change jpg to sharpemu videoOut test window

.gitignore

@@ -29,6 +29,8 @@ arm64/
 packages/
 *.nupkg
 .nuget/
+.dotnet-home/
+.cache/
 
 .DS_Store
 Thumbs.db

REUSE.toml

@@ -7,7 +7,8 @@ path = [
     "**/packages.lock.json",
     "scripts/ps5_names.txt",
     "src/SharpEmu.HLE/Aerolib/aerolib.bin",
-    "_logs/**"
+    "_logs/**",
+    ".github/images/**"
 ]
 precedence = "aggregate"
 SPDX-FileCopyrightText = "SharpEmu Emulator Project"

global.json

@@ -1,6 +1,6 @@
 {
     "sdk": {
         "version": "10.0.103",
-        "rollForward": "latestFeature"
+        "rollForward": "disable"
     }
 }

src/SharpEmu.CLI/SharpEmu.CLI.csproj

@@ -30,10 +30,6 @@ SPDX-License-Identifier: GPL-2.0-or-later
     <RestorePackagesWithLockFile>true</RestorePackagesWithLockFile>
   </PropertyGroup>
 
-  <ItemGroup>
-    <KnownILLinkPack Update="Microsoft.NET.ILLink.Tasks" ILLinkPackVersion="10.0.5" />
-  </ItemGroup>
-
   <ItemGroup>
     <Content Include="..\..\LICENSE.txt">
       <CopyToOutputDirectory>Always</CopyToOutputDirectory>

src/SharpEmu.CLI/packages.lock.json

@@ -4,9 +4,9 @@
     "net10.0": {
       "Microsoft.NET.ILLink.Tasks": {
         "type": "Direct",
-        "requested": "[10.0.5, )",
-        "resolved": "10.0.5",
-        "contentHash": "A+5ZuQ0f449tM+MQrhf6R9ZX7lYpjk/ODEwLYKrnF6111rtARx8fVsm4YznUnQiKnnXfaXNBqgxmil6RW3L3SA=="
+        "requested": "[10.0.3, )",
+        "resolved": "10.0.3",
+        "contentHash": "0B6nZyCHWXnvmlB559oduOspVdNOnpNXPjhpWVMovLPAsDVG7A4jJR9rzECf67JUzxP8/ee/wA8clwIzJcWNFA=="
       },
       "sharpemu.core": {
         "type": "Project",

src/SharpEmu.Core/Cpu/CpuDispatcher.cs

@@ -193,6 +193,11 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
             return FailEarly(OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
         }
 
+        if (!InitializeGuestFrameChainSentinel(context))
+        {
+            return FailEarly(OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
         if (!InitializeTls(context, tlsBase))
         {
             return FailEarly(OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
@@ -363,6 +368,23 @@ public sealed class CpuDispatcher : ICpuDispatcher, IDisposable
                context.TryWriteUInt64(tlsBase + 0x28, 0xC0DEC0DECAFEBABEUL);
     }
 
+    private static bool InitializeGuestFrameChainSentinel(CpuContext context)
+    {
+        var stackTop = context[CpuRegister.Rsp] + sizeof(ulong);
+        var sentinelFrame = AlignDown(stackTop - 0x20, 16);
+        var seedRsp = sentinelFrame - sizeof(ulong);
+        if (!context.TryWriteUInt64(sentinelFrame, 0) ||
+            !context.TryWriteUInt64(sentinelFrame + sizeof(ulong), 0) ||
+            !context.TryWriteUInt64(seedRsp, 0))
+        {
+            return false;
+        }
+
+        context[CpuRegister.Rbp] = sentinelFrame;
+        context[CpuRegister.Rsp] = seedRsp;
+        return true;
+    }
+
     private static bool InitializeProcessEntryFrame(
         CpuContext context,
         string processImageName,

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

@@ -302,6 +302,23 @@ public sealed partial class DirectExecutionBackend
 			_cpuContext[CpuRegister.R15] = value8;
 			_cpuContext[CpuRegister.Rdi] = value;
 			_cpuContext[CpuRegister.Rsi] = value2;
+			if (GuestThreadExecution.TryConsumeCurrentEntryExit(out var exitStatus, out var exitReason))
+			{
+				if (TryCompleteGuestEntryToHostStub(argPackPtr, num, num7, importStubEntry.Nid, exitReason, exitStatus))
+				{
+					_cpuContext[CpuRegister.Rax] = unchecked((ulong)exitStatus);
+				}
+				else
+				{
+					LastError = $"Failed to complete guest entry after {importStubEntry.Nid}: missing host return sentinel";
+					_cpuContext[CpuRegister.Rax] = 18446744071562199298uL;
+				}
+			}
+			if (GuestThreadExecution.TryConsumeCurrentThreadBlock(out var blockReason) &&
+				TryYieldGuestThreadToHostStub(argPackPtr, num, num7, importStubEntry.Nid, blockReason))
+			{
+				_cpuContext[CpuRegister.Rax] = 0uL;
+			}
 			if (flag || flag2 || flag3)
 			{
 				Console.Error.WriteLine($"[LOADER][TRACE] ImportRet#{num}: nid={importStubEntry.Nid} result={orbisGen2Result} rax=0x{_cpuContext[CpuRegister.Rax]:X16}");
@@ -342,6 +359,49 @@ public sealed partial class DirectExecutionBackend
 		return true;
 	}
 
+	private unsafe bool TryCompleteGuestEntryToHostStub(nint argPackPtr, long dispatchIndex, ulong returnRip, string nid, string reason, int status)
+	{
+		ulong hostExit = _entryReturnSentinelRip;
+		if (hostExit < 65536)
+		{
+			return false;
+		}
+		try
+		{
+			*(ulong*)(argPackPtr + 96) = hostExit;
+		}
+		catch
+		{
+			return false;
+		}
+		Console.Error.WriteLine(
+			$"[LOADER][INFO] Guest entry exit at import#{dispatchIndex}: nid={nid} ret=0x{returnRip:X16} reason={reason} status={status}");
+		return true;
+	}
+
+	private unsafe bool TryYieldGuestThreadToHostStub(nint argPackPtr, long dispatchIndex, ulong returnRip, string nid, string reason)
+	{
+		ulong hostExit = _entryReturnSentinelRip;
+		if (hostExit < 65536)
+		{
+			return false;
+		}
+		try
+		{
+			*(ulong*)(argPackPtr + 96) = hostExit;
+		}
+		catch
+		{
+			return false;
+		}
+
+		_guestThreadYieldRequested = true;
+		_guestThreadYieldReason = string.IsNullOrWhiteSpace(reason) ? nid : reason;
+		Console.Error.WriteLine(
+			$"[LOADER][INFO] Guest thread yield at import#{dispatchIndex}: nid={nid} ret=0x{returnRip:X16} reason={_guestThreadYieldReason}");
+		return true;
+	}
+
 	private bool ShouldForceGuestExitOnImportLoop(string nid, ulong returnRip, long dispatchIndex, ulong arg0, ulong arg1)
 	{
 		if (dispatchIndex < 1200)

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

@@ -7,11 +7,14 @@ using System.Diagnostics;
 using System.Linq;
 using System.Runtime.InteropServices;
 using System.Threading;
+using SharpEmu.Core.Cpu;
+using SharpEmu.Core.Loader;
+using SharpEmu.Core.Memory;
 using SharpEmu.HLE;
 
 namespace SharpEmu.Core.Cpu.Native;
 
-public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, IDisposable
+public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, IGuestThreadScheduler, IDisposable
 {
 	private const int ImportLoopHistoryLength = 2048;
 
@@ -87,6 +90,16 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private const ulong GuestImageScanEnd = 36507222016uL;
 
+	private const ulong GuestThreadStackBaseAddress = 0x7FFF_E000_0000UL;
+
+	private const ulong GuestThreadTlsBaseAddress = 0x7FFE_0000_0000UL;
+
+	private const ulong GuestThreadStackSize = 0x0020_0000UL;
+
+	private const ulong GuestThreadTlsSize = 0x0001_0000UL;
+
+	private const ulong GuestThreadRegionStride = 0x0100_0000UL;
+
 	private const uint PAGE_EXECUTE_READWRITE = 64u;
 
 	private const uint PAGE_READWRITE = 4u;
@@ -105,6 +118,8 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private nint _tlsBaseAddress;
 
+	private nint _ownedTlsBaseAddress;
+
 	private bool _ownsTlsBaseAddress;
 
 	private int _tlsPatchStubOffset;
@@ -197,6 +212,52 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private readonly Dictionary<string, ulong> _importNidHashCache = new Dictionary<string, ulong>(StringComparer.Ordinal);
 
+	private enum GuestThreadRunState
+	{
+		Ready,
+		Running,
+		Blocked,
+		Exited,
+		Faulted,
+	}
+
+	private enum GuestNativeCallExitReason
+	{
+		Returned,
+		Blocked,
+		ForcedExit,
+		Exception,
+	}
+
+	private sealed class GuestThreadState
+	{
+		public ulong ThreadHandle { get; init; }
+
+		public ulong EntryPoint { get; init; }
+
+		public ulong Argument { get; init; }
+
+		public string Name { get; init; } = string.Empty;
+
+		public CpuContext Context { get; init; } = null!;
+
+		public GuestThreadRunState State { get; set; }
+
+		public string? BlockReason { get; set; }
+	}
+
+	private readonly object _guestThreadGate = new object();
+
+	private readonly Queue<GuestThreadState> _readyGuestThreads = new Queue<GuestThreadState>();
+
+	private readonly Dictionary<ulong, GuestThreadState> _guestThreads = new Dictionary<ulong, GuestThreadState>();
+
+	private int _guestThreadPumpDepth;
+
+	private bool _guestThreadYieldRequested;
+
+	private string? _guestThreadYieldReason;
+
 	private bool _forcedGuestExit;
 
 	private ulong _lastAvTraceRip;
@@ -219,7 +280,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 
 	private nint _selfHandlePtr;
 
-	private static readonly byte[] TlsPattern = new byte[9] { 100, 72, 139, 4, 37, 0, 0, 0, 0 };
+	private const int MinTlsPatchInstructionBytes = 9;
 
 	private delegate ulong ImportGatewayDelegate(nint backendHandle, int importIndex, nint argPackPtr);
 	private delegate int RawExceptionHandlerDelegate(void* exceptionInfo);
@@ -336,6 +397,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		{
 			throw new OutOfMemoryException("Failed to allocate TLS base");
 		}
+		_ownedTlsBaseAddress = _tlsBaseAddress;
 		_ownsTlsBaseAddress = true;
 		SeedTlsLayout(_tlsBaseAddress);
 		_hostRspSlotStorage = (nint)VirtualAlloc(null, 4096u, 12288u, 4u);
@@ -376,12 +438,15 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		_importLoopSignatureWriteIndex = 0;
 		_importLoopPatternHits = 0;
 		_importNidHashCache.Clear();
+		ClearGuestThreads();
 		_contextualUnresolvedReturnSites.Clear();
 		_stallWatchdogTriggered = 0;
 		_stallWatchdogStop = false;
 		_patchedEa020eLookupCall = false;
 		MarkExecutionProgress();
 		BindTlsBase(context);
+		var previousGuestThreadScheduler = GuestThreadExecution.Scheduler;
+		GuestThreadExecution.Scheduler = this;
 		try
 		{
 			if (!SetupImportStubs(importStubs))
@@ -407,6 +472,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		}
 		finally
 		{
+			GuestThreadExecution.Scheduler = previousGuestThreadScheduler;
 			Console.Error.WriteLine("[LOADER][INFO] === Execute END (LastError: " + (LastError ?? "null") + ") ===");
 		}
 	}
@@ -645,18 +711,15 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		}
 		if (num != _tlsBaseAddress)
 		{
-			if (_ownsTlsBaseAddress && _tlsBaseAddress != 0)
-			{
-				VirtualFree((void*)_tlsBaseAddress, 0u, 32768u);
-				_ownsTlsBaseAddress = false;
-			}
 			_tlsBaseAddress = num;
+			_ownsTlsBaseAddress = _tlsBaseAddress == _ownedTlsBaseAddress;
 		}
 		if (_tlsBaseAddress != 0)
 		{
 			context.FsBase = (ulong)_tlsBaseAddress;
 			context.GsBase = (ulong)_tlsBaseAddress;
 			SeedTlsLayout(_tlsBaseAddress);
+			UpdateTlsHandlerBase(_tlsBaseAddress);
 		}
 	}
 
@@ -669,6 +732,30 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		*(ulong*)(tlsBase + 96) = num;
 	}
 
+	private unsafe void UpdateTlsHandlerBase(nint tlsBase)
+	{
+		if (_tlsHandlerAddress == 0)
+		{
+			return;
+		}
+
+		uint oldProtect = default;
+		if (!VirtualProtect((void*)_tlsHandlerAddress, 16u, 64u, &oldProtect))
+		{
+			return;
+		}
+
+		try
+		{
+			*(long*)((byte*)_tlsHandlerAddress + 2) = tlsBase;
+		}
+		finally
+		{
+			VirtualProtect((void*)_tlsHandlerAddress, 16u, oldProtect, &oldProtect);
+			FlushInstructionCache(GetCurrentProcess(), (void*)_tlsHandlerAddress, 16u);
+		}
+	}
+
 	private unsafe nint CreateImportHandlerTrampoline(int importIndex)
 	{
 		void* ptr = VirtualAlloc(null, 192u, 12288u, 64u);
@@ -966,19 +1053,19 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			uint num7 = lpBuffer.Protect & 0xFF;
 			bool flag = lpBuffer.State == 4096 && (lpBuffer.Protect & PAGE_GUARD) == 0 && num7 != PAGE_NOACCESS;
 			bool flag2 = num7 == PAGE_EXECUTE || num7 == 32 || num7 == 64 || num7 == PAGE_EXECUTE_WRITECOPY;
-			if (flag && flag2 && num6 > num5 + (ulong)TlsPattern.Length)
+			if (flag && flag2 && num6 > num5 + MinTlsPatchInstructionBytes)
 			{
 				byte* ptr = (byte*)num5;
-				int num8 = (int)(num6 - num5) - TlsPattern.Length;
-				for (int i = 0; i <= num8; i++)
+				int scanBytes = (int)(num6 - num5);
+				for (int i = 0; i <= scanBytes - MinTlsPatchInstructionBytes; i++)
 				{
 					nint address = (nint)(ptr + i);
-					if (IsPatternMatch(ptr + i, TlsPattern))
+					int remainingBytes = scanBytes - i;
+					if (TryPatchTlsLoadInstruction(address, ptr + i, remainingBytes))
 					{
 						num3++;
-						PatchTlsInstruction(address);
 					}
-					else if (TryPatchTlsImmediateStoreInstruction(address, ptr + i))
+					else if (remainingBytes >= 12 && TryPatchTlsImmediateStoreInstruction(address, ptr + i))
 					{
 						num9++;
 					}
@@ -1071,12 +1158,71 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		return true;
 	}
 
-	private unsafe void PatchTlsInstruction(nint address)
+	private unsafe bool TryPatchTlsLoadInstruction(nint address, byte* source, int availableLength)
+	{
+		if (availableLength < MinTlsPatchInstructionBytes)
+		{
+			return false;
+		}
+
+		var offset = 0;
+		while (offset < availableLength && source[offset] == 0x66)
+		{
+			offset++;
+		}
+
+		if (offset >= availableLength || source[offset] != 0x64)
+		{
+			return false;
+		}
+
+		offset++;
+		if (offset >= availableLength)
+		{
+			return false;
+		}
+
+		var rex = (byte)0;
+		if (source[offset] >= 0x40 && source[offset] <= 0x4F)
+		{
+			rex = source[offset];
+			offset++;
+		}
+
+		if (offset + 7 > availableLength || source[offset] != 0x8B)
+		{
+			return false;
+		}
+
+		var modRm = source[offset + 1];
+		var sib = source[offset + 2];
+		if ((modRm >> 6) != 0 || (modRm & 7) != 4 || sib != 0x25)
+		{
+			return false;
+		}
+
+		var displacement = *(int*)(source + offset + 3);
+		if (displacement != 0)
+		{
+			return false;
+		}
+
+		var destinationRegister = ((modRm >> 3) & 7) | (((rex & 4) != 0) ? 8 : 0);
+		var instructionLength = offset + 7;
+		if (instructionLength < MinTlsPatchInstructionBytes)
+		{
+			return false;
+		}
+
+		return PatchTlsLoadInstruction(address, instructionLength, destinationRegister);
+	}
+
+	private unsafe bool PatchTlsLoadInstruction(nint address, int instructionLength, int destinationRegister)
 	{
 		uint flNewProtect = default(uint);
-		if (!VirtualProtect((void*)address, 9u, 64u, &flNewProtect))
+		if (!VirtualProtect((void*)address, (nuint)instructionLength, 64u, &flNewProtect))
 		{
-			return;
+			return false;
 		}
 		try
 		{
@@ -1087,20 +1233,29 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			if (num3 < int.MinValue || num3 > int.MaxValue)
 			{
 				Console.Error.WriteLine($"[LOADER][WARNING] TLS patch out of rel32 range at 0x{address:X16}");
+				return false;
 			}
-			else
+
+			*(int*)(address + 1) = (int)num3;
+			var offset = 5;
+			if (destinationRegister != 0)
 			{
-				*(int*)(address + 1) = (int)num3;
-				*(sbyte*)(address + 5) = 72;
-				*(sbyte*)(address + 6) = -119;
-				*(sbyte*)(address + 7) = -64;
-				*(sbyte*)(address + 8) = -112;
+				*(byte*)(address + offset++) = (byte)(0x48 | (destinationRegister >= 8 ? 1 : 0));
+				*(byte*)(address + offset++) = 0x89;
+				*(byte*)(address + offset++) = (byte)(0xC0 | (destinationRegister & 7));
 			}
+
+			while (offset < instructionLength)
+			{
+				*(byte*)(address + offset++) = 0x90;
+			}
+
+			return true;
 		}
 		finally
 		{
-			VirtualProtect((void*)address, 9u, flNewProtect, &flNewProtect);
-			FlushInstructionCache(GetCurrentProcess(), (void*)address, 9u);
+			VirtualProtect((void*)address, (nuint)instructionLength, flNewProtect, &flNewProtect);
+			FlushInstructionCache(GetCurrentProcess(), (void*)address, (nuint)instructionLength);
 		}
 	}
 
