blob: 0506cd6e04cce267b9cc33bea3906d80b50e6f63 [file] [log] [blame]
Ingo Molnar07800602009-04-20 15:00:56 +02001/*
Ingo Molnarbf9e1872009-06-02 23:37:05 +02002 * 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)
Ingo Molnar07800602009-04-20 15:00:56 +020018 */
Ingo Molnarbf9e1872009-06-02 23:37:05 +020019#include "builtin.h"
Ingo Molnar07800602009-04-20 15:00:56 +020020
Peter Zijlstra1a482f32009-05-23 18:28:58 +020021#include "perf.h"
Ingo Molnarbf9e1872009-06-02 23:37:05 +020022
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -030023#include "util/symbol.h"
Ingo Molnar8fc03212009-06-04 15:19:47 +020024#include "util/color.h"
Ingo Molnar148be2c2009-04-27 08:02:14 +020025#include "util/util.h"
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -030026#include "util/rbtree.h"
Ingo Molnarb456bae2009-05-26 09:17:18 +020027#include "util/parse-options.h"
28#include "util/parse-events.h"
Ingo Molnar07800602009-04-20 15:00:56 +020029
Ingo Molnar07800602009-04-20 15:00:56 +020030#include <assert.h>
31#include <fcntl.h>
Ingo Molnar0e9b20b2009-05-26 09:17:18 +020032
Ingo Molnar07800602009-04-20 15:00:56 +020033#include <stdio.h>
Ingo Molnar0e9b20b2009-05-26 09:17:18 +020034
Ingo Molnar07800602009-04-20 15:00:56 +020035#include <errno.h>
Ingo Molnar07800602009-04-20 15:00:56 +020036#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
Ingo Molnara21ca2c2009-06-06 09:58:57 +020051static int fd[MAX_NR_CPUS][MAX_COUNTERS];
52
Ingo Molnar07800602009-04-20 15:00:56 +020053static int system_wide = 0;
54
Ingo Molnara21ca2c2009-06-06 09:58:57 +020055static int default_interval = 100000;
Ingo Molnar07800602009-04-20 15:00:56 +020056
Paul Mackerras9cffa8d2009-06-19 22:21:42 +100057static u64 count_filter = 5;
Ingo Molnar6e53cdf2009-06-04 08:53:05 +020058static int print_entries = 15;
Ingo Molnar07800602009-04-20 15:00:56 +020059
Ingo Molnar6e53cdf2009-06-04 08:53:05 +020060static int target_pid = -1;
Ingo Molnar07800602009-04-20 15:00:56 +020061static int profile_cpu = -1;
62static int nr_cpus = 0;
Ingo Molnar07800602009-04-20 15:00:56 +020063static unsigned int realtime_prio = 0;
64static int group = 0;
65static unsigned int page_size;
Ingo Molnarcf1f4572009-06-05 13:27:02 +020066static unsigned int mmap_pages = 16;
67static int freq = 0;
Ingo Molnar3da297a2009-06-07 17:39:02 +020068static int verbose = 0;
Ingo Molnar07800602009-04-20 15:00:56 +020069
Ingo Molnar07800602009-04-20 15:00:56 +020070static char *sym_filter;
71static unsigned long filter_start;
72static unsigned long filter_end;
73
74static int delay_secs = 2;
75static int zero;
76static int dump_symtab;
77
Ingo Molnar07800602009-04-20 15:00:56 +020078/*
79 * Symbols
80 */
81
Paul Mackerras9cffa8d2009-06-19 22:21:42 +100082static u64 min_ip;
83static u64 max_ip = -1ll;
Ingo Molnar07800602009-04-20 15:00:56 +020084
85struct sym_entry {
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -030086 struct rb_node rb_node;
87 struct list_head node;
Ingo Molnar07800602009-04-20 15:00:56 +020088 unsigned long count[MAX_COUNTERS];
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -030089 unsigned long snap_count;
90 double weight;
Ingo Molnar07800602009-04-20 15:00:56 +020091 int skip;
Ingo Molnar07800602009-04-20 15:00:56 +020092};
93
Ingo Molnar07800602009-04-20 15:00:56 +020094struct sym_entry *sym_filter_entry;
95
Ingo Molnara21ca2c2009-06-06 09:58:57 +020096struct dso *kernel_dso;
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -030097
98/*
99 * Symbols will be added here in record_ip and will get out
100 * after decayed.
