projects / blktrace.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
[PATCH] blkparse: allow a few more skips
[blktrace.git] / blkparse.c
... / ...
CommitLineData
1/*
2 * block queue tracing parse application
3 *
4 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 */
21#include <sys/types.h>
22#include <sys/stat.h>
23#include <unistd.h>
24#include <stdio.h>
25#include <fcntl.h>
26#include <stdlib.h>
27#include <string.h>
28#include <getopt.h>
29#include <errno.h>
30#include <signal.h>
31#include <locale.h>
32#include <limits.h>
33
34#include "blktrace.h"
35#include "rbtree.h"
36
37#define SECONDS(x) ((unsigned long long)(x) / 1000000000)
38#define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000)
39#define DOUBLE_TO_NANO_ULL(d) ((unsigned long long)((d) * 1000000000))
40
41#define MINORBITS 20
42#define MINORMASK ((1U << MINORBITS) - 1)
43#define MAJOR(dev) ((unsigned int) ((dev) >> MINORBITS))
44#define MINOR(dev) ((unsigned int) ((dev) & MINORMASK))
45
46#define min(a, b) ((a) < (b) ? (a) : (b))
47
48struct io_stats {
49 unsigned long qreads, qwrites, creads, cwrites, mreads, mwrites;
50 unsigned long ireads, iwrites;
51 unsigned long long qread_kb, qwrite_kb, cread_kb, cwrite_kb;
52 unsigned long long iread_kb, iwrite_kb;
53 unsigned long io_unplugs, timer_unplugs;
54};
55
56struct per_cpu_info {
57 int cpu;
58 int nelems;
59
60 int fd;
61 char fname[128];
62
63 struct io_stats io_stats;
64};
65
66struct per_dev_info {
67 dev_t id;
68 char *name;
69
70 int backwards;
71 unsigned long long events;
72 unsigned long long last_reported_time;
73 struct io_stats io_stats;
74 unsigned long last_sequence;
75 unsigned long skips;
76
77 int ncpus;
78 struct per_cpu_info *cpus;
79};
80
81struct per_process_info {
82 char name[16];
83 __u32 pid;
84 struct io_stats io_stats;
85 struct per_process_info *hash_next, *list_next;
86
87 /*
88 * individual io stats
89 */
90 unsigned long long longest_allocation_wait[2];
91 unsigned long long longest_dispatch_wait[2];
92 unsigned long long longest_completion_wait[2];
93};
94
95#define PPI_HASH_SHIFT (8)
96static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT];
97static struct per_process_info *ppi_list;
98
99#define S_OPTS "i:o:b:stqw:"
100static struct option l_opts[] = {
101 {
102 .name = "input",
103 .has_arg = 1,
104 .flag = NULL,
105 .val = 'i'
106 },
107 {
108 .name = "output",
109 .has_arg = 1,
110 .flag = NULL,
111 .val = 'o'
112 },
113 {
114 .name = "batch",
115 .has_arg = 1,
116 .flag = NULL,
117 .val = 'b'
118 },
119 {
120 .name = "per program stats",
121 .has_arg = 0,
122 .flag = NULL,
123 .val = 's'
124 },
125 {
126 .name = "track ios",
127 .has_arg = 0,
128 .flag = NULL,
129 .val = 't'
130 },
131 {
132 .name = "quiet",
133 .has_arg = 0,
134 .flag = NULL,
135 .val = 'q'
136 },
137 {
138 .name = "stopwatch",
139 .has_arg = 1,
140 .flag = NULL,
141 .val = 'w'
142 },
143 {
144 .name = NULL,
145 .has_arg = 0,
146 .flag = NULL,
147 .val = 0
148 }
149};
150
151/*
152 * for sorting the displayed output
153 */
154struct trace {
155 struct blk_io_trace *bit;
156 struct rb_node rb_node;
157 struct trace *next;
158 int skipped;
159};
160
161static struct rb_root rb_sort_root;
162static struct rb_root rb_track_root;
163
164static struct trace *trace_list;
165
166/*
167 * for tracking individual ios
168 */
169struct io_track {
170 struct rb_node rb_node;
171
172 dev_t device;
173 __u64 sector;
174 __u32 pid;
175 unsigned long long allocation_time;
176 unsigned long long queue_time;
177 unsigned long long dispatch_time;
178 unsigned long long completion_time;
179};
180
181static int ndevices;
182static struct per_dev_info *devices;
183static char *get_dev_name(struct per_dev_info *, char *, int);
184
185static FILE *ofp;
186static char *output_name;
187
188static unsigned long long genesis_time;
189static unsigned long long stopwatch_start; /* start from zero by default */
190static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
191
192static int per_process_stats;
193static int track_ios;
194
195#define RB_BATCH_DEFAULT (1024)
196static int rb_batch = RB_BATCH_DEFAULT;
197
198static int pipeline;
199
200#define is_done() (*(volatile int *)(&done))
201static volatile int done;
202
203static inline unsigned long hash_long(unsigned long val)
204{
205#if __WORDSIZE == 32
206 val *= 0x9e370001UL;
207#elif __WORDSIZE == 64
208 val *= 0x9e37fffffffc0001UL;
209#else
210#error unknown word size
211#endif
212
213 return val >> (__WORDSIZE - PPI_HASH_SHIFT);
214}
215
216static inline void add_process_to_hash(struct per_process_info *ppi)
217{
218 const int hash_idx = hash_long(ppi->pid);
219
220 ppi->hash_next = ppi_hash[hash_idx];