@@ -1268,6 +1423,439 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 		}
 	}
 
+	public bool TryStartThread(CpuContext creatorContext, GuestThreadStartRequest request, out string? error)
+	{
+		error = null;
+		if (request.ThreadHandle == 0 || request.EntryPoint < 65536)
+		{
+			error = $"invalid thread start request: handle=0x{request.ThreadHandle:X16} entry=0x{request.EntryPoint:X16}";
+			return false;
+		}
+		if (!TryCreateGuestThreadState(creatorContext, request, out var thread, out error))
+		{
+			return false;
+		}
+		lock (_guestThreadGate)
+		{
+			_guestThreads[request.ThreadHandle] = thread;
+			_readyGuestThreads.Enqueue(thread);
+		}
+		Console.Error.WriteLine(
+			$"[LOADER][INFO] Scheduled guest thread '{thread.Name}' handle=0x{thread.ThreadHandle:X16} entry=0x{thread.EntryPoint:X16} arg=0x{thread.Argument:X16}");
+		return true;
+	}
+
+	public void Pump(CpuContext callerContext, string reason)
+	{
+		_ = callerContext;
+		if (_guestThreadPumpDepth != 0)
+		{
+			return;
+		}
+		_guestThreadPumpDepth++;
+		try
+		{
+			for (int i = 0; i < 8; i++)
+			{
+				GuestThreadState? thread = null;
+				lock (_guestThreadGate)
+				{
+					while (_readyGuestThreads.Count > 0)
+					{
+						var candidate = _readyGuestThreads.Dequeue();
+						if (candidate.State == GuestThreadRunState.Ready)
+						{
+							thread = candidate;
+							thread.State = GuestThreadRunState.Running;
+							break;
+						}
+					}
+				}
+				if (thread == null)
+				{
+					return;
+				}
+				RunGuestThread(thread, reason);
+			}
+		}
+		finally
+		{
+			_guestThreadPumpDepth--;
+		}
+	}
+
+	private void ClearGuestThreads()
+	{
+		lock (_guestThreadGate)
+		{
+			_readyGuestThreads.Clear();
+			_guestThreads.Clear();
+		}
+	}
+
+	private bool TryCreateGuestThreadState(CpuContext creatorContext, GuestThreadStartRequest request, out GuestThreadState thread, out string? error)
+	{
+		thread = null!;
+		if (!TryGetVirtualMemory(creatorContext, out var virtualMemory))
+		{
+			error = "creator context memory is not backed by IVirtualMemory";
+			return false;
+		}
+		if (!TryMapGuestThreadRegion(virtualMemory, GuestThreadStackBaseAddress, GuestThreadStackSize, ProgramHeaderFlags.Read | ProgramHeaderFlags.Write, out var stackBase, out error))
+		{
+			return false;
+		}
+		if (!TryMapGuestThreadRegion(virtualMemory, GuestThreadTlsBaseAddress, GuestThreadTlsSize, ProgramHeaderFlags.Read | ProgramHeaderFlags.Write, out var tlsBase, out error))
+		{
+			return false;
+		}
+
+		var trackedMemory = new TrackedCpuMemory(virtualMemory);
+		var context = new CpuContext(trackedMemory, creatorContext.TargetGeneration)
+		{
+			Rip = request.EntryPoint,
+			Rflags = 0x202,
+			FsBase = tlsBase,
+			GsBase = tlsBase,
+		};
+		context[CpuRegister.Rsp] = stackBase + GuestThreadStackSize - sizeof(ulong);
+		context[CpuRegister.Rdi] = request.Argument;
+		context[CpuRegister.Rsi] = 0;
+		context[CpuRegister.Rdx] = 0;
+		context[CpuRegister.Rcx] = 0;
+		context[CpuRegister.R8] = 0;
+		context[CpuRegister.R9] = 0;
+		if (!InitializeGuestThreadFrame(context) || !InitializeGuestThreadTls(context, tlsBase))
+		{
+			error = "failed to initialize guest thread stack/TLS";
+			return false;
+		}
+
+		thread = new GuestThreadState
+		{
+			ThreadHandle = request.ThreadHandle,
+			EntryPoint = request.EntryPoint,
+			Argument = request.Argument,
+			Name = string.IsNullOrWhiteSpace(request.Name) ? $"Thread-{request.ThreadHandle:X}" : request.Name,
+			Context = context,
+			State = GuestThreadRunState.Ready,
+		};
+		error = null;
+		return true;
+	}
+
+	private static bool TryGetVirtualMemory(CpuContext context, out IVirtualMemory virtualMemory)
+	{
+		if (context.Memory is IVirtualMemory directMemory)
+		{
+			virtualMemory = directMemory;
+			return true;
+		}
+		if (context.Memory is TrackedCpuMemory trackedMemory && trackedMemory.Inner is IVirtualMemory trackedInner)
+		{
+			virtualMemory = trackedInner;
+			return true;
+		}
+
+		virtualMemory = null!;
+		return false;
+	}
+
+	private static bool TryMapGuestThreadRegion(
+		IVirtualMemory virtualMemory,
+		ulong baseAddress,
+		ulong size,
+		ProgramHeaderFlags protection,
+		out ulong mappedBase,
+		out string? error)
+	{
+		for (int i = 0; i < 64; i++)
+		{
+			var candidateBase = baseAddress - ((ulong)i * GuestThreadRegionStride);
+			if (!IsGuestThreadRegionFree(virtualMemory, candidateBase, size))
+			{
+				continue;
+			}
+			try
+			{
+				virtualMemory.Map(
+					candidateBase,
+					size,
+					fileOffset: 0,
+					fileData: ReadOnlySpan<byte>.Empty,
+					protection: protection);
+				mappedBase = candidateBase;
+				error = null;
+				return true;
+			}
+			catch (InvalidOperationException)
+			{
+			}
+		}
+
+		mappedBase = 0;
+		error = $"failed to map guest thread region near 0x{baseAddress:X16}";
+		return false;
+	}
+
+	private static bool IsGuestThreadRegionFree(IVirtualMemory virtualMemory, ulong candidateBase, ulong size)
+	{
+		var candidateEnd = candidateBase + size;
+		foreach (var region in virtualMemory.SnapshotRegions())
+		{
+			var regionStart = region.VirtualAddress;
+			var regionEnd = regionStart + region.MemorySize;
+			if (candidateBase < regionEnd && regionStart < candidateEnd)
+			{
+				return false;
+			}
+		}
+
+		return true;
+	}
+
+	private static bool InitializeGuestThreadFrame(CpuContext context)
+	{
+		var stackTop = context[CpuRegister.Rsp] + sizeof(ulong);
+		var sentinelFrame = AlignDown(stackTop - 0x20, 16);
+		var seedRsp = sentinelFrame - sizeof(ulong);
+		if (!context.TryWriteUInt64(sentinelFrame, 0) ||
+			!context.TryWriteUInt64(sentinelFrame + sizeof(ulong), 0) ||
+			!context.TryWriteUInt64(seedRsp, 0))
+		{
+			return false;
+		}
+
+		context[CpuRegister.Rbp] = sentinelFrame;
+		context[CpuRegister.Rsp] = seedRsp;
+		return true;
+	}
+
+	private static bool InitializeGuestThreadTls(CpuContext context, ulong tlsBase)
+	{
+		return context.TryWriteUInt64(tlsBase + 0x00, tlsBase) &&
+			context.TryWriteUInt64(tlsBase + 0x10, tlsBase) &&
+			context.TryWriteUInt64(tlsBase + 0x28, 0xC0DEC0DECAFEBABEUL);
+	}
+
+	private void RunGuestThread(GuestThreadState thread, string reason)
+	{
+		Console.Error.WriteLine(
+			$"[LOADER][INFO] Pumping guest thread '{thread.Name}' reason={reason} entry=0x{thread.EntryPoint:X16}");
+		var previousContext = _cpuContext;
+		var previousTlsBase = _tlsBaseAddress;
+		var previousOwnsTlsBase = _ownsTlsBaseAddress;
+		var previousForcedGuestExit = _forcedGuestExit;
+		var previousLastError = LastError;
+		var previousGuestThreadHandle = GuestThreadExecution.EnterGuestThread(thread.ThreadHandle);
+		try
+		{
+			_cpuContext = thread.Context;
+			_forcedGuestExit = false;
+			LastError = null;
+			BindTlsBase(thread.Context);
+			var exitReason = ExecuteGuestThreadEntry(thread.Context, thread.EntryPoint, thread.Name, out var blockReason);
+			lock (_guestThreadGate)
+			{
+				switch (exitReason)
+				{
+					case GuestNativeCallExitReason.Returned:
+						thread.State = GuestThreadRunState.Exited;
+						break;
+					case GuestNativeCallExitReason.Blocked:
+						thread.State = GuestThreadRunState.Blocked;
+						thread.BlockReason = blockReason;
+						break;
+					default:
+						thread.State = GuestThreadRunState.Faulted;
+						thread.BlockReason = blockReason;
+						break;
+				}
+			}
+			Console.Error.WriteLine(
+				$"[LOADER][INFO] Guest thread '{thread.Name}' state={thread.State} reason={blockReason ?? "none"}");
+		}
+		finally
+		{
+			GuestThreadExecution.RestoreGuestThread(previousGuestThreadHandle);
+			_cpuContext = previousContext;
+			_tlsBaseAddress = previousTlsBase;
+			_ownsTlsBaseAddress = previousOwnsTlsBase;
+			if (_tlsBaseAddress != 0)
+			{
+				SeedTlsLayout(_tlsBaseAddress);
+				UpdateTlsHandlerBase(_tlsBaseAddress);
+			}
+			_forcedGuestExit = previousForcedGuestExit;
+			LastError = previousLastError;
+		}
+	}
+
+	private unsafe GuestNativeCallExitReason ExecuteGuestThreadEntry(CpuContext context, ulong entryPoint, string name, out string? reason)
+	{
+		reason = null;
+		if (context[CpuRegister.Rsp] == 0)
+		{
+			reason = "guest thread stack pointer is zero";
+			return GuestNativeCallExitReason.Exception;
+		}
+		void* ptr = VirtualAlloc(null, 256u, 12288u, 64u);
+		if (ptr == null)
+		{
+			reason = "failed to allocate executable memory for guest thread stub";
+			return GuestNativeCallExitReason.Exception;
+		}
+		var previousSentinel = _entryReturnSentinelRip;
+		var previousHostRspSlotValue = _hostRspSlotStorage != 0 ? *(ulong*)_hostRspSlotStorage : 0;
+		var previousYieldRequested = _guestThreadYieldRequested;
+		var previousYieldReason = _guestThreadYieldReason;
+		try
+		{
+			byte* ptr2 = (byte*)ptr;
+			ulong hostRspSlot = (ulong)ptr + 224uL;
+			int offset = 0;
+			ptr2[offset++] = 83;
+			ptr2[offset++] = 85;
+			ptr2[offset++] = 87;
+			ptr2[offset++] = 86;
+			ptr2[offset++] = 73;
+			ptr2[offset++] = 186;
+			*(ulong*)(ptr2 + offset) = hostRspSlot;
+			offset += 8;
+			ptr2[offset++] = 73;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 34;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rsp];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 196;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 131;
+			ptr2[offset++] = 236;
+			ptr2[offset++] = 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 189;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rbp];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rdi];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 199;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rsi];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 198;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rdx];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 194;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = context[CpuRegister.Rcx];
+			offset += 8;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 137;
+			ptr2[offset++] = 193;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 184;
+			*(ulong*)(ptr2 + offset) = entryPoint;
+			offset += 8;
+			ptr2[offset++] = byte.MaxValue;
+			ptr2[offset++] = 208;
+			int sentinelOffset = offset + 4;
+			ptr2[offset++] = 72;
+			ptr2[offset++] = 131;
+			ptr2[offset++] = 196;
+			ptr2[offset++] = 8;
+			ptr2[offset++] = 73;
+			ptr2[offset++] = 186;
+			*(ulong*)(ptr2 + offset) = hostRspSlot;
+			offset += 8;
+			ptr2[offset++] = 73;
+			ptr2[offset++] = 139;
+			ptr2[offset++] = 34;
+			ptr2[offset++] = 94;
+			ptr2[offset++] = 95;
+			ptr2[offset++] = 93;
+			ptr2[offset++] = 91;
+			ptr2[offset++] = 195;
+			ulong sentinel = (ulong)ptr + (ulong)sentinelOffset;
+			_entryReturnSentinelRip = sentinel;
+			if (!context.TryWriteUInt64(context[CpuRegister.Rsp], sentinel))
+			{
+				reason = $"failed to patch guest thread return sentinel at 0x{context[CpuRegister.Rsp]:X16}";
+				return GuestNativeCallExitReason.Exception;
+			}
+			uint oldProtect = default(uint);
+			VirtualProtect(ptr, 256u, 64u, &oldProtect);
+			FlushInstructionCache(GetCurrentProcess(), ptr, 256u);
+			if (_hostRspSlotStorage != 0)
+			{
+				*(ulong*)_hostRspSlotStorage = hostRspSlot;
+			}
+			_guestThreadYieldRequested = false;
+			_guestThreadYieldReason = null;
+			try
+			{
+				var nativeReturn = Marshal.GetDelegateForFunctionPointer<NativeEntryDelegate>((nint)ptr)();
+				if (_guestThreadYieldRequested)
+				{
+					reason = _guestThreadYieldReason ?? "guest thread blocked";
+					return GuestNativeCallExitReason.Blocked;
+				}
+				if (_forcedGuestExit)
+				{
+					reason = LastError ?? "guest thread forced exit";
+					return GuestNativeCallExitReason.ForcedExit;
+				}
+				reason = $"returned 0x{nativeReturn:X8}";
+				return GuestNativeCallExitReason.Returned;
+			}
+			catch (AccessViolationException ex)
+			{
+				reason = "access violation: " + ex.Message;
+				return GuestNativeCallExitReason.Exception;
+			}
+			catch (Exception ex)
+			{
+				reason = ex.GetType().Name + ": " + ex.Message;
+				return GuestNativeCallExitReason.Exception;
+			}
+		}
+		finally
+		{
+			_entryReturnSentinelRip = previousSentinel;
+			if (_hostRspSlotStorage != 0)
+			{
+				*(ulong*)_hostRspSlotStorage = previousHostRspSlotValue;
+			}
+			_guestThreadYieldRequested = previousYieldRequested;
+			_guestThreadYieldReason = previousYieldReason;
+			VirtualFree(ptr, 0u, 32768u);
+		}
+	}
+
+	private static ulong AlignDown(ulong value, ulong alignment)
+	{
+		if (alignment == 0)
+		{
+			return value;
+		}
+		return value & ~(alignment - 1);
+	}
+
 	private unsafe bool ExecuteEntry(CpuContext context, ulong entryPoint, out OrbisGen2Result result)
 	{
 		Console.Error.WriteLine($"[LOADER][INFO] ExecuteEntry starting at 0x{entryPoint:X16}");
@@ -1315,6 +1903,10 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			ptr2[num3++] = 236;
 			ptr2[num3++] = 8;
 			ptr2[num3++] = 72;
+			ptr2[num3++] = 189;
+			*(ulong*)(ptr2 + num3) = context[CpuRegister.Rbp];
+			num3 += 8;
+			ptr2[num3++] = 72;
 			ptr2[num3++] = 184;
 			*(ulong*)(ptr2 + num3) = context[CpuRegister.Rdi];
 			num3 += 8;
@@ -1393,6 +1985,7 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			{
 				num6 = Marshal.GetDelegateForFunctionPointer<NativeEntryDelegate>((nint)ptr)();
 				Console.Error.WriteLine($"[LOADER][INFO] Guest returned: {num6}");
+				Pump(context, "entry_return");
 			}
 			catch (AccessViolationException ex)
 			{
@@ -1610,11 +2203,13 @@ public sealed unsafe partial class DirectExecutionBackend : INativeCpuBackend, I
 			_selfHandle.Free();
 			_selfHandlePtr = 0;
 		}
-		if (_ownsTlsBaseAddress && _tlsBaseAddress != 0)
+		if (_ownedTlsBaseAddress != 0)
 		{
-			VirtualFree((void*)_tlsBaseAddress, 0u, 32768u);
+			VirtualFree((void*)_ownedTlsBaseAddress, 0u, 32768u);
+			_ownedTlsBaseAddress = 0;
 		}
 		_tlsBaseAddress = 0;
+		_ownsTlsBaseAddress = false;
 		if (_tlsModuleBases.Count > 0)
 		{
 			foreach (var (_, num3) in _tlsModuleBases)

