aboutsummaryrefslogtreecommitdiffstats
path: root/tools/perf/builtin-top.c
diff options
context:
space:
mode:
Diffstat (limited to 'tools/perf/builtin-top.c')
-rw-r--r--tools/perf/builtin-top.c1038
1 files changed, 318 insertions, 720 deletions
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 05344c6210a..676b4fb0070 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -20,11 +20,16 @@
20 20
21#include "perf.h" 21#include "perf.h"
22 22
23#include "util/annotate.h"
24#include "util/cache.h"
23#include "util/color.h" 25#include "util/color.h"
26#include "util/evlist.h"
24#include "util/evsel.h" 27#include "util/evsel.h"
25#include "util/session.h" 28#include "util/session.h"
26#include "util/symbol.h" 29#include "util/symbol.h"
27#include "util/thread.h" 30#include "util/thread.h"
31#include "util/thread_map.h"
32#include "util/top.h"
28#include "util/util.h" 33#include "util/util.h"
29#include <linux/rbtree.h> 34#include <linux/rbtree.h>
30#include "util/parse-options.h" 35#include "util/parse-options.h"
@@ -40,11 +45,11 @@
40#include <stdio.h> 45#include <stdio.h>
41#include <termios.h> 46#include <termios.h>
42#include <unistd.h> 47#include <unistd.h>
48#include <inttypes.h>
43 49
44#include <errno.h> 50#include <errno.h>
45#include <time.h> 51#include <time.h>
46#include <sched.h> 52#include <sched.h>
47#include <pthread.h>
48 53
49#include <sys/syscall.h> 54#include <sys/syscall.h>
50#include <sys/ioctl.h> 55#include <sys/ioctl.h>
@@ -59,85 +64,42 @@
59 64
60#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) 65#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
61 66
67static struct perf_top top = {
68 .count_filter = 5,
69 .delay_secs = 2,
70 .display_weighted = -1,
71 .target_pid = -1,
72 .target_tid = -1,
73 .active_symbols = LIST_HEAD_INIT(top.active_symbols),
74 .active_symbols_lock = PTHREAD_MUTEX_INITIALIZER,
75 .active_symbols_cond = PTHREAD_COND_INITIALIZER,
76 .freq = 1000, /* 1 KHz */
77};
78
62static bool system_wide = false; 79static bool system_wide = false;
63 80
64static int default_interval = 0; 81static bool use_tui, use_stdio;
65 82
66static int count_filter = 5; 83static int default_interval = 0;
67static int print_entries;
68 84
69static int target_pid = -1;
70static int target_tid = -1;
71static struct thread_map *threads;
72static bool inherit = false; 85static bool inherit = false;
73static struct cpu_map *cpus;
74static int realtime_prio = 0; 86static int realtime_prio = 0;
75static bool group = false; 87static bool group = false;
76static unsigned int page_size; 88static unsigned int page_size;
77static unsigned int mmap_pages = 16; 89static unsigned int mmap_pages = 128;
78static int freq = 1000; /* 1 KHz */
79 90
80static int delay_secs = 2;
81static bool zero = false;
82static bool dump_symtab = false; 91static bool dump_symtab = false;
83 92
84static bool hide_kernel_symbols = false;
85static bool hide_user_symbols = false;
86static struct winsize winsize; 93static struct winsize winsize;
87 94
88/*
89 * Source
90 */
91
92struct source_line {
93 u64 eip;
94 unsigned long count[MAX_COUNTERS];
95 char *line;
96 struct source_line *next;
97};
98
99static const char *sym_filter = NULL; 95static const char *sym_filter = NULL;
100struct sym_entry *sym_filter_entry = NULL;
101struct sym_entry *sym_filter_entry_sched = NULL; 96struct sym_entry *sym_filter_entry_sched = NULL;
102static int sym_pcnt_filter = 5; 97static int sym_pcnt_filter = 5;
103static int sym_counter = 0;
104static struct perf_evsel *sym_evsel = NULL;
105static int display_weighted = -1;
106static const char *cpu_list;
107
108/*
109 * Symbols
110 */
111
112struct sym_entry_source {
113 struct source_line *source;
114 struct source_line *lines;
115 struct source_line **lines_tail;
116 pthread_mutex_t lock;
117};
118
119struct sym_entry {
120 struct rb_node rb_node;
121 struct list_head node;
122 unsigned long snap_count;
123 double weight;
124 int skip;
125 u16 name_len;
126 u8 origin;
127 struct map *map;
128 struct sym_entry_source *src;
129 unsigned long count[0];
130};
131 98
132/* 99/*
133 * Source functions 100 * Source functions
134 */ 101 */
135 102
136static inline struct symbol *sym_entry__symbol(struct sym_entry *self)
137{
138 return ((void *)self) + symbol_conf.priv_size;
139}
140
141void get_term_dimensions(struct winsize *ws) 103void get_term_dimensions(struct winsize *ws)
142{ 104{
143 char *s = getenv("LINES"); 105 char *s = getenv("LINES");
@@ -162,10 +124,10 @@ void get_term_dimensions(struct winsize *ws)
162 124
163static void update_print_entries(struct winsize *ws) 125static void update_print_entries(struct winsize *ws)
164{ 126{
165 print_entries = ws->ws_row; 127 top.print_entries = ws->ws_row;
166 128
167 if (print_entries > 9) 129 if (top.print_entries > 9)
168 print_entries -= 9; 130 top.print_entries -= 9;
169} 131}
170 132
171static void sig_winch_handler(int sig __used) 133static void sig_winch_handler(int sig __used)
@@ -177,12 +139,9 @@ static void sig_winch_handler(int sig __used)
177static int parse_source(struct sym_entry *syme) 139static int parse_source(struct sym_entry *syme)
178{ 140{
179 struct symbol *sym; 141 struct symbol *sym;
180 struct sym_entry_source *source; 142 struct annotation *notes;
181 struct map *map; 143 struct map *map;
182 FILE *file; 144 int err = -1;
183 char command[PATH_MAX*2];
184 const char *path;
185 u64 len;
186 145
187 if (!syme) 146 if (!syme)
188 return -1; 147 return -1;
@@ -193,411 +152,137 @@ static int parse_source(struct sym_entry *syme)
193 /* 152 /*
194 * We can't annotate with just /proc/kallsyms 153 * We can't annotate with just /proc/kallsyms
195 */ 154 */
196 if (map->dso->origin == DSO__ORIG_KERNEL) 155 if (map->dso->symtab_type == SYMTAB__KALLSYMS) {
156 pr_err("Can't annotate %s: No vmlinux file was found in the "
157 "path\n", sym->name);
158 sleep(1);
197 return -1; 159 return -1;
198
199 if (syme->src == NULL) {
200 syme->src = zalloc(sizeof(*source));
201 if (syme->src == NULL)
202 return -1;
203 pthread_mutex_init(&syme->src->lock, NULL);
204 } 160 }
205 161
206 source = syme->src; 162 notes = symbol__annotation(sym);
207 163 if (notes->src != NULL) {
208 if (source->lines) { 164 pthread_mutex_lock(&notes->lock);
