projects / blktrace.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
[PATCH] blkparse: improve unplug logging
[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};
54
55struct per_cpu_info {
56 int cpu;
57 int nelems;
58
59 int fd;
60 char fname[128];
61
62 struct io_stats io_stats;
63};
64
65struct per_dev_info {
66 dev_t id;
67 char *name;
68
69 int backwards;
70 unsigned long long events;
71 unsigned long long last_reported_time;
72 struct io_stats io_stats;
73
74 int ncpus;
75 struct per_cpu_info *cpus;
76};
77
78struct per_process_info {
79 char name[16];
80 __u32 pid;
81 struct io_stats io_stats;
82 struct per_process_info *hash_next, *list_next;
83
84 /*
85 * individual io stats
86 */
87 unsigned long long longest_allocation_wait[2];
88 unsigned long long longest_dispatch_wait[2];
89 unsigned long long longest_completion_wait[2];
90};
91
92#define PPI_HASH_SHIFT (8)
93static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT];
94static struct per_process_info *ppi_list;
95
96#define S_OPTS "i:o:b:stqw:"
97static struct option l_opts[] = {
98 {
99 .name = "input",
100 .has_arg = 1,
101 .flag = NULL,
102 .val = 'i'
103 },
104 {
105 .name = "output",
106 .has_arg = 1,
107 .flag = NULL,
108 .val = 'o'
109 },
110 {
111 .name = "batch",
112 .has_arg = 1,
113 .flag = NULL,
114 .val = 'b'
115 },
116 {
117 .name = "per program stats",
118 .has_arg = 0,
119 .flag = NULL,
120 .val = 's'
121 },
122 {
123 .name = "track ios",
124 .has_arg = 0,
125 .flag = NULL,
126 .val = 't'
127 },
128 {
129 .name = "quiet",
130 .has_arg = 0,
131 .flag = NULL,
132 .val = 'q'
133 },
134 {
135 .name = "stopwatch",
136 .has_arg = 1,
137 .flag = NULL,
138 .val = 'w'
139 },
140 {
141 .name = NULL,
142 .has_arg = 0,
143 .flag = NULL,
144 .val = 0
145 }
146};
147
148static struct rb_root rb_sort_root;
149static struct rb_root rb_track_root;
150
151/*
152 * for sorting the displayed output
153 */
154struct trace {
155 struct blk_io_trace *bit;
156 struct rb_node rb_node;
157};
158
159/*
160 * for tracking individual ios
161 */
162struct io_track {
163 struct rb_node rb_node;
164
165 dev_t device;
166 __u64 sector;
167 __u32 pid;
168 unsigned long long allocation_time;
169 unsigned long long queue_time;
170 unsigned long long dispatch_time;
171 unsigned long long completion_time;
172};
173
174static int ndevices;
175static struct per_dev_info *devices;
176static char *get_dev_name(struct per_dev_info *, char *, int);
177
178static FILE *ofp;
179static char *output_name;
180
181static unsigned long long genesis_time;
182static unsigned long long stopwatch_start; /* start from zero by default */
183static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
184
185static int per_process_stats;
186static int track_ios;
187
188#define RB_BATCH_DEFAULT (1024)
189static int rb_batch = RB_BATCH_DEFAULT;
190
191static int pipeline;
192
193#define is_done() (*(volatile int *)(&done))
194static volatile int done;
195
196static inline unsigned long hash_long(unsigned long val)
197{
198#if __WORDSIZE == 32
199 val *= 0x9e370001UL;
200#elif __WORDSIZE == 64
201 val *= 0x9e37fffffffc0001UL;
202#else
203#error unknown word size
204#endif
205
206 return val >> (__WORDSIZE - PPI_HASH_SHIFT);
207}
208
209static inline void add_process_to_hash(struct per_process_info *ppi)
210{
211 const int hash_idx = hash_long(ppi->pid);
212
213 ppi->hash_next = ppi_hash[hash_idx];
214 ppi_hash[hash_idx] = ppi;
215}
216
217static inline void add_process_to_list(struct per_process_info *ppi)
218{
219 ppi->list_next = ppi_list;
220 ppi_list = ppi;
221}
222
223static struct per_process_info *find_process_by_pid(__u32 pid)
224{
