It seems like CoreTiming's "external event" mutex is being used recursively. This fixes dsp lle jit on thread for me. YMMV

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@7342 8ced0084-cf51-0410-be5f-012b33b47a6e

5 files changed, 24 insertions, 58 deletions

diff --git a/Source/Core/Core/Src/CoreTiming.cpp b/Source/Core/Core/Src/CoreTiming.cpp
index 8095b56bde..9aacea5847 100644
--- a/Source/Core/Core/Src/CoreTiming.cpp
+++ b/Source/Core/Core/Src/CoreTiming.cpp
@@ -68,7 +68,7 @@ u64 fakeDecStartTicks;
 u64 fakeTBStartValue;
 u64 fakeTBStartTicks;
 
-static std::mutex externalEventSection;
+static std::recursive_mutex externalEventSection;
 
 void (*advanceCallback)(int cyclesExecuted) = NULL;
 
@@ -143,7 +143,7 @@ void Shutdown()
         delete ev;
     }
 
-    std::lock_guard<std::mutex> lk(externalEventSection);
+    std::lock_guard<std::recursive_mutex> lk(externalEventSection);
     while(eventTsPool)
     {
         Event *ev = eventTsPool;
@@ -154,7 +154,7 @@ void Shutdown()
 
 void DoState(PointerWrap &p)
 {
-	std::lock_guard<std::mutex> lk(externalEventSection);
+	std::lock_guard<std::recursive_mutex> lk(externalEventSection);
 	p.Do(downcount);
 	p.Do(slicelength);
 	p.Do(globalTimer);
@@ -225,7 +225,7 @@ u64 GetIdleTicks()
 // schedule things to be executed on the main thread.
 void ScheduleEvent_Threadsafe(int cyclesIntoFuture, int event_type, u64 userdata)
 {
-	std::lock_guard<std::mutex> lk(externalEventSection);
+	std::lock_guard<std::recursive_mutex> lk(externalEventSection);
 	Event *ne = GetNewTsEvent();
 	ne->time = globalTimer + cyclesIntoFuture;
 	ne->type = event_type;
@@ -244,7 +244,7 @@ void ScheduleEvent_Threadsafe_Immediate(int event_type, u64 userdata)
 {
 	if(Core::IsCPUThread())
 	{
-		std::lock_guard<std::mutex> lk(externalEventSection);
+		std::lock_guard<std::recursive_mutex> lk(externalEventSection);
 		event_types[event_type].callback(userdata, 0);
 	}
 	else
@@ -348,7 +348,7 @@ void RemoveEvent(int event_type)
 
 void RemoveThreadsafeEvent(int event_type)
 {
-	std::lock_guard<std::mutex> lk(externalEventSection);
+	std::lock_guard<std::recursive_mutex> lk(externalEventSection);
 	if (!tsFirst)
 	{
 		return;
@@ -431,7 +431,7 @@ void ProcessFifoWaitEvents()
 
 void MoveEvents()
 {
-	std::lock_guard<std::mutex> lk(externalEventSection);
+	std::lock_guard<std::recursive_mutex> lk(externalEventSection);
     // Move events from async queue into main queue
 	while (tsFirst)
 	{
diff --git a/Source/Core/Core/Src/DSP/DSPCore.cpp b/Source/Core/Core/Src/DSP/DSPCore.cpp
index bf115019fd..5085fb5f37 100644
--- a/Source/Core/Core/Src/DSP/DSPCore.cpp
+++ b/Source/Core/Core/Src/DSP/DSPCore.cpp
@@ -42,7 +42,6 @@ DSPCoreState core_state = DSPCORE_STOP;
 u16 cyclesLeft = 0;
 DSPEmitter *dspjit = NULL;
 Common::Event step_event;
-static std::mutex ExtIntCriticalSection;
 
 static bool LoadRom(const char *fname, int size_in_words, u16 *rom)
 {
@@ -204,7 +203,6 @@ void DSPCore_SetException(u8 level)
 // Notify that an external interrupt is pending (used by thread mode)
 void DSPCore_SetExternalInterrupt(bool val)
 {
-	std::lock_guard<std::mutex> lk(ExtIntCriticalSection);
 	g_dsp.external_interrupt_waiting = val;
 }
 
@@ -274,7 +272,6 @@ int DSPCore_RunCycles(int cycles)
 
 	while (cycles > 0)
 	{
-	reswitch:
 		switch (core_state)
 		{
 		case DSPCORE_RUNNING:
@@ -289,7 +286,7 @@ int DSPCore_RunCycles(int cycles)
 		case DSPCORE_STEPPING:
 			step_event.Wait();
 			if (core_state != DSPCORE_STEPPING)
-				goto reswitch;
+				continue;
 
 			DSPInterpreter::Step();
 			cycles--;
diff --git a/Source/Core/Core/Src/DSP/DSPCore.h b/Source/Core/Core/Src/DSP/DSPCore.h
index d15da70687..17b692b840 100644
--- a/Source/Core/Core/Src/DSP/DSPCore.h
+++ b/Source/Core/Core/Src/DSP/DSPCore.h
@@ -231,7 +231,7 @@ struct SDSP
 
 	u8 reg_stack_ptr[4];
 	u8 exceptions;   // pending exceptions
-	bool external_interrupt_waiting;
+	volatile bool external_interrupt_waiting;
 
 	// DSP hardware stacks. They're mapped to a bunch of registers, such that writes
 	// to them push and reads pop.
@@ -246,10 +246,7 @@ struct SDSP
 	u64 step_counter;
 
 	// Mailbox.
-	volatile u16 mbox[2][2];
-
-	// Mutex protecting the mailbox.
-	//std::mutex g_CriticalSection;
+	volatile u32 mbox[2];
 
