coreinit_Thread.cpp

#include "Cafe/OS/common/OSCommon.h"
#include "Cafe/HW/Espresso/PPCCallback.h"
#include "Cafe/OS/RPL/rpl.h"
#include "Cafe/OS/libs/coreinit/coreinit_Thread.h"
#include "Cafe/OS/libs/coreinit/coreinit_Time.h"
#include "Cafe/OS/libs/coreinit/coreinit_Alarm.h"
#include "Cafe/OS/libs/snd_core/ax.h"
#include "Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h"
#include "Cafe/HW/Espresso/Recompiler/PPCRecompiler.h"

#include "util/helpers/Semaphore.h"
#include "util/helpers/ConcurrentQueue.h"
#include "util/Fiber/Fiber.h"

#include "util/helpers/helpers.h"

SlimRWLock srwlock_activeThreadList;

// public list of active threads
MPTR activeThread[256];
sint32 activeThreadCount = 0;

MPTR exitThreadPtr = 0;

void nnNfp_update();

namespace coreinit
{
	void __OSFiberThreadEntry(void* thread);
	void __OSAddReadyThreadToRunQueue(OSThread_t* thread);
	void __OSRemoveThreadFromRunQueues(OSThread_t* thread);
};

namespace coreinit
{
	// scheduler state
	std::atomic<bool> sSchedulerActive;
	std::vector<std::thread> sSchedulerThreads;
	std::mutex sSchedulerStateMtx;

	SysAllocator<OSThreadQueue> g_activeThreadQueue; // list of all threads (can include non-detached inactive threads)

	SysAllocator<OSThreadQueue, 3> g_coreRunQueue;
	CounterSemaphore g_coreRunQueueThreadCount[3];

	bool g_isMulticoreMode;

	__declspec(thread) uint32 t_assignedCoreIndex;
	__declspec(thread) Fiber* t_schedulerFiber;

	struct OSHostThread
	{
		OSHostThread(OSThread_t* thread) : m_thread(thread), m_fiber(__OSFiberThreadEntry, this, this)
		{
		}

		~OSHostThread()
		{
		}

		OSThread_t* m_thread;
		Fiber m_fiber;
		// padding (used as stack memory in recompiler)
		uint8  padding[1024 * 128];
		PPCInterpreter_t ppcInstance;
		uint32 selectedCore;
	};

	std::unordered_map<OSThread_t*, OSHostThread*> s_threadToFiber;

	// create host thread (fiber) that will be used to run the PPC instance
	// note that host threads are fibers and not actual threads
	void __OSCreateHostThread(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		cemu_assert_debug(s_threadToFiber.find(thread) == s_threadToFiber.end());

		OSHostThread* hostThread = new OSHostThread(thread);
		s_threadToFiber.emplace(thread, hostThread);
	}

	// delete host thread
	void __OSDeleteHostThread(OSThread_t* thread)
	{
		static OSHostThread* _deleteQueue = nullptr;
		cemu_assert_debug(__OSHasSchedulerLock());

		if (_deleteQueue)
		{
			delete _deleteQueue;
			_deleteQueue = nullptr;
		}

		// delete with a delay (using queue of length 1)
		// since the fiber might still be in use right now we have to delay the deletion

		auto hostThread = s_threadToFiber[thread];
		s_threadToFiber.erase(thread);
		_deleteQueue = hostThread;
	}


	// add thread to active queue
	void __OSActivateThread(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());

		g_activeThreadQueue->addThread(thread, &thread->activeThreadChain); // todo - check if thread already in queue

		MPTR threadMPTR = memory_getVirtualOffsetFromPointer(thread);

		srwlock_activeThreadList.LockWrite();
		bool alreadyActive = false;
		for (sint32 i = 0; i < activeThreadCount; i++)
		{
			if (activeThread[i] == threadMPTR)
			{
				cemu_assert_debug(false); // should not happen
				alreadyActive = true;
			}
		}
		if (alreadyActive == false)
		{
			activeThread[activeThreadCount] = threadMPTR;
			activeThreadCount++;
		}

		__OSCreateHostThread(thread);

		srwlock_activeThreadList.UnlockWrite();
	}

	// remove thread from active queue. Reset id and state
	void __OSDeactivateThread(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());

		// remove thread from active thread list
		MPTR t = memory_getVirtualOffsetFromPointer(thread);
		srwlock_activeThreadList.LockWrite();
		bool noHit = true;
		for (sint32 i = 0; i < activeThreadCount; i++)
		{
			if (activeThread[i] == t)
			{
				activeThread[i] = activeThread[activeThreadCount - 1];
				activeThreadCount--;
				noHit = false;
				break;
			}
		}
		cemu_assert_debug(noHit == false);
		srwlock_activeThreadList.UnlockWrite();
		
		g_activeThreadQueue->removeThread(thread, &thread->activeThreadChain); // todo - check if thread in queue
		
		cemu_assert_debug(thread->state == OSThread_t::THREAD_STATE::STATE_NONE);
		thread->id = 0x8000;

		__OSDeleteHostThread(thread);
	}

	bool __OSIsThreadActive(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		MPTR threadMPTR = memory_getVirtualOffsetFromPointer(thread);
		srwlock_activeThreadList.LockWrite();
		bool isRunable = false;
		for (sint32 i = 0; i < activeThreadCount; i++)
		{
			if (activeThread[i] == threadMPTR)
			{
				srwlock_activeThreadList.UnlockWrite();
				return true;
			}
		}
		srwlock_activeThreadList.UnlockWrite();
		return false;
	}

	// thread
	OSThread_t* __currentCoreThread[3] = {};

	void OSSetCurrentThread(uint32 coreIndex, OSThread_t* thread)
	{
		if (coreIndex < 3)
			__currentCoreThread[coreIndex] = thread;
	}

	OSThread_t* OSGetCurrentThread()
	{
		PPCInterpreter_t* currentInstance = PPCInterpreter_getCurrentInstance();
		if (currentInstance == nullptr)
			return nullptr;
		return __currentCoreThread[currentInstance->spr.UPIR];
	}

	MPTR funcPtr_threadEntry = 0;

	void threadEntry(PPCInterpreter_t* hCPU)
	{
		OSThread_t* currentThread = coreinitThread_getCurrentThreadDepr(hCPU);
		uint32 r3 = hCPU->gpr[3];
		uint32 r4 = hCPU->gpr[4];
		uint32 lr = hCPU->spr.LR;

