aligned_stack:

* Added -fno-strict-aliasing to the ever growing list of gcc optimizations that can potentially cause perfectly good C code to generate entirely broken results.
 * Removed the preferred-stack=2, since this branch *should* run fine without it now.
 * Have Release builds in Linux SIGKILL when encountering an unexpected SIGSEGV.
 * Fixed some deadlock issues with the new console logger (it introduced cancel points, which is a good thing!  unless you're the EE recompiler and you swallow up any attempt by GCC to cancel a thread)
 * Compilation error fixes.

git-svn-id: http://pcsx2.googlecode.com/svn/branches/aligned_stack@2049 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
Jake.Stine 2009-10-21 08:57:05 +00:00
parent 51999656ab
commit f172829737
11 changed files with 108 additions and 78 deletions

View File

@ -66,8 +66,8 @@
<Add option="-fno-guess-branch-probability" />
<Add option="-fno-dse" />
<Add option="-fno-tree-dse" />
<Add option="-fno-strict-aliasing" />
<Add option="-pipe -msse -msse2" />
<Add option="-mpreferred-stack-boundary=2" />
<Add option="-m32" />
<Add directory="../../include/Utilities" />
<Add directory="../../include" />

View File

@ -67,8 +67,8 @@
<Add option="-fno-guess-branch-probability" />
<Add option="-fno-dse" />
<Add option="-fno-tree-dse" />
<Add option="-fno-strict-aliasing" />
<Add option="-pipe -msse -msse2" />
<Add option="-mpreferred-stack-boundary=2" />
<Add option="-m32" />
<Add directory="../../include/x86emitter" />
<Add directory="../../include" />

View File

@ -1,6 +1,6 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2009 PCSX2 Dev Team
*
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
@ -12,7 +12,7 @@
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include <semaphore.h>
@ -62,7 +62,7 @@ namespace Threading
// --------------------------------------------------------------------------------------
// The following set of documented functions have Linux/Win32 specific implementations,
// which are found in WinThreads.cpp and LnxThreads.cpp
// Returns the number of available logical CPUs (cores plus hyperthreaded cpus)
extern void CountLogicalCores( int LogicalCoresPerPhysicalCPU, int PhysicalCoresPerPhysicalCPU );
@ -146,6 +146,7 @@ namespace Threading
void WaitRaw();
bool WaitRaw( const wxTimeSpan& timeout );
void WaitNoCancel();
void WaitNoCancel( const wxTimeSpan& timeout );
int Count();
void Wait();
@ -161,7 +162,7 @@ namespace Threading
MutexLock();
virtual ~MutexLock() throw();
virtual bool IsRecursive() const { return false; }
void Recreate();
bool RecreateIfLocked();
void Detach();
@ -173,10 +174,10 @@ namespace Threading
void LockRaw();
bool LockRaw( const wxTimeSpan& timeout );
void Wait();
bool Wait( const wxTimeSpan& timeout );
protected:
// empty constructor used by MutexLockRecursive
MutexLock( bool ) {}
@ -217,7 +218,7 @@ namespace Threading
// --------------------------------------------------------------------------------------
// PersistentThread - Helper class for the basics of starting/managing persistent threads.
// --------------------------------------------------------------------------------------
// This class is meant to be a helper for the typical threading model of "start once and
// This class is meant to be a helper for the typical threading model of "start once and
// reuse many times." This class incorporates a lot of extra overhead in stopping and
// starting threads, but in turn provides most of the basic thread-safety and event-handling
// functionality needed for a threaded operation. In practice this model is usually an
@ -230,7 +231,7 @@ namespace Threading
// void OnStart();
// void ExecuteTaskInThread();
// void OnCleanupInThread();
//
//
// Use the public methods Start() and Cancel() to start and shutdown the thread, and use
// m_sem_event internally to post/receive events for the thread (make a public accessor for
// it in your derived class if your thread utilizes the post).
@ -247,14 +248,14 @@ namespace Threading
protected:
typedef int (*PlainJoeFP)();
wxString m_name; // diagnostic name for our thread.
pthread_t m_thread;
Semaphore m_sem_event; // general wait event that's needed by most threads.
MutexLock m_lock_InThread; // used for canceling and closing threads in a deadlock-safe manner
MutexLockRecursive m_lock_start; // used to lock the Start() code from starting simultaneous threads accidentally.
volatile long m_detached; // a boolean value which indicates if the m_thread handle is valid
volatile long m_running; // set true by Start(), and set false by Cancel(), Block(), etc.
@ -285,7 +286,7 @@ namespace Threading
// Start() once necessary locks have been obtained. Do not override Start() directly
// unless you're really sure that's what you need to do. ;)
virtual void OnStart();
virtual void OnStartInThread();
// This is called when the thread has been canceled or exits normally. The PersistentThread
@ -327,7 +328,7 @@ namespace Threading
static void* _internal_callback( void* func );
static void _pt_callback_cleanup( void* handle );
};
//////////////////////////////////////////////////////////////////////////////////////////
// ScopedLock: Helper class for using Mutexes.
// Using this class provides an exception-safe (and generally clean) method of locking
@ -370,7 +371,7 @@ namespace Threading
m_lock.Lock();
m_IsLocked = true;
}
bool IsLocked() const { return m_IsLocked; }
protected:
@ -380,9 +381,9 @@ namespace Threading
, m_IsLocked( isTryLock ? m_lock.TryLock() : false )
{
}
};
class ScopedTryLock : public ScopedLock
{
public:

