]>
Commit | Line | Data |
---|---|---|
07800601 | 1 | /* |
bf9e1876 IM |
2 | * builtin-top.c |
3 | * | |
4 | * Builtin top command: Display a continuously updated profile of | |
5 | * any workload, CPU or specific PID. | |
6 | * | |
7 | * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com> | |
8 | * | |
9 | * Improvements and fixes by: | |
10 | * | |
11 | * Arjan van de Ven <arjan@linux.intel.com> | |
12 | * Yanmin Zhang <yanmin.zhang@intel.com> | |
13 | * Wu Fengguang <fengguang.wu@intel.com> | |
14 | * Mike Galbraith <efault@gmx.de> | |
15 | * Paul Mackerras <paulus@samba.org> | |
16 | * | |
17 | * Released under the GPL v2. (and only v2, not any later version) | |
07800601 | 18 | */ |
bf9e1876 | 19 | #include "builtin.h" |
07800601 | 20 | |
1a482f38 | 21 | #include "perf.h" |
bf9e1876 | 22 | |
de04687f | 23 | #include "util/symbol.h" |
8fc0321f | 24 | #include "util/color.h" |
148be2c1 | 25 | #include "util/util.h" |
43cbcd8a | 26 | #include <linux/rbtree.h> |
b456bae0 IM |
27 | #include "util/parse-options.h" |
28 | #include "util/parse-events.h" | |
07800601 | 29 | |
07800601 IM |
30 | #include <assert.h> |
31 | #include <fcntl.h> | |
0e9b20b8 | 32 | |
07800601 | 33 | #include <stdio.h> |
0e9b20b8 | 34 | |
07800601 | 35 | #include <errno.h> |
07800601 IM |
36 | #include <time.h> |
37 | #include <sched.h> | |
38 | #include <pthread.h> | |
39 | ||
40 | #include <sys/syscall.h> | |
41 | #include <sys/ioctl.h> | |
42 | #include <sys/poll.h> | |
43 | #include <sys/prctl.h> | |
44 | #include <sys/wait.h> | |
45 | #include <sys/uio.h> | |
46 | #include <sys/mman.h> | |
47 | ||
48 | #include <linux/unistd.h> | |
49 | #include <linux/types.h> | |
50 | ||
a21ca2ca | 51 | static int fd[MAX_NR_CPUS][MAX_COUNTERS]; |
07800601 | 52 | |
a21ca2ca | 53 | static int system_wide = 0; |
07800601 | 54 | |
a21ca2ca | 55 | static int default_interval = 100000; |
07800601 | 56 | |
9cffa8d5 | 57 | static u64 count_filter = 5; |
6e53cdf1 | 58 | static int print_entries = 15; |
07800601 | 59 | |
6e53cdf1 | 60 | static int target_pid = -1; |
07800601 IM |
61 | static int profile_cpu = -1; |
62 | static int nr_cpus = 0; | |
07800601 IM |
63 | static unsigned int realtime_prio = 0; |
64 | static int group = 0; | |
65 | static unsigned int page_size; | |
cf1f4574 IM |
66 | static unsigned int mmap_pages = 16; |
67 | static int freq = 0; | |
3da297a6 | 68 | static int verbose = 0; |
42976487 | 69 | static char *vmlinux = NULL; |
07800601 | 70 | |
07800601 IM |
71 | static char *sym_filter; |
72 | static unsigned long filter_start; | |
73 | static unsigned long filter_end; | |
74 | ||
75 | static int delay_secs = 2; | |
76 | static int zero; | |
77 | static int dump_symtab; | |
78 | ||
07800601 IM |
79 | /* |
80 | * Symbols | |
81 | */ | |
82 | ||
9cffa8d5 PM |
83 | static u64 min_ip; |
84 | static u64 max_ip = -1ll; | |
07800601 IM |
85 | |
86 | struct sym_entry { | |
de04687f ACM |
87 | struct rb_node rb_node; |
88 | struct list_head node; | |
07800601 | 89 | unsigned long count[MAX_COUNTERS]; |
c44613a4 ACM |
90 | unsigned long snap_count; |
91 | double weight; | |
07800601 | 92 | int skip; |
07800601 IM |
93 | }; |
94 | ||
07800601 IM |
95 | struct sym_entry *sym_filter_entry; |
96 | ||
a21ca2ca | 97 | struct dso *kernel_dso; |
de04687f ACM |