		// cpp exception init callback
		//const uint32 im = OSDisableInterrupts(); -> on an actual Wii U interrupts are disabled for this callback, but there are games that yield the thread in the callback (see Angry Birds Star Wars)
		if (gCoreinitData->__cpp_exception_init_ptr != MPTR_NULL)
		{
			PPCCoreCallback(_swapEndianU32(gCoreinitData->__cpp_exception_init_ptr), &currentThread->crt.eh_globals);
		}
		//OSRestoreInterrupts(im);
		// forward to thread entrypoint
		hCPU->spr.LR = lr;
		hCPU->gpr[3] = r3;
		hCPU->gpr[4] = r4;
		hCPU->instructionPointer = _swapEndianU32(currentThread->entrypoint);
	}

	void coreinitExport_OSExitThreadDepr(PPCInterpreter_t* hCPU);

	void initFunctionPointers()
	{
		exitThreadPtr = PPCInterpreter_makeCallableExportDepr(coreinitExport_OSExitThreadDepr);
		funcPtr_threadEntry = PPCInterpreter_makeCallableExportDepr(threadEntry);
	}

	void OSCreateThreadInternal(OSThread_t* thread, uint32 entryPoint, MPTR stackLowerBaseAddr, uint32 stackSize, uint8 affinityMask, OSThread_t::THREAD_TYPE threadType)
	{
		cemu_assert_debug(thread != nullptr); // make thread struct mandatory. Caller can always use SysAllocator
		__OSLockScheduler();
		bool isThreadStillActive = __OSIsThreadActive(thread);
		if (isThreadStillActive)
		{
			// workaround for games that restart threads to quickly
			// seen in Fast Racing Neo at boot (0x020617BC OSCreateThread)
			forceLog_printf("Game attempting to re-initialize existing thread");
			while ((thread->state == OSThread_t::THREAD_STATE::STATE_READY || thread->state == OSThread_t::THREAD_STATE::STATE_RUNNING) && thread->suspendCounter == 0)
			{
				// wait for thread to finish
				__OSUnlockScheduler();
				OSSleepTicks(ESPRESSO_TIMER_CLOCK / 2000); // sleep 0.5ms
				__OSLockScheduler();
			}
			if (__OSIsThreadActive(thread) && thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND)
			{
				forceLog_printf("Calling OSCreateThread() on thread which is still active (Thread exited without detached flag). Forcing OSDetachThread()...");
				__OSUnlockScheduler();
				OSDetachThread(thread);
				__OSLockScheduler();
			}

		}
		cemu_assert_debug(__OSIsThreadActive(thread) == false);
		__OSUnlockScheduler();

		initFunctionPointers();
		if (thread == nullptr)
			thread = (OSThread_t*)memory_getPointerFromVirtualOffset(coreinit_allocFromSysArea(sizeof(OSThread_t), 32));
		memset(thread, 0x00, sizeof(OSThread_t));
		// init signatures
		thread->context.magic0 = OS_CONTEXT_MAGIC_0;
		thread->context.magic1 = OS_CONTEXT_MAGIC_1;
		thread->magic = 'tHrD';
		thread->type = threadType;
		thread->state = (entryPoint != MPTR_NULL) ? OSThread_t::THREAD_STATE::STATE_READY : OSThread_t::THREAD_STATE::STATE_NONE;
		thread->entrypoint = _swapEndianU32(entryPoint);
		__OSSetThreadBasePriority(thread, 0);
		__OSUpdateThreadEffectivePriority(thread);
		// untested, but seems to work (Batman Arkham City uses these values to calculate the stack size for duplicated threads)
		thread->stackBase = _swapEndianU32(stackLowerBaseAddr + stackSize); // these fields are quite important and lots of games rely on them being accurate (Examples: Darksiders 2, SMW3D, Batman Arkham City)
		thread->stackEnd = _swapEndianU32(stackLowerBaseAddr);
		// init stackpointer
		thread->context.gpr[GPR_SP] = _swapEndianU32(stackLowerBaseAddr + stackSize - 0x20); // how many free bytes should there be at the beginning of the stack?
		// init misc stuff
		thread->attr = affinityMask;
		thread->context.setAffinity(affinityMask);
		thread->context.srr0 = funcPtr_threadEntry;
		thread->context.lr = _swapEndianU32(exitThreadPtr);
		thread->id = 0x8000; // Warriors Orochi 3 softlocks if this is zero due to confusing threads (_OSActivateThread should set this?)
		// init ugqr
		thread->context.gqr[0] = 0x00000000;
		thread->context.gqr[1] = 0x00000000;
		thread->context.gqr[2] = 0x00040004;
		thread->context.gqr[3] = 0x00050005;
		thread->context.gqr[4] = 0x00060006;
		thread->context.gqr[5] = 0x00070007;
		thread->context.gqr[6] = 0x00000000;
		thread->context.gqr[7] = 0x00000000;
		// init r2 (SDA2) and r3 (SDA)
		thread->context.gpr[2] = _swapEndianU32(RPLLoader_GetSDA2Base());
		thread->context.gpr[13] = _swapEndianU32(RPLLoader_GetSDA1Base());
		// GHS related thread init?

		__OSLockScheduler();
		// if entrypoint is non-zero then put the thread on the active list and suspend it
		if (entryPoint != MPTR_NULL)
		{
			thread->suspendCounter = 1;
			__OSActivateThread(thread);
			thread->state = OSThread_t::THREAD_STATE::STATE_READY;
		}
		else
			thread->suspendCounter = 0;
		__OSUnlockScheduler();
	}

	bool OSCreateThreadType(OSThread_t* thread, MPTR entryPoint, sint32 numParam, void* ptrParam, void* stackTop2, sint32 stackSize, sint32 priority, uint32 attr, OSThread_t::THREAD_TYPE threadType)
	{
		OSCreateThreadInternal(thread, entryPoint, memory_getVirtualOffsetFromPointer(stackTop2) - stackSize, stackSize, attr, threadType);
		thread->context.gpr[3] = _swapEndianU32(numParam); // num arguments
		thread->context.gpr[4] = _swapEndianU32(memory_getVirtualOffsetFromPointer(ptrParam)); // arguments pointer
		__OSSetThreadBasePriority(thread, priority);
		__OSUpdateThreadEffectivePriority(thread);
		// set affinity
		uint8 affinityMask = 0;
		affinityMask = attr & 0x7;
		// if no core is selected -> set current one
		if (affinityMask == 0)
			affinityMask |= (1 << PPCInterpreter_getCoreIndex(ppcInterpreterCurrentInstance));
		// set attr
		// todo: Support for other attr bits
		thread->attr = (affinityMask & 0xFF) | (attr & OSThread_t::ATTR_BIT::ATTR_DETACHED);
		thread->context.setAffinity(affinityMask);
		// recompile entry point function
		if (entryPoint != MPTR_NULL)
			PPCRecompiler_recompileIfUnvisited(entryPoint);
		return true;
	}

	bool OSCreateThread(OSThread_t* thread, MPTR entryPoint, sint32 numParam, void* ptrParam, void* stackTop2, sint32 stackSize, sint32 priority, uint32 attr)
	{
		return OSCreateThreadType(thread, entryPoint, numParam, ptrParam, stackTop2, stackSize, priority, attr, OSThread_t::THREAD_TYPE::TYPE_APP);
	}

	bool OSRunThread(OSThread_t* thread, MPTR funcAddress, sint32 numParam, void* ptrParam)
	{
		__OSLockScheduler();

		cemu_assert_debug(ppcInterpreterCurrentInstance == nullptr || OSGetCurrentThread() != thread); // called on self, what should this function do?

