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 */
osnoise_free_top(struct osnoise_top_data * data)40 static void osnoise_free_top(struct osnoise_top_data *data)
41 {
42 free(data->cpu_data);
43 free(data);
44 }
45
osnoise_free_top_tool(struct osnoise_tool * tool)46 static void osnoise_free_top_tool(struct osnoise_tool *tool)
47 {
48 osnoise_free_top(tool->data);
49 }
50
51 /*
52 * osnoise_alloc_histogram - alloc runtime data
53 */
osnoise_alloc_top(int nr_cpus)54 static struct osnoise_top_data *osnoise_alloc_top(int nr_cpus)
55 {
56 struct osnoise_top_data *data;
57
58 data = calloc(1, sizeof(*data));
59 if (!data)
60 return NULL;
61
62 data->nr_cpus = nr_cpus;
63
64 /* one set of histograms per CPU */
65 data->cpu_data = calloc(1, sizeof(*data->cpu_data) * nr_cpus);
66 if (!data->cpu_data)
67 goto cleanup;
68
69 return data;
70
71 cleanup:
72 osnoise_free_top(data);
73 return NULL;
74 }
75
76 /*
77 * osnoise_top_handler - this is the handler for osnoise tracer events
78 */
79 static int
osnoise_top_handler(struct trace_seq * s,struct tep_record * record,struct tep_event * event,void * context)80 osnoise_top_handler(struct trace_seq *s, struct tep_record *record,
81 struct tep_event *event, void *context)
82 {
83 struct trace_instance *trace = context;
84 struct osnoise_tool *tool;
85 unsigned long long val;
86 struct osnoise_top_cpu *cpu_data;
87 struct osnoise_top_data *data;
88 int cpu = record->cpu;
89
90 tool = container_of(trace, struct osnoise_tool, trace);
91
92 data = tool->data;
93 cpu_data = &data->cpu_data[cpu];
94
95 cpu_data->sum_cycles++;
96
97 tep_get_field_val(s, event, "runtime", record, &val, 1);
98 update_sum(&cpu_data->sum_runtime, &val);
99
100 tep_get_field_val(s, event, "noise", record, &val, 1);
101 update_max(&cpu_data->max_noise, &val);
102 update_sum(&cpu_data->sum_noise, &val);
103
104 tep_get_field_val(s, event, "max_sample", record, &val, 1);
105 update_max(&cpu_data->max_sample, &val);
106
107 tep_get_field_val(s, event, "hw_count", record, &val, 1);
108 update_sum(&cpu_data->hw_count, &val);
109
110 tep_get_field_val(s, event, "nmi_count", record, &val, 1);
111 update_sum(&cpu_data->nmi_count, &val);
112
113 tep_get_field_val(s, event, "irq_count", record, &val, 1);
114 update_sum(&cpu_data->irq_count, &val);
115
116 tep_get_field_val(s, event, "softirq_count", record, &val, 1);
117 update_sum(&cpu_data->softirq_count, &val);
118
119 tep_get_field_val(s, event, "thread_count", record, &val, 1);
120 update_sum(&cpu_data->thread_count, &val);
121
122 return 0;
123 }
124
125 /*
126 * osnoise_top_header - print the header of the tool output
127 */
osnoise_top_header(struct osnoise_tool * top)128 static void osnoise_top_header(struct osnoise_tool *top)
129 {
130 struct osnoise_params *params = to_osnoise_params(top->params);
131 struct trace_seq *s = top->trace.seq;
132 bool pretty = params->common.pretty_output;
133 char duration[26];
134
135 get_duration(top->start_time, duration, sizeof(duration));
136
137 if (pretty)
138 trace_seq_printf(s, "\033[2;37;40m");
139
140 trace_seq_printf(s, " ");
141
142 if (params->mode == MODE_OSNOISE) {
143 trace_seq_printf(s, "Operating System Noise");
144 trace_seq_printf(s, " ");
145 } else if (params->mode == MODE_HWNOISE) {
146 trace_seq_printf(s, "Hardware-related Noise");
147 }
148
149 trace_seq_printf(s, " ");
150
151 if (pretty)
152 trace_seq_printf(s, "\033[0;0;0m");
