2 * intel_pt.c: Intel Processor Trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 #include <linux/kernel.h>
18 #include <linux/types.h>
19 #include <linux/bitops.h>
20 #include <linux/log2.h>
23 #include "../../perf.h"
24 #include "../../util/session.h"
25 #include "../../util/event.h"
26 #include "../../util/evlist.h"
27 #include "../../util/evsel.h"
28 #include "../../util/cpumap.h"
29 #include <subcmd/parse-options.h>
30 #include "../../util/parse-events.h"
31 #include "../../util/pmu.h"
32 #include "../../util/debug.h"
33 #include "../../util/auxtrace.h"
34 #include "../../util/tsc.h"
35 #include "../../util/intel-pt.h"
37 #define KiB(x) ((x) * 1024)
38 #define MiB(x) ((x) * 1024 * 1024)
39 #define KiB_MASK(x) (KiB(x) - 1)
40 #define MiB_MASK(x) (MiB(x) - 1)
42 #define INTEL_PT_DEFAULT_SAMPLE_SIZE KiB(4)
44 #define INTEL_PT_MAX_SAMPLE_SIZE KiB(60)
46 #define INTEL_PT_PSB_PERIOD_NEAR 256
48 struct intel_pt_snapshot_ref
{
54 struct intel_pt_recording
{
55 struct auxtrace_record itr
;
56 struct perf_pmu
*intel_pt_pmu
;
57 int have_sched_switch
;
58 struct perf_evlist
*evlist
;
60 bool snapshot_init_done
;
62 size_t snapshot_ref_buf_size
;
64 struct intel_pt_snapshot_ref
*snapshot_refs
;
67 static int intel_pt_parse_terms_with_default(struct list_head
*formats
,
71 struct list_head
*terms
;
72 struct perf_event_attr attr
= { .size
= 0, };
75 terms
= malloc(sizeof(struct list_head
));
79 INIT_LIST_HEAD(terms
);
81 err
= parse_events_terms(terms
, str
);
85 attr
.config
= *config
;
86 err
= perf_pmu__config_terms(formats
, &attr
, terms
, true, NULL
);
90 *config
= attr
.config
;
92 parse_events_terms__delete(terms
);
96 static int intel_pt_parse_terms(struct list_head
*formats
, const char *str
,
100 return intel_pt_parse_terms_with_default(formats
, str
, config
);
103 static u64
intel_pt_masked_bits(u64 mask
, u64 bits
)
105 const u64 top_bit
= 1ULL << 63;
109 for (i
= 0; i
< 64; i
++) {
110 if (mask
& top_bit
) {
122 static int intel_pt_read_config(struct perf_pmu
*intel_pt_pmu
, const char *str
,
123 struct perf_evlist
*evlist
, u64
*res
)
125 struct perf_evsel
*evsel
;
130 mask
= perf_pmu__format_bits(&intel_pt_pmu
->format
, str
);
134 evlist__for_each(evlist
, evsel
) {
135 if (evsel
->attr
.type
== intel_pt_pmu
->type
) {
136 *res
= intel_pt_masked_bits(mask
, evsel
->attr
.config
);
144 static size_t intel_pt_psb_period(struct perf_pmu
*intel_pt_pmu
,
145 struct perf_evlist
*evlist
)
148 int err
, topa_multiple_entries
;
151 if (perf_pmu__scan_file(intel_pt_pmu
, "caps/topa_multiple_entries",
152 "%d", &topa_multiple_entries
) != 1)
153 topa_multiple_entries
= 0;
156 * Use caps/topa_multiple_entries to indicate early hardware that had
157 * extra frequent PSBs.
