blktrace: accept -v (lower case) for version info as well
[blktrace.git] / blkparse.c
CommitLineData
d956a2cd
JA
1/*
2 * block queue tracing parse application
3 *
4 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
46e37c55 5 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
d956a2cd
JA
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 */
d0ca268b
JA
22#include <sys/types.h>
23#include <sys/stat.h>
24#include <unistd.h>
25#include <stdio.h>
26#include <fcntl.h>
27#include <stdlib.h>
8fc0abbc 28#include <string.h>
d5396421 29#include <getopt.h>
412819ce
JA
30#include <errno.h>
31#include <signal.h>
d69db225 32#include <locale.h>
6e0073ed 33#include <libgen.h>
d0ca268b 34
8fc0abbc
JA
35#include "blktrace.h"
36#include "rbtree.h"
bf0720af 37#include "jhash.h"
d0ca268b 38
a5981e2e 39static char blkparse_version[] = "0.99.3";
52724a0e 40
492da111
AB
41struct skip_info {
42 unsigned long start, end;
43 struct skip_info *prev, *next;
44};
45
e7c9f3ff 46struct per_dev_info {
f7bd1a9b 47 dev_t dev;
e7c9f3ff
NS
48 char *name;
49
50 int backwards;
51 unsigned long long events;
20ed6177 52 unsigned long long first_reported_time;
e7c9f3ff 53 unsigned long long last_reported_time;
287fa3d6 54 unsigned long long last_read_time;
e7c9f3ff 55 struct io_stats io_stats;
2990e589
JA
56 unsigned long skips;
57 unsigned long long seq_skips;
649c7b66
JA
58 unsigned int max_depth[2];
59 unsigned int cur_depth[2];
e7c9f3ff 60
f7bd1a9b
JA
61 struct rb_root rb_track;
62
73877e12 63 int nfiles;
e7c9f3ff 64 int ncpus;
824c2b39
JA
65
66 unsigned long *cpu_map;
67 unsigned int cpu_map_max;
68
e7c9f3ff
NS
69 struct per_cpu_info *cpus;
70};
71
2990e589
JA
72/*
73 * some duplicated effort here, we can unify this hash and the ppi hash later
74 */
75struct process_pid_map {
76 pid_t pid;
77 char comm[16];
78 struct process_pid_map *hash_next, *list_next;
79};
80
81#define PPM_HASH_SHIFT (8)
82#define PPM_HASH_SIZE (1 << PPM_HASH_SHIFT)
83#define PPM_HASH_MASK (PPM_HASH_SIZE - 1)
84static struct process_pid_map *ppm_hash_table[PPM_HASH_SIZE];
85
152f6476 86struct per_process_info {
2990e589 87 struct process_pid_map *ppm;
152f6476
JA
88 struct io_stats io_stats;
89 struct per_process_info *hash_next, *list_next;
715d8021 90 int more_than_one;
50adc0ba
JA
91
92 /*
93 * individual io stats
94 */
b9d40d6f
JA
95 unsigned long long longest_allocation_wait[2];
96 unsigned long long longest_dispatch_wait[2];
97 unsigned long long longest_completion_wait[2];
d0ca268b
JA
98};
99
152f6476 100#define PPI_HASH_SHIFT (8)
bf0720af
JA
101#define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT)
102#define PPI_HASH_MASK (PPI_HASH_SIZE - 1)
103static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE];
152f6476 104static struct per_process_info *ppi_list;
886ecf0e 105static int ppi_list_entries;
152f6476 106
234db09d 107#define S_OPTS "a:A:b:D:d:f:F:hi:o:Oqstw:vV"
d5396421 108static struct option l_opts[] = {
98f8386b
AB
109 {
110 .name = "act-mask",
111 .has_arg = required_argument,
112 .flag = NULL,
113 .val = 'a'
114 },
115 {
116 .name = "set-mask",
117 .has_arg = required_argument,
118 .flag = NULL,
119 .val = 'A'
120 },
d5396421 121 {
234db09d 122 .name = "batch",
428683db 123 .has_arg = required_argument,
d5396421 124 .flag = NULL,
234db09d 125 .val = 'b'
d5396421
JA
126 },
127 {
234db09d 128 .name = "input-directory",
428683db 129 .has_arg = required_argument,
d5396421 130 .flag = NULL,
234db09d 131 .val = 'D'
d5396421 132 },
79f19470 133 {
234db09d 134 .name = "dump-binary",
428683db 135 .has_arg = required_argument,
79f19470 136 .flag = NULL,
234db09d 137 .val = 'd'
79f19470 138 },
152f6476 139 {
234db09d
AB
140 .name = "format",
141 .has_arg = required_argument,
152f6476 142 .flag = NULL,
234db09d 143 .val = 'f'
152f6476 144 },
7997c5b0 145 {
234db09d
AB
146 .name = "format-spec",
147 .has_arg = required_argument,
7997c5b0 148 .flag = NULL,
234db09d 149 .val = 'F'
7997c5b0 150 },
1e1c60f1 151 {
234db09d 152 .name = "hash-by-name",
428683db 153 .has_arg = no_argument,
1e1c60f1 154 .flag = NULL,
234db09d 155 .val = 'h'
1e1c60f1 156 },
46e6968b 157 {
234db09d 158 .name = "input",
428683db 159 .has_arg = required_argument,
46e6968b 160 .flag = NULL,
234db09d 161 .val = 'i'
46e6968b 162 },
ab197ca7 163 {
234db09d 164 .name = "output",
428683db 165 .has_arg = required_argument,
ab197ca7 166 .flag = NULL,
234db09d 167 .val = 'o'
ab197ca7
AB
168 },
169 {
234db09d
AB
170 .name = "no-text-output",
171 .has_arg = no_argument,
ab197ca7 172 .flag = NULL,
234db09d 173 .val = 'O'
ab197ca7 174 },
bf0720af 175 {
234db09d 176 .name = "quiet",
bf0720af
JA
177 .has_arg = no_argument,
178 .flag = NULL,
234db09d 179 .val = 'q'
bf0720af 180 },
7d1c0411 181 {
234db09d 182 .name = "per-program-stats",
7d1c0411
JA
183 .has_arg = no_argument,
184 .flag = NULL,
234db09d 185 .val = 's'
7d1c0411 186 },
52724a0e 187 {
234db09d 188 .name = "track-ios",
52724a0e
JA
189 .has_arg = no_argument,
190 .flag = NULL,
234db09d 191 .val = 't'
52724a0e 192 },
d1d7f15f 193 {
234db09d 194 .name = "stopwatch",
d1d7f15f
JA
195 .has_arg = required_argument,
196 .flag = NULL,
234db09d 197 .val = 'w'
d1d7f15f 198 },
a2594911 199 {
234db09d
AB
200 .name = "verbose",
201 .has_arg = no_argument,
a2594911 202 .flag = NULL,
234db09d
AB
203 .val = 'v'
204 },
205 {
206 .name = "version",
207 .has_arg = no_argument,
208 .flag = NULL,
209 .val = 'V'
a2594911 210 },
71ef8b7c
JA
211 {
212 .name = NULL,
213 }
d5396421
JA
214};
215
7997c5b0
JA
216/*
217 * for sorting the displayed output
218 */
8fc0abbc
JA
219struct trace {
220 struct blk_io_trace *bit;
221 struct rb_node rb_node;
cb2a1a62 222 struct trace *next;
a43c1c17 223 unsigned long read_sequence;
8fc0abbc
JA
224};
225
cb2a1a62 226static struct rb_root rb_sort_root;
a649216c
JA
227static unsigned long rb_sort_entries;
228
cb2a1a62
JA
229static struct trace *trace_list;
230
d36421e4
JA
231/*
232 * allocation cache
233 */
234static struct blk_io_trace *bit_alloc_list;
235static struct trace *t_alloc_list;
236
7997c5b0
JA
237/*
238 * for tracking individual ios
239 */
240struct io_track {
241 struct rb_node rb_node;
242
2990e589 243 struct process_pid_map *ppm;
7997c5b0 244 __u64 sector;
95c15013 245 unsigned long long allocation_time;
7997c5b0
JA
246 unsigned long long queue_time;
247 unsigned long long dispatch_time;
248 unsigned long long completion_time;
249};
250
e7c9f3ff
NS