153 trace_seq_printf(s, "\n");
154
155 trace_seq_printf(s, "duration: %9s | time is in us\n", duration);
156
157 if (pretty)
158 trace_seq_printf(s, "\033[2;30;47m");
159
160 trace_seq_printf(s, "CPU Period Runtime ");
161 trace_seq_printf(s, " Noise ");
162 trace_seq_printf(s, " %% CPU Aval ");
163 trace_seq_printf(s, " Max Noise Max Single ");
164 trace_seq_printf(s, " HW NMI");
165
166 if (params->mode == MODE_HWNOISE)
167 goto eol;
168
169 trace_seq_printf(s, " IRQ Softirq Thread");
170
171 eol:
172 if (pretty)
173 trace_seq_printf(s, "\033[0;0;0m");
174 trace_seq_printf(s, "\n");
175 }
176
177 /*
178 * clear_terminal - clears the output terminal
179 */
clear_terminal(struct trace_seq * seq)180 static void clear_terminal(struct trace_seq *seq)
181 {
182 if (!config_debug)
183 trace_seq_printf(seq, "\033c");
184 }
185
186 /*
187 * osnoise_top_print - prints the output of a given CPU
188 */
osnoise_top_print(struct osnoise_tool * tool,int cpu)189 static void osnoise_top_print(struct osnoise_tool *tool, int cpu)
190 {
191 struct osnoise_params *params = to_osnoise_params(tool->params);
192 struct trace_seq *s = tool->trace.seq;
193 struct osnoise_top_cpu *cpu_data;
194 struct osnoise_top_data *data;
195 int percentage;
196 int decimal;
197
198 data = tool->data;
199 cpu_data = &data->cpu_data[cpu];
200
201 if (!cpu_data->sum_runtime)
202 return;
203
204 percentage = ((cpu_data->sum_runtime - cpu_data->sum_noise) * 10000000)
205 / cpu_data->sum_runtime;
206 decimal = percentage % 100000;
207 percentage = percentage / 100000;
208
209 trace_seq_printf(s, "%3d #%-6d %12llu ", cpu, cpu_data->sum_cycles, cpu_data->sum_runtime);
210 trace_seq_printf(s, "%12llu ", cpu_data->sum_noise);
211 trace_seq_printf(s, " %3d.%05d", percentage, decimal);
212 trace_seq_printf(s, "%12llu %12llu", cpu_data->max_noise, cpu_data->max_sample);
213
214 trace_seq_printf(s, "%12llu ", cpu_data->hw_count);
215 trace_seq_printf(s, "%12llu ", cpu_data->nmi_count);
216
217 if (params->mode == MODE_HWNOISE) {
218 trace_seq_printf(s, "\n");
219 return;
220 }
221
222 trace_seq_printf(s, "%12llu ", cpu_data->irq_count);
223 trace_seq_printf(s, "%12llu ", cpu_data->softirq_count);
224 trace_seq_printf(s, "%12llu\n", cpu_data->thread_count);
225 }
226
227 /*
228 * osnoise_print_stats - print data for all cpus
229 */
230 static void
osnoise_print_stats(struct osnoise_tool * top)231 osnoise_print_stats(struct osnoise_tool *top)
232 {
233 struct osnoise_params *params = to_osnoise_params(top->params);
234 struct trace_instance *trace = &top->trace;
235 static int nr_cpus = -1;
236 int i;
237
238 if (nr_cpus == -1)
239 nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
240
241 if (!params->common.quiet)
242 clear_terminal(trace->seq);
243
244 osnoise_top_header(top);
245
246 for (i = 0; i < nr_cpus; i++) {
247 if (params->common.cpus && !CPU_ISSET(i, ¶ms->common.monitored_cpus))
248 continue;
249 osnoise_top_print(top, i);
250 }
251
252 trace_seq_do_printf(trace->seq);
253 trace_seq_reset(trace->seq);
254 osnoise_report_missed_events(top);
255 }
256
257 /*
258 * osnoise_top_usage - prints osnoise top usage message
259 */
osnoise_top_usage(struct osnoise_params * params,char * usage)260 static void osnoise_top_usage(struct osnoise_params *params, char *usage)
261 {
262 int i;
263
264 static const char * const msg[] = {
265 " [-h] [-q] [-D] [-d s] [-a us] [-p us] [-r us] [-s us] [-S us] \\",
