail.
+        private const int VM_FLAGS_FIXED = 0;
+
+        private static readonly nint MAP_FAILED = -1;
+
+        private static readonly object Gate = new();
+        private static readonly SortedList<ulong, Region> Regions = new();
+
+        private sealed class Region
+        {
+            public ulong Base;
+            public ulong Size;
+            public uint DefaultProtect;
+            public Dictionary<ulong, uint>? PageProtects;
+            public bool UsesMachAllocation;
+
+            public ulong End => Base + Size;
+
+            public uint ProtectAt(ulong pageAddress)
+            {
+                if (PageProtects is not null && PageProtects.TryGetValue(pageAddress, out var overriden))
+                {
+                    return overriden;
+                }
+
+                return DefaultProtect;
+            }
+        }
+
+        public static void* Alloc(void* address, nuint size, uint allocationType, uint protect)
+        {
+            if (size == 0)
+            {
+                return null;
+            }
+
+            var alignedSize = AlignUp((ulong)size, PageSize);
+
+            lock (Gate)
+            {
+                if (allocationType == MEM_COMMIT && address != null &&
+                    TryFindRegionLocked((ulong)address, out var existing))
+                {
+                    // Note: MEM_RESERVE requests that overlap an existing
+                    // region must fail like Win32 does; only a pure commit
+                    // may target pages inside a tracked mapping.
+                    // Commit inside an existing mapping: the pages are already
+                    // backed (demand paged), so only apply the protection.
+                    var start = AlignDown((ulong)address, PageSize);
+                    var end = AlignUp((ulong)address + alignedSize, PageSize);
+                    if (end <= start || end > existing.End)
+                    {
+                        // Win32 fails a commit that runs past its reservation
+                        // instead of committing a prefix; committing partially
+                        // here would let callers believe the whole range is
+                        // usable.
+                        return null;
+                    }
+
+                    if (mprotect((nint)start, (nuint)(end - start), ToPosixProtect(protect)) != 0)
+                    {
+                        return null;
+                    }
+
+                    SetProtectRangeLocked(existing, start, end - start, protect);
+                    return address;
+                }
+
+                if ((allocationType & MEM_RESERVE) == 0)
+                {
+                    // MEM_COMMIT alone outside any known region is invalid here.
+                    return null;
+                }
+
+                var posixProtect = ToPosixProtect(protect);
+                var flags = MAP_PRIVATE | MAP_ANON;
+                if ((allocationType & MEM_COMMIT) == 0)
+                {
+                    // Reserve-only: keep the requested protection so the region
+                    // is usable without a separate commit step, but tell the
+                    // kernel not to account swap for it where supported.
+                    flags |= MAP_NORESERVE;
+                }
+
+                nint result;
+                var usesMachAllocation = false;
+                if (address != null)
+                {
+                    // Win32 maps at exactly the requested address or fails
+                    // without touching existing mappings. Fail up front on
+                    // any overlap we track, then place the mapping: Linux
+                    // gets MAP_FIXED_NOREPLACE (fails cleanly on host
+                    // mappings too). Darwin lacks NOREPLACE and plain
+                    // MAP_FIXED would silently clobber untracked host
+                    // memory (dyld, the runtime's JIT heap, Rosetta), so
+                    // pass the address as a hint instead -- the kernel
+                    // honors it when the range is free and relocates the
+                    // mapping otherwise, which we treat as failure.
+                    if (OverlapsTrackedRegionLocked((ulong)address, alignedSize))
+                    {
+                        Trace($"exact overlap: addr=0x{(ulong)address:X16} size=0x{alignedSize:X}");
+                        return null;
+                    }
+
+                    var exactFlags = OperatingSystem.IsMacOS() ? flags : flags | MAP_FIXED_NOREPLACE;
+                    result = mmap((nint)address, (nuint)alignedSize, posixProtect, exactFlags, -1, 0);
+                    if (result != MAP_FAILED && (ulong)result != (ulong)address)
+                    {
+                        munmap(result, (nuint)alignedSize);
+                        result = MAP_FAILED;
+                    }
+
+                    if (result == MAP_FAILED && OperatingSystem.IsMacOS())
+                    {
+                        // The hint-only attempt above didn't land at the requested
+                        // address. mach_vm_allocate with fixed placement and no
+                        // overwrite flag atomically maps the requested range or
+                        // fails if any host mapping already owns it. Unlike
+                        // MAP_FIXED, it cannot clobber CLR, dyld, JIT, or Rosetta
+                        // memory that is absent from our shadow table.
+                        Trace($"exact mmap hint failed, retrying with fixed Mach allocation: addr=0x{(ulong)address:X16}");
+                        result = AllocateDarwinFixed(
+                            (nint)address,
+                            alignedSize,
+                            posixProtect);
+                        usesMachAllocation = result != MAP_FAILED;
+                    }
+
+                    if (result == MAP_FAILED || (ulong)result != (ulong)address)
+                    {
+                        Trace($"exact mmap failed: addr=0x{(ulong)address:X16} got=0x{(ulong)result:X16} size=0x{alignedSize:X} errno={Marshal.GetLastPInvokeError()}");
+                        if (result != MAP_FAILED)
+                        {
+                            munmap(result, (nuint)alignedSize);
+                        }
+
+                        return null;
+                    }
+                }
+                else
+                {
+                    result = mmap(0, (nuint)alignedSize, posixProtect, flags, -1, 0);
+                    if (result == MAP_FAILED)
+                    {
+                        Trace($"mmap failed: size=0x{alignedSize:X} errno={Marshal.GetLastPInvokeError()}");
+                        return null;
+                    }
+                }
+
+                Regions[(ulong)result] = new Region
+                {
+                    Base = (ulong)result,
+                    Size = alignedSize,
+                    DefaultProtect = protect,
+                    UsesMachAllocation = usesMachAllocation,
+                };
+
+                return (void*)result;
+            }
+        }
+
+        public static bool Free(void* address, nuint size, uint freeType)
+        {
+            _ = size;
+            _ = freeType;
+
+            lock (Gate)
+            {
+                if (!Regions.TryGetValue((ulong)address, out var region))
+                {
+                    return false;
+                }
+
+                var released = region.UsesMachAllocation
+                    ? mach_vm_deallocate(mach_task_self(), (ulong)address, region.Size) == KERN_SUCCESS
+                    : munmap((nint)address, (nuint)region.Size) == 0;
+                if (released)
+                {
+                    Regions.Remove((ulong)address);
+                }
+
+                return released;
+            }
+        }
+
+        public static bool Protect(void* address, nuint size, uint newProtect, out uint oldProtect)
+        {
+            oldProtect = PAGE_NOACCESS;
+            if (size == 0)
+            {
+                return false;
+            }
+
+            var start = AlignDown((ulong)address, PageSize);
+            var end = AlignUp((ulong)address + size, PageSize);
+
+            lock (Gate)
+            {
+                if (!TryFindRegionLocked(start, out var region) || end > region.End)
+                {
+                    return false;
+                }
+
+                oldProtect = region.ProtectAt(start);
+                if (mprotect((nint)start, (nuint)(end - start), ToPosixProtect(newProtect)) != 0)
+                {
+                    return false;
+                }
+
+                SetProtectRangeLocked(region, start, end - start, newProtect);
+                return true;
+            }
+        }
+
+        public static nuint Query(void* address, out BasicInfo info)
+        {
+            info = default;
+            var pageAddress = AlignDown((ulong)address, PageSize);
+
+            lock (Gate)
+            {
+                if (TryFindRegionLocked(pageAddress, out var region))
+                {
+                    // Win32 VirtualQuery reports a run of pages sharing the
+                    // same protection, so stop the run where it changes.
+                    var protect = region.ProtectAt(pageAddress);
+                    var runEnd = pageAddress + PageSize;
+                    while (runEnd < region.End && region.ProtectAt(runEnd) == protect)
+                    {
+                        runEnd += PageSize;
+                    }
+
+                    info.BaseAddress = pageAddress;
+                    info.AllocationBase = region.Base;
+                    info.AllocationProtect = region.DefaultProtect;
+                    info.RegionSize = runEnd - pageAddress;
+                    info.State = MEM_COMMIT;
+                    info.Protect = protect;
+                    info.Type = MEM_PRIVATE;
+                    return (nuint)sizeof(BasicInfo);
+                }
+
+                // Untracked host memory (runtime heaps, stacks, libraries) is
+                // reported as a free block reaching to the next tracked region
+                // so scanning callers keep advancing.
+                var nextBase = ulong.MaxValue;
+                foreach (var regionBase in Regions.Keys)
+                {
+                    if (regionBase > pageAddress)
+                    {
+                        nextBase = regionBase;
+                        break;
+                    }
+                }
+
+                info.BaseAddress = pageAddress;
+                info.AllocationBase = 0;
+                info.AllocationProtect = PAGE_NOACCESS;
+                info.RegionSize = (nextBase == ulong.MaxValue ? pageAddress + PageSize : nextBase) - pageAddress;
+                info.State = MEM_FREE_STATE;
+                info.Protect = PAGE_NOACCESS;
+                info.Type = 0;
+                return (nuint)sizeof(BasicInfo);
+            }
+        }
+
+        private static bool OverlapsTrackedRegionLocked(ulong start, ulong size)
+        {
+            var end = start + size;
+            foreach (var region in Regions.Values)
+            {
+                if (region.Base < end && start < region.End)
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        private static bool TryFindRegionLocked(ulong address, out Region region)
+        {
+            region = null!;
+            var keys = Regions.Keys;
+            var low = 0;
+            var high = keys.Count - 1;
+            Region? candidate = null;
+            while (low <= high)
+            {
+                var middle = low + ((high - low) >> 1);
+                var entry = Regions.Values[middle];
+                if (entry.Base <= address)
+                {
+                    candidate = entry;
+                    low = middle + 1;
+                }
+                else
+                {
+                    high = middle - 1;
+                }
+            }
+
+            if (candidate is null || address >= candidate.End)
+            {
+                return false;
+            }
+
+            region = candidate;
+            return true;
+        }
+
+        private static void SetProtectRangeLocked(Region region, ulong start, ulong size, uint protect)
+        {
+            if (start == region.Base && size >= region.Size)
+            {
+                region.DefaultProtect = protect;
+                region.PageProtects = null;
+                return;
+            }
+
+            region.PageProtects ??= new Dictionary<ulong, uint>();
+            var end = start + size;
+            for (var pageAddress = start; pageAddress < end; pageAddress += PageSize)
+            {
+                if (protect == region.DefaultProtect)
+                {
+                    region.PageProtects.Remove(pageAddress);
+                }
+                else
+                {
+                    region.PageProtects[pageAddress] = protect;
+                }
+            }
+        }
+
+        private static int ToPosixProtect(uint win32Protect)
+        {
+            return win32Protect switch
+            {
+                PAGE_NOACCESS => PROT_NONE,
+                PAGE_READONLY => PROT_READ,
+                PAGE_READWRITE => PROT_READ | PROT_WRITE,
+                PAGE_EXECUTE => PROT_READ | PROT_EXEC,
+                PAGE_EXECUTE_READ => PROT_READ | PROT_EXEC,
+                PAGE_EXECUTE_READWRITE => PROT_READ | PROT_WRITE | PROT_EXEC,
+                _ => PROT_READ | PROT_WRITE
+            };
+        }
+
+        private static nint AllocateDarwinFixed(nint requestedAddress, ulong size, int posixProtect)
+        {
+            var address = (ulong)requestedAddress;
+            var result = mach_vm_allocate(
+                mach_task_self(),
+                ref address,
+                size,
+                VM_FLAGS_FIXED);
+            if (result != KERN_SUCCESS || address != (ulong)requestedAddress)
+            {
+                if (result == KERN_SUCCESS)
+                {
+                    _ = mach_vm_deallocate(mach_task_self(), address, size);
+                }
+
+                Trace(
+                    $"fixed Mach allocation failed: addr=0x{(ulong)requestedAddress:X16} " +
+                    $"size=0x{size:X} kern_return={result}");
+                return MAP_FAILED;
+            }
+
+            if (mprotect((nint)address, (nuint)size, posixProtect) == 0)
+            {
+                return (nint)address;
+            }
+
+            var error = Marshal.GetLastPInvokeError();
+            _ = mach_vm_deallocate(mach_task_self(), address, size);
+            Trace(
+                $"fixed Mach allocation protection failed: addr=0x{address:X16} " +
+                $"size=0x{size:X} errno={error}");
+            return MAP_FAILED;
+        }
+
+        private static void Trace(string message)
+        {
+            if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_VMEM"), "1", StringComparison.Ordinal))
+            {
+                Console.Error.WriteLine($"[HOSTMEM] {message}");
+            }
+        }
+
+        private static ulong AlignDown(ulong value, ulong alignment) => value & ~(alignment - 1);
+
+        private static ulong AlignUp(ulong value, ulong alignment) => checked((value + alignment - 1) & ~(alignment - 1));
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern nint mmap(nint addr, nuint length, int prot, int flags, int fd, long offset);
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern int munmap(nint addr, nuint length);
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern int mprotect(nint addr, nuint length, int prot);
+
+        [DllImport("libSystem.B.dylib")]
+        private static extern uint mach_task_self();
+
+        [DllImport("libSystem.B.dylib")]
+        private static extern int mach_vm_allocate(uint target, ref ulong address, ulong size, int flags);
+
+        [DllImport("libSystem.B.dylib")]
+        private static extern int mach_vm_deallocate(uint target, ulong address, ulong size);
+    }
+}

