x86 stack unwinding support (android)

core/linux/unwind_info.cpp

@@ -55,6 +55,14 @@ constexpr int dwarfRegRAId = 16;
 constexpr int dwarfRegXmmId = 17;
 constexpr int dwarfRegSP = dwarfRegId[4];	// RSP
 
+#elif HOST_CPU == CPU_X86
+
+inline static int registerId(int x86Id) {
+	return x86Id;
+}
+constexpr int dwarfRegRAId = 8;
+constexpr int dwarfRegSP = 4;	// ESP
+
 #elif HOST_CPU == CPU_ARM64
 // https://developer.arm.com/documentation/ihi0057/latest
 //
@@ -71,7 +79,7 @@ constexpr int dwarfRegSP = 31;
 
 #endif
 
-#if HOST_CPU == CPU_X64 || HOST_CPU == CPU_ARM64
+#if HOST_CPU == CPU_X64 || HOST_CPU == CPU_ARM64 || HOST_CPU == CPU_X86
 
 using ByteStream = std::vector<u8>;
 
@@ -120,10 +128,10 @@ static void writeSLEB128(ByteStream &stream, int32_t v)
 
 static void writePadding(ByteStream &stream)
 {
-	int padding = stream.size() % 8;
+	int padding = stream.size() % sizeof(uintptr_t);
 	if (padding != 0)
 	{
-		padding = 8 - padding;
+		padding = sizeof(uintptr_t) - padding;
 		for (int i = 0; i < padding; i++)
 			write<u8>(stream, 0);
 	}
@@ -160,8 +168,8 @@ static void writeFDE(ByteStream &stream, const ByteStream &fdeInstructions, u32
 	write<u32>(stream, offsetToCIE);
 
 	functionStart = stream.size();
-	write<u64>(stream, 0); // func start
+	write<uintptr_t>(stream, 0); // func start
-	write<u64>(stream, 0); // func size
+	write<uintptr_t>(stream, 0); // func size
 
 	writeULEB128(stream, 0); // LEB128 augmentation size
 
@@ -171,9 +179,9 @@ static void writeFDE(ByteStream &stream, const ByteStream &fdeInstructions, u32
 	writeLength(stream, lengthPos, stream.size() - lengthPos - 4);
 }
 
-static void writeAdvanceLoc(ByteStream &fdeInstructions, u64 offset, u64 &lastOffset)
+static void writeAdvanceLoc(ByteStream &fdeInstructions, uintptr_t offset, uintptr_t &lastOffset)
 {
-	u64 delta = offset - lastOffset;
+	uintptr_t delta = offset - lastOffset;
 	if (delta == 0)
 		return;
 	if (delta < (1 << 6))
@@ -227,8 +235,8 @@ static void writeRegisterStackLocationExtended(ByteStream &instructions, int dwa
 void UnwindInfo::start(void *address)
 {
 	startAddr = (u8 *)address;
-#if HOST_CPU == CPU_X64
+#if HOST_CPU == CPU_X64 || HOST_CPU == CPU_X86
-	stackOffset = 8;
+	stackOffset = sizeof(uintptr_t);
 #else
 	stackOffset = 0;
 #endif
@@ -243,7 +251,7 @@ void UnwindInfo::start(void *address)
 
 void UnwindInfo::pushReg(u32 offset, int reg)
 {
-	stackOffset += 8;
+	stackOffset += sizeof(uintptr_t);
 	writeAdvanceLoc(fdeInstructions, offset, lastOffset);
 	writeDefineStackOffset(fdeInstructions, stackOffset);
 	writeRegisterStackLocation(fdeInstructions, registerId(reg), stackOffset);
@@ -289,7 +297,7 @@ size_t UnwindInfo::end(u32 offset, ptrdiff_t rwRxOffset)
 