221 ppi_hash[hash_idx] = ppi;
222}
223
224static inline void add_process_to_list(struct per_process_info *ppi)
225{
226 ppi->list_next = ppi_list;
227 ppi_list = ppi;
228}
229
230static struct per_process_info *find_process_by_pid(__u32 pid)
231{
232 const int hash_idx = hash_long(pid);
233 struct per_process_info *ppi;
234
235 ppi = ppi_hash[hash_idx];
236 while (ppi) {
237 if (ppi->pid == pid)
238 return ppi;
239
240 ppi = ppi->hash_next;
241 }
242
243 return NULL;
244}
245
246static inline int trace_rb_insert(struct trace *t)
247{
248 struct rb_node **p = &rb_sort_root.rb_node;
249 struct rb_node *parent = NULL;
250 struct trace *__t;
251
252 if (genesis_time == 0 || t->bit->time < genesis_time)
253 genesis_time = t->bit->time;
254
255 while (*p) {
256 parent = *p;
257 __t = rb_entry(parent, struct trace, rb_node);
258
259 if (t->bit->time < __t->bit->time)
260 p = &(*p)->rb_left;
261 else if (t->bit->time > __t->bit->time)
262 p = &(*p)->rb_right;
263 else if (t->bit->device < __t->bit->device)
264 p = &(*p)->rb_left;
265 else if (t->bit->device > __t->bit->device)
266 p = &(*p)->rb_right;
267 else if (t->bit->sequence < __t->bit->sequence)
268 p = &(*p)->rb_left;
269 else if (t->bit->sequence > __t->bit->sequence)
270 p = &(*p)->rb_right;
271 else if (t->bit->device == __t->bit->device) {
272 fprintf(stderr,
273 "sequence alias (%d) on device %d,%d!\n",
274 t->bit->sequence,
275 MAJOR(t->bit->device), MINOR(t->bit->device));
276 return 1;
277 }
278 }
279
280 rb_link_node(&t->rb_node, parent, p);
281 rb_insert_color(&t->rb_node, &rb_sort_root);
282 return 0;
283}
284
285static inline int track_rb_insert(struct io_track *iot)
286{
287 struct rb_node **p = &rb_track_root.rb_node;
288 struct rb_node *parent = NULL;
289 struct io_track *__iot;
290
291 while (*p) {
292 parent = *p;
293
294 __iot = rb_entry(parent, struct io_track, rb_node);
295
296 if (iot->device < __iot->device)
297 p = &(*p)->rb_left;
298 else if (iot->device > __iot->device)
299 p = &(*p)->rb_right;
300 else if (iot->sector < __iot->sector)
301 p = &(*p)->rb_left;
302 else if (iot->sector > __iot->sector)
303 p = &(*p)->rb_right;
304 else {
305 fprintf(stderr,
306 "sector alias (%llu) on device %d,%d!\n",
307 iot->sector,
308 MAJOR(iot->device), MINOR(iot->device));
309 return 1;
310 }
311 }
312
313 rb_link_node(&iot->rb_node, parent, p);
314 rb_insert_color(&iot->rb_node, &rb_track_root);
315 return 0;
316}
317
318static struct io_track *__find_track(dev_t device, __u64 sector)
319{
320 struct rb_node **p = &rb_track_root.rb_node;
321 struct rb_node *parent = NULL;
322 struct io_track *__iot;
323
324 while (*p) {
325 parent = *p;
326
327 __iot = rb_entry(parent, struct io_track, rb_node);
328
329 if (device < __iot->device)
330 p = &(*p)->rb_left;
331 else if (device > __iot->device)
332 p = &(*p)->rb_right;
333 else if (sector < __iot->sector)
334 p = &(*p)->rb_left;
335 else if (sector > __iot->sector)
336 p = &(*p)->rb_right;
337 else
338 return __iot;
339 }
340
341 return NULL;
342}
343
344static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector)
345{
346 struct io_track *iot;
347
348 iot = __find_track(device, sector);
349 if (!iot) {
350 iot = malloc(sizeof(*iot));
351 iot->pid = pid;
352 iot->device = device;
353 iot->sector = sector;
354 track_rb_insert(iot);
355 }
356
357 return iot;
358}
359
360static void log_track_frontmerge(struct blk_io_trace *t)
361{
362 struct io_track *iot;
363
364 if (!track_ios)
365 return;
366
367 iot = __find_track(t->device, t->sector + (t->bytes >> 9));
368 if (!iot) {
369 fprintf(stderr, "failed to find mergeable event\n");
370 return;
371 }
372
373 rb_erase(&iot->rb_node, &rb_track_root);
374 iot->sector -= t->bytes >> 9;
375 track_rb_insert(iot);
376}
377
378static void log_track_getrq(struct blk_io_trace *t)
379{
380 struct io_track *iot;
381
382 if (!track_ios)
383 return;
384
385 iot = find_track(t->pid, t->device, t->sector);
386 iot->allocation_time = t->time;
387}
388
389
390/*
391 * return time between rq allocation and queue
392 */
393static unsigned long long log_track_queue(struct blk_io_trace *t)
394{
395 unsigned long long elapsed;
396 struct io_track *iot;
397
398 if (!track_ios)
399 return -1;
400
401 iot = find_track(t->pid, t->device, t->sector);
402 iot->queue_time = t->time;
403 elapsed = iot->queue_time - iot->allocation_time;
404
405 if (per_process_stats) {
406 struct per_process_info *ppi = find_process_by_pid(iot->pid);
407 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
408
409 if (ppi && elapsed > ppi->longest_allocation_wait[w])
410 ppi->longest_allocation_wait[w] = elapsed;
411 }
412
413 return elapsed;
414}
415
416/*
417 * return time between queue and issue
