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
20 #include <linux/kernel.h>
21 #include <linux/types.h>
36 #include "thread-stack.h"
38 #include "callchain.h"
46 #include "intel-pt-decoder/intel-pt-log.h"
47 #include "intel-pt-decoder/intel-pt-decoder.h"
48 #include "intel-pt-decoder/intel-pt-insn-decoder.h"
49 #include "intel-pt-decoder/intel-pt-pkt-decoder.h"
51 #define MAX_TIMESTAMP (~0ULL)
54 struct auxtrace auxtrace;
55 struct auxtrace_queues queues;
56 struct auxtrace_heap heap;
58 struct perf_session *session;
59 struct machine *machine;
60 struct perf_evsel *switch_evsel;
61 struct thread *unknown_thread;
62 bool timeless_decoding;
71 int have_sched_switch;
77 struct perf_tsc_conversion tc;
78 bool cap_user_time_zero;
80 struct itrace_synth_opts synth_opts;
82 bool sample_instructions;
83 u64 instructions_sample_type;
88 u64 branches_sample_type;
91 bool sample_transactions;
92 u64 transactions_sample_type;
96 u64 ptwrites_sample_type;
99 bool sample_pwr_events;
100 u64 pwr_events_sample_type;
114 unsigned max_non_turbo_ratio;
117 unsigned long num_events;
120 struct addr_filters filts;
124 INTEL_PT_SS_NOT_TRACING,
127 INTEL_PT_SS_EXPECTING_SWITCH_EVENT,
128 INTEL_PT_SS_EXPECTING_SWITCH_IP,
131 struct intel_pt_queue {
133 unsigned int queue_nr;
134 struct auxtrace_buffer *buffer;
135 struct auxtrace_buffer *old_buffer;
137 const struct intel_pt_state *state;
138 struct ip_callchain *chain;
139 struct branch_stack *last_branch;
140 struct branch_stack *last_branch_rb;
141 size_t last_branch_pos;
142 union perf_event *event_buf;
145 bool step_through_buffers;
146 bool use_buffer_pid_tid;
152 struct thread *thread;
160 char insn[INTEL_PT_INSN_BUF_SZ];
163 static void intel_pt_dump(struct intel_pt *pt __maybe_unused,
164 unsigned char *buf, size_t len)
166 struct intel_pt_pkt packet;
169 char desc[INTEL_PT_PKT_DESC_MAX];
170 const char *color = PERF_COLOR_BLUE;
172 color_fprintf(stdout, color,
173 ". ... Intel Processor Trace data: size %zu bytes\n",
177 ret = intel_pt_get_packet(buf, len, &packet);
183 color_fprintf(stdout, color, " %08x: ", pos);
184 for (i = 0; i < pkt_len; i++)
185 color_fprintf(stdout, color, " %02x", buf[i]);
187 color_fprintf(stdout, color, " ");
189 ret = intel_pt_pkt_desc(&packet, desc,
190 INTEL_PT_PKT_DESC_MAX);
192 color_fprintf(stdout, color, " %s\n", desc);
194 color_fprintf(stdout, color, " Bad packet!\n");
202 static void intel_pt_dump_event(struct intel_pt *pt, unsigned char *buf,
206 intel_pt_dump(pt, buf, len);
209 static int intel_pt_do_fix_overlap(struct intel_pt *pt, struct auxtrace_buffer *a,
210 struct auxtrace_buffer *b)
212 bool consecutive = false;
215 start = intel_pt_find_overlap(a->data, a->size, b->data, b->size,
216 pt->have_tsc, &consecutive);
219 b->use_size = b->data + b->size - start;
221 if (b->use_size && consecutive)
222 b->consecutive = true;
226 /* This function assumes data is processed sequentially only */
227 static int intel_pt_get_trace(struct intel_pt_buffer *b, void *data)
229 struct intel_pt_queue *ptq = data;
230 struct auxtrace_buffer *buffer = ptq->buffer;
231 struct auxtrace_buffer *old_buffer = ptq->old_buffer;
232 struct auxtrace_queue *queue;
240 queue = &ptq->pt->queues.queue_array[ptq->queue_nr];
242 buffer = auxtrace_buffer__next(queue, buffer);
245 auxtrace_buffer__drop_data(old_buffer);
250 ptq->buffer = buffer;
253 int fd = perf_data__fd(ptq->pt->session->data);
255 buffer->data = auxtrace_buffer__get_data(buffer, fd);
260 might_overlap = ptq->pt->snapshot_mode || ptq->pt->sampling_mode;
261 if (might_overlap && !buffer->consecutive && old_buffer &&
262 intel_pt_do_fix_overlap(ptq->pt, old_buffer, buffer))
265 if (buffer->use_data) {
266 b->len = buffer->use_size;
267 b->buf = buffer->use_data;
269 b->len = buffer->size;
270 b->buf = buffer->data;
272 b->ref_timestamp = buffer->reference;
274 if (!old_buffer || (might_overlap && !buffer->consecutive)) {
275 b->consecutive = false;
276 b->trace_nr = buffer->buffer_nr + 1;
278 b->consecutive = true;
281 if (ptq->step_through_buffers)
286 auxtrace_buffer__drop_data(old_buffer);
287 ptq->old_buffer = buffer;
289 auxtrace_buffer__drop_data(buffer);
290 return intel_pt_get_trace(b, data);
296 struct intel_pt_cache_entry {
297 struct auxtrace_cache_entry entry;
300 enum intel_pt_insn_op op;
301 enum intel_pt_insn_branch branch;
304 char insn[INTEL_PT_INSN_BUF_SZ];
307 static int intel_pt_config_div(const char *var, const char *value, void *data)
312 if (!strcmp(var, "intel-pt.cache-divisor")) {
313 val = strtol(value, NULL, 0);
314 if (val > 0 && val <= INT_MAX)
321 static int intel_pt_cache_divisor(void)
328 perf_config(intel_pt_config_div, &d);
336 static unsigned int intel_pt_cache_size(struct dso *dso,
337 struct machine *machine)
341 size = dso__data_size(dso, machine);
342 size /= intel_pt_cache_divisor();
345 if (size > (1 << 21))
347 return 32 - __builtin_clz(size);
350 static struct auxtrace_cache *intel_pt_cache(struct dso *dso,
351 struct machine *machine)
353 struct auxtrace_cache *c;
356 if (dso->auxtrace_cache)
357 return dso->auxtrace_cache;
359 bits = intel_pt_cache_size(dso, machine);
361 /* Ignoring cache creation failure */
362 c = auxtrace_cache__new(bits, sizeof(struct intel_pt_cache_entry), 200);
364 dso->auxtrace_cache = c;
369 static int intel_pt_cache_add(struct dso *dso, struct machine *machine,
370 u64 offset, u64 insn_cnt, u64 byte_cnt,
371 struct intel_pt_insn *intel_pt_insn)
373 struct auxtrace_cache *c = intel_pt_cache(dso, machine);
374 struct intel_pt_cache_entry *e;
380 e = auxtrace_cache__alloc_entry(c);
384 e->insn_cnt = insn_cnt;
385 e->byte_cnt = byte_cnt;
386 e->op = intel_pt_insn->op;
387 e->branch = intel_pt_insn->branch;
388 e->length = intel_pt_insn->length;
389 e->rel = intel_pt_insn->rel;
390 memcpy(e->insn, intel_pt_insn->buf, INTEL_PT_INSN_BUF_SZ);
392 err = auxtrace_cache__add(c, offset, &e->entry);
394 auxtrace_cache__free_entry(c, e);
399 static struct intel_pt_cache_entry *
400 intel_pt_cache_lookup(struct dso *dso, struct machine *machine, u64 offset)
402 struct auxtrace_cache *c = intel_pt_cache(dso, machine);
407 return auxtrace_cache__lookup(dso->auxtrace_cache, offset);
410 static inline u8 intel_pt_cpumode(struct intel_pt *pt, uint64_t ip)
412 return ip >= pt->kernel_start ?