101 */
102static LIST_HEAD(active_symbols);
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300103static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER;
Ingo Molnar07800602009-04-20 15:00:56 +0200104
Ingo Molnar07800602009-04-20 15:00:56 +0200105/*
106 * Ordering weight: count-1 * count-2 * ... / count-n
107 */
108static double sym_weight(const struct sym_entry *sym)
109{
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300110 double weight = sym->snap_count;
Ingo Molnar07800602009-04-20 15:00:56 +0200111 int counter;
112
Ingo Molnar07800602009-04-20 15:00:56 +0200113 for (counter = 1; counter < nr_counters-1; counter++)
114 weight *= sym->count[counter];
115
116 weight /= (sym->count[counter] + 1);
117
118 return weight;
119}
120
Ingo Molnar2debbc82009-06-05 14:29:10 +0200121static long samples;
122static long userspace_samples;
Ingo Molnar07800602009-04-20 15:00:56 +0200123static const char CONSOLE_CLEAR[] = "";
124
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300125static void __list_insert_active_sym(struct sym_entry *syme)
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300126{
127 list_add(&syme->node, &active_symbols);
128}
129
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300130static void list_remove_active_sym(struct sym_entry *syme)
131{
132 pthread_mutex_lock(&active_symbols_lock);
133 list_del_init(&syme->node);
134 pthread_mutex_unlock(&active_symbols_lock);
135}
136
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300137static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se)
138{
139 struct rb_node **p = &tree->rb_node;
140 struct rb_node *parent = NULL;
141 struct sym_entry *iter;
142
143 while (*p != NULL) {
144 parent = *p;
145 iter = rb_entry(parent, struct sym_entry, rb_node);
146
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300147 if (se->weight > iter->weight)
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300148 p = &(*p)->rb_left;
149 else
150 p = &(*p)->rb_right;
151 }
152
153 rb_link_node(&se->rb_node, parent, p);
154 rb_insert_color(&se->rb_node, tree);
155}
Ingo Molnar07800602009-04-20 15:00:56 +0200156
157static void print_sym_table(void)
158{
Ingo Molnar233f0b92009-06-03 21:48:40 +0200159 int printed = 0, j;
Ingo Molnar07800602009-04-20 15:00:56 +0200160 int counter;
Ingo Molnar2debbc82009-06-05 14:29:10 +0200161 float samples_per_sec = samples/delay_secs;
162 float ksamples_per_sec = (samples-userspace_samples)/delay_secs;
163 float sum_ksamples = 0.0;
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300164 struct sym_entry *syme, *n;
165 struct rb_root tmp = RB_ROOT;
166 struct rb_node *nd;
Ingo Molnar07800602009-04-20 15:00:56 +0200167
Ingo Molnar2debbc82009-06-05 14:29:10 +0200168 samples = userspace_samples = 0;
Ingo Molnar07800602009-04-20 15:00:56 +0200169
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300170 /* Sort the active symbols */
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300171 pthread_mutex_lock(&active_symbols_lock);
172 syme = list_entry(active_symbols.next, struct sym_entry, node);
173 pthread_mutex_unlock(&active_symbols_lock);
174
175 list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
176 syme->snap_count = syme->count[0];
177 if (syme->snap_count != 0) {
178 syme->weight = sym_weight(syme);
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300179 rb_insert_active_sym(&tmp, syme);
Ingo Molnar2debbc82009-06-05 14:29:10 +0200180 sum_ksamples += syme->snap_count;
Mike Galbraithd94b9432009-05-25 09:57:56 +0200181
182 for (j = 0; j < nr_counters; j++)
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300183 syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8;
184 } else
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300185 list_remove_active_sym(syme);
Mike Galbraithd94b9432009-05-25 09:57:56 +0200186 }
187
Frederic Weisbecker0f5486b2009-06-04 20:48:04 +0200188 puts(CONSOLE_CLEAR);
Ingo Molnar07800602009-04-20 15:00:56 +0200189
190 printf(
191"------------------------------------------------------------------------------\n");