225 const int hash_idx = hash_long(pid);
226 struct per_process_info *ppi;
227
228 ppi = ppi_hash[hash_idx];
229 while (ppi) {
230 if (ppi->pid == pid)
231 return ppi;
232
233 ppi = ppi->hash_next;
234 }
235
236 return NULL;
237}
238
239static inline int trace_rb_insert(struct trace *t)
240{
241 struct rb_node **p = &rb_sort_root.rb_node;
242 struct rb_node *parent = NULL;
243 struct trace *__t;
244
245 if (genesis_time == 0 || t->bit->time < genesis_time)
246 genesis_time = t->bit->time;
247
248 while (*p) {
249 parent = *p;
250 __t = rb_entry(parent, struct trace, rb_node);
251
252 if (t->bit->time < __t->bit->time)
253 p = &(*p)->rb_left;
254 else if (t->bit->time > __t->bit->time)
255 p = &(*p)->rb_right;
256 else if (t->bit->device < __t->bit->device)
257 p = &(*p)->rb_left;
258 else if (t->bit->device > __t->bit->device)
259 p = &(*p)->rb_right;
260 else if (t->bit->sequence < __t->bit->sequence)
261 p = &(*p)->rb_left;
262 else if (t->bit->sequence > __t->bit->sequence)
263 p = &(*p)->rb_right;
264 else if (t->bit->device == __t->bit->device) {
265 fprintf(stderr,
266 "sequence alias (%d) on device %d,%d!\n",
267 t->bit->sequence,
268 MAJOR(t->bit->device), MINOR(t->bit->device));
269 return 1;
270 }
271 }
272
273 rb_link_node(&t->rb_node, parent, p);
274 rb_insert_color(&t->rb_node, &rb_sort_root);
275 return 0;
276}
277
278static inline int track_rb_insert(struct io_track *iot)
279{
280 struct rb_node **p = &rb_track_root.rb_node;
281 struct rb_node *parent = NULL;
282 struct io_track *__iot;
283
284 while (*p) {
285 parent = *p;
286
287 __iot = rb_entry(parent, struct io_track, rb_node);
288
289 if (iot->device < __iot->device)
290 p = &(*p)->rb_left;
291 else if (iot->device > __iot->device)
292 p = &(*p)->rb_right;
293 else if (iot->sector < __iot->sector)
294 p = &(*p)->rb_left;
295 else if (iot->sector > __iot->sector)
296 p = &(*p)->rb_right;
297 else {
298 fprintf(stderr,
299 "sector alias (%llu) on device %d,%d!\n",
300 iot->sector,
301 MAJOR(iot->device), MINOR(iot->device));
302 return 1;
303 }
304 }
305
306 rb_link_node(&iot->rb_node, parent, p);
307 rb_insert_color(&iot->rb_node, &rb_track_root);
308 return 0;
309}
310
311static struct io_track *__find_track(dev_t device, __u64 sector)
312{
313 struct rb_node **p = &rb_track_root.rb_node;
314 struct rb_node *parent = NULL;
315 struct io_track *__iot;
316
317 while (*p) {
318 parent = *p;
319
320 __iot = rb_entry(parent, struct io_track, rb_node);
321
322 if (device < __iot->device)
323 p = &(*p)->rb_left;
324 else if (device > __iot->device)
325 p = &(*p)->rb_right;
326 else if (sector < __iot->sector)
327 p = &(*p)->rb_left;
328 else if (sector > __iot->sector)
329 p = &(*p)->rb_right;
330 else
331 return __iot;
332 }
333
334 return NULL;
335}
336
337static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector)
338{
339 struct io_track *iot;
340
341 iot = __find_track(device, sector);
342 if (!iot) {
343 iot = malloc(sizeof(*iot));
344 iot->pid = pid;
345 iot->device = device;
346 iot->sector = sector;
347 track_rb_insert(iot);
348 }
349
350 return iot;
351}
352
353static void log_track_merge(struct blk_io_trace *t)
354{
355 struct io_track *iot;
356
357 if (!track_ios)
358 return;
359 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
360 return;
361
362 iot = __find_track(t->device, t->sector - (t->bytes >> 10));
363 if (!iot) {
364 fprintf(stderr, "Trying to merge on non-existing request\n");
365 return;
366 }
367
368 rb_erase(&iot->rb_node, &rb_track_root);
369 iot->sector -= t->bytes >> 10;
370 track_rb_insert(iot);
371}
372
373static void log_track_getrq(struct blk_io_trace *t)
374{