413 PERF_RECORD_MISC_KERNEL :
414 PERF_RECORD_MISC_USER;
417 static int intel_pt_walk_next_insn(struct intel_pt_insn *intel_pt_insn,
418 uint64_t *insn_cnt_ptr, uint64_t *ip,
419 uint64_t to_ip, uint64_t max_insn_cnt,
422 struct intel_pt_queue *ptq = data;
423 struct machine *machine = ptq->pt->machine;
424 struct thread *thread;
425 struct addr_location al;
426 unsigned char buf[INTEL_PT_INSN_BUF_SZ];
430 u64 offset, start_offset, start_ip;
434 intel_pt_insn->length = 0;
436 if (to_ip && *ip == to_ip)
439 cpumode = intel_pt_cpumode(ptq->pt, *ip);
441 thread = ptq->thread;
443 if (cpumode != PERF_RECORD_MISC_KERNEL)
445 thread = ptq->pt->unknown_thread;
449 if (!thread__find_map(thread, cpumode, *ip, &al) || !al.map->dso)
452 if (al.map->dso->data.status == DSO_DATA_STATUS_ERROR &&
453 dso__data_status_seen(al.map->dso,
454 DSO_DATA_STATUS_SEEN_ITRACE))
457 offset = al.map->map_ip(al.map, *ip);
459 if (!to_ip && one_map) {
460 struct intel_pt_cache_entry *e;
462 e = intel_pt_cache_lookup(al.map->dso, machine, offset);
464 (!max_insn_cnt || e->insn_cnt <= max_insn_cnt)) {
465 *insn_cnt_ptr = e->insn_cnt;
467 intel_pt_insn->op = e->op;
468 intel_pt_insn->branch = e->branch;
469 intel_pt_insn->length = e->length;
470 intel_pt_insn->rel = e->rel;
471 memcpy(intel_pt_insn->buf, e->insn,
472 INTEL_PT_INSN_BUF_SZ);
473 intel_pt_log_insn_no_data(intel_pt_insn, *ip);
478 start_offset = offset;
481 /* Load maps to ensure dso->is_64_bit has been updated */
484 x86_64 = al.map->dso->is_64_bit;
487 len = dso__data_read_offset(al.map->dso, machine,
489 INTEL_PT_INSN_BUF_SZ);
493 if (intel_pt_get_insn(buf, len, x86_64, intel_pt_insn))
496 intel_pt_log_insn(intel_pt_insn, *ip);
500 if (intel_pt_insn->branch != INTEL_PT_BR_NO_BRANCH)
503 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
506 *ip += intel_pt_insn->length;
508 if (to_ip && *ip == to_ip) {
509 intel_pt_insn->length = 0;
513 if (*ip >= al.map->end)
516 offset += intel_pt_insn->length;
521 *insn_cnt_ptr = insn_cnt;
527 * Didn't lookup in the 'to_ip' case, so do it now to prevent duplicate
531 struct intel_pt_cache_entry *e;
533 e = intel_pt_cache_lookup(al.map->dso, machine, start_offset);
538 /* Ignore cache errors */
539 intel_pt_cache_add(al.map->dso, machine, start_offset, insn_cnt,
540 *ip - start_ip, intel_pt_insn);
545 *insn_cnt_ptr = insn_cnt;
549 static bool intel_pt_match_pgd_ip(struct intel_pt *pt, uint64_t ip,
550 uint64_t offset, const char *filename)
552 struct addr_filter *filt;
553 bool have_filter = false;
554 bool hit_tracestop = false;
555 bool hit_filter = false;
557 list_for_each_entry(filt, &pt->filts.head, list) {
561 if ((filename && !filt->filename) ||
562 (!filename && filt->filename) ||
563 (filename && strcmp(filename, filt->filename)))
566 if (!(offset >= filt->addr && offset < filt->addr + filt->size))
569 intel_pt_log("TIP.PGD ip %#"PRIx64" offset %#"PRIx64" in %s hit filter: %s offset %#"PRIx64" size %#"PRIx64"\n",
570 ip, offset, filename ? filename : "[kernel]",
571 filt->start ? "filter" : "stop",
572 filt->addr, filt->size);
577 hit_tracestop = true;
580 if (!hit_tracestop && !hit_filter)
581 intel_pt_log("TIP.PGD ip %#"PRIx64" offset %#"PRIx64" in %s is not in a filter region\n",
582 ip, offset, filename ? filename : "[kernel]");
584 return hit_tracestop || (have_filter && !hit_filter);
587 static int __intel_pt_pgd_ip(uint64_t ip, void *data)
589 struct intel_pt_queue *ptq = data;
590 struct thread *thread;
591 struct addr_location al;
595 if (ip >= ptq->pt->kernel_start)
596 return intel_pt_match_pgd_ip(ptq->pt, ip, ip, NULL);
598 cpumode = PERF_RECORD_MISC_USER;
600 thread = ptq->thread;
604 if (!thread__find_map(thread, cpumode, ip, &al) || !al.map->dso)
607 offset = al.map->map_ip(al.map, ip);
609 return intel_pt_match_pgd_ip(ptq->pt, ip, offset,
610 al.map->dso->long_name);
613 static bool intel_pt_pgd_ip(uint64_t ip, void *data)
615 return __intel_pt_pgd_ip(ip, data) > 0;
618 static bool intel_pt_get_config(struct intel_pt *pt,
619 struct perf_event_attr *attr, u64 *config)
621 if (attr->type == pt->pmu_type) {
623 *config = attr->config;
630 static bool intel_pt_exclude_kernel(struct intel_pt *pt)
632 struct perf_evsel *evsel;
634 evlist__for_each_entry(pt->session->evlist, evsel) {
635 if (intel_pt_get_config(pt, &evsel->attr, NULL) &&
636 !evsel->attr.exclude_kernel)
642 static bool intel_pt_return_compression(struct intel_pt *pt)
644 struct perf_evsel *evsel;
647 if (!pt->noretcomp_bit)
650 evlist__for_each_entry(pt->session->evlist, evsel) {
651 if (intel_pt_get_config(pt, &evsel->attr, &config) &&
652 (config & pt->noretcomp_bit))
658 static bool intel_pt_branch_enable(struct intel_pt *pt)
660 struct perf_evsel *evsel;
663 evlist__for_each_entry(pt->session->evlist, evsel) {
664 if (intel_pt_get_config(pt, &evsel->attr, &config) &&
665 (config & 1) && !(config & 0x2000))
671 static unsigned int intel_pt_mtc_period(struct intel_pt *pt)
673 struct perf_evsel *evsel;
677 if (!pt->mtc_freq_bits)
680 for (shift = 0, config = pt->mtc_freq_bits; !(config & 1); shift++)
683 evlist__for_each_entry(pt->session->evlist, evsel) {
684 if (intel_pt_get_config(pt, &evsel->attr, &config))
685 return (config & pt->mtc_freq_bits) >> shift;
690 static bool intel_pt_timeless_decoding(struct intel_pt *pt)
692 struct perf_evsel *evsel;
693 bool timeless_decoding = true;
696 if (!pt->tsc_bit || !pt->cap_user_time_zero)
699 evlist__for_each_entry(pt->session->evlist, evsel) {
700 if (!(evsel->attr.sample_type & PERF_SAMPLE_TIME))
702 if (intel_pt_get_config(pt, &evsel->attr, &config)) {
703 if (config & pt->tsc_bit)
704 timeless_decoding = false;
709 return timeless_decoding;
712 static bool intel_pt_tracing_kernel(struct intel_pt *pt)
714 struct perf_evsel *evsel;
716 evlist__for_each_entry(pt->session->evlist, evsel) {
717 if (intel_pt_get_config(pt, &evsel->attr, NULL) &&
718 !evsel->attr.exclude_kernel)
724 static bool intel_pt_have_tsc(struct intel_pt *pt)
726 struct perf_evsel *evsel;
727 bool have_tsc = false;
733 evlist__for_each_entry(pt->session->evlist, evsel) {
734 if (intel_pt_get_config(pt, &evsel->attr, &config)) {
735 if (config & pt->tsc_bit)
744 static u64 intel_pt_ns_to_ticks(const struct intel_pt *pt, u64 ns)
748 quot = ns / pt->tc.time_mult;
749 rem = ns % pt->tc.time_mult;
750 return (quot << pt->tc.time_shift) + (rem << pt->tc.time_shift) /
754 static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt,
755 unsigned int queue_nr)
757 struct intel_pt_params params = { .get_trace = 0, };
758 struct perf_env *env = pt->machine->env;
759 struct intel_pt_queue *ptq;
761 ptq = zalloc(sizeof(struct intel_pt_queue));
765 if (pt->synth_opts.callchain) {
766 size_t sz = sizeof(struct ip_callchain);
768 /* Add 1 to callchain_sz for callchain context */
769 sz += (pt->synth_opts.callchain_sz + 1) * sizeof(u64);
770 ptq->chain = zalloc(sz);
775 if (pt->synth_opts.last_branch) {
776 size_t sz = sizeof(struct branch_stack);
778 sz += pt->synth_opts.last_branch_sz *
779 sizeof(struct branch_entry);
780 ptq->last_branch = zalloc(sz);
781 if (!ptq->last_branch)
783 ptq->last_branch_rb = zalloc(sz);
784 if (!ptq->last_branch_rb)
788 ptq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE);
793 ptq->queue_nr = queue_nr;
794 ptq->exclude_kernel = intel_pt_exclude_kernel(pt);
800 params.get_trace = intel_pt_get_trace;
801 params.walk_insn = intel_pt_walk_next_insn;
803 params.return_compression = intel_pt_return_compression(pt);
804 params.branch_enable = intel_pt_branch_enable(pt);
805 params.max_non_turbo_ratio = pt->max_non_turbo_ratio;
806 params.mtc_period = intel_pt_mtc_period(pt);
807 params.tsc_ctc_ratio_n = pt->tsc_ctc_ratio_n;
