1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
|
/*-
* Copyright (c) 1982, 1986, 1993
* The Regents of the University of California. 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.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
*
* @(#)subr_prof.c 8.3 (Berkeley) 9/23/93
* $FreeBSD$
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/resourcevar.h>
#include <sys/sysctl.h>
#include <machine/cpu.h>
#ifdef GPROF
#include <sys/malloc.h>
#include <sys/gmon.h>
#undef MCOUNT
static MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer");
static void kmstartup(void *);
SYSINIT(kmem, SI_SUB_KPROF, SI_ORDER_FIRST, kmstartup, NULL)
struct gmonparam _gmonparam = { GMON_PROF_OFF };
#ifdef GUPROF
#include <machine/asmacros.h>
void
nullfunc_loop_profiled()
{
int i;
for (i = 0; i < CALIB_SCALE; i++)
nullfunc_profiled();
}
#define nullfunc_loop_profiled_end nullfunc_profiled /* XXX */
void
nullfunc_profiled()
{
}
#endif /* GUPROF */
/*
* Update the histograms to support extending the text region arbitrarily.
* This is done slightly naively (no sparse regions), so will waste slight
* amounts of memory, but will overall work nicely enough to allow profiling
* of KLDs.
*/
void
kmupetext(uintfptr_t nhighpc)
{
struct gmonparam np; /* slightly large */
struct gmonparam *p = &_gmonparam;
char *cp;
GIANT_REQUIRED;
bcopy(p, &np, sizeof(*p));
np.highpc = ROUNDUP(nhighpc, HISTFRACTION * sizeof(HISTCOUNTER));
if (np.highpc <= p->highpc)
return;
np.textsize = np.highpc - p->lowpc;
np.kcountsize = np.textsize / HISTFRACTION;
np.hashfraction = HASHFRACTION;
np.fromssize = np.textsize / HASHFRACTION;
np.tolimit = np.textsize * ARCDENSITY / 100;
if (np.tolimit < MINARCS)
np.tolimit = MINARCS;
else if (np.tolimit > MAXARCS)
np.tolimit = MAXARCS;
np.tossize = np.tolimit * sizeof(struct tostruct);
cp = malloc(np.kcountsize + np.fromssize + np.tossize,
M_GPROF, 0);
/*
* Check for something else extending highpc while we slept.
*/
if (np.highpc <= p->highpc) {
free(cp, M_GPROF);
return;
}
np.tos = (struct tostruct *)cp;
cp += np.tossize;
np.kcount = (HISTCOUNTER *)cp;
cp += np.kcountsize;
np.froms = (u_short *)cp;
#ifdef GUPROF
/* Reinitialize pointers to overhead counters. */
np.cputime_count = &KCOUNT(&np, PC_TO_I(&np, cputime));
np.mcount_count = &KCOUNT(&np, PC_TO_I(&np, mcount));
np.mexitcount_count = &KCOUNT(&np, PC_TO_I(&np, mexitcount));
#endif
critical_enter();
bcopy(p->tos, np.tos, p->tossize);
bzero((char *)np.tos + p->tossize, np.tossize - p->tossize);
bcopy(p->kcount, np.kcount, p->kcountsize);
bzero((char *)np.kcount + p->kcountsize, np.kcountsize -
p->kcountsize);
bcopy(p->froms, np.froms, p->fromssize);
bzero((char *)np.froms + p->fromssize, np.fromssize - p->fromssize);
cp = (char *)p->tos;
bcopy(&np, p, sizeof(*p));
critical_exit();
free(cp, M_GPROF);
}
static void
kmstartup(dummy)
void *dummy;
{
char *cp;
struct gmonparam *p = &_gmonparam;
#ifdef GUPROF
int cputime_overhead;
int empty_loop_time;
int i;
int mcount_overhead;
int mexitcount_overhead;
int nullfunc_loop_overhead;
int nullfunc_loop_profiled_time;
uintfptr_t tmp_addr;
#endif
/*
* Round lowpc and highpc to multiples of the density we're using
* so the rest of the scaling (here and in gprof) stays in ints.
