piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2024-11-14 13:27:45 -07:00
Code Issues Actions Packages Projects Releases Wiki Activity

Revert "Fix stack misalignment issues."

Browse Source 
This reverts commit d334a9bc23.

This breaks single core.
This commit is contained in:
Rachel Bryk 2013-09-22 14:29:35 -04:00
parent f3469c16a5
commit 9a6f28fce4
5 changed files with 182 additions and 134 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

275
Source/Core/Common/Src/x64ABI.cpp
View File

@ -10,51 +10,38 @@ using namespace Gen;
// Shared code between Win64 and Unix64
unsigned int XEmitter::ABI_GetAlignedFrameSize(unsigned int frameSize, bool noProlog) {
// On platforms other than Windows 32-bit: At the beginning of a function,
// the stack pointer is 4/8 bytes less than a multiple of 16; however, the
// function prolog immediately subtracts an appropriate amount to align
// it, so no alignment is required around a call.
// In the functions generated by ThunkManager::ProtectFunction, we add the
// necessary subtraction (and 0x20 bytes shadow space for Win64) into this
// rather than having a separate prolog.
// On Windows 32-bit, the required alignment is only 4 bytes, so we just
// ensure that the frame size isn't misaligned.
#ifdef _M_X64
// expect frameSize == 0
frameSize = noProlog ? 0x28 : 0;
#elif defined(_WIN32)
frameSize = (frameSize + 3) & -4;
#else
unsigned int existingAlignment = noProlog ? 0xc : 0;
frameSize -= existingAlignment;
frameSize = (frameSize + 15) & -16;
frameSize += existingAlignment;
// Sets up a __cdecl function.
void XEmitter::ABI_EmitPrologue(int maxCallParams)
{
#ifdef _M_IX86
// Don't really need to do anything
#elif defined(_M_X64)
#if _WIN32
int stacksize = ((maxCallParams + 1) & ~1) * 8 + 8;
// Set up a stack frame so that we can call functions
// TODO: use maxCallParams
SUB(64, R(RSP), Imm8(stacksize));
#endif
#else
#error Arch not supported
#endif
return frameSize;
}
void XEmitter::ABI_AlignStack(unsigned int frameSize, bool noProlog) {
unsigned int fillSize =
ABI_GetAlignedFrameSize(frameSize, noProlog) - frameSize;
if (fillSize != 0) {
#ifdef _M_X64
SUB(64, R(RSP), Imm8(fillSize));
#else
SUB(32, R(ESP), Imm8(fillSize));
void XEmitter::ABI_EmitEpilogue(int maxCallParams)
{
#ifdef _M_IX86
RET();
#elif defined(_M_X64)
#ifdef _WIN32
int stacksize = ((maxCallParams+1)&~1)*8 + 8;
ADD(64, R(RSP), Imm8(stacksize));
#endif
}
}
RET();
#else
#error Arch not supported
void XEmitter::ABI_RestoreStack(unsigned int frameSize, bool noProlog) {
unsigned int alignedSize = ABI_GetAlignedFrameSize(frameSize, noProlog);
if (alignedSize != 0) {
#ifdef _M_X64
ADD(64, R(RSP), Imm8(alignedSize));
#else
ADD(32, R(ESP), Imm8(alignedSize));
#endif
}
}
#ifdef _M_IX86 // All32
@ -78,7 +65,7 @@ void XEmitter::ABI_CallFunctionCC16(void *func, u32 param1, u16 param2) {
PUSH(16, Imm16(param2));
PUSH(32, Imm32(param1));
CALL(func);
ABI_AlignStack(1 * 2 + 1 * 4);
ABI_RestoreStack(1 * 2 + 1 * 4);
}
void XEmitter::ABI_CallFunctionC(void *func, u32 param1) {
@ -169,27 +156,60 @@ void XEmitter::ABI_CallFunctionA(void *func, const Gen::OpArg &arg1)
}
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
// Note: 4 * 4 = 16 bytes, so alignment is preserved.
PUSH(EBP);
MOV(32, R(EBP), R(ESP));
PUSH(EBX);
PUSH(ESI);
PUSH(EDI);
SUB(32, R(ESP), Imm8(0xc));
}
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
ADD(32, R(ESP), Imm8(0xc));
POP(EDI);
POP(ESI);
POP(EBX);
POP(EBP);
}
unsigned int XEmitter::ABI_GetAlignedFrameSize(unsigned int frameSize) {
frameSize += 4; // reserve space for return address
unsigned int alignedSize =
#ifdef __GNUC__
(frameSize + 15) & -16;
#else
(frameSize + 3) & -4;
#endif
return alignedSize;
}
void XEmitter::ABI_AlignStack(unsigned int frameSize) {
// Mac OS X requires the stack to be 16-byte aligned before every call.
// Linux requires the stack to be 16-byte aligned before calls that put SSE