266 " [-T us] [-t[file]] [-e sys[:event]] [--filter <filter>] [--trigger <trigger>] \\",
267 " [-c cpu-list] [-H cpu-list] [-P priority] [-C[=cgroup_name]] [--warm-up s]",
268 "",
269 " -h/--help: print this menu",
270 " -a/--auto: set automatic trace mode, stopping the session if argument in us sample is hit",
271 " -p/--period us: osnoise period in us",
272 " -r/--runtime us: osnoise runtime in us",
273 " -s/--stop us: stop trace if a single sample is higher than the argument in us",
274 " -S/--stop-total us: stop trace if the total sample is higher than the argument in us",
275 " -T/--threshold us: the minimum delta to be considered a noise",
276 " -c/--cpus cpu-list: list of cpus to run osnoise threads",
277 " -H/--house-keeping cpus: run rtla control threads only on the given cpus",
278 " -C/--cgroup[=cgroup_name]: set cgroup, if no cgroup_name is passed, the rtla's cgroup will be inherited",
279 " -d/--duration time[s|m|h|d]: duration of the session",
280 " -D/--debug: print debug info",
281 " -t/--trace[file]: save the stopped trace to [file|osnoise_trace.txt]",
282 " -e/--event <sys:event>: enable the <sys:event> in the trace instance, multiple -e are allowed",
283 " --filter <filter>: enable a trace event filter to the previous -e event",
284 " --trigger <trigger>: enable a trace event trigger to the previous -e event",
285 " -q/--quiet print only a summary at the end",
286 " -P/--priority o:prio|r:prio|f:prio|d:runtime:period : set scheduling parameters",
287 " o:prio - use SCHED_OTHER with prio",
288 " r:prio - use SCHED_RR with prio",
289 " f:prio - use SCHED_FIFO with prio",
290 " d:runtime[us|ms|s]:period[us|ms|s] - use SCHED_DEADLINE with runtime and period",
291 " in nanoseconds",
292 " --warm-up s: let the workload run for s seconds before collecting data",
293 " --trace-buffer-size kB: set the per-cpu trace buffer size in kB",
294 " --on-threshold <action>: define action to be executed at stop-total threshold, multiple are allowed",
295 " --on-end: define action to be executed at measurement end, multiple are allowed",
296 NULL,
297 };
298
299 if (usage)
300 fprintf(stderr, "%s\n", usage);
301
302 if (params->mode == MODE_OSNOISE) {
303 fprintf(stderr,
304 "rtla osnoise top: a per-cpu summary of the OS noise (version %s)\n",
305 VERSION);
306
307 fprintf(stderr, " usage: rtla osnoise [top]");
308 }
309
310 if (params->mode == MODE_HWNOISE) {
311 fprintf(stderr,
312 "rtla hwnoise: a summary of hardware-related noise (version %s)\n",
313 VERSION);
314
315 fprintf(stderr, " usage: rtla hwnoise");
316 }
317
318 for (i = 0; msg[i]; i++)
319 fprintf(stderr, "%s\n", msg[i]);
320
321 if (usage)
322 exit(EXIT_FAILURE);
323
324 exit(EXIT_SUCCESS);
325 }
326
327 /*
328 * osnoise_top_parse_args - allocs, parse and fill the cmd line parameters
329 */
osnoise_top_parse_args(int argc,char ** argv)330 struct common_params *osnoise_top_parse_args(int argc, char **argv)
331 {
332 struct osnoise_params *params;
333 struct trace_events *tevent;
334 int retval;
335 int c;
336 char *trace_output = NULL;
337
338 params = calloc(1, sizeof(*params));
339 if (!params)
340 exit(1);
341
342 actions_init(¶ms->common.threshold_actions);
343 actions_init(¶ms->common.end_actions);
344
345 if (strcmp(argv[0], "hwnoise") == 0) {
346 params->mode = MODE_HWNOISE;
347 /*
348 * Reduce CPU usage for 75% to avoid killing the system.
