rpcs3/Utilities/Thread.cpp

#include "stdafx.h"
#include "Emu/System.h"
#include "Emu/Cell/SPUThread.h"
#include "Emu/Cell/PPUThread.h"
#include "Emu/Cell/lv2/sys_mmapper.h"
#include "Emu/Cell/lv2/sys_event.h"
#include "Emu/Cell/lv2/sys_process.h"
#include "Emu/RSX/RSXThread.h"
#include "Thread.h"
#include "Utilities/JIT.h"
#include <thread>
#include <cfenv>

#ifdef _WIN32
#include <Windows.h>
#include <Psapi.h>
#include <process.h>
#include <sysinfoapi.h>

#include "util/dyn_lib.hpp"

DYNAMIC_IMPORT_RENAME("Kernel32.dll", SetThreadDescriptionImport, "SetThreadDescription", HRESULT(HANDLE hThread, PCWSTR lpThreadDescription));

#else
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#ifdef __APPLE__
#define _XOPEN_SOURCE
#define __USE_GNU
#include <mach/thread_act.h>
#include <mach/thread_policy.h>
#endif
#if defined(__DragonFly__) || defined(__FreeBSD__) || defined(__OpenBSD__)
#include <pthread_np.h>
#define cpu_set_t cpuset_t
#endif
#include <errno.h>
#include <signal.h>
#ifndef __OpenBSD__
#include <ucontext.h>
#endif
#include <pthread.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <time.h>
#endif
#ifdef __linux__
#include <sys/syscall.h>
#include <sys/timerfd.h>
#include <unistd.h>
#endif

#if defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
# include <sys/sysctl.h>
# include <unistd.h>
# if defined(__DragonFly__) || defined(__FreeBSD__)
#  include <sys/user.h>
# endif
# if defined(__OpenBSD__)
#  include <sys/param.h>
#  include <sys/proc.h>
# endif

# if defined(__NetBSD__)
#  undef KERN_PROC
#  define KERN_PROC KERN_PROC2
#  define kinfo_proc kinfo_proc2
# endif

# if defined(__APPLE__)
#  define KP_FLAGS kp_proc.p_flag
# elif defined(__DragonFly__)
#  define KP_FLAGS kp_flags
# elif defined(__FreeBSD__)
#  define KP_FLAGS ki_flag
# elif defined(__NetBSD__)
#  define KP_FLAGS p_flag
# elif defined(__OpenBSD__)
#  define KP_FLAGS p_psflags
#  define P_TRACED PS_TRACED
# endif
#endif

#include "util/vm.hpp"
#include "util/logs.hpp"
#include "util/asm.hpp"
#include "util/v128.hpp"
#include "util/simd.hpp"
#include "util/sysinfo.hpp"
#include "Emu/Memory/vm_locking.h"

LOG_CHANNEL(sig_log, "SIG");
LOG_CHANNEL(sys_log, "SYS");
LOG_CHANNEL(vm_log, "VM");

thread_local u64 g_tls_fault_all = 0;
thread_local u64 g_tls_fault_rsx = 0;
thread_local u64 g_tls_fault_spu = 0;
thread_local u64 g_tls_wait_time = 0;
thread_local u64 g_tls_wait_fail = 0;
thread_local bool g_tls_access_violation_recovered = false;
extern thread_local std::string(*g_tls_log_prefix)();

// Report error and call std::abort(), defined in main.cpp
[[noreturn]] void report_fatal_error(std::string_view text, bool is_html = false, bool include_help_text = true);

std::string dump_useful_thread_info()
{
	std::string result;

	if (auto cpu = get_current_cpu_thread())
	{
		cpu->dump_all(result);
	}

	return result;
}

#ifndef _WIN32
bool IsDebuggerPresent()
{
#if defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__)
	int mib[] = {
		CTL_KERN,
		KERN_PROC,
		KERN_PROC_PID,
		getpid(),
# if defined(__NetBSD__) || defined(__OpenBSD__)
		sizeof(struct kinfo_proc),
		1,
# endif
	};
	u_int miblen = std::size(mib);
	struct kinfo_proc info;
	usz size = sizeof(info);

	if (sysctl(mib, miblen, &info, &size, NULL, 0))
	{
		return false;
	}

	return info.KP_FLAGS & P_TRACED;
#else
	char buf[4096];
	fs::file status_fd("/proc/self/status");
	if (!status_fd)
	{
		std::fprintf(stderr, "Failed to open /proc/self/status\n");
		return false;
	}

	const auto num_read = status_fd.read(buf, sizeof(buf) - 1);
	if (num_read == 0 || num_read == umax)
	{
		std::fprintf(stderr, "Failed to read /proc/self/status (%d)\n", errno);
		return false;
	}

	buf[num_read] = '\0';
	std::string_view status = buf;

	const auto found = status.find("TracerPid:");
	if (found == umax)
	{
		std::fprintf(stderr, "Failed to find 'TracerPid:' in /proc/self/status\n");
		return false;
	}

	for (const char* cp = status.data() + found + 10; cp <= status.data() + num_read; ++cp)
	{
		if (!std::isspace(*cp))
		{
			return std::isdigit(*cp) != 0 && *cp != '0';
		}
	}

	return false;
#endif
}
#endif

bool is_debugger_present()
{
	if (g_cfg.core.external_debugger)
		return true;
	return IsDebuggerPresent();
}

#if defined(ARCH_X64)
enum x64_reg_t : u32
{
	X64R_RAX = 0,
	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 = 0,
	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,

	X64_BIT_O = 0x90,
	X64_BIT_NO,
	X64_BIT_C,
	X64_BIT_NC,
	X64_BIT_Z,
	X64_BIT_NZ,
	X64_BIT_BE,
	X64_BIT_NBE,
	X64_BIT_S,
	X64_BIT_NS,
	X64_BIT_P,
	X64_BIT_NP,
	X64_BIT_L,
	X64_BIT_NL,
	X64_BIT_LE,
	X64_BIT_NLE,

	X64R_ECX = X64R_CL,
};

enum x64_op_t : u32
{
	X64OP_NONE,
	X64OP_LOAD, // obtain and put the value into x64 register
	X64OP_LOAD_BE,
	X64OP_LOAD_CMP,
	X64OP_LOAD_TEST,
	X64OP_STORE, // take the value from x64 register or an immediate and use it
	X64OP_STORE_BE,
	X64OP_MOVS,
	X64OP_STOS,
	X64OP_XCHG,
	X64OP_CMPXCHG,
	X64OP_AND, // lock and [mem], ...
	X64OP_OR,  // lock or  [mem], ...
	X64OP_XOR, // lock xor [mem], ...
	X64OP_INC, // lock inc [mem]
	X64OP_DEC, // lock dec [mem]
	X64OP_ADD, // lock add [mem], ...
	X64OP_ADC, // lock adc [mem], ...
	X64OP_SUB, // lock sub [mem], ...
	X64OP_SBB, // lock sbb [mem], ...
	X64OP_BEXTR,
};

static thread_local x64_reg_t s_tls_reg3{};

void decode_x64_reg_op(const u8* code, x64_op_t& out_op, x64_reg_t& out_reg, usz& out_size, usz& 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)
			{
				sig_log.error("decode_x64_reg_op(%016llxh): LOCK prefix found twice", code - out_length);
			}

			lock = true;
			continue;
		}
		case REPNE: // group 1
		{
			if (repne)
			{
				sig_log.error("decode_x64_reg_op(%016llxh): REPNE/REPNZ prefix found twice", code - out_length);
			}

			repne = true;
			continue;
		}
		case REPE: // group 1
		{
			if (repe)
			{
				sig_log.error("decode_x64_reg_op(%016llxh): REP/REPE/REPZ prefix found twice", code - out_length);
			}

			repe = true;
			continue;
		}

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

		case 0x66: // group 3
		{
			if (oso)
			{
				sig_log.error("decode_x64_reg_op(%016llxh): operand-size override prefix found twice", code - out_length);
			}

			oso = true;
			continue;
		}

		case 0x67: // group 4
		{
			sig_log.error("decode_x64_reg_op(%016llxh): address-size override prefix found", code - out_length, prefix);
			out_op = X64OP_NONE;
			out_reg = X64_NOT_SET;
			out_size = 0;
			out_length = 0;
			return;
		}

		default:
		{
			if ((prefix & 0xf0) == 0x40) // check REX prefix
			{
				if (rex)
				{
					sig_log.error("decode_x64_reg_op(%016llxh): 0x%02x (REX prefix) found after 0x%02x", 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_RAX};
	};

	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_XMM0};
	};

	auto get_modRM_reg_lh = [](const u8* code) -> x64_reg_t
	{
		return x64_reg_t{((*code & 0x38) >> 3) + X64R_AL};
	};

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

	auto get_modRM_size = [](const u8* code) -> usz
	{
		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 0x11:
		case 0x29:
		{
			if (!repe && !repne) // MOVUPS/MOVAPS/MOVUPD/MOVAPD 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 0x7f:
		{
			if (repe != oso) // MOVDQU/MOVDQA 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;
		}
		case 0x90:
		case 0x91:
		case 0x92:
		case 0x93:
		case 0x94:
		case 0x95:
		case 0x96:
		case 0x97:
		case 0x98:
		case 0x9a:
		case 0x9b:
		case 0x9c:
		case 0x9d:
		case 0x9e:
		case 0x9f:
		{
			if (!lock) // SETcc
			{
				out_op = X64OP_STORE;
				out_reg = x64_reg_t(X64_BIT_O + op2 - 0x90); // 0x90 .. 0x9f
				out_size = 1;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		case 0x38:
		{
			out_length++, code++;

			switch (op3)
			{
			case 0xf0:
			case 0xf1:
			{
				if (!repne) // MOVBE
				{
					out_op = op3 == 0xf0 ? X64OP_LOAD_BE : X64OP_STORE_BE;
					out_reg = get_modRM_reg(code, rex);
					out_size = get_op_size(rex, oso);
					out_length += get_modRM_size(code);
					return;
				}

				break;
			}
			}

			break;
		}
		}

		break;
	}
	case 0x20:
	{
		if (!oso)
		{
			out_op = X64OP_AND;
			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 0x21:
	{
		if (true)
		{
			out_op = X64OP_AND;
			out_reg = get_modRM_reg(code, rex);
			out_size = get_op_size(rex, oso);
			out_length += get_modRM_size(code);
			return;
		}
		break;
	}
	case 0x80:
	{
		switch (get_modRM_reg(code, 0))
		{
		//case 0: out_op = X64OP_ADD; break; // TODO: strange info in instruction manual
		case 1: out_op = X64OP_OR; break;
		case 2: out_op = X64OP_ADC; break;
		case 3: out_op = X64OP_SBB; break;
		case 4: out_op = X64OP_AND; break;
		case 5: out_op = X64OP_SUB; break;
		case 6: out_op = X64OP_XOR; break;
		default: out_op = X64OP_LOAD_CMP; break;
		}

