rpcs3/Utilities/Thread.cpp

#include "stdafx.h"
#include "Log.h"
#include "rpcs3/Ini.h"
#include "Emu/System.h"
#include "Emu/CPU/CPUThread.h"
#include "Emu/SysCalls/SysCalls.h"
#include "Thread.h"

#ifdef _WIN32
#include <windows.h>
#else
#ifdef __APPLE__
#define _XOPEN_SOURCE
#define __USE_GNU
#endif
#include <signal.h>
#include <ucontext.h>
#endif

void SetCurrentThreadDebugName(const char* threadName)
{
#if defined(_MSC_VER) // this is VS-specific way to set thread names for the debugger

	#pragma pack(push,8)

	struct THREADNAME_INFO
	{
		DWORD dwType;
		LPCSTR szName;
		DWORD dwThreadID;
		DWORD dwFlags;
	} info;

	#pragma pack(pop)

	info.dwType = 0x1000;
	info.szName = threadName;
	info.dwThreadID = -1;
	info.dwFlags = 0;

	__try
	{
		RaiseException(0x406D1388, 0, sizeof(info) / sizeof(ULONG_PTR), (ULONG_PTR*)&info);
	}
	__except (EXCEPTION_EXECUTE_HANDLER)
	{
	}

#endif
}

enum x64_reg_t : u32
{
	X64R_RAX,
	X64R_RCX,
	X64R_RDX,
	X64R_RBX,
	X64R_RSP,
	X64R_RBP,
	X64R_RSI,
	X64R_RDI,
	X64R_R8,
	X64R_R9,
	X64R_R10,
	X64R_R11,
	X64R_R12,
	X64R_R13,
	X64R_R14,
	X64R_R15,

	X64R_XMM0,
	X64R_XMM1,
	X64R_XMM2,
	X64R_XMM3,
	X64R_XMM4,
	X64R_XMM5,
	X64R_XMM6,
	X64R_XMM7,
	X64R_XMM8,
	X64R_XMM9,
	X64R_XMM10,
	X64R_XMM11,
	X64R_XMM12,
	X64R_XMM13,
	X64R_XMM14,
	X64R_XMM15,
	
	X64R_AL,
	X64R_CL,
	X64R_DL,
	X64R_BL,
	X64R_AH,
	X64R_CH,
	X64R_DH,
	X64R_BH,
	
	X64_NOT_SET,
	X64_IMM8,
	X64_IMM16,
	X64_IMM32,

	X64R = X64R_RAX,
	X64R_XMM = X64R_XMM0,
	X64R_LH = X64R_AL,
	X64R_ECX = X64R_CL,
};

enum x64_op_t : u32
{
	X64OP_NOP,
	X64OP_LOAD, // obtain and put the value into x64 register (from Memory.ReadMMIO32, for example)
	X64OP_STORE, // take the value from x64 register or an immediate and use it (pass in Memory.WriteMMIO32, for example)
	// example: add eax,[rax] -> X64OP_LOAD_ADD (add the value to x64 register)
	// example: add [rax],eax -> X64OP_LOAD_ADD_STORE (this will probably never happen for MMIO registers)

	X64OP_MOVS,
	X64OP_XCHG,
	X64OP_CMPXCHG,
};

void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, u32& out_size, u32& out_length)
{
	// simple analysis of x64 code allows to reinterpret MOV or other instructions in any desired way
	out_length = 0;

	u8 rex = 0, pg2 = 0;

	bool oso = false, lock = false, repne = false, repe = false;

	enum : u8
	{
		LOCK  = 0xf0,
		REPNE = 0xf2,
		REPE  = 0xf3,
	};

	// check prefixes:
	for (;; code++, out_length++)
	{
		switch (const u8 prefix = *code)
		{
		case LOCK: // group 1
		{
			if (lock)
			{
				LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): LOCK prefix found twice", (size_t)code - out_length);
			}
			
			lock = true;
			continue;
		}
		case REPNE: // group 1
		{
			if (repne)
			{
				LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): REPNE/REPNZ prefix found twice", (size_t)code - out_length);
			}
			
			repne = true;
			continue;
		}
		case REPE: // group 1
		{
			if (repe)
			{
				LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): REP/REPE/REPZ prefix found twice", (size_t)code - out_length);
			}
			