251static int ndevices;
252static struct per_dev_info *devices;
253static char *get_dev_name(struct per_dev_info *, char *, int);
210824c3 254static int trace_rb_insert_last(struct per_dev_info *, struct trace *);
d0ca268b 255
71d5d4c9 256FILE *ofp = NULL;
e7c9f3ff 257static char *output_name;
d1d7f15f 258static char *input_dir;
e7c9f3ff
NS
259
260static unsigned long long genesis_time;
287fa3d6 261static unsigned long long last_allowed_time;
46e6968b 262static unsigned long long stopwatch_start; /* start from zero by default */
bc171579 263static unsigned long long stopwatch_end = -1ULL; /* "infinity" */
a43c1c17 264static unsigned long read_sequence;
152f6476
JA
265
266static int per_process_stats;
cbc927b6 267static int per_device_and_cpu_stats = 1;
7997c5b0 268static int track_ios;
bf0720af 269static int ppi_hash_by_pid = 1;
57ea8602 270static int verbose;
98f8386b 271static unsigned int act_mask = -1U;
cbc927b6 272static int stats_printed;
86368eb5 273int data_is_native = -1;
d0ca268b 274
346d8a74 275static FILE *dump_fp;
a2594911
AB
276static char *dump_binary;
277
1d24fc14
JA
278static unsigned int t_alloc_cache;
279static unsigned int bit_alloc_cache;
280
7d747d22 281#define RB_BATCH_DEFAULT (512)
e820abd7 282static unsigned int rb_batch = RB_BATCH_DEFAULT;
79f19470 283
e7c9f3ff 284static int pipeline;
67076cbc 285static char *pipename;
e7c9f3ff 286
234db09d
AB
287static int text_output = 1;
288
412819ce
JA
289#define is_done() (*(volatile int *)(&done))
290static volatile int done;
291
7bd4fd0a
OK
292struct timespec abs_start_time;
293static unsigned long long start_timestamp;
294
bf0720af
JA
295#define JHASH_RANDOM (0x3af5f2ee)
296
824c2b39
JA
297#define CPUS_PER_LONG (8 * sizeof(unsigned long))
298#define CPU_IDX(cpu) ((cpu) / CPUS_PER_LONG)
299#define CPU_BIT(cpu) ((cpu) & (CPUS_PER_LONG - 1))
300
a2594911
AB
301static void output_binary(void *buf, int len)
302{
303 if (dump_binary) {
346d8a74
AB
304 size_t n = fwrite(buf, len, 1, dump_fp);
305 if (n != 1) {
a2594911 306 perror(dump_binary);
346d8a74 307 fclose(dump_fp);
a2594911
AB
308 dump_binary = NULL;
309 }
310 }
311}
312
210824c3
JA
313static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
314{
315 struct per_cpu_info *cpus = pdi->cpus;
316 int ncpus = pdi->ncpus;
317 int new_count = cpu + 1;
318 int new_space, size;
319 char *new_start;
320
321 size = new_count * sizeof(struct per_cpu_info);
322 cpus = realloc(cpus, size);
323 if (!cpus) {
324 char name[20];
325 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
326 get_dev_name(pdi, name, sizeof(name)), size);
327 exit(1);
328 }
329
330 new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
331 new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
332 memset(new_start, 0, new_space);
333
334 pdi->ncpus = new_count;
335 pdi->cpus = cpus;
336
337 for (new_count = 0; new_count < pdi->ncpus; new_count++) {
338 struct per_cpu_info *pci = &pdi->cpus[new_count];
339
340 if (!pci->fd) {
341 pci->fd = -1;
342 memset(&pci->rb_last, 0, sizeof(pci->rb_last));
343 pci->rb_last_entries = 0;
344 pci->last_sequence = -1;
345 }
346 }
347}
348
349static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
350{
351 struct per_cpu_info *pci;
352
353 if (cpu >= pdi->ncpus)
354 resize_cpu_info(pdi, cpu);
355
356 pci = &pdi->cpus[cpu];
357 pci->cpu = cpu;
358 return pci;
359}
360
361
362static int resize_devices(char *name)
363{
364 int size = (ndevices + 1) * sizeof(struct per_dev_info);
365
366 devices = realloc(devices, size);
367 if (!devices) {
368 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
369 return 1;
370 }
371 memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
372 devices[ndevices].name = name;
373 ndevices++;
374 return 0;
375}
376
377static struct per_dev_info *get_dev_info(dev_t dev)
378{
379 struct per_dev_info *pdi;
380 int i;
381
382 for (i = 0; i < ndevices; i++) {
383 if (!devices[i].dev)
384 devices[i].dev = dev;
385 if (devices[i].dev == dev)
386 return &devices[i];
387 }
388
389 if (resize_devices(NULL))
390 return NULL;
391
392 pdi = &devices[ndevices - 1];
393 pdi->dev = dev;
394 pdi->first_reported_time = 0;
395 pdi->last_read_time = 0;
210824c3
JA
396
397 return pdi;
398}
399
66930177 400static void insert_skip(struct per_cpu_info *pci, unsigned long start,
492da111
AB
401 unsigned long end)
402{
403 struct skip_info *sip;
404
66930177 405 for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
492da111
AB
406 if (end == (sip->start - 1)) {
407 sip->start = start;
408 return;
409 } else if (start == (sip->end + 1)) {
410 sip->end = end;
411 return;
412 }
413 }
414
415 sip = malloc(sizeof(struct skip_info));
416 sip->start = start;
417 sip->end = end;
418 sip->prev = sip->next = NULL;
66930177
JA
419 if (pci->skips_tail == NULL)
420 pci->skips_head = pci->skips_tail = sip;
492da111 421 else {
66930177
JA
422 sip->prev = pci->skips_tail;
423 pci->skips_tail->next = sip;
424 pci->skips_tail = sip;
492da111
AB
425 }
426}
427
66930177 428static void remove_sip(struct per_cpu_info *pci, struct skip_info *sip)
492da111
AB
429{
430 if (sip->prev == NULL) {
431 if (sip->next == NULL)
66930177 432 pci->skips_head = pci->skips_tail = NULL;
492da111 433 else {
66930177 434 pci->skips_head = sip->next;
492da111
AB
435 sip->next->prev = NULL;
436 }
437 } else if (sip->next == NULL) {
66930177 438 pci->skips_tail = sip->prev;
492da111
AB
439 sip->prev->next = NULL;
440 } else {
441 sip->prev->next = sip->next;
442 sip->next->prev = sip->prev;
443 }
444
445 sip->prev = sip->next = NULL;
446 free(sip);
447}
448
449#define IN_SKIP(sip,seq) (((sip)->start <= (seq)) && ((seq) <= sip->end))
66930177 450static int check_current_skips(struct per_cpu_info *pci, unsigned long seq)
492da111
AB
451{
452 struct skip_info *sip;
453
66930177
JA
454 for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
455 if (IN_SKIP(sip, seq)) {
492da111
AB
456 if (sip->start == seq) {
457 if (sip->end == seq)
66930177 458 remove_sip(pci, sip);
492da111
AB
459 else
460 sip->start += 1;
461 } else if (sip->end == seq)
462 sip->end -= 1;
463 else {
464 sip->end = seq - 1;
66930177 465 insert_skip(pci, seq + 1, sip->end);
492da111
AB
466 }
467 return 1;
468 }
469 }
66930177 470
492da111
AB
471 return 0;
472}
473
474static void collect_pdi_skips(struct per_dev_info *pdi)
475{
476 struct skip_info *sip;
66930177 477 int cpu;
492da111
AB
478
479 pdi->skips = 0;