159 if (!topa_multiple_entries
) {
164 err
= intel_pt_read_config(intel_pt_pmu
, "psb_period", evlist
, &val
);
168 psb_period
= 1 << (val
+ 11);
170 pr_debug2("%s psb_period %zu\n", intel_pt_pmu
->name
, psb_period
);
174 static int intel_pt_pick_bit(int bits
, int target
)
178 for (pos
= 0; bits
; bits
>>= 1, pos
++) {
180 if (pos
<= target
|| pick
< 0)
190 static u64
intel_pt_default_config(struct perf_pmu
*intel_pt_pmu
)
193 int mtc
, mtc_periods
= 0, mtc_period
;
194 int psb_cyc
, psb_periods
, psb_period
;
198 pos
+= scnprintf(buf
+ pos
, sizeof(buf
) - pos
, "tsc");
200 if (perf_pmu__scan_file(intel_pt_pmu
, "caps/mtc", "%d",
205 if (perf_pmu__scan_file(intel_pt_pmu
, "caps/mtc_periods", "%x",
209 mtc_period
= intel_pt_pick_bit(mtc_periods
, 3);
210 pos
+= scnprintf(buf
+ pos
, sizeof(buf
) - pos
,
211 ",mtc,mtc_period=%d", mtc_period
);
215 if (perf_pmu__scan_file(intel_pt_pmu
, "caps/psb_cyc", "%d",
219 if (psb_cyc
&& mtc_periods
) {
220 if (perf_pmu__scan_file(intel_pt_pmu
, "caps/psb_periods", "%x",
224 psb_period
= intel_pt_pick_bit(psb_periods
, 3);
225 pos
+= scnprintf(buf
+ pos
, sizeof(buf
) - pos
,
226 ",psb_period=%d", psb_period
);
230 pr_debug2("%s default config: %s\n", intel_pt_pmu
->name
, buf
);
232 intel_pt_parse_terms(&intel_pt_pmu
->format
, buf
, &config
);
237 static int intel_pt_parse_snapshot_options(struct auxtrace_record
*itr
,
238 struct record_opts
*opts
,
241 struct intel_pt_recording
*ptr
=
242 container_of(itr
, struct intel_pt_recording
, itr
);
243 unsigned long long snapshot_size
= 0;
247 snapshot_size
= strtoull(str
, &endptr
, 0);
248 if (*endptr
|| snapshot_size
> SIZE_MAX
)
252 opts
->auxtrace_snapshot_mode
= true;
253 opts
->auxtrace_snapshot_size
= snapshot_size
;
255 ptr
->snapshot_size
= snapshot_size
;
260 struct perf_event_attr
*
261 intel_pt_pmu_default_config(struct perf_pmu
*intel_pt_pmu
)
263 struct perf_event_attr
*attr
;
265 attr
= zalloc(sizeof(struct perf_event_attr
));
269 attr
->config
= intel_pt_default_config(intel_pt_pmu
);
271 intel_pt_pmu
->selectable
= true;
277 intel_pt_info_priv_size(struct auxtrace_record
*itr __maybe_unused
,
278 struct perf_evlist
*evlist __maybe_unused
)
280 return INTEL_PT_AUXTRACE_PRIV_SIZE
;
283 static void intel_pt_tsc_ctc_ratio(u32
*n
, u32
*d
)
285 unsigned int eax
= 0, ebx
= 0, ecx
= 0, edx
= 0;
287 __get_cpuid(0x15, &eax
, &ebx
, &ecx
, &edx
);
292 static int intel_pt_info_fill(struct auxtrace_record
*itr
,
293 struct perf_session
*session
,
294 struct auxtrace_info_event
*auxtrace_info
,
297 struct intel_pt_recording
*ptr
=
298 container_of(itr
, struct intel_pt_recording
, itr
);
299 struct perf_pmu
*intel_pt_pmu
= ptr
->intel_pt_pmu
;
300 struct perf_event_mmap_page
*pc
;
301 struct perf_tsc_conversion tc
= { .time_mult
= 0, };
302 bool cap_user_time_zero
= false, per_cpu_mmaps
;
303 u64 tsc_bit
, mtc_bit
, mtc_freq_bits
, cyc_bit
, noretcomp_bit
;
304 u32 tsc_ctc_ratio_n
, tsc_ctc_ratio_d
;
307 if (priv_size
