1 // SPDX-License-Identifier: GPL-2.0 2 /* 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/coresight-pmu.h> 11 #include <linux/err.h> 12 #include <linux/kernel.h> 13 #include <linux/log2.h> 14 #include <linux/types.h> 15 #include <linux/zalloc.h> 16 17 #include <opencsd/ocsd_if_types.h> 18 #include <stdlib.h> 19 20 #include "auxtrace.h" 21 #include "color.h" 22 #include "cs-etm.h" 23 #include "cs-etm-decoder/cs-etm-decoder.h" 24 #include "debug.h" 25 #include "dso.h" 26 #include "evlist.h" 27 #include "intlist.h" 28 #include "machine.h" 29 #include "map.h" 30 #include "perf.h" 31 #include "session.h" 32 #include "map_symbol.h" 33 #include "branch.h" 34 #include "symbol.h" 35 #include "tool.h" 36 #include "thread.h" 37 #include "thread-stack.h" 38 #include "tsc.h" 39 #include <tools/libc_compat.h> 40 #include "util/synthetic-events.h" 41 42 struct 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 struct perf_tsc_conversion tc; 51 52 u8 timeless_decoding; 53 u8 snapshot_mode; 54 u8 data_queued; 55 u8 has_virtual_ts; /* Virtual/Kernel timestamps in the trace. */ 56 57 int num_cpu; 58 u64 latest_kernel_timestamp; 59 u32 auxtrace_type; 60 u64 branches_sample_type; 61 u64 branches_id; 62 u64 instructions_sample_type; 63 u64 instructions_sample_period; 64 u64 instructions_id; 65 u64 **metadata; 66 unsigned int pmu_type; 67 }; 68 69 struct cs_etm_traceid_queue { 70 u8 trace_chan_id; 71 pid_t pid, tid; 72 u64 period_instructions; 73 size_t last_branch_pos; 74 union perf_event *event_buf; 75 struct thread *thread; 76 struct branch_stack *last_branch; 77 struct branch_stack *last_branch_rb; 78 struct cs_etm_packet *prev_packet; 79 struct cs_etm_packet *packet; 80 struct cs_etm_packet_queue packet_queue; 81 }; 82 83 struct cs_etm_queue { 84 struct cs_etm_auxtrace *etm; 85 struct cs_etm_decoder *decoder; 86 struct auxtrace_buffer *buffer; 87 unsigned int queue_nr; 88 u8 pending_timestamp_chan_id; 89 u64 offset; 90 const unsigned char *buf; 91 size_t buf_len, buf_used; 92 /* Conversion between traceID and index in traceid_queues array */ 93 struct intlist *traceid_queues_list; 94 struct cs_etm_traceid_queue **traceid_queues; 95 }; 96 97 /* RB tree for quick conversion between traceID and metadata pointers */ 98 static struct intlist *traceid_list; 99 100 static int cs_etm__process_queues(struct cs_etm_auxtrace *etm); 101 static int cs_etm__process_timeless_queues(struct cs_etm_auxtrace *etm, 102 pid_t tid); 103 static int cs_etm__get_data_block(struct cs_etm_queue *etmq); 104 static int cs_etm__decode_data_block(struct cs_etm_queue *etmq); 105 106 /* PTMs ETMIDR [11:8] set to b0011 */ 107 #define ETMIDR_PTM_VERSION 0x00000300 108 109 /* 110 * A struct auxtrace_heap_item only has a queue_nr and a timestamp to 111 * work with. One option is to modify to auxtrace_heap_XYZ() API or simply 112 * encode the etm queue number as the upper 16 bit and the channel as 113 * the lower 16 bit. 114 */ 115 #define TO_CS_QUEUE_NR(queue_nr, trace_chan_id) \ 116 (queue_nr << 16 | trace_chan_id) 117 #define TO_QUEUE_NR(cs_queue_nr) (cs_queue_nr >> 16) 118 #define TO_TRACE_CHAN_ID(cs_queue_nr) (cs_queue_nr & 0x0000ffff) 119 120 static u32 cs_etm__get_v7_protocol_version(u32 etmidr) 121 { 122 etmidr &= ETMIDR_PTM_VERSION; 123 124 if (etmidr == ETMIDR_PTM_VERSION) 125 return CS_ETM_PROTO_PTM; 126 127 return CS_ETM_PROTO_ETMV3; 128 } 129 130 static int cs_etm__get_magic(u8 trace_chan_id, u64 *magic) 131 { 132 struct int_node *inode; 133 u64 *metadata; 134 135 inode = intlist__find(traceid_list, trace_chan_id); 136 if (!inode) 137 return -EINVAL; 138 139 metadata = inode->priv; 140 *magic = metadata[CS_ETM_MAGIC]; 141 return 0; 142 } 143 144 int cs_etm__get_cpu(u8 trace_chan_id, int *cpu) 145 { 146 struct int_node *inode; 147 u64 *metadata; 148 149 inode = intlist__find(traceid_list, trace_chan_id); 150 if (!inode) 151 return -EINVAL; 152 153 metadata = inode->priv; 154 *cpu = (int)metadata[CS_ETM_CPU]; 155 return 0; 156 } 157 158 /* 159 * The returned PID format is presented by two bits: 160 * 161 * Bit ETM_OPT_CTXTID: CONTEXTIDR or CONTEXTIDR_EL1 is traced; 162 * Bit ETM_OPT_CTXTID2: CONTEXTIDR_EL2 is traced. 163 * 164 * It's possible that the two bits ETM_OPT_CTXTID and ETM_OPT_CTXTID2 165 * are enabled at the same time when the session runs on an EL2 kernel. 166 * This means the CONTEXTIDR_EL1 and CONTEXTIDR_EL2 both will be 167 * recorded in the trace data, the tool will selectively use 168 * CONTEXTIDR_EL2 as PID. 169 */ 170 int cs_etm__get_pid_fmt(u8 trace_chan_id, u64 *pid_fmt) 171 { 172 struct int_node *inode; 173 u64 *metadata, val; 174 175 inode = intlist__find(traceid_list, trace_chan_id); 176 if (!inode) 177 return -EINVAL; 178 179 metadata = inode->priv; 180 181 if (metadata[CS_ETM_MAGIC] == __perf_cs_etmv3_magic) { 182 val = metadata[CS_ETM_ETMCR]; 183 /* CONTEXTIDR is traced */ 184 if (val & BIT(ETM_OPT_CTXTID)) 185 *pid_fmt = BIT(ETM_OPT_CTXTID); 186 } else { 187 val = metadata[CS_ETMV4_TRCCONFIGR]; 188 /* CONTEXTIDR_EL2 is traced */ 189 if (val & (BIT(ETM4_CFG_BIT_VMID) | BIT(ETM4_CFG_BIT_VMID_OPT))) 190 *pid_fmt = BIT(ETM_OPT_CTXTID2); 191 /* CONTEXTIDR_EL1 is traced */ 192 else if (val & BIT(ETM4_CFG_BIT_CTXTID)) 193 *pid_fmt = BIT(ETM_OPT_CTXTID); 194 } 195 196 return 0; 197 } 198 199 static int cs_etm__map_trace_id(u8 trace_chan_id, u64 *cpu_metadata) 200 { 201 struct int_node *inode; 202 203 /* Get an RB node for this CPU */ 204 inode = intlist__findnew(traceid_list, trace_chan_id); 205 206 /* Something went wrong, no need to continue */ 207 if (!inode) 208 return -ENOMEM; 209 210 /* 211 * The node for that CPU should not be taken. 212 * Back out if that's the case. 213 */ 214 if (inode->priv) 215 return -EINVAL; 216 217 /* All good, associate the traceID with the metadata pointer */ 218 inode->priv = cpu_metadata; 219 220 return 0; 221 } 222 223 static int cs_etm__metadata_get_trace_id(u8 *trace_chan_id, u64 *cpu_metadata) 224 { 225 u64 cs_etm_magic = cpu_metadata[CS_ETM_MAGIC]; 226 227 switch (cs_etm_magic) { 228 case __perf_cs_etmv3_magic: 229 *trace_chan_id = (u8)(cpu_metadata[CS_ETM_ETMTRACEIDR] & 230 CORESIGHT_TRACE_ID_VAL_MASK); 231 break; 232 case __perf_cs_etmv4_magic: 233 case __perf_cs_ete_magic: 234 *trace_chan_id = (u8)(cpu_metadata[CS_ETMV4_TRCTRACEIDR] & 235 CORESIGHT_TRACE_ID_VAL_MASK); 236 break; 237 default: 238 return -EINVAL; 239 } 240 return 0; 241 } 242 243 /* 244 * update metadata trace ID from the value found in the AUX_HW_INFO packet. 245 * This will also clear the CORESIGHT_TRACE_ID_UNUSED_FLAG flag if present. 246 */ 247 static int cs_etm__metadata_set_trace_id(u8 trace_chan_id, u64 *cpu_metadata) 248 { 249 u64 cs_etm_magic = cpu_metadata[CS_ETM_MAGIC]; 250 251 switch (cs_etm_magic) { 252 case __perf_cs_etmv3_magic: 253 cpu_metadata[CS_ETM_ETMTRACEIDR] = trace_chan_id; 254 break; 255 case __perf_cs_etmv4_magic: 256 case __perf_cs_ete_magic: 257 cpu_metadata[CS_ETMV4_TRCTRACEIDR] = trace_chan_id; 258 break; 259 260 default: 261 return -EINVAL; 262 } 263 return 0; 264 } 265 266 /* 267 * FIELD_GET (linux/bitfield.h) not available outside kernel code, 268 * and the header contains too many dependencies to just copy over, 269 * so roll our own based on the original 270 */ 271 #define __bf_shf(x) (__builtin_ffsll(x) - 1) 272 #define FIELD_GET(_mask, _reg) \ 273 ({ \ 274 (typeof(_mask))(((_reg) & (_mask)) >> __bf_shf(_mask)); \ 275 }) 276 277 /* 278 * Handle the PERF_RECORD_AUX_OUTPUT_HW_ID event. 279 * 280 * The payload associates the Trace ID and the CPU. 281 * The routine is tolerant of seeing multiple packets with the same association, 282 * but a CPU / Trace ID association changing during a session is an error. 283 */ 284 static int cs_etm__process_aux_output_hw_id(struct perf_session *session, 285 union perf_event *event) 286 { 287 struct cs_etm_auxtrace *etm; 288 struct perf_sample sample; 289 struct int_node *inode; 290 struct evsel *evsel; 291 u64 *cpu_data; 292 u64 hw_id; 293 int cpu, version, err; 294 u8 trace_chan_id, curr_chan_id; 295 296 /* extract and parse the HW ID */ 297 hw_id = event->aux_output_hw_id.hw_id; 298 version = FIELD_GET(CS_AUX_HW_ID_VERSION_MASK, hw_id); 299 trace_chan_id = FIELD_GET(CS_AUX_HW_ID_TRACE_ID_MASK, hw_id); 300 301 /* check that we can handle this version */ 302 if (version > CS_AUX_HW_ID_CURR_VERSION) 303 return -EINVAL; 304 305 /* get access to the etm metadata */ 306 etm = container_of(session->auxtrace, struct cs_etm_auxtrace, auxtrace); 307 if (!etm || !etm->metadata) 308 return -EINVAL; 309 310 /* parse the sample to get the CPU */ 311 evsel = evlist__event2evsel(session->evlist, event); 312 if (!evsel) 313 return -EINVAL; 314 err = evsel__parse_sample(evsel, event, &sample); 315 if (err) 316 return err; 317 cpu = sample.cpu; 318 if (cpu == -1) { 319 /* no CPU in the sample - possibly recorded with an old version of perf */ 320 pr_err("CS_ETM: no CPU AUX_OUTPUT_HW_ID sample. Use compatible perf to record."); 321 return -EINVAL; 322 } 323 324 /* See if the ID is mapped to a CPU, and it matches the current CPU */ 325 inode = intlist__find(traceid_list, trace_chan_id); 326 if (inode) { 327 cpu_data = inode->priv; 328 if ((int)cpu_data[CS_ETM_CPU] != cpu) { 329 pr_err("CS_ETM: map mismatch between HW_ID packet CPU and Trace ID\n"); 330 return -EINVAL; 331 } 332 333 /* check that the mapped ID matches */ 334 err = cs_etm__metadata_get_trace_id(&curr_chan_id, cpu_data); 335 if (err) 336 return err; 337 if (curr_chan_id != trace_chan_id) { 338 pr_err("CS_ETM: mismatch between CPU trace ID and HW_ID packet ID\n"); 339 return -EINVAL; 340 } 341 342 /* mapped and matched - return OK */ 343 return 0; 344 } 345 346 /* not one we've seen before - lets map it */ 347 cpu_data = etm->metadata[cpu]; 348 err = cs_etm__map_trace_id(trace_chan_id, cpu_data); 349 if (err) 350 return err; 351 352 /* 353 * if we are picking up the association from the packet, need to plug 354 * the correct trace ID into the metadata for setting up decoders later. 355 */ 356 err = cs_etm__metadata_set_trace_id(trace_chan_id, cpu_data); 357 return err; 358 } 359 360 void cs_etm__etmq_set_traceid_queue_timestamp(struct cs_etm_queue *etmq, 361 u8 trace_chan_id) 362 { 363 /* 364 * When a timestamp packet is encountered the backend code 365 * is stopped so that the front end has time to process packets 366 * that were accumulated in the traceID queue. Since there can 367 * be more than one channel per cs_etm_queue, we need to specify 368 * what traceID queue needs servicing. 369 */ 370 etmq->pending_timestamp_chan_id = trace_chan_id; 371 } 372 373 static u64 cs_etm__etmq_get_timestamp(struct cs_etm_queue *etmq, 374 u8 *trace_chan_id) 375 { 376 struct cs_etm_packet_queue *packet_queue; 377 378 if (!etmq->pending_timestamp_chan_id) 379 return 0; 380 381 if (trace_chan_id) 382 *trace_chan_id = etmq->pending_timestamp_chan_id; 383 384 packet_queue = cs_etm__etmq_get_packet_queue(etmq, 385 etmq->pending_timestamp_chan_id); 386 if (!packet_queue) 387 return 0; 388 389 /* Acknowledge pending status */ 390 etmq->pending_timestamp_chan_id = 0; 391 392 /* See function cs_etm_decoder__do_{hard|soft}_timestamp() */ 393 return packet_queue->cs_timestamp; 394 } 395 396 static void cs_etm__clear_packet_queue(struct cs_etm_packet_queue *queue) 397 { 398 int i; 399 400 queue->head = 0; 401 queue->tail = 0; 402 queue->packet_count = 0; 403 for (i = 0; i < CS_ETM_PACKET_MAX_BUFFER; i++) { 404 queue->packet_buffer[i].isa = CS_ETM_ISA_UNKNOWN; 405 queue->packet_buffer[i].start_addr = CS_ETM_INVAL_ADDR; 406 queue->packet_buffer[i].end_addr = CS_ETM_INVAL_ADDR; 407 queue->packet_buffer[i].instr_count = 0; 408 queue->packet_buffer[i].last_instr_taken_branch = false; 409 queue->packet_buffer[i].last_instr_size = 0; 410 queue->packet_buffer[i].last_instr_type = 0; 411 queue->packet_buffer[i].last_instr_subtype = 0; 412 queue->packet_buffer[i].last_instr_cond = 0; 413 queue->packet_buffer[i].flags = 0; 414 queue->packet_buffer[i].exception_number = UINT32_MAX; 415 queue->packet_buffer[i].trace_chan_id = UINT8_MAX; 416 queue->packet_buffer[i].cpu = INT_MIN; 417 } 418 } 419 420 static void cs_etm__clear_all_packet_queues(struct cs_etm_queue *etmq) 421 { 422 int idx; 423 struct int_node *inode; 424 struct cs_etm_traceid_queue *tidq; 425 struct intlist *traceid_queues_list = etmq->traceid_queues_list; 426 427 intlist__for_each_entry(inode, traceid_queues_list) { 428 idx = (int)(intptr_t)inode->priv; 429 tidq = etmq->traceid_queues[idx]; 430 cs_etm__clear_packet_queue(&tidq->packet_queue); 431 } 432 } 433 434 static int cs_etm__init_traceid_queue(struct cs_etm_queue *etmq, 435 struct cs_etm_traceid_queue *tidq, 436 u8 trace_chan_id) 437 { 438 int rc = -ENOMEM; 439 struct auxtrace_queue *queue; 440 struct cs_etm_auxtrace *etm = etmq->etm; 441 442 cs_etm__clear_packet_queue(&tidq->packet_queue); 443 444 queue = &etmq->etm->queues.queue_array[etmq->queue_nr]; 445 tidq->tid = queue->tid; 446 tidq->pid = -1; 447 tidq->trace_chan_id = trace_chan_id; 448 449 tidq->packet = zalloc(sizeof(struct cs_etm_packet)); 450 if (!tidq->packet) 451 goto out; 452 453 tidq->prev_packet = zalloc(sizeof(struct cs_etm_packet)); 454 if (!tidq->prev_packet) 455 goto out_free; 456 457 if (etm->synth_opts.last_branch) { 458 size_t sz = sizeof(struct branch_stack); 459 460 sz += etm->synth_opts.last_branch_sz * 461 sizeof(struct branch_entry); 462 tidq->last_branch = zalloc(sz); 463 if (!tidq->last_branch) 464 goto out_free; 465 tidq->last_branch_rb = zalloc(sz); 466 if (!tidq->last_branch_rb) 467 goto out_free; 468 } 469 470 tidq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE); 471 if (!tidq->event_buf) 472 goto out_free; 473 474 return 0; 475 476 out_free: 477 zfree(&tidq->last_branch_rb); 478 zfree(&tidq->last_branch); 479 zfree(&tidq->prev_packet); 480 zfree(&tidq->packet); 481 out: 482 return rc; 483 } 484 485 static struct cs_etm_traceid_queue 486 *cs_etm__etmq_get_traceid_queue(struct cs_etm_queue *etmq, u8 trace_chan_id) 487 { 488 int idx; 489 struct int_node *inode; 490 struct intlist *traceid_queues_list; 491 struct cs_etm_traceid_queue *tidq, **traceid_queues; 492 struct cs_etm_auxtrace *etm = etmq->etm; 493 494 if (etm->timeless_decoding) 495 trace_chan_id = CS_ETM_PER_THREAD_TRACEID; 496 497 traceid_queues_list = etmq->traceid_queues_list; 498 499 /* 500 * Check if the traceid_queue exist for this traceID by looking 501 * in the queue list. 502 */ 503 inode = intlist__find(traceid_queues_list, trace_chan_id); 504 if (inode) { 505 idx = (int)(intptr_t)inode->priv; 506 return etmq->traceid_queues[idx]; 507 } 508 509 /* We couldn't find a traceid_queue for this traceID, allocate one */ 510 tidq = malloc(sizeof(*tidq)); 511 if (!tidq) 512 return NULL; 513 514 memset(tidq, 0, sizeof(*tidq)); 515 516 /* Get a valid index for the new traceid_queue */ 517 idx = intlist__nr_entries(traceid_queues_list); 518 /* Memory for the inode is free'ed in cs_etm_free_traceid_queues () */ 519 inode = intlist__findnew(traceid_queues_list, trace_chan_id); 520 if (!inode) 521 goto out_free; 522 523 /* Associate this traceID with this index */ 524 inode->priv = (void *)(intptr_t)idx; 525 526 if (cs_etm__init_traceid_queue(etmq, tidq, trace_chan_id)) 527 goto out_free; 528 529 /* Grow the traceid_queues array by one unit */ 530 traceid_queues = etmq->traceid_queues; 531 traceid_queues = reallocarray(traceid_queues, 532 idx + 1, 533 sizeof(*traceid_queues)); 534 535 /* 536 * On failure reallocarray() returns NULL and the original block of 537 * memory is left untouched. 538 */ 539 if (!traceid_queues) 540 goto out_free; 541 542 traceid_queues[idx] = tidq; 543 etmq->traceid_queues = traceid_queues; 544 545 return etmq->traceid_queues[idx]; 546 547 out_free: 548 /* 549 * Function intlist__remove() removes the inode from the list 550 * and delete the memory associated to it. 551 */ 552 intlist__remove(traceid_queues_list, inode); 553 free(tidq); 554 555 return NULL; 556 } 557 558 struct cs_etm_packet_queue 559 *cs_etm__etmq_get_packet_queue(struct cs_etm_queue *etmq, u8 trace_chan_id) 560 { 561 struct cs_etm_traceid_queue *tidq; 562 563 tidq = cs_etm__etmq_get_traceid_queue(etmq, trace_chan_id); 564 if (tidq) 565 return &tidq->packet_queue; 566 567 return NULL; 568 } 569 570 static void cs_etm__packet_swap(struct cs_etm_auxtrace *etm, 571 struct cs_etm_traceid_queue *tidq) 572 { 573 struct cs_etm_packet *tmp; 574 575 if (etm->synth_opts.branches || etm->synth_opts.last_branch || 576 etm->synth_opts.instructions) { 577 /* 578 * Swap PACKET with PREV_PACKET: PACKET becomes PREV_PACKET for 579 * the next incoming packet. 580 */ 581 tmp = tidq->packet; 582 tidq->packet = tidq->prev_packet; 583 tidq->prev_packet = tmp; 584 } 585 } 586 587 static void cs_etm__packet_dump(const char *pkt_string) 588 { 589 const char *color = PERF_COLOR_BLUE; 590 int len = strlen(pkt_string); 591 592 if (len && (pkt_string[len-1] == '\n')) 593 color_fprintf(stdout, color, " %s", pkt_string); 594 else 595 color_fprintf(stdout, color, " %s\n", pkt_string); 596 597 fflush(stdout); 598 } 599 600 static void cs_etm__set_trace_param_etmv3(struct cs_etm_trace_params *t_params, 601 struct cs_etm_auxtrace *etm, int idx, 602 u32 etmidr) 603 { 604 u64 **metadata = etm->metadata; 605 606 t_params[idx].protocol = cs_etm__get_v7_protocol_version(etmidr); 607 t_params[idx].etmv3.reg_ctrl = metadata[idx][CS_ETM_ETMCR]; 608 t_params[idx].etmv3.reg_trc_id = metadata[idx][CS_ETM_ETMTRACEIDR]; 609 } 610 611 static void cs_etm__set_trace_param_etmv4(struct cs_etm_trace_params *t_params, 612 struct cs_etm_auxtrace *etm, int idx) 613 { 614 u64 **metadata = etm->metadata; 615 616 t_params[idx].protocol = CS_ETM_PROTO_ETMV4i; 617 t_params[idx].etmv4.reg_idr0 = metadata[idx][CS_ETMV4_TRCIDR0]; 618 t_params[idx].etmv4.reg_idr1 = metadata[idx][CS_ETMV4_TRCIDR1]; 619 t_params[idx].etmv4.reg_idr2 = metadata[idx][CS_ETMV4_TRCIDR2]; 620 t_params[idx].etmv4.reg_idr8 = metadata[idx][CS_ETMV4_TRCIDR8]; 621 t_params[idx].etmv4.reg_configr = metadata[idx][CS_ETMV4_TRCCONFIGR]; 622 t_params[idx].etmv4.reg_traceidr = metadata[idx][CS_ETMV4_TRCTRACEIDR]; 623 } 624 625 static void cs_etm__set_trace_param_ete(struct cs_etm_trace_params *t_params, 626 struct cs_etm_auxtrace *etm, int idx) 627 { 628 u64 **metadata = etm->metadata; 629 630 t_params[idx].protocol = CS_ETM_PROTO_ETE; 631 t_params[idx].ete.reg_idr0 = metadata[idx][CS_ETE_TRCIDR0]; 632 t_params[idx].ete.reg_idr1 = metadata[idx][CS_ETE_TRCIDR1]; 633 t_params[idx].ete.reg_idr2 = metadata[idx][CS_ETE_TRCIDR2]; 634 t_params[idx].ete.reg_idr8 = metadata[idx][CS_ETE_TRCIDR8]; 635 t_params[idx].ete.reg_configr = metadata[idx][CS_ETE_TRCCONFIGR]; 636 t_params[idx].ete.reg_traceidr = metadata[idx][CS_ETE_TRCTRACEIDR]; 637 t_params[idx].ete.reg_devarch = metadata[idx][CS_ETE_TRCDEVARCH]; 638 } 639 640 static int cs_etm__init_trace_params(struct cs_etm_trace_params *t_params, 641 struct cs_etm_auxtrace *etm, 642 int decoders) 643 { 644 int i; 645 u32 etmidr; 646 u64 architecture; 647 648 for (i = 0; i < decoders; i++) { 649 architecture = etm->metadata[i][CS_ETM_MAGIC]; 650 651 switch (architecture) { 652 case __perf_cs_etmv3_magic: 653 etmidr = etm->metadata[i][CS_ETM_ETMIDR]; 654 cs_etm__set_trace_param_etmv3(t_params, etm, i, etmidr); 655 break; 656 case __perf_cs_etmv4_magic: 657 cs_etm__set_trace_param_etmv4(t_params, etm, i); 658 break; 659 case __perf_cs_ete_magic: 660 cs_etm__set_trace_param_ete(t_params, etm, i); 661 break; 662 default: 663 return -EINVAL; 664 } 665 } 666 667 return 0; 668 } 669 670 static int cs_etm__init_decoder_params(struct cs_etm_decoder_params *d_params, 671 struct cs_etm_queue *etmq, 672 enum cs_etm_decoder_operation mode, 673 bool formatted) 674 { 675 int ret = -EINVAL; 676 677 if (!(mode < CS_ETM_OPERATION_MAX)) 678 goto out; 679 680 d_params->packet_printer = cs_etm__packet_dump; 681 d_params->operation = mode; 682 d_params->data = etmq; 683 d_params->formatted = formatted; 684 d_params->fsyncs = false; 685 d_params->hsyncs = false; 686 d_params->frame_aligned = true; 687 688 ret = 0; 689 out: 690 return ret; 691 } 692 693 static void cs_etm__dump_event(struct cs_etm_queue *etmq, 694 struct auxtrace_buffer *buffer) 695 { 696 int ret; 697 const char *color = PERF_COLOR_BLUE; 698 size_t buffer_used = 0; 699 700 fprintf(stdout, "\n"); 701 color_fprintf(stdout, color, 702 ". ... CoreSight %s Trace data: size %#zx bytes\n", 703 cs_etm_decoder__get_name(etmq->decoder), buffer->size); 704 705 do { 706 size_t consumed; 707 708 ret = cs_etm_decoder__process_data_block( 709 etmq->decoder, buffer->offset, 710 &((u8 *)buffer->data)[buffer_used], 711 buffer->size - buffer_used, &consumed); 712 if (ret) 713 break; 714 715 buffer_used += consumed; 716 } while (buffer_used < buffer->size); 717 718 cs_etm_decoder__reset(etmq->decoder); 719 } 720 721 static int cs_etm__flush_events(struct perf_session *session, 722 struct perf_tool *tool) 723 { 724 struct cs_etm_auxtrace *etm = container_of(session->auxtrace, 725 struct cs_etm_auxtrace, 726 auxtrace); 727 if (dump_trace) 728 return 0; 729 730 if (!tool->ordered_events) 731 return -EINVAL; 732 733 if (etm->timeless_decoding) 734 return cs_etm__process_timeless_queues(etm, -1); 735 736 return cs_etm__process_queues(etm); 737 } 738 739 static void cs_etm__free_traceid_queues(struct cs_etm_queue *etmq) 740 { 741 int idx; 742 uintptr_t priv; 743 struct int_node *inode, *tmp; 744 struct cs_etm_traceid_queue *tidq; 745 struct intlist *traceid_queues_list = etmq->traceid_queues_list; 746 747 intlist__for_each_entry_safe(inode, tmp, traceid_queues_list) { 748 priv = (uintptr_t)inode->priv; 749 idx = priv; 750 751 /* Free this traceid_queue from the array */ 752 tidq = etmq->traceid_queues[idx]; 753 thread__zput(tidq->thread); 754 zfree(&tidq->event_buf); 755 zfree(&tidq->last_branch); 756 zfree(&tidq->last_branch_rb); 757 zfree(&tidq->prev_packet); 758 zfree(&tidq->packet); 759 zfree(&tidq); 760 761 /* 762 * Function intlist__remove() removes the inode from the list 763 * and delete the memory associated to it. 764 */ 765 intlist__remove(traceid_queues_list, inode); 766 } 767 768 /* Then the RB tree itself */ 769 intlist__delete(traceid_queues_list); 770 etmq->traceid_queues_list = NULL; 771 772 /* finally free the traceid_queues array */ 773 zfree(&etmq->traceid_queues); 774 } 775 776 static void cs_etm__free_queue(void *priv) 777 { 778 struct cs_etm_queue *etmq = priv; 779 780 if (!etmq) 781 return; 782 783 cs_etm_decoder__free(etmq->decoder); 784 cs_etm__free_traceid_queues(etmq); 785 free(etmq); 786 } 787 788 static void cs_etm__free_events(struct perf_session *session) 789 { 790 unsigned int i; 791 struct cs_etm_auxtrace *aux = container_of(session->auxtrace, 792 struct cs_etm_auxtrace, 793 auxtrace); 794 struct auxtrace_queues *queues = &aux->queues; 795 796 for (i = 0; i < queues->nr_queues; i++) { 797 cs_etm__free_queue(queues->queue_array[i].priv); 798 queues->queue_array[i].priv = NULL; 799 } 800 801 auxtrace_queues__free(queues); 802 } 803 804 static void cs_etm__free(struct perf_session *session) 805 { 806 int i; 807 struct int_node *inode, *tmp; 808 struct cs_etm_auxtrace *aux = container_of(session->auxtrace, 809 struct cs_etm_auxtrace, 810 auxtrace); 811 cs_etm__free_events(session); 812 session->auxtrace = NULL; 813 814 /* First remove all traceID/metadata nodes for the RB tree */ 815 intlist__for_each_entry_safe(inode, tmp, traceid_list) 816 intlist__remove(traceid_list, inode); 817 /* Then the RB tree itself */ 818 intlist__delete(traceid_list); 819 820 for (i = 0; i < aux->num_cpu; i++) 821 zfree(&aux->metadata[i]); 822 823 thread__zput(aux->unknown_thread); 824 zfree(&aux->metadata); 825 zfree(&aux); 826 } 827 828 static bool cs_etm__evsel_is_auxtrace(struct perf_session *session, 829 struct evsel *evsel) 830 { 831 struct cs_etm_auxtrace *aux = container_of(session->auxtrace, 832 struct cs_etm_auxtrace, 833 auxtrace); 834 835 return evsel->core.attr.type == aux->pmu_type; 836 } 837 838 static u8 cs_etm__cpu_mode(struct cs_etm_queue *etmq, u64 address) 839 { 840 struct machine *machine; 841 842 machine = etmq->etm->machine; 843 844 if (address >= machine__kernel_start(machine)) { 845 if (machine__is_host(machine)) 846 return PERF_RECORD_MISC_KERNEL; 847 else 848 return PERF_RECORD_MISC_GUEST_KERNEL; 849 } else { 850 if (machine__is_host(machine)) 851 return PERF_RECORD_MISC_USER; 852 else if (perf_guest) 853 return PERF_RECORD_MISC_GUEST_USER; 854 else 855 return PERF_RECORD_MISC_HYPERVISOR; 856 } 857 } 858 859 static u32 cs_etm__mem_access(struct cs_etm_queue *etmq, u8 trace_chan_id, 860 u64 address, size_t size, u8 *buffer) 861 { 862 u8 cpumode; 863 u64 offset; 864 int len; 865 struct thread *thread; 866 struct machine *machine; 867 struct addr_location al; 868 struct dso *dso; 869 struct cs_etm_traceid_queue *tidq; 870 871 if (!etmq) 872 return 0; 873 874 machine = etmq->etm->machine; 875 cpumode = cs_etm__cpu_mode(etmq, address); 876 tidq = cs_etm__etmq_get_traceid_queue(etmq, trace_chan_id); 877 if (!tidq) 878 return 0; 879 880 thread = tidq->thread; 881 if (!thread) { 882 if (cpumode != PERF_RECORD_MISC_KERNEL) 883 return 0; 884 thread = etmq->etm->unknown_thread; 885 } 886 887 dso = map__dso(al.map); 888 889 if (!thread__find_map(thread, cpumode, address, &al) || !dso) 890 return 0; 891 892 if (dso->data.status == DSO_DATA_STATUS_ERROR && 893 dso__data_status_seen(dso, DSO_DATA_STATUS_SEEN_ITRACE)) 894 return 0; 895 896 offset = al.map->map_ip(al.map, address); 897 898 map__load(al.map); 899 900 len = dso__data_read_offset(dso, machine, offset, buffer, size); 901 902 if (len <= 0) { 903 ui__warning_once("CS ETM Trace: Missing DSO. Use 'perf archive' or debuginfod to export data from the traced system.