piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-23 22:29:39 -06:00
Code Issues Packages Projects Releases Wiki Activity

Finish replacing ThunkManager with ABI_PushRegistersAndAdjustStack.

Browse Source 
As part of that, change SafeLoadToEAX to SafeLoadToReg, and have JitIL
use that, which should fix fastmem on JitIL.

This should also fix a potential stack corruption issue with x86.
This commit is contained in:
comex
2013-09-29 22:51:07 -04:00
parent a53dc6f981
commit 5e4665301b
37 changed files with 220 additions and 406 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
Source/Core/Common/CMakeLists.txt
View File

@ -37,7 +37,6 @@ else()
if(NOT _M_GENERIC) #X86
set(SRCS ${SRCS}
Src/x64FPURoundMode.cpp
Src/x64Thunk.cpp
)
endif()
set(SRCS ${SRCS} Src/x64CPUDetect.cpp)

2
Source/Core/Common/Common.vcxproj
View File

@ -217,7 +217,6 @@
<ClCompile Include="Src\x64CPUDetect.cpp" />
<ClCompile Include="Src\x64Emitter.cpp" />
<ClCompile Include="Src\x64FPURoundMode.cpp" />
<ClCompile Include="Src\x64Thunk.cpp" />
</ItemGroup>
<ItemGroup>
<ClInclude Include="Src\Atomic.h" />
@ -263,7 +262,6 @@
<ClInclude Include="Src\SymbolDB.h" />
<ClInclude Include="Src\SysConf.h" />
<ClInclude Include="Src\Thread.h" />
<ClInclude Include="Src\Thunk.h" />
<ClInclude Include="Src\Timer.h" />
<ClInclude Include="Src\x64ABI.h" />
<ClInclude Include="Src\x64Analyzer.h" />

2
Source/Core/Common/Common.vcxproj.filters
View File

@ -47,7 +47,6 @@
<ClCompile Include="Src\x64ABI.cpp" />
<ClCompile Include="Src\x64CPUDetect.cpp" />
<ClCompile Include="Src\x64FPURoundMode.cpp" />
<ClCompile Include="Src\x64Thunk.cpp" />
</ItemGroup>
<ItemGroup>
<ClInclude Include="Src\Atomic.h" />
@ -84,7 +83,6 @@
<ClInclude Include="Src\SymbolDB.h" />
<ClInclude Include="Src\SysConf.h" />
<ClInclude Include="Src\Thread.h" />
<ClInclude Include="Src\Thunk.h" />
<ClInclude Include="Src\Timer.h" />
<ClInclude Include="Src\x64Analyzer.h" />
<ClInclude Include="Src\x64Emitter.h" />

46
Source/Core/Common/Src/Thunk.h
View File

@ -1,46 +0,0 @@
// Copyright 2013 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#ifndef _THUNK_H_
#define _THUNK_H_
#include <map>
#include "Common.h"
#include "x64Emitter.h"
// This simple class creates a wrapper around a C/C++ function that saves all fp state
// before entering it, and restores it upon exit. This is required to be able to selectively
// call functions from generated code, without inflicting the performance hit and increase
// of complexity that it means to protect the generated code from this problem.
// This process is called thunking.
// There will only ever be one level of thunking on the stack, plus,
// we don't want to pollute the stack, so we store away regs somewhere global.
// NOT THREAD SAFE. This may only be used from the CPU thread.
// Any other thread using this stuff will be FATAL.
class ThunkManager : public Gen::XCodeBlock
{
std::map<void *, const u8 *> thunks;
const u8 *save_regs;
const u8 *load_regs;
public:
ThunkManager() {
Init();
}
~ThunkManager() {
Shutdown();
}
void *ProtectFunction(void *function, int num_params);
private:
void Init();
void Shutdown();
void Reset();
};
#endif // _THUNK_H_

