1/*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2018, Matthew Macy 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 */ 28 29#include <sys/param.h> 30#include <sys/cpuset.h> 31#include <sys/event.h> 32#include <sys/queue.h> 33#include <sys/socket.h> 34#include <sys/stat.h> 35#include <sys/sysctl.h> 36#include <sys/time.h> 37#include <sys/ttycom.h> 38#include <sys/user.h> 39#include <sys/wait.h> 40 41#include <assert.h> 42#include <curses.h> 43#include <err.h> 44#include <errno.h> 45#include <fcntl.h> 46#include <getopt.h> 47#include <kvm.h> 48#include <libgen.h> 49#include <limits.h> 50#include <locale.h> 51#include <math.h> 52#include <pmc.h> 53#include <pmclog.h> 54#include <regex.h> 55#include <signal.h> 56#include <stdarg.h> 57#include <stdint.h> 58#include <stdio.h> 59#include <stdlib.h> 60#include <string.h> 61#include <sysexits.h> 62#include <unistd.h> 63 64#include <libpmcstat.h> 65#include "cmd_pmc.h" 66 67/* 68 * Return the frequency of the kernel's statistics clock. 69 */ 70static int 71getstathz(void) 72{ 73 int mib[2]; 74 size_t size; 75 struct clockinfo clockrate; 76 77 mib[0] = CTL_KERN; 78 mib[1] = KERN_CLOCKRATE; 79 size = sizeof clockrate; 80 if (sysctl(mib, 2, &clockrate, &size, NULL, 0) == -1) 81 err(1, "sysctl kern.clockrate"); 82 return clockrate.stathz; 83} 84 85#define STAT_MODE_NPMCS 6 86#define FIXED_MODE_NPMCS 2 87static struct timespec before_ts; 88#define CYCLES 0 89#define INST 1 90#define BR 2 91#define IAP_START BR 92#define BR_MISS 3 93#define CACHE 4 94#define CACHE_MISS 5 95static const char *pmc_stat_mode_names[] = { 96 "cycles", 97 "instructions", 98 "branches", 99 "branch-misses", 100 "cache-references", 101 "cache-misses", 102}; 103 104/* Common aliases for the desired stat counter */ 105static const char *pmc_stat_mode_aliases[] = { 106 "unhalted-cycles", 107 "instructions", 108 "branches", 109 "branch-mispredicts", 110 "LLC-REFERENCE", 111 "LLC-MISSES", 112}; 113 114static int pmcstat_sockpair[NSOCKPAIRFD]; 115 116static void __dead2 117usage(void) 118{ 119 errx(EX_USAGE, 120 "\t get basic stats from command line program\n" 121 "\t -j <eventlist>, --events <eventlist> comma-delimited list of event specifiers\n" 122 ); 123} 124 125static void 126showtime(FILE *out, struct timespec *before, struct timespec *after, 127 struct rusage *ru) 128{ 129 char decimal_point; 130 uint64_t real, user, sys; 131 132 (void)setlocale(LC_NUMERIC, ""); 133 decimal_point = localeconv()->decimal_point[0]; 134 135 after->tv_sec -= before->tv_sec; 136 after->tv_nsec -= before->tv_nsec; 137 if (after->tv_nsec < 0) { 138 after->tv_sec--; 139 after->tv_nsec += 1000000000; 140 } 141 142 real = (after->tv_sec * 1000000000 + after->tv_nsec) / 1000; 143 user = ru->ru_utime.tv_sec * 1000000 + ru->ru_utime.tv_usec; 144 sys = ru->ru_stime.tv_sec * 1000000 + ru->ru_stime.tv_usec; 145 fprintf(out, "%13jd%c%02ld real\t\t\t#\t%2.02f%% cpu\n", 146 (intmax_t)after->tv_sec, decimal_point, 147 after->tv_nsec / 10000000, 100 * (double)(sys + user + 1) / (double)(real + 1)); 148 fprintf(out, "%13jd%c%02ld user\t\t\t#\t%2.2f%% cpu\n", 149 (intmax_t)ru->ru_utime.tv_sec, decimal_point, 150 ru->ru_utime.tv_usec / 10000, 100 * (double)(user + 1) / (double)(real + 1)); 151 fprintf(out, "%13jd%c%02ld sys\t\t\t#\t%2.02f%% cpu\n", 152 (intmax_t)ru->ru_stime.tv_sec, decimal_point, 153 ru->ru_stime.tv_usec / 10000, 100 * (double)(sys + 1) / (double)(real + 1)); 154} 155 156static const char *stat_mode_cntrs[STAT_MODE_NPMCS]; 157static const char *stat_mode_names[STAT_MODE_NPMCS]; 158 159static void 160pmc_stat_setup_stat(int system_mode, const char *arg) 161{ 162 const char *new_cntrs[STAT_MODE_NPMCS]; 163 struct pmcstat_ev *ev; 164 char *counters, *counter; 165 int i, c, start, newcnt; 166 cpuset_t cpumask, rootmask; 167 168 if (cpuset_getaffinity(CPU_LEVEL_ROOT, CPU_WHICH_PID, -1, 169 sizeof(rootmask), &rootmask) == -1) 170 err(EX_OSERR, "ERROR: Cannot determine the root set of CPUs"); 171 CPU_COPY(&rootmask, &cpumask); 172 173 if (system_mode && geteuid() != 0) 174 errx(EX_USAGE, "ERROR: system mode counters can only be used as root"); 175 counters = NULL; 176 for (i = 0; i < STAT_MODE_NPMCS; i++) { 177 stat_mode_cntrs[i] = pmc_stat_mode_aliases[i]; 178 stat_mode_names[i] = pmc_stat_mode_names[i]; 179 } 180 if (arg) { 181 counters = strdup(arg); 182 newcnt = 0; 183 while ((counter = strsep(&counters, ",")) != NULL && 184 newcnt < STAT_MODE_NPMCS - IAP_START) { 185 new_cntrs[newcnt++] = counter; 186 if (pmc_pmu_sample_rate_get(counter) == DEFAULT_SAMPLE_COUNT) 187 errx(EX_USAGE, "ERROR: %s not recognized on host", counter); 188 } 189 start = IAP_START + STAT_MODE_NPMCS - FIXED_MODE_NPMCS - newcnt; 190 for (i = 0; i < newcnt; i++) { 191 stat_mode_cntrs[start + i] = new_cntrs[i]; 192 stat_mode_names[start + i] = new_cntrs[i]; 193 } 194 } 195 if (system_mode) 196 pmc_args.pa_flags |= FLAG_HAS_SYSTEM_PMCS; 197 else 198 pmc_args.pa_flags |= FLAG_HAS_PROCESS_PMCS; 199 pmc_args.pa_flags |= FLAG_HAS_COUNTING_PMCS; 200 pmc_args.pa_flags |= FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET; 201 pmc_args.pa_flags |= FLAG_HAS_PIPE; 202 pmc_args.pa_required |= FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET | FLAG_HAS_OUTPUT_LOGFILE; 203 pmc_args.pa_outputpath = strdup("/dev/null"); 204 pmc_args.pa_logfd = pmcstat_open_log(pmc_args.pa_outputpath, 205 PMCSTAT_OPEN_FOR_WRITE); 206 for (i = 0; i < STAT_MODE_NPMCS; i++) { 207 if ((ev = malloc(sizeof(*ev))) == NULL) 208 errx(EX_SOFTWARE, "ERROR: Out of memory."); 209 if (system_mode) 210 ev->ev_mode = PMC_MODE_SC; 211 else 212 ev->ev_mode = PMC_MODE_TC; 213 ev->ev_spec = strdup(stat_mode_cntrs[i]); 214 if (ev->ev_spec == NULL) 215 errx(EX_SOFTWARE, "ERROR: Out of memory."); 216 c = strcspn(strdup(stat_mode_cntrs[i]), ", \t"); 217 ev->ev_name = malloc(c + 1); 218 if (ev->ev_name == NULL) 219 errx(EX_SOFTWARE, "ERROR: Out of memory."); 220 (void)strncpy(ev->ev_name, stat_mode_cntrs[i], c); 221 *(ev->ev_name + c) = '\0'; 