View File

@ -163,6 +163,14 @@ void Threading::Semaphore::WaitNoCancel()
pthread_setcancelstate( oldstate, NULL );
}
void Threading::Semaphore::WaitNoCancel( const wxTimeSpan& timeout )
{
int oldstate;
pthread_setcancelstate( PTHREAD_CANCEL_DISABLE, &oldstate );
WaitRaw( timeout );
pthread_setcancelstate( oldstate, NULL );
}
int Threading::Semaphore::Count()
{
int retval;

View File

@ -42,25 +42,3 @@
extern void SetCPUState(u32 sseMXCSR, u32 sseVUMXCSR);
extern u32 g_sseVUMXCSR, g_sseMXCSR;
// SEH - "Built in" Structed Exception Handling support.
// This should be available on Windows, via Microsoft or Intel compilers (I'm pretty sure Intel
// supports native SEH model). GUNC in Windows, or any compiler in a non-windows platform, will
// ned to use setjmp/longjmp instead to exit recompiled code.
//
#if defined(_WIN32) && !defined(__GNUG__)
# define PCSX2_SEH
#endif
#ifndef PCSX2_SEH
# include <setjmp.h>
enum
{
SetJmp_Dispatcher = 1,
SetJmp_Exit,
};
extern __threadlocal jmp_buf SetJmp_RecExecute;
extern __threadlocal jmp_buf SetJmp_StateCheck;
#endif

View File

@ -44,15 +44,15 @@ void SysPageFaultExceptionFilter( int signal, siginfo_t *info, void * )
// get bad virtual address
uptr offset = (u8*)info->si_addr - psM;
DevCon.Status( "Protected memory cleanup. Offset 0x%x", offset );
if (offset>=Ps2MemSize::Base)
{
// Bad mojo! Completely invalid address.
// Instigate a crash or abort emulation or something.
wxTrap();
return;
if( !IsDebugBuild )
raise( SIGKILL );
}
DevCon.Status( "Protected memory cleanup. Offset 0x%x", offset );
mmap_ClearCpuBlock( offset & ~m_pagemask );
}

