projects / blktrace.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
[PATCH] Add TODO of action items before 1.0 release
[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
75 int ncpus;
76 struct per_cpu_info *cpus;
77};
78
79struct per_process_info {
80 char name[16];
81 __u32 pid;
82 struct io_stats io_stats;
83 struct per_process_info *hash_next, *list_next;
84
85 /*
86 * individual io stats
87 */
88 unsigned long long longest_allocation_wait[2];
89 unsigned long long longest_dispatch_wait[2];
90 unsigned long long longest_completion_wait[2];
91};
92
93#define PPI_HASH_SHIFT (8)
94static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT];
95static struct per_process_info *ppi_list;
96
97#define S_OPTS "i:o:b:stqw:"
98static struct option l_opts[] = {
99 {
100 .name = "input",
101 .has_arg = 1,
102 .flag = NULL,
103 .val = 'i'
104 },
105 {
106 .name = "output",
107 .has_arg = 1,
108 .flag = NULL,
109 .val = 'o'
110 },
111 {
112 .name = "batch",
113 .has_arg = 1,
114 .flag = NULL,
115 .val = 'b'
116 },
117 {
118 .name = "per program stats",
119 .has_arg = 0,
120 .flag = NULL,
121 .val = 's'
122 },
123 {
124 .name = "track ios",
125 .has_arg = 0,
126 .flag = NULL,
127 .val = 't'
128 },
129 {
130 .name = "quiet",
131 .has_arg = 0,
132 .flag = NULL,
133 .val = 'q'
134 },
135 {
136 .name = "stopwatch",
137 .has_arg = 1,
138 .flag = NULL,
139 .val = 'w'
140 },
141 {
142 .name = NULL,
143 .has_arg = 0,
144 .flag = NULL,
145 .val = 0
146 }
147};
148
149static struct rb_root rb_sort_root;
150static struct rb_root rb_track_root;
151
152/*
153 * for sorting the displayed output
154 */
155struct trace {
156 struct blk_io_trace *bit;
157 struct rb_node rb_node;
158};
159
160/*
161 * for tracking individual ios
162 */
163struct io_track {
164 struct rb_node rb_node;
165
166 dev_t device;
167 __u64 sector;
168 __u32 pid;
169 unsigned long long allocation_time;
170 unsigned long long queue_time;
171 unsigned long long dispatch_time;
172 unsigned long long completion_time;
173};
174
175static int ndevices;
176static struct per_dev_info *devices;
177static char *get_dev_name(struct per_dev_info *, char *, int);
178
179static FILE *ofp;
180static char *output_name;
181
182static unsigned long long genesis_time;
183static unsigned long long stopwatch_start; /* start from zero by default */
184static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
185
186static int per_process_stats;
187static int track_ios;
188
189#define RB_BATCH_DEFAULT (1024)
190static int rb_batch = RB_BATCH_DEFAULT;
191
192static int pipeline;
193
194#define is_done() (*(volatile int *)(&done))
195static volatile int done;
196
197static inline unsigned long hash_long(unsigned long val)
198{
199#if __WORDSIZE == 32
200 val *= 0x9e370001UL;
201#elif __WORDSIZE == 64
202 val *= 0x9e37fffffffc0001UL;
203#else
204#error unknown word size
205#endif
206
207 return val >> (__WORDSIZE - PPI_HASH_SHIFT);
208}
209
210static inline void add_process_to_hash(struct per_process_info *ppi)
211{
212 const int hash_idx = hash_long(ppi->pid);
213
214 ppi->hash_next = ppi_hash[hash_idx];
215 ppi_hash[hash_idx] = ppi;
216}
217
218static inline void add_process_to_list(struct per_process_info *ppi)
219{
220 ppi->list_next = ppi_list;
221 ppi_list = ppi;
222}
223
224static struct per_process_info *find_process_by_pid(__u32 pid)
225{
226 const int hash_idx = hash_long(pid);
227 struct per_process_info *ppi;
228
229 ppi = ppi_hash[hash_idx];
230 while (ppi) {
231 if (ppi->pid == pid)
232 return ppi;
233
234 ppi = ppi->hash_next;
235 }
236
237 return NULL;
238}
239
240static inline int trace_rb_insert(struct trace *t)
241{
242 struct rb_node **p = &rb_sort_root.rb_node;
243 struct rb_node *parent = NULL;
244 struct trace *__t;
245
246 if (genesis_time == 0 || t->bit->time < genesis_time)
247 genesis_time = t->bit->time;
248
249 while (*p) {
250 parent = *p;
251 __t = rb_entry(parent, struct trace, rb_node);
252
253 if (t->bit->time < __t->bit->time)
254 p = &(*p)->rb_left;
255 else if (t->bit->time > __t->bit->time)
256 p = &(*p)->rb_right;
257 else if (t->bit->device < __t->bit->device)
258 p = &(*p)->rb_left;
259 else if (t->bit->device > __t->bit->device)
260 p = &(*p)->rb_right;
261 else if (t->bit->sequence < __t->bit->sequence)
262 p = &(*p)->rb_left;
263 else if (t->bit->sequence > __t->bit->sequence)
264 p = &(*p)->rb_right;
265 else if (t->bit->device == __t->bit->device) {
266 fprintf(stderr,
267 "sequence alias (%d) on device %d,%d!\n",