375 struct io_track *iot;
376
377 if (!track_ios)
378 return;
379
380 iot = find_track(t->pid, t->device, t->sector);
381 iot->allocation_time = t->time;
382}
383
384
385/*
386 * return time between rq allocation and queue
387 */
388static unsigned long long log_track_queue(struct blk_io_trace *t)
389{
390 unsigned long long elapsed;
391 struct io_track *iot;
392
393 if (!track_ios)
394 return -1;
395
396 iot = find_track(t->pid, t->device, t->sector);
397 iot->queue_time = t->time;
398 elapsed = iot->queue_time - iot->allocation_time;
399
400 if (per_process_stats) {
401 struct per_process_info *ppi = find_process_by_pid(iot->pid);
402 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
403
404 if (ppi && elapsed > ppi->longest_allocation_wait[w])
405 ppi->longest_allocation_wait[w] = elapsed;
406 }
407
408 return elapsed;
409}
410
411/*
412 * return time between queue and issue
413 */
414static unsigned long long log_track_issue(struct blk_io_trace *t)
415{
416 unsigned long long elapsed;
417 struct io_track *iot;
418
419 if (!track_ios)
420 return -1;
421 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
422 return -1;
423
424 iot = __find_track(t->device, t->sector);
425 if (!iot) {
426 fprintf(stderr, "Trying to issue on non-existing request\n");
427 return -1;
428 }
429
430 iot->dispatch_time = t->time;
431 elapsed = iot->dispatch_time - iot->queue_time;
432
433 if (per_process_stats) {
434 struct per_process_info *ppi = find_process_by_pid(iot->pid);
435 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
436
437 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
438 ppi->longest_dispatch_wait[w] = elapsed;
439 }
440
441 return elapsed;
442}
443
444/*
445 * return time between dispatch and complete
446 */
447static unsigned long long log_track_complete(struct blk_io_trace *t)
448{
449 unsigned long long elapsed;
450 struct io_track *iot;
451
452 if (!track_ios)
453 return -1;
454 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
455 return -1;
456
457 iot = __find_track(t->device, t->sector);
458 if (!iot) {
459 fprintf(stderr, "Trying to dispatch on non-existing request\n");
460 return -1;
461 }
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 void output(struct per_cpu_info *pci, char *s)
678{
679 fprintf(ofp, "%s", s);
680}
681
682static char hstring[256];
683static char tstring[256];
684
685static inline char *setup_header(struct per_cpu_info *pci,
686 struct blk_io_trace *t, char act)
687{
688 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
689 int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER);
690 int s = t->action & BLK_TC_ACT(BLK_TC_SYNC);
691 char rwbs[4];
692 int i = 0;
693
694 if (w)
695 rwbs[i++] = 'W';
696 else
697 rwbs[i++] = 'R';
698 if (b)
699 rwbs[i++] = 'B';
700 if (s)
701 rwbs[i++] = 'S';
702
703 rwbs[i] = '\0';
704
705 sprintf(hstring, "%3d,%-3d %2d %8ld %5Lu.%09Lu %5u %c %3s",
706 MAJOR(t->device), MINOR(t->device), pci->cpu,
707 (unsigned long)t->sequence, SECONDS(t->time),
708 NANO_SECONDS(t->time), t->pid, act, rwbs);
709
710 return hstring;
711}
712
713static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t,
714 char act)
715{
716 unsigned long long elapsed = log_track_complete(t);
717
718 if (elapsed != -1ULL) {
719 unsigned long usec = elapsed / 1000;
720
721 sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n",
722 setup_header(pci, t, act),
723 (unsigned long long)t->sector, t->bytes >> 9,
724 usec, t->error);
725 } else {
726 sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act),
727 (unsigned long long)t->sector, t->bytes >> 9, t->error);
728 }
729
730 output(pci, tstring);
731}
732