View File

@ -96,8 +96,8 @@
<Add option="-fno-guess-branch-probability" />
<Add option="-fno-dse" />
<Add option="-fno-tree-dse" />
<Add option="-fno-strict-aliasing" />
<Add option="-pipe -msse -msse2" />
<Add option="-mpreferred-stack-boundary=2" />
<Add option="-m32" />
<Add option="-DWX_PRECOMP" />
<Add directory="$(SvnRootDir)/common/include/" />

View File

@ -1,6 +1,6 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2009 PCSX2 Dev Team
*
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
@ -61,8 +61,9 @@ extern void SysClearExecutionCache(); // clears recompiled execution caches!
extern u8 *SysMmapEx(uptr base, u32 size, uptr bounds, const char *caller="Unnamed");
extern void vSyncDebugStuff( uint frame );
//////////////////////////////////////////////////////////////////////////////////////////
//
// --------------------------------------------------------------------------------------
// Memory Protection (Used by VTLB, Recompilers, and Texture caches)
// --------------------------------------------------------------------------------------
#ifdef __LINUX__
# include <signal.h>
@ -87,6 +88,33 @@ extern void vSyncDebugStuff( uint frame );
# error PCSX2 - Unsupported operating system platform.
#endif
// --------------------------------------------------------------------------------------
// PCSX2_SEH - Defines existence of "built in" Structed Exception Handling support.
// --------------------------------------------------------------------------------------
// This should be available on Windows, via Microsoft or Intel compilers (I'm pretty sure Intel
// supports native SEH model). GNUC in Windows, or any compiler in a non-windows platform, will
// need to use setjmp/longjmp instead to exit recompiled code.
//
#if defined(_WIN32) && !defined(__GNUC__)
# define PCSX2_SEH
#else
# include <setjmp.h>
// Platforms without SEH need to use SetJmp / LongJmp to deal with exiting the recompiled
// code execution pipelines in an efficient manner, since standard C++ exceptions cannot
// unwind across dynamically recompiled code.
enum
{
SetJmp_Dispatcher = 1,
SetJmp_Exit,
};
extern jmp_buf SetJmp_RecExecute;
extern jmp_buf SetJmp_StateCheck;
#endif
class pxMessageBoxEvent;
//////////////////////////////////////////////////////////////////////////////////////////

View File

@ -242,7 +242,7 @@ ConsoleLogFrame::ConsoleLogFrame( MainEmuFrame *parent, const wxString& title, A
{
m_TextCtrl.SetBackgroundColour( wxColor( 230, 235, 242 ) );
m_TextCtrl.SetDefaultStyle( m_ColorTable[DefaultConsoleColor] );
// create Log menu (contains most options)
wxMenuBar *pMenuBar = new wxMenuBar();
wxMenu& menuLog = *new wxMenu();
@ -312,9 +312,7 @@ int m_pendingFlushes = 0;
// and this one will magically follow suite. :)
void ConsoleLogFrame::Write( ConsoleColors color, const wxString& text )
{
//#ifdef PCSX2_SEH
pthread_testcancel();
//#endif
ScopedLock lock( m_QueueLock );
@ -324,7 +322,7 @@ void ConsoleLogFrame::Write( ConsoleColors color, const wxString& text )
}
if( (m_QueueColorSection.GetLength() == 0) || ((color != Color_Current) && (m_QueueColorSection.GetLast().color != color)) )
{
{
++m_CurQueuePos; // Don't overwrite the NULL;
m_QueueColorSection.Add( ColorSection(color, m_CurQueuePos) );
}
@ -333,10 +331,10 @@ void ConsoleLogFrame::Write( ConsoleColors color, const wxString& text )
m_QueueBuffer.MakeRoomFor( endpos + 1 ); // and the null!!
memcpy_fast( &m_QueueBuffer[m_CurQueuePos], text.c_str(), sizeof(wxChar) * text.Length() );
m_CurQueuePos = endpos;
// this NULL may be overwritten if the next message sent doesn't perform a color change.
m_QueueBuffer[m_CurQueuePos] = 0;
// Idle events don't always pass (wx blocks them when moving windows or using menus, for
// example). So let's hackfix it so that an alternate message is posted if the queue is
// "piling up."
@ -355,7 +353,7 @@ void ConsoleLogFrame::Write( ConsoleColors color, const wxString& text )
++m_WaitingThreadsForFlush;
lock.Unlock();
if( !m_sem_QueueFlushed.WaitRaw( wxTimeSpan( 0,0,0,500 ) ) )
if( !m_sem_QueueFlushed.Wait( wxTimeSpan( 0,0,0,500 ) ) )
{
// Necessary since the main thread could grab the lock and process before
// the above function actually returns (gotta love threading!)
@ -528,7 +526,7 @@ void ConsoleLogFrame::OnFlushEvent( wxCommandEvent& evt )
// the textctrl has focus or not. The wxWidgets AppendText() function uses EM_LINESCROLL
// instead, which tends to be much faster for high-volume logs, but also ends up refreshing
// the console in sloppy fashion for normal logging.
// (both are needed, the WM_VSCROLL makes the scrolling smooth, and the EM_LINESCROLL avoids
// weird errors when the buffer reaches "max" and starts clearing old history)