*/
p->lowpc = ROUNDDOWN((u_long)btext, HISTFRACTION * sizeof(HISTCOUNTER));
p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
p->textsize = p->highpc - p->lowpc;
printf("Profiling kernel, textsize=%lu [%x..%x]\n",
p->textsize, p->lowpc, p->highpc);
p->kcountsize = p->textsize / HISTFRACTION;
p->hashfraction = HASHFRACTION;
p->fromssize = p->textsize / HASHFRACTION;
p->tolimit = p->textsize * ARCDENSITY / 100;
if (p->tolimit < MINARCS)
p->tolimit = MINARCS;
else if (p->tolimit > MAXARCS)
p->tolimit = MAXARCS;
p->tossize = p->tolimit * sizeof(struct tostruct);
cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
M_GPROF, M_ZERO);
p->tos = (struct tostruct *)cp;
cp += p->tossize;
p->kcount = (HISTCOUNTER *)cp;
cp += p->kcountsize;
p->froms = (u_short *)cp;
#ifdef GUPROF
/* Initialize pointers to overhead counters. */
p->cputime_count = &KCOUNT(p, PC_TO_I(p, cputime));
p->mcount_count = &KCOUNT(p, PC_TO_I(p, mcount));
p->mexitcount_count = &KCOUNT(p, PC_TO_I(p, mexitcount));
/*
* Disable interrupts to avoid interference while we calibrate
* things.
*/
critical_enter();
/*
* Determine overheads.
* XXX this needs to be repeated for each useful timer/counter.
*/
cputime_overhead = 0;
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
cputime_overhead += cputime();
empty_loop();
startguprof(p);
empty_loop();
empty_loop_time = cputime();
nullfunc_loop_profiled();
/*
* Start profiling. There won't be any normal function calls since
* interrupts are disabled, but we will call the profiling routines
* directly to determine their overheads.
*/
p->state = GMON_PROF_HIRES;
startguprof(p);
nullfunc_loop_profiled();
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
#if defined(__i386__) && __GNUC__ >= 2
__asm("pushl %0; call __mcount; popl %%ecx"
:
: "i" (profil)
: "ax", "bx", "cx", "dx", "memory");
#elif defined(lint)
#else
#error
#endif
mcount_overhead = KCOUNT(p, PC_TO_I(p, profil));
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
#if defined(__i386__) && __GNUC__ >= 2
__asm("call " __XSTRING(HIDENAME(mexitcount)) "; 1:"
: : : "ax", "bx", "cx", "dx", "memory");
__asm("movl $1b,%0" : "=rm" (tmp_addr));
#elif defined(lint)
#else
#error
#endif
mexitcount_overhead = KCOUNT(p, PC_TO_I(p, tmp_addr));
p->state = GMON_PROF_OFF;
stopguprof(p);
critical_exit();
nullfunc_loop_profiled_time = 0;
for (tmp_addr = (uintfptr_t)nullfunc_loop_profiled;
tmp_addr < (uintfptr_t)nullfunc_loop_profiled_end;
tmp_addr += HISTFRACTION * sizeof(HISTCOUNTER))
nullfunc_loop_profiled_time += KCOUNT(p, PC_TO_I(p, tmp_addr));
#define CALIB_DOSCALE(count) (((count) + CALIB_SCALE / 3) / CALIB_SCALE)
#define c2n(count, freq) ((int)((count) * 1000000000LL / freq))
printf("cputime %d, empty_loop %d, nullfunc_loop_profiled %d, mcount %d, mexitcount %d\n",
CALIB_DOSCALE(c2n(cputime_overhead, p->profrate)),
CALIB_DOSCALE(c2n(empty_loop_time, p->profrate)),
CALIB_DOSCALE(c2n(nullfunc_loop_profiled_time, p->profrate)),
CALIB_DOSCALE(c2n(mcount_overhead, p->profrate)),
CALIB_DOSCALE(c2n(mexitcount_overhead, p->profrate)));
cputime_overhead -= empty_loop_time;
mcount_overhead -= empty_loop_time;
mexitcount_overhead -= empty_loop_time;
/*-
* Profiling overheads are determined by the times between the
* following events:
* MC1: mcount() is called
* MC2: cputime() (called from mcount()) latches the timer
* MC3: mcount() completes
* ME1: mexitcount() is called
* ME2: cputime() (called from mexitcount()) latches the timer
* ME3: mexitcount() completes.
* The times between the events vary slightly depending on instruction
* combination and cache misses, etc. Attempt to determine the
* minimum times. These can be subtracted from the profiling times
* without much risk of reducing the profiling times below what they
* would be when profiling is not configured. Abbreviate:
* ab = minimum time between MC1 and MC3
* a = minumum time between MC1 and MC2
* b = minimum time between MC2 and MC3
* cd = minimum time between ME1 and ME3
* c = minimum time between ME1 and ME2
* d = minimum time between ME2 and ME3.
* These satisfy the relations:
* ab <= mcount_overhead (just measured)
* a + b <= ab
* cd <= mexitcount_overhead (just measured)
* c + d <= cd
* a + d <= nullfunc_loop_profiled_time (just measured)
* a >= 0, b >= 0, c >= 0, d >= 0.
