projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
perf cs-etm: Correct CPU mode for samples
[linux-2.6-block.git] / tools / perf / util / cs-etm.c
CommitLineData
8a9fd832 1// SPDX-License-Identifier: GPL-2.0
440a23b3 2/*
440a23b3
MP		 3 * Copyright(C) 2015-2018 Linaro Limited.
4 *
5 * Author: Tor Jeremiassen <tor@ti.com>
6 * Author: Mathieu Poirier <mathieu.poirier@linaro.org>
7 */
8
9#include <linux/bitops.h>
10#include <linux/err.h>
11#include <linux/kernel.h>
12#include <linux/log2.h>
13#include <linux/types.h>
14
15#include <stdlib.h>
16
17#include "auxtrace.h"
18#include "color.h"
19#include "cs-etm.h"
68ffe390 20#include "cs-etm-decoder/cs-etm-decoder.h"
440a23b3
MP		 21#include "debug.h"
22#include "evlist.h"
23#include "intlist.h"
24#include "machine.h"
25#include "map.h"
26#include "perf.h"
27#include "thread.h"
28#include "thread_map.h"
29#include "thread-stack.h"
30#include "util.h"
31
32#define MAX_TIMESTAMP (~0ULL)
33
e573e978
RW		 34/*
35 * A64 instructions are always 4 bytes
36 *
37 * Only A64 is supported, so can use this constant for converting between
38 * addresses and instruction counts, calculting offsets etc
39 */
40#define A64_INSTR_SIZE 4
41
440a23b3
MP		 42struct cs_etm_auxtrace {
43 struct auxtrace auxtrace;
44 struct auxtrace_queues queues;
45 struct auxtrace_heap heap;
46 struct itrace_synth_opts synth_opts;
47 struct perf_session *session;
48 struct machine *machine;
49 struct thread *unknown_thread;
50
51 u8 timeless_decoding;
52 u8 snapshot_mode;
53 u8 data_queued;
54 u8 sample_branches;
e573e978 55 u8 sample_instructions;
440a23b3
MP		 56
57 int num_cpu;
58 u32 auxtrace_type;
59 u64 branches_sample_type;
60 u64 branches_id;
e573e978
RW		 61 u64 instructions_sample_type;
62 u64 instructions_sample_period;
63 u64 instructions_id;
440a23b3
MP		 64 u64 **metadata;
65 u64 kernel_start;
66 unsigned int pmu_type;
67};
68
69struct cs_etm_queue {
70 struct cs_etm_auxtrace *etm;
71 struct thread *thread;
72 struct cs_etm_decoder *decoder;
73 struct auxtrace_buffer *buffer;
74 const struct cs_etm_state *state;
75 union perf_event *event_buf;
76 unsigned int queue_nr;
77 pid_t pid, tid;
78 int cpu;
79 u64 time;
80 u64 timestamp;
81 u64 offset;
e573e978
RW		 82 u64 period_instructions;
83 struct branch_stack *last_branch;
84 struct branch_stack *last_branch_rb;
85 size_t last_branch_pos;
86 struct cs_etm_packet *prev_packet;
87 struct cs_etm_packet *packet;
440a23b3
MP		 88};
89
9f878b29
MP		 90static int cs_etm__update_queues(struct cs_etm_auxtrace *etm);
91static int cs_etm__process_timeless_queues(struct cs_etm_auxtrace *etm,
92 pid_t tid, u64 time_);
93
68ffe390
MP		 94static void cs_etm__packet_dump(const char *pkt_string)
95{
96 const char *color = PERF_COLOR_BLUE;
97 int len = strlen(pkt_string);
98
99 if (len && (pkt_string[len-1] == '\n'))
100 color_fprintf(stdout, color, " %s", pkt_string);
101 else
102 color_fprintf(stdout, color, " %s\n", pkt_string);
103
104 fflush(stdout);
105}
106
107static void cs_etm__dump_event(struct cs_etm_auxtrace *etm,
108 struct auxtrace_buffer *buffer)
109{
110 int i, ret;
111 const char *color = PERF_COLOR_BLUE;
112 struct cs_etm_decoder_params d_params;
113 struct cs_etm_trace_params *t_params;
114 struct cs_etm_decoder *decoder;
115 size_t buffer_used = 0;
116
117 fprintf(stdout, "\n");
118 color_fprintf(stdout, color,
119 ". ... CoreSight ETM Trace data: size %zu bytes\n",
120 buffer->size);
121
122 /* Use metadata to fill in trace parameters for trace decoder */
123 t_params = zalloc(sizeof(*t_params) * etm->num_cpu);
124 for (i = 0; i < etm->num_cpu; i++) {
125 t_params[i].protocol = CS_ETM_PROTO_ETMV4i;
126 t_params[i].etmv4.reg_idr0 = etm->metadata[i][CS_ETMV4_TRCIDR0];
127 t_params[i].etmv4.reg_idr1 = etm->metadata[i][CS_ETMV4_TRCIDR1];
128 t_params[i].etmv4.reg_idr2 = etm->metadata[i][CS_ETMV4_TRCIDR2];
129 t_params[i].etmv4.reg_idr8 = etm->metadata[i][CS_ETMV4_TRCIDR8];
130 t_params[i].etmv4.reg_configr =
131 etm->metadata[i][CS_ETMV4_TRCCONFIGR];
132 t_params[i].etmv4.reg_traceidr =
133 etm->metadata[i][CS_ETMV4_TRCTRACEIDR];
134 }
135
136 /* Set decoder parameters to simply print the trace packets */
137 d_params.packet_printer = cs_etm__packet_dump;
138 d_params.operation = CS_ETM_OPERATION_PRINT;
139 d_params.formatted = true;
140 d_params.fsyncs = false;
141 d_params.hsyncs = false;
142 d_params.frame_aligned = true;
143
144 decoder = cs_etm_decoder__new(etm->num_cpu, &d_params, t_params);
145
146 zfree(&t_params);
147
148 if (!decoder)
149 return;
150 do {
151 size_t consumed;
152
153 ret = cs_etm_decoder__process_data_block(
154 decoder, buffer->offset,
155 &((u8 *)buffer->data)[buffer_used],
156 buffer->size - buffer_used, &consumed);
157 if (ret)
158 break;
159
160 buffer_used += consumed;
161 } while (buffer_used < buffer->size);
162
163 cs_etm_decoder__free(decoder);
164}
165
440a23b3
MP		 166static int cs_etm__flush_events(struct perf_session *session,
167 struct perf_tool *tool)
168{
9f878b29
MP		 169 int ret;
170 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
171 struct cs_etm_auxtrace,
172 auxtrace);
173 if (dump_trace)
174 return 0;
175
176 if (!tool->ordered_events)
177 return -EINVAL;
178
179 if (!etm->timeless_decoding)
180 return -EINVAL;
181
182 ret = cs_etm__update_queues(etm);
183
184 if (ret < 0)
185 return ret;
186
187 return cs_etm__process_timeless_queues(etm, -1, MAX_TIMESTAMP - 1);
440a23b3
MP		 188}
189
190static void cs_etm__free_queue(void *priv)
191{
192 struct cs_etm_queue *etmq = priv;
193
099c1130
MP		 194 if (!etmq)
195 return;
196
