Don't block the render thread while the CPU is paused. This is a prereq for imgui debuggers.

Core/Core.cpp

@@ -35,6 +35,7 @@
 #include "Core/Core.h"
 #include "Core/Config.h"
 #include "Core/MemMap.h"
+#include "Core/MIPS/MIPSDebugInterface.h"
 #include "Core/SaveState.h"
 #include "Core/System.h"
 #include "Core/MemFault.h"
@@ -51,16 +52,33 @@
 #include "Windows/InputDevice.h"
 #endif
 
-static std::condition_variable m_StepCond;
-static std::mutex m_hStepMutex;
+// Step command to execute next
+static std::mutex g_stepMutex;
+struct StepCommand {
+	CPUStepType type;
+	int param;
+	const char *reason;
+	u32 relatedAddr;
+	bool empty() const {
+		return type == CPUStepType::None;
+	}
+	void clear() {
+		type = CPUStepType::None;
+		param = 0;
+		reason = "";
+		relatedAddr = 0;
+	}
+};
+static StepCommand g_stepCommand;
+
+// This is so that external threads can wait for the CPU to become inactive.
 static std::condition_variable m_InactiveCond;
 static std::mutex m_hInactiveMutex;
-static int g_singleStepsPending = 0;
+
 static int steppingCounter = 0;
-static const char *steppingReason = "";
-static uint32_t steppingAddress = 0;
 static std::set<CoreLifecycleFunc> lifecycleFuncs;
 static std::set<CoreStopRequestFunc> stopFuncs;
+
 static bool windowHidden = false;
 static bool powerSaving = false;
 
@@ -126,14 +144,6 @@ bool Core_IsInactive() {
 	return coreState != CORE_RUNNING && coreState != CORE_NEXTFRAME && !coreStatePending;
 }
 
-static inline void Core_StateProcessed() {
-	if (coreStatePending) {
-		std::lock_guard<std::mutex> guard(m_hInactiveMutex);
-		coreStatePending = false;
-		m_InactiveCond.notify_all();
-	}
-}
-
 void Core_WaitInactive() {
 	while (Core_IsActive() && !GPUStepping::IsStepping()) {
 		std::unique_lock<std::mutex> guard(m_hInactiveMutex);
@@ -227,19 +237,23 @@ void Core_RunLoop(GraphicsContext *ctx) {
 	NativeFrame(ctx);
 }
 
-void Core_DoSingleStep() {
-	std::lock_guard<std::mutex> guard(m_hStepMutex);
-	g_singleStepsPending++;
-	m_StepCond.notify_all();
-}
-
-void Core_UpdateSingleStep() {
-	std::lock_guard<std::mutex> guard(m_hStepMutex);
-	m_StepCond.notify_all();
+bool Core_RequestSingleStep(CPUStepType type, int stepSize) {
+	std::lock_guard<std::mutex> guard(g_stepMutex);
+	if (g_stepCommand.type != CPUStepType::None) {
+		ERROR_LOG(Log::CPU, "Can't submit two steps in one frame");
+		return false;
+	}
+	g_stepCommand = { type, stepSize };
+	return true;
 }
 
 // See comment in header.
-void Core_PerformStep(DebugInterface *cpu, CPUStepType stepType, int stepSize) {
+// Handles more advanced step types (used by the debugger).
+// stepSize is to support stepping through compound instructions like fused lui+ladd (li).
+// Yes, our disassembler does support those.
+// Doesn't return the new address, as that's just mips->getPC().
+// Internal use.
+static void Core_PerformStep(MIPSDebugInterface *cpu, CPUStepType stepType, int stepSize) {
 	switch (stepType) {
 	case CPUStepType::Into:
 	{
@@ -247,8 +261,8 @@ void Core_PerformStep(DebugInterface *cpu, CPUStepType stepType, int stepSize) {
 		u32 newAddress = currentPc + stepSize;
 		// If the current PC is on a breakpoint, the user still wants the step to happen.
 		CBreakPoints::SetSkipFirst(currentPc);
-		for (int i = 0; i < (newAddress - currentPc) / 4; i++) {
-			Core_DoSingleStep();
+		for (int i = 0; i < (int)(newAddress - currentPc) / 4; i++) {
+			currentMIPS->SingleStep();
 		}
 		return;
 	}
@@ -316,22 +330,8 @@ void Core_PerformStep(DebugInterface *cpu, CPUStepType stepType, int stepSize) {
 	}
 }
 
-static inline int Core_WaitStepping() {
-	std::unique_lock<std::mutex> guard(m_hStepMutex);
-	// We only wait 16ms so that we can still draw UI or react to events.
-	double sleepStart = time_now_d();
-	if (!g_singleStepsPending && coreState == CORE_STEPPING)
-		m_StepCond.wait_for(guard, std::chrono::milliseconds(16));
-	double sleepEnd = time_now_d();
-	DisplayNotifySleep(sleepEnd - sleepStart);
-
-	int result = g_singleStepsPending;
-	g_singleStepsPending = 0;
-	return result;
-}
-
-void Core_ProcessStepping() {
-	Core_StateProcessed();
+void Core_ProcessStepping(MIPSDebugInterface *cpu) {
+	coreStatePending = false;
 