222 223 ev->ev_count = -1; 224 ev->ev_flags = 0; 225 ev->ev_flags |= PMC_F_DESCENDANTS; 226 ev->ev_cumulative = 1; 227 228 ev->ev_saved = 0LL; 229 ev->ev_pmcid = PMC_ID_INVALID; 230 STAILQ_INSERT_TAIL(&pmc_args.pa_events, ev, ev_next); 231 if (system_mode) { 232 ev->ev_cpu = CPU_FFS(&cpumask) - 1; 233 CPU_CLR(ev->ev_cpu, &cpumask); 234 pmcstat_clone_event_descriptor(ev, &cpumask, &pmc_args); 235 CPU_SET(ev->ev_cpu, &cpumask); 236 } else 237 ev->ev_cpu = PMC_CPU_ANY; 238 239 } 240 if (clock_gettime(CLOCK_MONOTONIC, &before_ts)) 241 err(1, "clock_gettime"); 242} 243 244static void 245pmc_stat_print_stat(struct rusage *ru) 246{ 247 struct pmcstat_ev *ev; 248 struct timespec after; 249 uint64_t cvals[STAT_MODE_NPMCS]; 250 uint64_t ticks, value; 251 int hz, i; 252 253 if (ru) { 254 hz = getstathz(); 255 ticks = hz * (ru->ru_utime.tv_sec + ru->ru_stime.tv_sec) + 256 hz * (ru->ru_utime.tv_usec + ru->ru_stime.tv_usec) / 1000000; 257 if (clock_gettime(CLOCK_MONOTONIC, &after)) 258 err(1, "clock_gettime"); 259 /* 260 * If our round-off on the tick calculation still puts us at 0, 261 * then always assume at least one tick. 262 */ 263 if (ticks == 0) 264 ticks = 1; 265 fprintf(pmc_args.pa_printfile, "%16ld %s\t\t#\t%02.03f M/sec\n", 266 ru->ru_minflt, "page faults", ((double)ru->ru_minflt / (double)ticks) / hz); 267 fprintf(pmc_args.pa_printfile, "%16ld %s\t\t#\t%02.03f M/sec\n", 268 ru->ru_nvcsw, "voluntary csw", ((double)ru->ru_nvcsw / (double)ticks) / hz); 269 fprintf(pmc_args.pa_printfile, "%16ld %s\t#\t%02.03f M/sec\n", 270 ru->ru_nivcsw, "involuntary csw", ((double)ru->ru_nivcsw / (double)ticks) / hz); 271 } 272 273 bzero(&cvals, sizeof(cvals)); 274 STAILQ_FOREACH(ev, &pmc_args.pa_events, ev_next) { 275 if (pmc_read(ev->ev_pmcid, &value) < 0) 276 err(EX_OSERR, "ERROR: Cannot read pmc \"%s\"", 277 ev->ev_name); 278 for (i = 0; i < STAT_MODE_NPMCS; i++) 279 if (strcmp(ev->ev_name, stat_mode_cntrs[i]) == 0) 280 cvals[i] += value; 281 } 282 283 fprintf(pmc_args.pa_printfile, "%16jd %s\n", (uintmax_t)cvals[CYCLES], stat_mode_names[CYCLES]); 284 fprintf(pmc_args.pa_printfile, "%16jd %s\t\t#\t%01.03f inst/cycle\n", (uintmax_t)cvals[INST], stat_mode_names[INST], 285 (double)cvals[INST] / cvals[CYCLES]); 286 fprintf(pmc_args.pa_printfile, "%16jd %s\n", (uintmax_t)cvals[BR], stat_mode_names[BR]); 287 if (stat_mode_names[BR_MISS] == pmc_stat_mode_names[BR_MISS]) 288 fprintf(pmc_args.pa_printfile, "%16jd %s\t\t#\t%.03f%%\n", 289 (uintmax_t)cvals[BR_MISS], stat_mode_names[BR_MISS], 290 100 * ((double)cvals[BR_MISS] / cvals[BR])); 291 else 292 fprintf(pmc_args.pa_printfile, "%16jd %s\n", 293 (uintmax_t)cvals[BR_MISS], stat_mode_names[BR_MISS]); 294 fprintf(pmc_args.pa_printfile, "%16jd %s%s", (uintmax_t)cvals[CACHE], stat_mode_names[CACHE], 295 stat_mode_names[CACHE] != pmc_stat_mode_names[CACHE] ? "\n" : ""); 296 if (stat_mode_names[CACHE] == pmc_stat_mode_names[CACHE]) 297 fprintf(pmc_args.pa_printfile, "\t#\t%.03f refs/inst\n", 298 ((double)cvals[CACHE] / cvals[INST])); 299 fprintf(pmc_args.pa_printfile, "%16jd %s%s", (uintmax_t)cvals[CACHE_MISS], stat_mode_names[CACHE_MISS], 300 stat_mode_names[CACHE_MISS] != pmc_stat_mode_names[CACHE_MISS] ? "\n" : ""); 301 if (stat_mode_names[CACHE_MISS] == pmc_stat_mode_names[CACHE_MISS]) 302 fprintf(pmc_args.pa_printfile, "\t\t#\t%.03f%%\n", 303 100 * ((double)cvals[CACHE_MISS] / cvals[CACHE])); 304 305 if (ru) 306 showtime(pmc_args.pa_printfile, &before_ts, &after, ru); 307} 308 309static struct option longopts[] = { 310 {"events", required_argument, NULL, 'j'}, 311 {NULL, 0, NULL, 0} 312}; 313 314static int 315pmc_stat_internal(int argc, char **argv, int system_mode) 316{ 317 char *event, *r; 318 struct sigaction sa; 319 struct kevent kev; 320 struct rusage ru; 321 struct winsize ws; 322 struct pmcstat_ev *ev; 323 int c, option, runstate; 324 int waitstatus, ru_valid, do_debug; 325 326 do_debug = ru_valid = 0; 327 r = event = NULL; 328 while ((option = getopt_long(argc, argv, "dj:", longopts, NULL)) != -1) { 329 switch (option) { 330 case 'j': 331 r = event = strdup(optarg); 332 break; 333 case 'd': 334 do_debug = 1; 335 break; 336 case '?': 337 default: 338 usage(); 339 } 340 } 341 pmc_args.pa_argc = (argc -= optind); 342 pmc_args.pa_argv = (argv += optind); 343 if (argc == 0) 344 usage(); 345 pmc_args.pa_flags |= FLAG_HAS_COMMANDLINE; 346 pmc_stat_setup_stat(system_mode, event); 347 free(r); 348 bzero(&ru, sizeof(ru)); 349 EV_SET(&kev, SIGINT, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL); 350 if (kevent(pmc_kq, &kev, 1, NULL, 0, NULL) < 0) 351 err(EX_OSERR, "ERROR: Cannot register kevent for SIGINT"); 352 353 EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL); 354 if (kevent(pmc_kq, &kev, 1, NULL, 0, NULL) < 0) 355 err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO"); 356 EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0, 1000, NULL); 357 if (kevent(pmc_kq, &kev, 1, NULL, 0, NULL) < 0) 358 err(EX_OSERR, 359 "ERROR: Cannot register kevent for timer"); 360 361 STAILQ_FOREACH(ev, &pmc_args.pa_events, ev_next) { 362 if (pmc_allocate(ev->ev_spec, ev->ev_mode, 363 ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid, ev->ev_count) < 0) 364 err(EX_OSERR, 365 "ERROR: Cannot allocate %s-mode pmc with specification \"%s\"", 366 PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 367 "system" : "process", ev->ev_spec); 368 369 if (PMC_IS_SAMPLING_MODE(ev->ev_mode) && 370 pmc_set(ev->ev_pmcid, ev->ev_count) < 0) 371 err(EX_OSERR, 372 "ERROR: Cannot set sampling count for PMC \"%s\"", 373 ev->ev_name); 374 } 375 376 /* 377 * An exec() failure of a forked child is signalled by the 378 * child sending the parent a SIGCHLD. We don't register an 379 * actual signal handler for SIGCHLD, but instead use our 380 * kqueue to pick up the signal. 