Lines Matching +full:sample +full:- +full:at +full:- +full:reset
1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright(C) 2015-2018 Linaro Limited.
12 #include <linux/coresight-pmu.h>
22 #include "cs-etm.h"
23 #include "cs-etm-decoder/cs-etm-decoder.h"
37 #include "thread-stack.h"
40 #include "util/synthetic-events.h"
61 * Per-thread ignores the trace channel ID and instead assumes that
151 #define SINK_UNSET ((u32) -1)
168 inode = intlist__find(etmq->traceid_list, trace_chan_id);
170 return -EINVAL;
172 metadata = inode->priv;
182 inode = intlist__find(etmq->traceid_list, trace_chan_id);
184 return -EINVAL;
186 metadata = inode->priv;
199 * are enabled at the same time when the session runs on an EL2 kernel.
204 * The result is cached in etm->pid_fmt so this function only needs to be called
231 return etmq->etm->pid_fmt;
238 struct int_node *inode = intlist__findnew(etmq->traceid_list, trace_chan_id);
242 return -ENOMEM;
244 /* Disallow re-mapping a different traceID to metadata pair. */
245 if (inode->priv) {
246 u64 *curr_cpu_data = inode->priv;
253 * are expected (but not supported) in per-thread mode,
256 if (etmq->etm->per_thread_decoding)
257 pr_err("CS_ETM: overlapping Trace IDs aren't currently supported in per-thread mode\n");
261 return -EINVAL;
271 return -EINVAL;
274 /* Skip re-adding the same mappings if everything matched */
279 inode->priv = cpu_metadata;
286 if (etm->per_thread_decoding)
287 return etm->queues.queue_array[0].priv;
289 return etm->queues.queue_array[cpu].priv;
302 if (etmq->format == UNFORMATTED)
310 for (unsigned int i = 0; i < etm->queues.nr_queues; ++i) {
313 etmq = etm->queues.queue_array[i].priv;
332 return -EINVAL;
355 * Check sink id hasn't changed in per-cpu mode. In per-thread mode,
359 if (!etmq->etm->per_thread_decoding && etmq->sink_id != SINK_UNSET &&
360 etmq->sink_id != sink_id) {
362 return -EINVAL;
365 etmq->sink_id = sink_id;
368 for (unsigned int i = 0; i < etm->queues.nr_queues; ++i) {
369 struct cs_etm_queue *other_etmq = etm->queues.queue_array[i].priv;
372 if (other_etmq->sink_id != etmq->sink_id)
376 if (other_etmq->traceid_list == etmq->traceid_list)
379 /* At the point of first linking, this one should be empty */
380 if (!intlist__empty(etmq->traceid_list)) {
382 return -EINVAL;
385 etmq->own_traceid_list = NULL;
386 intlist__delete(etmq->traceid_list);
387 etmq->traceid_list = other_etmq->traceid_list;
418 return -EINVAL;
440 return -EINVAL;
453 for (i = 0; i < etm->num_cpu; i++) {
454 if (etm->metadata[i][CS_ETM_CPU] == (u64)cpu) {
459 return -1;
470 return (idx != -1) ? etm->metadata[idx] : NULL;
484 struct perf_sample sample;
490 hw_id = event->aux_output_hw_id.hw_id;
497 return -EINVAL;
501 etm = container_of(session->auxtrace, struct cs_etm_auxtrace, auxtrace);
502 if (!etm || !etm->metadata)
503 return -EINVAL;
505 /* parse the sample to get the CPU */
506 evsel = evlist__event2evsel(session->evlist, event);
508 return -EINVAL;
509 perf_sample__init(&sample, /*all=*/false);
510 err = evsel__parse_sample(evsel, event, &sample);
513 cpu = sample.cpu;
514 if (cpu == -1) {
515 /* no CPU in the sample - possibly recorded with an old version of perf */
516 pr_err("CS_ETM: no CPU AUX_OUTPUT_HW_ID sample. Use compatible perf to record.");
517 err = -EINVAL;
528 perf_sample__exit(&sample);
542 etmq->pending_timestamp_chan_id = trace_chan_id;
550 if (!etmq->pending_timestamp_chan_id)
554 *trace_chan_id = etmq->pending_timestamp_chan_id;
557 etmq->pending_timestamp_chan_id);
562 etmq->pending_timestamp_chan_id = 0;
565 return packet_queue->cs_timestamp;
572 queue->head = 0;
573 queue->tail = 0;
574 queue->packet_count = 0;
576 queue->packet_buffer[i].isa = CS_ETM_ISA_UNKNOWN;
577 queue->packet_buffer[i].start_addr = CS_ETM_INVAL_ADDR;
578 queue->packet_buffer[i].end_addr = CS_ETM_INVAL_ADDR;
579 queue->packet_buffer[i].instr_count = 0;
580 queue->packet_buffer[i].last_instr_taken_branch = false;
581 queue->packet_buffer[i].last_instr_size = 0;
582 queue->packet_buffer[i].last_instr_type = 0;
583 queue->packet_buffer[i].last_instr_subtype = 0;
584 queue->packet_buffer[i].last_instr_cond = 0;
585 queue->packet_buffer[i].flags = 0;
586 queue->packet_buffer[i].exception_number = UINT32_MAX;
587 queue->packet_buffer[i].trace_chan_id = UINT8_MAX;
588 queue->packet_buffer[i].cpu = INT_MIN;
597 struct intlist *traceid_queues_list = etmq->traceid_queues_list;
600 idx = (int)(intptr_t)inode->priv;
601 tidq = etmq->traceid_queues[idx];
602 cs_etm__clear_packet_queue(&tidq->packet_queue);
610 int rc = -ENOMEM;
612 struct cs_etm_auxtrace *etm = etmq->etm;
614 cs_etm__clear_packet_queue(&tidq->packet_queue);
616 queue = &etmq->etm->queues.queue_array[etmq->queue_nr];
617 tidq->trace_chan_id = trace_chan_id;
618 tidq->el = tidq->prev_packet_el = ocsd_EL_unknown;
619 tidq->thread = machine__findnew_thread(&etm->session->machines.host, -1,
620 queue->tid);
621 tidq->prev_packet_thread = machine__idle_thread(&etm->session->machines.host);
623 tidq->packet = zalloc(sizeof(struct cs_etm_packet));
624 if (!tidq->packet)
627 tidq->prev_packet = zalloc(sizeof(struct cs_etm_packet));
628 if (!tidq->prev_packet)
631 if (etm->synth_opts.last_branch) {
634 sz += etm->synth_opts.last_branch_sz *
636 tidq->last_branch = zalloc(sz);
637 if (!tidq->last_branch)
639 tidq->last_branch_rb = zalloc(sz);
640 if (!tidq->last_branch_rb)
644 tidq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE);
645 if (!tidq->event_buf)
651 zfree(&tidq->last_branch_rb);
652 zfree(&tidq->last_branch);
653 zfree(&tidq->prev_packet);
654 zfree(&tidq->packet);
666 struct cs_etm_auxtrace *etm = etmq->etm;
668 if (etm->per_thread_decoding)
671 traceid_queues_list = etmq->traceid_queues_list;
679 idx = (int)(intptr_t)inode->priv;
680 return etmq->traceid_queues[idx];
698 inode->priv = (void *)(intptr_t)idx;
704 traceid_queues = etmq->traceid_queues;
717 etmq->traceid_queues = traceid_queues;
719 return etmq->traceid_queues[idx];
739 return &tidq->packet_queue;
749 if (etm->synth_opts.branches || etm->synth_opts.last_branch ||
750 etm->synth_opts.instructions) {
758 * thread at that time must also be assigned to that sample.