 	// Check if there's any pending save state actions.
 	SaveState::Process();
@@ -352,21 +352,26 @@ void Core_ProcessStepping() {
 	}
 
 	// Need to check inside the lock to avoid races.
-	int doSteps = Core_WaitStepping();
+	std::lock_guard<std::mutex> guard(g_stepMutex);
 
-	// We may still be stepping without singleStepPending to process a save state.
-	if (doSteps && coreState == CORE_STEPPING) {
-		Core_ResetException();
-
-		for (int i = 0; i < doSteps; i++) {
-			currentMIPS->SingleStep();
-			steppingCounter++;
-		}
-
-		// Update disasm dialog.
-		System_Notify(SystemNotification::DISASSEMBLY_AFTERSTEP);
-		System_Notify(SystemNotification::MEM_VIEW);
+	if (coreState != CORE_STEPPING || g_stepCommand.empty()) {
+		return;
 	}
+
+	Core_ResetException();
+
+	if (!g_stepCommand.empty()) {
+		Core_PerformStep(cpu, g_stepCommand.type, g_stepCommand.param);
+		if (g_stepCommand.type == CPUStepType::Into) {
+			// We're already done. The other step types will resume the CPU.
+			System_Notify(SystemNotification::DISASSEMBLY_AFTERSTEP);
+		}
+		g_stepCommand.clear();
+		steppingCounter++;
+	}
+
+	// Update disasm dialog.
+	System_Notify(SystemNotification::MEM_VIEW);
 }
 
 // Many platforms, like Android, do not call this function but handle things on their own.
@@ -375,7 +380,6 @@ bool Core_Run(GraphicsContext *ctx) {
 	System_Notify(SystemNotification::DISASSEMBLY);
 	while (true) {
 		if (GetUIState() != UISTATE_INGAME) {
-			Core_StateProcessed();
 			if (GetUIState() == UISTATE_EXIT) {
 				// Not sure why we do a final frame here?
 				NativeFrame(ctx);
@@ -388,11 +392,9 @@ bool Core_Run(GraphicsContext *ctx) {
 		switch (coreState) {
 		case CORE_RUNNING:
 		case CORE_STEPPING:
-			Core_StateProcessed();
 			// enter a fast runloop
 			Core_RunLoop(ctx);
 			if (coreState == CORE_POWERDOWN) {
-				Core_StateProcessed();
 				return true;
 			}
 			break;
@@ -402,7 +404,6 @@ bool Core_Run(GraphicsContext *ctx) {
 		case CORE_BOOT_ERROR:
 		case CORE_RUNTIME_ERROR:
 			// Exit loop!!
-			Core_StateProcessed();
 			return true;
 
 		case CORE_NEXTFRAME:
@@ -411,27 +412,30 @@ bool Core_Run(GraphicsContext *ctx) {
 	}
 }
 
+// Free-threaded (hm, possibly except tracing).
 void Core_Break(const char *reason, u32 relatedAddress) {
 	// Stop the tracer
 	mipsTracer.stop_tracing();
 
-	Core_UpdateState(CORE_STEPPING);
-	steppingCounter++;
-	_assert_msg_(reason != nullptr, "No reason specified for break");
-	steppingReason = reason;
-	steppingAddress = relatedAddress;
+	{
+		std::lock_guard<std::mutex> lock(g_stepMutex);
+		steppingCounter++;
+		_assert_msg_(reason != nullptr, "No reason specified for break");
+
+		Core_UpdateState(CORE_STEPPING);
+	}
 	System_Notify(SystemNotification::DEBUG_MODE_CHANGE);
 }
 
+// Free-threaded (or at least should be)
 void Core_Resume() {
 	// Clear the exception if we resume.
 	Core_ResetException();
 	coreState = CORE_RUNNING;
-	coreStatePending = false;
-	m_StepCond.notify_all();
 	System_Notify(SystemNotification::DEBUG_MODE_CHANGE);
 }
 
+// Should be called from the EmuThread.
 bool Core_NextFrame() {
 	if (coreState == CORE_RUNNING) {
 		coreState = CORE_NEXTFRAME;
@@ -447,8 +451,11 @@ int Core_GetSteppingCounter() {
 
 SteppingReason Core_GetSteppingReason() {
 	SteppingReason r;
-	r.reason = steppingReason;
-	r.relatedAddress = steppingAddress;
+	std::lock_guard<std::mutex> lock(g_stepMutex);
+	if (!g_stepCommand.empty()) {
+		r.reason = g_stepCommand.reason;
+		r.relatedAddress = g_stepCommand.relatedAddr;
+	}
 	return r;
 }
 

Core/Core.h

@@ -24,7 +24,6 @@
 #include "Core/CoreParameter.h"
 
 class GraphicsContext;
-class DebugInterface;
 
 // called from emu thread
 void UpdateRunLoop(GraphicsContext *ctx);
@@ -53,16 +52,9 @@ void Core_Break(const char *reason, u32 relatedAddress = 0);
 // void Core_Step(CPUStepType type);  // CPUStepType::None not allowed
 void Core_Resume();
 