418 */
419static unsigned long long log_track_issue(struct blk_io_trace *t)
420{
421 unsigned long long elapsed;
422 struct io_track *iot;
423
424 if (!track_ios)
425 return -1;
426 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
427 return -1;
428
429 iot = __find_track(t->device, t->sector);
430 if (!iot) {
431 fprintf(stderr, "failed to find issue event\n");
432 return -1;
433 }
434
435 iot->dispatch_time = t->time;
436 elapsed = iot->dispatch_time - iot->queue_time;
437
438 if (per_process_stats) {
439 struct per_process_info *ppi = find_process_by_pid(iot->pid);
440 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
441
442 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
443 ppi->longest_dispatch_wait[w] = elapsed;
444 }
445
446 return elapsed;
447}
448
449/*
450 * return time between dispatch and complete
451 */
452static unsigned long long log_track_complete(struct blk_io_trace *t)
453{
454 unsigned long long elapsed;
455 struct io_track *iot;
456
457 if (!track_ios)
458 return -1;
459 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
460 return -1;
461
462 iot = __find_track(t->device, t->sector);
463 if (!iot) {
464 fprintf(stderr, "failed to find complete event\n");
465 return -1;
466 }
467
468 iot->completion_time = t->time;
469 elapsed = iot->completion_time - iot->dispatch_time;
470
471 if (per_process_stats) {
472 struct per_process_info *ppi = find_process_by_pid(iot->pid);
473 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
474
475 if (ppi && elapsed > ppi->longest_completion_wait[w])
476 ppi->longest_completion_wait[w] = elapsed;
477 }
478
479 /*
480 * kill the trace, we don't need it after completion
481 */
482 rb_erase(&iot->rb_node, &rb_track_root);
483 free(iot);
484
485 return elapsed;
486}
487
488
489static struct io_stats *find_process_io_stats(__u32 pid, char *name)
490{
491 struct per_process_info *ppi = find_process_by_pid(pid);
492
493 if (!ppi) {
494 ppi = malloc(sizeof(*ppi));
495 memset(ppi, 0, sizeof(*ppi));
496 strncpy(ppi->name, name, sizeof(ppi->name));
497 ppi->pid = pid;
498 add_process_to_hash(ppi);
499 add_process_to_list(ppi);
500 }
501
502 return &ppi->io_stats;
503}
504
505
506static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
507{
508 struct per_cpu_info *cpus = pdi->cpus;
509 int ncpus = pdi->ncpus;
510 int new_count = cpu + 1;
511 int new_space, size;
512 char *new_start;
513
514 size = new_count * sizeof(struct per_cpu_info);
515 cpus = realloc(cpus, size);
516 if (!cpus) {
517 char name[20];
518 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
519 get_dev_name(pdi, name, sizeof(name)), size);
520 exit(1);
521 }
522
523 new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
524 new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
525 memset(new_start, 0, new_space);
526
527 pdi->ncpus = new_count;
528 pdi->cpus = cpus;
529}
530
531static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
532{
533 struct per_cpu_info *pci;
534
535 if (cpu >= pdi->ncpus)
536 resize_cpu_info(pdi, cpu);
537
538 pci = &pdi->cpus[cpu];
539 pci->cpu = cpu;
540 return pci;
541}
542
543
544static int resize_devices(char *name)
545{
546 int size = (ndevices + 1) * sizeof(struct per_dev_info);
547
548 devices = realloc(devices, size);
549 if (!devices) {
550 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
551 return 1;
552 }
553 memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
554 devices[ndevices].name = name;
555 ndevices++;
556 return 0;
557}
558
559static struct per_dev_info *get_dev_info(dev_t id)
560{
561 struct per_dev_info *pdi;
562 int i;
563
564 for (i = 0; i < ndevices; i++)
565 if (devices[i].id == id)
566 return &devices[i];
567
568 if (resize_devices(NULL) != 0)
569 return NULL;
570
571 pdi = &devices[ndevices - 1];
572 pdi->id = id;
573 return pdi;
574}
575
576static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
577{
578 if (pdi->name)
579 snprintf(buffer, size, "%s", pdi->name);
580 else
581 snprintf(buffer, size, "%d,%d", MAJOR(pdi->id), MINOR(pdi->id));
582 return buffer;
583}
584
585static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
586{
587 unsigned long long this = bit->time;
588 unsigned long long last = pdi->last_reported_time;
589
590 pdi->backwards = (this < last) ? 'B' : ' ';
591 pdi->last_reported_time = this;
592}
593
594static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
595 int rw)
596{
597 if (rw) {
598 ios->mwrites++;
599 ios->qwrite_kb += t->bytes >> 10;
600 } else {
601 ios->mreads++;
602 ios->qread_kb += t->bytes >> 10;
603 }
604}
605
606static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
607 int rw)
608{
609 __account_m(&pci->io_stats, t, rw);