!= INTEL_PT_AUXTRACE_PRIV_SIZE
)
310 intel_pt_parse_terms(&intel_pt_pmu
->format
, "tsc", &tsc_bit
);
311 intel_pt_parse_terms(&intel_pt_pmu
->format
, "noretcomp",
313 intel_pt_parse_terms(&intel_pt_pmu
->format
, "mtc", &mtc_bit
);
314 mtc_freq_bits
= perf_pmu__format_bits(&intel_pt_pmu
->format
,
316 intel_pt_parse_terms(&intel_pt_pmu
->format
, "cyc", &cyc_bit
);
318 intel_pt_tsc_ctc_ratio(&tsc_ctc_ratio_n
, &tsc_ctc_ratio_d
);
320 if (!session
->evlist
->nr_mmaps
)
323 pc
= session
->evlist
->mmap
[0].base
;
325 err
= perf_read_tsc_conversion(pc
, &tc
);
327 if (err
!= -EOPNOTSUPP
)
330 cap_user_time_zero
= tc
.time_mult
!= 0;
332 if (!cap_user_time_zero
)
333 ui__warning("Intel Processor Trace: TSC not available\n");
336 per_cpu_mmaps
= !cpu_map__empty(session
->evlist
->cpus
);
338 auxtrace_info
->type
= PERF_AUXTRACE_INTEL_PT
;
339 auxtrace_info
->priv
[INTEL_PT_PMU_TYPE
] = intel_pt_pmu
->type
;
340 auxtrace_info
->priv
[INTEL_PT_TIME_SHIFT
] = tc
.time_shift
;
341 auxtrace_info
->priv
[INTEL_PT_TIME_MULT
] = tc
.time_mult
;
342 auxtrace_info
->priv
[INTEL_PT_TIME_ZERO
] = tc
.time_zero
;
343 auxtrace_info
->priv
[INTEL_PT_CAP_USER_TIME_ZERO
] = cap_user_time_zero
;
344 auxtrace_info
->priv
[INTEL_PT_TSC_BIT
] = tsc_bit
;
345 auxtrace_info
->priv
[INTEL_PT_NORETCOMP_BIT
] = noretcomp_bit
;
346 auxtrace_info
->priv
[INTEL_PT_HAVE_SCHED_SWITCH
] = ptr
->have_sched_switch
;
347 auxtrace_info
->priv
[INTEL_PT_SNAPSHOT_MODE
] = ptr
->snapshot_mode
;
348 auxtrace_info
->priv
[INTEL_PT_PER_CPU_MMAPS
] = per_cpu_mmaps
;
349 auxtrace_info
->priv
[INTEL_PT_MTC_BIT
] = mtc_bit
;
350 auxtrace_info
->priv
[INTEL_PT_MTC_FREQ_BITS
] = mtc_freq_bits
;
351 auxtrace_info
->priv
[INTEL_PT_TSC_CTC_N
] = tsc_ctc_ratio_n
;
352 auxtrace_info
->priv
[INTEL_PT_TSC_CTC_D
] = tsc_ctc_ratio_d
;
353 auxtrace_info
->priv
[INTEL_PT_CYC_BIT
] = cyc_bit
;
358 static int intel_pt_track_switches(struct perf_evlist
*evlist
)
360 const char *sched_switch
= "sched:sched_switch";
361 struct perf_evsel
*evsel
;
364 if (!perf_evlist__can_select_event(evlist
, sched_switch
))
367 err
= parse_events(evlist
, sched_switch
, NULL
);
369 pr_debug2("%s: failed to parse %s, error %d\n",
370 __func__
, sched_switch
, err
);
374 evsel
= perf_evlist__last(evlist
);
376 perf_evsel__set_sample_bit(evsel
, CPU
);
377 perf_evsel__set_sample_bit(evsel
, TIME
);
379 evsel
->system_wide
= true;
380 evsel
->no_aux_samples
= true;
381 evsel
->immediate
= true;
386 static void intel_pt_valid_str(char *str
, size_t len
, u64 valid
)
388 unsigned int val
, last
= 0, state
= 1;
393 for (val
= 0; val
<= 64; val
++, valid
>>= 1) {
398 p
+= scnprintf(str
+ p
, len
- p
, ",");
401 p
+= scnprintf(str
+ p
, len
- p
, "%u", val
);
416 p
+= scnprintf(str
+ p
, len
- p
, ",%u", last
);
420 p
+= scnprintf(str
+ p
, len
- p
, "-%u", last
);
432 static int intel_pt_val_config_term(struct perf_pmu
*intel_pt_pmu
,
433 const char *caps
, const char *name
,
434 const char *supported