98 | |
99 | /* | |
100 | * Symbols will be added here in record_ip and will get out | |
101 | * after decayed. | |
102 | */ | |
103 | static LIST_HEAD(active_symbols); | |
c44613a4 | 104 | static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER; |
07800601 | 105 | |
07800601 IM |
106 | /* |
107 | * Ordering weight: count-1 * count-2 * ... / count-n | |
108 | */ | |
109 | static double sym_weight(const struct sym_entry *sym) | |
110 | { | |
c44613a4 | 111 | double weight = sym->snap_count; |
07800601 IM |
112 | int counter; |
113 | ||
07800601 IM |
114 | for (counter = 1; counter < nr_counters-1; counter++) |
115 | weight *= sym->count[counter]; | |
116 | ||
117 | weight /= (sym->count[counter] + 1); | |
118 | ||
119 | return weight; | |
120 | } | |
121 | ||
2debbc83 IM |
122 | static long samples; |
123 | static long userspace_samples; | |
07800601 IM |
124 | static const char CONSOLE_CLEAR[] = "\e[H\e[2J"; |
125 | ||
c44613a4 | 126 | static void __list_insert_active_sym(struct sym_entry *syme) |
de04687f ACM |
127 | { |
128 | list_add(&syme->node, &active_symbols); | |
129 | } | |
130 | ||
c44613a4 ACM |
131 | static void list_remove_active_sym(struct sym_entry *syme) |
132 | { | |
133 | pthread_mutex_lock(&active_symbols_lock); | |
134 | list_del_init(&syme->node); | |
135 | pthread_mutex_unlock(&active_symbols_lock); | |
136 | } | |
137 | ||
de04687f ACM |
138 | static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se) |
139 | { | |
140 | struct rb_node **p = &tree->rb_node; | |
141 | struct rb_node *parent = NULL; | |
142 | struct sym_entry *iter; | |
143 | ||
144 | while (*p != NULL) { | |
145 | parent = *p; | |
146 | iter = rb_entry(parent, struct sym_entry, rb_node); | |
147 | ||
c44613a4 | 148 | if (se->weight > iter->weight) |
de04687f ACM |
149 | p = &(*p)->rb_left; |
150 | else | |
151 | p = &(*p)->rb_right; | |
152 | } | |
153 | ||
154 | rb_link_node(&se->rb_node, parent, p); | |
155 | rb_insert_color(&se->rb_node, tree); | |
156 | } | |
07800601 IM |
157 | |
158 | static void print_sym_table(void) | |
159 | { | |
233f0b95 | 160 | int printed = 0, j; |
07800601 | 161 | int counter; |
2debbc83 IM |
162 | float samples_per_sec = samples/delay_secs; |
163 | float ksamples_per_sec = (samples-userspace_samples)/delay_secs; | |
164 | float sum_ksamples = 0.0; | |
de04687f ACM |
165 | struct sym_entry *syme, *n; |
166 | struct rb_root tmp = RB_ROOT; | |
167 | struct rb_node *nd; | |
07800601 | 168 | |
2debbc83 | 169 | samples = userspace_samples = 0; |
07800601 | 170 | |
de04687f | 171 | /* Sort the active symbols */ |
c44613a4 ACM |
172 | pthread_mutex_lock(&active_symbols_lock); |
173 | syme = list_entry(active_symbols.next, struct sym_entry, node); | |
174 | pthread_mutex_unlock(&active_symbols_lock); | |
175 | ||
176 | list_for_each_entry_safe_from(syme, n, &active_symbols, node) { | |
177 | syme->snap_count = syme->count[0]; | |
178 | if (syme->snap_count != 0) { | |
179 | syme->weight = sym_weight(syme); | |
de04687f | 180 | rb_insert_active_sym(&tmp, syme); |
2debbc83 | 181 | sum_ksamples += syme->snap_count; |
d94b9430 MG |
182 | |
183 | for (j = 0; j < nr_counters; j++) | |
de04687f ACM |
184 | syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8; |
185 | } else | |