\n" 904 " Enable CONFIG_PROC_KCORE or use option '-k /path/to/vmlinux' for kernel symbols.\n"); 905 if (!dso->auxtrace_warned) { 906 pr_err("CS ETM Trace: Debug data not found for address %#"PRIx64" in %s\n", 907 address, 908 dso->long_name ? dso->long_name : "Unknown"); 909 dso->auxtrace_warned = true; 910 } 911 return 0; 912 } 913 914 return len; 915 } 916 917 static struct cs_etm_queue *cs_etm__alloc_queue(struct cs_etm_auxtrace *etm, 918 bool formatted) 919 { 920 struct cs_etm_decoder_params d_params; 921 struct cs_etm_trace_params *t_params = NULL; 922 struct cs_etm_queue *etmq; 923 /* 924 * Each queue can only contain data from one CPU when unformatted, so only one decoder is 925 * needed. 926 */ 927 int decoders = formatted ? etm->num_cpu : 1; 928 929 etmq = zalloc(sizeof(*etmq)); 930 if (!etmq) 931 return NULL; 932 933 etmq->traceid_queues_list = intlist__new(NULL); 934 if (!etmq->traceid_queues_list) 935 goto out_free; 936 937 /* Use metadata to fill in trace parameters for trace decoder */ 938 t_params = zalloc(sizeof(*t_params) * decoders); 939 940 if (!t_params) 941 goto out_free; 942 943 if (cs_etm__init_trace_params(t_params, etm, decoders)) 944 goto out_free; 945 946 /* Set decoder parameters to decode trace packets */ 947 if (cs_etm__init_decoder_params(&d_params, etmq, 948 dump_trace ? CS_ETM_OPERATION_PRINT : 949 CS_ETM_OPERATION_DECODE, 950 formatted)) 951 goto out_free; 952 953 etmq->decoder = cs_etm_decoder__new(decoders, &d_params, 954 t_params); 955 956 if (!etmq->decoder) 957 goto out_free; 958 959 /* 960 * Register a function to handle all memory accesses required by 961 * the trace decoder library. 962 */ 963 if (cs_etm_decoder__add_mem_access_cb(etmq->decoder, 964 0x0L, ((u64) -1L), 965 cs_etm__mem_access)) 966 goto out_free_decoder; 967 968 zfree(&t_params); 969 return etmq; 970 971 out_free_decoder: 972 cs_etm_decoder__free(etmq->decoder); 973 out_free: 974 intlist__delete(etmq->traceid_queues_list); 975 free(etmq); 976 977 return NULL; 978 } 979 980 static int cs_etm__setup_queue(struct cs_etm_auxtrace *etm, 981 struct auxtrace_queue *queue, 982 unsigned int queue_nr, 983 bool formatted) 984 { 985 struct cs_etm_queue *etmq = queue->priv; 986 987 if (list_empty(&queue->head) || etmq) 988 return 0; 989 990 etmq = cs_etm__alloc_queue(etm, formatted); 991 992 if (!etmq) 993 return -ENOMEM; 994 995 queue->priv = etmq; 996 etmq->etm = etm; 997 etmq->queue_nr = queue_nr; 998 etmq->offset = 0; 999 1000 return 0; 1001 } 1002 1003 static int cs_etm__queue_first_cs_timestamp(struct cs_etm_auxtrace *etm, 1004 struct cs_etm_queue *etmq, 1005 unsigned int queue_nr) 1006 { 1007 int ret = 0; 1008 unsigned int cs_queue_nr; 1009 u8 trace_chan_id; 1010 u64 cs_timestamp; 1011 1012 /* 1013 * We are under a CPU-wide trace scenario. As such we need to know 1014 * when the code that generated the traces started to execute so that 1015 * it can be correlated with execution on other CPUs. So we get a 1016 * handle on the beginning of traces and decode until we find a 1017 * timestamp. The timestamp is then added to the auxtrace min heap 1018 * in order to know what nibble (of all the etmqs) to decode first. 1019 */ 1020 while (1) { 1021 /* 1022 * Fetch an aux_buffer from this etmq. Bail if no more 1023 * blocks or an error has been encountered. 1024 */ 1025 ret = cs_etm__get_data_block(etmq); 1026 if (ret <= 0) 1027 goto out; 1028 1029 /* 1030 * Run decoder on the trace block. The decoder will stop when 1031 * encountering a CS timestamp, a full packet queue or the end of 1032 * trace for that block. 1033 */ 1034 ret = cs_etm__decode_data_block(etmq); 1035 if (ret) 1036 goto out; 1037 1038 /* 1039 * Function cs_etm_decoder__do_{hard|soft}_timestamp() does all 1040 * the timestamp calculation for us. 1041 */ 1042 cs_timestamp = cs_etm__etmq_get_timestamp(etmq, &trace_chan_id); 1043 1044 /* We found a timestamp, no need to continue. */ 1045 if (cs_timestamp) 1046 break; 1047 1048 /* 1049 * We didn't find a timestamp so empty all the traceid packet 1050 * queues before looking for another timestamp packet, either 1051 * in the current data block or a new one. Packets that were 1052 * just decoded are useless since no timestamp has been 1053 * associated with them. As such simply discard them. 1054 */ 1055 cs_etm__clear_all_packet_queues(etmq); 1056 } 1057 1058 /* 1059 * We have a timestamp. Add it to the min heap to reflect when 1060 * instructions conveyed by the range packets of this traceID queue 1061 * started to execute. Once the same has been done for all the traceID 1062 * queues of each etmq, redenring and decoding can start in 1063 * chronological order. 1064 * 1065 * Note that packets decoded above are still in the traceID's packet 1066 * queue and will be processed in cs_etm__process_queues(). 1067 */ 1068 cs_queue_nr = TO_CS_QUEUE_NR(queue_nr, trace_chan_id); 1069 ret = auxtrace_heap__add(&etm->heap, cs_queue_nr, cs_timestamp); 1070 out: 1071 return ret; 1072 } 1073 1074 static inline 1075 void cs_etm__copy_last_branch_rb(struct cs_etm_queue *etmq, 1076 struct cs_etm_traceid_queue *tidq) 1077 { 1078 struct branch_stack *bs_src = tidq->last_branch_rb; 1079 struct branch_stack *bs_dst = tidq->last_branch; 1080 size_t nr = 0; 1081 1082 /* 1083 * Set the number of records before early exit: ->nr is used to 1084 * determine how many branches to copy from ->entries. 1085 */ 1086 bs_dst->nr = bs_src->nr; 1087 1088 /* 1089 * Early exit when there is nothing to copy. 1090 */ 1091 if (!bs_src->nr) 1092 return; 1093 1094 /* 1095 * As bs_src->entries is a circular buffer, we need to copy from it in 1096 * two steps. First, copy the branches from the most recently inserted 1097 * branch ->last_branch_pos until the end of bs_src->entries buffer. 1098 */ 1099 nr = etmq->etm->synth_opts.last_branch_sz - tidq->last_branch_pos; 1100 memcpy(&bs_dst->entries[0], 1101 &bs_src->entries[tidq->last_branch_pos], 1102 sizeof(struct branch_entry) * nr); 1103 1104 /* 1105 * If we wrapped around at least once, the branches from the beginning 1106 * of the bs_src->entries buffer and until the ->last_branch_pos element 1107 * are older valid branches: copy them over. The total number of 1108 * branches copied over will be equal to the number of branches asked by 1109 * the user in last_branch_sz. 1110 */ 1111 if (bs_src->nr >= etmq->etm->synth_opts.last_branch_sz) { 1112 memcpy(&bs_dst->entries[nr], 1113 &bs_src->entries[0], 1114 sizeof(struct branch_entry) * tidq->last_branch_pos); 1115 } 1116 } 1117 1118 static inline 1119 void cs_etm__reset_last_branch_rb(struct cs_etm_traceid_queue *tidq) 1120 { 1121 tidq->last_branch_pos = 0; 1122 tidq->last_branch_rb->nr = 0; 1123 } 1124 1125 static inline int cs_etm__t32_instr_size(struct cs_etm_queue *etmq, 1126 u8 trace_chan_id, u64 addr) 1127 { 1128 u8 instrBytes[2]; 1129 1130 cs_etm__mem_access(etmq, trace_chan_id, addr, 1131 ARRAY_SIZE(instrBytes), instrBytes); 1132 /* 1133 * T32 instruction size is indicated by bits[15:11] of the first 1134 * 16-bit word of the instruction: 0b11101, 0b11110 and 0b11111 1135 * denote a 32-bit instruction. 1136 */ 1137 return ((instrBytes[1] & 0xF8) >= 0xE8) ? 4 : 2; 1138 } 1139 1140 static inline u64 cs_etm__first_executed_instr(struct cs_etm_packet *packet) 1141 { 1142 /* Returns 0 for the CS_ETM_DISCONTINUITY packet */ 1143 if (packet->sample_type == CS_ETM_DISCONTINUITY) 1144 return 0; 1145 1146 return packet->start_addr; 1147 } 1148 1149 static inline 1150 u64 cs_etm__last_executed_instr(const struct cs_etm_packet *packet) 1151 { 1152 /* Returns 0 for the CS_ETM_DISCONTINUITY packet */ 1153 if (packet->sample_type == CS_ETM_DISCONTINUITY) 1154 return 0; 1155 1156 return packet->end_addr - packet->last_instr_size; 1157 } 1158 1159 static inline u64 cs_etm__instr_addr(struct cs_etm_queue *etmq, 1160 u64 trace_chan_id, 1161 const struct cs_etm_packet *packet, 1162 u64 offset) 1163 { 1164 if (packet->isa == CS_ETM_ISA_T32) { 1165 u64 addr = packet->start_addr; 1166 1167 while (offset) { 1168 addr += cs_etm__t32_instr_size(etmq, 1169 trace_chan_id, addr); 1170 offset--; 1171 } 1172 return addr; 1173 } 1174 1175 /* Assume a 4 byte instruction size (A32/A64) */ 1176 return packet->start_addr + offset * 4; 1177 } 1178 1179 static void cs_etm__update_last_branch_rb(struct cs_etm_queue *etmq, 1180 struct cs_etm_traceid_queue *tidq) 1181 { 1182 struct branch_stack *bs = tidq->last_branch_rb; 1183 struct branch_entry *be; 1184 1185 /* 1186 * The branches are recorded in a circular buffer in reverse 1187 * chronological order: we start recording from the last element of the 1188 * buffer down. After writing the first element of the stack, move the 1189 * insert position back to the end of the buffer. 1190 */ 1191 if (!tidq->last_branch_pos) 1192 tidq->last_branch_pos = etmq->etm->synth_opts.last_branch_sz; 1193 1194 tidq->last_branch_pos -= 1; 1195 1196 be = &bs->entries[tidq->last_branch_pos]; 1197 be->from = cs_etm__last_executed_instr(tidq->prev_packet); 1198 be->to = cs_etm__first_executed_instr(tidq->packet); 1199 /* No support for mispredict */ 1200 be->flags.mispred = 0; 1201 be->flags.predicted = 1; 1202 1203 /* 1204 * Increment bs->nr until reaching the number of last branches asked by 1205 * the user on the command line. 1206 */ 1207 if (bs->nr < etmq->etm->synth_opts.last_branch_sz) 1208 bs->nr += 1; 1209 } 1210 1211 static int cs_etm__inject_event(union perf_event *event, 1212 struct perf_sample *sample, u64 type) 1213 { 1214 event->header.size = perf_event__sample_event_size(sample, type, 0); 1215 return perf_event__synthesize_sample(event, type, 0, sample); 1216 } 1217 1218 1219 static int 1220 cs_etm__get_trace(struct cs_etm_queue *etmq) 1221 { 1222 struct auxtrace_buffer *aux_buffer = etmq->buffer; 1223 struct auxtrace_buffer *old_buffer = aux_buffer; 1224 struct auxtrace_queue *queue; 1225 1226 queue = &etmq->etm->queues.queue_array[etmq->queue_nr]; 1227 1228 aux_buffer = auxtrace_buffer__next(queue, aux_buffer); 1229 1230 /* If no more data, drop the previous auxtrace_buffer and return */ 1231 if (!aux_buffer) { 1232 if (old_buffer) 1233 auxtrace_buffer__drop_data(old_buffer); 1234 etmq->buf_len = 0; 1235 return 0; 1236 } 1237 1238 etmq->buffer = aux_buffer; 1239 1240 /* If the aux_buffer doesn't have data associated, try to load it */ 1241 if (!aux_buffer->data) { 1242 /* get the file desc associated with the perf data file */ 1243 int fd = perf_data__fd(etmq->etm->session->data); 1244 1245 aux_buffer->data = auxtrace_buffer__get_data(aux_buffer, fd); 1246 if (!aux_buffer->data) 1247 return -ENOMEM; 1248 } 1249 1250 /* If valid, drop the previous buffer */ 1251 if (old_buffer) 1252 auxtrace_buffer__drop_data(old_buffer); 1253 1254 etmq->buf_used = 0; 1255 etmq->buf_len = aux_buffer->size; 1256 etmq->buf = aux_buffer->data; 1257 1258 return etmq->buf_len; 1259 } 1260 1261 static void cs_etm__set_pid_tid_cpu(struct cs_etm_auxtrace *etm, 1262 struct cs_etm_traceid_queue *tidq) 1263 { 1264 if ((!tidq->thread) && (tidq->tid != -1)) 1265 tidq->thread = machine__find_thread(etm->machine, -1, 1266 tidq->tid); 1267 1268 if (tidq->thread) 1269 tidq->pid = tidq->thread->pid_; 1270 } 1271 1272 int cs_etm__etmq_set_tid(struct cs_etm_queue *etmq, 1273 pid_t tid, u8 trace_chan_id) 1274 { 1275 int cpu, err = -EINVAL; 1276 struct cs_etm_auxtrace *etm = etmq->etm; 1277 struct cs_etm_traceid_queue *tidq; 1278 1279 tidq = cs_etm__etmq_get_traceid_queue(etmq, trace_chan_id); 1280 if (!tidq) 1281 return err; 1282 1283 if (cs_etm__get_cpu(trace_chan_id, &cpu) < 0) 1284 return err; 1285 1286 err = machine__set_current_tid(etm->machine, cpu, tid, tid); 1287 if (err) 1288 return err; 1289 1290 tidq->tid = tid; 1291 thread__zput(tidq->thread); 1292 1293 cs_etm__set_pid_tid_cpu(etm, tidq); 1294 return 0; 1295 } 1296 1297 bool cs_etm__etmq_is_timeless(struct cs_etm_queue *etmq) 1298 { 1299 return !!etmq->etm->timeless_decoding; 1300 } 1301 1302 static void cs_etm__copy_insn(struct cs_etm_queue *etmq, 1303 u64 trace_chan_id, 1304 const struct cs_etm_packet *packet, 1305 struct perf_sample *sample) 1306 { 1307 /* 1308 * It's pointless to read instructions for the CS_ETM_DISCONTINUITY 1309 * packet, so directly bail out with 'insn_len' = 0. 