Ingo Molnarf2521b62009-06-03 19:17:25 +0200192 printf( " PerfTop:%8.0f irqs/sec kernel:%4.1f%% [",
Ingo Molnar2debbc82009-06-05 14:29:10 +0200193 samples_per_sec,
194 100.0 - (100.0*((samples_per_sec-ksamples_per_sec)/samples_per_sec)));
Ingo Molnar07800602009-04-20 15:00:56 +0200195
Ingo Molnarcf1f4572009-06-05 13:27:02 +0200196 if (nr_counters == 1) {
Paul Mackerras9cffa8d2009-06-19 22:21:42 +1000197 printf("%Ld", (u64)attrs[0].sample_period);
Ingo Molnarcf1f4572009-06-05 13:27:02 +0200198 if (freq)
199 printf("Hz ");
200 else
201 printf(" ");
202 }
Ingo Molnar07800602009-04-20 15:00:56 +0200203
204 for (counter = 0; counter < nr_counters; counter++) {
205 if (counter)
206 printf("/");
207
208 printf("%s", event_name(counter));
209 }
210
211 printf( "], ");
212
Ingo Molnarb456bae2009-05-26 09:17:18 +0200213 if (target_pid != -1)
214 printf(" (target_pid: %d", target_pid);
Ingo Molnar07800602009-04-20 15:00:56 +0200215 else
216 printf(" (all");
217
218 if (profile_cpu != -1)
219 printf(", cpu: %d)\n", profile_cpu);
220 else {
Ingo Molnarb456bae2009-05-26 09:17:18 +0200221 if (target_pid != -1)
Ingo Molnar07800602009-04-20 15:00:56 +0200222 printf(")\n");
223 else
224 printf(", %d CPUs)\n", nr_cpus);
225 }
226
227 printf("------------------------------------------------------------------------------\n\n");
228
229 if (nr_counters == 1)
Ingo Molnar2debbc82009-06-05 14:29:10 +0200230 printf(" samples pcnt");
Ingo Molnar07800602009-04-20 15:00:56 +0200231 else
Ingo Molnar2debbc82009-06-05 14:29:10 +0200232 printf(" weight samples pcnt");
Ingo Molnar07800602009-04-20 15:00:56 +0200233
234 printf(" RIP kernel function\n"
Ingo Molnar2debbc82009-06-05 14:29:10 +0200235 " ______ _______ _____ ________________ _______________\n\n"
Ingo Molnar07800602009-04-20 15:00:56 +0200236 );
237
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300238 for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
239 struct sym_entry *syme = rb_entry(nd, struct sym_entry, rb_node);
240 struct symbol *sym = (struct symbol *)(syme + 1);
Ingo Molnar8fc03212009-06-04 15:19:47 +0200241 char *color = PERF_COLOR_NORMAL;
242 double pcnt;
Ingo Molnar07800602009-04-20 15:00:56 +0200243
Ingo Molnar6e53cdf2009-06-04 08:53:05 +0200244 if (++printed > print_entries || syme->snap_count < count_filter)
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300245 continue;
Ingo Molnar07800602009-04-20 15:00:56 +0200246
Ingo Molnar2debbc82009-06-05 14:29:10 +0200247 pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) /
248 sum_ksamples));
Mike Galbraithd94b9432009-05-25 09:57:56 +0200249
Ingo Molnar8fc03212009-06-04 15:19:47 +0200250 /*
Ingo Molnaraefcf372009-06-08 23:15:28 +0200251 * We color high-overhead entries in red, mid-overhead
252 * entries in green - and keep the low overhead places
253 * normal:
Ingo Molnar8fc03212009-06-04 15:19:47 +0200254 */
Ingo Molnaraefcf372009-06-08 23:15:28 +0200255 if (pcnt >= 5.0) {
Ingo Molnar8fc03212009-06-04 15:19:47 +0200256 color = PERF_COLOR_RED;
Ingo Molnaraefcf372009-06-08 23:15:28 +0200257 } else {
258 if (pcnt >= 0.5)
259 color = PERF_COLOR_GREEN;
260 }
Ingo Molnar8fc03212009-06-04 15:19:47 +0200261
Mike Galbraithd94b9432009-05-25 09:57:56 +0200262 if (nr_counters == 1)
Ingo Molnar2debbc82009-06-05 14:29:10 +0200263 printf("%20.2f - ", syme->weight);
Mike Galbraithd94b9432009-05-25 09:57:56 +0200264 else
Ingo Molnar2debbc82009-06-05 14:29:10 +0200265 printf("%9.1f %10ld - ", syme->weight, syme->snap_count);
Ingo Molnar8fc03212009-06-04 15:19:47 +0200266
267 color_fprintf(stdout, color, "%4.1f%%", pcnt);
268 printf(" - %016llx : %s\n", sym->start, sym->name);
Ingo Molnar07800602009-04-20 15:00:56 +0200269 }
Ingo Molnar07800602009-04-20 15:00:56 +0200270}
271
272static void *display_thread(void *arg)
273{
Frederic Weisbecker0f5486b2009-06-04 20:48:04 +0200274 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
275 int delay_msecs = delay_secs * 1000;
276
