1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2021 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org> 4 */ 5 6 #include <getopt.h> 7 #include <stdlib.h> 8 #include <string.h> 9 #include <signal.h> 10 #include <unistd.h> 11 #include <stdio.h> 12 #include <time.h> 13 #include <errno.h> 14 15 #include "utils.h" 16 #include "osnoise.h" 17 #include "timerlat.h" 18 #include "timerlat_aa.h" 19 20 struct timerlat_top_params { 21 char *cpus; 22 char *monitored_cpus; 23 char *trace_output; 24 unsigned long long runtime; 25 long long stop_us; 26 long long stop_total_us; 27 long long timerlat_period_us; 28 long long print_stack; 29 int sleep_time; 30 int output_divisor; 31 int duration; 32 int quiet; 33 int set_sched; 34 int dma_latency; 35 int no_aa; 36 int dump_tasks; 37 struct sched_attr sched_param; 38 struct trace_events *events; 39 }; 40 41 struct timerlat_top_cpu { 42 int irq_count; 43 int thread_count; 44 45 unsigned long long cur_irq; 46 unsigned long long min_irq; 47 unsigned long long sum_irq; 48 unsigned long long max_irq; 49 50 unsigned long long cur_thread; 51 unsigned long long min_thread; 52 unsigned long long sum_thread; 53 unsigned long long max_thread; 54 }; 55 56 struct timerlat_top_data { 57 struct timerlat_top_cpu *cpu_data; 58 int nr_cpus; 59 }; 60 61 /* 62 * timerlat_free_top - free runtime data 63 */ 64 static void 65 timerlat_free_top(struct timerlat_top_data *data) 66 { 67 free(data->cpu_data); 68 free(data); 69 } 70 71 /* 72 * timerlat_alloc_histogram - alloc runtime data 73 */ 74 static struct timerlat_top_data *timerlat_alloc_top(int nr_cpus) 75 { 76 struct timerlat_top_data *data; 77 int cpu; 78 79 data = calloc(1, sizeof(*data)); 80 if (!data) 81 return NULL; 82 83 data->nr_cpus = nr_cpus; 84 85 /* one set of histograms per CPU */ 86 data->cpu_data = calloc(1, sizeof(*data->cpu_data) * nr_cpus); 87 if (!data->cpu_data) 88 goto cleanup; 89 90 /* set the min to max */ 91 for (cpu = 0; cpu < nr_cpus; cpu++) { 92 data->cpu_data[cpu].min_irq = ~0; 93 data->cpu_data[cpu].min_thread = ~0; 94 } 95 96 return data; 97 98 cleanup: 99 timerlat_free_top(data); 100 return NULL; 101 } 102 103 /* 104 * timerlat_hist_update - record a new timerlat occurent on cpu, updating data 105 */ 106 static void 107 timerlat_top_update(struct osnoise_tool *tool, int cpu, 108 unsigned long long thread, 109 unsigned long long latency) 110 { 111 struct timerlat_top_data *data = tool->data; 112 struct timerlat_top_cpu *cpu_data = &data->cpu_data[cpu]; 113 114 if (!thread) { 115 cpu_data->irq_count++; 116 cpu_data->cur_irq = latency; 117 update_min(&cpu_data->min_irq, &latency); 118 update_sum(&cpu_data->sum_irq, &latency); 119 update_max(&cpu_data->max_irq, &latency); 120 } else { 121 cpu_data->thread_count++; 122 cpu_data->cur_thread = latency; 123 update_min(&cpu_data->min_thread, &latency); 124 update_sum(&cpu_data->sum_thread, &latency); 125 update_max(&cpu_data->max_thread, &latency); 126 } 127 } 128 129 /* 130 * timerlat_top_handler - this is the handler for timerlat tracer events 131 */ 132 static int 133 timerlat_top_handler(struct trace_seq *s, struct tep_record *record, 134 struct tep_event *event, void *context) 135 { 136 struct trace_instance *trace = context; 137 struct timerlat_top_params *params; 138 unsigned long long latency, thread; 139 struct osnoise_tool *top; 140 int