~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/perf/builtin-kvm.c

Version: ~ [ linux-4.16-rc2 ] ~ [ linux-4.15.4 ] ~ [ linux-4.14.20 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.82 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.116 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.49 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.95 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.54 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.99 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 #include "builtin.h"
  3 #include "perf.h"
  4 
  5 #include "util/evsel.h"
  6 #include "util/evlist.h"
  7 #include "util/term.h"
  8 #include "util/util.h"
  9 #include "util/cache.h"
 10 #include "util/symbol.h"
 11 #include "util/thread.h"
 12 #include "util/header.h"
 13 #include "util/session.h"
 14 #include "util/intlist.h"
 15 #include <subcmd/parse-options.h>
 16 #include "util/trace-event.h"
 17 #include "util/debug.h"
 18 #include "util/tool.h"
 19 #include "util/stat.h"
 20 #include "util/top.h"
 21 #include "util/data.h"
 22 #include "util/ordered-events.h"
 23 
 24 #include <sys/prctl.h>
 25 #ifdef HAVE_TIMERFD_SUPPORT
 26 #include <sys/timerfd.h>
 27 #endif
 28 #include <sys/time.h>
 29 #include <sys/types.h>
 30 #include <sys/stat.h>
 31 #include <fcntl.h>
 32 
 33 #include <linux/kernel.h>
 34 #include <linux/time64.h>
 35 #include <errno.h>
 36 #include <inttypes.h>
 37 #include <poll.h>
 38 #include <termios.h>
 39 #include <semaphore.h>
 40 #include <signal.h>
 41 #include <math.h>
 42 
 43 static const char *get_filename_for_perf_kvm(void)
 44 {
 45         const char *filename;
 46 
 47         if (perf_host && !perf_guest)
 48                 filename = strdup("perf.data.host");
 49         else if (!perf_host && perf_guest)
 50                 filename = strdup("perf.data.guest");
 51         else
 52                 filename = strdup("perf.data.kvm");
 53 
 54         return filename;
 55 }
 56 
 57 #ifdef HAVE_KVM_STAT_SUPPORT
 58 #include "util/kvm-stat.h"
 59 
 60 void exit_event_get_key(struct perf_evsel *evsel,
 61                         struct perf_sample *sample,
 62                         struct event_key *key)
 63 {
 64         key->info = 0;
 65         key->key = perf_evsel__intval(evsel, sample, kvm_exit_reason);
 66 }
 67 
 68 bool kvm_exit_event(struct perf_evsel *evsel)
 69 {
 70         return !strcmp(evsel->name, kvm_exit_trace);
 71 }
 72 
 73 bool exit_event_begin(struct perf_evsel *evsel,
 74                       struct perf_sample *sample, struct event_key *key)
 75 {
 76         if (kvm_exit_event(evsel)) {
 77                 exit_event_get_key(evsel, sample, key);
 78                 return true;
 79         }
 80 
 81         return false;
 82 }
 83 
 84 bool kvm_entry_event(struct perf_evsel *evsel)
 85 {
 86         return !strcmp(evsel->name, kvm_entry_trace);
 87 }
 88 
 89 bool exit_event_end(struct perf_evsel *evsel,
 90                     struct perf_sample *sample __maybe_unused,
 91                     struct event_key *key __maybe_unused)
 92 {
 93         return kvm_entry_event(evsel);
 94 }
 95 
 96 static const char *get_exit_reason(struct perf_kvm_stat *kvm,
 97                                    struct exit_reasons_table *tbl,
 98                                    u64 exit_code)
 99 {
100         while (tbl->reason != NULL) {
101                 if (tbl->exit_code == exit_code)
102                         return tbl->reason;
103                 tbl++;
104         }
105 
106         pr_err("unknown kvm exit code:%lld on %s\n",
107                 (unsigned long long)exit_code, kvm->exit_reasons_isa);
108         return "UNKNOWN";
109 }
110 
111 void exit_event_decode_key(struct perf_kvm_stat *kvm,
112                            struct event_key *key,
113                            char *decode)
114 {
115         const char *exit_reason = get_exit_reason(kvm, key->exit_reasons,
116                                                   key->key);
117 
118         scnprintf(decode, decode_str_len, "%s", exit_reason);
119 }
120 
121 static bool register_kvm_events_ops(struct perf_kvm_stat *kvm)
122 {
123         struct kvm_reg_events_ops *events_ops = kvm_reg_events_ops;
124 
125         for (events_ops = kvm_reg_events_ops; events_ops->name; events_ops++) {
126                 if (!strcmp(events_ops->name, kvm->report_event)) {
127                         kvm->events_ops = events_ops->ops;
128                         return true;
129                 }
130         }
131 
132         return false;
133 }
134 
135 struct vcpu_event_record {
136         int vcpu_id;
137         u64 start_time;
138         struct kvm_event *last_event;
139 };
140 
141 
142 static void init_kvm_event_record(struct perf_kvm_stat *kvm)
143 {
144         unsigned int i;
145 
146         for (i = 0; i < EVENTS_CACHE_SIZE; i++)
147                 INIT_LIST_HEAD(&kvm->kvm_events_cache[i]);
148 }
149 
150 #ifdef HAVE_TIMERFD_SUPPORT
151 static void clear_events_cache_stats(struct list_head *kvm_events_cache)
152 {
153         struct list_head *head;
154         struct kvm_event *event;
155         unsigned int i;
156         int j;
157 
158         for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
159                 head = &kvm_events_cache[i];
160                 list_for_each_entry(event, head, hash_entry) {
161                         /* reset stats for event */
162                         event->total.time = 0;
163                         init_stats(&event->total.stats);
164 
165                         for (j = 0; j < event->max_vcpu; ++j) {
166                                 event->vcpu[j].time = 0;
167                                 init_stats(&event->vcpu[j].stats);
168                         }
169                 }
170         }
171 }
172 #endif
173 
174 static int kvm_events_hash_fn(u64 key)
175 {
176         return key & (EVENTS_CACHE_SIZE - 1);
177 }
178 
179 static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
180 {
181         int old_max_vcpu = event->max_vcpu;
182         void *prev;
183 
184         if (vcpu_id < event->max_vcpu)
185                 return true;
186 
187         while (event->max_vcpu <= vcpu_id)
188                 event->max_vcpu += DEFAULT_VCPU_NUM;
189 
190         prev = event->vcpu;
191         event->vcpu = realloc(event->vcpu,
192                               event->max_vcpu * sizeof(*event->vcpu));
193         if (!event->vcpu) {
194                 free(prev);
195                 pr_err("Not enough memory\n");
196                 return false;
197         }
198 
199         memset(event->vcpu + old_max_vcpu, 0,
200                (event->max_vcpu - old_max_vcpu) * sizeof(*event->vcpu));
201         return true;
202 }
203 
204 static struct kvm_event *kvm_alloc_init_event(struct event_key *key)
205 {
206         struct kvm_event *event;
207 
208         event = zalloc(sizeof(*event));
209         if (!event) {
210                 pr_err("Not enough memory\n");
211                 return NULL;
212         }
213 
214         event->key = *key;
215         init_stats(&event->total.stats);
216         return event;