// vectors on the stack, but since we do not keep track of which calls do that,
// it is effectively every call as well.
// Windows binaries compiled with MSVC do not have such a restriction*, but I
// expect that GCC on Windows acts the same as GCC on Linux in this respect.
// It would be nice if someone could verify this.
// *However, the MSVC optimizing compiler assumes a 4-byte-aligned stack at times.
unsigned int fillSize =
ABI_GetAlignedFrameSize(frameSize) - (frameSize + 4);
if (fillSize != 0) {
SUB(32, R(ESP), Imm8(fillSize));
}
}
void XEmitter::ABI_RestoreStack(unsigned int frameSize) {
unsigned int alignedSize = ABI_GetAlignedFrameSize(frameSize);
alignedSize -= 4; // return address is POPped at end of call
if (alignedSize != 0) {
ADD(32, R(ESP), Imm8(alignedSize));
}
}
#else //64bit
// Common functions
void XEmitter::ABI_CallFunction(void *func) {
ABI_AlignStack(0);
u64 distance = u64(func) - (u64(code) + 5);
if (distance >= 0x0000000080000000ULL
&& distance < 0xFFFFFFFF80000000ULL) {
@ -199,11 +219,9 @@ void XEmitter::ABI_CallFunction(void *func) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionC16(void *func, u16 param1) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32((u32)param1));
u64 distance = u64(func) - (u64(code) + 5);
if (distance >= 0x0000000080000000ULL
@ -214,11 +232,9 @@ void XEmitter::ABI_CallFunctionC16(void *func, u16 param1) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionCC16(void *func, u32 param1, u16 param2) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
MOV(32, R(ABI_PARAM2), Imm32((u32)param2));
u64 distance = u64(func) - (u64(code) + 5);
@ -230,11 +246,9 @@ void XEmitter::ABI_CallFunctionCC16(void *func, u32 param1, u16 param2) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionC(void *func, u32 param1) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
u64 distance = u64(func) - (u64(code) + 5);
if (distance >= 0x0000000080000000ULL
@ -245,11 +259,9 @@ void XEmitter::ABI_CallFunctionC(void *func, u32 param1) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionCC(void *func, u32 param1, u32 param2) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
MOV(32, R(ABI_PARAM2), Imm32(param2));
u64 distance = u64(func) - (u64(code) + 5);
@ -261,11 +273,9 @@ void XEmitter::ABI_CallFunctionCC(void *func, u32 param1, u32 param2) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionCCC(void *func, u32 param1, u32 param2, u32 param3) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
MOV(32, R(ABI_PARAM2), Imm32(param2));
MOV(32, R(ABI_PARAM3), Imm32(param3));
@ -278,11 +288,9 @@ void XEmitter::ABI_CallFunctionCCC(void *func, u32 param1, u32 param2, u32 param
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionCCP(void *func, u32 param1, u32 param2, void *param3) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
MOV(32, R(ABI_PARAM2), Imm32(param2));
MOV(64, R(ABI_PARAM3), Imm64((u64)param3));
@ -295,11 +303,9 @@ void XEmitter::ABI_CallFunctionCCP(void *func, u32 param1, u32 param2, void *par
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionCCCP(void *func, u32 param1, u32 param2, u32 param3, void *param4) {
ABI_AlignStack(0);
MOV(32, R(ABI_PARAM1), Imm32(param1));
MOV(32, R(ABI_PARAM2), Imm32(param2));
MOV(32, R(ABI_PARAM3), Imm32(param3));
@ -313,11 +319,9 @@ void XEmitter::ABI_CallFunctionCCCP(void *func, u32 param1, u32 param2, u32 para
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionPPC(void *func, void *param1, void *param2, u32 param3) {
ABI_AlignStack(0);
MOV(64, R(ABI_PARAM1), Imm64((u64)param1));
MOV(64, R(ABI_PARAM2), Imm64((u64)param2));
MOV(32, R(ABI_PARAM3), Imm32(param3));
@ -330,12 +334,10 @@ void XEmitter::ABI_CallFunctionPPC(void *func, void *param1, void *param2, u32 p
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
// Pass a register as a parameter.