197 thread__zput(etmq->thread);
198 cs_etm_decoder__free(etmq->decoder);
199 zfree(&etmq->event_buf);
e573e978
RW		 200 zfree(&etmq->last_branch);
201 zfree(&etmq->last_branch_rb);
202 zfree(&etmq->prev_packet);
203 zfree(&etmq->packet);
440a23b3
MP		 204 free(etmq);
205}
206
207static void cs_etm__free_events(struct perf_session *session)
208{
209 unsigned int i;
210 struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
211 struct cs_etm_auxtrace,
212 auxtrace);
213 struct auxtrace_queues *queues = &aux->queues;
214
215 for (i = 0; i < queues->nr_queues; i++) {
216 cs_etm__free_queue(queues->queue_array[i].priv);
217 queues->queue_array[i].priv = NULL;
218 }
219
220 auxtrace_queues__free(queues);
221}
222
223static void cs_etm__free(struct perf_session *session)
224{
cd8bfd8c
TJ		 225 int i;
226 struct int_node *inode, *tmp;
440a23b3
MP		 227 struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
228 struct cs_etm_auxtrace,
229 auxtrace);
230 cs_etm__free_events(session);
231 session->auxtrace = NULL;
232
cd8bfd8c
TJ		 233 /* First remove all traceID/CPU# nodes for the RB tree */
234 intlist__for_each_entry_safe(inode, tmp, traceid_list)
235 intlist__remove(traceid_list, inode);
236 /* Then the RB tree itself */
237 intlist__delete(traceid_list);
238
239 for (i = 0; i < aux->num_cpu; i++)
240 zfree(&aux->metadata[i]);
241
46d53620 242 thread__zput(aux->unknown_thread);
cd8bfd8c 243 zfree(&aux->metadata);
440a23b3
MP		 244 zfree(&aux);
245}
246
d6c9c05f
LY		 247static u8 cs_etm__cpu_mode(struct cs_etm_queue *etmq, u64 address)
248{
249 struct machine *machine;
250
251 machine = etmq->etm->machine;
252
253 if (address >= etmq->etm->kernel_start) {
254 if (machine__is_host(machine))
255 return PERF_RECORD_MISC_KERNEL;
256 else
257 return PERF_RECORD_MISC_GUEST_KERNEL;
258 } else {
259 if (machine__is_host(machine))
260 return PERF_RECORD_MISC_USER;
261 else if (perf_guest)
262 return PERF_RECORD_MISC_GUEST_USER;
263 else
264 return PERF_RECORD_MISC_HYPERVISOR;
265 }
266}
267
20d9c478
MP		 268static u32 cs_etm__mem_access(struct cs_etm_queue *etmq, u64 address,
269 size_t size, u8 *buffer)
270{
271 u8 cpumode;
272 u64 offset;
273 int len;
274 struct thread *thread;
275 struct machine *machine;
276 struct addr_location al;
277
278 if (!etmq)
279 return -1;
280
281 machine = etmq->etm->machine;
d6c9c05f 282 cpumode = cs_etm__cpu_mode(etmq, address);
20d9c478
MP		 283
284 thread = etmq->thread;
285 if (!thread) {
286 if (cpumode != PERF_RECORD_MISC_KERNEL)
287 return -EINVAL;
288 thread = etmq->etm->unknown_thread;
289 }
290
71a84b5a 291 if (!thread__find_map(thread, cpumode, address, &al) || !al.map->dso)
20d9c478
MP		 292 return 0;
293
294 if (al.map->dso->data.status == DSO_DATA_STATUS_ERROR &&
295 dso__data_status_seen(al.map->dso, DSO_DATA_STATUS_SEEN_ITRACE))
296 return 0;
297
298 offset = al.map->map_ip(al.map, address);
299
300 map__load(al.map);
301
302 len = dso__data_read_offset(al.map->dso, machine, offset, buffer, size);
303
304 if (len <= 0)
305 return 0;
306
307 return len;
308}
309
310static struct cs_etm_queue *cs_etm__alloc_queue(struct cs_etm_auxtrace *etm,
311 unsigned int queue_nr)
312{
313 int i;
314 struct cs_etm_decoder_params d_params;
315 struct cs_etm_trace_params *t_params;
316 struct cs_etm_queue *etmq;
e573e978 317 size_t szp = sizeof(struct cs_etm_packet);
20d9c478
MP		 318
319 etmq = zalloc(sizeof(*etmq));
320 if (!etmq)
321 return NULL;
322
e573e978
RW		 323 etmq->packet = zalloc(szp);
324 if (!etmq->packet)
325 goto out_free;
326
327 if (etm->synth_opts.last_branch || etm->sample_branches) {
328 etmq->prev_packet = zalloc(szp);
329 if (!etmq->prev_packet)
330 goto out_free;
331 }
332
333 if (etm->synth_opts.last_branch) {
334 size_t sz = sizeof(struct branch_stack);
335
336 sz += etm->synth_opts.last_branch_sz *
337 sizeof(struct branch_entry);
338 etmq->last_branch = zalloc(sz);
339 if (!etmq->last_branch)
340 goto out_free;
341 etmq->last_branch_rb = zalloc(sz);
342 if (!etmq->last_branch_rb)
343 goto out_free;
344 }
345
20d9c478
MP		 346 etmq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE);
347 if (!etmq->event_buf)
348 goto out_free;
349
350 etmq->etm = etm;
351 etmq->queue_nr = queue_nr;
352 etmq->pid = -1;
353 etmq->tid = -1;
354 etmq->cpu = -1;
355
356 /* Use metadata to fill in trace parameters for trace decoder */
357 t_params = zalloc(sizeof(*t_params) * etm->num_cpu);
358
359 if (!t_params)
360 goto out_free;
361
362 for (i = 0; i < etm->num_cpu; i++) {
363 t_params[i].protocol = CS_ETM_PROTO_ETMV4i;
364 t_params[i].etmv4.reg_idr0 = etm->metadata[i][CS_ETMV4_TRCIDR0];
365 t_params[i].etmv4.reg_idr1 = etm->metadata[i][CS_ETMV4_TRCIDR1];
366 t_params[i].etmv4.reg_idr2 = etm->metadata[i][CS_ETMV4_TRCIDR2];
367 t_params[i].etmv4.reg_idr8 = etm->metadata[i][CS_ETMV4_TRCIDR8];
368 t_params[i].etmv4.reg_configr =
369 etm->metadata[i][CS_ETMV4_TRCCONFIGR];
370 t_params[i].etmv4.reg_traceidr =
371 etm->metadata[i][CS_ETMV4_TRCTRACEIDR];
372 }
373
374 /* Set decoder parameters to simply print the trace packets */
375 d_params.packet_printer = cs_etm__packet_dump;
376 d_params.operation = CS_ETM_OPERATION_DECODE;
377 d_params.formatted = true;
378 d_params.fsyncs = false;
379 d_params.hsyncs = false;
380 d_params.frame_aligned = true;
381 d_params.data = etmq;
382
383 etmq->decoder = cs_etm_decoder__new(etm->num_cpu, &d_params, t_params);
384
385 zfree(&t_params);
386
387 if (!etmq->decoder)
388 goto out_free;
389
390 /*
391 * Register a function to handle all memory accesses required by
392 * the trace decoder library.