381 */ 382 EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL); 383 if (kevent(pmc_kq, &kev, 1, NULL, 0, NULL) < 0) 384 err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD"); 385 386 pmcstat_create_process(pmcstat_sockpair, &pmc_args, pmc_kq); 387 388 if (SLIST_EMPTY(&pmc_args.pa_targets)) 389 errx(EX_DATAERR, 390 "ERROR: No matching target processes."); 391 if (pmc_args.pa_flags & FLAG_HAS_PROCESS_PMCS) 392 pmcstat_attach_pmcs(&pmc_args); 393 394 /* start the pmcs */ 395 pmc_util_start_pmcs(&pmc_args); 396 397 /* start the (commandline) process if needed */ 398 pmcstat_start_process(pmcstat_sockpair); 399 400 /* Handle SIGINT using the kqueue loop */ 401 sa.sa_handler = SIG_IGN; 402 sa.sa_flags = 0; 403 (void)sigemptyset(&sa.sa_mask); 404 405 if (sigaction(SIGINT, &sa, NULL) < 0) 406 err(EX_OSERR, "ERROR: Cannot install signal handler"); 407 408 /* 409 * loop till either the target process (if any) exits, or we 410 * are killed by a SIGINT or we reached the time duration. 411 */ 412 runstate = PMCSTAT_RUNNING; 413 do { 414 if ((c = kevent(pmc_kq, NULL, 0, &kev, 1, NULL)) <= 0) { 415 if (errno != EINTR) 416 err(EX_OSERR, "ERROR: kevent failed"); 417 else 418 continue; 419 } 420 if (kev.flags & EV_ERROR) 421 errc(EX_OSERR, kev.data, "ERROR: kevent failed"); 422 423 switch (kev.filter) { 424 case EVFILT_PROC: /* target has exited */ 425 if (wait4(pmc_util_get_pid(&pmc_args), &waitstatus, 0, &ru) > 0) { 426 getrusage(RUSAGE_CHILDREN, &ru); 427 ru_valid = 1; 428 } 429 break; 430 431 case EVFILT_READ: /* log file data is present */ 432 break; 433 case EVFILT_TIMER: 434 if (do_debug) 435 pmc_stat_print_stat(NULL); 436 break; 437 case EVFILT_SIGNAL: 438 if (kev.ident == SIGCHLD) { 439 /* 440 * The child process sends us a 441 * SIGCHLD if its exec() failed. We 442 * wait for it to exit and then exit 443 * ourselves. 444 */ 445 (void)wait(&c); 446 runstate = PMCSTAT_FINISHED; 447 } else if (kev.ident == SIGIO) { 448 /* 449 * We get a SIGIO if a PMC loses all 450 * of its targets, or if logfile 451 * writes encounter an error. 452 */ 453 if (wait4(pmc_util_get_pid(&pmc_args), &waitstatus, 0, &ru) > 0) { 454 getrusage(RUSAGE_CHILDREN, &ru); 455 ru_valid = 1; 456 } 457 runstate = pmcstat_close_log(&pmc_args); 458 } else if (kev.ident == SIGINT) { 459 /* Kill the child process if we started it */ 460 if (pmc_args.pa_flags & FLAG_HAS_COMMANDLINE) 461 pmc_util_kill_process(&pmc_args); 462 runstate = pmcstat_close_log(&pmc_args); 463 } else if (kev.ident == SIGWINCH) { 464 if (ioctl(fileno(pmc_args.pa_printfile), 465 TIOCGWINSZ, &ws) < 0) 466 err(EX_OSERR, 467 "ERROR: Cannot determine window size"); 468 pmc_displayheight = ws.ws_row - 1; 469 pmc_displaywidth = ws.ws_col - 1; 470 } else 471 assert(0); 472 473 break; 474 } 475 } while (runstate != PMCSTAT_FINISHED); 476 if (!ru_valid) 477 warnx("couldn't get rusage"); 478 pmc_stat_print_stat(&ru); 479 pmc_util_cleanup(&pmc_args); 480 return (0); 481} 482 483int 484cmd_pmc_stat(int argc, char **argv) 485{ 486 return (pmc_stat_internal(argc, argv, 0)); 487} 488 489int 490cmd_pmc_stat_system(int argc, char **argv) 491{ 492 return (pmc_stat_internal(argc, argv, 1)); 493} 494