480 pdi->seq_skips = 0;
66930177
JA
481
482 for (cpu = 0; cpu < pdi->ncpus; cpu++) {
483 struct per_cpu_info *pci = &pdi->cpus[cpu];
484
485 for (sip = pci->skips_head; sip != NULL; sip = sip->next) {
486 pdi->skips++;
487 pdi->seq_skips += (sip->end - sip->start + 1);
488 if (verbose)
489 fprintf(stderr,"(%d,%d): skipping %lu -> %lu\n",
490 MAJOR(pdi->dev), MINOR(pdi->dev),
491 sip->start, sip->end);
492 }
492da111
AB
493 }
494}
495
824c2b39
JA
496static void cpu_mark_online(struct per_dev_info *pdi, unsigned int cpu)
497{
498 if (cpu >= pdi->cpu_map_max || !pdi->cpu_map) {
499 int new_max = (cpu + CPUS_PER_LONG) & ~(CPUS_PER_LONG - 1);
500 unsigned long *map = malloc(new_max / sizeof(long));
501
502 memset(map, 0, new_max / sizeof(long));
503
504 if (pdi->cpu_map) {
505 memcpy(map, pdi->cpu_map, pdi->cpu_map_max / sizeof(long));
506 free(pdi->cpu_map);
507 }
508
509 pdi->cpu_map = map;
510 pdi->cpu_map_max = new_max;
511 }
512
513 pdi->cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
514}
515
516static inline void cpu_mark_offline(struct per_dev_info *pdi, int cpu)
517{
518 pdi->cpu_map[CPU_IDX(cpu)] &= ~(1UL << CPU_BIT(cpu));
519}
520
521static inline int cpu_is_online(struct per_dev_info *pdi, int cpu)
522{
523 return (pdi->cpu_map[CPU_IDX(cpu)] & (1UL << CPU_BIT(cpu))) != 0;
524}
525
bfc70ad5
JA
526static inline int ppm_hash_pid(pid_t pid)
527{
528 return jhash_1word(pid, JHASH_RANDOM) & PPM_HASH_MASK;
529}
530
531static struct process_pid_map *find_ppm(pid_t pid)
532{
533 const int hash_idx = ppm_hash_pid(pid);
534 struct process_pid_map *ppm;
535
536 ppm = ppm_hash_table[hash_idx];
537 while (ppm) {
538 if (ppm->pid == pid)
539 return ppm;
540
541 ppm = ppm->hash_next;
542 }
543
544 return NULL;
545}
546
ebe2d1aa 547static struct process_pid_map *add_ppm_hash(pid_t pid, const char *name)
bfc70ad5
JA
548{
549 const int hash_idx = ppm_hash_pid(pid);
550 struct process_pid_map *ppm;
551
552 ppm = find_ppm(pid);
248eac8f
JA
553 if (!ppm) {
554 ppm = malloc(sizeof(*ppm));
555 memset(ppm, 0, sizeof(*ppm));
556 ppm->pid = pid;
557 strcpy(ppm->comm, name);
558 ppm->hash_next = ppm_hash_table[hash_idx];
559 ppm_hash_table[hash_idx] = ppm;
bfc70ad5 560 }
ebe2d1aa
JA
561
562 return ppm;
bfc70ad5
JA
563}
564
7bd4fd0a
OK
565static void handle_notify(struct blk_io_trace *bit)
566{
567 void *payload = (caddr_t) bit + sizeof(*bit);
568 __u32 two32[2];
569
570 switch (bit->action) {
571 case BLK_TN_PROCESS:
572 add_ppm_hash(bit->pid, payload);
573 break;
574
575 case BLK_TN_TIMESTAMP:
576 if (bit->pdu_len != sizeof(two32))
577 return;
578 memcpy(two32, payload, sizeof(two32));
579 if (!data_is_native) {
580 two32[0] = be32_to_cpu(two32[0]);
581 two32[1] = be32_to_cpu(two32[1]);
582 }
583 start_timestamp = bit->time;
584 abs_start_time.tv_sec = two32[0];
585 abs_start_time.tv_nsec = two32[1];
586 if (abs_start_time.tv_nsec < 0) {
587 abs_start_time.tv_sec--;
588 abs_start_time.tv_nsec += 1000000000;
589 }
590
591 break;
592
1a15f6a8
AB
593 case BLK_TN_MESSAGE:
594 if (bit->pdu_len > 0) {
595 char msg[bit->pdu_len+1];
596
597 memcpy(msg, (char *)payload, bit->pdu_len);
598 msg[bit->pdu_len] = '\0';
599
600 fprintf(ofp,
601 "%3d,%-3d %2d %8s %5d.%09lu %5u %2s %3s %s\n",
602 MAJOR(bit->device), MINOR(bit->device),
603 bit->cpu, "0", (int) SECONDS(bit->time),
604 (unsigned long) NANO_SECONDS(bit->time),
605 0, "m", "N", msg);
606 }
607 break;
608
7bd4fd0a
OK
609 default:
610 /* Ignore unknown notify events */
611 ;
612 }
613}
614
bfc70ad5
JA
615char *find_process_name(pid_t pid)
616{
617 struct process_pid_map *ppm = find_ppm(pid);
618
619 if (ppm)
620 return ppm->comm;
621
622 return NULL;
623}
624
9e4cd1b8 625static inline int ppi_hash_pid(pid_t pid)
bf0720af
JA
626{
627 return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK;
628}
629
630static inline int ppi_hash_name(const char *name)
152f6476 631{
bf0720af
JA
632 return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK;
633}
634
635static inline int ppi_hash(struct per_process_info *ppi)
636{
2990e589
JA
637 struct process_pid_map *ppm = ppi->ppm;
638
bf0720af 639 if (ppi_hash_by_pid)
2990e589 640 return ppi_hash_pid(ppm->pid);
bf0720af 641
2990e589 642 return ppi_hash_name(ppm->comm);
152f6476
JA
643}
644
bfc70ad5 645static inline void add_ppi_to_hash(struct per_process_info *ppi)
152f6476 646{
bf0720af 647 const int hash_idx = ppi_hash(ppi);
152f6476 648
bf0720af
JA
649 ppi->hash_next = ppi_hash_table[hash_idx];
650 ppi_hash_table[hash_idx] = ppi;
152f6476
JA
651}
652
bfc70ad5 653static inline void add_ppi_to_list(struct per_process_info *ppi)
152f6476
JA
654{
655 ppi->list_next = ppi_list;
656 ppi_list = ppi;
886ecf0e 657 ppi_list_entries++;
152f6476
JA
658}
659
bfc70ad5 660static struct per_process_info *find_ppi_by_name(char *name)
bf0720af
JA
661{
662 const int hash_idx = ppi_hash_name(name);
663 struct per_process_info *ppi;
664
665 ppi = ppi_hash_table[hash_idx];
666 while (ppi) {
2990e589
JA
667 struct process_pid_map *ppm = ppi->ppm;
668
669 if (!strcmp(ppm->comm, name))
bf0720af
JA
670 return ppi;
671
672 ppi = ppi->hash_next;
673 }
674
675 return NULL;
676}
677
9e4cd1b8 678static struct per_process_info *find_ppi_by_pid(pid_t pid)
152f6476 679{
bf0720af 680 const int hash_idx = ppi_hash_pid(pid);
152f6476
JA
681 struct per_process_info *ppi;
682
bf0720af 683 ppi = ppi_hash_table[hash_idx];
152f6476 684 while (ppi) {
2990e589
JA
685 struct process_pid_map *ppm = ppi->ppm;
686
687 if (ppm->pid == pid)
152f6476
JA
688 return ppi;
689
690 ppi = ppi->hash_next;
691 }
692
693 return NULL;
694}
695
9e4cd1b8 696static struct per_process_info *find_ppi(pid_t pid)
bf0720af 697{
715d8021 698 struct per_process_info *ppi;
bfc70ad5 699 char *name;
715d8021 700
bf0720af 701 if (ppi_hash_by_pid)
bfc70ad5
JA
702 return find_ppi_by_pid(pid);
703
704 name = find_process_name(pid);
705 if (!name)
706 return NULL;
bf0720af 707
bfc70ad5 708 ppi = find_ppi_by_name(name);
2990e589 709 if (ppi && ppi->ppm->pid != pid)
715d8021
JA
710 ppi->more_than_one = 1;
711
712 return ppi;
bf0720af
JA
713}
714
210824c3
JA
715/*
716 * struct trace and blktrace allocation cache, we do potentially
717 * millions of mallocs for these structures while only using at most
718 * a few thousand at the time
719 */
720static inline void t_free(struct trace *t)
721{
722 if (t_alloc_cache < 1024) {
723 t->next = t_alloc_list;
724 t_alloc_list = t;