c44613a4 | 186 | list_remove_active_sym(syme); |
d94b9430 MG |
187 | } |
188 | ||
0f5486b5 | 189 | puts(CONSOLE_CLEAR); |
07800601 IM |
190 | |
191 | printf( | |
192 | "------------------------------------------------------------------------------\n"); | |
f2521b6e | 193 | printf( " PerfTop:%8.0f irqs/sec kernel:%4.1f%% [", |
2debbc83 IM |
194 | samples_per_sec, |
195 | 100.0 - (100.0*((samples_per_sec-ksamples_per_sec)/samples_per_sec))); | |
07800601 | 196 | |
cf1f4574 | 197 | if (nr_counters == 1) { |
9cffa8d5 | 198 | printf("%Ld", (u64)attrs[0].sample_period); |
cf1f4574 IM |
199 | if (freq) |
200 | printf("Hz "); | |
201 | else | |
202 | printf(" "); | |
203 | } | |
07800601 IM |
204 | |
205 | for (counter = 0; counter < nr_counters; counter++) { | |
206 | if (counter) | |
207 | printf("/"); | |
208 | ||
209 | printf("%s", event_name(counter)); | |
210 | } | |
211 | ||
212 | printf( "], "); | |
213 | ||
b456bae0 IM |
214 | if (target_pid != -1) |
215 | printf(" (target_pid: %d", target_pid); | |
07800601 IM |
216 | else |
217 | printf(" (all"); | |
218 | ||
219 | if (profile_cpu != -1) | |
220 | printf(", cpu: %d)\n", profile_cpu); | |
221 | else { | |
b456bae0 | 222 | if (target_pid != -1) |
07800601 IM |
223 | printf(")\n"); |
224 | else | |
225 | printf(", %d CPUs)\n", nr_cpus); | |
226 | } | |
227 | ||
228 | printf("------------------------------------------------------------------------------\n\n"); | |
229 | ||
230 | if (nr_counters == 1) | |
2debbc83 | 231 | printf(" samples pcnt"); |
07800601 | 232 | else |
2debbc83 | 233 | printf(" weight samples pcnt"); |
07800601 IM |
234 | |
235 | printf(" RIP kernel function\n" | |
2debbc83 | 236 | " ______ _______ _____ ________________ _______________\n\n" |
07800601 IM |
237 | ); |
238 | ||
de04687f ACM |
239 | for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) { |
240 | struct sym_entry *syme = rb_entry(nd, struct sym_entry, rb_node); | |
241 | struct symbol *sym = (struct symbol *)(syme + 1); | |
8fc0321f | 242 | double pcnt; |
d94b9430 | 243 | |
6e53cdf1 | 244 | if (++printed > print_entries || syme->snap_count < count_filter) |
c44613a4 | 245 | continue; |
d94b9430 | 246 | |
2debbc83 IM |
247 | pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) / |
248 | sum_ksamples)); | |
d94b9430 MG |
249 | |
250 | if (nr_counters == 1) | |
2debbc83 | 251 | printf("%20.2f - ", syme->weight); |
d94b9430 | 252 | else |
2debbc83 | 253 | printf("%9.1f %10ld - ", syme->weight, syme->snap_count); |
8fc0321f | 254 | |
1e11fd82 | 255 | percent_color_fprintf(stdout, "%4.1f%%", pcnt); |
42976487 MG |
256 | printf(" - %016llx : %s", sym->start, sym->name); |
257 | if (sym->module) | |
258 | printf("\t[%s]", sym->module->name); | |
259 | printf("\n"); | |
07800601 | 260 | } |
07800601 IM |
261 | } |
262 | ||
f37a291c | 263 | static void *display_thread(void *arg __used) |
07800601 | 264 | { |
0f5486b5 FW |
265 | struct pollfd stdin_poll = { .fd = 0, .events = POLLIN }; |
266 | int delay_msecs = delay_secs * 1000; | |
267 | ||
f2521b6e | 268 | printf("PerfTop refresh period: %d seconds\n", delay_secs); |
07800601 | 269 | |
0f5486b5 | 270 | do { |
07800601 | 271 | print_sym_table(); |
0f5486b5 FW |
272 | } while (!poll(&stdin_poll, 1, delay_msecs) == 1); |
273 | ||
274 | printf("key pressed - exiting.\n"); | |
