| 1 | |
| 2 | #define _GNU_SOURCE
|
| 3 |
|
| 4 | #include <pthread.h>
|
| 5 | #include <signal.h>
|
| 6 | #include <stdlib.h>
|
| 7 | #include <string.h>
|
| 8 | #include <unistd.h>
|
| 9 | #include <sys/sysinfo.h>
|
| 10 |
|
| 11 | #include "common.h"
|
| 12 |
|
| 13 | struct osnoise_tool *trace_tool;
|
| 14 | volatile int stop_tracing;
|
| 15 | int nr_cpus;
|
| 16 |
|
| 17 | static void stop_trace(int sig)
|
| 18 | {
|
| 19 | if (stop_tracing) {
|
| 20 | |
| 21 | * Stop requested twice in a row; abort event processing and
|
| 22 | * exit immediately
|
| 23 |
|
| 24 | if (trace_tool)
|
| 25 | tracefs_iterate_stop(trace_tool->trace.inst);
|
| 26 | return;
|
| 27 | }
|
| 28 | stop_tracing = 1;
|
| 29 | if (trace_tool) {
|
| 30 | trace_instance_stop(&trace_tool->trace);
|
| 31 | if (trace_tool->record)
|
| 32 | trace_instance_stop(&trace_tool->record->trace);
|
| 33 | }
|
| 34 | }
|
| 35 |
|
| 36 | |
| 37 | * set_signals - handles the signal to stop the tool
|
| 38 |
|
| 39 | static void set_signals(struct common_params *params)
|
| 40 | {
|
| 41 | signal(SIGINT, stop_trace);
|
| 42 | if (params->duration) {
|
| 43 | signal(SIGALRM, stop_trace);
|
| 44 | alarm(params->duration);
|
| 45 | }
|
| 46 | }
|
| 47 |
|
| 48 | |
| 49 | * unset_signals - unsets the signals to stop the tool
|
| 50 |
|
| 51 | static void unset_signals(struct common_params *params)
|
| 52 | {
|
| 53 | signal(SIGINT, SIG_DFL);
|
| 54 | if (params->duration) {
|
| 55 | alarm(0);
|
| 56 | signal(SIGALRM, SIG_DFL);
|
| 57 | }
|
| 58 | }
|
| 59 |
|
| 60 | |
| 61 | * common_apply_config - apply common configs to the initialized tool
|
| 62 |
|
| 63 | int
|
| 64 | common_apply_config(struct osnoise_tool *tool, struct common_params *params)
|
| 65 | {
|
| 66 | int retval, i;
|
| 67 |
|
| 68 | if (!params->sleep_time)
|
| 69 | params->sleep_time = 1;
|
| 70 |
|
| 71 | retval = osnoise_set_cpus(tool->context, params->cpus ? params->cpus : "all");
|
| 72 | if (retval) {
|
| 73 | err_msg("Failed to apply CPUs config\n");
|
| 74 | goto out_err;
|
| 75 | }
|
| 76 |
|
| 77 | if (!params->cpus) {
|
| 78 | for (i = 0; i < nr_cpus; i++)
|
| 79 | CPU_SET(i, ¶ms->monitored_cpus);
|
| 80 | }
|
| 81 |
|
| 82 | if (params->hk_cpus) {
|
| 83 | retval = sched_setaffinity(getpid(), sizeof(params->hk_cpu_set),
|
| 84 | ¶ms->hk_cpu_set);
|
| 85 | if (retval == -1) {
|
| 86 | err_msg("Failed to set rtla to the house keeping CPUs\n");
|
| 87 | goto out_err;
|
| 88 | }
|
| 89 | } else if (params->cpus) {
|
| 90 | |
| 91 | * Even if the user do not set a house-keeping CPU, try to
|
| 92 | * move rtla to a CPU set different to the one where the user
|
| 93 | * set the workload to run.
|
| 94 | *
|
| 95 | * No need to check results as this is an automatic attempt.
|
| 96 |
|
| 97 | auto_house_keeping(¶ms->monitored_cpus);
|
| 98 | }
|
| 99 |
|
| 100 | |
| 101 | * Set workload according to type of thread if the kernel supports it.
|
| 102 | * On kernels without support, user threads will have already failed
|
| 103 | * on missing fd, and kernel threads do not need it.