610
611 if (per_process_stats) {
612 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
613
614 __account_m(ios, t, rw);
615 }
616}
617
618static inline void __account_q(struct io_stats *ios, struct blk_io_trace *t,
619 int rw)
620{
621 if (rw) {
622 ios->qwrites++;
623 ios->qwrite_kb += t->bytes >> 10;
624 } else {
625 ios->qreads++;
626 ios->qread_kb += t->bytes >> 10;
627 }
628}
629
630static inline void account_q(struct blk_io_trace *t, struct per_cpu_info *pci,
631 int rw)
632{
633 __account_q(&pci->io_stats, t, rw);
634
635 if (per_process_stats) {
636 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
637
638 __account_q(ios, t, rw);
639 }
640}
641
642static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes)
643{
644 if (rw) {
645 ios->cwrites++;
646 ios->cwrite_kb += bytes >> 10;
647 } else {
648 ios->creads++;
649 ios->cread_kb += bytes >> 10;
650 }
651}
652
653static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
654 int rw, int bytes)
655{
656 __account_c(&pci->io_stats, rw, bytes);
657
658 if (per_process_stats) {
659 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
660
661 __account_c(ios, rw, bytes);
662 }
663}
664
665static inline void __account_i(struct io_stats *ios, int rw, unsigned int bytes)
666{
667 if (rw) {
668 ios->iwrites++;
669 ios->iwrite_kb += bytes >> 10;
670 } else {
671 ios->ireads++;
672 ios->iread_kb += bytes >> 10;
673 }
674}
675
676static inline void account_i(struct blk_io_trace *t, struct per_cpu_info *pci,
677 int rw)
678{
679 __account_i(&pci->io_stats, rw, t->bytes);
680
681 if (per_process_stats) {
682 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
683
684 __account_i(ios, rw, t->bytes);
685 }
686}
687
688static inline void __account_unplug(struct io_stats *ios, int timer)
689{
690 if (timer)
691 ios->timer_unplugs++;
692 else
693 ios->io_unplugs++;
694}
695
696static inline void account_unplug(struct blk_io_trace *t,
697 struct per_cpu_info *pci, int timer)
698{
699 __account_unplug(&pci->io_stats, timer);
700
701 if (per_process_stats) {
702 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
703
704 __account_unplug(ios, timer);
705 }
706}
707
708static void output(struct per_cpu_info *pci, char *s)
709{
710 fprintf(ofp, "%s", s);
711}
712
713static char hstring[256];
714static char tstring[256];
715
716static inline char *setup_header(struct per_cpu_info *pci,
717 struct blk_io_trace *t, char *act)
718{
719 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
720 int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER);
721 int s = t->action & BLK_TC_ACT(BLK_TC_SYNC);
722 char rwbs[4];
723 int i = 0;
724
725 if (w)
726 rwbs[i++] = 'W';
727 else
728 rwbs[i++] = 'R';
729 if (b)
730 rwbs[i++] = 'B';
731 if (s)
732 rwbs[i++] = 'S';
733
734 rwbs[i] = '\0';
735
736 sprintf(hstring, "%3d,%-3d %2d %8ld %5Lu.%09Lu %5u %2s %3s",
737 MAJOR(t->device), MINOR(t->device), pci->cpu,
738 (unsigned long)t->sequence, SECONDS(t->time),
739 NANO_SECONDS(t->time), t->pid, act, rwbs);
740
741 return hstring;
742}
743
744static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t,
745 char *act)
746{
747 unsigned long long elapsed = log_track_complete(t);
748
749 if (elapsed != -1ULL) {
750 unsigned long usec = elapsed / 1000;
751
752 sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n",
753 setup_header(pci, t, act),
754 (unsigned long long)t->sector, t->bytes >> 9,
755 usec, t->error);
756 } else {
757 sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act),
758 (unsigned long long)t->sector, t->bytes >> 9, t->error);
759 }
760
761 output(pci, tstring);
762}
763
764static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
765 char *act)
766{
767 unsigned long long elapsed = log_track_queue(t);
768
769 if (elapsed != -1ULL) {
770 unsigned long usec = elapsed / 1000;
771
772 sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n",
773 setup_header(pci, t, act),
774 (unsigned long long)t->sector, t->bytes >> 9,
775 usec, t->comm);
776 } else {
777 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
778 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
779 }
780 output(pci, tstring);
781}
782
783static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t,
784 char *act)
785{
786 unsigned long long elapsed = log_track_issue(t);
787
788 if (elapsed != -1ULL) {
789 double usec = (double) elapsed / 1000;
790
791 sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n",
792 setup_header(pci, t, act),
793 (unsigned long long)t->sector, t->bytes >> 9,
794 usec, t->comm);
795 } else {
796 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
797 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
798 }
799
800 output(pci, tstring);
