Merge branch 'main' of https://github.com/par274/sharpemu into shader-decoder-part1

[ampr] new exports
[memory] trampoline fixes

.github/workflows/workflow.yml

@@ -9,11 +9,17 @@ on:
       - "**"
     paths-ignore:
       - "**/*.md"
+      - "**/*.png"
+      - "**/*.jpeg"
+      - "**/*.jpg"
   pull_request:
     branches:
       - main
     paths-ignore:
       - "**/*.md"
+      - "**/*.png"
+      - "**/*.jpeg"
+      - "**/*.jpg"
   workflow_dispatch:
 
 permissions:

Directory.Packages.props

@@ -9,5 +9,8 @@ SPDX-License-Identifier: GPL-2.0-or-later
   </PropertyGroup>
   <ItemGroup>
     <PackageVersion Include="Iced" Version="1.21.0" />
+    <PackageVersion Include="Silk.NET.Vulkan" Version="2.23.0" />
+    <PackageVersion Include="Silk.NET.Vulkan.Extensions.KHR" Version="2.23.0" />
+    <PackageVersion Include="Silk.NET.Windowing" Version="2.23.0" />
   </ItemGroup>
 </Project>

src/SharpEmu.CLI/packages.lock.json

@@ -8,6 +8,62 @@
         "resolved": "10.0.3",
         "contentHash": "0B6nZyCHWXnvmlB559oduOspVdNOnpNXPjhpWVMovLPAsDVG7A4jJR9rzECf67JUzxP8/ee/wA8clwIzJcWNFA=="
       },
+      "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.core": {
         "type": "Project",
         "dependencies": {
@@ -23,7 +79,10 @@
       "sharpemu.libs": {
         "type": "Project",
         "dependencies": {
-          "SharpEmu.HLE": "[1.0.0, )"
+          "SharpEmu.HLE": "[1.0.0, )",
+          "Silk.NET.Vulkan": "[2.23.0, )",
+          "Silk.NET.Vulkan.Extensions.KHR": "[2.23.0, )",
+          "Silk.NET.Windowing": "[2.23.0, )"
         }
       },
       "sharpemu.logging": {
@@ -34,10 +93,57 @@
         "requested": "[1.21.0, )",
         "resolved": "1.21.0",
         "contentHash": "dv5+81Q1TBQvVMSOOOmRcjJmvWcX3BZPZsIq31+RLc5cNft0IHAyNlkdb7ZarOWG913PyBoFDsDXoCIlKmLclg=="
+      },
+      "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.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"
+        }
       }
     },
-    "net10.0/linux-x64": {},
-    "net10.0/osx-arm64": {},
-    "net10.0/win-x64": {}
+    "net10.0/linux-x64": {
+      "Ultz.Native.GLFW": {
+        "type": "Transitive",
+        "resolved": "3.4.0",
+        "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
+      }
+    },
+    "net10.0/osx-arm64": {
+      "Ultz.Native.GLFW": {
+        "type": "Transitive",
+        "resolved": "3.4.0",
+        "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
+      }
+    },
+    "net10.0/win-x64": {
+      "Ultz.Native.GLFW": {
+        "type": "Transitive",
+        "resolved": "3.4.0",
+        "contentHash": "Iy22JopynbOJ32vA0lBhFEzGi65GQJBuJHYBYRBpydrDpNoTiHnjIXfA65Gu+8qsOr/ZEoIF8r9aHCgAXuO6DA=="
+      }
+    }
   }
 }

src/SharpEmu.Core/Cpu/CpuDispatcher.cs

@@ -22,6 +22,7 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
     private const ulong StackSize = 0x0020_0000UL;
     private const ulong TlsBaseAddress = 0x7FFE_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;
@@ -342,11 +343,12 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
         for (var i = 0; i < 32; i++)
         {
             var candidateBase = TlsBaseAddress - ((ulong)i * tlsStride);
+            var mappedBase = candidateBase - TlsPrefixSize;
             try
             {
                 _virtualMemory.Map(
-                    candidateBase,
-                    TlsSize,
+                    mappedBase,
+                    TlsSize + TlsPrefixSize,
                     fileOffset: 0,
                     fileData: ReadOnlySpan<byte>.Empty,
                     ProgramHeaderFlags.Read | ProgramHeaderFlags.Write);
@@ -363,9 +365,11 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
 
     private static bool InitializeTls(CpuContext context, ulong tlsBase)
     {
-        return context.TryWriteUInt64(tlsBase + 0x00, 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 + 0x28, 0xC0DEC0DECAFEBABEUL) &&
+               context.TryWriteUInt64(tlsBase + 0x60, tlsBase);
     }
 
     private static bool InitializeGuestFrameChainSentinel(CpuContext context)
@@ -420,6 +424,13 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
             return false;
         }
 
+        var entryStackPointer = entryParamsAddress - sizeof(ulong);
+        if (!context.TryWriteUInt64(entryStackPointer, 0))
+        {
+            return false;
+        }
+
+        context[CpuRegister.Rsp] = entryStackPointer;
         context[CpuRegister.Rdi] = entryParamsAddress;
         context[CpuRegister.Rsi] = programExitHandlerAddress;
         context[CpuRegister.Rdx] = 0;

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

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using System;
+using System.Collections.Concurrent;
 using System.Collections.Generic;
 using System.Linq;
 using System.Runtime.InteropServices;
@@ -12,10 +13,23 @@ namespace SharpEmu.Core.Cpu.Native;
 
 public sealed partial class DirectExecutionBackend
 {
+	private static readonly ConcurrentDictionary<ulong, byte> _knownExecutablePages = new();
 
-	private void RecordRecentImportTrace(string traceLine)
+	private void RecordRecentImportTrace(
+		long dispatchIndex,
+		string nid,
+		ulong returnRip,
+		ulong arg0,
+		ulong arg1,
+		ulong arg2)
 	{
-		_recentImportTrace[_recentImportTraceWriteIndex] = traceLine;
+		_recentImportTrace[_recentImportTraceWriteIndex] = new RecentImportTraceEntry(
+			dispatchIndex,
+			nid,
+			returnRip,
+			arg0,
+			arg1,
+			arg2);
 		_recentImportTraceWriteIndex = (_recentImportTraceWriteIndex + 1) % _recentImportTrace.Length;
 		if (_recentImportTraceCount < _recentImportTrace.Length)
 		{
@@ -34,10 +48,12 @@ public sealed partial class DirectExecutionBackend
 		for (int i = 0; i < _recentImportTraceCount; i++)
 		{
 			int num2 = (num + i) % _recentImportTrace.Length;
-			string text = _recentImportTrace[num2];
-			if (!string.IsNullOrEmpty(text))
+			var entry = _recentImportTrace[num2];
+			if (!string.IsNullOrEmpty(entry.Nid))
 			{
-				Console.Error.WriteLine("[LOADER][INFO]     " + text);
+				Console.Error.WriteLine(
+					$"[LOADER][INFO]     #{entry.DispatchIndex} nid={entry.Nid} ret=0x{entry.ReturnRip:X16} " +
+					$"rdi=0x{entry.Arg0:X16} rsi=0x{entry.Arg1:X16} rdx=0x{entry.Arg2:X16}");
 			}
 		}
 	}
@@ -97,12 +113,13 @@ public sealed partial class DirectExecutionBackend
 
 	private void ProbeReturnRip(ulong returnRip, long dispatchIndex)
 	{
-		if (_cpuContext == null || returnRip == 0)
+		var cpuContext = ActiveCpuContext;
+		if (cpuContext == null || returnRip == 0)
 		{
 			return;
 		}
 		Span<byte> destination = stackalloc byte[128];
-		if (!_cpuContext.Memory.TryRead(returnRip, destination))
+		if (!cpuContext.Memory.TryRead(returnRip, destination))
 		{
 			Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} return-rip probe: unreadable @0x{returnRip:X16}");
 			return;
@@ -113,7 +130,7 @@ public sealed partial class DirectExecutionBackend
 		{
 			int num = BitConverter.ToInt32(destination.Slice(2, 4));
 			ulong num2 = returnRip + 6 + (ulong)num;
-			if (_cpuContext.TryReadUInt64(num2, out var value2))
+			if (cpuContext.TryReadUInt64(num2, out var value2))
 			{
 				Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} return-rip slot: [0x{num2:X16}] = 0x{value2:X16}");
 			}
@@ -122,7 +139,7 @@ public sealed partial class DirectExecutionBackend
 		{
 			int num3 = BitConverter.ToInt32(destination.Slice(3, 4));
 			ulong num4 = returnRip + 7 + (ulong)num3;
-			if (_cpuContext.TryReadUInt64(num4, out var value3))
+			if (cpuContext.TryReadUInt64(num4, out var value3))
 			{
 				Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} return-rip mov-slot: [0x{num4:X16}] = 0x{value3:X16}");
 			}
@@ -134,12 +151,82 @@ public sealed partial class DirectExecutionBackend
 				int num5 = BitConverter.ToInt32(destination.Slice(i + 2, 4));
 				ulong num6 = returnRip + (ulong)i;
 				ulong num7 = num6 + 6 + (ulong)num5;
-				if (_cpuContext.TryReadUInt64(num7, out var value4))
+				if (cpuContext.TryReadUInt64(num7, out var value4))
 				{
 					Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} near-indirect @{num6:X16}: slot=0x{num7:X16} val=0x{value4:X16}");
 				}
 			}
 		}
+		Span<byte> targetBytes = stackalloc byte[32];
+		for (int i = 0; i + 5 <= destination.Length; i++)
+		{
+			if (destination[i] != 0xE8)
+			{
+				continue;
+			}
+
+			int rel32 = BitConverter.ToInt32(destination.Slice(i + 1, 4));
+			ulong callRip = returnRip + (ulong)i;
+			ulong target = unchecked((ulong)((long)(callRip + 5) + rel32));
+			Console.Error.WriteLine($"[LOADER][TRACE] Import#{dispatchIndex} near-call @{callRip:X16}: target=0x{target:X16}");
+			for (int importIndex = 0; importIndex < _importEntries.Length; importIndex++)
+			{
+				if (_importEntries[importIndex].Address != target)
+				{
+					continue;
+				}
+
+				string nid = _importEntries[importIndex].Nid;
+				if (_moduleManager.TryGetExport(nid, out var export))
+				{
+					Console.Error.WriteLine(
+						$"[LOADER][TRACE] Import#{dispatchIndex} near-call import: index={importIndex} {export.LibraryName}:{export.Name} ({nid})");
+				}
+				else
+				{
+					Console.Error.WriteLine(
+						$"[LOADER][TRACE] Import#{dispatchIndex} near-call import: index={importIndex} nid={nid}");
+				}
+				break;
+			}
+
+			if (cpuContext.Memory.TryRead(target, targetBytes))
+			{
+				Console.Error.WriteLine(
+					$"[LOADER][TRACE] Import#{dispatchIndex} near-call target bytes @0x{target:X16}: " +
+					BitConverter.ToString(targetBytes.ToArray()).Replace("-", " "));
+				if (targetBytes[0] == 0xFF && targetBytes[1] == 0x25)
+				{
+					int slotRel32 = BitConverter.ToInt32(targetBytes.Slice(2, 4));
+					ulong slot = unchecked((ulong)((long)(target + 6) + slotRel32));
+					if (cpuContext.TryReadUInt64(slot, out var slotTarget))
+					{
+						Console.Error.WriteLine(
+							$"[LOADER][TRACE] Import#{dispatchIndex} near-call PLT slot: [0x{slot:X16}] = 0x{slotTarget:X16}");
+						for (int importIndex = 0; importIndex < _importEntries.Length; importIndex++)
+						{
+							if (_importEntries[importIndex].Address != slotTarget)
+							{
+								continue;
+							}
+
+							string nid = _importEntries[importIndex].Nid;
+							if (_moduleManager.TryGetExport(nid, out var export))
+							{
+								Console.Error.WriteLine(
+									$"[LOADER][TRACE] Import#{dispatchIndex} near-call PLT import: index={importIndex} {export.LibraryName}:{export.Name} ({nid})");
+							}
+							else
+							{
+								Console.Error.WriteLine(
+									$"[LOADER][TRACE] Import#{dispatchIndex} near-call PLT import: index={importIndex} nid={nid}");
+							}
+							break;
+						}
+					}
+				}
+			}
+		}
 	}
 
 	private static bool IsUnresolvedSentinel(ulong value)
@@ -231,11 +318,24 @@ public sealed partial class DirectExecutionBackend
 
 	private unsafe static bool IsExecutableAddress(ulong address)
 	{
+		var pageAddress = address & ~0xFFFUL;
+		if (_knownExecutablePages.ContainsKey(pageAddress))
+		{
+			return true;
+		}
+
 		if (VirtualQuery((void*)address, out var lpBuffer, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
 		}
-		return lpBuffer.State == 4096 && IsExecutableProtection(lpBuffer.Protect);
+
+		var executable = lpBuffer.State == 4096 && IsExecutableProtection(lpBuffer.Protect);
+		if (executable)
+		{
+			_knownExecutablePages.TryAdd(pageAddress, 0);
+		}
+
+		return executable;
 	}
 
 	private static ulong AlignUp(ulong value, ulong alignment)

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

@@ -7,6 +7,7 @@ using System.Collections.Generic;
 using System.Globalization;
 using System.Reflection;
 using System.Runtime.InteropServices;
+using System.Threading;
 using SharpEmu.Core.Cpu.Disasm;
 using SharpEmu.HLE;
 
@@ -14,11 +15,19 @@ namespace SharpEmu.Core.Cpu.Native;
 
 public sealed partial class DirectExecutionBackend
 {
+	private const ulong LazyCommitWindowBytes = 0x0200_0000UL;
+	private static int _lazyCommitTraceCount;
+
 	private unsafe void SetupExceptionHandler()
 	{
 		if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_DISABLE_RAW_HANDLER"), "1", StringComparison.Ordinal))
 		{
-			_rawExceptionHandler = (nint)AddVectoredExceptionHandler(1u, RawVectoredHandlerPtrManaged);
+			_rawExceptionHandlerStub = CreateExceptionHandlerTrampoline(RawVectoredHandlerPtrManaged);
+			if (_rawExceptionHandlerStub == 0)
+			{
+				throw new InvalidOperationException("Failed to create raw exception handler trampoline");
+			}
+			_rawExceptionHandler = (nint)AddVectoredExceptionHandler(1u, _rawExceptionHandlerStub);
 			Console.Error.WriteLine($"[LOADER][INFO] Raw exception handler installed: 0x{_rawExceptionHandler:X16}");
 		}
 		else
@@ -28,12 +37,22 @@ public sealed partial class DirectExecutionBackend
 
 		_handlerDelegate = VectoredHandler;
 		_handlerHandle = GCHandle.Alloc(_handlerDelegate);
-		_exceptionHandler = (nint)AddVectoredExceptionHandler(1u, Marshal.GetFunctionPointerForDelegate(_handlerDelegate));
+		_exceptionHandlerStub = CreateExceptionHandlerTrampoline(Marshal.GetFunctionPointerForDelegate(_handlerDelegate));
+		if (_exceptionHandlerStub == 0)
+		{
+			throw new InvalidOperationException("Failed to create exception handler trampoline");
+		}
+		_exceptionHandler = (nint)AddVectoredExceptionHandler(1u, _exceptionHandlerStub);
 		Console.Error.WriteLine($"[LOADER][INFO] Exception handler installed: 0x{_exceptionHandler:X16}");
 
 		_unhandledFilterDelegate = UnhandledExceptionFilter;
 		_unhandledFilterHandle = GCHandle.Alloc(_unhandledFilterDelegate);
-		SetUnhandledExceptionFilter(Marshal.GetFunctionPointerForDelegate(_unhandledFilterDelegate));
+		_unhandledFilterStub = CreateExceptionHandlerTrampoline(Marshal.GetFunctionPointerForDelegate(_unhandledFilterDelegate));
+		if (_unhandledFilterStub == 0)
+		{
+			throw new InvalidOperationException("Failed to create unhandled exception filter trampoline");
+		}
+		SetUnhandledExceptionFilter(_unhandledFilterStub);
 	}
 
 	private unsafe int UnhandledExceptionFilter(void* exceptionInfo)
@@ -61,7 +80,7 @@ public sealed partial class DirectExecutionBackend
 	{
 		if (_vectoredHandlerDepth > 0)
 		{
-			Console.Error.WriteLine("[LOADER][TRACE] Nested VEH exception; passing through.");
+			LogNestedVectoredException(exceptionInfo);
 			Console.Error.Flush();
 			return 0;
 		}
@@ -84,10 +103,18 @@ public sealed partial class DirectExecutionBackend
 			ulong rip = ReadCtxU64(contextRecord, 248);
 			ulong rsp = ReadCtxU64(contextRecord, 152);
 
-			if (exceptionCode == 3221225477u && TryHandleLazyCommittedPage(exceptionRecord))
+			if (exceptionCode == 3221225477u && TryHandleLazyCommittedPage(exceptionRecord, rip, rsp))
 			{
 				return -1;
 			}
+			if (IsBenignHostDebugException(exceptionCode))
+			{
+				return -1;
+			}
+			if (exceptionCode == MSVC_CPP_EXCEPTION)
+			{
+				return 0;
+			}
 
 			switch (exceptionCode)
 			{
@@ -278,6 +305,37 @@ public sealed partial class DirectExecutionBackend
 		}
 	}
 
+	private static bool IsBenignHostDebugException(uint exceptionCode)
+	{
+		return exceptionCode is DBG_PRINTEXCEPTION_C or DBG_PRINTEXCEPTION_WIDE_C or MS_VC_THREADNAME_EXCEPTION;
+	}
+
+	private unsafe static void LogNestedVectoredException(void* exceptionInfo)
+	{
+		int count = Interlocked.Increment(ref _nestedVehTraceCount);
+		if (count > 16 && count % 128 != 0)
+		{
+			return;
+		}
+
+		try
+		{
+			EXCEPTION_POINTERS* pointers = (EXCEPTION_POINTERS*)exceptionInfo;
+			EXCEPTION_RECORD* record = pointers->ExceptionRecord;
+			void* contextRecord = pointers->ContextRecord;
+			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(
+				$"[LOADER][TRACE] Nested VEH exception#{count}: code=0x{record->ExceptionCode:X8} ex=0x{(ulong)record->ExceptionAddress:X16} rip=0x{rip:X16} rsp=0x{rsp:X16} type={accessType} target=0x{target:X16}; passing through.");
+		}
+		catch
+		{
+			Console.Error.WriteLine($"[LOADER][TRACE] Nested VEH exception#{count}; passing through.");
+		}
+	}
+
 	private unsafe void LogAccessViolationTrace(ulong exceptionAddress, EXCEPTION_RECORD* exceptionRecord)
 	{
 		ulong accessType = exceptionRecord->NumberParameters >= 1 ? (*exceptionRecord->ExceptionInformation) : 0;
@@ -828,7 +886,7 @@ public sealed partial class DirectExecutionBackend
 		return true;
 	}
 
-	private unsafe static bool TryHandleLazyCommittedPage(EXCEPTION_RECORD* exceptionRecord)
+	private unsafe bool TryHandleLazyCommittedPage(EXCEPTION_RECORD* exceptionRecord, ulong rip, ulong rsp)
 	{
 		if (exceptionRecord->NumberParameters < 2)
 		{
@@ -845,6 +903,10 @@ public sealed partial class DirectExecutionBackend
 		{
 			return false;
 		}
+		if (!IsGuestOwnedLazyCommitAddress(faultAddress, out var owner))
+		{
+			return false;
+		}
 		if (VirtualQuery((void*)faultAddress, out var mbi, (nuint)sizeof(MEMORY_BASIC_INFORMATION64)) == 0)
 		{
 			return false;
@@ -852,7 +914,21 @@ public sealed partial class DirectExecutionBackend
 
 		ulong pageBase = faultAddress & 0xFFFFFFFFFFFFF000uL;
 		uint commitProtect = ResolveLazyCommitProtection(accessType, mbi.AllocationProtect);
-		Console.Error.WriteLine($"[LOADER][TRACE] lazy-query: fault=0x{faultAddress:X16} state=0x{mbi.State:X08} base=0x{mbi.BaseAddress:X16} size=0x{mbi.RegionSize:X16} alloc=0x{mbi.AllocationProtect:X08} prot=0x{mbi.Protect:X08}");
+		int traceIndex = Interlocked.Increment(ref _lazyCommitTraceCount);
+		bool traceLazyCommit = ShouldTraceLazyCommit(traceIndex);
+		if (traceLazyCommit)
+		{
+			Console.Error.WriteLine($"[LOADER][TRACE] lazy-query#{traceIndex}: fault=0x{faultAddress:X16} owner={owner} rip=0x{rip:X16} rsp=0x{rsp:X16} state=0x{mbi.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 == 4096 && IsAccessCompatible(accessType, mbi.Protect))
+		{
+			if (traceLazyCommit)
+			{
+				Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit-race#{traceIndex}: fault=0x{faultAddress:X16} protect=0x{mbi.Protect:X08}");
+			}
+			return true;
+		}
 
 		bool committed = false;
 		ulong committedBase = 0;
@@ -860,14 +936,25 @@ public sealed partial class DirectExecutionBackend
 
 		if (mbi.State == 65536)
 		{
-			ulong largeBase = faultAddress & 0xFFFFFFFFFFE00000uL;
-			if (TryReserveThenCommit(largeBase, 2097152uL, largeBase, 2097152uL, commitProtect))
+			if (TryGetLazyCommitWindow(faultAddress, mbi.BaseAddress, mbi.RegionSize, out var windowBase, out var windowSize) &&
+				TryReserveThenCommit(windowBase, windowSize, windowBase, windowSize, commitProtect))
 			{
 				committed = true;
-				committedBase = largeBase;
-				committedSize = 2097152uL;
+				committedBase = windowBase;
+				committedSize = windowSize;
 			}
 			else
+			{
+				ulong largeBase = faultAddress & 0xFFFFFFFFFFE00000uL;
+				if (TryReserveThenCommit(largeBase, 2097152uL, largeBase, 2097152uL, commitProtect))
+				{
+					committed = true;
+					committedBase = largeBase;
+					committedSize = 2097152uL;
+				}
+			}
+
+			if (!committed)
 			{
 				ulong region64kBase = faultAddress & 0xFFFFFFFFFFFF0000uL;
 				if (TryReserveThenCommit(region64kBase, 65536uL, region64kBase, 65536uL, commitProtect))
@@ -890,7 +977,10 @@ public sealed partial class DirectExecutionBackend
 			}
 
 			TryCommitRange(pageBase + 4096, 4096uL, commitProtect);
-			Console.Error.WriteLine($"[LOADER][TRACE] lazy-reserve-commit: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
+			if (traceLazyCommit)
+			{
+				Console.Error.WriteLine($"[LOADER][TRACE] lazy-reserve-commit#{traceIndex}: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
+			}
 			return true;
 		}
 
@@ -899,14 +989,25 @@ public sealed partial class DirectExecutionBackend
 			return false;
 		}
 
-		ulong largeCommitBase = faultAddress & 0xFFFFFFFFFFE00000uL;
-		if (TryCommitRange(largeCommitBase, 2097152uL, commitProtect))
+		if (TryGetLazyCommitWindow(faultAddress, mbi.BaseAddress, mbi.RegionSize, out var commitWindowBase, out var commitWindowSize) &&
+			TryCommitRange(commitWindowBase, commitWindowSize, commitProtect))
 		{
 			committed = true;
-			committedBase = largeCommitBase;
-			committedSize = 2097152uL;
+			committedBase = commitWindowBase;
+			committedSize = commitWindowSize;
 		}
 		else
+		{
+			ulong largeCommitBase = faultAddress & 0xFFFFFFFFFFE00000uL;
+			if (TryCommitRange(largeCommitBase, 2097152uL, commitProtect))
+			{
+				committed = true;
+				committedBase = largeCommitBase;
+				committedSize = 2097152uL;
+			}
+		}
+
+		if (!committed)
 		{
 			ulong region64kBase = faultAddress & 0xFFFFFFFFFFFF0000uL;
 			if (TryCommitRange(region64kBase, 65536uL, commitProtect))
@@ -935,9 +1036,46 @@ public sealed partial class DirectExecutionBackend
 		}
 
 		TryCommitRange(pageBase + 4096, 4096uL, commitProtect);
-		Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
+		if (traceLazyCommit)
+		{
+			Console.Error.WriteLine($"[LOADER][TRACE] lazy-commit#{traceIndex}: addr=0x{committedBase:X16} size=0x{committedSize:X16} access={accessType} protect=0x{commitProtect:X8}");
+		}
 		return true;
 
+		static bool TryGetLazyCommitWindow(ulong fault, ulong regionBase, ulong regionSize, out ulong baseAddress, out ulong length)
+		{
+			baseAddress = 0;
+			length = 0;
+			if (regionSize == 0 || ulong.MaxValue - regionBase < regionSize)
+			{
+				return false;
+			}
+
+			ulong regionEnd = regionBase + regionSize;
+			ulong windowBase = fault & ~(LazyCommitWindowBytes - 1);
+			if (windowBase < regionBase)
+			{
+				windowBase = regionBase;
+			}
+
+			if (windowBase >= regionEnd)
+			{
+				return false;
+			}
+
+			ulong windowEnd = Math.Min(regionEnd, windowBase + LazyCommitWindowBytes);
+			ulong windowSize = windowEnd - windowBase;
+			windowSize &= 0xFFFFFFFFFFFFF000uL;
+			if (windowSize == 0)
+			{
+				return false;
+			}
+
+			baseAddress = windowBase;
+			length = windowSize;
+			return true;
+		}
+
 		static unsafe bool TryCommitRange(ulong baseAddress, ulong length, uint protection)
 		{
 			if (length == 0)
@@ -964,6 +1102,49 @@ public sealed partial class DirectExecutionBackend
 			}
 			return TryCommitRange(commitAddress, commitSize, protection);
 		}
+
+		static bool IsAccessCompatible(ulong accessType, uint protection)
+		{
+			const uint pageNoAccess = 0x01;
+			const uint pageReadOnly = 0x02;
+			const uint pageReadWrite = 0x04;
+			const uint pageWriteCopy = 0x08;
+			const uint pageExecute = 0x10;
+			const uint pageExecuteRead = 0x20;
+			const uint pageExecuteReadWrite = 0x40;
+			const uint pageExecuteWriteCopy = 0x80;
+			const uint pageGuard = 0x100;
+			const uint accessMask = 0xFF;
+
+			if ((protection & pageGuard) != 0)
+			{
+				return false;
+			}
+
+			uint access = protection & accessMask;
+			if (access == pageNoAccess)
+			{
+				return false;
+			}
+
+			return accessType switch
+			{
+				0 => access is pageReadOnly or pageReadWrite or pageWriteCopy or pageExecuteRead or pageExecuteReadWrite or pageExecuteWriteCopy,
+				1 => access is pageReadWrite or pageWriteCopy or pageExecuteReadWrite or pageExecuteWriteCopy,
+				8 => access is pageExecute or pageExecuteRead or pageExecuteReadWrite or pageExecuteWriteCopy,
+				_ => false
+			};
+		}
+	}
+
+	private static bool ShouldTraceLazyCommit(int traceIndex)
+	{
+		if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_LAZY_COMMIT"), "1", StringComparison.Ordinal))
+		{
+			return true;
+		}
+
+		return traceIndex <= 16 || traceIndex % 256 == 0;
 	}
 
 	private static uint ResolveLazyCommitProtection(ulong accessType, uint allocationProtect)

src/SharpEmu.Core/Cpu/TrackedCpuMemory.cs

@@ -5,7 +5,7 @@ using SharpEmu.HLE;
 
 namespace SharpEmu.Core.Cpu;
 
-public sealed class TrackedCpuMemory : ICpuMemory, ITrackedCpuMemory
+public sealed class TrackedCpuMemory : ICpuMemory, ITrackedCpuMemory, IGuestMemoryAllocator
 {
     private readonly ICpuMemory _inner;
 
@@ -39,4 +39,15 @@ public sealed class TrackedCpuMemory : ICpuMemory, ITrackedCpuMemory
 
         return result;
     }
+
+    public bool TryAllocateGuestMemory(ulong size, ulong alignment, out ulong address)
+    {
+        if (_inner is IGuestMemoryAllocator allocator)
+        {
+            return allocator.TryAllocateGuestMemory(size, alignment, out address);
+        }
+
+        address = 0;
+        return false;
+    }
 }

src/SharpEmu.Core/Loader/SelfImage.cs

@@ -21,7 +21,10 @@ public sealed class SelfImage
         IReadOnlyList<ulong>? initializerFunctions = null,
         ulong initFunctionEntryPoint = 0,
         ulong imageBase = 0,
-        ulong procParamAddress = 0)
+        ulong procParamAddress = 0,
+        string? title = null,
+        string? titleId = null,
+        string? version = null)
     {
         ArgumentNullException.ThrowIfNull(programHeaders);
         ArgumentNullException.ThrowIfNull(mappedRegions);
@@ -38,6 +41,9 @@ public sealed class SelfImage
         InitFunctionEntryPoint = initFunctionEntryPoint;
         _imageBase = imageBase;
         ProcParamAddress = procParamAddress;
+        Title = title;
+        TitleId = titleId;
+        Version = version;
     }
 
     public bool IsSelf { get; }
@@ -63,4 +69,10 @@ public sealed class SelfImage
     public ulong EntryPoint => ElfHeader.EntryPoint + _imageBase;
 
     public ulong ProcParamAddress { get; }
+
+    public string? Title { get; }
+
+    public string? TitleId { get; }
+
+    public string? Version { get; }
 }

src/SharpEmu.Core/Loader/SelfLoader.cs

@@ -144,10 +144,9 @@ public sealed class SelfLoader : ISelfLoader
             throw new InvalidDataException("Input image is empty.");
         }
 
-        if (readParamJson)
-        {
-            TryLoadParamJson(fs, mountRoot);
-        }
+        var applicationInfo = readParamJson
+            ? TryLoadParamJson(fs, mountRoot)
+            : default;
 
         if (clearVirtualMemory)
         {
@@ -270,15 +269,20 @@ public sealed class SelfLoader : ISelfLoader
             initializerFunctions,
             initFunctionEntryPoint,
             imageBase,
-            procParamAddress);
+            procParamAddress,
+            applicationInfo.Title,
+            applicationInfo.TitleId,
+            applicationInfo.Version);
     }
 
-    private static void TryLoadParamJson(IFileSystem? fs, string? mountRoot)
+    private static (string? Title, string? TitleId, string? Version) TryLoadParamJson(
+        IFileSystem? fs,
+        string? mountRoot)
     {
         if (fs == null)
         {
             Console.WriteLine("[LOADER] param.json not found (no filesystem provided).");
-            return;
+            return default;
         }
 
         string[] possiblePaths = string.IsNullOrEmpty(mountRoot)
@@ -298,12 +302,16 @@ public sealed class SelfLoader : ISelfLoader
         if (foundPath == null)
         {
             Console.WriteLine("[LOADER] param.json not found (no root path / unknown layout).");
-            return;
+            return default;
         }
 
-        var (title, titleId, ver) = Ps5ParamJsonReader.TryReadPs5Param(fs, foundPath);
+        var applicationInfo = Ps5ParamJsonReader.TryReadPs5Param(fs, foundPath);
         Console.WriteLine($"[LOADER] Loading param.json at {foundPath}");
-        Console.WriteLine($"[LOADER] Title: {title ?? "(unknown)"}, TitleId: {titleId ?? "(unknown)"}, Version: {ver ?? "(unknown)"}");
+        Console.WriteLine(
+            $"[LOADER] Title: {applicationInfo.Title ?? "(unknown)"}, " +
+            $"TitleId: {applicationInfo.TitleId ?? "(unknown)"}, " +
+            $"Version: {applicationInfo.Version ?? "(unknown)"}");
+        return applicationInfo;
     }
 
     private static LoadContext ParseLayout(ReadOnlySpan<byte> imageData)

src/SharpEmu.Core/Memory/PhysicalVirtualMemory.cs

@@ -7,13 +7,17 @@ using SharpEmu.HLE;
 
 namespace SharpEmu.Core.Memory;
 
-public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
+public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IGuestMemoryAllocator, IDisposable
 {
-    private readonly object _gate = new();
+    private readonly ReaderWriterLockSlim _gate = new(LockRecursionPolicy.SupportsRecursion);
+    private readonly object _guestAllocationGate = new();
     private readonly List<MemoryRegion> _regions = new();
     private readonly Dictionary<ulong, ProgramHeaderFlags> _pageProtections = new();
     private bool _disposed;
     private const ulong PageSize = 0x1000;
+    private const ulong GuestAllocationArenaAddress = 0x00006000_0000_0000;
+    private const ulong GuestAllocationArenaSize = 0x0100_0000;
+    private const ulong GuestAllocationArenaStartOffset = PageSize;
     private const ulong LargeDataReserveThreshold = 0x4000_0000UL; // 1 GiB
     private const ulong LazyReservePrimeBytes = 0x5000_0000UL; // 1.25 GiB
     private const ulong LazyReservePrimeChunkBytes = 0x0200_0000UL; // 32 MiB
@@ -29,6 +33,9 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
     private const uint PAGE_READWRITE = 0x04;
     private const uint PAGE_READONLY = 0x02;
 
+    private ulong _guestAllocationArenaBase;
+    private ulong _guestAllocationOffset;
+
     [DllImport("kernel32.dll", SetLastError = true)]
     private static extern void* VirtualAlloc(void* lpAddress, nuint dwSize, uint flAllocationType, uint flProtect);
 
@@ -71,7 +78,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
             return false;
         }
 
-        lock (_gate)
+        _gate.EnterWriteLock();
+        try
         {
             _regions.Add(new MemoryRegion
             {
@@ -82,6 +90,10 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
                 Protection = protection
             });
         }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
 
         var allocationKind = executable ? "executable memory" : "data memory";
         Console.Error.WriteLine($"[VMEM] Allocated exact {allocationKind}: 0x{actualAddress:X16} - 0x{actualAddress + alignedSize:X16} ({alignedSize} bytes)");
@@ -195,7 +207,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
             }
         }
 
-        lock (_gate)
+        _gate.EnterWriteLock();
+        try
         {
             _regions.Add(new MemoryRegion
             {
@@ -206,6 +219,10 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
                 Protection = protection
             });
         }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
 
         var allocationKind = reservedOnly
             ? "reserved data memory (lazy commit)"
@@ -215,16 +232,116 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
         return actualAddress;
     }
 
+    public bool TryAllocateAtOrAbove(
+        ulong desiredAddress,
+        ulong size,
+        bool executable,
+        ulong alignment,
+        out ulong actualAddress)
+    {
+        actualAddress = 0;
+        if (size == 0)
+        {
+            return false;
+        }
+
+        var alignedSize = AlignUp(size, PageSize);
+        var effectiveAlignment = Math.Max(PageSize, alignment == 0 ? PageSize : alignment);
+        var cursor = AlignUp(desiredAddress, effectiveAlignment);
+
+        for (var attempt = 0; attempt < 0x10000; attempt++)
+        {
+            if (cursor == 0 || ulong.MaxValue - cursor < alignedSize)
+            {
+                return false;
+            }
+
+            if (TryGetOverlappingRegionEnd(cursor, alignedSize, out var overlapEnd))
+            {
+                cursor = AlignUp(overlapEnd, effectiveAlignment);
+                continue;
+            }
+
+            try
+            {
+                actualAddress = AllocateAt(cursor, alignedSize, executable, allowAlternative: false);
+                if (actualAddress == cursor)
+                {
+                    return true;
+                }
+
+                actualAddress = 0;
+            }
+            catch
+            {
+            }
+
+            cursor = AlignUp(cursor + effectiveAlignment, effectiveAlignment);
+        }
+
+        return false;
+    }
+
+    public bool TryAllocateGuestMemory(ulong size, ulong alignment, out ulong address)
+    {
+        address = 0;
+        if (size == 0 || alignment == 0 || (alignment & (alignment - 1)) != 0)
+        {
+            return false;
+        }
+
+        lock (_guestAllocationGate)
+        {
+            if (_guestAllocationArenaBase == 0)
+            {
+                try
+                {
+                    _guestAllocationArenaBase = AllocateAt(
+                        GuestAllocationArenaAddress,
+                        GuestAllocationArenaSize,
+                        executable: false,
+                        allowAlternative: true);
+                    _guestAllocationOffset = GuestAllocationArenaStartOffset;
+                }
+                catch (Exception)
+                {
+                    return false;
+                }
+            }
+
+            var alignedOffset = AlignUp(_guestAllocationOffset, alignment);
+            if (alignedOffset > GuestAllocationArenaSize || size > GuestAllocationArenaSize - alignedOffset)
+            {
+                return false;
+            }
+
+            address = _guestAllocationArenaBase + alignedOffset;
+            _guestAllocationOffset = alignedOffset + size;
+            return true;
+        }
+    }
+
     public void Clear()
     {
-        lock (_gate)
+        lock (_guestAllocationGate)
         {
-            foreach (var region in _regions)
+            _gate.EnterWriteLock();
+            try
             {
-                VirtualFree((void*)region.VirtualAddress, 0, MEM_RELEASE);
+                foreach (var region in _regions)
+                {
+                    VirtualFree((void*)region.VirtualAddress, 0, MEM_RELEASE);
+                }
+                _regions.Clear();
+                _pageProtections.Clear();
             }
-            _regions.Clear();
-            _pageProtections.Clear();
+            finally
+            {
+                _gate.ExitWriteLock();
+            }
+
+            _guestAllocationArenaBase = 0;
+            _guestAllocationOffset = 0;
         }
     }
 
@@ -241,7 +358,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
         var mapEnd = AlignUp(segmentEnd, PageSize);
         var mapSize = checked(mapEnd - mapStart);
 
-        lock (_gate)
+        _gate.EnterWriteLock();
+        try
         {
             var existingRegion = FindRegion(mapStart, mapSize);
             if (existingRegion == null)
@@ -274,6 +392,10 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
 
             Console.Error.WriteLine($"[VMEM] Mapped segment: 0x{virtualAddress:X16} - 0x{virtualAddress + memorySize:X16} (file: {fileData.Length} bytes, prot: {protection})");
         }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
     }
 
     private void ApplySegmentProtection(ulong mapStart, ulong mapEnd, ProgramHeaderFlags flags)
@@ -323,7 +445,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
 
     public IReadOnlyList<VirtualMemoryRegion> SnapshotRegions()
     {
-        lock (_gate)
+        _gate.EnterReadLock();
+        try
         {
             var snapshot = new VirtualMemoryRegion[_regions.Count];
             for (var i = 0; i < _regions.Count; i++)
@@ -338,11 +461,17 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
             }
             return snapshot;
         }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
     }
 
     public bool TryRead(ulong virtualAddress, Span<byte> destination)
     {
-        lock (_gate)
+        var requiresExclusiveAccess = false;
+        _gate.EnterReadLock();
+        try
         {
             foreach (var region in _regions)
             {
@@ -354,38 +483,55 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
                         return true;
                     }
 
-                    if (!EnsureRangeCommitted((ulong)srcPtr, (ulong)destination.Length, region))
+                    if (region.IsReservedOnly)
                     {
-                        return false;
-                    }
-
-                    if (!TryTemporarilyProtectForRead((ulong)srcPtr, (ulong)destination.Length, region, out var touchedPages))
-                    {
-                        return false;
-                    }
-
-                    try
-                    {
-                        fixed (byte* destPtr = destination)
+                        if (!EnsureRangeCommitted((ulong)srcPtr, (ulong)destination.Length, region))
                         {
-                            Buffer.MemoryCopy(srcPtr, destPtr, (nuint)destination.Length, (nuint)destination.Length);
+                            return false;
                         }
                     }
-                    finally
+
+                    if (!CanReadWithoutProtectionChange((ulong)srcPtr, (ulong)destination.Length, region))
                     {
-                        RestorePageProtections(touchedPages);
+                        requiresExclusiveAccess = true;
+                        break;
+                    }
+
+                    fixed (byte* destPtr = destination)
+                    {
+                        Buffer.MemoryCopy(srcPtr, destPtr, (nuint)destination.Length, (nuint)destination.Length);
                     }
 
                     return true;
                 }
             }
+        }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
+
+        if (!requiresExclusiveAccess)
+        {
             return false;
         }
+
+        _gate.EnterWriteLock();
+        try
+        {
+            return TryReadExclusive(virtualAddress, destination);
+        }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
     }
 
     public bool TryWrite(ulong virtualAddress, ReadOnlySpan<byte> source)
     {
-        lock (_gate)
+        var requiresExclusiveAccess = false;
+        _gate.EnterReadLock();
+        try
         {
             foreach (var region in _regions)
             {
@@ -397,37 +543,148 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
                         return true;
                     }
 
-                    if (!EnsureRangeCommitted((ulong)destPtr, (ulong)source.Length, region))
+                    if (region.IsReservedOnly)
                     {
-                        return false;
-                    }
-
-                    if (!VirtualProtect(destPtr, (nuint)source.Length, PAGE_EXECUTE_READWRITE, out var oldProtect))
-                    {
-                        return false;
-                    }
-
-                    try
-                    {
-                        fixed (byte* srcPtr = source)
+                        if (!EnsureRangeCommitted((ulong)destPtr, (ulong)source.Length, region))
                         {
-                            Buffer.MemoryCopy(srcPtr, destPtr, (nuint)source.Length, (nuint)source.Length);
+                            return false;
                         }
                     }
-                    finally
+
+                    if (!CanWriteWithoutProtectionChange((ulong)destPtr, (ulong)source.Length, region))
                     {
-                        VirtualProtect(destPtr, (nuint)source.Length, oldProtect, out _);
-                        if (IsExecutableProtection(oldProtect))
-                        {
-                            FlushInstructionCache(null, destPtr, (nuint)source.Length);
-                        }
+                        requiresExclusiveAccess = true;
+                        break;
+                    }
+
+                    fixed (byte* srcPtr = source)
+                    {
+                        Buffer.MemoryCopy(srcPtr, destPtr, (nuint)source.Length, (nuint)source.Length);
                     }
 
                     return true;
                 }
             }
+        }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
+
+        if (!requiresExclusiveAccess)
+        {
             return false;
         }
+
+        _gate.EnterWriteLock();
+        try
+        {
+            return TryWriteExclusive(virtualAddress, source);
+        }
+        finally
+        {
+            _gate.ExitWriteLock();
+        }
+    }
+
+    private bool TryReadExclusive(ulong virtualAddress, Span<byte> destination)
+    {
+        foreach (var region in _regions)
+        {
+            if (!TryResolveRegionOffset(virtualAddress, (ulong)destination.Length, region, out var offset))
+            {
+                continue;
+            }
+
+            var srcPtr = (void*)(region.VirtualAddress + offset);
+            if (!EnsureRangeCommitted((ulong)srcPtr, (ulong)destination.Length, region))
+            {
+                return false;
+            }
+
+            if (CanReadWithoutProtectionChange((ulong)srcPtr, (ulong)destination.Length, region))
+            {
+                fixed (byte* destPtr = destination)
+                {
+                    Buffer.MemoryCopy(srcPtr, destPtr, (nuint)destination.Length, (nuint)destination.Length);
+                }
+
+                return true;
+            }
+
+            if (!TryTemporarilyProtectForRead((ulong)srcPtr, (ulong)destination.Length, region, out var touchedPages))
+            {
+                return false;
+            }
+
+            try
+            {
+                fixed (byte* destPtr = destination)
+                {
+                    Buffer.MemoryCopy(srcPtr, destPtr, (nuint)destination.Length, (nuint)destination.Length);
+                }
+            }
+            finally
+            {
+                RestorePageProtections(touchedPages);
+            }
+
+            return true;
+        }
+
+        return false;
+    }
+
+    private bool TryWriteExclusive(ulong virtualAddress, ReadOnlySpan<byte> source)
+    {
+        foreach (var region in _regions)
+        {
+            if (!TryResolveRegionOffset(virtualAddress, (ulong)source.Length, region, out var offset))
+            {
+                continue;
+            }
+
+            var destPtr = (void*)(region.VirtualAddress + offset);
+            if (!EnsureRangeCommitted((ulong)destPtr, (ulong)source.Length, region))
+            {
+                return false;
+            }
+
+            if (CanWriteWithoutProtectionChange((ulong)destPtr, (ulong)source.Length, region))
+            {
+                fixed (byte* srcPtr = source)
+                {
+                    Buffer.MemoryCopy(srcPtr, destPtr, (nuint)source.Length, (nuint)source.Length);
+                }
+
+                return true;
+            }
+
+            if (!VirtualProtect(destPtr, (nuint)source.Length, PAGE_EXECUTE_READWRITE, out var oldProtect))
+            {
+                return false;
+            }
+
+            try
+            {
+                fixed (byte* srcPtr = source)
+                {
+                    Buffer.MemoryCopy(srcPtr, destPtr, (nuint)source.Length, (nuint)source.Length);
+                }
+            }
+            finally
+            {
+                VirtualProtect(destPtr, (nuint)source.Length, oldProtect, out _);
+                if (IsExecutableProtection(oldProtect))
+                {
+                    FlushInstructionCache(null, destPtr, (nuint)source.Length);
+                }
+            }
+
+            return true;
+        }
+
+        return false;
     }
 
     public bool TryWriteUInt64(ulong virtualAddress, ulong value)
@@ -439,7 +696,8 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
 
     public void* GetPointer(ulong virtualAddress)
     {
-        lock (_gate)
+        _gate.EnterReadLock();
+        try
         {
             foreach (var region in _regions)
             {
@@ -451,11 +709,16 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
             }
             return null;
         }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
     }
 
     public bool IsAccessible(ulong virtualAddress, ulong size)
     {
-        lock (_gate)
+        _gate.EnterReadLock();
+        try
         {
             foreach (var region in _regions)
             {
@@ -466,6 +729,10 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
             }
             return false;
         }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
     }
 
     private MemoryRegion? FindRegion(ulong address, ulong size)
@@ -480,6 +747,35 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
         return null;
     }
 