808 params.tsc_ctc_ratio_d = pt->tsc_ctc_ratio_d;
810 if (pt->filts.cnt > 0)
811 params.pgd_ip = intel_pt_pgd_ip;
813 if (pt->synth_opts.instructions) {
814 if (pt->synth_opts.period) {
815 switch (pt->synth_opts.period_type) {
816 case PERF_ITRACE_PERIOD_INSTRUCTIONS:
818 INTEL_PT_PERIOD_INSTRUCTIONS;
819 params.period = pt->synth_opts.period;
821 case PERF_ITRACE_PERIOD_TICKS:
822 params.period_type = INTEL_PT_PERIOD_TICKS;
823 params.period = pt->synth_opts.period;
825 case PERF_ITRACE_PERIOD_NANOSECS:
826 params.period_type = INTEL_PT_PERIOD_TICKS;
827 params.period = intel_pt_ns_to_ticks(pt,
828 pt->synth_opts.period);
835 if (!params.period) {
836 params.period_type = INTEL_PT_PERIOD_INSTRUCTIONS;
841 if (env->cpuid && !strncmp(env->cpuid, "GenuineIntel,6,92,", 18))
842 params.flags |= INTEL_PT_FUP_WITH_NLIP;
844 ptq->decoder = intel_pt_decoder_new(¶ms);
851 zfree(&ptq->event_buf);
852 zfree(&ptq->last_branch);
853 zfree(&ptq->last_branch_rb);
859 static void intel_pt_free_queue(void *priv)
861 struct intel_pt_queue *ptq = priv;
865 thread__zput(ptq->thread);
866 intel_pt_decoder_free(ptq->decoder);
867 zfree(&ptq->event_buf);
868 zfree(&ptq->last_branch);
869 zfree(&ptq->last_branch_rb);
874 static void intel_pt_set_pid_tid_cpu(struct intel_pt *pt,
875 struct auxtrace_queue *queue)
877 struct intel_pt_queue *ptq = queue->priv;
879 if (queue->tid == -1 || pt->have_sched_switch) {
880 ptq->tid = machine__get_current_tid(pt->machine, ptq->cpu);
883 thread__zput(ptq->thread);
886 if (!ptq->thread && ptq->tid != -1)
887 ptq->thread = machine__find_thread(pt->machine, -1, ptq->tid);
890 ptq->pid = ptq->thread->pid_;
891 if (queue->cpu == -1)
892 ptq->cpu = ptq->thread->cpu;
896 static void intel_pt_sample_flags(struct intel_pt_queue *ptq)
899 if (ptq->state->flags & INTEL_PT_ABORT_TX) {
900 ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_TX_ABORT;
901 } else if (ptq->state->flags & INTEL_PT_ASYNC) {
902 if (ptq->state->to_ip)
903 ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_CALL |
905 PERF_IP_FLAG_INTERRUPT;
907 ptq->flags = PERF_IP_FLAG_BRANCH |
908 PERF_IP_FLAG_TRACE_END;
911 if (ptq->state->from_ip)
912 ptq->flags = intel_pt_insn_type(ptq->state->insn_op);
914 ptq->flags = PERF_IP_FLAG_BRANCH |
915 PERF_IP_FLAG_TRACE_BEGIN;
916 if (ptq->state->flags & INTEL_PT_IN_TX)
917 ptq->flags |= PERF_IP_FLAG_IN_TX;
918 ptq->insn_len = ptq->state->insn_len;
919 memcpy(ptq->insn, ptq->state->insn, INTEL_PT_INSN_BUF_SZ);
923 static int intel_pt_setup_queue(struct intel_pt *pt,
924 struct auxtrace_queue *queue,
925 unsigned int queue_nr)
927 struct intel_pt_queue *ptq = queue->priv;
929 if (list_empty(&queue->head))
933 ptq = intel_pt_alloc_queue(pt, queue_nr);
938 if (queue->cpu != -1)
939 ptq->cpu = queue->cpu;
940 ptq->tid = queue->tid;
942 if (pt->sampling_mode && !pt->snapshot_mode &&
943 pt->timeless_decoding)
944 ptq->step_through_buffers = true;
946 ptq->sync_switch = pt->sync_switch;
950 (!ptq->sync_switch ||
951 ptq->switch_state != INTEL_PT_SS_EXPECTING_SWITCH_EVENT)) {
952 const struct intel_pt_state *state;
955 if (pt->timeless_decoding)
958 intel_pt_log("queue %u getting timestamp\n", queue_nr);
959 intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n",
960 queue_nr, ptq->cpu, ptq->pid, ptq->tid);
962 state = intel_pt_decode(ptq->decoder);
964 if (state->err == INTEL_PT_ERR_NODATA) {
965 intel_pt_log("queue %u has no timestamp\n",
971 if (state->timestamp)
975 ptq->timestamp = state->timestamp;
976 intel_pt_log("queue %u timestamp 0x%" PRIx64 "\n",
977 queue_nr, ptq->timestamp);
979 ptq->have_sample = true;
980 intel_pt_sample_flags(ptq);
981 ret = auxtrace_heap__add(&pt->heap, queue_nr, ptq->timestamp);
990 static int intel_pt_setup_queues(struct intel_pt *pt)
995 for (i = 0; i < pt->queues.nr_queues; i++) {
996 ret = intel_pt_setup_queue(pt, &pt->queues.queue_array[i], i);
1003 static inline void intel_pt_copy_last_branch_rb(struct intel_pt_queue *ptq)
1005 struct branch_stack *bs_src = ptq->last_branch_rb;
1006 struct branch_stack *bs_dst = ptq->last_branch;
1009 bs_dst->nr = bs_src->nr;
1014 nr = ptq->pt->synth_opts.last_branch_sz - ptq->last_branch_pos;
1015 memcpy(&bs_dst->entries[0],
1016 &bs_src->entries[ptq->last_branch_pos],
1017 sizeof(struct branch_entry) * nr);
1019 if (bs_src->nr >= ptq->pt->synth_opts.last_branch_sz) {
1020 memcpy(&bs_dst->entries[nr],
1021 &bs_src->entries[0],
1022 sizeof(struct branch_entry) * ptq->last_branch_pos);
1026 static inline void intel_pt_reset_last_branch_rb(struct intel_pt_queue *ptq)
1028 ptq->last_branch_pos = 0;
1029 ptq->last_branch_rb->nr = 0;
1032 static void intel_pt_update_last_branch_rb(struct intel_pt_queue *ptq)
1034 const struct intel_pt_state *state = ptq->state;
1035 struct branch_stack *bs = ptq->last_branch_rb;
1036 struct branch_entry *be;
1038 if (!ptq->last_branch_pos)
1039 ptq->last_branch_pos = ptq->pt->synth_opts.last_branch_sz;
1041 ptq->last_branch_pos -= 1;
1043 be = &bs->entries[ptq->last_branch_pos];
1044 be->from = state->from_ip;
1045 be->to = state->to_ip;
1046 be->flags.abort = !!(state->flags & INTEL_PT_ABORT_TX);
1047 be->flags.in_tx = !!(state->flags & INTEL_PT_IN_TX);
1048 /* No support for mispredict */
1049 be->flags.mispred = ptq->pt->mispred_all;
1051 if (bs->nr < ptq->pt->synth_opts.last_branch_sz)
1055 static inline bool intel_pt_skip_event(struct intel_pt *pt)
1057 return pt->synth_opts.initial_skip &&
1058 pt->num_events++ < pt->synth_opts.initial_skip;
1061 static void intel_pt_prep_b_sample(struct intel_pt *pt,
1062 struct intel_pt_queue *ptq,
1063 union perf_event *event,
1064 struct perf_sample *sample)
1066 if (!pt->timeless_decoding)
1067 sample->time = tsc_to_perf_time(ptq->timestamp, &pt->tc);
1069 sample->ip = ptq->state->from_ip;
1070 sample->cpumode = intel_pt_cpumode(pt, sample->ip);
1071 sample->pid = ptq->pid;
1072 sample->tid = ptq->tid;
1073 sample->addr = ptq->state->to_ip;
1075 sample->cpu = ptq->cpu;
1076 sample->flags = ptq->flags;
1077 sample->insn_len = ptq->insn_len;
1078 memcpy(sample->insn, ptq->insn, INTEL_PT_INSN_BUF_SZ);
1080 event->sample.header.type = PERF_RECORD_SAMPLE;
1081 event->sample.header.misc = sample->cpumode;
1082 event->sample.header.size = sizeof(struct perf_event_header);
1085 static int intel_pt_inject_event(union perf_event *event,
1086 struct perf_sample *sample, u64 type)
1088 event->header.size = perf_event__sample_event_size(sample, type, 0);
1089 return perf_event__synthesize_sample(event, type, 0, sample);
1092 static inline int intel_pt_opt_inject(struct intel_pt *pt,
1093 union perf_event *event,
1094 struct perf_sample *sample, u64 type)
1096 if (!pt->synth_opts.inject)
1099 return intel_pt_inject_event(event, sample, type);
1102 static int intel_pt_deliver_synth_b_event(struct intel_pt *pt,
1103 union perf_event *event,
1104 struct perf_sample *sample, u64 type)
1108 ret = intel_pt_opt_inject(pt, event, sample, type);
1112 ret = perf_session__deliver_synth_event(pt->session, event, sample);
1114 pr_err("Intel PT: failed to deliver event, error %d\n", ret);
1119 static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
1121 struct intel_pt *pt = ptq->pt;
1122 union perf_event *event = ptq->event_buf;
1123 struct perf_sample sample = { .ip = 0, };
1124 struct dummy_branch_stack {
1126 struct branch_entry entries;
1129 if (pt->branches_filter && !(pt->branches_filter & ptq->flags))
1132 if (intel_pt_skip_event(pt))
1135 intel_pt_prep_b_sample(pt, ptq, event, &sample);
1137 sample.id = ptq->pt->branches_id;
1138 sample.stream_id = ptq->pt->branches_id;
1141 * perf report cannot handle events without a branch stack when using
1142 * SORT_MODE__BRANCH so make a dummy one.