760 * tracking the thread for the previous packet the branch sample
763 tmp = tidq->packet;
764 tidq->packet = tidq->prev_packet;
765 tidq->prev_packet = tmp;
766 tidq->prev_packet_el = tidq->el;
767 thread__put(tidq->prev_packet_thread);
768 tidq->prev_packet_thread = thread__get(tidq->thread);
780 snprintf(queue_nr, sizeof(queue_nr), "Qnr:%d; ", etmq->queue_nr);
784 if (len && (pkt_string[len-1] == '\n'))
795 t_params->protocol = cs_etm__get_v7_protocol_version(etmidr);
796 t_params->etmv3.reg_ctrl = metadata[CS_ETM_ETMCR];
797 t_params->etmv3.reg_trc_id = metadata[CS_ETM_ETMTRACEIDR];
803 t_params->protocol = CS_ETM_PROTO_ETMV4i;
804 t_params->etmv4.reg_idr0 = metadata[CS_ETMV4_TRCIDR0];
805 t_params->etmv4.reg_idr1 = metadata[CS_ETMV4_TRCIDR1];
806 t_params->etmv4.reg_idr2 = metadata[CS_ETMV4_TRCIDR2];
807 t_params->etmv4.reg_idr8 = metadata[CS_ETMV4_TRCIDR8];
808 t_params->etmv4.reg_configr = metadata[CS_ETMV4_TRCCONFIGR];
809 t_params->etmv4.reg_traceidr = metadata[CS_ETMV4_TRCTRACEIDR];
815 t_params->protocol = CS_ETM_PROTO_ETE;
816 t_params->ete.reg_idr0 = metadata[CS_ETE_TRCIDR0];
817 t_params->ete.reg_idr1 = metadata[CS_ETE_TRCIDR1];
818 t_params->ete.reg_idr2 = metadata[CS_ETE_TRCIDR2];
819 t_params->ete.reg_idr8 = metadata[CS_ETE_TRCIDR8];
820 t_params->ete.reg_configr = metadata[CS_ETE_TRCCONFIGR];
821 t_params->ete.reg_traceidr = metadata[CS_ETE_TRCTRACEIDR];
822 t_params->ete.reg_devarch = metadata[CS_ETE_TRCDEVARCH];
830 intlist__for_each_entry(inode, etmq->traceid_list) {
831 u64 *metadata = inode->priv;
847 return -EINVAL;
858 int ret = -EINVAL;
863 d_params->packet_printer = cs_etm__packet_dump;
864 d_params->operation = mode;
865 d_params->data = etmq;
866 d_params->formatted = etmq->format == FORMATTED;
867 d_params->fsyncs = false;
868 d_params->hsyncs = false;
869 d_params->frame_aligned = true;
886 cs_etm_decoder__get_name(etmq->decoder), buffer->size);
892 etmq->decoder, buffer->offset,
893 &((u8 *)buffer->data)[buffer_used],
894 buffer->size - buffer_used, &consumed);
899 } while (buffer_used < buffer->size);
901 cs_etm_decoder__reset(etmq->decoder);
907 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
913 if (!tool->ordered_events)
914 return -EINVAL;
916 if (etm->timeless_decoding) {
918 * Pass tid = -1 to process all queues. But likely they will have
921 return cs_etm__process_timeless_queues(etm, -1);
933 struct intlist *traceid_queues_list = etmq->traceid_queues_list;
936 priv = (uintptr_t)inode->priv;
940 tidq = etmq->traceid_queues[idx];
941 thread__zput(tidq->thread);
942 thread__zput(tidq->prev_packet_thread);
943 zfree(&tidq->event_buf);
944 zfree(&tidq->last_branch);
945 zfree(&tidq->last_branch_rb);
946 zfree(&tidq->prev_packet);
947 zfree(&tidq->packet);
959 etmq->traceid_queues_list = NULL;
962 zfree(&etmq->traceid_queues);
973 cs_etm_decoder__free(etmq->decoder);
976 if (etmq->own_traceid_list) {
978 intlist__for_each_entry_safe(inode, tmp, etmq->own_traceid_list)
979 intlist__remove(etmq->own_traceid_list, inode);
982 intlist__delete(etmq->own_traceid_list);
991 struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
994 struct auxtrace_queues *queues = &aux->queues;
996 for (i = 0; i < queues->nr_queues; i++) {
997 cs_etm__free_queue(queues->queue_array[i].priv);
998 queues->queue_array[i].priv = NULL;
1007 struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
1011 session->auxtrace = NULL;
1013 for (i = 0; i < aux->num_cpu; i++)
1014 zfree(&aux->metadata[i]);
1016 zfree(&aux->metadata);
1023 struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
1027 return evsel->core.attr.type == aux->pmu_type;
1037 * running at EL1 assume everything is the host.