+    private bool TryGetOverlappingRegionEnd(ulong address, ulong size, out ulong overlapEnd)
+    {
+        overlapEnd = 0;
+        if (size == 0 || ulong.MaxValue - address < size - 1)
+        {
+            return false;
+        }
+
+        var end = address + size;
+        _gate.EnterReadLock();
+        try
+        {
+            foreach (var region in _regions)
+            {
+                var regionEnd = region.VirtualAddress + region.Size;
+                if (address < regionEnd && region.VirtualAddress < end)
+                {
+                    overlapEnd = Math.Max(overlapEnd, regionEnd);
+                }
+            }
+        }
+        finally
+        {
+            _gate.ExitReadLock();
+        }
+
+        return overlapEnd != 0;
+    }
+
     private static bool TryResolveRegionOffset(ulong address, ulong size, MemoryRegion region, out ulong offset)
     {
         offset = 0;
@@ -507,6 +803,44 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
         return protection is PAGE_EXECUTE or PAGE_EXECUTE_READ or PAGE_EXECUTE_READWRITE or PAGE_EXECUTE_WRITECOPY;
     }
 
+    private bool CanReadWithoutProtectionChange(ulong address, ulong size, MemoryRegion region) =>
+        CanAccessWithoutProtectionChange(address, size, region, write: false);
+
+    private bool CanWriteWithoutProtectionChange(ulong address, ulong size, MemoryRegion region) =>
+        CanAccessWithoutProtectionChange(address, size, region, write: true);
+
+    private bool CanAccessWithoutProtectionChange(ulong address, ulong size, MemoryRegion region, bool write)
+    {
+        var startPage = AlignDown(address, PageSize);
+        var endPage = AlignUp(address + size, PageSize);
+        for (var pageAddress = startPage; pageAddress < endPage; pageAddress += PageSize)
+        {
+            if (_pageProtections.TryGetValue(pageAddress, out var flags))
+            {
+                if (write ? (flags & ProgramHeaderFlags.Write) == 0 : (flags & ProgramHeaderFlags.Read) == 0)
+                {
+                    return false;
+                }
+            }
+            else if (write ? !IsWritableProtection(region.Protection) : !IsReadableProtection(region.Protection))
+            {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
+    private static bool IsReadableProtection(uint protection)
+    {
+        return protection is PAGE_READONLY or PAGE_READWRITE or PAGE_EXECUTE_READ or PAGE_EXECUTE_READWRITE;
+    }
+
+    private static bool IsWritableProtection(uint protection)
+    {
+        return protection is PAGE_READWRITE or PAGE_EXECUTE_READWRITE;
+    }
+
     private static uint GetCommitProtection(MemoryRegion region)
     {
         return region.IsExecutable ? PAGE_EXECUTE_READWRITE : PAGE_READWRITE;
@@ -523,15 +857,26 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
         var endPage = AlignUp(address + size, PageSize);
         var commitProtection = GetCommitProtection(region);
 
-        for (var pageAddress = startPage; pageAddress < endPage; pageAddress += PageSize)
+        var pageAddress = startPage;
+        while (pageAddress < endPage)
         {
             if (VirtualQuery((void*)pageAddress, out var info, (nuint)sizeof(MemoryBasicInformation64)) == 0)
             {
                 return false;
             }
 
+            var queriedEnd = info.RegionSize > ulong.MaxValue - info.BaseAddress
+                ? ulong.MaxValue
+                : info.BaseAddress + info.RegionSize;
+            var rangeEnd = Math.Min(endPage, queriedEnd);
+            if (rangeEnd <= pageAddress)
+            {
+                return false;
+            }
+
             if (info.State == MEM_COMMIT)
             {
+                pageAddress = rangeEnd;
                 continue;
             }
 
@@ -540,10 +885,13 @@ public sealed unsafe class PhysicalVirtualMemory : IVirtualMemory, IDisposable
                 return false;
             }
 
-            if (VirtualAlloc((void*)pageAddress, (nuint)PageSize, MEM_COMMIT, commitProtection) == null)
+            var commitSize = rangeEnd - pageAddress;
+            if (VirtualAlloc((void*)pageAddress, (nuint)commitSize, MEM_COMMIT, commitProtection) == null)
             {
                 return false;
             }
+
+            pageAddress = rangeEnd;
         }
 
         return true;

src/SharpEmu.Core/Runtime/SharpEmuRuntime.cs

@@ -10,6 +10,7 @@ using SharpEmu.HLE;
 using SharpEmu.Libs.VideoOut;
 using SharpEmu.Libs.Kernel;
 using SharpEmu.Libs.AppContent;
+using SharpEmu.Libs.SaveData;
 using System.Buffers.Binary;
 using System.Collections.Generic;
 using System.Linq;
@@ -135,6 +136,8 @@ public sealed class SharpEmuRuntime : ISharpEmuRuntime
         LastMilestoneLog = null;
         KernelModuleRegistry.Reset();
         var image = LoadImage(normalizedEbootPath);
+        VideoOutExports.ConfigureApplicationInfo(image.Title, image.TitleId, image.Version);
+        SaveDataExports.ConfigureApplicationInfo(image.TitleId);
         RegisterLoadedModule(normalizedEbootPath, image, isMain: true, isSystemModule: false);
         KernelRuntimeCompatExports.ConfigureProcessProcParamAddress(image.ProcParamAddress);
         Console.Error.WriteLine($"[RUNTIME] Entry: 0x{image.EntryPoint:X16}");

src/SharpEmu.Core/packages.lock.json

@@ -8,17 +8,105 @@
         "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"
       },
       "sharpemu.libs": {
         "type": "Project",
         "dependencies": {
-          "SharpEmu.HLE": "[1.0.0, )"
+          "SharpEmu.HLE": "[1.0.0, )",
+          "Silk.NET.Vulkan": "[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.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.HLE/GuestThreadExecution.cs

@@ -1,6 +1,8 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+using System.Diagnostics;
+
 namespace SharpEmu.HLE;
 
 public readonly record struct GuestThreadStartRequest(
@@ -8,46 +10,165 @@ public readonly record struct GuestThreadStartRequest(
     ulong EntryPoint,
     ulong Argument,
     ulong AttributeAddress,
-    string Name);
+    string Name,
+    int Priority,
+    ulong AffinityMask);
+
+public readonly record struct GuestThreadSnapshot(
+    ulong ThreadHandle,
+    string Name,
+    string State,
+    long ImportCount,
+    string? LastImportNid,
+    ulong LastReturnRip,
+    string? BlockReason);
 
 public interface IGuestThreadScheduler
 {
+    bool SupportsGuestContextTransfer { get; }
+
     bool TryStartThread(CpuContext creatorContext, GuestThreadStartRequest request, out string? error);
 
+    bool TryJoinThread(
+        CpuContext callerContext,
+        ulong threadHandle,
+        out ulong returnValue,
+        out string? error);
+
     void Pump(CpuContext callerContext, string reason);
+
+    int WakeBlockedThreads(string wakeKey, int maxCount = int.MaxValue);
+
+    IReadOnlyList<GuestThreadSnapshot> SnapshotThreads();
+
+    bool TryCallGuestFunction(
+        CpuContext callerContext,
+        ulong entryPoint,
+        ulong arg0,
+        ulong arg1,
+        ulong stackAddress,
+        ulong stackSize,
+        string reason,
+        out string? error);
+
+    bool TryCallGuestContinuation(
+        CpuContext callerContext,
+        GuestCpuContinuation continuation,
+        string reason,
+        out string? error);
 }
 
+public readonly record struct GuestImportCallFrame(
+    bool IsValid,
+    ulong ReturnRip,
+    ulong ResumeRsp,
+    ulong ReturnSlotAddress);
+
+public readonly record struct GuestCpuContinuation(
+    ulong Rip,
+    ulong Rsp,
+    ulong ReturnSlotAddress,
+    ulong Rflags,
+    ulong FsBase,
+    ulong GsBase,
+    ulong Rax,
+    ulong Rcx,
+    ulong Rdx,
+    ulong Rbx,
+    ulong Rbp,
+    ulong Rsi,
+    ulong Rdi,
+    ulong R8,
+    ulong R9,
+    ulong R12,
+    ulong R13,
+    ulong R14,
+    ulong R15);
+
 public static class GuestThreadExecution
 {
     [ThreadStatic]
     private static ulong _currentGuestThreadHandle;
 
+    [ThreadStatic]
+    private static ulong _currentFiberAddress;
+
     [ThreadStatic]
     private static string? _pendingBlockReason;
 
+    [ThreadStatic]
+    private static bool _pendingBlockContinuationValid;
+
+    [ThreadStatic]
+    private static GuestCpuContinuation _pendingBlockContinuation;
+
+    [ThreadStatic]
+    private static string? _pendingBlockWakeKey;
+
+    [ThreadStatic]
+    private static Func<int>? _pendingBlockResumeHandler;
+
+    [ThreadStatic]
+    private static Func<bool>? _pendingBlockWakeHandler;
+
+    [ThreadStatic]
+    private static long _pendingBlockDeadlineTimestamp;
+
     [ThreadStatic]
     private static bool _pendingEntryExit;
 
     [ThreadStatic]
-    private static int _pendingEntryExitStatus;
+    private static ulong _pendingEntryExitValue;
 
     [ThreadStatic]
     private static string? _pendingEntryExitReason;
 
+    [ThreadStatic]
+    private static bool _pendingContextTransfer;
+
+    [ThreadStatic]
+    private static GuestCpuContinuation _pendingContextTransferTarget;
+
+    [ThreadStatic]
+    private static bool _hasCurrentImportCallFrame;
+
+    [ThreadStatic]
+    private static ulong _currentImportReturnRip;
+
+    [ThreadStatic]
+    private static ulong _currentImportResumeRsp;
+
+    [ThreadStatic]
+    private static ulong _currentImportReturnSlotAddress;
+
     public static IGuestThreadScheduler? Scheduler { get; set; }
 
     public static bool IsGuestThread => _currentGuestThreadHandle != 0;
 
     public static ulong CurrentGuestThreadHandle => _currentGuestThreadHandle;
 
+    public static ulong CurrentFiberAddress => _currentFiberAddress;
+
     public static ulong EnterGuestThread(ulong threadHandle)
     {
         var previous = _currentGuestThreadHandle;
         _currentGuestThreadHandle = threadHandle;
         _pendingBlockReason = null;
+        _pendingBlockContinuationValid = false;
+        _pendingBlockContinuation = default;
+        _pendingBlockWakeKey = null;
+        _pendingBlockResumeHandler = null;
+        _pendingBlockWakeHandler = null;
+        _pendingBlockDeadlineTimestamp = 0;
         _pendingEntryExit = false;
-        _pendingEntryExitStatus = 0;
+        _pendingEntryExitValue = 0;
         _pendingEntryExitReason = null;
+        _pendingContextTransfer = false;
+        _pendingContextTransferTarget = default;
+        _hasCurrentImportCallFrame = false;
+        _currentImportReturnRip = 0;
+        _currentImportResumeRsp = 0;
+        _currentImportReturnSlotAddress = 0;
         return previous;
     }
 
@@ -55,12 +176,44 @@ public static class GuestThreadExecution
     {
         _currentGuestThreadHandle = previousThreadHandle;
         _pendingBlockReason = null;
+        _pendingBlockContinuationValid = false;
+        _pendingBlockContinuation = default;
+        _pendingBlockWakeKey = null;
+        _pendingBlockResumeHandler = null;
+        _pendingBlockWakeHandler = null;
+        _pendingBlockDeadlineTimestamp = 0;
         _pendingEntryExit = false;
-        _pendingEntryExitStatus = 0;
+        _pendingEntryExitValue = 0;
         _pendingEntryExitReason = null;
+        _pendingContextTransfer = false;
+        _pendingContextTransferTarget = default;
+        _hasCurrentImportCallFrame = false;
+        _currentImportReturnRip = 0;
+        _currentImportResumeRsp = 0;
+        _currentImportReturnSlotAddress = 0;
     }
 
-    public static bool RequestCurrentThreadBlock(string reason)
+    public static ulong EnterFiber(ulong fiberAddress)
+    {
+        var previous = _currentFiberAddress;
+        _currentFiberAddress = fiberAddress;
+        return previous;
+    }
+
+    public static void RestoreFiber(ulong previousFiberAddress)
+    {
+        _currentFiberAddress = previousFiberAddress;
+    }
+
+    public static bool RequestCurrentThreadBlock(string reason) => RequestCurrentThreadBlock(null, reason);
+
+    public static bool RequestCurrentThreadBlock(
+        CpuContext? context,
+        string reason,
+        string? wakeKey = null,
+        Func<int>? resumeHandler = null,
+        Func<bool>? wakeHandler = null,
+        long blockDeadlineTimestamp = 0)
     {
         if (!IsGuestThread)
         {
@@ -68,31 +221,167 @@ public static class GuestThreadExecution
         }
 
         _pendingBlockReason = string.IsNullOrWhiteSpace(reason) ? "guest_thread_blocked" : reason;
+        _pendingBlockWakeKey = string.IsNullOrWhiteSpace(wakeKey) ? _pendingBlockReason : wakeKey;
+        _pendingBlockResumeHandler = resumeHandler;
+        _pendingBlockWakeHandler = wakeHandler;
+        _pendingBlockDeadlineTimestamp = blockDeadlineTimestamp;
+        if (context is not null && TryCaptureCurrentBlockContinuation(context, out var continuation))
+        {
+            _pendingBlockContinuation = continuation;
+            _pendingBlockContinuationValid = true;
+        }
+        else
+        {
+            _pendingBlockContinuation = default;
+            _pendingBlockContinuationValid = false;
+        }
+
         return true;
     }
 
     public static bool TryConsumeCurrentThreadBlock(out string reason)
+    {
+        return TryConsumeCurrentThreadBlock(out reason, out _, out _);
+    }
+
+    public static bool TryConsumeCurrentThreadBlock(
+        out string reason,
+        out GuestCpuContinuation continuation,
+        out bool hasContinuation)
+    {
+        return TryConsumeCurrentThreadBlock(
+            out reason,
+            out continuation,
+            out hasContinuation,
+            out _,
+            out _,
+            out _,
+            out _);
+    }
+
+    public static bool TryConsumeCurrentThreadBlock(
+        out string reason,
+        out GuestCpuContinuation continuation,
+        out bool hasContinuation,
+        out string wakeKey,
+        out Func<int>? resumeHandler,
+        out Func<bool>? wakeHandler)
+    {
+        return TryConsumeCurrentThreadBlock(
+            out reason,
+            out continuation,
+            out hasContinuation,
+            out wakeKey,
+            out resumeHandler,
+            out wakeHandler,
+            out _);
+    }
+
+    public static bool TryConsumeCurrentThreadBlock(
+        out string reason,
+        out GuestCpuContinuation continuation,
+        out bool hasContinuation,
+        out string wakeKey,
+        out Func<int>? resumeHandler,
+        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;
+    }
+
+    public static long ComputeDeadlineTimestamp(TimeSpan timeout)
+    {
+        if (timeout <= TimeSpan.Zero)
+        {
+            return Stopwatch.GetTimestamp();
+        }
+
+        var ticks = timeout.TotalSeconds >= long.MaxValue / (double)Stopwatch.Frequency
+            ? long.MaxValue
+            : (long)Math.Ceiling(timeout.TotalSeconds * Stopwatch.Frequency);
+        var now = Stopwatch.GetTimestamp();
+        if (long.MaxValue - now <= ticks)
+        {
+            return long.MaxValue;
+        }
+
+        return now + Math.Max(1, ticks);
+    }
+
+    private static bool TryCaptureCurrentBlockContinuation(CpuContext context, out GuestCpuContinuation continuation)
+    {
+        if (!TryGetCurrentImportCallFrame(out var frame) ||
+            frame.ReturnRip < 65536 ||
+            frame.ResumeRsp == 0 ||
+            frame.ReturnSlotAddress == 0)
+        {
+            continuation = default;
+            return false;
+        }
+
+        continuation = new GuestCpuContinuation(
+            frame.ReturnRip,
+            frame.ResumeRsp,
+            frame.ReturnSlotAddress,
+            context.Rflags,
+            context.FsBase,
+            context.GsBase,
+            0,
+            context[CpuRegister.Rcx],
+            context[CpuRegister.Rdx],
+            context[CpuRegister.Rbx],
+            context[CpuRegister.Rbp],
+            context[CpuRegister.Rsi],
+            context[CpuRegister.Rdi],
+            context[CpuRegister.R8],
+            context[CpuRegister.R9],
+            context[CpuRegister.R12],
+            context[CpuRegister.R13],
+            context[CpuRegister.R14],
+            context[CpuRegister.R15]);
         return true;
     }
 
     public static void RequestCurrentEntryExit(string reason, int status)
+    {
+        RequestCurrentEntryExit(reason, unchecked((ulong)(long)status));
+    }
+
+    public static void RequestCurrentEntryExit(string reason, ulong value)
     {
         _pendingEntryExit = true;
-        _pendingEntryExitStatus = status;
+        _pendingEntryExitValue = value;
         _pendingEntryExitReason = string.IsNullOrWhiteSpace(reason) ? "guest_entry_exit" : reason;
     }
 
-    public static bool TryConsumeCurrentEntryExit(out int status, out string reason)
+    public static bool TryConsumeCurrentEntryExit(out ulong value, out string reason)
     {
-        status = _pendingEntryExitStatus;
+        value = _pendingEntryExitValue;
         reason = _pendingEntryExitReason ?? string.Empty;
         if (!_pendingEntryExit)
         {
@@ -100,8 +389,68 @@ public static class GuestThreadExecution
         }
 
         _pendingEntryExit = false;
-        _pendingEntryExitStatus = 0;
+        _pendingEntryExitValue = 0;
         _pendingEntryExitReason = null;
         return true;
     }
+
+    public static void RequestCurrentContextTransfer(GuestCpuContinuation target)
+    {
+        _pendingContextTransferTarget = target;
+        _pendingContextTransfer = true;
+    }
+
+    public static bool TryConsumeCurrentContextTransfer(out GuestCpuContinuation target)
+    {
+        target = _pendingContextTransferTarget;
+        if (!_pendingContextTransfer)
+        {
+            return false;
+        }
+
+        _pendingContextTransfer = false;
+        _pendingContextTransferTarget = default;
+        return true;
+    }
+
+    public static GuestImportCallFrame EnterImportCallFrame(
+        ulong returnRip,
+        ulong resumeRsp,
+        ulong returnSlotAddress)
+    {
+        var previous = new GuestImportCallFrame(
+            _hasCurrentImportCallFrame,
+            _currentImportReturnRip,
+            _currentImportResumeRsp,
+            _currentImportReturnSlotAddress);
+        _hasCurrentImportCallFrame = true;
+        _currentImportReturnRip = returnRip;
+        _currentImportResumeRsp = resumeRsp;
+        _currentImportReturnSlotAddress = returnSlotAddress;
+        return previous;
+    }
+
+    public static void RestoreImportCallFrame(GuestImportCallFrame previous)
+    {
+        _hasCurrentImportCallFrame = previous.IsValid;
+        _currentImportReturnRip = previous.ReturnRip;
+        _currentImportResumeRsp = previous.ResumeRsp;
+        _currentImportReturnSlotAddress = previous.ReturnSlotAddress;
+    }
+
+    public static bool TryGetCurrentImportCallFrame(out GuestImportCallFrame frame)
+    {
+        if (!_hasCurrentImportCallFrame)
+        {
+            frame = default;
+            return false;
+        }
+
+        frame = new GuestImportCallFrame(
+            true,
+            _currentImportReturnRip,
+            _currentImportResumeRsp,
+            _currentImportReturnSlotAddress);
+        return true;
+    }
 }

src/SharpEmu.HLE/IGuestMemoryAllocator.cs

@@ -0,0 +1,9 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE;
+
+public interface IGuestMemoryAllocator
+{
+    bool TryAllocateGuestMemory(ulong size, ulong alignment, out ulong address);
+}

src/SharpEmu.Libs/Agc/Gen5ShaderIr.cs

@@ -0,0 +1,311 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.Libs.Agc;
+
+internal enum Gen5ShaderEncoding
+{
+    Sop1,
+    Sop2,
+    Sopc,
+    Sopp,
+    Sopk,
+    Smrd,
+    Smem,
+    Mubuf,
+    Mtbuf,
+    Vop1,
+    Vop2,
+    Vopc,
+    Vop3,
+    Vintrp,
+    Ds,
+    Flat,
+    Vop3p,
+    Mimg,
+    Exp,
+}
+
+internal enum Gen5OperandKind
+{
+    ScalarRegister,
+    VectorRegister,
+    EncodedConstant,
+    LiteralConstant,
+}
+
+internal enum Gen5ShaderResourceKind
+{
+    ReadOnlyTexture,
+    ReadWriteTexture,
+    Sampler,
+    ConstantBuffer,
+}
+
+internal enum Gen5PixelOutputKind
+{
+    Float,
+    Uint,
+    Sint,
+}
+
+internal enum Gen5SpirvStage
+{
+    Vertex,
+    Pixel,
+    Compute,
+}
+
+internal sealed record Gen5SpirvShader(
+    byte[] Spirv,
+    IReadOnlyList<Gen5GlobalMemoryBinding> GlobalMemoryBindings,
+    IReadOnlyList<Gen5ImageBinding> ImageBindings,
+    uint AttributeCount,
+    IReadOnlyList<Gen5VertexInputBinding> VertexInputs);
+
+internal readonly record struct Gen5ShaderResourceMapping(
+    Gen5ShaderResourceKind Kind,
+    uint Slot,
+    uint OffsetDwords,
+    bool SizeFlag);
+
+internal sealed record Gen5ShaderMetadata(
+    uint ExtendedUserDataSizeDwords,
+    uint ShaderResourceTableSizeDwords,
+    IReadOnlyDictionary<uint, uint> DirectResources,
+    IReadOnlyList<Gen5ShaderResourceMapping> Resources);
+
+internal readonly record struct Gen5ComputeSystemRegisters(
+    uint? WorkGroupXRegister,
+    uint? WorkGroupYRegister,
+    uint? WorkGroupZRegister,
+    uint? ThreadGroupSizeRegister)
+{
+    public bool TryGetExpression(uint scalarRegister, out string expression)
+    {
+        if (WorkGroupXRegister == scalarRegister)
+        {
+            expression = "gl_WorkGroupID.x";
+            return true;
+        }
+
+        if (WorkGroupYRegister == scalarRegister)
+        {
+            expression = "gl_WorkGroupID.y";
+            return true;
+        }
+
+        if (WorkGroupZRegister == scalarRegister)
+        {
+            expression = "gl_WorkGroupID.z";
+            return true;
+        }
+
+        if (ThreadGroupSizeRegister == scalarRegister)
+        {
+            expression = "(gl_WorkGroupSize.x * gl_WorkGroupSize.y * gl_WorkGroupSize.z)";
+            return true;
+        }
+
+        expression = string.Empty;
+        return false;
+    }
+
+    public void ClearStaticValues(Span<uint> scalarRegisters)
+    {
+        ClearStaticValue(scalarRegisters, WorkGroupXRegister);
+        ClearStaticValue(scalarRegisters, WorkGroupYRegister);
+        ClearStaticValue(scalarRegisters, WorkGroupZRegister);
+        ClearStaticValue(scalarRegisters, ThreadGroupSizeRegister);
+    }
+
+    private static void ClearStaticValue(Span<uint> scalarRegisters, uint? scalarRegister)
+    {
+        if (scalarRegister is { } register && register < scalarRegisters.Length)
+        {
+            scalarRegisters[(int)register] = 0;
+        }
+    }
+}
+
+internal sealed record Gen5ShaderState(
+    Gen5ShaderProgram Program,
+    IReadOnlyList<uint> UserData,
+    Gen5ShaderMetadata? Metadata,
+    Gen5ComputeSystemRegisters? ComputeSystemRegisters = null,
+    uint UserDataScalarRegisterBase = 0);
+
+internal readonly record struct Gen5Operand(Gen5OperandKind Kind, uint Value)
+{
+    public static Gen5Operand Scalar(uint index) =>
+        new(Gen5OperandKind.ScalarRegister, index);
+
+    public static Gen5Operand Vector(uint index) =>
+        new(Gen5OperandKind.VectorRegister, index);
+
+    public static Gen5Operand Source(uint encoded, uint? literal = null)
+    {
+        if (encoded >= 256)
+        {
+            return Vector(encoded - 256);
+        }
+
+        if (encoded is 249 or 255 && literal.HasValue)
+        {
+            return new(Gen5OperandKind.LiteralConstant, literal.Value);
+        }
+
+        if (encoded <= 105 || encoded is 106 or 107 or 124 or 126 or 127)
+        {
+            return Scalar(encoded);
+        }
+
+        return new(Gen5OperandKind.EncodedConstant, encoded);
+    }
+
+    public override string ToString() => Kind switch
+    {
+        Gen5OperandKind.ScalarRegister => $"s{Value}",
+        Gen5OperandKind.VectorRegister => $"v{Value}",
+        Gen5OperandKind.LiteralConstant => $"0x{Value:X8}",
+        _ => $"src[{Value}]",
+    };
+}
+
+internal abstract record Gen5InstructionControl;
+
+internal sealed record Gen5ImageControl(
+    uint Dmask,
+    uint VectorAddress,
+    IReadOnlyList<uint> AddressRegisters,
+    uint VectorData,
+    uint ScalarResource,
+    uint ScalarSampler,
+    uint Dimension,
+    bool IsArray,
+    bool Glc,
+    bool Slc) : Gen5InstructionControl
+{
+    public uint GetAddressRegister(int component) =>
+        component < AddressRegisters.Count
+            ? AddressRegisters[component]
+            : VectorAddress + (uint)component;
+}
+
+internal sealed record Gen5GlobalMemoryControl(
+    uint DwordCount,
+    uint VectorAddress,
+    uint VectorData,
+    uint ScalarAddress,
+    int OffsetBytes,
+    bool Glc,
+    bool Slc) : Gen5InstructionControl;
+
+internal sealed record Gen5BufferMemoryControl(
+    uint DwordCount,
+    uint VectorAddress,
+    uint VectorData,
+    uint ScalarResource,
+    int OffsetBytes,
+    bool IndexEnabled,
+    bool OffsetEnabled,
+    bool Glc,
+    bool Slc) : Gen5InstructionControl;
+
+internal sealed record Gen5ExportControl(
+    uint Target,
+    uint EnableMask,
+    bool Compressed,
+    bool Done,
+    bool ValidMask) : Gen5InstructionControl;
+
+internal sealed record Gen5InterpolationControl(
+    uint Attribute,
+    uint Channel) : Gen5InstructionControl;
+
+internal sealed record Gen5Vop3Control(
+    uint AbsoluteMask,
+    uint NegateMask,
+    uint OutputModifier,
+    bool Clamp,
+    uint? ScalarDestination) : Gen5InstructionControl;
+
+internal sealed record Gen5SdwaControl(
+    uint DestinationSelect,
+    uint Source0Select,
+    uint Source1Select,
+    uint AbsoluteMask,
+    uint NegateMask,
+    uint OutputModifier,
+    bool Clamp) : Gen5InstructionControl;
+
+internal sealed record Gen5DppControl(
+    uint Control,
+    bool FetchInactive,
+    bool BoundControl,
+    uint AbsoluteMask,
+    uint NegateMask,
+    uint BankMask,
+    uint RowMask) : Gen5InstructionControl;
+
+internal sealed record Gen5ScalarMemoryControl(
+    uint DestinationCount,
+    int ImmediateOffsetBytes,
+    uint? DynamicOffsetRegister) : Gen5InstructionControl;
+
+internal sealed record Gen5DataShareControl(
+    uint Offset0,
+    uint Offset1,
+    bool Gds) : Gen5InstructionControl;
+
+internal sealed record Gen5ImageBinding(
+    uint Pc,
+    string Opcode,
+    Gen5ImageControl Control,
+    IReadOnlyList<uint> ResourceDescriptor,
+    IReadOnlyList<uint> SamplerDescriptor,
+    uint? MipLevel);
+
+internal sealed record Gen5GlobalMemoryBinding(
+    uint ScalarAddress,
+    ulong BaseAddress,
+    IReadOnlyList<uint> InstructionPcs,
+    byte[] Data);
+
+internal sealed record Gen5VertexInputBinding(
+    uint Pc,
+    uint Location,
+    uint ComponentCount,
+    ulong BaseAddress,
+    uint Stride,
+    uint OffsetBytes,
+    byte[] Data);
+
+internal sealed record Gen5ShaderEvaluation(
+    IReadOnlyList<uint> InitialScalarRegisters,
+    IReadOnlyList<uint> ScalarRegisters,
+    IReadOnlyDictionary<uint, IReadOnlyList<uint>> ScalarRegistersByPc,
+    IReadOnlyList<Gen5ImageBinding> ImageBindings,
+    IReadOnlyList<Gen5GlobalMemoryBinding> GlobalMemoryBindings,
+    Gen5ComputeSystemRegisters? ComputeSystemRegisters = null,
+    IReadOnlySet<uint>? RuntimeScalarRegisters = null,
+    IReadOnlyList<Gen5VertexInputBinding>? VertexInputs = null);
+
+internal sealed record Gen5ShaderInstruction(
+    uint Pc,
+    Gen5ShaderEncoding Encoding,
+    string Opcode,
+    IReadOnlyList<uint> Words,
+    IReadOnlyList<Gen5Operand> Sources,
+    IReadOnlyList<Gen5Operand> Destinations,
+    Gen5InstructionControl? Control);
+
+internal sealed record Gen5ShaderProgram(
+    ulong Address,
+    IReadOnlyList<Gen5ShaderInstruction> Instructions)
+{
+    public IEnumerable<Gen5ImageControl> ImageResources =>
+        Instructions
+            .Select(instruction => instruction.Control)
+            .OfType<Gen5ImageControl>();
+}

src/SharpEmu.Libs/Agc/Gen5ShaderMetadataReader.cs

@@ -0,0 +1,154 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+
+namespace SharpEmu.Libs.Agc;
+
+internal static class Gen5ShaderMetadataReader
+{
+    private const ulong ShaderUserDataOffset = 0x08;
+    private const int ResourceClassCount = 4;
+    private const int MaxMetadataEntries = 4096;
+
+    public static bool TryRead(
+        CpuContext ctx,
+        ulong shaderHeaderAddress,
+        out Gen5ShaderMetadata metadata)
+    {
+        metadata = default!;
+        if (!TryReadUInt64(ctx, shaderHeaderAddress + ShaderUserDataOffset, out var userDataAddress) ||
+            userDataAddress == 0 ||
+            !TryReadUInt64(ctx, userDataAddress, out var directResourceOffsetsAddress))
+        {
+            return false;
+        }
+
+        var resourceOffsets = new ulong[ResourceClassCount];
+        for (var resourceClass = 0; resourceClass < ResourceClassCount; resourceClass++)
+        {
+            if (!TryReadUInt64(
+                    ctx,
+                    userDataAddress + 0x08 + (ulong)(resourceClass * sizeof(ulong)),
+                    out resourceOffsets[resourceClass]))
+            {
+                return false;
+            }
+        }
+
+        if (!TryReadUInt16(ctx, userDataAddress + 0x28, out var extendedUserDataSize) ||
+            !TryReadUInt16(ctx, userDataAddress + 0x2A, out var shaderResourceTableSize) ||
+            !TryReadUInt16(ctx, userDataAddress + 0x2C, out var directResourceCount) ||
+            directResourceCount > MaxMetadataEntries)
+        {
+            return false;
+        }
+
+        var resourceCounts = new ushort[ResourceClassCount];
+        for (var resourceClass = 0; resourceClass < ResourceClassCount; resourceClass++)
+        {
+            if (!TryReadUInt16(
+                    ctx,
+                    userDataAddress + 0x2E + (ulong)(resourceClass * sizeof(ushort)),
+                    out resourceCounts[resourceClass]) ||
+                resourceCounts[resourceClass] > MaxMetadataEntries)
+            {
+                return false;
+            }
+        }
+
+        var directResources = new Dictionary<uint, uint>();
+        if (directResourceCount != 0)
+        {
+            if (directResourceOffsetsAddress == 0)
+            {
+                return false;
+            }
+
+            for (uint type = 0; type < directResourceCount; type++)
+            {
+                if (!TryReadUInt16(ctx, directResourceOffsetsAddress + type * sizeof(ushort), out var offset))
+                {
+                    return false;
+                }
+
+                if (offset != ushort.MaxValue)
+                {
+                    directResources[type] = offset;
+                }
+            }
+        }
+
+        var resources = new List<Gen5ShaderResourceMapping>();
+        for (var resourceClass = 0; resourceClass < ResourceClassCount; resourceClass++)
+        {
+            var count = resourceCounts[resourceClass];
+            if (count == 0)
+            {
+                continue;
+            }
+
+            if (resourceOffsets[resourceClass] == 0)
+            {
+                return false;
+            }
+
+            for (uint slot = 0; slot < count; slot++)
+            {
+                if (!TryReadUInt16(
+                        ctx,
+                        resourceOffsets[resourceClass] + slot * sizeof(ushort),
+                        out var sharp))
+                {
+                    return false;
+                }
+
+                var offset = (uint)(sharp & 0x7FFF);
+                if (offset == 0x7FFF)
+                {
+                    continue;
+                }
+
+                resources.Add(new Gen5ShaderResourceMapping(
+                    (Gen5ShaderResourceKind)resourceClass,
+                    slot,
+                    offset,
+                    (sharp & 0x8000) != 0));
+            }
+        }
+
+        metadata = new Gen5ShaderMetadata(
+            extendedUserDataSize,
+            shaderResourceTableSize,
+            directResources,
+            resources);
+        return true;
+    }
+
+    private static bool TryReadUInt16(CpuContext ctx, ulong address, out ushort value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(ushort)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt16LittleEndian(bytes);
+        return true;
+    }
+
+    private static bool TryReadUInt64(CpuContext ctx, ulong address, out ulong value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(ulong)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt64LittleEndian(bytes);
+        return true;
+    }
+}

src/SharpEmu.Libs/Agc/SpirvFixedShaders.cs

@@ -0,0 +1,167 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.Libs.Agc;
+
+internal static class SpirvFixedShaders
+{
+    public static byte[] CreateFullscreenVertex(uint attributeCount)
+    {
+        var module = new SpirvModuleBuilder();
+        module.AddCapability(SpirvCapability.Shader);
+
+        var voidType = module.TypeVoid();
+        var boolType = module.TypeBool();
+        var uintType = module.TypeInt(32, signed: false);
+        var floatType = module.TypeFloat(32);
+        var vec4Type = module.TypeVector(floatType, 4);
+        var inputUintPointer = module.TypePointer(SpirvStorageClass.Input, uintType);
+        var outputVec4Pointer = module.TypePointer(SpirvStorageClass.Output, vec4Type);
+
+        var vertexIndex = module.AddGlobalVariable(inputUintPointer, SpirvStorageClass.Input);
+        module.AddName(vertexIndex, "vertexIndex");
+        module.AddDecoration(
+            vertexIndex,
+            SpirvDecoration.BuiltIn,
+            (uint)SpirvBuiltIn.VertexIndex);
+
+        var position = module.AddGlobalVariable(outputVec4Pointer, SpirvStorageClass.Output);
+        module.AddName(position, "position");
+        module.AddDecoration(position, SpirvDecoration.BuiltIn, (uint)SpirvBuiltIn.Position);
+
+        var attributes = new uint[attributeCount];
+        for (uint index = 0; index < attributeCount; index++)
+        {
+            attributes[index] =
+                module.AddGlobalVariable(outputVec4Pointer, SpirvStorageClass.Output);
+            module.AddName(attributes[index], $"attr{index}");
+            module.AddDecoration(attributes[index], SpirvDecoration.Location, index);
+            module.AddDecoration(attributes[index], SpirvDecoration.NoPerspective);
+        }
+
+        var functionType = module.TypeFunction(voidType);
+        var main = module.BeginFunction(voidType, functionType);
+        module.AddName(main, "main");
+        module.AddLabel();
+
+        var indexValue = module.AddInstruction(SpirvOp.Load, uintType, vertexIndex);
+        var one = module.Constant(uintType, 1);
+        var two = module.Constant(uintType, 2);
+        var shifted = module.AddInstruction(SpirvOp.ShiftLeftLogical, uintType, indexValue, one);
+        var xBits = module.AddInstruction(SpirvOp.BitwiseAnd, uintType, shifted, two);
+        var yBits = module.AddInstruction(SpirvOp.BitwiseAnd, uintType, indexValue, two);
+        var x = module.AddInstruction(SpirvOp.ConvertUToF, floatType, xBits);
+        var y = module.AddInstruction(SpirvOp.ConvertUToF, floatType, yBits);
+        var zero = module.ConstantFloat(floatType, 0f);
+        var oneFloat = module.ConstantFloat(floatType, 1f);
+        var twoFloat = module.ConstantFloat(floatType, 2f);
+        var xPosition = module.AddInstruction(SpirvOp.FMul, floatType, x, twoFloat);
+        xPosition = module.AddInstruction(SpirvOp.FSub, floatType, xPosition, oneFloat);
+        var yPosition = module.AddInstruction(SpirvOp.FMul, floatType, y, twoFloat);
+        yPosition = module.AddInstruction(SpirvOp.FSub, floatType, yPosition, oneFloat);
+        var positionValue = module.AddInstruction(
+            SpirvOp.CompositeConstruct,
+            vec4Type,
+            xPosition,
+            yPosition,
+            zero,
+            oneFloat);
+        module.AddStatement(SpirvOp.Store, position, positionValue);
+
+        var attributeValue = module.AddInstruction(
+            SpirvOp.CompositeConstruct,
+            vec4Type,
+            x,
+            y,
+            zero,
+            oneFloat);
+        foreach (var attribute in attributes)
+        {
+            module.AddStatement(SpirvOp.Store, attribute, attributeValue);
+        }
+
+        module.AddStatement(SpirvOp.Return);
+        module.EndFunction();
+
+        var interfaces = new uint[2 + attributes.Length];
+        interfaces[0] = vertexIndex;
+        interfaces[1] = position;
+        attributes.CopyTo(interfaces, 2);
+        module.AddEntryPoint(SpirvExecutionModel.Vertex, main, "main", interfaces);
+        _ = boolType;
+        return module.Build();
+    }
+
+    public static byte[] CreateCopyFragment()
+    {
+        var module = new SpirvModuleBuilder();
+        module.AddCapability(SpirvCapability.Shader);
+
+        var voidType = module.TypeVoid();
+        var floatType = module.TypeFloat(32);
+        var vec2Type = module.TypeVector(floatType, 2);
+        var vec4Type = module.TypeVector(floatType, 4);
+        var inputVec4Pointer = module.TypePointer(SpirvStorageClass.Input, vec4Type);
+        var outputVec4Pointer = module.TypePointer(SpirvStorageClass.Output, vec4Type);
+        var imageType = module.TypeImage(
+            floatType,
+            SpirvImageDim.Dim2D,
+            depth: false,
+            arrayed: false,
+            multisampled: false,
+            sampled: 1,
+            SpirvImageFormat.Unknown);
+        var sampledImageType = module.TypeSampledImage(imageType);
+        var sampledImagePointer =
+            module.TypePointer(SpirvStorageClass.UniformConstant, sampledImageType);
+
+        var attribute = module.AddGlobalVariable(inputVec4Pointer, SpirvStorageClass.Input);
+        module.AddName(attribute, "attr0");
+        module.AddDecoration(attribute, SpirvDecoration.Location, 0);
+
+        var texture = module.AddGlobalVariable(
+            sampledImagePointer,
+            SpirvStorageClass.UniformConstant);
+        module.AddName(texture, "tex0");
+        module.AddDecoration(texture, SpirvDecoration.DescriptorSet, 0);
+        module.AddDecoration(texture, SpirvDecoration.Binding, 1);
+
+        var output = module.AddGlobalVariable(outputVec4Pointer, SpirvStorageClass.Output);
+        module.AddName(output, "outColor");
+        module.AddDecoration(output, SpirvDecoration.Location, 0);
+
+        var functionType = module.TypeFunction(voidType);
+        var main = module.BeginFunction(voidType, functionType);
+        module.AddName(main, "main");
+        module.AddLabel();
+
+        var attributeValue = module.AddInstruction(SpirvOp.Load, vec4Type, attribute);
+        var coordinates = module.AddInstruction(
+            SpirvOp.VectorShuffle,
+            vec2Type,
+            attributeValue,
+            attributeValue,
+            0,
+            1);
+        var sampledImage = module.AddInstruction(SpirvOp.Load, sampledImageType, texture);
+        var lod = module.ConstantFloat(floatType, 0f);
+        var color = module.AddInstruction(
+            SpirvOp.ImageSampleExplicitLod,
+            vec4Type,
+            sampledImage,
+            coordinates,
+            2,
+            lod);
+        module.AddStatement(SpirvOp.Store, output, color);
+        module.AddStatement(SpirvOp.Return);
+        module.EndFunction();
+
+        module.AddEntryPoint(
+            SpirvExecutionModel.Fragment,
+            main,
+            "main",
+            [attribute, texture, output]);
+        module.AddExecutionMode(main, SpirvExecutionMode.OriginUpperLeft);
+        return module.Build();
+    }
+}