209 pthread_mutex_lock(&source->lock);
210 goto out_assign; 165 goto out_assign;
211 } 166 }
212 path = map->dso->long_name;
213
214 len = sym->end - sym->start;
215
216 sprintf(command,
217 "objdump --start-address=%#0*Lx --stop-address=%#0*Lx -dS %s",
218 BITS_PER_LONG / 4, map__rip_2objdump(map, sym->start),
219 BITS_PER_LONG / 4, map__rip_2objdump(map, sym->end), path);
220
221 file = popen(command, "r");
222 if (!file)
223 return -1;
224
225 pthread_mutex_lock(&source->lock);
226 source->lines_tail = &source->lines;
227 while (!feof(file)) {
228 struct source_line *src;
229 size_t dummy = 0;
230 char *c, *sep;
231 167
232 src = malloc(sizeof(struct source_line)); 168 pthread_mutex_lock(&notes->lock);
233 assert(src != NULL);
234 memset(src, 0, sizeof(struct source_line));
235 169
236 if (getline(&src->line, &dummy, file) < 0) 170 if (symbol__alloc_hist(sym, top.evlist->nr_entries) < 0) {
237 break; 171 pthread_mutex_unlock(&notes->lock);
238 if (!src->line) 172 pr_err("Not enough memory for annotating '%s' symbol!\n",
239 break; 173 sym->name);
240 174 sleep(1);
241 c = strchr(src->line, '\n'); 175 return err;
242 if (c)
243 *c = 0;
244
245 src->next = NULL;
246 *source->lines_tail = src;
247 source->lines_tail = &src->next;
248
249 src->eip = strtoull(src->line, &sep, 16);
250 if (*sep == ':')
251 src->eip = map__objdump_2ip(map, src->eip);
252 else /* this line has no ip info (e.g. source line) */
253 src->eip = 0;
254 } 176 }
255 pclose(file); 177
178 err = symbol__annotate(sym, syme->map, 0);
179 if (err == 0) {
256out_assign: 180out_assign:
257 sym_filter_entry = syme; 181 top.sym_filter_entry = syme;
258 pthread_mutex_unlock(&source->lock); 182 }
259 return 0; 183
184 pthread_mutex_unlock(&notes->lock);
185 return err;
260} 186}
261 187
262static void __zero_source_counters(struct sym_entry *syme) 188static void __zero_source_counters(struct sym_entry *syme)
263{ 189{
264 int i; 190 struct symbol *sym = sym_entry__symbol(syme);
265 struct source_line *line; 191 symbol__annotate_zero_histograms(sym);
266
267 line = syme->src->lines;
268 while (line) {
269 for (i = 0; i < nr_counters; i++)
270 line->count[i] = 0;
271 line = line->next;
272 }
273} 192}
274 193
275static void record_precise_ip(struct sym_entry *syme, int counter, u64 ip) 194static void record_precise_ip(struct sym_entry *syme, int counter, u64 ip)
276{ 195{
277 struct source_line *line; 196 struct annotation *notes;
278 197 struct symbol *sym;
279 if (syme != sym_filter_entry)
280 return;
281 198
282 if (pthread_mutex_trylock(&syme->src->lock)) 199 if (syme != top.sym_filter_entry)
283 return; 200 return;
284 201
285 if (syme->src == NULL || syme->src->source == NULL) 202 sym = sym_entry__symbol(syme);
286 goto out_unlock; 203 notes = symbol__annotation(sym);
287
288 for (line = syme->src->lines; line; line = line->next) {
289 /* skip lines without IP info */
290 if (line->eip == 0)
291 continue;
292 if (line->eip == ip) {
293 line->count[counter]++;
294 break;
295 }
296 if (line->eip > ip)
297 break;
298 }
299out_unlock:
300 pthread_mutex_unlock(&syme->src->lock);
301}
302
303#define PATTERN_LEN (BITS_PER_LONG / 4 + 2)
304
305static void lookup_sym_source(struct sym_entry *syme)
306{
307 struct symbol *symbol = sym_entry__symbol(syme);
308 struct source_line *line;
309 char pattern[PATTERN_LEN + 1];
310
311 sprintf(pattern, "%0*Lx <", BITS_PER_LONG / 4,
312 map__rip_2objdump(syme->map, symbol->start));
313
314 pthread_mutex_lock(&syme->src->lock);
315 for (line = syme->src->lines; line; line = line->next) {
316 if (memcmp(line->line, pattern, PATTERN_LEN) == 0) {
317 syme->src->source = line;
318 break;
319 }
320 }
321 pthread_mutex_unlock(&syme->src->lock);
322}
323 204
324static void show_lines(struct source_line *queue, int count, int total) 205 if (pthread_mutex_trylock(&notes->lock))
325{ 206 return;
326 int i;
327 struct source_line *line;
328 207
329 line = queue; 208 ip = syme->map->map_ip(syme->map, ip);
330 for (i = 0; i < count; i++) { 209 symbol__inc_addr_samples(sym, syme->map, counter, ip);
331 float pcnt = 100.0*(float)line->count[sym_counter]/(float)total;
332 210
333 printf("%8li %4.1f%%\t%s\n", line->count[sym_counter], pcnt, line->line); 211 pthread_mutex_unlock(&notes->lock);
334 line = line->next;
335 }
336} 212}
337 213
338#define TRACE_COUNT 3
339
340static void show_details(struct sym_entry *syme) 214static void show_details(struct sym_entry *syme)
341{ 215{
216 struct annotation *notes;
342 struct symbol *symbol; 217 struct symbol *symbol;
343 struct source_line *line; 218 int more;
344 struct source_line *line_queue = NULL;
345 int displayed = 0;
346 int line_queue_count = 0, total = 0, more = 0;
347 219
348 if (!syme) 220 if (!syme)
349 return; 221 return;
350 222
351 if (!syme->src->source)
352 lookup_sym_source(syme);
353
354 if (!syme->src->source)
355 return;
356
357 symbol = sym_entry__symbol(syme); 223 symbol = sym_entry__symbol(syme);
358 printf("Showing %s for %s\n", event_name(sym_evsel), symbol->name); 224 notes = symbol__annotation(symbol);
359 printf(" Events Pcnt (>=%d%%)\n", sym_pcnt_filter);
360
361 pthread_mutex_lock(&syme->src->lock);
362 line = syme->src->source;
363 while (line) {
364 total += line->count[sym_counter];
365 line = line->next;
366 }
367
368 line = syme->src->source;
369 while (line) {
370 float pcnt = 0.0;
371
372 if (!line_queue_count)
373 line_queue = line;
374 line_queue_count++;
375
376 if (line->count[sym_counter])
377 pcnt = 100.0 * line->count[sym_counter] / (float)total;
378 if (pcnt >= (float)sym_pcnt_filter) {
379 if (displayed <= print_entries)
380 show_lines(line_queue, line_queue_count, total);
381 else more++;
382 displayed += line_queue_count;
383 line_queue_count = 0;
384 line_queue = NULL;
385 } else if (line_queue_count > TRACE_COUNT) {
386 line_queue = line_queue->next;
387 line_queue_count--;
388 }
389
390 line->count[sym_counter] = zero ? 0 : line->count[sym_counter] * 7 / 8;
391 line = line->next;
392 }
393 pthread_mutex_unlock(&syme->src->lock);
394 if (more)
395 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
396}
397 225
398/* 226 pthread_mutex_lock(&notes->lock);
399 * Symbols will be added here in event__process_sample and will get out
400 * after decayed.