733static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
734 char act)
735{
736 unsigned long long elapsed = log_track_queue(t);
737
738 if (elapsed != -1ULL) {
739 unsigned long usec = elapsed / 1000;
740
741 sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n",
742 setup_header(pci, t, act),
743 (unsigned long long)t->sector, t->bytes >> 9,
744 usec, t->comm);
745 } else {
746 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
747 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
748 }
749 output(pci, tstring);
750}
751
752static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t,
753 char act)
754{
755 unsigned long long elapsed = log_track_issue(t);
756
757 if (elapsed != -1ULL) {
758 double usec = (double) elapsed / 1000;
759
760 sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n",
761 setup_header(pci, t, act),
762 (unsigned long long)t->sector, t->bytes >> 9,
763 usec, t->comm);
764 } else {
765 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
766 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
767 }
768
769 output(pci, tstring);
770}
771
772static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
773 char act)
774{
775 log_track_merge(t);
776
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 output(pci, tstring);
780}
781
782static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
783 char act)
784{
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 output(pci, tstring);
788}
789
790static int log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
791 char act)
792{
793 __u64 *depth;
794 int len;
795
796 len = sprintf(tstring,"%s ", setup_header(pci, t, act));
797 depth = (__u64 *) t + sizeof(*t);
798 sprintf(tstring + len, "%u\n", (unsigned int) be64_to_cpu(*depth));
799 output(pci, tstring);
800
801 return 0;
802}
803
804static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char act)
805{
806 unsigned char *buf;
807 int i;
808
809 sprintf(tstring,"%s ", setup_header(pci, t, act));
810 output(pci, tstring);
811
812 buf = (unsigned char *) t + sizeof(*t);
813 for (i = 0; i < t->pdu_len; i++) {
814 sprintf(tstring,"%02x ", buf[i]);
815 output(pci, tstring);
816 }
817
818 if (act == 'C') {
819 sprintf(tstring,"[%d]\n", t->error);
820 output(pci, tstring);
821 } else {
822 sprintf(tstring,"[%s]\n", t->comm);
823 output(pci, tstring);
824 }
825 return 0;
826}
827
828static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
829{
830 int ret = 0;
831
832 switch (t->action & 0xffff) {
833 case __BLK_TA_QUEUE:
834 log_generic(pci, t, 'Q');
835 break;
836 case __BLK_TA_GETRQ:
837 log_generic(pci, t, 'G');
838 break;
839 case __BLK_TA_SLEEPRQ:
840 log_generic(pci, t, 'S');
841 break;
842 case __BLK_TA_REQUEUE:
843 log_generic(pci, t, 'R');
844 break;
845 case __BLK_TA_ISSUE:
846 ret = log_pc(pci, t, 'D');
847 break;
848 case __BLK_TA_COMPLETE:
849 log_pc(pci, t, 'C');
850 break;
851 default:
852 fprintf(stderr, "Bad pc action %x\n", t->action);
853 ret = 1;
854 break;
855 }
856
857 return ret;
858}
859
860static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci)
861{
862 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
863 int act = t->action & 0xffff;
864
865 switch (act) {
866 case __BLK_TA_QUEUE:
867 account_q(t, pci, w);
868 log_queue(pci, t, 'Q');
869 break;
870 case __BLK_TA_BACKMERGE:
871 account_m(t, pci, w);
872 log_merge(pci, t, 'M');
873 break;
874 case __BLK_TA_FRONTMERGE:
875 account_m(t, pci, w);
876 log_merge(pci, t, 'F');
877 break;
878 case __BLK_TA_GETRQ:
879 log_track_getrq(t);
880 log_generic(pci, t, 'G');
881 break;
882 case __BLK_TA_SLEEPRQ:
883 log_generic(pci, t, 'S');
884 break;