		out_reg = X64_IMM8;
		out_size = 1;
		out_length += get_modRM_size(code) + 1;
		return;
	}
	case 0x81:
	{
		switch (get_modRM_reg(code, 0))
		{
		case 0: out_op = X64OP_ADD; break;
		case 1: out_op = X64OP_OR; break;
		case 2: out_op = X64OP_ADC; break;
		case 3: out_op = X64OP_SBB; break;
		case 4: out_op = X64OP_AND; break;
		case 5: out_op = X64OP_SUB; break;
		case 6: out_op = X64OP_XOR; break;
		default: out_op = X64OP_LOAD_CMP; break;
		}

		out_reg = oso ? X64_IMM16 : X64_IMM32;
		out_size = get_op_size(rex, oso);
		out_length += get_modRM_size(code) + (oso ? 2 : 4);
		return;
	}
	case 0x83:
	{
		switch (get_modRM_reg(code, 0))
		{
		case 0: out_op = X64OP_ADD; break;
		case 1: out_op = X64OP_OR; break;
		case 2: out_op = X64OP_ADC; break;
		case 3: out_op = X64OP_SBB; break;
		case 4: out_op = X64OP_AND; break;
		case 5: out_op = X64OP_SUB; break;
		case 6: out_op = X64OP_XOR; break;
		default: out_op = X64OP_LOAD_CMP; break;
		}

		out_reg = X64_IMM8;
		out_size = get_op_size(rex, oso);
		out_length += get_modRM_size(code) + 1;
		return;
	}
	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 0xaa:
	{
		if (!oso && !lock && !repe && !rex) // STOS
		{
			out_op = X64OP_STOS;
			out_reg = X64_NOT_SET;
			out_size = 1;
			return;
		}
		if (!oso && !lock && repe) // REP STOS
		{
			out_op = X64OP_STOS;
			out_reg = rex & 8 ? X64R_RCX : X64R_ECX;
			out_size = 1;
			return;
		}
		break;
	}
	case 0xc4: // 3-byte VEX prefix
	case 0xc5: // 2-byte VEX prefix
	{
		// Last prefix byte: op2 or op3
		const u8 opx = op1 == 0xc5 ? op2 : op3;

		// Implied prefixes
		rex |= op2 & 0x80 ? 0 : 0x4; // REX.R
		rex |= op1 == 0xc4 && op3 & 0x80 ? 0x8 : 0; // REX.W ???
		oso = (opx & 0x3) == 0x1;
		repe = (opx & 0x3) == 0x2;
		repne = (opx & 0x3) == 0x3;

		const u8 vopm = op1 == 0xc5 ? 1 : op2 & 0x1f;
		const u8 vop1 = op1 == 0xc5 ? op3 : code[2];
		const u8 vlen = (opx & 0x4) ? 32 : 16;
		const u8 vreg = (~opx >> 3) & 0xf;
		out_length += op1 == 0xc5 ? 2 : 3;
		code += op1 == 0xc5 ? 2 : 3;

		s_tls_reg3 = x64_reg_t{vreg};

		if (vopm == 0x1) switch (vop1) // Implied leading byte 0x0F
		{
		case 0x11:
		case 0x29:
		{
			if (!repe && !repne) // VMOVAPS/VMOVAPD/VMOVUPS/VMOVUPD mem,reg
			{
				out_op = X64OP_STORE;
				out_reg = get_modRM_reg_xmm(code, rex);
				out_size = vlen;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		case 0x7f:
		{
			if (repe || oso) // VMOVDQU/VMOVDQA mem,reg
			{
				out_op = X64OP_STORE;
				out_reg = get_modRM_reg_xmm(code, rex);
				out_size = vlen;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		}

		if (vopm == 0x2) switch (vop1) // Implied leading bytes 0x0F 0x38
		{
		case 0xf7:
		{
			if (!repe && !repne && vlen == 16) // BEXTR r32,mem,r32
			{
				out_op = X64OP_BEXTR;
				out_reg = get_modRM_reg_xmm(code, rex);
				out_size = opx & 0x80 ? 8 : 4;
				out_length += get_modRM_size(code);
				return;
			}
			break;
		}
		}

		break;
	}
	case 0xc6:
	{
		if (!lock && !oso && get_modRM_reg(code, 0) == 0) // 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) == 0) // 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;
	}
	case 0xf6:
	{
		switch (get_modRM_reg(code, 0))
		{
		case 0: out_op = X64OP_LOAD_TEST; break;
		default: out_op = X64OP_NONE; break; // TODO...
		}

		out_reg = X64_IMM8;
		out_size = 1;
		out_length += get_modRM_size(code) + 1;
		return;
	}
	case 0xf7:
	{
		switch (get_modRM_reg(code, 0))
		{
		case 0: out_op = X64OP_LOAD_TEST; break;
		default: out_op = X64OP_NONE; break; // TODO...
		}

		out_reg = oso ? X64_IMM16 : X64_IMM32;
		out_size = get_op_size(rex, oso);
		out_length += get_modRM_size(code) + (oso ? 2 : 4);
		return;
	}
	}

	out_op = X64OP_NONE;
	out_reg = X64_NOT_SET;
	out_size = 0;
	out_length = 0;
}

#ifdef _WIN32

typedef CONTEXT x64_context;
typedef CONTEXT ucontext_t;

#define X64REG(context, reg) (&(&(context)->Rax)[reg])
#define XMMREG(context, reg) (reinterpret_cast<v128*>(&(&(context)->Xmm0)[reg]))
#define EFLAGS(context) ((context)->EFlags)

#define ARG1(context) RCX(context)
#define ARG2(context) RDX(context)

#else

typedef ucontext_t x64_context;

#ifdef __APPLE__

#define X64REG(context, reg) (darwin_x64reg(context, reg))
#define XMMREG(context, reg) (reinterpret_cast<v128*>(&(context)->uc_mcontext->__fs.__fpu_xmm0.__xmm_reg[reg]))
#define EFLAGS(context) ((context)->uc_mcontext->__ss.__rflags)

u64* darwin_x64reg(x64_context *context, int reg)
{
	auto *state = &context->uc_mcontext->__ss;
	switch(reg)
	{
	case 0: return &state->__rax;
	case 1: return &state->__rcx;
	case 2: return &state->__rdx;
	case 3: return &state->__rbx;
	case 4: return &state->__rsp;
	case 5: return &state->__rbp;
	case 6: return &state->__rsi;
	case 7: return &state->__rdi;
	case 8: return &state->__r8;
	case 9: return &state->__r9;
	case 10: return &state->__r10;
	case 11: return &state->__r11;
	case 12: return &state->__r12;
	case 13: return &state->__r13;
	case 14: return &state->__r14;
	case 15: return &state->__r15;
	case 16: return &state->__rip;
	default:
		sig_log.error("Invalid register index: %d", reg);
		return nullptr;
	}
}

#elif defined(__DragonFly__) || defined(__FreeBSD__)

#define X64REG(context, reg) (freebsd_x64reg(context, reg))
#ifdef __DragonFly__
#  define XMMREG(context, reg) (reinterpret_cast<v128*>((reinterpret_cast<union savefpu*>(context)->uc_mcontext.mc_fpregs)->sv_xmm.sv_xmm[reg]))
#else
#  define XMMREG(context, reg) (reinterpret_cast<v128*>((reinterpret_cast<struct savefpu*>(context)->uc_mcontext.mc_fpstate)->sv_xmm[reg]))
#endif
#define EFLAGS(context) ((context)->uc_mcontext.mc_rflags)

register_t* freebsd_x64reg(x64_context *context, int reg)
{
	auto *state = &context->uc_mcontext;
	switch(reg)
	{
	case 0: return &state->mc_rax;
	case 1: return &state->mc_rcx;
	case 2: return &state->mc_rdx;
	case 3: return &state->mc_rbx;
	case 4: return &state->mc_rsp;
	case 5: return &state->mc_rbp;
	case 6: return &state->mc_rsi;
	case 7: return &state->mc_rdi;
	case 8: return &state->mc_r8;
	case 9: return &state->mc_r9;
	case 10: return &state->mc_r10;
	case 11: return &state->mc_r11;
	case 12: return &state->mc_r12;
	case 13: return &state->mc_r13;
	case 14: return &state->mc_r14;
	case 15: return &state->mc_r15;
	case 16: return &state->mc_rip;
	default:
		sig_log.error("Invalid register index: %d", reg);
		return nullptr;
	}
}

#elif defined(__OpenBSD__)

#define X64REG(context, reg) (openbsd_x64reg(context, reg))
#define XMMREG(context, reg) (reinterpret_cast<v128*>((context)->sc_fpstate->fx_xmm[reg]))
#define EFLAGS(context) ((context)->sc_rflags)

long* openbsd_x64reg(x64_context *context, int reg)
{
	auto *state = &context;
	switch(reg)
	{
	case 0: return &state->sc_rax;
	case 1: return &state->sc_rcx;
	case 2: return &state->sc_rdx;
	case 3: return &state->sc_rbx;
	case 4: return &state->sc_rsp;
	case 5: return &state->sc_rbp;
	case 6: return &state->sc_rsi;
	case 7: return &state->sc_rdi;
	case 8: return &state->sc_r8;
	case 9: return &state->sc_r9;
	case 10: return &state->sc_r10;
	case 11: return &state->sc_r11;
	case 12: return &state->sc_r12;
	case 13: return &state->sc_r13;
	case 14: return &state->sc_r14;
	case 15: return &state->sc_r15;
	case 16: return &state->sc_rip;
	default:
		sig_log.error("Invalid register index: %d", reg);
		return nullptr;
	}
}

#elif defined(__NetBSD__)

static const decltype(_REG_RAX) reg_table[] =
{
	_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]])
#define XMM_sig(context, reg) (reinterpret_cast<v128*>(((struct fxsave64*)(context)->uc_mcontext.__fpregs)->fx_xmm[reg]))
#define EFLAGS(context) ((context)->uc_mcontext.__gregs[_REG_RFL])

#else

static const int reg_table[] =
{
	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]])
#ifdef __sun
#define XMMREG(context, reg) (reinterpret_cast<v128*>(&(context)->uc_mcontext.fpregs.fp_reg_set.fpchip_state.xmm[reg_table[reg]]))
#else
#define XMMREG(context, reg) (reinterpret_cast<v128*>(&(context)->uc_mcontext.fpregs->_xmm[reg]))
#endif // __sun
#define EFLAGS(context) ((context)->uc_mcontext.gregs[REG_EFL])

#endif // __APPLE__

#define ARG1(context) RDI(context)
#define ARG2(context) RSI(context)