393 */
394 if (cs_etm_decoder__add_mem_access_cb(etmq->decoder,
395 0x0L, ((u64) -1L),
396 cs_etm__mem_access))
397 goto out_free_decoder;
398
399 etmq->offset = 0;
e573e978 400 etmq->period_instructions = 0;
20d9c478
MP		 401
402 return etmq;
403
404out_free_decoder:
405 cs_etm_decoder__free(etmq->decoder);
406out_free:
407 zfree(&etmq->event_buf);
e573e978
RW		 408 zfree(&etmq->last_branch);
409 zfree(&etmq->last_branch_rb);
410 zfree(&etmq->prev_packet);
411 zfree(&etmq->packet);
20d9c478
MP		 412 free(etmq);
413
414 return NULL;
415}
416
417static int cs_etm__setup_queue(struct cs_etm_auxtrace *etm,
418 struct auxtrace_queue *queue,
419 unsigned int queue_nr)
420{
421 struct cs_etm_queue *etmq = queue->priv;
422
423 if (list_empty(&queue->head) || etmq)
424 return 0;
425
426 etmq = cs_etm__alloc_queue(etm, queue_nr);
427
428 if (!etmq)
429 return -ENOMEM;
430
431 queue->priv = etmq;
432
433 if (queue->cpu != -1)
434 etmq->cpu = queue->cpu;
435
436 etmq->tid = queue->tid;
437
438 return 0;
439}
440
441static int cs_etm__setup_queues(struct cs_etm_auxtrace *etm)
442{
443 unsigned int i;
444 int ret;
445
446 for (i = 0; i < etm->queues.nr_queues; i++) {
447 ret = cs_etm__setup_queue(etm, &etm->queues.queue_array[i], i);
448 if (ret)
449 return ret;
450 }
451
452 return 0;
453}
454
455static int cs_etm__update_queues(struct cs_etm_auxtrace *etm)
456{
457 if (etm->queues.new_data) {
458 etm->queues.new_data = false;
459 return cs_etm__setup_queues(etm);
460 }
461
462 return 0;
463}
464
e573e978
RW		 465static inline void cs_etm__copy_last_branch_rb(struct cs_etm_queue *etmq)
466{
467 struct branch_stack *bs_src = etmq->last_branch_rb;
468 struct branch_stack *bs_dst = etmq->last_branch;
469 size_t nr = 0;
470
471 /*
472 * Set the number of records before early exit: ->nr is used to
473 * determine how many branches to copy from ->entries.
474 */
475 bs_dst->nr = bs_src->nr;
476
477 /*
478 * Early exit when there is nothing to copy.
479 */
480 if (!bs_src->nr)
481 return;
482
483 /*
484 * As bs_src->entries is a circular buffer, we need to copy from it in
485 * two steps. First, copy the branches from the most recently inserted
486 * branch ->last_branch_pos until the end of bs_src->entries buffer.
487 */
488 nr = etmq->etm->synth_opts.last_branch_sz - etmq->last_branch_pos;
489 memcpy(&bs_dst->entries[0],
490 &bs_src->entries[etmq->last_branch_pos],
491 sizeof(struct branch_entry) * nr);
492
493 /*
494 * If we wrapped around at least once, the branches from the beginning
495 * of the bs_src->entries buffer and until the ->last_branch_pos element
496 * are older valid branches: copy them over. The total number of
497 * branches copied over will be equal to the number of branches asked by
498 * the user in last_branch_sz.
499 */
500 if (bs_src->nr >= etmq->etm->synth_opts.last_branch_sz) {
501 memcpy(&bs_dst->entries[nr],
502 &bs_src->entries[0],
503 sizeof(struct branch_entry) * etmq->last_branch_pos);
504 }
505}
506
507static inline void cs_etm__reset_last_branch_rb(struct cs_etm_queue *etmq)
508{
509 etmq->last_branch_pos = 0;
510 etmq->last_branch_rb->nr = 0;
511}
512
513static inline u64 cs_etm__last_executed_instr(struct cs_etm_packet *packet)
514{
6035b680
LY		 515 /* Returns 0 for the CS_ETM_TRACE_ON packet */
516 if (packet->sample_type == CS_ETM_TRACE_ON)
517 return 0;
518
e573e978
RW		 519 /*
520 * The packet records the execution range with an exclusive end address
521 *
522 * A64 instructions are constant size, so the last executed
523 * instruction is A64_INSTR_SIZE before the end address
524 * Will need to do instruction level decode for T32 instructions as
525 * they can be variable size (not yet supported).
526 */
527 return packet->end_addr - A64_INSTR_SIZE;
528}
529
6035b680
LY		 530static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet)
531{
532 /* Returns 0 for the CS_ETM_TRACE_ON packet */
533 if (packet->sample_type == CS_ETM_TRACE_ON)
534 return 0;
535
536 return packet->start_addr;
537}
538
e573e978
RW		 539static inline u64 cs_etm__instr_count(const struct cs_etm_packet *packet)
540{
541 /*
542 * Only A64 instructions are currently supported, so can get
543 * instruction count by dividing.
544 * Will need to do instruction level decode for T32 instructions as
545 * they can be variable size (not yet supported).
546 */
547 return (packet->end_addr - packet->start_addr) / A64_INSTR_SIZE;
548}
549
550static inline u64 cs_etm__instr_addr(const struct cs_etm_packet *packet,
551 u64 offset)
552{
553 /*
554 * Only A64 instructions are currently supported, so can get
555 * instruction address by muliplying.
556 * Will need to do instruction level decode for T32 instructions as
557 * they can be variable size (not yet supported).
558 */
559 return packet->start_addr + offset * A64_INSTR_SIZE;
560}
561
562static void cs_etm__update_last_branch_rb(struct cs_etm_queue *etmq)
563{
564 struct branch_stack *bs = etmq->last_branch_rb;
565 struct branch_entry *be;
566
567 /*
568 * The branches are recorded in a circular buffer in reverse
569 * chronological order: we start recording from the last element of the
570 * buffer down. After writing the first element of the stack, move the
571 * insert position back to the end of the buffer.
572 */
573 if (!etmq->last_branch_pos)
574 etmq->last_branch_pos = etmq->etm->synth_opts.last_branch_sz;
575
576 etmq->last_branch_pos -= 1;
577
578 be = &bs->entries[etmq->last_branch_pos];
579 be->from = cs_etm__last_executed_instr(etmq->prev_packet);
6035b680 580 be->to = cs_etm__first_executed_instr(etmq->packet);
e573e978
RW		 581 /* No support for mispredict */
582 be->flags.mispred = 0;
583 be->flags.predicted = 1;
584
585 /*
586 * Increment bs->nr until reaching the number of last branches asked by
587 * the user on the command line.
588 */
589 if (bs->nr < etmq->etm->synth_opts.last_branch_sz)
590 bs->nr += 1;
591}
592
593static int cs_etm__inject_event(union perf_event *event,
594 struct perf_sample *sample, u64 type)
595{
596 event->header.size = perf_event__sample_event_size(sample, type, 0);
597 return perf_event__synthesize_sample(event, type, 0, sample);
598}
599
600
9f878b29
MP		 601static int
602cs_etm__get_trace(struct cs_etm_buffer *buff, struct cs_etm_queue *etmq)
603{
604 struct auxtrace_buffer *aux_buffer = etmq->buffer;
605 struct auxtrace_buffer *old_buffer = aux_buffer;
606 struct auxtrace_queue *queue;
607
608 queue = &etmq->etm->queues.queue_array[etmq->queue_nr];
609
610 aux_buffer = auxtrace_buffer__next(queue, aux_buffer);
611
612 /* If no more data, drop the previous auxtrace_buffer and return */
613 if (!aux_buffer) {
614 if (old_buffer)
615 auxtrace_buffer__drop_data(old_buffer);
616 buff->len = 0;
617 return 0;
618 }
619
620 etmq->buffer = aux_buffer;
621
622 /* If the aux_buffer doesn't have data associated, try to load it */
623 if (!aux_buffer->data) {
624 /* get the file desc associated with the perf data file */
625 int fd = perf_data__fd(etmq->etm->session->data);
626
627 aux_buffer->data = auxtrace_buffer__get_data(aux_buffer, fd);
628 if (!aux_buffer->data)
629 return -ENOMEM;
630 }
631
632 /* If valid, drop the previous buffer */
633 if (old_buffer)
634 auxtrace_buffer__drop_data(old_buffer);
635
636 buff->offset = aux_buffer->offset;
637 buff->len = aux_buffer->size;
638 buff->buf = aux_buffer->data;
639
640 buff->ref_timestamp = aux_buffer->reference;
641
642 return buff->len;
643}
644
3a088799
LY		 645static void cs_etm__set_pid_tid_cpu(struct cs_etm_auxtrace *etm,
646 struct auxtrace_queue *queue)
9f878b29
MP		 647{
648 struct cs_etm_queue *etmq = queue->priv;
649
650 /* CPU-wide tracing isn't supported yet */
651 if (queue->tid == -1)
652 return;
653
654 if ((!etmq->thread) && (etmq->tid != -1))
655 etmq->thread = machine__find_thread(etm->machine, -1,
656 etmq->tid);
657
658 if (etmq->thread) {
659 etmq->pid = etmq->thread->pid_;
660 if (queue->cpu == -1)
661 etmq->cpu = etmq->thread->cpu;
662 }
663}
664
e573e978
RW		 665static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq,
666 u64 addr, u64 period)
667{
668 int ret = 0;
669 struct cs_etm_auxtrace *etm = etmq->etm;
670 union perf_event *event = etmq->event_buf;
671 struct perf_sample sample = {.ip = 0,};
672
673 event->sample.header.type = PERF_RECORD_SAMPLE;
d6c9c05f 674 event->sample.header.misc = cs_etm__cpu_mode(etmq, addr);
e573e978
RW		 675 event->sample.header.size = sizeof(struct perf_event_header);
676
677 sample.ip = addr;
678 sample.pid = etmq->pid;
679 sample.tid = etmq->tid;
680 sample.id = etmq->etm->instructions_id;
681 sample.stream_id = etmq->etm->instructions_id;
682 sample.period = period;
683 sample.cpu = etmq->packet->cpu;
684 sample.flags = 0;
685 sample.insn_len = 1;
d6c9c05f 686 sample.cpumode = event->sample.header.misc;
e573e978
RW		 687
688 if (etm->synth_opts.last_branch) {
689 cs_etm__copy_last_branch_rb(etmq);
690 sample.branch_stack = etmq->last_branch;
691 }
692
693 if (etm->synth_opts.inject) {
694 ret = cs_etm__inject_event(event, &sample,
695 etm->instructions_sample_type);
696 if (ret)
697 return ret;
698 }
699
700 ret = perf_session__deliver_synth_event(etm->session, event, &sample);
701
702 if (ret)
703 pr_err(
704 "CS ETM Trace: failed to deliver instruction event, error %d\n",
705 ret);
706
707 if (etm->synth_opts.last_branch)
708 cs_etm__reset_last_branch_rb(etmq);
709
710 return ret;
711}
712
b12235b1
MP		 713/*
714 * The cs etm packet encodes an instruction range between a branch target
715 * and the next taken branch. Generate sample accordingly.
716 */
e573e978 717static int cs_etm__synth_branch_sample(struct cs_etm_queue *etmq)
b12235b1
MP		 718{
719 int ret = 0;
720 struct cs_etm_auxtrace *etm = etmq->etm;
721 struct perf_sample sample = {.ip = 0,};
722 union perf_event *event = etmq->event_buf;
e573e978
RW		 723 struct dummy_branch_stack {
724 u64 nr;
725 struct branch_entry entries;
726 } dummy_bs;
d6c9c05f
LY		 727 u64 ip;
728
729 ip = cs_etm__last_executed_instr(etmq->prev_packet);
b12235b1
MP		 730
731 event->sample.header.type = PERF_RECORD_SAMPLE;
d6c9c05f 732 event->sample.header.misc = cs_etm__cpu_mode(etmq, ip);
b12235b1
MP		 733 event->sample.header.size = sizeof(struct perf_event_header);
734
d6c9c05f 735 sample.ip = ip;
b12235b1
MP		 736 sample.pid = etmq->pid;
737 sample.tid = etmq->tid;
6035b680 738 sample.addr = cs_etm__first_executed_instr(etmq->packet);
b12235b1
MP		 739 sample.id = etmq->etm->branches_id;
740 sample.stream_id = etmq->etm->branches_id;
741 sample.period = 1;
e573e978 742 sample.cpu = etmq->packet->cpu;
b12235b1 743 sample.flags = 0;
d6c9c05f 744 sample.cpumode = event->sample.header.misc;
b12235b1 745
e573e978
RW		 746 /*
747 * perf report cannot handle events without a branch stack
748 */
749 if (etm->synth_opts.last_branch) {
750 dummy_bs = (struct dummy_branch_stack){
751 .nr = 1,
752 .entries = {
753 .from = sample.ip,
754 .to = sample.addr,
755 },
756 };
757 sample.branch_stack = (struct branch_stack *)&dummy_bs;
758 }
759
760 if (etm->synth_opts.inject) {
761 ret = cs_etm__inject_event(event, &sample,
762 etm->branches_sample_type);
763 if (ret)
764 return ret;
765 }
766
b12235b1
MP		 767 ret = perf_session__deliver_synth_event(etm->session, event, &sample);
768
769 if (ret)
770 pr_err(
771 "CS ETM Trace: failed to deliver instruction event, error %d\n",
772 ret);
773
774 return ret;
775}
776
777struct cs_etm_synth {
778 struct perf_tool dummy_tool;
779 struct perf_session *session;
780};
781
782static int cs_etm__event_synth(struct perf_tool *tool,
783 union perf_event *event,
784 struct perf_sample *sample __maybe_unused,
785 struct machine *machine __maybe_unused)
786{
787 struct cs_etm_synth *cs_etm_synth =
788 container_of(tool, struct cs_etm_synth, dummy_tool);
789
790 return perf_session__deliver_synth_event(cs_etm_synth->session,
791 event, NULL);
792}
793
794static int cs_etm__synth_event(struct perf_session *session,
795 struct perf_event_attr *attr, u64 id)
796{
797 struct cs_etm_synth cs_etm_synth;
798
799 memset(&cs_etm_synth, 0, sizeof(struct cs_etm_synth));
800 cs_etm_synth.session = session;
801
802 return perf_event__synthesize_attr(&cs_etm_synth.dummy_tool, attr, 1,
803 &id, cs_etm__event_synth);
804}
805
806static int cs_etm__synth_events(struct cs_etm_auxtrace *etm,
807 struct perf_session *session)
808{
809 struct perf_evlist *evlist = session->evlist;
810 struct perf_evsel *evsel;
811 struct perf_event_attr attr;
812 bool found = false;
813 u64 id;
814 int err;
815
816 evlist__for_each_entry(evlist, evsel) {
817 if (evsel->attr.type == etm->pmu_type) {
818 found = true;
819 break;
820 }
821 }
822
823 if (!found) {
824 pr_debug("No selected events with CoreSight Trace data\n");
825 return 0;
826 }
827
828 memset(&attr, 0, sizeof(struct perf_event_attr));
829 attr.size = sizeof(struct perf_event_attr);
830 attr.type = PERF_TYPE_HARDWARE;
831 attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK;
832 attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
833 PERF_SAMPLE_PERIOD;
834 if (etm->timeless_decoding)
835 attr.sample_type &= ~(u64)PERF_SAMPLE_TIME;
836 else
837 attr.sample_type |= PERF_SAMPLE_TIME;
838
839 attr.exclude_user = evsel->attr.exclude_user;
840 attr.exclude_kernel = evsel->attr.exclude_kernel;
841 attr.exclude_hv = evsel->attr.exclude_hv;
842 attr.exclude_host = evsel->attr.exclude_host;
843 attr.exclude_guest = evsel->attr.exclude_guest;
844 attr.sample_id_all = evsel->attr.sample_id_all;
845 attr.read_format = evsel->attr.read_format;
846
847 /* create new id val to be a fixed offset from evsel id */
848 id = evsel->id[0] + 1000000000;
849
850 if (!id)
851 id = 1;
852
853 if (etm->synth_opts.branches) {
854 attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
855 attr.sample_period = 1;
856 attr.sample_type |= PERF_SAMPLE_ADDR;
857 err = cs_etm__synth_event(session, &attr, id);
858 if (err)
859 return err;
860 etm->sample_branches = true;
861 etm->branches_sample_type = attr.sample_type;
862 etm->branches_id = id;
e573e978
RW		 863 id += 1;
864 attr.sample_type &= ~(u64)PERF_SAMPLE_ADDR;
865 }
866
867 if (etm->synth_opts.last_branch)
868 attr.sample_type |= PERF_SAMPLE_BRANCH_STACK;
869
870 if (etm->synth_opts.instructions) {
871 attr.config = PERF_COUNT_HW_INSTRUCTIONS;
872 attr.sample_period = etm->synth_opts.period;
873 etm->instructions_sample_period = attr.sample_period;
874 err = cs_etm__synth_event(session, &attr, id);
875 if (err)
876 return err;
877 etm->sample_instructions = true;
878 etm->instructions_sample_type = attr.sample_type;
879 etm->instructions_id = id;
880 id += 1;
b12235b1
MP		 881 }
882
883 return 0;
884}
885
886static int cs_etm__sample(struct cs_etm_queue *etmq)
887{
e573e978
RW		 888 struct cs_etm_auxtrace *etm = etmq->etm;
889 struct cs_etm_packet *tmp;
b12235b1 890 int ret;
e573e978 891 u64 instrs_executed;
b12235b1 892
e573e978
RW		 893 instrs_executed = cs_etm__instr_count(etmq->packet);
894 etmq->period_instructions += instrs_executed;
895
896 /*
897 * Record a branch when the last instruction in
898 * PREV_PACKET is a branch.
899 */
900 if (etm->synth_opts.last_branch &&
901 etmq->prev_packet &&
256e751c 902 etmq->prev_packet->sample_type == CS_ETM_RANGE &&
e573e978
RW		 903 etmq->prev_packet->last_instr_taken_branch)
904 cs_etm__update_last_branch_rb(etmq);
905
906 if (etm->sample_instructions &&
907 etmq->period_instructions >= etm->instructions_sample_period) {
908 /*
909 * Emit instruction sample periodically
910 * TODO: allow period to be defined in cycles and clock time
911 */
912
913 /* Get number of instructions executed after the sample point */
914 u64 instrs_over = etmq->period_instructions -
915 etm->instructions_sample_period;
916
917 /*
918 * Calculate the address of the sampled instruction (-1 as
919 * sample is reported as though instruction has just been
920 * executed, but PC has not advanced to next instruction)
921 */
922 u64 offset = (instrs_executed - instrs_over - 1);
923 u64 addr = cs_etm__instr_addr(etmq->packet, offset);
924
925 ret = cs_etm__synth_instruction_sample(
926 etmq, addr, etm->instructions_sample_period);
927 if (ret)
928 return ret;
929
930 /* Carry remaining instructions into next sample period */
931 etmq->period_instructions = instrs_over;
932 }
933
14a85b1e
LY		 934 if (etm->sample_branches && etmq->prev_packet) {
935 bool generate_sample = false;
936
937 /* Generate sample for tracing on packet */
938 if (etmq->prev_packet->sample_type == CS_ETM_TRACE_ON)
939 generate_sample = true;
940
941 /* Generate sample for branch taken packet */
942 if (etmq->prev_packet->sample_type == CS_ETM_RANGE &&
943 etmq->prev_packet->last_instr_taken_branch)
944 generate_sample = true;
945
946 if (generate_sample) {
947 ret = cs_etm__synth_branch_sample(etmq);
948 if (ret)
949 return ret;
950 }
e573e978 951 }
b12235b1 952
e573e978 953 if (etm->sample_branches || etm->synth_opts.last_branch) {
b12235b1 954 /*
e573e978
RW		 955 * Swap PACKET with PREV_PACKET: PACKET becomes PREV_PACKET for
956 * the next incoming packet.
b12235b1 957 */
e573e978
RW		 958 tmp = etmq->packet;
959 etmq->packet = etmq->prev_packet;
960 etmq->prev_packet = tmp;
b12235b1
MP		 961 }
962
963 return 0;
964}
965
256e751c
RW		 966static int cs_etm__flush(struct cs_etm_queue *etmq)
967{
968 int err = 0;
d603b4e9 969 struct cs_etm_auxtrace *etm = etmq->etm;
256e751c
RW		 970 struct cs_etm_packet *tmp;
971
3eb3e07b
LY		 972 if (!etmq->prev_packet)
973 return 0;
974
975 /* Handle start tracing packet */
976 if (etmq->prev_packet->sample_type == CS_ETM_EMPTY)
977 goto swap_packet;
978
256e751c 979 if (etmq->etm->synth_opts.last_branch &&
256e751c
RW		 980 etmq->prev_packet->sample_type == CS_ETM_RANGE) {
981 /*
982 * Generate a last branch event for the branches left in the
983 * circular buffer at the end of the trace.
984 *
985 * Use the address of the end of the last reported execution
986 * range
987 */
988 u64 addr = cs_etm__last_executed_instr(etmq->prev_packet);
989
990 err = cs_etm__synth_instruction_sample(
991 etmq, addr,
992 etmq->period_instructions);
6cd4ac6a
LY		 993 if (err)
994 return err;
995
256e751c
RW		 996 etmq->period_instructions = 0;
997
3eb3e07b
LY		 998 }
999
d603b4e9
LY		 1000 if (etm->sample_branches &&
1001 etmq->prev_packet->sample_type == CS_ETM_RANGE) {
1002 err = cs_etm__synth_branch_sample(etmq);
1003 if (err)
1004 return err;
1005 }
1006
3eb3e07b
LY		 1007swap_packet:
1008 if (etmq->etm->synth_opts.last_branch) {
256e751c
RW		 1009 /*
1010 * Swap PACKET with PREV_PACKET: PACKET becomes PREV_PACKET for
1011 * the next incoming packet.
1012 */
1013 tmp = etmq->packet;
1014 etmq->packet = etmq->prev_packet;
1015 etmq->prev_packet = tmp;
1016 }
1017
1018 return err;
1019}
1020
9f878b29
MP		 1021static int cs_etm__run_decoder(struct cs_etm_queue *etmq)
1022{
1023 struct cs_etm_auxtrace *etm = etmq->etm;
1024 struct cs_etm_buffer buffer;
1025 size_t buffer_used, processed;
1026 int err = 0;
1027
1028 if (!etm->kernel_start)
1029 etm->kernel_start = machine__kernel_start(etm->machine);
1030
1031 /* Go through each buffer in the queue and decode them one by one */
e573e978
RW		 1032 while (1) {
1033 buffer_used = 0;
1034 memset(&buffer, 0, sizeof(buffer));
1035 err = cs_etm__get_trace(&buffer, etmq);
1036 if (err <= 0)
1037 return err;
1038 /*
1039 * We cannot assume consecutive blocks in the data file are
1040 * contiguous, reset the decoder to force re-sync.
1041 */
1042 err = cs_etm_decoder__reset(etmq->decoder);
1043 if (err != 0)
9f878b29
MP		 1044 return err;
1045
e573e978
RW		 1046 /* Run trace decoder until buffer consumed or end of trace */
1047 do {
1048 processed = 0;
1049 err = cs_etm_decoder__process_data_block(
1050 etmq->decoder,
1051 etmq->offset,
1052 &buffer.buf[buffer_used],
1053 buffer.len - buffer_used,
1054 &processed);
1055 if (err)
1056 return err;
1057
1058 etmq->offset += processed;
1059 buffer_used += processed;
1060
1061 /* Process each packet in this chunk */
1062 while (1) {
1063 err = cs_etm_decoder__get_packet(etmq->decoder,
1064 etmq->packet);
1065 if (err <= 0)
1066 /*
1067 * Stop processing this chunk on
1068 * end of data or error
1069 */
1070 break;
1071
256e751c
RW		 1072 switch (etmq->packet->sample_type) {
1073 case CS_ETM_RANGE:
1074 /*
1075 * If the packet contains an instruction
1076 * range, generate instruction sequence
1077 * events.
1078 */
1079 cs_etm__sample(etmq);
1080 break;
1081 case CS_ETM_TRACE_ON:
1082 /*
1083 * Discontinuity in trace, flush
1084 * previous branch stack
1085 */
1086 cs_etm__flush(etmq);
1087 break;
3eb3e07b
LY		 1088 case CS_ETM_EMPTY:
1089 /*
1090 * Should not receive empty packet,
1091 * report error.
1092 */
1093 pr_err("CS ETM Trace: empty packet\n");
1094 return -EINVAL;
256e751c
RW		 1095 default:
1096 break;
1097 }
e573e978
RW		 1098 }
1099 } while (buffer.len > buffer_used);
b12235b1 1100
256e751c
RW		 1101 if (err == 0)
1102 /* Flush any remaining branch stack entries */
1103 err = cs_etm__flush(etmq);
e573e978 1104 }
9f878b29
MP		 1105
1106 return err;
1107}
1108
1109static int cs_etm__process_timeless_queues(struct cs_etm_auxtrace *etm,
1110 pid_t tid, u64 time_)
1111{
1112 unsigned int i;
1113 struct auxtrace_queues *queues = &etm->queues;
1114
1115 for (i = 0; i < queues->nr_queues; i++) {
1116 struct auxtrace_queue *queue = &etm->queues.queue_array[i];
1117 struct cs_etm_queue *etmq = queue->priv;
1118
1119 if (etmq && ((tid == -1) || (etmq->tid == tid))) {
1120 etmq->time = time_;
1121 cs_etm__set_pid_tid_cpu(etm, queue);
1122 cs_etm__run_decoder(etmq);
1123 }
1124 }
1125
1126 return 0;
1127}
1128
440a23b3
MP		 1129static int cs_etm__process_event(struct perf_session *session,
1130 union perf_event *event,
1131 struct perf_sample *sample,
1132 struct perf_tool *tool)
1133{
20d9c478
MP		 1134 int err = 0;
1135 u64 timestamp;
1136 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
1137 struct cs_etm_auxtrace,
1138 auxtrace);
1139
20d9c478
MP		 1140 if (dump_trace)
1141 return 0;
1142
1143 if (!tool->ordered_events) {
1144 pr_err("CoreSight ETM Trace requires ordered events\n");
1145 return -EINVAL;
1146 }
1147
1148 if (!etm->timeless_decoding)
1149 return -EINVAL;
1150
1151 if (sample->time && (sample->time != (u64) -1))
1152 timestamp = sample->time;
1153 else
1154 timestamp = 0;
1155
1156 if (timestamp || etm->timeless_decoding) {
1157 err = cs_etm__update_queues(etm);
1158 if (err)
1159 return err;
1160 }
1161
9f878b29
MP		 1162 if (event->header.type == PERF_RECORD_EXIT)
1163 return cs_etm__process_timeless_queues(etm,
1164 event->fork.tid,
1165 sample->time);
1166
440a23b3
MP		 1167 return 0;
1168}
1169
1170static int cs_etm__process_auxtrace_event(struct perf_session *session,
1171 union perf_event *event,
68ffe390 1172 struct perf_tool *tool __maybe_unused)
440a23b3 1173{
68ffe390
MP		 1174 struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
1175 struct cs_etm_auxtrace,
1176 auxtrace);
1177 if (!etm->data_queued) {
1178 struct auxtrace_buffer *buffer;
1179 off_t data_offset;
1180 int fd = perf_data__fd(session->data);
1181 bool is_pipe = perf_data__is_pipe(session->data);
1182 int err;
1183
1184 if (is_pipe)
1185 data_offset = 0;
1186 else {
1187 data_offset = lseek(fd, 0, SEEK_CUR);
1188 if (data_offset == -1)
1189 return -errno;
1190 }
1191
1192 err = auxtrace_queues__add_event(&etm->queues, session,
1193 event, data_offset, &buffer);
1194 if (err)
1195 return err;
1196
1197 if (dump_trace)
1198 if (auxtrace_buffer__get_data(buffer, fd)) {
1199 cs_etm__dump_event(etm, buffer);
1200 auxtrace_buffer__put_data(buffer);
1201 }
1202 }
1203
440a23b3
MP		 1204 return 0;
1205}
1206
1207static bool cs_etm__is_timeless_decoding(struct cs_etm_auxtrace *etm)
1208{
1209 struct perf_evsel *evsel;
1210 struct perf_evlist *evlist = etm->session->evlist;
1211 bool timeless_decoding = true;
1212
1213 /*
1214 * Circle through the list of event and complain if we find one
1215 * with the time bit set.