 	if (!unwindInfo.empty())
 	{
-		u64 *unwindfuncaddr = (u64 *)(unwindInfoDest + functionStart);
+		uintptr_t *unwindfuncaddr = (uintptr_t *)(unwindInfoDest + functionStart);
 		unwindfuncaddr[0] = (uintptr_t)startAddr + rwRxOffset;
 		unwindfuncaddr[1] = (ptrdiff_t)(endAddr - startAddr);
 

core/oslib/oslib.h

@@ -76,14 +76,14 @@ private:
 #endif
 #if defined(__unix__) || defined(__APPLE__) || defined(__SWITCH__)
 	int stackOffset = 0;
-	u64 lastOffset = 0;
+	uintptr_t lastOffset = 0;
 	std::vector<u8> cieInstructions;
 	std::vector<u8> fdeInstructions;
 	std::vector<u8 *> registeredFrames;
 #endif
 };
 
-#if HOST_CPU != CPU_X64 && HOST_CPU != CPU_ARM64
+#if HOST_CPU != CPU_X64 && HOST_CPU != CPU_ARM64 && (HOST_CPU != CPU_X86 || defined(_WIN32))
 inline void UnwindInfo::start(void *address) {
 }
 inline void UnwindInfo::pushReg(u32 offset, int reg) {

core/rec-x86/rec_x86.cpp

@@ -26,6 +26,7 @@
 #include "hw/sh4/sh4_interrupts.h"
 #include "hw/sh4/sh4_mem.h"
 #include "hw/mem/_vmem.h"
+#include "oslib/oslib.h"
 
 static int cycle_counter;
 static void (*mainloop)();
@@ -43,6 +44,7 @@ static X86Compiler* compiler;
 static Xbyak::Operand::Code alloc_regs[] {  Xbyak::Operand::EBX,  Xbyak::Operand::EBP,  Xbyak::Operand::ESI,  Xbyak::Operand::EDI, (Xbyak::Operand::Code)-1 };
 static s8 alloc_fregs[] = { 7, 6, 5, 4, -1 };
 alignas(16) static f32 thaw_regs[4];
+static UnwindInfo unwinder;
 
 void X86RegAlloc::doAlloc(RuntimeBlockInfo* block)
 {
@@ -359,14 +361,21 @@ void X86Compiler::thawXMM()
 
 void X86Compiler::genMainloop()
 {
+	unwinder.start((void *)getCurr());
 	push(esi);
+	unwinder.pushReg(getSize(), Xbyak::Operand::ESI);
 	push(edi);
+	unwinder.pushReg(getSize(), Xbyak::Operand::EDI);
 	push(ebp);
+	unwinder.pushReg(getSize(), Xbyak::Operand::EBP);
 	push(ebx);
+	unwinder.pushReg(getSize(), Xbyak::Operand::EBX);
 #ifndef _WIN32
 	// 16-byte alignment
 	sub(esp, 12);
+	unwinder.allocStack(getSize(), 12);
 #endif
+	unwinder.endProlog(getSize());
 
 	mov(ecx, dword[&Sh4cntx.pc]);
 
@@ -374,7 +383,6 @@ void X86Compiler::genMainloop()
 
 	mov(eax, 0);
 	//next_pc _MUST_ be on ecx
-	Xbyak::Label do_iter;
 	Xbyak::Label cleanup;
 //no_update:
 	Xbyak::Label no_updateLabel;
@@ -384,25 +392,6 @@ void X86Compiler::genMainloop()
 	and_(ecx, RAM_SIZE_MAX - 2);
 	jmp(dword[eax + ecx * 2]);
 
-//intc_sched:
-	Xbyak::Label intc_schedLabel;
-	L(intc_schedLabel);
-	add(dword[&cycle_counter], SH4_TIMESLICE);
-	call((void *)UpdateSystem);
-	cmp(eax, 0);
-	jnz(do_iter);
-	ret();
-
-//do_iter:
-	L(do_iter);
-	pop(ecx);
-	call((void *)rdv_DoInterrupts);
-	mov(ecx, eax);
-	mov(edx, dword[&Sh4cntx.CpuRunning]);
-	cmp(edx, 0);
-	jz(cleanup);
-	jmp(no_updateLabel);
-
 //cleanup:
 	L(cleanup);
 #ifndef _WIN32
@@ -416,6 +405,17 @@ void X86Compiler::genMainloop()
 
 	ret();
 