#endif

#define RAX(c) (*X64REG((c), 0))
#define RCX(c) (*X64REG((c), 1))
#define RDX(c) (*X64REG((c), 2))
#define RSP(c) (*X64REG((c), 4))
#define RSI(c) (*X64REG((c), 6))
#define RDI(c) (*X64REG((c), 7))
#define RIP(c) (*X64REG((c), 16))

bool get_x64_reg_value(x64_context* context, x64_reg_t reg, usz d_size, usz i_size, u64& out_value)
{
	// get x64 reg value (for store operations)
	if (reg - X64R_RAX < 16)
	{
		// load the value from x64 register
		const u64 reg_value = *X64REG(context, reg - X64R_RAX);

		switch (d_size)
		{
		case 1: out_value = static_cast<u8>(reg_value); return true;
		case 2: out_value = static_cast<u16>(reg_value); return true;
		case 4: out_value = static_cast<u32>(reg_value); return true;
		case 8: out_value = reg_value; return true;
		}
	}
	else if (reg - X64R_AL < 4 && d_size == 1)
	{
		out_value = static_cast<u8>(*X64REG(context, reg - X64R_AL));
		return true;
	}
	else if (reg - X64R_AH < 4 && d_size == 1)
	{
		out_value = static_cast<u8>(*X64REG(context, reg - X64R_AH) >> 8);
		return true;
	}
	else if (reg == X64_IMM8)
	{
		// load the immediate value (assuming it's at the end of the instruction)
		const s8 imm_value = *reinterpret_cast<s8*>(RIP(context) + i_size - 1);

		switch (d_size)
		{
		case 1: out_value = static_cast<u8>(imm_value); return true;
		case 2: out_value = static_cast<u16>(imm_value); return true; // sign-extended
		case 4: out_value = static_cast<u32>(imm_value); return true; // sign-extended
		case 8: out_value = static_cast<u64>(imm_value); return true; // sign-extended
		}
	}
	else if (reg == X64_IMM16)
	{
		const s16 imm_value = *reinterpret_cast<s16*>(RIP(context) + i_size - 2);

		switch (d_size)
		{
		case 2: out_value = static_cast<u16>(imm_value); return true;
		}
	}
	else if (reg == X64_IMM32)
	{
		const s32 imm_value = *reinterpret_cast<s32*>(RIP(context) + i_size - 4);

		switch (d_size)
		{
		case 4: out_value = static_cast<u32>(imm_value); return true;
		case 8: out_value = static_cast<u64>(imm_value); return true; // sign-extended
		}
	}
	else if (reg == X64R_ECX)
	{
		out_value = static_cast<u32>(RCX(context));
		return true;
	}
	else if (reg >= X64_BIT_O && reg <= X64_BIT_NLE)
	{
		const u32 _cf = EFLAGS(context) & 0x1;
		const u32 _zf = EFLAGS(context) & 0x40;
		const u32 _sf = EFLAGS(context) & 0x80;
		const u32 _of = EFLAGS(context) & 0x800;
		const u32 _pf = EFLAGS(context) & 0x4;
		const u32 _l = (_sf << 4) ^ _of; // SF != OF

		switch (reg & ~1)
		{
		case X64_BIT_O: out_value = !!_of ^ (reg & 1); break;
		case X64_BIT_C: out_value = !!_cf ^ (reg & 1); break;
		case X64_BIT_Z: out_value = !!_zf ^ (reg & 1); break;
		case X64_BIT_BE: out_value = !!(_cf | _zf) ^ (reg & 1); break;
		case X64_BIT_S: out_value = !!_sf ^ (reg & 1); break;
		case X64_BIT_P: out_value = !!_pf ^ (reg & 1); break;
		case X64_BIT_L: out_value = !!_l ^ (reg & 1); break;
		case X64_BIT_LE: out_value = !!(_l | _zf) ^ (reg & 1); break;
		}

		return true;
	}

	sig_log.error("get_x64_reg_value(): invalid arguments (reg=%d, d_size=%lld, i_size=%lld)", +reg, d_size, i_size);
	return false;
}

bool put_x64_reg_value(x64_context* context, x64_reg_t reg, usz d_size, u64 value)
{
	// save x64 reg value (for load operations)
	if (reg - X64R_RAX < 16)
	{
		// save the value into x64 register
		switch (d_size)
		{
		case 1: *X64REG(context, reg - X64R_RAX) = (value & 0xff) | (*X64REG(context, reg - X64R_RAX) & 0xffffff00); return true;
		case 2: *X64REG(context, reg - X64R_RAX) = (value & 0xffff) | (*X64REG(context, reg - X64R_RAX) & 0xffff0000); return true;
		case 4: *X64REG(context, reg - X64R_RAX) = value & 0xffffffff; return true;
		case 8: *X64REG(context, reg - X64R_RAX) = value; return true;
		}
	}

	sig_log.error("put_x64_reg_value(): invalid destination (reg=%d, d_size=%lld, value=0x%llx)", +reg, d_size, value);
	return false;
}

bool set_x64_cmp_flags(x64_context* context, usz d_size, u64 x, u64 y, bool carry = true)
{
	switch (d_size)
	{
	case 1: break;
	case 2: break;
	case 4: break;
	case 8: break;
	default: sig_log.error("set_x64_cmp_flags(): invalid d_size (%lld)", d_size); return false;
	}

	const u64 sign = 1ull << (d_size * 8 - 1); // sign mask
	const u64 diff = x - y;
	const u64 summ = x + y;

	if (carry && ((x & y) | ((x ^ y) & ~summ)) & sign)
	{
		EFLAGS(context) |= 0x1; // set CF
	}
	else if (carry)
	{
		EFLAGS(context) &= ~0x1; // clear CF
	}

	if (x == y)
	{
		EFLAGS(context) |= 0x40; // set ZF
	}
	else
	{
		EFLAGS(context) &= ~0x40; // clear ZF
	}

	if (diff & sign)
	{
		EFLAGS(context) |= 0x80; // set SF
	}
	else
	{
		EFLAGS(context) &= ~0x80; // clear SF
	}

	if ((x ^ summ) & (y ^ summ) & sign)
	{
		EFLAGS(context) |= 0x800; // set OF
	}
	else
	{
		EFLAGS(context) &= ~0x800; // clear OF
	}

	const u8 p1 = static_cast<u8>(diff) ^ (static_cast<u8>(diff) >> 4);
	const u8 p2 = p1 ^ (p1 >> 2);
	const u8 p3 = p2 ^ (p2 >> 1);

	if ((p3 & 1) == 0)
	{
		EFLAGS(context) |= 0x4; // set PF
	}
	else
	{
		EFLAGS(context) &= ~0x4; // clear PF
	}

	if (((x & y) | ((x ^ y) & ~summ)) & 0x8)
	{
		EFLAGS(context) |= 0x10; // set AF
	}
	else
	{
		EFLAGS(context) &= ~0x10; // clear AF
	}

	return true;
}

usz get_x64_access_size(x64_context* context, x64_op_t op, x64_reg_t reg, usz d_size, usz i_size)
{
	if (op == X64OP_MOVS || op == X64OP_STOS)
	{
		if (EFLAGS(context) & 0x400 /* direction flag */)
		{
			// TODO
			return 0;
		}

		if (reg != X64_NOT_SET) // get "full" access size from RCX register
		{
			u64 counter = 1;
			if (!get_x64_reg_value(context, reg, 8, i_size, counter))
			{
				return -1;
			}

			return d_size * counter;
		}
	}

	return d_size;
}

#elif defined(ARCH_ARM64)

#if defined(__APPLE__)
// https://github.com/bombela/backward-cpp/issues/200
#define RIP(context) ((context)->uc_mcontext->__ss.__pc)
#elif defined(__FreeBSD__)
#define RIP(context) ((context)->uc_mcontext.mc_gpregs.gp_elr)
#elif defined(__NetBSD__)
#define RIP(context) ((context)->uc_mcontext.__gregs[_REG_PC])
#elif defined(__OpenBSD__)
#define RIP(context) ((context)->sc_elr)
#else
#define RIP(context) ((context)->uc_mcontext.pc)
#endif

#endif /* ARCH_ */

namespace rsx
{
	extern std::function<bool(u32 addr, bool is_writing)> g_access_violation_handler;
}

bool handle_access_violation(u32 addr, bool is_writing, ucontext_t* context) noexcept
{
	g_tls_fault_all++;

	const auto cpu = get_current_cpu_thread();

	if (addr < RAW_SPU_BASE_ADDR && vm::check_addr(addr) && rsx::g_access_violation_handler)
	{
		bool state_changed = false;

		if (cpu)
		{
			state_changed = vm::temporary_unlock(*cpu);
		}

		bool handled = rsx::g_access_violation_handler(addr, is_writing);

		if (state_changed && (cpu->state += cpu_flag::temp, cpu->test_stopped()))
		{
			//
		}

		if (handled)
		{
			g_tls_fault_rsx++;
			return true;
		}
	}

#if defined(ARCH_X64)
	const u8* const code = reinterpret_cast<u8*>(RIP(context));

	x64_op_t op;
	x64_reg_t reg;
	usz d_size;
	usz i_size;

	// decode single x64 instruction that causes memory access
	decode_x64_reg_op(code, op, reg, d_size, i_size);

	auto report_opcode = [=]()
	{
		if (op == X64OP_NONE)
		{
			be_t<v128> dump;
			std::memcpy(&dump, code, sizeof(dump));
			sig_log.error("decode_x64_reg_op(%p): unsupported opcode: %s", code, dump);
		}
	};

	if (0x1'0000'0000ull - addr < d_size)
	{
		sig_log.error("Invalid d_size (0x%llx)", d_size);
		report_opcode();
		return false;
	}

	// get length of data being accessed
	usz a_size = get_x64_access_size(context, op, reg, d_size, i_size);

	if (0x1'0000'0000ull - addr < a_size)
	{
		sig_log.error("Invalid a_size (0x%llx)", a_size);
		report_opcode();
		return false;
	}