725 t_alloc_cache++;
726 } else
727 free(t);
728}
729
730static inline struct trace *t_alloc(void)
731{
732 struct trace *t = t_alloc_list;
733
734 if (t) {
735 t_alloc_list = t->next;
736 t_alloc_cache--;
737 return t;
738 }
739
740 return malloc(sizeof(*t));
741}
742
743static inline void bit_free(struct blk_io_trace *bit)
744{
745 if (bit_alloc_cache < 1024 && !bit->pdu_len) {
746 /*
747 * abuse a 64-bit field for a next pointer for the free item
748 */
749 bit->time = (__u64) (unsigned long) bit_alloc_list;
750 bit_alloc_list = (struct blk_io_trace *) bit;
751 bit_alloc_cache++;
752 } else
753 free(bit);
754}
755
756static inline struct blk_io_trace *bit_alloc(void)
757{
758 struct blk_io_trace *bit = bit_alloc_list;
759
760 if (bit) {
761 bit_alloc_list = (struct blk_io_trace *) (unsigned long) \
762 bit->time;
763 bit_alloc_cache--;
764 return bit;
765 }
766
767 return malloc(sizeof(*bit));
768}
769
770static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
771{
772 struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
773
774 rb_erase(&t->rb_node, &pci->rb_last);
775 pci->rb_last_entries--;
776
777 bit_free(t->bit);
778 t_free(t);
779}
780
781static void put_trace(struct per_dev_info *pdi, struct trace *t)
782{
783 rb_erase(&t->rb_node, &rb_sort_root);
784 rb_sort_entries--;
785
786 trace_rb_insert_last(pdi, t);
787}
788
89482da6 789static inline int trace_rb_insert(struct trace *t, struct rb_root *root)
7997c5b0 790{
2a1b3424 791 struct rb_node **p = &root->rb_node;
7997c5b0
JA
792 struct rb_node *parent = NULL;
793 struct trace *__t;
794
795 while (*p) {
796 parent = *p;
2a1b3424 797
7997c5b0
JA
798 __t = rb_entry(parent, struct trace, rb_node);
799
89482da6
JA
800 if (t->bit->time < __t->bit->time)
801 p = &(*p)->rb_left;
802 else if (t->bit->time > __t->bit->time)
803 p = &(*p)->rb_right;
804 else if (t->bit->device < __t->bit->device)
e7c9f3ff
NS
805 p = &(*p)->rb_left;
806 else if (t->bit->device > __t->bit->device)
807 p = &(*p)->rb_right;
dcf0f7ed
JA
808 else if (t->bit->sequence < __t->bit->sequence)
809 p = &(*p)->rb_left;
0b07f23e 810 else /* >= sequence */
dcf0f7ed 811 p = &(*p)->rb_right;
7997c5b0
JA
812 }
813
814 rb_link_node(&t->rb_node, parent, p);
2a1b3424 815 rb_insert_color(&t->rb_node, root);
7997c5b0
JA
816 return 0;
817}
818
2a1b3424 819static inline int trace_rb_insert_sort(struct trace *t)
e3556946 820{
89482da6 821 if (!trace_rb_insert(t, &rb_sort_root)) {
2a1b3424
JA
822 rb_sort_entries++;
823 return 0;
824 }
825
826 return 1;
827}
828
210824c3 829static int trace_rb_insert_last(struct per_dev_info *pdi, struct trace *t)
2a1b3424 830{
210824c3
JA
831 struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
832
833 if (trace_rb_insert(t, &pci->rb_last))
834 return 1;
835
836 pci->rb_last_entries++;
837
838 if (pci->rb_last_entries > rb_batch * pdi->nfiles) {
839 struct rb_node *n = rb_first(&pci->rb_last);
840
841 t = rb_entry(n, struct trace, rb_node);
842 __put_trace_last(pdi, t);
2a1b3424
JA
843 }
844
210824c3 845 return 0;
2a1b3424
JA
846}
847
848static struct trace *trace_rb_find(dev_t device, unsigned long sequence,
849 struct rb_root *root, int order)
850{
851 struct rb_node *n = root->rb_node;
852 struct rb_node *prev = NULL;
e3556946
JA
853 struct trace *__t;
854
2a1b3424
JA
855 while (n) {
856 __t = rb_entry(n, struct trace, rb_node);
857 prev = n;
e3556946 858
0583b6a2 859 if (device < __t->bit->device)
2a1b3424 860 n = n->rb_left;
0583b6a2 861 else if (device > __t->bit->device)
2a1b3424 862 n = n->rb_right;
0583b6a2 863 else if (sequence < __t->bit->sequence)
2a1b3424 864 n = n->rb_left;
e3556946 865 else if (sequence > __t->bit->sequence)
2a1b3424 866 n = n->rb_right;
e3556946
JA
867 else
868 return __t;
869 }
870
2a1b3424
JA
871 /*
872 * hack - the list may not be sequence ordered because some
873 * events don't have sequence and time matched. so we end up
874 * being a little off in the rb lookup here, because we don't
875 * know the time we are looking for. compensate by browsing
876 * a little ahead from the last entry to find the match
877 */
878 if (order && prev) {
879 int max = 5;
880
881 while (((n = rb_next(prev)) != NULL) && max--) {
882 __t = rb_entry(n, struct trace, rb_node);
492da111 883
2a1b3424
JA
884 if (__t->bit->device == device &&
885 __t->bit->sequence == sequence)
886 return __t;
887
888 prev = n;
889 }
890 }
492da111 891
e3556946
JA
892 return NULL;
893}
894
2a1b3424 895static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
210824c3 896 struct per_cpu_info *pci,
2a1b3424
JA
897 unsigned long seq)
898{
210824c3 899 return trace_rb_find(pdi->dev, seq, &pci->rb_last, 0);
2a1b3424
JA
900}
901
f7bd1a9b 902static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
7997c5b0 903{
f7bd1a9b 904 struct rb_node **p = &pdi->rb_track.rb_node;
7997c5b0
JA
905 struct rb_node *parent = NULL;
906 struct io_track *__iot;
907
908 while (*p) {
909 parent = *p;
7997c5b0
JA
910 __iot = rb_entry(parent, struct io_track, rb_node);
911
f7bd1a9b 912 if (iot->sector < __iot->sector)
7997c5b0
JA
913 p = &(*p)->rb_left;
914 else if (iot->sector > __iot->sector)
915 p = &(*p)->rb_right;
916 else {
e7c9f3ff 917 fprintf(stderr,
ab197ca7
AB
918 "sector alias (%Lu) on device %d,%d!\n",
919 (unsigned long long) iot->sector,
f7bd1a9b 920 MAJOR(pdi->dev), MINOR(pdi->dev));
7997c5b0
JA
921 return 1;
922 }
923 }
924
925 rb_link_node(&iot->rb_node, parent, p);
f7bd1a9b 926 rb_insert_color(&iot->rb_node, &pdi->rb_track);
7997c5b0
JA
927 return 0;
928}
929
f7bd1a9b 930static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector)
7997c5b0 931{
f7bd1a9b 932 struct rb_node *n = pdi->rb_track.rb_node;
7997c5b0
JA
933 struct io_track *__iot;
934
2a1b3424
JA
935 while (n) {
936 __iot = rb_entry(n, struct io_track, rb_node);
7997c5b0 937
f7bd1a9b 938 if (sector < __iot->sector)
2a1b3424 939 n = n->rb_left;
7997c5b0 940 else if (sector > __iot->sector)
2a1b3424 941 n = n->rb_right;
7997c5b0
JA
942 else
943 return __iot;
944 }
945
946 return NULL;
947}
948
9e4cd1b8 949static struct io_track *find_track(struct per_dev_info *pdi, pid_t pid,
bfc70ad5 950 __u64 sector)
7997c5b0 951{
916b5501 952 struct io_track *iot;
7997c5b0 953
f7bd1a9b 954 iot = __find_track(pdi, sector);
7997c5b0
JA
955 if (!iot) {
956 iot = malloc(sizeof(*iot));
2990e589 957 iot->ppm = find_ppm(pid);
ebe2d1aa
JA
958 if (!iot->ppm)