275 | exit(0); | |
07800601 IM |
276 | |
277 | return NULL; | |
278 | } | |
279 | ||
2ab52083 | 280 | /* Tag samples to be skipped. */ |
f37a291c | 281 | static const char *skip_symbols[] = { |
2ab52083 AB |
282 | "default_idle", |
283 | "cpu_idle", | |
284 | "enter_idle", | |
285 | "exit_idle", | |
286 | "mwait_idle", | |
3a3393ef AB |
287 | "ppc64_runlatch_off", |
288 | "pseries_dedicated_idle_sleep", | |
2ab52083 AB |
289 | NULL |
290 | }; | |
291 | ||
de04687f | 292 | static int symbol_filter(struct dso *self, struct symbol *sym) |
07800601 | 293 | { |
de04687f ACM |
294 | static int filter_match; |
295 | struct sym_entry *syme; | |
296 | const char *name = sym->name; | |
2ab52083 | 297 | int i; |
de04687f | 298 | |
3a3393ef AB |
299 | /* |
300 | * ppc64 uses function descriptors and appends a '.' to the | |
301 | * start of every instruction address. Remove it. | |
302 | */ | |
303 | if (name[0] == '.') | |
304 | name++; | |
305 | ||
de04687f ACM |
306 | if (!strcmp(name, "_text") || |
307 | !strcmp(name, "_etext") || | |
308 | !strcmp(name, "_sinittext") || | |
309 | !strncmp("init_module", name, 11) || | |
310 | !strncmp("cleanup_module", name, 14) || | |
311 | strstr(name, "_text_start") || | |
312 | strstr(name, "_text_end")) | |
07800601 | 313 | return 1; |
07800601 | 314 | |
de04687f | 315 | syme = dso__sym_priv(self, sym); |
2ab52083 AB |
316 | for (i = 0; skip_symbols[i]; i++) { |
317 | if (!strcmp(skip_symbols[i], name)) { | |
318 | syme->skip = 1; | |
319 | break; | |
320 | } | |
321 | } | |
07800601 IM |
322 | |
323 | if (filter_match == 1) { | |
de04687f | 324 | filter_end = sym->start; |
07800601 IM |
325 | filter_match = -1; |
326 | if (filter_end - filter_start > 10000) { | |
de04687f ACM |
327 | fprintf(stderr, |
328 | "hm, too large filter symbol <%s> - skipping.\n", | |
07800601 | 329 | sym_filter); |
de04687f ACM |
330 | fprintf(stderr, "symbol filter start: %016lx\n", |
331 | filter_start); | |
332 | fprintf(stderr, " end: %016lx\n", | |
333 | filter_end); | |
07800601 IM |
334 | filter_end = filter_start = 0; |
335 | sym_filter = NULL; | |
336 | sleep(1); | |
337 | } | |
338 | } | |
de04687f ACM |
339 | |
340 | if (filter_match == 0 && sym_filter && !strcmp(name, sym_filter)) { | |
07800601 | 341 | filter_match = 1; |
de04687f | 342 | filter_start = sym->start; |
07800601 IM |
343 | } |
344 | ||
de04687f | 345 | |
07800601 IM |
346 | return 0; |
347 | } | |
348 | ||
de04687f | 349 | static int parse_symbols(void) |
07800601 | 350 | { |
de04687f ACM |
351 | struct rb_node *node; |
352 | struct symbol *sym; | |
42976487 | 353 | int modules = vmlinux ? 1 : 0; |
07800601 | 354 | |
de04687f ACM |
355 | kernel_dso = dso__new("[kernel]", sizeof(struct sym_entry)); |
356 | if (kernel_dso == NULL) | |
357 | return -1; | |
07800601 | 358 | |
42976487 | 359 | if (dso__load_kernel(kernel_dso, vmlinux, symbol_filter, verbose, modules) <= 0) |
de04687f | 360 | goto out_delete_dso; |
07800601 | 361 | |
de04687f ACM |
362 | node = rb_first(&kernel_dso->syms); |
363 | sym = rb_entry(node, struct symbol, rb_node); | |
364 | min_ip = sym->start; | |
07800601 | 365 | |
de04687f ACM |
366 | node = rb_last(&kernel_dso->syms); |
367 | sym = rb_entry(node, struct symbol, rb_node); | |
da417a75 | 368 | max_ip = sym->end; |
