d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
5c32431080
freebsd-dev/sys/i386/isa/prof_machdep.c

351 lines
9.3 KiB
C
Raw Normal View History

Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
/*-
* Copyright (c) 1996 Bruce D. Evans.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
Added a $Id$ keyword. Bruce still needs to put a copyright notice on this file.
1996-04-08 16:41:06 +00:00
*
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
Ifdefed the declarations of conditionally used variables.
1998-12-14 18:21:34 +00:00
* $Id: prof_machdep.c,v 1.10 1998/04/19 15:41:06 bde Exp $
Added a $Id$ keyword. Bruce still needs to put a copyright notice on this file.
1996-04-08 16:41:06 +00:00
*/
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#ifdef GUPROF
#include "opt_i586_guprof.h"
#include "opt_perfmon.h"
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#include <sys/param.h>
#include <sys/systm.h>
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#include <sys/gmon.h>
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#include <sys/kernel.h>
#include <sys/sysctl.h>
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#include <machine/clock.h>
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#include <machine/perfmon.h>
#include <machine/profile.h>
#endif
More merge and update. (1) deleted #if 0 pc98/pc98/mse.c (2) hold per-unit I/O ports in ed_softc pc98/pc98/if_ed.c pc98/pc98/if_ed98.h (3) merge more files by segregating changes into headers. new file (moved from pc98/pc98): i386/isa/aic_98.h deleted: well, it's already in the commit message so I won't repeat the long list here ;) Submitted by: The FreeBSD(98) Development Team
1996-10-30 22:41:46 +00:00
#ifdef PC98
#include <pc98/pc98/pc98.h>
#else
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#include <i386/isa/isa.h>
More merge and update. (1) deleted #if 0 pc98/pc98/mse.c (2) hold per-unit I/O ports in ed_softc pc98/pc98/if_ed.c pc98/pc98/if_ed98.h (3) merge more files by segregating changes into headers. new file (moved from pc98/pc98): i386/isa/aic_98.h deleted: well, it's already in the commit message so I won't repeat the long list here ;) Submitted by: The FreeBSD(98) Development Team
1996-10-30 22:41:46 +00:00
#endif
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#include <i386/isa/timerreg.h>
#ifdef GUPROF
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#define CPUTIME_CLOCK_UNINITIALIZED 0
#define CPUTIME_CLOCK_I8254 1
Rename "i586_ctr" to "tsc" (both upper and lower case instances). Fix a couple of printfs too. Warning: This changes the names of a couple of kernel options!
1997-12-26 20:42:37 +00:00
#define CPUTIME_CLOCK_TSC 2
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#define CPUTIME_CLOCK_I586_PMC 3
#define CPUTIME_CLOCK_I8254_SHIFT 7
int cputime_bias = 1; /* initialize for locality of reference */
static int cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
#ifdef I586_PMC_GUPROF
static u_int cputime_clock_pmc_conf = I586_PMC_GUPROF;
static int cputime_clock_pmc_init;
static struct gmonparam saved_gmp;
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#endif
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#endif /* GUPROF */
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#ifdef __GNUC__
Support compiling with `gcc -pedantic' (don't use hard newlines in (asm) string constants).
1998-04-19 15:41:06 +00:00
__asm(" \n\
GM_STATE = 0 \n\
GMON_PROF_OFF = 3 \n\
\n\
.text \n\
.align 4,0x90 \n\
.globl __mcount \n\
__mcount: \n\
# \n\
# Check that we are profiling. Do it early for speed. \n\
# \n\
cmpl $GMON_PROF_OFF,__gmonparam+GM_STATE \n\
je Lmcount_exit \n\
# \n\
# __mcount is the same as mcount except the caller \n\
# hasn't changed the stack except to call here, so the \n\
# caller's raddr is above our raddr. \n\
# \n\
movl 4(%esp),%edx \n\
jmp Lgot_frompc \n\
\n\
.align 4,0x90 \n\
.globl mcount \n\
mcount: \n\
cmpl $GMON_PROF_OFF,__gmonparam+GM_STATE \n\
je Lmcount_exit \n\
# \n\
# The caller's stack frame has already been built, so \n\
# %ebp is the caller's frame pointer. The caller's \n\
# raddr is in the caller's frame following the caller's \n\
# caller's frame pointer. \n\
# \n\
movl 4(%ebp),%edx \n\
Lgot_frompc: \n\
# \n\
# Our raddr is the caller's pc. \n\
# \n\
movl (%esp),%eax \n\
\n\
pushfl \n\
pushl %eax \n\
pushl %edx \n\
cli \n\
call _mcount \n\
addl $8,%esp \n\
popfl \n\
Lmcount_exit: \n\
ret \n\
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
");
#else /* !__GNUC__ */
#error
#endif /* __GNUC__ */
#ifdef GUPROF
/*
* mexitcount saves the return register(s), loads selfpc and calls
* mexitcount(selfpc) to do the work. Someday it should be in a machine
* dependent file together with cputime(), __mcount and mcount. cputime()
* can't just be put in machdep.c because it has to be compiled without -pg.