217 }
218 
219 static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
220                                                struct event_key *key)
221 {
222         struct kvm_event *event;
223         struct list_head *head;
224 
225         BUG_ON(key->key == INVALID_KEY);
226 
227         head = &kvm->kvm_events_cache[kvm_events_hash_fn(key->key)];
228         list_for_each_entry(event, head, hash_entry) {
229                 if (event->key.key == key->key && event->key.info == key->info)
230                         return event;
231         }
232 
233         event = kvm_alloc_init_event(key);
234         if (!event)
235                 return NULL;
236 
237         list_add(&event->hash_entry, head);
238         return event;
239 }
240 
241 static bool handle_begin_event(struct perf_kvm_stat *kvm,
242                                struct vcpu_event_record *vcpu_record,
243                                struct event_key *key, u64 timestamp)
244 {
245         struct kvm_event *event = NULL;
246 
247         if (key->key != INVALID_KEY)
248                 event = find_create_kvm_event(kvm, key);
249 
250         vcpu_record->last_event = event;
251         vcpu_record->start_time = timestamp;
252         return true;
253 }
254 
255 static void
256 kvm_update_event_stats(struct kvm_event_stats *kvm_stats, u64 time_diff)
257 {
258         kvm_stats->time += time_diff;
259         update_stats(&kvm_stats->stats, time_diff);
260 }
261 
262 static double kvm_event_rel_stddev(int vcpu_id, struct kvm_event *event)
263 {
264         struct kvm_event_stats *kvm_stats = &event->total;
265 
266         if (vcpu_id != -1)
267                 kvm_stats = &event->vcpu[vcpu_id];
268 
269         return rel_stddev_stats(stddev_stats(&kvm_stats->stats),
270                                 avg_stats(&kvm_stats->stats));
271 }
272 
273 static bool update_kvm_event(struct kvm_event *event, int vcpu_id,
274                              u64 time_diff)
275 {
276         if (vcpu_id == -1) {
277                 kvm_update_event_stats(&event->total, time_diff);
278                 return true;
279         }
280 
281         if (!kvm_event_expand(event, vcpu_id))
282                 return false;
283 
284         kvm_update_event_stats(&event->vcpu[vcpu_id], time_diff);
285         return true;
286 }
287 
288 static bool is_child_event(struct perf_kvm_stat *kvm,
289                            struct perf_evsel *evsel,
290                            struct perf_sample *sample,
291                            struct event_key *key)
292 {
293         struct child_event_ops *child_ops;
294 
295         child_ops = kvm->events_ops->child_ops;
296 
297         if (!child_ops)
298                 return false;
299 
300         for (; child_ops->name; child_ops++) {
301                 if (!strcmp(evsel->name, child_ops->name)) {
302                         child_ops->get_key(evsel, sample, key);
303                         return true;
304                 }
305         }
306 
307         return false;
308 }
309 
310 static bool handle_child_event(struct perf_kvm_stat *kvm,
311                                struct vcpu_event_record *vcpu_record,
312                                struct event_key *key,
313                                struct perf_sample *sample __maybe_unused)
314 {
315         struct kvm_event *event = NULL;
316 
317         if (key->key != INVALID_KEY)
318                 event = find_create_kvm_event(kvm, key);
319 
320         vcpu_record->last_event = event;
321 
322         return true;
323 }
324 
325 static bool skip_event(const char *event)
326 {
327         const char * const *skip_events;
328 
329         for (skip_events = kvm_skip_events; *skip_events; skip_events++)
330                 if (!strcmp(event, *skip_events))
331                         return true;
332 
333         return false;
334 }
335 
336 static bool handle_end_event(struct perf_kvm_stat *kvm,
337                              struct vcpu_event_record *vcpu_record,
338                              struct event_key *key,
339                              struct perf_sample *sample)
340 {
341         struct kvm_event *event;
342         u64 time_begin, time_diff;
343         int vcpu;
344 
345         if (kvm->trace_vcpu == -1)
346                 vcpu = -1;
347         else
348                 vcpu = vcpu_record->vcpu_id;
349 
350         event = vcpu_record->last_event;
351         time_begin = vcpu_record->start_time;
352 
353         /* The begin event is not caught. */
354         if (!time_begin)
355                 return true;
356 
357         /*
358          * In some case, the 'begin event' only records the start timestamp,
359          * the actual event is recognized in the 'end event' (e.g. mmio-event).
360          */
361 
362         /* Both begin and end events did not get the key. */
363         if (!event && key->key == INVALID_KEY)
364                 return true;
365 
366         if (!event)
367                 event = find_create_kvm_event(kvm, key);
368 
369         if (!event)
370                 return false;
371 
372         vcpu_record->last_event = NULL;
373         vcpu_record->start_time = 0;
374 
375         /* seems to happen once in a while during live mode */
376         if (sample->time < time_begin) {
377                 pr_debug("End time before begin time; skipping event.\n");
378                 return true;
379         }
380 
381         time_diff = sample->time - time_begin;
382 
383         if (kvm->duration && time_diff > kvm->duration) {
384                 char decode[decode_str_len];
385 
386                 kvm->events_ops->decode_key(kvm, &event->key, decode);
387                 if (!skip_event(decode)) {
388                         pr_info("%" PRIu64 " VM %d, vcpu %d: %s event took %" PRIu64 "usec\n",
389                                  sample->time, sample->pid, vcpu_record->vcpu_id,
390                                  decode, time_diff / NSEC_PER_USEC);
391                 }
392         }
393 
394         return update_kvm_event(event, vcpu, time_diff);
395 }
396 
397 static
398 struct vcpu_event_record *per_vcpu_record(struct thread *thread,
399                                           struct perf_evsel *evsel,
400                                           struct perf_sample *sample)
401 {
402         /* Only kvm_entry records vcpu id. */
403         if (!thread__priv(thread) && kvm_entry_event(evsel)) {
404                 struct vcpu_event_record *vcpu_record;
405 
406                 vcpu_record = zalloc(sizeof(*vcpu_record));
407                 if (!vcpu_record) {
408                         pr_err("%s: Not enough memory\n", __func__);
409                         return NULL;
410                 }
411 
412                 vcpu_record->vcpu_id = perf_evsel__intval(evsel, sample,
413                                                           vcpu_id_str);
414                 thread__set_priv(thread, vcpu_record);
415         }
416 
417         return thread__priv(thread);
418 }
419 
420 static bool handle_kvm_event(struct perf_kvm_stat *kvm,
421                              struct thread *thread,