		if (thread->state != OSThread_t::THREAD_STATE::STATE_NONE && thread->state != OSThread_t::THREAD_STATE::STATE_MORIBUND)
		{
			// unsure about this case
			forceLogDebug_printf("OSRunThread called on thread which cannot be ran");
			__OSUnlockScheduler();
			return false;
		}

		if (thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND)
		{
			thread->state = OSThread_t::THREAD_STATE::STATE_NONE;
			coreinit::__OSDeactivateThread(thread);
			coreinit::__OSRemoveThreadFromRunQueues(thread);
		}

		// set thread state
		// todo - this should fully reinitialize the thread?

		thread->entrypoint = _swapEndianU32(funcAddress);
		thread->context.srr0 = coreinit::funcPtr_threadEntry;
		thread->context.lr = _swapEndianU32(exitThreadPtr);
		thread->context.gpr[3] = _swapEndianU32(numParam);
		thread->context.gpr[4] = _swapEndianU32(memory_getVirtualOffsetFromPointer(ptrParam));
		thread->suspendCounter = 0;	// verify

		MPTR threadMPTR = memory_getVirtualOffsetFromPointer(thread);

		coreinit::__OSActivateThread(thread);
		thread->state = OSThread_t::THREAD_STATE::STATE_READY;

		__OSAddReadyThreadToRunQueue(thread);

		__OSUnlockScheduler();

		return true;
	}

	void OSExitThread(sint32 exitValue)
	{
		PPCInterpreter_t* hCPU = PPCInterpreter_getCurrentInstance();
		hCPU->gpr[3] = exitValue;
		OSThread_t* threadBE = coreinitThread_getCurrentThreadDepr(hCPU);
		MPTR t = memory_getVirtualOffsetFromPointer(threadBE);

		// thread cleanup callback
		if (!threadBE->cleanupCallback2.IsNull())
		{
			threadBE->stateFlags = _swapEndianU32(_swapEndianU32(threadBE->stateFlags) | 0x00000001);
			PPCCoreCallback(threadBE->cleanupCallback2.GetMPTR(), threadBE, _swapEndianU32(threadBE->stackEnd));
		}
		// cpp exception cleanup
		if (gCoreinitData->__cpp_exception_cleanup_ptr != 0 && threadBE->crt.eh_globals != nullptr)
		{
			PPCCoreCallback(_swapEndianU32(gCoreinitData->__cpp_exception_cleanup_ptr), &threadBE->crt.eh_globals);
			threadBE->crt.eh_globals = nullptr;
		}
		// set exit code
		threadBE->exitValue = exitValue;

		__OSLockScheduler();

		// release held synchronization primitives
		if (!threadBE->mutexQueue.isEmpty())
		{
			cemuLog_force("OSExitThread: Thread is holding mutexes");
			while (true)
			{
				OSMutex* mutex = threadBE->mutexQueue.getFirst();
				if (!mutex)
					break;
				if (mutex->owner != threadBE)
				{
					cemuLog_force("OSExitThread: Thread is holding mutex which it doesn't own");
					threadBE->mutexQueue.removeMutex(mutex);
					continue;
				}
				coreinit::OSUnlockMutexInternal(mutex);
			}
		}
		// todo - release all fast mutexes

		// handle join queue
		if (!threadBE->joinQueue.isEmpty())
			threadBE->joinQueue.wakeupEntireWaitQueue(false);
	
		if ((threadBE->attr & 8) != 0)
		{
			// deactivate thread since it is detached
			threadBE->state = OSThread_t::THREAD_STATE::STATE_NONE;
			coreinit::__OSDeactivateThread(threadBE);
			// queue call to thread deallocator if set
			if (!threadBE->deallocatorFunc.IsNull())
				__OSQueueThreadDeallocation(threadBE);
		}
		else
		{
			// non-detached threads remain active
			threadBE->state = OSThread_t::THREAD_STATE::STATE_MORIBUND;
		}
		PPCCore_switchToSchedulerWithLock();
	}

	void OSSetThreadSpecific(uint32 index, void* value)
	{
		OSThread_t* currentThread = OSGetCurrentThread();
		if (index >= (uint32)currentThread->specificArray.size())
			return;
		currentThread->specificArray[index] = value;
	}

	void* OSGetThreadSpecific(uint32 index)
	{
		OSThread_t* currentThread = OSGetCurrentThread();
		if (index >= (uint32)currentThread->specificArray.size())
			return nullptr;
		return currentThread->specificArray[index].GetPtr();
	}

	void OSSetThreadName(OSThread_t* thread, char* name)
	{
		thread->threadName = name;
	}

	char* OSGetThreadName(OSThread_t* thread)
	{
		return thread->threadName.GetPtr();
	}

	void coreinitExport_OSExitThreadDepr(PPCInterpreter_t* hCPU)
	{
		ppcDefineParamU32(exitCode, 0);
		OSExitThread(exitCode);
	}

	void OSYieldThread()
	{
		PPCCore_switchToScheduler();
	}

	void _OSSleepTicks_alarmHandler(uint64 currentTick, void* context)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		OSThreadQueue* threadQueue = (OSThreadQueue*)context;
		threadQueue->wakeupEntireWaitQueue(false);
	}

	void OSSleepTicks(uint64 ticks)
	{
		cemu_assert_debug(__OSHasSchedulerLock() == false);
		StackAllocator<OSThreadQueue> _threadQueue;
		OSInitThreadQueue(_threadQueue.GetPointer());
		__OSLockScheduler();
		OSHostAlarm* hostAlarm = OSHostAlarmCreate(coreinit_getOSTime() + ticks, 0, _OSSleepTicks_alarmHandler, _threadQueue.GetPointer());
		_threadQueue.GetPointer()->queueAndWait(OSGetCurrentThread());
		OSHostAlarmDestroy(hostAlarm);
		__OSUnlockScheduler();
	}

	void OSDetachThread(OSThread_t* thread)
	{
		__OSLockScheduler();
		thread->attr |= OSThread_t::ATTR_BIT::ATTR_DETACHED;
		if (thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND)
		{
			// exit thread
			// ?