src/SharpEmu.Libs/Agc/SpirvModuleBuilder.cs

@@ -0,0 +1,885 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Buffers.Binary;
+using System.Text;
+
+namespace SharpEmu.Libs.Agc;
+
+internal enum SpirvOp : ushort
+{
+    Nop = 0,
+    Name = 5,
+    Extension = 10,
+    ExtInstImport = 11,
+    ExtInst = 12,
+    MemoryModel = 14,
+    EntryPoint = 15,
+    ExecutionMode = 16,
+    Capability = 17,
+    TypeVoid = 19,
+    TypeBool = 20,
+    TypeInt = 21,
+    TypeFloat = 22,
+    TypeVector = 23,
+    TypeImage = 25,
+    TypeSampler = 26,
+    TypeSampledImage = 27,
+    TypeArray = 28,
+    TypeRuntimeArray = 29,
+    TypeStruct = 30,
+    TypePointer = 32,
+    TypeFunction = 33,
+    ConstantTrue = 41,
+    ConstantFalse = 42,
+    Constant = 43,
+    ConstantComposite = 44,
+    ConstantNull = 46,
+    Function = 54,
+    FunctionParameter = 55,
+    FunctionEnd = 56,
+    FunctionCall = 57,
+    Variable = 59,
+    Load = 61,
+    Store = 62,
+    AccessChain = 65,
+    ArrayLength = 68,
+    Decorate = 71,
+    VectorExtractDynamic = 77,
+    VectorInsertDynamic = 78,
+    VectorShuffle = 79,
+    CompositeConstruct = 80,
+    CompositeExtract = 81,
+    CompositeInsert = 82,
+    CopyObject = 83,
+    SampledImage = 86,
+    ImageSampleImplicitLod = 87,
+    ImageSampleExplicitLod = 88,
+    ImageSampleDrefImplicitLod = 89,
+    ImageSampleDrefExplicitLod = 90,
+    ImageFetch = 95,
+    ImageGather = 96,
+    ImageDrefGather = 97,
+    ImageRead = 98,
+    ImageWrite = 99,
+    Image = 100,
+    ImageQuerySizeLod = 103,
+    ImageQuerySize = 104,
+    ImageQueryLod = 105,
+    ImageQueryLevels = 106,
+    ImageQuerySamples = 107,
+    ConvertFToU = 109,
+    ConvertFToS = 110,
+    ConvertSToF = 111,
+    ConvertUToF = 112,
+    UConvert = 113,
+    SConvert = 114,
+    FConvert = 115,
+    Bitcast = 124,
+    SNegate = 126,
+    FNegate = 127,
+    IAdd = 128,
+    FAdd = 129,
+    ISub = 130,
+    FSub = 131,
+    IMul = 132,
+    FMul = 133,
+    UDiv = 134,
+    SDiv = 135,
+    FDiv = 136,
+    UMod = 137,
+    SRem = 138,
+    SMod = 139,
+    FRem = 140,
+    FMod = 141,
+    IAddCarry = 149,
+    ISubBorrow = 150,
+    UMulExtended = 151,
+    SMulExtended = 152,
+    Any = 154,
+    All = 155,
+    IsNan = 156,
+    IsInf = 157,
+    LogicalEqual = 164,
+    LogicalNotEqual = 165,
+    LogicalOr = 166,
+    LogicalAnd = 167,
+    LogicalNot = 168,
+    Select = 169,
+    IEqual = 170,
+    INotEqual = 171,
+    UGreaterThan = 172,
+    SGreaterThan = 173,
+    UGreaterThanEqual = 174,
+    SGreaterThanEqual = 175,
+    ULessThan = 176,
+    SLessThan = 177,
+    ULessThanEqual = 178,
+    SLessThanEqual = 179,
+    FOrdEqual = 180,
+    FUnordEqual = 181,
+    FOrdNotEqual = 182,
+    FUnordNotEqual = 183,
+    FOrdLessThan = 184,
+    FUnordLessThan = 185,
+    FOrdGreaterThan = 186,
+    FUnordGreaterThan = 187,
+    FOrdLessThanEqual = 188,
+    FUnordLessThanEqual = 189,
+    FOrdGreaterThanEqual = 190,
+    FUnordGreaterThanEqual = 191,
+    ShiftRightLogical = 194,
+    ShiftRightArithmetic = 195,
+    ShiftLeftLogical = 196,
+    BitwiseOr = 197,
+    BitwiseXor = 198,
+    BitwiseAnd = 199,
+    Not = 200,
+    BitFieldInsert = 201,
+    BitFieldSExtract = 202,
+    BitFieldUExtract = 203,
+    BitReverse = 204,
+    BitCount = 205,
+    ControlBarrier = 224,
+    MemoryBarrier = 225,
+    AtomicIAdd = 234,
+    Phi = 245,
+    LoopMerge = 246,
+    SelectionMerge = 247,
+    Label = 248,
+    Branch = 249,
+    BranchConditional = 250,
+    Switch = 251,
+    Kill = 252,
+    Return = 253,
+    ReturnValue = 254,
+    Unreachable = 255,
+    GroupNonUniformElect = 333,
+    GroupNonUniformAll = 334,
+    GroupNonUniformAny = 335,
+    GroupNonUniformAllEqual = 336,
+    GroupNonUniformBroadcast = 337,
+    GroupNonUniformBroadcastFirst = 338,
+    GroupNonUniformBallot = 339,
+    GroupNonUniformShuffle = 345,
+    GroupNonUniformShuffleXor = 346,
+    GroupNonUniformShuffleUp = 347,
+    GroupNonUniformShuffleDown = 348,
+}
+
+internal enum SpirvCapability : uint
+{
+    Shader = 1,
+    Float16 = 9,
+    Float64 = 10,
+    Int64 = 11,
+    Int16 = 22,
+    ImageGatherExtended = 25,
+    StorageImageExtendedFormats = 49,
+    ImageQuery = 50,
+    StorageImageReadWithoutFormat = 55,
+    StorageImageWriteWithoutFormat = 56,
+    GroupNonUniform = 61,
+    GroupNonUniformVote = 62,
+    GroupNonUniformBallot = 64,
+    GroupNonUniformShuffle = 65,
+    RuntimeDescriptorArray = 5302,
+}
+
+internal enum SpirvStorageClass : uint
+{
+    UniformConstant = 0,
+    Input = 1,
+    Uniform = 2,
+    Output = 3,
+    Workgroup = 4,
+    Private = 6,
+    Function = 7,
+    PushConstant = 9,
+    Image = 11,
+    StorageBuffer = 12,
+}
+
+internal enum SpirvExecutionModel : uint
+{
+    Vertex = 0,
+    Fragment = 4,
+    GLCompute = 5,
+}
+
+internal enum SpirvExecutionMode : uint
+{
+    OriginUpperLeft = 7,
+    DepthReplacing = 12,
+    LocalSize = 17,
+}
+
+internal enum SpirvDecoration : uint
+{
+    Block = 2,
+    ArrayStride = 6,
+    BuiltIn = 11,
+    NoPerspective = 13,
+    Flat = 14,
+    Location = 30,
+    Binding = 33,
+    DescriptorSet = 34,
+    Offset = 35,
+}
+
+internal enum SpirvBuiltIn : uint
+{
+    Position = 0,
+    VertexIndex = 42,
+    InstanceIndex = 43,
+    FragCoord = 15,
+    FrontFacing = 17,
+    WorkgroupId = 26,
+    LocalInvocationId = 27,
+    GlobalInvocationId = 28,
+    LocalInvocationIndex = 29,
+    SubgroupLocalInvocationId = 41,
+}
+
+internal enum SpirvImageDim : uint
+{
+    Dim1D = 0,
+    Dim2D = 1,
+    Dim3D = 2,
+    Cube = 3,
+    Buffer = 5,
+}
+
+internal enum SpirvImageFormat : uint
+{
+    Unknown = 0,
+    Rgba32f = 1,
+    Rgba16f = 2,
+    R32f = 3,
+    Rgba8 = 4,
+    Rgba8Snorm = 5,
+    Rg32f = 6,
+    Rg16f = 7,
+    R11fG11fB10f = 8,
+    R16f = 9,
+    Rgba16 = 10,
+    Rgb10A2 = 11,
+    Rg16 = 12,
+    Rg8 = 13,
+    R16 = 14,
+    R8 = 15,
+    Rgba16Snorm = 16,
+    Rg16Snorm = 17,
+    Rg8Snorm = 18,
+    R16Snorm = 19,
+    R8Snorm = 20,
+    Rgba32i = 21,
+    Rgba16i = 22,
+    Rgba8i = 23,
+    R32i = 24,
+    Rg32i = 25,
+    Rg16i = 26,
+    Rg8i = 27,
+    R16i = 28,
+    R8i = 29,
+    Rgba32ui = 30,
+    Rgba16ui = 31,
+    Rgba8ui = 32,
+    R32ui = 33,
+    Rgb10A2ui = 34,
+    Rg32ui = 35,
+    Rg16ui = 36,
+    Rg8ui = 37,
+    R16ui = 38,
+    R8ui = 39,
+}
+
+internal sealed class SpirvModuleBuilder
+{
+    private const uint Magic = 0x07230203;
+    private const uint Version15 = 0x00010500;
+    private const uint Generator = 0x53504500; // "SPE"
+
+    private readonly List<uint> _capabilities = [];
+    private readonly List<uint> _extensions = [];
+    private readonly List<uint> _imports = [];
+    private readonly List<uint> _memoryModel = [];
+    private readonly List<uint> _entryPoints = [];
+    private readonly List<uint> _executionModes = [];
+    private readonly List<uint> _debug = [];
+    private readonly List<uint> _annotations = [];
+    private readonly List<uint> _typesConstantsGlobals = [];
+    private readonly List<uint> _functions = [];
+    private readonly Dictionary<(uint Width, bool Signed), uint> _integerTypes = [];
+    private readonly Dictionary<uint, uint> _floatTypes = [];
+    private readonly Dictionary<(uint Component, uint Count), uint> _vectorTypes = [];
+    private readonly Dictionary<
+        (
+            uint SampledType,
+            SpirvImageDim Dimension,
+            bool Depth,
+            bool Arrayed,
+            bool Multisampled,
+            uint Sampled,
+            SpirvImageFormat Format
+        ),
+        uint> _imageTypes = [];
+    private readonly Dictionary<uint, uint> _sampledImageTypes = [];
+    private readonly Dictionary<(SpirvStorageClass Storage, uint Type), uint> _pointerTypes = [];
+    private readonly Dictionary<(uint Element, uint Count), uint> _arrayTypes = [];
+    private readonly Dictionary<uint, uint> _runtimeArrayTypes = [];
+    private readonly Dictionary<string, uint> _functionTypes = [];
+    private readonly Dictionary<(uint Type, ulong Value), uint> _constants = [];
+    private readonly HashSet<SpirvCapability> _declaredCapabilities = [];
+    private readonly Dictionary<string, uint> _extInstImports = [];
+    private uint _nextId = 1;
+    private uint? _voidType;
+    private uint? _boolType;
+
+    public uint AllocateId() => _nextId++;
+
+    public void AddCapability(SpirvCapability capability)
+    {
+        if (_declaredCapabilities.Add(capability))
+        {
+            Emit(_capabilities, SpirvOp.Capability, (uint)capability);
+        }
+    }
+
+    public void AddExtension(string extension) =>
+        EmitWithString(_extensions, SpirvOp.Extension, [], extension);
+
+    public uint ImportExtInst(string name)
+    {
+        if (_extInstImports.TryGetValue(name, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        EmitWithString(_imports, SpirvOp.ExtInstImport, [id], name);
+        _extInstImports.Add(name, id);
+        return id;
+    }
+
+    public void SetLogicalGlsl450MemoryModel() =>
+        Emit(_memoryModel, SpirvOp.MemoryModel, 0, 1);
+
+    public void AddEntryPoint(
+        SpirvExecutionModel model,
+        uint function,
+        string name,
+        IReadOnlyList<uint> interfaces)
+    {
+        var prefix = new uint[2 + interfaces.Count];
+        prefix[0] = (uint)model;
+        prefix[1] = function;
+        for (var index = 0; index < interfaces.Count; index++)
+        {
+            prefix[index + 2] = interfaces[index];
+        }
+
+        EmitWithString(_entryPoints, SpirvOp.EntryPoint, prefix, name, stringBeforeTailCount: 2);
+    }
+
+    public void AddExecutionMode(uint function, SpirvExecutionMode mode, params uint[] operands)
+    {
+        var values = new uint[2 + operands.Length];
+        values[0] = function;
+        values[1] = (uint)mode;
+        operands.CopyTo(values, 2);
+        Emit(_executionModes, SpirvOp.ExecutionMode, values);
+    }
+
+    public void AddName(uint target, string name) =>
+        EmitWithString(_debug, SpirvOp.Name, [target], name);
+
+    public void AddDecoration(uint target, SpirvDecoration decoration, params uint[] operands)
+    {
+        var values = new uint[2 + operands.Length];
+        values[0] = target;
+        values[1] = (uint)decoration;
+        operands.CopyTo(values, 2);
+        Emit(_annotations, SpirvOp.Decorate, values);
+    }
+
+    public void AddMemberDecoration(
+        uint target,
+        uint member,
+        SpirvDecoration decoration,
+        params uint[] operands)
+    {
+        var values = new uint[3 + operands.Length];
+        values[0] = target;
+        values[1] = member;
+        values[2] = (uint)decoration;
+        operands.CopyTo(values, 3);
+        EmitRaw(_annotations, 72, values);
+    }
+
+    public uint TypeVoid()
+    {
+        if (_voidType is { } existing)
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeVoid, id);
+        _voidType = id;
+        return id;
+    }
+
+    public uint TypeBool()
+    {
+        if (_boolType is { } existing)
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeBool, id);
+        _boolType = id;
+        return id;
+    }
+
+    public uint TypeInt(uint width, bool signed)
+    {
+        var key = (width, signed);
+        if (_integerTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeInt, id, width, signed ? 1u : 0u);
+        _integerTypes.Add(key, id);
+        return id;
+    }
+
+    public uint TypeFloat(uint width)
+    {
+        if (_floatTypes.TryGetValue(width, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeFloat, id, width);
+        _floatTypes.Add(width, id);
+        return id;
+    }
+
+    public uint TypeVector(uint componentType, uint count)
+    {
+        var key = (componentType, count);
+        if (_vectorTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeVector, id, componentType, count);
+        _vectorTypes.Add(key, id);
+        return id;
+    }
+
+    public uint TypeImage(
+        uint sampledType,
+        SpirvImageDim dimension,
+        bool depth,
+        bool arrayed,
+        bool multisampled,
+        uint sampled,
+        SpirvImageFormat format)
+    {
+        var key = (
+            sampledType,
+            dimension,
+            depth,
+            arrayed,
+            multisampled,
+            sampled,
+            format);
+        if (_imageTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(
+            _typesConstantsGlobals,
+            SpirvOp.TypeImage,
+            id,
+            sampledType,
+            (uint)dimension,
+            depth ? 1u : 0u,
+            arrayed ? 1u : 0u,
+            multisampled ? 1u : 0u,
+            sampled,
+            (uint)format);
+        _imageTypes.Add(key, id);
+        return id;
+    }
+
+    public uint TypeSampledImage(uint imageType)
+    {
+        if (_sampledImageTypes.TryGetValue(imageType, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeSampledImage, id, imageType);
+        _sampledImageTypes.Add(imageType, id);
+        return id;
+    }
+
+    public uint TypeArray(uint elementType, uint count)
+    {
+        var key = (elementType, count);
+        if (_arrayTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var length = Constant(TypeInt(32, false), count);
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeArray, id, elementType, length);
+        _arrayTypes.Add(key, id);
+        return id;
+    }
+
+    public uint TypeRuntimeArray(uint elementType)
+    {
+        if (_runtimeArrayTypes.TryGetValue(elementType, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypeRuntimeArray, id, elementType);
+        _runtimeArrayTypes.Add(elementType, id);
+        return id;
+    }
+
+    public uint TypeStruct(params uint[] memberTypes)
+    {
+        var id = AllocateId();
+        var operands = new uint[memberTypes.Length + 1];
+        operands[0] = id;
+        memberTypes.CopyTo(operands, 1);
+        Emit(_typesConstantsGlobals, SpirvOp.TypeStruct, operands);
+        return id;
+    }
+
+    public uint TypePointer(SpirvStorageClass storageClass, uint type)
+    {
+        var key = (storageClass, type);
+        if (_pointerTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.TypePointer, id, (uint)storageClass, type);
+        _pointerTypes.Add(key, id);
+        return id;
+    }
+
+    public uint TypeFunction(uint returnType, params uint[] parameterTypes)
+    {
+        var key = returnType + ":" + string.Join(',', parameterTypes);
+        if (_functionTypes.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        var operands = new uint[parameterTypes.Length + 2];
+        operands[0] = id;
+        operands[1] = returnType;
+        parameterTypes.CopyTo(operands, 2);
+        Emit(_typesConstantsGlobals, SpirvOp.TypeFunction, operands);
+        _functionTypes.Add(key, id);
+        return id;
+    }
+
+    public uint ConstantBool(bool value)
+    {
+        var type = TypeBool();
+        var key = (type, value ? 1UL : 0UL);
+        if (_constants.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(
+            _typesConstantsGlobals,
+            value ? SpirvOp.ConstantTrue : SpirvOp.ConstantFalse,
+            type,
+            id);
+        _constants.Add(key, id);
+        return id;
+    }
+
+    public uint Constant(uint type, uint value)
+    {
+        var key = (type, (ulong)value);
+        if (_constants.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.Constant, type, id, value);
+        _constants.Add(key, id);
+        return id;
+    }
+
+    public uint Constant64(uint type, ulong value)
+    {
+        var key = (type, value);
+        if (_constants.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(
+            _typesConstantsGlobals,
+            SpirvOp.Constant,
+            type,
+            id,
+            (uint)value,
+            (uint)(value >> 32));
+        _constants.Add(key, id);
+        return id;
+    }
+
+    public uint ConstantFloat(uint type, float value) =>
+        Constant(type, BitConverter.SingleToUInt32Bits(value));
+
+    public uint ConstantComposite(uint type, params uint[] constituents)
+    {
+        var id = AllocateId();
+        var operands = new uint[constituents.Length + 2];
+        operands[0] = type;
+        operands[1] = id;
+        constituents.CopyTo(operands, 2);
+        Emit(_typesConstantsGlobals, SpirvOp.ConstantComposite, operands);
+        return id;
+    }
+
+    public uint ConstantNull(uint type)
+    {
+        var key = (type, ulong.MaxValue);
+        if (_constants.TryGetValue(key, out var existing))
+        {
+            return existing;
+        }
+
+        var id = AllocateId();
+        Emit(_typesConstantsGlobals, SpirvOp.ConstantNull, type, id);
+        _constants.Add(key, id);
+        return id;
+    }
+
+    public uint AddGlobalVariable(
+        uint pointerType,
+        SpirvStorageClass storageClass,
+        uint? initializer = null)
+    {
+        var id = AllocateId();
+        if (initializer.HasValue)
+        {
+            Emit(
+                _typesConstantsGlobals,
+                SpirvOp.Variable,
+                pointerType,
+                id,
+                (uint)storageClass,
+                initializer.Value);
+        }
+        else
+        {
+            Emit(_typesConstantsGlobals, SpirvOp.Variable, pointerType, id, (uint)storageClass);
+        }
+
+        return id;
+    }
+
+    public uint BeginFunction(uint returnType, uint functionType)
+    {
+        var id = AllocateId();
+        Emit(_functions, SpirvOp.Function, returnType, id, 0, functionType);
+        return id;
+    }
+
+    public uint AddFunctionParameter(uint type) =>
+        EmitResult(_functions, SpirvOp.FunctionParameter, type);
+
+    public uint AddLabel(uint? id = null)
+    {
+        var result = id ?? AllocateId();
+        Emit(_functions, SpirvOp.Label, result);
+        return result;
+    }
+
+    public uint AddFunctionVariable(uint pointerType, uint? initializer = null)
+    {
+        var id = AllocateId();
+        if (initializer.HasValue)
+        {
+            Emit(
+                _functions,
+                SpirvOp.Variable,
+                pointerType,
+                id,
+                (uint)SpirvStorageClass.Function,
+                initializer.Value);
+        }
+        else
+        {
+            Emit(
+                _functions,
+                SpirvOp.Variable,
+                pointerType,
+                id,
+                (uint)SpirvStorageClass.Function);
+        }
+
+        return id;
+    }
+
+    public uint AddInstruction(SpirvOp opcode, uint resultType, params uint[] operands) =>
+        EmitResult(_functions, opcode, resultType, operands);
+
+    public void AddStatement(SpirvOp opcode, params uint[] operands) =>
+        Emit(_functions, opcode, operands);
+
+    public void EndFunction() => Emit(_functions, SpirvOp.FunctionEnd);
+
+    public byte[] Build()
+    {
+        if (_memoryModel.Count == 0)
+        {
+            SetLogicalGlsl450MemoryModel();
+        }
+
+        var wordCount =
+            5 +
+            _capabilities.Count +
+            _extensions.Count +
+            _imports.Count +
+            _memoryModel.Count +
+            _entryPoints.Count +
+            _executionModes.Count +
+            _debug.Count +
+            _annotations.Count +
+            _typesConstantsGlobals.Count +
+            _functions.Count;
+        var words = new uint[wordCount];
+        var offset = 0;
+        WriteWord(Magic);
+        WriteWord(Version15);
+        WriteWord(Generator);
+        WriteWord(_nextId);
+        WriteWord(0);
+        WriteSection(_capabilities);
+        WriteSection(_extensions);
+        WriteSection(_imports);
+        WriteSection(_memoryModel);
+        WriteSection(_entryPoints);
+        WriteSection(_executionModes);
+        WriteSection(_debug);
+        WriteSection(_annotations);
+        WriteSection(_typesConstantsGlobals);
+        WriteSection(_functions);
+        var bytes = new byte[wordCount * sizeof(uint)];
+        Buffer.BlockCopy(words, 0, bytes, 0, bytes.Length);
+        return bytes;
+
+        void WriteWord(uint value)
+        {
+            words[offset++] = value;
+        }
+
+        void WriteSection(List<uint> section)
+        {
+            foreach (var value in section)
+            {
+                WriteWord(value);
+            }
+        }
+    }
+
+    private uint EmitResult(
+        List<uint> section,
+        SpirvOp opcode,
+        uint resultType,
+        params uint[] operands)
+    {
+        var result = AllocateId();
+        var values = new uint[operands.Length + 2];
+        values[0] = resultType;
+        values[1] = result;
+        operands.CopyTo(values, 2);
+        Emit(section, opcode, values);
+        return result;
+    }
+
+    private static void Emit(List<uint> section, SpirvOp opcode, params uint[] operands) =>
+        EmitRaw(section, (ushort)opcode, operands);
+
+    private static void EmitRaw(List<uint> section, ushort opcode, params uint[] operands)
+    {
+        section.Add(((uint)(operands.Length + 1) << 16) | opcode);
+        section.AddRange(operands);
+    }
+
+    private static void EmitWithString(
+        List<uint> section,
+        SpirvOp opcode,
+        IReadOnlyList<uint> prefix,
+        string value,
+        int stringBeforeTailCount = -1)
+    {
+        var encoded = EncodeString(value);
+        if (stringBeforeTailCount < 0)
+        {
+            var operands = new uint[prefix.Count + encoded.Length];
+            for (var index = 0; index < prefix.Count; index++)
+            {
+                operands[index] = prefix[index];
+            }
+
+            encoded.CopyTo(operands, prefix.Count);
+            Emit(section, opcode, operands);
+            return;
+        }
+
+        var result = new uint[prefix.Count + encoded.Length];
+        for (var index = 0; index < stringBeforeTailCount; index++)
+        {
+            result[index] = prefix[index];
+        }
+
+        encoded.CopyTo(result, stringBeforeTailCount);
+        for (var index = stringBeforeTailCount; index < prefix.Count; index++)
+        {
+            result[index + encoded.Length] = prefix[index];
+        }
+
+        Emit(section, opcode, result);
+    }
+
+    private static uint[] EncodeString(string value)
+    {
+        var byteCount = Encoding.UTF8.GetByteCount(value) + 1;
+        var words = new uint[(byteCount + 3) / 4];
+        Encoding.UTF8.GetBytes(value, System.Runtime.InteropServices.MemoryMarshal.AsBytes(words.AsSpan()));
+        return words;
+    }
+}

src/SharpEmu.Libs/Ampr/AmprExports.cs

@@ -2,6 +2,8 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Libs.Kernel;
+using System.Buffers;
 using System.Buffers.Binary;
 using System.Collections.Concurrent;
 
@@ -17,9 +19,16 @@ public static class AmprExports
     private const ulong CommandBufferAux1Offset = 0x20;
     private const ulong ReadFileRecordSize = 0x30;
     private const ulong KernelEventQueueRecordSize = 0x30;
+    private const ulong WriteAddressRecordSize = 0x20;
     private const uint ReadFileRecordType = 1;
     private const uint KernelEventQueueRecordType = 2;
+    private const uint WriteAddressRecordType = 3;
     private static readonly ConcurrentDictionary<ulong, CommandBufferState> _commandBuffers = new();
+    private static readonly bool _traceAmpr =
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal);
+    private static readonly bool _traceAmprReads =
+        _traceAmpr ||
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR_READS"), "1", StringComparison.Ordinal);
 
     private sealed class CommandBufferState
     {
@@ -72,12 +81,7 @@ public static class AmprExports
             return (int)OrbisGen2Result.ORBIS_GEN2_OK;
         }
 
-        var buffer = 0UL;
-        var size = 0UL;
-        _ = ctx.TryReadUInt64(commandBuffer + CommandBufferDataOffset, out buffer);
-        _ = ctx.TryReadUInt64(commandBuffer + CommandBufferSizeOffset, out size);
-
-        if (!InitializeCommandBuffer(ctx, commandBuffer, buffer, size, aux0, aux1, clear: false))
+        if (!InitializeCommandBuffer(ctx, commandBuffer, buffer: 0, size: 0, aux0, aux1, clear: false))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
@@ -101,8 +105,9 @@ public static class AmprExports
             return (int)OrbisGen2Result.ORBIS_GEN2_OK;
         }
 
-        if (!ctx.TryWriteUInt64(commandBuffer + CommandBufferAux0Offset, 0) ||
-            !ctx.TryWriteUInt64(commandBuffer + CommandBufferAux1Offset, 0))
+        Span<byte> auxiliaryPointers = stackalloc byte[sizeof(ulong) * 2];
+        auxiliaryPointers.Clear();
+        if (!ctx.Memory.TryWrite(commandBuffer + CommandBufferAux0Offset, auxiliaryPointers))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
@@ -176,8 +181,15 @@ public static class AmprExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        if (!ctx.TryReadUInt64(commandBuffer + CommandBufferDataOffset, out var buffer) ||
-            !ctx.TryReadUInt64(commandBuffer + CommandBufferSizeOffset, out var size) ||
+        Span<byte> bufferPointers = stackalloc byte[sizeof(ulong) * 2];
+        if (!ctx.Memory.TryRead(commandBuffer + CommandBufferDataOffset, bufferPointers))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        var buffer = BinaryPrimitives.ReadUInt64LittleEndian(bufferPointers);
+        var size = BinaryPrimitives.ReadUInt64LittleEndian(bufferPointers[sizeof(ulong)..]);
+        if (
             !WriteCommandBufferPointers(ctx, commandBuffer, buffer, size, writeOffset: 0))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
@@ -284,6 +296,18 @@ public static class AmprExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "C+IEj+BsAFM",
+        ExportName = "sceAmprMeasureCommandSizeWriteAddressOnCompletion",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int MeasureCommandSizeWriteAddressOnCompletion(CpuContext ctx)
+    {
+        TraceAmpr(ctx, "measure_write_address_complete", 0, WriteAddressRecordSize, 0);
+        ctx[CpuRegister.Rax] = WriteAddressRecordSize;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "tZDDEo2tE5k",
         ExportName = "sceAmprCommandBufferGetSize",
@@ -340,28 +364,116 @@ public static class AmprExports
         var commandBuffer = ctx[CpuRegister.Rdi];
         var equeue = ctx[CpuRegister.Rsi];
         var ident = ctx[CpuRegister.Rdx];
-        var filter = ctx[CpuRegister.Rcx];
+        var completionToken = ctx[CpuRegister.Rcx];
         var userData = ctx[CpuRegister.R8];
-        var data = ctx[CpuRegister.R9];
 
         if (commandBuffer == 0)
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        var extra = 0UL;
-        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp] + sizeof(ulong), out extra);
-
-        if (!AppendKernelEventQueueRecord(ctx, commandBuffer, equeue, ident, filter, userData, data, extra))
+        if (!AppendKernelEventQueueRecord(
+                ctx,
+                commandBuffer,
+                equeue,
+                ident,
+                completionToken,
+                userData))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        TraceAmpr(ctx, "write_equeue", commandBuffer, equeue, ident);
+        TraceAmpr(ctx, "write_equeue", commandBuffer, ident, completionToken);
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "sJXyWHjP-F8",
+        ExportName = "sceAmprCommandBufferWriteAddressOnCompletion",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferWriteAddressOnCompletion(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var address = ctx[CpuRegister.Rsi];
+        var value = ctx[CpuRegister.Rdx];
+
+        if (commandBuffer == 0 || address == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!AppendWriteAddressRecord(ctx, commandBuffer, address, value))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "write_address_complete", commandBuffer, address, value);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    public static int CompleteCommandBuffer(CpuContext ctx, ulong commandBuffer)
+    {
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!TryGetCommandBufferState(ctx, commandBuffer, out var buffer, out _, out var state) || state is null)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ulong writeOffset;
+        lock (state)
+        {
+            writeOffset = state.WriteOffset;
+        }
+
+        var offset = 0UL;
+        while (offset < writeOffset)
+        {
+            if (!TryReadUInt32(ctx, buffer + offset, out var recordType))
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+
+            switch (recordType)
+            {
+                case ReadFileRecordType:
+                    offset += ReadFileRecordSize;
+                    break;
+
+                case KernelEventQueueRecordType:
+                    if (!CompleteKernelEventQueueRecord(ctx, buffer + offset))
+                    {
+                        return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+                    }
+
+                    offset += KernelEventQueueRecordSize;
+                    break;
+
+                case WriteAddressRecordType:
+                    if (!CompleteWriteAddressRecord(ctx, buffer + offset))
+                    {
+                        return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+                    }
+
+                    offset += WriteAddressRecordSize;
+                    break;
+
+                default:
+                    TraceAmpr(ctx, "complete_unknown", commandBuffer, recordType, offset);
+                    return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+            }
+        }
+
+        TraceAmpr(ctx, "complete", commandBuffer, buffer, writeOffset);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     private static bool InitializeCommandBuffer(
         CpuContext ctx,
         ulong commandBuffer,
@@ -371,20 +483,34 @@ public static class AmprExports
         ulong aux1,
         bool clear)
     {
+        Span<byte> header = stackalloc byte[CommandBufferHeaderSize];
         if (clear)
         {
-            Span<byte> header = stackalloc byte[CommandBufferHeaderSize];
             header.Clear();
-            if (!ctx.Memory.TryWrite(commandBuffer, header))
+        }
+        else
+        {
+            if (!ctx.Memory.TryRead(commandBuffer, header))
             {
                 return false;
             }
+
+            buffer = BinaryPrimitives.ReadUInt64LittleEndian(header[(int)CommandBufferDataOffset..]);
+            size = BinaryPrimitives.ReadUInt64LittleEndian(header[(int)CommandBufferSizeOffset..]);
         }
 
-        return ctx.TryWriteUInt64(commandBuffer + CommandBufferSelfOffset, commandBuffer) &&
-               ctx.TryWriteUInt64(commandBuffer + CommandBufferAux0Offset, aux0) &&
-               ctx.TryWriteUInt64(commandBuffer + CommandBufferAux1Offset, aux1) &&
-               WriteCommandBufferPointers(ctx, commandBuffer, buffer, size, writeOffset: 0);
+        BinaryPrimitives.WriteUInt64LittleEndian(header[(int)CommandBufferSelfOffset..], commandBuffer);
+        BinaryPrimitives.WriteUInt64LittleEndian(header[(int)CommandBufferDataOffset..], buffer);
+        BinaryPrimitives.WriteUInt64LittleEndian(header[(int)CommandBufferSizeOffset..], size);
+        BinaryPrimitives.WriteUInt64LittleEndian(header[(int)CommandBufferAux0Offset..], aux0);
+        BinaryPrimitives.WriteUInt64LittleEndian(header[(int)CommandBufferAux1Offset..], aux1);
+        if (!ctx.Memory.TryWrite(commandBuffer, header))
+        {
+            return false;
+        }
+
+        UpdateCommandBufferState(commandBuffer, buffer, size, writeOffset: 0);
+        return true;
     }
 
     private static bool WriteCommandBufferPointers(CpuContext ctx, ulong commandBuffer, ulong buffer, ulong size)
@@ -399,6 +525,26 @@ public static class AmprExports
             return false;
         }
 
+        UpdateCommandBufferState(commandBuffer, buffer, size, writeOffset);
+
+        return true;
+    }
+
+    private static bool WriteVisibleCommandBufferPointers(CpuContext ctx, ulong commandBuffer, ulong buffer, ulong size)
+    {
+        Span<byte> pointers = stackalloc byte[sizeof(ulong) * 3];
+        BinaryPrimitives.WriteUInt64LittleEndian(pointers, commandBuffer);
+        BinaryPrimitives.WriteUInt64LittleEndian(pointers[sizeof(ulong)..], buffer);
+        BinaryPrimitives.WriteUInt64LittleEndian(pointers[(sizeof(ulong) * 2)..], size);
+        return ctx.Memory.TryWrite(commandBuffer + CommandBufferSelfOffset, pointers);
+    }
+
+    private static void UpdateCommandBufferState(
+        ulong commandBuffer,
+        ulong buffer,
+        ulong size,
+        ulong writeOffset)
+    {
         var state = _commandBuffers.GetOrAdd(commandBuffer, static _ => new CommandBufferState());
         lock (state)
         {
@@ -406,15 +552,6 @@ public static class AmprExports
             state.Size = size;
             state.WriteOffset = writeOffset;
         }
-
-        return true;
-    }
-
-    private static bool WriteVisibleCommandBufferPointers(CpuContext ctx, ulong commandBuffer, ulong buffer, ulong size)
-    {
-        return ctx.TryWriteUInt64(commandBuffer + CommandBufferSelfOffset, commandBuffer) &&
-               ctx.TryWriteUInt64(commandBuffer + CommandBufferDataOffset, buffer) &&
-               ctx.TryWriteUInt64(commandBuffer + CommandBufferSizeOffset, size);
     }
 
     private static bool TryGetCommandBufferState(
@@ -435,9 +572,11 @@ public static class AmprExports
             return true;
         }
 
-        if (ctx.TryReadUInt64(commandBuffer + CommandBufferDataOffset, out buffer) &&
-            ctx.TryReadUInt64(commandBuffer + CommandBufferSizeOffset, out size))
+        Span<byte> pointers = stackalloc byte[sizeof(ulong) * 2];
+        if (ctx.Memory.TryRead(commandBuffer + CommandBufferDataOffset, pointers))
         {
+            buffer = BinaryPrimitives.ReadUInt64LittleEndian(pointers);
+            size = BinaryPrimitives.ReadUInt64LittleEndian(pointers[sizeof(ulong)..]);
             state = _commandBuffers.GetOrAdd(commandBuffer, static _ => new CommandBufferState());
             lock (state)
             {
@@ -490,12 +629,18 @@ public static class AmprExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        const int ChunkSize = 64 * 1024;
-        var buffer = new byte[(int)Math.Min((ulong)ChunkSize, size)];
+        const int ChunkSize = 1024 * 1024;
+        var buffer = ArrayPool<byte>.Shared.Rent((int)Math.Min((ulong)ChunkSize, size));
 
         try
         {
-            using var stream = new FileStream(hostPath, FileMode.Open, FileAccess.Read, FileShare.ReadWrite | FileShare.Delete);
+            using var stream = new FileStream(
+                hostPath,
+                FileMode.Open,
+                FileAccess.Read,
+                FileShare.ReadWrite | FileShare.Delete,
+                ChunkSize,
+                FileOptions.SequentialScan);
             if (fileOffset >= (ulong)stream.Length)
             {
                 return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -528,6 +673,10 @@ public static class AmprExports
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
+        finally
+        {
+            ArrayPool<byte>.Shared.Return(buffer);
+        }
 
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -558,20 +707,28 @@ public static class AmprExports
         ulong commandBuffer,
         ulong equeue,
         ulong ident,
-        ulong filter,
-        ulong userData,
-        ulong data,
-        ulong extra)
+        ulong completionToken,
+        ulong userData)
     {
         Span<byte> record = stackalloc byte[(int)KernelEventQueueRecordSize];
         record.Clear();
         BinaryPrimitives.WriteUInt32LittleEndian(record[0x00..], KernelEventQueueRecordType);
-        BinaryPrimitives.WriteUInt32LittleEndian(record[0x04..], unchecked((uint)filter));
+        BinaryPrimitives.WriteInt16LittleEndian(record[0x04..], KernelEventQueueCompatExports.KernelEventFilterAmpr);
         BinaryPrimitives.WriteUInt64LittleEndian(record[0x08..], equeue);
         BinaryPrimitives.WriteUInt64LittleEndian(record[0x10..], ident);
         BinaryPrimitives.WriteUInt64LittleEndian(record[0x18..], userData);
-        BinaryPrimitives.WriteUInt64LittleEndian(record[0x20..], data);
-        BinaryPrimitives.WriteUInt64LittleEndian(record[0x28..], extra);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x20..], completionToken);
+
+        return AppendCommandBufferRecord(ctx, commandBuffer, record);
+    }
+
+    private static bool AppendWriteAddressRecord(CpuContext ctx, ulong commandBuffer, ulong address, ulong value)
+    {
+        Span<byte> record = stackalloc byte[(int)WriteAddressRecordSize];
+        record.Clear();
+        BinaryPrimitives.WriteUInt32LittleEndian(record[0x00..], WriteAddressRecordType);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x08..], address);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x10..], value);
 
         return AppendCommandBufferRecord(ctx, commandBuffer, record);
     }
@@ -604,9 +761,69 @@ public static class AmprExports
         return true;
     }
 
+    private static bool CompleteKernelEventQueueRecord(CpuContext ctx, ulong recordAddress)
+    {
+        Span<byte> record = stackalloc byte[(int)KernelEventQueueRecordSize];
+        if (!ctx.Memory.TryRead(recordAddress, record))
+        {
+            return false;
+        }
+
+        var filter = unchecked((short)BinaryPrimitives.ReadUInt32LittleEndian(record[0x04..]));
+        var equeue = BinaryPrimitives.ReadUInt64LittleEndian(record[0x08..]);
+        var ident = BinaryPrimitives.ReadUInt64LittleEndian(record[0x10..]);
+        var userData = BinaryPrimitives.ReadUInt64LittleEndian(record[0x18..]);
+        var data = BinaryPrimitives.ReadUInt64LittleEndian(record[0x20..]);
+        var extra = BinaryPrimitives.ReadUInt64LittleEndian(record[0x28..]);
+
+        var queuedEvent = new KernelEventQueueCompatExports.KernelQueuedEvent(
+            ident,
+            filter,
+            0x20,
+            unchecked((uint)extra),
+            data,
+            userData);
+
+        _ = KernelEventQueueCompatExports.EnqueueEvent(equeue, queuedEvent);
+        TraceAmpr(ctx, "complete_equeue", equeue, ident, data);
+        return true;
+    }
+
+    private static bool CompleteWriteAddressRecord(CpuContext ctx, ulong recordAddress)
+    {
+        Span<byte> record = stackalloc byte[(int)WriteAddressRecordSize];
+        if (!ctx.Memory.TryRead(recordAddress, record))
+        {
+            return false;
+        }
+
+        var address = BinaryPrimitives.ReadUInt64LittleEndian(record[0x08..]);
+        var value = BinaryPrimitives.ReadUInt64LittleEndian(record[0x10..]);
+        if (!ctx.TryWriteUInt64(address, value))
+        {
+            return false;
+        }
+
+        TraceAmpr(ctx, "complete_write_address", address, value, 0);
+        return true;
+    }
+
+    private static bool TryReadUInt32(CpuContext ctx, ulong address, out uint value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(uint)];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt32LittleEndian(buffer);
+        return true;
+    }
+
     private static void TraceAmpr(CpuContext ctx, string operation, ulong commandBuffer, ulong arg0, ulong arg1)
     {
-        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal))
+        if (!_traceAmpr)
         {
             return;
         }
@@ -628,7 +845,7 @@ public static class AmprExports
         string? hostPath,
         int result)
     {
-        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal))
+        if (!_traceAmprReads)
         {
             return;
         }

src/SharpEmu.Libs/AppContent/AppContentExports.cs

@@ -4,6 +4,7 @@
 using SharpEmu.HLE;
 using System.Buffers.Binary;
 using System.Text;
+using System.Text.Json;
 
 namespace SharpEmu.Libs.AppContent;
 
@@ -11,6 +12,8 @@ public static class AppContentExports
 {
     private const ulong BootParamAttrOffset = 4;
     private const string Temp0MountPoint = "/temp0";
+    private const uint AppParamSkuFlag = 0;
+    private const int AppParamSkuFlagFull = 3;
 
