|
|
//========= Copyright � 1996-2005, Valve Corporation, All rights reserved. ============//
//
// Purpose:
//
// $NoKeywords: $
//
//=============================================================================//
#ifndef P4PERFORMANCECOUNTERS_H
#define P4PERFORMANCECOUNTERS_H
#pragma once
// Pentium 4 support
/*
http://developer.intel.com/design/Pentium4/documentation.htm
IA-32 Intel Architecture Software Developer's Manual Volume 1: Basic Architecture
IA-32 Intel Architecture Software Developer's Manual Volume 2A: Instruction Set Reference, A-M
IA-32 Intel Architecture Software Developer's Manual Volume 2B: Instruction Set Reference, N-Z
IA-32 Intel Architecture Software Developer's Manual Volume 3: System Programming Guide
From Mikael Pettersson's perfctr: http://user.it.uu.se/~mikpe/linux/perfctr/
* Known quirks: - OVF_PMI+FORCE_OVF counters must have an ireset value of -1. This allows the regular overflow check to also handle FORCE_OVF counters. Not having this restriction would lead to MAJOR complications in the driver's "detect overflow counters" code. There is no loss of functionality since the ireset value doesn't affect the counter's PMI rate for FORCE_OVF counters.
- In experiments with FORCE_OVF counters, and regular OVF_PMI counters with small ireset values between -8 and -1, it appears that the faulting instruction is subjected to a new PMI before it can complete, ad infinitum. This occurs even though the driver clears the CCCR (and in testing also the ESCR) and invokes a user-space signal handler before restoring the CCCR and resuming the instruction. */
#define NCOUNTERS 18
// The 18 counters
enum Counters { MSR_BPU_COUNTER0, MSR_BPU_COUNTER1, MSR_BPU_COUNTER2, MSR_BPU_COUNTER3, MSR_MS_COUNTER0, MSR_MS_COUNTER1, MSR_MS_COUNTER2, MSR_MS_COUNTER3, MSR_FLAME_COUNTER0, MSR_FLAME_COUNTER1, MSR_FLAME_COUNTER2, MSR_FLAME_COUNTER3, MSR_IQ_COUNTER0, MSR_IQ_COUNTER1, MSR_IQ_COUNTER2, MSR_IQ_COUNTER3, MSR_IQ_COUNTER4, MSR_IQ_COUNTER5 };
// register base for counters
#define MSR_COUNTER_BASE 0x300
// register base for CCCR register
#define MSR_CCCR_BASE 0x360
#pragma pack(push, 1)
// access to these bits is through the methods
typedef union ESCR { struct { uint64 Reserved0_1 : 2; //
uint64 USR : 1; //
uint64 OS : 1; //
uint64 TagEnable : 1; //
uint64 TagValue : 4; //
uint64 EventMask : 16; // from event select
uint64 ESCREventSelect : 6; // 31:25 class of event
uint64 Reserved31 : 1; //
uint64 Reserved32_63 : 32; //
}; uint64 flat;
} ESCR;
typedef union CCCR { struct { uint64 Reserved0_11 : 12;// 0 -11
uint64 Enable : 1; // 12
uint64 CCCRSelect : 3; // 13-15
uint64 Reserved16_17 : 2; // 16 17
uint64 Compare : 1; // 18
uint64 Complement : 1; // 19
uint64 Threshold : 4; // 20-23
uint64 Edge : 1; // 24
uint64 FORCE_OVF : 1; // 25
uint64 OVF_PMI : 1; // 26
uint64 Reserved27_29 : 3; // 27-29
uint64 Cascade : 1; // 30
uint64 OVF : 1; // 31
uint64 Reserved32_63 : 32; //
}; uint64 flat;
} CCCR;
#pragma pack(pop)
extern const unsigned short cccr_escr_map[NCOUNTERS][8];
enum P4TagState { TagDisable, //
TagEnable, //
};
enum P4ForceOverflow { ForceOverflowDisable, ForceOverflowEnable, };
enum P4OverflowInterrupt { OverflowInterruptDisable, OverflowInterruptEnable, };
// Turn off the no return value warning in ReadCounter.