Ingo Molnarf2521b62009-06-03 19:17:25 +0200277 printf("PerfTop refresh period: %d seconds\n", delay_secs);
Ingo Molnar07800602009-04-20 15:00:56 +0200278
Frederic Weisbecker0f5486b2009-06-04 20:48:04 +0200279 do {
Ingo Molnar07800602009-04-20 15:00:56 +0200280 print_sym_table();
Frederic Weisbecker0f5486b2009-06-04 20:48:04 +0200281 } while (!poll(&stdin_poll, 1, delay_msecs) == 1);
282
283 printf("key pressed - exiting.\n");
284 exit(0);
Ingo Molnar07800602009-04-20 15:00:56 +0200285
286 return NULL;
287}
288
Anton Blanchard2ab52082009-07-01 09:00:46 +1000289/* Tag samples to be skipped. */
290char *skip_symbols[] = {
291 "default_idle",
292 "cpu_idle",
293 "enter_idle",
294 "exit_idle",
295 "mwait_idle",
Anton Blanchard3a3393e2009-07-01 09:00:47 +1000296 "ppc64_runlatch_off",
297 "pseries_dedicated_idle_sleep",
Anton Blanchard2ab52082009-07-01 09:00:46 +1000298 NULL
299};
300
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300301static int symbol_filter(struct dso *self, struct symbol *sym)
Ingo Molnar07800602009-04-20 15:00:56 +0200302{
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300303 static int filter_match;
304 struct sym_entry *syme;
305 const char *name = sym->name;
Anton Blanchard2ab52082009-07-01 09:00:46 +1000306 int i;
Ingo Molnar07800602009-04-20 15:00:56 +0200307
Anton Blanchard3a3393e2009-07-01 09:00:47 +1000308 /*
309 * ppc64 uses function descriptors and appends a '.' to the
310 * start of every instruction address. Remove it.
311 */
312 if (name[0] == '.')
313 name++;
314
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300315 if (!strcmp(name, "_text") ||
316 !strcmp(name, "_etext") ||
317 !strcmp(name, "_sinittext") ||
318 !strncmp("init_module", name, 11) ||
319 !strncmp("cleanup_module", name, 14) ||
320 strstr(name, "_text_start") ||
321 strstr(name, "_text_end"))
Ingo Molnar07800602009-04-20 15:00:56 +0200322 return 1;
323
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300324 syme = dso__sym_priv(self, sym);
Anton Blanchard2ab52082009-07-01 09:00:46 +1000325 for (i = 0; skip_symbols[i]; i++) {
326 if (!strcmp(skip_symbols[i], name)) {
327 syme->skip = 1;
328 break;
329 }
330 }
Ingo Molnar07800602009-04-20 15:00:56 +0200331
332 if (filter_match == 1) {
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300333 filter_end = sym->start;
Ingo Molnar07800602009-04-20 15:00:56 +0200334 filter_match = -1;
335 if (filter_end - filter_start > 10000) {
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300336 fprintf(stderr,
337 "hm, too large filter symbol <%s> - skipping.\n",
Ingo Molnar07800602009-04-20 15:00:56 +0200338 sym_filter);
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300339 fprintf(stderr, "symbol filter start: %016lx\n",
340 filter_start);
341 fprintf(stderr, " end: %016lx\n",
342 filter_end);
Ingo Molnar07800602009-04-20 15:00:56 +0200343 filter_end = filter_start = 0;
344 sym_filter = NULL;
345 sleep(1);
346 }
347 }
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300348
349 if (filter_match == 0 && sym_filter && !strcmp(name, sym_filter)) {
Ingo Molnar07800602009-04-20 15:00:56 +0200350 filter_match = 1;
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300351 filter_start = sym->start;
Ingo Molnar07800602009-04-20 15:00:56 +0200352 }
353
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300354
Ingo Molnar07800602009-04-20 15:00:56 +0200355 return 0;
356}
357
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300358static int parse_symbols(void)
Ingo Molnar07800602009-04-20 15:00:56 +0200359{
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300360 struct rb_node *node;
361 struct symbol *sym;
Ingo Molnar07800602009-04-20 15:00:56 +0200362
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300363 kernel_dso = dso__new("[kernel]", sizeof(struct sym_entry));
364 if (kernel_dso == NULL)
365 return -1;
Ingo Molnar07800602009-04-20 15:00:56 +0200366
Ingo Molnarbd741372009-06-04 14:13:04 +0200367 if (dso__load_kernel(kernel_dso, NULL, symbol_filter, 1) != 0)
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300368 goto out_delete_dso;