349 */
350 params->runtime = 750000;
351 params->period = 1000000;
352 }
353
354 while (1) {
355 static struct option long_options[] = {
356 {"auto", required_argument, 0, 'a'},
357 {"cpus", required_argument, 0, 'c'},
358 {"cgroup", optional_argument, 0, 'C'},
359 {"debug", no_argument, 0, 'D'},
360 {"duration", required_argument, 0, 'd'},
361 {"event", required_argument, 0, 'e'},
362 {"house-keeping", required_argument, 0, 'H'},
363 {"help", no_argument, 0, 'h'},
364 {"period", required_argument, 0, 'p'},
365 {"priority", required_argument, 0, 'P'},
366 {"quiet", no_argument, 0, 'q'},
367 {"runtime", required_argument, 0, 'r'},
368 {"stop", required_argument, 0, 's'},
369 {"stop-total", required_argument, 0, 'S'},
370 {"threshold", required_argument, 0, 'T'},
371 {"trace", optional_argument, 0, 't'},
372 {"trigger", required_argument, 0, '0'},
373 {"filter", required_argument, 0, '1'},
374 {"warm-up", required_argument, 0, '2'},
375 {"trace-buffer-size", required_argument, 0, '3'},
376 {"on-threshold", required_argument, 0, '4'},
377 {"on-end", required_argument, 0, '5'},
378 {0, 0, 0, 0}
379 };
380
381 /* getopt_long stores the option index here. */
382 int option_index = 0;
383
384 c = getopt_long(argc, argv, "a:c:C::d:De:hH:p:P:qr:s:S:t::T:0:1:2:3:",
385 long_options, &option_index);
386
387 /* Detect the end of the options. */
388 if (c == -1)
389 break;
390
391 switch (c) {
392 case 'a':
393 /* set sample stop to auto_thresh */
394 params->common.stop_us = get_llong_from_str(optarg);
395
396 /* set sample threshold to 1 */
397 params->threshold = 1;
398
399 /* set trace */
400 trace_output = "osnoise_trace.txt";
401
402 break;
403 case 'c':
404 retval = parse_cpu_set(optarg, ¶ms->common.monitored_cpus);
405 if (retval)
406 osnoise_top_usage(params, "\nInvalid -c cpu list\n");
407 params->common.cpus = optarg;
408 break;
409 case 'C':
410 params->common.cgroup = 1;
411 if (!optarg) {
412 /* will inherit this cgroup */
413 params->common.cgroup_name = NULL;
414 } else if (*optarg == '=') {
415 /* skip the = */
416 params->common.cgroup_name = ++optarg;
417 }
418 break;
419 case 'D':
420 config_debug = 1;
421 break;
422 case 'd':
423 params->common.duration = parse_seconds_duration(optarg);
424 if (!params->common.duration)
425 osnoise_top_usage(params, "Invalid -d duration\n");
426 break;
427 case 'e':
428 tevent = trace_event_alloc(optarg);
429 if (!tevent) {
430 err_msg("Error alloc trace event");
431 exit(EXIT_FAILURE);
432 }
433
434 if (params->common.events)
435 tevent->next = params->common.events;
436 params->common.events = tevent;
437
438 break;
439 case 'h':
440 case '?':
441 osnoise_top_usage(params, NULL);
442 break;
443 case 'H':
444 params->common.hk_cpus = 1;
445 retval = parse_cpu_set(optarg, ¶ms->common.hk_cpu_set);
446 if (retval) {
447 err_msg("Error parsing house keeping CPUs\n");
448 exit(EXIT_FAILURE);
449 }
450 break;
451 case 'p':
452 params->period = get_llong_from_str(optarg);
453 if (params->period > 10000000)
454 osnoise_top_usage(params, "Period longer than 10 s\n");
455 break;
456 case 'P':
457 retval = parse_prio(optarg, ¶ms->common.sched_param);
458 if (retval == -1)
459 osnoise_top_usage(params, "Invalid -P priority");
460 params->common.set_sched = 1;
461 break;
462 case 'q':
463 params->common.quiet = 1;
464 break;
465 case 'r':
466 params->runtime = get_llong_from_str(optarg);
467 if (params->runtime < 100)
468 osnoise_top_usage(params, "Runtime shorter than 100 us\n");
469 break;
470 case 's':
471 params->common.stop_us = get_llong_from_str(optarg);
472 break;
473 case 'S':
474 params->common.stop_total_us = get_llong_from_str(optarg);
475 break;
476 case 't':