#pragma warning( disable : 4035 )
class P4BaseEvent { int m_counter;
protected:
void SetCounter(int counter) { m_counter = counter; cccrPort = MSR_CCCR_BASE + m_counter; counterPort = MSR_COUNTER_BASE + m_counter; escrPort = cccr_escr_map[m_counter][cccr.CCCRSelect]; }
public:
unsigned short m_eventMask; const tchar *description; PME *pme; ESCR escr; CCCR cccr; int counterPort; int cccrPort; int escrPort;
P4BaseEvent() { pme = PME::Instance(); m_eventMask = 0; description = _T(""); escr.flat = 0; cccr.flat = 0; cccr.Reserved16_17 = 3; // must be set
escrPort = 0; m_counter = -1; }
void StartCounter() { cccr.Enable = 1; pme->WriteMSR( cccrPort, cccr.flat ); }
void StopCounter() { cccr.Enable = 0; pme->WriteMSR( cccrPort, cccr.flat ); }
void ClearCounter() { pme->WriteMSR( counterPort, 0ui64 ); // clear
}
void WriteCounter( int64 value ) { pme->WriteMSR( counterPort, value ); // clear
}
int64 ReadCounter() { #if PME_DEBUG
if ( escr.USR == 0 && escr.OS == 0 ) return -1; // no area to collect, use SetCaptureMode
if ( escr.EventMask == 0 ) return -2; // no event mask set
if ( m_counter == -1 ) return -3; // counter not legal
#endif
// ReadMSR should work here too, but RDPMC should be faster
int64 value = 0; pme->ReadMSR( counterPort, &value ); return value; #if 0
// we need to copy this into a temp for some reason
int temp = m_counter; _asm { mov ecx, temp RDPMC } #endif
}
void SetCaptureMode( PrivilegeCapture priv ) { switch ( priv ) { case OS_Only: { escr.USR = 0; escr.OS = 1; break; } case USR_Only: { escr.USR = 1; escr.OS = 0; break; } case OS_and_USR: { escr.USR = 1; escr.OS = 1; break; } }
escr.EventMask = m_eventMask; pme->WriteMSR( escrPort, escr.flat ); }
void SetTagging( P4TagState tagEnable, uint8 tagValue ) { escr.TagEnable = tagEnable; escr.TagValue = tagValue; pme->WriteMSR( escrPort, escr.flat ); }
void SetFiltering( CompareState compareEnable, CompareMethod compareMethod, uint8 threshold, EdgeState edgeEnable ) { cccr.Compare = compareEnable; cccr.Complement = compareMethod; cccr.Threshold = threshold; cccr.Edge = edgeEnable; pme->WriteMSR( cccrPort, cccr.flat ); }
void SetOverflowEnables( P4ForceOverflow overflowEnable, P4OverflowInterrupt overflowInterruptEnable ) { cccr.FORCE_OVF = overflowEnable; cccr.OVF_PMI = overflowInterruptEnable; pme->WriteMSR( cccrPort, cccr.flat ); }
void SetOverflow() { cccr.OVF = 1; pme->WriteMSR( cccrPort, cccr.flat ); }
void ClearOverflow() { cccr.OVF = 0; pme->WriteMSR( cccrPort, cccr.flat ); }
bool isOverflow() { CCCR cccr_temp; pme->ReadMSR( cccrPort, &cccr_temp.flat ); return cccr_temp.OVF; }
void SetCascade() { cccr.Cascade = 1; pme->WriteMSR( cccrPort, cccr.flat ); }
void ClearCascade() { cccr.Cascade = 0; pme->WriteMSR( cccrPort, cccr.flat ); } }; #pragma warning( default : 4035 )
#include "eventmasks.h"
#include "eventmodes.h"
#endif // P4PERFORMANCECOUNTERS_H
|