Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 1 | |
| 2 | |
Peter Zijlstra | 1a482f3 | 2009-05-23 18:28:58 +0200 | [diff] [blame] | 3 | #include "perf.h" |
Thomas Gleixner | 6eda583 | 2009-05-01 18:29:57 +0200 | [diff] [blame] | 4 | #include "util/util.h" |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 5 | #include "util/parse-options.h" |
Ingo Molnar | 8ad8db3 | 2009-05-26 11:10:09 +0200 | [diff] [blame^] | 6 | #include "util/parse-events.h" |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 7 | #include "util/exec_cmd.h" |
Thomas Gleixner | 6eda583 | 2009-05-01 18:29:57 +0200 | [diff] [blame] | 8 | |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 9 | #include <sched.h> |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 10 | |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 11 | #define ALIGN(x, a) __ALIGN_MASK(x, (typeof(x))(a)-1) |
| 12 | #define __ALIGN_MASK(x, mask) (((x)+(mask))&~(mask)) |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 13 | |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 14 | static int default_interval = 100000; |
| 15 | static int event_count[MAX_COUNTERS]; |
Ingo Molnar | 8ad8db3 | 2009-05-26 11:10:09 +0200 | [diff] [blame^] | 16 | |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 17 | static int fd[MAX_NR_CPUS][MAX_COUNTERS]; |
| 18 | static int nr_cpus = 0; |
| 19 | static unsigned int page_size; |
| 20 | static unsigned int mmap_pages = 16; |
| 21 | static int output; |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 22 | static const char *output_name = "output.perf"; |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 23 | static int group = 0; |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 24 | static unsigned int realtime_prio = 0; |
| 25 | static int system_wide = 0; |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 26 | static pid_t target_pid = -1; |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 27 | static int inherit = 1; |
| 28 | static int nmi = 1; |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 29 | |
| 30 | const unsigned int default_count[] = { |
| 31 | 1000000, |
| 32 | 1000000, |
| 33 | 10000, |
| 34 | 10000, |
| 35 | 1000000, |
| 36 | 10000, |
| 37 | }; |
| 38 | |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 39 | struct mmap_data { |
| 40 | int counter; |
| 41 | void *base; |
| 42 | unsigned int mask; |
| 43 | unsigned int prev; |
| 44 | }; |
| 45 | |
| 46 | static unsigned int mmap_read_head(struct mmap_data *md) |
| 47 | { |
| 48 | struct perf_counter_mmap_page *pc = md->base; |
| 49 | int head; |
| 50 | |
| 51 | head = pc->data_head; |
| 52 | rmb(); |
| 53 | |
| 54 | return head; |
| 55 | } |
| 56 | |
| 57 | static long events; |
| 58 | static struct timeval last_read, this_read; |
| 59 | |
| 60 | static void mmap_read(struct mmap_data *md) |
| 61 | { |
| 62 | unsigned int head = mmap_read_head(md); |
| 63 | unsigned int old = md->prev; |
| 64 | unsigned char *data = md->base + page_size; |
| 65 | unsigned long size; |
| 66 | void *buf; |
| 67 | int diff; |
| 68 | |
| 69 | gettimeofday(&this_read, NULL); |
| 70 | |
| 71 | /* |
| 72 | * If we're further behind than half the buffer, there's a chance |
| 73 | * the writer will bite our tail and screw up the events under us. |
| 74 | * |
| 75 | * If we somehow ended up ahead of the head, we got messed up. |
| 76 | * |