401 */
402static LIST_HEAD(active_symbols);
403static pthread_mutex_t active_symbols_lock = PTHREAD_MUTEX_INITIALIZER;
404
405/*
406 * Ordering weight: count-1 * count-2 * ... / count-n
407 */
408static double sym_weight(const struct sym_entry *sym)
409{
410 double weight = sym->snap_count;
411 int counter;
412
413 if (!display_weighted)
414 return weight;
415 227
416 for (counter = 1; counter < nr_counters-1; counter++) 228 if (notes->src == NULL)
417 weight *= sym->count[counter]; 229 goto out_unlock;
418 230
419 weight /= (sym->count[counter] + 1); 231 printf("Showing %s for %s\n", event_name(top.sym_evsel), symbol->name);
232 printf(" Events Pcnt (>=%d%%)\n", sym_pcnt_filter);
420 233
421 return weight; 234 more = symbol__annotate_printf(symbol, syme->map, top.sym_evsel->idx,
235 0, sym_pcnt_filter, top.print_entries, 4);
236 if (top.zero)
237 symbol__annotate_zero_histogram(symbol, top.sym_evsel->idx);
238 else
239 symbol__annotate_decay_histogram(symbol, top.sym_evsel->idx);
240 if (more != 0)
241 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
242out_unlock:
243 pthread_mutex_unlock(&notes->lock);
422} 244}
423 245
424static long samples;
425static long kernel_samples, us_samples;
426static long exact_samples;
427static long guest_us_samples, guest_kernel_samples;
428static const char CONSOLE_CLEAR[] = ""; 246static const char CONSOLE_CLEAR[] = "";
429 247
430static void __list_insert_active_sym(struct sym_entry *syme) 248static void __list_insert_active_sym(struct sym_entry *syme)
431{ 249{
432 list_add(&syme->node, &active_symbols); 250 list_add(&syme->node, &top.active_symbols);
433}
434
435static void list_remove_active_sym(struct sym_entry *syme)
436{
437 pthread_mutex_lock(&active_symbols_lock);
438 list_del_init(&syme->node);
439 pthread_mutex_unlock(&active_symbols_lock);
440} 251}
441 252
442static void rb_insert_active_sym(struct rb_root *tree, struct sym_entry *se) 253static void print_sym_table(struct perf_session *session)
443{ 254{
444 struct rb_node **p = &tree->rb_node; 255 char bf[160];
445 struct rb_node *parent = NULL; 256 int printed = 0;
446 struct sym_entry *iter;
447
448 while (*p != NULL) {
449 parent = *p;
450 iter = rb_entry(parent, struct sym_entry, rb_node);
451
452 if (se->weight > iter->weight)
453 p = &(*p)->rb_left;
454 else
455 p = &(*p)->rb_right;
456 }
457
458 rb_link_node(&se->rb_node, parent, p);
459 rb_insert_color(&se->rb_node, tree);
460}
461
462static void print_sym_table(void)
463{
464 int printed = 0, j;
465 struct perf_evsel *counter;
466 int snap = !display_weighted ? sym_counter : 0;
467 float samples_per_sec = samples/delay_secs;
468 float ksamples_per_sec = kernel_samples/delay_secs;
469 float us_samples_per_sec = (us_samples)/delay_secs;
470 float guest_kernel_samples_per_sec = (guest_kernel_samples)/delay_secs;
471 float guest_us_samples_per_sec = (guest_us_samples)/delay_secs;
472 float esamples_percent = (100.0*exact_samples)/samples;
473 float sum_ksamples = 0.0;
474 struct sym_entry *syme, *n;
475 struct rb_root tmp = RB_ROOT;
476 struct rb_node *nd; 257 struct rb_node *nd;
477 int sym_width = 0, dso_width = 0, dso_short_width = 0; 258 struct sym_entry *syme;
259 struct rb_root tmp = RB_ROOT;
478 const int win_width = winsize.ws_col - 1; 260 const int win_width = winsize.ws_col - 1;
479 261 int sym_width, dso_width, dso_short_width;
480 samples = us_samples = kernel_samples = exact_samples = 0; 262 float sum_ksamples = perf_top__decay_samples(&top, &tmp);
481 guest_kernel_samples = guest_us_samples = 0;
482
483 /* Sort the active symbols */
484 pthread_mutex_lock(&active_symbols_lock);
485 syme = list_entry(active_symbols.next, struct sym_entry, node);
486 pthread_mutex_unlock(&active_symbols_lock);
487
488 list_for_each_entry_safe_from(syme, n, &active_symbols, node) {
489 syme->snap_count = syme->count[snap];
490 if (syme->snap_count != 0) {
491
492 if ((hide_user_symbols &&
493 syme->origin == PERF_RECORD_MISC_USER) ||
494 (hide_kernel_symbols &&
495 syme->origin == PERF_RECORD_MISC_KERNEL)) {
496 list_remove_active_sym(syme);
497 continue;
498 }
499 syme->weight = sym_weight(syme);
500 rb_insert_active_sym(&tmp, syme);
501 sum_ksamples += syme->snap_count;
502
503 for (j = 0; j < nr_counters; j++)
504 syme->count[j] = zero ? 0 : syme->count[j] * 7 / 8;
505 } else
506 list_remove_active_sym(syme);
507 }
508 263
509 puts(CONSOLE_CLEAR); 264 puts(CONSOLE_CLEAR);
510 265
511 printf("%-*.*s\n", win_width, win_width, graph_dotted_line); 266 perf_top__header_snprintf(&top, bf, sizeof(bf));
512 if (!perf_guest) { 267 printf("%s\n", bf);
513 printf(" PerfTop:%8.0f irqs/sec kernel:%4.1f%%"
514 " exact: %4.1f%% [",
515 samples_per_sec,
516 100.0 - (100.0 * ((samples_per_sec - ksamples_per_sec) /
517 samples_per_sec)),
518 esamples_percent);
519 } else {
520 printf(" PerfTop:%8.0f irqs/sec kernel:%4.1f%% us:%4.1f%%"
521 " guest kernel:%4.1f%% guest us:%4.1f%%"
522 " exact: %4.1f%% [",
523 samples_per_sec,
524 100.0 - (100.0 * ((samples_per_sec-ksamples_per_sec) /
525 samples_per_sec)),
526 100.0 - (100.0 * ((samples_per_sec-us_samples_per_sec) /
527 samples_per_sec)),
528 100.0 - (100.0 * ((samples_per_sec -
529 guest_kernel_samples_per_sec) /
530 samples_per_sec)),
531 100.0 - (100.0 * ((samples_per_sec -
532 guest_us_samples_per_sec) /
533 samples_per_sec)),
534 esamples_percent);
535 }
536
537 if (nr_counters == 1 || !display_weighted) {
538 struct perf_evsel *first;
539 first = list_entry(evsel_list.next, struct perf_evsel, node);
540 printf("%Ld", first->attr.sample_period);
541 if (freq)
542 printf("Hz ");
543 else
544 printf(" ");
545 }
546
547 if (!display_weighted)
548 printf("%s", event_name(sym_evsel));
549 else list_for_each_entry(counter, &evsel_list, node) {
550 if (counter->idx)
551 printf("/");
552
553 printf("%s", event_name(counter));
554 }
555 268
556 printf( "], "); 269 perf_top__reset_sample_counters(&top);
557
558 if (target_pid != -1)
559 printf(" (target_pid: %d", target_pid);
560 else if (target_tid != -1)
561 printf(" (target_tid: %d", target_tid);
562 else
563 printf(" (all");
564
565 if (cpu_list)
566 printf(", CPU%s: %s)\n", cpus->nr > 1 ? "s" : "", cpu_list);
567 else {
568 if (target_tid != -1)
569 printf(")\n");
570 else
571 printf(", %d CPU%s)\n", cpus->nr, cpus->nr > 1 ? "s" : "");
572 }
573 270
574 printf("%-*.*s\n", win_width, win_width, graph_dotted_line); 271 printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
575 272
576 if (sym_filter_entry) { 273 if (session->hists.stats.total_lost != 0) {
577 show_details(sym_filter_entry); 274 color_fprintf(stdout, PERF_COLOR_RED, "WARNING:");
578 return; 275 printf(" LOST %" PRIu64 " events, Check IO/CPU overload\n",
276 session->hists.stats.total_lost);
579 } 277 }
580 278
581 /* 279 if (top.sym_filter_entry) {
582 * Find the longest symbol name that will be displayed 280 show_details(top.sym_filter_entry);
583 */ 281 return;
584 for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
585 syme = rb_entry(nd, struct sym_entry, rb_node);
586 if (++printed > print_entries ||
587 (int)syme->snap_count < count_filter)
588 continue;
589
590 if (syme->map->dso->long_name_len > dso_width)
591 dso_width = syme->map->dso->long_name_len;
592
593 if (syme->map->dso->short_name_len > dso_short_width)
594 dso_short_width = syme->map->dso->short_name_len;
595
596 if (syme->name_len > sym_width)
597 sym_width = syme->name_len;
598 } 282 }
599 283
600 printed = 0; 284 perf_top__find_widths(&top, &tmp, &dso_width, &dso_short_width,
285 &sym_width);
601 286
602 if (sym_width + dso_width > winsize.ws_col - 29) { 287 if (sym_width + dso_width > winsize.ws_col - 29) {
603 dso_width = dso_short_width; 288 dso_width = dso_short_width;
@@ -605,7 +290,7 @@ static void print_sym_table(void)
605 sym_width = winsize.ws_col - dso_width - 29; 290 sym_width = winsize.ws_col - dso_width - 29;
606 } 291 }
607 putchar('\n'); 292 putchar('\n');
608 if (nr_counters == 1) 293 if (top.evlist->nr_entries == 1)
609 printf(" samples pcnt"); 294 printf(" samples pcnt");
610 else 295 else
611 printf(" weight samples pcnt"); 296 printf(" weight samples pcnt");
@@ -614,7 +299,7 @@ static void print_sym_table(void)
614 printf(" RIP "); 299 printf(" RIP ");
615 printf(" %-*.*s DSO\n", sym_width, sym_width, "function"); 300 printf(" %-*.*s DSO\n", sym_width, sym_width, "function");
616 printf(" %s _______ _____", 301 printf(" %s _______ _____",
617 nr_counters == 1 ? " " : "______"); 302 top.evlist->nr_entries == 1 ? " " : "______");
618 if (verbose) 303 if (verbose)
619 printf(" ________________"); 304 printf(" ________________");
620 printf(" %-*.*s", sym_width, sym_width, graph_line); 305 printf(" %-*.*s", sym_width, sym_width, graph_line);
@@ -627,20 +312,21 @@ static void print_sym_table(void)
627 312
628 syme = rb_entry(nd, struct sym_entry, rb_node); 313 syme = rb_entry(nd, struct sym_entry, rb_node);
629 sym = sym_entry__symbol(syme); 314 sym = sym_entry__symbol(syme);
630 if (++printed > print_entries || (int)syme->snap_count < count_filter) 315 if (++printed > top.print_entries ||
316 (int)syme->snap_count < top.count_filter)
631 continue; 317 continue;
632 318
633 pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) / 319 pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) /
634 sum_ksamples)); 320 sum_ksamples));
635 321
636 if (nr_counters == 1 || !display_weighted) 322 if (top.evlist->nr_entries == 1 || !top.display_weighted)
637 printf("%20.2f ", syme->weight); 323 printf("%20.2f ", syme->weight);
638 else 324 else
639 printf("%9.1f %10ld ", syme->weight, syme->snap_count); 325 printf("%9.1f %10ld ", syme->weight, syme->snap_count);
640 326
641 percent_color_fprintf(stdout, "%4.1f%%", pcnt); 327 percent_color_fprintf(stdout, "%4.1f%%", pcnt);
642 if (verbose) 328 if (verbose)
643 printf(" %016llx", sym->start); 329 printf(" %016" PRIx64, sym->start);
644 printf(" %-*.*s", sym_width, sym_width, sym->name); 330 printf(" %-*.*s", sym_width, sym_width, sym->name);
645 printf(" %-*.*s\n", dso_width, dso_width, 331 printf(" %-*.*s\n", dso_width, dso_width,
646 dso_width >= syme->map->dso->long_name_len ? 332 dso_width >= syme->map->dso->long_name_len ?