885 case __BLK_TA_REQUEUE:
886 account_c(t, pci, w, -t->bytes);
887 log_queue(pci, t, 'R');
888 break;
889 case __BLK_TA_ISSUE:
890 account_i(t, pci, w);
891 log_issue(pci, t, 'D');
892 break;
893 case __BLK_TA_COMPLETE:
894 account_c(t, pci, w, t->bytes);
895 log_complete(pci, t, 'C');
896 break;
897 case __BLK_TA_PLUG:
898 log_generic(pci, t, 'P');
899 break;
900 case __BLK_TA_UNPLUG:
901 log_unplug(pci, t, 'U');
902 break;
903 default:
904 fprintf(stderr, "Bad fs action %x\n", t->action);
905 break;
906 }
907}
908
909static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
910 struct per_dev_info *pdi)
911{
912 int ret = 0;
913
914 if (t->action & BLK_TC_ACT(BLK_TC_PC))
915 ret = dump_trace_pc(t, pci);
916 else
917 dump_trace_fs(t, pci);
918
919 pdi->events++;
920 return ret;
921}
922
923static void dump_io_stats(struct io_stats *ios, char *msg)
924{
925 fprintf(ofp, "%s\n", msg);
926
927 fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
928 fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
929
930 fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
931 fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
932 fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
933 fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
934 fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
935
936 fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
937}
938
939static void dump_wait_stats(struct per_process_info *ppi)
940{
941 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
942 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
943 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
944 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
945 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
946 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
947
948 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
949 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
950 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
951 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
952 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
953 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
954}
955
956static void show_process_stats(void)
957{
958 struct per_process_info *ppi;
959
960 ppi = ppi_list;
961 while (ppi) {
962 dump_io_stats(&ppi->io_stats, ppi->name);
963 dump_wait_stats(ppi);
964 ppi = ppi->list_next;
965 }
966
967 fprintf(ofp, "\n");
968}
969
970static void show_device_and_cpu_stats(void)
971{
972 struct per_dev_info *pdi;
973 struct per_cpu_info *pci;
974 struct io_stats total, *ios;
975 int i, j, pci_events;
976 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
977 char name[32];
978
979 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
980
981 memset(&total, 0, sizeof(total));
982 pci_events = 0;
983
984 if (i > 0)
985 fprintf(ofp, "\n");
986
987 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
988 if (!pci->nelems)
989 continue;
990
991 ios = &pci->io_stats;
992 total.qreads += ios->qreads;
993 total.qwrites += ios->qwrites;
994 total.creads += ios->creads;
995 total.cwrites += ios->cwrites;
996 total.mreads += ios->mreads;
997 total.mwrites += ios->mwrites;
998 total.ireads += ios->ireads;
999 total.iwrites += ios->iwrites;
1000 total.qread_kb += ios->qread_kb;
1001 total.qwrite_kb += ios->qwrite_kb;
1002 total.cread_kb += ios->cread_kb;
1003 total.cwrite_kb += ios->cwrite_kb;
1004 total.iread_kb += ios->iread_kb;
1005 total.iwrite_kb += ios->iwrite_kb;
1006
1007 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1008 j, get_dev_name(pdi, name, sizeof(name)));
1009 dump_io_stats(ios, line);
1010 pci_events++;