959 iot->ppm = add_ppm_hash(pid, "unknown");
7997c5b0 960 iot->sector = sector;
f7bd1a9b 961 track_rb_insert(pdi, iot);
7997c5b0
JA
962 }
963
964 return iot;
965}
966
f7bd1a9b
JA
967static void log_track_frontmerge(struct per_dev_info *pdi,
968 struct blk_io_trace *t)
2e3e8ded
JA
969{
970 struct io_track *iot;
971
972 if (!track_ios)
973 return;
2e3e8ded 974
ae957cbc 975 iot = __find_track(pdi, t->sector + t_sec(t));
cb2a1a62 976 if (!iot) {
57ea8602
JA
977 if (verbose)
978 fprintf(stderr, "merge not found for (%d,%d): %llu\n",
979 MAJOR(pdi->dev), MINOR(pdi->dev),
980 (unsigned long long) t->sector + t_sec(t));
cb2a1a62 981 return;
2e3e8ded 982 }
cb2a1a62 983
f7bd1a9b 984 rb_erase(&iot->rb_node, &pdi->rb_track);
ae957cbc 985 iot->sector -= t_sec(t);
f7bd1a9b 986 track_rb_insert(pdi, iot);
2e3e8ded
JA
987}
988
f7bd1a9b 989static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t)
2e3e8ded
JA
990{
991 struct io_track *iot;
992
993 if (!track_ios)
994 return;
995
bfc70ad5 996 iot = find_track(pdi, t->pid, t->sector);
95c15013
JA
997 iot->allocation_time = t->time;
998}
999
753f9091
JA
1000static inline int is_remapper(struct per_dev_info *pdi)
1001{
1002 int major = MAJOR(pdi->dev);
1003
1004 return (major == 253 || major == 9);
1005}
1006
1007/*
1008 * for md/dm setups, the interesting cycle is Q -> C. So track queueing
1009 * time here, as dispatch time
1010 */
1011static void log_track_queue(struct per_dev_info *pdi, struct blk_io_trace *t)
1012{
1013 struct io_track *iot;
1014
1015 if (!track_ios)
1016 return;
1017 if (!is_remapper(pdi))
1018 return;
1019
bfc70ad5 1020 iot = find_track(pdi, t->pid, t->sector);
753f9091
JA
1021 iot->dispatch_time = t->time;
1022}
1023
95c15013 1024/*
b6076a9b 1025 * return time between rq allocation and insertion
95c15013 1026 */
f7bd1a9b
JA
1027static unsigned long long log_track_insert(struct per_dev_info *pdi,
1028 struct blk_io_trace *t)
95c15013 1029{
50adc0ba 1030 unsigned long long elapsed;
95c15013
JA
1031 struct io_track *iot;
1032
1033 if (!track_ios)
1034 return -1;
1035
bfc70ad5 1036 iot = find_track(pdi, t->pid, t->sector);
2e3e8ded 1037 iot->queue_time = t->time;
acd70d21
JA
1038
1039 if (!iot->allocation_time)
1040 return -1;
1041
50adc0ba
JA
1042 elapsed = iot->queue_time - iot->allocation_time;
1043
1044 if (per_process_stats) {
2990e589 1045 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
b9d40d6f 1046 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
50adc0ba 1047
b9d40d6f
JA
1048 if (ppi && elapsed > ppi->longest_allocation_wait[w])
1049 ppi->longest_allocation_wait[w] = elapsed;
50adc0ba
JA
1050 }
1051
1052 return elapsed;
2e3e8ded
JA
1053}
1054
1055/*
1056 * return time between queue and issue
1057 */
f7bd1a9b
JA
1058static unsigned long long log_track_issue(struct per_dev_info *pdi,
1059 struct blk_io_trace *t)
2e3e8ded 1060{
50adc0ba 1061 unsigned long long elapsed;
2e3e8ded
JA
1062 struct io_track *iot;
1063
1064 if (!track_ios)
1065 return -1;
1066 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
1067 return -1;
1068
f7bd1a9b 1069 iot = __find_track(pdi, t->sector);
cb2a1a62 1070 if (!iot) {
57ea8602
JA
1071 if (verbose)
1072 fprintf(stderr, "issue not found for (%d,%d): %llu\n",
1073 MAJOR(pdi->dev), MINOR(pdi->dev),
1074 (unsigned long long) t->sector);
2e3e8ded 1075 return -1;
cb2a1a62 1076 }
2e3e8ded
JA
1077
1078 iot->dispatch_time = t->time;
50adc0ba
JA
1079 elapsed = iot->dispatch_time - iot->queue_time;
1080
1081 if (per_process_stats) {
2990e589 1082 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
b9d40d6f 1083 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
50adc0ba 1084
b9d40d6f
JA
1085 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
1086 ppi->longest_dispatch_wait[w] = elapsed;
50adc0ba
JA
1087 }
1088
1089 return elapsed;
2e3e8ded
JA
1090}
1091
1092/*
1093 * return time between dispatch and complete
1094 */
f7bd1a9b
JA
1095static unsigned long long log_track_complete(struct per_dev_info *pdi,
1096 struct blk_io_trace *t)
2e3e8ded
JA
1097{
1098 unsigned long long elapsed;
1099 struct io_track *iot;
1100
1101 if (!track_ios)
1102 return -1;
2e3e8ded 1103
f7bd1a9b 1104 iot = __find_track(pdi, t->sector);
cb2a1a62 1105 if (!iot) {
57ea8602
JA
1106 if (verbose)
1107 fprintf(stderr,"complete not found for (%d,%d): %llu\n",
1108 MAJOR(pdi->dev), MINOR(pdi->dev),
1109 (unsigned long long) t->sector);
2e3e8ded 1110 return -1;
cb2a1a62 1111 }
2e3e8ded
JA
1112
1113 iot->completion_time = t->time;
1114 elapsed = iot->completion_time - iot->dispatch_time;
1115
50adc0ba 1116 if (per_process_stats) {
2990e589 1117 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
b9d40d6f 1118 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
50adc0ba 1119
b9d40d6f
JA
1120 if (ppi && elapsed > ppi->longest_completion_wait[w])
1121 ppi->longest_completion_wait[w] = elapsed;
50adc0ba
JA
1122 }
1123
2e3e8ded
JA
1124 /*
1125 * kill the trace, we don't need it after completion
1126 */
f7bd1a9b 1127 rb_erase(&iot->rb_node, &pdi->rb_track);
2e3e8ded
JA
1128 free(iot);
1129
1130 return elapsed;
1131}
1132
1133
9e4cd1b8 1134static struct io_stats *find_process_io_stats(pid_t pid)
152f6476 1135{
bfc70ad5 1136 struct per_process_info *ppi = find_ppi(pid);
152f6476
JA
1137
1138 if (!ppi) {
1139 ppi = malloc(sizeof(*ppi));
1140 memset(ppi, 0, sizeof(*ppi));
2990e589 1141 ppi->ppm = find_ppm(pid);
ebe2d1aa
JA
1142 if (!ppi->ppm)
1143 ppi->ppm = add_ppm_hash(pid, "unknown");
bfc70ad5
JA
1144 add_ppi_to_hash(ppi);
1145 add_ppi_to_list(ppi);
152f6476
JA
1146 }
1147
1148 return &ppi->io_stats;
1149}
1150
e7c9f3ff
NS
1151static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
1152{
1153 if (pdi->name)
1154 snprintf(buffer, size, "%s", pdi->name);
1155 else
f7bd1a9b 1156 snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev));
e7c9f3ff
NS
1157 return buffer;
1158}
1159
e7c9f3ff 1160static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
cfab07eb
AB
1161{
1162 unsigned long long this = bit->time;
e7c9f3ff 1163 unsigned long long last = pdi->last_reported_time;
cfab07eb 1164
e7c9f3ff
NS
1165 pdi->backwards = (this < last) ? 'B' : ' ';
1166 pdi->last_reported_time = this;
cfab07eb
AB
1167}
1168
fb2ec796
JA
1169static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
1170 int rw)
d0ca268b 1171{
fb2ec796 1172 if (rw) {
152f6476 1173 ios->mwrites++;
fb2ec796
JA
1174 ios->mwrite_kb += t_kb(t);
1175 } else {