     [SysAbiExport(
         Nid = "R9lA82OraNs",
@@ -59,6 +62,41 @@ public static class AppContentExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "99b82IKXpH4",
+        ExportName = "sceAppContentAppParamGetInt",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceAppContent")]
+    public static int AppContentAppParamGetInt(CpuContext ctx)
+    {
+        var paramId = (uint)ctx[CpuRegister.Rdi];
+        var valueAddress = ctx[CpuRegister.Rsi];
+        if (valueAddress == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        int value;
+        if (paramId == AppParamSkuFlag)
+        {
+            value = AppParamSkuFlagFull;
+        }
+        else if (!TryReadUserDefinedParam(paramId, out value))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> valueBytes = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(valueBytes, value);
+        if (!ctx.Memory.TryWrite(valueAddress, valueBytes))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        TraceAppContent($"app_param_get_int id={paramId} value={value}");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
     [SysAbiExport(
         Nid = "buYbeLOGWmA",
         ExportName = "sceAppContentTemporaryDataMount2",
@@ -83,6 +121,69 @@ public static class AppContentExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    private static bool TryReadUserDefinedParam(uint paramId, out int value)
+    {
+        value = 0;
+        if (paramId is < 1 or > 4)
+        {
+            return false;
+        }
+
+        var app0Root = Environment.GetEnvironmentVariable("SHARPEMU_APP0_DIR");
+        if (string.IsNullOrWhiteSpace(app0Root))
+        {
+            return true;
+        }
+
+        var paramJsonPath = Path.Combine(app0Root, "sce_sys", "param.json");
+        if (!File.Exists(paramJsonPath))
+        {
+            return true;
+        }
+
+        try
+        {
+            using var stream = File.OpenRead(paramJsonPath);
+            using var document = JsonDocument.Parse(stream);
+            var propertyName = $"userDefinedParam{paramId}";
+            if (document.RootElement.TryGetProperty(propertyName, out var element) &&
+                element.TryGetInt32(out var parsedValue))
+            {
+                value = parsedValue;
+            }
+
+            return true;
+        }
+        catch (IOException)
+        {
+            return true;
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return true;
+        }
+        catch (JsonException)
+        {
+            return true;
+        }
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+        return (int)result;
+    }
+
+    private static void TraceAppContent(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_APP_CONTENT"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] app_content.{message}");
+    }
+
     private static string ResolveTemp0Root()
     {
         const string temp0VariableName = "SHARPEMU_TEMP0_DIR";

src/SharpEmu.Libs/Audio/AudioOut2Exports.cs

@@ -0,0 +1,231 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Threading;
+
+namespace SharpEmu.Libs.Audio;
+
+public static class AudioOut2Exports
+{
+    private const int AudioOut2ContextParamSize = 0x80;
+    private const int AudioOut2ContextMemorySize = 0x10000;
+    private const int AudioOut2ContextMemoryAlignment = 0x10000;
+    private static long _nextContextHandle = 1;
+    private static long _nextUserHandle = 1;
+    private static int _nextPortId;
+
+    [SysAbiExport(
+        Nid = "g2tViFIohHE",
+        ExportName = "sceAudioOut2Initialize",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2Initialize(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "t5YrizufpQc",
+        ExportName = "sceAudioOut2ContextResetParam",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2ContextResetParam(CpuContext ctx)
+    {
+        var paramAddress = ctx[CpuRegister.Rdi];
+        if (paramAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> param = stackalloc byte[AudioOut2ContextParamSize];
+        param.Clear();
+        BinaryPrimitives.WriteUInt32LittleEndian(param[0x00..], AudioOut2ContextParamSize);
+        BinaryPrimitives.WriteUInt32LittleEndian(param[0x04..], 2);
+        BinaryPrimitives.WriteUInt32LittleEndian(param[0x08..], 48000);
+        BinaryPrimitives.WriteUInt32LittleEndian(param[0x0C..], 0x400);
+
+        return ctx.Memory.TryWrite(paramAddress, param)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "pDmme7Bgm6E",
+        ExportName = "sceAudioOut2ContextQueryMemory",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2ContextQueryMemory(CpuContext ctx)
+    {
+        var paramAddress = ctx[CpuRegister.Rdi];
+        var memoryInfoAddress = ctx[CpuRegister.Rsi];
+        if (paramAddress == 0 || memoryInfoAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> memoryInfo = stackalloc byte[0x20];
+        memoryInfo.Clear();
+        BinaryPrimitives.WriteUInt64LittleEndian(memoryInfo[0x00..], AudioOut2ContextMemorySize);
+        BinaryPrimitives.WriteUInt64LittleEndian(memoryInfo[0x08..], AudioOut2ContextMemoryAlignment);
+        BinaryPrimitives.WriteUInt64LittleEndian(memoryInfo[0x10..], AudioOut2ContextMemorySize);
+        BinaryPrimitives.WriteUInt64LittleEndian(memoryInfo[0x18..], AudioOut2ContextMemoryAlignment);
+
+        return ctx.Memory.TryWrite(memoryInfoAddress, memoryInfo)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "0x6o1VVAYSY",
+        ExportName = "sceAudioOut2ContextCreate",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2ContextCreate(CpuContext ctx)
+    {
+        var paramAddress = ctx[CpuRegister.Rdi];
+        var memoryAddress = ctx[CpuRegister.Rsi];
+        var memorySize = ctx[CpuRegister.Rdx];
+        var outContextAddress = ctx[CpuRegister.Rcx];
+        if (paramAddress == 0 || memoryAddress == 0 || memorySize == 0 || outContextAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        var handle = (ulong)Interlocked.Increment(ref _nextContextHandle);
+        return TryWriteUInt64(ctx, outContextAddress, handle)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "on6ZH7Abo10",
+        ExportName = "sceAudioOut2ContextDestroy",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2ContextDestroy(CpuContext ctx) => SetReturn(ctx, 0);
+
+    [SysAbiExport(
+        Nid = "JK2wamZPzwM",
+        ExportName = "sceAudioOut2PortCreate",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2PortCreate(CpuContext ctx)
+    {
+        var type = unchecked((int)ctx[CpuRegister.Rdi]);
+        var paramAddress = ctx[CpuRegister.Rsi];
+        var outPortAddress = ctx[CpuRegister.Rdx];
+        var contextAddress = ctx[CpuRegister.Rcx];
+        if (type < 0 || type > 255 || paramAddress == 0 || outPortAddress == 0 || contextAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        var portId = unchecked((uint)Interlocked.Increment(ref _nextPortId)) & 0xFF;
+        var handle = 0x2000_0000UL | ((ulong)(uint)type << 16) | portId;
+        return TryWriteUInt64(ctx, outPortAddress, handle)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "gatEUKG+Ea4",
+        ExportName = "sceAudioOut2PortGetState",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2PortGetState(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var stateAddress = ctx[CpuRegister.Rsi];
+        if (handle == 0 || stateAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        var type = (int)((handle >> 16) & 0xFF);
+        Span<byte> state = stackalloc byte[0x20];
+        state.Clear();
+        var output = type == 2 ? 0x40 : 0x01;
+        var channels = type == 2 ? 1 : 2;
+        BinaryPrimitives.WriteUInt16LittleEndian(state[0x00..], unchecked((ushort)output));
+        state[0x02] = unchecked((byte)channels);
+        BinaryPrimitives.WriteInt16LittleEndian(state[0x04..], -1);
+
+        return ctx.Memory.TryWrite(stateAddress, state)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "DImz2Ft9E2g",
+        ExportName = "sceAudioOut2GetSpeakerInfo",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2GetSpeakerInfo(CpuContext ctx)
+    {
+        var infoAddress = ctx[CpuRegister.Rdi];
+        if (infoAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> info = stackalloc byte[0x40];
+        info.Clear();
+        BinaryPrimitives.WriteUInt32LittleEndian(info[0x00..], 1);
+        BinaryPrimitives.WriteUInt32LittleEndian(info[0x04..], 2);
+        BinaryPrimitives.WriteUInt32LittleEndian(info[0x08..], 48000);
+
+        return ctx.Memory.TryWrite(infoAddress, info)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "cd+Rtw+D1x8",
+        ExportName = "sceAudioOut2PortDestroy",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2PortDestroy(CpuContext ctx) => SetReturn(ctx, 0);
+
+    [SysAbiExport(
+        Nid = "IaZXJ9M79uo",
+        ExportName = "sceAudioOut2UserDestroy",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2UserDestroy(CpuContext ctx) => SetReturn(ctx, 0);
+
+    [SysAbiExport(
+        Nid = "xywYcRB7nbQ",
+        ExportName = "sceAudioOut2UserCreate",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAudioOut2")]
+    public static int AudioOut2UserCreate(CpuContext ctx)
+    {
+        var userId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var outUserAddress = ctx[CpuRegister.Rsi];
+        if ((userId != 0 && userId != 1 && userId != 255) || outUserAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        var handle = (ulong)Interlocked.Increment(ref _nextUserHandle);
+        return TryWriteUInt64(ctx, outUserAddress, handle)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static bool TryWriteUInt64(CpuContext ctx, ulong address, ulong value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(ulong)];
+        BinaryPrimitives.WriteUInt64LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+}

src/SharpEmu.Libs/CommonDialog/CommonDialogExports.cs

@@ -0,0 +1,38 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.CommonDialog;
+
+public static class CommonDialogExports
+{
+    private const int AlreadySystemInitialized = unchecked((int)0x80B80002);
+    private static int _initialized;
+
+    [SysAbiExport(
+        Nid = "uoUpLGNkygk",
+        ExportName = "sceCommonDialogInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceCommonDialog")]
+    public static int CommonDialogInitialize(CpuContext ctx)
+    {
+        var result = Interlocked.Exchange(ref _initialized, 1) == 0
+            ? 0
+            : AlreadySystemInitialized;
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    [SysAbiExport(
+        Nid = "BQ3tey0JmQM",
+        ExportName = "sceCommonDialogIsUsed",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceCommonDialog")]
+    public static int CommonDialogIsUsed(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+}

src/SharpEmu.Libs/GameUpdate/GameUpdateExports.cs

@@ -0,0 +1,24 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.GameUpdate;
+
+public static class GameUpdateExports
+{
+    private static int _initialized;
+
+    [SysAbiExport(
+        Nid = "YJtKLttI9fM",
+        ExportName = "sceGameUpdateInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceGameUpdate")]
+    public static int GameUpdateInitialize(CpuContext ctx)
+    {
+        Interlocked.Exchange(ref _initialized, 1);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+}

src/SharpEmu.Libs/Json/JsonExports.cs

@@ -0,0 +1,91 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Json;
+
+public static class JsonExports
+{
+    [SysAbiExport(
+        Nid = "-hJRce8wn1U",
+        ExportName = "_ZN3sce4Json12MemAllocatorC2Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson")]
+    public static int MemAllocatorConstructor(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("MemAllocator.ctor", thisAddress, 0);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "OcAgPxcq5Vk",
+        ExportName = "_ZN3sce4Json12MemAllocatorD2Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson")]
+    public static int MemAllocatorDestructor(CpuContext ctx)
+    {
+        TraceJson("MemAllocator.dtor", ctx[CpuRegister.Rdi], 0);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "cK6bYHf-Q5E",
+        ExportName = "_ZN3sce4Json11InitializerC1Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson")]
+    public static int InitializerConstructor(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        TraceJson("Initializer.ctor", thisAddress, 0);
+        ctx[CpuRegister.Rax] = thisAddress;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "RujUxbr3haM",
+        ExportName = "_ZN3sce4Json11InitializerD1Ev",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson")]
+    public static int InitializerDestructor(CpuContext ctx)
+    {
+        TraceJson("Initializer.dtor", ctx[CpuRegister.Rdi], 0);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "Cxwy7wHq4J0",
+        ExportName = "_ZN3sce4Json11Initializer10initializeEPKNS0_13InitParameterE",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceJson")]
+    public static int InitializerInitialize(CpuContext ctx)
+    {
+        var thisAddress = ctx[CpuRegister.Rdi];
+        var initParameterAddress = ctx[CpuRegister.Rsi];
+        if (thisAddress == 0)
+        {
+            ctx[CpuRegister.Rax] = unchecked((ulong)(int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        TraceJson("Initializer.initialize", thisAddress, initParameterAddress);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static void TraceJson(string operation, ulong thisAddress, ulong argument)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_JSON"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] json.{operation} this=0x{thisAddress:X16} arg=0x{argument:X16}");
+    }
+}

src/SharpEmu.Libs/Kernel/KernelAprCompatExports.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Libs.Ampr;
 using System.Buffers.Binary;
 using System.Collections.Concurrent;
 using System.Threading;
@@ -10,8 +11,13 @@ namespace SharpEmu.Libs.Kernel;
 
 public static class KernelAprCompatExports
 {
-    private static readonly ConcurrentDictionary<uint, ulong> _submittedCommandBuffers = new();
+    private static readonly ConcurrentDictionary<uint, AprSubmission> _submittedCommandBuffers = new();
     private static int _nextSubmissionId;
+    private static int _aprWaitTraceCount;
+    private static readonly bool _traceApr =
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal);
+
+    private readonly record struct AprSubmission(ulong CommandBuffer, ulong Priority, ulong ResultAddress);
 
     [SysAbiExport(
         Nid = "ASoW5WE-UPo",
@@ -36,7 +42,13 @@ public static class KernelAprCompatExports
             submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
         }
 
-        _submittedCommandBuffers[submissionId] = commandBuffer;
+        _submittedCommandBuffers[submissionId] = new AprSubmission(commandBuffer, priority, resultAddress);
+
+        var completionResult = AmprExports.CompleteCommandBuffer(ctx, commandBuffer);
+        if (completionResult != (int)OrbisGen2Result.ORBIS_GEN2_OK)
+        {
+            return completionResult;
+        }
 
         if (outSubmissionId != 0 && !TryWriteUInt32(ctx, outSubmissionId, submissionId))
         {
@@ -60,20 +72,23 @@ public static class KernelAprCompatExports
     public static int KernelAprWaitCommandBuffer(CpuContext ctx)
     {
         var submissionId = unchecked((uint)ctx[CpuRegister.Rdi]);
-        var priority = ctx[CpuRegister.Rsi];
-        var resultAddress = ctx[CpuRegister.Rdx];
+        var waitArg1 = ctx[CpuRegister.Rsi];
+        var waitArg2 = ctx[CpuRegister.Rdx];
 
-        if (!_submittedCommandBuffers.TryRemove(submissionId, out var commandBuffer))
+        if (!_submittedCommandBuffers.TryRemove(submissionId, out var submission))
         {
+            TraceAprWaitFailure(ctx, "wait_missing", submissionId, commandBuffer: 0, waitArg1, waitArg2);
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
 
+        var resultAddress = ResolveWaitResultAddress(waitArg1, waitArg2, submission.ResultAddress);
         if (resultAddress != 0 && !TryWriteAprResult(ctx, resultAddress))
         {
+            TraceAprWaitFailure(ctx, "wait_result_fault", submissionId, submission.CommandBuffer, waitArg1, waitArg2);
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        TraceApr(ctx, "wait", submissionId, commandBuffer, priority, resultAddress);
+        TraceApr(ctx, "wait", submissionId, submission.CommandBuffer, waitArg1, resultAddress);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -91,7 +106,13 @@ public static class KernelAprCompatExports
         }
 
         var submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
-        _submittedCommandBuffers[submissionId] = commandBuffer;
+        _submittedCommandBuffers[submissionId] = new AprSubmission(commandBuffer, ctx[CpuRegister.Rsi], ResultAddress: 0);
+
+        var completionResult = AmprExports.CompleteCommandBuffer(ctx, commandBuffer);
+        if (completionResult != (int)OrbisGen2Result.ORBIS_GEN2_OK)
+        {
+            return completionResult;
+        }
 
         TraceApr(ctx, "submit", submissionId, commandBuffer, ctx[CpuRegister.Rsi], 0);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -112,7 +133,13 @@ public static class KernelAprCompatExports
         }
 
         var submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
-        _submittedCommandBuffers[submissionId] = commandBuffer;
+        _submittedCommandBuffers[submissionId] = new AprSubmission(commandBuffer, ctx[CpuRegister.Rsi], ResultAddress: 0);
+
+        var completionResult = AmprExports.CompleteCommandBuffer(ctx, commandBuffer);
+        if (completionResult != (int)OrbisGen2Result.ORBIS_GEN2_OK)
+        {
+            return completionResult;
+        }
 
         if (!TryWriteUInt32(ctx, outSubmissionId, submissionId))
         {
@@ -130,6 +157,27 @@ public static class KernelAprCompatExports
         return ctx.Memory.TryWrite(resultAddress, result);
     }
 
+    private static ulong ResolveWaitResultAddress(ulong waitArg1, ulong waitArg2, ulong submittedResultAddress)
+    {
+        if (waitArg2 == 0)
+        {
+            return submittedResultAddress;
+        }
+
+        if (IsAmprCompletionToken(waitArg1) && waitArg2 <= 0xFFFF)
+        {
+            return submittedResultAddress;
+        }
+
+        return waitArg2;
+    }
+
+    private static bool IsAmprCompletionToken(ulong value)
+    {
+        var tag = value >> 56;
+        return tag is 0x0C or 0x10;
+    }
+
     private static bool TryWriteUInt32(CpuContext ctx, ulong address, uint value)
     {
         Span<byte> buffer = stackalloc byte[sizeof(uint)];
@@ -145,7 +193,7 @@ public static class KernelAprCompatExports
         ulong priority,
         ulong aux)
     {
-        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal))
+        if (!_traceApr)
         {
             return;
         }
@@ -154,5 +202,54 @@ public static class KernelAprCompatExports
         _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out returnRip);
         Console.Error.WriteLine(
             $"[LOADER][TRACE] apr.{operation}: id=0x{submissionId:X8} cmd=0x{commandBuffer:X16} priority=0x{priority:X16} aux=0x{aux:X16} ret=0x{returnRip:X16}");
+        if (aux != 0 &&
+            ctx.TryReadUInt64(aux, out var result0) &&
+            ctx.TryReadUInt64(aux + sizeof(ulong), out var result1))
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][TRACE] apr.{operation}.result: addr=0x{aux:X16} q0=0x{result0:X16} q1=0x{result1:X16}");
+        }
+    }
+
+    private static void TraceAprWaitFailure(
+        CpuContext ctx,
+        string operation,
+        uint submissionId,
+        ulong commandBuffer,
+        ulong priority,
+        ulong resultAddress)
+    {
+        if (!_traceApr)
+        {
+            return;
+        }
+
+        var traceCount = Interlocked.Increment(ref _aprWaitTraceCount);
+        if (traceCount > 32 && (traceCount & 0x3FF) != 0)
+        {
+            return;
+        }
+
+        var returnRip = 0UL;
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] apr.{operation}: id=0x{submissionId:X8} cmd=0x{commandBuffer:X16} " +
+            $"rsi=0x{priority:X16} rdx=0x{resultAddress:X16} rcx=0x{ctx[CpuRegister.Rcx]:X16} " +
+            $"r8=0x{ctx[CpuRegister.R8]:X16} r9=0x{ctx[CpuRegister.R9]:X16} ret=0x{returnRip:X16}");
+        TraceReadableQword(ctx, operation, "rsi", priority);
+        TraceReadableQword(ctx, operation, "rdx", resultAddress);
+        TraceReadableQword(ctx, operation, "rcx", ctx[CpuRegister.Rcx]);
+        TraceReadableQword(ctx, operation, "r8", ctx[CpuRegister.R8]);
+    }
+
+    private static void TraceReadableQword(CpuContext ctx, string operation, string name, ulong address)
+    {
+        if (address == 0 || !ctx.TryReadUInt64(address, out var value))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] apr.{operation}.{name}: addr=0x{address:X16} q0=0x{value:X16}");
     }
 }

src/SharpEmu.Libs/Kernel/KernelEventFlagCompatExports.cs

@@ -0,0 +1,638 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Buffers.Binary;
+using System.Collections.Concurrent;
+using System.Text;
+using SharpEmu.HLE;
+using SharpEmu.Libs.Fiber;
+
+namespace SharpEmu.Libs.Kernel;
+
+public static class KernelEventFlagCompatExports
+{
+    private const int MaxEventFlagNameLength = 31;
+    private const int HostWaitPumpMilliseconds = 1;
+    private const uint AttrThreadFifo = 0x01;
+    private const uint AttrThreadPriority = 0x02;
+    private const uint AttrSingle = 0x10;
+    private const uint AttrMulti = 0x20;
+    private const uint WaitAnd = 0x01;
+    private const uint WaitOr = 0x02;
+    private const uint ClearAll = 0x10;
+    private const uint ClearPattern = 0x20;
+
+    private static readonly ConcurrentDictionary<ulong, EventFlagState> _eventFlags = new();
+    private static long _nextEventFlagHandle = 1;
+
+    private sealed class EventFlagState
+    {
+        public required string Name { get; init; }
+        public required uint Attributes { get; init; }
+        public ulong Bits { get; set; }
+        public int WaitingThreads { get; set; }
+        public object Gate { get; } = new();
+    }
+
+    [SysAbiExport(
+        Nid = "BpFoboUJoZU",
+        ExportName = "sceKernelCreateEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelCreateEventFlag(CpuContext ctx)
+    {
+        var outAddress = ctx[CpuRegister.Rdi];
+        var nameAddress = ctx[CpuRegister.Rsi];
+        var attributes = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var initialPattern = ctx[CpuRegister.Rcx];
+        var optionAddress = ctx[CpuRegister.R8];
+
+        if (outAddress == 0 ||
+            nameAddress == 0 ||
+            optionAddress != 0 ||
+            !IsValidAttributes(attributes))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        if (!TryReadNullTerminatedUtf8(ctx, nameAddress, MaxEventFlagNameLength + 1, out var name))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        if (Encoding.UTF8.GetByteCount(name) > MaxEventFlagNameLength)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        var handle = unchecked((ulong)Interlocked.Increment(ref _nextEventFlagHandle));
+        _eventFlags[handle] = new EventFlagState
+        {
+            Name = name,
+            Attributes = attributes,
+            Bits = initialPattern,
+        };
+
+        if (!ctx.TryWriteUInt64(outAddress, handle))
+        {
+            _eventFlags.TryRemove(handle, out _);
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        TraceEventFlag($"create handle=0x{handle:X16} name='{name}' attr=0x{attributes:X2} bits=0x{initialPattern:X16}");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "8mql9OcQnd4",
+        ExportName = "sceKernelDeleteEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        if (!_eventFlags.TryRemove(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        lock (state.Gate)
+        {
+            Monitor.PulseAll(state.Gate);
+        }
+
+        TraceEventFlag($"delete handle=0x{handle:X16} name='{state.Name}'");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "IOnSvHzqu6A",
+        ExportName = "sceKernelSetEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelSetEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var pattern = ctx[CpuRegister.Rsi];
+        var returnRip = GetCurrentReturnRip();
+        if (!_eventFlags.TryGetValue(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        lock (state.Gate)
+        {
+            state.Bits |= pattern;
+            Monitor.PulseAll(state.Gate);
+            TraceEventFlag($"set handle=0x{handle:X16} pattern=0x{pattern:X16} bits=0x{state.Bits:X16} ret=0x{returnRip:X16}");
+        }
+
+        _ = GuestThreadExecution.Scheduler?.WakeBlockedThreads(GetEventFlagWakeKey(handle));
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "7uhBFWRAS60",
+        ExportName = "sceKernelClearEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelClearEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var pattern = ctx[CpuRegister.Rsi];
+        if (!_eventFlags.TryGetValue(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        lock (state.Gate)
+        {
+            state.Bits &= pattern;
+            TraceEventFlag($"clear handle=0x{handle:X16} mask=0x{pattern:X16} bits=0x{state.Bits:X16}");
+        }
+
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "9lvj5DjHZiA",
+        ExportName = "sceKernelPollEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelPollEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var pattern = ctx[CpuRegister.Rsi];
+        var waitMode = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var resultAddress = ctx[CpuRegister.Rcx];
+
+        if (!_eventFlags.TryGetValue(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (pattern == 0 || !IsValidWaitMode(waitMode))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        lock (state.Gate)
+        {
+            if (!TryWriteResultPattern(ctx, resultAddress, state.Bits))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            if (!IsSatisfied(state.Bits, pattern, waitMode))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
+            }
+
+            ApplyClearMode(state, pattern, waitMode);
+            TraceEventFlag($"poll handle=0x{handle:X16} pattern=0x{pattern:X16} mode=0x{waitMode:X2} bits=0x{state.Bits:X16}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "JTvBflhYazQ",
+        ExportName = "sceKernelWaitEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelWaitEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var pattern = ctx[CpuRegister.Rsi];
+        var waitMode = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var resultAddress = ctx[CpuRegister.Rcx];
+        var timeoutAddress = ctx[CpuRegister.R8];
+        var returnRip = GetCurrentReturnRip();
+
+        if (!_eventFlags.TryGetValue(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (pattern == 0 || !IsValidWaitMode(waitMode))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        uint timeoutUsec = 0;
+        if (timeoutAddress != 0 && !TryReadUInt32(ctx, timeoutAddress, out timeoutUsec))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        Monitor.Enter(state.Gate);
+        try
+        {
+            if (TryCompleteSatisfiedWait(ctx, state, pattern, waitMode, resultAddress, out var immediateWaitResult))
+            {
+                return SetReturn(ctx, immediateWaitResult);
+            }
+
+            if (timeoutAddress != 0)
+            {
+                _ = TryWriteUInt32(ctx, timeoutAddress, 0);
+                _ = TryWriteResultPattern(ctx, resultAddress, state.Bits);
+                TraceEventFlag($"wait-timeout handle=0x{handle:X16} pattern=0x{pattern:X16} timeout={timeoutUsec} ret=0x{returnRip:X16}");
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TIMED_OUT);
+            }
+
+            var currentGuestThread = GuestThreadExecution.CurrentGuestThreadHandle;
+            var currentFiber = FiberExports.GetCurrentFiberAddressForDiagnostics(ctx);
+            var managedThread = Environment.CurrentManagedThreadId;
+            var blockedWaitResult = OrbisGen2Result.ORBIS_GEN2_OK;
+            var requestedBlock = GuestThreadExecution.RequestCurrentThreadBlock(
+                ctx,
+                "sceKernelWaitEventFlag",
+                GetEventFlagWakeKey(handle),
+                () => (int)blockedWaitResult,
+                () =>
+                {
+                    if (!TryPrepareBlockedWait(
+                            ctx,
+                            state,
+                            pattern,
+                            waitMode,
+                            resultAddress,
+                            out var preparedResult))
+                    {
+                        return false;
+                    }
+
+                    blockedWaitResult = preparedResult;
+                    return true;
+                });
+            TraceEventFlag($"wait-unsatisfied handle=0x{handle:X16} pattern=0x{pattern:X16} bits=0x{state.Bits:X16} guest_thread=0x{currentGuestThread:X16} fiber=0x{currentFiber:X16} managed={managedThread} block={requestedBlock} ret=0x{returnRip:X16} frames={FormatFrameChain(ctx)}");
+            TraceEventFlag($"wait-object handle=0x{handle:X16} name='{state.Name}' {FormatGuestWaitObject(ctx)}");
+            if (!requestedBlock)
+            {
+                var scheduler = GuestThreadExecution.Scheduler;
+                if (scheduler is null)
+                {
+                    return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN);
+                }
+
+                state.WaitingThreads++;
+                TraceEventFlag($"wait-pump handle=0x{handle:X16} pattern=0x{pattern:X16} waiters={state.WaitingThreads} guest_thread=0x{currentGuestThread:X16} fiber=0x{currentFiber:X16} managed={managedThread} ret=0x{returnRip:X16}");
+                var releaseWaiter = true;
+                try
+                {
+                    while (true)
+                    {
+                        Monitor.Exit(state.Gate);
+                        try
+                        {
+                            scheduler.Pump(ctx, "sceKernelWaitEventFlag");
+                        }
+                        finally
+                        {
+                            Monitor.Enter(state.Gate);
+                        }
+
+                        if (TryCompleteSatisfiedWait(ctx, state, pattern, waitMode, resultAddress, out var pumpedWaitResult))
+                        {
+                            state.WaitingThreads = Math.Max(0, state.WaitingThreads - 1);
+                            releaseWaiter = false;
+                            TraceEventFlag($"wait-wake handle=0x{handle:X16} pattern=0x{pattern:X16} bits=0x{state.Bits:X16} waiters={state.WaitingThreads} ret=0x{returnRip:X16}");
+                            return SetReturn(ctx, pumpedWaitResult);
+                        }
+
+                        Monitor.Wait(state.Gate, HostWaitPumpMilliseconds);
+                    }
+                }
+                finally
+                {
+                    if (releaseWaiter)
+                    {
+                        state.WaitingThreads = Math.Max(0, state.WaitingThreads - 1);
+                    }
+                }
+            }
+
+            state.WaitingThreads++;
+            TraceEventFlag($"wait-block handle=0x{handle:X16} pattern=0x{pattern:X16} waiters={state.WaitingThreads} guest_thread=0x{currentGuestThread:X16} fiber=0x{currentFiber:X16} managed={managedThread} ret=0x{returnRip:X16}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+        finally
+        {
+            Monitor.Exit(state.Gate);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "PZku4ZrXJqg",
+        ExportName = "sceKernelCancelEventFlag",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelCancelEventFlag(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var setPattern = ctx[CpuRegister.Rsi];
+        var waiterCountAddress = ctx[CpuRegister.Rdx];
+        if (!_eventFlags.TryGetValue(handle, out var state))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        lock (state.Gate)
+        {
+            if (waiterCountAddress != 0 &&
+                !TryWriteUInt32(ctx, waiterCountAddress, unchecked((uint)state.WaitingThreads)))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            state.Bits = setPattern;
+            state.WaitingThreads = 0;
+            Monitor.PulseAll(state.Gate);
+            TraceEventFlag($"cancel handle=0x{handle:X16} bits=0x{setPattern:X16}");
+        }
+
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    private static bool IsValidAttributes(uint attributes)
+    {
+        var queueMode = attributes & 0x0F;
+        var threadMode = attributes & 0xF0;
+        return (queueMode is 0 or AttrThreadFifo or AttrThreadPriority) &&
+            (threadMode is 0 or AttrSingle or AttrMulti) &&
+            (attributes & ~0x33u) == 0;
+    }
+
+    private static bool IsValidWaitMode(uint waitMode)
+    {
+        var condition = waitMode & 0x0F;
+        var clearMode = waitMode & 0xF0;
+        return condition is WaitAnd or WaitOr &&
+            clearMode is 0 or ClearAll or ClearPattern &&
+            (waitMode & ~0x33u) == 0;
+    }
+
+    private static bool IsSatisfied(ulong bits, ulong pattern, uint waitMode) =>
+        (waitMode & 0x0F) == WaitAnd
+            ? (bits & pattern) == pattern
+            : (bits & pattern) != 0;
+
+    private static void ApplyClearMode(EventFlagState state, ulong pattern, uint waitMode)
+    {
+        switch (waitMode & 0xF0)
+        {
+            case ClearAll:
+                state.Bits = 0;
+                break;
+            case ClearPattern:
+                state.Bits &= ~pattern;
+                break;
+        }
+    }
+
+    private static bool TryCompleteSatisfiedWait(
+    CpuContext ctx,
+    EventFlagState state,
+    ulong pattern,
+    uint waitMode,
+    ulong resultAddress,
+    out OrbisGen2Result result)
+    {
+        result = OrbisGen2Result.ORBIS_GEN2_OK;
+
+        if (!IsSatisfied(state.Bits, pattern, waitMode))
+        {
+            return false;
+        }
+
+        if (!TryWriteResultPattern(ctx, resultAddress, state.Bits))
+        {
+            result = OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            return true;
+        }
+
+        ApplyClearMode(state, pattern, waitMode);
+        return true;
+    }
+
+    private static bool TryPrepareBlockedWait(
+        CpuContext ctx,
+        EventFlagState state,
+        ulong pattern,
+        uint waitMode,
+        ulong resultAddress,
+        out OrbisGen2Result result)
+    {
+        lock (state.Gate)
+        {
+            result = OrbisGen2Result.ORBIS_GEN2_OK;
+            if (!IsSatisfied(state.Bits, pattern, waitMode))
+            {
+                return false;
+            }
+
+            if (!TryWriteResultPattern(ctx, resultAddress, state.Bits))
+            {
+                result = OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+            else
+            {
+                ApplyClearMode(state, pattern, waitMode);
+            }
+
+            state.WaitingThreads = Math.Max(0, state.WaitingThreads - 1);
+            TraceEventFlag(
+                $"wait-wake pattern=0x{pattern:X16} mode=0x{waitMode:X2} bits=0x{state.Bits:X16} waiters={state.WaitingThreads}");
+            return true;
+        }
+    }
+
+    private static string GetEventFlagWakeKey(ulong handle) =>
+        $"event_flag:0x{handle:X16}";
+
+    private static bool TryWriteResultPattern(CpuContext ctx, ulong address, ulong bits) =>
+        address == 0 || ctx.TryWriteUInt64(address, bits);
+
+    private static bool TryReadUInt32(CpuContext ctx, ulong address, out uint value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(uint)];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt32LittleEndian(buffer);
+        return true;
+    }
+
+    private static bool TryReadUInt64(CpuContext ctx, ulong address, out ulong value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(ulong)];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt64LittleEndian(buffer);
+        return true;
+    }
+
+    private static bool TryReadByte(CpuContext ctx, ulong address, out byte value)
+    {
+        Span<byte> buffer = stackalloc byte[1];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = buffer[0];
+        return true;
+    }
+
+    private static bool TryWriteUInt32(CpuContext ctx, ulong address, uint value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static bool TryReadNullTerminatedUtf8(CpuContext ctx, ulong address, int capacity, out string value)
+    {
+        var bytes = new byte[capacity];
+        for (var index = 0; index < bytes.Length; index++)
+        {
+            Span<byte> current = stackalloc byte[1];
+            if (!ctx.Memory.TryRead(address + (ulong)index, current))
+            {
+                value = string.Empty;
+                return false;
+            }
+
+            if (current[0] == 0)
+            {
+                value = Encoding.UTF8.GetString(bytes, 0, index);
+                return true;
+            }
+
+            bytes[index] = current[0];
+        }
+
+        value = Encoding.UTF8.GetString(bytes);
+        return true;
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        var value = (int)result;
+        ctx[CpuRegister.Rax] = unchecked((ulong)value);
+        return value;
+    }
+
+    private static void TraceEventFlag(string message)
+    {
+        if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_EVENT_FLAG"), "1", StringComparison.Ordinal))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] event_flag.{message}");
+        }
+    }
+
+    private static ulong GetCurrentReturnRip() =>
+        GuestThreadExecution.TryGetCurrentImportCallFrame(out var frame)
+            ? frame.ReturnRip
+            : 0UL;
+
+    private static string FormatFrameChain(CpuContext ctx)
+    {
+        Span<ulong> returns = stackalloc ulong[4];
+        var count = 0;
+        var frame = ctx[CpuRegister.Rbp];
+        for (var index = 0; index < returns.Length && frame != 0; index++)
+        {
+            if (!ctx.TryReadUInt64(frame, out var nextFrame) ||
+                !ctx.TryReadUInt64(frame + sizeof(ulong), out var returnAddress))
+            {
+                break;
+            }
+
+            returns[count++] = returnAddress;
+            if (nextFrame <= frame)
+            {
+                break;
+            }
+
+            frame = nextFrame;
+        }
+
+        return count switch
+        {
+            0 => "none",
+            1 => $"0x{returns[0]:X16}",
+            2 => $"0x{returns[0]:X16},0x{returns[1]:X16}",
+            3 => $"0x{returns[0]:X16},0x{returns[1]:X16},0x{returns[2]:X16}",
+            _ => $"0x{returns[0]:X16},0x{returns[1]:X16},0x{returns[2]:X16},0x{returns[3]:X16}",
+        };
+    }
+
+    private static string FormatGuestWaitObject(CpuContext ctx)
+    {
+        var r12 = ctx[CpuRegister.R12];
+        var r13 = ctx[CpuRegister.R13];
+        var objectAddress = r12 != 0
+            ? r12
+            : r13 >= 0xA8
+                ? r13 - 0xA8
+                : 0;
+
+        var builder = new StringBuilder(256);
+        builder.Append($"r12=0x{r12:X16} r13=0x{r13:X16}");
+        if (objectAddress == 0)
+        {
+            return builder.ToString();
+        }
+
+        builder.Append($" obj=0x{objectAddress:X16}");
+        AppendUInt32(builder, ctx, objectAddress + 0x58, "o58");
+        AppendUInt32(builder, ctx, objectAddress + 0x5C, "o5C");
+        AppendUInt64(builder, ctx, objectAddress + 0x60, "o60");
+        AppendByte(builder, ctx, objectAddress + 0x6C, "state6C");
+        AppendByte(builder, ctx, objectAddress + 0x6D, "o6D");
+        AppendByte(builder, ctx, objectAddress + 0xA0, "waitA0");
+        AppendByte(builder, ctx, objectAddress + 0xA1, "stateA1");
+        AppendByte(builder, ctx, objectAddress + 0xA2, "oA2");
+        AppendUInt64(builder, ctx, objectAddress + 0xA8, "eventA8");
+        if (r13 != 0)
+        {
+            AppendUInt64(builder, ctx, r13, "r13_0");
+            AppendUInt64(builder, ctx, r13 + 8, "r13_8");
+        }
+
+        return builder.ToString();
+    }
+
+    private static void AppendByte(StringBuilder builder, CpuContext ctx, ulong address, string name)
+    {
+        if (TryReadByte(ctx, address, out var value))
+        {
+            builder.Append($" {name}=0x{value:X2}");
+        }
+    }
+
+    private static void AppendUInt32(StringBuilder builder, CpuContext ctx, ulong address, string name)
+    {
+        if (TryReadUInt32(ctx, address, out var value))
+        {
+            builder.Append($" {name}=0x{value:X8}");
+        }
+    }
+
+    private static void AppendUInt64(StringBuilder builder, CpuContext ctx, ulong address, string name)
+    {
+        if (TryReadUInt64(ctx, address, out var value))
+        {
+            builder.Append($" {name}=0x{value:X16}");
+        }
+    }
+}

src/SharpEmu.Libs/Kernel/KernelEventQueueCompatExports.cs

@@ -10,10 +10,14 @@ namespace SharpEmu.Libs.Kernel;
 public static class KernelEventQueueCompatExports
 {
     private const int KernelEventSize = 0x20;
+    public const short KernelEventFilterGraphics = -14;
+    public const short KernelEventFilterAmpr = -16;
+    public const short KernelEventFilterAmprSystem = -17;
 
     private static readonly object _eventQueueGate = new();
     private static readonly HashSet<ulong> _eventQueues = new();
-    private static readonly Dictionary<ulong, Queue<KernelQueuedEvent>> _pendingEvents = new();
+    private static readonly Dictionary<ulong, LinkedList<KernelQueuedEvent>> _pendingEvents = new();
+    private static readonly Dictionary<ulong, Dictionary<(ulong Ident, short Filter), KernelEventRegistration>> _registeredEvents = new();
     private static long _nextEventQueueHandle = 1;
 
     public readonly record struct KernelQueuedEvent(
@@ -24,6 +28,11 @@ public static class KernelEventQueueCompatExports
         ulong Data,
         ulong UserData);
 
+    private readonly record struct KernelEventRegistration(
+        ulong Ident,
+        short Filter,
+        ulong UserData);
+
     [SysAbiExport(
         Nid = "D0OdFMjp46I",
         ExportName = "sceKernelCreateEqueue",
@@ -41,7 +50,8 @@ public static class KernelEventQueueCompatExports
         lock (_eventQueueGate)
         {
             _eventQueues.Add(handle);
-            _pendingEvents[handle] = new Queue<KernelQueuedEvent>();
+            _pendingEvents[handle] = new LinkedList<KernelQueuedEvent>();
+            _registeredEvents[handle] = new Dictionary<(ulong Ident, short Filter), KernelEventRegistration>();
         }
 
         if (!ctx.TryWriteUInt64(outAddress, handle))
@@ -65,6 +75,7 @@ public static class KernelEventQueueCompatExports
         {
             _eventQueues.Remove(handle);
             _pendingEvents.Remove(handle);
+            _registeredEvents.Remove(handle);
         }
 
         TraceEventQueue(ctx, "delete", handle);
@@ -122,8 +133,16 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelAddAmprEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "add_ampr", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var registered = RegisterEvent(
+            handle,
+            unchecked((uint)ctx[CpuRegister.Rsi]),
+            KernelEventFilterAmpr,
+            ctx[CpuRegister.Rdx]);
+        TraceEventQueue(ctx, "add_ampr", handle);
+        return registered
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -133,8 +152,16 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelAddAmprSystemEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "add_ampr_system", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var registered = RegisterEvent(
+            handle,
+            unchecked((uint)ctx[CpuRegister.Rsi]),
+            KernelEventFilterAmprSystem,
+            ctx[CpuRegister.Rdx]);
+        TraceEventQueue(ctx, "add_ampr_system", handle);
+        return registered
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -144,8 +171,15 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelDeleteAmprEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "delete_ampr", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var deleted = DeleteRegisteredEvent(
+            handle,
+            unchecked((uint)ctx[CpuRegister.Rsi]),
+            KernelEventFilterAmpr);
+        TraceEventQueue(ctx, "delete_ampr", handle);
+        return deleted
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -155,8 +189,15 @@ public static class KernelEventQueueCompatExports
         LibraryName = "libKernel")]
     public static int KernelDeleteAmprSystemEvent(CpuContext ctx)
     {
-        TraceEventQueue(ctx, "delete_ampr_system", ctx[CpuRegister.Rdi]);
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var handle = ctx[CpuRegister.Rdi];
+        var deleted = DeleteRegisteredEvent(
+            handle,
+            unchecked((uint)ctx[CpuRegister.Rsi]),
+            KernelEventFilterAmprSystem);
+        TraceEventQueue(ctx, "delete_ampr_system", handle);
+        return deleted
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
     }
 
     [SysAbiExport(
@@ -171,6 +212,57 @@ public static class KernelEventQueueCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "vz+pg2zdopI",
+        ExportName = "sceKernelGetEventUserData",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelGetEventUserData(CpuContext ctx)
+    {
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rdi] + 0x18, out var userData);
+        ctx[CpuRegister.Rax] = userData;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "mJ7aghmgvfc",
+        ExportName = "sceKernelGetEventId",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelGetEventId(CpuContext ctx)
+    {
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rdi], out var ident);
+        ctx[CpuRegister.Rax] = ident;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "23CPPI1tyBY",
+        ExportName = "sceKernelGetEventFilter",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelGetEventFilter(CpuContext ctx)
+    {
+        Span<byte> filterBytes = stackalloc byte[sizeof(short)];
+        var filter = ctx.Memory.TryRead(ctx[CpuRegister.Rdi] + 0x08, filterBytes)
+            ? BinaryPrimitives.ReadInt16LittleEndian(filterBytes)
+            : (short)0;
+        ctx[CpuRegister.Rax] = unchecked((uint)filter);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "kwGyyjohI50",
+        ExportName = "sceKernelGetEventData",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelGetEventData(CpuContext ctx)
+    {
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rdi] + 0x10, out var data);
+        ctx[CpuRegister.Rax] = data;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "fzyMKs9kim0",
         ExportName = "sceKernelWaitEqueue",
@@ -196,7 +288,13 @@ public static class KernelEventQueueCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_OK;
         }
 
-        if (timeoutAddress == 0 && GuestThreadExecution.RequestCurrentThreadBlock("sceKernelWaitEqueue"))
+        if (timeoutAddress == 0 &&
+            GuestThreadExecution.RequestCurrentThreadBlock(
+                ctx,
+                "sceKernelWaitEqueue",
+                GetEventQueueWakeKey(handle),
+                () => ResumeWaitEqueue(ctx, handle, eventsAddress, eventCapacity, outCountAddress),
+                () => HasPendingEvents(handle)))
         {
             TraceEventQueue(ctx, "wait-block", handle);
             return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -216,6 +314,7 @@ public static class KernelEventQueueCompatExports
 
     public static bool EnqueueEvent(ulong handle, KernelQueuedEvent queuedEvent)
     {
+        var queued = false;
         lock (_eventQueueGate)
         {
             if (!_eventQueues.Contains(handle))
@@ -225,15 +324,228 @@ public static class KernelEventQueueCompatExports
 
             if (!_pendingEvents.TryGetValue(handle, out var queue))
             {
-                queue = new Queue<KernelQueuedEvent>();
+                queue = new LinkedList<KernelQueuedEvent>();
                 _pendingEvents[handle] = queue;
             }
 
-            queue.Enqueue(queuedEvent);
+            queue.AddLast(queuedEvent);
+            queued = true;
+        }
+
+        if (queued)
+        {
+            WakeEventQueue(handle);
+        }
+
+        return queued;
+    }
+
+    public static bool RegisterEvent(
+        ulong handle,
+        ulong ident,
+        short filter,
+        ulong userData)
+    {
+        lock (_eventQueueGate)
+        {
+            if (!_eventQueues.Contains(handle))
+            {
+                return false;
+            }
+
+            if (!_registeredEvents.TryGetValue(handle, out var events))
+            {
+                events = new Dictionary<(ulong Ident, short Filter), KernelEventRegistration>();
+                _registeredEvents[handle] = events;
+            }
+
+            events[(ident, filter)] = new KernelEventRegistration(ident, filter, userData);
             return true;
         }
     }
 
+    public static bool DeleteRegisteredEvent(
+        ulong handle,
+        ulong ident,
+        short filter)
+    {
+        lock (_eventQueueGate)
+        {
+            return _registeredEvents.TryGetValue(handle, out var events) &&
+                   events.Remove((ident, filter));
+        }
+    }
+
+    public static int TriggerRegisteredEvents(
+        ulong ident,
+        short filter,
+        ulong data)
+    {
+        List<ulong>? wakeHandles = null;
+        var triggeredCount = 0;
+        lock (_eventQueueGate)
+        {
+            foreach (var (handle, registrations) in _registeredEvents)
+            {
+                if (!registrations.TryGetValue((ident, filter), out var registration))
+                {
+                    continue;
+                }
+
+                if (!_pendingEvents.TryGetValue(handle, out var queue))
+                {
+                    queue = new LinkedList<KernelQueuedEvent>();
+                    _pendingEvents[handle] = queue;
+                }
+
+                QueueOrUpdateEvent(
+                    queue,
+                    new KernelQueuedEvent(
+                        registration.Ident,
+                        registration.Filter,
+                        0,
+                        1,
+                        data,
+                        registration.UserData));
+                (wakeHandles ??= new List<ulong>()).Add(handle);
+                triggeredCount++;
+            }
+        }
+
+        if (wakeHandles is not null)
+        {
+            foreach (var handle in wakeHandles)
+            {
+                WakeEventQueue(handle);
+            }
+        }
+
+        return triggeredCount;
+    }
+
+    public static bool TriggerDisplayEvent(
+        ulong handle,
+        ulong ident,
+        short filter,
+        ulong eventHint,
+        ulong userData)
+    {
+        var triggered = false;
+        lock (_eventQueueGate)
+        {
+            if (!_eventQueues.Contains(handle))
+            {
+                return false;
+            }
+
+            if (!_pendingEvents.TryGetValue(handle, out var events))
+            {
+                events = new LinkedList<KernelQueuedEvent>();
+                _pendingEvents[handle] = events;
+            }
+
+            var count = 1UL;
+            var pendingNode = FindPendingEvent(events, ident, filter);
+            if (pendingNode is not null)
+            {
+                count = Math.Min(((pendingNode.Value.Data >> 12) & 0xFUL) + 1, 0xFUL);
+            }
+
+            var timeBits = unchecked((ulong)Environment.TickCount64) & 0xFFFUL;
+            var eventData = timeBits | (count << 12) | (eventHint & 0xFFFF_FFFF_FFFF_0000UL);
+            var triggeredEvent = new KernelQueuedEvent(
+                ident,
+                filter,
+                0x20,
+                0,
+                eventData,
+                userData);
+
+            if (pendingNode is not null)
+            {
+                pendingNode.Value = triggeredEvent;
+            }
+            else
+            {
+                events.AddLast(triggeredEvent);
+            }
+
+            triggered = true;
+        }
+
+        if (triggered)
+        {
+            WakeEventQueue(handle);
+        }
+
+        return triggered;
+    }
+
+    private static int ResumeWaitEqueue(
+        CpuContext ctx,
+        ulong handle,
+        ulong eventsAddress,
+        int eventCapacity,
+        ulong outCountAddress)
+    {
+        var deliveredCount = DequeueEvents(ctx, handle, eventsAddress, eventCapacity);
+        if (outCountAddress != 0 && !TryWriteUInt32(ctx, outCountAddress, (uint)deliveredCount))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static bool HasPendingEvents(ulong handle)
+    {
+        lock (_eventQueueGate)
+        {
+            return _pendingEvents.TryGetValue(handle, out var events) && events.Count != 0;
+        }
+    }
+
+    private static void QueueOrUpdateEvent(
+        LinkedList<KernelQueuedEvent> queue,
+        KernelQueuedEvent queuedEvent)
+    {
+        var pendingNode = FindPendingEvent(queue, queuedEvent.Ident, queuedEvent.Filter);
+        if (pendingNode is null)
+        {
+            queue.AddLast(queuedEvent);
+            return;
+        }
+
+        pendingNode.Value = queuedEvent with
+        {
+            Fflags = Math.Max(pendingNode.Value.Fflags + 1, queuedEvent.Fflags),
+        };
+    }
+
+    private static LinkedListNode<KernelQueuedEvent>? FindPendingEvent(
+        LinkedList<KernelQueuedEvent> queue,
+        ulong ident,
+        short filter)
+    {
+        for (var node = queue.First; node is not null; node = node.Next)
+        {
+            if (node.Value.Ident == ident && node.Value.Filter == filter)
+            {
+                return node;
+            }
+        }
+
+        return null;
+    }
+
+    private static string GetEventQueueWakeKey(ulong handle) =>
+        $"sceKernelWaitEqueue:{handle:X16}";
+
+    private static void WakeEventQueue(ulong handle)
+    {
+        _ = GuestThreadExecution.Scheduler?.WakeBlockedThreads(GetEventQueueWakeKey(handle));
+    }
+
     private static int DequeueEvents(CpuContext ctx, ulong handle, ulong eventsAddress, int eventCapacity)
     {
         if (eventsAddress == 0 || eventCapacity <= 0)
@@ -253,7 +565,8 @@ public static class KernelEventQueueCompatExports
             events = new KernelQueuedEvent[count];
             for (var i = 0; i < count; i++)
             {
-                events[i] = queue.Dequeue();
+                events[i] = queue.First!.Value;
+                queue.RemoveFirst();
             }
         }
 

src/SharpEmu.Libs/Kernel/KernelExports.cs

@@ -204,6 +204,16 @@ public static class KernelExports
         var nameAddress = ctx[CpuRegister.R8];
         var name = nameAddress == 0 ? string.Empty : ReadCString(ctx, nameAddress, 256);
         var threadHandle = KernelPthreadState.CreateThreadHandle(name);
+        KernelPthreadExtendedCompatExports.GetThreadStartScheduling(
+            ctx,
+            attrAddress,
+            out var priority,
+            out var affinityMask);
+        KernelPthreadExtendedCompatExports.RegisterThreadStart(
+            threadHandle,
+            name,
+            priority,
+            affinityMask);
         if (threadIdAddress != 0 && !ctx.TryWriteUInt64(threadIdAddress, threadHandle))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
@@ -212,13 +222,22 @@ public static class KernelExports
         if (ShouldTracePthread())
         {
             Console.Error.WriteLine(
-                $"[LOADER][TRACE] pthread_create: out=0x{threadIdAddress:X16} attr=0x{attrAddress:X16} entry=0x{entryAddress:X16} arg=0x{argument:X16} name_ptr=0x{nameAddress:X16} name='{name}' -> thread=0x{threadHandle:X16}");
+                $"[LOADER][TRACE] pthread_create: out=0x{threadIdAddress:X16} attr=0x{attrAddress:X16} " +
+                $"entry=0x{entryAddress:X16} arg=0x{argument:X16} name_ptr=0x{nameAddress:X16} " +
+                $"name='{name}' priority={priority} affinity=0x{affinityMask:X} -> thread=0x{threadHandle:X16}");
         }
 
         var scheduler = GuestThreadExecution.Scheduler;
         if (scheduler is not null && entryAddress != 0)
         {
-            var request = new GuestThreadStartRequest(threadHandle, entryAddress, argument, attrAddress, name);
+            var request = new GuestThreadStartRequest(
+                threadHandle,
+                entryAddress,
+                argument,
+                attrAddress,
+                name,
+                priority,
+                affinityMask);
             if (!scheduler.TryStartThread(ctx, request, out var error))
             {
                 Console.Error.WriteLine(
@@ -252,6 +271,32 @@ public static class KernelExports
         return PthreadCreate(ctx);
     }
 