* Assume that ab and cd are equal to the minimums.
*/
p->cputime_overhead = CALIB_DOSCALE(cputime_overhead);
p->mcount_overhead = CALIB_DOSCALE(mcount_overhead - cputime_overhead);
p->mexitcount_overhead = CALIB_DOSCALE(mexitcount_overhead
- cputime_overhead);
nullfunc_loop_overhead = nullfunc_loop_profiled_time - empty_loop_time;
p->mexitcount_post_overhead = CALIB_DOSCALE((mcount_overhead
- nullfunc_loop_overhead)
/ 4);
p->mexitcount_pre_overhead = p->mexitcount_overhead
+ p->cputime_overhead
- p->mexitcount_post_overhead;
p->mcount_pre_overhead = CALIB_DOSCALE(nullfunc_loop_overhead)
- p->mexitcount_post_overhead;
p->mcount_post_overhead = p->mcount_overhead
+ p->cputime_overhead
- p->mcount_pre_overhead;
printf(
"Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d nsec\n",
c2n(p->cputime_overhead, p->profrate),
c2n(p->mcount_overhead, p->profrate),
c2n(p->mcount_pre_overhead, p->profrate),
c2n(p->mcount_post_overhead, p->profrate),
c2n(p->cputime_overhead, p->profrate),
c2n(p->mexitcount_overhead, p->profrate),
c2n(p->mexitcount_pre_overhead, p->profrate),
c2n(p->mexitcount_post_overhead, p->profrate));
printf(
"Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d cycles\n",
p->cputime_overhead, p->mcount_overhead,
p->mcount_pre_overhead, p->mcount_post_overhead,
p->cputime_overhead, p->mexitcount_overhead,
p->mexitcount_pre_overhead, p->mexitcount_post_overhead);
#endif /* GUPROF */
}
/*
* Return kernel profiling information.
*/
static int
sysctl_kern_prof(SYSCTL_HANDLER_ARGS)
{
int *name = (int *) arg1;
u_int namelen = arg2;
struct gmonparam *gp = &_gmonparam;
int error;
int state;
/* all sysctl names at this level are terminal */
if (namelen != 1)
return (ENOTDIR); /* overloaded */
switch (name[0]) {
case GPROF_STATE:
state = gp->state;
error = sysctl_handle_int(oidp, &state, 0, req);
if (error)
return (error);
if (!req->newptr)
return (0);
if (state == GMON_PROF_OFF) {
gp->state = state;
PROC_LOCK(&proc0);
stopprofclock(&proc0);
PROC_UNLOCK(&proc0);
stopguprof(gp);
} else if (state == GMON_PROF_ON) {
gp->state = GMON_PROF_OFF;
stopguprof(gp);
gp->profrate = profhz;
startprofclock(&proc0);
gp->state = state;
#ifdef GUPROF
} else if (state == GMON_PROF_HIRES) {
gp->state = GMON_PROF_OFF;
PROC_LOCK(&proc0);
stopprofclock(&proc0);
PROC_UNLOCK(&proc0);
startguprof(gp);
gp->state = state;
#endif
} else if (state != gp->state)
return (EINVAL);
return (0);
case GPROF_COUNT:
return (sysctl_handle_opaque(oidp,
gp->kcount, gp->kcountsize, req));
case GPROF_FROMS:
return (sysctl_handle_opaque(oidp,
gp->froms, gp->fromssize, req));
case GPROF_TOS:
return (sysctl_handle_opaque(oidp,
gp->tos, gp->tossize, req));
case GPROF_GMONPARAM:
return (sysctl_handle_opaque(oidp, gp, sizeof *gp, req));
default:
return (EOPNOTSUPP);
}
/* NOTREACHED */
}
SYSCTL_NODE(_kern, KERN_PROF, prof, CTLFLAG_RW, sysctl_kern_prof, "");
#endif /* GPROF */
/*
* Profiling system call.
*
* The scale factor is a fixed point number with 16 bits of fraction, so that
* 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling.