152f6476 1176 ios->mreads++;
fb2ec796
JA
1177 ios->mread_kb += t_kb(t);
1178 }
152f6476
JA
1179}
1180
1181static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
1182 int rw)
1183{
fb2ec796 1184 __account_m(&pci->io_stats, t, rw);
152f6476
JA
1185
1186 if (per_process_stats) {
bfc70ad5 1187 struct io_stats *ios = find_process_io_stats(t->pid);
152f6476 1188
fb2ec796 1189 __account_m(ios, t, rw);
d0ca268b
JA
1190 }
1191}
1192
801646d6
CS
1193static inline void __account_pc_queue(struct io_stats *ios,
1194 struct blk_io_trace *t, int rw)
1195{
1196 if (rw) {
1197 ios->qwrites_pc++;
1198 ios->qwrite_kb_pc += t_kb(t);
1199 } else {
1200 ios->qreads_pc++;
1201 ios->qread_kb += t_kb(t);
1202 }
1203}
1204
1205static inline void account_pc_queue(struct blk_io_trace *t,
1206 struct per_cpu_info *pci, int rw)
1207{
1208 __account_pc_queue(&pci->io_stats, t, rw);
1209
1210 if (per_process_stats) {
1211 struct io_stats *ios = find_process_io_stats(t->pid);
1212
1213 __account_pc_queue(ios, t, rw);
1214 }
1215}
1216
1217static inline void __account_pc_issue(struct io_stats *ios, int rw,
1218 unsigned int bytes)
1219{
1220 if (rw) {
1221 ios->iwrites_pc++;
1222 ios->iwrite_kb_pc += bytes >> 10;
1223 } else {
1224 ios->ireads_pc++;
1225 ios->iread_kb_pc += bytes >> 10;
1226 }
1227}
1228
1229static inline void account_pc_issue(struct blk_io_trace *t,
1230 struct per_cpu_info *pci, int rw)
1231{
1232 __account_pc_issue(&pci->io_stats, rw, t->bytes);
1233
1234 if (per_process_stats) {
1235 struct io_stats *ios = find_process_io_stats(t->pid);
1236
1237 __account_pc_issue(ios, rw, t->bytes);
1238 }
1239}
1240
1241static inline void __account_pc_requeue(struct io_stats *ios,
1242 struct blk_io_trace *t, int rw)
1243{
1244 if (rw) {
1245 ios->wrqueue_pc++;
1246 ios->iwrite_kb_pc -= t_kb(t);
1247 } else {
1248 ios->rrqueue_pc++;
1249 ios->iread_kb_pc -= t_kb(t);
1250 }
1251}
1252
1253static inline void account_pc_requeue(struct blk_io_trace *t,
1254 struct per_cpu_info *pci, int rw)
1255{
1256 __account_pc_requeue(&pci->io_stats, t, rw);
1257
1258 if (per_process_stats) {
1259 struct io_stats *ios = find_process_io_stats(t->pid);
1260
1261 __account_pc_requeue(ios, t, rw);
1262 }
1263}
1264
1265static inline void __account_pc_c(struct io_stats *ios, int rw)
1266{
1267 if (rw)
1268 ios->cwrites_pc++;
1269 else
1270 ios->creads_pc++;
1271}
1272
1273static inline void account_pc_c(struct blk_io_trace *t,
1274 struct per_cpu_info *pci, int rw)
1275{
1276 __account_pc_c(&pci->io_stats, rw);
1277
1278 if (per_process_stats) {
1279 struct io_stats *ios = find_process_io_stats(t->pid);
1280
1281 __account_pc_c(ios, rw);
1282 }
1283}
1284
b6076a9b
JA
1285static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t,
1286 int rw)
d0ca268b
JA
1287{
1288 if (rw) {
152f6476 1289 ios->qwrites++;
ae957cbc 1290 ios->qwrite_kb += t_kb(t);
d0ca268b 1291 } else {
152f6476 1292 ios->qreads++;
ae957cbc 1293 ios->qread_kb += t_kb(t);
152f6476
JA
1294 }
1295}
1296
b6076a9b
JA
1297static inline void account_queue(struct blk_io_trace *t,
1298 struct per_cpu_info *pci, int rw)
152f6476 1299{
b6076a9b 1300 __account_queue(&pci->io_stats, t, rw);
152f6476
JA
1301
1302 if (per_process_stats) {
bfc70ad5 1303 struct io_stats *ios = find_process_io_stats(t->pid);
152f6476 1304
b6076a9b 1305 __account_queue(ios, t, rw);
d0ca268b
JA
1306 }
1307}
1308
e21dc4dd 1309static inline void __account_c(struct io_stats *ios, int rw, int bytes)
d0ca268b
JA
1310{
1311 if (rw) {
152f6476
JA
1312 ios->cwrites++;
1313 ios->cwrite_kb += bytes >> 10;
d0ca268b 1314 } else {
152f6476
JA
1315 ios->creads++;
1316 ios->cread_kb += bytes >> 10;
1317 }
1318}
1319
1320static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
1321 int rw, int bytes)
1322{
1323 __account_c(&pci->io_stats, rw, bytes);
1324
1325 if (per_process_stats) {
bfc70ad5 1326 struct io_stats *ios = find_process_io_stats(t->pid);
152f6476
JA
1327
1328 __account_c(ios, rw, bytes);
d0ca268b
JA
1329 }
1330}
1331
b6076a9b
JA
1332static inline void __account_issue(struct io_stats *ios, int rw,
1333 unsigned int bytes)
afd2d7ad 1334{
1335 if (rw) {
152f6476
JA
1336 ios->iwrites++;
1337 ios->iwrite_kb += bytes >> 10;
afd2d7ad 1338 } else {
152f6476
JA
1339 ios->ireads++;
1340 ios->iread_kb += bytes >> 10;
afd2d7ad 1341 }
1342}
1343
b6076a9b
JA
1344static inline void account_issue(struct blk_io_trace *t,
1345 struct per_cpu_info *pci, int rw)
d0ca268b 1346{
b6076a9b 1347 __account_issue(&pci->io_stats, rw, t->bytes);
152f6476
JA
1348
1349 if (per_process_stats) {
bfc70ad5 1350 struct io_stats *ios = find_process_io_stats(t->pid);
d5396421 1351
b6076a9b 1352 __account_issue(ios, rw, t->bytes);
152f6476
JA
1353 }
1354}
1355
06639b27
JA
1356static inline void __account_unplug(struct io_stats *ios, int timer)
1357{
1358 if (timer)
1359 ios->timer_unplugs++;
1360 else
1361 ios->io_unplugs++;
1362}
1363
1364static inline void account_unplug(struct blk_io_trace *t,
1365 struct per_cpu_info *pci, int timer)
1366{
1367 __account_unplug(&pci->io_stats, timer);
1368
1369 if (per_process_stats) {
bfc70ad5 1370 struct io_stats *ios = find_process_io_stats(t->pid);
06639b27
JA
1371
1372 __account_unplug(ios, timer);
1373 }
1374}
1375
4054070a
JA
1376static inline void __account_requeue(struct io_stats *ios,
1377 struct blk_io_trace *t, int rw)
1378{
1379 if (rw) {
1380 ios->wrqueue++;
1381 ios->iwrite_kb -= t_kb(t);
1382 } else {
1383 ios->rrqueue++;
1384 ios->iread_kb -= t_kb(t);
1385 }
1386}
1387
1388static inline void account_requeue(struct blk_io_trace *t,
1389 struct per_cpu_info *pci, int rw)
1390{
1391 __account_requeue(&pci->io_stats, t, rw);
1392
1393 if (per_process_stats) {
bfc70ad5 1394 struct io_stats *ios = find_process_io_stats(t->pid);
4054070a
JA
1395
1396 __account_requeue(ios, t, rw);
1397 }
1398}
1399
f7bd1a9b
JA
1400static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci,
1401 struct blk_io_trace *t, char *act)
ab197ca7 1402{
f7bd1a9b 1403 process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
ab197ca7
AB
1404}
1405
f7bd1a9b
JA
1406static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci,
1407 struct blk_io_trace *t, char *act)
b6076a9b 1408{
f7bd1a9b 1409 process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL);
b6076a9b
JA
1410}
1411
ab197ca7
AB
1412static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
1413 char *act)
1414{