			// todo -> call deallocator
			thread->state = OSThread_t::THREAD_STATE::STATE_NONE;
			thread->id = 0x8000;
			coreinit::__OSDeactivateThread(thread);
			if (!thread->joinQueue.isEmpty())
			{
				// handle join queue
				thread->joinQueue.wakeupEntireWaitQueue(true);
			}
		}
		__OSUnlockScheduler();
	}

	bool OSJoinThread(OSThread_t* thread, uint32be* exitValue)
	{
		__OSLockScheduler();

		if ((thread->attr & OSThread_t::ATTR_DETACHED) == 0 && thread->state != OSThread_t::THREAD_STATE::STATE_MORIBUND)
		{
			cemu_assert_debug(thread->joinQueue.isEmpty());
			// thread still running, wait in join queue
			thread->joinQueue.queueAndWait(OSGetCurrentThread());
		}
		else if (thread->state != OSThread_t::THREAD_STATE::STATE_MORIBUND)
		{
			// cannot join detached and active threads
			forceLogDebug_printf("Cannot join detached active thread");
			__OSUnlockScheduler();
			return false;
		}

		// thread already ended and is still attached, get return value
		cemu_assert_debug(thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND);
		cemu_assert_debug((thread->attr & OSThread_t::ATTR_DETACHED) == 0);
		if (exitValue)
			*exitValue = thread->exitValue;
		// end thread
		thread->state = OSThread_t::THREAD_STATE::STATE_NONE;
		__OSDeactivateThread(thread);
		coreinit::__OSRemoveThreadFromRunQueues(thread);
		thread->id = 0x8000;

		if (!thread->deallocatorFunc.IsNull())
			__OSQueueThreadDeallocation(thread);

		__OSUnlockScheduler();

		return true;
	}

	// adds the thread to each core's run queue if in runable state
	void __OSAddReadyThreadToRunQueue(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		if (thread->state != OSThread_t::THREAD_STATE::STATE_READY)
			return;
		if (thread->suspendCounter != 0)
			return;
		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
		{
			if (thread->currentRunQueue[i] != nullptr)
				continue; // already on the queue
			// check affinity
			if(!thread->context.hasCoreAffinitySet(i))
				continue;
			g_coreRunQueue.GetPtr()[i].addThread(thread, thread->linkRun + i);
			thread->currentRunQueue[i] = (g_coreRunQueue.GetPtr() + i);
			g_coreRunQueueThreadCount[i].increment();
		}
	}

	void __OSRemoveThreadFromRunQueues(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
		{
			if(thread->currentRunQueue[i] == nullptr)
				continue;
			g_coreRunQueue.GetPtr()[i].removeThread(thread, thread->linkRun + i);
			thread->currentRunQueue[i] = nullptr;
			g_coreRunQueueThreadCount[i].decrement();
		}
	}

	// returns true if thread runs on same core and has higher priority
	bool __OSCoreShouldSwitchToThread(OSThread_t* currentThread, OSThread_t* newThread)
	{
		uint32 coreIndex = OSGetCoreId();
		if (!newThread->context.hasCoreAffinitySet(coreIndex))
			return false;
		return newThread->effectivePriority < currentThread->effectivePriority;
	}

	sint32 __OSResumeThreadInternal(OSThread_t* thread, sint32 resumeCount)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		sint32 previousSuspendCount = thread->suspendCounter;
		cemu_assert_debug(previousSuspendCount >= 0);
		if (previousSuspendCount == 0)
			return 0;
		thread->suspendCounter = previousSuspendCount - resumeCount;
		if (thread->suspendCounter < 0)
			thread->suspendCounter = 0;
		if (thread->suspendCounter == 0)
		{
			__OSAddReadyThreadToRunQueue(thread);
			// set awakened time if thread is ready
			// todo - only set this once?
			thread->wakeUpTime = PPCInterpreter_getMainCoreCycleCounter();
			// reschedule if thread has higher priority
			if (ppcInterpreterCurrentInstance && __OSCoreShouldSwitchToThread(coreinit::OSGetCurrentThread(), thread))
				PPCCore_switchToSchedulerWithLock();
		}
		return previousSuspendCount;
	}

	sint32 OSResumeThread(OSThread_t* thread)
	{
		__OSLockScheduler();
		sint32 previousSuspendCount = __OSResumeThreadInternal(thread, 1);
		__OSUnlockScheduler();
		return previousSuspendCount;
	}

	void OSContinueThread(OSThread_t* thread)
	{
		__OSLockScheduler();
		__OSResumeThreadInternal(thread, thread->suspendCounter);
		__OSUnlockScheduler();
	}

	void __OSSuspendThreadInternal(OSThread_t* thread)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		cemu_assert_debug(thread->state != OSThread_t::THREAD_STATE::STATE_NONE && thread->state != OSThread_t::THREAD_STATE::STATE_MORIBUND); // how to handle these?

		sint32 previousSuspendCount = thread->suspendCounter;

		if (OSGetCurrentThread() == thread)
		{
			thread->suspendCounter = thread->suspendCounter + 1;
			PPCCore_switchToSchedulerWithLock();
		}
		else
		{
			thread->suspendCounter = thread->suspendCounter + 1;
			if (previousSuspendCount == 0)
			{
				__OSRemoveThreadFromRunQueues(thread);
				// todo - if thread is still running find a way to cancel it's timeslice immediately
			}
		}
	}

	void OSSuspendThread(OSThread_t* thread)
	{
		__OSLockScheduler();
		__OSSuspendThreadInternal(thread);
		__OSUnlockScheduler();
	}

	void OSSleepThread(OSThreadQueue* threadQueue)
	{
		__OSLockScheduler();
		threadQueue->queueAndWait(OSGetCurrentThread());
		__OSUnlockScheduler();
	}

	void OSWakeupThread(OSThreadQueue* threadQueue)
	{
		__OSLockScheduler();
		threadQueue->wakeupEntireWaitQueue(true);
		__OSUnlockScheduler();
	}

	bool OSSetThreadAffinity(OSThread_t* thread, uint32 affinityMask)
	{
		cemu_assert_debug((affinityMask & ~7) == 0);
		__OSLockScheduler();
		uint32 prevAffinityMask = thread->context.getAffinity();
		if (thread->state == OSThread_t::THREAD_STATE::STATE_RUNNING)
		{
			thread->attr = (thread->attr & ~7) | (affinityMask & 7);
			thread->context.setAffinity(affinityMask);
			// should this reschedule the thread?
		}
		else if (prevAffinityMask != affinityMask)
		{
			__OSRemoveThreadFromRunQueues(thread);
			thread->attr = (thread->attr & ~7) | (affinityMask & 7);
			thread->context.setAffinity(affinityMask);
			__OSAddReadyThreadToRunQueue(thread);
		}
		__OSUnlockScheduler();
		return true;
	}