| 77 | * In either case, truncate and restart at head. |
| 78 | */ |
| 79 | diff = head - old; |
| 80 | if (diff > md->mask / 2 || diff < 0) { |
| 81 | struct timeval iv; |
| 82 | unsigned long msecs; |
| 83 | |
| 84 | timersub(&this_read, &last_read, &iv); |
| 85 | msecs = iv.tv_sec*1000 + iv.tv_usec/1000; |
| 86 | |
| 87 | fprintf(stderr, "WARNING: failed to keep up with mmap data." |
| 88 | " Last read %lu msecs ago.\n", msecs); |
| 89 | |
| 90 | /* |
| 91 | * head points to a known good entry, start there. |
| 92 | */ |
| 93 | old = head; |
| 94 | } |
| 95 | |
| 96 | last_read = this_read; |
| 97 | |
| 98 | if (old != head) |
| 99 | events++; |
| 100 | |
| 101 | size = head - old; |
| 102 | |
| 103 | if ((old & md->mask) + size != (head & md->mask)) { |
| 104 | buf = &data[old & md->mask]; |
| 105 | size = md->mask + 1 - (old & md->mask); |
| 106 | old += size; |
| 107 | while (size) { |
| 108 | int ret = write(output, buf, size); |
| 109 | if (ret < 0) { |
| 110 | perror("failed to write"); |
| 111 | exit(-1); |
| 112 | } |
| 113 | size -= ret; |
| 114 | buf += ret; |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | buf = &data[old & md->mask]; |
| 119 | size = head - old; |
| 120 | old += size; |
| 121 | while (size) { |
| 122 | int ret = write(output, buf, size); |
| 123 | if (ret < 0) { |
| 124 | perror("failed to write"); |
| 125 | exit(-1); |
| 126 | } |
| 127 | size -= ret; |
| 128 | buf += ret; |
| 129 | } |
| 130 | |
| 131 | md->prev = old; |
| 132 | } |
| 133 | |
| 134 | static volatile int done = 0; |
| 135 | |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 136 | static void sig_handler(int sig) |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 137 | { |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 138 | done = 1; |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 139 | } |
| 140 | |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 141 | static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS]; |
| 142 | static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS]; |
| 143 | |
| 144 | static int nr_poll; |
| 145 | static int nr_cpu; |
| 146 | |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 147 | struct mmap_event { |
| 148 | struct perf_event_header header; |
| 149 | __u32 pid, tid; |
| 150 | __u64 start; |
| 151 | __u64 len; |
| 152 | __u64 pgoff; |
| 153 | char filename[PATH_MAX]; |
| 154 | }; |
| 155 | struct comm_event { |
| 156 | struct perf_event_header header; |
| 157 | __u32 pid,tid; |
| 158 | char comm[16]; |
| 159 | }; |
| 160 | |
| 161 | static pid_t pid_synthesize_comm_event(pid_t pid) |
| 162 | { |
| 163 | char filename[PATH_MAX]; |
| 164 | char bf[BUFSIZ]; |
| 165 | struct comm_event comm_ev; |
| 166 | size_t size; |
| 167 | int fd; |
| 168 | |
| 169 | snprintf(filename, sizeof(filename), "/proc/%d/stat", pid); |
| 170 | |
| 171 | fd = open(filename, O_RDONLY); |
| 172 | if (fd < 0) { |
| 173 | fprintf(stderr, "couldn't open %s\n", filename); |
| 174 | exit(EXIT_FAILURE); |
| 175 | } |
| 176 | if (read(fd, bf, sizeof(bf)) < 0) { |
| 177 | fprintf(stderr, "couldn't read %s\n", filename); |
| 178 | exit(EXIT_FAILURE); |
| 179 | } |
| 180 | close(fd); |
| 181 | |
| 182 | pid_t spid, ppid; |
| 183 | char state; |
| 184 | char comm[18]; |
| 185 | |
| 186 | memset(&comm_ev, 0, sizeof(comm_ev)); |
| 187 | int nr = sscanf(bf, "%d %s %c %d %d ", |
| 188 | &spid, comm, &state, &ppid, &comm_ev.pid); |