1040 return &etmq->etm->session->machines.host;
1052 return machines__find_guest(&etmq->etm->session->machines,
1059 return &etmq->etm->session->machines.host;
1078 * Can't really happen at the moment because
1118 assert(tidq->el == ocsd_EL1 || tidq->el == ocsd_EL0);
1120 assert(tidq->el == ocsd_EL2);
1122 assert(tidq->el == ocsd_EL3);
1125 cpumode = cs_etm__cpu_mode(etmq, address, tidq->el);
1127 if (!thread__find_map(tidq->thread, cpumode, address, &al))
1134 if (dso__data(dso)->status == DSO_DATA_STATUS_ERROR &&
1142 len = dso__data_read_offset(dso, maps__machine(thread__maps(tidq->thread)),
1147 " Enable CONFIG_PROC_KCORE or use option '-k /path/to/vmlinux' for kernel symbols.\n");
1168 etmq->traceid_queues_list = intlist__new(NULL);
1169 if (!etmq->traceid_queues_list)
1173 * Create an RB tree for traceID-metadata tuple. Since the conversion
1177 etmq->traceid_list = etmq->own_traceid_list = intlist__new(NULL);
1178 if (!etmq->traceid_list)
1184 intlist__delete(etmq->traceid_queues_list);
1194 struct cs_etm_queue *etmq = queue->priv;
1202 return -ENOMEM;
1204 queue->priv = etmq;
1205 etmq->etm = etm;
1206 etmq->queue_nr = queue_nr;
1207 queue->cpu = queue_nr; /* Placeholder, may be reset to -1 in per-thread mode */
1208 etmq->offset = 0;
1209 etmq->sink_id = SINK_UNSET;
1224 * We are under a CPU-wide trace scenario. As such we need to know
1280 ret = auxtrace_heap__add(&etm->heap, cs_queue_nr, cs_timestamp);
1289 struct branch_stack *bs_src = tidq->last_branch_rb;
1290 struct branch_stack *bs_dst = tidq->last_branch;
1294 * Set the number of records before early exit: ->nr is used to
1295 * determine how many branches to copy from ->entries.
1297 bs_dst->nr = bs_src->nr;
1302 if (!bs_src->nr)
1306 * As bs_src->entries is a circular buffer, we need to copy from it in
1308 * branch ->last_branch_pos until the end of bs_src->entries buffer.
1310 nr = etmq->etm->synth_opts.last_branch_sz - tidq->last_branch_pos;
1311 memcpy(&bs_dst->entries[0],
1312 &bs_src->entries[tidq->last_branch_pos],
1316 * If we wrapped around at least once, the branches from the beginning
1317 * of the bs_src->entries buffer and until the ->last_branch_pos element
1322 if (bs_src->nr >= etmq->etm->synth_opts.last_branch_sz) {
1323 memcpy(&bs_dst->entries[nr],
1324 &bs_src->entries[0],
1325 sizeof(struct branch_entry) * tidq->last_branch_pos);
1332 tidq->last_branch_pos = 0;
1333 tidq->last_branch_rb->nr = 0;
1345 * 16-bit word of the instruction: 0b11101, 0b11110 and 0b11111
1346 * denote a 32-bit instruction.
1357 if (packet->sample_type == CS_ETM_DISCONTINUITY ||
1358 packet->sample_type == CS_ETM_EXCEPTION)
1361 return packet->start_addr;
1368 if (packet->sample_type == CS_ETM_DISCONTINUITY)
1371 return packet->end_addr - packet->last_instr_size;
1379 if (packet->isa == CS_ETM_ISA_T32) {
1380 u64 addr = packet->start_addr;
1385 offset--;
1391 return packet->start_addr + offset * 4;
1397 struct branch_stack *bs = tidq->last_branch_rb;
1406 if (!tidq->last_branch_pos)
1407 tidq->last_branch_pos = etmq->etm->synth_opts.last_branch_sz;
1409 tidq->last_branch_pos -= 1;
1411 be = &bs->entries[tidq->last_branch_pos];
1412 be->from = cs_etm__last_executed_instr(tidq->prev_packet);
1413 be->to = cs_etm__first_executed_instr(tidq->packet);
1415 be->flags.mispred = 0;
1416 be->flags.predicted = 1;
1419 * Increment bs->nr until reaching the number of last branches asked by
1422 if (bs->nr < etmq->etm->synth_opts.last_branch_sz)
1423 bs->nr += 1;
1427 struct perf_sample *sample, u64 type)
1429 event->header.size = perf_event__sample_event_size(sample, type, 0);
1430 return perf_event__synthesize_sample(event, type, 0, sample);
1437 struct auxtrace_buffer *aux_buffer = etmq->buffer;
1441 queue = &etmq->etm->queues.queue_array[etmq->queue_nr];
1449 etmq->buf_len = 0;
1453 etmq->buffer = aux_buffer;
1456 if (!aux_buffer->data) {
1458 int fd = perf_data__fd(etmq->etm->session->data);
1460 aux_buffer->data = auxtrace_buffer__get_data(aux_buffer, fd);
1461 if (!aux_buffer->data)
1462 return -ENOMEM;
1469 etmq->buf_used = 0;
1470 etmq->buf_len = aux_buffer->size;
1471 etmq->buf = aux_buffer->data;
1473 return etmq->buf_len;
1482 if (tid != -1) {
1483 thread__zput(tidq->thread);
1484 tidq->thread = machine__find_thread(machine, -1, tid);
1488 if (!tidq->thread)
1489 tidq->thread = machine__idle_thread(machine);
1491 tidq->el = el;
1501 return -EINVAL;
1509 return !!etmq->etm->timeless_decoding;
1515 struct perf_sample *sample)
1521 if (packet->sample_type == CS_ETM_DISCONTINUITY) {
1522 sample->insn_len = 0;
1527 * T32 instruction size might be 32-bit or 16-bit, decide by calling
1530 if (packet->isa == CS_ETM_ISA_T32)
1531 sample->insn_len = cs_etm__t32_instr_size(etmq, trace_chan_id,
1532 sample->ip);
1533 /* Otherwise, A64 and A32 instruction size are always 32-bit. */
1535 sample->insn_len = 4;
1537 cs_etm__mem_access(etmq, trace_chan_id, sample->ip, sample->insn_len,
1538 (void *)sample->insn, 0);
1543 struct cs_etm_auxtrace *etm = etmq->etm;
1545 if (etm->has_virtual_ts)
1546 return tsc_to_perf_time(cs_timestamp, &etm->tc);