, u64 config
)
438 unsigned long long valid
;
442 if (perf_pmu__scan_file(intel_pt_pmu
, caps
, "%llx", &valid
) != 1)
446 perf_pmu__scan_file(intel_pt_pmu
, supported
, "%d", &ok
) == 1 && !ok
)
451 bits
= perf_pmu__format_bits(&intel_pt_pmu
->format
, name
);
455 for (shift
= 0; bits
&& !(bits
& 1); shift
++)
463 if (valid
& (1 << config
))
466 intel_pt_valid_str(valid_str
, sizeof(valid_str
), valid
);
467 pr_err("Invalid %s for %s. Valid values are: %s\n",
468 name
, INTEL_PT_PMU_NAME
, valid_str
);
472 static int intel_pt_validate_config(struct perf_pmu
*intel_pt_pmu
,
473 struct perf_evsel
*evsel
)
480 err
= intel_pt_val_config_term(intel_pt_pmu
, "caps/cycle_thresholds",
481 "cyc_thresh", "caps/psb_cyc",
486 err
= intel_pt_val_config_term(intel_pt_pmu
, "caps/mtc_periods",
487 "mtc_period", "caps/mtc",
492 return intel_pt_val_config_term(intel_pt_pmu
, "caps/psb_periods",
493 "psb_period", "caps/psb_cyc",
497 static int intel_pt_recording_options(struct auxtrace_record
*itr
,
498 struct perf_evlist
*evlist
,
499 struct record_opts
*opts
)
501 struct intel_pt_recording
*ptr
=
502 container_of(itr
, struct intel_pt_recording
, itr
);
503 struct perf_pmu
*intel_pt_pmu
= ptr
->intel_pt_pmu
;
504 bool have_timing_info
;
505 struct perf_evsel
*evsel
, *intel_pt_evsel
= NULL
;
506 const struct cpu_map
*cpus
= evlist
->cpus
;
507 bool privileged
= geteuid() == 0 || perf_event_paranoid() < 0;
511 ptr
->evlist
= evlist
;
512 ptr
->snapshot_mode
= opts
->auxtrace_snapshot_mode
;
514 evlist__for_each(evlist
, evsel
) {
515 if (evsel
->attr
.type
== intel_pt_pmu
->type
) {
516 if (intel_pt_evsel
) {
517 pr_err("There may be only one " INTEL_PT_PMU_NAME
" event\n");
520 evsel
->attr
.freq
= 0;
521 evsel
->attr
.sample_period
= 1;
522 intel_pt_evsel
= evsel
;
523 opts
->full_auxtrace
= true;
527 if (opts
->auxtrace_snapshot_mode
&& !opts
->full_auxtrace
) {
528 pr_err("Snapshot mode (-S option) requires " INTEL_PT_PMU_NAME
" PMU event (-e " INTEL_PT_PMU_NAME
")\n");
532 if (opts
->use_clockid
) {
533 pr_err("Cannot use clockid (-k option) with " INTEL_PT_PMU_NAME
"\n");
537 if (!opts
->full_auxtrace
)
540 err
= intel_pt_validate_config(intel_pt_pmu
, intel_pt_evsel
);
544 /* Set default sizes for snapshot mode */
545 if (opts
->auxtrace_snapshot_mode
) {
546 size_t psb_period
= intel_pt_psb_period(intel_pt_pmu
, evlist
);
548 if (!opts
->auxtrace_snapshot_size
&& !opts
->auxtrace_mmap_pages
) {
550 opts
->auxtrace_mmap_pages
= MiB(4) / page_size
;
552 opts
->auxtrace_mmap_pages
= KiB(128) / page_size
;
553 if (opts
->mmap_pages
== UINT_MAX
)
554 opts
->mmap_pages
= KiB(256) / page_size
;
556 } else if (!opts
->auxtrace_mmap_pages
&& !privileged
&&
557 opts
->mmap_pages
== UINT_MAX
) {
558 opts
->mmap_pages
= KiB(256) / page_size
;
560 if (!opts
->auxtrace_snapshot_size
)
561 opts
->auxtrace_snapshot_size
=
562 opts
->auxtrace_mmap_pages
* (size_t)page_size
;
563 if (!opts
->auxtrace_mmap_pages
) {
564 size_t sz