src/SharpEmu.HLE/GuestThreadExecution.cs

@@ -0,0 +1,107 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+namespace SharpEmu.HLE;
+
+public readonly record struct GuestThreadStartRequest(
+    ulong ThreadHandle,
+    ulong EntryPoint,
+    ulong Argument,
+    ulong AttributeAddress,
+    string Name);
+
+public interface IGuestThreadScheduler
+{
+    bool TryStartThread(CpuContext creatorContext, GuestThreadStartRequest request, out string? error);
+
+    void Pump(CpuContext callerContext, string reason);
+}
+
+public static class GuestThreadExecution
+{
+    [ThreadStatic]
+    private static ulong _currentGuestThreadHandle;
+
+    [ThreadStatic]
+    private static string? _pendingBlockReason;
+
+    [ThreadStatic]
+    private static bool _pendingEntryExit;
+
+    [ThreadStatic]
+    private static int _pendingEntryExitStatus;
+
+    [ThreadStatic]
+    private static string? _pendingEntryExitReason;
+
+    public static IGuestThreadScheduler? Scheduler { get; set; }
+
+    public static bool IsGuestThread => _currentGuestThreadHandle != 0;
+
+    public static ulong CurrentGuestThreadHandle => _currentGuestThreadHandle;
+
+    public static ulong EnterGuestThread(ulong threadHandle)
+    {
+        var previous = _currentGuestThreadHandle;
+        _currentGuestThreadHandle = threadHandle;
+        _pendingBlockReason = null;
+        _pendingEntryExit = false;
+        _pendingEntryExitStatus = 0;
+        _pendingEntryExitReason = null;
+        return previous;
+    }
+
+    public static void RestoreGuestThread(ulong previousThreadHandle)
+    {
+        _currentGuestThreadHandle = previousThreadHandle;
+        _pendingBlockReason = null;
+        _pendingEntryExit = false;
+        _pendingEntryExitStatus = 0;
+        _pendingEntryExitReason = null;
+    }
+
+    public static bool RequestCurrentThreadBlock(string reason)
+    {
+        if (!IsGuestThread)
+        {
+            return false;
+        }
+
+        _pendingBlockReason = string.IsNullOrWhiteSpace(reason) ? "guest_thread_blocked" : reason;
+        return true;
+    }
+
+    public static bool TryConsumeCurrentThreadBlock(out string reason)
+    {
+        reason = _pendingBlockReason ?? string.Empty;
+        if (string.IsNullOrEmpty(reason))
+        {
+            return false;
+        }
+
+        _pendingBlockReason = null;
+        return true;
+    }
+
+    public static void RequestCurrentEntryExit(string reason, int status)
+    {
+        _pendingEntryExit = true;
+        _pendingEntryExitStatus = status;
+        _pendingEntryExitReason = string.IsNullOrWhiteSpace(reason) ? "guest_entry_exit" : reason;
+    }
+
+    public static bool TryConsumeCurrentEntryExit(out int status, out string reason)
+    {
+        status = _pendingEntryExitStatus;
+        reason = _pendingEntryExitReason ?? string.Empty;
+        if (!_pendingEntryExit)
+        {
+            return false;
+        }
+
+        _pendingEntryExit = false;
+        _pendingEntryExitStatus = 0;
+        _pendingEntryExitReason = null;
+        return true;
+    }
+}

src/SharpEmu.Libs/Ampr/AmprExports.cs

@@ -0,0 +1,641 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Collections.Concurrent;
+
+namespace SharpEmu.Libs.Ampr;
+
+public static class AmprExports
+{
+    private const int CommandBufferHeaderSize = 0x28;
+    private const ulong CommandBufferSelfOffset = 0x00;
+    private const ulong CommandBufferDataOffset = 0x08;
+    private const ulong CommandBufferSizeOffset = 0x10;
+    private const ulong CommandBufferAux0Offset = 0x18;
+    private const ulong CommandBufferAux1Offset = 0x20;
+    private const ulong ReadFileRecordSize = 0x30;
+    private const ulong KernelEventQueueRecordSize = 0x30;
+    private const uint ReadFileRecordType = 1;
+    private const uint KernelEventQueueRecordType = 2;
+    private static readonly ConcurrentDictionary<ulong, CommandBufferState> _commandBuffers = new();
+
+    private sealed class CommandBufferState
+    {
+        public ulong Buffer;
+        public ulong Size;
+        public ulong WriteOffset;
+    }
+
+    [SysAbiExport(
+        Nid = "8aI7R7WaOlc",
+        ExportName = "sceAmprCommandBufferConstructor",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferConstructor(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var buffer = ctx[CpuRegister.Rsi];
+        var size = ctx[CpuRegister.Rdx];
+
+        if (commandBuffer == 0)
+        {
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (!InitializeCommandBuffer(ctx, commandBuffer, buffer, size, aux0: 0, aux1: 0, clear: true))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "ctor", commandBuffer, buffer, size);
+        ctx[CpuRegister.Rax] = commandBuffer;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "a8uLzYY--tM",
+        ExportName = "sceAmprAprCommandBufferConstructor",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int AprCommandBufferConstructor(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var aux0 = ctx[CpuRegister.Rsi];
+        var aux1 = ctx[CpuRegister.Rdx];
+
+        if (commandBuffer == 0)
+        {
+            ctx[CpuRegister.Rax] = 0;
+            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))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "apr_ctor", commandBuffer, aux0, aux1);
+        ctx[CpuRegister.Rax] = commandBuffer;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "Qs1xtplKo0U",
+        ExportName = "sceAmprAprCommandBufferDestructor",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int AprCommandBufferDestructor(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (!ctx.TryWriteUInt64(commandBuffer + CommandBufferAux0Offset, 0) ||
+            !ctx.TryWriteUInt64(commandBuffer + CommandBufferAux1Offset, 0))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "apr_dtor", commandBuffer, 0, 0);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "GuchCTefuZw",
+        ExportName = "sceAmprCommandBufferDestructor",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferDestructor(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (!WriteVisibleCommandBufferPointers(ctx, commandBuffer, buffer: 0, size: 0))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        _commandBuffers.TryRemove(commandBuffer, out _);
+        TraceAmpr(ctx, "dtor", commandBuffer, 0, 0);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "N-FSPA4S3nI",
+        ExportName = "sceAmprCommandBufferSetBuffer",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferSetBuffer(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var buffer = ctx[CpuRegister.Rsi];
+        var size = ctx[CpuRegister.Rdx];
+
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!WriteCommandBufferPointers(ctx, commandBuffer, buffer, size))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "set_buffer", commandBuffer, buffer, size);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "baQO9ez2gL4",
+        ExportName = "sceAmprCommandBufferReset",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferReset(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!ctx.TryReadUInt64(commandBuffer + CommandBufferDataOffset, out var buffer) ||
+            !ctx.TryReadUInt64(commandBuffer + CommandBufferSizeOffset, out var size) ||
+            !WriteCommandBufferPointers(ctx, commandBuffer, buffer, size, writeOffset: 0))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "reset", commandBuffer, buffer, size);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "ULvXMDz56po",
+        ExportName = "sceAmprCommandBufferClearBuffer",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferClearBuffer(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!TryGetCommandBufferState(ctx, commandBuffer, out var buffer, out var size, out _) ||
+            !WriteVisibleCommandBufferPointers(ctx, commandBuffer, buffer: 0, size: 0))
+        {
+            ctx[CpuRegister.Rax] = 0;
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        _commandBuffers.TryRemove(commandBuffer, out _);
+        TraceAmpr(ctx, "clear_buffer", commandBuffer, buffer, size);
+        ctx[CpuRegister.Rax] = buffer;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "mQ16-QdKv7k",
+        ExportName = "sceAmprAprCommandBufferReadFile",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int AprCommandBufferReadFile(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var fileId = unchecked((uint)ctx[CpuRegister.Rcx]);
+        var destination = ctx[CpuRegister.R8];
+        var size = ctx[CpuRegister.R9];
+
+        if (commandBuffer == 0 || (destination == 0 && size != 0))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!ctx.TryReadUInt64(ctx[CpuRegister.Rsp] + sizeof(ulong), out var fileOffset))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (!AmprFileRegistry.TryGetHostPath(fileId, out var hostPath) || !File.Exists(hostPath))
+        {
+            TraceAmprRead(ctx, commandBuffer, fileId, destination, size, fileOffset, bytesRead: 0, hostPath, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
+        }
+
+        var result = TryReadFileToGuestMemory(ctx, hostPath, fileOffset, destination, size, out var bytesRead);
+        if (result != (int)OrbisGen2Result.ORBIS_GEN2_OK)
+        {
+            TraceAmprRead(ctx, commandBuffer, fileId, destination, size, fileOffset, bytesRead, hostPath, result);
+            return result;
+        }
+
+        if (!AppendReadFileRecord(ctx, commandBuffer, fileId, destination, size, fileOffset, bytesRead))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmprRead(ctx, commandBuffer, fileId, destination, size, fileOffset, bytesRead, hostPath, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "vWU-odnS+fU",
+        ExportName = "sceAmprMeasureCommandSizeReadFile",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int MeasureCommandSizeReadFile(CpuContext ctx)
+    {
+        TraceAmpr(ctx, "measure_read_file", 0, ReadFileRecordSize, 0);
+        ctx[CpuRegister.Rax] = ReadFileRecordSize;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "sSAUCCU1dv4",
+        ExportName = "sceAmprMeasureCommandSizeWriteKernelEventQueue_04_00",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int MeasureCommandSizeWriteKernelEventQueue0400(CpuContext ctx)
+    {
+        TraceAmpr(ctx, "measure_write_equeue", 0, KernelEventQueueRecordSize, 0);
+        ctx[CpuRegister.Rax] = KernelEventQueueRecordSize;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "tZDDEo2tE5k",
+        ExportName = "sceAmprCommandBufferGetSize",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferGetSize(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!TryGetCommandBufferState(ctx, commandBuffer, out _, out var size, out _))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "get_size", commandBuffer, size, 0);
+        ctx[CpuRegister.Rax] = size;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "GnxKOHEawhk",
+        ExportName = "sceAmprCommandBufferGetCurrentOffset",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferGetCurrentOffset(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!TryGetCommandBufferOffset(ctx, commandBuffer, out var offset))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "get_offset", commandBuffer, offset, 0);
+        ctx[CpuRegister.Rax] = offset;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "H896Pt-yB4I",
+        ExportName = "sceAmprCommandBufferWriteKernelEventQueue_04_00",
+        Target = Generation.Gen5,
+        LibraryName = "libSceAmpr")]
+    public static int CommandBufferWriteKernelEventQueue0400(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var equeue = ctx[CpuRegister.Rsi];
+        var ident = ctx[CpuRegister.Rdx];
+        var filter = 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))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceAmpr(ctx, "write_equeue", commandBuffer, equeue, ident);
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static bool InitializeCommandBuffer(
+        CpuContext ctx,
+        ulong commandBuffer,
+        ulong buffer,
+        ulong size,
+        ulong aux0,
+        ulong aux1,
+        bool clear)
+    {
+        if (clear)
+        {
+            Span<byte> header = stackalloc byte[CommandBufferHeaderSize];
+            header.Clear();
+            if (!ctx.Memory.TryWrite(commandBuffer, header))
+            {
+                return false;
+            }
+        }
+
+        return ctx.TryWriteUInt64(commandBuffer + CommandBufferSelfOffset, commandBuffer) &&
+               ctx.TryWriteUInt64(commandBuffer + CommandBufferAux0Offset, aux0) &&
+               ctx.TryWriteUInt64(commandBuffer + CommandBufferAux1Offset, aux1) &&
+               WriteCommandBufferPointers(ctx, commandBuffer, buffer, size, writeOffset: 0);
+    }
+
+    private static bool WriteCommandBufferPointers(CpuContext ctx, ulong commandBuffer, ulong buffer, ulong size)
+    {
+        return WriteCommandBufferPointers(ctx, commandBuffer, buffer, size, writeOffset: 0);
+    }
+
+    private static bool WriteCommandBufferPointers(CpuContext ctx, ulong commandBuffer, ulong buffer, ulong size, ulong writeOffset)
+    {
+        if (!WriteVisibleCommandBufferPointers(ctx, commandBuffer, buffer, size))
+        {
+            return false;
+        }
+
+        var state = _commandBuffers.GetOrAdd(commandBuffer, static _ => new CommandBufferState());
+        lock (state)
+        {
+            state.Buffer = buffer;
+            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(
+        CpuContext ctx,
+        ulong commandBuffer,
+        out ulong buffer,
+        out ulong size,
+        out CommandBufferState? state)
+    {
+        if (_commandBuffers.TryGetValue(commandBuffer, out state))
+        {
+            lock (state)
+            {
+                buffer = state.Buffer;
+                size = state.Size;
+            }
+
+            return true;
+        }
+
+        if (ctx.TryReadUInt64(commandBuffer + CommandBufferDataOffset, out buffer) &&
+            ctx.TryReadUInt64(commandBuffer + CommandBufferSizeOffset, out size))
+        {
+            state = _commandBuffers.GetOrAdd(commandBuffer, static _ => new CommandBufferState());
+            lock (state)
+            {
+                state.Buffer = buffer;
+                state.Size = size;
+                state.WriteOffset = 0;
+            }
+
+            return true;
+        }
+
+        buffer = 0;
+        size = 0;
+        state = null;
+        return false;
+    }
+
+    private static bool TryGetCommandBufferOffset(CpuContext ctx, ulong commandBuffer, out ulong offset)
+    {
+        if (!TryGetCommandBufferState(ctx, commandBuffer, out _, out _, out var state) || state is null)
+        {
+            offset = 0;
+            return false;
+        }
+
+        lock (state)
+        {
+            offset = state.WriteOffset;
+        }
+
+        return true;
+    }
+
+    private static int TryReadFileToGuestMemory(
+        CpuContext ctx,
+        string hostPath,
+        ulong fileOffset,
+        ulong destination,
+        ulong size,
+        out ulong bytesRead)
+    {
+        bytesRead = 0;
+        if (size == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (fileOffset > long.MaxValue)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        const int ChunkSize = 64 * 1024;
+        var buffer = new byte[(int)Math.Min((ulong)ChunkSize, size)];
+
+        try
+        {
+            using var stream = new FileStream(hostPath, FileMode.Open, FileAccess.Read, FileShare.ReadWrite | FileShare.Delete);
+            if (fileOffset >= (ulong)stream.Length)
+            {
+                return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+            }
+
+            stream.Position = unchecked((long)fileOffset);
+
+            while (bytesRead < size)
+            {
+                var request = (int)Math.Min((ulong)buffer.Length, size - bytesRead);
+                var read = stream.Read(buffer, 0, request);
+                if (read <= 0)
+                {
+                    break;
+                }
+
+                if (!ctx.Memory.TryWrite(destination + bytesRead, buffer.AsSpan(0, read)))
+                {
+                    return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+                }
+
+                bytesRead += (ulong)read;
+            }
+        }
+        catch (UnauthorizedAccessException)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_PERMISSION_DENIED;
+        }
+        catch (IOException)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
+        }
+
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static bool AppendReadFileRecord(
+        CpuContext ctx,
+        ulong commandBuffer,
+        uint fileId,
+        ulong destination,
+        ulong size,
+        ulong fileOffset,
+        ulong bytesRead)
+    {
+        Span<byte> record = stackalloc byte[(int)ReadFileRecordSize];
+        record.Clear();
+        BinaryPrimitives.WriteUInt32LittleEndian(record[0x00..], ReadFileRecordType);
+        BinaryPrimitives.WriteUInt32LittleEndian(record[0x04..], fileId);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x08..], destination);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x10..], size);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x18..], fileOffset);
+        BinaryPrimitives.WriteUInt64LittleEndian(record[0x20..], bytesRead);
+
+        return AppendCommandBufferRecord(ctx, commandBuffer, record);
+    }
+
+    private static bool AppendKernelEventQueueRecord(
+        CpuContext ctx,
+        ulong commandBuffer,
+        ulong equeue,
+        ulong ident,
+        ulong filter,
+        ulong userData,
+        ulong data,
+        ulong extra)
+    {
+        Span<byte> record = stackalloc byte[(int)KernelEventQueueRecordSize];
+        record.Clear();
+        BinaryPrimitives.WriteUInt32LittleEndian(record[0x00..], KernelEventQueueRecordType);
+        BinaryPrimitives.WriteUInt32LittleEndian(record[0x04..], unchecked((uint)filter));
+        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);
+
+        return AppendCommandBufferRecord(ctx, commandBuffer, record);
+    }
+
+    private static bool AppendCommandBufferRecord(CpuContext ctx, ulong commandBuffer, ReadOnlySpan<byte> record)
+    {
+        if (!TryGetCommandBufferState(ctx, commandBuffer, out _, out _, out var state) || state is null)
+        {
+            return false;
+        }
+
+        var recordSize = (ulong)record.Length;
+        lock (state)
+        {
+            if (state.Buffer == 0 ||
+                state.WriteOffset > state.Size ||
+                recordSize > state.Size - state.WriteOffset)
+            {
+                return false;
+            }
+
+            if (!ctx.Memory.TryWrite(state.Buffer + state.WriteOffset, record))
+            {
+                return false;
+            }
+
+            state.WriteOffset += recordSize;
+        }
+
+        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))
+        {
+            return;
+        }
+
+        var returnRip = 0UL;
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] ampr.{operation}: cmd=0x{commandBuffer:X16} arg0=0x{arg0:X16} arg1=0x{arg1:X16} ret=0x{returnRip:X16}");
+    }
+
+    private static void TraceAmprRead(
+        CpuContext ctx,
+        ulong commandBuffer,
+        uint fileId,
+        ulong destination,
+        ulong size,
+        ulong fileOffset,
+        ulong bytesRead,
+        string? hostPath,
+        int result)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        var returnRip = 0UL;
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] ampr.read_file: cmd=0x{commandBuffer:X16} id=0x{fileId:X8} dst=0x{destination:X16} size=0x{size:X16} offset=0x{fileOffset:X16} read=0x{bytesRead:X16} result=0x{result:X8} path='{hostPath ?? string.Empty}' ret=0x{returnRip:X16}");
+    }
+}