|
| 104 |
|
| 105 | retval = osnoise_set_workload(tool->context, params->kernel_workload);
|
| 106 | if (retval < -1) {
|
| 107 | err_msg("Failed to set OSNOISE_WORKLOAD option\n");
|
| 108 | goto out_err;
|
| 109 | }
|
| 110 |
|
| 111 | return 0;
|
| 112 |
|
| 113 | out_err:
|
| 114 | return -1;
|
| 115 | }
|
| 116 |
|
| 117 |
|
| 118 | |
| 119 | * common_threshold_handler - handle latency threshold overflow
|
| 120 | * @tool: pointer to the osnoise_tool instance containing trace contexts
|
| 121 | *
|
| 122 | * Executes the configured threshold actions (e.g., saving trace, printing,
|
| 123 | * sending signals). If the continue flag is set (--on-threshold continue),
|
| 124 | * restarts the auxiliary trace instances to continue monitoring.
|
| 125 | *
|
| 126 | * Return: 0 for success, -1 for error.
|
| 127 |
|
| 128 | int
|
| 129 | common_threshold_handler(const struct osnoise_tool *tool)
|
| 130 | {
|
| 131 | actions_perform(&tool->params->threshold_actions);
|
| 132 |
|
| 133 | if (!should_continue_tracing(tool->params))
|
| 134 |
|
| 135 | return 0;
|
| 136 |
|
| 137 |
|
| 138 | if (tool->record && trace_instance_start(&tool->record->trace))
|
| 139 | goto err;
|
| 140 | if (tool->aa && trace_instance_start(&tool->aa->trace))
|
| 141 | goto err;
|
| 142 |
|
| 143 | return 0;
|
| 144 |
|
| 145 | err:
|
| 146 | err_msg("Error restarting trace\n");
|
| 147 | return -1;
|
| 148 | }
|
| 149 |
|
| 150 | int run_tool(struct tool_ops *ops, int argc, char *argv[])
|
| 151 | {
|
| 152 | struct common_params *params;
|
| 153 | enum result return_value = ERROR;
|
| 154 | struct osnoise_tool *tool;
|
| 155 | bool stopped;
|
| 156 | int retval;
|
| 157 |
|
| 158 | nr_cpus = get_nprocs_conf();
|
| 159 | params = ops->parse_args(argc, argv);
|
| 160 | if (!params)
|
| 161 | exit(1);
|
| 162 |
|
| 163 | tool = ops->init_tool(params);
|
| 164 | if (!tool) {
|
| 165 | err_msg("Could not init osnoise tool\n");
|
| 166 | goto out_exit;
|
| 167 | }
|
| 168 | tool->ops = ops;
|
| 169 | tool->params = params;
|
| 170 |
|
| 171 | |
| 172 | * Expose the tool to signal handlers so they can stop the trace.
|
| 173 | * Otherwise, rtla could loop indefinitely when overloaded.
|
| 174 |
|
| 175 | trace_tool = tool;
|
| 176 |
|
| 177 | retval = ops->apply_config(tool);
|
| 178 | if (retval) {
|
| 179 | err_msg("Could not apply config\n");
|
| 180 | goto out_free;
|
| 181 | }
|
| 182 |
|
| 183 | retval = enable_tracer_by_name(tool->trace.inst, ops->tracer);
|
| 184 | if (retval) {
|
| 185 | err_msg("Failed to enable %s tracer\n", ops->tracer);
|
| 186 | goto out_free;
|
| 187 | }
|
| 188 |
|
| 189 | if (params->set_sched) {
|
| 190 | retval = set_comm_sched_attr(ops->comm_prefix, ¶ms->sched_param);
|
| 191 | if (retval) {
|
| 192 | err_msg("Failed to set sched parameters\n");
|
| 193 | goto out_free;
|
| 194 | }
|
| 195 | }
|
| 196 |
|
| 197 | if (params->cgroup && !params->user_data) {
|
| 198 | retval = set_comm_cgroup(ops->comm_prefix, params->cgroup_name);
|
| 199 | if (!retval) {
|
| 200 | err_msg("Failed to move threads to cgroup\n");
|
| 201 | goto out_free;
|
| 202 | }
|
| 203 | }
|
| 204 |
|
| 205 |
|
| 206 | if (params->threshold_actions.present[ACTION_TRACE_OUTPUT] ||
|
| 207 | params->end_actions.present[ACTION_TRACE_OUTPUT]) {
|
| 208 | tool->record = osnoise_init_trace_tool(ops->tracer);
|
| 209 | if (!tool->record) {
|
| 210 | err_msg("Failed to enable the trace instance\n");
|
| 211 | goto out_free;
|
| 212 | }
|
| 213 | params->threshold_actions.trace_output_inst = tool->record->trace.inst;