07800601 | 369 | |
de04687f | 370 | if (dump_symtab) |
a3ec8d70 | 371 | dso__fprintf(kernel_dso, stderr); |
07800601 | 372 | |
de04687f | 373 | return 0; |
07800601 | 374 | |
de04687f ACM |
375 | out_delete_dso: |
376 | dso__delete(kernel_dso); | |
377 | kernel_dso = NULL; | |
378 | return -1; | |
07800601 IM |
379 | } |
380 | ||
07800601 IM |
381 | #define TRACE_COUNT 3 |
382 | ||
07800601 IM |
383 | /* |
384 | * Binary search in the histogram table and record the hit: | |
385 | */ | |
9cffa8d5 | 386 | static void record_ip(u64 ip, int counter) |
07800601 | 387 | { |
de04687f | 388 | struct symbol *sym = dso__find_symbol(kernel_dso, ip); |
07800601 | 389 | |
de04687f ACM |
390 | if (sym != NULL) { |
391 | struct sym_entry *syme = dso__sym_priv(kernel_dso, sym); | |
07800601 | 392 | |
de04687f ACM |
393 | if (!syme->skip) { |
394 | syme->count[counter]++; | |
c44613a4 | 395 | pthread_mutex_lock(&active_symbols_lock); |
de04687f | 396 | if (list_empty(&syme->node) || !syme->node.next) |
c44613a4 ACM |
397 | __list_insert_active_sym(syme); |
398 | pthread_mutex_unlock(&active_symbols_lock); | |
de04687f | 399 | return; |
07800601 | 400 | } |
07800601 IM |
401 | } |
402 | ||
2debbc83 | 403 | samples--; |
07800601 IM |
404 | } |
405 | ||
e6e18ec7 | 406 | static void process_event(u64 ip, int counter, int user) |
07800601 | 407 | { |
2debbc83 | 408 | samples++; |
07800601 | 409 | |
e6e18ec7 | 410 | if (user) { |
2debbc83 | 411 | userspace_samples++; |
07800601 IM |
412 | return; |
413 | } | |
414 | ||
415 | record_ip(ip, counter); | |
416 | } | |
417 | ||
07800601 | 418 | struct mmap_data { |
a21ca2ca IM |
419 | int counter; |
420 | void *base; | |
f37a291c | 421 | int mask; |
a21ca2ca | 422 | unsigned int prev; |
07800601 IM |
423 | }; |
424 | ||
425 | static unsigned int mmap_read_head(struct mmap_data *md) | |
426 | { | |
427 | struct perf_counter_mmap_page *pc = md->base; | |
428 | int head; | |
429 | ||
430 | head = pc->data_head; | |
431 | rmb(); | |
432 | ||
433 | return head; | |
434 | } | |
435 | ||
436 | struct timeval last_read, this_read; | |
437 | ||
2f01190a | 438 | static void mmap_read_counter(struct mmap_data *md) |
07800601 IM |
439 | { |
440 | unsigned int head = mmap_read_head(md); | |
441 | unsigned int old = md->prev; | |
442 | unsigned char *data = md->base + page_size; | |
443 | int diff; | |
444 | ||
445 | gettimeofday(&this_read, NULL); | |
446 | ||
447 | /* | |
448 | * If we're further behind than half the buffer, there's a chance | |
2debbc83 | 449 | * the writer will bite our tail and mess up the samples under us. |
07800601 IM |
450 | * |
451 | * If we somehow ended up ahead of the head, we got messed up. | |
452 | * | |
453 | * In either case, truncate and restart at head. | |
454 | */ | |
455 | diff = head - old; | |
456 | if (diff > md->mask / 2 || diff < 0) { | |
457 | struct timeval iv; | |
458 | unsigned long msecs; | |
459 | ||
460 | timersub(&this_read, &last_read, &iv); | |
461 | msecs = iv.tv_sec*1000 + iv.tv_usec/1000; | |
462 | ||
463 | fprintf(stderr, "WARNING: failed to keep up with mmap data." | |
464 | " Last read %lu msecs ago.\n", msecs); | |
465 | ||
466 | /* | |
467 | * head points to a known good entry, start there. | |
468 | */ | |
469 | old = head; | |