 	// Accelerator / DMA / other hardware registers. Not GPRs.
 	u16 ifx_regs[256];
diff --git a/Source/Core/Core/Src/DSP/DSPEmitter.cpp b/Source/Core/Core/Src/DSP/DSPEmitter.cpp
index 8fc2a94923..b6fa5bb5a9 100644
--- a/Source/Core/Core/Src/DSP/DSPEmitter.cpp
+++ b/Source/Core/Core/Src/DSP/DSPEmitter.cpp
@@ -368,7 +368,7 @@ void DSPEmitter::CompileDispatcher()
 	FixupBranch exceptionExit;
 	if (DSPHost_OnThread())
 	{
-		CMP(8, M(&g_dsp.external_interrupt_waiting), Imm8(0));
+		CMP(8, M(const_cast<bool*>(&g_dsp.external_interrupt_waiting)), Imm8(0));
 		exceptionExit = J_CC(CC_NE);
 	}
 
diff --git a/Source/Core/Core/Src/DSP/DSPHWInterface.cpp b/Source/Core/Core/Src/DSP/DSPHWInterface.cpp
index f3106df45c..17ec10e724 100644
--- a/Source/Core/Core/Src/DSP/DSPHWInterface.cpp
+++ b/Source/Core/Core/Src/DSP/DSPHWInterface.cpp
@@ -23,6 +23,7 @@
 
    ====================================================================*/
 
+#include "Atomic.h"
 #include "Thread.h"
 #include "MemoryUtil.h"
 
@@ -41,8 +42,6 @@
 
 static void gdsp_do_dma();
 
-static std::mutex g_CriticalSection;
-
 void gdsp_ifx_init()
 {
 	for (int i = 0; i < 256; i++)
@@ -50,40 +49,25 @@ void gdsp_ifx_init()
 		g_dsp.ifx_regs[i] = 0;
 	}
 
-	g_dsp.mbox[0][0] = 0;
-	g_dsp.mbox[0][1] = 0;
-	g_dsp.mbox[1][0] = 0;
-	g_dsp.mbox[1][1] = 0;
+	g_dsp.mbox[0] = 0;
+	g_dsp.mbox[1] = 0;
 }
 
 u32 gdsp_mbox_peek(u8 mbx)
 {
-	std::unique_lock<std::mutex> lk(g_CriticalSection, std::defer_lock);
-	if (DSPHost_OnThread())
-		lk.lock();
-
-	return ((g_dsp.mbox[mbx][0] << 16) | g_dsp.mbox[mbx][1]);
+	return Common::AtomicLoad(g_dsp.mbox[mbx]);
 }
 
 void gdsp_mbox_write_h(u8 mbx, u16 val)
 {
-	std::unique_lock<std::mutex> lk(g_CriticalSection, std::defer_lock);
-	if (DSPHost_OnThread())
-		lk.lock();
-
-	g_dsp.mbox[mbx][0] = val & 0x7fff;
+	const u32 new_value = (Common::AtomicLoadAcquire(g_dsp.mbox[mbx]) & 0xffff) | (val << 16);
+	Common::AtomicStoreRelease(g_dsp.mbox[mbx], new_value & ~0x80000000);
 }
 
 void gdsp_mbox_write_l(u8 mbx, u16 val)
 {
-	{
-	std::unique_lock<std::mutex> lk(g_CriticalSection, std::defer_lock);
-	if (DSPHost_OnThread())
-		lk.lock();
-
-	g_dsp.mbox[mbx][1]  = val;
-	g_dsp.mbox[mbx][0] |= 0x8000;
-	}
+	const u32 new_value = (Common::AtomicLoadAcquire(g_dsp.mbox[mbx]) & ~0xffff) | val;
+	Common::AtomicStoreRelease(g_dsp.mbox[mbx], new_value | 0x80000000);
 
 #if defined(_DEBUG) || defined(DEBUGFAST)
 	if (mbx == GDSP_MBOX_DSP)
@@ -97,25 +81,13 @@ void gdsp_mbox_write_l(u8 mbx, u16 val)
 
 u16 gdsp_mbox_read_h(u8 mbx)
 {
-	std::unique_lock<std::mutex> lk(g_CriticalSection, std::defer_lock);
-	if (DSPHost_OnThread())
-		lk.lock();
-
-	return g_dsp.mbox[mbx][0];  // TODO: mask away the top bit?
+	return (u16)(Common::AtomicLoad(g_dsp.mbox[mbx]) >> 16); // TODO: mask away the top bit?
 }
 
-
 u16 gdsp_mbox_read_l(u8 mbx)
 {
-	u16 val;
-	{
-	std::unique_lock<std::mutex> lk(g_CriticalSection, std::defer_lock);
-	if (DSPHost_OnThread())
-		lk.lock();
-
-	val = g_dsp.mbox[mbx][1];
-	g_dsp.mbox[mbx][0] &= ~0x8000;
-	}
+	const u32 value = Common::AtomicLoadAcquire(g_dsp.mbox[mbx]);
+	Common::AtomicStoreRelease(g_dsp.mbox[mbx], value & ~0x80000000);
 
 #if defined(_DEBUG) || defined(DEBUGFAST)
 	if (mbx == GDSP_MBOX_DSP)
@@ -124,9 +96,9 @@ u16 gdsp_mbox_read_l(u8 mbx)
 	} else {
 		NOTICE_LOG(DSP_MAIL, "CPU(RM) B:%i M:0x%08x (pc=0x%04x)", mbx, gdsp_mbox_peek(GDSP_MBOX_CPU), g_dsp.pc);
 	}
-#endif	
+#endif
 
-	return val;
+	return (u16)value;
 }
 
 void gdsp_ifx_write(u32 addr, u32 val)