1310 */ 1311 if (packet->sample_type == CS_ETM_DISCONTINUITY) { 1312 sample->insn_len = 0; 1313 return; 1314 } 1315 1316 /* 1317 * T32 instruction size might be 32-bit or 16-bit, decide by calling 1318 * cs_etm__t32_instr_size(). 1319 */ 1320 if (packet->isa == CS_ETM_ISA_T32) 1321 sample->insn_len = cs_etm__t32_instr_size(etmq, trace_chan_id, 1322 sample->ip); 1323 /* Otherwise, A64 and A32 instruction size are always 32-bit. */ 1324 else 1325 sample->insn_len = 4; 1326 1327 cs_etm__mem_access(etmq, trace_chan_id, sample->ip, 1328 sample->insn_len, (void *)sample->insn); 1329 } 1330 1331 u64 cs_etm__convert_sample_time(struct cs_etm_queue *etmq, u64 cs_timestamp) 1332 { 1333 struct cs_etm_auxtrace *etm = etmq->etm; 1334 1335 if (etm->has_virtual_ts) 1336 return tsc_to_perf_time(cs_timestamp, &etm->tc); 1337 else 1338 return cs_timestamp; 1339 } 1340 1341 static inline u64 cs_etm__resolve_sample_time(struct cs_etm_queue *etmq, 1342 struct cs_etm_traceid_queue *tidq) 1343 { 1344 struct cs_etm_auxtrace *etm = etmq->etm; 1345 struct cs_etm_packet_queue *packet_queue = &tidq->packet_queue; 1346 1347 if (etm->timeless_decoding) 1348 return 0; 1349 else if (etm->has_virtual_ts) 1350 return packet_queue->cs_timestamp; 1351 else 1352 return etm->latest_kernel_timestamp; 1353 } 1354 1355 static int cs_etm__synth_instruction_sample(struct cs_etm_queue *etmq, 1356 struct cs_etm_traceid_queue *tidq, 1357 u64 addr, u64 period) 1358 { 1359 int ret = 0; 1360 struct cs_etm_auxtrace *etm = etmq->etm; 1361 union perf_event *event = tidq->event_buf; 1362 struct perf_sample sample = {.ip = 0,}; 1363 1364 event->sample.header.type = PERF_RECORD_SAMPLE; 1365 event->sample.header.misc = cs_etm__cpu_mode(etmq, addr); 1366 event->sample.header.size = sizeof(struct perf_event_header); 1367 1368 /* Set time field based on etm auxtrace config. */ 1369 sample.time = cs_etm__resolve_sample_time(etmq, tidq); 1370 1371 sample.ip = addr; 1372 sample.pid = tidq->pid; 1373 sample.tid = tidq->tid; 1374 sample.id = etmq->etm->instructions_id; 1375 sample.stream_id = etmq->etm->instructions_id; 1376 sample.period = period; 1377 sample.cpu = tidq->packet->cpu; 1378 sample.flags = tidq->prev_packet->flags; 1379 sample.cpumode = event->sample.header.misc; 1380 1381 cs_etm__copy_insn(etmq, tidq->trace_chan_id, tidq->packet, &sample); 1382 1383 if (etm->synth_opts.last_branch) 1384 sample.branch_stack = tidq->last_branch; 1385 1386 if (etm->synth_opts.inject) { 1387 ret = cs_etm__inject_event(event, &sample, 1388 etm->instructions_sample_type); 1389 if (ret) 1390 return ret; 1391 } 1392 1393 ret = perf_session__deliver_synth_event(etm->session, event, &sample); 1394 1395 if (ret) 1396 pr_err( 1397 "CS ETM Trace: failed to deliver instruction event, error %d\n", 1398 ret); 1399 1400 return ret; 1401 } 1402 1403 /* 1404 * The cs etm packet encodes an instruction range between a branch target 1405 * and the next taken branch. Generate sample accordingly. 1406 */ 1407 static int cs_etm__synth_branch_sample(struct cs_etm_queue *etmq, 1408 struct cs_etm_traceid_queue *tidq) 1409 { 1410 int ret = 0; 1411 struct cs_etm_auxtrace *etm = etmq->etm; 1412 struct perf_sample sample = {.ip = 0,}; 1413 union perf_event *event = tidq->event_buf; 1414 struct dummy_branch_stack { 1415 u64 nr; 1416 u64 hw_idx; 1417 struct branch_entry entries; 1418 } dummy_bs; 1419 u64 ip; 1420 1421 ip = cs_etm__last_executed_instr(tidq->prev_packet); 1422 1423 event->sample.header.type = PERF_RECORD_SAMPLE; 1424 event->sample.header.misc = cs_etm__cpu_mode(etmq, ip); 1425 event->sample.header.size = sizeof(struct perf_event_header); 1426 1427 /* Set time field based on etm auxtrace config. */ 1428 sample.time = cs_etm__resolve_sample_time(etmq, tidq); 1429 1430 sample.ip = ip; 1431 sample.pid = tidq->pid; 1432 sample.tid = tidq->tid; 1433 sample.addr = cs_etm__first_executed_instr(tidq->packet); 1434 sample.id = etmq->etm->branches_id; 1435 sample.stream_id = etmq->etm->branches_id; 1436 sample.period = 1; 1437 sample.cpu = tidq->packet->cpu; 1438 sample.flags = tidq->prev_packet->flags; 1439 sample.cpumode = event->sample.header.misc; 1440 1441 cs_etm__copy_insn(etmq, tidq->trace_chan_id, tidq->prev_packet, 1442 &sample); 1443 1444 /* 1445 * perf report cannot handle events without a branch stack 1446 */ 1447 if (etm->synth_opts.last_branch) { 1448 dummy_bs = (struct dummy_branch_stack){ 1449 .nr = 1, 1450 .hw_idx = -1ULL, 1451 .entries = { 1452 .from = sample.ip, 1453 .to = sample.addr, 1454 }, 1455 }; 1456 sample.branch_stack = (struct branch_stack *)&dummy_bs; 1457 } 1458 1459 if (etm->synth_opts.inject) { 1460 ret = cs_etm__inject_event(event, &sample, 1461 etm->branches_sample_type); 1462 if (ret) 1463 return ret; 1464 } 1465 1466 ret = perf_session__deliver_synth_event(etm->session, event, &sample); 1467 1468 if (ret) 1469 pr_err( 1470 "CS ETM Trace: failed to deliver instruction event, error %d\n", 1471 ret); 1472 1473 return ret; 1474 } 1475 1476 struct cs_etm_synth { 1477 struct perf_tool dummy_tool; 1478 struct perf_session *session; 1479 }; 1480 1481 static int cs_etm__event_synth(struct perf_tool *tool, 1482 union perf_event *event, 1483 struct perf_sample *sample __maybe_unused, 1484 struct machine *machine __maybe_unused) 1485 { 1486 struct cs_etm_synth *cs_etm_synth = 1487 container_of(tool, struct cs_etm_synth, dummy_tool); 1488 1489 return perf_session__deliver_synth_event(cs_etm_synth->session, 1490 event, NULL); 1491 } 1492 1493 static int cs_etm__synth_event(struct perf_session *session, 1494 struct perf_event_attr *attr, u64 id) 1495 { 1496 struct cs_etm_synth cs_etm_synth; 1497 1498 memset(&cs_etm_synth, 0, sizeof(struct cs_etm_synth)); 1499 cs_etm_synth.session = session; 1500 1501 return perf_event__synthesize_attr(&cs_etm_synth.dummy_tool, attr, 1, 1502 &id, cs_etm__event_synth); 1503 } 1504 1505 static int cs_etm__synth_events(struct cs_etm_auxtrace *etm, 1506 struct perf_session *session) 1507 { 1508 struct evlist *evlist = session->evlist; 1509 struct evsel *evsel; 1510 struct perf_event_attr attr; 1511 bool found = false; 1512 u64 id; 1513 int err; 1514 1515 evlist__for_each_entry(evlist, evsel) { 1516 if (evsel->core.attr.type == etm->pmu_type) { 1517 found = true; 1518 break; 1519 } 1520 } 1521 1522 if (!found) { 1523 pr_debug("No selected events with CoreSight Trace data\n"); 1524 return 0; 1525 } 1526 1527 memset(&attr, 0, sizeof(struct perf_event_attr)); 1528 attr.size = sizeof(struct perf_event_attr); 1529 attr.type = PERF_TYPE_HARDWARE; 1530 attr.sample_type = evsel->core.attr.sample_type & PERF_SAMPLE_MASK; 1531 attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID | 1532 PERF_SAMPLE_PERIOD; 1533 if (etm->timeless_decoding) 1534 attr.sample_type &= ~(u64)PERF_SAMPLE_TIME; 1535 else 1536 attr.sample_type |= PERF_SAMPLE_TIME; 1537 1538 attr.exclude_user = evsel->core.attr.exclude_user; 1539 attr.exclude_kernel = evsel->core.attr.exclude_kernel; 1540 attr.exclude_hv = evsel->core.attr.exclude_hv; 1541 attr.exclude_host = evsel->core.attr.exclude_host; 1542 attr.exclude_guest = evsel->core.attr.exclude_guest; 1543 attr.sample_id_all = evsel->core.attr.sample_id_all; 1544 attr.read_format = evsel->core.attr.read_format; 1545 1546 /* create new id val to be a fixed offset from evsel id */ 1547 id = evsel->core.id[0] + 1000000000; 1548 1549 if (!id) 1550 id = 1; 1551 1552 if (etm->synth_opts.branches) { 1553 attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS; 1554 attr.sample_period = 1; 1555 attr.sample_type |= PERF_SAMPLE_ADDR; 1556 err = cs_etm__synth_event(session, &attr, id); 1557 if (err) 1558 return err; 1559 etm->branches_sample_type = attr.sample_type; 1560 etm->branches_id = id; 1561 id += 1; 1562 attr.sample_type &= ~(u64)PERF_SAMPLE_ADDR; 1563 } 1564 1565 if (etm->synth_opts.last_branch) { 1566 attr.sample_type |= PERF_SAMPLE_BRANCH_STACK; 1567 /* 1568 * We don't use the hardware index, but the sample generation 1569 * code uses the new format branch_stack with this field, 1570 * so the event attributes must indicate that it's present. 1571 */ 1572 attr.branch_sample_type |= PERF_SAMPLE_BRANCH_HW_INDEX; 1573 } 1574 1575 if (etm->synth_opts.instructions) { 1576 attr.config = PERF_COUNT_HW_INSTRUCTIONS; 1577 attr.sample_period = etm->synth_opts.period; 1578 etm->instructions_sample_period = attr.sample_period; 1579 err = cs_etm__synth_event(session, &attr, id); 1580 if (err) 1581 return err; 1582 etm->instructions_sample_type = attr.sample_type; 1583 etm->instructions_id = id; 1584 id += 1; 1585 } 1586 1587 return 0; 1588 } 1589 1590 static int cs_etm__sample(struct cs_etm_queue *etmq, 1591 struct cs_etm_traceid_queue *tidq) 1592 { 1593 struct cs_etm_auxtrace *etm = etmq->etm; 1594 int ret; 1595 u8 trace_chan_id = tidq->trace_chan_id; 1596 u64 instrs_prev; 1597 1598 /* Get instructions remainder from previous packet */ 1599 instrs_prev = tidq->period_instructions; 1600 1601 tidq->period_instructions += tidq->packet->instr_count; 1602 1603 /* 1604 * Record a branch when the last instruction in 1605 * PREV_PACKET is a branch. 1606 */ 1607 if (etm->synth_opts.last_branch && 1608 tidq->prev_packet->sample_type == CS_ETM_RANGE && 1609 tidq->prev_packet->last_instr_taken_branch) 1610 cs_etm__update_last_branch_rb(etmq, tidq); 1611 1612 if (etm->synth_opts.instructions && 1613 tidq->period_instructions >= etm->instructions_sample_period) { 1614 /* 1615 * Emit instruction sample periodically 1616 * TODO: allow period to be defined in cycles and clock time 1617 */ 1618 1619 /* 1620 * Below diagram demonstrates the instruction samples 1621 * generation flows: 1622 * 1623 * Instrs Instrs Instrs Instrs 1624 * Sample(n) Sample(n+1) Sample(n+2) Sample(n+3) 1625 * | | | | 1626 * V V V V 1627 * -------------------------------------------------- 1628 * ^ ^ 1629 * | | 1630 * Period Period 1631 * instructions(Pi) instructions(Pi') 1632 * 1633 * | | 1634 * \---------------- -----------------/ 1635 * V 1636 * tidq->packet->instr_count 1637 * 1638 * Instrs Sample(n...) are the synthesised samples occurring 1639 * every etm->instructions_sample_period instructions - as 1640 * defined on the perf command line. Sample(n) is being the 1641 * last sample before the current etm packet, n+1 to n+3 1642 * samples are generated from the current etm packet. 1643 * 1644 * tidq->packet->instr_count represents the number of 1645 * instructions in the current etm packet. 1646 * 1647 * Period instructions (Pi) contains the number of 1648 * instructions executed after the sample point(n) from the 1649 * previous etm packet. This will always be less than 1650 * etm->instructions_sample_period. 1651 * 1652 * When generate new samples, it combines with two parts 1653 * instructions, one is the tail of the old packet and another 1654 * is the head of the new coming packet, to generate 1655 * sample(n+1); sample(n+2) and sample(n+3) consume the 1656 * instructions with sample period. After sample(n+3), the rest 1657 * instructions will be used by later packet and it is assigned 1658 * to tidq->period_instructions for next round calculation. 1659 */ 1660 1661 /* 1662 * Get the initial offset into the current packet instructions; 1663 * entry conditions ensure that instrs_prev is less than 1664 * etm->instructions_sample_period. 1665 */ 1666 u64 offset = etm->instructions_sample_period - instrs_prev; 1667 u64 addr; 1668 1669 /* Prepare last branches for instruction sample */ 1670 if (etm->synth_opts.last_branch) 1671 cs_etm__copy_last_branch_rb(etmq, tidq); 1672 1673 while (tidq->period_instructions >= 1674 etm->instructions_sample_period) { 1675 /* 1676 * Calculate the address of the sampled instruction (-1 1677 * as sample is reported as though instruction has just 1678 * been executed, but PC has not advanced to next 1679 * instruction) 1680 */ 1681 addr = cs_etm__instr_addr(etmq, trace_chan_id, 1682 tidq->packet, offset - 1); 1683 ret = cs_etm__synth_instruction_sample( 1684 etmq, tidq, addr, 1685 etm->instructions_sample_period); 1686 if (ret) 1687 return ret; 1688 1689 offset += etm->instructions_sample_period; 1690 tidq->period_instructions -= 1691 etm->instructions_sample_period; 1692 } 1693 } 1694 1695 if (etm->synth_opts.branches) { 1696 bool generate_sample = false; 1697 1698 /* Generate sample for tracing on packet */ 1699 if (tidq->prev_packet->sample_type == CS_ETM_DISCONTINUITY) 1700 generate_sample = true; 1701 1702 /* Generate sample for branch taken packet */ 1703 if (tidq->prev_packet->sample_type == CS_ETM_RANGE && 1704 tidq->prev_packet->last_instr_taken_branch) 1705 generate_sample = true; 1706 1707 if (generate_sample) { 1708 ret = cs_etm__synth_branch_sample(etmq, tidq); 1709 if (ret) 1710 return ret; 1711 } 1712 } 1713 1714 cs_etm__packet_swap(etm, tidq); 1715 1716 return 0; 1717 } 1718 1719 static int cs_etm__exception(struct cs_etm_traceid_queue *tidq) 1720 { 1721 /* 1722 * When the exception packet is inserted, whether the last instruction 1723 * in previous range packet is taken branch or not, we need to force 1724 * to set 'prev_packet->last_instr_taken_branch' to true. This ensures 1725 * to generate branch sample for the instruction range before the 1726 * exception is trapped to kernel or before the exception returning. 