= opts
->auxtrace_snapshot_size
;
566 sz
= round_up(sz
, page_size
) / page_size
;
567 opts
->auxtrace_mmap_pages
= roundup_pow_of_two(sz
);
569 if (opts
->auxtrace_snapshot_size
>
570 opts
->auxtrace_mmap_pages
* (size_t)page_size
) {
571 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
572 opts
->auxtrace_snapshot_size
,
573 opts
->auxtrace_mmap_pages
* (size_t)page_size
);
576 if (!opts
->auxtrace_snapshot_size
|| !opts
->auxtrace_mmap_pages
) {
577 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
580 pr_debug2("Intel PT snapshot size: %zu\n",
581 opts
->auxtrace_snapshot_size
);
583 opts
->auxtrace_snapshot_size
<= psb_period
+
584 INTEL_PT_PSB_PERIOD_NEAR
)
585 ui__warning("Intel PT snapshot size (%zu) may be too small for PSB period (%zu)\n",
586 opts
->auxtrace_snapshot_size
, psb_period
);
589 /* Set default sizes for full trace mode */
590 if (opts
->full_auxtrace
&& !opts
->auxtrace_mmap_pages
) {
592 opts
->auxtrace_mmap_pages
= MiB(4) / page_size
;
594 opts
->auxtrace_mmap_pages
= KiB(128) / page_size
;
595 if (opts
->mmap_pages
== UINT_MAX
)
596 opts
->mmap_pages
= KiB(256) / page_size
;
600 /* Validate auxtrace_mmap_pages */
601 if (opts
->auxtrace_mmap_pages
) {
602 size_t sz
= opts
->auxtrace_mmap_pages
* (size_t)page_size
;
605 if (opts
->auxtrace_snapshot_mode
)
610 if (sz
< min_sz
|| !is_power_of_2(sz
)) {
611 pr_err("Invalid mmap size for Intel Processor Trace: must be at least %zuKiB and a power of 2\n",
617 intel_pt_parse_terms(&intel_pt_pmu
->format
, "tsc", &tsc_bit
);
619 if (opts
->full_auxtrace
&& (intel_pt_evsel
->attr
.config
& tsc_bit
))
620 have_timing_info
= true;
622 have_timing_info
= false;
625 * Per-cpu recording needs sched_switch events to distinguish different
628 if (have_timing_info
&& !cpu_map__empty(cpus
)) {
629 if (perf_can_record_switch_events()) {
630 bool cpu_wide
= !target__none(&opts
->target
) &&
631 !target__has_task(&opts
->target
);
633 if (!cpu_wide
&& perf_can_record_cpu_wide()) {
634 struct perf_evsel
*switch_evsel
;
636 err
= parse_events(evlist
, "dummy:u", NULL
);
640 switch_evsel
= perf_evlist__last(evlist
);
642 switch_evsel
->attr
.freq
= 0;
643 switch_evsel
->attr
.sample_period
= 1;
644 switch_evsel
->attr
.context_switch
= 1;
646 switch_evsel
->system_wide
= true;
647 switch_evsel
->no_aux_samples
= true;
648 switch_evsel
->immediate
= true;
650 perf_evsel__set_sample_bit(switch_evsel
, TID
);
651 perf_evsel__set_sample_bit(switch_evsel
, TIME
);
652 perf_evsel__set_sample_bit(switch_evsel
, CPU
);
654 opts
->record_switch_events
= false;
655 ptr
->have_sched_switch
= 3;
657 opts
->record_switch_events
= true;
659 ptr
->have_sched_switch
= 3;
661 ptr
->have_sched_switch
= 2;
664 err
= intel_pt_track_switches(evlist
);
666 pr_debug2("Unable to select sched:sched_switch\n");
670 ptr
->have_sched_switch
= 1;
674 if (intel_pt_evsel
) {
676 * To obtain the auxtrace buffer file descriptor, the auxtrace
677 * event must come first.