422                              struct perf_evsel *evsel,
423                              struct perf_sample *sample)
424 {
425         struct vcpu_event_record *vcpu_record;
426         struct event_key key = { .key = INVALID_KEY,
427                                  .exit_reasons = kvm->exit_reasons };
428 
429         vcpu_record = per_vcpu_record(thread, evsel, sample);
430         if (!vcpu_record)
431                 return true;
432 
433         /* only process events for vcpus user cares about */
434         if ((kvm->trace_vcpu != -1) &&
435             (kvm->trace_vcpu != vcpu_record->vcpu_id))
436                 return true;
437 
438         if (kvm->events_ops->is_begin_event(evsel, sample, &key))
439                 return handle_begin_event(kvm, vcpu_record, &key, sample->time);
440 
441         if (is_child_event(kvm, evsel, sample, &key))
442                 return handle_child_event(kvm, vcpu_record, &key, sample);
443 
444         if (kvm->events_ops->is_end_event(evsel, sample, &key))
445                 return handle_end_event(kvm, vcpu_record, &key, sample);
446 
447         return true;
448 }
449 
450 #define GET_EVENT_KEY(func, field)                                      \
451 static u64 get_event_ ##func(struct kvm_event *event, int vcpu)         \
452 {                                                                       \
453         if (vcpu == -1)                                                 \
454                 return event->total.field;                              \
455                                                                         \
456         if (vcpu >= event->max_vcpu)                                    \
457                 return 0;                                               \
458                                                                         \
459         return event->vcpu[vcpu].field;                                 \
460 }
461 
462 #define COMPARE_EVENT_KEY(func, field)                                  \
463 GET_EVENT_KEY(func, field)                                              \
464 static int compare_kvm_event_ ## func(struct kvm_event *one,            \
465                                         struct kvm_event *two, int vcpu)\
466 {                                                                       \
467         return get_event_ ##func(one, vcpu) >                           \
468                                 get_event_ ##func(two, vcpu);           \
469 }
470 
471 GET_EVENT_KEY(time, time);
472 COMPARE_EVENT_KEY(count, stats.n);
473 COMPARE_EVENT_KEY(mean, stats.mean);
474 GET_EVENT_KEY(max, stats.max);
475 GET_EVENT_KEY(min, stats.min);
476 
477 #define DEF_SORT_NAME_KEY(name, compare_key)                            \
478         { #name, compare_kvm_event_ ## compare_key }
479 
480 static struct kvm_event_key keys[] = {
481         DEF_SORT_NAME_KEY(sample, count),
482         DEF_SORT_NAME_KEY(time, mean),
483         { NULL, NULL }
484 };
485 
486 static bool select_key(struct perf_kvm_stat *kvm)
487 {
488         int i;
489 
490         for (i = 0; keys[i].name; i++) {
491                 if (!strcmp(keys[i].name, kvm->sort_key)) {
492                         kvm->compare = keys[i].key;
493                         return true;
494                 }
495         }
496 
497         pr_err("Unknown compare key:%s\n", kvm->sort_key);
498         return false;
499 }
500 
501 static void insert_to_result(struct rb_root *result, struct kvm_event *event,
502                              key_cmp_fun bigger, int vcpu)
503 {
504         struct rb_node **rb = &result->rb_node;
505         struct rb_node *parent = NULL;
506         struct kvm_event *p;
507 
508         while (*rb) {
509                 p = container_of(*rb, struct kvm_event, rb);
510                 parent = *rb;
511 
512                 if (bigger(event, p, vcpu))
513                         rb = &(*rb)->rb_left;
514                 else
515                         rb = &(*rb)->rb_right;
516         }
517 
518         rb_link_node(&event->rb, parent, rb);
519         rb_insert_color(&event->rb, result);
520 }
521 
522 static void
523 update_total_count(struct perf_kvm_stat *kvm, struct kvm_event *event)
524 {
525         int vcpu = kvm->trace_vcpu;
526 
527         kvm->total_count += get_event_count(event, vcpu);
528         kvm->total_time += get_event_time(event, vcpu);
529 }
530 
531 static bool event_is_valid(struct kvm_event *event, int vcpu)
532 {
533         return !!get_event_count(event, vcpu);
534 }
535 
536 static void sort_result(struct perf_kvm_stat *kvm)
537 {
538         unsigned int i;
539         int vcpu = kvm->trace_vcpu;
540         struct kvm_event *event;
541 
542         for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
543                 list_for_each_entry(event, &kvm->kvm_events_cache[i], hash_entry) {
544                         if (event_is_valid(event, vcpu)) {
545                                 update_total_count(kvm, event);
546                                 insert_to_result(&kvm->result, event,
547                                                  kvm->compare, vcpu);
548                         }
549                 }
550         }
551 }
552 
553 /* returns left most element of result, and erase it */
554 static struct kvm_event *pop_from_result(struct rb_root *result)
555 {
556         struct rb_node *node = rb_first(result);
557 
558         if (!node)
559                 return NULL;
560 
561         rb_erase(node, result);
562         return container_of(node, struct kvm_event, rb);
563 }
564 
565 static void print_vcpu_info(struct perf_kvm_stat *kvm)
566 {
567         int vcpu = kvm->trace_vcpu;
568 
569         pr_info("Analyze events for ");
570 
571         if (kvm->opts.target.system_wide)
572                 pr_info("all VMs, ");
573         else if (kvm->opts.target.pid)
574                 pr_info("pid(s) %s, ", kvm->opts.target.pid);
575         else
576                 pr_info("dazed and confused on what is monitored, ");
577 
578         if (vcpu == -1)
579                 pr_info("all VCPUs:\n\n");
580         else
581                 pr_info("VCPU %d:\n\n", vcpu);
582 }
583 
584 static void show_timeofday(void)
585 {
586         char date[64];
587         struct timeval tv;
588         struct tm ltime;
589 
590         gettimeofday(&tv, NULL);
591         if (localtime_r(&tv.tv_sec, &ltime)) {
592                 strftime(date, sizeof(date), "%H:%M:%S", &ltime);
593                 pr_info("%s.%06ld", date, tv.tv_usec);
594         } else
595                 pr_info("00:00:00.000000");
596 
597         return;
598 }
599 
600 static void print_result(struct perf_kvm_stat *kvm)
601 {
602         char decode[decode_str_len];
603         struct kvm_event *event;
604         int vcpu = kvm->trace_vcpu;
605 
606         if (kvm->live) {
607                 puts(CONSOLE_CLEAR);
608                 show_timeofday();
609         }
610 
611         pr_info("\n\n");
612         print_vcpu_info(kvm);
613         pr_info("%*s ", decode_str_len, kvm->events_ops->name);
614         pr_info("%10s ", "Samples");
615         pr_info("%9s ", "Samples%");