1727 * 1728 * The exception packet includes the dummy address values, so don't 1729 * swap PACKET with PREV_PACKET. This keeps PREV_PACKET to be useful 1730 * for generating instruction and branch samples. 1731 */ 1732 if (tidq->prev_packet->sample_type == CS_ETM_RANGE) 1733 tidq->prev_packet->last_instr_taken_branch = true; 1734 1735 return 0; 1736 } 1737 1738 static int cs_etm__flush(struct cs_etm_queue *etmq, 1739 struct cs_etm_traceid_queue *tidq) 1740 { 1741 int err = 0; 1742 struct cs_etm_auxtrace *etm = etmq->etm; 1743 1744 /* Handle start tracing packet */ 1745 if (tidq->prev_packet->sample_type == CS_ETM_EMPTY) 1746 goto swap_packet; 1747 1748 if (etmq->etm->synth_opts.last_branch && 1749 etmq->etm->synth_opts.instructions && 1750 tidq->prev_packet->sample_type == CS_ETM_RANGE) { 1751 u64 addr; 1752 1753 /* Prepare last branches for instruction sample */ 1754 cs_etm__copy_last_branch_rb(etmq, tidq); 1755 1756 /* 1757 * Generate a last branch event for the branches left in the 1758 * circular buffer at the end of the trace. 1759 * 1760 * Use the address of the end of the last reported execution 1761 * range 1762 */ 1763 addr = cs_etm__last_executed_instr(tidq->prev_packet); 1764 1765 err = cs_etm__synth_instruction_sample( 1766 etmq, tidq, addr, 1767 tidq->period_instructions); 1768 if (err) 1769 return err; 1770 1771 tidq->period_instructions = 0; 1772 1773 } 1774 1775 if (etm->synth_opts.branches && 1776 tidq->prev_packet->sample_type == CS_ETM_RANGE) { 1777 err = cs_etm__synth_branch_sample(etmq, tidq); 1778 if (err) 1779 return err; 1780 } 1781 1782 swap_packet: 1783 cs_etm__packet_swap(etm, tidq); 1784 1785 /* Reset last branches after flush the trace */ 1786 if (etm->synth_opts.last_branch) 1787 cs_etm__reset_last_branch_rb(tidq); 1788 1789 return err; 1790 } 1791 1792 static int cs_etm__end_block(struct cs_etm_queue *etmq, 1793 struct cs_etm_traceid_queue *tidq) 1794 { 1795 int err; 1796 1797 /* 1798 * It has no new packet coming and 'etmq->packet' contains the stale 1799 * packet which was set at the previous time with packets swapping; 1800 * so skip to generate branch sample to avoid stale packet. 1801 * 1802 * For this case only flush branch stack and generate a last branch 1803 * event for the branches left in the circular buffer at the end of 1804 * the trace. 1805 */ 1806 if (etmq->etm->synth_opts.last_branch && 1807 etmq->etm->synth_opts.instructions && 1808 tidq->prev_packet->sample_type == CS_ETM_RANGE) { 1809 u64 addr; 1810 1811 /* Prepare last branches for instruction sample */ 1812 cs_etm__copy_last_branch_rb(etmq, tidq); 1813 1814 /* 1815 * Use the address of the end of the last reported execution 1816 * range. 1817 */ 1818 addr = cs_etm__last_executed_instr(tidq->prev_packet); 1819 1820 err = cs_etm__synth_instruction_sample( 1821 etmq, tidq, addr, 1822 tidq->period_instructions); 1823 if (err) 1824 return err; 1825 1826 tidq->period_instructions = 0; 1827 } 1828 1829 return 0; 1830 } 1831 /* 1832 * cs_etm__get_data_block: Fetch a block from the auxtrace_buffer queue 1833 * if need be. 1834 * Returns: < 0 if error 1835 * = 0 if no more auxtrace_buffer to read 1836 * > 0 if the current buffer isn't empty yet 1837 */ 1838 static int cs_etm__get_data_block(struct cs_etm_queue *etmq) 1839 { 1840 int ret; 1841 1842 if (!etmq->buf_len) { 1843 ret = cs_etm__get_trace(etmq); 1844 if (ret <= 0) 1845 return ret; 1846 /* 1847 * We cannot assume consecutive blocks in the data file 1848 * are contiguous, reset the decoder to force re-sync. 1849 */ 1850 ret = cs_etm_decoder__reset(etmq->decoder); 1851 if (ret) 1852 return ret; 1853 } 1854 1855 return etmq->buf_len; 1856 } 1857 1858 static bool cs_etm__is_svc_instr(struct cs_etm_queue *etmq, u8 trace_chan_id, 1859 struct cs_etm_packet *packet, 1860 u64 end_addr) 1861 { 1862 /* Initialise to keep compiler happy */ 1863 u16 instr16 = 0; 1864 u32 instr32 = 0; 1865 u64 addr; 1866 1867 switch (packet->isa) { 1868 case CS_ETM_ISA_T32: 1869 /* 1870 * The SVC of T32 is defined in ARM DDI 0487D.a, F5.1.247: 1871 * 1872 * b'15 b'8 1873 * +-----------------+--------+ 1874 * | 1 1 0 1 1 1 1 1 | imm8 | 1875 * +-----------------+--------+ 1876 * 1877 * According to the specification, it only defines SVC for T32 1878 * with 16 bits instruction and has no definition for 32bits; 1879 * so below only read 2 bytes as instruction size for T32. 1880 */ 1881 addr = end_addr - 2; 1882 cs_etm__mem_access(etmq, trace_chan_id, addr, 1883 sizeof(instr16), (u8 *)&instr16); 1884 if ((instr16 & 0xFF00) == 0xDF00) 1885 return true; 1886 1887 break; 1888 case CS_ETM_ISA_A32: 1889 /* 1890 * The SVC of A32 is defined in ARM DDI 0487D.a, F5.1.247: 1891 * 1892 * b'31 b'28 b'27 b'24 1893 * +---------+---------+-------------------------+ 1894 * | !1111 | 1 1 1 1 | imm24 | 1895 * +---------+---------+-------------------------+ 1896 */ 1897 addr = end_addr - 4; 1898 cs_etm__mem_access(etmq, trace_chan_id, addr, 1899 sizeof(instr32), (u8 *)&instr32); 1900 if ((instr32 & 0x0F000000) == 0x0F000000 && 1901 (instr32 & 0xF0000000) != 0xF0000000) 1902 return true; 1903 1904 break; 1905 case CS_ETM_ISA_A64: 1906 /* 1907 * The SVC of A64 is defined in ARM DDI 0487D.a, C6.2.294: 1908 * 1909 * b'31 b'21 b'4 b'0 1910 * +-----------------------+---------+-----------+ 1911 * | 1 1 0 1 0 1 0 0 0 0 0 | imm16 | 0 0 0 0 1 | 1912 * +-----------------------+---------+-----------+ 1913 */ 1914 addr = end_addr - 4; 1915 cs_etm__mem_access(etmq, trace_chan_id, addr, 1916 sizeof(instr32), (u8 *)&instr32); 1917 if ((instr32 & 0xFFE0001F) == 0xd4000001) 1918 return true; 1919 1920 break; 1921 case CS_ETM_ISA_UNKNOWN: 1922 default: 1923 break; 1924 } 1925 1926 return false; 1927 } 1928 1929 static bool cs_etm__is_syscall(struct cs_etm_queue *etmq, 1930 struct cs_etm_traceid_queue *tidq, u64 magic) 1931 { 1932 u8 trace_chan_id = tidq->trace_chan_id; 1933 struct cs_etm_packet *packet = tidq->packet; 1934 struct cs_etm_packet *prev_packet = tidq->prev_packet; 1935 1936 if (magic == __perf_cs_etmv3_magic) 1937 if (packet->exception_number == CS_ETMV3_EXC_SVC) 1938 return true; 1939 1940 /* 1941 * ETMv4 exception type CS_ETMV4_EXC_CALL covers SVC, SMC and 1942 * HVC cases; need to check if it's SVC instruction based on 1943 * packet address. 1944 */ 1945 if (magic == __perf_cs_etmv4_magic) { 1946 if (packet->exception_number == CS_ETMV4_EXC_CALL && 1947 cs_etm__is_svc_instr(etmq, trace_chan_id, prev_packet, 1948 prev_packet->end_addr)) 1949 return true; 1950 } 1951 1952 return false; 1953 } 1954 1955 static bool cs_etm__is_async_exception(struct cs_etm_traceid_queue *tidq, 1956 u64 magic) 1957 { 1958 struct cs_etm_packet *packet = tidq->packet; 1959 1960 if (magic == __perf_cs_etmv3_magic) 1961 if (packet->exception_number == CS_ETMV3_EXC_DEBUG_HALT || 1962 packet->exception_number == CS_ETMV3_EXC_ASYNC_DATA_ABORT || 1963 packet->exception_number == CS_ETMV3_EXC_PE_RESET || 1964 packet->exception_number == CS_ETMV3_EXC_IRQ || 1965 packet->exception_number == CS_ETMV3_EXC_FIQ) 1966 return true; 1967 1968 if (magic == __perf_cs_etmv4_magic) 1969 if (packet->exception_number == CS_ETMV4_EXC_RESET || 1970 packet->exception_number == CS_ETMV4_EXC_DEBUG_HALT || 1971 packet->exception_number == CS_ETMV4_EXC_SYSTEM_ERROR || 1972 packet->exception_number == CS_ETMV4_EXC_INST_DEBUG || 1973 packet->exception_number == CS_ETMV4_EXC_DATA_DEBUG || 1974 packet->exception_number == CS_ETMV4_EXC_IRQ || 1975 packet->exception_number == CS_ETMV4_EXC_FIQ) 1976 return true; 1977 1978 return false; 1979 } 1980 1981 static bool cs_etm__is_sync_exception(struct cs_etm_queue *etmq, 1982 struct cs_etm_traceid_queue *tidq, 1983 u64 magic) 1984 { 1985 u8 trace_chan_id = tidq->trace_chan_id; 1986 struct cs_etm_packet *packet = tidq->packet; 1987 struct cs_etm_packet *prev_packet = tidq->prev_packet; 1988 1989 if (magic == __perf_cs_etmv3_magic) 1990 if (packet->exception_number == CS_ETMV3_EXC_SMC || 1991 packet->exception_number == CS_ETMV3_EXC_HYP || 1992 packet->exception_number == CS_ETMV3_EXC_JAZELLE_THUMBEE || 1993 packet->exception_number == CS_ETMV3_EXC_UNDEFINED_INSTR || 1994 packet->exception_number == CS_ETMV3_EXC_PREFETCH_ABORT || 1995 packet->exception_number == CS_ETMV3_EXC_DATA_FAULT || 1996 packet->exception_number == CS_ETMV3_EXC_GENERIC) 1997 return true; 1998 1999 if (magic == __perf_cs_etmv4_magic) { 2000 if (packet->exception_number == CS_ETMV4_EXC_TRAP || 2001 packet->exception_number == CS_ETMV4_EXC_ALIGNMENT || 2002 packet->exception_number == CS_ETMV4_EXC_INST_FAULT || 2003 packet->exception_number == CS_ETMV4_EXC_DATA_FAULT) 2004 return true; 2005 2006 /* 2007 * For CS_ETMV4_EXC_CALL, except SVC other instructions 2008 * (SMC, HVC) are taken as sync exceptions. 2009 */ 2010 if (packet->exception_number == CS_ETMV4_EXC_CALL && 2011 !cs_etm__is_svc_instr(etmq, trace_chan_id, prev_packet, 2012 prev_packet->end_addr)) 2013 return true; 2014 2015 /* 2016 * ETMv4 has 5 bits for exception number; if the numbers 2017 * are in the range ( CS_ETMV4_EXC_FIQ, CS_ETMV4_EXC_END ] 2018 * they are implementation defined exceptions. 2019 * 2020 * For this case, simply take it as sync exception. 2021 */ 2022 if (packet->exception_number > CS_ETMV4_EXC_FIQ && 2023 packet->exception_number <= CS_ETMV4_EXC_END) 2024 return true; 2025 } 2026 2027 return false; 2028 } 2029 2030 static int cs_etm__set_sample_flags(struct cs_etm_queue *etmq, 2031 struct cs_etm_traceid_queue *tidq) 2032 { 2033 struct cs_etm_packet *packet = tidq->packet; 2034 struct cs_etm_packet *prev_packet = tidq->prev_packet; 2035 u8 trace_chan_id = tidq->trace_chan_id; 2036 u64 magic; 2037 int ret; 2038 2039 switch (packet->sample_type) { 2040 case CS_ETM_RANGE: 2041 /* 2042 * Immediate branch instruction without neither link nor 2043 * return flag, it's normal branch instruction within 2044 * the function. 2045 */ 2046 if (packet->last_instr_type == OCSD_INSTR_BR && 2047 packet->last_instr_subtype == OCSD_S_INSTR_NONE) { 2048 packet->flags = PERF_IP_FLAG_BRANCH; 2049 2050 if (packet->last_instr_cond) 2051 packet->flags |= PERF_IP_FLAG_CONDITIONAL; 2052 } 2053 2054 /* 2055 * Immediate branch instruction with link (e.g. BL), this is 2056 * branch instruction for function call. 2057 */ 2058 if (packet->last_instr_type == OCSD_INSTR_BR && 2059 packet->last_instr_subtype == OCSD_S_INSTR_BR_LINK) 2060 packet->flags = PERF_IP_FLAG_BRANCH | 2061 PERF_IP_FLAG_CALL; 2062 2063 /* 2064 * Indirect branch instruction with link (e.g. BLR), this is 2065 * branch instruction for function call. 2066 */ 2067 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT && 2068 packet->last_instr_subtype == OCSD_S_INSTR_BR_LINK) 2069 packet->flags = PERF_IP_FLAG_BRANCH | 2070 PERF_IP_FLAG_CALL; 2071 2072 /* 2073 * Indirect branch instruction with subtype of 2074 * OCSD_S_INSTR_V7_IMPLIED_RET, this is explicit hint for 2075 * function return for A32/T32. 2076 */ 2077 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT && 2078 packet->last_instr_subtype == OCSD_S_INSTR_V7_IMPLIED_RET) 2079 packet->flags = PERF_IP_FLAG_BRANCH | 2080 PERF_IP_FLAG_RETURN; 2081 2082 /* 2083 * Indirect branch instruction without link (e.g. BR), usually 2084 * this is used for function return, especially for functions 2085 * within dynamic link lib. 2086 */ 2087 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT && 2088 packet->last_instr_subtype == OCSD_S_INSTR_NONE) 2089 packet->flags = PERF_IP_FLAG_BRANCH | 2090 PERF_IP_FLAG_RETURN; 2091 2092 /* Return instruction for function return. */ 2093 if (packet->last_instr_type == OCSD_INSTR_BR_INDIRECT && 2094 packet->last_instr_subtype == OCSD_S_INSTR_V8_RET) 2095 packet->flags = PERF_IP_FLAG_BRANCH | 2096 PERF_IP_FLAG_RETURN; 2097 2098 /* 2099 * Decoder might insert a discontinuity in the middle of 2100 * instruction packets, fixup prev_packet with flag 2101 * PERF_IP_FLAG_TRACE_BEGIN to indicate restarting trace. 