268 t->bit->sequence,
269 MAJOR(t->bit->device), MINOR(t->bit->device));
270 return 1;
271 }
272 }
273
274 rb_link_node(&t->rb_node, parent, p);
275 rb_insert_color(&t->rb_node, &rb_sort_root);
276 return 0;
277}
278
279static inline int track_rb_insert(struct io_track *iot)
280{
281 struct rb_node **p = &rb_track_root.rb_node;
282 struct rb_node *parent = NULL;
283 struct io_track *__iot;
284
285 while (*p) {
286 parent = *p;
287
288 __iot = rb_entry(parent, struct io_track, rb_node);
289
290 if (iot->device < __iot->device)
291 p = &(*p)->rb_left;
292 else if (iot->device > __iot->device)
293 p = &(*p)->rb_right;
294 else if (iot->sector < __iot->sector)
295 p = &(*p)->rb_left;
296 else if (iot->sector > __iot->sector)
297 p = &(*p)->rb_right;
298 else {
299 fprintf(stderr,
300 "sector alias (%llu) on device %d,%d!\n",
301 iot->sector,
302 MAJOR(iot->device), MINOR(iot->device));
303 return 1;
304 }
305 }
306
307 rb_link_node(&iot->rb_node, parent, p);
308 rb_insert_color(&iot->rb_node, &rb_track_root);
309 return 0;
310}
311
312static struct io_track *__find_track(dev_t device, __u64 sector)
313{
314 struct rb_node **p = &rb_track_root.rb_node;
315 struct rb_node *parent = NULL;
316 struct io_track *__iot;
317
318 while (*p) {
319 parent = *p;
320
321 __iot = rb_entry(parent, struct io_track, rb_node);
322
323 if (device < __iot->device)
324 p = &(*p)->rb_left;
325 else if (device > __iot->device)
326 p = &(*p)->rb_right;
327 else if (sector < __iot->sector)
328 p = &(*p)->rb_left;
329 else if (sector > __iot->sector)
330 p = &(*p)->rb_right;
331 else
332 return __iot;
333 }
334
335 return NULL;
336}
337
338static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector)
339{
340 struct io_track *iot;
341
342 iot = __find_track(device, sector);
343 if (!iot) {
344 iot = malloc(sizeof(*iot));
345 iot->pid = pid;
346 iot->device = device;
347 iot->sector = sector;
348 track_rb_insert(iot);
349 }
350
351 return iot;
352}
353
354static void log_track_merge(struct blk_io_trace *t)
355{
356 struct io_track *iot;
357
358 if (!track_ios)
359 return;
360 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
361 return;
362
363 /*
364 * this can happen if we lose events, so don't print an error
365 */
366 iot = __find_track(t->device, t->sector - (t->bytes >> 10));
367 if (iot) {
368 rb_erase(&iot->rb_node, &rb_track_root);
369 iot->sector -= t->bytes >> 10;
370 track_rb_insert(iot);
371 }
372}
373
374static void log_track_getrq(struct blk_io_trace *t)
375{
376 struct io_track *iot;
377
378 if (!track_ios)
379 return;
380
381 iot = find_track(t->pid, t->device, t->sector);
382 iot->allocation_time = t->time;
383}
384
385
386/*
387 * return time between rq allocation and queue
388 */
389static unsigned long long log_track_queue(struct blk_io_trace *t)
390{
391 unsigned long long elapsed;
392 struct io_track *iot;
393
394 if (!track_ios)
395 return -1;
396
397 iot = find_track(t->pid, t->device, t->sector);
398 iot->queue_time = t->time;
399 elapsed = iot->queue_time - iot->allocation_time;
400
401 if (per_process_stats) {
402 struct per_process_info *ppi = find_process_by_pid(iot->pid);
403 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
404
405 if (ppi && elapsed > ppi->longest_allocation_wait[w])
406 ppi->longest_allocation_wait[w] = elapsed;
407 }
408
409 return elapsed;
410}
411
412/*
413 * return time between queue and issue
414 */
415static unsigned long long log_track_issue(struct blk_io_trace *t)
416{
417 unsigned long long elapsed;
418 struct io_track *iot;
419
420 if (!track_ios)
421 return -1;
422 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
423 return -1;
424
425 /*
426 * this can happen if we lose events, so don't print an error
427 */
428 iot = __find_track(t->device, t->sector);
429 if (!iot)
430 return -1;
431
432 iot->dispatch_time = t->time;
433 elapsed = iot->dispatch_time - iot->queue_time;
434
435 if (per_process_stats) {
436 struct per_process_info *ppi = find_process_by_pid(iot->pid);
437 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
438
439 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
440 ppi->longest_dispatch_wait[w] = elapsed;
441 }
442
443 return elapsed;
444}
445
446/*
447 * return time between dispatch and complete
448 */
449static unsigned long long log_track_complete(struct blk_io_trace *t)
450{
451 unsigned long long elapsed;
452 struct io_track *iot;
453
454 if (!track_ios)
455 return -1;
456 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
457 return -1;
458
459 iot = __find_track(t->device, t->sector);
460 if (!iot)
461 return -1;
462
463 iot->completion_time = t->time;
464 elapsed = iot->completion_time - iot->dispatch_time;
465
466 if (per_process_stats) {
467 struct per_process_info *ppi = find_process_by_pid(iot->pid);
468 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
469
470 if (ppi && elapsed > ppi->longest_completion_wait[w])
471 ppi->longest_completion_wait[w] = elapsed;
472 }
473
474 /*
475 * kill the trace, we don't need it after completion
476 */
477 rb_erase(&iot->rb_node, &rb_track_root);
478 free(iot);
479
480 return elapsed;
481}
482
483
484static struct io_stats *find_process_io_stats(__u32 pid, char *name)
485{
486 struct per_process_info *ppi = find_process_by_pid(pid);
487
488 if (!ppi) {
489 ppi = malloc(sizeof(*ppi));
490 memset(ppi, 0, sizeof(*ppi));
491 strncpy(ppi->name, name, sizeof(ppi->name));
492 ppi->pid = pid;
493 add_process_to_hash(ppi);
494 add_process_to_list(ppi);
495 }
496
497 return &ppi->io_stats;
498}
499
500
501static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
502{
503 struct per_cpu_info *cpus = pdi->cpus;
504 int ncpus = pdi->ncpus;
505 int new_count = cpu + 1;
506 int new_space, size;
507 char *new_start;
508
509 size = new_count * sizeof(struct per_cpu_info);
510 cpus = realloc(cpus, size);
511 if (!cpus) {
512 char name[20];
513 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
514 get_dev_name(pdi, name, sizeof(name)), size);
515 exit(1);
516 }
517
518 new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
519 new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
520 memset(new_start, 0, new_space);
521
522 pdi->ncpus = new_count;
523 pdi->cpus = cpus;
524}
525
526static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
527{
528 if (cpu >= pdi->ncpus)
529 resize_cpu_info(pdi, cpu);
530 return &pdi->cpus[cpu];
531}
532
533
534static int resize_devices(char *name)
535{
536 int size = (ndevices + 1) * sizeof(struct per_dev_info);
537
538 devices = realloc(devices, size);
539 if (!devices) {
540 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
541 return 1;
542 }
543 memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
544 devices[ndevices].name = name;
545 ndevices++;
546 return 0;
547}
548
549static struct per_dev_info *get_dev_info(dev_t id, int create)
550{
551 int i;
552
553 for (i = 0; i < ndevices; i++)
554 if (devices[i].id == id)
555 return &devices[i];
556 if (!create)
557 return NULL;
558 if (resize_devices(NULL) != 0)
559 return NULL;
560 return &devices[ndevices-1];
561}
562
563static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
564{
565 if (pdi->name)
566 snprintf(buffer, size, "%s", pdi->name);
567 else
568 snprintf(buffer, size, "%d,%d", MAJOR(pdi->id), MINOR(pdi->id));
569 return buffer;
570}
571
572
573static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
574{
575 unsigned long long this = bit->time;
576 unsigned long long last = pdi->last_reported_time;
577
578 pdi->backwards = (this < last) ? 'B' : ' ';
579 pdi->last_reported_time = this;
580}
581
582
583static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
584 int rw)
585{
586 if (rw) {
587 ios->mwrites++;
588 ios->qwrite_kb += t->bytes >> 10;
589 } else {
590 ios->mreads++;
591 ios->qread_kb += t->bytes >> 10;
592 }
593}
594
595static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
596 int rw)
597{
598 __account_m(&pci->io_stats, t, rw);
599
600 if (per_process_stats) {
601 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
602
603 __account_m(ios, t, rw);
604 }
605}
606
607static inline void __account_q(struct io_stats *ios, struct blk_io_trace *t,
608 int rw)
609{
610 if (rw) {
611 ios->qwrites++;
612 ios->qwrite_kb += t->bytes >> 10;
613 } else {
614 ios->qreads++;
615 ios->qread_kb += t->bytes >> 10;
616 }
617}
618
619static inline void account_q(struct blk_io_trace *t, struct per_cpu_info *pci,
620 int rw)
621{
622 __account_q(&pci->io_stats, t, rw);
623
624 if (per_process_stats) {
625 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
626
627 __account_q(ios, t, rw);
628 }
629}
630
631static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes)
632{
633 if (rw) {
634 ios->cwrites++;
635 ios->cwrite_kb += bytes >> 10;
636 } else {
637 ios->creads++;
638 ios->cread_kb += bytes >> 10;
639 }
640}
641
642static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
643 int rw, int bytes)
644{
645 __account_c(&pci->io_stats, rw, bytes);
646
647 if (per_process_stats) {
648 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
649
650 __account_c(ios, rw, bytes);
651 }
652}
653
654static inline void __account_i(struct io_stats *ios, int rw, unsigned int bytes)
655{
656 if (rw) {
657 ios->iwrites++;
658 ios->iwrite_kb += bytes >> 10;
659 } else {
660 ios->ireads++;
661 ios->iread_kb += bytes >> 10;
662 }
663}
664
665static inline void account_i(struct blk_io_trace *t, struct per_cpu_info *pci,
666 int rw)
667{
668 __account_i(&pci->io_stats, rw, t->bytes);
669
670 if (per_process_stats) {
671 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
672
673 __account_i(ios, rw, t->bytes);
674 }
675}
676
677static inline void __account_unplug(struct io_stats *ios, int timer)
678{
679 if (timer)
680 ios->timer_unplugs++;
681 else
682 ios->io_unplugs++;
683}
684
685static inline void account_unplug(struct blk_io_trace *t,
686 struct per_cpu_info *pci, int timer)
687{
688 __account_unplug(&pci->io_stats, timer);
689
690 if (per_process_stats) {
691 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
692
693 __account_unplug(ios, timer);
694 }
695}
696
697static void output(struct per_cpu_info *pci, char *s)
698{
699 fprintf(ofp, "%s", s);
700}
701
702static char hstring[256];
703static char tstring[256];
704
705static inline char *setup_header(struct per_cpu_info *pci,
706 struct blk_io_trace *t, char *act)
707{
708 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
709 int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER);
710 int s = t->action & BLK_TC_ACT(BLK_TC_SYNC);
711 char rwbs[4];
712 int i = 0;
713
714 if (w)
715 rwbs[i++] = 'W';
716 else
717 rwbs[i++] = 'R';
718 if (b)
719 rwbs[i++] = 'B';
720 if (s)
721 rwbs[i++] = 'S';
722
723 rwbs[i] = '\0';
724
725 sprintf(hstring, "%3d,%-3d %2d %8ld %5Lu.%09Lu %5u %2s %3s",
726 MAJOR(t->device), MINOR(t->device), pci->cpu,
727 (unsigned long)t->sequence, SECONDS(t->time),
728 NANO_SECONDS(t->time), t->pid, act, rwbs);
729
730 return hstring;
731}
732
733static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t,
734 char *act)
735{
736 unsigned long long elapsed = log_track_complete(t);
737
738 if (elapsed != -1ULL) {
739 unsigned long usec = elapsed / 1000;
740
741 sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n",
742 setup_header(pci, t, act),
743 (unsigned long long)t->sector, t->bytes >> 9,
744 usec, t->error);
745 } else {
746 sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act),
747 (unsigned long long)t->sector, t->bytes >> 9, t->error);
748 }
749
750 output(pci, tstring);
751}
752
753static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
754 char *act)
755{
756 unsigned long long elapsed = log_track_queue(t);
757
758 if (elapsed != -1ULL) {
759 unsigned long usec = elapsed / 1000;
760
761 sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n",
762 setup_header(pci, t, act),
763 (unsigned long long)t->sector, t->bytes >> 9,
764 usec, t->comm);
765 } else {
766 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
767 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
768 }
769 output(pci, tstring);
770}
771
772static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t,
773 char *act)
774{
775 unsigned long long elapsed = log_track_issue(t);
776
777 if (elapsed != -1ULL) {
778 double usec = (double) elapsed / 1000;
779
780 sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n",
781 setup_header(pci, t, act),
782 (unsigned long long)t->sector, t->bytes >> 9,
783 usec, t->comm);
784 } else {
785 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
786 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
787 }
788
789 output(pci, tstring);
790}
791
792static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
793 char *act)
794{
795 log_track_merge(t);
796
797 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
798 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
799 output(pci, tstring);
800}
801
802static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
803 char *act)
804{
805 sprintf(tstring,"%s [%s]\n", setup_header(pci, t, act), t->comm);
806 output(pci, tstring);
807}
808
809static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
810 char *act)
811{
812 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
813 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
814 output(pci, tstring);
815}
816
817static int log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
818 char *act)
819{
820 __u64 *depth;
821 int len;
822
823 len = sprintf(tstring,"%s [%s] ", setup_header(pci, t, act), t->comm);
824 depth = (__u64 *) ((char *) t + sizeof(*t));
825 sprintf(tstring + len, "%u\n", (unsigned int) be64_to_cpu(*depth));
826 output(pci, tstring);
827
828 return 0;
829}
830
831static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
832{
833 unsigned char *buf;
834 int i;
835
836 sprintf(tstring,"%s ", setup_header(pci, t, act));
837 output(pci, tstring);
838
839 buf = (unsigned char *) t + sizeof(*t);
840 for (i = 0; i < t->pdu_len; i++) {
841 sprintf(tstring,"%02x ", buf[i]);
842 output(pci, tstring);
843 }
844
845 if (act[0] == 'C') {
846 sprintf(tstring,"[%d]\n", t->error);
847 output(pci, tstring);
848 } else {
849 sprintf(tstring,"[%s]\n", t->comm);
850 output(pci, tstring);
851 }
852 return 0;
853}
854
855static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
856{
857 int ret = 0;
858
859 switch (t->action & 0xffff) {
860 case __BLK_TA_QUEUE:
861 log_generic(pci, t, "Q");
862 break;
863 case __BLK_TA_GETRQ:
864 log_generic(pci, t, "G");
865 break;
866 case __BLK_TA_SLEEPRQ:
867 log_generic(pci, t, "S");
868 break;
869 case __BLK_TA_REQUEUE:
870 log_generic(pci, t, "R");
871 break;
872 case __BLK_TA_ISSUE:
873 ret = log_pc(pci, t, "D");
874 break;
875 case __BLK_TA_COMPLETE:
876 log_pc(pci, t, "C");
877 break;
878 default:
879 fprintf(stderr, "Bad pc action %x\n", t->action);
880 ret = 1;
881 break;
882 }
883
884 return ret;
885}
886
887static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci)
888{
889 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
890 int act = t->action & 0xffff;
891
892 switch (act) {
893 case __BLK_TA_QUEUE:
894 account_q(t, pci, w);
895 log_queue(pci, t, "Q");
896 break;
897 case __BLK_TA_BACKMERGE:
898 account_m(t, pci, w);
899 log_merge(pci, t, "M");
900 break;
901 case __BLK_TA_FRONTMERGE:
902 account_m(t, pci, w);
903 log_merge(pci, t, "F");
904 break;
905 case __BLK_TA_GETRQ:
906 log_track_getrq(t);
907 log_generic(pci, t, "G");
908 break;
909 case __BLK_TA_SLEEPRQ:
910 log_generic(pci, t, "S");
911 break;
912 case __BLK_TA_REQUEUE:
913 account_c(t, pci, w, -t->bytes);
914 log_queue(pci, t, "R");
915 break;
916 case __BLK_TA_ISSUE:
917 account_i(t, pci, w);
918 log_issue(pci, t, "D");
919 break;
920 case __BLK_TA_COMPLETE:
921 account_c(t, pci, w, t->bytes);
922 log_complete(pci, t, "C");
923 break;
924 case __BLK_TA_PLUG:
925 log_action(pci, t, "P");
926 break;
927 case __BLK_TA_UNPLUG_IO:
928 account_unplug(t, pci, 0);
929 log_unplug(pci, t, "U");
930 break;
931 case __BLK_TA_UNPLUG_TIMER:
932 account_unplug(t, pci, 1);
933 log_unplug(pci, t, "UT");
934 break;
935 default:
936 fprintf(stderr, "Bad fs action %x\n", t->action);
937 break;
938 }
939}
940
941static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
942 struct per_dev_info *pdi)
943{
944 int ret = 0;
945
946 if (t->action & BLK_TC_ACT(BLK_TC_PC))
947 ret = dump_trace_pc(t, pci);
948 else
949 dump_trace_fs(t, pci);
950
951 pdi->events++;
952 return ret;
953}
954
955static void dump_io_stats(struct io_stats *ios, char *msg)
956{
957 fprintf(ofp, "%s\n", msg);
958
959 fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
960 fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
961
962 fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
963 fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
964 fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
965 fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
966 fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
967 fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
968 fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
969 fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
970}
971
972static void dump_wait_stats(struct per_process_info *ppi)
973{
974 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
975 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
976 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
977 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
978 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
979 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
980
981 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
982 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
983 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
984 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
985 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
986 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
987}
988
989static void show_process_stats(void)
990{
991 struct per_process_info *ppi;
992
993 ppi = ppi_list;
994 while (ppi) {
995 dump_io_stats(&ppi->io_stats, ppi->name);
996 dump_wait_stats(ppi);
997 ppi = ppi->list_next;
998 }
999
1000 fprintf(ofp, "\n");
1001}
1002
1003static void show_device_and_cpu_stats(void)
1004{
1005 struct per_dev_info *pdi;
1006 struct per_cpu_info *pci;
1007 struct io_stats total, *ios;
1008 int i, j, pci_events;
1009 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1010 char name[32];
1011
1012 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1013
1014 memset(&total, 0, sizeof(total));
1015 pci_events = 0;
1016
1017 if (i > 0)
1018 fprintf(ofp, "\n");
1019
1020 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1021 if (!pci->nelems)
1022 continue;
1023
1024 ios = &pci->io_stats;
1025 total.qreads += ios->qreads;
1026 total.qwrites += ios->qwrites;
1027 total.creads += ios->creads;
1028 total.cwrites += ios->cwrites;
1029 total.mreads += ios->mreads;
1030 total.mwrites += ios->mwrites;
1031 total.ireads += ios->ireads;
1032 total.iwrites += ios->iwrites;
1033 total.qread_kb += ios->qread_kb;
1034 total.qwrite_kb += ios->qwrite_kb;
1035 total.cread_kb += ios->cread_kb;
1036 total.cwrite_kb += ios->cwrite_kb;
1037 total.iread_kb += ios->iread_kb;
1038 total.iwrite_kb += ios->iwrite_kb;
1039 total.timer_unplugs += ios->timer_unplugs;
1040 total.io_unplugs += ios->io_unplugs;
1041
1042 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1043 j, get_dev_name(pdi, name, sizeof(name)));
1044 dump_io_stats(ios, line);
1045 pci_events++;
1046 }
1047
1048 if (pci_events > 1) {
1049 fprintf(ofp, "\n");
1050 snprintf(line, sizeof(line) - 1, "Total (%s):",
1051 get_dev_name(pdi, name, sizeof(name)));
1052 dump_io_stats(&total, line);
1053 }
1054
1055 fprintf(ofp, "Events (%s): %'Lu\n",
1056 get_dev_name(pdi, line, sizeof(line)), pdi->events);
1057 }
1058}
1059
1060static struct blk_io_trace *find_trace(void *p, unsigned long offset, int nr)
1061{
1062 unsigned long max_offset = min(offset,nr * sizeof(struct blk_io_trace));
1063 unsigned long off;
1064 struct blk_io_trace *bit;
1065 __u32 magic;
1066
1067 for (off = 0; off < max_offset; off++) {
1068 bit = p + off;
1069
1070 magic = be32_to_cpu(bit->magic);
1071 if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
1072 return bit;
1073 }
1074
1075 return NULL;
1076}
1077
1078static int sort_entries(void *traces, unsigned long offset, int nr,
1079 struct per_dev_info *fpdi, struct per_cpu_info *fpci)
1080{
1081 struct per_dev_info *pdi;
1082 struct per_cpu_info *pci;
1083 struct blk_io_trace *bit;
1084 struct trace *t;
1085 void *start = traces;
1086
1087 while (traces - start <= offset - sizeof(*bit)) {
1088 if (!nr)
1089 break;
1090
1091 bit = find_trace(traces, offset - (traces - start), nr);
1092 if (!bit)
1093 break;
1094
1095 t = malloc(sizeof(*t));
1096 if (!t) {
1097 fprintf(stderr, "Out of memory, seq %d on dev %d,%d\n",
1098 bit->sequence,
1099 MAJOR(bit->device), MINOR(bit->device));
1100 return -1;
1101 }
1102 t->bit = bit;
1103 memset(&t->rb_node, 0, sizeof(t->rb_node));
1104
1105 trace_to_cpu(bit);
1106
1107 if (verify_trace(bit)) {
1108 free(t);
1109 break;
1110 }
1111
1112 pdi = fpdi ? fpdi : get_dev_info(bit->device, 1);
1113 pdi->id = bit->device;
1114 pci = fpci ? fpci : get_cpu_info(pdi, bit->cpu);
1115 pci->cpu = bit->cpu;
1116 pci->nelems++;
1117
1118 if (trace_rb_insert(t)) {
1119 free(t);
1120 return -1;
1121 }
1122
1123 traces += sizeof(*bit) + bit->pdu_len;
1124 nr--;
1125 }
1126
1127 return 0;
1128}
1129
1130static void free_entries_rb(void)
1131{
1132 struct rb_node *n;
1133
1134 while ((n = rb_first(&rb_sort_root)) != NULL) {
1135 struct trace *t = rb_entry(n, struct trace, rb_node);
1136
1137 rb_erase(&t->rb_node, &rb_sort_root);
1138 free(t);
1139 }
1140}
1141
1142static void show_entries_rb(void)
1143{
1144 struct per_dev_info *pdi;
1145 struct blk_io_trace *bit;
1146 struct rb_node *n;
1147 struct trace *t;
1148 int cpu;
1149
1150 n = rb_first(&rb_sort_root);
1151 if (!n)
1152 return;
1153
1154 do {
1155 t = rb_entry(n, struct trace, rb_node);
1156 bit = t->bit;
1157
1158 pdi = get_dev_info(bit->device, 0);
1159 if (!pdi) {
1160 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
1161 MAJOR(bit->device), MINOR(bit->device));
1162 break;
1163 }
1164 cpu = bit->cpu;
1165 if (cpu > pdi->ncpus) {
1166 fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
1167 cpu, MAJOR(bit->device), MINOR(bit->device));
1168 break;
1169 }
1170
1171 bit->time -= genesis_time;
1172 if (bit->time < stopwatch_start)
1173 continue;
1174 if (bit->time >= stopwatch_end)
1175 break;
1176
1177 check_time(pdi, bit);
1178
1179 if (dump_trace(bit, &pdi->cpus[cpu], pdi))
1180 break;
1181
1182 } while ((n = rb_next(n)) != NULL);
1183}
1184
1185static int read_data(int fd, void *buffer, int bytes, int block)
1186{
1187 int ret, bytes_left, fl;
1188 void *p;
1189
1190 fl = fcntl(fd, F_GETFL);
1191
1192 if (!block)
1193 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
1194 else
1195 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
1196
1197 bytes_left = bytes;
1198 p = buffer;
1199 while (bytes_left > 0) {
1200 ret = read(fd, p, bytes_left);
1201 if (!ret)
1202 return 1;
1203 else if (ret < 0) {
1204 if (errno != EAGAIN)
1205 perror("read");
1206 return -1;
1207 } else {
1208 p += ret;
1209 bytes_left -= ret;
1210 }
1211 }
1212
1213 return 0;
1214}
1215
1216static int do_file(void)
1217{
1218 struct per_dev_info *pdi;
1219 int i, j, nfiles = 0;
1220
1221 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1222 for (j = 0;; j++, nfiles++) {
1223 struct per_cpu_info *pci;
1224 struct stat st;
1225 void *tb;
1226
1227 pci = get_cpu_info(pdi, j);
1228 pci->cpu = j;
1229
1230 snprintf(pci->fname, sizeof(pci->fname)-1,
1231 "%s_out.%d", pdi->name, j);
1232 if (stat(pci->fname, &st) < 0)
1233 break;
1234 if (!st.st_size)
1235 continue;
1236
1237 printf("Processing %s\n", pci->fname);
1238
1239 tb = malloc(st.st_size);
1240 if (!tb) {
1241 fprintf(stderr, "Out of memory, skip file %s\n",
1242 pci->fname);
1243 continue;
1244 }
1245
1246 pci->fd = open(pci->fname, O_RDONLY);
1247 if (pci->fd < 0) {
1248 perror(pci->fname);
1249 free(tb);
1250 continue;
1251 }
1252
1253 if (read_data(pci->fd, tb, st.st_size, 1)) {
1254 close(pci->fd);
1255 free(tb);
1256 continue;
1257 }
1258
1259 if (sort_entries(tb, st.st_size, ~0U, pdi, pci) == -1) {
1260 close(pci->fd);
1261 free(tb);
1262 continue;
1263 }
1264
1265 printf("Completed %s (CPU%d %d, entries)\n",
1266 pci->fname, j, pci->nelems);
1267 close(pci->fd);
1268 }
1269 }
1270
1271 if (!nfiles) {
1272 fprintf(stderr, "No files found\n");
1273 return 1;
1274 }
1275
1276 show_entries_rb();
1277 return 0;
1278}
1279
1280static void resize_buffer(void **buffer, long *size, long offset)
1281{
1282 long old_size = *size;
1283
1284 if (*size == 0)
1285 *size = 64 * sizeof(struct blk_io_trace);
1286
1287 *size *= 2;
1288 *buffer = realloc(*buffer, *size);
1289
1290 if (old_size)
1291 memset(*buffer + offset, 0, *size - old_size);
1292}
1293
1294static int read_sort_events(int fd, void **buffer, long *max_offset)
1295{
1296 long offset;
1297 int events;
1298
1299 events = offset = 0;
1300 do {
1301 struct blk_io_trace *t;
1302 int pdu_len;
1303 __u32 magic;
1304
1305 if (*max_offset - offset < sizeof(*t))
1306 resize_buffer(buffer, max_offset, offset);
1307
1308 if (read_data(fd, *buffer + offset, sizeof(*t), !events))
1309 break;
1310
1311 t = *buffer + offset;
1312 offset += sizeof(*t);
1313
1314 magic = be32_to_cpu(t->magic);
1315 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1316 fprintf(stderr, "Bad magic %x\n", magic);
1317 break;
1318 }
1319
1320 pdu_len = be16_to_cpu(t->pdu_len);
1321 if (pdu_len) {
1322 if (*max_offset - offset <= pdu_len)
1323 resize_buffer(buffer, max_offset, offset);
1324
1325 if (read_data(fd, *buffer + offset, pdu_len, 1))
1326 break;
1327
1328 offset += pdu_len;
1329 }
1330
1331 events++;
1332 } while (!is_done() && events < rb_batch);
1333
1334 return events;
1335}
1336
1337static int do_stdin(void)
1338{
1339 int fd;
1340 void *ptr = NULL;
1341 long max_offset;
1342
1343 fd = dup(STDIN_FILENO);
1344 max_offset = 0;
1345 do {
1346 int events;
1347
1348 events = read_sort_events(fd, &ptr, &max_offset);
1349 if (!events)
1350 break;
1351
1352 if (sort_entries(ptr, ~0UL, events, NULL, NULL) == -1)
1353 break;
1354
1355 show_entries_rb();
1356 free_entries_rb();
1357 } while (1);
1358
1359 if (ptr)
1360 free(ptr);
1361
1362 close(fd);
1363 return 0;
1364}
1365
1366static void flush_output(void)
1367{
1368 fflush(ofp);
1369}
1370
1371static void handle_sigint(int sig)
1372{
1373 done = 1;
1374 flush_output();
1375}
1376
1377/*
1378 * Extract start and duration times from a string, allowing
1379 * us to specify a time interval of interest within a trace.
1380 * Format: "duration" (start is zero) or "start:duration".
1381 */
1382static int find_stopwatch_interval(char *string)
1383{
1384 double value;
1385 char *sp;
1386
1387 value = strtod(string, &sp);
1388 if (sp == string) {
1389 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
1390 return 1;
1391 }
1392 if (*sp == ':') {
1393 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
1394 string = sp + 1;
1395 value = strtod(string, &sp);
1396 if (sp == string || *sp != '\0') {
1397 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
1398 string);
1399 return 1;
1400 }
1401 } else if (*sp != '\0') {
1402 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
1403 return 1;
1404 }
1405 stopwatch_end = stopwatch_start + DOUBLE_TO_NANO_ULL(value);
1406 return 0;
1407}
1408
1409static void usage(char *prog)
1410{
1411 fprintf(stderr, "Usage: %s "
1412 "[-i <name>] [-o <output>] [-s] [-w N[:n]] <name>...\n",
1413 prog);
1414}
1415
1416int main(int argc, char *argv[])
1417{
1418 char *ofp_buffer;
1419 int c, ret, mode;
1420 int per_device_and_cpu_stats = 1;
1421
1422 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1423 switch (c) {
1424 case 'i':
1425 if (!strcmp(optarg, "-") && !pipeline)
1426 pipeline = 1;
1427 else if (resize_devices(optarg) != 0)
1428 return 1;
1429 break;
1430 case 'o':
1431 output_name = optarg;
1432 break;
1433 case 'b':
1434 rb_batch = atoi(optarg);
1435 if (rb_batch <= 0)
1436 rb_batch = RB_BATCH_DEFAULT;
1437 break;
1438 case 's':
1439 per_process_stats = 1;
1440 break;
1441 case 't':
1442 track_ios = 1;
1443 break;
1444 case 'q':
1445 per_device_and_cpu_stats = 0;
1446 break;
1447 case 'w':
1448 if (find_stopwatch_interval(optarg) != 0)
1449 return 1;
1450 break;
1451 default:
1452 usage(argv[0]);
1453 return 1;
1454 }
1455 }
1456
1457 while (optind < argc) {
1458 if (!strcmp(argv[optind], "-") && !pipeline)
1459 pipeline = 1;
1460 else if (resize_devices(argv[optind]) != 0)
1461 return 1;
1462 optind++;
1463 }
1464
1465 if (!pipeline && !ndevices) {
1466 usage(argv[0]);
1467 return 1;
1468 }
1469
1470 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
1471 memset(&rb_track_root, 0, sizeof(rb_track_root));
1472
1473 signal(SIGINT, handle_sigint);
1474 signal(SIGHUP, handle_sigint);
1475 signal(SIGTERM, handle_sigint);
1476
1477 setlocale(LC_NUMERIC, "en_US");
1478
1479 if (!output_name) {
1480 ofp = fdopen(STDOUT_FILENO, "w");
1481 mode = _IOLBF;
1482 } else {
1483 char ofname[128];
1484
1485 snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name);
1486 ofp = fopen(ofname, "w");
1487 mode = _IOFBF;
1488 }
1489
1490 if (!ofp) {
1491 perror("fopen");
1492 return 1;
1493 }
1494
1495 ofp_buffer = malloc(4096);
1496 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
1497 perror("setvbuf");
1498 return 1;
1499 }
1500
1501 if (pipeline)
1502 ret = do_stdin();
1503 else
1504 ret = do_file();
1505
1506 if (per_process_stats)
1507 show_process_stats();
1508
1509 if (per_device_and_cpu_stats)
1510 show_device_and_cpu_stats();
1511
1512 flush_output();
1513 return ret;
1514}
Block queue io tracer