801}
802
803static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
804 char *act)
805{
806 if (act[0] == 'F')
807 log_track_frontmerge(t);
808
809 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
810 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
811 output(pci, tstring);
812}
813
814static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
815 char *act)
816{
817 sprintf(tstring,"%s [%s]\n", setup_header(pci, t, act), t->comm);
818 output(pci, tstring);
819}
820
821static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
822 char *act)
823{
824 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
825 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
826 output(pci, tstring);
827}
828
829static int log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
830 char *act)
831{
832 __u64 *depth;
833 int len;
834
835 len = sprintf(tstring,"%s [%s] ", setup_header(pci, t, act), t->comm);
836 depth = (__u64 *) ((char *) t + sizeof(*t));
837 sprintf(tstring + len, "%u\n", (unsigned int) be64_to_cpu(*depth));
838 output(pci, tstring);
839
840 return 0;
841}
842
843static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
844{
845 unsigned char *buf;
846 int i;
847
848 sprintf(tstring,"%s ", setup_header(pci, t, act));
849 output(pci, tstring);
850
851 buf = (unsigned char *) t + sizeof(*t);
852 for (i = 0; i < t->pdu_len; i++) {
853 sprintf(tstring,"%02x ", buf[i]);
854 output(pci, tstring);
855 }
856
857 if (act[0] == 'C') {
858 sprintf(tstring,"[%d]\n", t->error);
859 output(pci, tstring);
860 } else {
861 sprintf(tstring,"[%s]\n", t->comm);
862 output(pci, tstring);
863 }
864 return 0;
865}
866
867static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
868{
869 int ret = 0;
870
871 switch (t->action & 0xffff) {
872 case __BLK_TA_QUEUE:
873 log_generic(pci, t, "Q");
874 break;
875 case __BLK_TA_GETRQ:
876 log_generic(pci, t, "G");
877 break;
878 case __BLK_TA_SLEEPRQ:
879 log_generic(pci, t, "S");
880 break;
881 case __BLK_TA_REQUEUE:
882 log_generic(pci, t, "R");
883 break;
884 case __BLK_TA_ISSUE:
885 ret = log_pc(pci, t, "D");
886 break;
887 case __BLK_TA_COMPLETE:
888 log_pc(pci, t, "C");
889 break;
890 default:
891 fprintf(stderr, "Bad pc action %x\n", t->action);
892 ret = 1;
893 break;
894 }
895
896 return ret;
897}
898
899static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci)
900{
901 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
902 int act = t->action & 0xffff;
903
904 switch (act) {
905 case __BLK_TA_QUEUE:
906 account_q(t, pci, w);
907 log_queue(pci, t, "Q");
908 break;
909 case __BLK_TA_BACKMERGE:
910 account_m(t, pci, w);
911 log_merge(pci, t, "M");
912 break;
913 case __BLK_TA_FRONTMERGE:
914 account_m(t, pci, w);
915 log_merge(pci, t, "F");
916 break;
917 case __BLK_TA_GETRQ:
918 log_track_getrq(t);
919 log_generic(pci, t, "G");
920 break;
921 case __BLK_TA_SLEEPRQ:
922 log_generic(pci, t, "S");
923 break;
924 case __BLK_TA_REQUEUE:
925 account_c(t, pci, w, -t->bytes);
926 log_queue(pci, t, "R");
927 break;
928 case __BLK_TA_ISSUE:
929 account_i(t, pci, w);
930 log_issue(pci, t, "D");
931 break;
932 case __BLK_TA_COMPLETE:
933 account_c(t, pci, w, t->bytes);
934 log_complete(pci, t, "C");
935 break;
936 case __BLK_TA_PLUG:
937 log_action(pci, t, "P");
938 break;
939 case __BLK_TA_UNPLUG_IO:
940 account_unplug(t, pci, 0);
941 log_unplug(pci, t, "U");
942 break;
943 case __BLK_TA_UNPLUG_TIMER:
944 account_unplug(t, pci, 1);
945 log_unplug(pci, t, "UT");
946 break;
947 default:
948 fprintf(stderr, "Bad fs action %x\n", t->action);
949 break;
950 }
951}
952
953static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
954 struct per_dev_info *pdi)
955{
956 int ret = 0;
957
958 if (t->action & BLK_TC_ACT(BLK_TC_PC))
959 ret = dump_trace_pc(t, pci);
960 else
961 dump_trace_fs(t, pci);
962
963 pdi->events++;
964 return ret;
965}
966
967static void dump_io_stats(struct io_stats *ios, char *msg)
968{
969 fprintf(ofp, "%s\n", msg);
970
971 fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
972 fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
973
974 fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
975 fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
976 fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
977 fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
978 fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
979 fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
980 fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
981 fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
982}
983
984static void dump_wait_stats(struct per_process_info *ppi)
985{
986 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