@@ -692,10 +378,8 @@ static void prompt_symbol(struct sym_entry **target, const char *msg)
692 378
693 /* zero counters of active symbol */ 379 /* zero counters of active symbol */
694 if (syme) { 380 if (syme) {
695 pthread_mutex_lock(&syme->src->lock);
696 __zero_source_counters(syme); 381 __zero_source_counters(syme);
697 *target = NULL; 382 *target = NULL;
698 pthread_mutex_unlock(&syme->src->lock);
699 } 383 }
700 384
701 fprintf(stdout, "\n%s: ", msg); 385 fprintf(stdout, "\n%s: ", msg);
@@ -706,11 +390,11 @@ static void prompt_symbol(struct sym_entry **target, const char *msg)
706 if (p) 390 if (p)
707 *p = 0; 391 *p = 0;
708 392
709 pthread_mutex_lock(&active_symbols_lock); 393 pthread_mutex_lock(&top.active_symbols_lock);
710 syme = list_entry(active_symbols.next, struct sym_entry, node); 394 syme = list_entry(top.active_symbols.next, struct sym_entry, node);
711 pthread_mutex_unlock(&active_symbols_lock); 395 pthread_mutex_unlock(&top.active_symbols_lock);
712 396
713 list_for_each_entry_safe_from(syme, n, &active_symbols, node) { 397 list_for_each_entry_safe_from(syme, n, &top.active_symbols, node) {
714 struct symbol *sym = sym_entry__symbol(syme); 398 struct symbol *sym = sym_entry__symbol(syme);
715 399
716 if (!strcmp(buf, sym->name)) { 400 if (!strcmp(buf, sym->name)) {
@@ -734,34 +418,34 @@ static void print_mapped_keys(void)
734{ 418{
735 char *name = NULL; 419 char *name = NULL;
736 420
737 if (sym_filter_entry) { 421 if (top.sym_filter_entry) {
738 struct symbol *sym = sym_entry__symbol(sym_filter_entry); 422 struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
739 name = sym->name; 423 name = sym->name;
740 } 424 }
741 425
742 fprintf(stdout, "\nMapped keys:\n"); 426 fprintf(stdout, "\nMapped keys:\n");
743 fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", delay_secs); 427 fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top.delay_secs);
744 fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", print_entries); 428 fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top.print_entries);
745 429
746 if (nr_counters > 1) 430 if (top.evlist->nr_entries > 1)
747 fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(sym_evsel)); 431 fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(top.sym_evsel));
748 432
749 fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", count_filter); 433 fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top.count_filter);
750 434
751 fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter); 435 fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter);
752 fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL"); 436 fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL");
753 fprintf(stdout, "\t[S] stop annotation.\n"); 437 fprintf(stdout, "\t[S] stop annotation.\n");
754 438
755 if (nr_counters > 1) 439 if (top.evlist->nr_entries > 1)
756 fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", display_weighted ? 1 : 0); 440 fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", top.display_weighted ? 1 : 0);
757 441
758 fprintf(stdout, 442 fprintf(stdout,
759 "\t[K] hide kernel_symbols symbols. \t(%s)\n", 443 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
760 hide_kernel_symbols ? "yes" : "no"); 444 top.hide_kernel_symbols ? "yes" : "no");
761 fprintf(stdout, 445 fprintf(stdout,
762 "\t[U] hide user symbols. \t(%s)\n", 446 "\t[U] hide user symbols. \t(%s)\n",
763 hide_user_symbols ? "yes" : "no"); 447 top.hide_user_symbols ? "yes" : "no");
764 fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", zero ? 1 : 0); 448 fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top.zero ? 1 : 0);
765 fprintf(stdout, "\t[qQ] quit.\n"); 449 fprintf(stdout, "\t[qQ] quit.\n");
766} 450}
767 451
@@ -782,7 +466,7 @@ static int key_mapped(int c)
782 return 1; 466 return 1;
783 case 'E': 467 case 'E':
784 case 'w': 468 case 'w':
785 return nr_counters > 1 ? 1 : 0; 469 return top.evlist->nr_entries > 1 ? 1 : 0;
786 default: 470 default:
787 break; 471 break;
788 } 472 }
@@ -817,47 +501,50 @@ static void handle_keypress(struct perf_session *session, int c)
817 501
818 switch (c) { 502 switch (c) {
819 case 'd': 503 case 'd':
820 prompt_integer(&delay_secs, "Enter display delay"); 504 prompt_integer(&top.delay_secs, "Enter display delay");
821 if (delay_secs < 1) 505 if (top.delay_secs < 1)
822 delay_secs = 1; 506 top.delay_secs = 1;
823 break; 507 break;
824 case 'e': 508 case 'e':
825 prompt_integer(&print_entries, "Enter display entries (lines)"); 509 prompt_integer(&top.print_entries, "Enter display entries (lines)");
826 if (print_entries == 0) { 510 if (top.print_entries == 0) {
827 sig_winch_handler(SIGWINCH); 511 sig_winch_handler(SIGWINCH);
828 signal(SIGWINCH, sig_winch_handler); 512 signal(SIGWINCH, sig_winch_handler);
829 } else 513 } else
830 signal(SIGWINCH, SIG_DFL); 514 signal(SIGWINCH, SIG_DFL);
831 break; 515 break;
832 case 'E': 516 case 'E':
833 if (nr_counters > 1) { 517 if (top.evlist->nr_entries > 1) {
518 /* Select 0 as the default event: */
519 int counter = 0;
520
834 fprintf(stderr, "\nAvailable events:"); 521 fprintf(stderr, "\nAvailable events:");
835 522
836 list_for_each_entry(sym_evsel, &evsel_list, node) 523 list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
837 fprintf(stderr, "\n\t%d %s", sym_evsel->idx, event_name(sym_evsel)); 524 fprintf(stderr, "\n\t%d %s", top.sym_evsel->idx, event_name(top.sym_evsel));
838 525
839 prompt_integer(&sym_counter, "Enter details event counter"); 526 prompt_integer(&counter, "Enter details event counter");
840 527
841 if (sym_counter >= nr_counters) { 528 if (counter >= top.evlist->nr_entries) {
842 sym_evsel = list_entry(evsel_list.next, struct perf_evsel, node); 529 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
843 sym_counter = 0; 530 fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(top.sym_evsel));
844 fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(sym_evsel));
845 sleep(1); 531 sleep(1);
846 break; 532 break;
847 } 533 }
848 list_for_each_entry(sym_evsel, &evsel_list, node) 534 list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
849 if (sym_evsel->idx == sym_counter) 535 if (top.sym_evsel->idx == counter)
850 break; 536 break;
851 } else sym_counter = 0; 537 } else
538 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
852 break; 539 break;
853 case 'f': 540 case 'f':
854 prompt_integer(&count_filter, "Enter display event count filter"); 541 prompt_integer(&top.count_filter, "Enter display event count filter");