View File

@ -241,13 +241,19 @@ static DynGenFunc* _DynGen_DispatcherReg()
#endif
// Set to 0 for a speedup in release builds.
// [doesn't apply to GCC/Mac, which must always align]
#define PCSX2_IOP_FORCED_ALIGN_STACK 0 //1
// For overriding stackframe generation options in Debug builds (possibly useful for troubleshooting)
// Typically this value should be the same as IsDevBuild.
static const bool GenerateStackFrame = IsDevBuild;
static DynGenFunc* _DynGen_EnterRecompiledCode()
{
u8* retval = xGetPtr();
bool allocatedStack = IsDevBuild || PCSX2_IOP_FORCED_ALIGN_STACK;
bool allocatedStack = GenerateStackFrame || PCSX2_IOP_FORCED_ALIGN_STACK;
// Optimization: The IOP never uses stack-based parameter invocation, so we can avoid
// allocating any room on the stack for it (which is important since the IOP's entry
@ -274,7 +280,7 @@ static DynGenFunc* _DynGen_EnterRecompiledCode()
// this is usually worthless because CALL+PUSH leaves us 8 byte aligned instead (fail). So
// we have to do the usual set of stackframe alignments and simulated callstack mess
// *regardless*.
// MSVC/Intel compilers:
// The PCSX2_IOP_FORCED_ALIGN_STACK setting is 0, so we don't care. Just push regs like
// the good old days! (stack alignment will be indeterminate)
@ -284,13 +290,12 @@ static DynGenFunc* _DynGen_EnterRecompiledCode()
xPUSH( ebx );
allocatedStack = false;
CannotUseCallBecauseItWillUnalignTheGodDamnedStack = !!PCSX2_ASSUME_ALIGNED_STACK;
}
uptr* imm = NULL;
if( allocatedStack )
{
if( IsDevBuild )
if( GenerateStackFrame )
{
// Simulate a CALL function by pushing the call address and EBP onto the stack.
// This retains proper stacktrace and stack unwinding (handy in devbuilds!)
@ -301,12 +306,15 @@ static DynGenFunc* _DynGen_EnterRecompiledCode()
// This part simulates the "normal" stackframe prep of "push ebp, mov ebp, esp"
xMOV( ptr32[esp+0x08], ebp );
xLEA( ebp, ptr32[esp+0x08] );
xMOV( &s_store_esp, esp );
xMOV( &s_store_ebp, ebp );
}
}
if( IsDevBuild )
{
xMOV( &s_store_esp, esp );
xMOV( &s_store_ebp, ebp );
}
xJMP( iopDispatcherReg );
if( imm != NULL )
*imm = (uptr)xGetPtr();
@ -319,7 +327,7 @@ static DynGenFunc* _DynGen_EnterRecompiledCode()
if( allocatedStack )
{
// pop the nested "simulated call" stackframe, if needed:
if( IsDevBuild ) xLEAVE();
if( GenerateStackFrame ) xLEAVE();
xMOV( edi, ptr[ebp-12] );
xMOV( esi, ptr[ebp-8] );
xMOV( ebx, ptr[ebp-4] );
@ -859,22 +867,22 @@ static __noinline s32 recExecuteBlock( s32 eeCycles )
// Register freezing note:
// The IOP does not use mmx/xmm registers, so we don't modify the status
// of the g_EEFreezeRegs here.
// [TODO] recExecuteBlock could be replaced by a direct call to the iopEnterRecompiledCode()
// (by assigning its address to the psxRec structure). But for that to happen, we need
// to move psxBreak/psxCycleEE update code to emitted assembly code. >_< --air
// Likely Disasm, as borrowed from MSVC:
// Entry:
// mov eax,dword ptr [esp+4]
// mov dword ptr [psxBreak (0E88DCCh)],0
// mov dword ptr [psxCycleEE (832A84h)],eax
// mov eax,dword ptr [esp+4]
// mov dword ptr [psxBreak (0E88DCCh)],0
// mov dword ptr [psxCycleEE (832A84h)],eax
// Exit:
// mov ecx,dword ptr [psxBreak (0E88DCCh)]
// mov edx,dword ptr [psxCycleEE (832A84h)]
// lea eax,[edx+ecx]
// mov ecx,dword ptr [psxBreak (0E88DCCh)]
// mov edx,dword ptr [psxCycleEE (832A84h)]
// lea eax,[edx+ecx]
iopEnterRecompiledCode();