void XEmitter::ABI_CallFunctionR(void *func, X64Reg reg1) {
ABI_AlignStack(0);
if (reg1 != ABI_PARAM1)
MOV(32, R(ABI_PARAM1), R(reg1));
u64 distance = u64(func) - (u64(code) + 5);
@ -347,12 +349,10 @@ void XEmitter::ABI_CallFunctionR(void *func, X64Reg reg1) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
// Pass two registers as parameters.
void XEmitter::ABI_CallFunctionRR(void *func, X64Reg reg1, X64Reg reg2) {
ABI_AlignStack(0);
if (reg2 != ABI_PARAM1) {
if (reg1 != ABI_PARAM1)
MOV(64, R(ABI_PARAM1), R(reg1));
@ -373,12 +373,10 @@ void XEmitter::ABI_CallFunctionRR(void *func, X64Reg reg1, X64Reg reg2) {
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionAC(void *func, const Gen::OpArg &arg1, u32 param2)
{
ABI_AlignStack(0);
if (!arg1.IsSimpleReg(ABI_PARAM1))
MOV(32, R(ABI_PARAM1), arg1);
MOV(32, R(ABI_PARAM2), Imm32(param2));
@ -391,12 +389,10 @@ void XEmitter::ABI_CallFunctionAC(void *func, const Gen::OpArg &arg1, u32 param2
} else {
CALL(func);
}
ABI_RestoreStack(0);
}
void XEmitter::ABI_CallFunctionA(void *func, const Gen::OpArg &arg1)
{
ABI_AlignStack(0);
if (!arg1.IsSimpleReg(ABI_PARAM1))
MOV(32, R(ABI_PARAM1), arg1);
u64 distance = u64(func) - (u64(code) + 5);
@ -408,7 +404,95 @@ void XEmitter::ABI_CallFunctionA(void *func, const Gen::OpArg &arg1)
} else {
CALL(func);
}
}
unsigned int XEmitter::ABI_GetAlignedFrameSize(unsigned int frameSize) {
return frameSize;
}
#ifdef _WIN32
// Win64 Specific Code
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
//we only want to do this once
PUSH(RBX);
PUSH(RSI);
PUSH(RDI);
PUSH(RBP);
PUSH(R12);
PUSH(R13);
PUSH(R14);
PUSH(R15);
//TODO: Also preserve XMM0-3?
ABI_AlignStack(0);
}
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
ABI_RestoreStack(0);
POP(R15);
POP(R14);
POP(R13);
POP(R12);
POP(RBP);
POP(RDI);
POP(RSI);
POP(RBX);
}
// Win64 Specific Code
void XEmitter::ABI_PushAllCallerSavedRegsAndAdjustStack() {
PUSH(RCX);
PUSH(RDX);
PUSH(RSI);
PUSH(RDI);
PUSH(R8);
PUSH(R9);
PUSH(R10);
PUSH(R11);
//TODO: Also preserve XMM0-15?
ABI_AlignStack(0);
}
void XEmitter::ABI_PopAllCallerSavedRegsAndAdjustStack() {
ABI_RestoreStack(0);
POP(R11);
POP(R10);
POP(R9);
POP(R8);
POP(RDI);
POP(RSI);
POP(RDX);
POP(RCX);
}
void XEmitter::ABI_AlignStack(unsigned int /*frameSize*/) {
SUB(64, R(RSP), Imm8(0x28));
}
void XEmitter::ABI_RestoreStack(unsigned int /*frameSize*/) {
ADD(64, R(RSP), Imm8(0x28));
}
#else
// Unix64 Specific Code
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
PUSH(RBX);
PUSH(RBP);
PUSH(R12);
PUSH(R13);
PUSH(R14);
PUSH(R15);
PUSH(R15); //just to align stack. duped push/pop doesn't hurt.
}
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
POP(R15);
POP(R15);
POP(R14);
POP(R13);
POP(R12);
POP(RBP);
POP(RBX);
}
void XEmitter::ABI_PushAllCallerSavedRegsAndAdjustStack() {
@ -435,59 +519,12 @@ void XEmitter::ABI_PopAllCallerSavedRegsAndAdjustStack() {
POP(RCX);
}
#ifdef _WIN32
// Win64 Specific Code
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
//we only want to do this once
PUSH(RBP);
MOV(64, R(RBP), R(RSP));
PUSH(RBX);
PUSH(RSI);
PUSH(RDI);
PUSH(R12);
PUSH(R13);
PUSH(R14);
PUSH(R15);
SUB(64, R(RSP), Imm8(0x28));
//TODO: Also preserve XMM0-3?
void XEmitter::ABI_AlignStack(unsigned int /*frameSize*/) {
SUB(64, R(RSP), Imm8(0x08));
}
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
ADD(64, R(RSP), Imm8(0x28));
POP(R15);
POP(R14);
POP(R13);
POP(R12);
POP(RDI);
POP(RSI);
POP(RBX);
POP(RBP);
}
#else
// Unix64 Specific Code
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
PUSH(RBP);
MOV(64, R(RBP), R(RSP));
PUSH(RBX);
PUSH(R12);
PUSH(R13);
PUSH(R14);
PUSH(R15);
SUB(64, R(RSP), Imm8(8));
}
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
ADD(64, R(RSP), Imm8(8));
POP(R15);
POP(R14);
POP(R13);
POP(R12);
POP(RBX);
POP(RBP);
void XEmitter::ABI_RestoreStack(unsigned int /*frameSize*/) {
ADD(64, R(RSP), Imm8(0x08));
}
#endif // WIN32