	uint32 OSGetThreadAffinity(OSThread_t* thread)
	{
		auto affinityMask = thread->context.getAffinity();
		cemu_assert_debug((affinityMask & ~7) == 0);
		return affinityMask;
	}

	void* OSSetThreadDeallocator(OSThread_t* thread, void* newDeallocatorFunc)
	{
		__OSLockScheduler();
		void* previousFunc = thread->deallocatorFunc.GetPtr();
		thread->deallocatorFunc = newDeallocatorFunc;
		__OSUnlockScheduler();
		return previousFunc;
	}

	void* OSSetThreadCleanupCallback(OSThread_t* thread, void* cleanupCallback)
	{
		__OSLockScheduler();
		void* previousFunc = thread->cleanupCallback2.GetPtr();
		thread->cleanupCallback2 = cleanupCallback;
		__OSUnlockScheduler();
		return previousFunc;
	}

	void __OSSetThreadBasePriority(OSThread_t* thread, sint32 newPriority)
	{
		cemu_assert_debug(newPriority >= 0 && newPriority < 32);
		newPriority += OSThread_t::GetTypePriorityBase(thread->type);
		thread->basePriority = newPriority;
	}

	void __OSUpdateThreadEffectivePriority(OSThread_t* thread)
	{
		if (thread->context.boostCount != 0)
		{
			// temporarily boosted threads have their priority set to 0 (maximum)
			thread->effectivePriority = 0;
			return;
		}
		thread->effectivePriority = thread->basePriority;
	}

	bool OSSetThreadPriority(OSThread_t* thread, sint32 newPriority)
	{
		if (newPriority < 0 || newPriority >= 0x20)
		{
			cemu_assert_debug(false);
			return false; // invalid priority value
		}
		__OSLockScheduler();
		__OSSetThreadBasePriority(thread, newPriority);
		__OSUpdateThreadEffectivePriority(thread);
		// reschedule if needed
		OSThread_t* currentThread = OSGetCurrentThread();
		if (currentThread && currentThread != thread)
		{
			if (__OSCoreShouldSwitchToThread(currentThread, thread))
				PPCCore_switchToSchedulerWithLock();
		}
		__OSUnlockScheduler();
		return true;
	}

	sint32 OSGetThreadPriority(OSThread_t* thread)
	{
		sint32 threadPriority = thread->basePriority;
		OSThread_t::THREAD_TYPE threadType = thread->type;
		threadPriority -= OSThread_t::GetTypePriorityBase(threadType);
		cemu_assert_debug(threadPriority >= 0 && threadPriority < 32);
		return threadPriority;
	}

	bool OSIsThreadTerminated(OSThread_t* thread)
	{
		__OSLockScheduler();
		bool isTerminated = false;
		if (thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND || thread->state == OSThread_t::THREAD_STATE::STATE_NONE)
			isTerminated = true;
		__OSUnlockScheduler();
		return isTerminated;
	}

	bool OSIsThreadSuspended(OSThread_t* thread)
	{
		__OSLockScheduler();
		sint32 suspendCounter = thread->suspendCounter;
		__OSUnlockScheduler();
		return suspendCounter > 0;
	}

	void OSCancelThread(OSThread_t* thread)
	{
		__OSLockScheduler();
		cemu_assert_debug(thread->requestFlags == 0 || thread->requestFlags == OSThread_t::REQUEST_FLAG_CANCEL); // todo - how to handle cases where other flags are already set?
		thread->requestFlags = OSThread_t::REQUEST_FLAG_CANCEL;
		__OSUnlockScheduler();
		// if the canceled thread is self, then exit immediately
		if (thread == OSGetCurrentThread())
			OSExitThread(-1);
	}

	void OSTestThreadCancelInternal()
	{
		// also handles suspend request ?
		cemu_assert_debug(__OSHasSchedulerLock());
		OSThread_t* thread = OSGetCurrentThread();
		if (thread->requestFlags == OSThread_t::REQUEST_FLAG_CANCEL)
		{
			__OSUnlockScheduler();
			OSExitThread(-1);
		}
	}

	void OSTestThreadCancel()
	{
		__OSLockScheduler();
		OSTestThreadCancelInternal();
		__OSUnlockScheduler();
	}

	void __OSSwitchToThreadFiber(OSThread_t* thread, uint32 coreIndex)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		cemu_assert_debug(s_threadToFiber.find(thread) != s_threadToFiber.end());

		OSHostThread* hostThread = s_threadToFiber.find(thread)->second;
		hostThread->selectedCore = coreIndex;
		Fiber::Switch(hostThread->m_fiber);
	}

	void __OSThreadLoadContext(PPCInterpreter_t* hCPU, OSThread_t* thread)
	{
		// load GPR
		for (uint32 i = 0; i < 32; i++)
			hCPU->gpr[i] = _swapEndianU32(thread->context.gpr[i]);

		// load SPR cr, lr, ctr, xer
		ppc_setCR(hCPU, thread->context.cr);
		hCPU->spr.LR = _swapEndianU32(thread->context.lr);
		hCPU->spr.CTR = thread->context.ctr;
		PPCInterpreter_setXER(hCPU, thread->context.xer);

		hCPU->fpscr = thread->context.fpscr.fpscr;

		// store floating point and Gekko registers
		for (uint32 i = 0; i < 32; i++)
		{
			hCPU->fpr[i].fp0int = thread->context.fp_ps0[i];
			hCPU->fpr[i].fp1int = thread->context.fp_ps1[i];
		}

		for (uint32 i = 0; i < 8; i++)
		{
			hCPU->spr.UGQR[0 + i] = thread->context.gqr[i];
		}

		hCPU->instructionPointer = thread->context.srr0;
	}

	void __OSThreadStoreContext(PPCInterpreter_t* hCPU, OSThread_t* thread)
	{
		// store GPR
		for (uint32 i = 0; i < 32; i++)
			thread->context.gpr[i] = _swapEndianU32(hCPU->gpr[i]);
		// store SPRs
		thread->context.cr = ppc_getCR(hCPU);
		thread->context.lr = _swapEndianU32(hCPU->spr.LR);
		thread->context.ctr = hCPU->spr.CTR;
		thread->context.xer = PPCInterpreter_getXER(hCPU);

		thread->context.fpscr.fpscr = hCPU->fpscr;
		thread->context.fpscr.padding = 0;