	// check if address is RawSPU MMIO register
	do if (addr - RAW_SPU_BASE_ADDR < (6 * RAW_SPU_OFFSET) && (addr % RAW_SPU_OFFSET) >= RAW_SPU_PROB_OFFSET)
	{
		auto thread = idm::get<named_thread<spu_thread>>(spu_thread::find_raw_spu((addr - RAW_SPU_BASE_ADDR) / RAW_SPU_OFFSET));

		if (!thread)
		{
			break;
		}

		if (!a_size || !d_size || !i_size)
		{
			sig_log.error("Invalid or unsupported instruction (op=%d, reg=%d, d_size=%lld, a_size=0x%llx, i_size=%lld)", +op, +reg, d_size, a_size, i_size);
			report_opcode();
			return false;
		}

		if (a_size != 4)
		{
			// Might be unimplemented, such as writing MFC proxy EAL+EAH using 64-bit store
			break;
		}

		switch (op)
		{
		case X64OP_LOAD:
		case X64OP_LOAD_BE:
		case X64OP_LOAD_CMP:
		case X64OP_LOAD_TEST:
		{
			u32 value;
			if (is_writing || !thread->read_reg(addr, value))
			{
				return false;
			}

			if (op != X64OP_LOAD_BE)
			{
				value = stx::se_storage<u32>::swap(value);
			}

			if (op == X64OP_LOAD_CMP)
			{
				u64 rvalue;
				if (!get_x64_reg_value(context, reg, d_size, i_size, rvalue) || !set_x64_cmp_flags(context, d_size, value, rvalue))
				{
					return false;
				}

				break;
			}

			if (op == X64OP_LOAD_TEST)
			{
				u64 rvalue;
				if (!get_x64_reg_value(context, reg, d_size, i_size, rvalue) || !set_x64_cmp_flags(context, d_size, value & rvalue, 0))
				{
					return false;
				}

				break;
			}

			if (!put_x64_reg_value(context, reg, d_size, value))
			{
				return false;
			}

			break;
		}
		case X64OP_BEXTR:
		{
			u32 value;
			if (is_writing || !thread->read_reg(addr, value))
			{
				return false;
			}

			value = stx::se_storage<u32>::swap(value);

			u64 ctrl;
			if (!get_x64_reg_value(context, s_tls_reg3, d_size, i_size, ctrl))
			{
				return false;
			}

			u8 start = ctrl & 0xff;
			u8 _len = (ctrl & 0xff00) >> 8;
			if (_len > 32)
				_len = 32;
			if (start > 32)
				start = 32;
			value = (u64{value} >> start) & ~(u64{umax} << _len);

			if (!put_x64_reg_value(context, reg, d_size, value) || !set_x64_cmp_flags(context, d_size, value, 0))
			{
				return false;
			}

			break;
		}
		case X64OP_STORE:
		case X64OP_STORE_BE:
		{
			u64 reg_value;
			if (!is_writing || !get_x64_reg_value(context, reg, d_size, i_size, reg_value))
			{
				return false;
			}

			u32 val32 = static_cast<u32>(reg_value);
			if (!thread->write_reg(addr, op == X64OP_STORE ? stx::se_storage<u32>::swap(val32) : val32))
			{
				return false;
			}

			break;
		}
		case X64OP_MOVS: // possibly, TODO
		case X64OP_STOS:
		default:
		{
			sig_log.error("Invalid or unsupported operation (op=%d, reg=%d, d_size=%lld, i_size=%lld)", +op, +reg, d_size, i_size);
			report_opcode();
			return false;
		}
		}

		// skip processed instruction
		RIP(context) += i_size;
		g_tls_fault_spu++;
		return true;
	} while (0);
#else
	static_cast<void>(context);
#endif /* ARCH_ */

	if (vm::check_addr(addr, is_writing ? vm::page_writable : vm::page_readable))
	{
		return true;
	}

	// Hack: allocate memory in case the emulator is stopping
	const auto hack_alloc = [&]()
	{
		g_tls_access_violation_recovered = true;

		if (vm::check_addr(addr, is_writing ? vm::page_writable : vm::page_readable))
		{
			return true;
		}

		const auto area = vm::reserve_map(vm::any, addr & -0x10000, 0x10000);

		if (!area)
		{
			return false;
		}

		if (vm::writer_lock mlock; area->flags & vm::preallocated || vm::check_addr(addr, 0))
		{
			// For allocated memory with protection lower than required (such as protection::no or read-only while writing to it)
			utils::memory_protect(vm::base(addr & -0x1000), 0x1000, utils::protection::rw);
			return true;
		}

		return area->falloc(addr & -0x10000, 0x10000) || vm::check_addr(addr, is_writing ? vm::page_writable : vm::page_readable);
	};

	if (cpu && (cpu->id_type() == 1 || cpu->id_type() == 2))
	{
		vm::temporary_unlock(*cpu);
		u32 pf_port_id = 0;

		if (auto& pf_entries = g_fxo->get<page_fault_notification_entries>(); true)
		{
			if (auto mem = vm::get(vm::any, addr))
			{
				reader_lock lock(pf_entries.mutex);

				for (const auto& entry : pf_entries.entries)
				{
					if (entry.start_addr == mem->addr)
					{
						pf_port_id = entry.port_id;
						break;
					}
				}
			}
		}

		if (auto pf_port = idm::get<lv2_obj, lv2_event_port>(pf_port_id); pf_port && pf_port->queue)
		{
			// We notify the game that a page fault occurred so it can rectify it.
			// Note, for data3, were the memory readable AND we got a page fault, it must be due to a write violation since reads are allowed.
			u64 data1 = addr;
			u64 data2 = 0;

			if (cpu->try_get<ppu_thread>())
			{
				data2 = (SYS_MEMORY_PAGE_FAULT_TYPE_PPU_THREAD << 32) | cpu->id;
			}
			else if (auto spu = cpu->try_get<spu_thread>())
			{
				const u64 type = spu->get_type() == spu_type::threaded ?
					SYS_MEMORY_PAGE_FAULT_TYPE_SPU_THREAD :
					SYS_MEMORY_PAGE_FAULT_TYPE_RAW_SPU;

				data2 = (type << 32) | spu->lv2_id;
			}

			u64 data3;
			{
				vm::writer_lock rlock;
				if (vm::check_addr(addr, is_writing ? vm::page_writable : vm::page_readable))
				{
					// Memory was allocated inbetween, retry
					return true;
				}
				else if (vm::check_addr(addr))
				{
					data3 = SYS_MEMORY_PAGE_FAULT_CAUSE_READ_ONLY; // TODO
				}
				else
				{
					data3 = SYS_MEMORY_PAGE_FAULT_CAUSE_NON_MAPPED;
				}
			}


			// Now, place the page fault event onto table so that other functions [sys_mmapper_free_address and pagefault recovery funcs etc]
			// know that this thread is page faulted and where.

			auto& pf_events = g_fxo->get<page_fault_event_entries>();

			// De-schedule
			if (cpu->id_type() == 1)
			{
				cpu->state -= cpu_flag::signal; // Cannot use check_state here and signal must be removed if exists
				lv2_obj::sleep(*cpu);
			}

			auto send_event = [&]() -> error_code
			{
				lv2_obj::notify_all_t notify_later{};

				std::lock_guard pf_lock(pf_events.pf_mutex);

				if (auto error = pf_port->queue->send(pf_port->name ? pf_port->name : ((u64{process_getpid() + 0u} << 32) | u64{pf_port_id}), data1, data2, data3))
				{
					return error;
				}

				pf_events.events.emplace(cpu, addr);
				return {};
			};

			sig_log.warning("Page_fault %s location 0x%x because of %s memory", is_writing ? "writing" : "reading",
				addr, data3 == SYS_MEMORY_PAGE_FAULT_CAUSE_READ_ONLY ? "writing read-only" : "using unmapped");

			if (cpu->id_type() == 1)
			{
				if (const auto func = static_cast<ppu_thread*>(cpu)->current_function)
				{
					sig_log.warning("Page_fault while in function %s", func);
				}
			}

			error_code sending_error = not_an_error(CELL_EBUSY);

			// If we fail due to being busy, wait a bit and try again.
			for (; static_cast<u32>(sending_error) == CELL_EBUSY; thread_ctrl::wait_for(1000))
			{
				sending_error = send_event();

				if (cpu->is_stopped())
				{
					sending_error = {};
					break;
				}
			}

			if (sending_error)
			{
				vm_log.error("Unknown error 0x%x while trying to pass page fault.", +sending_error);
				return false;
			}
			else
			{
				// Wait until the thread is recovered
				while (auto state = cpu->state.fetch_sub(cpu_flag::signal))
				{
					if (is_stopped(state) || state & cpu_flag::signal)
					{
						break;
					}

					thread_ctrl::wait_on(cpu->state, state);
				}
			}

			// Reschedule, test cpu state and try recovery if stopped
			if (cpu->test_stopped() && !hack_alloc())
			{
				return false;
			}

			return true;
		}

		if (cpu->id_type() == 2)
		{
			if (!g_tls_access_violation_recovered)
			{
				vm_log.notice("\n%s", dump_useful_thread_info());
				vm_log.error("[%s] Access violation %s location 0x%x (%s)", cpu->get_name(), is_writing ? "writing" : "reading", addr, (is_writing && vm::check_addr(addr)) ? "read-only memory" : "unmapped memory");
			}

			// TODO:
			// RawSPU: Send appropriate interrupt
			// SPUThread: Send sys_spu exception event
			cpu->state += cpu_flag::dbg_pause;

			if (cpu->check_state() && !hack_alloc())
			{
				return false;
			}

			return true;
		}
		else
		{
			if (auto last_func = static_cast<ppu_thread*>(cpu)->current_function)
			{
				ppu_log.fatal("Function aborted: %s", last_func);
			}

			lv2_obj::sleep(*cpu);
		}
	}

	if (cpu)
	{
		cpu->state += cpu_flag::wait;
	}

	Emu.Pause(true);

	if (!g_tls_access_violation_recovered)
	{
		vm_log.notice("\n%s", dump_useful_thread_info());
	}

	// Note: a thread may access violate more than once after hack_alloc recovery
	// Do not log any further access violations in this case.
	if (!g_tls_access_violation_recovered)
	{
		vm_log.fatal("Access violation %s location 0x%x (%s)", is_writing ? "writing" : (cpu && cpu->id_type() == 1 && cpu->get_pc() == addr ? "executing" : "reading"), addr, (is_writing && vm::check_addr(addr)) ? "read-only memory" : "unmapped memory");
	}

	while (Emu.IsPaused())
	{
		thread_ctrl::wait();
	}

	if (Emu.IsStopped() && !hack_alloc())
	{
		return false;
	}

	return true;
}

static void append_thread_name(std::string& msg)
{
	if (thread_ctrl::get_current())
	{
		fmt::append(msg, "Emu Thread Name: '%s'.\n", thread_ctrl::get_name());
	}
	else if (thread_ctrl::is_main())
	{
		fmt::append(msg, "Thread: Main Thread.\n");
	}
	else
	{
		fmt::append(msg, "Thread id = %u.\n", thread_ctrl::get_tid());
	}
}