855 break; 542 break;
856 case 'F': 543 case 'F':
857 prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)"); 544 prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)");
858 break; 545 break;
859 case 'K': 546 case 'K':
860 hide_kernel_symbols = !hide_kernel_symbols; 547 top.hide_kernel_symbols = !top.hide_kernel_symbols;
861 break; 548 break;
862 case 'q': 549 case 'q':
863 case 'Q': 550 case 'Q':
@@ -866,34 +553,50 @@ static void handle_keypress(struct perf_session *session, int c)
866 perf_session__fprintf_dsos(session, stderr); 553 perf_session__fprintf_dsos(session, stderr);
867 exit(0); 554 exit(0);
868 case 's': 555 case 's':
869 prompt_symbol(&sym_filter_entry, "Enter details symbol"); 556 prompt_symbol(&top.sym_filter_entry, "Enter details symbol");
870 break; 557 break;
871 case 'S': 558 case 'S':
872 if (!sym_filter_entry) 559 if (!top.sym_filter_entry)
873 break; 560 break;
874 else { 561 else {
875 struct sym_entry *syme = sym_filter_entry; 562 struct sym_entry *syme = top.sym_filter_entry;
876 563
877 pthread_mutex_lock(&syme->src->lock); 564 top.sym_filter_entry = NULL;
878 sym_filter_entry = NULL;
879 __zero_source_counters(syme); 565 __zero_source_counters(syme);
880 pthread_mutex_unlock(&syme->src->lock);
881 } 566 }
882 break; 567 break;
883 case 'U': 568 case 'U':
884 hide_user_symbols = !hide_user_symbols; 569 top.hide_user_symbols = !top.hide_user_symbols;
885 break; 570 break;
886 case 'w': 571 case 'w':
887 display_weighted = ~display_weighted; 572 top.display_weighted = ~top.display_weighted;
888 break; 573 break;
889 case 'z': 574 case 'z':
890 zero = !zero; 575 top.zero = !top.zero;
891 break; 576 break;
892 default: 577 default:
893 break; 578 break;
894 } 579 }
895} 580}
896 581
582static void *display_thread_tui(void *arg __used)
583{
584 int err = 0;
585 pthread_mutex_lock(&top.active_symbols_lock);
586 while (list_empty(&top.active_symbols)) {
587 err = pthread_cond_wait(&top.active_symbols_cond,
588 &top.active_symbols_lock);
589 if (err)
590 break;
591 }
592 pthread_mutex_unlock(&top.active_symbols_lock);
593 if (!err)
594 perf_top__tui_browser(&top);
595 exit_browser(0);
596 exit(0);
597 return NULL;
598}
599
897static void *display_thread(void *arg __used) 600static void *display_thread(void *arg __used)
898{ 601{
899 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN }; 602 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
@@ -908,13 +611,13 @@ static void *display_thread(void *arg __used)
908 tc.c_cc[VTIME] = 0; 611 tc.c_cc[VTIME] = 0;
909 612
910repeat: 613repeat:
911 delay_msecs = delay_secs * 1000; 614 delay_msecs = top.delay_secs * 1000;
912 tcsetattr(0, TCSANOW, &tc); 615 tcsetattr(0, TCSANOW, &tc);
913 /* trash return*/ 616 /* trash return*/
914 getc(stdin); 617 getc(stdin);
915 618
916 do { 619 do {
917 print_sym_table(); 620 print_sym_table(session);
918 } while (!poll(&stdin_poll, 1, delay_msecs) == 1); 621 } while (!poll(&stdin_poll, 1, delay_msecs) == 1);
919 622
920 c = getc(stdin); 623 c = getc(stdin);
@@ -929,6 +632,7 @@ repeat:
929/* Tag samples to be skipped. */ 632/* Tag samples to be skipped. */
930static const char *skip_symbols[] = { 633static const char *skip_symbols[] = {
931 "default_idle", 634 "default_idle",
635 "native_safe_halt",
932 "cpu_idle", 636 "cpu_idle",
933 "enter_idle", 637 "enter_idle",
934 "exit_idle", 638 "exit_idle",
@@ -964,9 +668,9 @@ static int symbol_filter(struct map *map, struct symbol *sym)
964 668
965 syme = symbol__priv(sym); 669 syme = symbol__priv(sym);
966 syme->map = map; 670 syme->map = map;
967 syme->src = NULL; 671 symbol__annotate_init(map, sym);
968 672
969 if (!sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) { 673 if (!top.sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) {
970 /* schedule initial sym_filter_entry setup */ 674 /* schedule initial sym_filter_entry setup */
971 sym_filter_entry_sched = syme; 675 sym_filter_entry_sched = syme;
972 sym_filter = NULL; 676 sym_filter = NULL;
@@ -974,49 +678,45 @@ static int symbol_filter(struct map *map, struct symbol *sym)
974 678
975 for (i = 0; skip_symbols[i]; i++) { 679 for (i = 0; skip_symbols[i]; i++) {
976 if (!strcmp(skip_symbols[i], name)) { 680 if (!strcmp(skip_symbols[i], name)) {
977 syme->skip = 1; 681 sym->ignore = true;
978 break; 682 break;
979 } 683 }
980 } 684 }
981 685
982 if (!syme->skip)
983 syme->name_len = strlen(sym->name);
984
985 return 0; 686 return 0;
986} 687}
987 688
988static void event__process_sample(const event_t *self, 689static void perf_event__process_sample(const union perf_event *event,
989 struct sample_data *sample, 690 struct perf_sample *sample,
990 struct perf_session *session, 691 struct perf_session *session)
991 struct perf_evsel *evsel)
992{ 692{
993 u64 ip = self->ip.ip; 693 u64 ip = event->ip.ip;
994 struct sym_entry *syme; 694 struct sym_entry *syme;
995 struct addr_location al; 695 struct addr_location al;
996 struct machine *machine; 696 struct machine *machine;
997 u8 origin = self->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 697 u8 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
998 698
999 ++samples; 699 ++top.samples;
1000 700
1001 switch (origin) { 701 switch (origin) {
1002 case PERF_RECORD_MISC_USER: 702 case PERF_RECORD_MISC_USER:
1003 ++us_samples; 703 ++top.us_samples;
1004 if (hide_user_symbols) 704 if (top.hide_user_symbols)
1005 return; 705 return;
1006 machine = perf_session__find_host_machine(session); 706 machine = perf_session__find_host_machine(session);
1007 break; 707 break;
1008 case PERF_RECORD_MISC_KERNEL: 708 case PERF_RECORD_MISC_KERNEL:
1009 ++kernel_samples; 709 ++top.kernel_samples;
1010 if (hide_kernel_symbols) 710 if (top.hide_kernel_symbols)
1011 return; 711 return;
1012 machine = perf_session__find_host_machine(session); 712 machine = perf_session__find_host_machine(session);
1013 break; 713 break;
1014 case PERF_RECORD_MISC_GUEST_KERNEL: 714 case PERF_RECORD_MISC_GUEST_KERNEL:
1015 ++guest_kernel_samples; 715 ++top.guest_kernel_samples;
1016 machine = perf_session__find_machine(session, self->ip.pid); 716 machine = perf_session__find_machine(session, event->ip.pid);
1017 break; 717 break;
1018 case PERF_RECORD_MISC_GUEST_USER: 718 case PERF_RECORD_MISC_GUEST_USER:
1019 ++guest_us_samples; 719 ++top.guest_us_samples;
1020 /* 720 /*
1021 * TODO: we don't process guest user from host side 721 * TODO: we don't process guest user from host side
1022 * except simple counting. 722 * except simple counting.