			repe = true;
			continue;
		}

		case 0x2e: // group 2
		case 0x36:
		case 0x3e:
		case 0x26:
		case 0x64:
		case 0x65:
		{
			if (pg2)
			{
				LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): 0x%02x (group 2 prefix) found after 0x%02x", (size_t)code - out_length, prefix, pg2);
			}
			else
			{
				pg2 = prefix; // probably, segment register
			}
			continue;
		}

		case 0x66: // group 3
		{
			if (oso)
			{
				LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): operand-size override prefix found twice", (size_t)code - out_length);
			}
			
			oso = true;
			continue;
		}

		case 0x67: // group 4
		{
			LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): address-size override prefix found", (size_t)code - out_length, prefix);
			out_op = X64OP_NOP;
			out_reg = X64_NOT_SET;
			out_size = 0;
			out_length = 0;
			return;
		}

		default:
		{
			if ((prefix & 0xf0) == 0x40) // check REX prefix
			{
				if (rex)
				{
					LOG_ERROR(GENERAL, "decode_x64_reg_op(%016llxh): 0x%02x (REX prefix) found after 0x%02x", (size_t)code - out_length, prefix, rex);
				}
				else
				{
					rex = prefix;
				}
				continue;
			}
		}
		}

		break;
	}

	auto get_modRM_reg = [](const u8* code, const u8 rex) -> x64_reg_t
	{
		return (x64_reg_t)(((*code & 0x38) >> 3 | (/* check REX.R bit */ rex & 4 ? 8 : 0)) + X64R);
	};

	auto get_modRM_reg_xmm = [](const u8* code, const u8 rex) -> x64_reg_t
	{
		return (x64_reg_t)(((*code & 0x38) >> 3 | (/* check REX.R bit */ rex & 4 ? 8 : 0)) + X64R_XMM);
	};

	auto get_modRM_reg_lh = [](const u8* code) -> x64_reg_t
	{
		return (x64_reg_t)(((*code & 0x38) >> 3) + X64R_LH);
	};

	auto get_op_size = [](const u8 rex, const bool oso) -> u32
	{
		return rex & 8 ? 8 : (oso ? 2 : 4);
	};

	auto get_modRM_size = [](const u8* code) -> u32
	{
		switch (*code >> 6) // check Mod
		{
		case 0: return (*code & 0x07) == 4 ? 2 : 1; // check SIB
		case 1: return (*code & 0x07) == 4 ? 3 : 2; // check SIB (disp8)
		case 2: return (*code & 0x07) == 4 ? 6 : 5; // check SIB (disp32)
		default: return 1;
		}
	};

	const u8 op1 = (out_length++, *code++), op2 = code[0], op3 = code[1];

	switch (op1)
	{
	case 0x0f:
	{
		out_length++, code++;

		switch (op2)
		{
		case 0x7f:
		{
			if (repe && !oso) // MOVDQU xmm/m, xmm
			{
				out_op = X64OP_STORE;
				out_reg = get_modRM_reg_xmm(code, rex);
				out_size = 16;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		case 0xb0:
		{
			if (!oso) // CMPXCHG r8/m8, r8
			{
				out_op = X64OP_CMPXCHG;
				out_reg = rex & 8 ? get_modRM_reg(code, rex) : get_modRM_reg_lh(code);
				out_size = 1;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		case 0xb1:
		{
			if (true) // CMPXCHG r/m, r (16, 32, 64)
			{
				out_op = X64OP_CMPXCHG;
				out_reg = get_modRM_reg(code, rex);
				out_size = get_op_size(rex, oso);
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		}