80
Source/Core/Common/Src/x64ABI.cpp
View File

@ -57,6 +57,86 @@ void XEmitter::ABI_RestoreStack(unsigned int frameSize, bool noProlog) {
}
}
void XEmitter::ABI_PushRegistersAndAdjustStack(u32 mask, bool noProlog)
{
int regSize =
#ifdef _M_X64
8;
#else
4;
#endif
int shadow = 0;
#if defined(_WIN32) && defined(_M_X64)
shadow = 0x20;
#endif
int count = 0;
for (int r = 0; r < 16; r++)
{
if (mask & (1 << r))
{
PUSH((X64Reg) r);
count++;
}
}
int size = ((noProlog ? -regSize : 0) - (count * regSize)) & 0xf;
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
size += 16;
}
size += shadow;
if (size)
SUB(regSize * 8, R(RSP), size >= 0x80 ? Imm32(size) : Imm8(size));
int offset = shadow;
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
{
MOVAPD(MDisp(RSP, offset), (X64Reg) x);
offset += 16;
}
}
}
void XEmitter::ABI_PopRegistersAndAdjustStack(u32 mask, bool noProlog)
{
int regSize =
#ifdef _M_X64
8;
#else
4;
#endif
int size = 0;
#if defined(_WIN32) && defined(_M_X64)
size += 0x20;
#endif
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
{
MOVAPD((X64Reg) x, MDisp(RSP, size));
size += 16;
}
}
int count = 0;
for (int r = 0; r < 16; r++)
{
if (mask & (1 << r))
count++;
}
size += ((noProlog ? -regSize : 0) - (count * regSize)) & 0xf;
if (size)
ADD(regSize * 8, R(RSP), size >= 0x80 ? Imm32(size) : Imm8(size));
for (int r = 15; r >= 0; r--)
{
if (mask & (1 << r))
{
POP((X64Reg) r);
}
}
}
#ifdef _M_IX86 // All32
// Shared code between Win32 and Unix32

68
Source/Core/Common/Src/x64Emitter.cpp
View File

@ -1634,74 +1634,6 @@ void XEmitter::___CallCdeclImport6(void* impptr, u32 arg0, u32 arg1, u32 arg2, u
CALLptr(M(impptr));
}
void XEmitter::PushRegistersAndAlignStack(u32 mask)
{
int shadow = 0;
#ifdef _WIN32
shadow = 0x20;
#endif
int count = 0;
for (int r = 0; r < 16; r++)
{
if (mask & (1 << r))
{
PUSH((X64Reg) r);
count++;
}
}
int size = (count & 1) ? 0 : 8;
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
size += 16;
}
size += shadow;
if (size)
SUB(64, R(RSP), size >= 0x80 ? Imm32(size) : Imm8(size));
int offset = shadow;
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
{
MOVAPD(MDisp(RSP, offset), (X64Reg) x);
offset += 16;
}
}
}
void XEmitter::PopRegistersAndAlignStack(u32 mask)
{
int size = 0;
#ifdef _WIN32
size += 0x20;
#endif
for (int x = 0; x < 16; x++)
{
if (mask & (1 << (16 + x)))
{
MOVAPD((X64Reg) x, MDisp(RSP, size));
size += 16;
}
}
int count = 0;
for (int r = 0; r < 16; r++)
{
if (mask & (1 << r))
count++;
}
size += (count & 1) ? 0 : 8;
if (size)
ADD(64, R(RSP), size >= 0x80 ? Imm32(size) : Imm8(size));
for (int r = 15; r >= 0; r--)
{
if (mask & (1 << r))
{
POP((X64Reg) r);
}
}
}
#endif
}

7
Source/Core/Common/Src/x64Emitter.h
View File

@ -646,6 +646,10 @@ public:
void ABI_PushAllCalleeSavedRegsAndAdjustStack();
void ABI_PopAllCalleeSavedRegsAndAdjustStack();
// A more flexible version of the above.
void ABI_PushRegistersAndAdjustStack(u32 mask, bool noProlog);
void ABI_PopRegistersAndAdjustStack(u32 mask, bool noProlog);
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);
@ -691,9 +695,6 @@ public:
#define DECLARE_IMPORT(x) extern "C" void *__imp_##x
void PushRegistersAndAlignStack(u32 mask);
void PopRegistersAndAlignStack(u32 mask);
#endif
}; // class XEmitter