src/SharpEmu.Libs/Ampr/AmprFileRegistry.cs

@@ -0,0 +1,40 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using System.Collections.Concurrent;
+
+namespace SharpEmu.Libs.Ampr;
+
+internal static class AmprFileRegistry
+{
+    private static readonly ConcurrentDictionary<uint, string> _hostPathsById = new();
+
+    public static uint Register(string guestPath, string hostPath)
+    {
+        var id = ComputeFileId(guestPath);
+        _hostPathsById[id] = hostPath;
+        return id;
+    }
+
+    public static bool TryGetHostPath(uint id, out string hostPath)
+    {
+        return _hostPathsById.TryGetValue(id, out hostPath!);
+    }
+
+    private static uint ComputeFileId(string guestPath)
+    {
+        var bytes = System.Text.Encoding.UTF8.GetBytes(guestPath);
+
+        const uint offsetBasis = 2166136261;
+        const uint prime = 16777619;
+
+        var hash = offsetBasis;
+        foreach (var b in bytes)
+        {
+            hash ^= b;
+            hash *= prime;
+        }
+
+        return hash;
+    }
+}

src/SharpEmu.Libs/Kernel/KernelAprCompatExports.cs

@@ -0,0 +1,158 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Collections.Concurrent;
+using System.Threading;
+
+namespace SharpEmu.Libs.Kernel;
+
+public static class KernelAprCompatExports
+{
+    private static readonly ConcurrentDictionary<uint, ulong> _submittedCommandBuffers = new();
+    private static int _nextSubmissionId;
+
+    [SysAbiExport(
+        Nid = "ASoW5WE-UPo",
+        ExportName = "sceKernelAprSubmitCommandBufferAndGetResult",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAprSubmitCommandBufferAndGetResult(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var priority = ctx[CpuRegister.Rsi];
+        var resultAddress = ctx[CpuRegister.Rdx];
+        var outSubmissionId = ctx[CpuRegister.Rcx];
+
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
+        if (submissionId == 0)
+        {
+            submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
+        }
+
+        _submittedCommandBuffers[submissionId] = commandBuffer;
+
+        if (outSubmissionId != 0 && !TryWriteUInt32(ctx, outSubmissionId, submissionId))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (resultAddress != 0 && !TryWriteAprResult(ctx, resultAddress))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceApr(ctx, "submit_get_result", submissionId, commandBuffer, priority, resultAddress);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "rqwFKI4PAiM",
+        ExportName = "sceKernelAprWaitCommandBuffer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAprWaitCommandBuffer(CpuContext ctx)
+    {
+        var submissionId = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var priority = ctx[CpuRegister.Rsi];
+        var resultAddress = ctx[CpuRegister.Rdx];
+
+        if (!_submittedCommandBuffers.TryRemove(submissionId, out var commandBuffer))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
+        }
+
+        if (resultAddress != 0 && !TryWriteAprResult(ctx, resultAddress))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceApr(ctx, "wait", submissionId, commandBuffer, priority, resultAddress);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "eE4Szl8sil8",
+        ExportName = "sceKernelAprSubmitCommandBuffer",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAprSubmitCommandBuffer(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        if (commandBuffer == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
+        _submittedCommandBuffers[submissionId] = commandBuffer;
+
+        TraceApr(ctx, "submit", submissionId, commandBuffer, ctx[CpuRegister.Rsi], 0);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "qvMUCyyaCSI",
+        ExportName = "sceKernelAprSubmitCommandBufferAndGetId",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAprSubmitCommandBufferAndGetId(CpuContext ctx)
+    {
+        var commandBuffer = ctx[CpuRegister.Rdi];
+        var outSubmissionId = ctx[CpuRegister.Rdx];
+        if (commandBuffer == 0 || outSubmissionId == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var submissionId = unchecked((uint)Interlocked.Increment(ref _nextSubmissionId));
+        _submittedCommandBuffers[submissionId] = commandBuffer;
+
+        if (!TryWriteUInt32(ctx, outSubmissionId, submissionId))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceApr(ctx, "submit_get_id", submissionId, commandBuffer, ctx[CpuRegister.Rsi], outSubmissionId);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static bool TryWriteAprResult(CpuContext ctx, ulong resultAddress)
+    {
+        Span<byte> result = stackalloc byte[sizeof(ulong)];
+        result.Clear();
+        return ctx.Memory.TryWrite(resultAddress, result);
+    }
+
+    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 void TraceApr(
+        CpuContext ctx,
+        string operation,
+        uint submissionId,
+        ulong commandBuffer,
+        ulong priority,
+        ulong aux)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_AMPR"), "1", StringComparison.Ordinal))
+        {
+            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} priority=0x{priority:X16} aux=0x{aux:X16} ret=0x{returnRip:X16}");
+    }
+}

src/SharpEmu.Libs/Kernel/KernelEventQueueCompatExports.cs

@@ -0,0 +1,302 @@
+// 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.Kernel;
+
+public static class KernelEventQueueCompatExports
+{
+    private const int KernelEventSize = 0x20;
+
+    private static readonly object _eventQueueGate = new();
+    private static readonly HashSet<ulong> _eventQueues = new();
+    private static readonly Dictionary<ulong, Queue<KernelQueuedEvent>> _pendingEvents = new();
+    private static long _nextEventQueueHandle = 1;
+
+    public readonly record struct KernelQueuedEvent(
+        ulong Ident,
+        short Filter,
+        ushort Flags,
+        uint Fflags,
+        ulong Data,
+        ulong UserData);
+
+    [SysAbiExport(
+        Nid = "D0OdFMjp46I",
+        ExportName = "sceKernelCreateEqueue",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelCreateEqueue(CpuContext ctx)
+    {
+        var outAddress = ctx[CpuRegister.Rdi];
+        if (outAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        var handle = unchecked((ulong)Interlocked.Increment(ref _nextEventQueueHandle));
+        lock (_eventQueueGate)
+        {
+            _eventQueues.Add(handle);
+            _pendingEvents[handle] = new Queue<KernelQueuedEvent>();
+        }
+
+        if (!ctx.TryWriteUInt64(outAddress, handle))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        TraceEventQueue(ctx, "create", handle);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "jpFjmgAC5AE",
+        ExportName = "sceKernelDeleteEqueue",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteEqueue(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        lock (_eventQueueGate)
+        {
+            _eventQueues.Remove(handle);
+            _pendingEvents.Remove(handle);
+        }
+
+        TraceEventQueue(ctx, "delete", handle);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "WDszmSbWuDk",
+        ExportName = "sceKernelAddUserEventEdge",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAddUserEventEdge(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "add_user_edge", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "4R6-OvI2cEA",
+        ExportName = "sceKernelAddUserEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAddUserEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "add_user", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "LJDwdSNTnDg",
+        ExportName = "sceKernelDeleteUserEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteUserEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "delete_user", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "F6e0kwo4cnk",
+        ExportName = "sceKernelTriggerUserEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelTriggerUserEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "trigger_user", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "bBfz7kMF2Ho",
+        ExportName = "sceKernelAddAmprEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAddAmprEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "add_ampr", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "vuae5JPNt9A",
+        ExportName = "sceKernelAddAmprSystemEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelAddAmprSystemEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "add_ampr_system", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "bMmid3pfyjo",
+        ExportName = "sceKernelDeleteAmprEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteAmprEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "delete_ampr", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "Ij+ryuEClXQ",
+        ExportName = "sceKernelDeleteAmprSystemEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteAmprSystemEvent(CpuContext ctx)
+    {
+        TraceEventQueue(ctx, "delete_ampr_system", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "QyrxcdBrb0M",
+        ExportName = "sceKernelGetKqueueFromEqueue",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelGetKqueueFromEqueue(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = ctx[CpuRegister.Rdi];
+        TraceEventQueue(ctx, "get_kqueue", ctx[CpuRegister.Rdi]);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "fzyMKs9kim0",
+        ExportName = "sceKernelWaitEqueue",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelWaitEqueue(CpuContext ctx)
+    {
+        var handle = ctx[CpuRegister.Rdi];
+        var eventsAddress = ctx[CpuRegister.Rsi];
+        var eventCapacity = (int)Math.Min(ctx[CpuRegister.Rdx], int.MaxValue);
+        var outCountAddress = ctx[CpuRegister.Rcx];
+        var timeoutAddress = ctx[CpuRegister.R8];
+
+        var deliveredCount = DequeueEvents(ctx, handle, eventsAddress, eventCapacity);
+        if (outCountAddress != 0 && !TryWriteUInt32(ctx, outCountAddress, (uint)deliveredCount))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (deliveredCount > 0)
+        {
+            TraceEventQueue(ctx, "wait-deliver", handle);
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        if (timeoutAddress == 0 && GuestThreadExecution.RequestCurrentThreadBlock("sceKernelWaitEqueue"))
+        {
+            TraceEventQueue(ctx, "wait-block", handle);
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        }
+
+        TraceEventQueue(ctx, "wait", handle);
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    public static bool IsValidEqueue(ulong handle)
+    {
+        lock (_eventQueueGate)
+        {
+            return _eventQueues.Contains(handle);
+        }
+    }
+
+    public static bool EnqueueEvent(ulong handle, KernelQueuedEvent queuedEvent)
+    {
+        lock (_eventQueueGate)
+        {
+            if (!_eventQueues.Contains(handle))
+            {
+                return false;
+            }
+
+            if (!_pendingEvents.TryGetValue(handle, out var queue))
+            {
+                queue = new Queue<KernelQueuedEvent>();
+                _pendingEvents[handle] = queue;
+            }
+
+            queue.Enqueue(queuedEvent);
+            return true;
+        }
+    }
+
+    private static int DequeueEvents(CpuContext ctx, ulong handle, ulong eventsAddress, int eventCapacity)
+    {
+        if (eventsAddress == 0 || eventCapacity <= 0)
+        {
+            return 0;
+        }
+
+        KernelQueuedEvent[] events;
+        lock (_eventQueueGate)
+        {
+            if (!_pendingEvents.TryGetValue(handle, out var queue) || queue.Count == 0)
+            {
+                return 0;
+            }
+
+            var count = Math.Min(eventCapacity, queue.Count);
+            events = new KernelQueuedEvent[count];
+            for (var i = 0; i < count; i++)
+            {
+                events[i] = queue.Dequeue();
+            }
+        }
+
+        for (var i = 0; i < events.Length; i++)
+        {
+            if (!WriteKernelEvent(ctx, eventsAddress + ((ulong)i * KernelEventSize), events[i]))
+            {
+                return i;
+            }
+        }
+
+        return events.Length;
+    }
+
+    private static bool WriteKernelEvent(CpuContext ctx, ulong address, KernelQueuedEvent queuedEvent)
+    {
+        Span<byte> eventBytes = stackalloc byte[KernelEventSize];
+        BinaryPrimitives.WriteUInt64LittleEndian(eventBytes[0x00..], queuedEvent.Ident);
+        BinaryPrimitives.WriteInt16LittleEndian(eventBytes[0x08..], queuedEvent.Filter);
+        BinaryPrimitives.WriteUInt16LittleEndian(eventBytes[0x0A..], queuedEvent.Flags);
+        BinaryPrimitives.WriteUInt32LittleEndian(eventBytes[0x0C..], queuedEvent.Fflags);
+        BinaryPrimitives.WriteUInt64LittleEndian(eventBytes[0x10..], queuedEvent.Data);
+        BinaryPrimitives.WriteUInt64LittleEndian(eventBytes[0x18..], queuedEvent.UserData);
+        return ctx.Memory.TryWrite(address, eventBytes);
+    }
+
+    private static void TraceEventQueue(CpuContext ctx, string operation, ulong handle)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_EQUEUE"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        var returnRip = 0UL;
+        _ = ctx.TryReadUInt64(ctx[CpuRegister.Rsp], out returnRip);
+        Console.Error.WriteLine(
+            $"[LOADER][TRACE] equeue.{operation}: handle=0x{handle:X16} rsi=0x{ctx[CpuRegister.Rsi]:X16} rdx=0x{ctx[CpuRegister.Rdx]:X16} ret=0x{returnRip:X16}");
+    }
+
+    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);
+    }
+}

src/SharpEmu.Libs/Kernel/KernelExports.cs

@@ -55,7 +55,10 @@ public static class KernelExports
         LibraryName = "libKernel")]
     public static int Exit(CpuContext ctx)
     {
-        _ = ctx;
+        var status = unchecked((int)ctx[CpuRegister.Rdi]);
+        Console.Error.WriteLine($"[LOADER][INFO] exit(status={status})");
+        GuestThreadExecution.RequestCurrentEntryExit("exit", status);
+        ctx[CpuRegister.Rax] = unchecked((ulong)status);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -68,7 +71,8 @@ public static class KernelExports
     {
         var status = unchecked((int)ctx[CpuRegister.Rdi]);
         Console.Error.WriteLine($"[LOADER][INFO] catchReturnFromMain(status={status})");
-        ctx[CpuRegister.Rax] = 0;
+        GuestThreadExecution.RequestCurrentEntryExit("catchReturnFromMain", status);
+        ctx[CpuRegister.Rax] = unchecked((ulong)status);
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -211,6 +215,19 @@ public static class KernelExports
                 $"[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}");
         }
 
+        var scheduler = GuestThreadExecution.Scheduler;
+        if (scheduler is not null && entryAddress != 0)
+        {
+            var request = new GuestThreadStartRequest(threadHandle, entryAddress, argument, attrAddress, name);
+            if (!scheduler.TryStartThread(ctx, request, out var error))
+            {
+                Console.Error.WriteLine(
+                    $"[LOADER][ERROR] pthread_create: failed to schedule guest thread '{name}' entry=0x{entryAddress:X16}: {error}");
+                ctx[CpuRegister.Rax] = unchecked((ulong)(int)OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN);
+                return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN;
+            }
+        }
+
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -225,6 +242,16 @@ public static class KernelExports
         return PthreadCreate(ctx);
     }
 
+    [SysAbiExport(
+        Nid = "Jmi+9w9u0E4",
+        ExportName = "pthread_create_name_np",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadCreateNameNp(CpuContext ctx)
+    {
+        return PthreadCreate(ctx);
+    }
+
     [SysAbiExport(
         Nid = "onNY9Byn-W8",
         ExportName = "scePthreadJoin",
@@ -264,22 +291,14 @@ public static class KernelExports
         ExportName = "open",
         Target = Generation.Gen4 | Generation.Gen5,
         LibraryName = "libKernel")]
-    public static int Open(CpuContext ctx)
-    {
-        ctx[CpuRegister.Rax] = unchecked((ulong)Interlocked.Increment(ref _nextFileDescriptor));
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
-    }
+    public static int Open(CpuContext ctx) => KernelMemoryCompatExports.KernelOpenUnderscore(ctx);
 
     [SysAbiExport(
         Nid = "1G3lF1Gg1k8",
         ExportName = "sceKernelOpen",
         Target = Generation.Gen4 | Generation.Gen5,
         LibraryName = "libKernel")]
-    public static int KernelOpen(CpuContext ctx)
-    {
-        ctx[CpuRegister.Rax] = unchecked((ulong)Interlocked.Increment(ref _nextFileDescriptor));
-        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
-    }
+    public static int KernelOpen(CpuContext ctx) => KernelMemoryCompatExports.KernelOpenUnderscore(ctx);
 