987 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
988 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
989 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
990 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
991 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
992
993 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
994 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
995 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
996 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
997 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
998 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
999}
1000
1001static void show_process_stats(void)
1002{
1003 struct per_process_info *ppi;
1004
1005 ppi = ppi_list;
1006 while (ppi) {
1007 dump_io_stats(&ppi->io_stats, ppi->name);
1008 dump_wait_stats(ppi);
1009 ppi = ppi->list_next;
1010 }
1011
1012 fprintf(ofp, "\n");
1013}
1014
1015static void show_device_and_cpu_stats(void)
1016{
1017 struct per_dev_info *pdi;
1018 struct per_cpu_info *pci;
1019 struct io_stats total, *ios;
1020 int i, j, pci_events;
1021 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1022 char name[32];
1023
1024 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1025
1026 memset(&total, 0, sizeof(total));
1027 pci_events = 0;
1028
1029 if (i > 0)
1030 fprintf(ofp, "\n");
1031
1032 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1033 if (!pci->nelems)
1034 continue;
1035
1036 ios = &pci->io_stats;
1037 total.qreads += ios->qreads;
1038 total.qwrites += ios->qwrites;
1039 total.creads += ios->creads;
1040 total.cwrites += ios->cwrites;
1041 total.mreads += ios->mreads;
1042 total.mwrites += ios->mwrites;
1043 total.ireads += ios->ireads;
1044 total.iwrites += ios->iwrites;
1045 total.qread_kb += ios->qread_kb;
1046 total.qwrite_kb += ios->qwrite_kb;
1047 total.cread_kb += ios->cread_kb;
1048 total.cwrite_kb += ios->cwrite_kb;
1049 total.iread_kb += ios->iread_kb;
1050 total.iwrite_kb += ios->iwrite_kb;
1051 total.timer_unplugs += ios->timer_unplugs;
1052 total.io_unplugs += ios->io_unplugs;
1053
1054 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1055 j, get_dev_name(pdi, name, sizeof(name)));
1056 dump_io_stats(ios, line);
1057 pci_events++;
1058 }
1059
1060 if (pci_events > 1) {
1061 fprintf(ofp, "\n");
1062 snprintf(line, sizeof(line) - 1, "Total (%s):",
1063 get_dev_name(pdi, name, sizeof(name)));
1064 dump_io_stats(&total, line);
1065 }
1066
1067 fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n",
1068 get_dev_name(pdi, line, sizeof(line)), pdi->events,
1069 pdi->skips);
1070 }
1071}
1072
1073static struct blk_io_trace *find_trace(void *p, unsigned long offset)
1074{
1075 unsigned long max_offset = offset;
1076 unsigned long off;
1077 struct blk_io_trace *bit;
1078 __u32 magic;
1079
1080 for (off = 0; off < max_offset; off++) {
1081 bit = p + off;
1082
1083 magic = be32_to_cpu(bit->magic);
1084 if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
1085 return bit;
1086 }
1087
1088 return NULL;
1089}
1090
1091static int sort_entries(void)
1092{
1093 struct per_dev_info *pdi;
1094 struct per_cpu_info *pci;
1095 struct blk_io_trace *bit;
1096 struct trace *t;
1097 int nr = 0;
1098
1099 while ((t = trace_list) != NULL) {
1100
1101 trace_list = t->next;
1102 bit = t->bit;
1103
1104 memset(&t->rb_node, 0, sizeof(t->rb_node));
1105
1106 if (verify_trace(bit))
1107 break;
1108 if (trace_rb_insert(t))
1109 return -1;
1110
1111 pdi = get_dev_info(bit->device);
1112 pci = get_cpu_info(pdi, bit->cpu);
1113 pci->nelems++;
1114
1115 nr++;
1116 }
1117
1118 return nr;
1119}
1120
1121static void show_entries_rb(int piped)
1122{
1123 struct per_dev_info *pdi;
1124 struct blk_io_trace *bit;
1125 struct rb_node *n;
1126 struct trace *t;
1127 int cpu;
1128
1129 while ((n = rb_first(&rb_sort_root)) != NULL) {
1130
1131 t = rb_entry(n, struct trace, rb_node);
1132 bit = t->bit;
1133
1134 pdi = get_dev_info(bit->device);
1135 if (!pdi) {
1136 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
1137 MAJOR(bit->device), MINOR(bit->device));
1138 break;
1139 }
1140 cpu = bit->cpu;
1141 if (cpu > pdi->ncpus) {
1142 fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
1143 cpu, MAJOR(bit->device), MINOR(bit->device));
1144 break;
1145 }
1146
1147 /*
1148 * back off displaying more info if we are out of sync
1149 * on SMP systems. to prevent stalling on lost events,
1150 * only allow an event to us a few times
1151 */
1152 if (piped && bit->sequence != (pdi->last_sequence + 1)) {
1153 if (t->skipped < 5) {
1154 t->skipped++;
1155 break;
1156 } else {
1157 fprintf(stderr, "skipping from %lu to %u\n", pdi->last_sequence, bit->sequence);