		// store floating point and Gekko registers
		for (uint32 i = 0; i < 32; i++)
		{
			thread->context.fp_ps0[i] = hCPU->fpr[i].fp0int;
			thread->context.fp_ps1[i] = hCPU->fpr[i].fp1int;
		}

		for (uint32 i = 0; i < 8; i++)
		{
			thread->context.gqr[i] = hCPU->spr.UGQR[0 + i];
		}

		thread->context.srr0 = hCPU->instructionPointer;
	}

	void __OSStoreThread(OSThread_t* thread, PPCInterpreter_t* hCPU)
	{
		if (thread->state == OSThread_t::THREAD_STATE::STATE_RUNNING)
		{
			thread->state = OSThread_t::THREAD_STATE::STATE_READY;
			__OSAddReadyThreadToRunQueue(thread);
		}
		else if (thread->state == OSThread_t::THREAD_STATE::STATE_MORIBUND || thread->state == OSThread_t::THREAD_STATE::STATE_NONE || thread->state == OSThread_t::THREAD_STATE::STATE_WAITING)
		{
			// thread exited or suspending/waiting
		}
		else
		{
			cemu_assert_debug(false);
		}

		thread->requestFlags = (OSThread_t::REQUEST_FLAG_BIT)(thread->requestFlags & OSThread_t::REQUEST_FLAG_CANCEL); // remove all flags except cancel flag

		// update total cycles
		uint64 remainingCycles = std::min((uint64)ppcInterpreterCurrentInstance->remainingCycles, (uint64)thread->quantumTicks);
		uint64 executedCycles = thread->quantumTicks - remainingCycles;
		if (executedCycles < ppcInterpreterCurrentInstance->skippedCycles)
			executedCycles = 0;
		else
			executedCycles -= ppcInterpreterCurrentInstance->skippedCycles;
		thread->totalCycles += executedCycles;
		// store context and set current thread to null
		__OSThreadStoreContext(hCPU, thread);
		OSSetCurrentThread(OSGetCoreId(), nullptr);
		ppcInterpreterCurrentInstance = nullptr;
	}

	void __OSLoadThread(OSThread_t* thread, PPCInterpreter_t* hCPU, uint32 coreIndex)
	{
		hCPU->LSQE = 1;
		hCPU->PSE = 1;
		hCPU->reservedMemAddr = MPTR_NULL;
		hCPU->reservedMemValue = 0;
		hCPU->spr.UPIR = coreIndex;
		hCPU->coreInterruptMask = 1;
		ppcInterpreterCurrentInstance = hCPU;
		OSSetCurrentThread(OSGetCoreId(), thread);
		__OSThreadLoadContext(hCPU, thread);
		thread->context.upir = coreIndex;
		thread->quantumTicks = ppcThreadQuantum;
		// statistics
		thread->wakeUpTime = PPCInterpreter_getMainCoreCycleCounter();
		thread->wakeUpCount = thread->wakeUpCount + 1;
	}

	OSThread_t* __OSGetNextRunableThread(uint32 coreIndex)
	{
		cemu_assert_debug(__OSHasSchedulerLock());
		// pick thread, then remove from run queue
		OSThreadQueue* runQueue = g_coreRunQueue.GetPtr() + coreIndex;

		OSThread_t* threadItr = runQueue->head.GetPtr();
		OSThread_t* selectedThread = nullptr;
		if (!threadItr)
			return nullptr;
		selectedThread = threadItr;

		while (threadItr)
		{
			if (threadItr->effectivePriority < selectedThread->effectivePriority)
				selectedThread = threadItr;
			threadItr = threadItr->linkRun[coreIndex].next.GetPtr();
		}

		cemu_assert_debug(selectedThread->state == OSThread_t::THREAD_STATE::STATE_READY);

		__OSRemoveThreadFromRunQueues(selectedThread);
		selectedThread->state = OSThread_t::THREAD_STATE::STATE_RUNNING;

		return selectedThread;
	}

	void __OSCheckSystemEvents()
	{
		// AX update
		snd_core::AXOut_update();
		// alarm update
		coreinit::alarm_update();
		// nfp update
		nnNfp_update();
	}

	Fiber* g_idleLoopFiber[3]{};

	// idle fiber per core if no thread is runnable
	// this is necessary since we can't block in __OSThreadSwitchToNext() (__OSStoreThread + thread switch must happen inside same scheduler lock)
	void __OSThreadCoreIdle(void* unusedParam)
	{
		bool isMainCore = g_isMulticoreMode == false || t_assignedCoreIndex == 1;
		sint32 coreIndex = t_assignedCoreIndex;
		__OSUnlockScheduler();
		while (true)
		{
			if (!g_coreRunQueueThreadCount[coreIndex].isZero()) // avoid hammering the lock on the main core if there is no runable thread
			{
				__OSLockScheduler();
				OSThread_t* nextThread = __OSGetNextRunableThread(coreIndex);
				if (nextThread)
				{
					cemu_assert_debug(nextThread->state == OSThread_t::THREAD_STATE::STATE_RUNNING);
					__OSSwitchToThreadFiber(nextThread, coreIndex);
				}
				__OSUnlockScheduler();
			}
			if (isMainCore)
			{
				__OSCheckSystemEvents();
				if(g_isMulticoreMode == false)
					coreIndex = (coreIndex + 1) % 3;
			}
			else
			{
				// wait for semaphore (only in multicore mode)
				g_coreRunQueueThreadCount[t_assignedCoreIndex].waitUntilNonZero();
				if (!sSchedulerActive.load(std::memory_order::relaxed))
					Fiber::Switch(*t_schedulerFiber); // switch back to original thread to exit
			}
		}
	}

	void __OSThreadSwitchToNext()
	{
		cemu_assert_debug(__OSHasSchedulerLock());

		OSHostThread* hostThread = (OSHostThread*)Fiber::GetFiberPrivateData();
        cemu_assert_debug(hostThread);

		//if (ppcInterpreterCurrentInstance)
		//	debug_printf("Core %d store thread %08x (t = %d)\n", hostThread->ppcInstance.sprNew.UPIR, memory_getVirtualOffsetFromPointer(hostThread->thread), t_assignedCoreIndex);

		// store context of current thread
		__OSStoreThread(OSGetCurrentThread(), &hostThread->ppcInstance);
		cemu_assert_debug(ppcInterpreterCurrentInstance == nullptr);

		if (!sSchedulerActive.load(std::memory_order::relaxed))
		{
			__OSUnlockScheduler();
			Fiber::Switch(*t_schedulerFiber); // switch back to original thread entry for it to exit
		}
		// choose core
		static uint32 _coreCounter = 0;

		sint32 coreIndex;
		if (g_isMulticoreMode)
			coreIndex = t_assignedCoreIndex;
		else
		{
			coreIndex = _coreCounter;
			_coreCounter = (_coreCounter + 1) % 3;
		}

		// if main thread then dont forget to do update checks
		bool isMainThread = g_isMulticoreMode == false || t_assignedCoreIndex == 1;