1554 struct cs_etm_auxtrace *etm = etmq->etm;
1555 struct cs_etm_packet_queue *packet_queue = &tidq->packet_queue;
1557 if (!etm->timeless_decoding && etm->has_virtual_ts)
1558 return packet_queue->cs_timestamp;
1560 return etm->latest_kernel_timestamp;
1568 struct cs_etm_auxtrace *etm = etmq->etm;
1569 union perf_event *event = tidq->event_buf;
1570 struct perf_sample sample;
1572 perf_sample__init(&sample, /*all=*/true);
1573 event->sample.header.type = PERF_RECORD_SAMPLE;
1574 event->sample.header.misc = cs_etm__cpu_mode(etmq, addr, tidq->el);
1575 event->sample.header.size = sizeof(struct perf_event_header);
1578 sample.time = cs_etm__resolve_sample_time(etmq, tidq);
1580 sample.ip = addr;
1581 sample.pid = thread__pid(tidq->thread);
1582 sample.tid = thread__tid(tidq->thread);
1583 sample.id = etmq->etm->instructions_id;
1584 sample.stream_id = etmq->etm->instructions_id;
1585 sample.period = period;
1586 sample.cpu = tidq->packet->cpu;
1587 sample.flags = tidq->prev_packet->flags;
1588 sample.cpumode = event->sample.header.misc;
1590 cs_etm__copy_insn(etmq, tidq->trace_chan_id, tidq->packet, &sample);
1592 if (etm->synth_opts.last_branch)
1593 sample.branch_stack = tidq->last_branch;
1595 if (etm->synth_opts.inject) {
1596 ret = cs_etm__inject_event(event, &sample,
1597 etm->instructions_sample_type);
1602 ret = perf_session__deliver_synth_event(etm->session, event, &sample);
1609 perf_sample__exit(&sample);
1615 * and the next taken branch. Generate sample accordingly.
1621 struct cs_etm_auxtrace *etm = etmq->etm;
1622 struct perf_sample sample = {.ip = 0,};
1623 union perf_event *event = tidq->event_buf;
1631 ip = cs_etm__last_executed_instr(tidq->prev_packet);
1633 event->sample.header.type = PERF_RECORD_SAMPLE;
1634 event->sample.header.misc = cs_etm__cpu_mode(etmq, ip,
1635 tidq->prev_packet_el);
1636 event->sample.header.size = sizeof(struct perf_event_header);
1639 sample.time = cs_etm__resolve_sample_time(etmq, tidq);
1641 sample.ip = ip;
1642 sample.pid = thread__pid(tidq->prev_packet_thread);
1643 sample.tid = thread__tid(tidq->prev_packet_thread);
1644 sample.addr = cs_etm__first_executed_instr(tidq->packet);
1645 sample.id = etmq->etm->branches_id;
1646 sample.stream_id = etmq->etm->branches_id;
1647 sample.period = 1;
1648 sample.cpu = tidq->packet->cpu;
1649 sample.flags = tidq->prev_packet->flags;
1650 sample.cpumode = event->sample.header.misc;
1652 cs_etm__copy_insn(etmq, tidq->trace_chan_id, tidq->prev_packet,
1653 &sample);
1658 if (etm->synth_opts.last_branch) {
1661 .hw_idx = -1ULL,
1663 .from = sample.ip,
1664 .to = sample.addr,
1667 sample.branch_stack = (struct branch_stack *)&dummy_bs;
1670 if (etm->synth_opts.inject) {
1671 ret = cs_etm__inject_event(event, &sample,
1672 etm->branches_sample_type);
1677 ret = perf_session__deliver_synth_event(etm->session, event, &sample);
1690 struct evlist *evlist = session->evlist;
1698 if (evsel->core.attr.type == etm->pmu_type) {
1712 attr.sample_type = evsel->core.attr.sample_type & PERF_SAMPLE_MASK;
1715 if (etm->timeless_decoding)
1720 attr.exclude_user = evsel->core.attr.exclude_user;
1721 attr.exclude_kernel = evsel->core.attr.exclude_kernel;
1722 attr.exclude_hv = evsel->core.attr.exclude_hv;
1723 attr.exclude_host = evsel->core.attr.exclude_host;
1724 attr.exclude_guest = evsel->core.attr.exclude_guest;
1725 attr.sample_id_all = evsel->core.attr.sample_id_all;
1726 attr.read_format = evsel->core.attr.read_format;
1729 id = evsel->core.id[0] + 1000000000;
1734 if (etm->synth_opts.branches) {
1741 etm->branches_sample_type = attr.sample_type;
1742 etm->branches_id = id;
1747 if (etm->synth_opts.last_branch) {
1750 * We don't use the hardware index, but the sample generation
1757 if (etm->synth_opts.instructions) {
1759 attr.sample_period = etm->synth_opts.period;
1760 etm->instructions_sample_period = attr.sample_period;
1764 etm->instructions_sample_type = attr.sample_type;
1765 etm->instructions_id = id;
1775 struct cs_etm_auxtrace *etm = etmq->etm;
1777 u8 trace_chan_id = tidq->trace_chan_id;
1781 instrs_prev = tidq->period_instructions;
1783 tidq->period_instructions += tidq->packet->instr_count;
1789 if (etm->synth_opts.last_branch &&
1790 tidq->prev_packet->sample_type == CS_ETM_RANGE &&
1791 tidq->prev_packet->last_instr_taken_branch)
1794 if (etm->synth_opts.instructions &&
1795 tidq->period_instructions >= etm->instructions_sample_period) {
1797 * Emit instruction sample periodically
1806 * Sample(n) Sample(n+1) Sample(n+2) Sample(n+3)
1809 * --------------------------------------------------
1816 * \---------------- -----------------/
1818 * tidq->packet->instr_count
1820 * Instrs Sample(n...) are the synthesised samples occurring
1821 * every etm->instructions_sample_period instructions - as
1822 * defined on the perf command line. Sample(n) is being the
1823 * last sample before the current etm packet, n+1 to n+3
1826 * tidq->packet->instr_count represents the number of
1830 * instructions executed after the sample point(n) from the
1832 * etm->instructions_sample_period.
1837 * sample(n+1); sample(n+2) and sample(n+3) consume the
1838 * instructions with sample period. After sample(n+3), the rest
1840 * to tidq->period_instructions for next round calculation.
1846 * etm->instructions_sample_period.
1848 u64 offset = etm->instructions_sample_period - instrs_prev;
1851 /* Prepare last branches for instruction sample */
1852 if (etm->synth_opts.last_branch)
1855 while (tidq->period_instructions >=
1856 etm->instructions_sample_period) {
1858 * Calculate the address of the sampled instruction (-1
1859 * as sample is reported as though instruction has just
1864 tidq->packet, offset - 1);
1867 etm->instructions_sample_period);
1871 offset += etm->instructions_sample_period;
1872 tidq->period_instructions -=
1873 etm->instructions_sample_period;
1877 if (etm->synth_opts.branches) {
1880 /* Generate sample for tracing on packet */
1881 if (tidq->prev_packet->sample_type == CS_ETM_DISCONTINUITY)
1884 /* Generate sample for branch taken packet */
1885 if (tidq->prev_packet->sample_type == CS_ETM_RANGE &&
1886 tidq->prev_packet->last_instr_taken_branch)
1906 * to set 'prev_packet->last_instr_taken_branch' to true. This ensures
1907 * to generate branch sample for the instruction range before the
1914 if (tidq->prev_packet->sample_type == CS_ETM_RANGE)
1915 tidq->prev_packet->last_instr_taken_branch = true;
1924 struct cs_etm_auxtrace *etm = etmq->etm;
1927 if (tidq->prev_packet->sample_type == CS_ETM_EMPTY)
1930 if (etmq->etm->synth_opts.last_branch &&
1931 etmq->etm->synth_opts.instructions &&
1932 tidq->prev_packet->sample_type == CS_ETM_RANGE) {
1935 /* Prepare last branches for instruction sample */
1940 * circular buffer at the end of the trace.
1945 addr = cs_etm__last_executed_instr(tidq->prev_packet);
1949 tidq->period_instructions);
1953 tidq->period_instructions = 0;
1957 if (etm->synth_opts.branches &&
1958 tidq->prev_packet->sample_type == CS_ETM_RANGE) {
1967 /* Reset last branches after flush the trace */
1968 if (etm->synth_opts.last_branch)
1980 * It has no new packet coming and 'etmq->packet' contains the stale
1981 * packet which was set at the previous time with packets swapping;
1982 * so skip to generate branch sample to avoid stale packet.
1985 * event for the branches left in the circular buffer at the end of
1988 if (etmq->etm->synth_opts.last_branch &&
1989 etmq->etm->synth_opts.instructions &&
1990 tidq->prev_packet->sample_type == CS_ETM_RANGE) {
1993 /* Prepare last branches for instruction sample */
2000 addr = cs_etm__last_executed_instr(tidq->prev_packet);
2004 tidq->period_instructions);
2008 tidq->period_instructions = 0;
2024 if (!etmq->buf_len) {
2030 * are contiguous, reset the decoder to force re-sync.
2032 ret = cs_etm_decoder__reset(etmq->decoder);
2037 return etmq->buf_len;
2049 switch (packet->isa) {
2055 * +-----------------+--------+
2057 * +-----------------+--------+
2063 addr = end_addr - 2;
2075 * +---------+---------+-------------------------+
2077 * +---------+---------+-------------------------+
2079 addr = end_addr - 4;
2092 * +-----------------------+---------+-----------+
2094 * +-----------------------+---------+-----------+
2096 addr = end_addr - 4;
2114 u8 trace_chan_id = tidq->trace_chan_id;
2115 struct cs_etm_packet *packet = tidq->packet;
2116 struct cs_etm_packet *prev_packet = tidq->prev_packet;
2119 if (packet->exception_number == CS_ETMV3_EXC_SVC)
2128 if (packet->exception_number == CS_ETMV4_EXC_CALL &&
2130 prev_packet->end_addr))
2140 struct cs_etm_packet *packet = tidq->packet;
2143 if (packet->exception_number == CS_ETMV3_EXC_DEBUG_HALT ||
2144 packet->exception_number == CS_ETMV3_EXC_ASYNC_DATA_ABORT ||
2145 packet->exception_number == CS_ETMV3_EXC_PE_RESET ||
2146 packet->exception_number == CS_ETMV3_EXC_IRQ ||
2147 packet->exception_number == CS_ETMV3_EXC_FIQ)
2151 if (packet->exception_number == CS_ETMV4_EXC_RESET ||
2152 packet->exception_number == CS_ETMV4_EXC_DEBUG_HALT ||
2153 packet->exception_number == CS_ETMV4_EXC_SYSTEM_ERROR ||
2154 packet->exception_number == CS_ETMV4_EXC_INST_DEBUG ||
2155 packet->exception_number == CS_ETMV4_EXC_DATA_DEBUG ||
2156 packet->exception_number == CS_ETMV4_EXC_IRQ ||
2157 packet->exception_number == CS_ETMV4_EXC_FIQ)
2167 u8 trace_chan_id = tidq->trace_chan_id;
2168 struct cs_etm_packet *packet = tidq->packet;
2169 struct cs_etm_packet *prev_packet = tidq->prev_packet;
2172 if (packet->exception_number == CS_ETMV3_EXC_SMC ||
2173 packet->exception_number == CS_ETMV3_EXC_HYP ||
2174 packet->exception_number == CS_ETMV3_EXC_JAZELLE_THUMBEE ||
2175 packet->exception_number == CS_ETMV3_EXC_UNDEFINED_INSTR ||
2176 packet->exception_number == CS_ETMV3_EXC_PREFETCH_ABORT ||
2177 packet->exception_number == CS_ETMV3_EXC_DATA_FAULT ||
2178 packet->exception_number == CS_ETMV3_EXC_GENERIC)
2182 if (packet->exception_number == CS_ETMV4_EXC_TRAP ||
2183 packet->exception_number == CS_ETMV4_EXC_ALIGNMENT ||
2184 packet->exception_number == CS_ETMV4_EXC_INST_FAULT ||
2185 packet->exception_number == CS_ETMV4_EXC_DATA_FAULT)
2192 if (packet->exception_number == CS_ETMV4_EXC_CALL &&
2194 prev_packet->end_addr))
2204 if (packet->exception_number > CS_ETMV4_EXC_FIQ &&
2205 packet->exception_number <= CS_ETMV4_EXC_END)
2215 struct cs_etm_packet *packet = tidq->packet;
2216 struct cs_etm_packet *prev_packet = tidq->prev_packet;
2217 u8 trace_chan_id = tidq->trace_chan_id;
2221 switch (packet->sample_type) {
2228 if (packet->last_instr_type == OCSD_INSTR_BR &&
2229 packet->last_instr_subtype == OCSD_S_INSTR_NONE) {
2230 packet->flags = PERF_IP_FLAG_BRANCH;
2232 if (packet->last_instr_cond)
2233 packet->flags |= PERF_IP_FLAG_CONDITIONAL;
2240 if (packet->last_instr_type == OCSD_INSTR_BR &&
2241 packet->last_instr_subtype == OCSD_S_INSTR_BR_LINK)
2242 packet->flags = PERF_IP_FLAG_BRANCH |
2249 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT &&
2250 packet->last_instr_subtype == OCSD_S_INSTR_BR_LINK)
2251 packet->flags = PERF_IP_FLAG_BRANCH |
2259 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT &&
2260 packet->last_instr_subtype == OCSD_S_INSTR_V7_IMPLIED_RET)
2261 packet->flags = PERF_IP_FLAG_BRANCH |
2269 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT &&
2270 packet->last_instr_subtype == OCSD_S_INSTR_NONE)
2271 packet->flags = PERF_IP_FLAG_BRANCH |
2275 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT &&
2276 packet->last_instr_subtype == OCSD_S_INSTR_V8_RET)
2277 packet->flags = PERF_IP_FLAG_BRANCH |
2285 if (prev_packet->sample_type == CS_ETM_DISCONTINUITY)
2286 prev_packet->flags |= PERF_IP_FLAG_BRANCH |
2292 * it needs to calibrate the previous packet sample flags
2295 if (prev_packet->flags == (PERF_IP_FLAG_BRANCH |
2299 packet, packet->start_addr))
2300 prev_packet->flags = PERF_IP_FLAG_BRANCH |
2310 if (prev_packet->sample_type == CS_ETM_RANGE)
2311 prev_packet->flags |= PERF_IP_FLAG_BRANCH |
2315 ret = cs_etm__get_magic(etmq, packet->trace_chan_id, &magic);
2321 packet->flags = PERF_IP_FLAG_BRANCH |
2326 * interrupt controller, debug module, PE reset or halt.
2329 packet->flags = PERF_IP_FLAG_BRANCH |
2338 packet->flags = PERF_IP_FLAG_BRANCH |
2349 if (prev_packet->sample_type == CS_ETM_RANGE)
2350 prev_packet->flags = packet->flags;
2375 * system call instruction and then calibrate the sample flag
2378 if (prev_packet->sample_type == CS_ETM_RANGE)
2379 prev_packet->flags = PERF_IP_FLAG_BRANCH |
2403 ret = cs_etm_decoder__process_data_block(etmq->decoder,
2404 etmq->offset,
2405 &etmq->buf[etmq->buf_used],
2406 etmq->buf_len,
2411 etmq->offset += processed;
2412 etmq->buf_used += processed;
2413 etmq->buf_len -= processed;
2425 packet_queue = &tidq->packet_queue;
2430 tidq->packet);
2441 * thus setting sample flags must be called
2449 switch (tidq->packet->sample_type) {
2480 return -EINVAL;
2494 struct intlist *traceid_queues_list = etmq->traceid_queues_list;
2497 idx = (int)(intptr_t)inode->priv;
2498 tidq = etmq->traceid_queues[idx];
2512 return -EINVAL;
2533 } while (etmq->buf_len);
2564 * buffer. But here in per-cpu mode we need to iterate
2568 etmq->traceid_queues_list) {
2569 idx = (int)(intptr_t)inode->priv;
2570 tidq = etmq->traceid_queues[idx];
2573 } while (etmq->buf_len);
2575 intlist__for_each_entry(inode, etmq->traceid_queues_list) {
2576 idx = (int)(intptr_t)inode->priv;
2577 tidq = etmq->traceid_queues[idx];
2592 struct auxtrace_queues *queues = &etm->queues;
2594 for (i = 0; i < queues->nr_queues; i++) {
2595 struct auxtrace_queue *queue = &etm->queues.queue_array[i];
2596 struct cs_etm_queue *etmq = queue->priv;
2602 if (etm->per_thread_decoding) {
2609 if (tid == -1 || thread__tid(tidq->thread) == tid)
2629 * Pre-populate the heap with one entry from each queue so that we can
2632 for (i = 0; i < etm->queues.nr_queues; i++) {
2633 etmq = etm->queues.queue_array[i].priv;
2643 if (!etm->heap.heap_cnt)
2646 /* Take the entry at the top of the min heap */
2647 cs_queue_nr = etm->heap.heap_array[0].queue_nr;
2650 queue = &etm->queues.queue_array[queue_nr];
2651 etmq = queue->priv;
2657 auxtrace_heap__pop(&etm->heap);
2666 ret = -EINVAL;
2724 ret = auxtrace_heap__add(&etm->heap, cs_queue_nr, cs_timestamp);
2727 for (i = 0; i < etm->queues.nr_queues; i++) {
2730 etmq = etm->queues.queue_array[i].priv;
2734 intlist__for_each_entry(inode, etmq->traceid_queues_list) {