+    [SysAbiExport(
+        Nid = "3kg7rT0NQIs",
+        ExportName = "scePthreadExit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PthreadExit(CpuContext ctx)
+    {
+        var value = ctx[CpuRegister.Rdi];
+        GuestThreadExecution.RequestCurrentEntryExit("scePthreadExit", value);
+        ctx[CpuRegister.Rax] = value;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "FJrT5LuUBAU",
+        ExportName = "pthread_exit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePosix")]
+    public static int PosixPthreadExit(CpuContext ctx)
+    {
+        var value = ctx[CpuRegister.Rdi];
+        GuestThreadExecution.RequestCurrentEntryExit("pthread_exit", value);
+        ctx[CpuRegister.Rax] = value;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "onNY9Byn-W8",
         ExportName = "scePthreadJoin",
@@ -261,10 +306,6 @@ public static class KernelExports
     {
         var threadId = ctx[CpuRegister.Rdi];
         var returnValueAddress = ctx[CpuRegister.Rsi];
-        if (returnValueAddress != 0 && !ctx.TryWriteUInt64(returnValueAddress, 0))
-        {
-            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
-        }
 
         if (ShouldTracePthread())
         {
@@ -272,6 +313,30 @@ public static class KernelExports
                 $"[LOADER][TRACE] pthread_join: thread=0x{threadId:X16} retval_out=0x{returnValueAddress:X16}");
         }
 
+        var returnValue = 0UL;
+        if (GuestThreadExecution.Scheduler is { } scheduler &&
+            !scheduler.TryJoinThread(ctx, threadId, out returnValue, out var error))
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][ERROR] pthread_join: thread=0x{threadId:X16}: {error}");
+            var result = string.Equals(
+                error,
+                "thread cannot join itself",
+                StringComparison.Ordinal)
+                ? OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT
+                : OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
+            ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+            return (int)result;
+        }
+
+        if (returnValueAddress != 0 &&
+            !ctx.TryWriteUInt64(returnValueAddress, returnValue))
+        {
+            ctx[CpuRegister.Rax] =
+                unchecked((ulong)(int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }

src/SharpEmu.Libs/Kernel/KernelPthreadCompatExports.cs

@@ -1,7 +1,9 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+using System.Collections.Concurrent;
 using SharpEmu.HLE;
+using System.Buffers.Binary;
 using System.Threading;
 using System.Diagnostics.CodeAnalysis;
 
@@ -9,35 +11,48 @@ namespace SharpEmu.Libs.Kernel;
 
 public static class KernelPthreadCompatExports
 {
-    private const int MutexTypeDefault = 1;
     private const int MutexTypeErrorCheck = 1;
     private const int MutexTypeRecursive = 2;
     private const int MutexTypeNormal = 3;
     private const int MutexTypeAdaptiveNp = 4;
     private const ulong StaticAdaptiveMutexInitializer = 1;
-    private const ulong SyntheticMutexHandleBase = 0x00006000_0000_0000;
-    private const ulong SyntheticMutexAttrHandleBase = 0x00006001_0000_0000;
-    private const ulong SyntheticCondHandleBase = 0x00006002_0000_0000;
+    private const int MutexObjectSize = 0x100;
+    private const int MutexAttrObjectSize = 0x40;
+    private const int CondObjectSize = 0x100;
     private const int DefaultSpuriousCondWakeMilliseconds = 1;
+    private const int PthreadOnceUninitialized = 0;
+    private const int PthreadOnceInProgress = 1;
+    private const int PthreadOnceDone = 2;
 
     private static readonly object _stateGate = new();
-    private static readonly Dictionary<ulong, PthreadMutexState> _mutexStates = new();
+    private static readonly ConcurrentDictionary<ulong, PthreadMutexState> _mutexStates = new();
     private static readonly Dictionary<ulong, PthreadMutexAttrState> _mutexAttrStates = new();
     private static readonly Dictionary<ulong, PthreadCondState> _condStates = new();
+    private static readonly Dictionary<ulong, object> _onceGates = new();
     private static readonly HashSet<ulong> _condAttrStates = new();
-    private static long _nextSyntheticMutexHandleId = 1;
-    private static long _nextSyntheticMutexAttrHandleId = 1;
-    private static long _nextSyntheticCondHandleId = 1;
+    private static readonly bool _tracePthreads =
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PTHREADS"), "1", StringComparison.Ordinal);
+    private static readonly bool _tracePthreadConds =
+        _tracePthreads ||
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PTHREAD_CONDS"), "1", StringComparison.Ordinal);
+    private static readonly HashSet<ulong>? _tracePthreadMutexFilter = ParseTraceAddressFilter(
+        Environment.GetEnvironmentVariable("SHARPEMU_LOG_PTHREAD_MUTEX_FILTER"));
 
     private sealed class PthreadMutexState
     {
         public SemaphoreSlim Semaphore { get; } = new(1, 1);
         public ulong OwnerThreadId { get; set; }
         public int RecursionCount { get; set; }
-        public int Type { get; set; } = MutexTypeDefault;
+        public int Type { get; set; } = MutexTypeErrorCheck;
         public int Protocol { get; set; }
     }
 
+    private sealed class PthreadMutexWaiter
+    {
+        public required ulong ThreadId { get; init; }
+        public int Reserved;
+    }
+
     private sealed class PthreadCondState
     {
         public object SyncRoot { get; } = new();
@@ -60,6 +75,13 @@ public static class KernelPthreadCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "EotR8a3ASf4",
+        ExportName = "pthread_self",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadSelf(CpuContext ctx) => PthreadSelf(ctx);
+
     [SysAbiExport(
         Nid = "3PtV6p3QNX4",
         ExportName = "scePthreadEqual",
@@ -343,6 +365,93 @@ public static class KernelPthreadCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "14bOACANTBo",
+        ExportName = "scePthreadOnce",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PthreadOnce(CpuContext ctx)
+    {
+        var onceAddress = ctx[CpuRegister.Rdi];
+        var initRoutine = ctx[CpuRegister.Rsi];
+        if (onceAddress == 0 || initRoutine == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        if (!TryReadInt32(ctx, onceAddress, out var onceValue))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        if (onceValue == PthreadOnceDone)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+
+        var gate = GetPthreadOnceGate(onceAddress);
+        var shouldCall = false;
+        lock (gate)
+        {
+            if (!TryReadInt32(ctx, onceAddress, out onceValue))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            while (onceValue == PthreadOnceInProgress)
+            {
+                Monitor.Wait(gate, TimeSpan.FromMilliseconds(1));
+                if (!TryReadInt32(ctx, onceAddress, out onceValue))
+                {
+                    return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+                }
+            }
+
+            if (onceValue != PthreadOnceDone)
+            {
+                if (!TryWriteInt32(ctx, onceAddress, PthreadOnceInProgress))
+                {
+                    return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+                }
+
+                shouldCall = true;
+            }
+        }
+
+        if (shouldCall)
+        {
+            var scheduler = GuestThreadExecution.Scheduler;
+            string? error = null;
+            if (scheduler is null ||
+                !scheduler.TryCallGuestFunction(ctx, initRoutine, 0, 0, 0, 0, "pthread_once", out error))
+            {
+                lock (gate)
+                {
+                    _ = TryWriteInt32(ctx, onceAddress, PthreadOnceUninitialized);
+                    Monitor.PulseAll(gate);
+                }
+
+                TracePthreadOnce(onceAddress, initRoutine, "failed", error);
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN);
+            }
+
+            lock (gate)
+            {
+                if (!TryWriteInt32(ctx, onceAddress, PthreadOnceDone))
+                {
+                    _ = TryWriteInt32(ctx, onceAddress, PthreadOnceUninitialized);
+                    Monitor.PulseAll(gate);
+                    return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+                }
+
+                Monitor.PulseAll(gate);
+            }
+        }
+
+        TracePthreadOnce(onceAddress, initRoutine, shouldCall ? "call" : "done", null);
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
     private static int PthreadMutexInitCore(CpuContext ctx, ulong mutexAddress, ulong attrAddress)
     {
         if (mutexAddress == 0)
@@ -357,14 +466,28 @@ public static class KernelPthreadCompatExports
             Protocol = attr.Protocol,
         };
 
-        var syntheticHandle = AllocateSyntheticHandle(SyntheticMutexHandleBase, ref _nextSyntheticMutexHandleId);
-        lock (_stateGate)
+        if (!TryAllocateOpaqueObject(ctx, MutexObjectSize, out var handle))
         {
-            _mutexStates[mutexAddress] = state;
-            _mutexStates[syntheticHandle] = state;
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+        if (!InitializeMutexObject(ctx, handle, state))
+        {
+            state.Semaphore.Dispose();
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        _mutexStates[mutexAddress] = state;
+        _mutexStates[handle] = state;
+
+        if (!KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, mutexAddress, handle))
+        {
+            _mutexStates.TryRemove(mutexAddress, out _);
+            _mutexStates.TryRemove(handle, out _);
+
+            state.Semaphore.Dispose();
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        _ = ctx.TryWriteUInt64(mutexAddress, syntheticHandle);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -376,15 +499,10 @@ public static class KernelPthreadCompatExports
         }
 
         var resolvedAddress = ResolveMutexHandle(ctx, mutexAddress);
-        PthreadMutexState? state;
-        lock (_stateGate)
+        _mutexStates.TryRemove(resolvedAddress, out var state);
+        if (resolvedAddress != mutexAddress)
         {
-            _mutexStates.TryGetValue(resolvedAddress, out state);
-            _mutexStates.Remove(resolvedAddress);
-            if (resolvedAddress != mutexAddress)
-            {
-                _mutexStates.Remove(mutexAddress);
-            }
+            _mutexStates.TryRemove(mutexAddress, out _);
         }
 
         if (state is null)
@@ -392,7 +510,7 @@ public static class KernelPthreadCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
 
-        _ = ctx.TryWriteUInt64(mutexAddress, 0);
+        _ = KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, mutexAddress, 0);
         state.Semaphore.Dispose();
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -447,15 +565,33 @@ public static class KernelPthreadCompatExports
             }
         }
 
-        var acquired = true;
-        if (tryOnly)
+        var acquired = state.Semaphore.Wait(0);
+        if (!acquired)
         {
-            acquired = state.Semaphore.Wait(0);
-        }
-        else
-        {
-            state.Semaphore.Wait();
+            // Guest-thread blocking for pthread_mutex_lock is currently disabled.
+            // Demon's Souls deadlocks during PS5SyncEvent initialization.
+            /* var waiter = new PthreadMutexWaiter { ThreadId = currentThreadId };
+            if (!tryOnly &&
+                GuestThreadExecution.IsGuestThread &&
+                GuestThreadExecution.TryGetCurrentImportCallFrame(out _) &&
+                GuestThreadExecution.RequestCurrentThreadBlock(
+                    ctx,
+                    "pthread_mutex_lock",
+                    GetMutexWakeKey(resolvedAddress),
+                    () => CompleteBlockedMutexLock(ctx, mutexAddress, resolvedAddress, state, waiter),
+                    () => TryReserveBlockedMutexLock(ctx, mutexAddress, resolvedAddress, state, waiter)))
+            {
+                TracePthreadMutex(ctx, "lock-block", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+                return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+            } */
+
+            if (!tryOnly)
+            {
+                state.Semaphore.Wait();
+                acquired = true;
+            }
         }
+
         if (!acquired)
         {
             TracePthreadMutex(ctx, tryOnly ? "trylock" : "lock", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
@@ -514,6 +650,7 @@ public static class KernelPthreadCompatExports
             try
             {
                 state.Semaphore.Release();
+                _ = GuestThreadExecution.Scheduler?.WakeBlockedThreads(GetMutexWakeKey(resolvedAddress), 1);
             }
             catch (SemaphoreFullException)
             {
@@ -533,14 +670,34 @@ public static class KernelPthreadCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        var syntheticHandle = AllocateSyntheticHandle(SyntheticMutexAttrHandleBase, ref _nextSyntheticMutexAttrHandleId);
-        lock (_stateGate)
+        if (!TryAllocateOpaqueObject(ctx, MutexAttrObjectSize, out var handle))
         {
-            _mutexAttrStates[attrAddress] = new PthreadMutexAttrState(MutexTypeDefault, 0);
-            _mutexAttrStates[syntheticHandle] = new PthreadMutexAttrState(MutexTypeDefault, 0);
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        var initialState = new PthreadMutexAttrState(MutexTypeErrorCheck, 0);
+        if (!WriteMutexAttrObject(ctx, handle, initialState))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        lock (_stateGate)
+        {
+            _mutexAttrStates[attrAddress] = initialState;
+            _mutexAttrStates[handle] = initialState;
+        }
+
+        if (!KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, attrAddress, handle))
+        {
+            lock (_stateGate)
+            {
+                _mutexAttrStates.Remove(attrAddress);
+                _mutexAttrStates.Remove(handle);
+            }
+
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        _ = ctx.TryWriteUInt64(attrAddress, syntheticHandle);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -572,21 +729,25 @@ public static class KernelPthreadCompatExports
         }
 
         var resolvedAddress = ResolveMutexAttrHandle(ctx, attrAddress);
+        PthreadMutexAttrState updatedState;
         lock (_stateGate)
         {
             if (!_mutexAttrStates.TryGetValue(resolvedAddress, out var state))
             {
-                state = new PthreadMutexAttrState(MutexTypeDefault, 0);
+                state = new PthreadMutexAttrState(MutexTypeErrorCheck, 0);
             }
 
-            _mutexAttrStates[resolvedAddress] = state with { Type = NormalizeMutexType(type) };
+            updatedState = state with { Type = NormalizeMutexType(type) };
+            _mutexAttrStates[resolvedAddress] = updatedState;
             if (resolvedAddress != attrAddress)
             {
-                _mutexAttrStates[attrAddress] = _mutexAttrStates[resolvedAddress];
+                _mutexAttrStates[attrAddress] = updatedState;
             }
         }
 
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        return WriteMutexAttrObject(ctx, resolvedAddress, updatedState)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
     }
 
     private static int PthreadMutexattrSetprotocolCore(CpuContext ctx, ulong attrAddress, int protocol)
@@ -597,21 +758,25 @@ public static class KernelPthreadCompatExports
         }
 
         var resolvedAddress = ResolveMutexAttrHandle(ctx, attrAddress);
+        PthreadMutexAttrState updatedState;
         lock (_stateGate)
         {
             if (!_mutexAttrStates.TryGetValue(resolvedAddress, out var state))
             {
-                state = new PthreadMutexAttrState(MutexTypeDefault, 0);
+                state = new PthreadMutexAttrState(MutexTypeErrorCheck, 0);
             }
 
-            _mutexAttrStates[resolvedAddress] = state with { Protocol = protocol };
+            updatedState = state with { Protocol = protocol };
+            _mutexAttrStates[resolvedAddress] = updatedState;
             if (resolvedAddress != attrAddress)
             {
-                _mutexAttrStates[attrAddress] = _mutexAttrStates[resolvedAddress];
+                _mutexAttrStates[attrAddress] = updatedState;
             }
         }
 
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        return WriteMutexAttrObject(ctx, resolvedAddress, updatedState)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
     }
 
     private static ulong ResolveMutexHandle(CpuContext ctx, ulong mutexAddress)
@@ -621,22 +786,16 @@ public static class KernelPthreadCompatExports
             return 0;
         }
 
-        lock (_stateGate)
+        if (_mutexStates.ContainsKey(mutexAddress))
         {
-            if (_mutexStates.ContainsKey(mutexAddress))
-            {
-                return mutexAddress;
-            }
+            return mutexAddress;
         }
 
-        if (ctx.TryReadUInt64(mutexAddress, out var pointedHandle) && pointedHandle != 0)
+        if (KernelMemoryCompatExports.TryReadUInt64Compat(ctx, mutexAddress, out var pointedHandle) && pointedHandle != 0)
         {
-            lock (_stateGate)
+            if (_mutexStates.ContainsKey(pointedHandle))
             {
-                if (_mutexStates.ContainsKey(pointedHandle))
-                {
-                    return pointedHandle;
-                }
+                return pointedHandle;
             }
         }
 
@@ -652,16 +811,13 @@ public static class KernelPthreadCompatExports
             return false;
         }
 
-        lock (_stateGate)
+        if (_mutexStates.TryGetValue(mutexAddress, out state))
         {
-            if (_mutexStates.TryGetValue(mutexAddress, out state))
-            {
-                resolvedAddress = mutexAddress;
-                return true;
-            }
+            resolvedAddress = mutexAddress;
+            return true;
         }
 
-        if (!ctx.TryReadUInt64(mutexAddress, out var pointedHandle))
+        if (!KernelMemoryCompatExports.TryReadUInt64Compat(ctx, mutexAddress, out var pointedHandle))
         {
             return false;
         }
@@ -673,14 +829,11 @@ public static class KernelPthreadCompatExports
 
         if (pointedHandle != 0)
         {
-            lock (_stateGate)
+            if (_mutexStates.TryGetValue(pointedHandle, out state))
             {
-                if (_mutexStates.TryGetValue(pointedHandle, out state))
-                {
-                    _mutexStates[mutexAddress] = state;
-                    resolvedAddress = pointedHandle;
-                    return true;
-                }
+                _mutexStates.TryAdd(mutexAddress, state);
+                resolvedAddress = pointedHandle;
+                return true;
             }
 
             resolvedAddress = pointedHandle;
@@ -693,7 +846,7 @@ public static class KernelPthreadCompatExports
             return false;
         }
 
-        return CreateImplicitMutexState(ctx, mutexAddress, MutexTypeDefault, out resolvedAddress, out state);
+        return CreateImplicitMutexState(ctx, mutexAddress, MutexTypeErrorCheck, out resolvedAddress, out state);
     }
 
     private static ulong ResolveMutexAttrHandle(CpuContext ctx, ulong attrAddress)
@@ -703,15 +856,7 @@ public static class KernelPthreadCompatExports
             return 0;
         }
 
-        lock (_stateGate)
-        {
-            if (_mutexAttrStates.ContainsKey(attrAddress))
-            {
-                return attrAddress;
-            }
-        }
-
-        if (ctx.TryReadUInt64(attrAddress, out var pointedHandle) && pointedHandle != 0)
+        if (KernelMemoryCompatExports.TryReadUInt64Compat(ctx, attrAddress, out var pointedHandle) && pointedHandle != 0)
         {
             lock (_stateGate)
             {
@@ -722,6 +867,14 @@ public static class KernelPthreadCompatExports
             }
         }
 
+        lock (_stateGate)
+        {
+            if (_mutexAttrStates.ContainsKey(attrAddress))
+            {
+                return attrAddress;
+            }
+        }
+
         return attrAddress;
     }
 
@@ -737,7 +890,7 @@ public static class KernelPthreadCompatExports
         {
             return _mutexAttrStates.TryGetValue(resolvedAddress, out var state)
                 ? state
-                : new PthreadMutexAttrState(MutexTypeDefault, 0);
+                : new PthreadMutexAttrState(MutexTypeErrorCheck, 0);
         }
     }
 
@@ -756,7 +909,7 @@ public static class KernelPthreadCompatExports
             }
         }
 
-        if (ctx.TryReadUInt64(condAddress, out var pointedHandle) && pointedHandle != 0)
+        if (KernelMemoryCompatExports.TryReadUInt64Compat(ctx, condAddress, out var pointedHandle) && pointedHandle != 0)
         {
             lock (_stateGate)
             {
@@ -788,7 +941,7 @@ public static class KernelPthreadCompatExports
             }
         }
 
-        if (ctx is null || !ctx.TryReadUInt64(condAddress, out var pointedHandle))
+        if (ctx is null || !KernelMemoryCompatExports.TryReadUInt64Compat(ctx, condAddress, out var pointedHandle))
         {
             return false;
         }
@@ -816,23 +969,60 @@ public static class KernelPthreadCompatExports
         }
 
         var createdState = new PthreadCondState();
-        var syntheticHandle = AllocateSyntheticHandle(SyntheticCondHandleBase, ref _nextSyntheticCondHandleId);
+        if (!TryAllocateOpaqueObject(ctx, CondObjectSize, out var handle))
+        {
+            return false;
+        }
+
         lock (_stateGate)
         {
             _condStates[condAddress] = createdState;
-            _condStates[syntheticHandle] = createdState;
+            _condStates[handle] = createdState;
         }
 
-        _ = ctx.TryWriteUInt64(condAddress, syntheticHandle);
-        resolvedAddress = syntheticHandle;
+        if (!KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, condAddress, handle))
+        {
+            lock (_stateGate)
+            {
+                _condStates.Remove(condAddress);
+                _condStates.Remove(handle);
+            }
+
+            return false;
+        }
+
+        resolvedAddress = handle;
         state = createdState;
         return true;
     }
 
-    private static ulong AllocateSyntheticHandle(ulong baseAddress, ref long nextId)
+    private static bool TryAllocateOpaqueObject(CpuContext ctx, int size, out ulong address)
     {
-        var id = unchecked((ulong)Interlocked.Increment(ref nextId));
-        return baseAddress + (id << 4);
+        address = 0;
+        if (ctx.Memory is not IGuestMemoryAllocator allocator ||
+            !allocator.TryAllocateGuestMemory((ulong)size, alignment: 0x10, out address))
+        {
+            return false;
+        }
+
+        Span<byte> initialData = stackalloc byte[size];
+        initialData.Clear();
+        return ctx.Memory.TryWrite(address, initialData);
+    }
+
+    private static bool InitializeMutexObject(CpuContext ctx, ulong address, PthreadMutexState state) =>
+        TryWriteUInt32(ctx, address + 0x20, unchecked((uint)state.Type)) &&
+        TryWriteUInt32(ctx, address + 0x3C, unchecked((uint)state.Protocol));
+
+    private static bool WriteMutexAttrObject(CpuContext ctx, ulong address, PthreadMutexAttrState state) =>
+        TryWriteUInt32(ctx, address, unchecked((uint)state.Type)) &&
+        TryWriteUInt32(ctx, address + 4, unchecked((uint)state.Protocol));
+
+    private static bool TryWriteUInt32(CpuContext ctx, ulong address, uint value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(uint)];
+        BitConverter.TryWriteBytes(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes);
     }
 
     private static int PthreadCondInitCore(CpuContext ctx, ulong condAddress)
@@ -842,15 +1032,29 @@ public static class KernelPthreadCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        var syntheticHandle = AllocateSyntheticHandle(SyntheticCondHandleBase, ref _nextSyntheticCondHandleId);
+        if (!TryAllocateOpaqueObject(ctx, CondObjectSize, out var handle))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
         lock (_stateGate)
         {
             var state = new PthreadCondState();
             _condStates[condAddress] = state;
-            _condStates[syntheticHandle] = state;
+            _condStates[handle] = state;
+        }
+
+        if (!KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, condAddress, handle))
+        {
+            lock (_stateGate)
+            {
+                _condStates.Remove(condAddress);
+                _condStates.Remove(handle);
+            }
+
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        _ = ctx.TryWriteUInt64(condAddress, syntheticHandle);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -871,7 +1075,7 @@ public static class KernelPthreadCompatExports
             }
         }
 
-        _ = ctx.TryWriteUInt64(condAddress, 0);
+        _ = KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, condAddress, 0);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -1007,6 +1211,63 @@ public static class KernelPthreadCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    private static string GetMutexWakeKey(ulong resolvedMutexAddress) =>
+        $"pthread_mutex:0x{resolvedMutexAddress:X16}";
+
+    private static bool TryReserveBlockedMutexLock(
+        CpuContext ctx,
+        ulong mutexAddress,
+        ulong resolvedAddress,
+        PthreadMutexState state,
+        PthreadMutexWaiter waiter)
+    {
+        lock (state)
+        {
+            if (state.OwnerThreadId != 0 || state.RecursionCount != 0)
+            {
+                TracePthreadMutex(ctx, "lock-reserve-busy", mutexAddress, resolvedAddress, state, waiter.ThreadId, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
+                return false;
+            }
+
+            state.OwnerThreadId = waiter.ThreadId;
+            state.RecursionCount = 1;
+            Interlocked.Exchange(ref waiter.Reserved, 1);
+        }
+
+        TracePthreadMutex(ctx, "lock-reserve", mutexAddress, resolvedAddress, state, waiter.ThreadId, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+        return true;
+    }
+
+    private static int CompleteBlockedMutexLock(
+        CpuContext ctx,
+        ulong mutexAddress,
+        ulong resolvedAddress,
+        PthreadMutexState state,
+        PthreadMutexWaiter waiter)
+    {
+        var currentThreadId = KernelPthreadState.GetCurrentThreadHandle();
+        if (Interlocked.Exchange(ref waiter.Reserved, 0) == 1)
+        {
+            TracePthreadMutex(ctx, "lock-resume", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (!state.Semaphore.Wait(0))
+        {
+            TracePthreadMutex(ctx, "lock-resume-busy", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY;
+        }
+
+        lock (state)
+        {
+            state.OwnerThreadId = currentThreadId;
+            state.RecursionCount = 1;
+        }
+
+        TracePthreadMutex(ctx, "lock-resume", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     private static TimeSpan GetCondWaitTimeout(uint timeoutUsec)
     {
         if (timeoutUsec == 0)
@@ -1031,15 +1292,55 @@ public static class KernelPthreadCompatExports
     {
         return type switch
         {
-            0 => MutexTypeDefault,
+            0 => MutexTypeErrorCheck,
             1 => MutexTypeErrorCheck,
             2 => MutexTypeRecursive,
             3 => MutexTypeNormal,
             4 => MutexTypeAdaptiveNp,
-            _ => MutexTypeDefault,
+            _ => MutexTypeErrorCheck,
         };
     }
 
+    private static object GetPthreadOnceGate(ulong onceAddress)
+    {
+        lock (_stateGate)
+        {
+            if (!_onceGates.TryGetValue(onceAddress, out var gate))
+            {
+                gate = new object();
+                _onceGates[onceAddress] = gate;
+            }
+
+            return gate;
+        }
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+        return (int)result;
+    }
+
+    private static bool TryReadInt32(CpuContext ctx, ulong address, out int value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(int)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadInt32LittleEndian(bytes);
+        return true;
+    }
+
+    private static bool TryWriteInt32(CpuContext ctx, ulong address, int value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes);
+    }
+
     private static bool CreateImplicitMutexState(CpuContext ctx, ulong mutexAddress, int type, out ulong resolvedAddress, [NotNullWhen(true)] out PthreadMutexState? state)
     {
         var createdState = new PthreadMutexState
@@ -1047,7 +1348,19 @@ public static class KernelPthreadCompatExports
             Type = type,
         };
 
-        var syntheticHandle = AllocateSyntheticHandle(SyntheticMutexHandleBase, ref _nextSyntheticMutexHandleId);
+        if (!TryAllocateOpaqueObject(ctx, MutexObjectSize, out var handle))
+        {
+            resolvedAddress = 0;
+            state = null;
+            return false;
+        }
+        if (!InitializeMutexObject(ctx, handle, createdState))
+        {
+            resolvedAddress = 0;
+            state = null;
+            return false;
+        }
+
         lock (_stateGate)
         {
             if (_mutexStates.TryGetValue(mutexAddress, out state))
@@ -1056,18 +1369,27 @@ public static class KernelPthreadCompatExports
                 return true;
             }
 
-            if (_mutexStates.TryGetValue(syntheticHandle, out state))
+            if (_mutexStates.TryGetValue(handle, out state))
             {
-                resolvedAddress = syntheticHandle;
+                resolvedAddress = handle;
                 return true;
             }
 
             _mutexStates[mutexAddress] = createdState;
-            _mutexStates[syntheticHandle] = createdState;
+            _mutexStates[handle] = createdState;
         }
 
-        _ = ctx.TryWriteUInt64(mutexAddress, syntheticHandle);
-        resolvedAddress = syntheticHandle;
+        if (!KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, mutexAddress, handle))
+        {
+            _mutexStates.TryRemove(mutexAddress, out _);
+            _mutexStates.TryRemove(handle, out _);
+
+            resolvedAddress = 0;
+            state = null;
+            return false;
+        }
+
+        resolvedAddress = handle;
         state = createdState;
         return true;
     }
@@ -1084,15 +1406,27 @@ public static class KernelPthreadCompatExports
             $"[LOADER][TRACE] pthread_self: stale_rdi=0x{ctx[CpuRegister.Rdi]:X16} thread=0x{currentThreadHandle:X16} tid=0x{currentThreadId:X16}");
     }
 
-    private static void TracePthreadMutex(CpuContext ctx, string operation, ulong mutexAddress, ulong resolvedAddress, PthreadMutexState? state, ulong currentThreadId, int result)
+    private static void TracePthreadOnce(ulong onceAddress, ulong initRoutine, string operation, string? error)
     {
         if (!ShouldTracePthread())
         {
             return;
         }
 
-        _ = ctx.TryReadUInt64(mutexAddress, out var guestWord0);
-        _ = ctx.TryReadUInt64(mutexAddress + 8, out var guestWord1);
+        var suffix = string.IsNullOrWhiteSpace(error) ? string.Empty : $" error={error}";
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] pthread_once_{operation}: once=0x{onceAddress:X16} init=0x{initRoutine:X16}{suffix}");
+    }
+
+    private static void TracePthreadMutex(CpuContext ctx, string operation, ulong mutexAddress, ulong resolvedAddress, PthreadMutexState? state, ulong currentThreadId, int result)
+    {
+        if (!ShouldTracePthreadMutex(mutexAddress, resolvedAddress))
+        {
+            return;
+        }
+
+        _ = KernelMemoryCompatExports.TryReadUInt64Compat(ctx, mutexAddress, out var guestWord0);
+        _ = KernelMemoryCompatExports.TryReadUInt64Compat(ctx, mutexAddress + 8, out var guestWord1);
         Console.Error.WriteLine(
             $"[LOADER][TRACE] pthread_{operation}: mutex=0x{mutexAddress:X16} resolved=0x{resolvedAddress:X16} " +
             $"guest[0]=0x{guestWord0:X16} guest[8]=0x{guestWord1:X16} " +
@@ -1102,7 +1436,7 @@ public static class KernelPthreadCompatExports
 
     private static void TracePthreadCond(string operation, ulong condAddress, ulong mutexAddress, PthreadCondState? state, bool timed, int result)
     {
-        if (!ShouldTracePthread())
+        if (!_tracePthreadConds)
         {
             return;
         }
@@ -1114,6 +1448,45 @@ public static class KernelPthreadCompatExports
 
     private static bool ShouldTracePthread()
     {
-        return string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PTHREADS"), "1", StringComparison.Ordinal);
+        return _tracePthreads;
+    }
+
+    private static bool ShouldTracePthreadMutex(ulong mutexAddress, ulong resolvedAddress)
+    {
+        if (_tracePthreadMutexFilter is null || _tracePthreadMutexFilter.Count == 0)
+        {
+            return _tracePthreads;
+        }
+
+        return _tracePthreadMutexFilter.Contains(mutexAddress) ||
+            _tracePthreadMutexFilter.Contains(resolvedAddress);
+    }
+
+    private static HashSet<ulong>? ParseTraceAddressFilter(string? filter)
+    {
+        if (string.IsNullOrWhiteSpace(filter))
+        {
+            return null;
+        }
+
+        var addresses = new HashSet<ulong>();
+        foreach (var token in filter.Split(new[] { ',', ';', ' ' }, StringSplitOptions.RemoveEmptyEntries | StringSplitOptions.TrimEntries))
+        {
+            var normalized = token.StartsWith("0x", StringComparison.OrdinalIgnoreCase)
+                ? token[2..]
+                : token;
+            normalized = normalized.TrimStart('0');
+
+            if (ulong.TryParse(
+                    normalized.Length == 0 ? "0" : normalized,
+                    System.Globalization.NumberStyles.HexNumber,
+                    System.Globalization.CultureInfo.InvariantCulture,
+                    out var address))
+            {
+                addresses.Add(address);
+            }
+        }
+
+        return addresses.Count == 0 ? null : addresses;
     }
 }

src/SharpEmu.Libs/Kernel/KernelPthreadExtendedCompatExports.cs

@@ -3,6 +3,7 @@
 
 using SharpEmu.HLE;
 using System.Buffers.Binary;
+using System.Collections.Concurrent;
 using System.Text;
 using System.Threading;
 using System.Diagnostics.CodeAnalysis;
@@ -12,13 +13,13 @@ namespace SharpEmu.Libs.Kernel;
 public static class KernelPthreadExtendedCompatExports
 {
     private const int DefaultThreadPriority = 700;
-    private const ulong DefaultThreadAffinityMask = ulong.MaxValue;
+    private const ulong DefaultThreadAffinityMask = 0x7FUL;
     private const int DefaultDetachState = 0;
     private const ulong DefaultGuardSize = 0x1000UL;
     private const ulong DefaultStackSize = 0x1_00000UL;
-    private const int DefaultInheritSched = 0;
-    private const int DefaultSchedPolicy = 0;
-    private const int DefaultSchedPriority = 0;
+    private const int DefaultInheritSched = 4;
+    private const int DefaultSchedPolicy = 1;
+    private const int DefaultSchedPriority = DefaultThreadPriority;
     private const ulong SyntheticRwlockHandleBase = 0x00006003_0000_0000;
     private const ulong SyntheticPthreadAttrHandleBase = 0x00006004_0000_0000;
 
@@ -26,12 +27,54 @@ public static class KernelPthreadExtendedCompatExports
     private static readonly Dictionary<ulong, ThreadState> _threadStates = new();
     private static readonly Dictionary<ulong, PthreadAttrState> _attrStates = new();
     private static readonly Dictionary<ulong, PthreadRwlockState> _rwlockStates = new();
-    private static readonly Dictionary<int, TlsKeyState> _tlsKeys = new();
+    private static readonly ConcurrentDictionary<int, TlsKeyState> _tlsKeys = new();
     private static int _nextTlsKey = 1;
     private static long _nextSyntheticRwlockHandleId = 1;
     private static long _nextSyntheticPthreadAttrHandleId = 1;
 
-    private static readonly Dictionary<ulong, Dictionary<int, ulong>> _threadLocalSpecific = new();
+    private static readonly ConcurrentDictionary<ulong, ConcurrentDictionary<int, ulong>> _threadLocalSpecific = new();
+
+    internal static void GetThreadStartScheduling(
+        CpuContext ctx,
+        ulong attrAddress,
+        out int priority,
+        out ulong affinityMask)
+    {
+        if (attrAddress == 0)
+        {
+            priority = DefaultThreadPriority;
+            affinityMask = DefaultThreadAffinityMask;
+            return;
+        }
+
+        var resolvedAddress = ResolvePthreadAttrHandle(ctx, attrAddress);
+        lock (_stateGate)
+        {
+            var attributes = GetOrCreateAttrStateLocked(resolvedAddress);
+            priority = attributes.SchedPriority;
+            affinityMask = attributes.AffinityMask;
+        }
+    }
+
+    internal static void RegisterThreadStart(
+        ulong thread,
+        string name,
+        int priority,
+        ulong affinityMask)
+    {
+        lock (_stateGate)
+        {
+            var state = GetOrCreateThreadStateLocked(thread);
+            state.Name = name;
+            state.Priority = priority;
+            state.AffinityMask = affinityMask;
+            state.Attributes = state.Attributes with
+            {
+                SchedPriority = priority,
+                AffinityMask = affinityMask,
+            };
+        }
+    }
 
     private sealed class ThreadState
     {
@@ -185,6 +228,35 @@ public static class KernelPthreadExtendedCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "rcrVFJsQWRY",
+        ExportName = "scePthreadGetaffinity",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PthreadGetaffinity(CpuContext ctx)
+    {
+        var thread = ctx[CpuRegister.Rdi];
+        var outMaskAddress = ctx[CpuRegister.Rsi];
+        if (thread == 0 || outMaskAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        ulong affinityMask;
+        lock (_stateGate)
+        {
+            affinityMask = GetOrCreateThreadStateLocked(thread).AffinityMask;
+        }
+
+        if (!ctx.TryWriteUInt64(outMaskAddress, affinityMask))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "1tKyG7RlMJo",
         ExportName = "scePthreadGetprio",
@@ -489,6 +561,37 @@ public static class KernelPthreadExtendedCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "-quPa4SEJUw",
+        ExportName = "scePthreadAttrGetstack",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PthreadAttrGetstack(CpuContext ctx)
+    {
+        var attrAddress = ctx[CpuRegister.Rdi];
+        var outStackAddressPointer = ctx[CpuRegister.Rsi];
+        var outStackSizeAddress = ctx[CpuRegister.Rdx];
+        if (attrAddress == 0 || outStackAddressPointer == 0 || outStackSizeAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        PthreadAttrState state;
+        lock (_stateGate)
+        {
+            state = GetOrCreateAttrStateLocked(attrAddress);
+        }
+
+        if (!ctx.TryWriteUInt64(outStackAddressPointer, state.StackAddress) ||
+            !ctx.TryWriteUInt64(outStackSizeAddress, state.StackSize))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "-fA+7ZlGDQs",
         ExportName = "scePthreadAttrGetstacksize",
@@ -733,7 +836,7 @@ public static class KernelPthreadExtendedCompatExports
             _rwlockStates[syntheticHandle] = rwlock;
         }
 
-        _ = ctx.TryWriteUInt64(rwlockAddress, syntheticHandle);
+        _ = KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, rwlockAddress, syntheticHandle);
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -787,7 +890,7 @@ public static class KernelPthreadExtendedCompatExports
             }
         }
 
-        _ = ctx.TryWriteUInt64(rwlockAddress, 0);
+        _ = KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, rwlockAddress, 0);
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -899,15 +1002,13 @@ public static class KernelPthreadExtendedCompatExports
         }
 
         int key;
-        lock (_stateGate)
+        while (true)
         {
-            while (_tlsKeys.ContainsKey(_nextTlsKey))
+            key = Interlocked.Increment(ref _nextTlsKey) - 1;
+            if (_tlsKeys.TryAdd(key, new TlsKeyState(destructor)))
             {
-                _nextTlsKey++;
+                break;
             }
-
-            key = _nextTlsKey++;
-            _tlsKeys[key] = new TlsKeyState(destructor);
         }
 
         if (!TryWriteInt32(ctx, outKeyAddress, key))
@@ -934,17 +1035,14 @@ public static class KernelPthreadExtendedCompatExports
     public static int PosixPthreadKeyDelete(CpuContext ctx)
     {
         var key = unchecked((int)ctx[CpuRegister.Rdi]);
-        lock (_stateGate)
+        if (!_tlsKeys.TryRemove(key, out _))
         {
-            if (!_tlsKeys.Remove(key))
-            {
-                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
-            }
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
+        }
 
-            foreach (var entry in _threadLocalSpecific)
-            {
-                entry.Value.Remove(key);
-            }
+        foreach (var values in _threadLocalSpecific.Values)
+        {
+            values.TryRemove(key, out _);
         }
 
         ctx[CpuRegister.Rax] = 0;
@@ -968,22 +1066,15 @@ public static class KernelPthreadExtendedCompatExports
         var key = unchecked((int)ctx[CpuRegister.Rdi]);
         var value = ctx[CpuRegister.Rsi];
         var currentThreadHandle = KernelPthreadState.GetCurrentThreadHandle();
-        lock (_stateGate)
+        if (!_tlsKeys.ContainsKey(key))
         {
-            if (!_tlsKeys.TryGetValue(key, out _))
-            {
-                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
-            }
-
-            if (!_threadLocalSpecific.TryGetValue(currentThreadHandle, out var values))
-            {
-                values = new Dictionary<int, ulong>();
-                _threadLocalSpecific[currentThreadHandle] = values;
-            }
-
-            values[key] = value;
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
 
+        var values = _threadLocalSpecific.GetOrAdd(
+            currentThreadHandle,
+            static _ => new ConcurrentDictionary<int, ulong>());
+        values[key] = value;
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -1005,19 +1096,16 @@ public static class KernelPthreadExtendedCompatExports
         var key = unchecked((int)ctx[CpuRegister.Rdi]);
         var currentThreadHandle = KernelPthreadState.GetCurrentThreadHandle();
         ulong value = 0;
-        lock (_stateGate)
+        if (!_tlsKeys.ContainsKey(key))
         {
-            if (!_tlsKeys.TryGetValue(key, out _))
-            {
-                ctx[CpuRegister.Rax] = 0;
-                return (int)OrbisGen2Result.ORBIS_GEN2_OK;
-            }
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
 
-            if (_threadLocalSpecific.TryGetValue(currentThreadHandle, out var values) &&
-                values.TryGetValue(key, out var storedValue))
-            {
-                value = storedValue;
-            }
+        if (_threadLocalSpecific.TryGetValue(currentThreadHandle, out var values) &&
+            values.TryGetValue(key, out var storedValue))
+        {
+            value = storedValue;
         }
 
         ctx[CpuRegister.Rax] = value;
@@ -1103,7 +1191,7 @@ public static class KernelPthreadExtendedCompatExports
             }
         }
 