679 perf_evlist__to_front(evlist
, intel_pt_evsel
);
681 * In the case of per-cpu mmaps, we need the CPU on the
684 if (!cpu_map__empty(cpus
))
685 perf_evsel__set_sample_bit(intel_pt_evsel
, CPU
);
688 /* Add dummy event to keep tracking */
689 if (opts
->full_auxtrace
) {
690 struct perf_evsel
*tracking_evsel
;
692 err
= parse_events(evlist
, "dummy:u", NULL
);
696 tracking_evsel
= perf_evlist__last(evlist
);
698 perf_evlist__set_tracking_event(evlist
, tracking_evsel
);
700 tracking_evsel
->attr
.freq
= 0;
701 tracking_evsel
->attr
.sample_period
= 1;
703 /* In per-cpu case, always need the time of mmap events etc */
704 if (!cpu_map__empty(cpus
)) {
705 perf_evsel__set_sample_bit(tracking_evsel
, TIME
);
706 /* And the CPU for switch events */
707 perf_evsel__set_sample_bit(tracking_evsel
, CPU
);
712 * Warn the user when we do not have enough information to decode i.e.
713 * per-cpu with no sched_switch (except workload-only).
715 if (!ptr
->have_sched_switch
&& !cpu_map__empty(cpus
) &&
716 !target__none(&opts
->target
))
717 ui__warning("Intel Processor Trace decoding will not be possible except for kernel tracing!\n");
722 static int intel_pt_snapshot_start(struct auxtrace_record
*itr
)
724 struct intel_pt_recording
*ptr
=
725 container_of(itr
, struct intel_pt_recording
, itr
);
726 struct perf_evsel
*evsel
;
728 evlist__for_each(ptr
->evlist
, evsel
) {
729 if (evsel
->attr
.type
== ptr
->intel_pt_pmu
->type
)
730 return perf_evsel__disable(evsel
);
735 static int intel_pt_snapshot_finish(struct auxtrace_record
*itr
)
737 struct intel_pt_recording
*ptr
=
738 container_of(itr
, struct intel_pt_recording
, itr
);
739 struct perf_evsel
*evsel
;
741 evlist__for_each(ptr
->evlist
, evsel
) {
742 if (evsel
->attr
.type
== ptr
->intel_pt_pmu
->type
)
743 return perf_evsel__enable(evsel
);
748 static int intel_pt_alloc_snapshot_refs(struct intel_pt_recording
*ptr
, int idx
)
750 const size_t sz
= sizeof(struct intel_pt_snapshot_ref
);
751 int cnt
= ptr
->snapshot_ref_cnt
, new_cnt
= cnt
* 2;
752 struct intel_pt_snapshot_ref
*refs
;
757 while (new_cnt
<= idx
)
760 refs
= calloc(new_cnt
, sz
);
764 memcpy(refs
, ptr
->snapshot_refs
, cnt
* sz
);
766 ptr
->snapshot_refs
= refs
;
767 ptr
->snapshot_ref_cnt
= new_cnt
;
772 static void intel_pt_free_snapshot_refs(struct intel_pt_recording
*ptr
)
776 for (i
= 0; i
< ptr
->snapshot_ref_cnt
; i
++)
777 zfree(&ptr
->snapshot_refs
[i
].ref_buf
);
778 zfree(&ptr
->snapshot_refs
);
781 static void intel_pt_recording_free(struct auxtrace_record
*itr
)
783 struct intel_pt_recording
*ptr
=
784 container_of(itr
, struct intel_pt_recording
, itr
);
786 intel_pt_free_snapshot_refs(ptr
);
790 static int intel_pt_alloc_snapshot_ref(struct intel_pt_recording
*ptr
, int idx
,
791 size_t snapshot_buf_size
)
793 size_t ref_buf_size
= ptr
->snapshot_ref_buf_size