616 
617         pr_info("%9s ", "Time%");
618         pr_info("%11s ", "Min Time");
619         pr_info("%11s ", "Max Time");
620         pr_info("%16s ", "Avg time");
621         pr_info("\n\n");
622 
623         while ((event = pop_from_result(&kvm->result))) {
624                 u64 ecount, etime, max, min;
625 
626                 ecount = get_event_count(event, vcpu);
627                 etime = get_event_time(event, vcpu);
628                 max = get_event_max(event, vcpu);
629                 min = get_event_min(event, vcpu);
630 
631                 kvm->events_ops->decode_key(kvm, &event->key, decode);
632                 pr_info("%*s ", decode_str_len, decode);
633                 pr_info("%10llu ", (unsigned long long)ecount);
634                 pr_info("%8.2f%% ", (double)ecount / kvm->total_count * 100);
635                 pr_info("%8.2f%% ", (double)etime / kvm->total_time * 100);
636                 pr_info("%9.2fus ", (double)min / NSEC_PER_USEC);
637                 pr_info("%9.2fus ", (double)max / NSEC_PER_USEC);
638                 pr_info("%9.2fus ( +-%7.2f%% )", (double)etime / ecount / NSEC_PER_USEC,
639                         kvm_event_rel_stddev(vcpu, event));
640                 pr_info("\n");
641         }
642 
643         pr_info("\nTotal Samples:%" PRIu64 ", Total events handled time:%.2fus.\n\n",
644                 kvm->total_count, kvm->total_time / (double)NSEC_PER_USEC);
645 
646         if (kvm->lost_events)
647                 pr_info("\nLost events: %" PRIu64 "\n\n", kvm->lost_events);
648 }
649 
650 #ifdef HAVE_TIMERFD_SUPPORT
651 static int process_lost_event(struct perf_tool *tool,
652                               union perf_event *event __maybe_unused,
653                               struct perf_sample *sample __maybe_unused,
654                               struct machine *machine __maybe_unused)
655 {
656         struct perf_kvm_stat *kvm = container_of(tool, struct perf_kvm_stat, tool);
657 
658         kvm->lost_events++;
659         return 0;
660 }
661 #endif
662 
663 static bool skip_sample(struct perf_kvm_stat *kvm,
664                         struct perf_sample *sample)
665 {
666         if (kvm->pid_list && intlist__find(kvm->pid_list, sample->pid) == NULL)
667                 return true;
668 
669         return false;
670 }
671 
672 static int process_sample_event(struct perf_tool *tool,
673                                 union perf_event *event,
674                                 struct perf_sample *sample,
675                                 struct perf_evsel *evsel,
676                                 struct machine *machine)
677 {
678         int err = 0;
679         struct thread *thread;
680         struct perf_kvm_stat *kvm = container_of(tool, struct perf_kvm_stat,
681                                                  tool);
682 
683         if (skip_sample(kvm, sample))
684                 return 0;
685 
686         thread = machine__findnew_thread(machine, sample->pid, sample->tid);
687         if (thread == NULL) {
688                 pr_debug("problem processing %d event, skipping it.\n",
689                         event->header.type);
690                 return -1;
691         }
692 
693         if (!handle_kvm_event(kvm, thread, evsel, sample))
694                 err = -1;
695 
696         thread__put(thread);
697         return err;
698 }
699 
700 static int cpu_isa_config(struct perf_kvm_stat *kvm)
701 {
702         char buf[64], *cpuid;
703         int err;
704 
705         if (kvm->live) {
706                 err = get_cpuid(buf, sizeof(buf));
707                 if (err != 0) {
708                         pr_err("Failed to look up CPU type\n");
709                         return err;
710                 }
711                 cpuid = buf;
712         } else
713                 cpuid = kvm->session->header.env.cpuid;
714 
715         if (!cpuid) {
716                 pr_err("Failed to look up CPU type\n");
717                 return -EINVAL;
718         }
719 
720         err = cpu_isa_init(kvm, cpuid);
721         if (err == -ENOTSUP)
722                 pr_err("CPU %s is not supported.\n", cpuid);
723 
724         return err;
725 }
726 
727 static bool verify_vcpu(int vcpu)
728 {
729         if (vcpu != -1 && vcpu < 0) {
730                 pr_err("Invalid vcpu:%d.\n", vcpu);
731                 return false;
732         }
733 
734         return true;
735 }
736 
737 #ifdef HAVE_TIMERFD_SUPPORT
738 /* keeping the max events to a modest level to keep
739  * the processing of samples per mmap smooth.
740  */
741 #define PERF_KVM__MAX_EVENTS_PER_MMAP  25
742 
743 static s64 perf_kvm__mmap_read_idx(struct perf_kvm_stat *kvm, int idx,
744                                    u64 *mmap_time)
745 {
746         union perf_event *event;
747         u64 timestamp;
748         s64 n = 0;
749         int err;
750 
751         *mmap_time = ULLONG_MAX;
752         while ((event = perf_evlist__mmap_read(kvm->evlist, idx)) != NULL) {
753                 err = perf_evlist__parse_sample_timestamp(kvm->evlist, event, &timestamp);
754                 if (err) {
755                         perf_evlist__mmap_consume(kvm->evlist, idx);
756                         pr_err("Failed to parse sample\n");
757                         return -1;
758                 }
759 
760                 err = perf_session__queue_event(kvm->session, event, timestamp, 0);
761                 /*
762                  * FIXME: Here we can't consume the event, as perf_session__queue_event will
763                  *        point to it, and it'll get possibly overwritten by the kernel.
764                  */
765                 perf_evlist__mmap_consume(kvm->evlist, idx);
766 
767                 if (err) {
768                         pr_err("Failed to enqueue sample: %d\n", err);
769                         return -1;
770                 }
771 
772                 /* save time stamp of our first sample for this mmap */
773                 if (n == 0)
774                         *mmap_time = timestamp;
775 
776                 /* limit events per mmap handled all at once */
777                 n++;
778                 if (n == PERF_KVM__MAX_EVENTS_PER_MMAP)
779                         break;
780         }
781 
782         return n;
783 }
784 
785 static int perf_kvm__mmap_read(struct perf_kvm_stat *kvm)
786 {
787         int i, err, throttled = 0;
788         s64 n, ntotal = 0;
789         u64 flush_time = ULLONG_MAX, mmap_time;
790 
791         for (i = 0; i < kvm->evlist->nr_mmaps; i++) {
792                 n = perf_kvm__mmap_read_idx(kvm, i, &mmap_time);
793                 if (n < 0)
794                         return -1;
795 
796                 /* flush time is going to be the minimum of all the individual
797                  * mmap times. Essentially, we flush all the samples queued up
798                  * from the last pass under our minimal start time -- that leaves
799                  * a very small race for samples to come in with a lower timestamp.
800                  * The ioctl to return the perf_clock timestamp should close the
801                  * race entirely.