470 | } | |
471 | ||
472 | last_read = this_read; | |
473 | ||
474 | for (; old != head;) { | |
475 | struct ip_event { | |
476 | struct perf_event_header header; | |
9cffa8d5 PM |
477 | u64 ip; |
478 | u32 pid, target_pid; | |
07800601 IM |
479 | }; |
480 | struct mmap_event { | |
481 | struct perf_event_header header; | |
9cffa8d5 PM |
482 | u32 pid, target_pid; |
483 | u64 start; | |
484 | u64 len; | |
485 | u64 pgoff; | |
07800601 IM |
486 | char filename[PATH_MAX]; |
487 | }; | |
488 | ||
489 | typedef union event_union { | |
490 | struct perf_event_header header; | |
491 | struct ip_event ip; | |
492 | struct mmap_event mmap; | |
493 | } event_t; | |
494 | ||
495 | event_t *event = (event_t *)&data[old & md->mask]; | |
496 | ||
497 | event_t event_copy; | |
498 | ||
6f06ccbc | 499 | size_t size = event->header.size; |
07800601 IM |
500 | |
501 | /* | |
502 | * Event straddles the mmap boundary -- header should always | |
503 | * be inside due to u64 alignment of output. | |
504 | */ | |
505 | if ((old & md->mask) + size != ((old + size) & md->mask)) { | |
506 | unsigned int offset = old; | |
507 | unsigned int len = min(sizeof(*event), size), cpy; | |
508 | void *dst = &event_copy; | |
509 | ||
510 | do { | |
511 | cpy = min(md->mask + 1 - (offset & md->mask), len); | |
512 | memcpy(dst, &data[offset & md->mask], cpy); | |
513 | offset += cpy; | |
514 | dst += cpy; | |
515 | len -= cpy; | |
516 | } while (len); | |
517 | ||
518 | event = &event_copy; | |
519 | } | |
520 | ||
521 | old += size; | |
522 | ||
e6e18ec7 PZ |
523 | if (event->header.type == PERF_EVENT_SAMPLE) { |
524 | int user = | |
525 | (event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK) == PERF_EVENT_MISC_USER; | |
526 | process_event(event->ip.ip, md->counter, user); | |
07800601 IM |
527 | } |
528 | } | |
529 | ||
530 | md->prev = old; | |
531 | } | |
532 | ||
c2990a2a MG |
533 | static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS]; |
534 | static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS]; | |
535 | ||
2f01190a FW |
536 | static void mmap_read(void) |
537 | { | |
538 | int i, counter; | |
539 | ||
540 | for (i = 0; i < nr_cpus; i++) { | |
541 | for (counter = 0; counter < nr_counters; counter++) | |
542 | mmap_read_counter(&mmap_array[i][counter]); | |
543 | } | |
544 | } | |
545 | ||
716c69fe IM |
546 | int nr_poll; |
547 | int group_fd; | |
548 | ||
549 | static void start_counter(int i, int counter) | |
07800601 | 550 | { |
a21ca2ca | 551 | struct perf_counter_attr *attr; |
07800601 | 552 | unsigned int cpu; |
716c69fe IM |
553 | |
554 | cpu = profile_cpu; | |
555 | if (target_pid == -1 && profile_cpu == -1) | |
556 | cpu = i; | |
557 | ||
558 | attr = attrs + counter; | |
559 | ||
560 | attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; | |
561 | attr->freq = freq; | |
562 | ||
563 | try_again: | |
564 | fd[i][counter] = sys_perf_counter_open(attr, target_pid, cpu, group_fd, 0); | |
565 | ||
566 | if (fd[i][counter] < 0) { | |
567 | int err = errno; | |
568 | ||
716c69fe | 569 | if (err == EPERM) |
3da297a6 | 570 | die("No permission - are you root?\n"); |
716c69fe IM |
571 | /* |
572 | * If it's cycles then fall back to hrtimer | |
573 | * based cpu-clock-tick sw counter, which | |
574 | * is always available even if no PMU support: | |
575 | */ | |
576 | if (attr->type == PERF_TYPE_HARDWARE | |
f4dbfa8f | 577 | && attr->config == PERF_COUNT_HW_CPU_CYCLES) { |
716c69fe | 578 | |
3da297a6 IM |
579 | if (verbose) |
580 | warning(" ... trying to fall back to cpu-clock-ticks\n"); | |
581 | ||
716c69fe | 582 | attr->type = PERF_TYPE_SOFTWARE; |
f4dbfa8f | 583 | attr->config = PERF_COUNT_SW_CPU_CLOCK; |
716c69fe IM |
584 | goto try_again; |
585 | } | |
30c806a0 IM |
586 | printf("\n"); |
587 | error("perfcounter syscall returned with %d (%s)\n", | |
588 | fd[i][counter], strerror(err)); | |
589 | die("No CONFIG_PERF_COUNTERS=y kernel support configured?\n"); | |
716c69fe IM |
590 | exit(-1); |
591 | } | |
592 | assert(fd[i][counter] >= 0); | |
593 | fcntl(fd[i][counter], F_SETFL, O_NONBLOCK); | |
594 | ||
595 | /* | |
596 | * First counter acts as the group leader: | |
597 | */ | |
598 | if (group && group_fd == -1) | |
599 | group_fd = fd[i][counter]; | |
600 | ||
601 | event_array[nr_poll].fd = fd[i][counter]; | |
602 | event_array[nr_poll].events = POLLIN; | |
603 | nr_poll++; | |
604 | ||
605 | mmap_array[i][counter].counter = counter; | |
606 | mmap_array[i][counter].prev = 0; | |
607 | mmap_array[i][counter].mask = mmap_pages*page_size - 1; | |
608 | mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size, | |
609 | PROT_READ, MAP_SHARED, fd[i][counter], 0); | |
610 | if (mmap_array[i][counter].base == MAP_FAILED) | |
611 | die("failed to mmap with %d (%s)\n", errno, strerror(errno)); | |
612 | } | |
613 | ||
614 | static int __cmd_top(void) | |
615 | { | |
616 | pthread_t thread; | |
617 | int i, counter; | |
07800601 IM |
618 | int ret; |
619 | ||
07800601 IM |
620 | for (i = 0; i < nr_cpus; i++) { |
621 | group_fd = -1; | |
716c69fe IM |
622 | for (counter = 0; counter < nr_counters; counter++) |
623 | start_counter(i, counter); | |
07800601 IM |
624 | } |
625 | ||
2f01190a FW |
626 | /* Wait for a minimal set of events before starting the snapshot */ |
627 | poll(event_array, nr_poll, 100); | |
628 | ||
629 | mmap_read(); | |
630 | ||
07800601 IM |
631 | if (pthread_create(&thread, NULL, display_thread, NULL)) { |
632 | printf("Could not create display thread.\n"); | |
633 | exit(-1); | |
634 | } | |
635 | ||
636 | if (realtime_prio) { | |
637 | struct sched_param param; | |
638 | ||
639 | param.sched_priority = realtime_prio; | |
640 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { | |
641 | printf("Could not set realtime priority.\n"); | |
642 | exit(-1); | |
643 | } | |
644 | } | |
645 | ||
646 | while (1) { | |
2debbc83 | 647 | int hits = samples; |
07800601 | 648 | |
2f01190a | 649 | mmap_read(); |
07800601 | 650 | |
2debbc83 | 651 | if (hits == samples) |
07800601 IM |
652 | ret = poll(event_array, nr_poll, 100); |
653 | } | |
654 | ||
655 | return 0; | |
656 | } | |
b456bae0 IM |
657 | |
658 | static const char * const top_usage[] = { | |
659 | "perf top [<options>]", | |
660 | NULL | |
661 | }; | |
662 | ||
b456bae0 IM |
663 | static const struct option options[] = { |
664 | OPT_CALLBACK('e', "event", NULL, "event", | |
86847b62 TG |
665 | "event selector. use 'perf list' to list available events", |
666 | parse_events), | |
b456bae0 IM |
667 | OPT_INTEGER('c', "count", &default_interval, |