| 189 | if (nr != 5) { |
| 190 | fprintf(stderr, "couldn't get COMM and pgid, malformed %s\n", |
| 191 | filename); |
| 192 | exit(EXIT_FAILURE); |
| 193 | } |
| 194 | comm_ev.header.type = PERF_EVENT_COMM; |
| 195 | comm_ev.tid = pid; |
| 196 | size = strlen(comm); |
| 197 | comm[--size] = '\0'; /* Remove the ')' at the end */ |
| 198 | --size; /* Remove the '(' at the begin */ |
| 199 | memcpy(comm_ev.comm, comm + 1, size); |
| 200 | size = ALIGN(size, sizeof(uint64_t)); |
| 201 | comm_ev.header.size = sizeof(comm_ev) - (sizeof(comm_ev.comm) - size); |
| 202 | int ret = write(output, &comm_ev, comm_ev.header.size); |
| 203 | if (ret < 0) { |
| 204 | perror("failed to write"); |
| 205 | exit(-1); |
| 206 | } |
| 207 | return comm_ev.pid; |
| 208 | } |
| 209 | |
| 210 | static void pid_synthesize_mmap_events(pid_t pid, pid_t pgid) |
| 211 | { |
| 212 | char filename[PATH_MAX]; |
| 213 | FILE *fp; |
| 214 | |
| 215 | snprintf(filename, sizeof(filename), "/proc/%d/maps", pid); |
| 216 | |
| 217 | fp = fopen(filename, "r"); |
| 218 | if (fp == NULL) { |
| 219 | fprintf(stderr, "couldn't open %s\n", filename); |
| 220 | exit(EXIT_FAILURE); |
| 221 | } |
| 222 | while (1) { |
| 223 | char bf[BUFSIZ]; |
| 224 | unsigned char vm_read, vm_write, vm_exec, vm_mayshare; |
| 225 | struct mmap_event mmap_ev = { |
| 226 | .header.type = PERF_EVENT_MMAP, |
| 227 | }; |
| 228 | unsigned long ino; |
| 229 | int major, minor; |
| 230 | size_t size; |
| 231 | if (fgets(bf, sizeof(bf), fp) == NULL) |
| 232 | break; |
| 233 | |
| 234 | /* 00400000-0040c000 r-xp 00000000 fd:01 41038 /bin/cat */ |
| 235 | sscanf(bf, "%llx-%llx %c%c%c%c %llx %x:%x %lu", |
| 236 | &mmap_ev.start, &mmap_ev.len, |
| 237 | &vm_read, &vm_write, &vm_exec, &vm_mayshare, |
| 238 | &mmap_ev.pgoff, &major, &minor, &ino); |
| 239 | if (vm_exec == 'x') { |
| 240 | char *execname = strrchr(bf, ' '); |
| 241 | |
| 242 | if (execname == NULL || execname[1] != '/') |
| 243 | continue; |
| 244 | |
| 245 | execname += 1; |
| 246 | size = strlen(execname); |
| 247 | execname[size - 1] = '\0'; /* Remove \n */ |
| 248 | memcpy(mmap_ev.filename, execname, size); |
| 249 | size = ALIGN(size, sizeof(uint64_t)); |
| 250 | mmap_ev.len -= mmap_ev.start; |
| 251 | mmap_ev.header.size = (sizeof(mmap_ev) - |
| 252 | (sizeof(mmap_ev.filename) - size)); |
| 253 | mmap_ev.pid = pgid; |
| 254 | mmap_ev.tid = pid; |
| 255 | |
| 256 | if (write(output, &mmap_ev, mmap_ev.header.size) < 0) { |
| 257 | perror("failed to write"); |
| 258 | exit(-1); |
| 259 | } |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | fclose(fp); |
| 264 | } |
| 265 | |
| 266 | static void open_counters(int cpu, pid_t pid) |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 267 | { |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 268 | struct perf_counter_hw_event hw_event; |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 269 | int counter, group_fd; |
| 270 | int track = 1; |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 271 | |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 272 | if (pid > 0) { |
| 273 | pid_t pgid = pid_synthesize_comm_event(pid); |
| 274 | pid_synthesize_mmap_events(pid, pgid); |
| 275 | } |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 276 | |
| 277 | group_fd = -1; |
| 278 | for (counter = 0; counter < nr_counters; counter++) { |
| 279 | |