b6076a9b 1415 process_fmt(act, pci, t, -1, 0, NULL);
ab197ca7 1416}
2e3e8ded 1417
f7bd1a9b
JA
1418static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci,
1419 struct blk_io_trace *t, char *act)
ab197ca7 1420{
f7bd1a9b 1421 process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
d0ca268b
JA
1422}
1423
f7bd1a9b
JA
1424static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci,
1425 struct blk_io_trace *t, char *act)
d0ca268b 1426{
a01516de 1427 if (act[0] == 'F')
f7bd1a9b 1428 log_track_frontmerge(pdi, t);
2e3e8ded 1429
ab197ca7 1430 process_fmt(act, pci, t, -1ULL, 0, NULL);
d0ca268b
JA
1431}
1432
dfe34da1 1433static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
3639a11e 1434 char *act)
dfe34da1 1435{
ab197ca7 1436 process_fmt(act, pci, t, -1ULL, 0, NULL);
dfe34da1
JA
1437}
1438
d5396421 1439static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
3639a11e 1440 char *act)
d0ca268b 1441{
ab197ca7 1442 process_fmt(act, pci, t, -1ULL, 0, NULL);
d0ca268b
JA
1443}
1444
ab197ca7 1445static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
3639a11e 1446 char *act)
67e14fdc 1447{
ab197ca7 1448 process_fmt(act, pci, t, -1ULL, 0, NULL);
67e14fdc
JA
1449}
1450
93f1c611
JA
1451static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t,
1452 char *act)
1453{
1454 process_fmt(act, pci, t, -1ULL, 0, NULL);
1455}
1456
ab197ca7 1457static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
d0ca268b 1458{
ab197ca7 1459 unsigned char *buf = (unsigned char *) t + sizeof(*t);
d0ca268b 1460
ab197ca7 1461 process_fmt(act, pci, t, -1ULL, t->pdu_len, buf);
d0ca268b
JA
1462}
1463
c82a8c9d
CS
1464static void dump_trace_pc(struct blk_io_trace *t, struct per_dev_info *pdi,
1465 struct per_cpu_info *pci)
d0ca268b 1466{
c82a8c9d 1467 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
56f2af81
JA
1468 int act = t->action & 0xffff;
1469
1470 switch (act) {
d0ca268b 1471 case __BLK_TA_QUEUE:
3639a11e 1472 log_generic(pci, t, "Q");
801646d6 1473 account_pc_queue(t, pci, w);
d0ca268b
JA
1474 break;
1475 case __BLK_TA_GETRQ:
3639a11e 1476 log_generic(pci, t, "G");
d0ca268b
JA
1477 break;
1478 case __BLK_TA_SLEEPRQ:
3639a11e 1479 log_generic(pci, t, "S");
d0ca268b
JA
1480 break;
1481 case __BLK_TA_REQUEUE:
c82a8c9d
CS
1482 /*
1483 * can happen if we miss traces, don't let it go
1484 * below zero
1485 */
1486 if (pdi->cur_depth[w])
1487 pdi->cur_depth[w]--;
801646d6 1488 account_pc_requeue(t, pci, w);
3639a11e 1489 log_generic(pci, t, "R");
d0ca268b
JA
1490 break;
1491 case __BLK_TA_ISSUE:
801646d6 1492 account_pc_issue(t, pci, w);
c82a8c9d
CS
1493 pdi->cur_depth[w]++;
1494 if (pdi->cur_depth[w] > pdi->max_depth[w])
1495 pdi->max_depth[w] = pdi->cur_depth[w];
ab197ca7 1496 log_pc(pci, t, "D");
d0ca268b
JA
1497 break;
1498 case __BLK_TA_COMPLETE:
c82a8c9d
CS
1499 if (pdi->cur_depth[w])
1500 pdi->cur_depth[w]--;
3639a11e 1501 log_pc(pci, t, "C");
801646d6 1502 account_pc_c(t, pci, w);
d0ca268b 1503 break;
56f2af81
JA
1504 case __BLK_TA_INSERT:
1505 log_pc(pci, t, "I");
1506 break;
d0ca268b 1507 default:
56f2af81 1508 fprintf(stderr, "Bad pc action %x\n", act);
87b72777 1509 break;
d0ca268b 1510 }
d0ca268b
JA
1511}
1512
f7bd1a9b
JA
1513static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi,
1514 struct per_cpu_info *pci)
d0ca268b 1515{
649c7b66 1516 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
7997c5b0 1517 int act = t->action & 0xffff;
d0ca268b 1518
7997c5b0 1519 switch (act) {
d0ca268b 1520 case __BLK_TA_QUEUE:
753f9091 1521 log_track_queue(pdi, t);
b6076a9b 1522 account_queue(t, pci, w);
3639a11e 1523 log_queue(pci, t, "Q");
d0ca268b 1524 break;
b6076a9b 1525 case __BLK_TA_INSERT:
f7bd1a9b 1526 log_insert(pdi, pci, t, "I");
b6076a9b 1527 break;
d0ca268b 1528 case __BLK_TA_BACKMERGE:
152f6476 1529 account_m(t, pci, w);
f7bd1a9b 1530 log_merge(pdi, pci, t, "M");
d0ca268b
JA
1531 break;
1532 case __BLK_TA_FRONTMERGE:
152f6476 1533 account_m(t, pci, w);
f7bd1a9b 1534 log_merge(pdi, pci, t, "F");
d0ca268b
JA
1535 break;
1536 case __BLK_TA_GETRQ:
f7bd1a9b 1537 log_track_getrq(pdi, t);
3639a11e 1538 log_generic(pci, t, "G");
d0ca268b
JA
1539 break;
1540 case __BLK_TA_SLEEPRQ:
3639a11e 1541 log_generic(pci, t, "S");
d0ca268b
JA
1542 break;
1543 case __BLK_TA_REQUEUE:
65f2deb5
JA
1544 /*
1545 * can happen if we miss traces, don't let it go
1546 * below zero
1547 */
1548 if (pdi->cur_depth[w])
1549 pdi->cur_depth[w]--;
4054070a 1550 account_requeue(t, pci, w);
3639a11e 1551 log_queue(pci, t, "R");
d0ca268b
JA
1552 break;
1553 case __BLK_TA_ISSUE:
b6076a9b 1554 account_issue(t, pci, w);
649c7b66
JA
1555 pdi->cur_depth[w]++;
1556 if (pdi->cur_depth[w] > pdi->max_depth[w])
1557 pdi->max_depth[w] = pdi->cur_depth[w];
f7bd1a9b 1558 log_issue(pdi, pci, t, "D");
d0ca268b
JA
1559 break;
1560 case __BLK_TA_COMPLETE:
65f2deb5
JA
1561 if (pdi->cur_depth[w])
1562 pdi->cur_depth[w]--;
152f6476 1563 account_c(t, pci, w, t->bytes);
f7bd1a9b 1564 log_complete(pdi, pci, t, "C");
d0ca268b 1565 break;
88b1a526 1566 case __BLK_TA_PLUG:
3639a11e 1567 log_action(pci, t, "P");
88b1a526 1568 break;
3639a11e 1569 case __BLK_TA_UNPLUG_IO:
06639b27 1570 account_unplug(t, pci, 0);
3639a11e
JA
1571 log_unplug(pci, t, "U");
1572 break;
1573 case __BLK_TA_UNPLUG_TIMER:
06639b27 1574 account_unplug(t, pci, 1);
3639a11e 1575 log_unplug(pci, t, "UT");
88b1a526 1576 break;
93f1c611
JA
1577 case __BLK_TA_SPLIT:
1578 log_split(pci, t, "X");
1579 break;
1580 case __BLK_TA_BOUNCE:
1581 log_generic(pci, t, "B");
1582 break;
a8f30e64
JA
1583 case __BLK_TA_REMAP:
1584 log_generic(pci, t, "A");
1585 break;
d0ca268b
JA
1586 default:
1587 fprintf(stderr, "Bad fs action %x\n", t->action);
1f79c4a0 1588 break;
d0ca268b 1589 }
d0ca268b
JA
1590}
1591
ff3a732c
JA
1592static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
1593 struct per_dev_info *pdi)
d0ca268b 1594{
234db09d 1595 if (text_output) {
1a15f6a8
AB
1596 if (t->action == BLK_TN_MESSAGE)
1597 handle_notify(t);
1598 else if (t->action & BLK_TC_ACT(BLK_TC_PC))
c82a8c9d 1599 dump_trace_pc(t, pdi, pci);
234db09d
AB
1600 else
1601 dump_trace_fs(t, pdi, pci);
1602 }
87b72777 1603
20ed6177
JA
1604 if (!pdi->events)
1605 pdi->first_reported_time = t->time;
1606
e7c9f3ff 1607 pdi->events++;
a2594911
AB
1608
1609 output_binary(t, sizeof(*t) + t->pdu_len);
d0ca268b
JA
1610}
1611
4c523165
JA
1612/*
1613 * print in a proper way, not too small and not too big. if more than