121
Source/Core/Common/Src/x64Thunk.cpp
View File

@ -1,121 +0,0 @@
// Copyright 2013 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#include <map>
#include "Common.h"
#include "MemoryUtil.h"
#include "x64ABI.h"
#include "Thunk.h"
#define THUNK_ARENA_SIZE 1024*1024*1
namespace
{
static u8 GC_ALIGNED32(saved_fp_state[16 * 4 * 4]);
static u8 GC_ALIGNED32(saved_gpr_state[16 * 8]);
static u16 saved_mxcsr;
} // namespace
using namespace Gen;
void ThunkManager::Init()
{
AllocCodeSpace(THUNK_ARENA_SIZE);
save_regs = GetCodePtr();
for (int i = 2; i < ABI_GetNumXMMRegs(); i++)
MOVAPS(M(saved_fp_state + i * 16), (X64Reg)(XMM0 + i));
STMXCSR(M(&saved_mxcsr));
#ifdef _M_X64
MOV(64, M(saved_gpr_state + 0 ), R(RCX));
MOV(64, M(saved_gpr_state + 8 ), R(RDX));
MOV(64, M(saved_gpr_state + 16), R(R8) );
MOV(64, M(saved_gpr_state + 24), R(R9) );
MOV(64, M(saved_gpr_state + 32), R(R10));
MOV(64, M(saved_gpr_state + 40), R(R11));
#ifndef _WIN32
MOV(64, M(saved_gpr_state + 48), R(RSI));
MOV(64, M(saved_gpr_state + 56), R(RDI));
#endif
MOV(64, M(saved_gpr_state + 64), R(RBX));
#else
MOV(32, M(saved_gpr_state + 0 ), R(RCX));
MOV(32, M(saved_gpr_state + 4 ), R(RDX));
#endif
RET();
load_regs = GetCodePtr();
LDMXCSR(M(&saved_mxcsr));
for (int i = 2; i < ABI_GetNumXMMRegs(); i++)
MOVAPS((X64Reg)(XMM0 + i), M(saved_fp_state + i * 16));
#ifdef _M_X64
MOV(64, R(RCX), M(saved_gpr_state + 0 ));
MOV(64, R(RDX), M(saved_gpr_state + 8 ));
MOV(64, R(R8) , M(saved_gpr_state + 16));
MOV(64, R(R9) , M(saved_gpr_state + 24));
MOV(64, R(R10), M(saved_gpr_state + 32));
MOV(64, R(R11), M(saved_gpr_state + 40));
#ifndef _WIN32
MOV(64, R(RSI), M(saved_gpr_state + 48));
MOV(64, R(RDI), M(saved_gpr_state + 56));
#endif
MOV(64, R(RBX), M(saved_gpr_state + 64));
#else
MOV(32, R(RCX), M(saved_gpr_state + 0 ));
MOV(32, R(RDX), M(saved_gpr_state + 4 ));
#endif
RET();
}
void ThunkManager::Reset()
{
thunks.clear();
ResetCodePtr();
}
void ThunkManager::Shutdown()
{
Reset();
FreeCodeSpace();
}
void *ThunkManager::ProtectFunction(void *function, int num_params)
{
std::map<void *, const u8 *>::iterator iter;
iter = thunks.find(function);
if (iter != thunks.end())
return (void *)iter->second;
if (!region)
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);
CALL((void*)save_regs);
CALL((void*)function);
CALL((void*)load_regs);
ABI_RestoreStack(0, true);
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);
unsigned int alignedSize = ABI_GetAlignedFrameSize(num_params * 4, true);
for (int i = 0; i < num_params; i++) {
// ESP is changing, so we do not need i
PUSH(32, MDisp(ESP, alignedSize));
}
CALL(function);
ABI_RestoreStack(num_params * 4, true);
CALL((void*)load_regs);
RET();
#endif
thunks[function] = call_point;
return (void *)call_point;
}