192 lines
3.3 KiB
C
192 lines
3.3 KiB
C
/*++
|
||
|
||
Copyright (c) 1995 Intel Corporation
|
||
|
||
Module Name:
|
||
|
||
i64perfc.c copied from simperfc.c
|
||
|
||
Abstract:
|
||
|
||
This module implements the routines to support performance counters.
|
||
|
||
Author:
|
||
|
||
14-Apr-1995
|
||
|
||
Environment:
|
||
|
||
Kernel mode
|
||
|
||
Revision History:
|
||
|
||
|
||
--*/
|
||
|
||
#include "halp.h"
|
||
#include "eisa.h"
|
||
|
||
//
|
||
// Define and initialize the 64-bit count of total system cycles used
|
||
// as the performance counter.
|
||
//
|
||
|
||
ULONGLONG HalpCycleCount = 0;
|
||
|
||
BOOLEAN HalpITCCalibrate = TRUE; // XXTF
|
||
|
||
extern ULONGLONG HalpITCFrequency;
|
||
|
||
extern ULONGLONG HalpClockCount;
|
||
|
||
#if 0
|
||
|
||
VOID
|
||
HalpCheckPerformanceCounter(
|
||
VOID
|
||
)
|
||
|
||
Routine Description:
|
||
|
||
This function is called every system clock interrupt in order to
|
||
check for wrap of the performance counter. The function must handle
|
||
a wrap if it is detected.
|
||
|
||
N.B. - This function was from the Alpha HAL.
|
||
This function must be called at CLOCK_LEVEL.
|
||
|
||
Arguments:
|
||
|
||
None.
|
||
|
||
Return Value:
|
||
|
||
None.
|
||
|
||
{
|
||
|
||
return;
|
||
|
||
} // HalpCheckPerformanceCounter()
|
||
|
||
#endif // 0
|
||
|
||
|
||
LARGE_INTEGER
|
||
KeQueryPerformanceCounter (
|
||
OUT PLARGE_INTEGER PerformanceFrequency OPTIONAL
|
||
)
|
||
|
||
/*++
|
||
|
||
Routine Description:
|
||
|
||
This routine returns current 64-bit performance counter and,
|
||
optionally, the Performance Frequency.
|
||
|
||
Arguments:
|
||
|
||
PerformanceFrequency - optionally, supplies the address
|
||
of a variable to receive the performance counter frequency.
|
||
|
||
Return Value:
|
||
|
||
Current value of the performance counter will be returned.
|
||
|
||
--*/
|
||
|
||
{
|
||
LARGE_INTEGER result;
|
||
|
||
#ifndef DISABLE_ITC_WORKAROUND
|
||
result.QuadPart = __getReg(CV_IA64_ApITC);
|
||
while ((result.QuadPart & 0xFFFFFFFF) == 0xFFFFFFFF) {
|
||
result.QuadPart = __getReg(CV_IA64_ApITC);
|
||
}
|
||
#else
|
||
result.QuadPart = __getReg(CV_IA64_ApITC);
|
||
#endif
|
||
|
||
if (ARGUMENT_PRESENT(PerformanceFrequency)) {
|
||
PerformanceFrequency->QuadPart = HalpITCFrequency;
|
||
}
|
||
|
||
return result;
|
||
|
||
} // KeQueryPerformanceCounter()
|
||
|
||
|
||
|
||
VOID
|
||
HalCalibratePerformanceCounter (
|
||
IN LONG volatile *Number,
|
||
IN ULONGLONG NewCount
|
||
)
|
||
|
||
/*++
|
||
|
||
Routine Description:
|
||
|
||
This routine sets the performance counter value for the current
|
||
processor to the specified valueo.
|
||
The reset is done such that the resulting value is closely
|
||
synchronized with other processors in the configuration.
|
||
|
||
Arguments:
|
||
|
||
Number - Supplies a pointer to count of the number of processors in
|
||
the configuration.
|
||
|
||
Return Value:
|
||
|
||
None.
|
||
|
||
--*/
|
||
{
|
||
KSPIN_LOCK Lock;
|
||
KIRQL OldIrql;
|
||
|
||
if ( HalpITCCalibrate ) {
|
||
|
||
//
|
||
// Raise IRQL to HIGH_LEVEL, decrement the number of processors, and
|
||
// wait until the number is zero.
|
||
//
|
||
|
||
KeInitializeSpinLock(&Lock);
|
||
KeRaiseIrql(HIGH_LEVEL, &OldIrql);
|
||
|
||
PCR->HalReserved[CURRENT_ITM_VALUE_INDEX] = NewCount + HalpClockCount;
|
||
HalpWriteITM( PCR->HalReserved[CURRENT_ITM_VALUE_INDEX] );
|
||
|
||
if (ExInterlockedDecrementLong((PLONG)Number, &Lock) != RESULT_ZERO) {
|
||
do {
|
||
} while (*((LONG volatile *)Number) !=0);
|
||
}
|
||
|
||
//
|
||
// Write the compare register with defined current ITM value,
|
||
// and set the performance counter for the current processor
|
||
// with the passed count.
|
||
//
|
||
|
||
HalpWriteITC( NewCount );
|
||
|
||
//
|
||
// Restore IRQL to its previous value and return.
|
||
//
|
||
|
||
KeLowerIrql(OldIrql);
|
||
|
||
}
|
||
else {
|
||
|
||
*Number = 0;
|
||
|
||
}
|
||
|
||
return;
|
||
|
||
} // HalCalibratePerformanceCounter()
|
||
|