*/
#ifdef __GNUC__
Support compiling with `gcc -pedantic' (don't use hard newlines in (asm) string constants).
1998-04-19 15:41:06 +00:00
__asm(" \n\
.text \n\
# \n\
# Dummy label to be seen when gprof -u hides mexitcount. \n\
# \n\
.align 4,0x90 \n\
.globl __mexitcount \n\
__mexitcount: \n\
nop \n\
\n\
GMON_PROF_HIRES = 4 \n\
\n\
.align 4,0x90 \n\
.globl mexitcount \n\
mexitcount: \n\
cmpl $GMON_PROF_HIRES,__gmonparam+GM_STATE \n\
jne Lmexitcount_exit \n\
pushl %edx \n\
pushl %eax \n\
movl 8(%esp),%eax \n\
pushfl \n\
pushl %eax \n\
cli \n\
call _mexitcount \n\
addl $4,%esp \n\
popfl \n\
popl %eax \n\
popl %edx \n\
Lmexitcount_exit: \n\
ret \n\
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
");
#else /* !__GNUC__ */
#error
#endif /* __GNUC__ */
/*
* Return the time elapsed since the last call. The units are machine-
* dependent.
*/
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
int
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
cputime()
{
u_int count;
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
int delta;
Ifdefed the declarations of conditionally used variables.
1998-12-14 18:21:34 +00:00
#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP) && \
defined(PERFMON) && defined(I586_PMC_GUPROF)
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
u_quad_t event_count;
#endif
u_char high, low;
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
static u_int prev_count;
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
Rename "i586_ctr" to "tsc" (both upper and lower case instances). Fix a couple of printfs too. Warning: This changes the names of a couple of kernel options!
1997-12-26 20:42:37 +00:00
if (cputime_clock == CPUTIME_CLOCK_TSC) {
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
count = (u_int)rdtsc();
delta = (int)(count - prev_count);
prev_count = count;
return (delta);
}
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
/*
* XXX permon_read() should be inlined so that the
* perfmon module doesn't need to be compiled with
* profiling disabled and so that it is fast.
*/
perfmon_read(0, &event_count);
count = (u_int)event_count;
delta = (int)(count - prev_count);
prev_count = count;
return (delta);
}
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#endif /* PERFMON && I586_PMC_GUPROF */
#endif /* (I586_CPU || I686_CPU) && !SMP */
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
/*
* Read the current value of the 8254 timer counter 0.
*/
outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
low = inb(TIMER_CNTR0);
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
high = inb(TIMER_CNTR0);
count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
/*
* The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
* While profiling is enabled, this routine is called at least twice
* per timer reset (for mcounting and mexitcounting hardclock()),
* so at most one reset has occurred since the last call, and one
* has occurred iff the current count is larger than the previous
* count. This allows counter underflow to be detected faster
* than in microtime().
*/
delta = prev_count - count;
prev_count = count;
if ((int) delta <= 0)
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
return (delta + (timer0_max_count << CPUTIME_CLOCK_I8254_SHIFT));
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
return (delta);
}
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
static int
sysctl_machdep_cputime_clock SYSCTL_HANDLER_ARGS
{
int clock;
int error;
Ifdefed the declarations of conditionally used variables.
1998-12-14 18:21:34 +00:00
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
int event;
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
struct pmc pmc;
Ifdefed the declarations of conditionally used variables.
1998-12-14 18:21:34 +00:00
#endif
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
clock = cputime_clock;
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (clock == CPUTIME_CLOCK_I586_PMC) {
pmc.pmc_val = cputime_clock_pmc_conf;
clock += pmc.pmc_event;
}
#endif
error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
if (error == 0 && req->newptr != NULL) {
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (clock >= CPUTIME_CLOCK_I586_PMC) {
event = clock - CPUTIME_CLOCK_I586_PMC;
if (event >= 256)
return (EINVAL);
pmc.pmc_num = 0;
pmc.pmc_event = event;
pmc.pmc_unit = 0;
pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
pmc.pmc_mask = 0;
cputime_clock_pmc_conf = pmc.pmc_val;
cputime_clock = CPUTIME_CLOCK_I586_PMC;
} else
#endif
{
if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
return (EINVAL);
cputime_clock = clock;
}
}
return (error);
}
SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
/*
* The start and stop routines need not be here since we turn off profiling
* before calling them. They are here for convenience.
*/
void
startguprof(gp)
struct gmonparam *gp;
{
if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
cputime_clock = CPUTIME_CLOCK_I8254;
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
Rename "i586_ctr" to "tsc" (both upper and lower case instances). Fix a couple of printfs too. Warning: This changes the names of a couple of kernel options!
1997-12-26 20:42:37 +00:00
if (tsc_freq != 0)
cputime_clock = CPUTIME_CLOCK_TSC;
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
#endif
}
gp->profrate = timer_freq << CPUTIME_CLOCK_I8254_SHIFT;
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
Rename "i586_ctr" to "tsc" (both upper and lower case instances). Fix a couple of printfs too. Warning: This changes the names of a couple of kernel options!