cpu = record->cpu; 141 142 top = container_of(trace, struct osnoise_tool, trace); 143 params = top->params; 144 145 tep_get_field_val(s, event, "context", record, &thread, 1); 146 tep_get_field_val(s, event, "timer_latency", record, &latency, 1); 147 148 timerlat_top_update(top, cpu, thread, latency); 149 150 if (!params->no_aa) 151 timerlat_aa_handler(s, record, event, context); 152 153 return 0; 154 } 155 156 /* 157 * timerlat_top_header - print the header of the tool output 158 */ 159 static void timerlat_top_header(struct osnoise_tool *top) 160 { 161 struct timerlat_top_params *params = top->params; 162 struct trace_seq *s = top->trace.seq; 163 char duration[26]; 164 165 get_duration(top->start_time, duration, sizeof(duration)); 166 167 trace_seq_printf(s, "\033[2;37;40m"); 168 trace_seq_printf(s, " Timer Latency "); 169 trace_seq_printf(s, "\033[0;0;0m"); 170 trace_seq_printf(s, "\n"); 171 172 trace_seq_printf(s, "%-6s | IRQ Timer Latency (%s) | Thread Timer Latency (%s)\n", duration, 173 params->output_divisor == 1 ? "ns" : "us", 174 params->output_divisor == 1 ? "ns" : "us"); 175 176 trace_seq_printf(s, "\033[2;30;47m"); 177 trace_seq_printf(s, "CPU COUNT | cur min avg max | cur min avg max"); 178 trace_seq_printf(s, "\033[0;0;0m"); 179 trace_seq_printf(s, "\n"); 180 } 181 182 /* 183 * timerlat_top_print - prints the output of a given CPU 184 */ 185 static void timerlat_top_print(struct osnoise_tool *top, int cpu) 186 { 187 188 struct timerlat_top_params *params = top->params; 189 struct timerlat_top_data *data = top->data; 190 struct timerlat_top_cpu *cpu_data = &data->cpu_data[cpu]; 191 int divisor = params->output_divisor; 192 struct trace_seq *s = top->trace.seq; 193 194 if (divisor == 0) 195 return; 196 197 /* 198 * Skip if no data is available: is this cpu offline? 199 */ 200 if (!cpu_data->irq_count && !cpu_data->thread_count) 201 return; 202 203 /* 204 * Unless trace is being lost, IRQ counter is always the max. 205 */ 206 trace_seq_printf(s, "%3d #%-9d |", cpu, cpu_data->irq_count); 207 208 if (!cpu_data->irq_count) { 209 trace_seq_printf(s, " - "); 210 trace_seq_printf(s, " - "); 211 trace_seq_printf(s, " - "); 212 trace_seq_printf(s, " - |"); 213 } else { 214 trace_seq_printf(s, "%9llu ", cpu_data->cur_irq / params->output_divisor); 215 trace_seq_printf(s, "%9llu ", cpu_data->min_irq / params->output_divisor); 216 trace_seq_printf(s, "%9llu ", (cpu_data->sum_irq / cpu_data->irq_count) / divisor); 217 trace_seq_printf(s, "%9llu |", cpu_data->max_irq / divisor); 218 } 219 220 if (!cpu_data->thread_count) { 221 trace_seq_printf(s, " - "); 222 trace_seq_printf(s, " - "); 223 trace_seq_printf(s, " - "); 224 trace_seq_printf(s, " -\n"); 225 } else { 226 trace_seq_printf(s, "%9llu ", cpu_data->cur_thread / divisor); 227 trace_seq_printf(s, "%9llu ", cpu_data->min_thread / divisor); 228 trace_seq_printf(s, "%9llu ", 229 (cpu_data->sum_thread / cpu_data->thread_count) / divisor); 230 trace_seq_printf(s, "%9llu\n", cpu_data->max_thread / divisor); 231 } 232 } 233 234 /* 235 * clear_terminal - clears the output terminal 236 */ 237 static void clear_terminal(struct trace_seq *seq) 238 { 239 if (!config_debug) 240 trace_seq_printf(seq, "\033c"); 241 } 242 243 /* 244 * timerlat_print_stats - print data for all cpus 245 */ 246 static void 247 timerlat_print_stats(struct timerlat_top_params *params, struct osnoise_tool *top) 248 { 249 struct trace_instance *trace = &top->trace; 250 static int nr_cpus = -1; 251 int i; 252 253 if (nr_cpus == -1) 254 nr_cpus = sysconf(_SC_NPROCESSORS_CONF); 255 256 if (!params->quiet) 257 clear_terminal(trace->seq); 258 259 timerlat_top_header(top); 260 261 for (i = 0; i < nr_cpus; i++) { 262 if (params->cpus && !params->monitored_cpus[i]) 263 continue; 264 timerlat_top_print(top, i); 265 } 266 267 trace_seq_do_printf(trace->seq); 268 trace_seq_reset(trace->seq); 269 } 270 271 /* 272 * timerlat_top_usage - prints timerlat top usage message 273 */ 274 static void timerlat_top_usage(char *usage) 275 { 276 int i; 277 278 static const char *const msg[] = { 279 "", 280 " usage: rtla timerlat [top] [-h] [-q] [-a us] [-d s] [-D] [-n] [-p us] [-i us] [-T us] [-s us] \\", 281 " [[-t[=file]] [-e sys[:event]] [--filter <filter>] [--trigger <trigger>] [-c cpu-list] \\", 282 " [-P priority] [--dma-latency us]", 283 "", 284 " -h/--help: print this menu", 285 " -a/--auto: set automatic trace mode, stopping the session if argument in us latency is hit", 286 " -p/--period us: timerlat period in us", 287 " -i/--irq us: stop trace if the irq latency is higher than the argument in us", 288 " -T/--thread us: stop trace if the thread latency is higher than the argument in us", 289 " -s/--stack us: save the stack trace at the IRQ if a thread latency is higher than the argument in us", 290 " -c/--cpus cpus: run the tracer only on the given cpus", 291 " -d/--duration time[m|h|d]: duration of the session in seconds", 292 " -D/--debug: print debug info", 293 " --dump-tasks: prints the task running on all CPUs if stop conditions are met (depends on !--no-aa)", 294 " -t/--trace[=file]: save the stopped trace to [file|timerlat_trace.txt]", 295 " -e/--event <sys:event>: enable the <sys:event> in the trace instance, multiple -e are allowed", 296 " --filter <command>: enable a trace event filter to the previous -e event", 297 " --trigger <command>: enable a trace event trigger to the previous -e event", 298 " -n/--nano: display data in nanoseconds", 299 " --no-aa: disable auto-analysis, reducing rtla timerlat cpu usage", 300 " -q/--quiet print only a summary at the end", 301 " --dma-latency us: set /dev/cpu_dma_latency latency <us> to reduce exit from idle latency", 302 " -P/--priority o:prio|r:prio|f:prio|d:runtime:period : set scheduling parameters", 303 " o:prio - use SCHED_OTHER with prio", 304 " r:prio - use SCHED_RR with prio", 305 " f:prio - use SCHED_FIFO with prio", 306 " d:runtime[us|ms|s]:period[us|ms|s] - use SCHED_DEADLINE with runtime and period", 307 " in nanoseconds", 308 NULL, 309 }; 310 311 if (usage) 312 fprintf(stderr, "%s\n", usage); 313 314 fprintf(stderr, "rtla timerlat top: a per-cpu summary of the timer latency (version %s)\n", 315 VERSION); 316 317 for (i = 0; msg[i]; i++) 318 fprintf(stderr, "%s\n", msg[i]); 319 exit(1); 320 } 321 322 /* 323 * timerlat_top_parse_args - allocs, parse and fill the cmd line parameters 324 */ 325 static struct timerlat_top_params 326 *timerlat_top_parse_args(int argc, char **argv) 327 { 328 struct timerlat_top_params *params; 329 struct trace_events *tevent; 330 long long auto_thresh; 331 int retval; 332 int c; 333 334 params = calloc(1, sizeof(*params)); 335 if (!params) 336 exit(1); 337 338 /* disabled by default */ 339 params->dma_latency = -1; 340 341 /* display data in microseconds */ 342 params->output_divisor = 1000; 343 344 while (1) { 345 static struct option long_options[] = { 346 {"auto", required_argument, 0, 'a'}, 347 {"cpus", required_argument, 0, 'c'}, 348 {"debug", no_argument, 0, 'D'}, 349 {"duration", required_argument, 0, 'd'}, 350 {"event", required_argument, 0, 'e'}, 351 {"help", no_argument, 0, 'h'}, 352 {"irq", required_argument, 0, 'i'}, 353 {"nano", no_argument, 0, 'n'}, 354 {"period", required_argument, 0, 'p'}, 355 {"priority", required_argument, 0, 'P'}, 356 {"quiet", no_argument, 0, 'q'}, 357 {"stack", required_argument, 0, 's'}, 358 {"thread", required_argument, 0, 'T'}, 359 {"trace", optional_argument, 0, 't'}, 360 {"trigger", required_argument, 0, '0'}, 361 {"filter", required_argument, 0, '1'}, 362 {"dma-latency", required_argument, 0, '2'}, 363 {"no-aa", no_argument, 0, '3'}, 364 {"dump-tasks", no_argument, 0, '4'}, 365 {0, 0, 0, 0} 366 }; 367 368 /* getopt_long stores the option index here. */ 369 int option_index = 0; 370 371 c = getopt_long(argc, argv, "a:c:d:De:hi:np:P:qs:t::T:0:1:2:34", 372 long_options, &option_index); 373 374 /* detect the end of the options. */ 375 if (c == -1) 376 break; 377 378 switch (c) { 379 case 'a': 380 auto_thresh = get_llong_from_str(optarg); 381 382 /* set thread stop to auto_thresh */ 383 params->stop_total_us = auto_thresh; 384 params->stop_us = auto_thresh; 385 386 /* get stack trace */ 387 params->print_stack = auto_thresh; 388 389 /* set trace */ 390 params->trace_output = "timerlat_trace.txt"; 391 break; 392 case 'c': 393 retval = parse_cpu_list(optarg, ¶ms->monitored_cpus); 394 if (retval) 395 timerlat_top_usage("\nInvalid -c cpu list\n"); 396 params->cpus = optarg; 397 break; 398 case 'D': 399 config_debug = 1; 400 break; 401 case 'd': 402 params->duration = parse_seconds_duration(optarg); 403 if (!params->duration) 404 timerlat_top_usage("Invalid -D duration\n"); 405 break; 406 case 'e': 407 tevent = trace_event_alloc(optarg); 408 if (!tevent) { 409 err_msg("Error alloc trace event"); 410 exit(EXIT_FAILURE); 411 } 412 413 if (params->events) 414 tevent->next = params->events; 415 params->events = tevent; 416 break; 417 case 'h': 418 case '?': 419 timerlat_top_usage(NULL); 420 break; 421 case 'i': 422 params->stop_us = get_llong_from_str(optarg); 423 break; 424 case 'n': 425 params->output_divisor = 1; 426 break; 427 case 'p': 428 params->timerlat_period_us = get_llong_from_str(optarg); 429 if (params->timerlat_period_us > 1000000) 430 timerlat_top_usage("Period longer than 1 s\n"); 431 break; 432 case 'P': 433 retval = parse_prio(optarg, ¶ms->sched_param); 434 if (retval == -1) 435 timerlat_top_usage("Invalid -P priority"); 436 params->set_sched = 1; 437 break; 438 case 'q': 439 params->quiet = 1; 440 break; 441 case 's': 442 params->print_stack = get_llong_from_str(optarg); 443 break; 444 case 'T': 445 params->stop_total_us = get_llong_from_str(optarg); 446 break; 447 case 't': 448 if (optarg) 449 /* skip = */ 450 params->trace_output = &optarg[1]; 451 else 452 params->trace_output = "timerlat_trace.txt"; 453 454 break; 455 case '0': /* trigger */ 456 if (params->events) { 457 retval = trace_event_add_trigger(params->events, optarg); 458 if (retval) { 459 err_msg("Error adding trigger %s\n", optarg); 460 exit(EXIT_FAILURE); 461 } 462 } else { 463 timerlat_top_usage("--trigger requires a previous -e\n"); 464 } 465 break; 466 case '1': /* filter */ 467 if (params->events) { 468 retval = trace_event_add_filter(params->events, optarg); 469 if (retval) { 470 err_msg("Error adding filter %s\n", optarg); 471 exit(EXIT_FAILURE); 472 } 473 } else { 474 timerlat_top_usage("--filter requires a previous -e\n"); 475 } 476 break; 477 case '2': /* dma-latency */ 478 params->dma_latency = get_llong_from_str(optarg); 479 if (params->dma_latency < 0 || params->dma_latency > 10000) { 480 err_msg("--dma-latency needs to be >= 0 and < 10000"); 481 exit(EXIT_FAILURE); 482 } 483 break; 484 case '3': /* no-aa */ 485 params->no_aa = 1; 486 break; 487 case '4': 488 params->dump_tasks = 1; 489 break; 490 default: 491 timerlat_top_usage("Invalid option"); 492 } 493 } 494 495 if (geteuid()) { 496 err_msg("rtla needs root permission\n"); 497 exit(EXIT_FAILURE); 498 } 499 500 /* 501 * Auto analysis only happens if stop tracing, thus: 502 */ 503 if (!params->stop_us && !params->stop_total_us) 504 params->no_aa = 1; 505 506 return params; 507 } 508 509 /* 510 * timerlat_top_apply_config - apply the top configs to the initialized tool 511 */ 512 static int 513 timerlat_top_apply_config(struct osnoise_tool *top, struct timerlat_top_params *params) 514 { 515 int retval; 516 517 if (!params->sleep_time) 518 params->sleep_time = 1; 519 520 if (params->cpus) { 521 retval = osnoise_set_cpus(top->context, params->cpus); 522 if (retval) { 523 err_msg("Failed to apply CPUs config\n"); 524 goto out_err; 525 } 526 } 527 528 if (params->stop_us) { 529 retval = osnoise_set_stop_us(top->context, params->stop_us); 530 if (retval) { 531 err_msg("Failed to set stop us\n"); 532 goto out_err; 533 } 534 } 535 536 if (params->stop_total_us) { 537 retval = osnoise_set_stop_total_us(top->context, params->stop_total_us); 538 if (retval) { 539 err_msg("Failed to set stop total us\n"); 540 goto out_err; 541 } 542 } 543 544 545 if (params->timerlat_period_us) { 546 retval = osnoise_set_timerlat_period_us(top->context, params->timerlat_period_us); 547 if (retval) { 548 err_msg("Failed to set timerlat period\n"); 549 goto out_err; 550 } 551 } 552 553 554 if (params->print_stack) { 555 retval = osnoise_set_print_stack(top->context, params->print_stack); 556 if (retval) { 557 err_msg("Failed to set print stack\n"); 558 goto out_err; 559 } 560 } 561 562 return 0; 563 564 out_err: 565 return -1; 566 } 567 568 /* 569 * timerlat_init_top - initialize a timerlat top tool with parameters 570 */ 571 static struct osnoise_tool 572 *timerlat_init_top(struct timerlat_top_params *params) 573 { 574 struct osnoise_tool *top; 575 int nr_cpus; 576 int retval; 577 578 nr_cpus = sysconf(_SC_NPROCESSORS_CONF); 579 580 top = osnoise_init_tool("timerlat_top"); 581 if (!top) 582 return NULL; 583 584 top->data = timerlat_alloc_top(nr_cpus); 585 if (!top->data) 586 goto out_err; 587 588 top->params = params; 589 590 tep_register_event_handler(top->trace.tep, -1, "ftrace", "timerlat", 591 timerlat_top_handler, top); 592 593 /* 594 * If no auto analysis, we are ready. 595 */ 596 if (params->no_aa) 597 return top; 598 599 retval = timerlat_aa_init(top, nr_cpus, params->dump_tasks); 600 if (retval) 601 goto out_err; 602 603 return top; 604 605 out_err: 606 osnoise_destroy_tool(top); 607 return NULL; 608 } 609 610 static int stop_tracing; 611 static void stop_top(int sig) 612 { 613 stop_tracing = 1; 614 } 615 616 /* 617 * timerlat_top_set_signals - handles the signal to stop the tool 618 */ 619 static void 620 timerlat_top_set_signals(struct timerlat_top_params *params) 621 { 622 signal(SIGINT, stop_top); 623 if (params->duration) { 624 signal(SIGALRM, stop_top); 625 alarm(params->duration); 626 } 627 } 628 629 int timerlat_top_main(int argc, char *argv[]) 630 { 631 struct timerlat_top_params *params; 632 struct osnoise_tool *record = NULL; 633 struct osnoise_tool *top = NULL; 634 struct trace_instance *trace; 635 int dma_latency_fd = -1; 636 int return_value = 1; 637 int retval; 638 639 params = timerlat_top_parse_args(argc, argv); 640 if (!params) 641 exit(1); 642 643 top = timerlat_init_top(params); 644 if (!top) { 645 err_msg("Could not init osnoise top\n"); 646 goto out_exit; 647 } 648 649 retval = timerlat_top_apply_config(top, params); 650 if (retval) { 651 err_msg("Could not apply config\n"); 652 goto out_free; 653 } 654 655 trace = &top->trace; 656 657 retval = enable_timerlat(trace); 658 if (retval) { 659 err_msg("Failed to enable timerlat tracer\n"); 660 goto out_free; 661 } 662 663 if (params->set_sched) { 664 retval = set_comm_sched_attr("timerlat/", ¶ms->sched_param); 665 if (retval) { 666 err_msg("Failed to set sched parameters\n"); 667 goto out_free; 668 } 669 } 670 671 if (params->dma_latency >= 0) { 672 dma_latency_fd = set_cpu_dma_latency(params->dma_latency); 673 if (dma_latency_fd < 0) { 674 err_msg("Could not set /dev/cpu_dma_latency.\n"); 675 goto out_free; 676 } 677 } 678 679 trace_instance_start(trace); 680 681 if (params->trace_output) { 682 record = osnoise_init_trace_tool("timerlat"); 683 if (!record) { 684 err_msg("Failed to enable the trace instance\n"); 685 goto out_free; 686 } 687 688 if (params->events) { 689 retval = trace_events_enable(&record->trace, params->events); 690 if (retval) 691 goto out_top; 692 } 693 694 trace_instance_start(&record->trace); 695 } 696 697 top->start_time = time(NULL); 698 timerlat_top_set_signals(params); 699 700 while (!stop_tracing) { 701 sleep(params->sleep_time); 702 703 retval = tracefs_iterate_raw_events(trace->tep, 704 trace->inst, 705 NULL, 706 0, 707 collect_registered_events, 708 trace); 709 if (retval < 0) { 710 err_msg("Error iterating on events\n"); 711 goto out_top; 712 } 713 714 if (!params->quiet) 715 timerlat_print_stats(params, top); 716 717 if (trace_is_off(&top->trace, &record->trace)) 718 break; 719 720 } 721 722 timerlat_print_stats(params, top); 723 724 return_value = 0; 725 726 if (trace_is_off(&top->trace, &record->trace)) { 727 printf("rtla timerlat hit stop tracing\n"); 728 729 if (!params->no_aa) 730 timerlat_auto_analysis(params->stop_us, params->stop_total_us); 731 732 if (params->trace_output) { 733 printf(" Saving trace to %s\n", params->trace_output); 734 save_trace_to_file(record->trace.inst, params->trace_output); 735 } 736 } 737 738 out_top: 739 if (dma_latency_fd >= 0) 740 close(dma_latency_fd); 741 trace_events_destroy(&record->trace, params->events); 742 params->events = NULL; 743 out_free: 744 timerlat_free_top(top->data); 745 timerlat_aa_destroy(); 746 osnoise_destroy_tool(record); 747 osnoise_destroy_tool(top); 748 free(params); 749 out_exit: 750 exit(return_value); 751 } 752