477 if (optarg) {
478 if (optarg[0] == '=')
479 trace_output = &optarg[1];
480 else
481 trace_output = &optarg[0];
482 } else if (optind < argc && argv[optind][0] != '-')
483 trace_output = argv[optind];
484 else
485 trace_output = "osnoise_trace.txt";
486 break;
487 case 'T':
488 params->threshold = get_llong_from_str(optarg);
489 break;
490 case '0': /* trigger */
491 if (params->common.events) {
492 retval = trace_event_add_trigger(params->common.events, optarg);
493 if (retval) {
494 err_msg("Error adding trigger %s\n", optarg);
495 exit(EXIT_FAILURE);
496 }
497 } else {
498 osnoise_top_usage(params, "--trigger requires a previous -e\n");
499 }
500 break;
501 case '1': /* filter */
502 if (params->common.events) {
503 retval = trace_event_add_filter(params->common.events, optarg);
504 if (retval) {
505 err_msg("Error adding filter %s\n", optarg);
506 exit(EXIT_FAILURE);
507 }
508 } else {
509 osnoise_top_usage(params, "--filter requires a previous -e\n");
510 }
511 break;
512 case '2':
513 params->common.warmup = get_llong_from_str(optarg);
514 break;
515 case '3':
516 params->common.buffer_size = get_llong_from_str(optarg);
517 break;
518 case '4':
519 retval = actions_parse(¶ms->common.threshold_actions, optarg,
520 "osnoise_trace.txt");
521 if (retval) {
522 err_msg("Invalid action %s\n", optarg);
523 exit(EXIT_FAILURE);
524 }
525 break;
526 case '5':
527 retval = actions_parse(¶ms->common.end_actions, optarg,
528 "osnoise_trace.txt");
529 if (retval) {
530 err_msg("Invalid action %s\n", optarg);
531 exit(EXIT_FAILURE);
532 }
533 break;
534 default:
535 osnoise_top_usage(params, "Invalid option");
536 }
537 }
538
539 if (trace_output)
540 actions_add_trace_output(¶ms->common.threshold_actions, trace_output);
541
542 if (geteuid()) {
543 err_msg("osnoise needs root permission\n");
544 exit(EXIT_FAILURE);
545 }
546
547 return ¶ms->common;
548 }
549
550 /*
551 * osnoise_top_apply_config - apply the top configs to the initialized tool
552 */
553 static int
osnoise_top_apply_config(struct osnoise_tool * tool)554 osnoise_top_apply_config(struct osnoise_tool *tool)
555 {
556 struct osnoise_params *params = to_osnoise_params(tool->params);
557 int retval;
558
559 retval = osnoise_apply_config(tool, params);
560 if (retval)
561 goto out_err;
562
563 if (params->mode == MODE_HWNOISE) {
564 retval = osnoise_set_irq_disable(tool->context, 1);
565 if (retval) {
566 err_msg("Failed to set OSNOISE_IRQ_DISABLE option\n");
567 goto out_err;
568 }
569 }
570
571 if (isatty(STDOUT_FILENO) && !params->common.quiet)
572 params->common.pretty_output = 1;
573
574 return 0;
575
576 out_err:
577 return -1;
578 }
579
580 /*
581 * osnoise_init_top - initialize a osnoise top tool with parameters
582 */
osnoise_init_top(struct common_params * params)583 struct osnoise_tool *osnoise_init_top(struct common_params *params)
584 {
585 struct osnoise_tool *tool;
586 int nr_cpus;
587
588 nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
589
590 tool = osnoise_init_tool("osnoise_top");
591 if (!tool)
592 return NULL;
593
594 tool->data = osnoise_alloc_top(nr_cpus);
595 if (!tool->data) {
596 osnoise_destroy_tool(tool);
597 return NULL;
598 }
599
600 tep_register_event_handler(tool->trace.tep, -1, "ftrace", "osnoise",
601 osnoise_top_handler, NULL);
602
603 return tool;
604 }
605
606 struct tool_ops osnoise_top_ops = {
607 .tracer = "osnoise",
608 .comm_prefix = "osnoise/",
609 .parse_args = osnoise_top_parse_args,
610 .init_tool = osnoise_init_top,
611 .apply_config = osnoise_top_apply_config,
612 .enable = osnoise_enable,
613 .main = top_main_loop,
614 .print_stats = osnoise_print_stats,
615 .free = osnoise_free_top_tool,
616 };
617