View File

@ -684,8 +684,17 @@ static void StateThreadCheck_LongJmp()
{
setjmp( SetJmp_StateCheck );
int oldstate;
// Important! Most of the console logging and such has cancel points in it. This is great
// in Windows, where SEH lets us safely kill a thread from anywhere we want. This is bad
// in Linux, which cannot have a C++ exception cross the recompiler. Hence the changing
// of the cancelstate here!
pthread_setcancelstate( PTHREAD_CANCEL_ENABLE, &oldstate );
mtgsThread.RethrowException();
SysCoreThread::Get().StateCheckInThread();
pthread_setcancelstate( PTHREAD_CANCEL_DISABLE, &oldstate );
}
static void recExecute()
@ -847,17 +856,17 @@ void recClear(u32 addr, u32 size)
}
#ifdef __GNUG__
__threadlocal jmp_buf SetJmp_RecExecute;
__threadlocal jmp_buf SetJmp_StateCheck;
#ifndef PCSX2_SEH
jmp_buf SetJmp_RecExecute;
jmp_buf SetJmp_StateCheck;
#endif
static void ExitRec()
{
#ifdef __GNUG__
longjmp( SetJmp_RecExecute, SetJmp_Exit );
#else
#ifdef PCSX2_SEH
throw Exception::ExitRecExecute();
#else
longjmp( SetJmp_RecExecute, SetJmp_Exit );
#endif
}
@ -1085,7 +1094,7 @@ static u32 eeScaleBlockCycles()
static void iBranchTest(u32 newpc)
{
_DynGen_StackFrameCheck();
if( g_ExecBiosHack ) CheckForBIOSEnd();
// Check the Event scheduler if our "cycle target" has been reached.
@ -1313,7 +1322,7 @@ void __fastcall dyna_block_discard(u32 start,u32 sz)
recClear(start, sz);
// Stack trick: This function was invoked via a direct jmp, so manually pop the
// EBP/stackframe before issuing a RET, else esp/ebp will be incorrect.
// EBP/stackframe before issuing a RET, else esp/ebp will be incorrect.
#ifdef _MSC_VER
__asm leave __asm jmp [ExitRecompiledCode]