1144 if (pt->synth_opts.last_branch && sort__mode == SORT_MODE__BRANCH) {
1145 dummy_bs = (struct dummy_branch_stack){
1152 sample.branch_stack = (struct branch_stack *)&dummy_bs;
1155 return intel_pt_deliver_synth_b_event(pt, event, &sample,
1156 pt->branches_sample_type);
1159 static void intel_pt_prep_sample(struct intel_pt *pt,
1160 struct intel_pt_queue *ptq,
1161 union perf_event *event,
1162 struct perf_sample *sample)
1164 intel_pt_prep_b_sample(pt, ptq, event, sample);
1166 if (pt->synth_opts.callchain) {
1167 thread_stack__sample(ptq->thread, ptq->chain,
1168 pt->synth_opts.callchain_sz + 1,
1169 sample->ip, pt->kernel_start);
1170 sample->callchain = ptq->chain;
1173 if (pt->synth_opts.last_branch) {
1174 intel_pt_copy_last_branch_rb(ptq);
1175 sample->branch_stack = ptq->last_branch;
1179 static inline int intel_pt_deliver_synth_event(struct intel_pt *pt,
1180 struct intel_pt_queue *ptq,
1181 union perf_event *event,
1182 struct perf_sample *sample,
1187 ret = intel_pt_deliver_synth_b_event(pt, event, sample, type);
1189 if (pt->synth_opts.last_branch)
1190 intel_pt_reset_last_branch_rb(ptq);
1195 static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq)
1197 struct intel_pt *pt = ptq->pt;
1198 union perf_event *event = ptq->event_buf;
1199 struct perf_sample sample = { .ip = 0, };
1201 if (intel_pt_skip_event(pt))
1204 intel_pt_prep_sample(pt, ptq, event, &sample);
1206 sample.id = ptq->pt->instructions_id;
1207 sample.stream_id = ptq->pt->instructions_id;
1208 sample.period = ptq->state->tot_insn_cnt - ptq->last_insn_cnt;
1210 ptq->last_insn_cnt = ptq->state->tot_insn_cnt;
1212 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1213 pt->instructions_sample_type);
1216 static int intel_pt_synth_transaction_sample(struct intel_pt_queue *ptq)
1218 struct intel_pt *pt = ptq->pt;
1219 union perf_event *event = ptq->event_buf;
1220 struct perf_sample sample = { .ip = 0, };
1222 if (intel_pt_skip_event(pt))
1225 intel_pt_prep_sample(pt, ptq, event, &sample);
1227 sample.id = ptq->pt->transactions_id;
1228 sample.stream_id = ptq->pt->transactions_id;
1230 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1231 pt->transactions_sample_type);
1234 static void intel_pt_prep_p_sample(struct intel_pt *pt,
1235 struct intel_pt_queue *ptq,
1236 union perf_event *event,
1237 struct perf_sample *sample)
1239 intel_pt_prep_sample(pt, ptq, event, sample);
1242 * Zero IP is used to mean "trace start" but that is not the case for
1243 * power or PTWRITE events with no IP, so clear the flags.
1249 static int intel_pt_synth_ptwrite_sample(struct intel_pt_queue *ptq)
1251 struct intel_pt *pt = ptq->pt;
1252 union perf_event *event = ptq->event_buf;
1253 struct perf_sample sample = { .ip = 0, };
1254 struct perf_synth_intel_ptwrite raw;
1256 if (intel_pt_skip_event(pt))
1259 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1261 sample.id = ptq->pt->ptwrites_id;
1262 sample.stream_id = ptq->pt->ptwrites_id;
1265 raw.ip = !!(ptq->state->flags & INTEL_PT_FUP_IP);
1266 raw.payload = cpu_to_le64(ptq->state->ptw_payload);
1268 sample.raw_size = perf_synth__raw_size(raw);
1269 sample.raw_data = perf_synth__raw_data(&raw);
1271 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1272 pt->ptwrites_sample_type);
1275 static int intel_pt_synth_cbr_sample(struct intel_pt_queue *ptq)
1277 struct intel_pt *pt = ptq->pt;
1278 union perf_event *event = ptq->event_buf;
1279 struct perf_sample sample = { .ip = 0, };
1280 struct perf_synth_intel_cbr raw;
1283 if (intel_pt_skip_event(pt))
1286 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1288 sample.id = ptq->pt->cbr_id;
1289 sample.stream_id = ptq->pt->cbr_id;
1291 flags = (u16)ptq->state->cbr_payload | (pt->max_non_turbo_ratio << 16);
1292 raw.flags = cpu_to_le32(flags);
1293 raw.freq = cpu_to_le32(raw.cbr * pt->cbr2khz);
1296 sample.raw_size = perf_synth__raw_size(raw);
1297 sample.raw_data = perf_synth__raw_data(&raw);
1299 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1300 pt->pwr_events_sample_type);
1303 static int intel_pt_synth_mwait_sample(struct intel_pt_queue *ptq)
1305 struct intel_pt *pt = ptq->pt;
1306 union perf_event *event = ptq->event_buf;
1307 struct perf_sample sample = { .ip = 0, };
1308 struct perf_synth_intel_mwait raw;
1310 if (intel_pt_skip_event(pt))
1313 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1315 sample.id = ptq->pt->mwait_id;
1316 sample.stream_id = ptq->pt->mwait_id;
1319 raw.payload = cpu_to_le64(ptq->state->mwait_payload);
1321 sample.raw_size = perf_synth__raw_size(raw);
1322 sample.raw_data = perf_synth__raw_data(&raw);
1324 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1325 pt->pwr_events_sample_type);
1328 static int intel_pt_synth_pwre_sample(struct intel_pt_queue *ptq)
1330 struct intel_pt *pt = ptq->pt;
1331 union perf_event *event = ptq->event_buf;
1332 struct perf_sample sample = { .ip = 0, };
1333 struct perf_synth_intel_pwre raw;
1335 if (intel_pt_skip_event(pt))
1338 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1340 sample.id = ptq->pt->pwre_id;
1341 sample.stream_id = ptq->pt->pwre_id;
1344 raw.payload = cpu_to_le64(ptq->state->pwre_payload);
1346 sample.raw_size = perf_synth__raw_size(raw);
1347 sample.raw_data = perf_synth__raw_data(&raw);
1349 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1350 pt->pwr_events_sample_type);
1353 static int intel_pt_synth_exstop_sample(struct intel_pt_queue *ptq)
1355 struct intel_pt *pt = ptq->pt;
1356 union perf_event *event = ptq->event_buf;
1357 struct perf_sample sample = { .ip = 0, };
1358 struct perf_synth_intel_exstop raw;
1360 if (intel_pt_skip_event(pt))
1363 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1365 sample.id = ptq->pt->exstop_id;
1366 sample.stream_id = ptq->pt->exstop_id;
1369 raw.ip = !!(ptq->state->flags & INTEL_PT_FUP_IP);
1371 sample.raw_size = perf_synth__raw_size(raw);
1372 sample.raw_data = perf_synth__raw_data(&raw);
1374 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1375 pt->pwr_events_sample_type);
1378 static int intel_pt_synth_pwrx_sample(struct intel_pt_queue *ptq)
1380 struct intel_pt *pt = ptq->pt;
1381 union perf_event *event = ptq->event_buf;
1382 struct perf_sample sample = { .ip = 0, };
1383 struct perf_synth_intel_pwrx raw;
1385 if (intel_pt_skip_event(pt))
1388 intel_pt_prep_p_sample(pt, ptq, event, &sample);
1390 sample.id = ptq->pt->pwrx_id;
1391 sample.stream_id = ptq->pt->pwrx_id;
1394 raw.payload = cpu_to_le64(ptq->state->pwrx_payload);
1396 sample.raw_size = perf_synth__raw_size(raw);
1397 sample.raw_data = perf_synth__raw_data(&raw);
1399 return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
1400 pt->pwr_events_sample_type);
1403 static int intel_pt_synth_error(struct intel_pt *pt, int code, int cpu,
1404 pid_t pid, pid_t tid, u64 ip)
1406 union perf_event event;
1407 char msg[MAX_AUXTRACE_ERROR_MSG];
1410 intel_pt__strerror(code, msg, MAX_AUXTRACE_ERROR_MSG);
1412 auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
1413 code, cpu, pid, tid, ip, msg);
1415 err = perf_session__deliver_synth_event(pt->session, &event, NULL);
1417 pr_err("Intel Processor Trace: failed to deliver error event, error %d\n",
