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