#ifdef _WIN32

static LONG exception_handler(PEXCEPTION_POINTERS pExp) noexcept
{
	if (pExp->ExceptionRecord->ExceptionCode == EXCEPTION_BREAKPOINT)
	{
		return EXCEPTION_CONTINUE_SEARCH;
	}

	const auto ptr = reinterpret_cast<u8*>(pExp->ExceptionRecord->ExceptionInformation[1]);
	const bool is_writing = pExp->ExceptionRecord->ExceptionInformation[0] == 1;
	const bool is_executing = pExp->ExceptionRecord->ExceptionInformation[0] == 8;

	if (pExp->ExceptionRecord->ExceptionCode == EXCEPTION_ACCESS_VIOLATION && !is_executing)
	{
		u32 addr = 0;

		if (auto [addr0, ok] = vm::try_get_addr(ptr); ok)
		{
			addr = addr0;
		}
		else if (const usz exec64 = (ptr - vm::g_exec_addr) / 2; exec64 <= u32{umax})
		{
			addr = static_cast<u32>(exec64);
		}
		else
		{
			return EXCEPTION_CONTINUE_SEARCH;
		}

		if (thread_ctrl::get_current() && handle_access_violation(addr, is_writing, pExp->ContextRecord))
		{
			return EXCEPTION_CONTINUE_EXECUTION;
		}
	}

	switch (pExp->ExceptionRecord->ExceptionCode)
	{
	case EXCEPTION_ACCESS_VIOLATION:
	case EXCEPTION_ARRAY_BOUNDS_EXCEEDED:
	case EXCEPTION_DATATYPE_MISALIGNMENT:
	case EXCEPTION_ILLEGAL_INSTRUCTION:
	case EXCEPTION_IN_PAGE_ERROR:
	case EXCEPTION_INT_DIVIDE_BY_ZERO:
	case EXCEPTION_NONCONTINUABLE_EXCEPTION:
	case EXCEPTION_PRIV_INSTRUCTION:
	//case EXCEPTION_STACK_OVERFLOW:
	{
		sys_log.notice("\n%s", dump_useful_thread_info());
		logs::listener::sync_all();
		break;
	}
	default:
	{
		break;
	}
	}

	return EXCEPTION_CONTINUE_SEARCH;
}

static LONG exception_filter(PEXCEPTION_POINTERS pExp) noexcept
{
	std::string msg = fmt::format("Unhandled Win32 exception 0x%08X.\n", pExp->ExceptionRecord->ExceptionCode);

	if (pExp->ExceptionRecord->ExceptionCode == EXCEPTION_ACCESS_VIOLATION)
	{
		const auto cause =
			pExp->ExceptionRecord->ExceptionInformation[0] == 8 ? "executing" :
			pExp->ExceptionRecord->ExceptionInformation[0] == 1 ? "writing" : "reading";

		fmt::append(msg, "Segfault %s location %p at %p.\n", cause, pExp->ExceptionRecord->ExceptionInformation[1], pExp->ExceptionRecord->ExceptionAddress);
	}
	else
	{
		fmt::append(msg, "Exception address: %p.\n", pExp->ExceptionRecord->ExceptionAddress);

		for (DWORD i = 0; i < pExp->ExceptionRecord->NumberParameters; i++)
		{
			fmt::append(msg, "ExceptionInformation[0x%x]: %p.\n", i, pExp->ExceptionRecord->ExceptionInformation[i]);
		}
	}

	append_thread_name(msg);

	std::vector<HMODULE> modules;
	for (DWORD size = 256; modules.size() != size; size /= sizeof(HMODULE))
	{
		modules.resize(size);
		if (!EnumProcessModules(GetCurrentProcess(), modules.data(), size * sizeof(HMODULE), &size))
		{
			modules.clear();
			break;
		}
	}

	fmt::append(msg, "Instruction address: %p.\n", pExp->ContextRecord->Rip);

	DWORD64 unwind_base;
	if (const auto rtf = RtlLookupFunctionEntry(pExp->ContextRecord->Rip, &unwind_base, nullptr))
	{
		// Get function address
		const DWORD64 func_addr = rtf->BeginAddress + unwind_base;
		fmt::append(msg, "Function address: %p (base+0x%x).\n", func_addr, rtf->BeginAddress);

		// Access UNWIND_INFO structure
		//const auto uw = (u8*)(unwind_base + rtf->UnwindData);
	}

	for (HMODULE _module : modules)
	{
		MODULEINFO info;
		if (GetModuleInformation(GetCurrentProcess(), _module, &info, sizeof(info)))
		{
			const DWORD64 base = reinterpret_cast<DWORD64>(info.lpBaseOfDll);

			if (pExp->ContextRecord->Rip >= base && pExp->ContextRecord->Rip < base + info.SizeOfImage)
			{
				std::string module_name;
				for (DWORD size = 15; module_name.size() != size;)
				{
					module_name.resize(size);
					size = GetModuleBaseNameA(GetCurrentProcess(), _module, &module_name.front(), size + 1);
					if (!size)
					{
						module_name.clear();
						break;
					}
				}

				fmt::append(msg, "Module name: '%s'.\n", module_name);
				fmt::append(msg, "Module base: %p.\n", info.lpBaseOfDll);
			}
		}
	}

	fmt::append(msg, "RPCS3 image base: %p.\n", GetModuleHandle(NULL));

	// TODO: print registers and the callstack

	sys_log.fatal("\n%s", msg);
	logs::listener::sync_all();

	thread_ctrl::emergency_exit(msg);
}

const bool s_exception_handler_set = []() -> bool
{
#ifdef USE_ASAN
	if (!AddVectoredExceptionHandler(FALSE, static_cast<PVECTORED_EXCEPTION_HANDLER>(exception_handler)))
#else
	if (!AddVectoredExceptionHandler(1, static_cast<PVECTORED_EXCEPTION_HANDLER>(exception_handler)))
#endif
	{
		report_fatal_error("AddVectoredExceptionHandler() failed.");
	}

	if (!SetUnhandledExceptionFilter(static_cast<LPTOP_LEVEL_EXCEPTION_FILTER>(exception_filter)))
	{
		report_fatal_error("SetUnhandledExceptionFilter() failed.");
	}

	return true;
}();

#else

static void signal_handler(int /*sig*/, siginfo_t* info, void* uct) noexcept
{
	ucontext_t* context = static_cast<ucontext_t*>(uct);

#if defined(ARCH_X64)
#ifdef __APPLE__
	const u64 err = context->uc_mcontext->__es.__err;
#elif defined(__DragonFly__) || defined(__FreeBSD__)
	const u64 err = context->uc_mcontext.mc_err;
#elif defined(__OpenBSD__)
	const u64 err = context->sc_err;
#elif defined(__NetBSD__)
	const u64 err = context->uc_mcontext.__gregs[_REG_ERR];
#else
	const u64 err = context->uc_mcontext.gregs[REG_ERR];
#endif

	const bool is_executing = err & 0x10;
	const bool is_writing = err & 0x2;
#elif defined(ARCH_ARM64)
	const bool is_executing = uptr(info->si_addr) == uptr(RIP(context));
	const u32 insn = is_executing ? 0 : *reinterpret_cast<u32*>(RIP(context));
	const bool is_writing = (insn & 0xbfff0000) == 0x0c000000
		|| (insn & 0xbfe00000) == 0x0c800000
		|| (insn & 0xbfdf0000) == 0x0d000000
		|| (insn & 0xbfc00000) == 0x0d800000
		|| (insn & 0x3f400000) == 0x08000000
		|| (insn & 0x3bc00000) == 0x39000000
		|| (insn & 0x3fc00000) == 0x3d800000
		|| (insn & 0x3bc00000) == 0x38000000
		|| (insn & 0x3fe00000) == 0x3c800000
		|| (insn & 0x3a400000) == 0x28000000;

#else
#error "signal_handler not implemented"
#endif

	const u64 exec64 = (reinterpret_cast<u64>(info->si_addr) - reinterpret_cast<u64>(vm::g_exec_addr)) / 2;
	const auto cause = is_executing ? "executing" : is_writing ? "writing" : "reading";

	if (auto [addr, ok] = vm::try_get_addr(info->si_addr); ok && !is_executing)
	{
		// Try to process access violation
		if (thread_ctrl::get_current() && handle_access_violation(addr, is_writing, context))
		{
			return;
		}
	}

	if (exec64 < 0x100000000ull && !is_executing)
	{
		if (thread_ctrl::get_current() && handle_access_violation(static_cast<u32>(exec64), is_writing, context))
		{
			return;
		}
	}

	std::string msg = fmt::format("Segfault %s location %p at %p.\n", cause, info->si_addr, RIP(context));

	append_thread_name(msg);

	sys_log.fatal("\n%s", msg);
	sys_log.notice("\n%s", dump_useful_thread_info());
	logs::listener::sync_all();

	if (IsDebuggerPresent())
	{
		// Convert to SIGTRAP
		raise(SIGTRAP);
		return;
	}

	thread_ctrl::emergency_exit(msg);
}

static void sigill_handler(int /*sig*/, siginfo_t* info, void* /*uct*/) noexcept
{
	std::string msg = fmt::format("Illegal instruction at %p (%s).\n", info->si_addr, *reinterpret_cast<be_t<u128>*>(info->si_addr));

	append_thread_name(msg);

	sys_log.fatal("\n%s", msg);
	sys_log.notice("\n%s", dump_useful_thread_info());
	logs::listener::sync_all();

	if (IsDebuggerPresent())
	{
		// Convert to SIGTRAP
		raise(SIGTRAP);
		return;
	}

	thread_ctrl::emergency_exit(msg);
}

void sigpipe_signaling_handler(int)
{
}

const bool s_exception_handler_set = []() -> bool
{
	struct ::sigaction sa;
	sa.sa_flags = SA_SIGINFO;
	sigemptyset(&sa.sa_mask);
	sa.sa_sigaction = signal_handler;

	if (::sigaction(SIGSEGV, &sa, NULL) == -1)
	{
		std::fprintf(stderr, "sigaction(SIGSEGV) failed (%d).\n", errno);
		std::abort();
	}

#ifdef __APPLE__
	if (::sigaction(SIGBUS, &sa, NULL) == -1)
	{
		std::fprintf(stderr, "sigaction(SIGBUS) failed (%d).\n", errno);
		std::abort();
	}
#endif

	sa.sa_sigaction = sigill_handler;
	if (::sigaction(SIGILL, &sa, NULL) == -1)
	{
		std::fprintf(stderr, "sigaction(SIGILL) failed (%d).\n", errno);
		std::abort();
	}

	sa.sa_handler = sigpipe_signaling_handler;
	if (::sigaction(SIGPIPE, &sa, NULL) == -1)
	{
		std::fprintf(stderr, "sigaction(SIGPIPE) failed (%d).\n", errno);
		std::abort();
	}

	std::printf("Debugger: %d\n", +IsDebuggerPresent());
	return true;
}();

#endif

const bool s_terminate_handler_set = []() -> bool
{
	std::set_terminate([]()
	{
		if (IsDebuggerPresent())
		{
			logs::listener::sync_all();
			utils::trap();
		}

		report_fatal_error("RPCS3 has abnormally terminated.");
	});

	return true;
}();

thread_local DECLARE(thread_ctrl::g_tls_this_thread) = nullptr;

thread_local DECLARE(thread_ctrl::g_tls_error_callback) = nullptr;

