2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1996 Bruce D. Evans.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
33 #include "opt_i586_guprof.h"
34 #include "opt_perfmon.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
40 #include <sys/eventhandler.h>
42 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
46 #include <machine/clock.h>
47 #include <machine/perfmon.h>
48 #include <machine/timerreg.h>
50 #define CPUTIME_CLOCK_UNINITIALIZED 0
51 #define CPUTIME_CLOCK_I8254 1
52 #define CPUTIME_CLOCK_TSC 2
53 #define CPUTIME_CLOCK_I586_PMC 3
54 #define CPUTIME_CLOCK_I8254_SHIFT 7
56 int cputime_bias = 1; /* initialize for locality of reference */
58 static int cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
59 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
60 static u_int cputime_clock_pmc_conf = I586_PMC_GUPROF;
61 static int cputime_clock_pmc_init;
62 static struct gmonparam saved_gmp;
64 #if defined(I586_CPU) || defined(I686_CPU)
65 static int cputime_prof_active;
77 .type __mcount,@function \n\
80 # Check that we are profiling. Do it early for speed. \n\
82 cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
85 # __mcount is the same as [.]mcount except the caller \n\
86 # hasn't changed the stack except to call here, so the \n\
87 # caller's raddr is above our raddr. \n\
95 cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
98 # The caller's stack frame has already been built, so \n\
99 # %ebp is the caller's frame pointer. The caller's \n\
100 # raddr is in the caller's frame following the caller's \n\
101 # caller's frame pointer. \n\
103 movl 4(%ebp),%edx \n\
106 # Our raddr is the caller's pc. \n\
120 #else /* !__GNUCLIKE_ASM */
121 #error "this file needs to be ported to your compiler"
122 #endif /* __GNUCLIKE_ASM */
126 * [.]mexitcount saves the return register(s), loads selfpc and calls
127 * mexitcount(selfpc) to do the work. Someday it should be in a machine
128 * dependent file together with cputime(), __mcount and [.]mcount. cputime()
129 * can't just be put in machdep.c because it has to be compiled without -pg.
131 #ifdef __GNUCLIKE_ASM
135 # Dummy label to be seen when gprof -u hides [.]mexitcount. \n\
138 .globl __mexitcount \n\
139 .type __mexitcount,@function \n\
143 GMON_PROF_HIRES = 4 \n\
146 .globl .mexitcount \n\
148 cmpl $GMON_PROF_HIRES,_gmonparam+GM_STATE \n\
149 jne .mexitcount_exit \n\
152 movl 8(%esp),%eax \n\
161 .mexitcount_exit: \n\
164 #endif /* __GNUCLIKE_ASM */
167 * Return the time elapsed since the last call. The units are machine-
175 #if (defined(I586_CPU) || defined(I686_CPU)) && \
176 defined(PERFMON) && defined(I586_PMC_GUPROF) && !defined(SMP)
177 u_quad_t event_count;
180 static u_int prev_count;
182 #if defined(I586_CPU) || defined(I686_CPU)
183 if (cputime_clock == CPUTIME_CLOCK_TSC) {
185 * Scale the TSC a little to make cputime()'s frequency
186 * fit in an int, assuming that the TSC frequency fits
187 * in a u_int. Use a fixed scale since dynamic scaling
188 * would be slower and we can't really use the low bit
191 count = (u_int)rdtsc() & ~1u;
192 delta = (int)(count - prev_count) >> 1;
196 #if defined(PERFMON) && defined(I586_PMC_GUPROF) && !defined(SMP)
197 if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
199 * XXX permon_read() should be inlined so that the
200 * perfmon module doesn't need to be compiled with
201 * profiling disabled and so that it is fast.
203 perfmon_read(0, &event_count);
205 count = (u_int)event_count;
206 delta = (int)(count - prev_count);
210 #endif /* PERFMON && I586_PMC_GUPROF && !SMP */
211 #endif /* I586_CPU || I686_CPU */
214 * Read the current value of the 8254 timer counter 0.