;
796 ref_buf
= zalloc(ref_buf_size
);
800 ptr
->snapshot_refs
[idx
].ref_buf
= ref_buf
;
801 ptr
->snapshot_refs
[idx
].ref_offset
= snapshot_buf_size
- ref_buf_size
;
806 static size_t intel_pt_snapshot_ref_buf_size(struct intel_pt_recording
*ptr
,
807 size_t snapshot_buf_size
)
809 const size_t max_size
= 256 * 1024;
810 size_t buf_size
= 0, psb_period
;
812 if (ptr
->snapshot_size
<= 64 * 1024)
815 psb_period
= intel_pt_psb_period(ptr
->intel_pt_pmu
, ptr
->evlist
);
817 buf_size
= psb_period
* 2;
819 if (!buf_size
|| buf_size
> max_size
)
822 if (buf_size
>= snapshot_buf_size
)
825 if (buf_size
>= ptr
->snapshot_size
/ 2)
831 static int intel_pt_snapshot_init(struct intel_pt_recording
*ptr
,
832 size_t snapshot_buf_size
)
834 if (ptr
->snapshot_init_done
)
837 ptr
->snapshot_init_done
= true;
839 ptr
->snapshot_ref_buf_size
= intel_pt_snapshot_ref_buf_size(ptr
,
846 * intel_pt_compare_buffers - compare bytes in a buffer to a circular buffer.
847 * @buf1: first buffer
848 * @compare_size: number of bytes to compare
849 * @buf2: second buffer (a circular buffer)
850 * @offs2: offset in second buffer
851 * @buf2_size: size of second buffer
853 * The comparison allows for the possibility that the bytes to compare in the
854 * circular buffer are not contiguous. It is assumed that @compare_size <=
855 * @buf2_size. This function returns %false if the bytes are identical, %true
858 static bool intel_pt_compare_buffers(void *buf1
, size_t compare_size
,
859 void *buf2
, size_t offs2
, size_t buf2_size
)
861 size_t end2
= offs2
+ compare_size
, part_size
;
863 if (end2
<= buf2_size
)
864 return memcmp(buf1
, buf2
+ offs2
, compare_size
);
866 part_size
= end2
- buf2_size
;
867 if (memcmp(buf1
, buf2
+ offs2
, part_size
))
870 compare_size
-= part_size
;
872 return memcmp(buf1
+ part_size
, buf2
, compare_size
);
875 static bool intel_pt_compare_ref(void *ref_buf
, size_t ref_offset
,
876 size_t ref_size
, size_t buf_size
,
877 void *data
, size_t head
)
879 size_t ref_end
= ref_offset
+ ref_size
;
881 if (ref_end
> buf_size
) {
882 if (head
> ref_offset
|| head
< ref_end
- buf_size
)
884 } else if (head
> ref_offset
&& head
< ref_end
) {
888 return intel_pt_compare_buffers(ref_buf
, ref_size
, data
, ref_offset
,
892 static void intel_pt_copy_ref(void *ref_buf
, size_t ref_size
, size_t buf_size
,
893 void *data
, size_t head
)
895 if (head
>= ref_size
) {
896 memcpy(ref_buf
, data
+ head
- ref_size
, ref_size
);
898 memcpy(ref_buf
, data
, head
);
900 memcpy(ref_buf
+ head
, data
+ buf_size
- ref_size
, ref_size
);
904 static bool intel_pt_wrapped(struct intel_pt_recording
*ptr
, int idx
,
905 struct auxtrace_mmap
*mm
, unsigned char *data
,
908 struct intel_pt_snapshot_ref
*ref
= &ptr
->snapshot_refs
[idx
];
911 wrapped
= intel_pt_compare_ref(ref
->ref_buf
, ref
->ref_offset