DECLARE(thread_ctrl::g_native_core_layout) { native_core_arrangement::undefined };

void thread_base::start()
{
	m_sync.atomic_op([&](u32& v)
	{
		v &= ~static_cast<u32>(thread_state::mask);
		v |= static_cast<u32>(thread_state::created);
	});

#ifdef _WIN32
	m_thread = ::_beginthreadex(nullptr, 0, entry_point, this, CREATE_SUSPENDED, nullptr);
	ensure(m_thread);
	ensure(::ResumeThread(reinterpret_cast<HANDLE>(+m_thread)) != static_cast<DWORD>(-1));
#elif defined(__APPLE__)
	pthread_attr_t stack_size_attr;
	pthread_attr_init(&stack_size_attr);
	pthread_attr_setstacksize(&stack_size_attr, 0x800000);
	ensure(pthread_create(reinterpret_cast<pthread_t*>(&m_thread.raw()), &stack_size_attr, entry_point, this) == 0);
#else
	ensure(pthread_create(reinterpret_cast<pthread_t*>(&m_thread.raw()), nullptr, entry_point, this) == 0);
#endif
}

void thread_base::initialize(void (*error_cb)())
{
#ifndef _WIN32
	m_thread.release(reinterpret_cast<u64>(pthread_self()));
#endif

	// Initialize TLS variables
	thread_ctrl::g_tls_this_thread = this;

	thread_ctrl::g_tls_error_callback = error_cb;

	g_tls_log_prefix = []
	{
		return thread_ctrl::get_name_cached();
	};

	atomic_wait_engine::set_wait_callback([](const void*, u64 attempts, u64 stamp0) -> bool
	{
		if (attempts == umax)
		{
			g_tls_wait_time += utils::get_tsc() - stamp0;
		}
		else if (attempts > 1)
		{
			g_tls_wait_fail += attempts - 1;
		}

		return true;
	});

	set_name(thread_ctrl::get_name_cached());
}

void thread_base::set_name(std::string name)
{
#ifdef _WIN32
	if (SetThreadDescriptionImport)
	{
		SetThreadDescriptionImport(GetCurrentThread(), utf8_to_wchar(name).c_str());
	}
#endif

#ifdef _MSC_VER
	struct THREADNAME_INFO
	{
		DWORD dwType;
		LPCSTR szName;
		DWORD dwThreadID;
		DWORD dwFlags;
	};

	// Set thread name for VS debugger
	if (IsDebuggerPresent()) [&]() NEVER_INLINE
	{
		THREADNAME_INFO info;
		info.dwType = 0x1000;
		info.szName = name.c_str();
		info.dwThreadID = -1;
		info.dwFlags = 0;

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

#if defined(__APPLE__)
	name.resize(std::min<usz>(15, name.size()));
	pthread_setname_np(name.c_str());
#elif defined(__DragonFly__) || defined(__FreeBSD__) || defined(__OpenBSD__)
	pthread_set_name_np(pthread_self(), name.c_str());
#elif defined(__NetBSD__)
	pthread_setname_np(pthread_self(), "%s", name.data());
#elif !defined(_WIN32)
	name.resize(std::min<usz>(15, name.size()));
	pthread_setname_np(pthread_self(), name.c_str());
#endif
}

u64 thread_base::finalize(thread_state result_state) noexcept
{
	// Report pending errors
	error_code::error_report(0, nullptr, nullptr, nullptr, nullptr);

#ifdef _WIN32
	static thread_local ULONG64 tls_cycles{};
	static thread_local u64 tls_time{};
	ULONG64 cycles{};
	QueryThreadCycleTime(GetCurrentThread(), &cycles);
	cycles -= tls_cycles;
	tls_cycles += cycles;
	FILETIME ctime, etime, ktime, utime;
	GetThreadTimes(GetCurrentThread(), &ctime, &etime, &ktime, &utime);
	const u64 time = ((ktime.dwLowDateTime | static_cast<u64>(ktime.dwHighDateTime) << 32) + (utime.dwLowDateTime | static_cast<u64>(utime.dwHighDateTime) << 32)) * 100ull - tls_time;
	tls_time += time;
	const u64 fsoft = 0;
	const u64 fhard = 0;
	const u64 ctxvol = 0;
	const u64 ctxinv = 0;
#elif defined(RUSAGE_THREAD)
	static thread_local u64 tls_time{}, tls_fsoft{}, tls_fhard{}, tls_ctxvol{}, tls_ctxinv{};
	const u64 cycles = 0; // Not supported
	struct ::rusage stats{};
	::getrusage(RUSAGE_THREAD, &stats);
	const u64 time = (stats.ru_utime.tv_sec + stats.ru_stime.tv_sec) * 1000000000ull + (stats.ru_utime.tv_usec + stats.ru_stime.tv_usec) * 1000ull - tls_time;
	tls_time += time;
	const u64 fsoft = stats.ru_minflt - tls_fsoft;
	tls_fsoft += fsoft;
	const u64 fhard = stats.ru_majflt - tls_fhard;
	tls_fhard += fhard;
	const u64 ctxvol = stats.ru_nvcsw - tls_ctxvol;
	tls_ctxvol += ctxvol;
	const u64 ctxinv = stats.ru_nivcsw - tls_ctxinv;
	tls_ctxinv += ctxinv;
#else
	const u64 cycles = 0;
	const u64 time = 0;
	const u64 fsoft = 0;
	const u64 fhard = 0;
	const u64 ctxvol = 0;
	const u64 ctxinv = 0;
#endif

	g_tls_log_prefix = []
	{
		return thread_ctrl::get_name_cached();
	};

	const bool is_cpu_thread = !!cpu_thread::get_current();
	auto& thread_log = (is_cpu_thread || g_tls_fault_all ? sig_log.notice : sig_log.trace);

	thread_log("Thread time: %fs (%fGc); Faults: %u [rsx:%u, spu:%u]; [soft:%u hard:%u]; Switches:[vol:%u unvol:%u]; Wait:[%.3fs, spur:%u]",
		time / 1000000000.,
		cycles / 1000000000.,
		g_tls_fault_all,
		g_tls_fault_rsx,
		g_tls_fault_spu,
		fsoft, fhard, ctxvol, ctxinv,
		g_tls_wait_time / (utils::get_tsc_freq() / 1.),
		g_tls_wait_fail);

	atomic_wait_engine::set_wait_callback(nullptr);

	// Avoid race with the destructor
	const u64 _self = m_thread;

	// Set result state (errored or finalized)
	m_sync.fetch_op([&](u32& v)
	{
		v &= -4;
		v |= static_cast<u32>(result_state);
	});

	// Signal waiting threads
	m_sync.notify_all();

	return _self;
}

thread_base::native_entry thread_base::finalize(u64 _self) noexcept
{
	g_tls_fault_all = 0;
	g_tls_fault_rsx = 0;
	g_tls_fault_spu = 0;
	g_tls_wait_time = 0;
	g_tls_wait_fail = 0;
	g_tls_access_violation_recovered = false;

	g_tls_log_prefix = []() -> std::string { return {}; };

	if (_self == umax)
	{
		thread_ctrl::g_tls_this_thread = nullptr;
		return nullptr;
	}

#ifdef _WIN32
	_endthreadex(0);
#else
	pthread_exit(nullptr);
#endif

	return nullptr;
}

thread_base::native_entry thread_base::make_trampoline(u64(*entry)(thread_base* _base))
{
	return build_function_asm<native_entry>("", [&](native_asm& c, auto& args)
	{
		using namespace asmjit;

#if defined(ARCH_X64)
		Label _ret = c.newLabel();
		c.push(x86::rbp);
		c.sub(x86::rsp, 0x20);

		// Call entry point (TODO: support for detached threads missing?)
		c.call(entry);

		// Call finalize, return if zero
		c.mov(args[0], x86::rax);
		c.call(static_cast<native_entry(*)(u64)>(&finalize));
		c.test(x86::rax, x86::rax);
		c.jz(_ret);

		// Otherwise, call it as an entry point with first arg = new current thread
		c.mov(x86::rbp, x86::rax);
		c.call(thread_ctrl::get_current);
		c.mov(args[0], x86::rax);
		c.add(x86::rsp, 0x28);
		c.jmp(x86::rbp);

		c.bind(_ret);
		c.add(x86::rsp, 0x28);
		c.ret();
#endif
	});
}

thread_state thread_ctrl::state()
{
	auto _this = g_tls_this_thread;

	// Guard for recursive calls (TODO: may be more effective to reuse one of m_sync bits)
	static thread_local bool s_tls_exec = false;

	// Drain execution queue
	if (!s_tls_exec)
	{
		s_tls_exec = true;
		_this->exec();
		s_tls_exec = false;
	}

	return static_cast<thread_state>(_this->m_sync & 3);
}

void thread_ctrl::wait_for(u64 usec, [[maybe_unused]] bool alert /* true */)
{
	if (!usec)
	{
		return;
	}

	auto _this = g_tls_this_thread;

	if (!alert && usec > 50000)
	{
		usec = 50000;
	}

#ifdef __linux__
	static thread_local struct linux_timer_handle_t
	{
		// Allocate timer only if needed (i.e. someone calls wait_for with alert and short period)
		const int m_timer = timerfd_create(CLOCK_MONOTONIC, 0);

		linux_timer_handle_t() noexcept
		{
			if (m_timer == -1)
			{
				sig_log.error("Linux timer allocation failed, using the fallback instead.");
			}
		}

		operator int() const
		{
			return m_timer;
		}

		~linux_timer_handle_t()
		{
			if (m_timer != -1)
			{
				close(m_timer);
			}
		}
	} fd_timer;

	if (!alert && fd_timer != -1)
	{
		struct itimerspec timeout;
		u64 missed;

		timeout.it_value.tv_nsec = usec % 1'000'000 * 1'000ull;
		timeout.it_value.tv_sec = usec / 1'000'000;
		timeout.it_interval.tv_sec = 0;
		timeout.it_interval.tv_nsec = 0;
		timerfd_settime(fd_timer, 0, &timeout, NULL);
		if (read(fd_timer, &missed, sizeof(missed)) != sizeof(missed))
			sig_log.error("timerfd: read() failed");
		return;
	}
#endif

	if (alert)
	{
		if (_this->m_sync.bit_test_reset(2) || _this->m_taskq)
		{
			return;
		}
	}