802                  */
803                 if (mmap_time < flush_time)
804                         flush_time = mmap_time;
805 
806                 ntotal += n;
807                 if (n == PERF_KVM__MAX_EVENTS_PER_MMAP)
808                         throttled = 1;
809         }
810 
811         /* flush queue after each round in which we processed events */
812         if (ntotal) {
813                 struct ordered_events *oe = &kvm->session->ordered_events;
814 
815                 oe->next_flush = flush_time;
816                 err = ordered_events__flush(oe, OE_FLUSH__ROUND);
817                 if (err) {
818                         if (kvm->lost_events)
819                                 pr_info("\nLost events: %" PRIu64 "\n\n",
820                                         kvm->lost_events);
821                         return err;
822                 }
823         }
824 
825         return throttled;
826 }
827 
828 static volatile int done;
829 
830 static void sig_handler(int sig __maybe_unused)
831 {
832         done = 1;
833 }
834 
835 static int perf_kvm__timerfd_create(struct perf_kvm_stat *kvm)
836 {
837         struct itimerspec new_value;
838         int rc = -1;
839 
840         kvm->timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK);
841         if (kvm->timerfd < 0) {
842                 pr_err("timerfd_create failed\n");
843                 goto out;
844         }
845 
846         new_value.it_value.tv_sec = kvm->display_time;
847         new_value.it_value.tv_nsec = 0;
848         new_value.it_interval.tv_sec = kvm->display_time;
849         new_value.it_interval.tv_nsec = 0;
850 
851         if (timerfd_settime(kvm->timerfd, 0, &new_value, NULL) != 0) {
852                 pr_err("timerfd_settime failed: %d\n", errno);
853                 close(kvm->timerfd);
854                 goto out;
855         }
856 
857         rc = 0;
858 out:
859         return rc;
860 }
861 
862 static int perf_kvm__handle_timerfd(struct perf_kvm_stat *kvm)
863 {
864         uint64_t c;
865         int rc;
866 
867         rc = read(kvm->timerfd, &c, sizeof(uint64_t));
868         if (rc < 0) {
869                 if (errno == EAGAIN)
870                         return 0;
871 
872                 pr_err("Failed to read timer fd: %d\n", errno);
873                 return -1;
874         }
875 
876         if (rc != sizeof(uint64_t)) {
877                 pr_err("Error reading timer fd - invalid size returned\n");
878                 return -1;
879         }
880 
881         if (c != 1)
882                 pr_debug("Missed timer beats: %" PRIu64 "\n", c-1);
883 
884         /* update display */
885         sort_result(kvm);
886         print_result(kvm);
887 
888         /* reset counts */
889         clear_events_cache_stats(kvm->kvm_events_cache);
890         kvm->total_count = 0;
891         kvm->total_time = 0;
892         kvm->lost_events = 0;
893 
894         return 0;
895 }
896 
897 static int fd_set_nonblock(int fd)
898 {
899         long arg = 0;
900 
901         arg = fcntl(fd, F_GETFL);
902         if (arg < 0) {
903                 pr_err("Failed to get current flags for fd %d\n", fd);
904                 return -1;
905         }
906 
907         if (fcntl(fd, F_SETFL, arg | O_NONBLOCK) < 0) {
908                 pr_err("Failed to set non-block option on fd %d\n", fd);
909                 return -1;
910         }
911 
912         return 0;
913 }
914 
915 static int perf_kvm__handle_stdin(void)
916 {
917         int c;
918 
919         c = getc(stdin);
920         if (c == 'q')
921                 return 1;
922 
923         return 0;
924 }
925 
926 static int kvm_events_live_report(struct perf_kvm_stat *kvm)
927 {
928         int nr_stdin, ret, err = -EINVAL;
929         struct termios save;
930 
931         /* live flag must be set first */
932         kvm->live = true;
933 
934         ret = cpu_isa_config(kvm);
935         if (ret < 0)
936                 return ret;
937 
938         if (!verify_vcpu(kvm->trace_vcpu) ||
939             !select_key(kvm) ||
940             !register_kvm_events_ops(kvm)) {
941                 goto out;
942         }
943 
944         set_term_quiet_input(&save);
945         init_kvm_event_record(kvm);
946 
947         signal(SIGINT, sig_handler);
948         signal(SIGTERM, sig_handler);
949 
950         /* add timer fd */
951         if (perf_kvm__timerfd_create(kvm) < 0) {
952                 err = -1;
953                 goto out;
954         }
955 
956         if (perf_evlist__add_pollfd(kvm->evlist, kvm->timerfd) < 0)
957                 goto out;
958 
959         nr_stdin = perf_evlist__add_pollfd(kvm->evlist, fileno(stdin));
960         if (nr_stdin < 0)
961                 goto out;
962 
963         if (fd_set_nonblock(fileno(stdin)) != 0)
964                 goto out;
965 
966         /* everything is good - enable the events and process */
967         perf_evlist__enable(kvm->evlist);
968 
969         while (!done) {
970                 struct fdarray *fda = &kvm->evlist->pollfd;
971                 int rc;
972 
973                 rc = perf_kvm__mmap_read(kvm);
974                 if (rc < 0)
975                         break;
976 
977                 err = perf_kvm__handle_timerfd(kvm);
978                 if (err)
979                         goto out;
980 
981                 if (fda->entries[nr_stdin].revents & POLLIN)
982                         done = perf_kvm__handle_stdin();
983 
984                 if (!rc && !done)
985                         err = fdarray__poll(fda, 100);
986         }
987 
988         perf_evlist__disable(kvm->evlist);
989 
990         if (err == 0) {
991                 sort_result(kvm);
992                 print_result(kvm);
993         }
994 
995 out:
996         if (kvm->timerfd >= 0)
997                 close(kvm->timerfd);
998 
999         tcsetattr(0, TCSAFLUSH, &save);
1000         return err;
1001 }
1002 
1003 static int kvm_live_open_events(struct perf_kvm_stat *kvm)
1004 {
1005         int err, rc = -1;
1006         struct perf_evsel *pos;
1007         struct perf_evlist *evlist = kvm->evlist;
1008         char sbuf[STRERR_BUFSIZE];
1009 
1010         perf_evlist__config(evlist, &kvm->opts, NULL);
1011 
1012         /*
1013          * Note: exclude_{guest,host} do not apply here.
1014          *       This command processes KVM tracepoints from host only
1015          */
1016         evlist__for_each_entry(evlist, pos) {
1017                 struct perf_event_attr *attr = &pos->attr;
1018 
1019                 /* make sure these *are* set */
1020                 perf_evsel__set_sample_bit(pos, TID);
1021                 perf_evsel__set_sample_bit(pos, TIME);
1022                 perf_evsel__set_sample_bit(pos, CPU);
1023                 perf_evsel__set_sample_bit(pos, RAW);
1024                 /* make sure these are *not*; want as small a sample as possible */
1025                 perf_evsel__reset_sample_bit(pos, PERIOD);
1026                 perf_evsel__reset_sample_bit(pos, IP);
1027                 perf_evsel__reset_sample_bit(pos, CALLCHAIN);
1028                 perf_evsel__reset_sample_bit(pos, ADDR);
1029                 perf_evsel__reset_sample_bit(pos, READ);
1030                 attr->mmap = 0;
1031                 attr->comm = 0;
1032                 attr->task = 0;
1033 
1034                 attr->sample_period = 1;
1035 
1036                 attr->watermark = 0;
1037                 attr->wakeup_events = 1000;
1038 
1039                 /* will enable all once we are ready */
1040                 attr->disabled = 1;
1041         }
1042 
1043         err = perf_evlist__open(evlist);
1044         if (err < 0) {
1045                 printf("Couldn't create the events: %s\n",
1046                        str_error_r(errno, sbuf, sizeof(sbuf)));
1047                 goto out;
1048         }
1049 
1050         if (perf_evlist__mmap(evlist, kvm->opts.mmap_pages) < 0) {
1051                 ui__error("Failed to mmap the events: %s\n",
1052                           str_error_r(errno, sbuf, sizeof(sbuf)));
1053                 perf_evlist__close(evlist);
1054                 goto out;
1055         }
1056 
1057         rc = 0;
1058 
1059 out:
1060         return rc;
1061 }
1062 #endif
1063 
1064 static int read_events(struct perf_kvm_stat *kvm)
1065 {
1066         int ret;
1067 
1068         struct perf_tool eops = {
1069                 .sample                 = process_sample_event,
1070                 .comm                   = perf_event__process_comm,
1071                 .namespaces             = perf_event__process_namespaces,
1072                 .ordered_events         = true,
1073         };
1074         struct perf_data file = {
1075                 .file      = {
1076                         .path = kvm->file_name,
1077                 },
1078                 .mode      = PERF_DATA_MODE_READ,
1079                 .force     = kvm->force,
1080         };
1081 
1082         kvm->tool = eops;
1083         kvm->session = perf_session__new(&file, false, &kvm->tool);
1084         if (!kvm->session) {
1085                 pr_err("Initializing perf session failed\n");
1086                 return -1;
1087         }
1088 
1089         symbol__init(&kvm->session->header.env);
1090 
1091         if (!perf_session__has_traces(kvm->session, "kvm record")) {
1092                 ret = -EINVAL;
1093                 goto out_delete;
1094         }
1095 
1096         /*
1097          * Do not use 'isa' recorded in kvm_exit tracepoint since it is not
1098          * traced in the old kernel.