		break;
	}
	case 0x86:
	{
		if (!oso) // XCHG r8/m8, r8
		{
			out_op = X64OP_XCHG;
			out_reg = rex & 8 ? get_modRM_reg(code, rex) : get_modRM_reg_lh(code);
			out_size = 1;
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x87:
	{
		if (true) // XCHG r/m, r (16, 32, 64)
		{
			out_op = X64OP_XCHG;
			out_reg = get_modRM_reg(code, rex);
			out_size = get_op_size(rex, oso);
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x88:
	{
		if (!lock && !oso) // MOV r8/m8, r8
		{
			out_op = X64OP_STORE;
			out_reg = rex & 8 ? get_modRM_reg(code, rex) : get_modRM_reg_lh(code);
			out_size = 1;
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x89:
	{
		if (!lock) // MOV r/m, r (16, 32, 64)
		{
			out_op = X64OP_STORE;
			out_reg = get_modRM_reg(code, rex);
			out_size = get_op_size(rex, oso);
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x8a:
	{
		if (!lock && !oso) // MOV r8, r8/m8
		{
			out_op = X64OP_LOAD;
			out_reg = rex & 8 ? get_modRM_reg(code, rex) : get_modRM_reg_lh(code);
			out_size = 1;
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x8b:
	{
		if (!lock) // MOV r, r/m (16, 32, 64)
		{
			out_op = X64OP_LOAD;
			out_reg = get_modRM_reg(code, rex);
			out_size = get_op_size(rex, oso);
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0xa4:
	{
		if (!oso && !lock && !repe && !rex) // MOVS
		{
			out_op = X64OP_MOVS;
			out_reg = X64_NOT_SET;
			out_size = 1;
			return;
		}
		if (!oso && !lock && repe) // REP MOVS
		{
			out_op = X64OP_MOVS;
			out_reg = rex & 8 ? X64R_RCX : X64R_ECX;
			out_size = 1;
			return;
		}
		break;
	}
	case 0xc6:
	{
		if (!lock && !oso && get_modRM_reg(code, 0) == X64R_RAX) // MOV r8/m8, imm8
		{
			out_op = X64OP_STORE;
			out_reg = X64_IMM8;
			out_size = 1;
			out_length += get_modRM_size(code) + 1;
			return;
		}
		break;
	}
	case 0xc7:
	{
		if (!lock && get_modRM_reg(code, 0) == X64R_RAX) // MOV r/m, imm16/imm32 (16, 32, 64)
		{
			out_op = X64OP_STORE;
			out_reg = oso ? X64_IMM16 : X64_IMM32;
			out_size = get_op_size(rex, oso);
			out_length += get_modRM_size(code) + (oso ? 2 : 4);
			return;
		}
		break;
	}
	}

	LOG_WARNING(GENERAL, "decode_x64_reg_op(%016llxh): unsupported opcode found (%llX%llX)", (size_t)code - out_length, *(be_t<u64>*)(code - out_length), *(be_t<u64>*)(code - out_length + 8));
	out_op = X64OP_NOP;
	out_reg = X64_NOT_SET;
	out_size = 0;
	out_length = 0;
}

#ifdef _WIN32

typedef CONTEXT x64_context;

#define RIP 16
#define X64REG(context, reg) (&(&context->Rax)[reg])

#else

typedef ucontext_t x64_context;
#define RIP 16

#ifdef __APPLE__

#define X64REG(context, reg) (darwin_x64reg(context, reg))

uint64_t* darwin_x64reg(x64_context *context, int reg)
{
	auto *state = &context->uc_mcontext->__ss;
	switch(reg)
	{
	case 0: // RAX
		return &state->__rax;
	case 1: // RCX
		return &state->__rcx;
	case 2: // RDX
		return &state->__rdx;
	case 3: // RBX
		return &state->__rbx;
	case 4: // RSP
		return &state->__rsp;
	case 5: // RBP
		return &state->__rbp;
	case 6: // RSI
		return &state->__rsi;
	case 7: // RDI
		return &state->__rdi;
	case 8: // R8
		return &state->__r8;
	case 9: // R9
		return &state->__r9;
	case 10: // R10
		return &state->__r10;
	case 11: // R11
		return &state->__r11;
	case 12: // R12
		return &state->__r12;
	case 13: // R13
		return &state->__r13;
	case 14: // R14
		return &state->__r14;
	case 15: // R15
		return &state->__r15;
	case 16: // RIP
		return &state->__rip;
	default: // FAIL
		assert(0);
	}
}

#else

typedef decltype(REG_RIP) reg_table_t;

static const reg_table_t reg_table[17] =
{
	REG_RAX, REG_RCX, REG_RDX, REG_RBX, REG_RSP, REG_RBP, REG_RSI, REG_RDI,
	REG_R8, REG_R9, REG_R10, REG_R11, REG_R12, REG_R13, REG_R14, REG_R15, REG_RIP
};

#define X64REG(context, reg) (&context->uc_mcontext.gregs[reg_table[reg]])

#endif // __APPLE__

#endif

bool handle_access_violation(const u32 addr, bool is_writing, x64_context* context)
{
	x64_op_t op;
	x64_reg_t reg;
	u32 d_size;
	u32 i_size;

	// decode single x64 instruction that causes memory access
	decode_x64_reg_op((const u8*)(*X64REG(context, RIP)), op, reg, d_size, i_size);