1158 pdi->skips++;
1159 }
1160 }
1161
1162 pdi->last_sequence = bit->sequence;
1163
1164 bit->time -= genesis_time;
1165 if (bit->time < stopwatch_start)
1166 continue;
1167 if (bit->time >= stopwatch_end)
1168 break;
1169
1170 check_time(pdi, bit);
1171
1172 if (dump_trace(bit, &pdi->cpus[cpu], pdi))
1173 break;
1174
1175 rb_erase(&t->rb_node, &rb_sort_root);
1176
1177 if (piped) {
1178 free(bit);
1179 free(t);
1180 }
1181 }
1182}
1183
1184static int read_data(int fd, void *buffer, int bytes, int block)
1185{
1186 int ret, bytes_left, fl;
1187 void *p;
1188
1189 fl = fcntl(fd, F_GETFL);
1190
1191 if (!block)
1192 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
1193 else
1194 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
1195
1196 bytes_left = bytes;
1197 p = buffer;
1198 while (bytes_left > 0) {
1199 ret = read(fd, p, bytes_left);
1200 if (!ret)
1201 return 1;
1202 else if (ret < 0) {
1203 if (errno != EAGAIN)
1204 perror("read");
1205 return -1;
1206 } else {
1207 p += ret;
1208 bytes_left -= ret;
1209 }
1210 }
1211
1212 return 0;
1213}
1214
1215/*
1216 * Find the traces in 'tb' and add them to the list for sorting and
1217 * displaying
1218 */
1219static int find_entries(void *tb, unsigned long size)
1220{
1221 struct blk_io_trace *bit;
1222 struct trace *t;
1223 void *start = tb;
1224
1225 while (tb - start <= size - sizeof(*bit)) {
1226 bit = find_trace(tb, size - (tb - start));
1227 if (!bit)
1228 break;
1229
1230 t = malloc(sizeof(*t));
1231 memset(t, 0, sizeof(*t));
1232 t->bit = bit;
1233
1234 trace_to_cpu(bit);
1235 t->next = trace_list;
1236 trace_list = t;
1237
1238 tb += sizeof(*bit) + bit->pdu_len;
1239 }
1240
1241 return 0;
1242}
1243
1244static int do_file(void)
1245{
1246 int i, j, nfiles = 0, nelems;
1247
1248 for (i = 0; i < ndevices; i++) {
1249 for (j = 0;; j++, nfiles++) {
1250 struct per_dev_info *pdi;
1251 struct per_cpu_info *pci;
1252 struct stat st;
1253 void *tb;
1254
1255 pdi = &devices[i];
1256 pci = get_cpu_info(pdi, j);
1257 pci->cpu = j;
1258
1259 snprintf(pci->fname, sizeof(pci->fname)-1,
1260 "%s_out.%d", pdi->name, j);
1261 if (stat(pci->fname, &st) < 0)
1262 break;
1263 if (!st.st_size)
1264 continue;
1265
1266 printf("Processing %s\n", pci->fname);
1267
1268 tb = malloc(st.st_size);
1269 if (!tb) {
1270 fprintf(stderr, "Out of memory, skip file %s\n",
1271 pci->fname);
1272 continue;
1273 }
1274
1275 pci->fd = open(pci->fname, O_RDONLY);
1276 if (pci->fd < 0) {
1277 perror(pci->fname);
1278 free(tb);
1279 continue;
1280 }
1281
1282 if (read_data(pci->fd, tb, st.st_size, 1)) {
1283 close(pci->fd);
1284 free(tb);
1285 continue;
1286 }
1287
1288 if (find_entries(tb, st.st_size)) {
1289 close(pci->fd);
1290 free(tb);
1291 }
1292
1293 nelems = sort_entries();
1294 if (nelems == -1) {
1295 close(pci->fd);
1296 free(tb);
1297 continue;
1298 }
1299
1300 printf("Completed %s (CPU%d %d, entries)\n",
1301 pci->fname, j, nelems);
1302 close(pci->fd);
1303 }
1304 }
1305
1306 if (!nfiles) {
1307 fprintf(stderr, "No files found\n");
1308 return 1;
1309 }
1310
1311 show_entries_rb(0);
1312 return 0;
1313}
1314
1315static int read_sort_events(int fd)
1316{
1317 int events = 0;
1318
1319 do {
1320 struct blk_io_trace *bit;
1321 struct trace *t;
1322 int pdu_len;
1323 __u32 magic;
1324
1325 bit = malloc(sizeof(*bit));
1326
1327 if (read_data(fd, bit, sizeof(*bit), !events))
1328 break;
1329
1330 magic = be32_to_cpu(bit->magic);
1331 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1332 fprintf(stderr, "Bad magic %x\n", magic);
1333 break;
1334 }
1335
1336 pdu_len = be16_to_cpu(bit->pdu_len);
1337 if (pdu_len) {
1338 void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
1339
1340 if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1))
1341 break;
1342
1343 bit = ptr;
1344 }
1345
1346 t = malloc(sizeof(*t));
1347 memset(t, 0, sizeof(*t));
1348 t->bit = bit;
1349
1350 trace_to_cpu(bit);
1351 t->next = trace_list;
1352 trace_list = t;
1353
1354 events++;
1355 } while (!is_done() && events < rb_batch);
1356
1357 return events;
1358}
1359
1360static int do_stdin(void)
1361{
1362 int fd;
1363
1364 fd = dup(STDIN_FILENO);
1365 do {
1366 int events;
1367
1368 events = read_sort_events(fd);
1369 if (!events)
1370 break;
1371
1372 if (sort_entries() == -1)
1373 break;
1374
1375 show_entries_rb(1);
1376 } while (1);
1377
1378 close(fd);
1379 return 0;
1380}
1381
1382static void flush_output(void)
1383{
1384 fflush(ofp);
1385}
1386
1387static void handle_sigint(int sig)
1388{
1389 done = 1;
1390 flush_output();
1391}
1392
1393/*
1394 * Extract start and duration times from a string, allowing
1395 * us to specify a time interval of interest within a trace.