1423 static int intel_pt_next_tid(struct intel_pt *pt, struct intel_pt_queue *ptq)
1425 struct auxtrace_queue *queue;
1426 pid_t tid = ptq->next_tid;
1432 intel_pt_log("switch: cpu %d tid %d\n", ptq->cpu, tid);
1434 err = machine__set_current_tid(pt->machine, ptq->cpu, -1, tid);
1436 queue = &pt->queues.queue_array[ptq->queue_nr];
1437 intel_pt_set_pid_tid_cpu(pt, queue);
1444 static inline bool intel_pt_is_switch_ip(struct intel_pt_queue *ptq, u64 ip)
1446 struct intel_pt *pt = ptq->pt;
1448 return ip == pt->switch_ip &&
1449 (ptq->flags & PERF_IP_FLAG_BRANCH) &&
1450 !(ptq->flags & (PERF_IP_FLAG_CONDITIONAL | PERF_IP_FLAG_ASYNC |
1451 PERF_IP_FLAG_INTERRUPT | PERF_IP_FLAG_TX_ABORT));
1454 #define INTEL_PT_PWR_EVT (INTEL_PT_MWAIT_OP | INTEL_PT_PWR_ENTRY | \
1455 INTEL_PT_EX_STOP | INTEL_PT_PWR_EXIT | \
1458 static int intel_pt_sample(struct intel_pt_queue *ptq)
1460 const struct intel_pt_state *state = ptq->state;
1461 struct intel_pt *pt = ptq->pt;
1464 if (!ptq->have_sample)
1467 ptq->have_sample = false;
1469 if (pt->sample_pwr_events && (state->type & INTEL_PT_PWR_EVT)) {
1470 if (state->type & INTEL_PT_CBR_CHG) {
1471 err = intel_pt_synth_cbr_sample(ptq);
1475 if (state->type & INTEL_PT_MWAIT_OP) {
1476 err = intel_pt_synth_mwait_sample(ptq);
1480 if (state->type & INTEL_PT_PWR_ENTRY) {
1481 err = intel_pt_synth_pwre_sample(ptq);
1485 if (state->type & INTEL_PT_EX_STOP) {
1486 err = intel_pt_synth_exstop_sample(ptq);
1490 if (state->type & INTEL_PT_PWR_EXIT) {
1491 err = intel_pt_synth_pwrx_sample(ptq);
1497 if (pt->sample_instructions && (state->type & INTEL_PT_INSTRUCTION)) {
1498 err = intel_pt_synth_instruction_sample(ptq);
1503 if (pt->sample_transactions && (state->type & INTEL_PT_TRANSACTION)) {
1504 err = intel_pt_synth_transaction_sample(ptq);
1509 if (pt->sample_ptwrites && (state->type & INTEL_PT_PTW)) {
1510 err = intel_pt_synth_ptwrite_sample(ptq);
1515 if (!(state->type & INTEL_PT_BRANCH))
1518 if (pt->synth_opts.callchain || pt->synth_opts.thread_stack)
1519 thread_stack__event(ptq->thread, ptq->flags, state->from_ip,
1520 state->to_ip, ptq->insn_len,
1523 thread_stack__set_trace_nr(ptq->thread, state->trace_nr);
1525 if (pt->sample_branches) {
1526 err = intel_pt_synth_branch_sample(ptq);
1531 if (pt->synth_opts.last_branch)
1532 intel_pt_update_last_branch_rb(ptq);
1534 if (!ptq->sync_switch)
1537 if (intel_pt_is_switch_ip(ptq, state->to_ip)) {
1538 switch (ptq->switch_state) {
1539 case INTEL_PT_SS_NOT_TRACING:
1540 case INTEL_PT_SS_UNKNOWN:
1541 case INTEL_PT_SS_EXPECTING_SWITCH_IP:
1542 err = intel_pt_next_tid(pt, ptq);
1545 ptq->switch_state = INTEL_PT_SS_TRACING;
1548 ptq->switch_state = INTEL_PT_SS_EXPECTING_SWITCH_EVENT;
1551 } else if (!state->to_ip) {
1552 ptq->switch_state = INTEL_PT_SS_NOT_TRACING;
1553 } else if (ptq->switch_state == INTEL_PT_SS_NOT_TRACING) {
1554 ptq->switch_state = INTEL_PT_SS_UNKNOWN;
1555 } else if (ptq->switch_state == INTEL_PT_SS_UNKNOWN &&
1556 state->to_ip == pt->ptss_ip &&
1557 (ptq->flags & PERF_IP_FLAG_CALL)) {
1558 ptq->switch_state = INTEL_PT_SS_TRACING;
1564 static u64 intel_pt_switch_ip(struct intel_pt *pt, u64 *ptss_ip)
1566 struct machine *machine = pt->machine;
1568 struct symbol *sym, *start;
1569 u64 ip, switch_ip = 0;
1575 map = machine__kernel_map(machine);
1582 start = dso__first_symbol(map->dso);
1584 for (sym = start; sym; sym = dso__next_symbol(sym)) {
1585 if (sym->binding == STB_GLOBAL &&
1586 !strcmp(sym->name, "__switch_to")) {
1587 ip = map->unmap_ip(map, sym->start);
1588 if (ip >= map->start && ip < map->end) {
1595 if (!switch_ip || !ptss_ip)
1598 if (pt->have_sched_switch == 1)
1599 ptss = "perf_trace_sched_switch";
1601 ptss = "__perf_event_task_sched_out";
1603 for (sym = start; sym; sym = dso__next_symbol(sym)) {
1604 if (!strcmp(sym->name, ptss)) {
1605 ip = map->unmap_ip(map, sym->start);
1606 if (ip >= map->start && ip < map->end) {
1616 static void intel_pt_enable_sync_switch(struct intel_pt *pt)
1620 pt->sync_switch = true;
1622 for (i = 0; i < pt->queues.nr_queues; i++) {
1623 struct auxtrace_queue *queue = &pt->queues.queue_array[i];
1624 struct intel_pt_queue *ptq = queue->priv;
1627 ptq->sync_switch = true;
1631 static int intel_pt_run_decoder(struct intel_pt_queue *ptq, u64 *timestamp)
1633 const struct intel_pt_state *state = ptq->state;
1634 struct intel_pt *pt = ptq->pt;
1637 if (!pt->kernel_start) {
1638 pt->kernel_start = machine__kernel_start(pt->machine);
1639 if (pt->per_cpu_mmaps &&
1640 (pt->have_sched_switch == 1 || pt->have_sched_switch == 3) &&
1641 !pt->timeless_decoding && intel_pt_tracing_kernel(pt) &&
1642 !pt->sampling_mode) {
1643 pt->switch_ip = intel_pt_switch_ip(pt, &pt->ptss_ip);
1644 if (pt->switch_ip) {
1645 intel_pt_log("switch_ip: %"PRIx64" ptss_ip: %"PRIx64"\n",
1646 pt->switch_ip, pt->ptss_ip);
1647 intel_pt_enable_sync_switch(pt);
1652 intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n",
1653 ptq->queue_nr, ptq->cpu, ptq->pid, ptq->tid);
1655 err = intel_pt_sample(ptq);
1659 state = intel_pt_decode(ptq->decoder);
1661 if (state->err == INTEL_PT_ERR_NODATA)
1663 if (ptq->sync_switch &&
1664 state->from_ip >= pt->kernel_start) {
1665 ptq->sync_switch = false;
1666 intel_pt_next_tid(pt, ptq);
1668 if (pt->synth_opts.errors) {
1669 err = intel_pt_synth_error(pt, state->err,
1680 ptq->have_sample = true;
1681 intel_pt_sample_flags(ptq);
1683 /* Use estimated TSC upon return to user space */
1685 (state->from_ip >= pt->kernel_start || !state->from_ip) &&
1686 state->to_ip && state->to_ip < pt->kernel_start) {
1687 intel_pt_log("TSC %"PRIx64" est. TSC %"PRIx64"\n",
1688 state->timestamp, state->est_timestamp);
1689 ptq->timestamp = state->est_timestamp;
1690 /* Use estimated TSC in unknown switch state */
1691 } else if (ptq->sync_switch &&
1692 ptq->switch_state == INTEL_PT_SS_UNKNOWN &&
1693 intel_pt_is_switch_ip(ptq, state->to_ip) &&
1694 ptq->next_tid == -1) {
1695 intel_pt_log("TSC %"PRIx64" est. TSC %"PRIx64"\n",
1696 state->timestamp, state->est_timestamp);
1697 ptq->timestamp = state->est_timestamp;
1698 } else if (state->timestamp > ptq->timestamp) {
1699 ptq->timestamp = state->timestamp;
1702 if (!pt->timeless_decoding && ptq->timestamp >= *timestamp) {
1703 *timestamp = ptq->timestamp;
1710 static inline int intel_pt_update_queues(struct intel_pt *pt)
1712 if (pt->queues.new_data) {
1713 pt->queues.new_data = false;
1714 return intel_pt_setup_queues(pt);
1719 static int intel_pt_process_queues(struct intel_pt *pt, u64 timestamp)
1721 unsigned int queue_nr;
1726 struct auxtrace_queue *queue;
1727 struct intel_pt_queue *ptq;
1729 if (!pt->heap.heap_cnt)
1732 if (pt->heap.heap_array[0].ordinal >= timestamp)
1735 queue_nr = pt->heap.heap_array[0].queue_nr;
1736 queue = &pt->queues.queue_array[queue_nr];
1739 intel_pt_log("queue %u processing 0x%" PRIx64 " to 0x%" PRIx64 "\n",
1740 queue_nr, pt->heap.heap_array[0].ordinal,
1743 auxtrace_heap__pop(&pt->heap);
1745 if (pt->heap.heap_cnt) {
1746 ts = pt->heap.heap_array[0].ordinal + 1;
1753 intel_pt_set_pid_tid_cpu(pt, queue);
1755 ret = intel_pt_run_decoder(ptq, &ts);
1758 auxtrace_heap__add(&pt->heap, queue_nr, ts);