@@ -1028,15 +728,15 @@ static void event__process_sample(const event_t *self,
1028 728
1029 if (!machine && perf_guest) { 729 if (!machine && perf_guest) {
1030 pr_err("Can't find guest [%d]'s kernel information\n", 730 pr_err("Can't find guest [%d]'s kernel information\n",
1031 self->ip.pid); 731 event->ip.pid);
1032 return; 732 return;
1033 } 733 }
1034 734
1035 if (self->header.misc & PERF_RECORD_MISC_EXACT_IP) 735 if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
1036 exact_samples++; 736 top.exact_samples++;
1037 737
1038 if (event__preprocess_sample(self, session, &al, sample, 738 if (perf_event__preprocess_sample(event, session, &al, sample,
1039 symbol_filter) < 0 || 739 symbol_filter) < 0 ||
1040 al.filtered) 740 al.filtered)
1041 return; 741 return;
1042 742
@@ -1054,8 +754,9 @@ static void event__process_sample(const event_t *self,
1054 */ 754 */
1055 if (al.map == machine->vmlinux_maps[MAP__FUNCTION] && 755 if (al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
1056 RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) { 756 RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
1057 pr_err("The %s file can't be used\n", 757 ui__warning("The %s file can't be used\n",
1058 symbol_conf.vmlinux_name); 758 symbol_conf.vmlinux_name);
759 exit_browser(0);
1059 exit(1); 760 exit(1);
1060 } 761 }
1061 762
@@ -1064,13 +765,13 @@ static void event__process_sample(const event_t *self,
1064 765
1065 /* let's see, whether we need to install initial sym_filter_entry */ 766 /* let's see, whether we need to install initial sym_filter_entry */
1066 if (sym_filter_entry_sched) { 767 if (sym_filter_entry_sched) {
1067 sym_filter_entry = sym_filter_entry_sched; 768 top.sym_filter_entry = sym_filter_entry_sched;
1068 sym_filter_entry_sched = NULL; 769 sym_filter_entry_sched = NULL;
1069 if (parse_source(sym_filter_entry) < 0) { 770 if (parse_source(top.sym_filter_entry) < 0) {
1070 struct symbol *sym = sym_entry__symbol(sym_filter_entry); 771 struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
1071 772
1072 pr_err("Can't annotate %s", sym->name); 773 pr_err("Can't annotate %s", sym->name);
1073 if (sym_filter_entry->map->dso->origin == DSO__ORIG_KERNEL) { 774 if (top.sym_filter_entry->map->dso->symtab_type == SYMTAB__KALLSYMS) {
1074 pr_err(": No vmlinux file was found in the path:\n"); 775 pr_err(": No vmlinux file was found in the path:\n");
1075 machine__fprintf_vmlinux_path(machine, stderr); 776 machine__fprintf_vmlinux_path(machine, stderr);
1076 } else 777 } else
@@ -1080,167 +781,73 @@ static void event__process_sample(const event_t *self,
1080 } 781 }
1081 782
1082 syme = symbol__priv(al.sym); 783 syme = symbol__priv(al.sym);
1083 if (!syme->skip) { 784 if (!al.sym->ignore) {
785 struct perf_evsel *evsel;
786
787 evsel = perf_evlist__id2evsel(top.evlist, sample->id);
788 assert(evsel != NULL);
1084 syme->count[evsel->idx]++; 789 syme->count[evsel->idx]++;
1085 syme->origin = origin;
1086 record_precise_ip(syme, evsel->idx, ip); 790 record_precise_ip(syme, evsel->idx, ip);
1087 pthread_mutex_lock(&active_symbols_lock); 791 pthread_mutex_lock(&top.active_symbols_lock);
1088 if (list_empty(&syme->node) || !syme->node.next) 792 if (list_empty(&syme->node) || !syme->node.next) {
793 static bool first = true;
1089 __list_insert_active_sym(syme); 794 __list_insert_active_sym(syme);
1090 pthread_mutex_unlock(&active_symbols_lock); 795 if (first) {
796 pthread_cond_broadcast(&top.active_symbols_cond);
797 first = false;
798 }
799 }
800 pthread_mutex_unlock(&top.active_symbols_lock);
1091 } 801 }
1092} 802}
1093 803
1094struct mmap_data { 804static void perf_session__mmap_read_cpu(struct perf_session *self, int cpu)
1095 void *base;
1096 int mask;
1097 unsigned int prev;
1098};
1099
1100static int perf_evsel__alloc_mmap_per_thread(struct perf_evsel *evsel,
1101 int ncpus, int nthreads)
1102{
1103 evsel->priv = xyarray__new(ncpus, nthreads, sizeof(struct mmap_data));
1104 return evsel->priv != NULL ? 0 : -ENOMEM;
1105}
1106
1107static void perf_evsel__free_mmap(struct perf_evsel *evsel)
1108{
1109 xyarray__delete(evsel->priv);
1110 evsel->priv = NULL;
1111}
1112
1113static unsigned int mmap_read_head(struct mmap_data *md)
1114{
1115 struct perf_event_mmap_page *pc = md->base;
1116 int head;
1117
1118 head = pc->data_head;
1119 rmb();
1120
1121 return head;
1122}
1123
1124static void perf_session__mmap_read_counter(struct perf_session *self,
1125 struct perf_evsel *evsel,
1126 int cpu, int thread_idx)
1127{ 805{
1128 struct xyarray *mmap_array = evsel->priv; 806 struct perf_sample sample;
1129 struct mmap_data *md = xyarray__entry(mmap_array, cpu, thread_idx); 807 union perf_event *event;
1130 unsigned int head = mmap_read_head(md);
1131 unsigned int old = md->prev;
1132 unsigned char *data = md->base + page_size;
1133 struct sample_data sample;
1134 int diff;
1135
1136 /*
1137 * If we're further behind than half the buffer, there's a chance
1138 * the writer will bite our tail and mess up the samples under us.
1139 *
1140 * If we somehow ended up ahead of the head, we got messed up.
1141 *
1142 * In either case, truncate and restart at head.
1143 */
1144 diff = head - old;
1145 if (diff > md->mask / 2 || diff < 0) {
1146 fprintf(stderr, "WARNING: failed to keep up with mmap data.\n");
1147
1148 /*
1149 * head points to a known good entry, start there.
1150 */
1151 old = head;
1152 }
1153
1154 for (; old != head;) {
1155 event_t *event = (event_t *)&data[old & md->mask];
1156
1157 event_t event_copy;
1158 808
1159 size_t size = event->header.size; 809 while ((event = perf_evlist__read_on_cpu(top.evlist, cpu)) != NULL) {
810 perf_session__parse_sample(self, event, &sample);
1160 811
1161 /*
1162 * Event straddles the mmap boundary -- header should always
1163 * be inside due to u64 alignment of output.
1164 */
1165 if ((old & md->mask) + size != ((old + size) & md->mask)) {
1166 unsigned int offset = old;
1167 unsigned int len = min(sizeof(*event), size), cpy;
1168 void *dst = &event_copy;
1169
1170 do {
1171 cpy = min(md->mask + 1 - (offset & md->mask), len);
1172 memcpy(dst, &data[offset & md->mask], cpy);
1173 offset += cpy;
1174 dst += cpy;
1175 len -= cpy;
1176 } while (len);
1177
1178 event = &event_copy;
1179 }
1180
1181 event__parse_sample(event, self, &sample);
1182 if (event->header.type == PERF_RECORD_SAMPLE) 812 if (event->header.type == PERF_RECORD_SAMPLE)
1183 event__process_sample(event, &sample, self, evsel); 813 perf_event__process_sample(event, &sample, self);
1184 else 814 else
1185 event__process(event, &sample, self); 815 perf_event__process(event, &sample, self);
1186 old += size;
1187 } 816 }
1188
1189 md->prev = old;
1190} 817}
1191 818
1192static struct pollfd *event_array;
1193
1194static void perf_session__mmap_read(struct perf_session *self) 819static void perf_session__mmap_read(struct perf_session *self)
1195{ 820{
1196 struct perf_evsel *counter; 821 int i;
1197 int i, thread_index;
1198
1199 for (i = 0; i < cpus->nr; i++) {
1200 list_for_each_entry(counter, &evsel_list, node) {
1201 for (thread_index = 0;
1202 thread_index < threads->nr;
1203 thread_index++) {
1204 perf_session__mmap_read_counter(self,
1205 counter, i, thread_index);
1206 }
1207 }
1208 }
1209}
1210 822
1211int nr_poll; 823 for (i = 0; i < top.evlist->cpus->nr; i++)
1212int group_fd; 824 perf_session__mmap_read_cpu(self, i);
825}
1213 826
1214static void start_counter(int i, struct perf_evsel *evsel) 827static void start_counters(struct perf_evlist *evlist)
1215{ 828{
1216 struct xyarray *mmap_array = evsel->priv; 829 struct perf_evsel *counter;
1217 struct mmap_data *mm;
1218 struct perf_event_attr *attr;