src/SharpEmu.HLE/Host/Posix/PosixHostPlatform.cs

@@ -0,0 +1,17 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host.Posix;
+
+internal sealed class PosixHostPlatform : IHostPlatform
+{
+    public IHostMemory Memory { get; } = new PosixHostMemory();
+
+    public IHostThreading Threading { get; } = new PosixHostThreading();
+
+    public IHostSymbolResolver Symbols { get; } = new PosixHostSymbolResolver();
+
+    public IHostAudioOutput Audio { get; } = new PosixHostAudio();
+
+    public IHostInput Input { get; } = new PosixHostInput();
+}

src/SharpEmu.HLE/Host/Posix/PosixHostStubs.cs

@@ -0,0 +1,657 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Posix;
+
+/// <summary>
+/// POSIX replacements for the kernel32 helpers the native backend embeds in
+/// emitted x86-64 code. Every stub exposed here follows the Win64 calling
+/// convention the emitted call sites were written for (first argument in
+/// ECX, result in RAX, Win64 non-volatile registers preserved), so the
+/// emission code stays identical across platforms.
+/// </summary>
+internal static unsafe class PosixHostStubs
+{
+    private static readonly object Gate = new();
+    private static bool _initialized;
+    private static nint _tlsGetValueStub;
+    private static nint _queryPerformanceCounterStub;
+    private static nint _switchToThreadStub;
+    private static nint _sleepStub;
+    private static nint _waitForSingleObjectStub;
+    private static nint _setEventStub;
+    private static nint _exitThreadStub;
+
+    public static nint TlsGetValueStubAddress
+    {
+        get { EnsureInitialized(); return _tlsGetValueStub; }
+    }
+
+    public static nint QueryPerformanceCounterStubAddress
+    {
+        get { EnsureInitialized(); return _queryPerformanceCounterStub; }
+    }
+
+    public static nint SwitchToThreadStubAddress
+    {
+        get { EnsureInitialized(); return _switchToThreadStub; }
+    }
+
+    public static nint SleepStubAddress
+    {
+        get { EnsureInitialized(); return _sleepStub; }
+    }
+
+    /// <summary>
+    /// Win64-convention replacements for the kernel32 event/thread helpers the
+    /// native guest worker loop embeds. The "handle" they take is a worker
+    /// event created by <see cref="CreateWorkerEvent"/>: a dispatch semaphore
+    /// on macOS, an unnamed POSIX semaphore on Linux. The wait stub always
+    /// waits forever (the worker loop passes INFINITE) and retries EINTR.
+    /// </summary>
+    public static nint WaitForSingleObjectStubAddress
+    {
+        get { EnsureInitialized(); return _waitForSingleObjectStub; }
+    }
+
+    public static nint SetEventStubAddress
+    {
+        get { EnsureInitialized(); return _setEventStub; }
+    }
+
+    public static nint ExitThreadStubAddress
+    {
+        get { EnsureInitialized(); return _exitThreadStub; }
+    }
+
+    /// <summary>
+    /// Creates a binary-semaphore worker event signalable/waitable both from
+    /// managed code and from emitted native code (via the stub addresses
+    /// above). Returns 0 on failure.
+    /// </summary>
+    public static nint CreateWorkerEvent()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            return dispatch_semaphore_create(0);
+        }
+
+        var semaphore = Marshal.AllocHGlobal(64);
+        if (sem_init(semaphore, 0, 0) != 0)
+        {
+            Marshal.FreeHGlobal(semaphore);
+            return 0;
+        }
+
+        return semaphore;
+    }
+
+    public static bool SignalWorkerEvent(nint handle)
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            _ = dispatch_semaphore_signal(handle);
+            return true;
+        }
+
+        return sem_post(handle) == 0;
+    }
+
+    /// <summary>Waits for a worker event; a negative timeout waits forever.</summary>
+    public static bool WaitWorkerEvent(nint handle, int timeoutMilliseconds)
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            if (timeoutMilliseconds < 0)
+            {
+                return dispatch_semaphore_wait(handle, ulong.MaxValue) == 0;
+            }
+
+            var deadline = dispatch_time(0, timeoutMilliseconds * 1_000_000L);
+            return dispatch_semaphore_wait(handle, deadline) == 0;
+        }
+
+        if (timeoutMilliseconds < 0)
+        {
+            while (sem_wait(handle) != 0)
+            {
+                // EINTR: retry.
+            }
+
+            return true;
+        }
+
+        var deadlineTicks = Environment.TickCount64 + timeoutMilliseconds;
+        while (sem_trywait(handle) != 0)
+        {
+            if (Environment.TickCount64 >= deadlineTicks)
+            {
+                return false;
+            }
+
+            System.Threading.Thread.Sleep(1);
+        }
+
+        return true;
+    }
+
+    public static void DestroyWorkerEvent(nint handle)
+    {
+        if (handle == 0)
+        {
+            return;
+        }
+
+        if (OperatingSystem.IsMacOS())
+        {
+            dispatch_release(handle);
+            return;
+        }
+
+        _ = sem_destroy(handle);
+        Marshal.FreeHGlobal(handle);
+    }
+
+    /// <summary>
+    /// Starts a raw pthread at a native entry point (pthread entries take their
+    /// argument in RDI; the worker loop stub ignores it). Returns an opaque
+    /// handle for <see cref="WaitForWorkerThreadExit"/>/<see cref="CloseWorkerThreadHandle"/>,
+    /// or 0 on failure.
+    /// </summary>
+    public static nint CreateWorkerThread(nint entry, nint parameter, nuint stackReserveBytes, out uint threadId)
+    {
+        threadId = 0;
+        byte* attr = stackalloc byte[512];
+        if (pthread_attr_init(attr) != 0)
+        {
+            return 0;
+        }
+
+        try
+        {
+            if (stackReserveBytes != 0)
+            {
+                _ = pthread_attr_setstacksize(attr, nuint.Max(stackReserveBytes, 512 * 1024));
+            }
+
+            nint thread;
+            if (pthread_create(&thread, attr, entry, parameter) != 0)
+            {
+                return 0;
+            }
+
+            if (OperatingSystem.IsMacOS())
+            {
+                ulong numericId;
+                if (pthread_threadid_np(thread, &numericId) == 0)
+                {
+                    threadId = unchecked((uint)numericId);
+                }
+            }
+            else
+            {
+                threadId = unchecked((uint)thread);
+            }
+
+            var holder = (nint*)Marshal.AllocHGlobal(sizeof(nint) * 2);
+            holder[0] = thread;
+            holder[1] = 0; // joined flag
+            return (nint)holder;
+        }
+        finally
+        {
+            _ = pthread_attr_destroy(attr);
+        }
+    }
+
+    /// <summary>
+    /// Waits for a worker thread to exit. Liveness is probed with
+    /// pthread_kill(thread, 0) (ESRCH once the thread has terminated) because
+    /// neither platform offers a portable timed join; the exited thread is then
+    /// joined so its resources are reclaimed.
+    /// </summary>
+    public static bool WaitForWorkerThreadExit(nint threadHandle, uint timeoutMilliseconds)
+    {
+        var holder = (nint*)threadHandle;
+        if (holder == null)
+        {
+            return false;
+        }
+
+        if (holder[1] != 0)
+        {
+            return true;
+        }
+
+        var thread = holder[0];
+        var deadline = Environment.TickCount64 + timeoutMilliseconds;
+        while (pthread_kill(thread, 0) == 0)
+        {
+            if (Environment.TickCount64 >= deadline)
+            {
+                return false;
+            }
+
+            System.Threading.Thread.Sleep(1);
+        }
+
+        _ = pthread_join(thread, null);
+        holder[1] = 1;
+        return true;
+    }
+
+    public static void CloseWorkerThreadHandle(nint threadHandle)
+    {
+        var holder = (nint*)threadHandle;
+        if (holder == null)
+        {
+            return;
+        }
+
+        if (holder[1] == 0)
+        {
+            // Never observed exiting: detach so the thread does not leak a
+            // zombie join target when it eventually terminates.
+            _ = pthread_detach(holder[0]);
+        }
+
+        Marshal.FreeHGlobal(threadHandle);
+    }
+
+    /// <summary>Allocates a pthread TLS key, mirroring kernel32!TlsAlloc.</summary>
+    public static uint TlsAlloc()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            nuint key;
+            return pthread_key_create_mac(&key, 0) == 0 ? (uint)key : uint.MaxValue;
+        }
+
+        uint key32;
+        return pthread_key_create_linux(&key32, 0) == 0 ? key32 : uint.MaxValue;
+    }
+
+    public static bool TlsFree(uint key)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_key_delete_mac((nuint)key) == 0
+            : pthread_key_delete_linux(key) == 0;
+    }
+
+    public static bool TlsSetValue(uint key, nint value)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_setspecific_mac((nuint)key, value) == 0
+            : pthread_setspecific_linux(key, value) == 0;
+    }
+
+    public static nint TlsGetValue(uint key)
+    {
+        return OperatingSystem.IsMacOS()
+            ? pthread_getspecific_mac((nuint)key)
+            : pthread_getspecific_linux(key);
+    }
+
+    /// <summary>Stable numeric id of the calling thread (kernel32!GetCurrentThreadId).</summary>
+    public static uint GetCurrentThreadId()
+    {
+        if (OperatingSystem.IsMacOS())
+        {
+            ulong tid;
+            return pthread_threadid_np(0, &tid) == 0 ? unchecked((uint)tid) : 0u;
+        }
+
+        return unchecked((uint)gettid());
+    }
+
+    /// <summary>
+    /// Wraps a managed callback (compiled for the SysV ABI on POSIX .NET) in a
+    /// thunk that accepts up to four integer arguments in the Win64 ABI the
+    /// emitted x86-64 call sites use. Win64 passes args in rcx/rdx/r8/r9 and
+    /// treats rdi/rsi as non-volatile; SysV expects rdi/rsi/rdx/rcx and
+    /// clobbers them, so the thunk saves rdi/rsi, shuffles the registers, keeps
+    /// the stack 16-byte aligned for the call, and forwards the rax result.
+    /// </summary>
+    public static nint CreateWin64ToSysVThunk(nint sysvTarget)
+    {
+        var memory = HostPlatform.Current.Memory;
+        var page = (byte*)memory.Allocate(
+            0,
+            4096,
+            HostPageProtection.ReadWriteExecute);
+        if (page == null)
+        {
+            throw new OutOfMemoryException("Failed to allocate Win64->SysV thunk page");
+        }
+
+        var offset = 0;
+        Emit(page, ref offset, 0x57);                   // push rdi
+        Emit(page, ref offset, 0x56);                   // push rsi
+        Emit(page, ref offset, 0x48, 0x89, 0xCF);       // mov rdi, rcx
+        Emit(page, ref offset, 0x48, 0x89, 0xD6);       // mov rsi, rdx
+        Emit(page, ref offset, 0x4C, 0x89, 0xC2);       // mov rdx, r8
+        Emit(page, ref offset, 0x4C, 0x89, 0xC9);       // mov rcx, r9
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08); // sub rsp, 8 (realign to 16)
+        EmitMovRaxImm64(page, ref offset, sysvTarget);  // mov rax, target
+        Emit(page, ref offset, 0xFF, 0xD0);             // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08); // add rsp, 8
+        Emit(page, ref offset, 0x5E);                   // pop rsi
+        Emit(page, ref offset, 0x5F);                   // pop rdi
+        Emit(page, ref offset, 0xC3);                   // ret
+
+        if (!memory.Protect((ulong)page, 4096, HostPageProtection.ReadExecute, out _))
+        {
+            throw new InvalidOperationException("Failed to protect Win64->SysV thunk page");
+        }
+
+        memory.FlushInstructionCache((ulong)page, (ulong)offset);
+        return (nint)page;
+    }
+
+    private static void EnsureInitialized()
+    {
+        if (_initialized)
+        {
+            return;
+        }
+
+        lock (Gate)
+        {
+            if (_initialized)
+            {
+                return;
+            }
+
+            BuildStubs();
+            _initialized = true;
+        }
+    }
+
+    private static void BuildStubs()
+    {
+        var memory = HostPlatform.Current.Memory;
+        var page = (byte*)memory.Allocate(
+            0,
+            4096,
+            HostPageProtection.ReadWriteExecute);
+        if (page == null)
+        {
+            throw new OutOfMemoryException("Failed to allocate POSIX host helper stub page");
+        }
+
+        var offset = 0;
+        _tlsGetValueStub = EmitTlsGetValue(page, ref offset);
+        _queryPerformanceCounterStub = EmitQueryPerformanceCounter(page, ref offset);
+        _switchToThreadStub = EmitSwitchToThread(page, ref offset);