*/
#ifndef _SYS_SYSPROTO_H_
struct profil_args {
caddr_t samples;
size_t size;
size_t offset;
u_int scale;
};
#endif
/*
* MPSAFE
*/
/* ARGSUSED */
int
profil(td, uap)
struct thread *td;
register struct profil_args *uap;
{
struct uprof *upp;
int s;
int error = 0;
mtx_lock(&Giant);
if (uap->scale > (1 << 16)) {
error = EINVAL;
goto done2;
}
if (uap->scale == 0) {
PROC_LOCK(td->td_proc);
stopprofclock(td->td_proc);
PROC_UNLOCK(td->td_proc);
goto done2;
}
upp = &td->td_proc->p_stats->p_prof;
/* Block profile interrupts while changing state. */
s = splstatclock();
upp->pr_off = uap->offset;
upp->pr_scale = uap->scale;
upp->pr_base = uap->samples;
upp->pr_size = uap->size;
startprofclock(td->td_proc);
splx(s);
done2:
mtx_unlock(&Giant);
return (error);
}
/*
* Scale is a fixed-point number with the binary point 16 bits
* into the value, and is <= 1.0. pc is at most 32 bits, so the
* intermediate result is at most 48 bits.
*/
#define PC_TO_INDEX(pc, prof) \
((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
(u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
/*
* Collect user-level profiling statistics; called on a profiling tick,
* when a process is running in user-mode. This routine may be called
* from an interrupt context. We try to update the user profiling buffers
* cheaply with fuswintr() and suswintr(). If that fails, we revert to
* an AST that will vector us to trap() with a context in which copyin
* and copyout will work. Trap will then call addupc_task().
*
* Note that we may (rarely) not get around to the AST soon enough, and
* lose profile ticks when the next tick overwrites this one, but in this
* case the system is overloaded and the profile is probably already
* inaccurate.
*/
void
addupc_intr(struct thread *td, uintptr_t pc, u_int ticks)
{
struct uprof *prof;
caddr_t addr;
u_int i;
int v;
if (ticks == 0)
return;
prof = &td->td_proc->p_stats->p_prof;
if (pc < prof->pr_off ||
(i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
return; /* out of range; ignore */
addr = prof->pr_base + i;
if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) {
mtx_lock_spin(&sched_lock);
td->td_praddr = pc;
td->td_prticks = ticks;
td->td_flags |= (TDF_OWEUPC | TDF_ASTPENDING);
mtx_unlock_spin(&sched_lock);
}
}
/*
* Much like before, but we can afford to take faults here. If the
* update fails, we simply turn off profiling.
* XXXKSE, don't use kse unless we got sched lock.
*/
void
addupc_task(struct thread *td, uintptr_t pc, u_int ticks)
{
struct proc *p = td->td_proc;
register struct uprof *prof;
register caddr_t addr;
register u_int i;
u_short v;
int stop = 0;
if (ticks == 0)
return;
PROC_LOCK(p);
mtx_lock_spin(&sched_lock);
if (!(p->p_sflag & PS_PROFIL)) {
mtx_unlock_spin(&sched_lock);
PROC_UNLOCK(p);
return;
}
p->p_profthreads++;
mtx_unlock_spin(&sched_lock);
PROC_UNLOCK(p);
prof = &p->p_stats->p_prof;
if (pc < prof->pr_off ||
(i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) {
goto out;
}
addr = prof->pr_base + i;
if (copyin(addr, &v, sizeof(v)) == 0) {
v += ticks;
if (copyout(&v, addr, sizeof(v)) == 0)
goto out;
}
stop = 1;
out:
PROC_LOCK(p);
if (--p->p_profthreads == 0) {
if (p->p_sflag & PS_STOPPROF) {
wakeup(&p->p_profthreads);
stop = 0;
}
}
if (stop)
stopprofclock(p);
PROC_UNLOCK(p);
}
#if defined(__i386__) && __GNUC__ >= 2
/*
* Support for "--test-coverage --profile-arcs" in GCC.
*
* We need to call all the functions in the .ctor section, in order
* to get all the counter-arrays strung into a list.
*
* XXX: the .ctors call __bb_init_func which is located in over in
* XXX: i386/i386/support.s for historical reasons. There is probably
* XXX: no reason for that to be assembler anymore, but doing it right
* XXX: in MI C code requires one to reverse-engineer the type-selection
* XXX: inside GCC. Have fun.
*
* XXX: Worrisome perspective: Calling the .ctors may make C++ in the
* XXX: kernel feasible. Don't.
*/
typedef void (*ctor_t)(void);
extern ctor_t _start_ctors, _stop_ctors;
static void
tcov_init(void *foo __unused)
{
ctor_t *p, q;
for (p = &_start_ctors; p < &_stop_ctors; p++) {
q = *p;
q();
}
}
SYSINIT(tcov_init, SI_SUB_KPROF, SI_ORDER_SECOND, tcov_init, NULL)
/*
* GCC contains magic to recognize calls to for instance execve() and
* puts in calls to this function to preserve the profile counters.
* XXX: Put zinging punchline here.
*/
void __bb_fork_func(void);
void
__bb_fork_func(void)
{
}
#endif
|