     [SysAbiExport(
         Nid = "hcuQgD53UxM",

src/SharpEmu.Libs/Kernel/KernelPthreadCompatExports.cs

@@ -18,6 +18,7 @@ public static class KernelPthreadCompatExports
     private const ulong SyntheticMutexHandleBase = 0x00006000_0000_0000;
     private const ulong SyntheticMutexAttrHandleBase = 0x00006001_0000_0000;
     private const ulong SyntheticCondHandleBase = 0x00006002_0000_0000;
+    private const int DefaultSpuriousCondWakeMilliseconds = 1;
 
     private static readonly object _stateGate = new();
     private static readonly Dictionary<ulong, PthreadMutexState> _mutexStates = new();
@@ -237,6 +238,13 @@ public static class KernelPthreadCompatExports
         LibraryName = "libKernel")]
     public static int PthreadCondInit(CpuContext ctx) => PthreadCondInitCore(ctx, ctx[CpuRegister.Rdi]);
 
+    [SysAbiExport(
+        Nid = "0TyVk4MSLt0",
+        ExportName = "pthread_cond_init",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadCondInit(CpuContext ctx) => PthreadCondInitCore(ctx, ctx[CpuRegister.Rdi]);
+
     [SysAbiExport(
         Nid = "g+PZd2hiacg",
         ExportName = "scePthreadCondDestroy",
@@ -286,6 +294,13 @@ public static class KernelPthreadCompatExports
         LibraryName = "libKernel")]
     public static int PosixPthreadCondBroadcast(CpuContext ctx) => PthreadCondSignalCore(ctx, ctx[CpuRegister.Rdi], broadcast: true);
 
+    [SysAbiExport(
+        Nid = "2MOy+rUfuhQ",
+        ExportName = "pthread_cond_signal",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadCondSignal(CpuContext ctx) => PthreadCondSignalCore(ctx, ctx[CpuRegister.Rdi], broadcast: false);
+
     [SysAbiExport(
         Nid = "m5-2bsNfv7s",
         ExportName = "scePthreadCondattrInit",
@@ -407,6 +422,23 @@ public static class KernelPthreadCompatExports
                     return (int)OrbisGen2Result.ORBIS_GEN2_OK;
                 }
 
+                if (state.Type is MutexTypeNormal or MutexTypeAdaptiveNp)
+                {
+                    if (tryOnly)
+                    {
+                        TracePthreadMutex(ctx, "trylock", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
+                        return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY;
+                    }
+
+                    // Normal/adaptive mutexes do not report EDEADLK on self-lock.
+                    // Under the current single-host-thread guest execution model,
+                    // treating them as nested ownership keeps init paths moving
+                    // without turning a would-block path into a hard error.
+                    state.RecursionCount++;
+                    TracePthreadMutex(ctx, "lock", mutexAddress, resolvedAddress, state, currentThreadId, (int)OrbisGen2Result.ORBIS_GEN2_OK);
+                    return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+                }
+
                 var ownedResult = tryOnly
                     ? (int)OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY
                     : (int)OrbisGen2Result.ORBIS_GEN2_ERROR_DEADLOCK;
@@ -856,6 +888,7 @@ public static class KernelPthreadCompatExports
         }
 
         var waitResult = (int)OrbisGen2Result.ORBIS_GEN2_OK;
+        var spuriousWake = false;
         lock (state.SyncRoot)
         {
             state.Waiters++;
@@ -870,11 +903,36 @@ public static class KernelPthreadCompatExports
                 return unlockResult;
             }
 
+            var scheduler = GuestThreadExecution.Scheduler;
+            if (!timed && GuestThreadExecution.RequestCurrentThreadBlock("pthread_cond_wait"))
+            {
+                TracePthreadCond("wait-block", condAddress, mutexAddress, state, timed, waitResult);
+                return waitResult;
+            }
+
+            if (scheduler is not null)
+            {
+                Monitor.Exit(state.SyncRoot);
+                try
+                {
+                    scheduler.Pump(ctx, "pthread_cond_wait");
+                }
+                finally
+                {
+                    Monitor.Enter(state.SyncRoot);
+                }
+            }
+
             while (state.SignalEpoch == observedEpoch)
             {
                 if (!timed)
                 {
-                    Monitor.Wait(state.SyncRoot);
+                    if (!Monitor.Wait(state.SyncRoot, GetCondSpuriousWakeTimeout()))
+                    {
+                        spuriousWake = true;
+                        break;
+                    }
+
                     continue;
                 }
 
@@ -886,7 +944,15 @@ public static class KernelPthreadCompatExports
             }
 
             state.Waiters = Math.Max(0, state.Waiters - 1);
-            TracePthreadCond(waitResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? "wait-wake" : "wait-timeout", condAddress, mutexAddress, state, timed, waitResult);
+            TracePthreadCond(
+                spuriousWake
+                    ? "wait-spurious"
+                    : (waitResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? "wait-wake" : "wait-timeout"),
+                condAddress,
+                mutexAddress,
+                state,
+                timed,
+                waitResult);
         }
 
         var lockResult = PthreadMutexLockCore(ctx, mutexAddress, tryOnly: false);
@@ -896,7 +962,15 @@ public static class KernelPthreadCompatExports
             return lockResult;
         }
 
-        TracePthreadCond(waitResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? "wait-exit" : "wait-exit-timeout", condAddress, mutexAddress, state, timed, waitResult);
+        TracePthreadCond(
+            spuriousWake
+                ? "wait-exit-spurious"
+                : (waitResult == (int)OrbisGen2Result.ORBIS_GEN2_OK ? "wait-exit" : "wait-exit-timeout"),
+            condAddress,
+            mutexAddress,
+            state,
+            timed,
+            waitResult);
         return waitResult;
     }
 
@@ -943,6 +1017,16 @@ public static class KernelPthreadCompatExports
         return TimeSpan.FromTicks((long)timeoutUsec * 10L);
     }
 
+    private static TimeSpan GetCondSpuriousWakeTimeout()
+    {
+        if (int.TryParse(Environment.GetEnvironmentVariable("SHARPEMU_PTHREAD_COND_SPURIOUS_WAKE_MS"), out var milliseconds))
+        {
+            return TimeSpan.FromMilliseconds(Math.Max(1, milliseconds));
+        }
+
+        return TimeSpan.FromMilliseconds(DefaultSpuriousCondWakeMilliseconds);
+    }
+
     private static int NormalizeMutexType(int type)
     {
         return type switch

src/SharpEmu.Libs/Kernel/KernelPthreadExtendedCompatExports.cs

@@ -20,6 +20,7 @@ public static class KernelPthreadExtendedCompatExports
     private const int DefaultSchedPolicy = 0;
     private const int DefaultSchedPriority = 0;
     private const ulong SyntheticRwlockHandleBase = 0x00006003_0000_0000;
+    private const ulong SyntheticPthreadAttrHandleBase = 0x00006004_0000_0000;
 
     private static readonly object _stateGate = new();
     private static readonly Dictionary<ulong, ThreadState> _threadStates = new();
@@ -28,9 +29,9 @@ public static class KernelPthreadExtendedCompatExports
     private static readonly Dictionary<int, TlsKeyState> _tlsKeys = new();
     private static int _nextTlsKey = 1;
     private static long _nextSyntheticRwlockHandleId = 1;
+    private static long _nextSyntheticPthreadAttrHandleId = 1;
 
-    [ThreadStatic]
-    private static Dictionary<int, ulong>? _threadLocalSpecific;
+    private static readonly Dictionary<ulong, Dictionary<int, ulong>> _threadLocalSpecific = new();
 
     private sealed class ThreadState
     {
@@ -236,6 +237,41 @@ public static class KernelPthreadExtendedCompatExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "Xs9hdiD7sAA",
+        ExportName = "pthread_setschedparam",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadSetschedparam(CpuContext ctx)
+    {
+        var thread = ctx[CpuRegister.Rdi];
+        var policy = unchecked((int)ctx[CpuRegister.Rsi]);
+        var schedParamAddress = ctx[CpuRegister.Rdx];
+        if (thread == 0 || schedParamAddress == 0)
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
+        }
+
+        if (!TryReadInt32(ctx, schedParamAddress, out var schedPriority))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        lock (_stateGate)
+        {
+            var state = GetOrCreateThreadStateLocked(thread);
+            state.Priority = schedPriority;
+            state.Attributes = state.Attributes with
+            {
+                SchedPolicy = policy,
+                SchedPriority = schedPriority,
+            };
+        }
+
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
     [SysAbiExport(
         Nid = "nsYoNRywwNg",
         ExportName = "scePthreadAttrInit",
@@ -249,15 +285,32 @@ public static class KernelPthreadExtendedCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
+        var syntheticHandle = AllocateSyntheticHandle(SyntheticPthreadAttrHandleBase, ref _nextSyntheticPthreadAttrHandleId);
+        if (!ctx.TryWriteUInt64(attrAddress, syntheticHandle))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
         lock (_stateGate)
         {
             _attrStates[attrAddress] = PthreadAttrState.Default;
+            _attrStates[syntheticHandle] = PthreadAttrState.Default;
         }
 
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "wtkt-teR1so",
+        ExportName = "pthread_attr_init",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadAttrInit(CpuContext ctx)
+    {
+        return PthreadAttrInit(ctx);
+    }
+
     [SysAbiExport(
         Nid = "62KCwEMmzcM",
         ExportName = "scePthreadAttrDestroy",
@@ -271,15 +324,31 @@ public static class KernelPthreadExtendedCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
+        var resolvedAddress = ResolvePthreadAttrHandle(ctx, attrAddress);
         lock (_stateGate)
         {
             _attrStates.Remove(attrAddress);
+            if (resolvedAddress != attrAddress)
+            {
+                _attrStates.Remove(resolvedAddress);
+            }
         }
 
+        _ = ctx.TryWriteUInt64(attrAddress, 0);
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "zHchY8ft5pk",
+        ExportName = "pthread_attr_destroy",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadAttrDestroy(CpuContext ctx)
+    {
+        return PthreadAttrDestroy(ctx);
+    }
+
     [SysAbiExport(
         Nid = "x1X76arYMxU",
         ExportName = "scePthreadAttrGet",
@@ -612,16 +681,32 @@ public static class KernelPthreadExtendedCompatExports
             return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT;
         }
 
+        var resolvedAddress = ResolvePthreadAttrHandle(ctx, attrAddress);
         lock (_stateGate)
         {
-            var state = GetOrCreateAttrStateLocked(attrAddress);
-            _attrStates[attrAddress] = state with { StackSize = stackSize };
+            var state = GetOrCreateAttrStateLocked(resolvedAddress);
+            var updated = state with { StackSize = stackSize };
+            _attrStates[resolvedAddress] = updated;
+            if (resolvedAddress != attrAddress)
+            {
+                _attrStates[attrAddress] = updated;
+            }
         }
 
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "2Q0z6rnBrTE",
+        ExportName = "pthread_attr_setstacksize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int PosixPthreadAttrSetstacksize(CpuContext ctx)
+    {
+        return PthreadAttrSetstacksize(ctx);
+    }
+
     [SysAbiExport(
         Nid = "6ULAa0fq4jA",
         ExportName = "scePthreadRwlockInit",
@@ -855,9 +940,13 @@ public static class KernelPthreadExtendedCompatExports
             {
                 return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND;
             }
+
+            foreach (var entry in _threadLocalSpecific)
+            {
+                entry.Value.Remove(key);
+            }
         }
 
-        _threadLocalSpecific?.Remove(key);
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -878,16 +967,23 @@ public static class KernelPthreadExtendedCompatExports
     {
         var key = unchecked((int)ctx[CpuRegister.Rdi]);
         var value = ctx[CpuRegister.Rsi];
+        var currentThreadHandle = KernelPthreadState.GetCurrentThreadHandle();
         lock (_stateGate)
         {
             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;
         }
 
-        _threadLocalSpecific ??= new Dictionary<int, ulong>();
-        _threadLocalSpecific[key] = value;
         ctx[CpuRegister.Rax] = 0;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -907,6 +1003,8 @@ public static class KernelPthreadExtendedCompatExports
     public static int PosixPthreadGetspecific(CpuContext ctx)
     {
         var key = unchecked((int)ctx[CpuRegister.Rdi]);
+        var currentThreadHandle = KernelPthreadState.GetCurrentThreadHandle();
+        ulong value = 0;
         lock (_stateGate)
         {
             if (!_tlsKeys.TryGetValue(key, out _))
@@ -914,12 +1012,15 @@ public static class KernelPthreadExtendedCompatExports
                 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;
+            }
         }
 
-        ctx[CpuRegister.Rax] =
-            _threadLocalSpecific is not null && _threadLocalSpecific.TryGetValue(key, out var value)
-                ? value
-                : 0UL;
+        ctx[CpuRegister.Rax] = value;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
@@ -1116,6 +1217,35 @@ public static class KernelPthreadExtendedCompatExports
         return state;
     }
 
+    private static ulong ResolvePthreadAttrHandle(CpuContext ctx, ulong attrAddress)
+    {
+        if (attrAddress == 0)
+        {
+            return 0;
+        }
+
+        lock (_stateGate)
+        {
+            if (_attrStates.ContainsKey(attrAddress))
+            {
+                return attrAddress;
+            }
+        }
+
+        if (ctx.TryReadUInt64(attrAddress, out var pointedHandle) && pointedHandle != 0)
+        {
+            lock (_stateGate)
+            {
+                if (_attrStates.ContainsKey(pointedHandle))
+                {
+                    return pointedHandle;
+                }
+            }
+        }
+
+        return attrAddress;
+    }
+
     private static bool TryWriteFixedUtf8CString(CpuContext ctx, ulong address, string value, int maxBytes)
     {
         if (maxBytes <= 0)

src/SharpEmu.Libs/Kernel/KernelPthreadState.cs

@@ -3,6 +3,7 @@
 
 using System.Runtime.InteropServices;
 using System.Threading;
+using SharpEmu.HLE;
 
 namespace SharpEmu.Libs.Kernel;
 
@@ -25,12 +26,24 @@ internal static class KernelPthreadState
 
     internal static ulong GetCurrentThreadHandle()
     {
+        var guestThreadHandle = GuestThreadExecution.CurrentGuestThreadHandle;
+        if (guestThreadHandle != 0 && TryGetThreadIdentity(guestThreadHandle, out _))
+        {
+            return guestThreadHandle;
+        }
+
         EnsureCurrentThreadRegistered();
         return _currentThreadHandle;
     }
 
     internal static ulong GetCurrentThreadUniqueId()
     {
+        var guestThreadHandle = GuestThreadExecution.CurrentGuestThreadHandle;
+        if (guestThreadHandle != 0 && TryGetThreadIdentity(guestThreadHandle, out var identity))
+        {
+            return identity.UniqueId;
+        }
+
         EnsureCurrentThreadRegistered();
         return _currentThreadUniqueId;
     }

src/SharpEmu.Libs/Kernel/KernelRuntimeCompatExports.cs

@@ -258,6 +258,8 @@ public static class KernelRuntimeCompatExports
             }
         }
 
+        TraceProcParam(ctx, address);
+
         ctx[CpuRegister.Rax] = address;
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
@@ -270,6 +272,223 @@ public static class KernelRuntimeCompatExports
         }
     }
 