1396 * Format: "duration" (start is zero) or "start:duration".
1397 */
1398static int find_stopwatch_interval(char *string)
1399{
1400 double value;
1401 char *sp;
1402
1403 value = strtod(string, &sp);
1404 if (sp == string) {
1405 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
1406 return 1;
1407 }
1408 if (*sp == ':') {
1409 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
1410 string = sp + 1;
1411 value = strtod(string, &sp);
1412 if (sp == string || *sp != '\0') {
1413 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
1414 string);
1415 return 1;
1416 }
1417 } else if (*sp != '\0') {
1418 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
1419 return 1;
1420 }
1421 stopwatch_end = stopwatch_start + DOUBLE_TO_NANO_ULL(value);
1422 return 0;
1423}
1424
1425static void usage(char *prog)
1426{
1427 fprintf(stderr, "Usage: %s "
1428 "[-i <name>] [-o <output>] [-s] [-w N[:n]] <name>...\n",
1429 prog);
1430}
1431
1432int main(int argc, char *argv[])
1433{
1434 char *ofp_buffer;
1435 int c, ret, mode;
1436 int per_device_and_cpu_stats = 1;
1437
1438 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1439 switch (c) {
1440 case 'i':
1441 if (!strcmp(optarg, "-") && !pipeline)
1442 pipeline = 1;
1443 else if (resize_devices(optarg) != 0)
1444 return 1;
1445 break;
1446 case 'o':
1447 output_name = optarg;
1448 break;
1449 case 'b':
1450 rb_batch = atoi(optarg);
1451 if (rb_batch <= 0)
1452 rb_batch = RB_BATCH_DEFAULT;
1453 break;
1454 case 's':
1455 per_process_stats = 1;
1456 break;
1457 case 't':
1458 track_ios = 1;
1459 break;
1460 case 'q':
1461 per_device_and_cpu_stats = 0;
1462 break;
1463 case 'w':
1464 if (find_stopwatch_interval(optarg) != 0)
1465 return 1;
1466 break;
1467 default:
1468 usage(argv[0]);
1469 return 1;
1470 }
1471 }
1472
1473 while (optind < argc) {
1474 if (!strcmp(argv[optind], "-") && !pipeline)
1475 pipeline = 1;
1476 else if (resize_devices(argv[optind]) != 0)
1477 return 1;
1478 optind++;
1479 }
1480
1481 if (!pipeline && !ndevices) {
1482 usage(argv[0]);
1483 return 1;
1484 }
1485
1486 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
1487 memset(&rb_track_root, 0, sizeof(rb_track_root));
1488
1489 signal(SIGINT, handle_sigint);
1490 signal(SIGHUP, handle_sigint);
1491 signal(SIGTERM, handle_sigint);
1492
1493 setlocale(LC_NUMERIC, "en_US");
1494
1495 if (!output_name) {
1496 ofp = fdopen(STDOUT_FILENO, "w");
1497 mode = _IOLBF;
1498 } else {
1499 char ofname[128];
1500
1501 snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name);
1502 ofp = fopen(ofname, "w");
1503 mode = _IOFBF;
1504 }
1505
1506 if (!ofp) {
1507 perror("fopen");
1508 return 1;
1509 }
1510
1511 ofp_buffer = malloc(4096);
1512 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
1513 perror("setvbuf");
1514 return 1;
1515 }
1516
1517 if (pipeline)
1518 ret = do_stdin();
1519 else
1520 ret = do_file();
1521
1522 if (per_process_stats)
1523 show_process_stats();
1524
1525 if (per_device_and_cpu_stats)
1526 show_device_and_cpu_stats();
1527
1528 flush_output();
1529 return ret;
1530}
Block queue io tracer