216 outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
217 low = inb(TIMER_CNTR0);
218 high = inb(TIMER_CNTR0);
219 count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
222 * The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
223 * While profiling is enabled, this routine is called at least twice
224 * per timer reset (for mcounting and mexitcounting hardclock()),
225 * so at most one reset has occurred since the last call, and one
226 * has occurred iff the current count is larger than the previous
227 * count. This allows counter underflow to be detected faster
228 * than in microtime().
230 delta = prev_count - count;
232 if ((int) delta <= 0)
233 return (delta + (i8254_max_count << CPUTIME_CLOCK_I8254_SHIFT));
238 sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
242 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
247 clock = cputime_clock;
248 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
249 if (clock == CPUTIME_CLOCK_I586_PMC) {
250 pmc.pmc_val = cputime_clock_pmc_conf;
251 clock += pmc.pmc_event;
254 error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
255 if (error == 0 && req->newptr != NULL) {
256 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
257 if (clock >= CPUTIME_CLOCK_I586_PMC) {
258 event = clock - CPUTIME_CLOCK_I586_PMC;
262 pmc.pmc_event = event;
264 pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
266 cputime_clock_pmc_conf = pmc.pmc_val;
267 cputime_clock = CPUTIME_CLOCK_I586_PMC;
271 if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
273 cputime_clock = clock;
279 SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
280 0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
283 * The start and stop routines need not be here since we turn off profiling
284 * before calling them. They are here for convenience.
289 struct gmonparam *gp;
291 #if defined(I586_CPU) || defined(I686_CPU)
294 freq = atomic_load_acq_64(&tsc_freq);
295 if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
296 if (freq != 0 && mp_ncpus == 1)
297 cputime_clock = CPUTIME_CLOCK_TSC;
299 cputime_clock = CPUTIME_CLOCK_I8254;
301 if (cputime_clock == CPUTIME_CLOCK_TSC) {
302 gp->profrate = freq >> 1;
303 cputime_prof_active = 1;
305 gp->profrate = i8254_freq << CPUTIME_CLOCK_I8254_SHIFT;
306 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
307 if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
308 if (perfmon_avail() &&
309 perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
310 if (perfmon_start(0) != 0)
313 /* XXX 1 event == 1 us. */
314 gp->profrate = 1000000;
318 /* Zap overheads. They are invalid. */
319 gp->cputime_overhead = 0;
320 gp->mcount_overhead = 0;
321 gp->mcount_post_overhead = 0;
322 gp->mcount_pre_overhead = 0;
323 gp->mexitcount_overhead = 0;
324 gp->mexitcount_post_overhead = 0;
325 gp->mexitcount_pre_overhead = 0;
327 cputime_clock_pmc_init = TRUE;
331 #endif /* PERFMON && I586_PMC_GUPROF */
332 #else /* !(I586_CPU || I686_CPU) */
333 if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED)
334 cputime_clock = CPUTIME_CLOCK_I8254;
335 gp->profrate = i8254_freq << CPUTIME_CLOCK_I8254_SHIFT;
336 #endif /* I586_CPU || I686_CPU */
343 struct gmonparam *gp;
345 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
346 if (cputime_clock_pmc_init) {
349 cputime_clock_pmc_init = FALSE;
352 #if defined(I586_CPU) || defined(I686_CPU)
353 if (cputime_clock == CPUTIME_CLOCK_TSC)
354 cputime_prof_active = 0;
358 #if defined(I586_CPU) || defined(I686_CPU)
359 /* If the cpu frequency changed while profiling, report a warning. */
361 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
365 * If there was an error during the transition or
366 * TSC is P-state invariant, don't do anything.
368 if (status != 0 || tsc_is_invariant)
370 if (cputime_prof_active && cputime_clock == CPUTIME_CLOCK_TSC)
371 printf("warning: cpu freq changed while profiling active\n");
374 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
375 EVENTHANDLER_PRI_ANY);
376 #endif /* I586_CPU || I686_CPU */