1763 ret = auxtrace_heap__add(&pt->heap, queue_nr, ts);
1767 ptq->on_heap = false;
1774 static int intel_pt_process_timeless_queues(struct intel_pt *pt, pid_t tid,
1777 struct auxtrace_queues *queues = &pt->queues;
1781 for (i = 0; i < queues->nr_queues; i++) {
1782 struct auxtrace_queue *queue = &pt->queues.queue_array[i];
1783 struct intel_pt_queue *ptq = queue->priv;
1785 if (ptq && (tid == -1 || ptq->tid == tid)) {
1787 intel_pt_set_pid_tid_cpu(pt, queue);
1788 intel_pt_run_decoder(ptq, &ts);
1794 static int intel_pt_lost(struct intel_pt *pt, struct perf_sample *sample)
1796 return intel_pt_synth_error(pt, INTEL_PT_ERR_LOST, sample->cpu,
1797 sample->pid, sample->tid, 0);
1800 static struct intel_pt_queue *intel_pt_cpu_to_ptq(struct intel_pt *pt, int cpu)
1804 if (cpu < 0 || !pt->queues.nr_queues)
1807 if ((unsigned)cpu >= pt->queues.nr_queues)
1808 i = pt->queues.nr_queues - 1;
1812 if (pt->queues.queue_array[i].cpu == cpu)
1813 return pt->queues.queue_array[i].priv;
1815 for (j = 0; i > 0; j++) {
1816 if (pt->queues.queue_array[--i].cpu == cpu)
1817 return pt->queues.queue_array[i].priv;
1820 for (; j < pt->queues.nr_queues; j++) {
1821 if (pt->queues.queue_array[j].cpu == cpu)
1822 return pt->queues.queue_array[j].priv;
1828 static int intel_pt_sync_switch(struct intel_pt *pt, int cpu, pid_t tid,
1831 struct intel_pt_queue *ptq;
1834 if (!pt->sync_switch)
1837 ptq = intel_pt_cpu_to_ptq(pt, cpu);
1838 if (!ptq || !ptq->sync_switch)
1841 switch (ptq->switch_state) {
1842 case INTEL_PT_SS_NOT_TRACING:
1845 case INTEL_PT_SS_UNKNOWN:
1846 case INTEL_PT_SS_TRACING:
1847 ptq->next_tid = tid;
1848 ptq->switch_state = INTEL_PT_SS_EXPECTING_SWITCH_IP;
1850 case INTEL_PT_SS_EXPECTING_SWITCH_EVENT:
1851 if (!ptq->on_heap) {
1852 ptq->timestamp = perf_time_to_tsc(timestamp,
1854 err = auxtrace_heap__add(&pt->heap, ptq->queue_nr,
1858 ptq->on_heap = true;
1860 ptq->switch_state = INTEL_PT_SS_TRACING;
1862 case INTEL_PT_SS_EXPECTING_SWITCH_IP:
1863 ptq->next_tid = tid;
1864 intel_pt_log("ERROR: cpu %d expecting switch ip\n", cpu);
1873 static int intel_pt_process_switch(struct intel_pt *pt,
1874 struct perf_sample *sample)
1876 struct perf_evsel *evsel;
1880 evsel = perf_evlist__id2evsel(pt->session->evlist, sample->id);
1881 if (evsel != pt->switch_evsel)
1884 tid = perf_evsel__intval(evsel, sample, "next_pid");
1887 intel_pt_log("sched_switch: cpu %d tid %d time %"PRIu64" tsc %#"PRIx64"\n",
1888 cpu, tid, sample->time, perf_time_to_tsc(sample->time,
1891 ret = intel_pt_sync_switch(pt, cpu, tid, sample->time);
1895 return machine__set_current_tid(pt->machine, cpu, -1, tid);
1898 static int intel_pt_context_switch(struct intel_pt *pt, union perf_event *event,
1899 struct perf_sample *sample)
1901 bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT;
1907 if (pt->have_sched_switch == 3) {
1910 if (event->header.type != PERF_RECORD_SWITCH_CPU_WIDE) {
1911 pr_err("Expecting CPU-wide context switch event\n");
1914 pid = event->context_switch.next_prev_pid;
1915 tid = event->context_switch.next_prev_tid;
1924 intel_pt_log("context_switch event has no tid\n");
1926 intel_pt_log("context_switch: cpu %d pid %d tid %d time %"PRIu64" tsc %#"PRIx64"\n",
1927 cpu, pid, tid, sample->time, perf_time_to_tsc(sample->time,
1930 ret = intel_pt_sync_switch(pt, cpu, tid, sample->time);
1934 return machine__set_current_tid(pt->machine, cpu, pid, tid);
1937 static int intel_pt_process_itrace_start(struct intel_pt *pt,
1938 union perf_event *event,
1939 struct perf_sample *sample)
1941 if (!pt->per_cpu_mmaps)
1944 intel_pt_log("itrace_start: cpu %d pid %d tid %d time %"PRIu64" tsc %#"PRIx64"\n",
1945 sample->cpu, event->itrace_start.pid,
1946 event->itrace_start.tid, sample->time,
1947 perf_time_to_tsc(sample->time, &pt->tc));
1949 return machine__set_current_tid(pt->machine, sample->cpu,
1950 event->itrace_start.pid,
1951 event->itrace_start.tid);
1954 static int intel_pt_process_event(struct perf_session *session,
1955 union perf_event *event,
1956 struct perf_sample *sample,
1957 struct perf_tool *tool)
1959 struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt,
1967 if (!tool->ordered_events) {
1968 pr_err("Intel Processor Trace requires ordered events\n");
1972 if (sample->time && sample->time != (u64)-1)
1973 timestamp = perf_time_to_tsc(sample->time, &pt->tc);
1977 if (timestamp || pt->timeless_decoding) {
1978 err = intel_pt_update_queues(pt);
1983 if (pt->timeless_decoding) {
1984 if (event->header.type == PERF_RECORD_EXIT) {
1985 err = intel_pt_process_timeless_queues(pt,
1989 } else if (timestamp) {
1990 err = intel_pt_process_queues(pt, timestamp);
1995 if (event->header.type == PERF_RECORD_AUX &&
1996 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) &&
1997 pt->synth_opts.errors) {
1998 err = intel_pt_lost(pt, sample);
2003 if (pt->switch_evsel && event->header.type == PERF_RECORD_SAMPLE)
2004 err = intel_pt_process_switch(pt, sample);
2005 else if (event->header.type == PERF_RECORD_ITRACE_START)
2006 err = intel_pt_process_itrace_start(pt, event, sample);
2007 else if (event->header.type == PERF_RECORD_SWITCH ||
2008 event->header.type == PERF_RECORD_SWITCH_CPU_WIDE)
2009 err = intel_pt_context_switch(pt, event, sample);
2011 intel_pt_log("event %s (%u): cpu %d time %"PRIu64" tsc %#"PRIx64"\n",
2012 perf_event__name(event->header.type), event->header.type,
2013 sample->cpu, sample->time, timestamp);
2018 static int intel_pt_flush(struct perf_session *session, struct perf_tool *tool)
2020 struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt,
2027 if (!tool->ordered_events)
2030 ret = intel_pt_update_queues(pt);
2034 if (pt->timeless_decoding)
2035 return intel_pt_process_timeless_queues(pt, -1,
2038 return intel_pt_process_queues(pt, MAX_TIMESTAMP);
2041 static void intel_pt_free_events(struct perf_session *session)
2043 struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt,
2045 struct auxtrace_queues *queues = &pt->queues;
2048 for (i = 0; i < queues->nr_queues; i++) {
2049 intel_pt_free_queue(queues->queue_array[i].priv);
2050 queues->queue_array[i].priv = NULL;
2052 intel_pt_log_disable();
2053 auxtrace_queues__free(queues);
2056 static void intel_pt_free(struct perf_session *session)
2058 struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt,
2061 auxtrace_heap__free(&pt->heap);
2062 intel_pt_free_events(session);
2063 session->auxtrace = NULL;
2064 thread__put(pt->unknown_thread);
2065 addr_filters__exit(&pt->filts);
2070 static int intel_pt_process_auxtrace_event(struct perf_session *session,
2071 union perf_event *event,
2072 struct perf_tool *tool __maybe_unused)
2074 struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt,
2077 if (!pt->data_queued) {
2078 struct auxtrace_buffer *buffer;
2080 int fd = perf_data__fd(session->data);
2083 if (perf_data__is_pipe(session->data)) {
2086 data_offset = lseek(fd, 0, SEEK_CUR);
2087 if (data_offset == -1)
2091 err = auxtrace_queues__add_event(&pt->queues, session, event,
2092 data_offset, &buffer);
2096 /* Dump here now we have copied a piped trace out of the pipe */
2098 if (auxtrace_buffer__get_data(buffer, fd)) {
2099 intel_pt_dump_event(pt, buffer->data,
2101 auxtrace_buffer__put_data(buffer);