1099          */
1100         ret = cpu_isa_config(kvm);
1101         if (ret < 0)
1102                 goto out_delete;
1103 
1104         ret = perf_session__process_events(kvm->session);
1105 
1106 out_delete:
1107         perf_session__delete(kvm->session);
1108         return ret;
1109 }
1110 
1111 static int parse_target_str(struct perf_kvm_stat *kvm)
1112 {
1113         if (kvm->opts.target.pid) {
1114                 kvm->pid_list = intlist__new(kvm->opts.target.pid);
1115                 if (kvm->pid_list == NULL) {
1116                         pr_err("Error parsing process id string\n");
1117                         return -EINVAL;
1118                 }
1119         }
1120 
1121         return 0;
1122 }
1123 
1124 static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
1125 {
1126         int ret = -EINVAL;
1127         int vcpu = kvm->trace_vcpu;
1128 
1129         if (parse_target_str(kvm) != 0)
1130                 goto exit;
1131 
1132         if (!verify_vcpu(vcpu))
1133                 goto exit;
1134 
1135         if (!select_key(kvm))
1136                 goto exit;
1137 
1138         if (!register_kvm_events_ops(kvm))
1139                 goto exit;
1140 
1141         init_kvm_event_record(kvm);
1142         setup_pager();
1143 
1144         ret = read_events(kvm);
1145         if (ret)
1146                 goto exit;
1147 
1148         sort_result(kvm);
1149         print_result(kvm);
1150 
1151 exit:
1152         return ret;
1153 }
1154 
1155 #define STRDUP_FAIL_EXIT(s)             \
1156         ({      char *_p;               \
1157         _p = strdup(s);         \
1158                 if (!_p)                \
1159                         return -ENOMEM; \
1160                 _p;                     \
1161         })
1162 
1163 int __weak setup_kvm_events_tp(struct perf_kvm_stat *kvm __maybe_unused)
1164 {
1165         return 0;
1166 }
1167 
1168 static int
1169 kvm_events_record(struct perf_kvm_stat *kvm, int argc, const char **argv)
1170 {
1171         unsigned int rec_argc, i, j, events_tp_size;
1172         const char **rec_argv;
1173         const char * const record_args[] = {
1174                 "record",
1175                 "-R",
1176                 "-m", "1024",
1177                 "-c", "1",
1178         };
1179         const char * const kvm_stat_record_usage[] = {
1180                 "perf kvm stat record [<options>]",
1181                 NULL
1182         };
1183         const char * const *events_tp;
1184         int ret;
1185 
1186         events_tp_size = 0;
1187         ret = setup_kvm_events_tp(kvm);
1188         if (ret < 0) {
1189                 pr_err("Unable to setup the kvm tracepoints\n");
1190                 return ret;
1191         }
1192 
1193         for (events_tp = kvm_events_tp; *events_tp; events_tp++)
1194                 events_tp_size++;
1195 
1196         rec_argc = ARRAY_SIZE(record_args) + argc + 2 +
1197                    2 * events_tp_size;
1198         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1199 
1200         if (rec_argv == NULL)
1201                 return -ENOMEM;
1202 
1203         for (i = 0; i < ARRAY_SIZE(record_args); i++)
1204                 rec_argv[i] = STRDUP_FAIL_EXIT(record_args[i]);
1205 
1206         for (j = 0; j < events_tp_size; j++) {
1207                 rec_argv[i++] = "-e";
1208                 rec_argv[i++] = STRDUP_FAIL_EXIT(kvm_events_tp[j]);
1209         }
1210 
1211         rec_argv[i++] = STRDUP_FAIL_EXIT("-o");
1212         rec_argv[i++] = STRDUP_FAIL_EXIT(kvm->file_name);
1213 
1214         for (j = 1; j < (unsigned int)argc; j++, i++)
1215                 rec_argv[i] = argv[j];
1216 
1217         set_option_flag(record_options, 'e', "event", PARSE_OPT_HIDDEN);
1218         set_option_flag(record_options, 0, "filter", PARSE_OPT_HIDDEN);
1219         set_option_flag(record_options, 'R', "raw-samples", PARSE_OPT_HIDDEN);
1220 
1221         set_option_flag(record_options, 'F', "freq", PARSE_OPT_DISABLED);
1222         set_option_flag(record_options, 0, "group", PARSE_OPT_DISABLED);
1223         set_option_flag(record_options, 'g', NULL, PARSE_OPT_DISABLED);
1224         set_option_flag(record_options, 0, "call-graph", PARSE_OPT_DISABLED);
1225         set_option_flag(record_options, 'd', "data", PARSE_OPT_DISABLED);
1226         set_option_flag(record_options, 'T', "timestamp", PARSE_OPT_DISABLED);
1227         set_option_flag(record_options, 'P', "period", PARSE_OPT_DISABLED);
1228         set_option_flag(record_options, 'n', "no-samples", PARSE_OPT_DISABLED);
1229         set_option_flag(record_options, 'N', "no-buildid-cache", PARSE_OPT_DISABLED);
1230         set_option_flag(record_options, 'B', "no-buildid", PARSE_OPT_DISABLED);
1231         set_option_flag(record_options, 'G', "cgroup", PARSE_OPT_DISABLED);
1232         set_option_flag(record_options, 'b', "branch-any", PARSE_OPT_DISABLED);
1233         set_option_flag(record_options, 'j', "branch-filter", PARSE_OPT_DISABLED);
1234         set_option_flag(record_options, 'W', "weight", PARSE_OPT_DISABLED);
1235         set_option_flag(record_options, 0, "transaction", PARSE_OPT_DISABLED);
1236 
1237         record_usage = kvm_stat_record_usage;
1238         return cmd_record(i, rec_argv);
1239 }
1240 
1241 static int
1242 kvm_events_report(struct perf_kvm_stat *kvm, int argc, const char **argv)
1243 {
1244         const struct option kvm_events_report_options[] = {
1245                 OPT_STRING(0, "event", &kvm->report_event, "report event",
1246                            "event for reporting: vmexit, "
1247                            "mmio (x86 only), ioport (x86 only)"),
1248                 OPT_INTEGER(0, "vcpu", &kvm->trace_vcpu,
1249                             "vcpu id to report"),
1250                 OPT_STRING('k', "key", &kvm->sort_key, "sort-key",
1251                             "key for sorting: sample(sort by samples number)"
1252                             " time (sort by avg time)"),
1253                 OPT_STRING('p', "pid", &kvm->opts.target.pid, "pid",
1254                            "analyze events only for given process id(s)"),
1255                 OPT_BOOLEAN('f', "force", &kvm->force, "don't complain, do it"),
1256                 OPT_END()
1257         };
1258 
1259         const char * const kvm_events_report_usage[] = {
1260                 "perf kvm stat report [<options>]",
1261                 NULL
1262         };
1263 
1264         if (argc) {
1265                 argc = parse_options(argc, argv,
1266                                      kvm_events_report_options,
1267                                      kvm_events_report_usage, 0);
1268                 if (argc)
1269                         usage_with_options(kvm_events_report_usage,
1270                                            kvm_events_report_options);
1271         }
1272 
1273         if (!kvm->opts.target.pid)
1274                 kvm->opts.target.system_wide = true;
1275 
1276         return kvm_events_report_vcpu(kvm);