+        _sleepStub = EmitSleep(page, ref offset);
+        _waitForSingleObjectStub = EmitWaitForSingleObject(page, ref offset);
+        _setEventStub = EmitSetEvent(page, ref offset);
+        _exitThreadStub = EmitExitThread(page, ref offset);
+
+        if (!memory.Protect((ulong)page, 4096, HostPageProtection.ReadExecute, out _))
+        {
+            throw new InvalidOperationException("Failed to protect POSIX host helper stub page");
+        }
+
+        memory.FlushInstructionCache((ulong)page, (ulong)offset);
+    }
+
+    private static nint EmitTlsGetValue(byte* page, ref int offset)
+    {
+        var start = (nint)(page + offset);
+        if (OperatingSystem.IsMacOS())
+        {
+            // On macOS x86-64 pthread keys index the gs-based thread specific
+            // data array directly, so TlsGetValue(index in ecx) collapses to a
+            // single load that clobbers nothing but RAX.
+            Emit(page, ref offset, 0x89, 0xC8);                                     // mov eax, ecx
+            Emit(page, ref offset, 0x65, 0x48, 0x8B, 0x04, 0xC5, 0, 0, 0, 0);       // mov rax, gs:[rax*8]
+            Emit(page, ref offset, 0xC3);                                           // ret
+            return start;
+        }
+
+        // Linux: call pthread_getspecific, preserving the registers that are
+        // volatile in SysV but non-volatile in Win64 (rsi, rdi).
+        var pthreadGetSpecific = ResolveLibcExport("pthread_getspecific");
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        Emit(page, ref offset, 0x89, 0xCF);                                         // mov edi, ecx
+        EmitMovRaxImm64(page, ref offset, pthreadGetSpecific);                      // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitQueryPerformanceCounter(byte* page, ref int offset)
+    {
+        // BOOL QueryPerformanceCounter(LARGE_INTEGER* out in rcx): the emitted
+        // consumers only need a monotonically increasing counter, which rdtsc
+        // provides without leaving Win64-safe registers.
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x0F, 0x31);                                         // rdtsc
+        Emit(page, ref offset, 0x48, 0xC1, 0xE2, 0x20);                             // shl rdx, 32
+        Emit(page, ref offset, 0x48, 0x09, 0xD0);                                   // or rax, rdx
+        Emit(page, ref offset, 0x48, 0x89, 0x01);                                   // mov [rcx], rax
+        Emit(page, ref offset, 0xB8, 0x01, 0x00, 0x00, 0x00);                       // mov eax, 1
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitSwitchToThread(byte* page, ref int offset)
+    {
+        var schedYield = ResolveLibcExport("sched_yield");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        EmitMovRaxImm64(page, ref offset, schedYield);                              // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xB8, 0x01, 0x00, 0x00, 0x00);                       // mov eax, 1
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitSleep(byte* page, ref int offset)
+    {
+        // void Sleep(DWORD milliseconds in ecx) -> usleep(microseconds in edi).
+        var usleep = ResolveLibcExport("usleep");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                               // push rsi
+        Emit(page, ref offset, 0x57);                                               // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                             // sub rsp, 8
+        Emit(page, ref offset, 0x89, 0xCF);                                         // mov edi, ecx
+        Emit(page, ref offset, 0x81, 0xFF, 0xFF, 0x0F, 0x00, 0x00);                 // cmp edi, 0xFFF
+        Emit(page, ref offset, 0x76, 0x05);                                         // jbe +5
+        Emit(page, ref offset, 0xBF, 0xFF, 0x0F, 0x00, 0x00);                       // mov edi, 0xFFF (cap at ~4s)
+        Emit(page, ref offset, 0x69, 0xFF, 0xE8, 0x03, 0x00, 0x00);                 // imul edi, edi, 1000
+        EmitMovRaxImm64(page, ref offset, usleep);                                  // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                         // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                             // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                               // pop rdi
+        Emit(page, ref offset, 0x5E);                                               // pop rsi
+        Emit(page, ref offset, 0xC3);                                               // ret
+        return start;
+    }
+
+    private static nint EmitWaitForSingleObject(byte* page, ref int offset)
+    {
+        // DWORD WaitForSingleObject(worker event in rcx, timeout in edx): the
+        // worker loop only ever waits forever, so the timeout is ignored.
+        // macOS waits on a dispatch semaphore (needs DISPATCH_TIME_FOREVER in
+        // rsi), Linux on a sem_t; both retry until the wait succeeds (EINTR).
+        var wait = ResolveLibcExport(
+            OperatingSystem.IsMacOS() ? "dispatch_semaphore_wait" : "sem_wait");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                           // push rsi
+        Emit(page, ref offset, 0x57);                                           // push rdi
+        Emit(page, ref offset, 0x53);                                           // push rbx
+        Emit(page, ref offset, 0x48, 0x89, 0xCB);                               // mov rbx, rcx
+        var retry = offset;
+        Emit(page, ref offset, 0x48, 0x89, 0xDF);                               // mov rdi, rbx
+        if (OperatingSystem.IsMacOS())
+        {
+            Emit(page, ref offset, 0x48, 0xC7, 0xC6, 0xFF, 0xFF, 0xFF, 0xFF);   // mov rsi, DISPATCH_TIME_FOREVER
+        }
+        EmitMovRaxImm64(page, ref offset, wait);                                // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                     // call rax
+        Emit(page, ref offset, 0x85, 0xC0);                                     // test eax, eax
+        Emit(page, ref offset, 0x75, unchecked((byte)(retry - (offset + 2))));  // jnz retry
+        Emit(page, ref offset, 0x31, 0xC0);                                     // xor eax, eax (WAIT_OBJECT_0)
+        Emit(page, ref offset, 0x5B);                                           // pop rbx
+        Emit(page, ref offset, 0x5F);                                           // pop rdi
+        Emit(page, ref offset, 0x5E);                                           // pop rsi
+        Emit(page, ref offset, 0xC3);                                           // ret
+        return start;
+    }
+
+    private static nint EmitSetEvent(byte* page, ref int offset)
+    {
+        // BOOL SetEvent(worker event in rcx) -> dispatch_semaphore_signal /
+        // sem_post.
+        var signal = ResolveLibcExport(
+            OperatingSystem.IsMacOS() ? "dispatch_semaphore_signal" : "sem_post");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x56);                                           // push rsi
+        Emit(page, ref offset, 0x57);                                           // push rdi
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                         // sub rsp, 8
+        Emit(page, ref offset, 0x48, 0x89, 0xCF);                               // mov rdi, rcx
+        EmitMovRaxImm64(page, ref offset, signal);                              // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                     // call rax
+        Emit(page, ref offset, 0x48, 0x83, 0xC4, 0x08);                         // add rsp, 8
+        Emit(page, ref offset, 0x5F);                                           // pop rdi
+        Emit(page, ref offset, 0x5E);                                           // pop rsi
+        Emit(page, ref offset, 0xB8, 0x01, 0x00, 0x00, 0x00);                   // mov eax, 1
+        Emit(page, ref offset, 0xC3);                                           // ret
+        return start;
+    }
+
+    private static nint EmitExitThread(byte* page, ref int offset)
+    {
+        // void ExitThread(code in ecx) -> pthread_exit(NULL); never returns,
+        // so no registers need preserving. pthread_exit runs the thread's TSD
+        // destructors, which detaches the CLR if the thread lazily attached.
+        var pthreadExit = ResolveLibcExport("pthread_exit");
+        var start = (nint)(page + offset);
+        Emit(page, ref offset, 0x48, 0x83, 0xEC, 0x08);                         // sub rsp, 8
+        Emit(page, ref offset, 0x31, 0xFF);                                     // xor edi, edi
+        EmitMovRaxImm64(page, ref offset, pthreadExit);                         // mov rax, imm64
+        Emit(page, ref offset, 0xFF, 0xD0);                                     // call rax
+        Emit(page, ref offset, 0xCC);                                           // int3 (never returns)
+        return start;
+    }
+
+    private static nint ResolveLibcExport(string name)
+    {
+        var libc = NativeLibrary.Load(OperatingSystem.IsMacOS() ? "libSystem.dylib" : "libc.so.6");
+        return NativeLibrary.GetExport(libc, name);
+    }
+
+    private static void Emit(byte* page, ref int offset, params byte[] bytes)
+    {
+        foreach (var value in bytes)
+        {
+            page[offset++] = value;
+        }
+    }
+
+    private static void EmitMovRaxImm64(byte* page, ref int offset, nint value)
+    {
+        Emit(page, ref offset, 0x48, 0xB8);
+        *(long*)(page + offset) = value;
+        offset += sizeof(long);
+    }
+
+    [DllImport("libc", EntryPoint = "pthread_key_create", SetLastError = true)]
+    private static extern int pthread_key_create_mac(nuint* key, nint destructor);
+
+    [DllImport("libc", EntryPoint = "pthread_key_create", SetLastError = true)]
+    private static extern int pthread_key_create_linux(uint* key, nint destructor);
+
+    [DllImport("libc", EntryPoint = "pthread_key_delete")]
+    private static extern int pthread_key_delete_mac(nuint key);
+
+    [DllImport("libc", EntryPoint = "pthread_key_delete")]
+    private static extern int pthread_key_delete_linux(uint key);
+
+    [DllImport("libc", EntryPoint = "pthread_setspecific")]
+    private static extern int pthread_setspecific_mac(nuint key, nint value);
+
+    [DllImport("libc", EntryPoint = "pthread_setspecific")]
+    private static extern int pthread_setspecific_linux(uint key, nint value);
+
+    [DllImport("libc", EntryPoint = "pthread_getspecific")]
+    private static extern nint pthread_getspecific_mac(nuint key);
+
+    [DllImport("libc", EntryPoint = "pthread_getspecific")]
+    private static extern nint pthread_getspecific_linux(uint key);
+
+    [DllImport("libc")]
+    private static extern int pthread_threadid_np(nint thread, ulong* threadId);
+
+    [DllImport("libc")]
+    private static extern int gettid();
+
+    [DllImport("libc")]
+    private static extern int pthread_attr_init(byte* attr);
+
+    [DllImport("libc")]
+    private static extern int pthread_attr_destroy(byte* attr);
+
+    [DllImport("libc")]
+    private static extern int pthread_attr_setstacksize(byte* attr, nuint stackSize);
+
+    [DllImport("libc")]
+    private static extern int pthread_create(nint* thread, byte* attr, nint startRoutine, nint arg);
+
+    [DllImport("libc")]
+    private static extern int pthread_join(nint thread, nint* returnValue);
+
+    [DllImport("libc")]
+    private static extern int pthread_detach(nint thread);
+
+    [DllImport("libc")]
+    private static extern int pthread_kill(nint thread, int signal);
+
+    // macOS: dispatch semaphores back the worker events (unnamed sem_init is
+    // unsupported on Darwin). libSystem reexports libdispatch, so "libc"
+    // resolves these like the pthread imports above.
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_create(long value);
+
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_signal(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern nint dispatch_semaphore_wait(nint semaphore, ulong timeout);
+
+    [DllImport("libc")]
+    private static extern ulong dispatch_time(ulong when, long deltaNanoseconds);
+
+    [DllImport("libc")]
+    private static extern void dispatch_release(nint handle);
+
+    // Linux: unnamed POSIX semaphores.
+    [DllImport("libc")]
+    private static extern int sem_init(nint semaphore, int shared, uint value);
+
+    [DllImport("libc")]
+    private static extern int sem_post(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_wait(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_trywait(nint semaphore);
+
+    [DllImport("libc")]
+    private static extern int sem_destroy(nint semaphore);
+}

src/SharpEmu.HLE/Host/Posix/PosixHostSymbolResolver.cs

@@ -0,0 +1,19 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host.Posix;
+
+internal sealed class PosixHostSymbolResolver : IHostSymbolResolver
+{
+    public nint GetAddress(HostRuntimeFunction function) => function switch
+    {
+        HostRuntimeFunction.TlsGetValue => PosixHostStubs.TlsGetValueStubAddress,
+        HostRuntimeFunction.QueryPerformanceCounter => PosixHostStubs.QueryPerformanceCounterStubAddress,
+        HostRuntimeFunction.SwitchToThread => PosixHostStubs.SwitchToThreadStubAddress,
+        HostRuntimeFunction.Sleep => PosixHostStubs.SleepStubAddress,
+        HostRuntimeFunction.WaitForSingleObject => PosixHostStubs.WaitForSingleObjectStubAddress,
+        HostRuntimeFunction.SetEvent => PosixHostStubs.SetEventStubAddress,
+        HostRuntimeFunction.ExitThread => PosixHostStubs.ExitThreadStubAddress,
+        _ => throw new ArgumentOutOfRangeException(nameof(function), function, null),
+    };
+}

src/SharpEmu.HLE/Host/Posix/PosixHostThreading.cs

@@ -0,0 +1,57 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE.Host.Posix;
+
+internal sealed class PosixHostThreading : IHostThreading
+{
+    public uint AllocateTlsSlot() => PosixHostStubs.TlsAlloc();
+
+    public bool FreeTlsSlot(uint slot) => PosixHostStubs.TlsFree(slot);
+
+    public bool SetTlsValue(uint slot, nint value) => PosixHostStubs.TlsSetValue(slot, value);
+
+    public nint GetTlsValue(uint slot) => PosixHostStubs.TlsGetValue(slot);
+
+    public uint CurrentThreadId => PosixHostStubs.GetCurrentThreadId();
+
+    public void RequestTimerResolution()
+    {
+        // POSIX sleep primitives are already high-resolution; there is no
+        // timeBeginPeriod equivalent to request.
+    }
+
+    // Thread affinity is advisory on POSIX hosts (macOS offers no
+    // pthread-level affinity API); callers treat false as "not applied".
+    public bool TrySetCurrentThreadAffinity(nuint affinityMask)
+    {
+        _ = affinityMask;
+        return false;
+    }
+
+    public nint CreateNativeThread(
+        nint entry,
+        nint parameter,
+        nuint stackReserveBytes,
+        out uint threadId)
+    {
+        return PosixHostStubs.CreateWorkerThread(entry, parameter, stackReserveBytes, out threadId);
+    }
+
+    public bool WaitForThreadExit(nint threadHandle, uint timeoutMilliseconds)
+    {
+        return PosixHostStubs.WaitForWorkerThreadExit(threadHandle, timeoutMilliseconds);
+    }
+
+    public void CloseThreadHandle(nint threadHandle)
+    {
+        PosixHostStubs.CloseWorkerThreadHandle(threadHandle);
+    }
+
+    public bool TryCaptureThreadRegisters(uint threadId, out HostCapturedRegisters registers)
+    {
+        _ = threadId;
+        registers = default;
+        return false;
+    }
+}

src/SharpEmu.HLE/Host/Windows/WindowsDualSenseReader.cs

@@ -3,21 +3,21 @@
 
 using Microsoft.Win32.SafeHandles;
 
-namespace SharpEmu.Libs.Pad;
+namespace SharpEmu.HLE.Host.Windows;
 
 /// <summary>
 /// Reads a DualSense controller over raw HID on a background thread.
 /// Supports USB (input report 0x01) and Bluetooth (extended report 0x31,
 /// activated by requesting feature report 0x05), with hot-plug retry.
 /// </summary>
-internal static class DualSenseReader
+public static class WindowsDualSenseReader
 {
     private const ushort SonyVendorId = 0x054C;
     private const ushort DualSenseProductId = 0x0CE6;
     private const ushort DualSenseEdgeProductId = 0x0DF2;
 
     private static readonly object Gate = new();
-    private static PadState _state;
+    private static HostGamepadState _state;
     private static bool _started;
 
     // Output (rumble/lightbar) state, all guarded by Gate.
@@ -35,8 +35,10 @@ internal static class DualSenseReader
     private static byte _playerLeds = 0x04; // center LED = player 1
 
     /// <summary>Starts the background reader once; safe to call repeatedly.</summary>
-    internal static void EnsureStarted()
+    public static void EnsureStarted()
     {
+        // The GUI source-links this reader and calls it directly, without the
+        // host-platform resolution that otherwise guarantees Windows.
         if (!OperatingSystem.IsWindows())
         {
             return;
@@ -59,7 +61,7 @@ internal static class DualSenseReader
         }
     }
 
-    internal static bool TryGetState(out PadState state)
+    public static bool TryGetState(out HostGamepadState state)
     {
         lock (Gate)
         {
@@ -69,7 +71,7 @@ internal static class DualSenseReader
         return state.Connected;
     }
 
-    private static void SetState(in PadState state)
+    private static void SetState(in HostGamepadState state)
     {
         lock (Gate)
         {
@@ -148,11 +150,11 @@ internal static class DualSenseReader
         {
             if (_outputStream is null)
             {
-                var handle = HidNative.CreateFile(
+                var handle = WindowsHidNative.CreateFile(
                     _devicePath,
-                    HidNative.GenericRead | HidNative.GenericWrite,
-                    HidNative.FileShareRead | HidNative.FileShareWrite,
-                    0, HidNative.OpenExisting, 0, 0);
+                    WindowsHidNative.GenericRead | WindowsHidNative.GenericWrite,
+                    WindowsHidNative.FileShareRead | WindowsHidNative.FileShareWrite,
+                    0, WindowsHidNative.OpenExisting, 0, 0);
                 if (handle.IsInvalid)
                 {
                     handle.Dispose();
@@ -262,7 +264,7 @@ internal static class DualSenseReader
                 // to the full 0x31 input report. Harmless over USB.
                 var feature = new byte[41];
                 feature[0] = 0x05;
-                _ = HidNative.HidD_GetFeature(handle, feature, feature.Length);
+                _ = WindowsHidNative.HidD_GetFeature(handle, feature, feature.Length);
 
                 if (!announcedConnect)
                 {
@@ -320,18 +322,18 @@ internal static class DualSenseReader
     private static SafeFileHandle? OpenDualSense(out string? devicePath)
     {
         devicePath = null;
-        foreach (var path in HidNative.EnumerateHidDevicePaths())
+        foreach (var path in WindowsHidNative.EnumerateHidDevicePaths())
         {
             // Open without access rights just to query VID/PID.
-            using var probe = HidNative.CreateFile(
-                path, 0, HidNative.FileShareRead | HidNative.FileShareWrite, 0, HidNative.OpenExisting, 0, 0);
+            using var probe = WindowsHidNative.CreateFile(
+                path, 0, WindowsHidNative.FileShareRead | WindowsHidNative.FileShareWrite, 0, WindowsHidNative.OpenExisting, 0, 0);
             if (probe.IsInvalid)
             {
                 continue;
             }
 
-            var attributes = new HidNative.HiddAttributes { Size = 12 };
-            if (!HidNative.HidD_GetAttributes(probe, ref attributes) ||
+            var attributes = new WindowsHidNative.HiddAttributes { Size = 12 };
+            if (!WindowsHidNative.HidD_GetAttributes(probe, ref attributes) ||
                 attributes.VendorId != SonyVendorId ||
                 (attributes.ProductId != DualSenseProductId && attributes.ProductId != DualSenseEdgeProductId))
             {
@@ -339,19 +341,19 @@ internal static class DualSenseReader
             }
 
             // Read+write so feature reports work; fall back to read-only.
-            var handle = HidNative.CreateFile(
+            var handle = WindowsHidNative.CreateFile(
                 path,
-                HidNative.GenericRead | HidNative.GenericWrite,
-                HidNative.FileShareRead | HidNative.FileShareWrite,
-                0, HidNative.OpenExisting, 0, 0);
+                WindowsHidNative.GenericRead | WindowsHidNative.GenericWrite,
+                WindowsHidNative.FileShareRead | WindowsHidNative.FileShareWrite,
+                0, WindowsHidNative.OpenExisting, 0, 0);
             if (handle.IsInvalid)
             {
                 handle.Dispose();
-                handle = HidNative.CreateFile(
+                handle = WindowsHidNative.CreateFile(
                     path,
-                    HidNative.GenericRead,
-                    HidNative.FileShareRead | HidNative.FileShareWrite,
-                    0, HidNative.OpenExisting, 0, 0);
+                    WindowsHidNative.GenericRead,
+                    WindowsHidNative.FileShareRead | WindowsHidNative.FileShareWrite,
+                    0, WindowsHidNative.OpenExisting, 0, 0);
             }
 
             if (!handle.IsInvalid)
@@ -366,7 +368,7 @@ internal static class DualSenseReader
         return null;
     }
 
-    private static bool TryParseReport(ReadOnlySpan<byte> report, out PadState state)
+    private static bool TryParseReport(ReadOnlySpan<byte> report, out HostGamepadState state)
     {
         // USB: report id 0x01, payload starts at [1].
         // Bluetooth extended: report id 0x31, sequence byte at [1], payload at [2].
@@ -395,43 +397,43 @@ internal static class DualSenseReader
         var buttons1 = report[offset + 8];
         var buttons2 = report[offset + 9];
 
-        uint buttons = 0;
-        buttons |= (buttons0 & 0x10) != 0 ? OrbisPadButton.Square : 0;
-        buttons |= (buttons0 & 0x20) != 0 ? OrbisPadButton.Cross : 0;
-        buttons |= (buttons0 & 0x40) != 0 ? OrbisPadButton.Circle : 0;
-        buttons |= (buttons0 & 0x80) != 0 ? OrbisPadButton.Triangle : 0;
+        var buttons = HostGamepadButtons.None;
+        buttons |= (buttons0 & 0x10) != 0 ? HostGamepadButtons.Square : 0;
+        buttons |= (buttons0 & 0x20) != 0 ? HostGamepadButtons.Cross : 0;
+        buttons |= (buttons0 & 0x40) != 0 ? HostGamepadButtons.Circle : 0;
+        buttons |= (buttons0 & 0x80) != 0 ? HostGamepadButtons.Triangle : 0;
         buttons |= HatToButtons(buttons0 & 0x0F);
-        buttons |= (buttons1 & 0x01) != 0 ? OrbisPadButton.L1 : 0;
-        buttons |= (buttons1 & 0x02) != 0 ? OrbisPadButton.R1 : 0;
-        buttons |= (buttons1 & 0x04) != 0 ? OrbisPadButton.L2 : 0;
-        buttons |= (buttons1 & 0x08) != 0 ? OrbisPadButton.R2 : 0;
-        buttons |= (buttons1 & 0x20) != 0 ? OrbisPadButton.Options : 0;
-        buttons |= (buttons1 & 0x40) != 0 ? OrbisPadButton.L3 : 0;
-        buttons |= (buttons1 & 0x80) != 0 ? OrbisPadButton.R3 : 0;
-        buttons |= (buttons2 & 0x02) != 0 ? OrbisPadButton.TouchPad : 0;
+        buttons |= (buttons1 & 0x01) != 0 ? HostGamepadButtons.L1 : 0;
+        buttons |= (buttons1 & 0x02) != 0 ? HostGamepadButtons.R1 : 0;
+        buttons |= (buttons1 & 0x04) != 0 ? HostGamepadButtons.L2 : 0;
+        buttons |= (buttons1 & 0x08) != 0 ? HostGamepadButtons.R2 : 0;
+        buttons |= (buttons1 & 0x20) != 0 ? HostGamepadButtons.Options : 0;
+        buttons |= (buttons1 & 0x40) != 0 ? HostGamepadButtons.L3 : 0;
+        buttons |= (buttons1 & 0x80) != 0 ? HostGamepadButtons.R3 : 0;
+        buttons |= (buttons2 & 0x02) != 0 ? HostGamepadButtons.TouchPad : 0;
 
-        state = new PadState(
+        state = new HostGamepadState(
             Connected: true,
             Buttons: buttons,
             LeftX: leftX,
             LeftY: leftY,
             RightX: rightX,
             RightY: rightY,
-            L2: l2,
-            R2: r2);
+            LeftTrigger: l2,
+            RightTrigger: r2);
         return true;
     }
 
-    private static uint HatToButtons(int hat) => hat switch
+    private static HostGamepadButtons HatToButtons(int hat) => hat switch
     {
-        0 => OrbisPadButton.Up,
-        1 => OrbisPadButton.Up | OrbisPadButton.Right,
-        2 => OrbisPadButton.Right,
-        3 => OrbisPadButton.Right | OrbisPadButton.Down,
-        4 => OrbisPadButton.Down,
-        5 => OrbisPadButton.Down | OrbisPadButton.Left,
-        6 => OrbisPadButton.Left,
-        7 => OrbisPadButton.Left | OrbisPadButton.Up,
+        0 => HostGamepadButtons.Up,
+        1 => HostGamepadButtons.Up | HostGamepadButtons.Right,
+        2 => HostGamepadButtons.Right,
+        3 => HostGamepadButtons.Right | HostGamepadButtons.Down,
+        4 => HostGamepadButtons.Down,
+        5 => HostGamepadButtons.Down | HostGamepadButtons.Left,
+        6 => HostGamepadButtons.Left,
+        7 => HostGamepadButtons.Left | HostGamepadButtons.Up,
         _ => 0,
     };
 }

src/SharpEmu.HLE/Host/Windows/WindowsHidNative.cs

@@ -4,13 +4,13 @@
 using System.Runtime.InteropServices;
 using Microsoft.Win32.SafeHandles;
 
-namespace SharpEmu.Libs.Pad;
+namespace SharpEmu.HLE.Host.Windows;
 
 /// <summary>
 /// Minimal Win32 HID interop used to talk to a DualSense controller
 /// directly, without any external input library.
 /// </summary>
-internal static partial class HidNative
+internal static partial class WindowsHidNative
 {
     internal const int DigcfPresent = 0x02;
     internal const int DigcfDeviceInterface = 0x10;
@@ -38,28 +38,32 @@ internal static partial class HidNative
         public ushort VersionNumber;
     }
 
-    [DllImport("hid.dll")]
-    internal static extern void HidD_GetHidGuid(out Guid hidGuid);
+    [LibraryImport("hid.dll")]
+    internal static partial void HidD_GetHidGuid(out Guid hidGuid);
 
-    [DllImport("hid.dll")]
-    internal static extern bool HidD_GetAttributes(SafeFileHandle hidDeviceObject, ref HiddAttributes attributes);
+    [LibraryImport("hid.dll")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    internal static partial bool HidD_GetAttributes(SafeFileHandle hidDeviceObject, ref HiddAttributes attributes);
 
-    [DllImport("hid.dll")]
-    internal static extern bool HidD_GetFeature(SafeFileHandle hidDeviceObject, byte[] reportBuffer, int reportBufferLength);
+    [LibraryImport("hid.dll")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    internal static partial bool HidD_GetFeature(SafeFileHandle hidDeviceObject, [In, Out] byte[] reportBuffer, int reportBufferLength);
 
-    [DllImport("setupapi.dll", CharSet = CharSet.Unicode)]
-    internal static extern nint SetupDiGetClassDevs(ref Guid classGuid, nint enumerator, nint hwndParent, int flags);
+    [LibraryImport("setupapi.dll", EntryPoint = "SetupDiGetClassDevsW")]
+    internal static partial nint SetupDiGetClassDevs(ref Guid classGuid, nint enumerator, nint hwndParent, int flags);
 
-    [DllImport("setupapi.dll")]
-    internal static extern bool SetupDiEnumDeviceInterfaces(
+    [LibraryImport("setupapi.dll")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    internal static partial bool SetupDiEnumDeviceInterfaces(
         nint deviceInfoSet,
         nint deviceInfoData,
         ref Guid interfaceClassGuid,
         int memberIndex,
         ref SpDeviceInterfaceData deviceInterfaceData);
 
-    [DllImport("setupapi.dll", CharSet = CharSet.Unicode)]
-    internal static extern bool SetupDiGetDeviceInterfaceDetail(
+    [LibraryImport("setupapi.dll", EntryPoint = "SetupDiGetDeviceInterfaceDetailW")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    internal static partial bool SetupDiGetDeviceInterfaceDetail(
         nint deviceInfoSet,
         ref SpDeviceInterfaceData deviceInterfaceData,
         nint deviceInterfaceDetailData,
@@ -67,11 +71,12 @@ internal static partial class HidNative
         out int requiredSize,
         nint deviceInfoData);
 
-    [DllImport("setupapi.dll")]
-    internal static extern bool SetupDiDestroyDeviceInfoList(nint deviceInfoSet);
+    [LibraryImport("setupapi.dll")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    internal static partial bool SetupDiDestroyDeviceInfoList(nint deviceInfoSet);
 
-    [DllImport("kernel32.dll", CharSet = CharSet.Unicode, SetLastError = true)]
-    internal static extern SafeFileHandle CreateFile(
+    [LibraryImport("kernel32.dll", EntryPoint = "CreateFileW", SetLastError = true, StringMarshalling = StringMarshalling.Utf16)]
+    internal static partial SafeFileHandle CreateFile(
         string fileName,
         uint desiredAccess,
         uint shareMode,

src/SharpEmu.HLE/Host/Windows/WindowsHostInput.cs

@@ -0,0 +1,101 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Windows;
+
+/// <summary>
+/// Windows input backend: DualSense over raw HID plus XInput controllers for gamepads,
+/// user32 for the keyboard-fallback queries. Rumble fans out to every reader; lightbar
+/// only exists on the DualSense.
+/// </summary>
+internal sealed partial class WindowsHostInput : IHostInput
+{
+    public void EnsureStarted()
+    {
+        WindowsDualSenseReader.EnsureStarted();
+        WindowsXInputReader.EnsureStarted();
+    }
+
+    public int GetGamepadStates(Span<HostGamepadState> destination)
+    {
+        var count = 0;
+        if (count < destination.Length && WindowsDualSenseReader.TryGetState(out var dualSense))
+        {
+            destination[count++] = dualSense;
+        }
+
+        if (count < destination.Length && WindowsXInputReader.TryGetState(out var xinput))
+        {
+            destination[count++] = xinput;
+        }
+
+        return count;
+    }
+
+    public string? DescribeConnectedGamepad()
+    {
+        if (WindowsDualSenseReader.TryGetState(out _))
+        {
+            return "DualSense";
+        }
+
+        return WindowsXInputReader.TryGetState(out _) ? "Xbox controller" : null;
+    }
+
+    public void SetRumble(byte largeMotor, byte smallMotor)
+    {
+        WindowsDualSenseReader.SetRumble(largeMotor, smallMotor);
+        WindowsXInputReader.SetRumble(largeMotor, smallMotor);
+    }
+
+    public void SetTriggerRumble(byte? leftTrigger, byte? rightTrigger) =>
+        WindowsXInputReader.SetTriggerRumble(leftTrigger, rightTrigger);
+
+    public void SetLightbar(byte red, byte green, byte blue) =>
+        WindowsDualSenseReader.SetLightbar(red, green, blue);
+
+    public void ResetLightbar() => WindowsDualSenseReader.ResetLightbar();
+
+    public bool IsHostWindowFocused()
+    {
+        var foregroundWindow = GetForegroundWindow();
+        if (foregroundWindow == 0)
+        {
+            return false;
+        }
+
+        GetWindowThreadProcessId(foregroundWindow, out var processId);
+        if (processId == (uint)Environment.ProcessId)
+        {
+            return true;
+        }
+
+        // The GUI runs the emulator in an isolated child process. Its native
+        // Vulkan surface is a child of the GUI window, so the foreground
+        // window belongs to the launcher process rather than this one.
+        var embeddedHostWindow = HostSessionControl.EmbeddedHostWindow;
+        var hostTopLevelWindow = embeddedHostWindow == 0
+            ? 0
+            : GetAncestor(embeddedHostWindow, GetAncestorRoot);
+        return hostTopLevelWindow != 0 && foregroundWindow == hostTopLevelWindow;
+    }
+
+    public bool IsKeyDown(int virtualKey) =>
+        (GetAsyncKeyState(virtualKey) & 0x8000) != 0;
+
+    [LibraryImport("user32.dll")]
+    private static partial short GetAsyncKeyState(int vKey);
+
+    [LibraryImport("user32.dll")]
+    private static partial nint GetForegroundWindow();
+
+    [LibraryImport("user32.dll")]
+    private static partial uint GetWindowThreadProcessId(nint hWnd, out uint processId);
+
+    [LibraryImport("user32.dll")]
+    private static partial nint GetAncestor(nint hWnd, uint gaFlags);
+
+    private const uint GetAncestorRoot = 2;
+}

src/SharpEmu.HLE/Host/Windows/WindowsHostPlatform.cs

@@ -10,4 +10,8 @@ internal sealed class WindowsHostPlatform : IHostPlatform
     public IHostThreading Threading { get; } = new WindowsHostThreading();
 
     public IHostSymbolResolver Symbols { get; } = new WindowsHostSymbolResolver();
+
+    public IHostAudioOutput Audio { get; } = new WindowsWaveOutAudio();
+
+    public IHostInput Input { get; } = new WindowsHostInput();
 }

src/SharpEmu.HLE/Host/Windows/WindowsHostThreading.cs

@@ -23,6 +23,31 @@ internal sealed unsafe partial class WindowsHostThreading : IHostThreading
     private const int CtxRbp = 160;
     private const int CtxRip = 248;
 
+    private static int _timerResolutionRequested;
+
+    public void RequestTimerResolution()
+    {
+        if (Interlocked.Exchange(ref _timerResolutionRequested, 1) != 0)
+        {
+            return;
+        }
+
+        try
+        {
+            if (TimeBeginPeriod(1) != 0)
+            {
+                Console.Error.WriteLine(
+                    "[LOADER][WARN] Host timer resolution request rejected; " +
+                    "timed waits keep the default ~15.6 ms granularity.");
+            }
+        }
+        catch (DllNotFoundException exception)
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][WARN] Host timer resolution unavailable: {exception.Message}");
+        }
+    }
+
     public uint AllocateTlsSlot() => TlsAlloc();
 
     public bool FreeTlsSlot(uint slot) => TlsFree(slot);
@@ -162,4 +187,7 @@ internal sealed unsafe partial class WindowsHostThreading : IHostThreading
     [LibraryImport("kernel32.dll", SetLastError = true)]
     [return: MarshalAs(UnmanagedType.Bool)]
     private static partial bool CloseHandle(nint hObject);
+
+    [LibraryImport("winmm.dll", EntryPoint = "timeBeginPeriod")]
+    private static partial uint TimeBeginPeriod(uint uPeriod);
 }

src/SharpEmu.HLE/Host/Windows/WindowsWaveOutAudio.cs

@@ -0,0 +1,243 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE.Host.Windows;
+
+internal sealed partial class WindowsWaveOutAudio : IHostAudioOutput
+{
+    public string BackendName => "winmm";
+
+    public IHostAudioStream OpenStereoPcm16Stream(uint sampleRate) => new WaveOutStream(sampleRate);
+
+    private sealed partial class WaveOutStream : IHostAudioStream
+    {
+        private const uint WaveMapper = uint.MaxValue;
+        private const uint CallbackEvent = 0x0005_0000;
+        private const ushort WaveFormatPcm = 1;
+        private const uint WaveHeaderDone = 0x0000_0001;
+        private const int MaximumQueuedPcmBytes = 32 * 1024;
+
+        private readonly object _gate = new();
+        private readonly AutoResetEvent _completion = new(false);
+        private readonly Queue<NativeBuffer> _buffers = new();
+        private IntPtr _device;
+        private int _queuedPcmBytes;
+        private bool _disposed;
+
+        public WaveOutStream(uint sampleRate)
+        {
+            var format = new WaveFormat
+            {
+                FormatTag = WaveFormatPcm,
+                Channels = 2,
+                SamplesPerSecond = sampleRate,
+                AverageBytesPerSecond = checked(sampleRate * 4),
+                BlockAlign = 4,
+                BitsPerSample = 16,
+                ExtraSize = 0,
+            };
+            var result = WaveOutOpen(
+                out _device,
+                WaveMapper,
+                ref format,
+                _completion.SafeWaitHandle.DangerousGetHandle(),
+                IntPtr.Zero,
+                CallbackEvent);
+            if (result != 0)
+            {
+                throw new InvalidOperationException($"waveOutOpen failed with MMRESULT {result}.");
+            }
+        }
+
+        public bool Submit(ReadOnlySpan<byte> stereoPcm16)
+        {
+            lock (_gate)
+            {
+                if (_disposed)
+                {
+                    return false;
+                }
+
+                ReapCompletedBuffers();
+                while (_queuedPcmBytes != 0 &&
+                       _queuedPcmBytes + stereoPcm16.Length > MaximumQueuedPcmBytes)
+                {
+                    if (!_completion.WaitOne(TimeSpan.FromSeconds(1)))
+                    {
+                        return false;
+                    }
+
+                    ReapCompletedBuffers();
+                }
+
+                return QueueBuffer(stereoPcm16);
+            }
+        }
+
+        public void Dispose()
+        {
+            lock (_gate)
+            {
+                if (_disposed)
+                {
+                    return;
+                }
+
+                _disposed = true;
+                if (_device != IntPtr.Zero)
+                {
+                    WaveOutReset(_device);
+                    while (_buffers.TryDequeue(out var buffer))
+                    {
+                        ReleaseBuffer(buffer);
+                    }
+
+                    WaveOutClose(_device);
+                    _device = IntPtr.Zero;
+                }
+
+                _completion.Dispose();
+            }
+        }
+
+        private bool QueueBuffer(ReadOnlySpan<byte> data)
+        {
+            var dataAddress = Marshal.AllocHGlobal(data.Length);
+            var headerAddress = IntPtr.Zero;
+            try
+            {
+                unsafe
+                {
+                    data.CopyTo(new Span<byte>((void*)dataAddress, data.Length));
+                }
+
+                var header = new WaveHeader
+                {
+                    Data = dataAddress,
+                    BufferLength = checked((uint)data.Length),
+                };
+                headerAddress = Marshal.AllocHGlobal(Marshal.SizeOf<WaveHeader>());
+                Marshal.StructureToPtr(header, headerAddress, false);
+
+                var result = WaveOutPrepareHeader(
+                    _device,
+                    headerAddress,
+                    checked((uint)Marshal.SizeOf<WaveHeader>()));
+                if (result != 0)
+                {
+                    return false;
+                }
+
+                result = WaveOutWrite(
+                    _device,
+                    headerAddress,
+                    checked((uint)Marshal.SizeOf<WaveHeader>()));
+                if (result != 0)
+                {
+                    WaveOutUnprepareHeader(
+                        _device,
+                        headerAddress,
+                        checked((uint)Marshal.SizeOf<WaveHeader>()));
+                    return false;
+                }
+
+                _buffers.Enqueue(new NativeBuffer(dataAddress, headerAddress, data.Length));
+                _queuedPcmBytes += data.Length;
+                dataAddress = IntPtr.Zero;
+                headerAddress = IntPtr.Zero;
+                return true;
+            }
+            finally
+            {
+                if (headerAddress != IntPtr.Zero)
+                {
+                    Marshal.FreeHGlobal(headerAddress);
+                }
+
+                if (dataAddress != IntPtr.Zero)
+                {
+                    Marshal.FreeHGlobal(dataAddress);
+                }
+            }
+        }
+
+        private void ReapCompletedBuffers()
+        {
+            while (_buffers.TryPeek(out var buffer))
+            {
+                var header = Marshal.PtrToStructure<WaveHeader>(buffer.Header);
+                if ((header.Flags & WaveHeaderDone) == 0)
+                {
+                    return;
+                }
+
+                _buffers.Dequeue();
+                ReleaseBuffer(buffer);
+            }
+        }
+
+        private void ReleaseBuffer(NativeBuffer buffer)
+        {
+            WaveOutUnprepareHeader(
+                _device,
+                buffer.Header,
+                checked((uint)Marshal.SizeOf<WaveHeader>()));
+            _queuedPcmBytes -= buffer.Length;
+            Marshal.FreeHGlobal(buffer.Header);
+            Marshal.FreeHGlobal(buffer.Data);
+        }
+
+        private readonly record struct NativeBuffer(IntPtr Data, IntPtr Header, int Length);
+
+        [StructLayout(LayoutKind.Sequential, Pack = 2)]
+        private struct WaveFormat
+        {
+            public ushort FormatTag;
+            public ushort Channels;
+            public uint SamplesPerSecond;
+            public uint AverageBytesPerSecond;
+            public ushort BlockAlign;
+            public ushort BitsPerSample;
+            public ushort ExtraSize;
+        }
+
+        [StructLayout(LayoutKind.Sequential)]
+        private struct WaveHeader
+        {
+            public IntPtr Data;
+            public uint BufferLength;
+            public uint BytesRecorded;
+            public nuint User;
+            public uint Flags;
+            public uint Loops;
+            public IntPtr Next;
+            public nuint Reserved;
+        }
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutOpen")]
+        private static partial uint WaveOutOpen(
+            out IntPtr device,
+            uint deviceId,
+            ref WaveFormat format,
+            IntPtr callback,
+            IntPtr instance,
+            uint flags);
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutPrepareHeader")]
+        private static partial uint WaveOutPrepareHeader(IntPtr device, IntPtr header, uint headerSize);
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutWrite")]
+        private static partial uint WaveOutWrite(IntPtr device, IntPtr header, uint headerSize);
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutUnprepareHeader")]
+        private static partial uint WaveOutUnprepareHeader(IntPtr device, IntPtr header, uint headerSize);
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutReset")]
+        private static partial uint WaveOutReset(IntPtr device);
+
+        [LibraryImport("winmm.dll", EntryPoint = "waveOutClose")]
+        private static partial uint WaveOutClose(IntPtr device);
+    }
+}

src/SharpEmu.HLE/Host/Windows/WindowsXInputReader.cs

@@ -3,15 +3,15 @@
 
 using System.Runtime.InteropServices;
 
-namespace SharpEmu.Libs.Pad;
+namespace SharpEmu.HLE.Host.Windows;
 
 /// <summary>
 /// Reads Xbox 360 / Xbox One (and other XInput-compatible) controllers via
-/// the Windows XInput API on a background thread, translated to the same
-/// ORBIS pad conventions as <see cref="DualSenseReader"/>. Supports rumble
-/// and hot-plug retry; the first connected slot (of four) is used.
+/// the Windows XInput API on a background thread, translated to
+/// <see cref="HostGamepadState"/> conventions. Supports rumble and hot-plug
+/// retry; the first connected slot (of four) is used.
 /// </summary>
-internal static class XInputReader
+public static partial class WindowsXInputReader
 {
     private const uint ErrorSuccess = 0;
     private const int SlotCount = 4;
@@ -34,7 +34,7 @@ internal static class XInputReader
     private const ushort XinputY = 0x8000;
 
     private static readonly object Gate = new();
-    private static PadState _state;
+    private static HostGamepadState _state;
     private static bool _started;
     private static int _slot = -1; // connected XInput user index, -1 when none
     private static byte _motorLeft;
@@ -43,8 +43,10 @@ internal static class XInputReader
     private static byte _triggerRight;
 
     /// <summary>Starts the background reader once; safe to call repeatedly.</summary>
-    internal static void EnsureStarted()
+    public static void EnsureStarted()
     {
+        // The GUI source-links this reader and calls it directly, without the
+        // host-platform resolution that otherwise guarantees Windows.
         if (!OperatingSystem.IsWindows())
         {
             return;
@@ -67,7 +69,7 @@ internal static class XInputReader
         }
     }
 
-    internal static bool TryGetState(out PadState state)
+    public static bool TryGetState(out HostGamepadState state)
     {
         lock (Gate)
         {
@@ -77,7 +79,7 @@ internal static class XInputReader
         return state.Connected;
     }
 
-    private static void SetState(in PadState state)
+    private static void SetState(in HostGamepadState state)
     {
         lock (Gate)
         {
@@ -204,40 +206,40 @@ internal static class XInputReader
         return -1;
     }
 
-    private static PadState Translate(in XInputGamepad pad)
+    private static HostGamepadState Translate(in XInputGamepad pad)
     {
-        uint buttons = 0;
-        buttons |= (pad.Buttons & XinputDpadUp) != 0 ? OrbisPadButton.Up : 0;
-        buttons |= (pad.Buttons & XinputDpadDown) != 0 ? OrbisPadButton.Down : 0;
-        buttons |= (pad.Buttons & XinputDpadLeft) != 0 ? OrbisPadButton.Left : 0;
-        buttons |= (pad.Buttons & XinputDpadRight) != 0 ? OrbisPadButton.Right : 0;
-        buttons |= (pad.Buttons & XinputStart) != 0 ? OrbisPadButton.Options : 0;
-        buttons |= (pad.Buttons & XinputBack) != 0 ? OrbisPadButton.TouchPad : 0;
-        buttons |= (pad.Buttons & XinputLeftThumb) != 0 ? OrbisPadButton.L3 : 0;
-        buttons |= (pad.Buttons & XinputRightThumb) != 0 ? OrbisPadButton.R3 : 0;
-        buttons |= (pad.Buttons & XinputLeftShoulder) != 0 ? OrbisPadButton.L1 : 0;
-        buttons |= (pad.Buttons & XinputRightShoulder) != 0 ? OrbisPadButton.R1 : 0;
-        buttons |= (pad.Buttons & XinputA) != 0 ? OrbisPadButton.Cross : 0;
-        buttons |= (pad.Buttons & XinputB) != 0 ? OrbisPadButton.Circle : 0;
-        buttons |= (pad.Buttons & XinputX) != 0 ? OrbisPadButton.Square : 0;
-        buttons |= (pad.Buttons & XinputY) != 0 ? OrbisPadButton.Triangle : 0;
-        buttons |= pad.LeftTrigger > TriggerThreshold ? OrbisPadButton.L2 : 0;
-        buttons |= pad.RightTrigger > TriggerThreshold ? OrbisPadButton.R2 : 0;
+        var buttons = HostGamepadButtons.None;
+        buttons |= (pad.Buttons & XinputDpadUp) != 0 ? HostGamepadButtons.Up : 0;
+        buttons |= (pad.Buttons & XinputDpadDown) != 0 ? HostGamepadButtons.Down : 0;
+        buttons |= (pad.Buttons & XinputDpadLeft) != 0 ? HostGamepadButtons.Left : 0;
+        buttons |= (pad.Buttons & XinputDpadRight) != 0 ? HostGamepadButtons.Right : 0;
+        buttons |= (pad.Buttons & XinputStart) != 0 ? HostGamepadButtons.Options : 0;
+        buttons |= (pad.Buttons & XinputBack) != 0 ? HostGamepadButtons.TouchPad : 0;
+        buttons |= (pad.Buttons & XinputLeftThumb) != 0 ? HostGamepadButtons.L3 : 0;
+        buttons |= (pad.Buttons & XinputRightThumb) != 0 ? HostGamepadButtons.R3 : 0;
+        buttons |= (pad.Buttons & XinputLeftShoulder) != 0 ? HostGamepadButtons.L1 : 0;
+        buttons |= (pad.Buttons & XinputRightShoulder) != 0 ? HostGamepadButtons.R1 : 0;
+        buttons |= (pad.Buttons & XinputA) != 0 ? HostGamepadButtons.Cross : 0;
+        buttons |= (pad.Buttons & XinputB) != 0 ? HostGamepadButtons.Circle : 0;
+        buttons |= (pad.Buttons & XinputX) != 0 ? HostGamepadButtons.Square : 0;
+        buttons |= (pad.Buttons & XinputY) != 0 ? HostGamepadButtons.Triangle : 0;
+        buttons |= pad.LeftTrigger > TriggerThreshold ? HostGamepadButtons.L2 : 0;
+        buttons |= pad.RightTrigger > TriggerThreshold ? HostGamepadButtons.R2 : 0;
 
-        return new PadState(
+        return new HostGamepadState(
             Connected: true,
             Buttons: buttons,
             LeftX: AxisToByte(pad.ThumbLX),
             LeftY: AxisToByteInverted(pad.ThumbLY),
             RightX: AxisToByte(pad.ThumbRX),
             RightY: AxisToByteInverted(pad.ThumbRY),
-            L2: pad.LeftTrigger,
-            R2: pad.RightTrigger);
+            LeftTrigger: pad.LeftTrigger,
+            RightTrigger: pad.RightTrigger);
     }
 
     private static byte AxisToByte(short value) => (byte)((value + 32768) >> 8);
 
-    // XInput Y grows upward, ORBIS pads report Y growing downward.
+    // XInput Y grows upward, host pad conventions report Y growing downward.
     private static byte AxisToByteInverted(short value) => (byte)(255 - ((value + 32768) >> 8));
 
     [StructLayout(LayoutKind.Sequential)]
@@ -267,9 +269,9 @@ internal static class XInputReader
     }
 
     // xinput1_4.dll ships with Windows 8 and later.
-    [DllImport("xinput1_4.dll")]
-    private static extern uint XInputGetState(uint userIndex, out XInputState state);
+    [LibraryImport("xinput1_4.dll")]
+    private static partial uint XInputGetState(uint userIndex, out XInputState state);
 
-    [DllImport("xinput1_4.dll")]
-    private static extern uint XInputSetState(uint userIndex, ref XInputVibration vibration);
+    [LibraryImport("xinput1_4.dll")]
+    private static partial uint XInputSetState(uint userIndex, ref XInputVibration vibration);
 }

src/SharpEmu.HLE/HostMainThread.cs

@@ -0,0 +1,78 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Collections.Concurrent;
+
+namespace SharpEmu.HLE;
+
+/// <summary>
+/// Runs work on the real process main thread. macOS only allows AppKit (and
+/// therefore GLFW windowing) on that thread, so the CLI moves emulation onto
+/// a worker thread, parks the main thread in <see cref="Pump"/>, and the
+/// video presenter posts its window loop here. On other platforms
+/// <see cref="IsAvailable"/> stays false and nothing changes.
+/// </summary>
+public static class HostMainThread
+{
+    private static readonly BlockingCollection<Action> _work = new();
+    private static Action? _shutdownRequestHandler;
+
+    public static bool IsAvailable { get; private set; }
+
+    /// <summary>
+    /// Registers a callback invoked by <see cref="Shutdown"/> so a
+    /// long-running posted work item (the presenter's window loop) can be
+    /// asked to return to the pump.
+    /// </summary>
+    public static void SetShutdownRequestHandler(Action handler) =>
+        _shutdownRequestHandler = handler;
+
+    /// <summary>Marks the pump as present. Call before guest code can run.</summary>
+    public static void Enable() => IsAvailable = true;
+
+    public static void Post(Action work)
+    {
+        try
+        {
+            _work.Add(work);
+        }
+        catch (InvalidOperationException)
+        {
+            // Shutdown already requested; the process is exiting.
+        }
+    }
+
+    /// <summary>
+    /// Services posted work on the calling (main) thread until
+    /// <see cref="Shutdown"/> is called and the queue drains.
+    /// </summary>
+    public static void Pump()
+    {
+        foreach (var work in _work.GetConsumingEnumerable())
+        {
+            try
+            {
+                work();
+            }
+            catch (Exception exception)
+            {
+                Console.Error.WriteLine($"[LOADER][ERROR] Main-thread work failed: {exception}");
+            }
+        }
+    }
+
+    public static void Shutdown()
+    {
+        IsAvailable = false;
+        try
+        {
+            _shutdownRequestHandler?.Invoke();
+        }
+        catch (Exception exception)
+        {
+            Console.Error.WriteLine($"[LOADER][WARN] Main-thread shutdown handler failed: {exception.Message}");
+        }
+
+        _work.CompleteAdding();
+    }
+}