	// check if address is RawSPU MMIO register
	if (addr - RAW_SPU_BASE_ADDR < (6 * RAW_SPU_OFFSET) && (addr % RAW_SPU_OFFSET) >= RAW_SPU_PROB_OFFSET)
	{
		if (d_size != 4 || !i_size)
		{
			LOG_ERROR(GENERAL, "Invalid instruction (op=%d, reg=%d, d_size=0x%x, i_size=0x%x)", op, reg, d_size, i_size);
			return false;
		}

		// get x64 reg value (for store operations)
		u64 reg_value;
		if (reg - X64R < 16)
		{
			// load the value from x64 register
			reg_value = (u32)*X64REG(context, reg - X64R);
		}
		else if (reg == X64_IMM32)
		{
			// load the immediate value (assuming it's at the end of the instruction)
			reg_value = *(u32*)(*X64REG(context, RIP) + i_size - 4);
		}
		else
		{
			LOG_ERROR(GENERAL, "Invalid source (reg=%d)", reg);
			return false;
		}

		bool save_reg = false;

		switch (op)
		{
		case X64OP_LOAD:
		{
			reg_value = re32(Memory.ReadMMIO32(addr));
			save_reg = true;
			break;
		}
		case X64OP_STORE:
		{
			Memory.WriteMMIO32(addr, re32((u32)reg_value));
			break;
		}
		default:
		{
			LOG_ERROR(GENERAL, "Invalid operation (op=%d)", op);
			return false;
		}
		}

		// save x64 reg value (for load operations)
		if (save_reg)
		{
			if (reg - X64R < 16)
			{
				// store the value into x64 register
				*X64REG(context, reg - X64R) = (u32)reg_value;
			}
			else
			{
				LOG_ERROR(GENERAL, "Invalid destination (reg=%d, reg_value=0x%llx)", reg, reg_value);
				return false;
			}
		}

		// skip decoded instruction
		*X64REG(context, RIP) += i_size;
		return true;
	}

	// check if fault is caused by reservation
	if (vm::reservation_query(addr, is_writing))
	{
		return true;
	}

	// TODO: allow recovering from a page fault as a feature of PS3 virtual memory
	return false;
}

#ifdef _WIN32

void _se_translator(unsigned int u, EXCEPTION_POINTERS* pExp)
{
	const u64 addr64 = (u64)pExp->ExceptionRecord->ExceptionInformation[1] - (u64)vm::g_base_addr;
	const bool is_writing = pExp->ExceptionRecord->ExceptionInformation[0] != 0;

	if (u == EXCEPTION_ACCESS_VIOLATION && (u32)addr64 == addr64)
	{
		throw fmt::format("Access violation %s location 0x%llx", is_writing ? "writing" : "reading", addr64);
	}
}

const PVOID exception_handler = (atexit([]{ RemoveVectoredExceptionHandler(exception_handler); }), AddVectoredExceptionHandler(1, [](PEXCEPTION_POINTERS pExp) -> LONG
{
	const u64 addr64 = (u64)pExp->ExceptionRecord->ExceptionInformation[1] - (u64)vm::g_base_addr;
	const bool is_writing = pExp->ExceptionRecord->ExceptionInformation[0] != 0;

	if (pExp->ExceptionRecord->ExceptionCode == EXCEPTION_ACCESS_VIOLATION &&
		(u32)addr64 == addr64 &&
		GetCurrentNamedThread() &&
		handle_access_violation((u32)addr64, is_writing, pExp->ContextRecord))
	{
		return EXCEPTION_CONTINUE_EXECUTION;
	}
	else
	{
		return EXCEPTION_CONTINUE_SEARCH;
	}
}));

#else

void signal_handler(int sig, siginfo_t* info, void* uct)
{
	const u64 addr64 = (u64)info->si_addr - (u64)vm::g_base_addr;

#ifdef __APPLE__
	const bool is_writing = ((ucontext_t*)uct)->uc_mcontext->__es.__err & 0x2;
#else
	const bool is_writing = ((ucontext_t*)uct)->uc_mcontext.gregs[REG_ERR] & 0x2;
#endif

	if ((u32)addr64 == addr64 && GetCurrentNamedThread())
	{
		if (handle_access_violation((u32)addr64, is_writing, (ucontext_t*)uct))
		{
			return; // proceed execution
		}

		// TODO: this may be wrong
		throw fmt::format("Access violation %s location 0x%llx", is_writing ? "writing" : "reading", addr64);
	}

	// else some fatal error
	exit(EXIT_FAILURE);
}

const int sigaction_result = []() -> int
{
	struct sigaction sa;

	sa.sa_flags = SA_SIGINFO;
	sigemptyset(&sa.sa_mask);
	sa.sa_sigaction = signal_handler;
	return sigaction(SIGSEGV, &sa, NULL);
}();