-// Handles more advanced step types (used by the debugger).
-// stepSize is to support stepping through compound instructions like fused lui+ladd (li).
-// Yes, our disassembler does support those.
-// Doesn't return the new address, as that's just mips->getPC().
-void Core_PerformStep(DebugInterface *cpu, CPUStepType stepType, int stepSize);
-
-// Refactor.
-void Core_DoSingleStep();
-void Core_UpdateSingleStep();
-void Core_ProcessStepping();
+// This should be called externally.
+// Can fail if another step type was requested this frame.
+bool Core_RequestSingleStep(CPUStepType stepType, int stepSize);
 
 bool Core_ShouldRunBehind();
 bool Core_MustRunBehind();

Core/Debugger/WebSocket/CPUCoreSubscriber.cpp

@@ -90,7 +90,7 @@ void WebSocketCPUResume(DebuggerRequest &req) {
 
 	CBreakPoints::SetSkipFirst(currentMIPS->pc);
 	if (currentMIPS->inDelaySlot) {
-		Core_DoSingleStep();
+		Core_RequestSingleStep(CPUStepType::Into, 1);
 	}
 	Core_Resume();
 }

Core/Debugger/WebSocket/SteppingSubscriber.cpp

@@ -107,9 +107,7 @@ void WebSocketSteppingState::Into(DebuggerRequest &req) {
 		CBreakPoints::SetSkipFirst(currentMIPS->pc);
 
 		int c = GetNextInstructionCount(cpuDebug);
-		for (int i = 0; i < c; ++i) {
-			Core_DoSingleStep();
-		}
+		Core_RequestSingleStep(CPUStepType::Into, c);
 	} else {
 		uint32_t breakpointAddress = cpuDebug->GetPC();
 		PrepareResume();
@@ -278,7 +276,8 @@ int WebSocketSteppingState::GetNextInstructionCount(DebugInterface *cpuDebug) {
 
 void WebSocketSteppingState::PrepareResume() {
 	if (currentMIPS->inDelaySlot) {
-		Core_DoSingleStep();
+		// Delay slot instructions are never joined, so we pass 1.
+		Core_RequestSingleStep(CPUStepType::Into, 1);
 	} else {
 		// If the current PC is on a breakpoint, the user doesn't want to do nothing.
 		CBreakPoints::SetSkipFirst(currentMIPS->pc);

Core/SaveState.cpp

@@ -419,7 +419,6 @@ namespace SaveState
 		// Don't actually run it until next frame.
 		// It's possible there might be a duplicate but it won't hurt us.
 		needsProcess = true;
-		Core_UpdateSingleStep();
 	}
 
 	void Load(const Path &filename, int slot, Callback callback, void *cbUserData)

Core/System.cpp

@@ -118,6 +118,9 @@ static volatile bool pspIsIniting = false;
 static volatile bool pspIsQuitting = false;
 static volatile bool pspIsRebooting = false;
 
+// This is called on EmuThread before RunLoop.
+void Core_ProcessStepping(MIPSDebugInterface *cpu);
+
 void ResetUIState() {
 	globalUIState = UISTATE_MENU;
 }
@@ -386,7 +389,6 @@ void Core_UpdateState(CoreState newState) {
 	if ((coreState == CORE_RUNNING || coreState == CORE_NEXTFRAME) && newState != CORE_RUNNING)
 		coreStatePending = true;
 	coreState = newState;
-	Core_UpdateSingleStep();
 }
 
 void Core_UpdateDebugStats(bool collectStats) {
@@ -629,7 +631,7 @@ void PSP_RunLoopUntil(u64 globalticks) {
 	if (coreState == CORE_POWERDOWN || coreState == CORE_BOOT_ERROR || coreState == CORE_RUNTIME_ERROR) {
 		return;
 	} else if (coreState == CORE_STEPPING) {
-		Core_ProcessStepping();
+		Core_ProcessStepping(currentDebugMIPS);
 		return;
 	}
 

GPU/Debugger/Stepping.cpp

@@ -74,8 +74,8 @@ static void SetPauseAction(PauseAction act, bool waitComplete = true) {
 	pauseAction = act;
 	pauseLock.unlock();
 
-	if (coreState == CORE_STEPPING && act != PAUSE_CONTINUE)
-		Core_UpdateSingleStep();
+	// if (coreState == CORE_STEPPING && act != PAUSE_CONTINUE)
+	// 	Core_UpdateSingleStep();
 
 	actionComplete = false;
 	pauseWait.notify_all();

Windows/Debugger/Debugger_Disasm.cpp

@@ -208,7 +208,7 @@ void CDisasm::step(CPUStepType stepType) {
 	lastTicks_ = CoreTiming::GetTicks();
 
 	u32 stepSize = ptr->getInstructionSizeAt(cpu->GetPC());
-	Core_PerformStep(cpu, stepType, stepSize);
+	Core_RequestSingleStep(stepType, stepSize);
 }
 
 void CDisasm::runToLine() {