1011 }
1012
1013 if (pci_events > 1) {
1014 fprintf(ofp, "\n");
1015 snprintf(line, sizeof(line) - 1, "Total (%s):",
1016 get_dev_name(pdi, name, sizeof(name)));
1017 dump_io_stats(&total, line);
1018 }
1019
1020 fprintf(ofp, "Events (%s): %'Lu\n",
1021 get_dev_name(pdi, line, sizeof(line)), pdi->events);
1022 }
1023}
1024
1025static struct blk_io_trace *find_trace(void *p, unsigned long offset, int nr)
1026{
1027 unsigned long max_offset = min(offset,nr * sizeof(struct blk_io_trace));
1028 unsigned long off;
1029 struct blk_io_trace *bit;
1030 __u32 magic;
1031
1032 for (off = 0; off < max_offset; off++) {
1033 bit = p + off;
1034
1035 magic = be32_to_cpu(bit->magic);
1036 if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
1037 return bit;
1038 }
1039
1040 return NULL;
1041}
1042
1043static int sort_entries(void *traces, unsigned long offset, int nr,
1044 struct per_dev_info *fpdi, struct per_cpu_info *fpci)
1045{
1046 struct per_dev_info *pdi;
1047 struct per_cpu_info *pci;
1048 struct blk_io_trace *bit;
1049 struct trace *t;
1050 void *start = traces;
1051
1052 while (traces - start <= offset - sizeof(*bit)) {
1053 if (!nr)
1054 break;
1055
1056 bit = find_trace(traces, offset - (traces - start), nr);
1057 if (!bit)
1058 break;
1059
1060 t = malloc(sizeof(*t));
1061 if (!t) {
1062 fprintf(stderr, "Out of memory, seq %d on dev %d,%d\n",
1063 bit->sequence,
1064 MAJOR(bit->device), MINOR(bit->device));
1065 return -1;
1066 }
1067 t->bit = bit;
1068 memset(&t->rb_node, 0, sizeof(t->rb_node));
1069
1070 trace_to_cpu(bit);
1071
1072 if (verify_trace(bit)) {
1073 free(t);
1074 break;
1075 }
1076
1077 pdi = fpdi ? fpdi : get_dev_info(bit->device, 1);
1078 pdi->id = bit->device;
1079 pci = fpci ? fpci : get_cpu_info(pdi, bit->cpu);
1080 pci->cpu = bit->cpu;
1081 pci->nelems++;
1082
1083 if (trace_rb_insert(t)) {
1084 free(t);
1085 return -1;
1086 }
1087
1088 traces += sizeof(*bit) + bit->pdu_len;
1089 nr--;
1090 }
1091
1092 return 0;
1093}
1094
1095static void free_entries_rb(void)
1096{
1097 struct rb_node *n;
1098
1099 while ((n = rb_first(&rb_sort_root)) != NULL) {
1100 struct trace *t = rb_entry(n, struct trace, rb_node);
1101
1102 rb_erase(&t->rb_node, &rb_sort_root);
1103 free(t);
1104 }
1105}
1106
1107static void show_entries_rb(void)
1108{
1109 struct per_dev_info *pdi;
1110 struct blk_io_trace *bit;
1111 struct rb_node *n;
1112 struct trace *t;
1113 int cpu;
1114
1115 n = rb_first(&rb_sort_root);
1116 if (!n)
1117 return;
1118
1119 do {
1120 t = rb_entry(n, struct trace, rb_node);
1121 bit = t->bit;
1122
1123 pdi = get_dev_info(bit->device, 0);
1124 if (!pdi) {
1125 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
1126 MAJOR(bit->device), MINOR(bit->device));
1127 break;
1128 }
1129 cpu = bit->cpu;
1130 if (cpu > pdi->ncpus) {
1131 fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
1132 cpu, MAJOR(bit->device), MINOR(bit->device));
1133 break;
1134 }
1135
1136 bit->time -= genesis_time;
1137 if (bit->time < stopwatch_start)
1138 continue;
1139 if (bit->time >= stopwatch_end)
1140 break;
1141
1142 check_time(pdi, bit);
1143
1144 if (dump_trace(bit, &pdi->cpus[cpu], pdi))
1145 break;
1146
1147 } while ((n = rb_next(n)) != NULL);
1148}
1149
1150static int read_data(int fd, void *buffer, int bytes, int block)
1151{
1152 int ret, bytes_left, fl;
1153 void *p;
1154
1155 fl = fcntl(fd, F_GETFL);
1156
1157 if (!block)
1158 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
1159 else
1160 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
1161
1162 bytes_left = bytes;
1163 p = buffer;
1164 while (bytes_left > 0) {
1165 ret = read(fd, p, bytes_left);
1166 if (!ret)
1167 return 1;
1168 else if (ret < 0) {
1169 if (errno != EAGAIN)
1170 perror("read");
1171 return -1;
1172 } else {
1173 p += ret;
1174 bytes_left -= ret;
1175 }
1176 }
1177
1178 return 0;
1179}
1180
1181static int do_file(void)
1182{
1183 struct per_dev_info *pdi;
1184 int i, j, nfiles = 0;
1185
1186 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1187 for (j = 0;; j++, nfiles++) {
1188 struct per_cpu_info *pci;
1189 struct stat st;
1190 void *tb;
1191
1192 pci = get_cpu_info(pdi, j);
1193 pci->cpu = j;
1194
1195 snprintf(pci->fname, sizeof(pci->fname)-1,
1196 "%s_out.%d", pdi->name, j);
1197 if (stat(pci->fname, &st) < 0)
1198 break;
1199 if (!st.st_size)
1200 continue;
1201
1202 printf("Processing %s\n", pci->fname);
1203
1204 tb = malloc(st.st_size);
1205 if (!tb) {
1206 fprintf(stderr, "Out of memory, skip file %s\n",
1207 pci->fname);
1208 continue;
1209 }
1210
1211 pci->fd = open(pci->fname, O_RDONLY);
1212 if (pci->fd < 0) {
1213 perror(pci->fname);
1214 free(tb);
1215 continue;
1216 }
1217
1218 if (read_data(pci->fd, tb, st.st_size, 1)) {
1219 close(pci->fd);
1220 free(tb);
1221 continue;
1222 }
1223
1224 if (sort_entries(tb, st.st_size, ~0U, pdi, pci) == -1) {
1225 close(pci->fd);
1226 free(tb);
1227 continue;
1228 }
1229
1230 printf("Completed %s (CPU%d %d, entries)\n",
1231 pci->fname, j, pci->nelems);
1232 close(pci->fd);
1233 }
1234 }
1235
1236 if (!nfiles) {
1237 fprintf(stderr, "No files found\n");
1238 return 1;
1239 }
1240
1241 show_entries_rb();
1242 return 0;
1243}
1244
1245static void resize_buffer(void **buffer, long *size, long offset)
1246{
1247 long old_size = *size;
1248
1249 if (*size == 0)
1250 *size = 64 * sizeof(struct blk_io_trace);
1251
1252 *size *= 2;
1253 *buffer = realloc(*buffer, *size);
1254
1255 if (old_size)
1256 memset(*buffer + offset, 0, *size - old_size);
1257}
1258
1259static int read_sort_events(int fd, void **buffer, long *max_offset)
1260{
1261 long offset;
1262 int events;
1263
1264 events = offset = 0;
1265 do {
1266 struct blk_io_trace *t;
1267 int pdu_len;
1268 __u32 magic;
1269
1270 if (*max_offset - offset < sizeof(*t))
1271 resize_buffer(buffer, max_offset, offset);
1272
1273 if (read_data(fd, *buffer + offset, sizeof(*t), !events))
1274 break;
1275
1276 t = *buffer + offset;
1277 offset += sizeof(*t);
1278
1279 magic = be32_to_cpu(t->magic);
1280 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1281 fprintf(stderr, "Bad magic %x\n", magic);
1282 break;
1283 }
1284
1285 pdu_len = be16_to_cpu(t->pdu_len);
1286 if (pdu_len) {
1287 if (*max_offset - offset <= pdu_len)
1288 resize_buffer(buffer, max_offset, offset);
1289
1290 if (read_data(fd, *buffer + offset, pdu_len, 1))
1291 break;
1292
1293 offset += pdu_len;
1294 }
1295
1296 events++;
1297 } while (!is_done() && events < rb_batch);
1298
1299 return events;
1300}
1301
1302static int do_stdin(void)
1303{
1304 int fd;
1305 void *ptr = NULL;
1306 long max_offset;
1307
1308 fd = dup(STDIN_FILENO);
1309 max_offset = 0;
1310 do {
1311 int events;
1312
1313 events = read_sort_events(fd, &ptr, &max_offset);
1314 if (!events)
1315 break;
1316
1317 if (sort_entries(ptr, ~0UL, events, NULL, NULL) == -1)
1318 break;
1319
1320 show_entries_rb();
1321 free_entries_rb();
1322 } while (1);
1323
1324 if (ptr)
1325 free(ptr);
1326
1327 close(fd);
1328 return 0;
1329}
1330
1331static void flush_output(void)
1332{
1333 fflush(ofp);
1334}
1335
1336static void handle_sigint(int sig)
1337{
1338 done = 1;
1339 flush_output();
1340}
1341
1342/*
1343 * Extract start and duration times from a string, allowing
1344 * us to specify a time interval of interest within a trace.