|
| 214 | params->end_actions.trace_output_inst = tool->record->trace.inst;
|
| 215 |
|
| 216 | if (params->events) {
|
| 217 | retval = trace_events_enable(&tool->record->trace, params->events);
|
| 218 | if (retval)
|
| 219 | goto out_trace;
|
| 220 | }
|
| 221 |
|
| 222 | if (params->buffer_size > 0) {
|
| 223 | retval = trace_set_buffer_size(&tool->record->trace, params->buffer_size);
|
| 224 | if (retval)
|
| 225 | goto out_trace;
|
| 226 | }
|
| 227 | }
|
| 228 |
|
| 229 | if (params->user_workload) {
|
| 230 | pthread_t user_thread;
|
| 231 |
|
| 232 |
|
| 233 | params->user.should_run = 1;
|
| 234 |
|
| 235 | params->user.stopped_running = 0;
|
| 236 |
|
| 237 | params->user.set = ¶ms->monitored_cpus;
|
| 238 | if (params->set_sched)
|
| 239 | params->user.sched_param = ¶ms->sched_param;
|
| 240 | else
|
| 241 | params->user.sched_param = NULL;
|
| 242 |
|
| 243 | params->user.cgroup_name = params->cgroup_name;
|
| 244 |
|
| 245 | retval = pthread_create(&user_thread, NULL, timerlat_u_dispatcher, ¶ms->user);
|
| 246 | if (retval) {
|
| 247 | err_msg("Error creating timerlat user-space threads\n");
|
| 248 | goto out_trace;
|
| 249 | }
|
| 250 | }
|
| 251 |
|
| 252 | retval = ops->enable(tool);
|
| 253 | if (retval)
|
| 254 | goto out_trace;
|
| 255 |
|
| 256 | tool->start_time = time(NULL);
|
| 257 | set_signals(params);
|
| 258 |
|
| 259 | retval = ops->main(tool);
|
| 260 | if (retval)
|
| 261 | goto out_signals;
|
| 262 |
|
| 263 | if (params->user_workload && !params->user.stopped_running) {
|
| 264 | params->user.should_run = 0;
|
| 265 | sleep(1);
|
| 266 | }
|
| 267 |
|
| 268 | ops->print_stats(tool);
|
| 269 |
|
| 270 | actions_perform(¶ms->end_actions);
|
| 271 |
|
| 272 | return_value = PASSED;
|
| 273 |
|
| 274 | stopped = osnoise_trace_is_off(tool, tool->record) && !stop_tracing;
|
| 275 | if (stopped) {
|
| 276 | printf("%s hit stop tracing\n", ops->tracer);
|
| 277 | return_value = FAILED;
|
| 278 | }
|
| 279 |
|
| 280 | if (ops->analyze)
|
| 281 | ops->analyze(tool, stopped);
|
| 282 |
|
| 283 | out_signals:
|
| 284 | unset_signals(params);
|
| 285 | out_trace:
|
| 286 | trace_events_destroy(&tool->record->trace, params->events);
|
| 287 | params->events = NULL;
|
| 288 | out_free:
|
| 289 | ops->free(tool);
|
| 290 | osnoise_destroy_tool(tool->record);
|
| 291 | osnoise_destroy_tool(tool);
|
| 292 | actions_destroy(¶ms->threshold_actions);
|
| 293 | actions_destroy(¶ms->end_actions);
|
| 294 | free(params);
|
| 295 | out_exit:
|
| 296 | exit(return_value);
|
| 297 | }
|
| 298 |
|
| 299 | int top_main_loop(struct osnoise_tool *tool)
|
| 300 | {
|
| 301 | struct common_params *params = tool->params;
|
| 302 | struct trace_instance *trace = &tool->trace;
|
| 303 | struct osnoise_tool *record = tool->record;
|
| 304 | int retval;
|
| 305 |
|
| 306 | while (!stop_tracing) {
|
| 307 | sleep(params->sleep_time);
|
| 308 |
|
| 309 | if (params->aa_only && !osnoise_trace_is_off(tool, record))
|
| 310 | continue;
|
| 311 |
|
| 312 | retval = tracefs_iterate_raw_events(trace->tep,
|
| 313 | trace->inst,
|
| 314 | NULL,
|
| 315 | 0,
|
| 316 | collect_registered_events,
|
| 317 | trace);
|
| 318 | if (retval < 0) {
|
| 319 | err_msg("Error iterating on events\n");
|
| 320 | return retval;
|
| 321 | }
|
| 322 |
|
| 323 | if (!params->quiet)
|
| 324 | tool->ops->print_stats(tool);
|
| 325 |
|
| 326 | if (osnoise_trace_is_off(tool, record)) {
|
| 327 | if (stop_tracing)
|
| 328 |
|
| 329 | return 0;
|
| 330 |
|