1277 }
1278 
1279 #ifdef HAVE_TIMERFD_SUPPORT
1280 static struct perf_evlist *kvm_live_event_list(void)
1281 {
1282         struct perf_evlist *evlist;
1283         char *tp, *name, *sys;
1284         int err = -1;
1285         const char * const *events_tp;
1286 
1287         evlist = perf_evlist__new();
1288         if (evlist == NULL)
1289                 return NULL;
1290 
1291         for (events_tp = kvm_events_tp; *events_tp; events_tp++) {
1292 
1293                 tp = strdup(*events_tp);
1294                 if (tp == NULL)
1295                         goto out;
1296 
1297                 /* split tracepoint into subsystem and name */
1298                 sys = tp;
1299                 name = strchr(tp, ':');
1300                 if (name == NULL) {
1301                         pr_err("Error parsing %s tracepoint: subsystem delimiter not found\n",
1302                                *events_tp);
1303                         free(tp);
1304                         goto out;
1305                 }
1306                 *name = '\0';
1307                 name++;
1308 
1309                 if (perf_evlist__add_newtp(evlist, sys, name, NULL)) {
1310                         pr_err("Failed to add %s tracepoint to the list\n", *events_tp);
1311                         free(tp);
1312                         goto out;
1313                 }
1314 
1315                 free(tp);
1316         }
1317 
1318         err = 0;
1319 
1320 out:
1321         if (err) {
1322                 perf_evlist__delete(evlist);
1323                 evlist = NULL;
1324         }
1325 
1326         return evlist;
1327 }
1328 
1329 static int kvm_events_live(struct perf_kvm_stat *kvm,
1330                            int argc, const char **argv)
1331 {
1332         char errbuf[BUFSIZ];
1333         int err;
1334 
1335         const struct option live_options[] = {
1336                 OPT_STRING('p', "pid", &kvm->opts.target.pid, "pid",
1337                         "record events on existing process id"),
1338                 OPT_CALLBACK('m', "mmap-pages", &kvm->opts.mmap_pages, "pages",
1339                         "number of mmap data pages",
1340                         perf_evlist__parse_mmap_pages),
1341                 OPT_INCR('v', "verbose", &verbose,
1342                         "be more verbose (show counter open errors, etc)"),
1343                 OPT_BOOLEAN('a', "all-cpus", &kvm->opts.target.system_wide,
1344                         "system-wide collection from all CPUs"),
1345                 OPT_UINTEGER('d', "display", &kvm->display_time,
1346                         "time in seconds between display updates"),
1347                 OPT_STRING(0, "event", &kvm->report_event, "report event",
1348                         "event for reporting: "
1349                         "vmexit, mmio (x86 only), ioport (x86 only)"),
1350                 OPT_INTEGER(0, "vcpu", &kvm->trace_vcpu,
1351                         "vcpu id to report"),
1352                 OPT_STRING('k', "key", &kvm->sort_key, "sort-key",
1353                         "key for sorting: sample(sort by samples number)"
1354                         " time (sort by avg time)"),
1355                 OPT_U64(0, "duration", &kvm->duration,
1356                         "show events other than"
1357                         " HLT (x86 only) or Wait state (s390 only)"
1358                         " that take longer than duration usecs"),
1359                 OPT_UINTEGER(0, "proc-map-timeout", &kvm->opts.proc_map_timeout,
1360                                 "per thread proc mmap processing timeout in ms"),
1361                 OPT_END()
1362         };
1363         const char * const live_usage[] = {
1364                 "perf kvm stat live [<options>]",
1365                 NULL
1366         };
1367         struct perf_data data = {
1368                 .mode = PERF_DATA_MODE_WRITE,
1369         };
1370 
1371 
1372         /* event handling */
1373         kvm->tool.sample = process_sample_event;
1374         kvm->tool.comm   = perf_event__process_comm;
1375         kvm->tool.exit   = perf_event__process_exit;
1376         kvm->tool.fork   = perf_event__process_fork;
1377         kvm->tool.lost   = process_lost_event;
1378         kvm->tool.namespaces  = perf_event__process_namespaces;
1379         kvm->tool.ordered_events = true;
1380         perf_tool__fill_defaults(&kvm->tool);
1381 
1382         /* set defaults */
1383         kvm->display_time = 1;
1384         kvm->opts.user_interval = 1;
1385         kvm->opts.mmap_pages = 512;
1386         kvm->opts.target.uses_mmap = false;
1387         kvm->opts.target.uid_str = NULL;
1388         kvm->opts.target.uid = UINT_MAX;
1389         kvm->opts.proc_map_timeout = 500;
1390 
1391         symbol__init(NULL);
1392         disable_buildid_cache();
1393 
1394         use_browser = 0;
1395 
1396         if (argc) {
1397                 argc = parse_options(argc, argv, live_options,
1398                                      live_usage, 0);
1399                 if (argc)
1400                         usage_with_options(live_usage, live_options);
1401         }
1402 
1403         kvm->duration *= NSEC_PER_USEC;   /* convert usec to nsec */
1404 
1405         /*
1406          * target related setups
1407          */
1408         err = target__validate(&kvm->opts.target);
1409         if (err) {
1410                 target__strerror(&kvm->opts.target, err, errbuf, BUFSIZ);
1411                 ui__warning("%s", errbuf);
1412         }
1413 
1414         if (target__none(&kvm->opts.target))
1415                 kvm->opts.target.system_wide = true;
1416 
1417 
1418         /*
1419          * generate the event list
1420          */
1421         err = setup_kvm_events_tp(kvm);
1422         if (err < 0) {
1423                 pr_err("Unable to setup the kvm tracepoints\n");
1424                 return err;
1425         }
1426 
1427         kvm->evlist = kvm_live_event_list();
1428         if (kvm->evlist == NULL) {
1429                 err = -1;
1430                 goto out;
1431         }
1432 
1433         symbol_conf.nr_events = kvm->evlist->nr_entries;
1434 
1435         if (perf_evlist__create_maps(kvm->evlist, &kvm->opts.target) < 0)
1436                 usage_with_options(live_usage, live_options);
1437 
1438         /*
1439          * perf session
1440          */
1441         kvm->session = perf_session__new(&data, false, &kvm->tool);
1442         if (kvm->session == NULL) {
1443                 err = -1;
1444                 goto out;
1445         }
1446         kvm->session->evlist = kvm->evlist;
1447         perf_session__set_id_hdr_size(kvm->session);
1448         ordered_events__set_copy_on_queue(&kvm->session->ordered_events, true);
1449         machine__synthesize_threads(&kvm->session->machines.host, &kvm->opts.target,
1450                                     kvm->evlist->threads, false,
1451                                     kvm->opts.proc_map_timeout, 1);
1452         err = kvm_live_open_events(kvm);