1219 int cpu = -1;
1220 int thread_index;
1221
1222 if (target_tid == -1)
1223 cpu = cpus->map[i];
1224 830
1225 attr = &evsel->attr; 831 list_for_each_entry(counter, &evlist->entries, node) {
832 struct perf_event_attr *attr = &counter->attr;
1226 833
1227 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID; 834 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
1228 835
1229 if (freq) { 836 if (top.freq) {
1230 attr->sample_type |= PERF_SAMPLE_PERIOD; 837 attr->sample_type |= PERF_SAMPLE_PERIOD;
1231 attr->freq = 1; 838 attr->freq = 1;
1232 attr->sample_freq = freq; 839 attr->sample_freq = top.freq;
1233 } 840 }
1234 841
1235 attr->inherit = (cpu < 0) && inherit; 842 if (evlist->nr_entries > 1) {
1236 attr->mmap = 1; 843 attr->sample_type |= PERF_SAMPLE_ID;
844 attr->read_format |= PERF_FORMAT_ID;
845 }
1237 846
1238 for (thread_index = 0; thread_index < threads->nr; thread_index++) { 847 attr->mmap = 1;
1239try_again: 848try_again:
1240 FD(evsel, i, thread_index) = sys_perf_event_open(attr, 849 if (perf_evsel__open(counter, top.evlist->cpus,
1241 threads->map[thread_index], cpu, group_fd, 0); 850 top.evlist->threads, group, inherit) < 0) {
1242
1243 if (FD(evsel, i, thread_index) < 0) {
1244 int err = errno; 851 int err = errno;
1245 852
1246 if (err == EPERM || err == EACCES) 853 if (err == EPERM || err == EACCES)
@@ -1252,8 +859,8 @@ try_again:
1252 * based cpu-clock-tick sw counter, which 859 * based cpu-clock-tick sw counter, which
1253 * is always available even if no PMU support: 860 * is always available even if no PMU support:
1254 */ 861 */
1255 if (attr->type == PERF_TYPE_HARDWARE 862 if (attr->type == PERF_TYPE_HARDWARE &&
1256 && attr->config == PERF_COUNT_HW_CPU_CYCLES) { 863 attr->config == PERF_COUNT_HW_CPU_CYCLES) {
1257 864
1258 if (verbose) 865 if (verbose)
1259 warning(" ... trying to fall back to cpu-clock-ticks\n"); 866 warning(" ... trying to fall back to cpu-clock-ticks\n");
@@ -1263,39 +870,22 @@ try_again:
1263 goto try_again; 870 goto try_again;
1264 } 871 }
1265 printf("\n"); 872 printf("\n");
1266 error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", 873 error("sys_perf_event_open() syscall returned with %d "
1267 FD(evsel, i, thread_index), strerror(err)); 874 "(%s). /bin/dmesg may provide additional information.\n",
875 err, strerror(err));
1268 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); 876 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
1269 exit(-1); 877 exit(-1);
1270 } 878 }
1271 assert(FD(evsel, i, thread_index) >= 0);
1272 fcntl(FD(evsel, i, thread_index), F_SETFL, O_NONBLOCK);
1273
1274 /*
1275 * First counter acts as the group leader:
1276 */
1277 if (group && group_fd == -1)
1278 group_fd = FD(evsel, i, thread_index);
1279
1280 event_array[nr_poll].fd = FD(evsel, i, thread_index);
1281 event_array[nr_poll].events = POLLIN;
1282 nr_poll++;
1283
1284 mm = xyarray__entry(mmap_array, i, thread_index);
1285 mm->prev = 0;
1286 mm->mask = mmap_pages*page_size - 1;
1287 mm->base = mmap(NULL, (mmap_pages+1)*page_size,
1288 PROT_READ, MAP_SHARED, FD(evsel, i, thread_index), 0);
1289 if (mm->base == MAP_FAILED)
1290 die("failed to mmap with %d (%s)\n", errno, strerror(errno));
1291 } 879 }
880
881 if (perf_evlist__mmap(evlist, mmap_pages, false) < 0)
882 die("failed to mmap with %d (%s)\n", errno, strerror(errno));
1292} 883}
1293 884
1294static int __cmd_top(void) 885static int __cmd_top(void)
1295{ 886{
1296 pthread_t thread; 887 pthread_t thread;
1297 struct perf_evsel *counter; 888 int ret __used;
1298 int i, ret;
1299 /* 889 /*
1300 * FIXME: perf_session__new should allow passing a O_MMAP, so that all this 890 * FIXME: perf_session__new should allow passing a O_MMAP, so that all this
1301 * mmap reading, etc is encapsulated in it. Use O_WRONLY for now. 891 * mmap reading, etc is encapsulated in it. Use O_WRONLY for now.
@@ -1304,23 +894,23 @@ static int __cmd_top(void)
1304 if (session == NULL) 894 if (session == NULL)
1305 return -ENOMEM; 895 return -ENOMEM;
1306 896
1307 if (target_tid != -1) 897 if (top.target_tid != -1)
1308 event__synthesize_thread(target_tid, event__process, session); 898 perf_event__synthesize_thread_map(top.evlist->threads,
899 perf_event__process, session);
1309 else 900 else
1310 event__synthesize_threads(event__process, session); 901 perf_event__synthesize_threads(perf_event__process, session);
1311 902
1312 for (i = 0; i < cpus->nr; i++) { 903 start_counters(top.evlist);
1313 group_fd = -1; 904 session->evlist = top.evlist;
1314 list_for_each_entry(counter, &evsel_list, node) 905 perf_session__update_sample_type(session);
1315 start_counter(i, counter);
1316 }
1317 906
1318 /* Wait for a minimal set of events before starting the snapshot */ 907 /* Wait for a minimal set of events before starting the snapshot */
1319 poll(&event_array[0], nr_poll, 100); 908 poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
1320 909
1321 perf_session__mmap_read(session); 910 perf_session__mmap_read(session);
1322 911
1323 if (pthread_create(&thread, NULL, display_thread, session)) { 912 if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
913 display_thread), session)) {
1324 printf("Could not create display thread.\n"); 914 printf("Could not create display thread.\n");
1325 exit(-1); 915 exit(-1);
1326 } 916 }
@@ -1336,12 +926,12 @@ static int __cmd_top(void)
1336 } 926 }
1337 927
1338 while (1) { 928 while (1) {
1339 int hits = samples; 929 u64 hits = top.samples;
1340 930
1341 perf_session__mmap_read(session); 931 perf_session__mmap_read(session);
1342 932
1343 if (hits == samples) 933 if (hits == top.samples)
1344 ret = poll(event_array, nr_poll, 100); 934 ret = poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
1345 } 935 }
1346 936
1347 return 0; 937 return 0;
@@ -1353,31 +943,31 @@ static const char * const top_usage[] = {
1353}; 943};
1354 944
1355static const struct option options[] = { 945static const struct option options[] = {
1356 OPT_CALLBACK('e', "event", NULL, "event", 946 OPT_CALLBACK('e', "event", &top.evlist, "event",
1357 "event selector. use 'perf list' to list available events", 947 "event selector. use 'perf list' to list available events",
1358 parse_events), 948 parse_events),
1359 OPT_INTEGER('c', "count", &default_interval, 949 OPT_INTEGER('c', "count", &default_interval,
1360 "event period to sample"), 950 "event period to sample"),
1361 OPT_INTEGER('p', "pid", &target_pid, 951 OPT_INTEGER('p', "pid", &top.target_pid,
1362 "profile events on existing process id"), 952 "profile events on existing process id"),
1363 OPT_INTEGER('t', "tid", &target_tid, 953 OPT_INTEGER('t', "tid", &top.target_tid,
1364 "profile events on existing thread id"), 954 "profile events on existing thread id"),
1365 OPT_BOOLEAN('a', "all-cpus", &system_wide, 955 OPT_BOOLEAN('a', "all-cpus", &system_wide,
1366 "system-wide collection from all CPUs"), 956 "system-wide collection from all CPUs"),
1367 OPT_STRING('C', "cpu", &cpu_list, "cpu", 957 OPT_STRING('C', "cpu", &top.cpu_list, "cpu",
1368 "list of cpus to monitor"), 958 "list of cpus to monitor"),
1369 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 959 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1370 "file", "vmlinux pathname"), 960 "file", "vmlinux pathname"),
1371 OPT_BOOLEAN('K', "hide_kernel_symbols", &hide_kernel_symbols, 961 OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1372 "hide kernel symbols"), 962 "hide kernel symbols"),
1373 OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), 963 OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
1374 OPT_INTEGER('r', "realtime", &realtime_prio, 964 OPT_INTEGER('r', "realtime", &realtime_prio,