+//do_iter:
+	Xbyak::Label do_iter;
+	L(do_iter);
+	pop(ecx);
+	call((void *)rdv_DoInterrupts);
+	mov(ecx, eax);
+	mov(edx, dword[&Sh4cntx.CpuRunning]);
+	cmp(edx, 0);
+	jz(cleanup);
+	jmp(no_updateLabel);
+
 //ngen_LinkBlock_Shared_stub:
 	Xbyak::Label ngen_LinkBlock_Shared_stub;
 	L(ngen_LinkBlock_Shared_stub);
@@ -424,6 +424,23 @@ void X86Compiler::genMainloop()
 	call((void *)rdv_LinkBlock);
 	jmp(eax);
 
+	size_t unwindSize = unwinder.end(getSize());
+	setSize(getSize() + unwindSize);
+
+	// Functions called by blocks
+
+//intc_sched:
+	unwinder.start((void *)getCurr());
+	size_t startOffset = getSize();
+	unwinder.endProlog(0);
+	Xbyak::Label intc_schedLabel;
+	L(intc_schedLabel);
+	add(dword[&cycle_counter], SH4_TIMESLICE);
+	call((void *)UpdateSystem);
+	cmp(eax, 0);
+	jnz(do_iter);
+	ret();
+
 //ngen_LinkBlock_cond_Next_stub:
 	Xbyak::Label ngen_LinkBlock_cond_Next_label;
 	L(ngen_LinkBlock_cond_Next_label);
@@ -442,6 +459,25 @@ void X86Compiler::genMainloop()
 	mov(edx, dword[&Sh4cntx.jdyn]);
 	jmp(ngen_LinkBlock_Shared_stub);
 
+	genMemHandlers();
+
+	unwindSize = unwinder.end(getSize() - startOffset);
+	setSize(getSize() + unwindSize);
+
+	// The following code and all code blocks use the same stack frame as mainloop()
+	// (direct jump from there or from a block)
+	unwinder.start((void *)getCurr());
+	startOffset = getSize();
+	unwinder.pushReg(0, Xbyak::Operand::ESI);
+	unwinder.pushReg(0, Xbyak::Operand::EDI);
+	unwinder.pushReg(0, Xbyak::Operand::EBP);
+	unwinder.pushReg(0, Xbyak::Operand::EBX);
+#ifndef _WIN32
+	// 16-byte alignment
+	unwinder.allocStack(0, 12);
+#endif
+	unwinder.endProlog(0);
+
 //ngen_FailedToFindBlock_:
 	Xbyak::Label failedToFindBlock;
 	L(failedToFindBlock);
@@ -455,7 +491,8 @@ void X86Compiler::genMainloop()
 	call((void *)rdv_BlockCheckFail);
 	jmp(eax);
 
-	genMemHandlers();
+	unwindSize = unwinder.end(CODE_SIZE - 128 - startOffset);
+	verify(unwindSize <= 128);
 
 	ready();
 
@@ -711,6 +748,7 @@ void ngen_ResetBlocks()
 	if (mainloop != nullptr)
 		return;
 
+	unwinder.clear();
 	compiler = new X86Compiler();
 
 	try {
@@ -732,8 +770,6 @@ void ngen_mainloop(void* v_cntx)
 		mainloop();
 	} catch (const SH4ThrownException&) {
 		ERROR_LOG(DYNAREC, "SH4ThrownException in mainloop");
-	} catch (...) {
-		ERROR_LOG(DYNAREC, "Uncaught unknown exception in mainloop");
 	}
 }