-        if (ctx.TryReadUInt64(rwlockAddress, out var pointedHandle) && pointedHandle != 0)
+        if (KernelMemoryCompatExports.TryReadUInt64Compat(ctx, rwlockAddress, out var pointedHandle) && pointedHandle != 0)
         {
             lock (_stateGate)
             {
@@ -1135,7 +1223,7 @@ public static class KernelPthreadExtendedCompatExports
             }
         }
 
-        if (!ctx.TryReadUInt64(rwlockAddress, out var pointedHandle))
+        if (!KernelMemoryCompatExports.TryReadUInt64Compat(ctx, rwlockAddress, out var pointedHandle))
         {
             return false;
         }
@@ -1170,7 +1258,7 @@ public static class KernelPthreadExtendedCompatExports
             _rwlockStates[syntheticHandle] = createdRwlock;
         }
 
-        _ = ctx.TryWriteUInt64(rwlockAddress, syntheticHandle);
+        _ = KernelMemoryCompatExports.TryWriteUInt64Compat(ctx, rwlockAddress, syntheticHandle);
         resolvedAddress = syntheticHandle;
         rwlock = createdRwlock;
         return true;

src/SharpEmu.Libs/Kernel/KernelPthreadState.cs

@@ -1,6 +1,7 @@
 // Copyright (C) 2026 SharpEmu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+using System.Collections.Concurrent;
 using System.Runtime.InteropServices;
 using System.Threading;
 using SharpEmu.HLE;
@@ -11,8 +12,7 @@ internal static class KernelPthreadState
 {
     private const int ThreadObjectSize = 0x1000;
 
-    private static readonly object Gate = new();
-    private static readonly Dictionary<ulong, ThreadIdentity> Threads = new();
+    private static readonly ConcurrentDictionary<ulong, ThreadIdentity> Threads = new();
     private static readonly byte[] ZeroThreadObject = new byte[ThreadObjectSize];
     private static long _nextUniqueThreadId = 1;
 
@@ -56,10 +56,7 @@ internal static class KernelPthreadState
 
     internal static bool TryGetThreadIdentity(ulong threadHandle, out ThreadIdentity identity)
     {
-        lock (Gate)
-        {
-            return Threads.TryGetValue(threadHandle, out identity);
-        }
+        return Threads.TryGetValue(threadHandle, out identity);
     }
 
     private static void EnsureCurrentThreadRegistered()
@@ -81,10 +78,7 @@ internal static class KernelPthreadState
         Marshal.Copy(ZeroThreadObject, 0, pointer, ThreadObjectSize);
 
         var handle = unchecked((ulong)pointer.ToInt64());
-        lock (Gate)
-        {
-            Threads[handle] = new ThreadIdentity(uniqueId, string.IsNullOrWhiteSpace(name) ? $"Thread-{uniqueId:X}" : name);
-        }
+        Threads[handle] = new ThreadIdentity(uniqueId, string.IsNullOrWhiteSpace(name) ? $"Thread-{uniqueId:X}" : name);
 
         return handle;
     }

src/SharpEmu.Libs/Kernel/KernelRuntimeCompatExports.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Libs.Fiber;
 using System.Buffers.Binary;
 using System.Collections.Generic;
 using System.Diagnostics;
@@ -25,12 +26,15 @@ public static class KernelRuntimeCompatExports
     private const ulong TlsNewReplaceOffset = 0x300;
     private const int MallocReplaceSize = 0x70;
     private const int NewReplaceSize = 0x68;
+    private const int OrbisTimesecSize = sizeof(long) + sizeof(uint) + sizeof(uint);
     private const ulong ModuleInfoHandleOffset = 0x108;
     private const ulong ModuleInfoNameOffset = 0x10;
     private const int ModuleInfoNameMaxBytes = 64;
     private const ulong DefaultKernelTscFrequency = 10_000_000UL;
     private const ulong PrtAreaStartAddress = 0x0000001000000000UL;
     private const ulong PrtAreaSize = 0x000000EC00000000UL;
+    private const int MapFlagFixed = 0x10;
+    private const ulong DefaultVirtualRangeAlignment = 0x4000UL;
     private const int AioInitParamSize = 0x3C;
     private const uint MemCommit = 0x1000;
     private const uint MemReserve = 0x2000;
@@ -52,6 +56,14 @@ public static class KernelRuntimeCompatExports
     private static readonly object _prtApertureGate = new();
     private static readonly (ulong Base, ulong Size)[] _prtApertures = new (ulong Base, ulong Size)[3];
     private static int _stackChkFailCount;
+    private static long _usleepTraceCount;
+    private static readonly bool _traceUsleep =
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_USLEEP"), "1", StringComparison.Ordinal);
+    private static readonly bool _traceGuestThreads =
+        string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_GUEST_THREADS"), "1", StringComparison.Ordinal);
+
+    [ThreadStatic]
+    private static int _shortUsleepCount;
 
     [UnmanagedFunctionPointer(CallingConvention.Cdecl)]
     private delegate ulong RdtscDelegate();
@@ -64,33 +76,104 @@ public static class KernelRuntimeCompatExports
     public static int KernelUsleep(CpuContext ctx)
     {
         var micros = ctx[CpuRegister.Rdi];
+        TraceUsleepSpin(ctx, micros);
         if (micros == 0)
         {
             ctx[CpuRegister.Rax] = 0;
             return (int)OrbisGen2Result.ORBIS_GEN2_OK;
         }
 
-        var sleepMilliseconds = (int)Math.Min(micros / 1000UL, int.MaxValue);
-        if (sleepMilliseconds > 0)
-        {
-            Thread.Sleep(sleepMilliseconds);
-        }
+        GuestThreadExecution.Scheduler?.Pump(ctx, "sceKernelUsleep");
 
-        var remainingMicros = micros % 1000UL;
-        if (remainingMicros > 0)
+        if (micros < 1000)
         {
-            var targetTicks = (long)((double)remainingMicros * Stopwatch.Frequency / 1_000_000.0);
-            var spin = Stopwatch.StartNew();
-            while (spin.ElapsedTicks < targetTicks)
+            // Guest worker pools use usleep(1) as a polling backoff. Periodically
+            // relinquish a full host time slice so spin workers cannot starve producers.
+            if ((++_shortUsleepCount & 31) == 0)
             {
-                Thread.SpinWait(16);
+                Thread.Sleep(1);
             }
+            else
+            {
+                Thread.Yield();
+            }
+        }
+        else
+        {
+            _shortUsleepCount = 0;
+            var sleepMilliseconds = (int)Math.Min((micros + 999UL) / 1000UL, int.MaxValue);
+            Thread.Sleep(sleepMilliseconds);
         }
 
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    private static void TraceUsleepSpin(CpuContext ctx, ulong micros)
+    {
+        if (!_traceUsleep)
+        {
+            return;
+        }
+
+        var count = Interlocked.Increment(ref _usleepTraceCount);
+        if (count > 32 && count % 10000 != 0)
+        {
+            return;
+        }
+
+        var rbx = ctx[CpuRegister.Rbx];
+        var r12 = ctx[CpuRegister.R12];
+        var r13 = ctx[CpuRegister.R13];
+        var lockAddress = rbx == 0 ? 0 : rbx + 0xF78;
+        var lockText = "unreadable";
+        if (lockAddress != 0 && ctx.TryReadUInt64(lockAddress, out var lockValue))
+        {
+            lockText = $"0x{lockValue:X16}";
+        }
+
+        var schedulerText = "unreadable";
+        if (r12 != 0 && ctx.TryReadUInt64(r12 + 8, out var schedulerAddress))
+        {
+            schedulerText = $"0x{schedulerAddress:X16}";
+        }
+
+        var waitValueText = "unreadable";
+        if (r13 != 0 && ctx.TryReadUInt64(r13, out var waitValue))
+        {
+            waitValueText = $"0x{waitValue:X16}";
+        }
+
+        var callerReturnText = "unreadable";
+        var rbp = ctx[CpuRegister.Rbp];
+        if (rbp != 0 && ctx.TryReadUInt64(rbp + 8, out var callerReturn))
+        {
+            callerReturnText = $"0x{callerReturn:X16}";
+        }
+
+        var returnRip = GuestThreadExecution.TryGetCurrentImportCallFrame(out var frame)
+            ? frame.ReturnRip
+            : 0UL;
+        var thread = GuestThreadExecution.CurrentGuestThreadHandle;
+        var fiber = FiberExports.GetCurrentFiberAddressForDiagnostics(ctx);
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] usleep#{count}: usec={micros} ret=0x{returnRip:X16} caller={callerReturnText} thread=0x{thread:X16} fiber=0x{fiber:X16} rbx=0x{rbx:X16} lock@+F78=0x{lockAddress:X16}:{lockText} r12=0x{r12:X16} scheduler@+8={schedulerText} r13=0x{r13:X16}:{waitValueText} r14=0x{ctx[CpuRegister.R14]:X16} r15=0x{ctx[CpuRegister.R15]:X16}");
+
+        if (count % 100000 == 0 &&
+            _traceGuestThreads &&
+            GuestThreadExecution.Scheduler is { } scheduler)
+        {
+            foreach (var snapshot in scheduler.SnapshotThreads())
+            {
+                Console.Error.WriteLine(
+                    $"[LOADER][TRACE] guest_thread.snapshot handle=0x{snapshot.ThreadHandle:X16} name='{snapshot.Name}' " +
+                    $"state={snapshot.State} imports={snapshot.ImportCount} nid={snapshot.LastImportNid ?? "none"} " +
+                    $"ret=0x{snapshot.LastReturnRip:X16} block={snapshot.BlockReason ?? "none"}");
+            }
+        }
+    }
+
     [SysAbiExport(
         Nid = "QBi7HCK03hw",
         ExportName = "sceKernelClockGettime",
@@ -156,6 +239,37 @@ public static class KernelRuntimeCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "n88vx3C5nW8",
+        ExportName = "gettimeofday",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixGettimeofday(CpuContext ctx)
+    {
+        var timeAddress = ctx[CpuRegister.Rdi];
+        var timezoneAddress = ctx[CpuRegister.Rsi];
+        var now = DateTimeOffset.UtcNow;
+        var seconds = now.ToUnixTimeSeconds();
+        var microseconds = (now.Ticks % TimeSpan.TicksPerSecond) / 10;
+
+        if (timeAddress != 0 &&
+            (!ctx.TryWriteUInt64(timeAddress, unchecked((ulong)seconds)) ||
+             !ctx.TryWriteUInt64(timeAddress + sizeof(long), unchecked((ulong)microseconds))))
+        {
+            return -1;
+        }
+
+        if (timezoneAddress != 0 &&
+            (!TryWriteInt32(ctx, timezoneAddress, 0) ||
+             !TryWriteInt32(ctx, timezoneAddress + sizeof(int), 0)))
+        {
+            return -1;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return 0;
+    }
+
     [SysAbiExport(
         Nid = "-2IRUCO--PM",
         ExportName = "sceKernelReadTsc",
@@ -497,19 +611,16 @@ public static class KernelRuntimeCompatExports
     public static int ErrorAddress(CpuContext ctx)
     {
         var address = GetTlsScratchAddress(ctx, TlsErrnoOffset);
-        if (address != 0)
-        {
-            Span<byte> zero = stackalloc byte[sizeof(int)];
-            if (!ctx.Memory.TryWrite(address, zero))
-            {
-                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
-            }
-        }
-
         ctx[CpuRegister.Rax] = address;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    internal static bool TrySetErrno(CpuContext ctx, int value)
+    {
+        var address = GetTlsScratchAddress(ctx, TlsErrnoOffset);
+        return address != 0 && TryWriteInt32(ctx, address, value);
+    }
+
     [SysAbiExport(
         Nid = "bnZxYgAFeA0",
         ExportName = "sceKernelGetSanitizerNewReplaceExternal",
@@ -552,6 +663,17 @@ public static class KernelRuntimeCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "jh+8XiK4LeE",
+        ExportName = "sceKernelIsAddressSanitizerEnabled",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelIsAddressSanitizerEnabled(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "ca7v6Cxulzs",
         ExportName = "sceKernelSetGPO",
@@ -590,7 +712,7 @@ public static class KernelRuntimeCompatExports
     {
         var inOutAddressPointer = ctx[CpuRegister.Rdi];
         var length = ctx[CpuRegister.Rsi];
-        var _flags = ctx[CpuRegister.Rdx];
+        var flags = unchecked((int)ctx[CpuRegister.Rdx]);
         var alignment = ctx[CpuRegister.Rcx];
         if (inOutAddressPointer == 0 || length == 0)
         {
@@ -602,7 +724,8 @@ public static class KernelRuntimeCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
 
-        var effectiveAlignment = alignment == 0 ? 0x10000UL : alignment;
+        var effectiveAlignment = alignment == 0 ? DefaultVirtualRangeAlignment : alignment;
+        var fixedMapping = (flags & MapFlagFixed) != 0;
         ulong desiredAddress;
         lock (_stateGate)
         {
@@ -611,13 +734,13 @@ public static class KernelRuntimeCompatExports
                 : AlignUp(_nextReservedVirtualBase, effectiveAlignment);
         }
 
-        if (!TryReserveVirtualRange(ctx, desiredAddress, length, out var mappedAddress))
+        if (!TryReserveVirtualRange(ctx, desiredAddress, length, effectiveAlignment, allowSearch: !fixedMapping, out var mappedAddress))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
         }
 
         Console.Error.WriteLine(
-            $"[LOADER][TRACE] reserve_virtual_range: req=0x{requestedAddress:X16} desired=0x{desiredAddress:X16} mapped=0x{mappedAddress:X16} len=0x{length:X16}");
+            $"[LOADER][TRACE] reserve_virtual_range: req=0x{requestedAddress:X16} desired=0x{desiredAddress:X16} mapped=0x{mappedAddress:X16} len=0x{length:X16} flags=0x{flags:X8} align=0x{effectiveAlignment:X16}");
 
         if (!ctx.TryWriteUInt64(inOutAddressPointer, mappedAddress))
         {
@@ -901,14 +1024,12 @@ public static class KernelRuntimeCompatExports
     public static int StackCheckFail(CpuContext ctx)
     {
         var count = Interlocked.Increment(ref _stackChkFailCount);
-        if (count <= 8)
-        {
-            Console.Error.WriteLine(
-                $"[LOADER][WARNING] __stack_chk_fail recovery#{count}: rip=0x{ctx.Rip:X16} rdi=0x{ctx[CpuRegister.Rdi]:X16}");
-        }
-
-        ctx[CpuRegister.Rax] = 0;
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        Console.Error.WriteLine(
+            $"[LOADER][ERROR] __stack_chk_fail#{count}: rip=0x{ctx.Rip:X16} rdi=0x{ctx[CpuRegister.Rdi]:X16}");
+        var result = (int)OrbisGen2Result.ORBIS_GEN2_ERROR_CPU_TRAP;
+        GuestThreadExecution.RequestCurrentEntryExit("__stack_chk_fail", result);
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
     }
 
     [SysAbiExport(
@@ -1024,6 +1145,111 @@ public static class KernelRuntimeCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "-o5uEDpN+oY",
+        ExportName = "sceKernelConvertUtcToLocaltime",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelConvertUtcToLocaltime(CpuContext ctx)
+    {
+        var utcSeconds = unchecked((long)ctx[CpuRegister.Rdi]);
+        var localTimeAddress = ctx[CpuRegister.Rsi];
+        var timesecAddress = ctx[CpuRegister.Rdx];
+        var dstSecondsAddress = ctx[CpuRegister.Rcx];
+
+        if (localTimeAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var utc = DateTimeOffset.FromUnixTimeSeconds(utcSeconds);
+        var local = TimeZoneInfo.ConvertTime(utc, TimeZoneInfo.Local);
+        var offset = local.Offset;
+        var localSeconds = utcSeconds + (long)offset.TotalSeconds;
+        var dstSeconds = TimeZoneInfo.Local.IsDaylightSavingTime(local.DateTime)
+            ? (uint)Math.Max(0, TimeZoneInfo.Local.GetAdjustmentRules()
+                .Where(rule => rule.DateStart <= local.Date && rule.DateEnd >= local.Date)
+                .Select(rule => rule.DaylightDelta.TotalSeconds)
+                .DefaultIfEmpty(0)
+                .Max())
+            : 0u;
+        var westSeconds = unchecked((uint)(int)offset.TotalSeconds);
+
+        if (!ctx.TryWriteUInt64(localTimeAddress, unchecked((ulong)localSeconds)))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (timesecAddress != 0)
+        {
+            Span<byte> timesec = stackalloc byte[OrbisTimesecSize];
+            BinaryPrimitives.WriteInt64LittleEndian(timesec, utcSeconds);
+            BinaryPrimitives.WriteUInt32LittleEndian(timesec.Slice(sizeof(long), sizeof(uint)), westSeconds);
+            BinaryPrimitives.WriteUInt32LittleEndian(timesec.Slice(sizeof(long) + sizeof(uint), sizeof(uint)), dstSeconds);
+            if (!ctx.Memory.TryWrite(timesecAddress, timesec))
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+        }
+
+        if (dstSecondsAddress != 0 && !ctx.TryWriteUInt64(dstSecondsAddress, dstSeconds))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "0NTHN1NKONI",
+        ExportName = "sceKernelConvertLocaltimeToUtc",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelConvertLocaltimeToUtc(CpuContext ctx)
+    {
+        var localSeconds = unchecked((long)ctx[CpuRegister.Rdi]);
+        var utcTimeAddress = ctx[CpuRegister.Rdx];
+        var timezoneAddress = ctx[CpuRegister.Rcx];
+        var dstSecondsAddress = ctx[CpuRegister.R8];
+
+        if (timezoneAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var localDate = DateTimeOffset.FromUnixTimeSeconds(localSeconds).DateTime;
+        var offset = TimeZoneInfo.Local.GetUtcOffset(localDate);
+        var utcSeconds = localSeconds - (long)offset.TotalSeconds;
+        var dstSeconds = TimeZoneInfo.Local.IsDaylightSavingTime(localDate)
+            ? (int)Math.Max(0, TimeZoneInfo.Local.GetAdjustmentRules()
+                .Where(rule => rule.DateStart <= localDate.Date && rule.DateEnd >= localDate.Date)
+                .Select(rule => rule.DaylightDelta.TotalSeconds)
+                .DefaultIfEmpty(0)
+                .Max())
+            : 0;
+        var minutesWest = unchecked((int)-offset.TotalMinutes);
+
+        if (!TryWriteInt32(ctx, timezoneAddress, minutesWest) ||
+            !TryWriteInt32(ctx, timezoneAddress + sizeof(int), dstSeconds / 60))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (utcTimeAddress != 0 && !ctx.TryWriteUInt64(utcTimeAddress, unchecked((ulong)utcSeconds)))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (dstSecondsAddress != 0 && !TryWriteInt32(ctx, dstSecondsAddress, dstSeconds))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "vYU8P9Td2Zo",
         ExportName = "sceKernelAioInitializeImpl",
@@ -1306,30 +1532,40 @@ public static class KernelRuntimeCompatExports
             ulong.TryParse(overrideHzText, out var overrideHz) &&
             overrideHz >= minSane)
         {
+            TraceKernelTscFrequency("env", overrideHz);
             return overrideHz;
         }
 
-        if (TryResolveCpuidTscFrequency(out ulong cpuidHz) && cpuidHz >= minSane)
-        {
-            return cpuidHz;
-        }
-
         if (TryCalibrateHostTscFrequency(out ulong calibratedHz) && calibratedHz >= minSane)
         {
+            TraceKernelTscFrequency("calibrated-rdtsc", calibratedHz);
             return calibratedHz;
         }
 
+        if (TryResolveCpuidTscFrequency(out ulong cpuidHz) && cpuidHz >= minSane)
+        {
+            TraceKernelTscFrequency("cpuid", cpuidHz);
+            return cpuidHz;
+        }
+
         var hostQpc = Stopwatch.Frequency > 0
             ? unchecked((ulong)Stopwatch.Frequency)
             : DefaultKernelTscFrequency;
         if (hostQpc >= minSane)
         {
+            TraceKernelTscFrequency("qpc", hostQpc);
             return hostQpc;
         }
 
+        TraceKernelTscFrequency("default", DefaultKernelTscFrequency);
         return DefaultKernelTscFrequency;
     }
 
+    private static void TraceKernelTscFrequency(string source, ulong frequencyHz)
+    {
+        Console.Error.WriteLine($"[LOADER][INFO] Kernel TSC frequency: {frequencyHz} Hz ({source})");
+    }
+
     private static bool TryResolveCpuidTscFrequency(out ulong frequencyHz)
     {
         frequencyHz = 0;
@@ -1488,7 +1724,13 @@ public static class KernelRuntimeCompatExports
     [DllImport("kernel32.dll", SetLastError = true)]
     private static extern nint VirtualAlloc(nint lpAddress, nuint dwSize, uint flAllocationType, uint flProtect);
 
-    private static bool TryReserveVirtualRange(CpuContext ctx, ulong desiredAddress, ulong length, out ulong mappedAddress)
+    private static bool TryReserveVirtualRange(
+        CpuContext ctx,
+        ulong desiredAddress,
+        ulong length,
+        ulong alignment,
+        bool allowSearch,
+        out ulong mappedAddress)
     {
         mappedAddress = 0;
         if (length == 0)
@@ -1500,40 +1742,51 @@ public static class KernelRuntimeCompatExports
         {
             object memoryObject = ctx.Memory;
             MethodInfo? allocateAt = null;
+            MethodInfo? allocateAtOrAbove = null;
             var allocateAtHasAllowAlternativeArg = false;
             for (var depth = 0; depth < 4; depth++)
             {
                 foreach (var candidate in memoryObject.GetType().GetMethods(BindingFlags.Public | BindingFlags.Instance))
                 {
-                    if (!string.Equals(candidate.Name, "AllocateAt", StringComparison.Ordinal))
-                    {
-                        continue;
-                    }
-
                     var parameters = candidate.GetParameters();
-                    if (parameters.Length == 3 &&
-                        parameters[0].ParameterType == typeof(ulong) &&
-                        parameters[1].ParameterType == typeof(ulong) &&
-                        parameters[2].ParameterType == typeof(bool))
-                    {
-                        allocateAt = candidate;
-                        allocateAtHasAllowAlternativeArg = false;
-                        break;
-                    }
-
-                    if (parameters.Length == 4 &&
+                    if (string.Equals(candidate.Name, "TryAllocateAtOrAbove", StringComparison.Ordinal) &&
+                        parameters.Length == 5 &&
                         parameters[0].ParameterType == typeof(ulong) &&
                         parameters[1].ParameterType == typeof(ulong) &&
                         parameters[2].ParameterType == typeof(bool) &&
-                        parameters[3].ParameterType == typeof(bool))
+                        parameters[3].ParameterType == typeof(ulong) &&
+                        parameters[4].ParameterType == typeof(ulong).MakeByRefType())
+                    {
+                        allocateAtOrAbove = candidate;
+                    }
+                    else if (string.Equals(candidate.Name, "AllocateAt", StringComparison.Ordinal))
+                    {
+                        if (parameters.Length == 3 &&
+                            parameters[0].ParameterType == typeof(ulong) &&
+                            parameters[1].ParameterType == typeof(ulong) &&
+                            parameters[2].ParameterType == typeof(bool))
+                        {
+                            allocateAt = candidate;
+                            allocateAtHasAllowAlternativeArg = false;
+                        }
+                        else if (parameters.Length == 4 &&
+                            parameters[0].ParameterType == typeof(ulong) &&
+                            parameters[1].ParameterType == typeof(ulong) &&
+                            parameters[2].ParameterType == typeof(bool) &&
+                            parameters[3].ParameterType == typeof(bool))
+                        {
+                            allocateAt = candidate;
+                            allocateAtHasAllowAlternativeArg = true;
+                        }
+                    }
+
+                    if (allocateAtOrAbove is not null && allocateAt is not null)
                     {
-                        allocateAt = candidate;
-                        allocateAtHasAllowAlternativeArg = true;
                         break;
                     }
                 }
 
-                if (allocateAt is not null)
+                if (allocateAtOrAbove is not null || allocateAt is not null)
                 {
                     break;
                 }
@@ -1553,6 +1806,18 @@ public static class KernelRuntimeCompatExports
                 memoryObject = innerValue;
             }
 
+            if (allowSearch && allocateAtOrAbove is not null)
+            {
+                var searchArgs = new object[] { desiredAddress, length, false, alignment, 0UL };
+                var searchResult = allocateAtOrAbove.Invoke(memoryObject, searchArgs);
+                if (searchResult is bool trueValue && trueValue &&
+                    searchArgs[4] is ulong searchedAddress && searchedAddress != 0)
+                {
+                    mappedAddress = searchedAddress;
+                    return true;
+                }
+            }
+
             if (allocateAt is null)
             {
                 Console.Error.WriteLine($"[LOADER][TRACE] reserve_virtual_range: AllocateAt missing on {ctx.Memory.GetType().FullName}");
@@ -1560,7 +1825,7 @@ public static class KernelRuntimeCompatExports
             }
 
             var invokeArgs = allocateAtHasAllowAlternativeArg
-                ? new object[] { desiredAddress, length, false, true }
+                ? new object[] { desiredAddress, length, false, allowSearch }
                 : new object[] { desiredAddress, length, false };
             var result = allocateAt.Invoke(memoryObject, invokeArgs);
             if (result is not ulong allocated || allocated == 0)

src/SharpEmu.Libs/Kernel/KernelSemaphoreCompatExports.cs

@@ -120,7 +120,7 @@ public static class KernelSemaphoreCompatExports
                 return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TIMED_OUT);
             }
 
-            if (!GuestThreadExecution.RequestCurrentThreadBlock("sceKernelWaitSema"))
+            if (!GuestThreadExecution.RequestCurrentThreadBlock(ctx, "sceKernelWaitSema"))
             {
                 TraceSemaphore($"wait-would-block handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
                 return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN);

src/SharpEmu.Libs/Network/Http2Exports.cs

@@ -0,0 +1,79 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using System.Collections.Concurrent;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Network;
+
+public static class Http2Exports
+{
+    private const int Http2ErrorInvalidId = unchecked((int)0x80436004);
+    private const int Http2ErrorInvalidArgument = unchecked((int)0x80436016);
+
+    private static readonly ConcurrentDictionary<int, Http2Context> _contexts = new();
+    private static int _nextContextId;
+
+    private sealed record Http2Context(int NetId, int SslId, ulong PoolSize, int MaxRequests);
+
+    [SysAbiExport(
+        Nid = "3JCe3lCbQ8A",
+        ExportName = "sceHttp2Init",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp2")]
+    public static int Http2Init(CpuContext ctx)
+    {
+        var netId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var sslId = unchecked((int)ctx[CpuRegister.Rsi]);
+        var poolSize = ctx[CpuRegister.Rdx];
+        var maxRequests = unchecked((int)ctx[CpuRegister.Rcx]);
+
+        if (poolSize == 0 || maxRequests <= 0)
+        {
+            return SetReturn(ctx, Http2ErrorInvalidArgument);
+        }
+
+        var id = Interlocked.Increment(ref _nextContextId);
+        _contexts[id] = new Http2Context(netId, sslId, poolSize, maxRequests);
+
+        TraceHttp2("init", id, unchecked((ulong)netId), unchecked((ulong)sslId), poolSize, unchecked((ulong)maxRequests));
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "YiBUtz-pGkc",
+        ExportName = "sceHttp2Term",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp2")]
+    public static int Http2Term(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!_contexts.TryRemove(id, out _))
+        {
+            return SetReturn(ctx, Http2ErrorInvalidId);
+        }
+
+        TraceHttp2("term", id, 0, 0, 0, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceHttp2(string operation, int id, ulong arg0, ulong arg1, ulong arg2, ulong arg3)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_HTTP2"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] http2.{operation} id={id} arg0=0x{arg0:X16} arg1=0x{arg1:X16} arg2=0x{arg2:X16} arg3=0x{arg3:X16}");
+    }
+}

src/SharpEmu.Libs/Network/HttpExports.cs

@@ -0,0 +1,122 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Collections.Concurrent;
+using System.Threading;
+
+namespace SharpEmu.Libs.Network;
+
+public static class HttpExports
+{
+    private const int HttpErrorInvalidId = unchecked((int)0x80431100);
+    private const int HttpErrorInvalidValue = unchecked((int)0x804311FE);
+
+    private static readonly ConcurrentDictionary<int, HttpContext> Contexts = new();
+    private static readonly ConcurrentDictionary<int, HttpTemplate> Templates = new();
+    private static int _nextContextId;
+    private static int _nextTemplateId = 0x1000;
+
+    private sealed record HttpContext(int NetMemoryId, int SslContextId, ulong PoolSize);
+
+    private sealed record HttpTemplate(int ContextId, ulong UserAgentAddress, int HttpVersion, bool AutoProxyConfig);
+
+    [SysAbiExport(
+        Nid = "A9cVMUtEp4Y",
+        ExportName = "sceHttpInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp")]
+    public static int HttpInit(CpuContext ctx)
+    {
+        var netMemoryId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var sslContextId = unchecked((int)ctx[CpuRegister.Rsi]);
+        var poolSize = ctx[CpuRegister.Rdx];
+        if (poolSize == 0)
+        {
+            return SetReturn(ctx, HttpErrorInvalidValue);
+        }
+
+        var id = Interlocked.Increment(ref _nextContextId);
+        Contexts[id] = new HttpContext(netMemoryId, sslContextId, poolSize);
+        TraceHttp("init", id, unchecked((ulong)netMemoryId), unchecked((ulong)sslContextId), poolSize, 0);
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "0gYjPTR-6cY",
+        ExportName = "sceHttpCreateTemplate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp")]
+    public static int HttpCreateTemplate(CpuContext ctx)
+    {
+        var contextId = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!Contexts.ContainsKey(contextId))
+        {
+            return SetReturn(ctx, HttpErrorInvalidId);
+        }
+
+        var userAgentAddress = ctx[CpuRegister.Rsi];
+        var httpVersion = unchecked((int)ctx[CpuRegister.Rdx]);
+        var autoProxyConfig = ctx[CpuRegister.Rcx] != 0;
+        var id = Interlocked.Increment(ref _nextTemplateId);
+        Templates[id] = new HttpTemplate(contextId, userAgentAddress, httpVersion, autoProxyConfig);
+        TraceHttp("create_template", id, unchecked((ulong)contextId), userAgentAddress, unchecked((ulong)httpVersion), autoProxyConfig ? 1UL : 0UL);
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "4I8vEpuEhZ8",
+        ExportName = "sceHttpDeleteTemplate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp")]
+    public static int HttpDeleteTemplate(CpuContext ctx)
+    {
+        var templateId = unchecked((int)ctx[CpuRegister.Rdi]);
+        return Templates.TryRemove(templateId, out _)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, HttpErrorInvalidId);
+    }
+
+    [SysAbiExport(
+        Nid = "Ik-KpLTlf7Q",
+        ExportName = "sceHttpTerm",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceHttp")]
+    public static int HttpTerm(CpuContext ctx)
+    {
+        var contextId = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!Contexts.TryRemove(contextId, out _))
+        {
+            return SetReturn(ctx, HttpErrorInvalidId);
+        }
+
+        foreach (var pair in Templates)
+        {
+            if (pair.Value.ContextId == contextId)
+            {
+                Templates.TryRemove(pair.Key, out _);
+            }
+        }
+
+        return SetReturn(ctx, 0);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceHttp(string operation, int id, ulong arg0, ulong arg1, ulong arg2, ulong arg3)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_HTTP"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] http.{operation} id={id} arg0=0x{arg0:X16} arg1=0x{arg1:X16} arg2=0x{arg2:X16} arg3=0x{arg3:X16}");
+    }
+}

src/SharpEmu.Libs/Network/NetCtlExports.cs

@@ -0,0 +1,229 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+
+namespace SharpEmu.Libs.Network;
+
+public static class NetCtlExports
+{
+    private const int MaxCallbacks = 8;
+    private const int NatInfoSize = 16;
+    private const int NetCtlErrorNoSpace = unchecked((int)0x80412103);
+    private const int NetCtlErrorInvalidAddress = unchecked((int)0x80412107);
+    private const int NetCtlErrorNotConnected = unchecked((int)0x80412108);
+    private const int NetCtlInfoDevice = 1;
+    private const int NetCtlInfoEtherAddress = 2;
+    private const int NetCtlInfoMtu = 3;
+    private const int NetCtlInfoLink = 4;
+    private const int NetCtlInfoIpConfig = 11;
+    private const int NetCtlInfoDhcpHostname = 12;
+    private const int NetCtlInfoPppoeAuthName = 13;
+    private const int NetCtlInfoIpAddress = 14;
+    private const int NetCtlInfoNetmask = 15;
+    private const int NetCtlInfoDefaultRoute = 16;
+    private const int NetCtlInfoPrimaryDns = 17;
+    private const int NetCtlInfoSecondaryDns = 18;
+    private const int NetCtlInfoHttpProxyConfig = 19;
+    private const int NetCtlInfoHttpProxyServer = 20;
+    private const int NetCtlInfoHttpProxyPort = 21;
+    private const int NetCtlDeviceWired = 0;
+    private const int NetCtlLinkDisconnected = 0;
+    private const int NetCtlIpConfigStatic = 0;
+    private static readonly object CallbackGate = new();
+    private static readonly CallbackRegistration[] Callbacks = new CallbackRegistration[MaxCallbacks];
+
+    private readonly record struct CallbackRegistration(ulong Function, ulong Argument);
+
+    [SysAbiExport(
+        Nid = "gky0+oaNM4k",
+        ExportName = "sceNetCtlInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlInit(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "JO4yuTuMoKI",
+        ExportName = "sceNetCtlGetNatInfo",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlGetNatInfo(CpuContext ctx)
+    {
+        var natInfoAddress = ctx[CpuRegister.Rdi];
+        if (natInfoAddress == 0)
+        {
+            return SetReturn(ctx, NetCtlErrorInvalidAddress);
+        }
+
+        Span<byte> natInfo = stackalloc byte[NatInfoSize];
+        if (!ctx.Memory.TryRead(natInfoAddress, natInfo))
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        var size = BinaryPrimitives.ReadUInt32LittleEndian(natInfo[..sizeof(uint)]);
+        if (size != NatInfoSize)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        BinaryPrimitives.WriteInt32LittleEndian(natInfo[4..], 1);
+        BinaryPrimitives.WriteInt32LittleEndian(natInfo[8..], 3);
+        BinaryPrimitives.WriteUInt32LittleEndian(natInfo[12..], 0x7F000001);
+        return ctx.Memory.TryWrite(natInfoAddress, natInfo)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "iQw3iQPhvUQ",
+        ExportName = "sceNetCtlCheckCallback",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlCheckCallback(CpuContext ctx)
+    {
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "uBPlr0lbuiI",
+        ExportName = "sceNetCtlGetState",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlGetState(CpuContext ctx)
+    {
+        var stateAddress = ctx[CpuRegister.Rdi];
+        if (stateAddress == 0)
+        {
+            return SetReturn(ctx, NetCtlErrorInvalidAddress);
+        }
+
+        Span<byte> stateBytes = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(stateBytes, 0);
+        return ctx.Memory.TryWrite(stateAddress, stateBytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "UJ+Z7Q+4ck0",
+        ExportName = "sceNetCtlRegisterCallback",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlRegisterCallback(CpuContext ctx)
+    {
+        var function = ctx[CpuRegister.Rdi];
+        var argument = ctx[CpuRegister.Rsi];
+        var callbackIdAddress = ctx[CpuRegister.Rdx];
+        if (function == 0 || callbackIdAddress == 0)
+        {
+            return SetReturn(ctx, NetCtlErrorInvalidAddress);
+        }
+
+        lock (CallbackGate)
+        {
+            var callbackId = Array.FindIndex(Callbacks, static callback => callback.Function == 0);
+            if (callbackId < 0)
+            {
+                return SetReturn(ctx, NetCtlErrorNoSpace);
+            }
+
+            Span<byte> callbackIdBytes = stackalloc byte[sizeof(uint)];
+            BinaryPrimitives.WriteUInt32LittleEndian(callbackIdBytes, unchecked((uint)callbackId));
+            if (!ctx.Memory.TryWrite(callbackIdAddress, callbackIdBytes))
+            {
+                return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            Callbacks[callbackId] = new CallbackRegistration(function, argument);
+        }
+
+        return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "obuxdTiwkF8",
+        ExportName = "sceNetCtlGetInfo",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNetCtl")]
+    public static int NetCtlGetInfo(CpuContext ctx)
+    {
+        var code = unchecked((int)ctx[CpuRegister.Rdi]);
+        var infoAddress = ctx[CpuRegister.Rsi];
+        if (infoAddress == 0)
+        {
+            return SetReturn(ctx, NetCtlErrorInvalidAddress);
+        }
+
+        return code switch
+        {
+            NetCtlInfoDevice => WriteUInt32(ctx, infoAddress, NetCtlDeviceWired),
+            NetCtlInfoEtherAddress => WriteZeroBytes(ctx, infoAddress, 6),
+            NetCtlInfoMtu => WriteUInt32(ctx, infoAddress, 1500),
+            NetCtlInfoLink => WriteUInt32(ctx, infoAddress, NetCtlLinkDisconnected),
+            NetCtlInfoIpConfig => WriteUInt32(ctx, infoAddress, NetCtlIpConfigStatic),
+            NetCtlInfoDhcpHostname => WriteAsciiZ(ctx, infoAddress, string.Empty, 256),
+            NetCtlInfoPppoeAuthName => WriteAsciiZ(ctx, infoAddress, string.Empty, 128),
+            NetCtlInfoIpAddress => WriteAsciiZ(ctx, infoAddress, "127.0.0.1", 16),
+            NetCtlInfoNetmask => WriteAsciiZ(ctx, infoAddress, "255.0.0.0", 16),
+            NetCtlInfoDefaultRoute => WriteAsciiZ(ctx, infoAddress, "127.0.0.1", 16),
+            NetCtlInfoPrimaryDns => WriteAsciiZ(ctx, infoAddress, "1.1.1.1", 16),
+            NetCtlInfoSecondaryDns => WriteAsciiZ(ctx, infoAddress, "1.1.1.1", 16),
+            NetCtlInfoHttpProxyConfig => WriteUInt32(ctx, infoAddress, 0),
+            NetCtlInfoHttpProxyServer => WriteAsciiZ(ctx, infoAddress, string.Empty, 256),
+            NetCtlInfoHttpProxyPort => WriteUInt16(ctx, infoAddress, 0),
+            _ => SetReturn(ctx, NetCtlErrorNotConnected),
+        };
+    }
+
+    private static int WriteZeroBytes(CpuContext ctx, ulong address, int count)
+    {
+        Span<byte> bytes = stackalloc byte[count];
+        return ctx.Memory.TryWrite(address, bytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static int WriteUInt32(CpuContext ctx, ulong address, uint value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static int WriteUInt16(CpuContext ctx, ulong address, ushort value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(ushort)];
+        BinaryPrimitives.WriteUInt16LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static int WriteAsciiZ(CpuContext ctx, ulong address, string value, int byteCount)
+    {
+        Span<byte> bytes = stackalloc byte[byteCount];
+        var copyCount = Math.Min(value.Length, byteCount - 1);
+        for (var i = 0; i < copyCount; i++)
+        {
+            bytes[i] = (byte)value[i];
+        }
+
+        return ctx.Memory.TryWrite(address, bytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(long)result);
+        return result;
+    }
+}

src/SharpEmu.Libs/Network/NetExports.cs

@@ -0,0 +1,210 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using System.Buffers.Binary;
+using System.Collections.Concurrent;
+using System.Text;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Network;
+
+public static class NetExports
+{
+    private const int NetErrorBadFileDescriptor = unchecked((int)0x80410109);
+    private const int NetErrorInvalidArgument = unchecked((int)0x80410116);
+    private const int MaxNameLength = 256;
+
+    private static readonly ConcurrentDictionary<int, NetPool> _pools = new();
+    private static readonly ConcurrentDictionary<int, ResolverContext> _resolvers = new();
+    private static int _nextPoolId;
+    private static int _nextResolverId = 0x2000;
+    private static bool _initialized;
+
+    private sealed record NetPool(string Name, int Size, int Flags);
+
+    private sealed record ResolverContext(string Name, int PoolId, int Flags, int LastError);
+
+    [SysAbiExport(
+        Nid = "Nlev7Lg8k3A",
+        ExportName = "sceNetInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetInit(CpuContext ctx)
+    {
+        _initialized = true;
+        TraceNet("init", 0, 0, 0, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "cTGkc6-TBlI",
+        ExportName = "sceNetTerm",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetTerm(CpuContext ctx)
+    {
+        _initialized = false;
+        _pools.Clear();
+        _resolvers.Clear();
+        TraceNet("term", 0, 0, 0, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "dgJBaeJnGpo",
+        ExportName = "sceNetPoolCreate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetPoolCreate(CpuContext ctx)
+    {
+        var nameAddress = ctx[CpuRegister.Rdi];
+        var size = unchecked((int)ctx[CpuRegister.Rsi]);
+        var flags = unchecked((int)ctx[CpuRegister.Rdx]);
+
+        if (size <= 0)
+        {
+            return SetReturn(ctx, NetErrorInvalidArgument);
+        }
+
+        var name = TryReadUtf8Z(ctx, nameAddress, MaxNameLength, out var value)
+            ? value
+            : string.Empty;
+
+        var id = Interlocked.Increment(ref _nextPoolId);
+        _pools[id] = new NetPool(name, size, flags);
+
+        TraceNet("pool.create", id, unchecked((ulong)size), unchecked((ulong)flags), _initialized ? 1UL : 0UL);
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "K7RlrTkI-mw",
+        ExportName = "sceNetPoolDestroy",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetPoolDestroy(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!_pools.TryRemove(id, out _))
+        {
+            return SetReturn(ctx, NetErrorBadFileDescriptor);
+        }
+
+        TraceNet("pool.destroy", id, 0, 0, _initialized ? 1UL : 0UL);
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "C4UgDHHPvdw",
+        ExportName = "sceNetResolverCreate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetResolverCreate(CpuContext ctx)
+    {
+        var nameAddress = ctx[CpuRegister.Rdi];
+        var poolId = unchecked((int)ctx[CpuRegister.Rsi]);
+        var flags = unchecked((int)ctx[CpuRegister.Rdx]);
+        if (flags != 0)
+        {
+            return SetReturn(ctx, NetErrorInvalidArgument);
+        }
+
+        var name = TryReadUtf8Z(ctx, nameAddress, MaxNameLength, out var value)
+            ? value
+            : string.Empty;
+        var id = Interlocked.Increment(ref _nextResolverId);
+        _resolvers[id] = new ResolverContext(name, poolId, flags, 0);
+        TraceNet("resolver.create", id, unchecked((ulong)poolId), unchecked((ulong)flags), _initialized ? 1UL : 0UL);
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "kJlYH5uMAWI",
+        ExportName = "sceNetResolverDestroy",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetResolverDestroy(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        return _resolvers.TryRemove(id, out _)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, NetErrorBadFileDescriptor);
+    }
+
+    [SysAbiExport(
+        Nid = "J5i3hiLJMPk",
+        ExportName = "sceNetResolverGetError",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNet")]
+    public static int NetResolverGetError(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        var statusAddress = ctx[CpuRegister.Rsi];
+        if (statusAddress == 0)
+        {
+            return SetReturn(ctx, NetErrorInvalidArgument);
+        }
+
+        if (!_resolvers.TryGetValue(id, out var resolver))
+        {
+            return SetReturn(ctx, NetErrorBadFileDescriptor);
+        }
+
+        Span<byte> status = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(status, resolver.LastError);
+        return ctx.Memory.TryWrite(statusAddress, status)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static bool TryReadUtf8Z(CpuContext ctx, ulong address, int maxLength, out string value)
+    {
+        value = string.Empty;
+        if (address == 0)
+        {
+            return true;
+        }
+
+        Span<byte> one = stackalloc byte[1];
+        var bytes = new byte[maxLength];
+        var count = 0;
+        for (; count < maxLength; count++)
+        {
+            if (!ctx.Memory.TryRead(address + (ulong)count, one))
+            {
+                return false;
+            }
+
+            if (one[0] == 0)
+            {
+                break;
+            }
+
+            bytes[count] = one[0];
+        }
+
+        value = Encoding.UTF8.GetString(bytes, 0, count);
+        return true;
+    }
+
+    private static void TraceNet(string operation, int id, ulong arg0, ulong arg1, ulong arg2)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_NET"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] net.{operation} id={id} arg0=0x{arg0:X16} arg1=0x{arg1:X16} arg2=0x{arg2:X16}");
+    }
+}