2109 struct intel_pt_synth {
2110 struct perf_tool dummy_tool;
2111 struct perf_session *session;
2114 static int intel_pt_event_synth(struct perf_tool *tool,
2115 union perf_event *event,
2116 struct perf_sample *sample __maybe_unused,
2117 struct machine *machine __maybe_unused)
2119 struct intel_pt_synth *intel_pt_synth =
2120 container_of(tool, struct intel_pt_synth, dummy_tool);
2122 return perf_session__deliver_synth_event(intel_pt_synth->session, event,
2126 static int intel_pt_synth_event(struct perf_session *session, const char *name,
2127 struct perf_event_attr *attr, u64 id)
2129 struct intel_pt_synth intel_pt_synth;
2132 pr_debug("Synthesizing '%s' event with id %" PRIu64 " sample type %#" PRIx64 "\n",
2133 name, id, (u64)attr->sample_type);
2135 memset(&intel_pt_synth, 0, sizeof(struct intel_pt_synth));
2136 intel_pt_synth.session = session;
2138 err = perf_event__synthesize_attr(&intel_pt_synth.dummy_tool, attr, 1,
2139 &id, intel_pt_event_synth);
2141 pr_err("%s: failed to synthesize '%s' event type\n",
2147 static void intel_pt_set_event_name(struct perf_evlist *evlist, u64 id,
2150 struct perf_evsel *evsel;
2152 evlist__for_each_entry(evlist, evsel) {
2153 if (evsel->id && evsel->id[0] == id) {
2155 zfree(&evsel->name);
2156 evsel->name = strdup(name);
2162 static struct perf_evsel *intel_pt_evsel(struct intel_pt *pt,
2163 struct perf_evlist *evlist)
2165 struct perf_evsel *evsel;
2167 evlist__for_each_entry(evlist, evsel) {
2168 if (evsel->attr.type == pt->pmu_type && evsel->ids)
2175 static int intel_pt_synth_events(struct intel_pt *pt,
2176 struct perf_session *session)
2178 struct perf_evlist *evlist = session->evlist;
2179 struct perf_evsel *evsel = intel_pt_evsel(pt, evlist);
2180 struct perf_event_attr attr;
2185 pr_debug("There are no selected events with Intel Processor Trace data\n");
2189 memset(&attr, 0, sizeof(struct perf_event_attr));
2190 attr.size = sizeof(struct perf_event_attr);
2191 attr.type = PERF_TYPE_HARDWARE;
2192 attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK;
2193 attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
2195 if (pt->timeless_decoding)
2196 attr.sample_type &= ~(u64)PERF_SAMPLE_TIME;
2198 attr.sample_type |= PERF_SAMPLE_TIME;
2199 if (!pt->per_cpu_mmaps)
2200 attr.sample_type &= ~(u64)PERF_SAMPLE_CPU;
2201 attr.exclude_user = evsel->attr.exclude_user;
2202 attr.exclude_kernel = evsel->attr.exclude_kernel;
2203 attr.exclude_hv = evsel->attr.exclude_hv;
2204 attr.exclude_host = evsel->attr.exclude_host;
2205 attr.exclude_guest = evsel->attr.exclude_guest;
2206 attr.sample_id_all = evsel->attr.sample_id_all;
2207 attr.read_format = evsel->attr.read_format;
2209 id = evsel->id[0] + 1000000000;
2213 if (pt->synth_opts.branches) {
2214 attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
2215 attr.sample_period = 1;
2216 attr.sample_type |= PERF_SAMPLE_ADDR;
2217 err = intel_pt_synth_event(session, "branches", &attr, id);
2220 pt->sample_branches = true;
2221 pt->branches_sample_type = attr.sample_type;
2222 pt->branches_id = id;
2224 attr.sample_type &= ~(u64)PERF_SAMPLE_ADDR;
2227 if (pt->synth_opts.callchain)
2228 attr.sample_type |= PERF_SAMPLE_CALLCHAIN;
2229 if (pt->synth_opts.last_branch)
2230 attr.sample_type |= PERF_SAMPLE_BRANCH_STACK;
2232 if (pt->synth_opts.instructions) {
2233 attr.config = PERF_COUNT_HW_INSTRUCTIONS;
2234 if (pt->synth_opts.period_type == PERF_ITRACE_PERIOD_NANOSECS)
2235 attr.sample_period =
2236 intel_pt_ns_to_ticks(pt, pt->synth_opts.period);
2238 attr.sample_period = pt->synth_opts.period;
2239 err = intel_pt_synth_event(session, "instructions", &attr, id);
2242 pt->sample_instructions = true;
2243 pt->instructions_sample_type = attr.sample_type;
2244 pt->instructions_id = id;
2248 attr.sample_type &= ~(u64)PERF_SAMPLE_PERIOD;
2249 attr.sample_period = 1;
2251 if (pt->synth_opts.transactions) {
2252 attr.config = PERF_COUNT_HW_INSTRUCTIONS;
2253 err = intel_pt_synth_event(session, "transactions", &attr, id);
2256 pt->sample_transactions = true;
2257 pt->transactions_sample_type = attr.sample_type;
2258 pt->transactions_id = id;
2259 intel_pt_set_event_name(evlist, id, "transactions");
2263 attr.type = PERF_TYPE_SYNTH;
2264 attr.sample_type |= PERF_SAMPLE_RAW;
2266 if (pt->synth_opts.ptwrites) {
2267 attr.config = PERF_SYNTH_INTEL_PTWRITE;
2268 err = intel_pt_synth_event(session, "ptwrite", &attr, id);
2271 pt->sample_ptwrites = true;
2272 pt->ptwrites_sample_type = attr.sample_type;
2273 pt->ptwrites_id = id;
2274 intel_pt_set_event_name(evlist, id, "ptwrite");
2278 if (pt->synth_opts.pwr_events) {
2279 pt->sample_pwr_events = true;
2280 pt->pwr_events_sample_type = attr.sample_type;
2282 attr.config = PERF_SYNTH_INTEL_CBR;
2283 err = intel_pt_synth_event(session, "cbr", &attr, id);
2287 intel_pt_set_event_name(evlist, id, "cbr");
2291 if (pt->synth_opts.pwr_events && (evsel->attr.config & 0x10)) {
2292 attr.config = PERF_SYNTH_INTEL_MWAIT;
2293 err = intel_pt_synth_event(session, "mwait", &attr, id);
2297 intel_pt_set_event_name(evlist, id, "mwait");
2300 attr.config = PERF_SYNTH_INTEL_PWRE;
2301 err = intel_pt_synth_event(session, "pwre", &attr, id);
2305 intel_pt_set_event_name(evlist, id, "pwre");
2308 attr.config = PERF_SYNTH_INTEL_EXSTOP;
2309 err = intel_pt_synth_event(session, "exstop", &attr, id);
2313 intel_pt_set_event_name(evlist, id, "exstop");
2316 attr.config = PERF_SYNTH_INTEL_PWRX;
2317 err = intel_pt_synth_event(session, "pwrx", &attr, id);
2321 intel_pt_set_event_name(evlist, id, "pwrx");
2328 static struct perf_evsel *intel_pt_find_sched_switch(struct perf_evlist *evlist)
2330 struct perf_evsel *evsel;
2332 evlist__for_each_entry_reverse(evlist, evsel) {
2333 const char *name = perf_evsel__name(evsel);
2335 if (!strcmp(name, "sched:sched_switch"))
2342 static bool intel_pt_find_switch(struct perf_evlist *evlist)
2344 struct perf_evsel *evsel;
2346 evlist__for_each_entry(evlist, evsel) {
2347 if (evsel->attr.context_switch)
2354 static int intel_pt_perf_config(const char *var, const char *value, void *data)
2356 struct intel_pt *pt = data;
2358 if (!strcmp(var, "intel-pt.mispred-all"))
2359 pt->mispred_all = perf_config_bool(var, value);
2364 static const char * const intel_pt_info_fmts[] = {
2365 [INTEL_PT_PMU_TYPE] = " PMU Type %"PRId64"\n",
2366 [INTEL_PT_TIME_SHIFT] = " Time Shift %"PRIu64"\n",
2367 [INTEL_PT_TIME_MULT] = " Time Muliplier %"PRIu64"\n",
2368 [INTEL_PT_TIME_ZERO] = " Time Zero %"PRIu64"\n",
2369 [INTEL_PT_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n",
2370 [INTEL_PT_TSC_BIT] = " TSC bit %#"PRIx64"\n",
2371 [INTEL_PT_NORETCOMP_BIT] = " NoRETComp bit %#"PRIx64"\n",
2372 [INTEL_PT_HAVE_SCHED_SWITCH] = " Have sched_switch %"PRId64"\n",
2373 [INTEL_PT_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n",
2374 [INTEL_PT_PER_CPU_MMAPS] = " Per-cpu maps %"PRId64"\n",
2375 [INTEL_PT_MTC_BIT] = " MTC bit %#"PRIx64"\n",
2376 [INTEL_PT_TSC_CTC_N] = " TSC:CTC numerator %"PRIu64"\n",
2377 [INTEL_PT_TSC_CTC_D] = " TSC:CTC denominator %"PRIu64"\n",
2378 [INTEL_PT_CYC_BIT] = " CYC bit %#"PRIx64"\n",
2379 [INTEL_PT_MAX_NONTURBO_RATIO] = " Max non-turbo ratio %"PRIu64"\n",