1345 * Format: "duration" (start is zero) or "start:duration".
1346 */
1347static int find_stopwatch_interval(char *string)
1348{
1349 double value;
1350 char *sp;
1351
1352 value = strtod(string, &sp);
1353 if (sp == string) {
1354 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
1355 return 1;
1356 }
1357 if (*sp == ':') {
1358 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
1359 string = sp + 1;
1360 value = strtod(string, &sp);
1361 if (sp == string || *sp != '\0') {
1362 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
1363 string);
1364 return 1;
1365 }
1366 } else if (*sp != '\0') {
1367 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
1368 return 1;
1369 }
1370 stopwatch_end = stopwatch_start + DOUBLE_TO_NANO_ULL(value);
1371 return 0;
1372}
1373
1374static void usage(char *prog)
1375{
1376 fprintf(stderr, "Usage: %s "
1377 "[-i <name>] [-o <output>] [-s] [-w N[:n]] <name>...\n",
1378 prog);
1379}
1380
1381int main(int argc, char *argv[])
1382{
1383 char *ofp_buffer;
1384 int c, ret, mode;
1385 int per_device_and_cpu_stats = 1;
1386
1387 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1388 switch (c) {
1389 case 'i':
1390 if (!strcmp(optarg, "-") && !pipeline)
1391 pipeline = 1;
1392 else if (resize_devices(optarg) != 0)
1393 return 1;
1394 break;
1395 case 'o':
1396 output_name = optarg;
1397 break;
1398 case 'b':
1399 rb_batch = atoi(optarg);
1400 if (rb_batch <= 0)
1401 rb_batch = RB_BATCH_DEFAULT;
1402 break;
1403 case 's':
1404 per_process_stats = 1;
1405 break;
1406 case 't':
1407 track_ios = 1;
1408 break;
1409 case 'q':
1410 per_device_and_cpu_stats = 0;
1411 break;
1412 case 'w':
1413 if (find_stopwatch_interval(optarg) != 0)
1414 return 1;
1415 break;
1416 default:
1417 usage(argv[0]);
1418 return 1;
1419 }
1420 }
1421
1422 while (optind < argc) {
1423 if (!strcmp(argv[optind], "-") && !pipeline)
1424 pipeline = 1;
1425 else if (resize_devices(argv[optind]) != 0)
1426 return 1;
1427 optind++;
1428 }
1429
1430 if (!pipeline && !ndevices) {
1431 usage(argv[0]);
1432 return 1;
1433 }
1434
1435 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
1436 memset(&rb_track_root, 0, sizeof(rb_track_root));
1437
1438 signal(SIGINT, handle_sigint);
1439 signal(SIGHUP, handle_sigint);
1440 signal(SIGTERM, handle_sigint);
1441
1442 setlocale(LC_NUMERIC, "en_US");
1443
1444 if (!output_name) {
1445 ofp = fdopen(STDOUT_FILENO, "w");
1446 mode = _IOLBF;
1447 } else {
1448 char ofname[128];
1449
1450 snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name);
1451 ofp = fopen(ofname, "w");
1452 mode = _IOFBF;
1453 }
1454
1455 if (!ofp) {
1456 perror("fopen");
1457 return 1;
1458 }
1459
1460 ofp_buffer = malloc(4096);
1461 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
1462 perror("setvbuf");
1463 return 1;
1464 }
1465
1466 if (pipeline)
1467 ret = do_stdin();
1468 else
1469 ret = do_file();
1470
1471 if (per_process_stats)
1472 show_process_stats();
1473
1474 if (per_device_and_cpu_stats)
1475 show_device_and_cpu_stats();
1476
1477 flush_output();
1478 return ret;
1479}
Block queue io tracer