2102 */ 2103 if (prev_packet->sample_type == CS_ETM_DISCONTINUITY) 2104 prev_packet->flags |= PERF_IP_FLAG_BRANCH | 2105 PERF_IP_FLAG_TRACE_BEGIN; 2106 2107 /* 2108 * If the previous packet is an exception return packet 2109 * and the return address just follows SVC instruction, 2110 * it needs to calibrate the previous packet sample flags 2111 * as PERF_IP_FLAG_SYSCALLRET. 2112 */ 2113 if (prev_packet->flags == (PERF_IP_FLAG_BRANCH | 2114 PERF_IP_FLAG_RETURN | 2115 PERF_IP_FLAG_INTERRUPT) && 2116 cs_etm__is_svc_instr(etmq, trace_chan_id, 2117 packet, packet->start_addr)) 2118 prev_packet->flags = PERF_IP_FLAG_BRANCH | 2119 PERF_IP_FLAG_RETURN | 2120 PERF_IP_FLAG_SYSCALLRET; 2121 break; 2122 case CS_ETM_DISCONTINUITY: 2123 /* 2124 * The trace is discontinuous, if the previous packet is 2125 * instruction packet, set flag PERF_IP_FLAG_TRACE_END 2126 * for previous packet. 2127 */ 2128 if (prev_packet->sample_type == CS_ETM_RANGE) 2129 prev_packet->flags |= PERF_IP_FLAG_BRANCH | 2130 PERF_IP_FLAG_TRACE_END; 2131 break; 2132 case CS_ETM_EXCEPTION: 2133 ret = cs_etm__get_magic(packet->trace_chan_id, &magic); 2134 if (ret) 2135 return ret; 2136 2137 /* The exception is for system call. */ 2138 if (cs_etm__is_syscall(etmq, tidq, magic)) 2139 packet->flags = PERF_IP_FLAG_BRANCH | 2140 PERF_IP_FLAG_CALL | 2141 PERF_IP_FLAG_SYSCALLRET; 2142 /* 2143 * The exceptions are triggered by external signals from bus, 2144 * interrupt controller, debug module, PE reset or halt. 2145 */ 2146 else if (cs_etm__is_async_exception(tidq, magic)) 2147 packet->flags = PERF_IP_FLAG_BRANCH | 2148 PERF_IP_FLAG_CALL | 2149 PERF_IP_FLAG_ASYNC | 2150 PERF_IP_FLAG_INTERRUPT; 2151 /* 2152 * Otherwise, exception is caused by trap, instruction & 2153 * data fault, or alignment errors. 2154 */ 2155 else if (cs_etm__is_sync_exception(etmq, tidq, magic)) 2156 packet->flags = PERF_IP_FLAG_BRANCH | 2157 PERF_IP_FLAG_CALL | 2158 PERF_IP_FLAG_INTERRUPT; 2159 2160 /* 2161 * When the exception packet is inserted, since exception 2162 * packet is not used standalone for generating samples 2163 * and it's affiliation to the previous instruction range 2164 * packet; so set previous range packet flags to tell perf 2165 * it is an exception taken branch. 2166 */ 2167 if (prev_packet->sample_type == CS_ETM_RANGE) 2168 prev_packet->flags = packet->flags; 2169 break; 2170 case CS_ETM_EXCEPTION_RET: 2171 /* 2172 * When the exception return packet is inserted, since 2173 * exception return packet is not used standalone for 2174 * generating samples and it's affiliation to the previous 2175 * instruction range packet; so set previous range packet 2176 * flags to tell perf it is an exception return branch. 2177 * 2178 * The exception return can be for either system call or 2179 * other exception types; unfortunately the packet doesn't 2180 * contain exception type related info so we cannot decide 2181 * the exception type purely based on exception return packet. 2182 * If we record the exception number from exception packet and 2183 * reuse it for exception return packet, this is not reliable 2184 * due the trace can be discontinuity or the interrupt can 2185 * be nested, thus the recorded exception number cannot be 2186 * used for exception return packet for these two cases. 2187 * 2188 * For exception return packet, we only need to distinguish the 2189 * packet is for system call or for other types. Thus the 2190 * decision can be deferred when receive the next packet which 2191 * contains the return address, based on the return address we 2192 * can read out the previous instruction and check if it's a 2193 * system call instruction and then calibrate the sample flag 2194 * as needed. 2195 */ 2196 if (prev_packet->sample_type == CS_ETM_RANGE) 2197 prev_packet->flags = PERF_IP_FLAG_BRANCH | 2198 PERF_IP_FLAG_RETURN | 2199 PERF_IP_FLAG_INTERRUPT; 2200 break; 2201 case CS_ETM_EMPTY: 2202 default: 2203 break; 2204 } 2205 2206 return 0; 2207 } 2208 2209 static int cs_etm__decode_data_block(struct cs_etm_queue *etmq) 2210 { 2211 int ret = 0; 2212 size_t processed = 0; 2213 2214 /* 2215 * Packets are decoded and added to the decoder's packet queue 2216 * until the decoder packet processing callback has requested that 2217 * processing stops or there is nothing left in the buffer. Normal 2218 * operations that stop processing are a timestamp packet or a full 2219 * decoder buffer queue. 2220 */ 2221 ret = cs_etm_decoder__process_data_block(etmq->decoder, 2222 etmq->offset, 2223 &etmq->buf[etmq->buf_used], 2224 etmq->buf_len, 2225 &processed); 2226 if (ret) 2227 goto out; 2228 2229 etmq->offset += processed; 2230 etmq->buf_used += processed; 2231 etmq->buf_len -= processed; 2232 2233 out: 2234 return ret; 2235 } 2236 2237 static int cs_etm__process_traceid_queue(struct cs_etm_queue *etmq, 2238 struct cs_etm_traceid_queue *tidq) 2239 { 2240 int ret; 2241 struct cs_etm_packet_queue *packet_queue; 2242 2243 packet_queue = &tidq->packet_queue; 2244 2245 /* Process each packet in this chunk */ 2246 while (1) { 2247 ret = cs_etm_decoder__get_packet(packet_queue, 2248 tidq->packet); 2249 if (ret <= 0) 2250 /* 2251 * Stop processing this chunk on 2252 * end of data or error 2253 */ 2254 break; 2255 2256 /* 2257 * Since packet addresses are swapped in packet 2258 * handling within below switch() statements, 2259 * thus setting sample flags must be called 2260 * prior to switch() statement to use address 2261 * information before packets swapping. 2262 */ 2263 ret = cs_etm__set_sample_flags(etmq, tidq); 2264 if (ret < 0) 2265 break; 2266 2267 switch (tidq->packet->sample_type) { 2268 case CS_ETM_RANGE: 2269 /* 2270 * If the packet contains an instruction 2271 * range, generate instruction sequence 2272 * events. 2273 */ 2274 cs_etm__sample(etmq, tidq); 2275 break; 2276 case CS_ETM_EXCEPTION: 2277 case CS_ETM_EXCEPTION_RET: 2278 /* 2279 * If the exception packet is coming, 2280 * make sure the previous instruction 2281 * range packet to be handled properly. 2282 */ 2283 cs_etm__exception(tidq); 2284 break; 2285 case CS_ETM_DISCONTINUITY: 2286 /* 2287 * Discontinuity in trace, flush 2288 * previous branch stack 2289 */ 2290 cs_etm__flush(etmq, tidq); 2291 break; 2292 case CS_ETM_EMPTY: 2293 /* 2294 * Should not receive empty packet, 2295 * report error. 2296 */ 2297 pr_err("CS ETM Trace: empty packet\n"); 2298 return -EINVAL; 2299 default: 2300 break; 2301 } 2302 } 2303 2304 return ret; 2305 } 2306 2307 static void cs_etm__clear_all_traceid_queues(struct cs_etm_queue *etmq) 2308 { 2309 int idx; 2310 struct int_node *inode; 2311 struct cs_etm_traceid_queue *tidq; 2312 struct intlist *traceid_queues_list = etmq->traceid_queues_list; 2313 2314 intlist__for_each_entry(inode, traceid_queues_list) { 2315 idx = (int)(intptr_t)inode->priv; 2316 tidq = etmq->traceid_queues[idx]; 2317 2318 /* Ignore return value */ 2319 cs_etm__process_traceid_queue(etmq, tidq); 2320 2321 /* 2322 * Generate an instruction sample with the remaining 2323 * branchstack entries. 2324 */ 2325 cs_etm__flush(etmq, tidq); 2326 } 2327 } 2328 2329 static int cs_etm__run_decoder(struct cs_etm_queue *etmq) 2330 { 2331 int err = 0; 2332 struct cs_etm_traceid_queue *tidq; 2333 2334 tidq = cs_etm__etmq_get_traceid_queue(etmq, CS_ETM_PER_THREAD_TRACEID); 2335 if (!tidq) 2336 return -EINVAL; 2337 2338 /* Go through each buffer in the queue and decode them one by one */ 2339 while (1) { 2340 err = cs_etm__get_data_block(etmq); 2341 if (err <= 0) 2342 return err; 2343 2344 /* Run trace decoder until buffer consumed or end of trace */ 2345 do { 2346 err = cs_etm__decode_data_block(etmq); 2347 if (err) 2348 return err; 2349 2350 /* 2351 * Process each packet in this chunk, nothing to do if 2352 * an error occurs other than hoping the next one will 2353 * be better. 2354 */ 2355 err = cs_etm__process_traceid_queue(etmq, tidq); 2356 2357 } while (etmq->buf_len); 2358 2359 if (err == 0) 2360 /* Flush any remaining branch stack entries */ 2361 err = cs_etm__end_block(etmq, tidq); 2362 } 2363 2364 return err; 2365 } 2366 2367 static int cs_etm__process_timeless_queues(struct cs_etm_auxtrace *etm, 2368 pid_t tid) 2369 { 2370 unsigned int i; 2371 struct auxtrace_queues *queues = &etm->queues; 2372 2373 for (i = 0; i < queues->nr_queues; i++) { 2374 struct auxtrace_queue *queue = &etm->queues.queue_array[i]; 2375 struct cs_etm_queue *etmq = queue->priv; 2376 struct cs_etm_traceid_queue *tidq; 2377 2378 if (!etmq) 2379 continue; 2380 2381 tidq = cs_etm__etmq_get_traceid_queue(etmq, 2382 CS_ETM_PER_THREAD_TRACEID); 2383 2384 if (!tidq) 2385 continue; 2386 2387 if ((tid == -1) || (tidq->tid == tid)) { 2388 cs_etm__set_pid_tid_cpu(etm, tidq); 2389 cs_etm__run_decoder(etmq); 2390 } 2391 } 2392 2393 return 0; 2394 } 2395 2396 static int cs_etm__process_queues(struct cs_etm_auxtrace *etm) 2397 { 2398 int ret = 0; 2399 unsigned int cs_queue_nr, queue_nr, i; 2400 u8 trace_chan_id; 2401 u64 cs_timestamp; 2402 struct auxtrace_queue *queue; 2403 struct cs_etm_queue *etmq; 2404 struct cs_etm_traceid_queue *tidq; 2405 2406 /* 2407 * Pre-populate the heap with one entry from each queue so that we can 2408 * start processing in time order across all queues. 2409 */ 2410 for (i = 0; i < etm->queues.nr_queues; i++) { 2411 etmq = etm->queues.queue_array[i].priv; 2412 if (!etmq) 2413 continue; 2414 2415 ret = cs_etm__queue_first_cs_timestamp(etm, etmq, i); 2416 if (ret) 2417 return ret; 2418 } 2419 2420 while (1) { 2421 if (!etm->heap.heap_cnt) 2422 goto out; 2423 2424 /* Take the entry at the top of the min heap */ 2425 cs_queue_nr = etm->heap.heap_array[0].queue_nr; 2426 queue_nr = TO_QUEUE_NR(cs_queue_nr); 2427 trace_chan_id = TO_TRACE_CHAN_ID(cs_queue_nr); 2428 queue = &etm->queues.queue_array[queue_nr]; 2429 etmq = queue->priv; 2430 2431 /* 2432 * Remove the top entry from the heap since we are about 2433 * to process it. 2434 */ 2435 auxtrace_heap__pop(&etm->heap); 2436 2437 tidq = cs_etm__etmq_get_traceid_queue(etmq, trace_chan_id); 2438 if (!tidq) { 2439 /* 2440 * No traceID queue has been allocated for this traceID, 2441 * which means something somewhere went very wrong. No 2442 * other choice than simply exit. 2443 */ 2444 ret = -EINVAL; 2445 goto out; 2446 } 2447 2448 /* 2449 * Packets associated with this timestamp are already in 2450 * the etmq's traceID queue, so process them. 2451 */ 2452 ret = cs_etm__process_traceid_queue(etmq, tidq); 2453 if (ret < 0) 2454 goto out; 2455 2456 /* 2457 * Packets for this timestamp have been processed, time to 2458 * move on to the next timestamp, fetching a new auxtrace_buffer 2459 * if need be. 2460 */ 2461 refetch: 2462 ret = cs_etm__get_data_block(etmq); 2463 if (ret < 0) 2464 goto out; 2465 2466 /* 2467 * No more auxtrace_buffers to process in this etmq, simply 2468 * move on to another entry in the auxtrace_heap. 2469 */ 2470 if (!ret) 2471 continue; 2472 2473 ret = cs_etm__decode_data_block(etmq); 2474 if (ret) 2475 goto out; 2476 2477 cs_timestamp = cs_etm__etmq_get_timestamp(etmq, &trace_chan_id); 2478 2479 if (!cs_timestamp) { 2480 /* 2481 * Function cs_etm__decode_data_block() returns when 2482 * there is no more traces to decode in the current 2483 * auxtrace_buffer OR when a timestamp has been 2484 * encountered on any of the traceID queues. Since we 2485 * did not get a timestamp, there is no more traces to 2486 * process in this auxtrace_buffer. As such empty and 2487 * flush all traceID queues. 2488 */ 2489 cs_etm__clear_all_traceid_queues(etmq); 2490 2491 /* Fetch another auxtrace_buffer for this etmq */ 2492 goto refetch; 2493 } 2494 2495 /* 2496 * Add to the min heap the timestamp for packets that have 2497 * just been decoded. They will be processed and synthesized 2498 * during the next call to cs_etm__process_traceid_queue() for 2499 * this queue/traceID. 2500 */ 2501 cs_queue_nr = TO_CS_QUEUE_NR(queue_nr, trace_chan_id); 2502 ret = auxtrace_heap__add(&etm->heap, cs_queue_nr, cs_timestamp); 2503 } 2504 2505 out: 2506 return ret; 2507 } 2508 2509 static int cs_etm__process_itrace_start(struct cs_etm_auxtrace *etm, 2510 union perf_event *event) 2511 { 2512 struct thread *th; 2513 2514 if (etm->timeless_decoding) 2515 return 0; 2516 2517 /* 2518 * Add the tid/pid to the log so that we can get a match when 2519 * we get a contextID from the decoder. 2520 */ 2521 th = machine__findnew_thread(etm->machine, 2522 event->itrace_start.pid, 2523 event->itrace_start.tid); 2524 if (!th) 2525 return -ENOMEM; 2526 2527 thread__put(th); 2528 2529 return 0; 2530 } 2531 2532 static int cs_etm__process_switch_cpu_wide(struct cs_etm_auxtrace *etm, 2533 union perf_event *event) 2534 { 2535 struct thread *th; 2536 bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT; 2537 2538 /* 2539 * Context switch in per-thread mode are irrelevant since perf 2540 * will start/stop tracing as the process is scheduled. 