src/SharpEmu.HLE/HostMemory.cs

@@ -0,0 +1,498 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Runtime.InteropServices;
+
+namespace SharpEmu.HLE;
+
+/// <summary>
+/// Cross-platform host virtual memory API with Win32 semantics.
+/// On Windows this forwards directly to kernel32. On POSIX systems it is
+/// implemented over mmap/mprotect/munmap with a shadow region table that
+/// answers VirtualQuery-style questions and tracks page protections.
+/// POSIX anonymous mappings are demand-paged by the kernel, so Win32
+/// "reserve-only" regions are mapped as committed memory directly and
+/// commit requests become protection changes.
+/// </summary>
+public static unsafe class HostMemory
+{
+    public const uint MEM_COMMIT = 0x1000;
+    public const uint MEM_RESERVE = 0x2000;
+    public const uint MEM_RELEASE = 0x8000;
+    public const uint MEM_FREE_STATE = 0x10000;
+    public const uint MEM_PRIVATE = 0x20000;
+
+    public const uint PAGE_NOACCESS = 0x01;
+    public const uint PAGE_READONLY = 0x02;
+    public const uint PAGE_READWRITE = 0x04;
+    public const uint PAGE_EXECUTE = 0x10;
+    public const uint PAGE_EXECUTE_READ = 0x20;
+    public const uint PAGE_EXECUTE_READWRITE = 0x40;
+
+    private const ulong PageSize = 0x1000;
+
+    /// <summary>Win32 MEMORY_BASIC_INFORMATION (64-bit) layout.</summary>
+    public struct BasicInfo
+    {
+        public ulong BaseAddress;
+        public ulong AllocationBase;
+        public uint AllocationProtect;
+        public uint Alignment1;
+        public ulong RegionSize;
+        public uint State;
+        public uint Protect;
+        public uint Type;
+        public uint Alignment2;
+    }
+
+    public static void* Alloc(void* address, nuint size, uint allocationType, uint protect)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            return Win32VirtualAlloc(address, size, allocationType, protect);
+        }
+
+        return Posix.Alloc(address, size, allocationType, protect);
+    }
+
+    public static bool Free(void* address, nuint size, uint freeType)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            return Win32VirtualFree(address, size, freeType);
+        }
+
+        return Posix.Free(address, size, freeType);
+    }
+
+    public static bool Protect(void* address, nuint size, uint newProtect, out uint oldProtect)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            return Win32VirtualProtect(address, size, newProtect, out oldProtect);
+        }
+
+        return Posix.Protect(address, size, newProtect, out oldProtect);
+    }
+
+    public static nuint Query(void* address, out BasicInfo info)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            return Win32VirtualQuery(address, out info, (nuint)sizeof(BasicInfo));
+        }
+
+        return Posix.Query(address, out info);
+    }
+
+    public static void FlushInstructionCache(void* address, nuint size)
+    {
+        if (OperatingSystem.IsWindows())
+        {
+            Win32FlushInstructionCache(Win32GetCurrentProcess(), address, size);
+            return;
+        }
+
+        // The emulator only executes x86-64 guest code, so a non-Windows host
+        // is either x86-64 (including Rosetta 2 translation) with a coherent
+        // instruction cache, or would need sys_icache_invalidate for a future
+        // arm64 recompiler. Nothing to do today.
+    }
+
+    [DllImport("kernel32.dll", EntryPoint = "VirtualAlloc", SetLastError = true)]
+    private static extern void* Win32VirtualAlloc(void* lpAddress, nuint dwSize, uint flAllocationType, uint flProtect);
+
+    [DllImport("kernel32.dll", EntryPoint = "VirtualFree", SetLastError = true)]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool Win32VirtualFree(void* lpAddress, nuint dwSize, uint dwFreeType);
+
+    [DllImport("kernel32.dll", EntryPoint = "VirtualProtect", SetLastError = true)]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool Win32VirtualProtect(void* lpAddress, nuint dwSize, uint flNewProtect, out uint lpflOldProtect);
+
+    [DllImport("kernel32.dll", EntryPoint = "VirtualQuery")]
+    private static extern nuint Win32VirtualQuery(void* lpAddress, out BasicInfo lpBuffer, nuint dwLength);
+
+    [DllImport("kernel32.dll", EntryPoint = "GetCurrentProcess")]
+    private static extern void* Win32GetCurrentProcess();
+
+    [DllImport("kernel32.dll", EntryPoint = "FlushInstructionCache")]
+    [return: MarshalAs(UnmanagedType.Bool)]
+    private static extern bool Win32FlushInstructionCache(void* hProcess, void* lpBaseAddress, nuint dwSize);
+
+    private static class Posix
+    {
+        private const int PROT_NONE = 0x0;
+        private const int PROT_READ = 0x1;
+        private const int PROT_WRITE = 0x2;
+        private const int PROT_EXEC = 0x4;
+
+        private const int MAP_PRIVATE = 0x02;
+        private const int MAP_FIXED = 0x10;
+        private static readonly int MAP_ANON = OperatingSystem.IsMacOS() ? 0x1000 : 0x20;
+        private static readonly int MAP_NORESERVE = OperatingSystem.IsMacOS() ? 0 : 0x4000;
+
+        // Linux-only: fail instead of clobbering an existing mapping.
+        private const int MAP_FIXED_NOREPLACE = 0x100000;
+
+        private static readonly nint MAP_FAILED = -1;
+
+        private static readonly object Gate = new();
+        private static readonly SortedList<ulong, Region> Regions = new();
+
+        private sealed class Region
+        {
+            public ulong Base;
+            public ulong Size;
+            public uint DefaultProtect;
+            public Dictionary<ulong, uint>? PageProtects;
+
+            public ulong End => Base + Size;
+
+            public uint ProtectAt(ulong pageAddress)
+            {
+                if (PageProtects is not null && PageProtects.TryGetValue(pageAddress, out var overriden))
+                {
+                    return overriden;
+                }
+
+                return DefaultProtect;
+            }
+        }
+
+        public static void* Alloc(void* address, nuint size, uint allocationType, uint protect)
+        {
+            if (size == 0)
+            {
+                return null;
+            }
+
+            var alignedSize = AlignUp((ulong)size, PageSize);
+
+            lock (Gate)
+            {
+                if (allocationType == MEM_COMMIT && address != null &&
+                    TryFindRegionLocked((ulong)address, out var existing))
+                {
+                    // Note: MEM_RESERVE requests that overlap an existing
+                    // region must fail like Win32 does; only a pure commit
+                    // may target pages inside a tracked mapping.
+                    // Commit inside an existing mapping: the pages are already
+                    // backed (demand paged), so only apply the protection.
+                    var start = AlignDown((ulong)address, PageSize);
+                    var end = Math.Min(existing.End, AlignUp((ulong)address + alignedSize, PageSize));
+                    if (end <= start)
+                    {
+                        return null;
+                    }
+
+                    if (mprotect((nint)start, (nuint)(end - start), ToPosixProtect(protect)) != 0)
+                    {
+                        return null;
+                    }
+
+                    SetProtectRangeLocked(existing, start, end - start, protect);
+                    return address;
+                }
+
+                if ((allocationType & MEM_RESERVE) == 0)
+                {
+                    // MEM_COMMIT alone outside any known region is invalid here.
+                    return null;
+                }
+
+                var posixProtect = ToPosixProtect(protect);
+                var flags = MAP_PRIVATE | MAP_ANON;
+                if ((allocationType & MEM_COMMIT) == 0)
+                {
+                    // Reserve-only: keep the requested protection so the region
+                    // is usable without a separate commit step, but tell the
+                    // kernel not to account swap for it where supported.
+                    flags |= MAP_NORESERVE;
+                }
+
+                nint result;
+                if (address != null)
+                {
+                    // Win32 maps at exactly the requested address or fails
+                    // without touching existing mappings. Fail up front on
+                    // any overlap we track, then place the mapping: Linux
+                    // gets MAP_FIXED_NOREPLACE (fails cleanly on host
+                    // mappings too). Darwin lacks NOREPLACE and plain
+                    // MAP_FIXED would silently clobber untracked host
+                    // memory (dyld, the runtime's JIT heap, Rosetta), so
+                    // pass the address as a hint instead -- the kernel
+                    // honors it when the range is free and relocates the
+                    // mapping otherwise, which we treat as failure.
+                    if (OverlapsTrackedRegionLocked((ulong)address, alignedSize))
+                    {
+                        Trace($"exact overlap: addr=0x{(ulong)address:X16} size=0x{alignedSize:X}");
+                        return null;
+                    }
+
+                    var exactFlags = OperatingSystem.IsMacOS() ? flags : flags | MAP_FIXED_NOREPLACE;
+                    result = mmap((nint)address, (nuint)alignedSize, posixProtect, exactFlags, -1, 0);
+                    if (result == MAP_FAILED || (ulong)result != (ulong)address)
+                    {
+                        Trace($"exact mmap failed: addr=0x{(ulong)address:X16} got=0x{(ulong)result:X16} size=0x{alignedSize:X} errno={Marshal.GetLastPInvokeError()}");
+                        if (result != MAP_FAILED)
+                        {
+                            munmap(result, (nuint)alignedSize);
+                        }
+
+                        return null;
+                    }
+                }
+                else
+                {
+                    result = mmap(0, (nuint)alignedSize, posixProtect, flags, -1, 0);
+                    if (result == MAP_FAILED)
+                    {
+                        Trace($"mmap failed: size=0x{alignedSize:X} errno={Marshal.GetLastPInvokeError()}");
+                        return null;
+                    }
+                }
+
+                Regions[(ulong)result] = new Region
+                {
+                    Base = (ulong)result,
+                    Size = alignedSize,
+                    DefaultProtect = protect
+                };
+
+                return (void*)result;
+            }
+        }
+
+        public static bool Free(void* address, nuint size, uint freeType)
+        {
+            _ = size;
+            _ = freeType;
+
+            lock (Gate)
+            {
+                if (!Regions.TryGetValue((ulong)address, out var region))
+                {
+                    return false;
+                }
+
+                Regions.Remove((ulong)address);
+                return munmap((nint)address, (nuint)region.Size) == 0;
+            }
+        }
+
+        public static bool Protect(void* address, nuint size, uint newProtect, out uint oldProtect)
+        {
+            oldProtect = PAGE_NOACCESS;
+            if (size == 0)
+            {
+                return false;
+            }
+
+            var start = AlignDown((ulong)address, PageSize);
+            var end = AlignUp((ulong)address + size, PageSize);
+
+            lock (Gate)
+            {
+                if (!TryFindRegionLocked(start, out var region) || end > region.End)
+                {
+                    return false;
+                }
+
+                oldProtect = region.ProtectAt(start);
+                if (mprotect((nint)start, (nuint)(end - start), ToPosixProtect(newProtect)) != 0)
+                {
+                    return false;
+                }
+
+                SetProtectRangeLocked(region, start, end - start, newProtect);
+                return true;
+            }
+        }
+
+        public static nuint Query(void* address, out BasicInfo info)
+        {
+            info = default;
+            var pageAddress = AlignDown((ulong)address, PageSize);
+
+            lock (Gate)
+            {
+                if (TryFindRegionLocked(pageAddress, out var region))
+                {
+                    // Win32 VirtualQuery reports a run of pages sharing the
+                    // same protection, so stop the run where it changes. Do
+                    // not walk the whole mapping page-by-page here: PS5 GPU
+                    // apertures can span hundreds of GiB, while protection
+                    // overrides are sparse. A tiny read inside such a mapping
+                    // otherwise performs tens of millions of dictionary
+                    // lookups before it can return.
+                    var pageProtects = region.PageProtects;
+                    uint protect;
+                    ulong runEnd;
+                    if (pageProtects is null || pageProtects.Count == 0)
+                    {
+                        protect = region.DefaultProtect;
+                        runEnd = region.End;
+                    }
+                    else if (pageProtects.TryGetValue(pageAddress, out protect))
+                    {
+                        runEnd = pageAddress + PageSize;
+                        while (runEnd < region.End &&
+                            pageProtects.TryGetValue(runEnd, out var nextProtect) &&
+                            nextProtect == protect)
+                        {
+                            runEnd += PageSize;
+                        }
+                    }
+                    else
+                    {
+                        protect = region.DefaultProtect;
+                        runEnd = region.End;
+                        foreach (var overrideAddress in pageProtects.Keys)
+                        {
+                            if (overrideAddress > pageAddress && overrideAddress < runEnd)
+                            {
+                                runEnd = overrideAddress;
+                            }
+                        }
+                    }
+
+                    info.BaseAddress = pageAddress;
+                    info.AllocationBase = region.Base;
+                    info.AllocationProtect = region.DefaultProtect;
+                    info.RegionSize = runEnd - pageAddress;
+                    info.State = MEM_COMMIT;
+                    info.Protect = protect;
+                    info.Type = MEM_PRIVATE;
+                    return (nuint)sizeof(BasicInfo);
+                }
+
+                // Untracked host memory (runtime heaps, stacks, libraries) is
+                // reported as a free block reaching to the next tracked region
+                // so scanning callers keep advancing.
+                var nextBase = ulong.MaxValue;
+                foreach (var regionBase in Regions.Keys)
+                {
+                    if (regionBase > pageAddress)
+                    {
+                        nextBase = regionBase;
+                        break;
+                    }
+                }
+
+                info.BaseAddress = pageAddress;
+                info.AllocationBase = 0;
+                info.AllocationProtect = PAGE_NOACCESS;
+                info.RegionSize = (nextBase == ulong.MaxValue ? pageAddress + PageSize : nextBase) - pageAddress;
+                info.State = MEM_FREE_STATE;
+                info.Protect = PAGE_NOACCESS;
+                info.Type = 0;
+                return (nuint)sizeof(BasicInfo);
+            }
+        }
+
+        private static bool OverlapsTrackedRegionLocked(ulong start, ulong size)
+        {
+            var end = start + size;
+            foreach (var region in Regions.Values)
+            {
+                if (region.Base < end && start < region.End)
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        private static bool TryFindRegionLocked(ulong address, out Region region)
+        {
+            region = null!;
+            var keys = Regions.Keys;
+            var low = 0;
+            var high = keys.Count - 1;
+            Region? candidate = null;
+            while (low <= high)
+            {
+                var middle = low + ((high - low) >> 1);
+                var entry = Regions.Values[middle];
+                if (entry.Base <= address)
+                {
+                    candidate = entry;
+                    low = middle + 1;
+                }
+                else
+                {
+                    high = middle - 1;
+                }
+            }
+
+            if (candidate is null || address >= candidate.End)
+            {
+                return false;
+            }
+
+            region = candidate;
+            return true;
+        }
+
+        private static void SetProtectRangeLocked(Region region, ulong start, ulong size, uint protect)
+        {
+            if (start == region.Base && size >= region.Size)
+            {
+                region.DefaultProtect = protect;
+                region.PageProtects = null;
+                return;
+            }
+
+            region.PageProtects ??= new Dictionary<ulong, uint>();
+            var end = start + size;
+            for (var pageAddress = start; pageAddress < end; pageAddress += PageSize)
+            {
+                if (protect == region.DefaultProtect)
+                {
+                    region.PageProtects.Remove(pageAddress);
+                }
+                else
+                {
+                    region.PageProtects[pageAddress] = protect;
+                }
+            }
+        }
+
+        private static int ToPosixProtect(uint win32Protect)
+        {
+            return win32Protect switch
+            {
+                PAGE_NOACCESS => PROT_NONE,
+                PAGE_READONLY => PROT_READ,
+                PAGE_READWRITE => PROT_READ | PROT_WRITE,
+                PAGE_EXECUTE => PROT_READ | PROT_EXEC,
+                PAGE_EXECUTE_READ => PROT_READ | PROT_EXEC,
+                PAGE_EXECUTE_READWRITE => PROT_READ | PROT_WRITE | PROT_EXEC,
+                _ => PROT_READ | PROT_WRITE
+            };
+        }
+
+        private static void Trace(string message)
+        {
+            if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_VMEM"), "1", StringComparison.Ordinal))
+            {
+                Console.Error.WriteLine($"[HOSTMEM] {message}");
+            }
+        }
+
+        private static ulong AlignDown(ulong value, ulong alignment) => value & ~(alignment - 1);
+
+        private static ulong AlignUp(ulong value, ulong alignment) => checked((value + alignment - 1) & ~(alignment - 1));
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern nint mmap(nint addr, nuint length, int prot, int flags, int fd, long offset);
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern int munmap(nint addr, nuint length);
+
+        [DllImport("libc", SetLastError = true)]
+        private static extern int mprotect(nint addr, nuint length, int prot);
+    }
+}