	// Wait for signal and thread state abort
	atomic_wait::list<2> list{};

	if (alert)
	{
		list.set<0>(_this->m_sync, 0);
		list.set<1>(utils::bless<atomic_t<u32>>(&_this->m_taskq)[1], 0);
	}
	else
	{
		list.set<0>(_this->m_dummy, 0);
	}

	list.wait(atomic_wait_timeout{usec <= 0xffff'ffff'ffff'ffff / 1000 ? usec * 1000 : 0xffff'ffff'ffff'ffff});
}

void thread_ctrl::wait_for_accurate(u64 usec)
{
	if (!usec)
	{
		return;
	}

	if (usec > 50000)
	{
		fmt::throw_exception("thread_ctrl::wait_for_accurate: unsupported amount");
	}

#ifdef __linux__
	return wait_for(usec, false);
#else
	using namespace std::chrono_literals;

	const auto until = std::chrono::steady_clock::now() + 1us * usec;

	while (true)
	{
		// Host scheduler quantum for windows (worst case)
		constexpr u64 host_min_quantum = 500;

		if (usec >= host_min_quantum)
		{
			// Wait on multiple of min quantum for large durations to avoid overloading low thread cpus
			wait_for(usec - (usec % host_min_quantum), false);
		}
		// TODO: Determine best value for yield delay
		else if (usec >= host_min_quantum / 2)
		{
			std::this_thread::yield();
		}
		else
		{
			busy_wait(100);
		}

		const auto current = std::chrono::steady_clock::now();

		if (current >= until)
		{
			break;
		}

		usec = std::chrono::duration_cast<std::chrono::microseconds>(until - current).count();
	}
#endif
}

std::string thread_ctrl::get_name_cached()
{
	auto _this = thread_ctrl::g_tls_this_thread;

	if (!_this)
	{
		return {};
	}

	static thread_local shared_ptr<std::string> name_cache;

	if (!_this->m_tname.is_equal(name_cache)) [[unlikely]]
	{
		_this->m_tname.peek_op([&](const shared_ptr<std::string>& ptr)
		{
			if (ptr != name_cache)
			{
				name_cache = ptr;
			}
		});
	}

	return *name_cache;
}

thread_base::thread_base(native_entry entry, std::string name)
	: entry_point(entry)
	, m_tname(make_single_value(std::move(name)))
{
}

thread_base::~thread_base()
{
	// Cleanup abandoned tasks: initialize default results and signal
	this->exec();

	// Cleanup
	{
#ifdef _WIN32
		const HANDLE handle0 = reinterpret_cast<HANDLE>(m_thread.load());
		WaitForSingleObject(handle0, INFINITE);
		CloseHandle(handle0);
#else
		pthread_join(reinterpret_cast<pthread_t>(m_thread.load()), nullptr);
#endif
	}
}

bool thread_base::join(bool dtor) const
{
	// Check if already finished
	if (m_sync & 2)
	{
		return (m_sync & 3) == 3;
	}

	// Hacked for too sleepy threads (1ms) TODO: make sure it's unneeded and remove
	const auto timeout = dtor && Emu.IsStopped() ? atomic_wait_timeout{1'000'000} : atomic_wait_timeout::inf;

	auto stamp0 = utils::get_tsc();

	for (u64 i = 0; (m_sync & 3) <= 1; i++)
	{
		m_sync.wait(m_sync & ~2, timeout);

		if (m_sync & 2)
		{
			break;
		}

		if (i >= 16 && !(i & (i - 1)) && timeout != atomic_wait_timeout::inf)
		{
			sig_log.error(u8"Thread [%s] is too sleepy. Waiting for it %.3fµs already!", *m_tname.load(), (utils::get_tsc() - stamp0) / (utils::get_tsc_freq() / 1000000.));
		}
	}

	return (m_sync & 3) == 3;
}

void thread_base::notify()
{
	// Set notification
	m_sync |= 4;
	m_sync.notify_all();
}

u64 thread_base::get_native_id() const
{
#ifdef _WIN32
	return GetThreadId(reinterpret_cast<HANDLE>(m_thread.load()));
#else
	return m_thread.load();
#endif
}

u64 thread_base::get_cycles()
{
	u64 cycles = 0;

	const u64 handle = m_thread;

#ifdef _WIN32
	if (QueryThreadCycleTime(reinterpret_cast<HANDLE>(handle), &cycles))
	{
#elif __APPLE__
	mach_port_name_t port = pthread_mach_thread_np(reinterpret_cast<pthread_t>(handle));
	mach_msg_type_number_t count = THREAD_BASIC_INFO_COUNT;
	thread_basic_info_data_t info;
	kern_return_t ret = thread_info(port, THREAD_BASIC_INFO, reinterpret_cast<thread_info_t>(&info), &count);
	if (ret == KERN_SUCCESS)
	{
		cycles = static_cast<u64>(info.user_time.seconds + info.system_time.seconds) * 1'000'000'000 +
			static_cast<u64>(info.user_time.microseconds + info.system_time.microseconds) * 1'000;
#else
	clockid_t _clock;
	struct timespec thread_time;
	if (!pthread_getcpuclockid(reinterpret_cast<pthread_t>(handle), &_clock) && !clock_gettime(_clock, &thread_time))
	{
		cycles = static_cast<u64>(thread_time.tv_sec) * 1'000'000'000 + thread_time.tv_nsec;
#endif
		if (const u64 old_cycles = m_cycles.exchange(cycles))
		{
			return cycles - old_cycles;
		}

		// Report 0 the first time this function is called
		return 0;
	}
	else
	{
		return m_cycles;
	}
}

void thread_base::push(shared_ptr<thread_future> task)
{
	const auto next = &task->next;
	m_taskq.push_head(*next, std::move(task));
	m_taskq.notify_one();
}

void thread_base::exec()
{
	if (!m_taskq) [[likely]]
	{
		return;
	}

	while (shared_ptr<thread_future> head = m_taskq.exchange(null_ptr))
	{
		// TODO: check if adapting reverse algorithm is feasible here
		thread_future* prev_head{head.get()};

		for (thread_future* prev{};;)
		{
			utils::prefetch_exec(prev_head->exec.load());

			if (auto next = prev_head->next.get())
			{
				prev = std::exchange(prev_head, next);
				prev_head->prev = prev;
			}
			else
			{
				break;
			}
		}

		for (auto ptr = prev_head; ptr; ptr = ptr->prev)
		{
			if (auto task = ptr->exec.load()) [[likely]]
			{
				// Execute or discard (if aborting)
				if ((m_sync & 3) == 0) [[likely]]
				{
					task(this, ptr);
				}
				else
				{
					task(nullptr, ptr);
				}

				// Notify waiters
				ptr->done.release(1);
				ptr->done.notify_all();
			}

			if (ptr->next)
			{
				// Partial cleanup
				ptr->next.reset();
			}
		}

		if (!m_taskq) [[likely]]
		{
			return;
		}
	}
}

[[noreturn]] void thread_ctrl::emergency_exit(std::string_view reason)
{
	if (std::string info = dump_useful_thread_info(); !info.empty())
	{
		sys_log.notice("\n%s", info);
	}

	std::string reason_buf;

	if (auto ppu = cpu_thread::get_current<ppu_thread>())
	{
		if (auto func = ppu->current_function)
		{
			fmt::append(reason_buf, "%s (PPU: %s)", reason, func);
		}
	}

	if (!reason_buf.empty())
	{
		reason = reason_buf;
	}

	sig_log.fatal("Thread terminated due to fatal error: %s", reason);

	logs::listener::sync_all();

	if (IsDebuggerPresent())
	{
		utils::trap();
	}

	if (const auto _this = g_tls_this_thread)
	{
		g_tls_error_callback();

		u64 _self = _this->finalize(thread_state::errored);

		if (_self == umax)
		{
			// Unused, detached thread support remnant
			delete _this;
		}

		thread_base::finalize(umax);

#ifdef _WIN32
		_endthreadex(0);
#else
		pthread_exit(nullptr);
#endif
	}

	report_fatal_error(reason);
}

void thread_ctrl::detect_cpu_layout()
{
	if (!g_native_core_layout.compare_and_swap_test(native_core_arrangement::undefined, native_core_arrangement::generic))
		return;

	const auto system_id = utils::get_cpu_brand();
	if (system_id.find("Ryzen") != umax)
	{
		g_native_core_layout.store(native_core_arrangement::amd_ccx);
	}
	else if (system_id.find("Intel") != umax)
	{
#ifdef _WIN32
		const LOGICAL_PROCESSOR_RELATIONSHIP relationship = LOGICAL_PROCESSOR_RELATIONSHIP::RelationProcessorCore;
		DWORD buffer_size = 0;

		// If buffer size is set to 0 bytes, it will be overwritten with the required size
		if (GetLogicalProcessorInformationEx(relationship, nullptr, &buffer_size))
		{
			sig_log.error("GetLogicalProcessorInformationEx returned 0 bytes");
			return;
		}
		DWORD error_code = GetLastError();
		if (error_code != ERROR_INSUFFICIENT_BUFFER)
		{
			sig_log.error("Unexpected windows error code when detecting CPU layout: %u", error_code);
			return;
		}

		std::vector<u8> buffer(buffer_size);

		if (!GetLogicalProcessorInformationEx(relationship,
			reinterpret_cast<SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *>(buffer.data()), &buffer_size))
		{
			sig_log.error("GetLogicalProcessorInformationEx failed (size=%u, error=%s)", buffer_size, fmt::win_error{GetLastError(), nullptr});
		}
		else
		{
			// Iterate through the buffer until a core with hyperthreading is found
			auto ptr = reinterpret_cast<uptr>(buffer.data());
			const uptr end = ptr + buffer_size;

			while (ptr < end)
			{
				auto info = reinterpret_cast<SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *>(ptr);
				if (info->Relationship == relationship && info->Processor.Flags == LTP_PC_SMT)
				{
					g_native_core_layout.store(native_core_arrangement::intel_ht);
					break;
				}
				ptr += info->Size;
			}
		}
#else
		sig_log.todo("Thread scheduler is not implemented for Intel and this OS");
#endif
	}
}

u64 thread_ctrl::get_affinity_mask(thread_class group)
{
	detect_cpu_layout();

	if (const auto thread_count = utils::get_thread_count())
	{
		const u64 all_cores_mask = process_affinity_mask;

		switch (g_native_core_layout)
		{
		default:
		case native_core_arrangement::generic:
		{
			return all_cores_mask;
		}
		case native_core_arrangement::amd_ccx:
		{
			if (thread_count <= 8)
			{
				// Single CCX or not enough threads, do nothing
				return all_cores_mask;
			}

			u64 spu_mask, ppu_mask, rsx_mask;
			spu_mask = ppu_mask = rsx_mask = all_cores_mask; // Fallback, in case someone is messing with core config

			const auto system_id = utils::get_cpu_brand();
			const auto family_id = utils::get_cpu_family();
			const auto model_id = utils::get_cpu_model();