| 280 | memset(&hw_event, 0, sizeof(hw_event)); |
| 281 | hw_event.config = event_id[counter]; |
| 282 | hw_event.irq_period = event_count[counter]; |
| 283 | hw_event.record_type = PERF_RECORD_IP | PERF_RECORD_TID; |
| 284 | hw_event.nmi = nmi; |
| 285 | hw_event.mmap = track; |
| 286 | hw_event.comm = track; |
| 287 | hw_event.inherit = (cpu < 0) && inherit; |
| 288 | |
| 289 | track = 0; // only the first counter needs these |
| 290 | |
| 291 | fd[nr_cpu][counter] = |
| 292 | sys_perf_counter_open(&hw_event, pid, cpu, group_fd, 0); |
| 293 | |
| 294 | if (fd[nr_cpu][counter] < 0) { |
| 295 | int err = errno; |
| 296 | printf("kerneltop error: syscall returned with %d (%s)\n", |
| 297 | fd[nr_cpu][counter], strerror(err)); |
| 298 | if (err == EPERM) |
| 299 | printf("Are you root?\n"); |
| 300 | exit(-1); |
| 301 | } |
| 302 | assert(fd[nr_cpu][counter] >= 0); |
| 303 | fcntl(fd[nr_cpu][counter], F_SETFL, O_NONBLOCK); |
| 304 | |
| 305 | /* |
| 306 | * First counter acts as the group leader: |
| 307 | */ |
| 308 | if (group && group_fd == -1) |
| 309 | group_fd = fd[nr_cpu][counter]; |
| 310 | |
| 311 | event_array[nr_poll].fd = fd[nr_cpu][counter]; |
| 312 | event_array[nr_poll].events = POLLIN; |
| 313 | nr_poll++; |
| 314 | |
| 315 | mmap_array[nr_cpu][counter].counter = counter; |
| 316 | mmap_array[nr_cpu][counter].prev = 0; |
| 317 | mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1; |
| 318 | mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size, |
| 319 | PROT_READ, MAP_SHARED, fd[nr_cpu][counter], 0); |
| 320 | if (mmap_array[nr_cpu][counter].base == MAP_FAILED) { |
| 321 | printf("kerneltop error: failed to mmap with %d (%s)\n", |
| 322 | errno, strerror(errno)); |
| 323 | exit(-1); |
| 324 | } |
| 325 | } |
| 326 | nr_cpu++; |
| 327 | } |
| 328 | |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 329 | static int __cmd_record(int argc, const char **argv) |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 330 | { |
| 331 | int i, counter; |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 332 | pid_t pid; |
| 333 | int ret; |
| 334 | |
| 335 | page_size = sysconf(_SC_PAGE_SIZE); |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 336 | nr_cpus = sysconf(_SC_NPROCESSORS_ONLN); |
| 337 | assert(nr_cpus <= MAX_NR_CPUS); |
| 338 | assert(nr_cpus >= 0); |
| 339 | |
| 340 | output = open(output_name, O_CREAT|O_RDWR, S_IRWXU); |
| 341 | if (output < 0) { |
| 342 | perror("failed to create output file"); |
| 343 | exit(-1); |
| 344 | } |
| 345 | |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 346 | if (!system_wide) { |
| 347 | open_counters(-1, target_pid != -1 ? target_pid : 0); |
| 348 | } else for (i = 0; i < nr_cpus; i++) |
| 349 | open_counters(i, target_pid); |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 350 | |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 351 | signal(SIGCHLD, sig_handler); |
| 352 | signal(SIGINT, sig_handler); |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 353 | |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 354 | if (target_pid == -1) { |
| 355 | pid = fork(); |
| 356 | if (pid < 0) |
| 357 | perror("failed to fork"); |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 358 | |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 359 | if (!pid) { |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 360 | if (execvp(argv[0], (char **)argv)) { |