+    private static void TraceProcParam(CpuContext ctx, ulong address)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PROC_PARAM"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        if (address == 0)
+        {
+            Console.Error.WriteLine("[LOADER][TRACE] proc_param: address=0");
+            return;
+        }
+
+        const int dumpSize = 0x200;
+        var buffer = GC.AllocateUninitializedArray<byte>(dumpSize);
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param: address=0x{address:X16} unreadable");
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] proc_param: address=0x{address:X16} size=0x{dumpSize:X}");
+        for (var offset = 0; offset < dumpSize; offset += 16)
+        {
+            var slice = buffer.AsSpan(offset, 16);
+            var hex = Convert.ToHexString(slice);
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param[{offset:X3}]: {hex}");
+        }
+
+        TraceProcParamPointers(ctx, address, buffer);
+    }
+
+    private static void TraceProcParamPointers(CpuContext ctx, ulong baseAddress, ReadOnlySpan<byte> buffer)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_PROC_PARAM_PTRS"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        if (buffer.Length < 0x50)
+        {
+            return;
+        }
+
+        for (var offset = 0x20; offset <= 0x48; offset += 8)
+        {
+            var ptr = BinaryPrimitives.ReadUInt64LittleEndian(buffer.Slice(offset, 8));
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr@{offset:X2}: 0x{ptr:X16}");
+            if (ptr == 0)
+            {
+                continue;
+            }
+
+            TraceProcParamPointerTarget(ctx, ptr);
+        }
+    }
+
+    private static void TraceProcParamPointerTarget(CpuContext ctx, ulong address)
+    {
+        const int maxAsciiBytes = 256;
+        const int maxWideChars = 128;
+
+        if (TryReadUtf8CString(ctx, address, maxAsciiBytes, out var asciiValue))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.target ascii@0x{address:X16}: \"{asciiValue}\"");
+            return;
+        }
+
+        if (TryReadUtf16CString(ctx, address, maxWideChars, out var wideValue))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.target wide@0x{address:X16}: \"{wideValue}\"");
+            return;
+        }
+
+        var preview = GC.AllocateUninitializedArray<byte>(64);
+        if (ctx.Memory.TryRead(address, preview))
+        {
+            var hex = Convert.ToHexString(preview);
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.target hex@0x{address:X16}: {hex}");
+            TraceProcParamEmbeddedPointers(ctx, address, preview);
+        }
+        else
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.target unreadable@0x{address:X16}");
+        }
+    }
+
+    private static bool TryReadUtf8CString(CpuContext ctx, ulong address, int maxBytes, out string value)
+    {
+        value = string.Empty;
+        var buffer = GC.AllocateUninitializedArray<byte>(maxBytes);
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            return false;
+        }
+
+        var length = Array.IndexOf(buffer, (byte)0);
+        if (length < 0)
+        {
+            length = maxBytes;
+        }
+
+        if (length == 0)
+        {
+            return false;
+        }
+
+        var text = Encoding.UTF8.GetString(buffer, 0, length);
+        if (!IsMostlyPrintable(text))
+        {
+            return false;
+        }
+
+        value = text;
+        return true;
+    }
+
+    private static bool TryReadUtf16CString(CpuContext ctx, ulong address, int maxChars, out string value)
+    {
+        value = string.Empty;
+        var maxBytes = maxChars * 2;
+        var buffer = GC.AllocateUninitializedArray<byte>(maxBytes);
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            return false;
+        }
+
+        var lengthBytes = -1;
+        for (var i = 0; i + 1 < buffer.Length; i += 2)
+        {
+            if (buffer[i] == 0 && buffer[i + 1] == 0)
+            {
+                lengthBytes = i;
+                break;
+            }
+        }
+
+        if (lengthBytes <= 0)
+        {
+            return false;
+        }
+
+        var text = Encoding.Unicode.GetString(buffer, 0, lengthBytes);
+        if (!IsMostlyPrintable(text))
+        {
+            return false;
+        }
+
+        value = text;
+        return true;
+    }
+
+    private static bool IsMostlyPrintable(string text)
+    {
+        if (string.IsNullOrWhiteSpace(text))
+        {
+            return false;
+        }
+
+        var printable = 0;
+        for (var i = 0; i < text.Length; i++)
+        {
+            var ch = text[i];
+            if (ch == '\0')
+            {
+                continue;
+            }
+
+            if (!char.IsControl(ch) || ch == '\r' || ch == '\n' || ch == '\t')
+            {
+                printable++;
+            }
+        }
+
+        return printable >= Math.Max(4, text.Length * 3 / 4);
+    }
+
+    private static void TraceProcParamEmbeddedPointers(CpuContext ctx, ulong baseAddress, ReadOnlySpan<byte> data)
+    {
+        const int maxCandidates = 12;
+        var found = 0;
+        Span<byte> probe = stackalloc byte[2];
+
+        for (var offset = 0; offset + 8 <= data.Length; offset += 8)
+        {
+            var candidate = BinaryPrimitives.ReadUInt64LittleEndian(data.Slice(offset, 8));
+            if (candidate == 0)
+            {
+                continue;
+            }
+
+            if (!ctx.Memory.TryRead(candidate, probe))
+            {
+                continue;
+            }
+
+            if (TryReadUtf8CString(ctx, candidate, 256, out var ascii))
+            {
+                Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.embed@0x{baseAddress:X16}+0x{offset:X2} -> 0x{candidate:X16} ascii \"{ascii}\"");
+                if (++found >= maxCandidates)
+                {
+                    return;
+                }
+                continue;
+            }
+
+            if (TryReadUtf16CString(ctx, candidate, 128, out var wide))
+            {
+                Console.Error.WriteLine($"[LOADER][TRACE] proc_param.ptr.embed@0x{baseAddress:X16}+0x{offset:X2} -> 0x{candidate:X16} wide \"{wide}\"");
+                if (++found >= maxCandidates)
+                {
+                    return;
+                }
+            }
+        }
+    }
+
     [SysAbiExport(
         Nid = "9BcDykPmo1I",
         ExportName = "__error",

src/SharpEmu.Libs/Kernel/KernelSemaphoreCompatExports.cs

@@ -0,0 +1,318 @@
+// 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;
+
+namespace SharpEmu.Libs.Kernel;
+
+public static class KernelSemaphoreCompatExports
+{
+    private const int MaxSemaphoreNameLength = 128;
+    private static readonly ConcurrentDictionary<uint, KernelSemaphoreState> _semaphores = new();
+    private static int _nextSemaphoreHandle = 1;
+
+    private sealed class KernelSemaphoreState
+    {
+        public required string Name { get; init; }
+        public required int InitialCount { get; init; }
+        public required int MaxCount { get; init; }
+        public int Count { get; set; }
+        public int WaitingThreads { get; set; }
+        public object Gate { get; } = new();
+    }
+
+    [SysAbiExport(
+        Nid = "188x57JYp0g",
+        ExportName = "sceKernelCreateSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelCreateSema(CpuContext ctx)
+    {
+        var semaphoreAddress = ctx[CpuRegister.Rdi];
+        var nameAddress = ctx[CpuRegister.Rsi];
+        var attr = unchecked((uint)ctx[CpuRegister.Rdx]);
+        var initialCount = unchecked((int)ctx[CpuRegister.Rcx]);
+        var maxCount = unchecked((int)ctx[CpuRegister.R8]);
+        var optionAddress = ctx[CpuRegister.R9];
+
+        if (semaphoreAddress == 0 ||
+            nameAddress == 0 ||
+            attr > 2 ||
+            initialCount < 0 ||
+            maxCount <= 0 ||
+            initialCount > maxCount ||
+            optionAddress != 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        if (!TryReadNullTerminatedUtf8(ctx, nameAddress, MaxSemaphoreNameLength, out var name))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        var handle = unchecked((uint)Interlocked.Increment(ref _nextSemaphoreHandle));
+        if (handle == 0)
+        {
+            handle = unchecked((uint)Interlocked.Increment(ref _nextSemaphoreHandle));
+        }
+
+        _semaphores[handle] = new KernelSemaphoreState
+        {
+            Name = name,
+            InitialCount = initialCount,
+            MaxCount = maxCount,
+            Count = initialCount,
+        };
+
+        if (!TryWriteUInt32(ctx, semaphoreAddress, handle))
+        {
+            _semaphores.TryRemove(handle, out _);
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+        }
+
+        TraceSemaphore($"create handle=0x{handle:X8} name='{name}' attr=0x{attr:X} init={initialCount} max={maxCount}");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    [SysAbiExport(
+        Nid = "Zxa0VhQVTsk",
+        ExportName = "sceKernelWaitSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelWaitSema(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var needCount = unchecked((int)ctx[CpuRegister.Rsi]);
+        var timeoutAddress = ctx[CpuRegister.Rdx];
+
+        if (!_semaphores.TryGetValue(handle, out var semaphore))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (needCount < 1 || needCount > semaphore.MaxCount)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        lock (semaphore.Gate)
+        {
+            if (semaphore.Count >= needCount)
+            {
+                semaphore.Count -= needCount;
+                TraceSemaphore($"wait handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+            }
+
+            if (timeoutAddress != 0)
+            {
+                if (!TryReadUInt32(ctx, timeoutAddress, out _))
+                {
+                    return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+                }
+
+                _ = TryWriteUInt32(ctx, timeoutAddress, 0);
+                TraceSemaphore($"wait-timeout handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TIMED_OUT);
+            }
+
+            if (!GuestThreadExecution.RequestCurrentThreadBlock("sceKernelWaitSema"))
+            {
+                TraceSemaphore($"wait-would-block handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_TRY_AGAIN);
+            }
+
+            semaphore.WaitingThreads++;
+            TraceSemaphore($"wait-block handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count} waiters={semaphore.WaitingThreads}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "12wOHk8ywb0",
+        ExportName = "sceKernelPollSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelPollSema(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var needCount = unchecked((int)ctx[CpuRegister.Rsi]);
+
+        if (!_semaphores.TryGetValue(handle, out var semaphore))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (needCount < 1 || needCount > semaphore.MaxCount)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        lock (semaphore.Gate)
+        {
+            if (semaphore.Count < needCount)
+            {
+                TraceSemaphore($"poll-busy handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_BUSY);
+            }
+
+            semaphore.Count -= needCount;
+            TraceSemaphore($"poll handle=0x{handle:X8} name='{semaphore.Name}' need={needCount} count={semaphore.Count}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "4czppHBiriw",
+        ExportName = "sceKernelSignalSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelSignalSema(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var signalCount = unchecked((int)ctx[CpuRegister.Rsi]);
+
+        if (!_semaphores.TryGetValue(handle, out var semaphore))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (signalCount <= 0)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        lock (semaphore.Gate)
+        {
+            if (semaphore.Count > semaphore.MaxCount - signalCount)
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+            }
+
+            semaphore.Count += signalCount;
+            TraceSemaphore($"signal handle=0x{handle:X8} name='{semaphore.Name}' signal={signalCount} count={semaphore.Count} waiters={semaphore.WaitingThreads}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "4DM06U2BNEY",
+        ExportName = "sceKernelCancelSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelCancelSema(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        var setCount = unchecked((int)ctx[CpuRegister.Rsi]);
+        var waitingThreadsAddress = ctx[CpuRegister.Rdx];
+
+        if (!_semaphores.TryGetValue(handle, out var semaphore))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        if (setCount > semaphore.MaxCount)
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        lock (semaphore.Gate)
+        {
+            if (waitingThreadsAddress != 0 && !TryWriteUInt32(ctx, waitingThreadsAddress, unchecked((uint)semaphore.WaitingThreads)))
+            {
+                return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+            }
+
+            semaphore.Count = setCount < 0 ? semaphore.InitialCount : setCount;
+            semaphore.WaitingThreads = 0;
+            TraceSemaphore($"cancel handle=0x{handle:X8} name='{semaphore.Name}' set={setCount} count={semaphore.Count}");
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+        }
+    }
+
+    [SysAbiExport(
+        Nid = "R1Jvn8bSCW8",
+        ExportName = "sceKernelDeleteSema",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libKernel")]
+    public static int KernelDeleteSema(CpuContext ctx)
+    {
+        var handle = unchecked((uint)ctx[CpuRegister.Rdi]);
+        if (!_semaphores.TryRemove(handle, out var semaphore))
+        {
+            return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_ERROR_NOT_FOUND);
+        }
+
+        TraceSemaphore($"delete handle=0x{handle:X8} name='{semaphore.Name}'");
+        return SetReturn(ctx, OrbisGen2Result.ORBIS_GEN2_OK);
+    }
+
+    private static int SetReturn(CpuContext ctx, OrbisGen2Result result)
+    {
+        var value = (int)result;
+        ctx[CpuRegister.Rax] = unchecked((ulong)value);
+        return value;
+    }
+
+    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 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 maxLength, out string value)
+    {
+        value = string.Empty;
+        if (address == 0 || maxLength <= 0)
+        {
+            return false;
+        }
+
+        var bytes = new byte[Math.Min(maxLength, 4096)];
+        for (var i = 0; i < bytes.Length; i++)
+        {
+            Span<byte> current = stackalloc byte[1];
+            if (!ctx.Memory.TryRead(address + (ulong)i, current))
+            {
+                return false;
+            }
+
+            if (current[0] == 0)
+            {
+                value = Encoding.UTF8.GetString(bytes, 0, i);
+                return true;
+            }
+
+            bytes[i] = current[0];
+        }
+
+        value = Encoding.UTF8.GetString(bytes);
+        return true;
+    }
+
+    private static void TraceSemaphore(string message)
+    {
+        if (string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_SEMA"), "1", StringComparison.Ordinal))
+        {
+            Console.Error.WriteLine($"[LOADER][TRACE] sema.{message}");
+        }
+    }
+}

src/SharpEmu.Libs/Pad/PadExports.cs

@@ -0,0 +1,158 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+using System.Diagnostics;
+
+namespace SharpEmu.Libs.Pad;
+
+public static class PadExports
+{
+    private const int OrbisPadErrorInvalidHandle = unchecked((int)0x80920003);
+    private const int OrbisPadErrorNotInitialized = unchecked((int)0x80920005);
+    private const int OrbisPadErrorDeviceNotConnected = unchecked((int)0x80920007);
+    private const int OrbisPadErrorDeviceNoHandle = unchecked((int)0x80920008);
+    private const int PrimaryUserId = 1;
+    private const int StandardPortType = 0;
+    private const int PrimaryPadHandle = 1;
+    private const int ControllerInformationSize = 0x1C;
+    private const int PadDataSize = 0x78;
+
+    private static bool _initialized;
+
+    [SysAbiExport(
+        Nid = "hv1luiJrqQM",
+        ExportName = "scePadInit",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadInit(CpuContext ctx)
+    {
+        _initialized = true;
+        return SetReturn(ctx, 0);
+    }
+
+    [SysAbiExport(
+        Nid = "xk0AcarP3V4",
+        ExportName = "scePadOpen",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadOpen(CpuContext ctx)
+    {
+        var userId = unchecked((int)ctx[CpuRegister.Rdi]);
+        var type = unchecked((int)ctx[CpuRegister.Rsi]);
+        var index = unchecked((int)ctx[CpuRegister.Rdx]);
+        var parameterAddress = ctx[CpuRegister.Rcx];
+        if (!_initialized)
+        {
+            return SetReturn(ctx, OrbisPadErrorNotInitialized);
+        }
+
+        if (userId == -1)
+        {
+            return SetReturn(ctx, OrbisPadErrorDeviceNoHandle);
+        }
+
+        if (userId != PrimaryUserId || type != StandardPortType || index != 0 || parameterAddress != 0)
+        {
+            return SetReturn(ctx, OrbisPadErrorDeviceNotConnected);
+        }
+
+        return SetReturn(ctx, PrimaryPadHandle);
+    }
+
+    [SysAbiExport(
+        Nid = "clVvL4ZDntw",
+        ExportName = "scePadSetMotionSensorState",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadSetMotionSensorState(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        return handle == PrimaryPadHandle
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, OrbisPadErrorInvalidHandle);
+    }
+
+    [SysAbiExport(
+        Nid = "gjP9-KQzoUk",
+        ExportName = "scePadGetControllerInformation",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadGetControllerInformation(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        var informationAddress = ctx[CpuRegister.Rsi];
+        if (handle != PrimaryPadHandle)
+        {
+            return SetReturn(ctx, OrbisPadErrorInvalidHandle);
+        }
+
+        if (informationAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> information = stackalloc byte[ControllerInformationSize];
+        BinaryPrimitives.WriteSingleLittleEndian(information[0x00..], 44.86f);
+        BinaryPrimitives.WriteUInt16LittleEndian(information[0x04..], 1920);
+        BinaryPrimitives.WriteUInt16LittleEndian(information[0x06..], 943);
+        information[0x08] = 30;
+        information[0x09] = 30;
+        information[0x0A] = StandardPortType;
+        information[0x0B] = 1;
+        information[0x0C] = 1;
+        BinaryPrimitives.WriteInt32LittleEndian(information[0x10..], 0);
+
+        return ctx.Memory.TryWrite(informationAddress, information)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "YndgXqQVV7c",
+        ExportName = "scePadReadState",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libScePad")]
+    public static int PadReadState(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        var dataAddress = ctx[CpuRegister.Rsi];
+        if (handle != PrimaryPadHandle)
+        {
+            return SetReturn(ctx, OrbisPadErrorInvalidHandle);
+        }
+
+        if (dataAddress == 0)
+        {
+            return SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_INVALID_ARGUMENT);
+        }
+
+        Span<byte> data = stackalloc byte[PadDataSize];
+        data.Clear();
+        data[0x04] = 128;
+        data[0x05] = 128;
+        data[0x06] = 128;
+        data[0x07] = 128;
+        BinaryPrimitives.WriteSingleLittleEndian(data[0x18..], 1.0f);
+        data[0x4C] = 1;
+        var timestampTicks = Stopwatch.GetTimestamp();
+        var timestampMicroseconds =
+            ((ulong)(timestampTicks / Stopwatch.Frequency) * 1_000_000UL) +
+            ((ulong)(timestampTicks % Stopwatch.Frequency) * 1_000_000UL / (ulong)Stopwatch.Frequency);
+        BinaryPrimitives.WriteUInt64LittleEndian(
+            data[0x50..],
+            timestampMicroseconds);
+        data[0x68] = 1;
+
+        return ctx.Memory.TryWrite(dataAddress, data)
+            ? 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;
+    }
+}