2735 int idx = (int)(intptr_t)inode->priv;
2738 tidq = etmq->traceid_queues[idx];
2753 if (etm->timeless_decoding)
2762 th = machine__findnew_thread(&etm->session->machines.host,
2763 event->itrace_start.pid,
2764 event->itrace_start.tid);
2766 return -ENOMEM;
2777 bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT;
2780 * Context switch in per-thread mode are irrelevant since perf
2783 if (etm->timeless_decoding)
2800 th = machine__findnew_thread(&etm->session->machines.host,
2801 event->context_switch.next_prev_pid,
2802 event->context_switch.next_prev_tid);
2804 return -ENOMEM;
2813 struct perf_sample *sample,
2816 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
2823 if (!tool->ordered_events) {
2825 return -EINVAL;
2828 switch (event->header.type) {
2831 * Don't need to wait for cs_etm__flush_events() in per-thread mode to
2838 if (etm->per_thread_decoding && etm->timeless_decoding)
2840 event->fork.tid);
2855 if (sample->time && (sample->time != (u64)-1))
2856 etm->latest_kernel_timestamp = sample->time;
2876 for (i = 0; i < etm->queues.nr_queues; ++i)
2877 list_for_each_entry(buf, &etm->queues.queue_array[i].head, list)
2878 if (buf->reference == event->reference)
2879 cs_etm__dump_event(etm->queues.queue_array[i].priv, buf);
2886 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
2889 if (!etm->data_queued) {
2892 int fd = perf_data__fd(session->data);
2893 bool is_pipe = perf_data__is_pipe(session->data);
2895 int idx = event->auxtrace.idx;
2901 if (data_offset == -1)
2902 return -errno;
2905 err = auxtrace_queues__add_event(&etm->queues, session,
2912 cs_etm__dump_event(etm->queues.queue_array[idx].priv, buffer);
2916 dump_queued_data(etm, &event->auxtrace);
2924 struct evlist *evlist = etm->session->evlist;
2926 /* Override timeless mode with user input from --itrace=Z */
2927 if (etm->synth_opts.timeless_decoding) {
2928 etm->timeless_decoding = true;
2933 * Find the cs_etm evsel and look at what its timestamp setting was
2936 if (cs_etm__evsel_is_auxtrace(etm->session, evsel)) {
2937 etm->timeless_decoding =
2938 !(evsel->core.attr.config & BIT(ETM_OPT_TS));
2943 return -EINVAL;
2978 /* remaining block params at offset +1 from source */
2979 for (k = CS_ETM_COMMON_BLK_MAX_V1 - 1; k < nr_in_params; k++)
2984 /* read version 1 info block - input and output nr_params may differ */
2989 /* if input has more params than output - skip excess */
2998 metadata[CS_ETM_NR_TRC_PARAMS] = nr_out_params - nr_cmn_params;
3009 * on the bounds of aux_event, if it matches with the buffer that's at
3013 * want to reset the decoder for every PERF_RECORD_AUX event, and the decoder
3014 * is reset across each buffer, so splitting the buffers up in advance has
3018 struct perf_record_aux *aux_event, struct perf_sample *sample)
3028 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
3033 * There should be a PERF_RECORD_AUXTRACE event at the file_offset that we got
3040 auxtrace_event = &auxtrace_event_union->auxtrace;
3041 if (auxtrace_event->header.type != PERF_RECORD_AUXTRACE)
3042 return -EINVAL;
3044 if (auxtrace_event->header.size < sizeof(struct perf_record_auxtrace) ||
3045 auxtrace_event->header.size != sz) {
3046 return -EINVAL;
3050 * In per-thread mode, auxtrace CPU is set to -1, but TID will be set instead. See
3051 * auxtrace_mmap_params__set_idx(). However, the sample AUX event will contain a
3053 * So now compare only TIDs if auxtrace CPU is -1, and CPUs if auxtrace CPU is not -1.
3056 if (auxtrace_event->cpu == (__u32) -1) {
3057 etm->per_thread_decoding = true;
3058 if (auxtrace_event->tid != sample->tid)
3060 } else if (auxtrace_event->cpu != sample->cpu) {
3061 if (etm->per_thread_decoding) {
3063 * Found a per-cpu buffer after a per-thread one was
3066 pr_err("CS ETM: Inconsistent per-thread/per-cpu mode.\n");
3067 return -EINVAL;
3072 if (aux_event->flags & PERF_AUX_FLAG_OVERWRITE) {
3078 aux_size = min(aux_event->aux_size, auxtrace_event->size);
3084 aux_offset = aux_event->aux_offset - aux_size;
3086 aux_size = aux_event->aux_size;
3087 aux_offset = aux_event->aux_offset;
3090 if (aux_offset >= auxtrace_event->offset &&
3091 aux_offset + aux_size <= auxtrace_event->offset + auxtrace_event->size) {
3092 struct cs_etm_queue *etmq = etm->queues.queue_array[auxtrace_event->idx].priv;
3101 file_offset += aux_offset - auxtrace_event->offset + auxtrace_event->header.size;
3104 " tid: %d cpu: %d\n", aux_size, aux_offset, sample->tid, sample->cpu);
3105 err = auxtrace_queues__add_event(&etm->queues, session, &auxtrace_fragment,
3110 format = (aux_event->flags & PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW) ?