src/SharpEmu.Libs/Network/SslExports.cs

@@ -0,0 +1,87 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using System.Collections.Concurrent;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Network;
+
+public static class SslExports
+{
+    private const int SslErrorInvalidId = unchecked((int)0x8095F006);
+    private const int SslErrorOutOfSize = unchecked((int)0x8095F008);
+
+    private static readonly ConcurrentDictionary<int, SslContext> _contexts = new();
+    private static int _nextContextId;
+
+    private sealed record SslContext(ulong PoolSize);
+
+    [SysAbiExport(
+        Nid = "hdpVEUDFW3s",
+        ExportName = "sceSslInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSsl")]
+    public static int SslInit(CpuContext ctx)
+    {
+        var poolSize = ctx[CpuRegister.Rdi];
+        if (poolSize == 0)
+        {
+            return SetReturn(ctx, SslErrorOutOfSize);
+        }
+
+        var id = Interlocked.Increment(ref _nextContextId);
+        _contexts[id] = new SslContext(poolSize);
+
+        TraceSsl("init", id, poolSize);
+        ctx[CpuRegister.Rax] = unchecked((ulong)id);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "0K1yQ6Lv-Yc",
+        ExportName = "sceSslTerm",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSsl")]
+    public static int SslTerm(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!_contexts.TryRemove(id, out _))
+        {
+            return SetReturn(ctx, SslErrorInvalidId);
+        }
+
+        TraceSsl("term", id, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "viRXSHZYd0c",
+        ExportName = "sceSslClose",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSsl")]
+    public static int SslClose(CpuContext ctx)
+    {
+        var id = unchecked((int)ctx[CpuRegister.Rdi]);
+        TraceSsl("close", id, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceSsl(string operation, int id, ulong arg0)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_SSL"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] ssl.{operation} id={id} arg0=0x{arg0:X16}");
+    }
+}

src/SharpEmu.Libs/Np/NpEntitlementAccessExports.cs

@@ -0,0 +1,76 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Np;
+
+public static class NpEntitlementAccessExports
+{
+    private const int BootParamClearSize = 0x20;
+    private const int EmptyAddcontInfoListSize = 0x10;
+
+    [SysAbiExport(
+        Nid = "jO8DM8oyego",
+        ExportName = "sceNpEntitlementAccessInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpEntitlementAccess")]
+    public static int NpEntitlementAccessInitialize(CpuContext ctx)
+    {
+        var initParam = ctx[CpuRegister.Rdi];
+        var bootParam = ctx[CpuRegister.Rsi];
+
+        if (bootParam != 0)
+        {
+            Span<byte> clear = stackalloc byte[BootParamClearSize];
+            clear.Clear();
+            if (!ctx.Memory.TryWrite(bootParam, clear))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+        }
+
+        TraceNpEntitlementAccess($"initialize init=0x{initParam:X16} boot=0x{bootParam:X16}");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "TFyU+KFBv54",
+        ExportName = "sceNpEntitlementAccessGetAddcontEntitlementInfoList",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpEntitlementAccess")]
+    public static int NpEntitlementAccessGetAddcontEntitlementInfoList(CpuContext ctx)
+    {
+        var listAddress = ctx[CpuRegister.Rsi];
+        if (listAddress != 0)
+        {
+            Span<byte> emptyList = stackalloc byte[EmptyAddcontInfoListSize];
+            emptyList.Clear();
+            if (!ctx.Memory.TryWrite(listAddress, emptyList))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+        }
+
+        TraceNpEntitlementAccess(
+            $"get_addcont_info_list service=0x{ctx[CpuRegister.Rdi]:X16} list=0x{listAddress:X16} " +
+            $"max={ctx[CpuRegister.Rdx]} flags=0x{ctx[CpuRegister.Rcx]:X16} -> empty");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+        return (int)result;
+    }
+
+    private static void TraceNpEntitlementAccess(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_NP"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] np.entitlement.{message}");
+    }
+}

src/SharpEmu.Libs/Np/NpManagerExports.cs

@@ -0,0 +1,143 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+
+namespace SharpEmu.Libs.Np;
+
+public static class NpManagerExports
+{
+    private const int NpTitleIdSize = 16;
+    private const int NpTitleSecretSize = 128;
+
+    [SysAbiExport(
+        Nid = "3Zl8BePTh9Y",
+        ExportName = "sceNpCheckCallback",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpCheckCallback(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "JELHf4xPufo",
+        ExportName = "sceNpCheckCallbackForLib",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpCheckCallbackForLib(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "VfRSmPmj8Q8",
+        ExportName = "sceNpRegisterStateCallback",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpRegisterStateCallback(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "qQJfO8HAiaY",
+        ExportName = "sceNpRegisterStateCallbackA",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpRegisterStateCallbackA(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "0c7HbXRKUt4",
+        ExportName = "sceNpRegisterStateCallbackForToolkit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManagerForToolkit")]
+    public static int NpRegisterStateCallbackForToolkit(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "eQH7nWPcAgc",
+        ExportName = "sceNpGetState",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpGetState(CpuContext ctx)
+    {
+        var stateAddress = ctx[CpuRegister.Rsi];
+        if (stateAddress == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> stateBytes = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(stateBytes, 1);
+        return ctx.Memory.TryWrite(stateAddress, stateBytes)
+            ? SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK)
+            : SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "Ec63y59l9tw",
+        ExportName = "sceNpSetNpTitleId",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpManager")]
+    public static int NpSetNpTitleId(CpuContext ctx)
+    {
+        var titleIdAddress = ctx[CpuRegister.Rdi];
+        var titleSecretAddress = ctx[CpuRegister.Rsi];
+        if (titleIdAddress == 0 || titleSecretAddress == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> titleId = stackalloc byte[NpTitleIdSize];
+        Span<byte> titleSecret = stackalloc byte[NpTitleSecretSize];
+        if (!ctx.Memory.TryRead(titleIdAddress, titleId) ||
+            !ctx.Memory.TryRead(titleSecretAddress, titleSecret))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        TraceNp($"set_np_title_id title='{ReadTitleId(titleId)}'");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+        return (int)result;
+    }
+
+    private static string ReadTitleId(ReadOnlySpan<byte> bytes)
+    {
+        var length = 0;
+        while (length < 12 && length < bytes.Length && bytes[length] != 0)
+        {
+            length++;
+        }
+
+        return length == 0
+            ? string.Empty
+            : System.Text.Encoding.ASCII.GetString(bytes[..length]);
+    }
+
+    private static void TraceNp(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_NP"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] np.{message}");
+    }
+}

src/SharpEmu.Libs/Np/NpSessionSignalingExports.cs

@@ -0,0 +1,20 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Np;
+
+public static class NpSessionSignalingExports
+{
+    [SysAbiExport(
+        Nid = "ysmw6J-P8Ak",
+        ExportName = "sceNpSessionSignalingInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpSessionSignaling")]
+    public static int NpSessionSignalingInitialize(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+}

src/SharpEmu.Libs/Np/NpUniversalDataSystemExports.cs

@@ -0,0 +1,98 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Threading;
+
+namespace SharpEmu.Libs.Np;
+
+public static class NpUniversalDataSystemExports
+{
+    private const int NpUniversalDataSystemErrorInvalidArgument = unchecked((int)0x80553102);
+    private static int _nextHandle = 1;
+
+    [SysAbiExport(
+        Nid = "sjaobBgqeB4",
+        ExportName = "sceNpUniversalDataSystemInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpUniversalDataSystem")]
+    public static int NpUniversalDataSystemInitialize(CpuContext ctx)
+    {
+        var parameterAddress = ctx[CpuRegister.Rdi];
+        if (parameterAddress == 0)
+        {
+            return SetReturn(ctx, NpUniversalDataSystemErrorInvalidArgument);
+        }
+
+        Span<byte> parameters = stackalloc byte[16];
+        return ctx.Memory.TryRead(parameterAddress, parameters)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "5zBnau1uIEo",
+        ExportName = "sceNpUniversalDataSystemCreateContext",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpUniversalDataSystem")]
+    public static int NpUniversalDataSystemCreateContext(CpuContext ctx)
+    {
+        var contextAddress = ctx[CpuRegister.Rdi];
+        if (contextAddress == 0)
+        {
+            return SetReturn(ctx, 0);
+        }
+
+        Span<byte> context = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(context, 1);
+        return ctx.Memory.TryWrite(contextAddress, context)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "hT0IAEvN+M0",
+        ExportName = "sceNpUniversalDataSystemCreateHandle",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpUniversalDataSystem")]
+    public static int NpUniversalDataSystemCreateHandle(CpuContext ctx)
+    {
+        var handle = Interlocked.Increment(ref _nextHandle);
+        if (TryWriteInt32(ctx, ctx[CpuRegister.Rdi], handle) ||
+            TryWriteInt32(ctx, ctx[CpuRegister.Rsi], handle))
+        {
+            return SetReturn(ctx, 0);
+        }
+
+        return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "tpFJ8LIKvPw",
+        ExportName = "sceNpUniversalDataSystemRegisterContext",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpUniversalDataSystem")]
+    public static int NpUniversalDataSystemRegisterContext(CpuContext ctx)
+    {
+        return SetReturn(ctx, 0);
+    }
+
+    private static bool TryWriteInt32(CpuContext ctx, ulong address, int value)
+    {
+        if (address == 0)
+        {
+            return false;
+        }
+
+        Span<byte> bytes = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(long)result);
+        return result;
+    }
+}

src/SharpEmu.Libs/Np/NpWebApi2Exports.cs

@@ -0,0 +1,65 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Np;
+
+public static class NpWebApi2Exports
+{
+    private const int NpWebApi2ErrorInvalidArgument = unchecked((int)0x80553402);
+
+    private static int _initialized;
+
+    [SysAbiExport(
+        Nid = "+o9816YQhqQ",
+        ExportName = "sceNpWebApi2Initialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpWebApi2")]
+    public static int NpWebApi2Initialize(CpuContext ctx)
+    {
+        var httpContextId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var poolSize = ctx[CpuRegister.Rsi];
+
+        if (httpContextId <= 0 || poolSize == 0)
+        {
+            return SetReturn(ctx, NpWebApi2ErrorInvalidArgument);
+        }
+
+        Interlocked.Exchange(ref _initialized, 1);
+        TraceNpWebApi2("init", httpContextId, poolSize);
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "bEvXpcEk200",
+        ExportName = "sceNpWebApi2Terminate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpWebApi2")]
+    public static int NpWebApi2Terminate(CpuContext ctx)
+    {
+        var libraryContextId = unchecked((int)ctx[CpuRegister.Rdi]);
+        Interlocked.Exchange(ref _initialized, 0);
+        TraceNpWebApi2("term", libraryContextId, 0);
+        return SetReturn(ctx, 0);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceNpWebApi2(string operation, int id, ulong arg0)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_NP_WEB_API2"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] npwebapi2.{operation} id={id} arg0=0x{arg0:X16} initialized={Volatile.Read(ref _initialized)}");
+    }
+}

src/SharpEmu.Libs/NpGameIntent/NpGameIntentExports.cs

@@ -0,0 +1,24 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.NpGameIntent;
+
+public static class NpGameIntentExports
+{
+    private static int _initialized;
+
+    [SysAbiExport(
+        Nid = "m87BHxt-H60",
+        ExportName = "sceNpGameIntentInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceNpGameIntent")]
+    public static int NpGameIntentInitialize(CpuContext ctx)
+    {
+        Interlocked.Exchange(ref _initialized, 1);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+}

src/SharpEmu.Libs/Pad/PadExports.cs

@@ -128,6 +128,38 @@ public static class PadExports
             return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
         }
 
+        return WriteNeutralPadData(ctx, dataAddress)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "q1cHNfGycLI",
+        ExportName = "scePadRead",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadRead(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        var dataAddress = ctx[CpuRegister.Rsi];
+        var count = unchecked((int)ctx[CpuRegister.Rdx]);
+        if (handle != PrimaryPadHandle)
+        {
+            return SetReturn(ctx, OrbisPadErrorInvalidHandle);
+        }
+
+        if (dataAddress == 0 || count < 1 || count > 64)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        return WriteNeutralPadData(ctx, dataAddress)
+            ? SetReturn(ctx, 1)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    private static bool WriteNeutralPadData(CpuContext ctx, ulong dataAddress)
+    {
         Span<byte> data = stackalloc byte[PadDataSize];
         data.Clear();
         data[0x04] = 128;
@@ -145,9 +177,7 @@ public static class PadExports
             timestampMicroseconds);
         data[0x68] = 1;
 
-        return ctx.Memory.TryWrite(dataAddress, data)
-            ? SetReturn(ctx, 0)
-            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        return ctx.Memory.TryWrite(dataAddress, data);
     }
 
     private static int SetReturn(CpuContext ctx, int result)

src/SharpEmu.Libs/PlayGo/PlayGoExports.cs

@@ -3,6 +3,7 @@
 
 using SharpEmu.HLE;
 using System.Buffers.Binary;
+using System.Text.RegularExpressions;
 
 namespace SharpEmu.Libs.PlayGo;
 
@@ -14,15 +15,38 @@ public static class PlayGoExports
     private const int OrbisPlayGoErrorBadHandle = unchecked((int)0x80B20009);
     private const int OrbisPlayGoErrorBadPointer = unchecked((int)0x80B2000A);
     private const int OrbisPlayGoErrorBadSize = unchecked((int)0x80B2000B);
+    private const int OrbisPlayGoErrorBadChunkId = unchecked((int)0x80B2000C);
     private const int OrbisPlayGoErrorNotSupportPlayGo = unchecked((int)0x80B2000E);
+    private const int OrbisPlayGoErrorBadLocus = unchecked((int)0x80B20010);
     private const ulong PlayGoInitBufAddrOffset = 0;
     private const ulong PlayGoInitBufSizeOffset = 8;
     private const uint PlayGoMinimumInitBufferSize = 0x200000;
     private const uint PlayGoHandle = 1;
+    private const int PlayGoLocusNotDownloaded = 0;
+    private const int PlayGoLocusLocalSlow = 2;
+    private const int PlayGoLocusLocalFast = 3;
+    private const int PlayGoInstallSpeedSuspended = 0;
+    private const int PlayGoInstallSpeedTrickle = 1;
+    private const int PlayGoInstallSpeedFull = 2;
+    private const uint MaxPlayGoQueryEntries = 0x4000;
+    private const uint PlayGoAllEntriesSentinel = uint.MaxValue;
+
+    private static readonly Regex ChunkIdPattern = new(
+        @"<chunk\s+[^>]*\bid\s*=\s*""(?<id>\d+)""",
+        RegexOptions.CultureInvariant);
+
+    private static readonly Regex DefaultChunkPattern = new(
+        @"default_chunk\s*=\s*""(?<id>\d+)""",
+        RegexOptions.CultureInvariant);
 
     private static readonly object _stateGate = new();
     private static bool _initialized;
-    private static bool _hasPlayGoData;
+    private static bool _opened;
+    private static PlayGoMetadata _metadata = PlayGoMetadata.Empty;
+    private static int _installSpeed = PlayGoInstallSpeedTrickle;
+    private static ulong _languageMask = ulong.MaxValue;
+    private static int _unknownChunkDiagnostics;
+    private static int _locusTraceDiagnostics;
 
     [SysAbiExport(
         Nid = "ts6GlZOKRrE",
@@ -67,8 +91,12 @@ public static class PlayGoExports
                 return OrbisPlayGoErrorAlreadyInitialized;
             }
 
-            _hasPlayGoData = HasPlayGoChunkData();
+            _metadata = LoadPlayGoMetadata();
+            _installSpeed = PlayGoInstallSpeedTrickle;
+            _languageMask = ulong.MaxValue;
+            _opened = false;
             _initialized = true;
+            TracePlayGo($"initialize chunks={_metadata.ChunkIds.Length} available={_metadata.Available}");
         }
 
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -100,10 +128,13 @@ public static class PlayGoExports
                 return OrbisPlayGoErrorNotInitialized;
             }
 
-            if (!_hasPlayGoData)
+            if (!_metadata.Available)
             {
                 return OrbisPlayGoErrorNotSupportPlayGo;
             }
+
+            _opened = true;
+            TracePlayGo($"open handle={PlayGoHandle} chunks={_metadata.ChunkIds.Length}");
         }
 
         Span<byte> handleBytes = stackalloc byte[sizeof(uint)];
@@ -132,7 +163,8 @@ public static class PlayGoExports
             }
 
             _initialized = false;
-            _hasPlayGoData = false;
+            _opened = false;
+            _metadata = PlayGoMetadata.Empty;
         }
 
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
@@ -157,20 +189,611 @@ public static class PlayGoExports
             {
                 return OrbisPlayGoErrorBadHandle;
             }
+
+            _opened = false;
         }
 
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
-    private static bool HasPlayGoChunkData()
+    [SysAbiExport(
+        Nid = "73fF1MFU8hA",
+        ExportName = "scePlayGoGetChunkId",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetChunkId(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var outChunkIdList = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var outEntries = ctx[CpuRegister.Rcx];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (outEntries == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (outChunkIdList != 0 && numberOfEntries == 0)
+        {
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        ushort[] chunkIds;
+        lock (_stateGate)
+        {
+            chunkIds = _metadata.ChunkIds;
+        }
+
+        var availableEntries = chunkIds.Length == 0 ? 1u : (uint)chunkIds.Length;
+        if (outChunkIdList == 0)
+        {
+            TracePlayGo($"get_chunk_id count_only entries={availableEntries} out_entries=0x{outEntries:X16}");
+            return TryWriteUInt32(ctx, outEntries, availableEntries)
+                ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+                : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        var entriesToWrite = Math.Min(numberOfEntries, availableEntries);
+        if (entriesToWrite > MaxPlayGoQueryEntries)
+        {
+            TracePlayGo($"get_chunk_id bad_size requested={numberOfEntries} available={availableEntries}");
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        for (uint i = 0; i < entriesToWrite; i++)
+        {
+            var chunkId = chunkIds.Length == 0 ? (ushort)0 : chunkIds[i];
+            if (!TryWriteUInt16(ctx, outChunkIdList + (i * sizeof(ushort)), chunkId))
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+        }
+
+        TracePlayGo($"get_chunk_id write requested={numberOfEntries} wrote={entriesToWrite} available={availableEntries}");
+        return TryWriteUInt32(ctx, outEntries, entriesToWrite)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "v6EZ-YWRdMs",
+        ExportName = "scePlayGoGetEta",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetEta(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var chunkIds = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var outEta = ctx[CpuRegister.Rcx];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (chunkIds == 0 || outEta == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (numberOfEntries == 0 || numberOfEntries > MaxPlayGoQueryEntries)
+        {
+            TracePlayGo($"get_eta bad_size entries={numberOfEntries} chunk_ids=0x{chunkIds:X16} out=0x{outEta:X16}");
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        return ValidateChunkIds(ctx, chunkIds, numberOfEntries) is { } chunkError && chunkError != 0
+            ? chunkError
+            : TryWriteInt64(ctx, outEta, 0)
+                ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+                : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "rvBSfTimejE",
+        ExportName = "scePlayGoGetInstallSpeed",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetInstallSpeed(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var outSpeed = ctx[CpuRegister.Rsi];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (outSpeed == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        int speed;
+        lock (_stateGate)
+        {
+            speed = _installSpeed;
+        }
+
+        return TryWriteInt32(ctx, outSpeed, speed)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "3OMbYZBaa50",
+        ExportName = "scePlayGoGetLanguageMask",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetLanguageMask(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var outLanguageMask = ctx[CpuRegister.Rsi];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (outLanguageMask == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        ulong languageMask;
+        lock (_stateGate)
+        {
+            languageMask = _languageMask;
+        }
+
+        return ctx.TryWriteUInt64(outLanguageMask, languageMask)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "uWIYLFkkwqk",
+        ExportName = "scePlayGoGetLocus",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetLocus(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var chunkIds = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var outLoci = ctx[CpuRegister.Rcx];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (chunkIds == 0 || outLoci == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (numberOfEntries == PlayGoAllEntriesSentinel)
+        {
+            TracePlayGo($"get_locus sentinel entries={numberOfEntries} chunk_ids=0x{chunkIds:X16} out=0x{outLoci:X16}");
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (numberOfEntries == 0 || numberOfEntries > MaxPlayGoQueryEntries)
+        {
+            TracePlayGo($"get_locus bad_size entries={numberOfEntries} chunk_ids=0x{chunkIds:X16} out=0x{outLoci:X16}");
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        var loci = new byte[numberOfEntries];
+        for (uint i = 0; i < numberOfEntries; i++)
+        {
+            if (!TryReadUInt16(ctx, chunkIds + (i * sizeof(ushort)), out var chunkId))
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+
+            if (!IsKnownChunkId(chunkId))
+            {
+                if (Interlocked.Increment(ref _unknownChunkDiagnostics) <= 8)
+                {
+                    ushort[] knownChunkIds;
+                    lock (_stateGate)
+                    {
+                        knownChunkIds = _metadata.ChunkIds;
+                    }
+
+                    Console.Error.WriteLine(
+                        $"[LOADER][TRACE] playgo.unknown_chunk_id id={chunkId} entries={numberOfEntries} " +
+                        $"known=[{string.Join(',', knownChunkIds)}]");
+                }
+            }
+
+            loci[i] = PlayGoLocusLocalFast;
+        }
+
+        TracePlayGoLocus(numberOfEntries, chunkIds, outLoci);
+        return ctx.Memory.TryWrite(outLoci, loci)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "-RJWNMK3fC8",
+        ExportName = "scePlayGoGetProgress",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetProgress(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var chunkIds = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var outProgress = ctx[CpuRegister.Rcx];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (chunkIds == 0 || outProgress == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (numberOfEntries == 0 || numberOfEntries > MaxPlayGoQueryEntries)
+        {
+            TracePlayGo($"get_progress bad_size entries={numberOfEntries} chunk_ids=0x{chunkIds:X16} out=0x{outProgress:X16}");
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        var chunkError = ValidateChunkIds(ctx, chunkIds, numberOfEntries);
+        if (chunkError != 0)
+        {
+            return chunkError;
+        }
+
+        TracePlayGo($"get_progress entries={numberOfEntries}");
+        Span<byte> progress = stackalloc byte[sizeof(ulong) * 2];
+        BinaryPrimitives.WriteUInt64LittleEndian(progress, 0);
+        BinaryPrimitives.WriteUInt64LittleEndian(progress[sizeof(ulong)..], 0);
+        return ctx.Memory.TryWrite(outProgress, progress)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "Nn7zKwnA5q0",
+        ExportName = "scePlayGoGetToDoList",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoGetToDoList(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var outTodoList = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var outEntries = ctx[CpuRegister.Rcx];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (outTodoList == 0 || outEntries == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (numberOfEntries == 0)
+        {
+            TracePlayGo("get_todo bad_size entries=0");
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        TracePlayGo($"get_todo requested={numberOfEntries} wrote=0");
+        return TryWriteUInt32(ctx, outEntries, 0)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "-Q1-u1a7p0g",
+        ExportName = "scePlayGoPrefetch",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoPrefetch(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var chunkIds = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var minimumLocus = unchecked((int)ctx[CpuRegister.Rcx]);
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (chunkIds == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        if (numberOfEntries == 0 || numberOfEntries > MaxPlayGoQueryEntries)
+        {
+            return OrbisPlayGoErrorBadSize;
+        }
+
+        if (minimumLocus is not PlayGoLocusNotDownloaded and not PlayGoLocusLocalSlow and not PlayGoLocusLocalFast)
+        {
+            return OrbisPlayGoErrorBadLocus;
+        }
+
+        return ValidateChunkIds(ctx, chunkIds, numberOfEntries) is { } chunkError && chunkError != 0
+            ? chunkError
+            : (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "4AAcTU9R3XM",
+        ExportName = "scePlayGoSetInstallSpeed",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoSetInstallSpeed(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var speed = unchecked((int)ctx[CpuRegister.Rsi]);
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (speed is not PlayGoInstallSpeedSuspended and not PlayGoInstallSpeedTrickle and not PlayGoInstallSpeedFull)
+        {
+            return OrbisPlayGoErrorInvalidArgument;
+        }
+
+        lock (_stateGate)
+        {
+            _installSpeed = speed;
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "LosLlHOpNqQ",
+        ExportName = "scePlayGoSetLanguageMask",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoSetLanguageMask(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var languageMask = ctx[CpuRegister.Rsi];
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        lock (_stateGate)
+        {
+            _languageMask = languageMask;
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "gUPGiOQ1tmQ",
+        ExportName = "scePlayGoSetToDoList",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePlayGo")]
+    public static int PlayGoSetToDoList(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var todoList = ctx[CpuRegister.Rsi];
+        var numberOfEntries = unchecked((uint)ctx[CpuRegister.Rdx]);
+
+        var validation = ValidateHandle(handle);
+        if (validation != 0)
+        {
+            return validation;
+        }
+
+        if (todoList == 0)
+        {
+            return OrbisPlayGoErrorBadPointer;
+        }
+
+        return numberOfEntries == 0
+            ? OrbisPlayGoErrorBadSize
+            : (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static int ValidateHandle(uint handle)
+    {
+        lock (_stateGate)
+        {
+            if (!_initialized)
+            {
+                return OrbisPlayGoErrorNotInitialized;
+            }
+
+            if (handle != PlayGoHandle || !_opened)
+            {
+                return OrbisPlayGoErrorBadHandle;
+            }
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static int ValidateChunkIds(CpuContext ctx, ulong chunkIds, uint numberOfEntries)
+    {
+        for (uint i = 0; i < numberOfEntries; i++)
+        {
+            if (!TryReadUInt16(ctx, chunkIds + (i * sizeof(ushort)), out var chunkId))
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+            }
+
+            _ = chunkId;
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static bool IsKnownChunkId(ushort chunkId)
+    {
+        lock (_stateGate)
+        {
+            return _metadata.ChunkIds.Length == 0 || Array.BinarySearch(_metadata.ChunkIds, chunkId) >= 0;
+        }
+    }
+
+    private static PlayGoMetadata LoadPlayGoMetadata()
     {
         var app0Root = Environment.GetEnvironmentVariable("SHARPEMU_APP0_DIR");
         if (string.IsNullOrWhiteSpace(app0Root))
         {
+            return PlayGoMetadata.Empty;
+        }
+
+        var playGoDat = Path.Combine(app0Root, "sce_sys", "playgo-chunk.dat");
+        var scenarioJson = Path.Combine(app0Root, "sce_sys", "playgo-scenario.json");
+        var chunkDefsXml = Path.Combine(app0Root, "playgo-chunkdefs.xml");
+
+        var hasMetadata = File.Exists(playGoDat) || File.Exists(scenarioJson) || File.Exists(chunkDefsXml);
+        if (!hasMetadata)
+        {
+            return PlayGoMetadata.Empty;
+        }
+
+        var chunkIds = LoadChunkIds(chunkDefsXml);
+        return new PlayGoMetadata(true, chunkIds);
+    }
+
+    private static ushort[] LoadChunkIds(string chunkDefsXml)
+    {
+        if (!File.Exists(chunkDefsXml))
+        {
+            return Array.Empty<ushort>();
+        }
+
+        try
+        {
+            var xml = File.ReadAllText(chunkDefsXml);
+            var chunkIds = new HashSet<ushort>();
+            AddChunkIds(xml, DefaultChunkPattern, chunkIds);
+            AddChunkIds(xml, ChunkIdPattern, chunkIds);
+
+            var sorted = chunkIds.ToArray();
+            Array.Sort(sorted);
+            return sorted;
+        }
+        catch (IOException)
+        {
+            return Array.Empty<ushort>();
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return Array.Empty<ushort>();
+        }
+    }
+
+    private static void AddChunkIds(string xml, Regex pattern, HashSet<ushort> chunkIds)
+    {
+        foreach (Match match in pattern.Matches(xml))
+        {
+            if (ushort.TryParse(match.Groups["id"].Value, out var chunkId))
+            {
+                chunkIds.Add(chunkId);
+            }
+        }
+    }
+
+    private static bool TryReadUInt16(CpuContext ctx, ulong address, out ushort value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(ushort)];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
             return false;
         }
 
-        var hostPath = Path.Combine(app0Root, "sce_sys", "playgo-chunk.dat");
-        return File.Exists(hostPath);
+        value = BinaryPrimitives.ReadUInt16LittleEndian(buffer);
+        return true;
+    }
+
+    private static bool TryWriteUInt16(CpuContext ctx, ulong address, ushort value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(ushort)];
+        BinaryPrimitives.WriteUInt16LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static bool TryWriteUInt32(CpuContext ctx, ulong address, uint value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static bool TryWriteInt32(CpuContext ctx, ulong address, int value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static bool TryWriteInt64(CpuContext ctx, ulong address, long value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(long)];
+        BinaryPrimitives.WriteInt64LittleEndian(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static void TracePlayGo(string message)
+    {
+        if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PLAYGO"), "1", StringComparison.Ordinal))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] playgo.{message}");
+        }
+    }
+
+    private static void TracePlayGoLocus(uint entries, ulong chunkIds, ulong outLoci)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PLAYGO"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        var count = Interlocked.Increment(ref _locusTraceDiagnostics);
+        if (entries != 1 || count <= 32 || count % 1000 == 0)
+        {
+            Console.Error.WriteLine(
+                $"[LOADER][TRACE] playgo.get_locus entries={entries} chunk_ids=0x{chunkIds:X16} out=0x{outLoci:X16}");
+        }
+    }
+
+    private sealed record PlayGoMetadata(bool Available, ushort[] ChunkIds)
+    {
+        public static readonly PlayGoMetadata Empty = new(false, Array.Empty<ushort>());
     }
 }

src/SharpEmu.Libs/Rtc/RtcExports.cs

@@ -8,6 +8,8 @@ namespace SharpEmu.Libs.Rtc;
 
 public static class RtcExports
 {
+    private const long DateTimeTicksPerMicrosecond = 10;
+
     [SysAbiExport(
         Nid = "ZPD1YOKI+Kw",
         ExportName = "sceRtcGetCurrentClockLocalTime",
@@ -42,6 +44,29 @@ public static class RtcExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "18B2NS1y9UU",
+        ExportName = "sceRtcGetCurrentTick",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceRtc")]
+    public static int RtcGetCurrentTick(CpuContext ctx)
+    {
+        var tickAddress = ctx[CpuRegister.Rdi];
+        if (tickAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var tickValue = unchecked((ulong)(DateTime.UtcNow.Ticks / DateTimeTicksPerMicrosecond));
+        if (!ctx.TryWriteUInt64(tickAddress, tickValue))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "8w-H19ip48I",
         ExportName = "sceRtcGetTick",
@@ -72,14 +97,57 @@ public static class RtcExports
                 rtcDateTime.Minute,
                 rtcDateTime.Second,
                 DateTimeKind.Utc);
-            tickValue = checked((ulong)((baseDateTime.Ticks / 10) + rtcDateTime.Microsecond));
+            tickValue = checked((ulong)((baseDateTime.Ticks / DateTimeTicksPerMicrosecond) + rtcDateTime.Microsecond));
+        }
+        catch (Exception ex) when (ex is ArgumentOutOfRangeException or OverflowException)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!ctx.TryWriteUInt64(tickAddress, tickValue))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "ueega6v3GUw",
+        ExportName = "sceRtcSetTick",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceRtc")]
+    public static int RtcSetTick(CpuContext ctx)
+    {
+        var dateTimeAddress = ctx[CpuRegister.Rdi];
+        var tickAddress = ctx[CpuRegister.Rsi];
+        if (dateTimeAddress == 0 || tickAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!ctx.TryReadUInt64(tickAddress, out var tickValue))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (tickValue > long.MaxValue / DateTimeTicksPerMicrosecond)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        DateTime dateTime;
+        try
+        {
+            dateTime = new DateTime(checked((long)tickValue * DateTimeTicksPerMicrosecond), DateTimeKind.Utc);
         }
         catch (ArgumentOutOfRangeException)
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
-        if (!ctx.TryWriteUInt64(tickAddress, tickValue))
+        if (!TryWriteRtcDateTime(ctx, dateTimeAddress, dateTime))
         {
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
         }
@@ -108,6 +176,21 @@ public static class RtcExports
         return true;
     }
 
+    private static bool TryWriteRtcDateTime(CpuContext ctx, ulong address, DateTime dateTime)
+    {
+        Span<byte> rtcDateTime = stackalloc byte[16];
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[0..2], checked((ushort)dateTime.Year));
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[2..4], checked((ushort)dateTime.Month));
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[4..6], checked((ushort)dateTime.Day));
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[6..8], checked((ushort)dateTime.Hour));
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[8..10], checked((ushort)dateTime.Minute));
+        BinaryPrimitives.WriteUInt16LittleEndian(rtcDateTime[10..12], checked((ushort)dateTime.Second));
+        BinaryPrimitives.WriteUInt32LittleEndian(
+            rtcDateTime[12..16],
+            checked((uint)((dateTime.Ticks % TimeSpan.TicksPerSecond) / DateTimeTicksPerMicrosecond)));
+        return ctx.Memory.TryWrite(address, rtcDateTime);
+    }
+
     private readonly record struct RtcDateTime(
         ushort Year,
         ushort Month,

src/SharpEmu.Libs/SaveData/SaveDataDialogExports.cs

@@ -0,0 +1,213 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Threading;
+
+namespace SharpEmu.Libs.SaveData;
+
+public static class SaveDataDialogExports
+{
+    private const int StatusNone = 0;
+    private const int StatusInitialized = 1;
+    private const int StatusRunning = 2;
+    private const int StatusFinished = 3;
+
+    private const int ErrorOk = 0;
+    private const int ErrorNotInitialized = unchecked((int)0x80B80003);
+    private const int ErrorAlreadyInitialized = unchecked((int)0x80B80004);
+    private const int ErrorNotFinished = unchecked((int)0x80B80005);
+    private const int ErrorNotRunning = unchecked((int)0x80B8000B);
+    private const int ErrorArgNull = unchecked((int)0x80B8000D);
+
+    private const int ResultSize = 0x48;
+    private static int _status;
+    private static int _lastMode;
+    private static ulong _lastUserData;
+
+    [SysAbiExport(
+        Nid = "s9e3+YpRnzw",
+        ExportName = "sceSaveDataDialogInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogInitialize(CpuContext ctx)
+    {
+        if (Interlocked.CompareExchange(ref _status, StatusInitialized, StatusNone) != StatusNone)
+        {
+            return SetReturn(ctx, ErrorAlreadyInitialized);
+        }
+
+        return SetReturn(ctx, ErrorOk);
+    }
+
+    [SysAbiExport(
+        Nid = "4tPhsP6FpDI",
+        ExportName = "sceSaveDataDialogOpen",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogOpen(CpuContext ctx)
+    {
+        var paramAddress = ctx[CpuRegister.Rdi];
+        if (paramAddress == 0)
+        {
+            return SetReturn(ctx, ErrorArgNull);
+        }
+
+        if (_status is not (StatusInitialized or StatusFinished))
+        {
+            return SetReturn(ctx, ErrorNotInitialized);
+        }
+
+        _lastMode = TryReadInt32(ctx, paramAddress, out var mode) ? mode : 0;
+        _lastUserData = TryReadUInt64(ctx, paramAddress + 0xC8, out var userData) ? userData : 0;
+
+        // There is no host save dialog yet. Complete immediately with OK so
+        // guest polling sees a finished dialog instead of spinning forever.
+        Interlocked.Exchange(ref _status, StatusFinished);
+        TraceSaveDataDialog($"open mode={_lastMode} userData=0x{_lastUserData:X16} -> finished");
+        return SetReturn(ctx, ErrorOk);
+    }
+
+    [SysAbiExport(
+        Nid = "ERKzksauAJA",
+        ExportName = "sceSaveDataDialogGetStatus",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogGetStatus(CpuContext ctx) => SetReturn(ctx, Volatile.Read(ref _status));
+
+    [SysAbiExport(
+        Nid = "KK3Bdg1RWK0",
+        ExportName = "sceSaveDataDialogUpdateStatus",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogUpdateStatus(CpuContext ctx) => SetReturn(ctx, Volatile.Read(ref _status));
+
+    [SysAbiExport(
+        Nid = "en7gNVnh878",
+        ExportName = "sceSaveDataDialogIsReadyToDisplay",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogIsReadyToDisplay(CpuContext ctx) => SetReturn(ctx, 1);
+
+    [SysAbiExport(
+        Nid = "yEiJ-qqr6Cg",
+        ExportName = "sceSaveDataDialogGetResult",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogGetResult(CpuContext ctx)
+    {
+        var resultAddress = ctx[CpuRegister.Rdi];
+        if (resultAddress == 0)
+        {
+            return SetReturn(ctx, ErrorArgNull);
+        }
+
+        if (Volatile.Read(ref _status) != StatusFinished)
+        {
+            return SetReturn(ctx, ErrorNotFinished);
+        }
+
+        Span<byte> result = stackalloc byte[ResultSize];
+        result.Clear();
+        BinaryPrimitives.WriteInt32LittleEndian(result[0x00..], _lastMode);
+        BinaryPrimitives.WriteInt32LittleEndian(result[0x04..], 0);
+        BinaryPrimitives.WriteInt32LittleEndian(result[0x08..], 0);
+        BinaryPrimitives.WriteUInt64LittleEndian(result[0x20..], _lastUserData);
+
+        if (!ctx.Memory.TryWrite(resultAddress, result))
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        return SetReturn(ctx, ErrorOk);
+    }
+
+    [SysAbiExport(
+        Nid = "fH46Lag88XY",
+        ExportName = "sceSaveDataDialogClose",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogClose(CpuContext ctx)
+    {
+        if (Interlocked.CompareExchange(ref _status, StatusFinished, StatusRunning) != StatusRunning)
+        {
+            return SetReturn(ctx, ErrorNotRunning);
+        }
+
+        return SetReturn(ctx, ErrorOk);
+    }
+
+    [SysAbiExport(
+        Nid = "YuH2FA7azqQ",
+        ExportName = "sceSaveDataDialogTerminate",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogTerminate(CpuContext ctx)
+    {
+        if (Interlocked.Exchange(ref _status, StatusNone) == StatusNone)
+        {
+            return SetReturn(ctx, ErrorNotInitialized);
+        }
+
+        _lastMode = 0;
+        _lastUserData = 0;
+        return SetReturn(ctx, ErrorOk);
+    }
+
+    [SysAbiExport(
+        Nid = "V-uEeFKARJU",
+        ExportName = "sceSaveDataDialogProgressBarInc",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogProgressBarInc(CpuContext ctx) => SetReturn(ctx, ErrorOk);
+
+    [SysAbiExport(
+        Nid = "hay1CfTmLyA",
+        ExportName = "sceSaveDataDialogProgressBarSetValue",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveDataDialog")]
+    public static int SaveDataDialogProgressBarSetValue(CpuContext ctx) => SetReturn(ctx, ErrorOk);
+
+    private static bool TryReadInt32(CpuContext ctx, ulong address, out int value)
+    {
+        value = 0;
+        Span<byte> bytes = stackalloc byte[sizeof(int)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadInt32LittleEndian(bytes);
+        return true;
+    }
+
+    private static bool TryReadUInt64(CpuContext ctx, ulong address, out ulong value)
+    {
+        value = 0;
+        Span<byte> bytes = stackalloc byte[sizeof(ulong)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt64LittleEndian(bytes);
+        return true;
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceSaveDataDialog(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_SAVEDATA"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] save_data_dialog.{message}");
+    }
+}