1453         if (err)
1454                 goto out;
1455 
1456         err = kvm_events_live_report(kvm);
1457 
1458 out:
1459         perf_session__delete(kvm->session);
1460         kvm->session = NULL;
1461         perf_evlist__delete(kvm->evlist);
1462 
1463         return err;
1464 }
1465 #endif
1466 
1467 static void print_kvm_stat_usage(void)
1468 {
1469         printf("Usage: perf kvm stat <command>\n\n");
1470 
1471         printf("# Available commands:\n");
1472         printf("\trecord: record kvm events\n");
1473         printf("\treport: report statistical data of kvm events\n");
1474         printf("\tlive:   live reporting of statistical data of kvm events\n");
1475 
1476         printf("\nOtherwise, it is the alias of 'perf stat':\n");
1477 }
1478 
1479 static int kvm_cmd_stat(const char *file_name, int argc, const char **argv)
1480 {
1481         struct perf_kvm_stat kvm = {
1482                 .file_name = file_name,
1483 
1484                 .trace_vcpu     = -1,
1485                 .report_event   = "vmexit",
1486                 .sort_key       = "sample",
1487 
1488         };
1489 
1490         if (argc == 1) {
1491                 print_kvm_stat_usage();
1492                 goto perf_stat;
1493         }
1494 
1495         if (!strncmp(argv[1], "rec", 3))
1496                 return kvm_events_record(&kvm, argc - 1, argv + 1);
1497 
1498         if (!strncmp(argv[1], "rep", 3))
1499                 return kvm_events_report(&kvm, argc - 1 , argv + 1);
1500 
1501 #ifdef HAVE_TIMERFD_SUPPORT
1502         if (!strncmp(argv[1], "live", 4))
1503                 return kvm_events_live(&kvm, argc - 1 , argv + 1);
1504 #endif
1505 
1506 perf_stat:
1507         return cmd_stat(argc, argv);
1508 }
1509 #endif /* HAVE_KVM_STAT_SUPPORT */
1510 
1511 static int __cmd_record(const char *file_name, int argc, const char **argv)
1512 {
1513         int rec_argc, i = 0, j;
1514         const char **rec_argv;
1515 
1516         rec_argc = argc + 2;
1517         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1518         rec_argv[i++] = strdup("record");
1519         rec_argv[i++] = strdup("-o");
1520         rec_argv[i++] = strdup(file_name);
1521         for (j = 1; j < argc; j++, i++)
1522                 rec_argv[i] = argv[j];
1523 
1524         BUG_ON(i != rec_argc);
1525 
1526         return cmd_record(i, rec_argv);
1527 }
1528 
1529 static int __cmd_report(const char *file_name, int argc, const char **argv)
1530 {
1531         int rec_argc, i = 0, j;
1532         const char **rec_argv;
1533 
1534         rec_argc = argc + 2;
1535         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1536         rec_argv[i++] = strdup("report");
1537         rec_argv[i++] = strdup("-i");
1538         rec_argv[i++] = strdup(file_name);
1539         for (j = 1; j < argc; j++, i++)
1540                 rec_argv[i] = argv[j];
1541 
1542         BUG_ON(i != rec_argc);
1543 
1544         return cmd_report(i, rec_argv);
1545 }
1546 
1547 static int
1548 __cmd_buildid_list(const char *file_name, int argc, const char **argv)
1549 {
1550         int rec_argc, i = 0, j;
1551         const char **rec_argv;
1552 
1553         rec_argc = argc + 2;
1554         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1555         rec_argv[i++] = strdup("buildid-list");
1556         rec_argv[i++] = strdup("-i");
1557         rec_argv[i++] = strdup(file_name);
1558         for (j = 1; j < argc; j++, i++)
1559                 rec_argv[i] = argv[j];
1560 
1561         BUG_ON(i != rec_argc);
1562 
1563         return cmd_buildid_list(i, rec_argv);
1564 }
1565 
1566 int cmd_kvm(int argc, const char **argv)
1567 {
1568         const char *file_name = NULL;
1569         const struct option kvm_options[] = {
1570                 OPT_STRING('i', "input", &file_name, "file",
1571                            "Input file name"),
1572                 OPT_STRING('o', "output", &file_name, "file",
1573                            "Output file name"),
1574                 OPT_BOOLEAN(0, "guest", &perf_guest,
1575                             "Collect guest os data"),
1576                 OPT_BOOLEAN(0, "host", &perf_host,
1577                             "Collect host os data"),
1578                 OPT_STRING(0, "guestmount", &symbol_conf.guestmount, "directory",
1579                            "guest mount directory under which every guest os"
1580                            " instance has a subdir"),
1581                 OPT_STRING(0, "guestvmlinux", &symbol_conf.default_guest_vmlinux_name,
1582                            "file", "file saving guest os vmlinux"),
1583                 OPT_STRING(0, "guestkallsyms", &symbol_conf.default_guest_kallsyms,
1584                            "file", "file saving guest os /proc/kallsyms"),
1585                 OPT_STRING(0, "guestmodules", &symbol_conf.default_guest_modules,
1586                            "file", "file saving guest os /proc/modules"),
1587                 OPT_INCR('v', "verbose", &verbose,
1588                             "be more verbose (show counter open errors, etc)"),
1589                 OPT_END()
1590         };
1591 
1592         const char *const kvm_subcommands[] = { "top", "record", "report", "diff",
1593                                                 "buildid-list", "stat", NULL };
1594         const char *kvm_usage[] = { NULL, NULL };
1595 
1596         perf_host  = 0;
1597         perf_guest = 1;
1598 
1599         argc = parse_options_subcommand(argc, argv, kvm_options, kvm_subcommands, kvm_usage,
1600                                         PARSE_OPT_STOP_AT_NON_OPTION);
1601         if (!argc)
1602                 usage_with_options(kvm_usage, kvm_options);
1603 
1604         if (!perf_host)
1605                 perf_guest = 1;
1606 
1607         if (!file_name) {
1608                 file_name = get_filename_for_perf_kvm();
1609 
1610                 if (!file_name) {
1611                         pr_err("Failed to allocate memory for filename\n");
1612                         return -ENOMEM;
1613                 }
1614         }
1615 
1616         if (!strncmp(argv[0], "rec", 3))
1617                 return __cmd_record(file_name, argc, argv);
1618         else if (!strncmp(argv[0], "rep", 3))
1619                 return __cmd_report(file_name, argc, argv);
1620         else if (!strncmp(argv[0], "diff", 4))
1621                 return cmd_diff(argc, argv);
1622         else if (!strncmp(argv[0], "top", 3))
1623                 return cmd_top(argc, argv);
1624         else if (!strncmp(argv[0], "buildid-list", 12))
1625                 return __cmd_buildid_list(file_name, argc, argv);
1626 #ifdef HAVE_KVM_STAT_SUPPORT
1627         else if (!strncmp(argv[0], "stat", 4))
1628                 return kvm_cmd_stat(file_name, argc, argv);
1629 #endif
1630         else
1631                 usage_with_options(kvm_usage, kvm_options);
1632 
1633         return 0;
1634 }
1635 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp