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 #define _GNU_SOURCE
7 #include <getopt.h>
8 #include <stdlib.h>
9 #include <string.h>
10 #include <signal.h>
11 #include <unistd.h>
12 #include <stdio.h>
13 #include <time.h>
14
15 #include "osnoise.h"
16
17 struct osnoise_top_cpu {
18 unsigned long long sum_runtime;
19 unsigned long long sum_noise;
20 unsigned long long max_noise;
21 unsigned long long max_sample;
22
23 unsigned long long hw_count;
24 unsigned long long nmi_count;
25 unsigned long long irq_count;
26 unsigned long long softirq_count;
27 unsigned long long thread_count;
28
29 int sum_cycles;
30 };
31
32 struct osnoise_top_data {
33 struct osnoise_top_cpu *cpu_data;
34 int nr_cpus;
35 };
36
37 /*
38 * osnoise_free_top - free runtime data
39 */
40 static void
osnoise_free_top(struct osnoise_top_data * data)41 osnoise_free_top(struct osnoise_top_data *data)
42 {
43 free(data->cpu_data);
44 free(data);
45 }
46
47 /*
48 * osnoise_alloc_histogram - alloc runtime data
49 */
osnoise_alloc_top(int nr_cpus)50 static struct osnoise_top_data *osnoise_alloc_top(int nr_cpus)
51 {
52 struct osnoise_top_data *data;
53
54 data = calloc(1, sizeof(*data));
55 if (!data)
56 return NULL;
57
58 data->nr_cpus = nr_cpus;
59
60 /* one set of histograms per CPU */
61 data->cpu_data = calloc(1, sizeof(*data->cpu_data) * nr_cpus);
62 if (!data->cpu_data)
63 goto cleanup;
64
65 return data;
66
67 cleanup:
68 osnoise_free_top(data);
69 return NULL;
70 }
71
72 /*
73 * osnoise_top_handler - this is the handler for osnoise tracer events
74 */
75 static int
osnoise_top_handler(struct trace_seq * s,struct tep_record * record,struct tep_event * event,void * context)76 osnoise_top_handler(struct trace_seq *s, struct tep_record *record,
77 struct tep_event *event, void *context)
78 {
79 struct trace_instance *trace = context;
80 struct osnoise_tool *tool;
81 unsigned long long val;
82 struct osnoise_top_cpu *cpu_data;
83 struct osnoise_top_data *data;
84 int cpu = record->cpu;
85
86 tool = container_of(trace, struct osnoise_tool, trace);
87
88 data = tool->data;
89 cpu_data = &data->cpu_data[cpu];
90
91 cpu_data->sum_cycles++;
92
93 tep_get_field_val(s, event, "runtime", record, &val, 1);
94 update_sum(&cpu_data->sum_runtime, &val);
95
96 tep_get_field_val(s, event, "noise", record, &val, 1);
97 update_max(&cpu_data->max_noise, &val);
98 update_sum(&cpu_data->sum_noise, &val);
99
100 tep_get_field_val(s, event, "max_sample", record, &val, 1);
101 update_max(&cpu_data->max_sample, &val);
102
103 tep_get_field_val(s, event, "hw_count", record, &val, 1);
104 update_sum(&cpu_data->hw_count, &val);
105
106 tep_get_field_val(s, event, "nmi_count", record, &val, 1);
107 update_sum(&cpu_data->nmi_count, &val);
108
109 tep_get_field_val(s, event, "irq_count", record, &val, 1);
110 update_sum(&cpu_data->irq_count, &val);
111
112 tep_get_field_val(s, event, "softirq_count", record, &val, 1);
113 update_sum(&cpu_data->softirq_count, &val);
114
115 tep_get_field_val(s, event, "thread_count", record, &val, 1);
116 update_sum(&cpu_data->thread_count, &val);
117
118 return 0;
119 }
120
121 /*
122 * osnoise_top_header - print the header of the tool output
123 */
osnoise_top_header(struct osnoise_tool * top)124 static void osnoise_top_header(struct osnoise_tool *top)
125 {
126 struct osnoise_params *params = top->params;
127 struct trace_seq *s = top->trace.seq;
128 char duration[26];
129
130 get_duration(top->start_time, duration, sizeof(duration));
131
132 if (params->pretty_output)
133 trace_seq_printf(s, "\033[2;37;40m");
134
135 trace_seq_printf(s, " ");
136
137 if (params->mode == MODE_OSNOISE) {
138 trace_seq_printf(s, "Operating System Noise");
139 trace_seq_printf(s, " ");
140 } else if (params->mode == MODE_HWNOISE) {
141 trace_seq_printf(s, "Hardware-related Noise");
142 }
143
144 trace_seq_printf(s, " ");
145
146 if (params->pretty_output)
147 trace_seq_printf(s, "\033[0;0;0m");
148 trace_seq_printf(s, "\n");
149
150 trace_seq_printf(s, "duration: %9s | time is in us\n", duration);
151
152 if (params->pretty_output)
153 trace_seq_printf(s, "\033[2;30;47m");
154
155 trace_seq_printf(s, "CPU Period Runtime ");
156 trace_seq_printf(s, " Noise ");
157 trace_seq_printf(s, " %% CPU Aval ");
158 trace_seq_printf(s, " Max Noise Max Single ");
159 trace_seq_printf(s, " HW NMI");
160
161 if (params->mode == MODE_HWNOISE)
162 goto eol;
163
164 trace_seq_printf(s, " IRQ Softirq Thread");
165
166 eol:
167 if (params->pretty_output)
168 trace_seq_printf(s, "\033[0;0;0m");
169 trace_seq_printf(s, "\n");
170 }
171
172 /*
173 * clear_terminal - clears the output terminal
174 */
clear_terminal(struct trace_seq * seq)175 static void clear_terminal(struct trace_seq *seq)
176 {
177 if (!config_debug)
178 trace_seq_printf(seq, "\033c");
179 }
180
181 /*
182 * osnoise_top_print - prints the output of a given CPU
183 */
osnoise_top_print(struct osnoise_tool * tool,int cpu)184 static void osnoise_top_print(struct osnoise_tool *tool, int cpu)
185 {
186 struct osnoise_params *params = tool->params;
187 struct trace_seq *s = tool->trace.seq;
188 struct osnoise_top_cpu *cpu_data;
189 struct osnoise_top_data *data;
190 int percentage;
191 int decimal;
192
193 data = tool->data;
194 cpu_data = &data->cpu_data[cpu];
195
196 if (!cpu_data->sum_runtime)
197 return;
198
199 percentage = ((cpu_data->sum_runtime - cpu_data->sum_noise) * 10000000)
200 / cpu_data->sum_runtime;
201 decimal = percentage % 100000;
202 percentage = percentage / 100000;
203
204 trace_seq_printf(s, "%3d #%-6d %12llu ", cpu, cpu_data->sum_cycles, cpu_data->sum_runtime);
205 trace_seq_printf(s, "%12llu ", cpu_data->sum_noise);
206 trace_seq_printf(s, " %3d.%05d", percentage, decimal);
207 trace_seq_printf(s, "%12llu %12llu", cpu_data->max_noise, cpu_data->max_sample);
208
209 trace_seq_printf(s, "%12llu ", cpu_data->hw_count);
210 trace_seq_printf(s, "%12llu ", cpu_data->nmi_count);
211
212 if (params->mode == MODE_HWNOISE) {
213 trace_seq_printf(s, "\n");
214 return;
215 }
216
217 trace_seq_printf(s, "%12llu ", cpu_data->irq_count);
218 trace_seq_printf(s, "%12llu ", cpu_data->softirq_count);
219 trace_seq_printf(s, "%12llu\n", cpu_data->thread_count);
220 }
221
222 /*
223 * osnoise_print_stats - print data for all cpus
224 */
225 static void
osnoise_print_stats(struct osnoise_params * params,struct osnoise_tool * top)226 osnoise_print_stats(struct osnoise_params *params, struct osnoise_tool *top)
227 {
228 struct trace_instance *trace = &top->trace;
229 static int nr_cpus = -1;
230 int i;
231
232 if (nr_cpus == -1)
233 nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
234
235 if (!params->quiet)
236 clear_terminal(trace->seq);
237
238 osnoise_top_header(top);
239
240 for (i = 0; i < nr_cpus; i++) {
241 if (params->cpus && !CPU_ISSET(i, ¶ms->monitored_cpus))
242 continue;
243 osnoise_top_print(top, i);
244 }
245
246 trace_seq_do_printf(trace->seq);
247 trace_seq_reset(trace->seq);
248 osnoise_report_missed_events(top);
249 }
250
251 /*
252 * osnoise_top_usage - prints osnoise top usage message
253 */
osnoise_top_usage(struct osnoise_params * params,char * usage)254 static void osnoise_top_usage(struct osnoise_params *params, char *usage)
255 {
256 int i;
257
258 static const char * const msg[] = {
259 " [-h] [-q] [-D] [-d s] [-a us] [-p us] [-r us] [-s us] [-S us] \\",
260 " [-T us] [-t[file]] [-e sys[:event]] [--filter <filter>] [--trigger <trigger>] \\",
261 " [-c cpu-list] [-H cpu-list] [-P priority] [-C[=cgroup_name]] [--warm-up s]",
262 "",
263 " -h/--help: print this menu",
264 " -a/--auto: set automatic trace mode, stopping the session if argument in us sample is hit",
265 " -p/--period us: osnoise period in us",
266 " -r/--runtime us: osnoise runtime in us",
267 " -s/--stop us: stop trace if a single sample is higher than the argument in us",
268 " -S/--stop-total us: stop trace if the total sample is higher than the argument in us",
269 " -T/--threshold us: the minimum delta to be considered a noise",
270 " -c/--cpus cpu-list: list of cpus to run osnoise threads",
271 " -H/--house-keeping cpus: run rtla control threads only on the given cpus",
272 " -C/--cgroup[=cgroup_name]: set cgroup, if no cgroup_name is passed, the rtla's cgroup will be inherited",
273 " -d/--duration time[s|m|h|d]: duration of the session",
274 " -D/--debug: print debug info",
275 " -t/--trace[file]: save the stopped trace to [file|osnoise_trace.txt]",
276 " -e/--event <sys:event>: enable the <sys:event> in the trace instance, multiple -e are allowed",
277 " --filter <filter>: enable a trace event filter to the previous -e event",
278 " --trigger <trigger>: enable a trace event trigger to the previous -e event",
279 " -q/--quiet print only a summary at the end",
280 " -P/--priority o:prio|r:prio|f:prio|d:runtime:period : set scheduling parameters",
281 " o:prio - use SCHED_OTHER with prio",
282 " r:prio - use SCHED_RR with prio",
283 " f:prio - use SCHED_FIFO with prio",
284 " d:runtime[us|ms|s]:period[us|ms|s] - use SCHED_DEADLINE with runtime and period",
285 " in nanoseconds",
286 " --warm-up s: let the workload run for s seconds before collecting data",
287 " --trace-buffer-size kB: set the per-cpu trace buffer size in kB",
288 NULL,
289 };
290
291 if (usage)
292 fprintf(stderr, "%s\n", usage);
293
294 if (params->mode == MODE_OSNOISE) {
295 fprintf(stderr,
296 "rtla osnoise top: a per-cpu summary of the OS noise (version %s)\n",
297 VERSION);
298
299 fprintf(stderr, " usage: rtla osnoise [top]");
300 }
301
302 if (params->mode == MODE_HWNOISE) {
303 fprintf(stderr,
304 "rtla hwnoise: a summary of hardware-related noise (version %s)\n",
305 VERSION);
306
307 fprintf(stderr, " usage: rtla hwnoise");
308 }
309
310 for (i = 0; msg[i]; i++)
311 fprintf(stderr, "%s\n", msg[i]);
312
313 if (usage)
314 exit(EXIT_FAILURE);
315
316 exit(EXIT_SUCCESS);
317 }
318
319 /*
320 * osnoise_top_parse_args - allocs, parse and fill the cmd line parameters
321 */
osnoise_top_parse_args(int argc,char ** argv)322 struct osnoise_params *osnoise_top_parse_args(int argc, char **argv)
323 {
324 struct osnoise_params *params;
325 struct trace_events *tevent;
326 int retval;
327 int c;
328
329 params = calloc(1, sizeof(*params));
330 if (!params)
331 exit(1);
332
333 if (strcmp(argv[0], "hwnoise") == 0) {
334 params->mode = MODE_HWNOISE;
335 /*
336 * Reduce CPU usage for 75% to avoid killing the system.
337 */
338 params->runtime = 750000;
339 params->period = 1000000;
340 }
341
342 while (1) {
343 static struct option long_options[] = {
344 {"auto", required_argument, 0, 'a'},
345 {"cpus", required_argument, 0, 'c'},
346 {"cgroup", optional_argument, 0, 'C'},
347 {"debug", no_argument, 0, 'D'},
348 {"duration", required_argument, 0, 'd'},
349 {"event", required_argument, 0, 'e'},
350 {"house-keeping", required_argument, 0, 'H'},
351 {"help", no_argument, 0, 'h'},
352 {"period", required_argument, 0, 'p'},
353 {"priority", required_argument, 0, 'P'},
354 {"quiet", no_argument, 0, 'q'},
355 {"runtime", required_argument, 0, 'r'},
356 {"stop", required_argument, 0, 's'},
357 {"stop-total", required_argument, 0, 'S'},
358 {"threshold", required_argument, 0, 'T'},
359 {"trace", optional_argument, 0, 't'},
360 {"trigger", required_argument, 0, '0'},
361 {"filter", required_argument, 0, '1'},
362 {"warm-up", required_argument, 0, '2'},
363 {"trace-buffer-size", required_argument, 0, '3'},
364 {0, 0, 0, 0}
365 };
366
367 /* getopt_long stores the option index here. */
368 int option_index = 0;
369
370 c = getopt_long(argc, argv, "a:c:C::d:De:hH:p:P:qr:s:S:t::T:0:1:2:3:",
371 long_options, &option_index);
372
373 /* Detect the end of the options. */
374 if (c == -1)
375 break;
376
377 switch (c) {
378 case 'a':
379 /* set sample stop to auto_thresh */
380 params->stop_us = get_llong_from_str(optarg);
381
382 /* set sample threshold to 1 */
383 params->threshold = 1;
384
385 /* set trace */
386 params->trace_output = "osnoise_trace.txt";
387
388 break;
389 case 'c':
390 retval = parse_cpu_set(optarg, ¶ms->monitored_cpus);
391 if (retval)
392 osnoise_top_usage(params, "\nInvalid -c cpu list\n");
393 params->cpus = optarg;
394 break;
395 case 'C':
396 params->cgroup = 1;
397 if (!optarg) {
398 /* will inherit this cgroup */
399 params->cgroup_name = NULL;
400 } else if (*optarg == '=') {
401 /* skip the = */
402 params->cgroup_name = ++optarg;
403 }
404 break;
405 case 'D':
406 config_debug = 1;
407 break;
408 case 'd':
409 params->duration = parse_seconds_duration(optarg);
410 if (!params->duration)
411 osnoise_top_usage(params, "Invalid -d duration\n");
412 break;
413 case 'e':
414 tevent = trace_event_alloc(optarg);
415 if (!tevent) {
416 err_msg("Error alloc trace event");
417 exit(EXIT_FAILURE);
418 }
419
420 if (params->events)
421 tevent->next = params->events;
422 params->events = tevent;
423
424 break;
425 case 'h':
426 case '?':
427 osnoise_top_usage(params, NULL);
428 break;
429 case 'H':
430 params->hk_cpus = 1;
431 retval = parse_cpu_set(optarg, ¶ms->hk_cpu_set);
432 if (retval) {
433 err_msg("Error parsing house keeping CPUs\n");
434 exit(EXIT_FAILURE);
435 }
436 break;
437 case 'p':
438 params->period = get_llong_from_str(optarg);
439 if (params->period > 10000000)
440 osnoise_top_usage(params, "Period longer than 10 s\n");
441 break;
442 case 'P':
443 retval = parse_prio(optarg, ¶ms->sched_param);
444 if (retval == -1)
445 osnoise_top_usage(params, "Invalid -P priority");
446 params->set_sched = 1;
447 break;
448 case 'q':
449 params->quiet = 1;
450 break;
451 case 'r':
452 params->runtime = get_llong_from_str(optarg);
453 if (params->runtime < 100)
454 osnoise_top_usage(params, "Runtime shorter than 100 us\n");
455 break;
456 case 's':
457 params->stop_us = get_llong_from_str(optarg);
458 break;
459 case 'S':
460 params->stop_total_us = get_llong_from_str(optarg);
461 break;
462 case 't':
463 if (optarg) {
464 if (optarg[0] == '=')
465 params->trace_output = &optarg[1];
466 else
467 params->trace_output = &optarg[0];
468 } else if (optind < argc && argv[optind][0] != '-')
469 params->trace_output = argv[optind];
470 else
471 params->trace_output = "osnoise_trace.txt";
472 break;
473 case 'T':
474 params->threshold = get_llong_from_str(optarg);
475 break;
476 case '0': /* trigger */
477 if (params->events) {
478 retval = trace_event_add_trigger(params->events, optarg);
479 if (retval) {
480 err_msg("Error adding trigger %s\n", optarg);
481 exit(EXIT_FAILURE);
482 }
483 } else {
484 osnoise_top_usage(params, "--trigger requires a previous -e\n");
485 }
486 break;
487 case '1': /* filter */
488 if (params->events) {
489 retval = trace_event_add_filter(params->events, optarg);
490 if (retval) {
491 err_msg("Error adding filter %s\n", optarg);
492 exit(EXIT_FAILURE);
493 }
494 } else {
495 osnoise_top_usage(params, "--filter requires a previous -e\n");
496 }
497 break;
498 case '2':
499 params->warmup = get_llong_from_str(optarg);
500 break;
501 case '3':
502 params->buffer_size = get_llong_from_str(optarg);
503 break;
504 default:
505 osnoise_top_usage(params, "Invalid option");
506 }
507 }
508
509 if (geteuid()) {
510 err_msg("osnoise needs root permission\n");
511 exit(EXIT_FAILURE);
512 }
513
514 return params;
515 }
516
517 /*
518 * osnoise_top_apply_config - apply the top configs to the initialized tool
519 */
520 static int
osnoise_top_apply_config(struct osnoise_tool * tool,struct osnoise_params * params)521 osnoise_top_apply_config(struct osnoise_tool *tool, struct osnoise_params *params)
522 {
523 int retval;
524
525 retval = osnoise_apply_config(tool, params);
526 if (retval)
527 goto out_err;
528
529 if (params->mode == MODE_HWNOISE) {
530 retval = osnoise_set_irq_disable(tool->context, 1);
531 if (retval) {
532 err_msg("Failed to set OSNOISE_IRQ_DISABLE option\n");
533 goto out_err;
534 }
535 }
536
537 if (isatty(STDOUT_FILENO) && !params->quiet)
538 params->pretty_output = 1;
539
540 return 0;
541
542 out_err:
543 return -1;
544 }
545
546 /*
547 * osnoise_init_top - initialize a osnoise top tool with parameters
548 */
osnoise_init_top(struct osnoise_params * params)549 struct osnoise_tool *osnoise_init_top(struct osnoise_params *params)
550 {
551 struct osnoise_tool *tool;
552 int nr_cpus;
553
554 nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
555
556 tool = osnoise_init_tool("osnoise_top");
557 if (!tool)
558 return NULL;
559
560 tool->data = osnoise_alloc_top(nr_cpus);
561 if (!tool->data) {
562 osnoise_destroy_tool(tool);
563 return NULL;
564 }
565
566 tool->params = params;
567
568 tep_register_event_handler(tool->trace.tep, -1, "ftrace", "osnoise",
569 osnoise_top_handler, NULL);
570
571 return tool;
572 }
573
574 static int stop_tracing;
stop_top(int sig)575 static void stop_top(int sig)
576 {
577 stop_tracing = 1;
578 }
579
580 /*
581 * osnoise_top_set_signals - handles the signal to stop the tool
582 */
osnoise_top_set_signals(struct osnoise_params * params)583 static void osnoise_top_set_signals(struct osnoise_params *params)
584 {
585 signal(SIGINT, stop_top);
586 if (params->duration) {
587 signal(SIGALRM, stop_top);
588 alarm(params->duration);
589 }
590 }
591
osnoise_top_main(int argc,char ** argv)592 int osnoise_top_main(int argc, char **argv)
593 {
594 struct osnoise_params *params;
595 struct osnoise_tool *record = NULL;
596 struct osnoise_tool *tool = NULL;
597 struct trace_instance *trace;
598 int return_value = 1;
599 int retval;
600
601 params = osnoise_top_parse_args(argc, argv);
602 if (!params)
603 exit(1);
604
605 tool = osnoise_init_top(params);
606 if (!tool) {
607 err_msg("Could not init osnoise top\n");
608 goto out_exit;
609 }
610
611 retval = osnoise_top_apply_config(tool, params);
612 if (retval) {
613 err_msg("Could not apply config\n");
614 goto out_free;
615 }
616
617 trace = &tool->trace;
618
619 retval = enable_osnoise(trace);
620 if (retval) {
621 err_msg("Failed to enable osnoise tracer\n");
622 goto out_free;
623 }
624
625 if (params->set_sched) {
626 retval = set_comm_sched_attr("osnoise/", ¶ms->sched_param);
627 if (retval) {
628 err_msg("Failed to set sched parameters\n");
629 goto out_free;
630 }
631 }
632
633 if (params->cgroup) {
634 retval = set_comm_cgroup("osnoise/", params->cgroup_name);
635 if (!retval) {
636 err_msg("Failed to move threads to cgroup\n");
637 goto out_free;
638 }
639 }
640
641 if (params->trace_output) {
642 record = osnoise_init_trace_tool("osnoise");
643 if (!record) {
644 err_msg("Failed to enable the trace instance\n");
645 goto out_free;
646 }
647
648 if (params->events) {
649 retval = trace_events_enable(&record->trace, params->events);
650 if (retval)
651 goto out_top;
652 }
653
654 if (params->buffer_size > 0) {
655 retval = trace_set_buffer_size(&record->trace, params->buffer_size);
656 if (retval)
657 goto out_top;
658 }
659 }
660
661 /*
662 * Start the tracer here, after having set all instances.
663 *
664 * Let the trace instance start first for the case of hitting a stop
665 * tracing while enabling other instances. The trace instance is the
666 * one with most valuable information.
667 */
668 if (params->trace_output)
669 trace_instance_start(&record->trace);
670 trace_instance_start(trace);
671
672 if (params->warmup > 0) {
673 debug_msg("Warming up for %d seconds\n", params->warmup);
674 sleep(params->warmup);
675 if (stop_tracing)
676 goto out_top;
677
678 /*
679 * Clean up the buffer. The osnoise workload do not run
680 * with tracing off to avoid creating a performance penalty
681 * when not needed.
682 */
683 retval = tracefs_instance_file_write(trace->inst, "trace", "");
684 if (retval < 0) {
685 debug_msg("Error cleaning up the buffer");
686 goto out_top;
687 }
688
689 }
690
691 tool->start_time = time(NULL);
692 osnoise_top_set_signals(params);
693
694 while (!stop_tracing) {
695 sleep(params->sleep_time);
696
697 retval = tracefs_iterate_raw_events(trace->tep,
698 trace->inst,
699 NULL,
700 0,
701 collect_registered_events,
702 trace);
703 if (retval < 0) {
704 err_msg("Error iterating on events\n");
705 goto out_top;
706 }
707
708 if (!params->quiet)
709 osnoise_print_stats(params, tool);
710
711 if (osnoise_trace_is_off(tool, record))
712 break;
713
714 }
715
716 osnoise_print_stats(params, tool);
717
718 return_value = 0;
719
720 if (osnoise_trace_is_off(tool, record)) {
721 printf("osnoise hit stop tracing\n");
722 save_trace_to_file(record ? record->trace.inst : NULL,
723 params->trace_output);
724 }
725
726 out_top:
727 trace_events_destroy(&record->trace, params->events);
728 params->events = NULL;
729 out_free:
730 osnoise_free_top(tool->data);
731 osnoise_destroy_tool(record);
732 osnoise_destroy_tool(tool);
733 free(params);
734 out_exit:
735 exit(return_value);
736 }
737