1216 */
1217 evlist__for_each_entry(evlist, evsel) {
1218 if ((evsel->attr.sample_type & PERF_SAMPLE_TIME))
1219 timeless_decoding = false;
1220 }
1221
1222 return timeless_decoding;
1223}
1224
cd8bfd8c
TJ		 1225static const char * const cs_etm_global_header_fmts[] = {
1226 [CS_HEADER_VERSION_0] = " Header version %llx\n",
1227 [CS_PMU_TYPE_CPUS] = " PMU type/num cpus %llx\n",
1228 [CS_ETM_SNAPSHOT] = " Snapshot %llx\n",
1229};
1230
1231static const char * const cs_etm_priv_fmts[] = {
1232 [CS_ETM_MAGIC] = " Magic number %llx\n",
1233 [CS_ETM_CPU] = " CPU %lld\n",
1234 [CS_ETM_ETMCR] = " ETMCR %llx\n",
1235 [CS_ETM_ETMTRACEIDR] = " ETMTRACEIDR %llx\n",
1236 [CS_ETM_ETMCCER] = " ETMCCER %llx\n",
1237 [CS_ETM_ETMIDR] = " ETMIDR %llx\n",
1238};
1239
1240static const char * const cs_etmv4_priv_fmts[] = {
1241 [CS_ETM_MAGIC] = " Magic number %llx\n",
1242 [CS_ETM_CPU] = " CPU %lld\n",
1243 [CS_ETMV4_TRCCONFIGR] = " TRCCONFIGR %llx\n",
1244 [CS_ETMV4_TRCTRACEIDR] = " TRCTRACEIDR %llx\n",
1245 [CS_ETMV4_TRCIDR0] = " TRCIDR0 %llx\n",
1246 [CS_ETMV4_TRCIDR1] = " TRCIDR1 %llx\n",
1247 [CS_ETMV4_TRCIDR2] = " TRCIDR2 %llx\n",
1248 [CS_ETMV4_TRCIDR8] = " TRCIDR8 %llx\n",
1249 [CS_ETMV4_TRCAUTHSTATUS] = " TRCAUTHSTATUS %llx\n",
1250};
1251
1252static void cs_etm__print_auxtrace_info(u64 *val, int num)
1253{
1254 int i, j, cpu = 0;
1255
1256 for (i = 0; i < CS_HEADER_VERSION_0_MAX; i++)
1257 fprintf(stdout, cs_etm_global_header_fmts[i], val[i]);
1258
1259 for (i = CS_HEADER_VERSION_0_MAX; cpu < num; cpu++) {
1260 if (val[i] == __perf_cs_etmv3_magic)
1261 for (j = 0; j < CS_ETM_PRIV_MAX; j++, i++)
1262 fprintf(stdout, cs_etm_priv_fmts[j], val[i]);
1263 else if (val[i] == __perf_cs_etmv4_magic)
1264 for (j = 0; j < CS_ETMV4_PRIV_MAX; j++, i++)
1265 fprintf(stdout, cs_etmv4_priv_fmts[j], val[i]);
1266 else
1267 /* failure.. return */
1268 return;
1269 }
1270}
1271
440a23b3
MP		 1272int cs_etm__process_auxtrace_info(union perf_event *event,
1273 struct perf_session *session)
1274{
1275 struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info;
1276 struct cs_etm_auxtrace *etm = NULL;
cd8bfd8c
TJ		 1277 struct int_node *inode;
1278 unsigned int pmu_type;
440a23b3
MP		 1279 int event_header_size = sizeof(struct perf_event_header);
1280 int info_header_size;
1281 int total_size = auxtrace_info->header.size;
cd8bfd8c
TJ		 1282 int priv_size = 0;
1283 int num_cpu;
1284 int err = 0, idx = -1;
1285 int i, j, k;
1286 u64 *ptr, *hdr = NULL;
1287 u64 **metadata = NULL;
440a23b3
MP		 1288
1289 /*
1290 * sizeof(auxtrace_info_event::type) +
1291 * sizeof(auxtrace_info_event::reserved) == 8
1292 */
1293 info_header_size = 8;
1294
1295 if (total_size < (event_header_size + info_header_size))
1296 return -EINVAL;
1297
cd8bfd8c
TJ		 1298 priv_size = total_size - event_header_size - info_header_size;
1299
1300 /* First the global part */
1301 ptr = (u64 *) auxtrace_info->priv;
1302
1303 /* Look for version '0' of the header */
1304 if (ptr[0] != 0)
1305 return -EINVAL;
1306
1307 hdr = zalloc(sizeof(*hdr) * CS_HEADER_VERSION_0_MAX);
1308 if (!hdr)
1309 return -ENOMEM;
1310
1311 /* Extract header information - see cs-etm.h for format */
1312 for (i = 0; i < CS_HEADER_VERSION_0_MAX; i++)
1313 hdr[i] = ptr[i];
1314 num_cpu = hdr[CS_PMU_TYPE_CPUS] & 0xffffffff;
1315 pmu_type = (unsigned int) ((hdr[CS_PMU_TYPE_CPUS] >> 32) &
1316 0xffffffff);
1317
1318 /*
1319 * Create an RB tree for traceID-CPU# tuple. Since the conversion has
1320 * to be made for each packet that gets decoded, optimizing access in
1321 * anything other than a sequential array is worth doing.