1997-12-26 20:42:37 +00:00
if (cputime_clock == CPUTIME_CLOCK_TSC)
gp->profrate = tsc_freq;
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
else if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
if (perfmon_avail() &&
perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
if (perfmon_start(0) != 0)
perfmon_fini(0);
else {
/* XXX 1 event == 1 us. */
gp->profrate = 1000000;
saved_gmp = *gp;
/* Zap overheads. They are invalid. */
gp->cputime_overhead = 0;
gp->mcount_overhead = 0;
gp->mcount_post_overhead = 0;
gp->mcount_pre_overhead = 0;
gp->mexitcount_overhead = 0;
gp->mexitcount_post_overhead = 0;
gp->mexitcount_pre_overhead = 0;
cputime_clock_pmc_init = TRUE;
}
}
}
Added a sysctl (machdep.cputime_clock) to select the clock used by "high resolution" profiling. The available clocks are: - the i8254 clock - on non-SMP i586's and i686's: the TSC - on systems with I586_PMC_GUPROF configured, and PERFMON configured and available: all the performance counters. This is unfinshed (there are problems with locking out the PERFMON device driver, and with losing calibration after switching the clock), but better than static configuration or writing to kmem. Changed ifdefs to avoid generating code for non-working option combinations.
1997-11-24 18:16:23 +00:00
#endif /* PERFMON && I586_PMC_GUPROF */
#endif /* (I586_CPU || I686_CPU) && !SMP */
Improved non-statistical (GUPROF) profiling: - use a more accurate and more efficient method of compensating for overheads. The old method counted too much time against leaf functions. - normally use the Pentium timestamp counter if available. On Pentiums, the times are now accurate to within a couple of cpu clock cycles per function call in the (unlikely) event that there are no cache misses in or caused by the profiling code. - optionally use an arbitrary Pentium event counter if available. - optionally regress to using the i8254 counter. - scaled the i8254 counter by a factor of 128. Now the i8254 counters overflow slightly faster than the TSC counters for a 150MHz Pentium :-) (after about 16 seconds). This is to avoid fractional overheads. files.i386: permon.c temporarily has to be classified as a profiling-routine because a couple of functions in it may be called from profiling code. options.i386: - I586_CTR_GUPROF is currently unused (oops). - I586_PMC_GUPROF should be something like 0x70000 to enable (but not use unless prof_machdep.c is changed) support for Pentium event counters. 7 is a control mode and the counter number 0 is somewhere in the 0000 bits (see perfmon.h for the encoding). profile.h: - added declarations. - cleaned up separation of user mode declarations. prof_machdep.c: Mostly clock-select changes. The default clock can be changed by editing kmem. There should be a sysctl for this. subr_prof.c: - added copyright. - calibrate overheads for the new method. - documented new method. - fixed races and and machine dependencies in start/stop code. mcount.c: Use the new overhead compensation method. gmon.h: - changed GPROF4 counter type from unsigned to int. Oops, this should be machine-dependent and/or int32_t. - reorganized overhead counters. Submitted by: Pentium event counter changes mostly by wollman
1996-10-17 19:32:31 +00:00
cputime_bias = 0;
cputime();
}
void
stopguprof(gp)
struct gmonparam *gp;
{
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (cputime_clock_pmc_init) {
*gp = saved_gmp;
perfmon_fini(0);
cputime_clock_pmc_init = FALSE;
}
#endif
}
#else /* !GUPROF */
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
#ifdef __GNUC__
Support compiling with `gcc -pedantic' (don't use hard newlines in (asm) string constants).
1998-04-19 15:41:06 +00:00
__asm(" \n\
.text \n\
.align 4,0x90 \n\
.globl mexitcount \n\
mexitcount: \n\
ret \n\
Implemented non-statistical kernel profiling. This is based on looking at a high resolution clock for each of the following events: function call, function return, interrupt entry, interrupt exit, and interesting branches. The differences between the times of these events are added at appropriate places in a ordinary histogram (as if very fast statistical profiling sampled the pc at those places) so that ordinary gprof can be used to analyze the times. gmon.h: Histogram counters need to be 4 bytes for microsecond resolutions. They will need to be larger for the 586 clock. The comments were vax-centric and wrong even on vaxes. Does anyone disagree? gprof4.c: The standard gprof should support counters of all integral sizes and the size of the counter should be in the gmon header. This hack will do until then. (Use gprof4 -u to examine the results of non-statistical profiling.) config/*: Non-statistical profiling is configured with `config -pp'. `config -p' still gives ordinary profiling. kgmon/*: Non-statistical profiling is enabled with `kgmon -B'. `kgmon -b' still enables ordinary profiling (and distables non-statistical profiling) if non-statistical profiling is configured.
1995-12-29 15:30:05 +00:00
");
#else /* !__GNUC__ */
#error
#endif /* __GNUC__ */
#endif /* GUPROF */
Reference in New Issue Copy Permalink