		// find next thread to run
		// for main thread we force switching to the idle loop since it calls __OSCheckSystemEvents()
		if(isMainThread)
			Fiber::Switch(*g_idleLoopFiber[t_assignedCoreIndex]);
		else if (OSThread_t* nextThread = __OSGetNextRunableThread(coreIndex))
		{
			cemu_assert_debug(nextThread->state == OSThread_t::THREAD_STATE::STATE_RUNNING);
			__OSSwitchToThreadFiber(nextThread, coreIndex);
		}
		else
		{
			Fiber::Switch(*g_idleLoopFiber[t_assignedCoreIndex]);
		}

		cemu_assert_debug(__OSHasSchedulerLock());	
		cemu_assert_debug(g_isMulticoreMode == false || hostThread->selectedCore == t_assignedCoreIndex);

		// load self thread
		__OSLoadThread(hostThread->m_thread, &hostThread->ppcInstance, hostThread->selectedCore);
	}

	void __OSFiberThreadEntry(void* _thread)
	{
		OSHostThread* hostThread = (OSHostThread*)_thread;

		_mm_setcsr(_mm_getcsr() | 0x8000); // flush denormals to zero

		uint32 lehmer_lcg = 12345;

		PPCInterpreter_t* hCPU = &hostThread->ppcInstance;
		__OSLoadThread(hostThread->m_thread, hCPU, hostThread->selectedCore);
		__OSUnlockScheduler(); // lock is always held when switching to a fiber, so we need to unlock it here
		while (true)
		{
			// run one timeslice
			hCPU->remainingCycles = ppcThreadQuantum;
			hCPU->skippedCycles = 0;
			// we add a slight randomized variance to the thread quantum to avoid getting stuck in repeated code sequences where the duration matches the thread quantum perfectly
			// this was seen in Mario Party 10 on launch where several OSLockMutex operations would align in such a way that one thread would never successfully acquire the lock
			hCPU->remainingCycles += (lehmer_lcg & 0x7F);
			lehmer_lcg = (uint32)((uint64)lehmer_lcg * 279470273ull % 0xfffffffbull);

			if (hCPU->remainingCycles > 0)
			{
				// try to enter recompiler immediately
				PPCRecompiler_attemptEnterWithoutRecompile(hCPU, hCPU->instructionPointer);
				// keep executing as long as there are cycles left
				while ((--hCPU->remainingCycles) >= 0)
					PPCInterpreterSlim_executeInstruction(hCPU);
			}

			// reset reservation
			hCPU->reservedMemAddr = 0;
			hCPU->reservedMemValue = 0;

			// reschedule
			__OSLockScheduler();
			__OSThreadSwitchToNext();
			__OSUnlockScheduler();
		}
	}

	void OSSchedulerCoreEmulationThread(void* _assignedCoreIndex)
	{
		SetThreadName(fmt::format("OSSchedulerThread[core={}]", (uintptr_t)_assignedCoreIndex).c_str());
		t_assignedCoreIndex = (sint32)(uintptr_t)_assignedCoreIndex;
		_mm_setcsr(_mm_getcsr() | 0x8000); // flush denormals to zero
		t_schedulerFiber = Fiber::PrepareCurrentThread();
		
		// create scheduler idle fiber and switch to it
		g_idleLoopFiber[t_assignedCoreIndex] = new Fiber(__OSThreadCoreIdle, nullptr, nullptr);
		cemu_assert_debug(ppcInterpreterCurrentInstance == nullptr);
		__OSLockScheduler();
		Fiber::Switch(*g_idleLoopFiber[t_assignedCoreIndex]);
		// returned from scheduler loop, exit thread
		cemu_assert_debug(!__OSHasSchedulerLock());
	}

	std::vector<std::thread::native_handle_type> g_schedulerThreadHandles;

	std::vector<std::thread::native_handle_type>& OSGetSchedulerThreads()
	{
		return g_schedulerThreadHandles;
	}

	// starts PPC core emulation
	void OSSchedulerBegin(sint32 numCPUEmulationThreads)
	{
		std::unique_lock _lock(sSchedulerStateMtx);
		if (sSchedulerActive.exchange(true))
			return;
		cemu_assert_debug(numCPUEmulationThreads == 1 || numCPUEmulationThreads == 3);
		g_isMulticoreMode = numCPUEmulationThreads > 1;
		if (numCPUEmulationThreads == 1)
			sSchedulerThreads.emplace_back(OSSchedulerCoreEmulationThread, (void*)0);
		else if (numCPUEmulationThreads == 3)
		{
			for (size_t i = 0; i < 3; i++)
				sSchedulerThreads.emplace_back(OSSchedulerCoreEmulationThread, (void*)i);
		}
		else
			cemu_assert_debug(false);
		for (auto& it : sSchedulerThreads)
			g_schedulerThreadHandles.emplace_back(it.native_handle());
	}

	// shuts down all scheduler host threads and deletes all fibers and their state
	void OSSchedulerEnd()
	{
		std::unique_lock _lock(sSchedulerStateMtx);
		sSchedulerActive.store(false);
		for (size_t i = 0; i < Espresso::CORE_COUNT; i++)
			g_coreRunQueueThreadCount[i].increment(); // make sure to wake up cores if they are paused and waiting for runnable threads
		// wait for threads to stop execution
		for (auto& threadItr : sSchedulerThreads)
			threadItr.join();
		sSchedulerThreads.clear();
		g_schedulerThreadHandles.clear();
		// clean up all fibers
		for (auto& it : g_idleLoopFiber)
		{
			delete it;
			it = nullptr;
		}
		for (auto& it : s_threadToFiber)
		{
			OSHostThread* hostThread = it.second;
			delete hostThread;
		}
		s_threadToFiber.clear();
	}

	SysAllocator<OSThread_t, PPC_CORE_COUNT> s_defaultThreads;
	SysAllocator<uint8, PPC_CORE_COUNT * 1024 * 1024> s_stack;

	OSThread_t* OSGetDefaultThread(sint32 coreIndex)
	{
		if (coreIndex < 0 || coreIndex >= PPC_CORE_COUNT)
			return nullptr;
		return s_defaultThreads.GetPtr() + coreIndex;
	}

	void* OSGetDefaultThreadStack(sint32 coreIndex, uint32& size)
	{
		if (coreIndex < 0 || coreIndex >= PPC_CORE_COUNT)
			return nullptr;
		size = 1024 * 1024;
		return s_stack.GetPtr() + coreIndex * size;
	}

	void OSInitThreadQueue(OSThreadQueue* threadQueue)
	{
		threadQueue->head = nullptr;
		threadQueue->tail = nullptr;
		threadQueue->userData = nullptr;
		threadQueue->ukn0C = 0;
	}
	
	void OSInitThreadQueueEx(OSThreadQueue* threadQueue, void* userData)
	{
		threadQueue->head = nullptr;
		threadQueue->tail = nullptr;
		threadQueue->userData = userData;
		threadQueue->ukn0C = 0;
	}
	