3112 if (etmq->format != UNSET && format != etmq->format) {
3114 return -EINVAL;
3116 etmq->format = format;
3128 if (event->header.type == PERF_RECORD_AUX_OUTPUT_HW_ID) {
3138 struct perf_sample sample;
3146 if (event->header.type != PERF_RECORD_AUX)
3149 if (event->header.size < sizeof(struct perf_record_aux))
3150 return -EINVAL;
3153 if (!event->aux.aux_size)
3157 * Parse the sample, we need the sample_id_all data that comes after the event so that the
3160 evsel = evlist__event2evsel(session->evlist, event);
3162 return -EINVAL;
3163 perf_sample__init(&sample, /*all=*/false);
3164 ret = evsel__parse_sample(evsel, event, &sample);
3171 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
3172 for (i = 0; i < auxtrace_index->nr; i++) {
3173 ent = &auxtrace_index->entries[i];
3174 ret = cs_etm__queue_aux_fragment(session, ent->file_offset,
3175 ent->sz, &event->aux, &sample);
3190 " tid: %d cpu: %d\n", event->aux.aux_offset, sample.tid, sample.cpu);
3193 perf_sample__exit(&sample);
3199 struct auxtrace_index *index = list_first_entry_or_null(&session->auxtrace_index,
3201 if (index && index->nr > 0)
3202 return perf_session__peek_events(session, session->header.data_offset,
3203 session->header.data_size,
3208 * buffers or no index at all). Fail silently as there is the possibility of
3209 * queueing them in cs_etm__process_auxtrace_event() if etm->data_queued is still
3218 (CS_##type##_##param - CS_ETM_COMMON_BLK_MAX_V1))
3221 * Loop through the ETMs and complain if we find at least one where ts_source != 1 (virtual
3268 return -EINVAL;
3285 int decoders = intlist__nr_entries(etmq->traceid_list);
3294 if (etmq->format == UNFORMATTED)
3312 etmq->decoder = cs_etm_decoder__new(decoders, &d_params,
3315 if (!etmq->decoder)
3322 if (cs_etm_decoder__add_mem_access_cb(etmq->decoder,
3323 0x0L, ((u64) -1L),
3331 cs_etm_decoder__free(etmq->decoder);
3334 return -EINVAL;
3339 struct auxtrace_queues *queues = &etm->queues;
3341 for (unsigned int i = 0; i < queues->nr_queues; i++) {
3342 bool empty = list_empty(&queues->queue_array[i].head);
3343 struct cs_etm_queue *etmq = queues->queue_array[i].priv;
3348 * etmq->format is unknown for empty queues.
3350 assert(empty || etmq->format != UNSET);
3364 struct perf_record_auxtrace_info *auxtrace_info = &event->auxtrace_info;
3366 struct perf_record_time_conv *tc = &session->time_conv;
3368 int total_size = auxtrace_info->header.size;
3378 ptr = (u64 *) auxtrace_info->priv;
3382 return -ENOMEM;
3405 metadata[j] = cs_etm__create_meta_blk(ptr, &i, CS_ETE_PRIV_MAX, -1);
3409 err = -EINVAL;
3414 err = -ENOMEM;
3429 priv_size = total_size - event_header_size - INFO_HEADER_SIZE;
3431 err = -EINVAL;
3438 err = -ENOMEM;
3443 * As all the ETMs run at the same exception level, the system should
3447 etm->pid_fmt = cs_etm__init_pid_fmt(metadata[0]);
3449 err = auxtrace_queues__init_nr(&etm->queues, max_cpu + 1);
3453 for (unsigned int j = 0; j < etm->queues.nr_queues; ++j) {
3454 err = cs_etm__setup_queue(etm, &etm->queues.queue_array[j], j);
3459 if (session->itrace_synth_opts->set) {
3460 etm->synth_opts = *session->itrace_synth_opts;
3462 itrace_synth_opts__set_default(&etm->synth_opts,
3463 session->itrace_synth_opts->default_no_sample);
3464 etm->synth_opts.callchain = false;
3467 etm->session = session;
3469 etm->num_cpu = num_cpu;
3470 etm->pmu_type = (unsigned int) ((ptr[CS_PMU_TYPE_CPUS] >> 32) & 0xffffffff);
3471 etm->snapshot_mode = (ptr[CS_ETM_SNAPSHOT] != 0);
3472 etm->metadata = metadata;
3473 etm->auxtrace_type = auxtrace_info->type;
3475 if (etm->synth_opts.use_timestamp)
3485 etm->has_virtual_ts = true;
3488 etm->has_virtual_ts = cs_etm__has_virtual_ts(metadata, num_cpu);
3490 if (!etm->has_virtual_ts)
3497 etm->auxtrace.process_event = cs_etm__process_event;
3498 etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event;
3499 etm->auxtrace.flush_events = cs_etm__flush_events;
3500 etm->auxtrace.free_events = cs_etm__free_events;
3501 etm->auxtrace.free = cs_etm__free;
3502 etm->auxtrace.evsel_is_auxtrace = cs_etm__evsel_is_auxtrace;
3503 session->auxtrace = &etm->auxtrace;
3509 etm->tc.time_shift = tc->time_shift;
3510 etm->tc.time_mult = tc->time_mult;
3511 etm->tc.time_zero = tc->time_zero;
3513 etm->tc.time_cycles = tc->time_cycles;
3514 etm->tc.time_mask = tc->time_mask;
3515 etm->tc.cap_user_time_zero = tc->cap_user_time_zero;
3516 etm->tc.cap_user_time_short = tc->cap_user_time_short;
3542 * If no AUX_HW_ID packets are present - which means a file recorded on an old kernel
3549 err = perf_session__peek_events(session, session->header.data_offset,
3550 session->header.data_size,
3566 etm->data_queued = etm->queues.populated;
3570 auxtrace_queues__free(&etm->queues);
3571 session->auxtrace = NULL;