1375 "collect data with this RT SCHED_FIFO priority"), 965 "collect data with this RT SCHED_FIFO priority"),
1376 OPT_INTEGER('d', "delay", &delay_secs, 966 OPT_INTEGER('d', "delay", &top.delay_secs,
1377 "number of seconds to delay between refreshes"), 967 "number of seconds to delay between refreshes"),
1378 OPT_BOOLEAN('D', "dump-symtab", &dump_symtab, 968 OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
1379 "dump the symbol table used for profiling"), 969 "dump the symbol table used for profiling"),
1380 OPT_INTEGER('f', "count-filter", &count_filter, 970 OPT_INTEGER('f', "count-filter", &top.count_filter,
1381 "only display functions with more events than this"), 971 "only display functions with more events than this"),
1382 OPT_BOOLEAN('g', "group", &group, 972 OPT_BOOLEAN('g', "group", &group,
1383 "put the counters into a counter group"), 973 "put the counters into a counter group"),
@@ -1385,14 +975,16 @@ static const struct option options[] = {
1385 "child tasks inherit counters"), 975 "child tasks inherit counters"),
1386 OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name", 976 OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name",
1387 "symbol to annotate"), 977 "symbol to annotate"),
1388 OPT_BOOLEAN('z', "zero", &zero, 978 OPT_BOOLEAN('z', "zero", &top.zero,
1389 "zero history across updates"), 979 "zero history across updates"),
1390 OPT_INTEGER('F', "freq", &freq, 980 OPT_INTEGER('F', "freq", &top.freq,
1391 "profile at this frequency"), 981 "profile at this frequency"),
1392 OPT_INTEGER('E', "entries", &print_entries, 982 OPT_INTEGER('E', "entries", &top.print_entries,
1393 "display this many functions"), 983 "display this many functions"),
1394 OPT_BOOLEAN('U', "hide_user_symbols", &hide_user_symbols, 984 OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1395 "hide user symbols"), 985 "hide user symbols"),
986 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
987 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
1396 OPT_INCR('v', "verbose", &verbose, 988 OPT_INCR('v', "verbose", &verbose,
1397 "be more verbose (show counter open errors, etc)"), 989 "be more verbose (show counter open errors, etc)"),
1398 OPT_END() 990 OPT_END()
@@ -1403,64 +995,68 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
1403 struct perf_evsel *pos; 995 struct perf_evsel *pos;
1404 int status = -ENOMEM; 996 int status = -ENOMEM;
1405 997
998 top.evlist = perf_evlist__new(NULL, NULL);
999 if (top.evlist == NULL)
1000 return -ENOMEM;
1001
1406 page_size = sysconf(_SC_PAGE_SIZE); 1002 page_size = sysconf(_SC_PAGE_SIZE);
1407 1003
1408 argc = parse_options(argc, argv, options, top_usage, 0); 1004 argc = parse_options(argc, argv, options, top_usage, 0);
1409 if (argc) 1005 if (argc)
1410 usage_with_options(top_usage, options); 1006 usage_with_options(top_usage, options);
1411 1007
1412 if (target_pid != -1) 1008 /*
1413 target_tid = target_pid; 1009 * XXX For now start disabled, only using TUI if explicitely asked for.
1010 * Change that when handle_keys equivalent gets written, live annotation
1011 * done, etc.
1012 */
1013 use_browser = 0;
1414 1014
1415 threads = thread_map__new(target_pid, target_tid); 1015 if (use_stdio)
1416 if (threads == NULL) { 1016 use_browser = 0;
1417 pr_err("Problems finding threads of monitor\n"); 1017 else if (use_tui)
1418 usage_with_options(top_usage, options); 1018 use_browser = 1;
1419 }
1420 1019
1421 event_array = malloc((sizeof(struct pollfd) * 1020 setup_browser(false);
1422 MAX_NR_CPUS * MAX_COUNTERS * threads->nr));
1423 if (!event_array)
1424 return -ENOMEM;
1425 1021
1426 /* CPU and PID are mutually exclusive */ 1022 /* CPU and PID are mutually exclusive */
1427 if (target_tid > 0 && cpu_list) { 1023 if (top.target_tid > 0 && top.cpu_list) {
1428 printf("WARNING: PID switch overriding CPU\n"); 1024 printf("WARNING: PID switch overriding CPU\n");
1429 sleep(1); 1025 sleep(1);
1430 cpu_list = NULL; 1026 top.cpu_list = NULL;
1431 } 1027 }
1432 1028
1433 if (!nr_counters && perf_evsel_list__create_default() < 0) { 1029 if (top.target_pid != -1)
1030 top.target_tid = top.target_pid;
1031
1032 if (perf_evlist__create_maps(top.evlist, top.target_pid,
1033 top.target_tid, top.cpu_list) < 0)
1034 usage_with_options(top_usage, options);
1035
1036 if (!top.evlist->nr_entries &&
1037 perf_evlist__add_default(top.evlist) < 0) {
1434 pr_err("Not enough memory for event selector list\n"); 1038 pr_err("Not enough memory for event selector list\n");
1435 return -ENOMEM; 1039 return -ENOMEM;
1436 } 1040 }
1437 1041
1438 if (delay_secs < 1) 1042 if (top.delay_secs < 1)
1439 delay_secs = 1; 1043 top.delay_secs = 1;
1440 1044
1441 /* 1045 /*
1442 * User specified count overrides default frequency. 1046 * User specified count overrides default frequency.
1443 */ 1047 */
1444 if (default_interval) 1048 if (default_interval)
1445 freq = 0; 1049 top.freq = 0;
1446 else if (freq) { 1050 else if (top.freq) {
1447 default_interval = freq; 1051 default_interval = top.freq;
1448 } else { 1052 } else {
1449 fprintf(stderr, "frequency and count are zero, aborting\n"); 1053 fprintf(stderr, "frequency and count are zero, aborting\n");
1450 exit(EXIT_FAILURE); 1054 exit(EXIT_FAILURE);
1451 } 1055 }
1452 1056
1453 if (target_tid != -1) 1057 list_for_each_entry(pos, &top.evlist->entries, node) {
1454 cpus = cpu_map__dummy_new(); 1058 if (perf_evsel__alloc_fd(pos, top.evlist->cpus->nr,
1455 else 1059 top.evlist->threads->nr) < 0)
1456 cpus = cpu_map__new(cpu_list);
1457
1458 if (cpus == NULL)
1459 usage_with_options(top_usage, options);
1460
1461 list_for_each_entry(pos, &evsel_list, node) {
1462 if (perf_evsel__alloc_mmap_per_thread(pos, cpus->nr, threads->nr) < 0 ||
1463 perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0)
1464 goto out_free_fd; 1060 goto out_free_fd;
1465 /* 1061 /*
1466 * Fill in the ones not specifically initialized via -c: 1062 * Fill in the ones not specifically initialized via -c:
@@ -1471,26 +1067,28 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
1471 pos->attr.sample_period = default_interval; 1067 pos->attr.sample_period = default_interval;
1472 } 1068 }
1473 1069
1474 sym_evsel = list_entry(evsel_list.next, struct perf_evsel, node); 1070 if (perf_evlist__alloc_pollfd(top.evlist) < 0 ||
1071 perf_evlist__alloc_mmap(top.evlist) < 0)
1072 goto out_free_fd;
1073
1074 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
1475 1075
1476 symbol_conf.priv_size = (sizeof(struct sym_entry) + 1076 symbol_conf.priv_size = (sizeof(struct sym_entry) + sizeof(struct annotation) +
1477 (nr_counters + 1) * sizeof(unsigned long)); 1077 (top.evlist->nr_entries + 1) * sizeof(unsigned long));
1478 1078
1479 symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL); 1079 symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1480 if (symbol__init() < 0) 1080 if (symbol__init() < 0)
1481 return -1; 1081 return -1;
1482 1082
1483 get_term_dimensions(&winsize); 1083 get_term_dimensions(&winsize);
1484 if (print_entries == 0) { 1084 if (top.print_entries == 0) {
1485 update_print_entries(&winsize); 1085 update_print_entries(&winsize);
1486 signal(SIGWINCH, sig_winch_handler); 1086 signal(SIGWINCH, sig_winch_handler);
1487 } 1087 }
1488 1088
1489 status = __cmd_top(); 1089 status = __cmd_top();
1490out_free_fd: 1090out_free_fd:
1491 list_for_each_entry(pos, &evsel_list, node) 1091 perf_evlist__delete(top.evlist);
1492 perf_evsel__free_mmap(pos);
1493 perf_evsel_list__delete();
1494 1092
1495 return status; 1093 return status;
1496} 1094}
generated by cgit v1.2.2 (git 2.25.0) at 2025-04-25 08:22:19 -0400