2541 */ 2542 if (etm->timeless_decoding) 2543 return 0; 2544 2545 /* 2546 * SWITCH_IN events carry the next process to be switched out while 2547 * SWITCH_OUT events carry the process to be switched in. As such 2548 * we don't care about IN events. 2549 */ 2550 if (!out) 2551 return 0; 2552 2553 /* 2554 * Add the tid/pid to the log so that we can get a match when 2555 * we get a contextID from the decoder. 2556 */ 2557 th = machine__findnew_thread(etm->machine, 2558 event->context_switch.next_prev_pid, 2559 event->context_switch.next_prev_tid); 2560 if (!th) 2561 return -ENOMEM; 2562 2563 thread__put(th); 2564 2565 return 0; 2566 } 2567 2568 static int cs_etm__process_event(struct perf_session *session, 2569 union perf_event *event, 2570 struct perf_sample *sample, 2571 struct perf_tool *tool) 2572 { 2573 u64 sample_kernel_timestamp; 2574 struct cs_etm_auxtrace *etm = container_of(session->auxtrace, 2575 struct cs_etm_auxtrace, 2576 auxtrace); 2577 2578 if (dump_trace) 2579 return 0; 2580 2581 if (!tool->ordered_events) { 2582 pr_err("CoreSight ETM Trace requires ordered events\n"); 2583 return -EINVAL; 2584 } 2585 2586 if (sample->time && (sample->time != (u64) -1)) 2587 sample_kernel_timestamp = sample->time; 2588 else 2589 sample_kernel_timestamp = 0; 2590 2591 /* 2592 * Don't wait for cs_etm__flush_events() in per-thread/timeless mode to start the decode. We 2593 * need the tid of the PERF_RECORD_EXIT event to assign to the synthesised samples because 2594 * ETM_OPT_CTXTID is not enabled. 2595 */ 2596 if (etm->timeless_decoding && 2597 event->header.type == PERF_RECORD_EXIT) 2598 return cs_etm__process_timeless_queues(etm, 2599 event->fork.tid); 2600 2601 if (event->header.type == PERF_RECORD_ITRACE_START) 2602 return cs_etm__process_itrace_start(etm, event); 2603 else if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) 2604 return cs_etm__process_switch_cpu_wide(etm, event); 2605 2606 if (!etm->timeless_decoding && event->header.type == PERF_RECORD_AUX) { 2607 /* 2608 * Record the latest kernel timestamp available in the header 2609 * for samples so that synthesised samples occur from this point 2610 * onwards. 2611 */ 2612 etm->latest_kernel_timestamp = sample_kernel_timestamp; 2613 } 2614 2615 return 0; 2616 } 2617 2618 static void dump_queued_data(struct cs_etm_auxtrace *etm, 2619 struct perf_record_auxtrace *event) 2620 { 2621 struct auxtrace_buffer *buf; 2622 unsigned int i; 2623 /* 2624 * Find all buffers with same reference in the queues and dump them. 2625 * This is because the queues can contain multiple entries of the same 2626 * buffer that were split on aux records. 2627 */ 2628 for (i = 0; i < etm->queues.nr_queues; ++i) 2629 list_for_each_entry(buf, &etm->queues.queue_array[i].head, list) 2630 if (buf->reference == event->reference) 2631 cs_etm__dump_event(etm->queues.queue_array[i].priv, buf); 2632 } 2633 2634 static int cs_etm__process_auxtrace_event(struct perf_session *session, 2635 union perf_event *event, 2636 struct perf_tool *tool __maybe_unused) 2637 { 2638 struct cs_etm_auxtrace *etm = container_of(session->auxtrace, 2639 struct cs_etm_auxtrace, 2640 auxtrace); 2641 if (!etm->data_queued) { 2642 struct auxtrace_buffer *buffer; 2643 off_t data_offset; 2644 int fd = perf_data__fd(session->data); 2645 bool is_pipe = perf_data__is_pipe(session->data); 2646 int err; 2647 int idx = event->auxtrace.idx; 2648 2649 if (is_pipe) 2650 data_offset = 0; 2651 else { 2652 data_offset = lseek(fd, 0, SEEK_CUR); 2653 if (data_offset == -1) 2654 return -errno; 2655 } 2656 2657 err = auxtrace_queues__add_event(&etm->queues, session, 2658 event, data_offset, &buffer); 2659 if (err) 2660 return err; 2661 2662 /* 2663 * Knowing if the trace is formatted or not requires a lookup of 2664 * the aux record so only works in non-piped mode where data is 2665 * queued in cs_etm__queue_aux_records(). Always assume 2666 * formatted in piped mode (true). 2667 */ 2668 err = cs_etm__setup_queue(etm, &etm->queues.queue_array[idx], 2669 idx, true); 2670 if (err) 2671 return err; 2672 2673 if (dump_trace) 2674 if (auxtrace_buffer__get_data(buffer, fd)) { 2675 cs_etm__dump_event(etm->queues.queue_array[idx].priv, buffer); 2676 auxtrace_buffer__put_data(buffer); 2677 } 2678 } else if (dump_trace) 2679 dump_queued_data(etm, &event->auxtrace); 2680 2681 return 0; 2682 } 2683 2684 static bool cs_etm__is_timeless_decoding(struct cs_etm_auxtrace *etm) 2685 { 2686 struct evsel *evsel; 2687 struct evlist *evlist = etm->session->evlist; 2688 bool timeless_decoding = true; 2689 2690 /* Override timeless mode with user input from --itrace=Z */ 2691 if (etm->synth_opts.timeless_decoding) 2692 return true; 2693 2694 /* 2695 * Circle through the list of event and complain if we find one 2696 * with the time bit set. 2697 */ 2698 evlist__for_each_entry(evlist, evsel) { 2699 if ((evsel->core.attr.sample_type & PERF_SAMPLE_TIME)) 2700 timeless_decoding = false; 2701 } 2702 2703 return timeless_decoding; 2704 } 2705 2706 /* 2707 * Read a single cpu parameter block from the auxtrace_info priv block. 2708 * 2709 * For version 1 there is a per cpu nr_params entry. If we are handling 2710 * version 1 file, then there may be less, the same, or more params 2711 * indicated by this value than the compile time number we understand. 2712 * 2713 * For a version 0 info block, there are a fixed number, and we need to 2714 * fill out the nr_param value in the metadata we create. 2715 */ 2716 static u64 *cs_etm__create_meta_blk(u64 *buff_in, int *buff_in_offset, 2717 int out_blk_size, int nr_params_v0) 2718 { 2719 u64 *metadata = NULL; 2720 int hdr_version; 2721 int nr_in_params, nr_out_params, nr_cmn_params; 2722 int i, k; 2723 2724 metadata = zalloc(sizeof(*metadata) * out_blk_size); 2725 if (!metadata) 2726 return NULL; 2727 2728 /* read block current index & version */ 2729 i = *buff_in_offset; 2730 hdr_version = buff_in[CS_HEADER_VERSION]; 2731 2732 if (!hdr_version) { 2733 /* read version 0 info block into a version 1 metadata block */ 2734 nr_in_params = nr_params_v0; 2735 metadata[CS_ETM_MAGIC] = buff_in[i + CS_ETM_MAGIC]; 2736 metadata[CS_ETM_CPU] = buff_in[i + CS_ETM_CPU]; 2737 metadata[CS_ETM_NR_TRC_PARAMS] = nr_in_params; 2738 /* remaining block params at offset +1 from source */ 2739 for (k = CS_ETM_COMMON_BLK_MAX_V1 - 1; k < nr_in_params; k++) 2740 metadata[k + 1] = buff_in[i + k]; 2741 /* version 0 has 2 common params */ 2742 nr_cmn_params = 2; 2743 } else { 2744 /* read version 1 info block - input and output nr_params may differ */ 2745 /* version 1 has 3 common params */ 2746 nr_cmn_params = 3; 2747 nr_in_params = buff_in[i + CS_ETM_NR_TRC_PARAMS]; 2748 2749 /* if input has more params than output - skip excess */ 2750 nr_out_params = nr_in_params + nr_cmn_params; 2751 if (nr_out_params > out_blk_size) 2752 nr_out_params = out_blk_size; 2753 2754 for (k = CS_ETM_MAGIC; k < nr_out_params; k++) 2755 metadata[k] = buff_in[i + k]; 2756 2757 /* record the actual nr params we copied */ 2758 metadata[CS_ETM_NR_TRC_PARAMS] = nr_out_params - nr_cmn_params; 2759 } 2760 2761 /* adjust in offset by number of in params used */ 2762 i += nr_in_params + nr_cmn_params; 2763 *buff_in_offset = i; 2764 return metadata; 2765 } 2766 2767 /** 2768 * Puts a fragment of an auxtrace buffer into the auxtrace queues based 2769 * on the bounds of aux_event, if it matches with the buffer that's at 2770 * file_offset. 2771 * 2772 * Normally, whole auxtrace buffers would be added to the queue. But we 2773 * want to reset the decoder for every PERF_RECORD_AUX event, and the decoder 2774 * is reset across each buffer, so splitting the buffers up in advance has 2775 * the same effect. 2776 */ 2777 static int cs_etm__queue_aux_fragment(struct perf_session *session, off_t file_offset, size_t sz, 2778 struct perf_record_aux *aux_event, struct perf_sample *sample) 2779 { 2780 int err; 2781 char buf[PERF_SAMPLE_MAX_SIZE]; 2782 union perf_event *auxtrace_event_union; 2783 struct perf_record_auxtrace *auxtrace_event; 2784 union perf_event auxtrace_fragment; 2785 __u64 aux_offset, aux_size; 2786 __u32 idx; 2787 bool formatted; 2788 2789 struct cs_etm_auxtrace *etm = container_of(session->auxtrace, 2790 struct cs_etm_auxtrace, 2791 auxtrace); 2792 2793 /* 2794 * There should be a PERF_RECORD_AUXTRACE event at the file_offset that we got 2795 * from looping through the auxtrace index. 2796 */ 2797 err = perf_session__peek_event(session, file_offset, buf, 2798 PERF_SAMPLE_MAX_SIZE, &auxtrace_event_union, NULL); 2799 if (err) 2800 return err; 2801 auxtrace_event = &auxtrace_event_union->auxtrace; 2802 if (auxtrace_event->header.type != PERF_RECORD_AUXTRACE) 2803 return -EINVAL; 2804 2805 if (auxtrace_event->header.size < sizeof(struct perf_record_auxtrace) || 2806 auxtrace_event->header.size != sz) { 2807 return -EINVAL; 2808 } 2809 2810 /* 2811 * In per-thread mode, auxtrace CPU is set to -1, but TID will be set instead. See 2812 * auxtrace_mmap_params__set_idx(). However, the sample AUX event will contain a 2813 * CPU as we set this always for the AUX_OUTPUT_HW_ID event. 2814 * So now compare only TIDs if auxtrace CPU is -1, and CPUs if auxtrace CPU is not -1. 2815 * Return 'not found' if mismatch. 2816 */ 2817 if (auxtrace_event->cpu == (__u32) -1) { 2818 if (auxtrace_event->tid != sample->tid) 2819 return 1; 2820 } else if (auxtrace_event->cpu != sample->cpu) 2821 return 1; 2822 2823 if (aux_event->flags & PERF_AUX_FLAG_OVERWRITE) { 2824 /* 2825 * Clamp size in snapshot mode. The buffer size is clamped in 2826 * __auxtrace_mmap__read() for snapshots, so the aux record size doesn't reflect 2827 * the buffer size. 2828 */ 2829 aux_size = min(aux_event->aux_size, auxtrace_event->size); 2830 2831 /* 2832 * In this mode, the head also points to the end of the buffer so aux_offset 2833 * needs to have the size subtracted so it points to the beginning as in normal mode 2834 */ 2835 aux_offset = aux_event->aux_offset - aux_size; 2836 } else { 2837 aux_size = aux_event->aux_size; 2838 aux_offset = aux_event->aux_offset; 2839 } 2840 2841 if (aux_offset >= auxtrace_event->offset && 2842 aux_offset + aux_size <= auxtrace_event->offset + auxtrace_event->size) { 2843 /* 2844 * If this AUX event was inside this buffer somewhere, create a new auxtrace event 2845 * based on the sizes of the aux event, and queue that fragment. 2846 */ 2847 auxtrace_fragment.auxtrace = *auxtrace_event; 2848 auxtrace_fragment.auxtrace.size = aux_size; 2849 auxtrace_fragment.auxtrace.offset = aux_offset; 2850 file_offset += aux_offset - auxtrace_event->offset + auxtrace_event->header.size; 2851 2852 pr_debug3("CS ETM: Queue buffer size: %#"PRI_lx64" offset: %#"PRI_lx64 2853 " tid: %d cpu: %d\n", aux_size, aux_offset, sample->tid, sample->cpu); 2854 err = auxtrace_queues__add_event(&etm->queues, session, &auxtrace_fragment, 2855 file_offset, NULL); 2856 if (err) 2857 return err; 2858 2859 idx = auxtrace_event->idx; 2860 formatted = !(aux_event->flags & PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW); 2861 return cs_etm__setup_queue(etm, &etm->queues.queue_array[idx], 2862 idx, formatted); 2863 } 2864 2865 /* Wasn't inside this buffer, but there were no parse errors. 1 == 'not found' */ 2866 return 1; 2867 } 2868 2869 static int cs_etm__process_aux_hw_id_cb(struct perf_session *session, union perf_event *event, 2870 u64 offset __maybe_unused, void *data __maybe_unused) 2871 { 2872 /* look to handle PERF_RECORD_AUX_OUTPUT_HW_ID early to ensure decoders can be set up */ 2873 if (event->header.type == PERF_RECORD_AUX_OUTPUT_HW_ID) { 2874 (*(int *)data)++; /* increment found count */ 2875 return cs_etm__process_aux_output_hw_id(session, event); 2876 } 2877 return 0; 2878 } 2879 2880 static int cs_etm__queue_aux_records_cb(struct perf_session *session, union perf_event *event, 2881 u64 offset __maybe_unused, void *data __maybe_unused) 2882 { 2883 struct perf_sample sample; 2884 int ret; 2885 struct auxtrace_index_entry *ent; 2886 struct auxtrace_index *auxtrace_index; 2887 struct evsel *evsel; 2888 size_t i; 2889 2890 /* Don't care about any other events, we're only queuing buffers for AUX events */ 2891 if (event->header.type != PERF_RECORD_AUX) 2892 return 0; 2893 2894 if (event->header.size < sizeof(struct perf_record_aux)) 2895 return -EINVAL; 2896 2897 /* Truncated Aux records can have 0 size and shouldn't result in anything being queued. */ 2898 if (!event->aux.aux_size) 2899 return 0; 2900 2901 /* 2902 * Parse the sample, we need the sample_id_all data that comes after the event so that the 2903 * CPU or PID can be matched to an AUXTRACE buffer's CPU or PID. 2904 */ 2905 evsel = evlist__event2evsel(session->evlist, event); 2906 if (!evsel) 2907 return -EINVAL; 2908 ret = evsel__parse_sample(evsel, event, &sample); 2909 if (ret) 2910 return ret; 2911 2912 /* 2913 * Loop through the auxtrace index to find the buffer that matches up with this aux event. 