src/SharpEmu.Libs/SystemService/SystemServiceExports.cs

@@ -0,0 +1,42 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+
+namespace SharpEmu.Libs.SystemService;
+
+public static class SystemServiceExports
+{
+    private const int OrbisSystemServiceErrorParameter = unchecked((int)0x80A10003);
+    private const int SystemServiceStatusSize = 0x0C;
+
+    [SysAbiExport(
+        Nid = "rPo6tV8D9bM",
+        ExportName = "sceSystemServiceGetStatus",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceSystemService")]
+    public static int SystemServiceGetStatus(CpuContext ctx)
+    {
+        var statusAddress = ctx[CpuRegister.Rdi];
+        if (statusAddress == 0)
+        {
+            return SetReturn(ctx, OrbisSystemServiceErrorParameter);
+        }
+
+        Span<byte> status = stackalloc byte[SystemServiceStatusSize];
+        status.Clear();
+        BinaryPrimitives.WriteInt32LittleEndian(status, 0);
+        status[0x06] = 1;
+
+        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;
+    }
+}

src/SharpEmu.Libs/UserService/UserServiceExports.cs

@@ -0,0 +1,79 @@
+// Copyright (C) 2026 SharpEmu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+using SharpEmu.HLE;
+using System.Buffers.Binary;
+
+namespace SharpEmu.Libs.UserService;
+
+public static class UserServiceExports
+{
+    private const int OrbisUserServiceErrorInvalidArgument = unchecked((int)0x80960005);
+    private const int PrimaryUserId = 1;
+    private const int InvalidUserId = -1;
+
+    [SysAbiExport(
+        Nid = "j3YMu1MVNNo",
+        ExportName = "sceUserServiceInitialize",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceInitialize(CpuContext ctx)
+    {
+        ctx[CpuRegister.Rax] = 0;
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "CdWp0oHWGr0",
+        ExportName = "sceUserServiceGetInitialUser",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetInitialUser(CpuContext ctx)
+    {
+        var userIdAddress = ctx[CpuRegister.Rdi];
+        if (userIdAddress == 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidArgument);
+        }
+
+        return TryWriteInt32(ctx, userIdAddress, PrimaryUserId)
+            ? SetReturn(ctx, 0)
+            : SetReturn(ctx, (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT);
+    }
+
+    [SysAbiExport(
+        Nid = "fPhymKNvK-A",
+        ExportName = "sceUserServiceGetLoginUserIdList",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceUserService")]
+    public static int UserServiceGetLoginUserIdList(CpuContext ctx)
+    {
+        var userIdListAddress = ctx[CpuRegister.Rdi];
+        if (userIdListAddress == 0)
+        {
+            return SetReturn(ctx, OrbisUserServiceErrorInvalidArgument);
+        }
+
+        Span<byte> userIds = stackalloc byte[sizeof(int) * 4];
+        BinaryPrimitives.WriteInt32LittleEndian(userIds[0x00..], PrimaryUserId);
+        BinaryPrimitives.WriteInt32LittleEndian(userIds[0x04..], InvalidUserId);
+        BinaryPrimitives.WriteInt32LittleEndian(userIds[0x08..], InvalidUserId);
+        BinaryPrimitives.WriteInt32LittleEndian(userIds[0x0C..], InvalidUserId);
+        return ctx.Memory.TryWrite(userIdListAddress, userIds)
+            ? 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)];
+        BinaryPrimitives.WriteInt32LittleEndian(bytes, value);
+        return ctx.Memory.TryWrite(address, bytes);
+    }
+
+    private static int SetReturn(CpuContext ctx, int result)
+    {
+        ctx[CpuRegister.Rax] = unchecked((ulong)result);
+        return result;
+    }
+}

src/SharpEmu.Libs/VideoOut/VideoOutExports.cs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 using SharpEmu.HLE;
+using SharpEmu.Libs.Kernel;
 using System.Buffers.Binary;
 using System.Diagnostics.CodeAnalysis;
 using System.Threading;
@@ -15,28 +16,79 @@ public static class VideoOutExports
     private const int OrbisVideoOutErrorResourceBusy = unchecked((int)0x80290009);
     private const int OrbisVideoOutErrorInvalidIndex = unchecked((int)0x8029000A);
     private const int OrbisVideoOutErrorInvalidHandle = unchecked((int)0x8029000B);
+    private const int OrbisVideoOutErrorInvalidEventQueue = unchecked((int)0x8029000C);
+    private const int OrbisVideoOutErrorInvalidEvent = unchecked((int)0x8029000D);
     private const int OrbisVideoOutErrorInvalidOption = unchecked((int)0x8029001A);
     private const int SceVideoOutBusTypeMain = 0;
     private const int SceVideoOutBufferAttributeOptionNone = 0;
+    private const int SceVideoOutTilingModeLinear = 1;
     private const int MaxOpenPorts = 4;
     private const int MaxDisplayBuffers = 16;
-    private const int VideoOutBufferAttributeSize = 0x24;
+    private const int MaxDisplayBufferGroups = 4;
+    private const int MaxFrameDumps = 8;
+    private const int VideoOutBufferAttributeSize = 0x28;
     private const int VideoOutBufferAttribute2Size = 0x50;
     private const int VideoOutBuffersEntrySize = 0x20;
+    private const ulong SceVideoOutPixelFormatA8R8G8B8Srgb = 0x80000000;
+    private const ulong SceVideoOutPixelFormatA8B8G8R8Srgb = 0x80002200;
+    private const ulong SceVideoOutPixelFormatA2R10G10B10 = 0x88060000;
+    private const ulong SceVideoOutPixelFormatA2R10G10B10Srgb = 0x88000000;
+    private const ulong SceVideoOutPixelFormatA2R10G10B10Bt2020Pq = 0x88740000;
+    private const ulong SceVideoOutInternalEventFlip = 0x6;
+    private const short OrbisKernelEventFilterVideoOut = -13;
 
     private static readonly object _stateGate = new();
+    private static readonly object _frameDumpGate = new();
     private static readonly Dictionary<int, VideoOutPortState> _ports = new();
+    private static readonly Dictionary<(int Handle, int BufferIndex, ulong Address), ulong> _lastFrameFingerprints = new();
     private static int _nextHandle = 1;
+    private static int _frameDumpCount;
+    private static long _nextFrameDumpIndex;
 
     private sealed class VideoOutPortState
     {
         public required int Handle { get; init; }
         public int FlipRate { get; set; }
         public ulong VblankCount { get; set; }
-        public int NextSetId { get; set; } = 1;
-        public HashSet<int> RegisteredSetIds { get; } = new();
+        public ulong FlipCount { get; set; }
+        public int CurrentBuffer { get; set; } = -1;
+        public VideoOutBufferGroup?[] Groups { get; } = new VideoOutBufferGroup?[MaxDisplayBufferGroups];
+        public VideoOutBufferSlot[] BufferSlots { get; } = CreateBufferSlots();
+        public List<FlipEventRegistration> FlipEvents { get; } = new();
     }
 
+    private sealed class VideoOutBufferGroup
+    {
+        public required int Index { get; init; }
+        public required BufferAttribute Attribute { get; init; }
+    }
+
+    private sealed class VideoOutBufferSlot
+    {
+        public int GroupIndex { get; set; } = -1;
+        public ulong AddressLeft { get; set; }
+        public ulong AddressRight { get; set; }
+    }
+
+    private readonly record struct FlipEventRegistration(ulong Equeue, ulong UserData);
+
+    private readonly record struct BufferAttribute(
+        ulong PixelFormat,
+        uint TilingMode,
+        uint AspectRatio,
+        uint Width,
+        uint Height,
+        uint PitchInPixel,
+        ulong Option);
+
+    internal readonly record struct DisplayBufferInfo(
+        ulong Address,
+        ulong PixelFormat,
+        uint TilingMode,
+        uint Width,
+        uint Height,
+        uint PitchInPixel);
+
     [SysAbiExport(
         Nid = "Up36PTk687E",
         ExportName = "sceVideoOutOpen",
@@ -136,6 +188,146 @@ public static class VideoOutExports
         return (int)OrbisGen2Result.ORBIS_GEN2_OK;
     }
 
+    [SysAbiExport(
+        Nid = "HXzjK9yI30k",
+        ExportName = "sceVideoOutAddFlipEvent",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutAddFlipEvent(CpuContext ctx)
+    {
+        var equeue = ctx[CpuRegister.Rdi];
+        var handle = unchecked((int)ctx[CpuRegister.Rsi]);
+        var userData = ctx[CpuRegister.Rdx];
+        if (!TryGetPort(handle, out var port))
+        {
+            return OrbisVideoOutErrorInvalidHandle;
+        }
+
+        if (!KernelEventQueueCompatExports.IsValidEqueue(equeue))
+        {
+            return OrbisVideoOutErrorInvalidEventQueue;
+        }
+
+        lock (_stateGate)
+        {
+            port.FlipEvents.Add(new FlipEventRegistration(equeue, userData));
+        }
+
+        TraceVideoOut($"videoout.add_flip_event eq=0x{equeue:X16} handle={handle} udata=0x{userData:X16}");
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    [SysAbiExport(
+        Nid = "U46NwOiJpys",
+        ExportName = "sceVideoOutSubmitFlip",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutSubmitFlip(CpuContext ctx)
+    {
+        var handle = unchecked((int)ctx[CpuRegister.Rdi]);
+        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);
+    }
+
+    [SysAbiExport(
+        Nid = "U2JJtSqNKZI",
+        ExportName = "sceVideoOutGetEventId",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutGetEventId(CpuContext ctx)
+    {
+        var eventAddress = ctx[CpuRegister.Rdi];
+        if (eventAddress == 0)
+        {
+            return OrbisVideoOutErrorInvalidAddress;
+        }
+
+        if (!ctx.TryReadUInt64(eventAddress, out var ident) ||
+            !TryReadInt16(ctx, eventAddress + 0x08, out var filter))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (filter != OrbisKernelEventFilterVideoOut || ident != SceVideoOutInternalEventFlip)
+        {
+            return OrbisVideoOutErrorInvalidEvent;
+        }
+
+        return 0;
+    }
+
+    [SysAbiExport(
+        Nid = "rWUTcKdkUzQ",
+        ExportName = "sceVideoOutGetEventData",
+        Target = Generation.Gen4 | Generation.Gen5,
+        LibraryName = "libSceVideoOut")]
+    public static int VideoOutGetEventData(CpuContext ctx)
+    {
+        var eventAddress = ctx[CpuRegister.Rdi];
+        var dataAddress = ctx[CpuRegister.Rsi];
+        if (eventAddress == 0 || dataAddress == 0)
+        {
+            return OrbisVideoOutErrorInvalidAddress;
+        }
+
+        if (!ctx.TryReadUInt64(eventAddress, out var ident) ||
+            !TryReadInt16(ctx, eventAddress + 0x08, out var filter) ||
+            !ctx.TryReadUInt64(eventAddress + 0x10, out var data))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        if (filter != OrbisKernelEventFilterVideoOut || ident != SceVideoOutInternalEventFlip)
+        {
+            return OrbisVideoOutErrorInvalidEvent;
+        }
+
+        return ctx.TryWriteUInt64(dataAddress, data >> 16)
+            ? (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);
+
+    internal static bool TryGetDisplayBufferInfo(int handle, int bufferIndex, out DisplayBufferInfo info)
+    {
+        info = default;
+        if (bufferIndex < 0 || bufferIndex >= MaxDisplayBuffers)
+        {
+            return false;
+        }
+
+        lock (_stateGate)
+        {
+            if (!_ports.TryGetValue(handle, out var port))
+            {
+                return false;
+            }
+
+            var slot = port.BufferSlots[bufferIndex];
+            if (slot.AddressLeft == 0 ||
+                slot.GroupIndex < 0 ||
+                slot.GroupIndex >= port.Groups.Length ||
+                port.Groups[slot.GroupIndex] is not { } group)
+            {
+                return false;
+            }
+
+            var attribute = group.Attribute;
+            info = new DisplayBufferInfo(
+                slot.AddressLeft,
+                attribute.PixelFormat,
+                attribute.TilingMode,
+                attribute.Width,
+                attribute.Height,
+                attribute.PitchInPixel);
+            return true;
+        }
+    }
+
     [SysAbiExport(
         Nid = "MTxxrOCeSig",
         ExportName = "sceVideoOutSetWindowModeMargins",
@@ -173,9 +365,23 @@ public static class VideoOutExports
 
         lock (_stateGate)
         {
-            return port.RegisteredSetIds.Remove(attributeIndex)
-                ? (int)OrbisGen2Result.ORBIS_GEN2_OK
-                : OrbisVideoOutErrorInvalidValue;
+            if (attributeIndex >= port.Groups.Length || port.Groups[attributeIndex] is null)
+            {
+                return OrbisVideoOutErrorInvalidValue;
+            }
+
+            port.Groups[attributeIndex] = null;
+            foreach (var slot in port.BufferSlots)
+            {
+                if (slot.GroupIndex == attributeIndex)
+                {
+                    slot.GroupIndex = -1;
+                    slot.AddressLeft = 0;
+                    slot.AddressRight = 0;
+                }
+            }
+
+            return (int)OrbisGen2Result.ORBIS_GEN2_OK;
         }
     }
 
@@ -292,16 +498,21 @@ public static class VideoOutExports
             return OrbisVideoOutErrorInvalidValue;
         }
 
+        if (!TryReadBufferAttribute(ctx, attributeAddress, false, out var attribute))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        Span<ulong> addresses = stackalloc ulong[Math.Min(bufferNum, MaxDisplayBuffers)];
         for (var i = 0; i < bufferNum; i++)
         {
-            if (!ctx.TryReadUInt64(addressesAddress + ((ulong)i * 8), out _))
+            if (!ctx.TryReadUInt64(addressesAddress + ((ulong)i * 8), out addresses[i]))
             {
                 return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
             }
         }
 
-        var setId = AllocateBufferSet(port);
-        return setId;
+        return RegisterBufferRange(port, startIndex, addresses[..bufferNum], attribute);
     }
 
     [SysAbiExport(
@@ -348,34 +559,458 @@ public static class VideoOutExports
             return OrbisVideoOutErrorInvalidValue;
         }
 
+        if (!TryReadBufferAttribute(ctx, attributeAddress, true, out var attribute))
+        {
+            return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
+        }
+
+        Span<ulong> addresses = stackalloc ulong[Math.Min(bufferNum, MaxDisplayBuffers)];
         for (var i = 0; i < bufferNum; i++)
         {
             var entryAddress = buffersAddress + ((ulong)i * VideoOutBuffersEntrySize);
-            if (!ctx.TryReadUInt64(entryAddress + 0x00, out _) ||
+            if (!ctx.TryReadUInt64(entryAddress + 0x00, out addresses[i]) ||
                 !ctx.TryReadUInt64(entryAddress + 0x08, out _))
             {
                 return (int)OrbisGen2Result.ORBIS_GEN2_ERROR_MEMORY_FAULT;
             }
         }
 
-        lock (_stateGate)
-        {
-            port.RegisteredSetIds.Add(setIndex);
-        }
-
-        return setIndex;
+        var groupIndex = RegisterBufferRange(port, bufferIndexStart, addresses[..bufferNum], attribute, setIndex);
+        return groupIndex < 0 ? groupIndex : setIndex;
     }
 
-    private static int AllocateBufferSet(VideoOutPortState port)
+    private static int SubmitFlip(CpuContext ctx, int handle, int bufferIndex, int flipMode, long flipArg)
+    {
+        if (!TryGetPort(handle, out var port))
+        {
+            return OrbisVideoOutErrorInvalidHandle;
+        }
+
+        if (bufferIndex < -1 || bufferIndex >= MaxDisplayBuffers)
+        {
+            return OrbisVideoOutErrorInvalidIndex;
+        }
+
+        ulong eventData;
+        List<FlipEventRegistration> flipEvents;
+        lock (_stateGate)
+        {
+            if (bufferIndex != -1 && port.BufferSlots[bufferIndex].GroupIndex < 0)
+            {
+                return OrbisVideoOutErrorInvalidIndex;
+            }
+
+            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);
+            flipEvents = new List<FlipEventRegistration>(port.FlipEvents);
+        }
+
+        _ = TryDumpFrame(ctx, port, bufferIndex, flipMode, flipArg);
+
+        foreach (var flipEvent in flipEvents)
+        {
+            _ = KernelEventQueueCompatExports.EnqueueEvent(
+                flipEvent.Equeue,
+                new KernelEventQueueCompatExports.KernelQueuedEvent(
+                    SceVideoOutInternalEventFlip,
+                    OrbisKernelEventFilterVideoOut,
+                    0,
+                    0,
+                    eventData,
+                    flipEvent.UserData));
+        }
+
+        TraceVideoOut($"videoout.submit_flip handle={handle} index={bufferIndex} mode={flipMode} arg={flipArg} events={flipEvents.Count}");
+        return (int)OrbisGen2Result.ORBIS_GEN2_OK;
+    }
+
+    private static int RegisterBufferRange(VideoOutPortState port, int startIndex, ReadOnlySpan<ulong> addresses, BufferAttribute attribute, int requestedGroupIndex = -1)
     {
         lock (_stateGate)
         {
-            var setId = port.NextSetId++;
-            port.RegisteredSetIds.Add(setId);
-            return setId;
+            var groupIndex = requestedGroupIndex >= 0 ? requestedGroupIndex : FindFreeGroupIndex(port);
+            if (groupIndex < 0 || groupIndex >= MaxDisplayBufferGroups)
+            {
+                return OrbisVideoOutErrorInvalidValue;
+            }
+
+            if (port.Groups[groupIndex] is not null)
+            {
+                return OrbisVideoOutErrorResourceBusy;
+            }
+
+            for (var i = 0; i < addresses.Length; i++)
+            {
+                if (port.BufferSlots[startIndex + i].GroupIndex >= 0)
+                {
+                    return OrbisVideoOutErrorResourceBusy;
+                }
+            }
+
+            port.Groups[groupIndex] = new VideoOutBufferGroup
+            {
+                Index = groupIndex,
+                Attribute = attribute,
+            };
+
+            for (var i = 0; i < addresses.Length; i++)
+            {
+                var slot = port.BufferSlots[startIndex + i];
+                slot.GroupIndex = groupIndex;
+                slot.AddressLeft = addresses[i];
+                slot.AddressRight = 0;
+            }
+
+            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}");
+            return groupIndex;
         }
     }
 