Ingo Molnar07800602009-04-20 15:00:56 +0200369
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300370 node = rb_first(&kernel_dso->syms);
371 sym = rb_entry(node, struct symbol, rb_node);
372 min_ip = sym->start;
Ingo Molnar07800602009-04-20 15:00:56 +0200373
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300374 node = rb_last(&kernel_dso->syms);
375 sym = rb_entry(node, struct symbol, rb_node);
Mike Galbraithda417a72009-05-29 08:23:16 +0200376 max_ip = sym->end;
Ingo Molnar07800602009-04-20 15:00:56 +0200377
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300378 if (dump_symtab)
Mike Galbraitha3ec8d72009-05-29 06:46:46 +0200379 dso__fprintf(kernel_dso, stderr);
Ingo Molnar07800602009-04-20 15:00:56 +0200380
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300381 return 0;
Ingo Molnar07800602009-04-20 15:00:56 +0200382
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300383out_delete_dso:
384 dso__delete(kernel_dso);
385 kernel_dso = NULL;
386 return -1;
Ingo Molnar07800602009-04-20 15:00:56 +0200387}
388
Ingo Molnar07800602009-04-20 15:00:56 +0200389#define TRACE_COUNT 3
390
Ingo Molnar07800602009-04-20 15:00:56 +0200391/*
392 * Binary search in the histogram table and record the hit:
393 */
Paul Mackerras9cffa8d2009-06-19 22:21:42 +1000394static void record_ip(u64 ip, int counter)
Ingo Molnar07800602009-04-20 15:00:56 +0200395{
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300396 struct symbol *sym = dso__find_symbol(kernel_dso, ip);
Ingo Molnar07800602009-04-20 15:00:56 +0200397
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300398 if (sym != NULL) {
399 struct sym_entry *syme = dso__sym_priv(kernel_dso, sym);
Ingo Molnar07800602009-04-20 15:00:56 +0200400
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300401 if (!syme->skip) {
402 syme->count[counter]++;
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300403 pthread_mutex_lock(&active_symbols_lock);
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300404 if (list_empty(&syme->node) || !syme->node.next)
Arnaldo Carvalho de Meloc44613a2009-05-29 17:03:07 -0300405 __list_insert_active_sym(syme);
406 pthread_mutex_unlock(&active_symbols_lock);
Arnaldo Carvalho de Melode046872009-05-28 14:55:41 -0300407 return;
Ingo Molnar07800602009-04-20 15:00:56 +0200408 }
Ingo Molnar07800602009-04-20 15:00:56 +0200409 }
410
Ingo Molnar2debbc82009-06-05 14:29:10 +0200411 samples--;
Ingo Molnar07800602009-04-20 15:00:56 +0200412}
413
Peter Zijlstrae6e18ec2009-06-25 11:27:12 +0200414static void process_event(u64 ip, int counter, int user)
Ingo Molnar07800602009-04-20 15:00:56 +0200415{
Ingo Molnar2debbc82009-06-05 14:29:10 +0200416 samples++;
Ingo Molnar07800602009-04-20 15:00:56 +0200417
Peter Zijlstrae6e18ec2009-06-25 11:27:12 +0200418 if (user) {
Ingo Molnar2debbc82009-06-05 14:29:10 +0200419 userspace_samples++;
Ingo Molnar07800602009-04-20 15:00:56 +0200420 return;
421 }
422
423 record_ip(ip, counter);
424}
425
Ingo Molnar07800602009-04-20 15:00:56 +0200426struct mmap_data {
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200427 int counter;
428 void *base;
429 unsigned int mask;
430 unsigned int prev;
Ingo Molnar07800602009-04-20 15:00:56 +0200431};
432
433static unsigned int mmap_read_head(struct mmap_data *md)
434{
435 struct perf_counter_mmap_page *pc = md->base;
436 int head;
437
438 head = pc->data_head;
439 rmb();
440
441 return head;
442}
443
444struct timeval last_read, this_read;
445
Frederic Weisbecker2f011902009-06-06 23:10:43 +0200446static void mmap_read_counter(struct mmap_data *md)
Ingo Molnar07800602009-04-20 15:00:56 +0200447{
448 unsigned int head = mmap_read_head(md);
449 unsigned int old = md->prev;
450 unsigned char *data = md->base + page_size;
451 int diff;
452
453 gettimeofday(&this_read, NULL);
454
455 /*
456 * If we're further behind than half the buffer, there's a chance
Ingo Molnar2debbc82009-06-05 14:29:10 +0200457 * the writer will bite our tail and mess up the samples under us.
Ingo Molnar07800602009-04-20 15:00:56 +0200458 *
459 * If we somehow ended up ahead of the head, we got messed up.
460 *
461 * In either case, truncate and restart at head.
462 */
463 diff = head - old;
464 if (diff > md->mask / 2 || diff < 0) {
465 struct timeval iv;
466 unsigned long msecs;
467
468 timersub(&this_read, &last_read, &iv);
469 msecs = iv.tv_sec*1000 + iv.tv_usec/1000;
470
471 fprintf(stderr, "WARNING: failed to keep up with mmap data."
472 " Last read %lu msecs ago.\n", msecs);
473
474 /*
475 * head points to a known good entry, start there.
476 */
477 old = head;
478 }
479
480 last_read = this_read;
481
482 for (; old != head;) {
483 struct ip_event {
484 struct perf_event_header header;
Paul Mackerras9cffa8d2009-06-19 22:21:42 +1000485 u64 ip;
486 u32 pid, target_pid;
Ingo Molnar07800602009-04-20 15:00:56 +0200487 };
488 struct mmap_event {
489 struct perf_event_header header;
Paul Mackerras9cffa8d2009-06-19 22:21:42 +1000490 u32 pid, target_pid;
491 u64 start;
492 u64 len;
493 u64 pgoff;
Ingo Molnar07800602009-04-20 15:00:56 +0200494 char filename[PATH_MAX];
495 };
496
497 typedef union event_union {
498 struct perf_event_header header;
499 struct ip_event ip;
500 struct mmap_event mmap;
501 } event_t;
502
503 event_t *event = (event_t *)&data[old & md->mask];
504
505 event_t event_copy;
506
Ingo Molnar6f06ccb2009-04-20 15:22:22 +0200507 size_t size = event->header.size;
Ingo Molnar07800602009-04-20 15:00:56 +0200508
509 /*
510 * Event straddles the mmap boundary -- header should always
511 * be inside due to u64 alignment of output.
512 */
513 if ((old & md->mask) + size != ((old + size) & md->mask)) {
514 unsigned int offset = old;
515 unsigned int len = min(sizeof(*event), size), cpy;
516 void *dst = &event_copy;
517
518 do {
519 cpy = min(md->mask + 1 - (offset & md->mask), len);
520 memcpy(dst, &data[offset & md->mask], cpy);
521 offset += cpy;
522 dst += cpy;
523 len -= cpy;
524 } while (len);
525
526 event = &event_copy;
527 }
528
529 old += size;
530
Peter Zijlstrae6e18ec2009-06-25 11:27:12 +0200531 if (event->header.type == PERF_EVENT_SAMPLE) {
532 int user =
533 (event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK) == PERF_EVENT_MISC_USER;
534 process_event(event->ip.ip, md->counter, user);
Ingo Molnar07800602009-04-20 15:00:56 +0200535 }
536 }
537
538 md->prev = old;
539}
540
Mike Galbraithc2990a22009-05-24 08:35:49 +0200541static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
542static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];
543
Frederic Weisbecker2f011902009-06-06 23:10:43 +0200544static void mmap_read(void)
545{
546 int i, counter;
547
548 for (i = 0; i < nr_cpus; i++) {
549 for (counter = 0; counter < nr_counters; counter++)
550 mmap_read_counter(&mmap_array[i][counter]);
551 }
552}
553
Ingo Molnar716c69f2009-06-07 17:31:52 +0200554int nr_poll;
555int group_fd;
556
557static void start_counter(int i, int counter)
Ingo Molnar07800602009-04-20 15:00:56 +0200558{
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200559 struct perf_counter_attr *attr;
Ingo Molnar07800602009-04-20 15:00:56 +0200560 unsigned int cpu;
Ingo Molnar716c69f2009-06-07 17:31:52 +0200561
562 cpu = profile_cpu;
563 if (target_pid == -1 && profile_cpu == -1)
564 cpu = i;
565
566 attr = attrs + counter;
567
568 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
569 attr->freq = freq;
570
571try_again:
572 fd[i][counter] = sys_perf_counter_open(attr, target_pid, cpu, group_fd, 0);
573
574 if (fd[i][counter] < 0) {
575 int err = errno;
576
Ingo Molnar716c69f2009-06-07 17:31:52 +0200577 if (err == EPERM)
Ingo Molnar3da297a2009-06-07 17:39:02 +0200578 die("No permission - are you root?\n");
Ingo Molnar716c69f2009-06-07 17:31:52 +0200579 /*
580 * If it's cycles then fall back to hrtimer
581 * based cpu-clock-tick sw counter, which
582 * is always available even if no PMU support:
583 */
584 if (attr->type == PERF_TYPE_HARDWARE
Peter Zijlstraf4dbfa82009-06-11 14:06:28 +0200585 && attr->config == PERF_COUNT_HW_CPU_CYCLES) {
Ingo Molnar716c69f2009-06-07 17:31:52 +0200586
Ingo Molnar3da297a2009-06-07 17:39:02 +0200587 if (verbose)
588 warning(" ... trying to fall back to cpu-clock-ticks\n");
589
Ingo Molnar716c69f2009-06-07 17:31:52 +0200590 attr->type = PERF_TYPE_SOFTWARE;
Peter Zijlstraf4dbfa82009-06-11 14:06:28 +0200591 attr->config = PERF_COUNT_SW_CPU_CLOCK;
Ingo Molnar716c69f2009-06-07 17:31:52 +0200592 goto try_again;
593 }
Ingo Molnar30c806a2009-06-07 17:46:24 +0200594 printf("\n");
595 error("perfcounter syscall returned with %d (%s)\n",
596 fd[i][counter], strerror(err));
597 die("No CONFIG_PERF_COUNTERS=y kernel support configured?\n");
Ingo Molnar716c69f2009-06-07 17:31:52 +0200598 exit(-1);
599 }
600 assert(fd[i][counter] >= 0);
601 fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);
602
603 /*
604 * First counter acts as the group leader:
605 */
606 if (group && group_fd == -1)
607 group_fd = fd[i][counter];
608
609 event_array[nr_poll].fd = fd[i][counter];
610 event_array[nr_poll].events = POLLIN;
611 nr_poll++;
612
613 mmap_array[i][counter].counter = counter;
614 mmap_array[i][counter].prev = 0;
615 mmap_array[i][counter].mask = mmap_pages*page_size - 1;
616 mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
617 PROT_READ, MAP_SHARED, fd[i][counter], 0);
618 if (mmap_array[i][counter].base == MAP_FAILED)
619 die("failed to mmap with %d (%s)\n", errno, strerror(errno));
620}
621
622static int __cmd_top(void)
623{
624 pthread_t thread;
625 int i, counter;
Ingo Molnar07800602009-04-20 15:00:56 +0200626 int ret;
627
Ingo Molnar07800602009-04-20 15:00:56 +0200628 for (i = 0; i < nr_cpus; i++) {
629 group_fd = -1;
Ingo Molnar716c69f2009-06-07 17:31:52 +0200630 for (counter = 0; counter < nr_counters; counter++)
631 start_counter(i, counter);
Ingo Molnar07800602009-04-20 15:00:56 +0200632 }
633
Frederic Weisbecker2f011902009-06-06 23:10:43 +0200634 /* Wait for a minimal set of events before starting the snapshot */
635 poll(event_array, nr_poll, 100);
636
637 mmap_read();
638
Ingo Molnar07800602009-04-20 15:00:56 +0200639 if (pthread_create(&thread, NULL, display_thread, NULL)) {
640 printf("Could not create display thread.\n");
641 exit(-1);
642 }
643
644 if (realtime_prio) {
645 struct sched_param param;
646
647 param.sched_priority = realtime_prio;
648 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
649 printf("Could not set realtime priority.\n");
650 exit(-1);
651 }
652 }
653
654 while (1) {
Ingo Molnar2debbc82009-06-05 14:29:10 +0200655 int hits = samples;
Ingo Molnar07800602009-04-20 15:00:56 +0200656
Frederic Weisbecker2f011902009-06-06 23:10:43 +0200657 mmap_read();
Ingo Molnar07800602009-04-20 15:00:56 +0200658
Ingo Molnar2debbc82009-06-05 14:29:10 +0200659 if (hits == samples)
Ingo Molnar07800602009-04-20 15:00:56 +0200660 ret = poll(event_array, nr_poll, 100);
661 }
662
663 return 0;
664}
Ingo Molnarb456bae2009-05-26 09:17:18 +0200665
666static const char * const top_usage[] = {
667 "perf top [<options>]",
668 NULL
669};
670
Ingo Molnarb456bae2009-05-26 09:17:18 +0200671static const struct option options[] = {
672 OPT_CALLBACK('e', "event", NULL, "event",
Thomas Gleixner86847b62009-06-06 12:24:17 +0200673 "event selector. use 'perf list' to list available events",
674 parse_events),
Ingo Molnarb456bae2009-05-26 09:17:18 +0200675 OPT_INTEGER('c', "count", &default_interval,
676 "event period to sample"),
677 OPT_INTEGER('p', "pid", &target_pid,
678 "profile events on existing pid"),
679 OPT_BOOLEAN('a', "all-cpus", &system_wide,
680 "system-wide collection from all CPUs"),
681 OPT_INTEGER('C', "CPU", &profile_cpu,
682 "CPU to profile on"),
683 OPT_INTEGER('m', "mmap-pages", &mmap_pages,
684 "number of mmap data pages"),
685 OPT_INTEGER('r', "realtime", &realtime_prio,
686 "collect data with this RT SCHED_FIFO priority"),
Mike Galbraithdb20c002009-05-26 15:25:34 +0200687 OPT_INTEGER('d', "delay", &delay_secs,
Ingo Molnarb456bae2009-05-26 09:17:18 +0200688 "number of seconds to delay between refreshes"),
689 OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
690 "dump the symbol table used for profiling"),
Ingo Molnar6e53cdf2009-06-04 08:53:05 +0200691 OPT_INTEGER('f', "count-filter", &count_filter,
Ingo Molnarb456bae2009-05-26 09:17:18 +0200692 "only display functions with more events than this"),
693 OPT_BOOLEAN('g', "group", &group,
694 "put the counters into a counter group"),
695 OPT_STRING('s', "sym-filter", &sym_filter, "pattern",
696 "only display symbols matchig this pattern"),
Anton Blanchard1f208ea2009-07-01 09:00:44 +1000697 OPT_BOOLEAN('z', "zero", &zero,
Ingo Molnarb456bae2009-05-26 09:17:18 +0200698 "zero history across updates"),
Ingo Molnar6e53cdf2009-06-04 08:53:05 +0200699 OPT_INTEGER('F', "freq", &freq,
Ingo Molnarb456bae2009-05-26 09:17:18 +0200700 "profile at this frequency"),
Ingo Molnar6e53cdf2009-06-04 08:53:05 +0200701 OPT_INTEGER('E', "entries", &print_entries,
702 "display this many functions"),
Ingo Molnar3da297a2009-06-07 17:39:02 +0200703 OPT_BOOLEAN('v', "verbose", &verbose,
704 "be more verbose (show counter open errors, etc)"),
Ingo Molnarb456bae2009-05-26 09:17:18 +0200705 OPT_END()
706};
707
708int cmd_top(int argc, const char **argv, const char *prefix)
709{
710 int counter;
711
712 page_size = sysconf(_SC_PAGE_SIZE);
713
Ingo Molnarb456bae2009-05-26 09:17:18 +0200714 argc = parse_options(argc, argv, options, top_usage, 0);
715 if (argc)
716 usage_with_options(top_usage, options);
717
718 if (freq) {
719 default_interval = freq;
720 freq = 1;
721 }
722
723 /* CPU and PID are mutually exclusive */
724 if (target_pid != -1 && profile_cpu != -1) {
725 printf("WARNING: PID switch overriding CPU\n");
726 sleep(1);
727 profile_cpu = -1;
728 }
729
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200730 if (!nr_counters)
Ingo Molnarb456bae2009-05-26 09:17:18 +0200731 nr_counters = 1;
Ingo Molnarb456bae2009-05-26 09:17:18 +0200732
Frederic Weisbecker2f335a02009-06-05 19:31:01 +0200733 if (delay_secs < 1)
734 delay_secs = 1;
735
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200736 parse_symbols();
737
738 /*
739 * Fill in the ones not specifically initialized via -c:
740 */
Ingo Molnarb456bae2009-05-26 09:17:18 +0200741 for (counter = 0; counter < nr_counters; counter++) {
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200742 if (attrs[counter].sample_period)
Ingo Molnarb456bae2009-05-26 09:17:18 +0200743 continue;
744
Ingo Molnara21ca2c2009-06-06 09:58:57 +0200745 attrs[counter].sample_period = default_interval;
Ingo Molnarb456bae2009-05-26 09:17:18 +0200746 }
747
748 nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
749 assert(nr_cpus <= MAX_NR_CPUS);
750 assert(nr_cpus >= 0);
751
752 if (target_pid != -1 || profile_cpu != -1)
753 nr_cpus = 1;
754
Ingo Molnarb456bae2009-05-26 09:17:18 +0200755 return __cmd_top();
756}