#endif

thread_local NamedThreadBase* g_tls_this_thread = nullptr;
std::atomic<u32> g_thread_count(0);

NamedThreadBase* GetCurrentNamedThread()
{
	return g_tls_this_thread;
}

void SetCurrentNamedThread(NamedThreadBase* value)
{
	const auto old_value = g_tls_this_thread;

	if (old_value == value)
	{
		return;
	}

	if (old_value)
	{
		vm::reservation_free();
	}

	if (value && value->m_tls_assigned.exchange(true))
	{
		LOG_ERROR(GENERAL, "Thread '%s' was already assigned to g_tls_this_thread of another thread", value->GetThreadName());
		g_tls_this_thread = nullptr;
	}
	else
	{
		g_tls_this_thread = value;
	}

	if (old_value)
	{
		old_value->m_tls_assigned = false;
	}
}

std::string NamedThreadBase::GetThreadName() const
{
	return m_name;
}

void NamedThreadBase::SetThreadName(const std::string& name)
{
	m_name = name;
}

void NamedThreadBase::WaitForAnySignal(u64 time) // wait for Notify() signal or sleep
{
	std::unique_lock<std::mutex> lock(m_signal_mtx);
	m_signal_cv.wait_for(lock, std::chrono::milliseconds(time));
}

void NamedThreadBase::Notify() // wake up waiting thread or nothing
{
	m_signal_cv.notify_one();
}

ThreadBase::ThreadBase(const std::string& name)
	: NamedThreadBase(name)
	, m_executor(nullptr)
	, m_destroy(false)
	, m_alive(false)
{
}

ThreadBase::~ThreadBase()
{
	if(IsAlive())
		Stop(false);

	delete m_executor;
	m_executor = nullptr;
}

void ThreadBase::Start()
{
	if(m_executor) Stop();

	std::lock_guard<std::mutex> lock(m_main_mutex);

	m_destroy = false;
	m_alive = true;

	m_executor = new std::thread([this]()
	{
		SetCurrentThreadDebugName(GetThreadName().c_str());

#ifdef _WIN32
		auto old_se_translator = _set_se_translator(_se_translator);
		if (!exception_handler)
		{
			LOG_ERROR(GENERAL, "exception_handler not set");
			return;
		}
#else
		if (sigaction_result == -1)
		{
			printf("sigaction() failed");
			exit(EXIT_FAILURE);
		}
#endif

		SetCurrentNamedThread(this);
		g_thread_count++;

		try
		{
			Task();
		}
		catch (const char* e)
		{
			LOG_ERROR(GENERAL, "%s: %s", GetThreadName().c_str(), e);
		}
		catch (const std::string& e)
		{
			LOG_ERROR(GENERAL, "%s: %s", GetThreadName().c_str(), e.c_str());
		}

		m_alive = false;
		SetCurrentNamedThread(nullptr);
		g_thread_count--;

#ifdef _WIN32
		_set_se_translator(old_se_translator);
#endif
	});
}

void ThreadBase::Stop(bool wait, bool send_destroy)
{
	std::lock_guard<std::mutex> lock(m_main_mutex);

	if (send_destroy)
		m_destroy = true;

	if(!m_executor)
		return;

	if(wait && m_executor->joinable() && m_alive)
	{
		m_executor->join();
	}
	else
	{
		m_executor->detach();
	}

	delete m_executor;
	m_executor = nullptr;
}

bool ThreadBase::Join() const
{
	std::lock_guard<std::mutex> lock(m_main_mutex);
	if(m_executor->joinable() && m_alive && m_executor != nullptr)
	{
		m_executor->join();
		return true;
	}

	return false;
}

bool ThreadBase::IsAlive() const
{
	std::lock_guard<std::mutex> lock(m_main_mutex);
	return m_alive;
}

bool ThreadBase::TestDestroy() const
{
	return m_destroy;
}

thread_t::thread_t(const std::string& name, bool autojoin, std::function<void()> func)
	: m_name(name)
	, m_state(TS_NON_EXISTENT)
	, m_autojoin(autojoin)
{
	start(func);
}

thread_t::thread_t(const std::string& name, std::function<void()> func)
	: m_name(name)
	, m_state(TS_NON_EXISTENT)
	, m_autojoin(false)
{
	start(func);
}