+    private static int FindFreeGroupIndex(VideoOutPortState port)
+    {
+        for (var i = 0; i < port.Groups.Length; i++)
+        {
+            if (port.Groups[i] is null)
+            {
+                return i;
+            }
+        }
+
+        return -1;
+    }
+
+    private static bool TryReadBufferAttribute(CpuContext ctx, ulong attributeAddress, bool attribute2, out BufferAttribute attribute)
+    {
+        attribute = default;
+        if (!TryReadUInt32(ctx, attributeAddress + 0x04, out var tilingMode) ||
+            !TryReadUInt32(ctx, attributeAddress + 0x0C, out var width) ||
+            !TryReadUInt32(ctx, attributeAddress + 0x10, out var height))
+        {
+            return false;
+        }
+
+        if (attribute2)
+        {
+            if (!ctx.TryReadUInt64(attributeAddress + 0x18, out var option) ||
+                !ctx.TryReadUInt64(attributeAddress + 0x20, out var pixelFormat))
+            {
+                return false;
+            }
+
+            attribute = new BufferAttribute(NormalizePixelFormat(pixelFormat), tilingMode, 0, width, height, width, option);
+            return true;
+        }
+
+        if (!TryReadUInt32(ctx, attributeAddress + 0x00, out var pixelFormat32) ||
+            !TryReadUInt32(ctx, attributeAddress + 0x08, out var aspectRatio) ||
+            !TryReadUInt32(ctx, attributeAddress + 0x14, out var pitchInPixel) ||
+            !TryReadUInt32(ctx, attributeAddress + 0x18, out var option32))
+        {
+            return false;
+        }
+
+        attribute = new BufferAttribute(NormalizePixelFormat(pixelFormat32), tilingMode, aspectRatio, width, height, pitchInPixel, option32);
+        return true;
+    }
+
+    private static bool TryDumpFrame(CpuContext ctx, VideoOutPortState port, int bufferIndex, int flipMode, long flipArg)
+    {
+        if (bufferIndex < 0)
+        {
+            return false;
+        }
+
+        VideoOutBufferSlot slot;
+        VideoOutBufferGroup? group;
+        lock (_stateGate)
+        {
+            slot = port.BufferSlots[bufferIndex];
+            group = slot.GroupIndex >= 0 && slot.GroupIndex < port.Groups.Length
+                ? port.Groups[slot.GroupIndex]
+                : null;
+        }
+
+        if (group is null || slot.AddressLeft == 0)
+        {
+            return false;
+        }
+
+        var attribute = group.Attribute;
+        if (attribute.Width == 0 || attribute.Height == 0 || attribute.Width > 8192 || attribute.Height > 8192)
+        {
+            return false;
+        }
+
+        var bytesPerPixel = GetBytesPerPixel(attribute.PixelFormat);
+        if (bytesPerPixel == 0)
+        {
+            return DumpRawFrame(ctx, port.Handle, slot.AddressLeft, attribute, bufferIndex, flipMode, flipArg, "unsupported-format");
+        }
+
+        var pitch = attribute.PitchInPixel == 0 ? attribute.Width : attribute.PitchInPixel;
+        var rowBytes = checked((int)(pitch * bytesPerPixel));
+        var visibleRowBytes = checked((int)(attribute.Width * bytesPerPixel));
+        var frameBytes = checked((ulong)rowBytes * attribute.Height);
+        if (frameBytes > 256UL * 1024UL * 1024UL)
+        {
+            return false;
+        }
+
+        lock (_frameDumpGate)
+        {
+            if (_frameDumpCount >= MaxFrameDumps)
+            {
+                return false;
+            }
+        }
+
+        const ulong fnvOffsetBasis = 14695981039346656037UL;
+        const ulong fnvPrime = 1099511628211UL;
+        var fingerprint = fnvOffsetBasis;
+        var row = new byte[rowBytes];
+        for (uint y = 0; y < attribute.Height; y++)
+        {
+            if (!ctx.Memory.TryRead(slot.AddressLeft + ((ulong)y * (ulong)rowBytes), row))
+            {
+                return false;
+            }
+
+            foreach (var value in row.AsSpan(0, visibleRowBytes))
+            {
+                fingerprint = (fingerprint ^ value) * fnvPrime;
+            }
+        }
+
+        var fingerprintKey = (port.Handle, bufferIndex, slot.AddressLeft);
+        lock (_frameDumpGate)
+        {
+            if (_lastFrameFingerprints.TryGetValue(fingerprintKey, out var previousFingerprint) &&
+                previousFingerprint == fingerprint)
+            {
+                return false;
+            }
+
+            if (_frameDumpCount >= MaxFrameDumps)
+            {
+                return false;
+            }
+
+            _lastFrameFingerprints[fingerprintKey] = fingerprint;
+            _frameDumpCount++;
+        }
+
+        var rgb = new byte[checked((int)(attribute.Width * attribute.Height * 3))];
+        var rgbOffset = 0;
+        for (uint y = 0; y < attribute.Height; y++)
+        {
+            if (!ctx.Memory.TryRead(slot.AddressLeft + ((ulong)y * (ulong)rowBytes), row))
+            {
+                return false;
+            }
+
+            ConvertRowToRgb(row.AsSpan(0, visibleRowBytes), rgb.AsSpan(rgbOffset, (int)attribute.Width * 3), attribute.PixelFormat);
+            rgbOffset += (int)attribute.Width * 3;
+        }
+
+        var frameIndex = Interlocked.Increment(ref _nextFrameDumpIndex);
+        var basePath = GetFrameDumpBasePath(frameIndex, port.Handle, bufferIndex);
+        WriteBmp(basePath + ".bmp", attribute.Width, attribute.Height, rgb);
+        WriteFrameMetadata(basePath + ".txt", slot.AddressLeft, attribute, bufferIndex, flipMode, flipArg, "bmp-linear-read", fingerprint);
+        TraceVideoOut($"videoout.dump_frame path={basePath}.bmp addr=0x{slot.AddressLeft:X16} {attribute.Width}x{attribute.Height} fmt=0x{attribute.PixelFormat:X} fingerprint=0x{fingerprint:X16}");
+        return true;
+    }
+
+    private static bool DumpRawFrame(CpuContext ctx, int handle, ulong address, BufferAttribute attribute, int bufferIndex, int flipMode, long flipArg, string reason)
+    {
+        var bytesPerPixel = Math.Max(GetBytesPerPixel(attribute.PixelFormat), 4u);
+        var pitch = attribute.PitchInPixel == 0 ? attribute.Width : attribute.PitchInPixel;
+        var byteCount = checked((ulong)pitch * attribute.Height * bytesPerPixel);
+        if (byteCount == 0 || byteCount > 256UL * 1024UL * 1024UL)
+        {
+            return false;
+        }
+
+        var bytes = new byte[(int)byteCount];
+        if (!ctx.Memory.TryRead(address, bytes))
+        {
+            return false;
+        }
+
+        var fingerprint = ComputeFingerprint(bytes);
+        var fingerprintKey = (handle, bufferIndex, address);
+        lock (_frameDumpGate)
+        {
+            if ((_lastFrameFingerprints.TryGetValue(fingerprintKey, out var previousFingerprint) &&
+                 previousFingerprint == fingerprint) ||
+                _frameDumpCount >= MaxFrameDumps)
+            {
+                return false;
+            }
+
+            _lastFrameFingerprints[fingerprintKey] = fingerprint;
+            _frameDumpCount++;
+        }
+
+        var frameIndex = Interlocked.Increment(ref _nextFrameDumpIndex);
+        var basePath = GetFrameDumpBasePath(frameIndex, handle, bufferIndex);
+        File.WriteAllBytes(basePath + ".raw", bytes);
+        WriteFrameMetadata(basePath + ".txt", address, attribute, bufferIndex, flipMode, flipArg, reason, fingerprint);
+        TraceVideoOut($"videoout.dump_frame path={basePath}.raw addr=0x{address:X16} bytes={byteCount} reason={reason} fingerprint=0x{fingerprint:X16}");
+        return true;
+    }
+
+    private static ulong ComputeFingerprint(ReadOnlySpan<byte> bytes)
+    {
+        const ulong fnvOffsetBasis = 14695981039346656037UL;
+        const ulong fnvPrime = 1099511628211UL;
+        var fingerprint = fnvOffsetBasis;
+        foreach (var value in bytes)
+        {
+            fingerprint = (fingerprint ^ value) * fnvPrime;
+        }
+
+        return fingerprint;
+    }
+
+    private static uint GetBytesPerPixel(ulong pixelFormat) =>
+        pixelFormat is SceVideoOutPixelFormatA8R8G8B8Srgb or
+            SceVideoOutPixelFormatA8B8G8R8Srgb or
+            SceVideoOutPixelFormatA2R10G10B10 or
+            SceVideoOutPixelFormatA2R10G10B10Srgb or
+            SceVideoOutPixelFormatA2R10G10B10Bt2020Pq
+            ? 4u
+            : 0u;
+
+    private static ulong NormalizePixelFormat(ulong pixelFormat)
+    {
+        if (GetBytesPerPixel(pixelFormat) != 0)
+        {
+            return pixelFormat;
+        }
+
+        var low = (uint)(pixelFormat & 0xFFFF_FFFFUL);
+        if (GetBytesPerPixel(low) != 0)
+        {
+            return low;
+        }
+
+        var high = (uint)(pixelFormat >> 32);
+        if (GetBytesPerPixel(high) != 0)
+        {
+            return high;
+        }
+
+        var packed = high | (low >> 16);
+        return GetBytesPerPixel(packed) != 0 ? packed : pixelFormat;
+    }
+
+    private static void ConvertRowToRgb(ReadOnlySpan<byte> source, Span<byte> destination, ulong pixelFormat)
+    {
+        var dst = 0;
+        for (var src = 0; src + 3 < source.Length; src += 4)
+        {
+            if (pixelFormat == SceVideoOutPixelFormatA8B8G8R8Srgb)
+            {
+                destination[dst++] = source[src + 0];
+                destination[dst++] = source[src + 1];
+                destination[dst++] = source[src + 2];
+            }
+            else if (pixelFormat is SceVideoOutPixelFormatA2R10G10B10 or SceVideoOutPixelFormatA2R10G10B10Srgb or SceVideoOutPixelFormatA2R10G10B10Bt2020Pq)
+            {
+                var value = BinaryPrimitives.ReadUInt32LittleEndian(source[src..(src + 4)]);
+                destination[dst++] = (byte)(((value >> 20) & 0x3FF) >> 2);
+                destination[dst++] = (byte)(((value >> 10) & 0x3FF) >> 2);
+                destination[dst++] = (byte)((value & 0x3FF) >> 2);
+            }
+            else
+            {
+                destination[dst++] = source[src + 2];
+                destination[dst++] = source[src + 1];
+                destination[dst++] = source[src + 0];
+            }
+        }
+    }
+
+    private static string GetFrameDumpBasePath(long frameIndex, int handle, int bufferIndex)
+    {
+        var directory = GetLogsDirectory();
+        Directory.CreateDirectory(directory);
+        return Path.Combine(directory, $"videoout_frame_{frameIndex:D4}_h{handle}_b{bufferIndex}");
+    }
+
+    private static string GetLogsDirectory()
+    {
+        var current = new DirectoryInfo(AppContext.BaseDirectory);
+        while (current is not null)
+        {
+            if (File.Exists(Path.Combine(current.FullName, "SharpEmu.slnx")))
+            {
+                return Path.Combine(current.FullName, "logs");
+            }
+
+            current = current.Parent;
+        }
+
+        return Path.Combine(Directory.GetCurrentDirectory(), "logs");
+    }
+
+    private static void WriteBmp(string path, uint width, uint height, byte[] rgb)
+    {
+        var rowStride = checked((int)(((width * 3u) + 3u) & ~3u));
+        var pixelBytes = checked(rowStride * (int)height);
+        var fileSize = 54 + pixelBytes;
+        using var stream = File.Create(path);
+        Span<byte> header = stackalloc byte[54];
+        header[0] = (byte)'B';
+        header[1] = (byte)'M';
+        BinaryPrimitives.WriteUInt32LittleEndian(header[0x02..], (uint)fileSize);
+        BinaryPrimitives.WriteUInt32LittleEndian(header[0x0A..], 54);
+        BinaryPrimitives.WriteUInt32LittleEndian(header[0x0E..], 40);
+        BinaryPrimitives.WriteInt32LittleEndian(header[0x12..], (int)width);
+        BinaryPrimitives.WriteInt32LittleEndian(header[0x16..], -(int)height);
+        BinaryPrimitives.WriteUInt16LittleEndian(header[0x1A..], 1);
+        BinaryPrimitives.WriteUInt16LittleEndian(header[0x1C..], 24);
+        BinaryPrimitives.WriteUInt32LittleEndian(header[0x22..], (uint)pixelBytes);
+        stream.Write(header);
+
+        var row = new byte[rowStride];
+        var sourceStride = (int)width * 3;
+        var heightInt = (int)height;
+        var widthInt = (int)width;
+        for (var y = 0; y < heightInt; y++)
+        {
+            row.AsSpan().Clear();
+            var src = rgb.AsSpan(y * sourceStride, sourceStride);
+            for (var x = 0; x < widthInt; x++)
+            {
+                row[(x * 3) + 0] = src[(x * 3) + 2];
+                row[(x * 3) + 1] = src[(x * 3) + 1];
+                row[(x * 3) + 2] = src[(x * 3) + 0];
+            }
+
+            stream.Write(row);
+        }
+    }
+
+    private static void WriteFrameMetadata(
+        string path,
+        ulong address,
+        BufferAttribute attribute,
+        int bufferIndex,
+        int flipMode,
+        long flipArg,
+        string kind,
+        ulong fingerprint)
+    {
+        File.WriteAllText(
+            path,
+            $"kind={kind}\naddress=0x{address:X16}\nbuffer_index={bufferIndex}\nflip_mode={flipMode}\nflip_arg={flipArg}\nfingerprint=0x{fingerprint:X16}\npixel_format=0x{attribute.PixelFormat:X}\ntiling_mode={attribute.TilingMode}\nwidth={attribute.Width}\nheight={attribute.Height}\npitch_in_pixel={attribute.PitchInPixel}\noption=0x{attribute.Option:X}\n");
+    }
+
     private static bool IsValidBufferRange(int startIndex, int bufferNum)
     {
         return startIndex >= 0 &&
@@ -393,6 +1028,17 @@ public static class VideoOutExports
         }
     }
 
+    private static VideoOutBufferSlot[] CreateBufferSlots()
+    {
+        var slots = new VideoOutBufferSlot[MaxDisplayBuffers];
+        for (var i = 0; i < slots.Length; i++)
+        {
+            slots[i] = new VideoOutBufferSlot();
+        }
+
+        return slots;
+    }
+
     private static bool TryReadStackUInt32(CpuContext ctx, int stackIndex, out uint value)
     {
         var address = ctx[CpuRegister.Rsp] + 0x08 + ((ulong)stackIndex * 0x08);
@@ -406,4 +1052,40 @@ public static class VideoOutExports
         value = BinaryPrimitives.ReadUInt32LittleEndian(buffer);
         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 bool TryReadInt16(CpuContext ctx, ulong address, out short value)
+    {
+        Span<byte> buffer = stackalloc byte[sizeof(short)];
+        if (!ctx.Memory.TryRead(address, buffer))
+        {
+            value = 0;
+            return false;
+        }
+
+        value = BinaryPrimitives.ReadInt16LittleEndian(buffer);
+        return true;
+    }
+
+    private static void TraceVideoOut(string message)
+    {
+        if (!string.Equals(Environment.GetEnvironmentVariable("SHARPEMU_LOG_VIDEOOUT"), "1", StringComparison.Ordinal))
+        {
+            return;
+        }
+
+        Console.Error.WriteLine($"[LOADER][TRACE] {message}");
+    }
 }