	/* Thread terminator threads (for calling thread deallocators) */
	struct TerminatorThread 
	{
		struct DeallocatorQueueEntry
		{
			DeallocatorQueueEntry() {};
			DeallocatorQueueEntry(OSThread_t* thread, MEMPTR<void> stack, MEMPTR<void> deallocatorFunc) : thread(thread), stack(stack), deallocatorFunc(deallocatorFunc) {};

			OSThread_t* thread{};
			MEMPTR<void> stack;
			MEMPTR<void> deallocatorFunc;
		};

		SysAllocator<OSThread_t> terminatorThread;
		SysAllocator<uint8, 16 * 1024> threadStack;
		SysAllocator<char, 64> threadName;
		SysAllocator<OSSemaphore> semaphoreQueuedDeallocators;
		ConcurrentQueue<DeallocatorQueueEntry> queueDeallocators;
	};

	TerminatorThread s_terminatorThreads[PPC_CORE_COUNT];

	void __OSTerminatorThreadFunc(PPCInterpreter_t* hCPU)
	{
		uint32 coreIndex = OSGetCoreId();
		while (OSWaitSemaphore(s_terminatorThreads[coreIndex].semaphoreQueuedDeallocators.GetPtr()))
		{
			TerminatorThread::DeallocatorQueueEntry queueEntry;
			s_terminatorThreads[coreIndex].queueDeallocators.pop(queueEntry);
			PPCCoreCallback(queueEntry.deallocatorFunc, queueEntry.thread, queueEntry.stack);
		}
		osLib_returnFromFunction(hCPU, 0);
	}

	// queue thread deallocation to run after current thread finishes
	// the termination threads run at a higher priority on the same threads
	void __OSQueueThreadDeallocation(OSThread_t* thread)
	{
		uint32 coreIndex = OSGetCoreId();
		TerminatorThread::DeallocatorQueueEntry queueEntry(thread, memory_getPointerFromVirtualOffset(_swapEndianU32(thread->stackEnd)), thread->deallocatorFunc);
		s_terminatorThreads[coreIndex].queueDeallocators.push(queueEntry);
		OSSignalSemaphoreInternal(s_terminatorThreads[coreIndex].semaphoreQueuedDeallocators.GetPtr(), false); // do not reschedule here! Current thread must not be interrupted otherwise deallocator will run too early
	}

	void __OSInitTerminatorThreads()
	{
		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
		{
			OSInitSemaphore(s_terminatorThreads[i].semaphoreQueuedDeallocators.GetPtr(), 0);
			sprintf(s_terminatorThreads[i].threadName.GetPtr(), "{SYS Thread Terminator Core %d}", i);
			OSCreateThreadType(s_terminatorThreads[i].terminatorThread.GetPtr(), PPCInterpreter_makeCallableExportDepr(__OSTerminatorThreadFunc), 0, nullptr, (uint8*)s_terminatorThreads[i].threadStack.GetPtr() + s_terminatorThreads[i].threadStack.GetByteSize(), (sint32)s_terminatorThreads[i].threadStack.GetByteSize(), 0, 1 << i, OSThread_t::THREAD_TYPE::TYPE_IO);
			OSSetThreadName(s_terminatorThreads[i].terminatorThread.GetPtr(), s_terminatorThreads[i].threadName.GetPtr());
			OSResumeThread(s_terminatorThreads[i].terminatorThread.GetPtr());
		}
	}

	SysAllocator<char, 64> _defaultThreadName[PPC_CORE_COUNT];

	void __OSInitDefaultThreads()
	{
		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
		{
			sprintf(_defaultThreadName[i].GetPtr(), "Default Core %d", i);
			OSThread_t* coreDefaultThread = coreinit::OSGetDefaultThread(i);
			uint32 stackSize;
			void* stackBase = coreinit::OSGetDefaultThreadStack(i, stackSize);
			coreinit::OSCreateThreadType(coreDefaultThread, MPTR_NULL, 0, nullptr, (uint8*)stackBase + stackSize, stackSize, 16, 1 << i, OSThread_t::THREAD_TYPE::TYPE_APP);
			coreinit::OSSetThreadName(coreDefaultThread, _defaultThreadName[i].GetPtr());
		}
	}

	void InitializeThread()
	{
		cafeExportRegister("coreinit", OSCreateThreadType, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSCreateThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSExitThread, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSGetCurrentThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSetThreadSpecific, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSGetThreadSpecific, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSetThreadName, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSGetThreadName, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSRunThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSDetachThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSJoinThread, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSResumeThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSContinueThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSuspendThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSleepThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSWakeupThread, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSYieldThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSleepTicks, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSSetThreadDeallocator, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSetThreadCleanupCallback, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSetThreadPriority, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSGetThreadPriority, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSSetThreadAffinity, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSGetThreadAffinity, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSIsThreadTerminated, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSIsThreadSuspended, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSCancelThread, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSTestThreadCancel, LogType::CoreinitThread);

		cafeExportRegister("coreinit", OSGetDefaultThread, LogType::CoreinitThread);

		// OSThreadQueue
		cafeExportRegister("coreinit", OSInitThreadQueue, LogType::CoreinitThread);
		cafeExportRegister("coreinit", OSInitThreadQueueEx, LogType::CoreinitThread);

		OSInitThreadQueue(g_activeThreadQueue.GetPtr());
		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
			OSInitThreadQueue(g_coreRunQueue.GetPtr() + i);

		for (sint32 i = 0; i < PPC_CORE_COUNT; i++)
			__currentCoreThread[i] = nullptr;

		__OSInitDefaultThreads();
		__OSInitTerminatorThreads();
	}
}

void coreinit_suspendThread(OSThread_t* OSThreadBE, sint32 count)
{
	// for legacy source
	OSThreadBE->suspendCounter += count;
}

void coreinit_resumeThread(OSThread_t* OSThreadBE, sint32 count)
{
	__OSLockScheduler();
	coreinit::__OSResumeThreadInternal(OSThreadBE, count);
	__OSUnlockScheduler();
}

MPTR coreinitThread_getCurrentThreadMPTRDepr(PPCInterpreter_t* hCPU)
{
	return memory_getVirtualOffsetFromPointer(coreinit::__currentCoreThread[PPCInterpreter_getCoreIndex(hCPU)]);
}

OSThread_t* coreinitThread_getCurrentThreadDepr(PPCInterpreter_t* hCPU)
{
	return coreinit::__currentCoreThread[PPCInterpreter_getCoreIndex(hCPU)];
}