src/SharpEmu.Libs/SaveData/SaveDataExports.cs

@@ -0,0 +1,539 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using SharpEmu.Libs.Kernel;
+using System.Buffers.Binary;
+using System.Text;
+
+namespace SharpEmu.Libs.SaveData;
+
+public static class SaveDataExports
+{
+    private const int OrbisSaveDataErrorParameter = unchecked((int)0x809F0000);
+    private const int OrbisSaveDataErrorExists = unchecked((int)0x809F0007);
+    private const int OrbisSaveDataErrorNotFound = unchecked((int)0x809F0008);
+    private const int OrbisSaveDataErrorInternal = unchecked((int)0x809F000B);
+    private const int SaveDataTitleIdSize = 10;
+    private const int SaveDataDirNameSize = 32;
+    private const int SaveDataParamSize = 0x530;
+    private const int SaveDataSearchInfoSize = 0x30;
+    private const ulong ResultHitNumOffset = 0x00;
+    private const ulong ResultDirNamesOffset = 0x08;
+    private const ulong ResultDirNamesNumOffset = 0x10;
+    private const ulong ResultSetNumOffset = 0x14;
+    private const ulong ResultParamsOffset = 0x18;
+    private const ulong ResultInfosOffset = 0x20;
+    private const uint SortKeyFreeBlocks = 5;
+    private const uint SortOrderDescent = 1;
+    private const uint MountModeCreate = 1u << 2;
+    private const uint MountModeCreate2 = 1u << 5;
+    private const int MountResultSize = 0x40;
+    private static readonly object _stateGate = new();
+    private static string? _titleId;
+
+    public static void ConfigureApplicationInfo(string? titleId)
+    {
+        lock (_stateGate)
+        {
+            _titleId = string.IsNullOrWhiteSpace(titleId) ? null : SanitizePathSegment(titleId.Trim());
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "TywrFKCoLGY",
+        ExportName = "sceSaveDataInitialize3",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveData")]
+    public static int SaveDataInitialize3(CpuContext ctx)
+    {
+        try
+        {
+            Directory.CreateDirectory(ResolveSaveDataRoot());
+            return SetReturn(ctx, 0);
+        }
+        catch (IOException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "dyIhnXq-0SM",
+        ExportName = "sceSaveDataDirNameSearch",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSaveData")]
+    public static int SaveDataDirNameSearch(CpuContext ctx)
+    {
+        var condAddress = ctx[CpuRegister.Rdi];
+        var resultAddress = ctx[CpuRegister.Rsi];
+        if (condAddress == 0 || resultAddress == 0)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorParameter);
+        }
+
+        if (!TryReadSearchCond(ctx, condAddress, out var cond) ||
+            !TryReadSearchResult(ctx, resultAddress, out var result))
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        if (cond.UserId < 0 || cond.SortKey > SortKeyFreeBlocks || cond.SortOrder > SortOrderDescent)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorParameter);
+        }
+
+        try
+        {
+            string titleId;
+            if (cond.TitleIdAddress == 0)
+            {
+                titleId = ResolveConfiguredTitleId();
+            }
+            else if (!TryReadFixedAscii(ctx, cond.TitleIdAddress, SaveDataTitleIdSize, out titleId))
+            {
+                return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            var root = ResolveTitleSaveRoot(cond.UserId, titleId);
+            var entries = Directory.Exists(root)
+                ? EnumerateSaveDirectories(root, cond.Pattern)
+                : [];
+
+            entries = SortEntries(entries, cond.SortKey, cond.SortOrder);
+            var setNum = result.DirNamesNum == 0
+                ? 0
+                : Math.Min(result.DirNamesNum, entries.Count);
+            if (!TryWriteUInt32(ctx, resultAddress + ResultHitNumOffset, checked((uint)entries.Count)) ||
+                !TryWriteUInt32(ctx, resultAddress + ResultSetNumOffset, checked((uint)setNum)))
+            {
+                return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            if (setNum == 0)
+            {
+                TraceSaveData($"dir_name_search user={cond.UserId} title={titleId} hits={entries.Count} set=0 root='{root}'");
+                return SetReturn(ctx, 0);
+            }
+
+            if (result.DirNamesAddress == 0)
+            {
+                return SetReturn(ctx, OrbisSaveDataErrorParameter);
+            }
+
+            for (var i = 0; i < setNum; i++)
+            {
+                var entry = entries[i];
+                if (!TryWriteFixedAscii(
+                        ctx,
+                        result.DirNamesAddress + ((ulong)i * SaveDataDirNameSize),
+                        SaveDataDirNameSize,
+                        entry.Name) ||
+                    (result.ParamsAddress != 0 &&
+                     !TryWriteParam(ctx, result.ParamsAddress + ((ulong)i * SaveDataParamSize), entry)) ||
+                    (result.InfosAddress != 0 &&
+                     !TryWriteSearchInfo(ctx, result.InfosAddress + ((ulong)i * SaveDataSearchInfoSize), entry)))
+                {
+                    return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+                }
+            }
+
+            TraceSaveData($"dir_name_search user={cond.UserId} title={titleId} hits={entries.Count} set={setNum} root='{root}'");
+            return SetReturn(ctx, 0);
+        }
+        catch (IOException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "ZP4e7rlzOUk",
+        ExportName = "sceSaveDataMount3",
+        Target = Generation.Gen5,
+        LibraryName = "libSceSaveData")]
+    public static int SaveDataMount3(CpuContext ctx)
+    {
+        var mountAddress = ctx[CpuRegister.Rdi];
+        var resultAddress = ctx[CpuRegister.Rsi];
+        if (mountAddress == 0 || resultAddress == 0)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorParameter);
+        }
+
+        if (!TryReadInt32(ctx, mountAddress, out var userId) ||
+            !ctx.TryReadUInt64(mountAddress + 0x08, out var dirNameAddress) ||
+            !ctx.TryReadUInt64(mountAddress + 0x10, out var blocks) ||
+            !ctx.TryReadUInt64(mountAddress + 0x18, out var systemBlocks) ||
+            !TryReadUInt32(ctx, mountAddress + 0x20, out var mountMode) ||
+            !TryReadUInt32(ctx, mountAddress + 0x24, out var resource) ||
+            !TryReadUInt32(ctx, mountAddress + 0x28, out var mode) ||
+            dirNameAddress == 0 ||
+            !TryReadFixedAscii(ctx, dirNameAddress, SaveDataDirNameSize, out var dirName))
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        if (userId < 0 || string.IsNullOrWhiteSpace(dirName))
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorParameter);
+        }
+
+        try
+        {
+            var titleId = ResolveConfiguredTitleId();
+            var savePath = Path.Combine(
+                ResolveTitleSaveRoot(userId, titleId),
+                SanitizePathSegment(dirName));
+            var existed = Directory.Exists(savePath);
+            var create = (mountMode & MountModeCreate) != 0;
+            var createIfMissing = (mountMode & MountModeCreate2) != 0;
+
+            if (!existed && !create && !createIfMissing)
+            {
+                return SetReturn(ctx, OrbisSaveDataErrorNotFound);
+            }
+
+            if (existed && create)
+            {
+                return SetReturn(ctx, OrbisSaveDataErrorExists);
+            }
+
+            if (!existed)
+            {
+                Directory.CreateDirectory(savePath);
+            }
+
+            const string mountPoint = "/savedata0";
+            KernelMemoryCompatExports.RegisterGuestPathMount(mountPoint, savePath);
+
+            Span<byte> result = stackalloc byte[MountResultSize];
+            result.Clear();
+            WriteAscii(result[..16], mountPoint);
+            BinaryPrimitives.WriteUInt32LittleEndian(result[0x1C..], createIfMissing && !existed ? 1u : 0u);
+            if (!ctx.Memory.TryWrite(resultAddress, result))
+            {
+                return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            TraceSaveData(
+                $"mount3 user={userId} title={titleId} dir={dirName} blocks={blocks} " +
+                $"system_blocks={systemBlocks} mount_mode=0x{mountMode:X} resource={resource} mode={mode} " +
+                $"mount_point={mountPoint} created={!existed} root='{savePath}'");
+            return SetReturn(ctx, 0);
+        }
+        catch (IOException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorInternal);
+        }
+        catch (ArgumentException)
+        {
+            return SetReturn(ctx, OrbisSaveDataErrorParameter);
+        }
+    }
+
+    private static bool TryReadSearchCond(CpuContext ctx, ulong address, out SearchCond cond)
+    {
+        cond = default;
+        if (!TryReadInt32(ctx, address, out var userId) ||
+            !ctx.TryReadUInt64(address + 0x08, out var titleIdAddress) ||
+            !ctx.TryReadUInt64(address + 0x10, out var dirNameAddress) ||
+            !TryReadUInt32(ctx, address + 0x18, out var sortKey) ||
+            !TryReadUInt32(ctx, address + 0x1C, out var sortOrder))
+        {
+            return false;
+        }
+
+        string pattern;
+        if (dirNameAddress == 0)
+        {
+            pattern = string.Empty;
+        }
+        else if (!TryReadFixedAscii(ctx, dirNameAddress, SaveDataDirNameSize, out pattern))
+        {
+            return false;
+        }
+
+        cond = new SearchCond(userId, titleIdAddress, pattern, sortKey, sortOrder);
+        return true;
+    }
+
+    private static bool TryReadSearchResult(CpuContext ctx, ulong address, out SearchResult result)
+    {
+        result = default;
+        if (!ctx.TryReadUInt64(address + ResultDirNamesOffset, out var dirNamesAddress) ||
+            !TryReadUInt32(ctx, address + ResultDirNamesNumOffset, out var dirNamesNum) ||
+            !ctx.TryReadUInt64(address + ResultParamsOffset, out var paramsAddress) ||
+            !ctx.TryReadUInt64(address + ResultInfosOffset, out var infosAddress))
+        {
+            return false;
+        }
+
+        result = new SearchResult(dirNamesAddress, dirNamesNum, paramsAddress, infosAddress);
+        return true;
+    }
+
+    private static List<SaveEntry> EnumerateSaveDirectories(string root, string pattern)
+    {
+        var entries = new List<SaveEntry>();
+        foreach (var directory in Directory.EnumerateDirectories(root))
+        {
+            var name = Path.GetFileName(directory);
+            if (string.IsNullOrWhiteSpace(name) ||
+                name.StartsWith("sce_", StringComparison.OrdinalIgnoreCase) ||
+                (!string.IsNullOrEmpty(pattern) && !MatchPattern(name, pattern)))
+            {
+                continue;
+            }
+
+            var info = new DirectoryInfo(directory);
+            entries.Add(new SaveEntry(name, directory, info.LastWriteTimeUtc));
+        }
+
+        return entries;
+    }
+
+    private static List<SaveEntry> SortEntries(List<SaveEntry> entries, uint sortKey, uint sortOrder)
+    {
+        IOrderedEnumerable<SaveEntry> sorted = sortKey switch
+        {
+            3 => entries.OrderBy(entry => entry.LastWriteUtc),
+            _ => entries.OrderBy(entry => entry.Name, StringComparer.OrdinalIgnoreCase),
+        };
+
+        var list = sorted.ToList();
+        if (sortOrder == SortOrderDescent)
+        {
+            list.Reverse();
+        }
+
+        return list;
+    }
+
+    private static bool TryWriteParam(CpuContext ctx, ulong address, SaveEntry entry)
+    {
+        var param = new byte[SaveDataParamSize];
+        WriteAscii(param.AsSpan(0x00, 128), "Saved Data");
+        WriteAscii(param.AsSpan(0x100, 1024), entry.Name);
+        BinaryPrimitives.WriteInt64LittleEndian(
+            param.AsSpan(0x508, sizeof(long)),
+            new DateTimeOffset(entry.LastWriteUtc).ToUnixTimeSeconds());
+        return ctx.Memory.TryWrite(address, param);
+    }
+
+    private static bool TryWriteSearchInfo(CpuContext ctx, ulong address, SaveEntry entry)
+    {
+        var size = GetDirectorySize(entry.Path);
+        var usedBlocks = checked((ulong)((size + 32767) / 32768));
+        var blocks = Math.Max(96UL, usedBlocks);
+        Span<byte> info = stackalloc byte[SaveDataSearchInfoSize];
+        info.Clear();
+        BinaryPrimitives.WriteUInt64LittleEndian(info[0x00..], blocks);
+        BinaryPrimitives.WriteUInt64LittleEndian(info[0x08..], blocks - usedBlocks);
+        return ctx.Memory.TryWrite(address, info);
+    }
+
+    private static long GetDirectorySize(string root)
+    {
+        long total = 0;
+        foreach (var file in Directory.EnumerateFiles(root, "*", SearchOption.AllDirectories))
+        {
+            total += new FileInfo(file).Length;
+        }
+
+        return total;
+    }
+
+    private static bool MatchPattern(string value, string pattern) =>
+        MatchPattern(value.AsSpan(), pattern.AsSpan());
+
+    private static bool MatchPattern(ReadOnlySpan<char> value, ReadOnlySpan<char> pattern)
+    {
+        if (pattern.IsEmpty)
+        {
+            return value.IsEmpty;
+        }
+
+        if (pattern[0] == '%')
+        {
+            for (var i = 0; i <= value.Length; i++)
+            {
+                if (MatchPattern(value[i..], pattern[1..]))
+                {
+                    return true;
+                }
+            }
+
+            return false;
+        }
+
+        if (value.IsEmpty)
+        {
+            return false;
+        }
+
+        if (pattern[0] == '_' ||
+            char.ToUpperInvariant(pattern[0]) == char.ToUpperInvariant(value[0]))
+        {
+            return MatchPattern(value[1..], pattern[1..]);
+        }
+
+        return false;
+    }
+
+    private static string ResolveTitleSaveRoot(int userId, string titleId) =>
+        Path.Combine(ResolveSaveDataRoot(), userId.ToString(), SanitizePathSegment(titleId));
+
+    private static string ResolveSaveDataRoot()
+    {
+        var configured = Environment.GetEnvironmentVariable("SHARPEMU_SAVEDATA_DIR");
+        var root = string.IsNullOrWhiteSpace(configured)
+            ? Path.Combine(Environment.CurrentDirectory, "user", "savedata")
+            : configured;
+        return Path.GetFullPath(root);
+    }
+
+    private static string ResolveConfiguredTitleId()
+    {
+        lock (_stateGate)
+        {
+            if (!string.IsNullOrWhiteSpace(_titleId))
+            {
+                return _titleId;
+            }
+        }
+
+        var app0Root = Environment.GetEnvironmentVariable("SHARPEMU_APP0_DIR");
+        var app0Name = string.IsNullOrWhiteSpace(app0Root)
+            ? null
+            : Path.GetFileName(Path.TrimEndingDirectorySeparator(app0Root));
+        if (!string.IsNullOrWhiteSpace(app0Name))
+        {
+            var candidate = app0Name.Split('-', StringSplitOptions.RemoveEmptyEntries)[0];
+            if (!string.IsNullOrWhiteSpace(candidate))
+            {
+                return SanitizePathSegment(candidate);
+            }
+        }
+
+        return "default";
+    }
+
+    private static string SanitizePathSegment(string value)
+    {
+        var invalid = Path.GetInvalidFileNameChars();
+        var sanitized = new string(value.Select(ch => invalid.Contains(ch) ? '_' : ch).ToArray());
+        return string.IsNullOrWhiteSpace(sanitized) ? "default" : sanitized;
+    }
+
+    private static bool TryReadFixedAscii(CpuContext ctx, ulong address, int length, out string value)
+    {
+        value = string.Empty;
+        Span<byte> buffer = stackalloc byte[length];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            return false;
+        }
+
+        var stringLength = buffer.IndexOf((byte)0);
+        if (stringLength < 0)
+        {
+            stringLength = buffer.Length;
+        }
+
+        value = Encoding.ASCII.GetString(buffer[..stringLength]);
+        return true;
+    }
+
+    private static bool TryWriteFixedAscii(CpuContext ctx, ulong address, int length, string value)
+    {
+        Span<byte> buffer = stackalloc byte[length];
+        buffer.Clear();
+        WriteAscii(buffer, value);
+        return ctx.Memory.TryWrite(address, buffer);
+    }
+
+    private static void WriteAscii(Span<byte> destination, string value)
+    {
+        var count = Math.Min(value.Length, Math.Max(0, destination.Length - 1));
+        for (var i = 0; i < count; i++)
+        {
+            var ch = value[i];
+            destination[i] = ch <= 0x7F ? (byte)ch : (byte)'?';
+        }
+    }
+
+    private static bool TryReadInt32(CpuContext ctx, ulong address, out int value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(int)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadInt32LittleEndian(bytes);
+        return true;
+    }
+
+    private static bool TryReadUInt32(CpuContext ctx, ulong address, out uint value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(uint)];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadUInt32LittleEndian(bytes);
+        return true;
+    }
+
+    private static bool TryWriteUInt32(CpuContext ctx, ulong address, uint value)
+    {
+        Span<byte> bytes = stackalloc byte[sizeof(uint)];
+        BinaryPrimitives.WriteUInt32LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+
+    private static void TraceSaveData(string message)
+    {
+        if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_SAVEDATA"), "1", StringComparison.Ordinal))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] savedata.{message}");
+        }
+    }
+
+    private readonly record struct SearchCond(
+        int UserId,
+        ulong TitleIdAddress,
+        string Pattern,
+        uint SortKey,
+        uint SortOrder);
+
+    private readonly record struct SearchResult(
+        ulong DirNamesAddress,
+        uint DirNamesNum,
+        ulong ParamsAddress,
+        ulong InfosAddress);
+
+    private readonly record struct SaveEntry(string Name, string Path, DateTime LastWriteUtc);
+}

src/SharpEmu.Libs/Share/ShareExports.cs

@@ -0,0 +1,112 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Text;
+using System.Threading;
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.Share;
+
+public static class ShareExports
+{
+    private const int MaxContentParamBytes = 4096;
+
+    private static int _initialized;
+    private static string _contentParam = string.Empty;
+
+    [SysAbiExport(
+        Nid = "nBDD66kiFW8",
+        ExportName = "sceShareInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceShareUtility")]
+    public static int ShareInitialize(CpuContext ctx)
+    {
+        var memorySize = ctx[CpuRegister.Rdi];
+        var priority = unchecked((int)ctx[CpuRegister.Rsi]);
+        var affinityMask = ctx[CpuRegister.Rdx];
+        if (memorySize == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Interlocked.Exchange(ref _initialized, 1);
+
+        TraceShare($"initialize memory=0x{memorySize:X} priority={priority} affinity=0x{affinityMask:X}");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "7QZtURYnXG4",
+        ExportName = "sceShareSetContentParam",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceShareUtility")]
+    public static int ShareSetContentParam(CpuContext ctx)
+    {
+        var contentParamAddress = ctx[CpuRegister.Rdi];
+        if (contentParamAddress == 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        if (!TryReadNullTerminatedUtf8(ctx, contentParamAddress, MaxContentParamBytes, out var contentParam))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        _contentParam = contentParam;
+        if (Volatile.Read(ref _initialized) == 0)
+        {
+            TraceShare("set_content_param before initialize");
+        }
+
+        TraceShare($"set_content_param len={contentParam.Length} preview='{FormatTraceString(contentParam)}'");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)(int)result);
+        return (int)result;
+    }
+
+    private static bool TryReadNullTerminatedUtf8(CpuContext ctx, ulong address, int maxLength, out string value)
+    {
+        Span<byte> bytes = stackalloc byte[maxLength];
+        Span<byte> one = stackalloc byte[1];
+        for (var index = 0; index < maxLength; index++)
+        {
+            if (!ctx.Memory.TryRead(address + (ulong)index, one))
+            {
+                value = string.Empty;
+                return false;
+            }
+
+            if (one[0] == 0)
+            {
+                value = Encoding.UTF8.GetString(bytes[..index]);
+                return true;
+            }
+
+            bytes[index] = one[0];
+        }
+
+        value = string.Empty;
+        return false;
+    }
+
+    private static string FormatTraceString(string value)
+    {
+        var normalized = value.Replace("\r", "\\r", StringComparison.Ordinal).Replace("\n", "\\n", StringComparison.Ordinal);
+        return normalized.Length <= 120 ? normalized : string.Concat(normalized.AsSpan(0, 120), "...");
+    }
+
+    private static void TraceShare(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_SHARE"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] share.{message}");
+    }
+}

src/SharpEmu.Libs/SharpEmu.Libs.csproj

@@ -8,6 +8,12 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <ProjectReference Include="..\SharpEmu.HLE\SharpEmu.HLE.csproj" />
   </ItemGroup>
 
+  <ItemGroup>
+    <PackageReference Include="Silk.NET.Vulkan" />
+    <PackageReference Include="Silk.NET.Vulkan.Extensions.KHR" />
+    <PackageReference Include="Silk.NET.Windowing" />
+  </ItemGroup>
+
   <PropertyGroup>
     <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
   </PropertyGroup>

src/SharpEmu.Libs/SystemGesture/SystemGestureExports.cs

@@ -0,0 +1,75 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+
+namespace SharpEmu.Libs.SystemGesture;
+
+public static class SystemGestureExports
+{
+    [SysAbiExport(
+        Nid = "3pcAvmwKCvM",
+        ExportName = "sceSystemGestureInitializePrimitiveTouchRecognizer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureInitializePrimitiveTouchRecognizer(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "FWF8zkhr854",
+        ExportName = "sceSystemGestureCreateTouchRecognizer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureCreateTouchRecognizer(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "qpo-mEOwje0",
+        ExportName = "sceSystemGestureOpen",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureOpen(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "GgFMb22sbbI",
+        ExportName = "sceSystemGestureUpdatePrimitiveTouchRecognizer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureUpdatePrimitiveTouchRecognizer(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "j4h82CQWENo",
+        ExportName = "sceSystemGestureUpdateTouchRecognizer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureUpdateTouchRecognizer(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "h8uongcBNVs",
+        ExportName = "sceSystemGestureGetTouchEventsCount",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemGesture")]
+    public static int SystemGestureGetTouchEventsCount(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+}

src/SharpEmu.Libs/SystemService/SystemServiceExports.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Libs.VideoOut;
 using System.Buffers.Binary;
 
 namespace SharpEmu.Libs.SystemService;
@@ -10,6 +11,35 @@ public static class SystemServiceExports
 {
     private const int OrbisSystemServiceErrorParameter = unchecked((int)0x80A10003);
     private const int SystemServiceStatusSize = 0x0C;
+    private const int DisplaySafeAreaInfoSize = sizeof(float) + 128;
+
+    [SysAbiExport(
+        Nid = "fZo48un7LK4",
+        ExportName = "sceSystemServiceParamGetInt",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemService")]
+    public static int SystemServiceParamGetInt(CpuContext ctx)
+    {
+        var parameterId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var valueAddress = ctx[CpuRegister.Rsi];
+        if (valueAddress == 0)
+        {
+            return SetReturn(ctx, OrbisSystemServiceErrorParameter);
+        }
+
+        var value = parameterId switch
+        {
+            1 or 2 or 3 or 1000 => 1,
+            4 => 180,
+            _ => 0,
+        };
+
+        Span<byte> valueBytes = stackalloc byte[sizeof(int)];
+        BinaryPrimitives.WriteInt32LittleEndian(valueBytes, value);
+        return ctx.Memory.TryWrite(valueAddress, valueBytes)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
 
     [SysAbiExport(
         Nid = "rPo6tV8D9bM",
@@ -34,6 +64,39 @@ public static class SystemServiceExports
             : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
     }
 
+    [SysAbiExport(
+        Nid = "1n37q1Bvc5Y",
+        ExportName = "sceSystemServiceGetDisplaySafeAreaInfo",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemService")]
+    public static int SystemServiceGetDisplaySafeAreaInfo(CpuContext ctx)
+    {
+        var infoAddress = ctx[CpuRegister.Rdi];
+        if (infoAddress == 0)
+        {
+            return SetReturn(ctx, OrbisSystemServiceErrorParameter);
+        }
+
+        Span<byte> info = stackalloc byte[DisplaySafeAreaInfoSize];
+        info.Clear();
+        BinaryPrimitives.WriteSingleLittleEndian(info, 1.0f);
+
+        return ctx.Memory.TryWrite(infoAddress, info)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "Vo5V8KAwCmk",
+        ExportName = "sceSystemServiceHideSplashScreen",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemService")]
+    public static int SystemServiceHideSplashScreen(CpuContext ctx)
+    {
+        VulkanVideoPresenter.HideSplashScreen();
+        return SetReturn(ctx, 0);
+    }
+
     private static int SetReturn(CpuContext ctx, int result)
     {
         ctx[CpuRegister.Rax] = unchecked((ulong)result);

src/SharpEmu.Libs/UserService/UserServiceExports.cs

@@ -3,14 +3,21 @@
 
 using SharpEmu.HLE;
 using System.Buffers.Binary;
+using System.Text;
+using System.Threading;
 
 namespace SharpEmu.Libs.UserService;
 
 public static class UserServiceExports
 {
     private const int OrbisUserServiceErrorInvalidArgument = unchecked((int)0x80960005);
+    private const int OrbisUserServiceErrorNoEvent = unchecked((int)0x80960007);
+    private const int OrbisUserServiceErrorInvalidParameter = unchecked((int)0x80960009);
+    private const int OrbisUserServiceErrorBufferTooShort = unchecked((int)0x8096000A);
     private const int PrimaryUserId = 1;
     private const int InvalidUserId = -1;
+    private const string PrimaryUserName = "SharpEmu";
+    private static int _loginEventDelivered;
 
     [SysAbiExport(
         Nid = "j3YMu1MVNNo",
@@ -64,6 +71,89 @@ public static class UserServiceExports
             : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
     }
 
+    [SysAbiExport(
+        Nid = "yH17Q6NWtVg",
+        ExportName = "sceUserServiceGetEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetEvent(CpuContext ctx)
+    {
+        var eventAddress = ctx[CpuRegister.Rdi];
+        if (eventAddress == 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidArgument);
+        }
+
+        if (Interlocked.Exchange(ref _loginEventDelivered, 1) != 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorNoEvent);
+        }
+
+        Span<byte> payload = stackalloc byte[sizeof(int) * 2];
+        BinaryPrimitives.WriteInt32LittleEndian(payload[0..], 0);
+        BinaryPrimitives.WriteInt32LittleEndian(payload[sizeof(int)..], PrimaryUserId);
+        return ctx.Memory.TryWrite(eventAddress, payload)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "1xxcMiGu2fo",
+        ExportName = "sceUserServiceGetUserName",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetUserName(CpuContext ctx)
+    {
+        var userId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var nameAddress = ctx[CpuRegister.Rsi];
+        var capacity = ctx[CpuRegister.Rdx];
+        if (userId != PrimaryUserId)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidParameter);
+        }
+
+        if (nameAddress == 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidArgument);
+        }
+
+        var nameBytes = Encoding.UTF8.GetBytes(PrimaryUserName);
+        if (capacity <= (ulong)nameBytes.Length)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorBufferTooShort);
+        }
+
+        Span<byte> output = stackalloc byte[nameBytes.Length + 1];
+        nameBytes.CopyTo(output);
+        return ctx.Memory.TryWrite(nameAddress, output)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "D-CzAxQL0XI",
+        ExportName = "sceUserServiceGetPlatformPrivacySetting",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetPlatformPrivacySetting(CpuContext ctx)
+    {
+        var parameterId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var valueAddress = ctx[CpuRegister.Rsi];
+        if (parameterId != 1000)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidParameter);
+        }
+
+        if (valueAddress == 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidArgument);
+        }
+
+        return TryWriteInt32(ctx, valueAddress, 0)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
     private static bool TryWriteInt32(CpuContext ctx, ulong address, int value)
     {
         Span<byte> bytes = stackalloc byte[sizeof(int)];

src/SharpEmu.Libs/VideoOut/PngSplashLoader.cs

@@ -0,0 +1,213 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Buffers.Binary;
+using System.IO.Compression;
+
+namespace SharpEmu.Libs.VideoOut;
+
+internal static class PngSplashLoader
+{
+    private static ReadOnlySpan<byte> PngSignature =>
+    [
+        0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A,
+    ];
+
+    public static bool TryLoad(out byte[] pixels, out uint width, out uint height)
+    {
+        pixels = [];
+        width = 0;
+        height = 0;
+
+        try
+        {
+            var app0Root = Environment.GetEnvironmentVariable("SHARPEMU_APP0_DIR");
+            if (string.IsNullOrWhiteSpace(app0Root))
+            {
+                return false;
+            }
+
+            var path = Path.Combine(app0Root, "sce_sys", "pic0.png");
+            if (!File.Exists(path))
+            {
+                return false;
+            }
+
+            return TryDecode(File.ReadAllBytes(path), out pixels, out width, out height);
+        }
+        catch
+        {
+            pixels = [];
+            width = 0;
+            height = 0;
+            return false;
+        }
+    }
+
+    private static bool TryDecode(
+        ReadOnlySpan<byte> png,
+        out byte[] pixels,
+        out uint width,
+        out uint height)
+    {
+        pixels = [];
+        width = 0;
+        height = 0;
+        if (png.Length < 33 || !png[..8].SequenceEqual(PngSignature))
+        {
+            return false;
+        }
+
+        byte bitDepth = 0;
+        byte colorType = 0;
+        byte interlace = 0;
+        using var compressed = new MemoryStream();
+        var offset = 8;
+        while (offset <= png.Length - 12)
+        {
+            var chunkLength = BinaryPrimitives.ReadUInt32BigEndian(png.Slice(offset, 4));
+            if (chunkLength > int.MaxValue || offset > png.Length - 12 - (int)chunkLength)
+            {
+                return false;
+            }
+
+            var chunkType = png.Slice(offset + 4, 4);
+            var chunkData = png.Slice(offset + 8, (int)chunkLength);
+            if (chunkType.SequenceEqual("IHDR"u8))
+            {
+                if (chunkData.Length != 13)
+                {
+                    return false;
+                }
+
+                width = BinaryPrimitives.ReadUInt32BigEndian(chunkData[..4]);
+                height = BinaryPrimitives.ReadUInt32BigEndian(chunkData.Slice(4, 4));
+                bitDepth = chunkData[8];
+                colorType = chunkData[9];
+                interlace = chunkData[12];
+            }
+            else if (chunkType.SequenceEqual("IDAT"u8))
+            {
+                compressed.Write(chunkData);
+            }
+            else if (chunkType.SequenceEqual("IEND"u8))
+            {
+                break;
+            }
+
+            offset += checked((int)chunkLength + 12);
+        }
+
+        var sourceBytesPerPixel = colorType switch
+        {
+            2 => 3,
+            6 => 4,
+            _ => 0,
+        };
+        if (width == 0 ||
+            height == 0 ||
+            width > 16384 ||
+            height > 16384 ||
+            bitDepth != 8 ||
+            interlace != 0 ||
+            sourceBytesPerPixel == 0 ||
+            compressed.Length == 0)
+        {
+            return false;
+        }
+
+        var stride = checked((int)width * sourceBytesPerPixel);
+        var scanlineLength = checked(stride + 1);
+        var decompressedLength = checked(scanlineLength * (int)height);
+        var scanlines = GC.AllocateUninitializedArray<byte>(decompressedLength);
+        compressed.Position = 0;
+        using (var zlib = new ZLibStream(compressed, CompressionMode.Decompress))
+        {
+            zlib.ReadExactly(scanlines);
+            if (zlib.ReadByte() != -1)
+            {
+                return false;
+            }
+        }
+
+        var reconstructed = GC.AllocateUninitializedArray<byte>(checked(stride * (int)height));
+        for (var y = 0; y < (int)height; y++)
+        {
+            var sourceLine = scanlines.AsSpan(y * scanlineLength + 1, stride);
+            var targetLine = reconstructed.AsSpan(y * stride, stride);
+            var previousLine = y == 0
+                ? ReadOnlySpan<byte>.Empty
+                : reconstructed.AsSpan((y - 1) * stride, stride);
+            if (!TryUnfilter(
+                    scanlines[y * scanlineLength],
+                    sourceLine,
+                    previousLine,
+                    targetLine,
+                    sourceBytesPerPixel))
+            {
+                return false;
+            }
+        }
+
+        pixels = GC.AllocateUninitializedArray<byte>(checked((int)width * (int)height * 4));
+        for (int sourceOffset = 0, targetOffset = 0;
+             sourceOffset < reconstructed.Length;
+             sourceOffset += sourceBytesPerPixel, targetOffset += 4)
+        {
+            pixels[targetOffset] = reconstructed[sourceOffset + 2];
+            pixels[targetOffset + 1] = reconstructed[sourceOffset + 1];
+            pixels[targetOffset + 2] = reconstructed[sourceOffset];
+            pixels[targetOffset + 3] = sourceBytesPerPixel == 4
+                ? reconstructed[sourceOffset + 3]
+                : (byte)0xFF;
+        }
+
+        return true;
+    }
+
+    private static bool TryUnfilter(
+        byte filter,
+        ReadOnlySpan<byte> source,
+        ReadOnlySpan<byte> previous,
+        Span<byte> target,
+        int bytesPerPixel)
+    {
+        for (var x = 0; x < source.Length; x++)
+        {
+            var left = x >= bytesPerPixel ? target[x - bytesPerPixel] : (byte)0;
+            var above = previous.IsEmpty ? (byte)0 : previous[x];
+            var upperLeft = !previous.IsEmpty && x >= bytesPerPixel
+                ? previous[x - bytesPerPixel]
+                : (byte)0;
+            target[x] = filter switch
+            {
+                0 => source[x],
+                1 => unchecked((byte)(source[x] + left)),
+                2 => unchecked((byte)(source[x] + above)),
+                3 => unchecked((byte)(source[x] + ((left + above) >> 1))),
+                4 => unchecked((byte)(source[x] + Paeth(left, above, upperLeft))),
+                _ => source[x],
+            };
+
+            if (filter > 4)
+            {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
+    private static byte Paeth(byte left, byte above, byte upperLeft)
+    {
+        var estimate = left + above - upperLeft;
+        var leftDistance = Math.Abs(estimate - left);
+        var aboveDistance = Math.Abs(estimate - above);
+        var upperLeftDistance = Math.Abs(estimate - upperLeft);
+        return leftDistance <= aboveDistance && leftDistance <= upperLeftDistance
+            ? left
+            : aboveDistance <= upperLeftDistance
+                ? above
+                : upperLeft;
+    }
+}

src/SharpEmu.Libs/VideoOut/VideoOutExports.cs

@@ -29,8 +29,11 @@ public static class VideoOutExports
     private const int VideoOutBufferAttributeSize = 0x28;
     private const int VideoOutBufferAttribute2Size = 0x50;
     private const int VideoOutBuffersEntrySize = 0x20;
+    private const int VideoOutOutputStatusSize = 0x30;
     private const ulong SceVideoOutPixelFormatA8R8G8B8Srgb = 0x80000000;
     private const ulong SceVideoOutPixelFormatA8B8G8R8Srgb = 0x80002200;
+    private const ulong SceVideoOutPixelFormatB8G8R8A8Unorm = 0x8100000000000000;
+    private const ulong SceVideoOutPixelFormatR8G8B8A8Unorm = 0x8100000022000000;
     private const ulong SceVideoOutPixelFormatA2R10G10B10 = 0x88060000;
     private const ulong SceVideoOutPixelFormatA2R10G10B10Srgb = 0x88000000;
     private const ulong SceVideoOutPixelFormatA2R10G10B10Bt2020Pq = 0x88740000;
@@ -44,6 +47,36 @@ public static class VideoOutExports
     private static int _nextHandle = 1;
     private static int _frameDumpCount;
     private static long _nextFrameDumpIndex;
+    private static string _windowTitle = "SharpEmu VideoOut";
+
+    public static void ConfigureApplicationInfo(string? title, string? titleId, string? version)
+    {
+        var parts = new List<string>();
+        if (!string.IsNullOrWhiteSpace(title))
+        {
+            parts.Add(title.Trim());
+        }
+
+        if (!string.IsNullOrWhiteSpace(titleId))
+        {
+            parts.Add($"[{titleId.Trim()}]");
+        }
+
+        var application = parts.Count == 0 ? "VideoOut" : string.Join(' ', parts);
+        var versionSuffix = string.IsNullOrWhiteSpace(version) ? string.Empty : $" v{version.Trim()}";
+        lock (_stateGate)
+        {
+            _windowTitle = $"SharpEmu - {application}{versionSuffix}";
+        }
+    }
+
+    internal static string GetWindowTitle()
+    {
+        lock (_stateGate)
+        {
+            return _windowTitle;
+        }
+    }
 
     private sealed class VideoOutPortState
     {
@@ -52,6 +85,10 @@ public static class VideoOutExports
         public ulong VblankCount { get; set; }
         public ulong FlipCount { get; set; }
         public int CurrentBuffer { get; set; } = -1;
+        public uint OutputWidth { get; set; } = 1920;
+        public uint OutputHeight { get; set; } = 1080;
+        public uint RefreshRate { get; set; } = 60;
+        public float Gamma { get; set; } = 1.0f;
         public VideoOutBufferGroup?[] Groups { get; } = new VideoOutBufferGroup?[MaxDisplayBufferGroups];
         public VideoOutBufferSlot[] BufferSlots { get; } = CreateBufferSlots();
         public List<FlipEventRegistration> FlipEvents { get; } = new();
@@ -166,6 +203,93 @@ public static class VideoOutExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "utPrVdxio-8",
+        ExportName = "sceVideoOutGetOutputStatus",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutGetOutputStatus(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        var statusAddress = ctx[CpuRegister.Rsi];
+        if (statusAddress == 0)
+        {
+            return OrbisVideoOutErrorInvalidAddress;
+        }
+
+        if (!TryGetPort(handle, out var port))
+        {
+            return OrbisVideoOutErrorInvalidHandle;
+        }
+
+        Span<byte> status = stackalloc byte[VideoOutOutputStatusSize];
+        status.Clear();
+        var resolutionClass = port.OutputWidth >= 3840 || port.OutputHeight >= 2160 ? 2 : 1;
+        BinaryPrimitives.WriteInt32LittleEndian(status[0x00..0x04], resolutionClass);
+        BinaryPrimitives.WriteInt32LittleEndian(status[0x04..0x08], 1);
+        BinaryPrimitives.WriteUInt64LittleEndian(status[0x08..0x10], port.RefreshRate);
+        return ctx.Memory.TryWrite(statusAddress, status)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "DYhhWbJSeRg",
+        ExportName = "sceVideoOutColorSettingsSetGamma_",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutColorSettingsSetGamma(CpuContext ctx)
+    {
+        var settingsAddress = ctx[CpuRegister.Rdi];
+        if (settingsAddress == 0)
+        {
+            return OrbisVideoOutErrorInvalidAddress;
+        }
+
+        ctx.GetXmmRegister(0, out var xmm0Low, out _);
+        var gamma = BitConverter.Int32BitsToSingle(unchecked((int)xmm0Low));
+        if (!float.IsFinite(gamma) || gamma is < 0.1f or > 2.0f)
+        {
+            return OrbisVideoOutErrorInvalidValue;
+        }
+
+        Span<byte> gammaBytes = stackalloc byte[sizeof(float)];
+        BinaryPrimitives.WriteInt32LittleEndian(gammaBytes, BitConverter.SingleToInt32Bits(gamma));
+        return ctx.Memory.TryWrite(settingsAddress, gammaBytes)
+            ? (int)OrbisGen2Result.ORBIS_GEN2_OK
+            : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+    }
+
+    [SysAbiExport(
+        Nid = "pv9CI5VC+R0",
+        ExportName = "sceVideoOutAdjustColor_",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutAdjustColor(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        var settingsAddress = ctx[CpuRegister.Rsi];
+        if (settingsAddress == 0)
+        {
+            return OrbisVideoOutErrorInvalidAddress;
+        }
+
+        if (!TryGetPort(handle, out var port))
+        {
+            return OrbisVideoOutErrorInvalidHandle;
+        }
+
+        Span<byte> gammaBytes = stackalloc byte[sizeof(float)];
+        if (!ctx.Memory.TryRead(settingsAddress, gammaBytes))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        port.Gamma = BitConverter.Int32BitsToSingle(
+            BinaryPrimitives.ReadInt32LittleEndian(gammaBytes));
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "j6RaAUlaLv0",
         ExportName = "sceVideoOutWaitVblank",
@@ -210,7 +334,15 @@ public static class VideoOutExports
 
         lock (_stateGate)
         {
-            port.FlipEvents.Add(new FlipEventRegistration(equeue, userData));
+            var existingIndex = port.FlipEvents.FindIndex(registration => registration.Equeue == equeue);
+            if (existingIndex >= 0)
+            {
+                port.FlipEvents[existingIndex] = new FlipEventRegistration(equeue, userData);
+            }
+            else
+            {
+                port.FlipEvents.Add(new FlipEventRegistration(equeue, userData));
+            }
         }
 
         TraceVideoOut($"videoout.add_flip_event eq=0x{equeue:X16} handle={handle} udata=0x{userData:X16}");
@@ -228,7 +360,24 @@ public static class VideoOutExports
         var bufferIndex = unchecked((int)ctx[CpuRegister.Rsi]);
         var flipMode = unchecked((int)ctx[CpuRegister.Rdx]);
         var flipArg = unchecked((long)ctx[CpuRegister.Rcx]);
-        return SubmitFlip(ctx, handle, bufferIndex, flipMode, flipArg);
+        return SubmitFlip(ctx, handle, bufferIndex, flipMode, flipArg, submitGpuImage: true);
+    }
+
+    [SysAbiExport(
+        Nid = "zgXifHT9ErY",
+        ExportName = "sceVideoOutIsFlipPending",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutIsFlipPending(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        if (!TryGetPort(handle, out _))
+        {
+            return OrbisVideoOutErrorInvalidHandle;
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
     [SysAbiExport(
@@ -284,13 +433,33 @@ public static class VideoOutExports
             return OrbisVideoOutErrorInvalidEvent;
         }
 
-        return ctx.TryWriteUInt64(dataAddress, data >> 16)
+        var decodedData = unchecked((ulong)(unchecked((long)data) >> 16));
+        return ctx.TryWriteUInt64(dataAddress, decodedData)
             ? (int)OrbisGen2Result.ORBIS_GEN2_OK
             : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
     }
 
     public static int SubmitFlipFromAgc(CpuContext ctx, int handle, int bufferIndex, int flipMode, long flipArg) =>
-        SubmitFlip(ctx, handle, bufferIndex, flipMode, flipArg);
+        SubmitFlip(ctx, handle, bufferIndex, flipMode, flipArg, submitGpuImage: false);
+
+    internal static void SubmitHostRgbaFrame(ReadOnlySpan<byte> rgbaFrame, uint width, uint height)
+    {
+        if (rgbaFrame.Length != checked((int)(width * height * 4)))
+        {
+            return;
+        }
+
+        var bgraFrame = new byte[rgbaFrame.Length];
+        for (var offset = 0; offset < rgbaFrame.Length; offset += 4)
+        {
+            bgraFrame[offset + 0] = rgbaFrame[offset + 2];
+            bgraFrame[offset + 1] = rgbaFrame[offset + 1];
+            bgraFrame[offset + 2] = rgbaFrame[offset + 0];
+            bgraFrame[offset + 3] = rgbaFrame[offset + 3];
+        }
+
+        VulkanVideoPresenter.Submit(bgraFrame, width, height);
+    }
 
     internal static bool TryGetDisplayBufferInfo(int handle, int bufferIndex, out DisplayBufferInfo info)
     {
@@ -579,7 +748,13 @@ public static class VideoOutExports
         return groupIndex < 0 ? groupIndex : setIndex;
     }
 
-    private static int SubmitFlip(CpuContext ctx, int handle, int bufferIndex, int flipMode, long flipArg)
+    private static int SubmitFlip(
+        CpuContext ctx,
+        int handle,
+        int bufferIndex,
+        int flipMode,
+        long flipArg,
+        bool submitGpuImage)
     {
         if (!TryGetPort(handle, out var port))
         {
@@ -591,7 +766,7 @@ public static class VideoOutExports
             return OrbisVideoOutErrorInvalidIndex;
         }
 
-        ulong eventData;
+        ulong eventHint;
         List<FlipEventRegistration> flipEvents;
         lock (_stateGate)
         {
@@ -602,25 +777,38 @@ public static class VideoOutExports
 
             port.CurrentBuffer = bufferIndex;
             port.FlipCount++;
-            var eventCount = Math.Min(port.FlipCount, 0xFUL);
-            var timeBits = (ulong)Environment.TickCount64 & 0xFFFUL;
-            eventData = timeBits | (eventCount << 12) | ((unchecked((ulong)flipArg) & 0x0000_FFFF_FFFF_FFFFUL) << 16);
+            eventHint = SceVideoOutInternalEventFlip |
+                ((unchecked((ulong)flipArg) & 0x0000_FFFF_FFFF_FFFFUL) << 16);
             flipEvents = new List<FlipEventRegistration>(port.FlipEvents);
         }
 
-        _ = TryDumpFrame(ctx, port, bufferIndex, flipMode, flipArg);
+        if (submitGpuImage &&
+            bufferIndex >= 0 &&
+            TryGetDisplayBufferInfo(handle, bufferIndex, out var displayBuffer))
+        {
+            _ = VulkanVideoPresenter.TrySubmitGuestImage(
+                displayBuffer.Address,
+                displayBuffer.Width,
+                displayBuffer.Height,
+                displayBuffer.PitchInPixel);
+        }
+
+        if (string.Equals(
+                Environment.GetEnvironmentVariable("SHARPEMU_DUMP_VIDEOOUT"),
+                "1",
+                StringComparison.Ordinal))
+        {
+            _ = TryDumpFrame(ctx, port, bufferIndex, flipMode, flipArg);
+        }
 
         foreach (var flipEvent in flipEvents)
         {
-            _ = KernelEventQueueCompatExports.EnqueueEvent(
+            _ = KernelEventQueueCompatExports.TriggerDisplayEvent(
                 flipEvent.Equeue,
-                new KernelEventQueueCompatExports.KernelQueuedEvent(
-                    SceVideoOutInternalEventFlip,
-                    OrbisKernelEventFilterVideoOut,
-                    0,
-                    0,
-                    eventData,
-                    flipEvent.UserData));
+                SceVideoOutInternalEventFlip,
+                OrbisKernelEventFilterVideoOut,
+                eventHint,
+                flipEvent.UserData);
         }
 
         TraceVideoOut($"videoout.submit_flip handle={handle} index={bufferIndex} mode={flipMode} arg={flipArg} events={flipEvents.Count}");
@@ -655,6 +843,8 @@ public static class VideoOutExports
                 Index = groupIndex,
                 Attribute = attribute,
             };
+            port.OutputWidth = attribute.Width;
+            port.OutputHeight = attribute.Height;
 
             for (var i = 0; i < addresses.Length; i++)
             {
@@ -666,6 +856,7 @@ public static class VideoOutExports
 
             TraceVideoOut(
                 $"videoout.register_buffers handle={port.Handle} group={groupIndex} start={startIndex} count={addresses.Length} fmt=0x{attribute.PixelFormat:X} tile={attribute.TilingMode} {attribute.Width}x{attribute.Height} pitch={attribute.PitchInPixel}");
+            VulkanVideoPresenter.EnsureStarted(attribute.Width, attribute.Height);
             return groupIndex;
         }
     }
@@ -879,6 +1070,8 @@ public static class VideoOutExports
     private static uint GetBytesPerPixel(ulong pixelFormat) =>
         pixelFormat is SceVideoOutPixelFormatA8R8G8B8Srgb or
             SceVideoOutPixelFormatA8B8G8R8Srgb or
+            SceVideoOutPixelFormatB8G8R8A8Unorm or
+            SceVideoOutPixelFormatR8G8B8A8Unorm or
             SceVideoOutPixelFormatA2R10G10B10 or
             SceVideoOutPixelFormatA2R10G10B10Srgb or
             SceVideoOutPixelFormatA2R10G10B10Bt2020Pq
@@ -913,7 +1106,7 @@ public static class VideoOutExports
         var dst = 0;
         for (var src = 0; src + 3 < source.Length; src += 4)
         {
-            if (pixelFormat == SceVideoOutPixelFormatA8B8G8R8Srgb)
+            if (pixelFormat is SceVideoOutPixelFormatA8B8G8R8Srgb or SceVideoOutPixelFormatR8G8B8A8Unorm)
             {
                 destination[dst++] = source[src + 0];
                 destination[dst++] = source[src + 1];

src/SharpEmu.Libs/packages.lock.json

@@ -2,6 +2,91 @@
   "version": 2,
   "dependencies": {
     "net10.0": {
+      "Silk.NET.Vulkan": {
+        "type": "Direct",
+        "requested": "[2.23.0, )",
+        "resolved": "2.23.0",
+        "contentHash": "3/irtlSWXZ3eTi8N6nelI6L34NTB8ZJHpqVMNzZx2aX7Ek9YEQ34NoQW8/Tljrtmkg8KRhHW8hKTEzZaKV8PgA==",
+        "dependencies": {
+          "Silk.NET.Core": "2.23.0"
+        }
+      },
+      "Silk.NET.Vulkan.Extensions.KHR": {
+        "type": "Direct",
+        "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": "Direct",
+        "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"
+        }
+      },
+      "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"
       }