thread_t::thread_t(const std::string& name)
	: m_name(name)
	, m_state(TS_NON_EXISTENT)
	, m_autojoin(false)
{
}

thread_t::thread_t()
	: m_state(TS_NON_EXISTENT)
	, m_autojoin(false)
{
}

void thread_t::set_name(const std::string& name)
{
	m_name = name;
}

thread_t::~thread_t()
{
	if (m_state == TS_JOINABLE)
	{
		if (m_autojoin)
		{
			m_thr.join();
		}
		else
		{
			m_thr.detach();
		}
	}
}

void thread_t::start(std::function<void()> func)
{
	if (m_state.exchange(TS_NON_EXISTENT) == TS_JOINABLE)
	{
		m_thr.join(); // forcefully join previously created thread
	}

	std::string name = m_name;
	m_thr = std::thread([func, name]()
	{
		SetCurrentThreadDebugName(name.c_str());

#ifdef _WIN32
		auto old_se_translator = _set_se_translator(_se_translator);
#endif

		NamedThreadBase info(name);
		SetCurrentNamedThread(&info);
		g_thread_count++;

		if (Ini.HLELogging.GetValue())
		{
			LOG_NOTICE(HLE, name + " started");
		}

		try
		{
			func();
		}
		catch (const char* e)
		{
			LOG_ERROR(GENERAL, "%s: %s", name.c_str(), e);
		}
		catch (const std::string& e)
		{
			LOG_ERROR(GENERAL, "%s: %s", name.c_str(), e.c_str());
		}

		if (Emu.IsStopped())
		{
			LOG_NOTICE(HLE, name + " aborted");
		}
		else if (Ini.HLELogging.GetValue())
		{
			LOG_NOTICE(HLE, name + " ended");
		}

		SetCurrentNamedThread(nullptr);
		g_thread_count--;

#ifdef _WIN32
		_set_se_translator(old_se_translator);
#endif
	});

	if (m_state.exchange(TS_JOINABLE) == TS_JOINABLE)
	{
		assert(!"thread_t::start() failed"); // probably started from another thread
	}
}

void thread_t::detach()
{
	if (m_state.exchange(TS_NON_EXISTENT) == TS_JOINABLE)
	{
		m_thr.detach();
	}
	else
	{
		assert(!"thread_t::detach() failed"); // probably joined or detached
	}
}

void thread_t::join()
{
	if (m_state.exchange(TS_NON_EXISTENT) == TS_JOINABLE)
	{
		m_thr.join();
	}
	else
	{
		assert(!"thread_t::join() failed"); // probably joined or detached
	}
}

bool thread_t::joinable() const
{
	//return m_thr.joinable();
	return m_state == TS_JOINABLE;
}

bool waiter_map_t::is_stopped(u64 signal_id)
{
	if (Emu.IsStopped())
	{
		LOG_WARNING(Log::HLE, "%s: waiter_op() aborted (signal_id=0x%llx)", m_name.c_str(), signal_id);
		return true;
	}
	return false;
}

void waiter_map_t::waiter_reg_t::init()
{
	if (!thread)
	{
		thread = GetCurrentNamedThread();

		std::lock_guard<std::mutex> lock(map.m_mutex);

		// add waiter
		map.m_waiters.push_back({ signal_id, thread });
	}
}

waiter_map_t::waiter_reg_t::~waiter_reg_t()
{
	if (thread)
	{
		std::lock_guard<std::mutex> lock(map.m_mutex);

		// remove waiter
		for (s64 i = map.m_waiters.size() - 1; i >= 0; i--)
		{
			if (map.m_waiters[i].signal_id == signal_id && map.m_waiters[i].thread == thread)
			{
				map.m_waiters.erase(map.m_waiters.begin() + i);
				return;
			}
		}

		LOG_ERROR(HLE, "%s(): waiter not found (signal_id=0x%llx, map='%s')", __FUNCTION__, signal_id, map.m_name.c_str());
		Emu.Pause();
	}
}

void waiter_map_t::notify(u64 signal_id)
{
	if (m_waiters.size())
	{
		std::lock_guard<std::mutex> lock(m_mutex);

		// find waiter and signal
		for (auto& v : m_waiters)
		{
			if (v.signal_id == signal_id)
			{
				v.thread->Notify();
			}
		}
	}
}

const std::function<bool()> SQUEUE_ALWAYS_EXIT = [](){ return true; };
const std::function<bool()> SQUEUE_NEVER_EXIT = [](){ return false; };

bool squeue_test_exit()
{
	return Emu.IsStopped();
}