1322 */
1323 traceid_list = intlist__new(NULL);
1324 if (!traceid_list) {
1325 err = -ENOMEM;
1326 goto err_free_hdr;
1327 }
1328
1329 metadata = zalloc(sizeof(*metadata) * num_cpu);
1330 if (!metadata) {
1331 err = -ENOMEM;
1332 goto err_free_traceid_list;
1333 }
1334
1335 /*
1336 * The metadata is stored in the auxtrace_info section and encodes
1337 * the configuration of the ARM embedded trace macrocell which is
1338 * required by the trace decoder to properly decode the trace due
1339 * to its highly compressed nature.
1340 */
1341 for (j = 0; j < num_cpu; j++) {
1342 if (ptr[i] == __perf_cs_etmv3_magic) {
1343 metadata[j] = zalloc(sizeof(*metadata[j]) *
1344 CS_ETM_PRIV_MAX);
1345 if (!metadata[j]) {
1346 err = -ENOMEM;
1347 goto err_free_metadata;
1348 }
1349 for (k = 0; k < CS_ETM_PRIV_MAX; k++)
1350 metadata[j][k] = ptr[i + k];
1351
1352 /* The traceID is our handle */
1353 idx = metadata[j][CS_ETM_ETMTRACEIDR];
1354 i += CS_ETM_PRIV_MAX;
1355 } else if (ptr[i] == __perf_cs_etmv4_magic) {
1356 metadata[j] = zalloc(sizeof(*metadata[j]) *
1357 CS_ETMV4_PRIV_MAX);
1358 if (!metadata[j]) {
1359 err = -ENOMEM;
1360 goto err_free_metadata;
1361 }
1362 for (k = 0; k < CS_ETMV4_PRIV_MAX; k++)
1363 metadata[j][k] = ptr[i + k];
1364
1365 /* The traceID is our handle */
1366 idx = metadata[j][CS_ETMV4_TRCTRACEIDR];
1367 i += CS_ETMV4_PRIV_MAX;
1368 }
1369
1370 /* Get an RB node for this CPU */
1371 inode = intlist__findnew(traceid_list, idx);
1372
1373 /* Something went wrong, no need to continue */
1374 if (!inode) {
1375 err = PTR_ERR(inode);
1376 goto err_free_metadata;
1377 }
1378
1379 /*
1380 * The node for that CPU should not be taken.
1381 * Back out if that's the case.
1382 */
1383 if (inode->priv) {
1384 err = -EINVAL;
1385 goto err_free_metadata;
1386 }
1387 /* All good, associate the traceID with the CPU# */
1388 inode->priv = &metadata[j][CS_ETM_CPU];
1389 }
1390
1391 /*
1392 * Each of CS_HEADER_VERSION_0_MAX, CS_ETM_PRIV_MAX and
1393 * CS_ETMV4_PRIV_MAX mark how many double words are in the
1394 * global metadata, and each cpu's metadata respectively.
1395 * The following tests if the correct number of double words was
1396 * present in the auxtrace info section.
1397 */
1398 if (i * 8 != priv_size) {
1399 err = -EINVAL;
1400 goto err_free_metadata;
1401 }
1402
440a23b3
MP		 1403 etm = zalloc(sizeof(*etm));
1404
cd8bfd8c 1405 if (!etm) {
440a23b3 1406 err = -ENOMEM;
cd8bfd8c
TJ		 1407 goto err_free_metadata;
1408 }
440a23b3
MP		 1409
1410 err = auxtrace_queues__init(&etm->queues);
1411 if (err)
1412 goto err_free_etm;
1413
1414 etm->session = session;
1415 etm->machine = &session->machines.host;
1416
cd8bfd8c
TJ		 1417 etm->num_cpu = num_cpu;
1418 etm->pmu_type = pmu_type;
1419 etm->snapshot_mode = (hdr[CS_ETM_SNAPSHOT] != 0);
1420 etm->metadata = metadata;
440a23b3
MP		 1421 etm->auxtrace_type = auxtrace_info->type;
1422 etm->timeless_decoding = cs_etm__is_timeless_decoding(etm);
1423
1424 etm->auxtrace.process_event = cs_etm__process_event;
1425 etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event;
1426 etm->auxtrace.flush_events = cs_etm__flush_events;
1427 etm->auxtrace.free_events = cs_etm__free_events;
1428 etm->auxtrace.free = cs_etm__free;
1429 session->auxtrace = &etm->auxtrace;
1430
46d53620
LY		 1431 etm->unknown_thread = thread__new(999999999, 999999999);
1432 if (!etm->unknown_thread)
1433 goto err_free_queues;
1434
1435 /*
1436 * Initialize list node so that at thread__zput() we can avoid
1437 * segmentation fault at list_del_init().
1438 */
1439 INIT_LIST_HEAD(&etm->unknown_thread->node);
1440
1441 err = thread__set_comm(etm->unknown_thread, "unknown", 0);
1442 if (err)
1443 goto err_delete_thread;
1444
1445 if (thread__init_map_groups(etm->unknown_thread, etm->machine))
1446 goto err_delete_thread;
1447
cd8bfd8c
TJ		 1448 if (dump_trace) {
1449 cs_etm__print_auxtrace_info(auxtrace_info->priv, num_cpu);
440a23b3 1450 return 0;
cd8bfd8c 1451 }
440a23b3 1452
b12235b1
MP		 1453 if (session->itrace_synth_opts && session->itrace_synth_opts->set) {
1454 etm->synth_opts = *session->itrace_synth_opts;
1455 } else {
4eb06815
AK		 1456 itrace_synth_opts__set_default(&etm->synth_opts,
1457 session->itrace_synth_opts->default_no_sample);
b12235b1
MP		 1458 etm->synth_opts.callchain = false;
1459 }
1460
1461 err = cs_etm__synth_events(etm, session);
1462 if (err)
46d53620 1463 goto err_delete_thread;
b12235b1 1464
440a23b3
MP		 1465 err = auxtrace_queues__process_index(&etm->queues, session);
1466 if (err)
46d53620 1467 goto err_delete_thread;
440a23b3
MP		 1468
1469 etm->data_queued = etm->queues.populated;
1470
1471 return 0;
1472
46d53620
LY		 1473err_delete_thread:
1474 thread__zput(etm->unknown_thread);
440a23b3
MP		 1475err_free_queues:
1476 auxtrace_queues__free(&etm->queues);
1477 session->auxtrace = NULL;
1478err_free_etm:
1479 zfree(&etm);
cd8bfd8c
TJ		 1480err_free_metadata:
1481 /* No need to check @metadata[j], free(NULL) is supported */
1482 for (j = 0; j < num_cpu; j++)
1483 free(metadata[j]);
1484 zfree(&metadata);
1485err_free_traceid_list:
1486 intlist__delete(traceid_list);
1487err_free_hdr:
1488 zfree(&hdr);
440a23b3
MP		 1489
1490 return -EINVAL;
1491}
Linux 6.x block layer and io_uring tree(s)