,
912 ptr
->snapshot_ref_buf_size
, mm
->len
,
915 intel_pt_copy_ref(ref
->ref_buf
, ptr
->snapshot_ref_buf_size
, mm
->len
,
921 static bool intel_pt_first_wrap(u64
*data
, size_t buf_size
)
930 for (i
= a
; i
< b
; i
++) {
938 static int intel_pt_find_snapshot(struct auxtrace_record
*itr
, int idx
,
939 struct auxtrace_mmap
*mm
, unsigned char *data
,
942 struct intel_pt_recording
*ptr
=
943 container_of(itr
, struct intel_pt_recording
, itr
);
947 pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
948 __func__
, idx
, (size_t)*old
, (size_t)*head
);
950 err
= intel_pt_snapshot_init(ptr
, mm
->len
);
954 if (idx
>= ptr
->snapshot_ref_cnt
) {
955 err
= intel_pt_alloc_snapshot_refs(ptr
, idx
);
960 if (ptr
->snapshot_ref_buf_size
) {
961 if (!ptr
->snapshot_refs
[idx
].ref_buf
) {
962 err
= intel_pt_alloc_snapshot_ref(ptr
, idx
, mm
->len
);
966 wrapped
= intel_pt_wrapped(ptr
, idx
, mm
, data
, *head
);
968 wrapped
= ptr
->snapshot_refs
[idx
].wrapped
;
969 if (!wrapped
&& intel_pt_first_wrap((u64
*)data
, mm
->len
)) {
970 ptr
->snapshot_refs
[idx
].wrapped
= true;
976 * In full trace mode 'head' continually increases. However in snapshot
977 * mode 'head' is an offset within the buffer. Here 'old' and 'head'
978 * are adjusted to match the full trace case which expects that 'old' is
979 * always less than 'head'.
993 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
994 __func__
, wrapped
? "" : "not ", (size_t)*old
, (size_t)*head
);
999 pr_err("%s: failed, error %d\n", __func__
, err
);
1003 static u64
intel_pt_reference(struct auxtrace_record
*itr __maybe_unused
)
1008 static int intel_pt_read_finish(struct auxtrace_record
*itr
, int idx
)
1010 struct intel_pt_recording
*ptr
=
1011 container_of(itr
, struct intel_pt_recording
, itr
);
1012 struct perf_evsel
*evsel
;
1014 evlist__for_each(ptr
->evlist
, evsel
) {
1015 if (evsel
->attr
.type
== ptr
->intel_pt_pmu
->type
)
1016 return perf_evlist__enable_event_idx(ptr
->evlist
, evsel
,
1022 struct auxtrace_record
*intel_pt_recording_init(int *err
)
1024 struct perf_pmu
*intel_pt_pmu
= perf_pmu__find(INTEL_PT_PMU_NAME
);
1025 struct intel_pt_recording
*ptr
;
1030 ptr
= zalloc(sizeof(struct intel_pt_recording
));
1036 ptr
->intel_pt_pmu
= intel_pt_pmu
;
1037 ptr
->itr
.recording_options
= intel_pt_recording_options
;
1038 ptr
->itr
.info_priv_size
= intel_pt_info_priv_size
;
1039 ptr
->itr
.info_fill
= intel_pt_info_fill
;
1040 ptr
->itr
.free
= intel_pt_recording_free
;
1041 ptr
->itr
.snapshot_start
= intel_pt_snapshot_start
;
1042 ptr
->itr
.snapshot_finish
= intel_pt_snapshot_finish
;
1043 ptr
->itr
.find_snapshot
= intel_pt_find_snapshot
;
1044 ptr
->itr
.parse_snapshot_options
= intel_pt_parse_snapshot_options
;
1045 ptr
->itr
.reference
= intel_pt_reference
;
1046 ptr
->itr
.read_finish
= intel_pt_read_finish
;