11
Source/Core/Common/Src/x64Emitter.h
View File

@ -652,9 +652,14 @@ public:
void ABI_PushAllCallerSavedRegsAndAdjustStack();
void ABI_PopAllCallerSavedRegsAndAdjustStack();
unsigned int ABI_GetAlignedFrameSize(unsigned int frameSize, bool noProlog = false);
void ABI_AlignStack(unsigned int frameSize, bool noProlog = false);
void ABI_RestoreStack(unsigned int frameSize, bool noProlog = false);
unsigned int ABI_GetAlignedFrameSize(unsigned int frameSize);
void ABI_AlignStack(unsigned int frameSize);
void ABI_RestoreStack(unsigned int frameSize);
// Sets up a __cdecl function.
// Only x64 really needs the parameter count.
void ABI_EmitPrologue(int maxCallParams);
void ABI_EmitEpilogue(int maxCallParams);
#ifdef _M_IX86
inline int ABI_GetNumXMMRegs() { return 8; }

20
Source/Core/Common/Src/x64Thunk.cpp
View File

@ -91,27 +91,35 @@ void *ThunkManager::ProtectFunction(void *function, int num_params)
PanicAlert("Trying to protect functions before the emu is started. Bad bad bad.");
const u8 *call_point = GetCodePtr();
#ifdef _M_X64
// Make sure to align stack.
ABI_AlignStack(0, true);
#ifdef _M_X64
#ifdef _WIN32
SUB(64, R(ESP), Imm8(0x28));
#else
SUB(64, R(ESP), Imm8(0x8));
#endif
CALL((void*)save_regs);
CALL((void*)function);
CALL((void*)load_regs);
ABI_RestoreStack(0, true);
#ifdef _WIN32
ADD(64, R(ESP), Imm8(0x28));
#else
ADD(64, R(ESP), Imm8(0x8));
#endif
RET();
#else
CALL((void*)save_regs);
// Since parameters are in the previous stack frame, not in registers, this takes some
// trickery : we simply re-push the parameters. might not be optimal, but that doesn't really
// matter.
ABI_AlignStack(num_params * 4, true);
ABI_AlignStack(num_params * 4);
unsigned int alignedSize = ABI_GetAlignedFrameSize(num_params * 4);
for (int i = 0; i < num_params; i++) {
// ESP is changing, so we do not need i
PUSH(32, MDisp(ESP, alignedSize));
PUSH(32, MDisp(ESP, alignedSize - 4));
}
CALL(function);
ABI_RestoreStack(num_params * 4, true);
ABI_RestoreStack(num_params * 4);
CALL((void*)load_regs);
RET();
#endif

5
Source/Core/VideoCommon/Src/VertexLoader.cpp
View File

@ -217,7 +217,7 @@ void VertexLoader::CompileVertexTranslator()
PanicAlert("Trying to recompile a vertex translator");
m_compiledCode = GetCodePtr();
ABI_PushAllCalleeSavedRegsAndAdjustStack();
ABI_EmitPrologue(4);
// Start loop here
const u8 *loop_start = GetCodePtr();
@ -499,8 +499,7 @@ void VertexLoader::CompileVertexTranslator()
#endif
J_CC(CC_NZ, loop_start, true);
ABI_PopAllCalleeSavedRegsAndAdjustStack();
RET();
ABI_EmitEpilogue(4);
#endif
m_NativeFmt->Initialize(vtx_decl);
}

5
Source/Core/VideoCommon/Src/x64DLCache.cpp
View File

@ -409,7 +409,7 @@ void CompileAndRunDisplayList(u32 address, u32 size, CachedDisplayList *dl)
emitter.AlignCode4();
dl->compiled_code = emitter.GetCodePtr();
emitter.ABI_PushAllCalleeSavedRegsAndAdjustStack();
emitter.ABI_EmitPrologue(4);
while (g_pVideoData < end)
{
@ -572,8 +572,7 @@ void CompileAndRunDisplayList(u32 address, u32 size, CachedDisplayList *dl)
break;
}
}
emitter.ABI_PopAllCalleeSavedRegsAndAdjustStack();
emitter.RET();
emitter.ABI_EmitEpilogue(4);
INCSTAT(stats.numDListsCalled);
INCSTAT(stats.thisFrame.numDListsCalled);
Statistics::SwapDL();