			switch (family_id)
			{
			case 0x17: // Zen, Zen+, Zen2
			case 0x18: // Dhyana core (Zen)
			{
				if (model_id > 0x30)
				{
					// Zen2 (models 49, 96, 113, 144)
					// Much improved inter-CCX latency
					switch (thread_count)
					{
					case 128:
					case 64:
					case 48:
					case 32:
						// TR 3000 series, or R9 3950X, Assign threads 9-32
						ppu_mask = 0b11111111000000000000000000000000;
						spu_mask = 0b00000000111111110000000000000000;
						rsx_mask = 0b00000000000000001111111100000000;
						break;
					case 24:
						// 3900X, Assign threads 7-24
						ppu_mask = 0b111111000000000000000000;
						spu_mask = 0b000000111111000000000000;
						rsx_mask = 0b000000000000111111000000;
						break;
					case 16:
						// 3700, 3800 family, Assign threads 1-16
						ppu_mask = 0b0000000011110000;
						spu_mask = 0b1111111100000000;
						rsx_mask = 0b0000000000001111;
						break;
					case 12:
						// 3600 family, Assign threads 1-12
						ppu_mask = 0b000000111000;
						spu_mask = 0b111111000000;
						rsx_mask = 0b000000000111;
						break;
					default:
						break;
					}
				}
				else
				{
					// Zen, Zen+ (models 1, 8(+), 17, 24(+), 32)
					switch (thread_count)
					{
					case 64:
						// TR 2990WX, Assign threads 17-32
						ppu_mask = 0b00000000111111110000000000000000;
						spu_mask = ppu_mask;
						rsx_mask = 0b11111111000000000000000000000000;
						break;
					case 48:
						// TR 2970WX, Assign threads 9-24
						ppu_mask = 0b000000111111000000000000;
						spu_mask = ppu_mask;
						rsx_mask = 0b111111000000000000000000;
						break;
					case 32:
						// TR 2950X, TR 1950X, Assign threads 17-32
						ppu_mask = 0b00000000111111110000000000000000;
						spu_mask = ppu_mask;
						rsx_mask = 0b11111111000000000000000000000000;
						break;
					case 24:
						// TR 1920X, 2920X, Assign threads 13-24
						ppu_mask = 0b000000111111000000000000;
						spu_mask = ppu_mask;
						rsx_mask = 0b111111000000000000000000;
						break;
					case 16:
						// 1700, 1800, 2700, TR 1900X family
						if (g_cfg.core.thread_scheduler == thread_scheduler_mode::alt)
						{
							ppu_mask = 0b0010000010000000;
							spu_mask = 0b0000101010101010;
							rsx_mask = 0b1000000000000000;
						}
						else
						{
							ppu_mask = 0b1111111100000000;
							spu_mask = ppu_mask;
							rsx_mask = 0b0000000000111100;
						}
						break;
					case 12:
						// 1600, 2600 family, Assign threads 3-12
						ppu_mask = 0b111111000000;
						spu_mask = ppu_mask;
						rsx_mask = 0b000000111100;
						break;
					default:
						break;
					}
				}
				break;
			}
			case 0x19: // Zen3
			{
				// Single-CCX architecture, just disable SMT if wide enough
				// CCX now holds upto 16 threads
				// Lack of hw availability makes testing difficult
				switch (thread_count)
				{
				case 24:
					// 5900X, Use same scheduler as 3900X
					// Unverified on windows, may be worse than just disabling SMT and scheduler
					ppu_mask = 0b111111000000000000000000;
					spu_mask = 0b000000111111000000000000;
					rsx_mask = 0b000000000000111111000000;
					break;
				case 16:
					// 5800X
					if (g_cfg.core.thread_scheduler == thread_scheduler_mode::alt)
					{
						ppu_mask = 0b0000000011110000;
						spu_mask = 0b1111111100000000;
						rsx_mask = 0b0000000000001111;
					}
					else
					{
						// Verified by more than one windows user on 16-thread CPU
						ppu_mask = spu_mask = rsx_mask = (0b10101010101010101010101010101010 & all_cores_mask);
					}
					break;
				case 12:
					// 5600X
					if (g_cfg.core.thread_scheduler == thread_scheduler_mode::alt)
					{
						ppu_mask = 0b000000001100;
						spu_mask = 0b111111110000;
						rsx_mask = 0b000000000011;
					}
					else
					{
						ppu_mask = spu_mask = rsx_mask = all_cores_mask;
					}
					break;
				default:
					if (thread_count > 24)
					{
						ppu_mask = spu_mask = rsx_mask = (0b10101010101010101010101010101010 & all_cores_mask);
					}
					break;
				}
				break;
			}
			default:
			{
				break;
			}
			}

			switch (group)
			{
			default:
			case thread_class::general:
				return all_cores_mask;
			case thread_class::rsx:
				return rsx_mask;
			case thread_class::ppu:
				return ppu_mask;
			case thread_class::spu:
				return spu_mask;
			}
		}
		case native_core_arrangement::intel_ht:
		{
			if (thread_count >= 12 && g_cfg.core.thread_scheduler == thread_scheduler_mode::alt)
				return (0b10101010101010101010101010101010 & all_cores_mask); // Potentially improves performance by mimicking HT off
			return all_cores_mask;
		}
		}
	}

	return -1;
}

void thread_ctrl::set_native_priority(int priority)
{
#ifdef _WIN32
	HANDLE _this_thread = GetCurrentThread();
	INT native_priority = THREAD_PRIORITY_NORMAL;

	if (priority > 0)
		native_priority = THREAD_PRIORITY_ABOVE_NORMAL;
	if (priority < 0)
		native_priority = THREAD_PRIORITY_BELOW_NORMAL;

	if (!SetThreadPriority(_this_thread, native_priority))
	{
		sig_log.error("SetThreadPriority() failed: %s", fmt::win_error{GetLastError(), nullptr});
	}
#else
	int policy;
	struct sched_param param;

	pthread_getschedparam(pthread_self(), &policy, &param);

	if (priority > 0)
		param.sched_priority = sched_get_priority_max(policy);
	if (priority < 0)
		param.sched_priority = sched_get_priority_min(policy);

	if (int err = pthread_setschedparam(pthread_self(), policy, &param))
	{
		sig_log.error("pthread_setschedparam() failed: %d", err);
	}
#endif
}

u64 thread_ctrl::get_process_affinity_mask()
{
	static const u64 mask = []() -> u64
	{
#ifdef _WIN32
		DWORD_PTR res, _sys;
		if (!GetProcessAffinityMask(GetCurrentProcess(), &res, &_sys))
		{
			sig_log.error("Failed to get process affinity mask.");
			return 0;
		}

		return res;
#else
		// Assume it's called from the main thread (this is a bit shaky)
		return thread_ctrl::get_thread_affinity_mask();
#endif
	}();

	return mask;
}

DECLARE(thread_ctrl::process_affinity_mask) = get_process_affinity_mask();

void thread_ctrl::set_thread_affinity_mask(u64 mask)
{
	sig_log.trace("set_thread_affinity_mask called with mask=0x%x", mask);

#ifdef _WIN32
	HANDLE _this_thread = GetCurrentThread();
	if (!SetThreadAffinityMask(_this_thread, !mask ? process_affinity_mask : mask))
	{
		sig_log.error("Failed to set thread affinity 0x%x: error: %s", mask, fmt::win_error{GetLastError(), nullptr});
	}
#elif __APPLE__
	// Supports only one core
	thread_affinity_policy_data_t policy = { static_cast<integer_t>(std::countr_zero(mask)) };
	thread_port_t mach_thread = pthread_mach_thread_np(pthread_self());
	thread_policy_set(mach_thread, THREAD_AFFINITY_POLICY, reinterpret_cast<thread_policy_t>(&policy), !mask ? 0 : 1);
#elif defined(__linux__) || defined(__DragonFly__) || defined(__FreeBSD__)
	if (!mask)
	{
		// Reset affinity mask
		mask = process_affinity_mask;
	}

	cpu_set_t cs;
	CPU_ZERO(&cs);

	for (u32 core = 0; core < 64u; ++core)
	{
		const u64 shifted = mask >> core;

		if (shifted & 1)
		{
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wold-style-cast"
			CPU_SET(core, &cs);
#pragma GCC diagnostic pop
		}

		if (shifted <= 1)
		{
			break;
		}
	}

	if (int err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cs))
	{
		sig_log.error("Failed to set thread affinity 0x%x: error %d.", mask, err);
	}
#endif
}

u64 thread_ctrl::get_thread_affinity_mask()
{
#ifdef _WIN32
	const u64 res = process_affinity_mask;

	if (DWORD_PTR result = SetThreadAffinityMask(GetCurrentThread(), res))
	{
		if (res != result)
		{
			SetThreadAffinityMask(GetCurrentThread(), result);
		}

		return result;
	}

	sig_log.error("Failed to get thread affinity mask.");
	return 0;
#elif defined(__linux__) || defined(__DragonFly__) || defined(__FreeBSD__)
	cpu_set_t cs;
	CPU_ZERO(&cs);

	if (int err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), &cs))
	{
		sig_log.error("Failed to get thread affinity mask: error %d.", err);
		return 0;
	}

	u64 result = 0;

	for (u32 core = 0; core < 64u; core++)
	{
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wold-style-cast"
		if (CPU_ISSET(core, &cs))
#pragma GCC diagnostic pop
		{
			result |= 1ull << core;
		}
	}

	if (result == 0)
	{
		sig_log.error("Thread affinity mask is out of u64 range.");
		return 0;
	}

	return result;
#else
	return -1;
#endif
}

std::pair<void*, usz> thread_ctrl::get_thread_stack()
{
#ifdef _WIN32
	ULONG_PTR _min = 0;
	ULONG_PTR _max = 0;
	GetCurrentThreadStackLimits(&_min, &_max);
	const usz ssize = _max - _min;
	const auto saddr = reinterpret_cast<void*>(_min);
#else
	void* saddr = 0;
	usz ssize = 0;
#if defined(__linux__)
	pthread_attr_t attr;
	pthread_getattr_np(pthread_self(), &attr);
	pthread_attr_getstack(&attr, &saddr, &ssize);
#elif defined(__APPLE__)
	saddr = pthread_get_stackaddr_np(pthread_self());
	ssize = pthread_get_stacksize_np(pthread_self());
#else
	pthread_attr_t attr;
	pthread_attr_get_np(pthread_self(), &attr);
	pthread_attr_getstackaddr(&attr, &saddr);
	pthread_attr_getstacksize(&attr, &ssize);
#endif
#endif
	return {saddr, ssize};
}

u64 thread_ctrl::get_tid()
{
#ifdef _WIN32
	return GetCurrentThreadId();
#elif defined(__linux__)
	return syscall(SYS_gettid);
#else
	return reinterpret_cast<u64>(pthread_self());
#endif
}

bool thread_ctrl::is_main()
{
	return get_tid() == utils::main_tid;
}