2380 [INTEL_PT_FILTER_STR_LEN] = " Filter string len. %"PRIu64"\n",
2383 static void intel_pt_print_info(u64 *arr, int start, int finish)
2390 for (i = start; i <= finish; i++)
2391 fprintf(stdout, intel_pt_info_fmts[i], arr[i]);
2394 static void intel_pt_print_info_str(const char *name, const char *str)
2399 fprintf(stdout, " %-20s%s\n", name, str ? str : "");
2402 static bool intel_pt_has(struct auxtrace_info_event *auxtrace_info, int pos)
2404 return auxtrace_info->header.size >=
2405 sizeof(struct auxtrace_info_event) + (sizeof(u64) * (pos + 1));
2408 int intel_pt_process_auxtrace_info(union perf_event *event,
2409 struct perf_session *session)
2411 struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info;
2412 size_t min_sz = sizeof(u64) * INTEL_PT_PER_CPU_MMAPS;
2413 struct intel_pt *pt;
2418 if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) +
2422 pt = zalloc(sizeof(struct intel_pt));
2426 addr_filters__init(&pt->filts);
2428 err = perf_config(intel_pt_perf_config, pt);
2432 err = auxtrace_queues__init(&pt->queues);
2436 intel_pt_log_set_name(INTEL_PT_PMU_NAME);
2438 pt->session = session;
2439 pt->machine = &session->machines.host; /* No kvm support */
2440 pt->auxtrace_type = auxtrace_info->type;
2441 pt->pmu_type = auxtrace_info->priv[INTEL_PT_PMU_TYPE];
2442 pt->tc.time_shift = auxtrace_info->priv[INTEL_PT_TIME_SHIFT];
2443 pt->tc.time_mult = auxtrace_info->priv[INTEL_PT_TIME_MULT];
2444 pt->tc.time_zero = auxtrace_info->priv[INTEL_PT_TIME_ZERO];
2445 pt->cap_user_time_zero = auxtrace_info->priv[INTEL_PT_CAP_USER_TIME_ZERO];
2446 pt->tsc_bit = auxtrace_info->priv[INTEL_PT_TSC_BIT];
2447 pt->noretcomp_bit = auxtrace_info->priv[INTEL_PT_NORETCOMP_BIT];
2448 pt->have_sched_switch = auxtrace_info->priv[INTEL_PT_HAVE_SCHED_SWITCH];
2449 pt->snapshot_mode = auxtrace_info->priv[INTEL_PT_SNAPSHOT_MODE];
2450 pt->per_cpu_mmaps = auxtrace_info->priv[INTEL_PT_PER_CPU_MMAPS];
2451 intel_pt_print_info(&auxtrace_info->priv[0], INTEL_PT_PMU_TYPE,
2452 INTEL_PT_PER_CPU_MMAPS);
2454 if (intel_pt_has(auxtrace_info, INTEL_PT_CYC_BIT)) {
2455 pt->mtc_bit = auxtrace_info->priv[INTEL_PT_MTC_BIT];
2456 pt->mtc_freq_bits = auxtrace_info->priv[INTEL_PT_MTC_FREQ_BITS];
2457 pt->tsc_ctc_ratio_n = auxtrace_info->priv[INTEL_PT_TSC_CTC_N];
2458 pt->tsc_ctc_ratio_d = auxtrace_info->priv[INTEL_PT_TSC_CTC_D];
2459 pt->cyc_bit = auxtrace_info->priv[INTEL_PT_CYC_BIT];
2460 intel_pt_print_info(&auxtrace_info->priv[0], INTEL_PT_MTC_BIT,
2464 if (intel_pt_has(auxtrace_info, INTEL_PT_MAX_NONTURBO_RATIO)) {
2465 pt->max_non_turbo_ratio =
2466 auxtrace_info->priv[INTEL_PT_MAX_NONTURBO_RATIO];
2467 intel_pt_print_info(&auxtrace_info->priv[0],
2468 INTEL_PT_MAX_NONTURBO_RATIO,
2469 INTEL_PT_MAX_NONTURBO_RATIO);
2472 info = &auxtrace_info->priv[INTEL_PT_FILTER_STR_LEN] + 1;
2473 info_end = (void *)info + auxtrace_info->header.size;
2475 if (intel_pt_has(auxtrace_info, INTEL_PT_FILTER_STR_LEN)) {
2478 len = auxtrace_info->priv[INTEL_PT_FILTER_STR_LEN];
2479 intel_pt_print_info(&auxtrace_info->priv[0],
2480 INTEL_PT_FILTER_STR_LEN,
2481 INTEL_PT_FILTER_STR_LEN);
2483 const char *filter = (const char *)info;
2485 len = roundup(len + 1, 8);
2487 if ((void *)info > info_end) {
2488 pr_err("%s: bad filter string length\n", __func__);
2490 goto err_free_queues;
2492 pt->filter = memdup(filter, len);
2495 goto err_free_queues;
2497 if (session->header.needs_swap)
2498 mem_bswap_64(pt->filter, len);
2499 if (pt->filter[len - 1]) {
2500 pr_err("%s: filter string not null terminated\n", __func__);
2502 goto err_free_queues;
2504 err = addr_filters__parse_bare_filter(&pt->filts,
2507 goto err_free_queues;
2509 intel_pt_print_info_str("Filter string", pt->filter);
2512 pt->timeless_decoding = intel_pt_timeless_decoding(pt);
2513 if (pt->timeless_decoding && !pt->tc.time_mult)
2514 pt->tc.time_mult = 1;
2515 pt->have_tsc = intel_pt_have_tsc(pt);
2516 pt->sampling_mode = false;
2517 pt->est_tsc = !pt->timeless_decoding;
2519 pt->unknown_thread = thread__new(999999999, 999999999);
2520 if (!pt->unknown_thread) {
2522 goto err_free_queues;
2526 * Since this thread will not be kept in any rbtree not in a
2527 * list, initialize its list node so that at thread__put() the
2528 * current thread lifetime assuption is kept and we don't segfault
2529 * at list_del_init().
2531 INIT_LIST_HEAD(&pt->unknown_thread->node);
2533 err = thread__set_comm(pt->unknown_thread, "unknown", 0);
2535 goto err_delete_thread;
2536 if (thread__init_map_groups(pt->unknown_thread, pt->machine)) {
2538 goto err_delete_thread;
2541 pt->auxtrace.process_event = intel_pt_process_event;
2542 pt->auxtrace.process_auxtrace_event = intel_pt_process_auxtrace_event;
2543 pt->auxtrace.flush_events = intel_pt_flush;
2544 pt->auxtrace.free_events = intel_pt_free_events;
2545 pt->auxtrace.free = intel_pt_free;
2546 session->auxtrace = &pt->auxtrace;
2551 if (pt->have_sched_switch == 1) {
2552 pt->switch_evsel = intel_pt_find_sched_switch(session->evlist);
2553 if (!pt->switch_evsel) {
2554 pr_err("%s: missing sched_switch event\n", __func__);
2556 goto err_delete_thread;
2558 } else if (pt->have_sched_switch == 2 &&
2559 !intel_pt_find_switch(session->evlist)) {
2560 pr_err("%s: missing context_switch attribute flag\n", __func__);
2562 goto err_delete_thread;
2565 if (session->itrace_synth_opts && session->itrace_synth_opts->set) {
2566 pt->synth_opts = *session->itrace_synth_opts;
2568 itrace_synth_opts__set_default(&pt->synth_opts);
2569 if (use_browser != -1) {
2570 pt->synth_opts.branches = false;
2571 pt->synth_opts.callchain = true;
2573 if (session->itrace_synth_opts)
2574 pt->synth_opts.thread_stack =
2575 session->itrace_synth_opts->thread_stack;
2578 if (pt->synth_opts.log)
2579 intel_pt_log_enable();
2581 /* Maximum non-turbo ratio is TSC freq / 100 MHz */
2582 if (pt->tc.time_mult) {
2583 u64 tsc_freq = intel_pt_ns_to_ticks(pt, 1000000000);
2585 if (!pt->max_non_turbo_ratio)
2586 pt->max_non_turbo_ratio =
2587 (tsc_freq + 50000000) / 100000000;
2588 intel_pt_log("TSC frequency %"PRIu64"\n", tsc_freq);
2589 intel_pt_log("Maximum non-turbo ratio %u\n",
2590 pt->max_non_turbo_ratio);
2591 pt->cbr2khz = tsc_freq / pt->max_non_turbo_ratio / 1000;
2594 if (pt->synth_opts.calls)
2595 pt->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC |
2596 PERF_IP_FLAG_TRACE_END;
2597 if (pt->synth_opts.returns)
2598 pt->branches_filter |= PERF_IP_FLAG_RETURN |
2599 PERF_IP_FLAG_TRACE_BEGIN;
2601 if (pt->synth_opts.callchain && !symbol_conf.use_callchain) {
2602 symbol_conf.use_callchain = true;
2603 if (callchain_register_param(&callchain_param) < 0) {
2604 symbol_conf.use_callchain = false;
2605 pt->synth_opts.callchain = false;
2609 err = intel_pt_synth_events(pt, session);
2611 goto err_delete_thread;
2613 err = auxtrace_queues__process_index(&pt->queues, session);
2615 goto err_delete_thread;
2617 if (pt->queues.populated)
2618 pt->data_queued = true;
2620 if (pt->timeless_decoding)
2621 pr_debug2("Intel PT decoding without timestamps\n");
2626 thread__zput(pt->unknown_thread);
2628 intel_pt_log_disable();
2629 auxtrace_queues__free(&pt->queues);
2630 session->auxtrace = NULL;
2632 addr_filters__exit(&pt->filts);