Arnaldo Carvalho de Melo | 1a853e3 | 2009-05-14 22:50:46 -0300 | [diff] [blame] | 361 | perror(argv[0]); |
| 362 | exit(-1); |
| 363 | } |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 364 | } |
| 365 | } |
| 366 | |
| 367 | if (realtime_prio) { |
| 368 | struct sched_param param; |
| 369 | |
| 370 | param.sched_priority = realtime_prio; |
| 371 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { |
| 372 | printf("Could not set realtime priority.\n"); |
| 373 | exit(-1); |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | /* |
| 378 | * TODO: store the current /proc/$/maps information somewhere |
| 379 | */ |
| 380 | |
| 381 | while (!done) { |
| 382 | int hits = events; |
| 383 | |
Peter Zijlstra | 16c8a10 | 2009-05-05 17:50:27 +0200 | [diff] [blame] | 384 | for (i = 0; i < nr_cpu; i++) { |
Peter Zijlstra | de9ac07 | 2009-04-08 15:01:31 +0200 | [diff] [blame] | 385 | for (counter = 0; counter < nr_counters; counter++) |
| 386 | mmap_read(&mmap_array[i][counter]); |
| 387 | } |
| 388 | |
| 389 | if (hits == events) |
| 390 | ret = poll(event_array, nr_poll, 100); |
| 391 | } |
| 392 | |
| 393 | return 0; |
| 394 | } |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 395 | |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 396 | static const char * const record_usage[] = { |
| 397 | "perf record [<options>] <command>", |
| 398 | NULL |
| 399 | }; |
| 400 | |
Ingo Molnar | 8ad8db3 | 2009-05-26 11:10:09 +0200 | [diff] [blame^] | 401 | static char events_help_msg[EVENTS_HELP_MAX]; |
| 402 | |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 403 | const struct option options[] = { |
| 404 | OPT_CALLBACK('e', "event", NULL, "event", |
Ingo Molnar | 8ad8db3 | 2009-05-26 11:10:09 +0200 | [diff] [blame^] | 405 | events_help_msg, parse_events), |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 406 | OPT_INTEGER('c', "count", &default_interval, |
| 407 | "event period to sample"), |
| 408 | OPT_INTEGER('m', "mmap-pages", &mmap_pages, |
| 409 | "number of mmap data pages"), |
| 410 | OPT_STRING('o', "output", &output_name, "file", |
| 411 | "output file name"), |
| 412 | OPT_BOOLEAN('i', "inherit", &inherit, |
| 413 | "child tasks inherit counters"), |
| 414 | OPT_INTEGER('p', "pid", &target_pid, |
| 415 | "record events on existing pid"), |
| 416 | OPT_INTEGER('r', "realtime", &realtime_prio, |
| 417 | "collect data with this RT SCHED_FIFO priority"), |
| 418 | OPT_BOOLEAN('a', "all-cpus", &system_wide, |
| 419 | "system-wide collection from all CPUs"), |
| 420 | OPT_END() |
| 421 | }; |
| 422 | |
| 423 | int cmd_record(int argc, const char **argv, const char *prefix) |
| 424 | { |
| 425 | int counter; |
| 426 | |
Ingo Molnar | 8ad8db3 | 2009-05-26 11:10:09 +0200 | [diff] [blame^] | 427 | create_events_help(events_help_msg); |
Ingo Molnar | 0e9b20b | 2009-05-26 09:17:18 +0200 | [diff] [blame] | 428 | |
| 429 | argc = parse_options(argc, argv, options, record_usage, 0); |
| 430 | if (!argc) |
| 431 | usage_with_options(record_usage, options); |
| 432 | |
| 433 | if (!nr_counters) { |
| 434 | nr_counters = 1; |
| 435 | event_id[0] = 0; |
| 436 | } |
| 437 | |
| 438 | for (counter = 0; counter < nr_counters; counter++) { |
| 439 | if (event_count[counter]) |
| 440 | continue; |
| 441 | |
| 442 | event_count[counter] = default_interval; |
| 443 | } |
| 444 | |
| 445 | return __cmd_record(argc, argv); |
| 446 | } |