| 331 | retval = common_threshold_handler(tool);
|
| 332 | if (retval)
|
| 333 | return retval;
|
| 334 |
|
| 335 |
|
| 336 | if (!should_continue_tracing(params))
|
| 337 | return 0;
|
| 338 |
|
| 339 | trace_instance_start(trace);
|
| 340 | }
|
| 341 |
|
| 342 |
|
| 343 | if (params->user_workload) {
|
| 344 | if (params->user.stopped_running) {
|
| 345 | debug_msg("timerlat user space threads stopped!\n");
|
| 346 | break;
|
| 347 | }
|
| 348 | }
|
| 349 | }
|
| 350 |
|
| 351 | return 0;
|
| 352 | }
|
| 353 |
|
| 354 | int hist_main_loop(struct osnoise_tool *tool)
|
| 355 | {
|
| 356 | struct common_params *params = tool->params;
|
| 357 | struct trace_instance *trace = &tool->trace;
|
| 358 | int retval = 0;
|
| 359 |
|
| 360 | while (!stop_tracing) {
|
| 361 | sleep(params->sleep_time);
|
| 362 |
|
| 363 | retval = tracefs_iterate_raw_events(trace->tep,
|
| 364 | trace->inst,
|
| 365 | NULL,
|
| 366 | 0,
|
| 367 | collect_registered_events,
|
| 368 | trace);
|
| 369 | if (retval < 0) {
|
| 370 | err_msg("Error iterating on events\n");
|
| 371 | break;
|
| 372 | }
|
| 373 |
|
| 374 | if (osnoise_trace_is_off(tool, tool->record)) {
|
| 375 | if (stop_tracing)
|
| 376 |
|
| 377 | break;
|
| 378 |
|
| 379 | retval = common_threshold_handler(tool);
|
| 380 | if (retval)
|
| 381 | return retval;
|
| 382 |
|
| 383 | if (!should_continue_tracing(params))
|
| 384 | return 0;
|
| 385 |
|
| 386 | trace_instance_start(trace);
|
| 387 | }
|
| 388 |
|
| 389 |
|
| 390 | if (params->user_workload) {
|
| 391 | if (params->user.stopped_running) {
|
| 392 | debug_msg("user-space threads stopped!\n");
|
| 393 | break;
|
| 394 | }
|
| 395 | }
|
| 396 | }
|
| 397 |
|
| 398 | return retval;
|
| 399 | }
|
| 400 |
|
| 401 | int osn_set_stop(struct osnoise_tool *tool)
|
| 402 | {
|
| 403 | struct common_params *params = tool->params;
|
| 404 | int retval;
|
| 405 |
|
| 406 | retval = osnoise_set_stop_us(tool->context, params->stop_us);
|
| 407 | if (retval) {
|
| 408 | err_msg("Failed to set stop us\n");
|
| 409 | return retval;
|
| 410 | }
|
| 411 |
|
| 412 | retval = osnoise_set_stop_total_us(tool->context, params->stop_total_us);
|
| 413 | if (retval) {
|
| 414 | err_msg("Failed to set stop total us\n");
|
| 415 | return retval;
|
| 416 | }
|
| 417 |
|
| 418 | return 0;
|
| 419 | }
|
| 420 |
|
| 421 | static void print_msg_array(const char * const *msgs)
|
| 422 | {
|
| 423 | if (!msgs)
|
| 424 | return;
|
| 425 |
|
| 426 | for (int i = 0; msgs[i]; i++)
|
| 427 | fprintf(stderr, "%s\n", msgs[i]);
|
| 428 | }
|
| 429 |
|
| 430 | |
| 431 | * common_usage - print complete usage information
|
| 432 |
|
| 433 | void common_usage(const char *tool, const char *mode,
|
| 434 | const char *desc, const char * const *start_msgs, const char * const *opt_msgs)
|
| 435 | {
|
| 436 | static const char * const common_options[] = {
|
| 437 | " -h/--help: print this menu",
|
| 438 | NULL
|
| 439 | };
|
| 440 | fprintf(stderr, "rtla %s", tool);
|
| 441 | if (strcmp(mode, ""))
|
| 442 | fprintf(stderr, " %s", mode);
|
| 443 | fprintf(stderr, ": %s (version %s)\n\n", desc, VERSION);
|
| 444 | fprintf(stderr, " usage: [rtla] %s ", tool);
|
| 445 |
|
| 446 | if (strcmp(mode, "top") == 0)
|
| 447 | fprintf(stderr, "[top] [-h] ");
|
| 448 | else
|
| 449 | fprintf(stderr, "%s [-h] ", mode);
|
| 450 |
|
| 451 | print_msg_array(start_msgs);
|
| 452 | fprintf(stderr, "\n");
|
| 453 | print_msg_array(common_options);
|
| 454 | print_msg_array(opt_msgs);
|
| 455 |
|
| 456 | exit(EXIT_SUCCESS);
|
| 457 | }
|