668 | "event period to sample"), | |
669 | OPT_INTEGER('p', "pid", &target_pid, | |
670 | "profile events on existing pid"), | |
671 | OPT_BOOLEAN('a', "all-cpus", &system_wide, | |
672 | "system-wide collection from all CPUs"), | |
673 | OPT_INTEGER('C', "CPU", &profile_cpu, | |
674 | "CPU to profile on"), | |
42976487 | 675 | OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"), |
b456bae0 IM |
676 | OPT_INTEGER('m', "mmap-pages", &mmap_pages, |
677 | "number of mmap data pages"), | |
678 | OPT_INTEGER('r', "realtime", &realtime_prio, | |
679 | "collect data with this RT SCHED_FIFO priority"), | |
db20c003 | 680 | OPT_INTEGER('d', "delay", &delay_secs, |
b456bae0 IM |
681 | "number of seconds to delay between refreshes"), |
682 | OPT_BOOLEAN('D', "dump-symtab", &dump_symtab, | |
683 | "dump the symbol table used for profiling"), | |
6e53cdf1 | 684 | OPT_INTEGER('f', "count-filter", &count_filter, |
b456bae0 IM |
685 | "only display functions with more events than this"), |
686 | OPT_BOOLEAN('g', "group", &group, | |
687 | "put the counters into a counter group"), | |
688 | OPT_STRING('s', "sym-filter", &sym_filter, "pattern", | |
689 | "only display symbols matchig this pattern"), | |
1f208ea6 | 690 | OPT_BOOLEAN('z', "zero", &zero, |
b456bae0 | 691 | "zero history across updates"), |
6e53cdf1 | 692 | OPT_INTEGER('F', "freq", &freq, |
b456bae0 | 693 | "profile at this frequency"), |
6e53cdf1 IM |
694 | OPT_INTEGER('E', "entries", &print_entries, |
695 | "display this many functions"), | |
3da297a6 IM |
696 | OPT_BOOLEAN('v', "verbose", &verbose, |
697 | "be more verbose (show counter open errors, etc)"), | |
b456bae0 IM |
698 | OPT_END() |
699 | }; | |
700 | ||
f37a291c | 701 | int cmd_top(int argc, const char **argv, const char *prefix __used) |
b456bae0 IM |
702 | { |
703 | int counter; | |
704 | ||
42976487 MG |
705 | symbol__init(); |
706 | ||
b456bae0 IM |
707 | page_size = sysconf(_SC_PAGE_SIZE); |
708 | ||
b456bae0 IM |
709 | argc = parse_options(argc, argv, options, top_usage, 0); |
710 | if (argc) | |
711 | usage_with_options(top_usage, options); | |
712 | ||
713 | if (freq) { | |
714 | default_interval = freq; | |
715 | freq = 1; | |
716 | } | |
717 | ||
718 | /* CPU and PID are mutually exclusive */ | |
719 | if (target_pid != -1 && profile_cpu != -1) { | |
720 | printf("WARNING: PID switch overriding CPU\n"); | |
721 | sleep(1); | |
722 | profile_cpu = -1; | |
723 | } | |
724 | ||
a21ca2ca | 725 | if (!nr_counters) |
b456bae0 | 726 | nr_counters = 1; |
b456bae0 | 727 | |
2f335a02 FW |
728 | if (delay_secs < 1) |
729 | delay_secs = 1; | |
730 | ||
a21ca2ca IM |
731 | parse_symbols(); |
732 | ||
733 | /* | |
734 | * Fill in the ones not specifically initialized via -c: | |
735 | */ | |
b456bae0 | 736 | for (counter = 0; counter < nr_counters; counter++) { |
a21ca2ca | 737 | if (attrs[counter].sample_period) |
b456bae0 IM |
738 | continue; |
739 | ||
a21ca2ca | 740 | attrs[counter].sample_period = default_interval; |
b456bae0 IM |
741 | } |
742 | ||
743 | nr_cpus = sysconf(_SC_NPROCESSORS_ONLN); | |
744 | assert(nr_cpus <= MAX_NR_CPUS); | |
745 | assert(nr_cpus >= 0); | |
746 | ||
747 | if (target_pid != -1 || profile_cpu != -1) | |
748 | nr_cpus = 1; | |
749 | ||
b456bae0 IM |
750 | return __cmd_top(); |
751 | } |