2914 */ 2915 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) { 2916 for (i = 0; i < auxtrace_index->nr; i++) { 2917 ent = &auxtrace_index->entries[i]; 2918 ret = cs_etm__queue_aux_fragment(session, ent->file_offset, 2919 ent->sz, &event->aux, &sample); 2920 /* 2921 * Stop search on error or successful values. Continue search on 2922 * 1 ('not found') 2923 */ 2924 if (ret != 1) 2925 return ret; 2926 } 2927 } 2928 2929 /* 2930 * Couldn't find the buffer corresponding to this aux record, something went wrong. Warn but 2931 * don't exit with an error because it will still be possible to decode other aux records. 2932 */ 2933 pr_err("CS ETM: Couldn't find auxtrace buffer for aux_offset: %#"PRI_lx64 2934 " tid: %d cpu: %d\n", event->aux.aux_offset, sample.tid, sample.cpu); 2935 return 0; 2936 } 2937 2938 static int cs_etm__queue_aux_records(struct perf_session *session) 2939 { 2940 struct auxtrace_index *index = list_first_entry_or_null(&session->auxtrace_index, 2941 struct auxtrace_index, list); 2942 if (index && index->nr > 0) 2943 return perf_session__peek_events(session, session->header.data_offset, 2944 session->header.data_size, 2945 cs_etm__queue_aux_records_cb, NULL); 2946 2947 /* 2948 * We would get here if there are no entries in the index (either no auxtrace 2949 * buffers or no index at all). Fail silently as there is the possibility of 2950 * queueing them in cs_etm__process_auxtrace_event() if etm->data_queued is still 2951 * false. 2952 * 2953 * In that scenario, buffers will not be split by AUX records. 2954 */ 2955 return 0; 2956 } 2957 2958 #define HAS_PARAM(j, type, param) (metadata[(j)][CS_ETM_NR_TRC_PARAMS] <= \ 2959 (CS_##type##_##param - CS_ETM_COMMON_BLK_MAX_V1)) 2960 2961 /* 2962 * Loop through the ETMs and complain if we find at least one where ts_source != 1 (virtual 2963 * timestamps). 2964 */ 2965 static bool cs_etm__has_virtual_ts(u64 **metadata, int num_cpu) 2966 { 2967 int j; 2968 2969 for (j = 0; j < num_cpu; j++) { 2970 switch (metadata[j][CS_ETM_MAGIC]) { 2971 case __perf_cs_etmv4_magic: 2972 if (HAS_PARAM(j, ETMV4, TS_SOURCE) || metadata[j][CS_ETMV4_TS_SOURCE] != 1) 2973 return false; 2974 break; 2975 case __perf_cs_ete_magic: 2976 if (HAS_PARAM(j, ETE, TS_SOURCE) || metadata[j][CS_ETE_TS_SOURCE] != 1) 2977 return false; 2978 break; 2979 default: 2980 /* Unknown / unsupported magic number. */ 2981 return false; 2982 } 2983 } 2984 return true; 2985 } 2986 2987 /* map trace ids to correct metadata block, from information in metadata */ 2988 static int cs_etm__map_trace_ids_metadata(int num_cpu, u64 **metadata) 2989 { 2990 u64 cs_etm_magic; 2991 u8 trace_chan_id; 2992 int i, err; 2993 2994 for (i = 0; i < num_cpu; i++) { 2995 cs_etm_magic = metadata[i][CS_ETM_MAGIC]; 2996 switch (cs_etm_magic) { 2997 case __perf_cs_etmv3_magic: 2998 metadata[i][CS_ETM_ETMTRACEIDR] &= CORESIGHT_TRACE_ID_VAL_MASK; 2999 trace_chan_id = (u8)(metadata[i][CS_ETM_ETMTRACEIDR]); 3000 break; 3001 case __perf_cs_etmv4_magic: 3002 case __perf_cs_ete_magic: 3003 metadata[i][CS_ETMV4_TRCTRACEIDR] &= CORESIGHT_TRACE_ID_VAL_MASK; 3004 trace_chan_id = (u8)(metadata[i][CS_ETMV4_TRCTRACEIDR]); 3005 break; 3006 default: 3007 /* unknown magic number */ 3008 return -EINVAL; 3009 } 3010 err = cs_etm__map_trace_id(trace_chan_id, metadata[i]); 3011 if (err) 3012 return err; 3013 } 3014 return 0; 3015 } 3016 3017 /* 3018 * If we found AUX_HW_ID packets, then set any metadata marked as unused to the 3019 * unused value to reduce the number of unneeded decoders created. 3020 */ 3021 static int cs_etm__clear_unused_trace_ids_metadata(int num_cpu, u64 **metadata) 3022 { 3023 u64 cs_etm_magic; 3024 int i; 3025 3026 for (i = 0; i < num_cpu; i++) { 3027 cs_etm_magic = metadata[i][CS_ETM_MAGIC]; 3028 switch (cs_etm_magic) { 3029 case __perf_cs_etmv3_magic: 3030 if (metadata[i][CS_ETM_ETMTRACEIDR] & CORESIGHT_TRACE_ID_UNUSED_FLAG) 3031 metadata[i][CS_ETM_ETMTRACEIDR] = CORESIGHT_TRACE_ID_UNUSED_VAL; 3032 break; 3033 case __perf_cs_etmv4_magic: 3034 case __perf_cs_ete_magic: 3035 if (metadata[i][CS_ETMV4_TRCTRACEIDR] & CORESIGHT_TRACE_ID_UNUSED_FLAG) 3036 metadata[i][CS_ETMV4_TRCTRACEIDR] = CORESIGHT_TRACE_ID_UNUSED_VAL; 3037 break; 3038 default: 3039 /* unknown magic number */ 3040 return -EINVAL; 3041 } 3042 } 3043 return 0; 3044 } 3045 3046 int cs_etm__process_auxtrace_info_full(union perf_event *event, 3047 struct perf_session *session) 3048 { 3049 struct perf_record_auxtrace_info *auxtrace_info = &event->auxtrace_info; 3050 struct cs_etm_auxtrace *etm = NULL; 3051 struct perf_record_time_conv *tc = &session->time_conv; 3052 int event_header_size = sizeof(struct perf_event_header); 3053 int total_size = auxtrace_info->header.size; 3054 int priv_size = 0; 3055 int num_cpu; 3056 int err = 0; 3057 int aux_hw_id_found; 3058 int i, j; 3059 u64 *ptr = NULL; 3060 u64 **metadata = NULL; 3061 3062 /* 3063 * Create an RB tree for traceID-metadata tuple. Since the conversion 3064 * has to be made for each packet that gets decoded, optimizing access 3065 * in anything other than a sequential array is worth doing. 3066 */ 3067 traceid_list = intlist__new(NULL); 3068 if (!traceid_list) 3069 return -ENOMEM; 3070 3071 /* First the global part */ 3072 ptr = (u64 *) auxtrace_info->priv; 3073 num_cpu = ptr[CS_PMU_TYPE_CPUS] & 0xffffffff; 3074 metadata = zalloc(sizeof(*metadata) * num_cpu); 3075 if (!metadata) { 3076 err = -ENOMEM; 3077 goto err_free_traceid_list; 3078 } 3079 3080 /* Start parsing after the common part of the header */ 3081 i = CS_HEADER_VERSION_MAX; 3082 3083 /* 3084 * The metadata is stored in the auxtrace_info section and encodes 3085 * the configuration of the ARM embedded trace macrocell which is 3086 * required by the trace decoder to properly decode the trace due 3087 * to its highly compressed nature. 3088 */ 3089 for (j = 0; j < num_cpu; j++) { 3090 if (ptr[i] == __perf_cs_etmv3_magic) { 3091 metadata[j] = 3092 cs_etm__create_meta_blk(ptr, &i, 3093 CS_ETM_PRIV_MAX, 3094 CS_ETM_NR_TRC_PARAMS_V0); 3095 } else if (ptr[i] == __perf_cs_etmv4_magic) { 3096 metadata[j] = 3097 cs_etm__create_meta_blk(ptr, &i, 3098 CS_ETMV4_PRIV_MAX, 3099 CS_ETMV4_NR_TRC_PARAMS_V0); 3100 } else if (ptr[i] == __perf_cs_ete_magic) { 3101 metadata[j] = cs_etm__create_meta_blk(ptr, &i, CS_ETE_PRIV_MAX, -1); 3102 } else { 3103 ui__error("CS ETM Trace: Unrecognised magic number %#"PRIx64". File could be from a newer version of perf.\n", 3104 ptr[i]); 3105 err = -EINVAL; 3106 goto err_free_metadata; 3107 } 3108 3109 if (!metadata[j]) { 3110 err = -ENOMEM; 3111 goto err_free_metadata; 3112 } 3113 } 3114 3115 /* 3116 * Each of CS_HEADER_VERSION_MAX, CS_ETM_PRIV_MAX and 3117 * CS_ETMV4_PRIV_MAX mark how many double words are in the 3118 * global metadata, and each cpu's metadata respectively. 3119 * The following tests if the correct number of double words was 3120 * present in the auxtrace info section. 3121 */ 3122 priv_size = total_size - event_header_size - INFO_HEADER_SIZE; 3123 if (i * 8 != priv_size) { 3124 err = -EINVAL; 3125 goto err_free_metadata; 3126 } 3127 3128 etm = zalloc(sizeof(*etm)); 3129 3130 if (!etm) { 3131 err = -ENOMEM; 3132 goto err_free_metadata; 3133 } 3134 3135 err = auxtrace_queues__init(&etm->queues); 3136 if (err) 3137 goto err_free_etm; 3138 3139 if (session->itrace_synth_opts->set) { 3140 etm->synth_opts = *session->itrace_synth_opts; 3141 } else { 3142 itrace_synth_opts__set_default(&etm->synth_opts, 3143 session->itrace_synth_opts->default_no_sample); 3144 etm->synth_opts.callchain = false; 3145 } 3146 3147 etm->session = session; 3148 etm->machine = &session->machines.host; 3149 3150 etm->num_cpu = num_cpu; 3151 etm->pmu_type = (unsigned int) ((ptr[CS_PMU_TYPE_CPUS] >> 32) & 0xffffffff); 3152 etm->snapshot_mode = (ptr[CS_ETM_SNAPSHOT] != 0); 3153 etm->metadata = metadata; 3154 etm->auxtrace_type = auxtrace_info->type; 3155 etm->timeless_decoding = cs_etm__is_timeless_decoding(etm); 3156 3157 /* Use virtual timestamps if all ETMs report ts_source = 1 */ 3158 etm->has_virtual_ts = cs_etm__has_virtual_ts(metadata, num_cpu); 3159 3160 if (!etm->has_virtual_ts) 3161 ui__warning("Virtual timestamps are not enabled, or not supported by the traced system.\n" 3162 "The time field of the samples will not be set accurately.\n\n"); 3163 3164 etm->auxtrace.process_event = cs_etm__process_event; 3165 etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event; 3166 etm->auxtrace.flush_events = cs_etm__flush_events; 3167 etm->auxtrace.free_events = cs_etm__free_events; 3168 etm->auxtrace.free = cs_etm__free; 3169 etm->auxtrace.evsel_is_auxtrace = cs_etm__evsel_is_auxtrace; 3170 session->auxtrace = &etm->auxtrace; 3171 3172 etm->unknown_thread = thread__new(999999999, 999999999); 3173 if (!etm->unknown_thread) { 3174 err = -ENOMEM; 3175 goto err_free_queues; 3176 } 3177 3178 /* 3179 * Initialize list node so that at thread__zput() we can avoid 3180 * segmentation fault at list_del_init(). 3181 */ 3182 INIT_LIST_HEAD(&etm->unknown_thread->node); 3183 3184 err = thread__set_comm(etm->unknown_thread, "unknown", 0); 3185 if (err) 3186 goto err_delete_thread; 3187 3188 if (thread__init_maps(etm->unknown_thread, etm->machine)) { 3189 err = -ENOMEM; 3190 goto err_delete_thread; 3191 } 3192 3193 etm->tc.time_shift = tc->time_shift; 3194 etm->tc.time_mult = tc->time_mult; 3195 etm->tc.time_zero = tc->time_zero; 3196 if (event_contains(*tc, time_cycles)) { 3197 etm->tc.time_cycles = tc->time_cycles; 3198 etm->tc.time_mask = tc->time_mask; 3199 etm->tc.cap_user_time_zero = tc->cap_user_time_zero; 3200 etm->tc.cap_user_time_short = tc->cap_user_time_short; 3201 } 3202 err = cs_etm__synth_events(etm, session); 3203 if (err) 3204 goto err_delete_thread; 3205 3206 /* 3207 * Map Trace ID values to CPU metadata. 3208 * 3209 * Trace metadata will always contain Trace ID values from the legacy algorithm. If the 3210 * files has been recorded by a "new" perf updated to handle AUX_HW_ID then the metadata 3211 * ID value will also have the CORESIGHT_TRACE_ID_UNUSED_FLAG set. 3212 * 3213 * The updated kernel drivers that use AUX_HW_ID to sent Trace IDs will attempt to use 3214 * the same IDs as the old algorithm as far as is possible, unless there are clashes 3215 * in which case a different value will be used. This means an older perf may still 3216 * be able to record and read files generate on a newer system. 3217 * 3218 * For a perf able to interpret AUX_HW_ID packets we first check for the presence of 3219 * those packets. If they are there then the values will be mapped and plugged into 3220 * the metadata. We then set any remaining metadata values with the used flag to a 3221 * value CORESIGHT_TRACE_ID_UNUSED_VAL - which indicates no decoder is required. 3222 * 3223 * If no AUX_HW_ID packets are present - which means a file recorded on an old kernel 3224 * then we map Trace ID values to CPU directly from the metadata - clearing any unused 3225 * flags if present. 3226 */ 3227 3228 /* first scan for AUX_OUTPUT_HW_ID records to map trace ID values to CPU metadata */ 3229 aux_hw_id_found = 0; 3230 err = perf_session__peek_events(session, session->header.data_offset, 3231 session->header.data_size, 3232 cs_etm__process_aux_hw_id_cb, &aux_hw_id_found); 3233 if (err) 3234 goto err_delete_thread; 3235 3236 /* if HW ID found then clear any unused metadata ID values */ 3237 if (aux_hw_id_found) 3238 err = cs_etm__clear_unused_trace_ids_metadata(num_cpu, metadata); 3239 /* otherwise, this is a file with metadata values only, map from metadata */ 3240 else 3241 err = cs_etm__map_trace_ids_metadata(num_cpu, metadata); 3242 3243 if (err) 3244 goto err_delete_thread; 3245 3246 err = cs_etm__queue_aux_records(session); 3247 if (err) 3248 goto err_delete_thread; 3249 3250 etm->data_queued = etm->queues.populated; 3251 return 0; 3252 3253 err_delete_thread: 3254 thread__zput(etm->unknown_thread); 3255 err_free_queues: 3256 auxtrace_queues__free(&etm->queues); 3257 session->auxtrace = NULL; 3258 err_free_etm: 3259 zfree(&etm); 3260 err_free_metadata: 3261 /* No need to check @metadata[j], free(NULL) is supported */ 3262 for (j = 0; j < num_cpu; j++) 3263 zfree(&metadata[j]); 3264 zfree(&metadata); 3265 err_free_traceid_list: 3266 intlist__delete(traceid_list); 3267 return err; 3268 } 3269