1614 * 1000,000K, turn into M and so on
1615 */
1616static char *size_cnv(char *dst, unsigned long long num, int in_kb)
1617{
da19e768 1618 char suff[] = { '\0', 'K', 'M', 'G', 'P' };
0dc3602c 1619 unsigned int i = 0;
4c523165
JA
1620
1621 if (in_kb)
1622 i++;
1623
0dc3602c 1624 while (num > 1000 * 1000ULL && (i < sizeof(suff) - 1)) {
4c523165
JA
1625 i++;
1626 num /= 1000;
1627 }
1628
1629 sprintf(dst, "%'8Lu%c", num, suff[i]);
1630 return dst;
1631}
1632
649c7b66
JA
1633static void dump_io_stats(struct per_dev_info *pdi, struct io_stats *ios,
1634 char *msg)
5c017e4b 1635{
4c523165
JA
1636 static char x[256], y[256];
1637
152f6476
JA
1638 fprintf(ofp, "%s\n", msg);
1639
4c523165
JA
1640 fprintf(ofp, " Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads, 0), size_cnv(y, ios->qread_kb, 1));
1641 fprintf(ofp, " Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites, 0), size_cnv(y, ios->qwrite_kb, 1));
4c523165
JA
1642 fprintf(ofp, " Read Dispatches: %s, %siB\t", size_cnv(x, ios->ireads, 0), size_cnv(y, ios->iread_kb, 1));
1643 fprintf(ofp, " Write Dispatches: %s, %siB\n", size_cnv(x, ios->iwrites, 0), size_cnv(y, ios->iwrite_kb, 1));
4054070a
JA
1644 fprintf(ofp, " Reads Requeued: %s\t\t", size_cnv(x, ios->rrqueue, 0));
1645 fprintf(ofp, " Writes Requeued: %s\n", size_cnv(x, ios->wrqueue, 0));
4c523165
JA
1646 fprintf(ofp, " Reads Completed: %s, %siB\t", size_cnv(x, ios->creads, 0), size_cnv(y, ios->cread_kb, 1));
1647 fprintf(ofp, " Writes Completed: %s, %siB\n", size_cnv(x, ios->cwrites, 0), size_cnv(y, ios->cwrite_kb, 1));
fb2ec796
JA
1648 fprintf(ofp, " Read Merges: %s, %siB\t", size_cnv(x, ios->mreads, 0), size_cnv(y, ios->mread_kb, 1));
1649 fprintf(ofp, " Write Merges: %s, %siB\n", size_cnv(x, ios->mwrites, 0), size_cnv(y, ios->mwrite_kb, 1));
649c7b66
JA
1650 if (pdi) {
1651 fprintf(ofp, " Read depth: %'8u%8c\t", pdi->max_depth[0], ' ');
1652 fprintf(ofp, " Write depth: %'8u\n", pdi->max_depth[1]);
1653 }
801646d6
CS
1654 if (ios->qreads_pc || ios->qwrites_pc || ios->ireads_pc || ios->iwrites_pc ||
1655 ios->rrqueue_pc || ios->wrqueue_pc || ios->creads_pc || ios->cwrites_pc) {
1656 fprintf(ofp, " PC Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads_pc, 0), size_cnv(y, ios->qread_kb_pc, 1));
1657 fprintf(ofp, " PC Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites_pc, 0), size_cnv(y, ios->qwrite_kb_pc, 1));
1658 fprintf(ofp, " PC Read Disp.: %s, %siB\t", size_cnv(x, ios->ireads_pc, 0), size_cnv(y, ios->iread_kb_pc, 1));
1659 fprintf(ofp, " PC Write Disp.: %s, %siB\n", size_cnv(x, ios->iwrites_pc, 0), size_cnv(y, ios->iwrite_kb_pc, 1));
1660 fprintf(ofp, " PC Reads Req.: %s\t\t", size_cnv(x, ios->rrqueue_pc, 0));
1661 fprintf(ofp, " PC Writes Req.: %s\n", size_cnv(x, ios->wrqueue_pc, 0));
1662 fprintf(ofp, " PC Reads Compl.: %s\t\t", size_cnv(x, ios->creads_pc, 0));
1663 fprintf(ofp, " PC Writes Compl.: %s\n", size_cnv(x, ios->cwrites, 0));
1664 }
06639b27
JA
1665 fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
1666 fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
5c017e4b
JA
1667}
1668
50adc0ba
JA
1669static void dump_wait_stats(struct per_process_info *ppi)
1670{
b9d40d6f
JA
1671 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
1672 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
1673 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
1674 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
1675 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
1676 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
1677
1678 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
1679 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
1680 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
1681 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
1682 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
1683 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
50adc0ba
JA
1684}
1685
886ecf0e
JA
1686static int ppi_name_compare(const void *p1, const void *p2)
1687{
1688 struct per_process_info *ppi1 = *((struct per_process_info **) p1);
1689 struct per_process_info *ppi2 = *((struct per_process_info **) p2);
1690 int res;
1691
2990e589 1692 res = strverscmp(ppi1->ppm->comm, ppi2->ppm->comm);
886ecf0e 1693 if (!res)
2990e589 1694 res = ppi1->ppm->pid > ppi2->ppm->pid;
886ecf0e
JA
1695
1696 return res;
1697}
1698
1699static void sort_process_list(void)
1700{
1701 struct per_process_info **ppis;
1702 struct per_process_info *ppi;
1703 int i = 0;
1704
1705 ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *));
1706
1707 ppi = ppi_list;
1708 while (ppi) {
06e6f286 1709 ppis[i++] = ppi;
886ecf0e
JA
1710 ppi = ppi->list_next;
1711 }
1712
06e6f286 1713 qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare);
886ecf0e
JA
1714
1715 i = ppi_list_entries - 1;
1716 ppi_list = NULL;
1717 while (i >= 0) {
1718 ppi = ppis[i];
1719
1720 ppi->list_next = ppi_list;
1721 ppi_list = ppi;
1722 i--;
1723 }
50c38702
JA
1724
1725 free(ppis);
886ecf0e
JA
1726}
1727
152f6476
JA
1728static void show_process_stats(void)
1729{
1730 struct per_process_info *ppi;
1731
886ecf0e
JA
1732 sort_process_list();
1733
152f6476
JA
1734 ppi = ppi_list;
1735 while (ppi) {
2990e589 1736 struct process_pid_map *ppm = ppi->ppm;
ce8b6b4f
JA
1737 char name[64];
1738
715d8021 1739 if (ppi->more_than_one)
2990e589 1740 sprintf(name, "%s (%u, ...)", ppm->comm, ppm->pid);
715d8021 1741 else
2990e589 1742 sprintf(name, "%s (%u)", ppm->comm, ppm->pid);
bf0720af 1743
649c7b66 1744 dump_io_stats(NULL, &ppi->io_stats, name);
50adc0ba 1745 dump_wait_stats(ppi);
152f6476
JA
1746 ppi = ppi->list_next;
1747 }
1748
1749 fprintf(ofp, "\n");
1750}
1751
e7c9f3ff 1752static void show_device_and_cpu_stats(void)
d0ca268b 1753{
e7c9f3ff
NS
1754 struct per_dev_info *pdi;
1755 struct per_cpu_info *pci;
1756 struct io_stats total, *ios;
20ed6177 1757 unsigned long long rrate, wrate, msec;
e7c9f3ff
NS
1758 int i, j, pci_events;
1759 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1760 char name[32];
8a82e321 1761 double ratio;
e7c9f3ff
NS
1762
1763 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1764
1765 memset(&total, 0, sizeof(total));
1766 pci_events = 0;
1767
1768 if (i > 0)
1769 fprintf(ofp, "\n");
1770
1771 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1772 if (!pci->nelems)
1773 continue;
1774
1775 ios = &pci->io_stats;
1776 total.qreads += ios->qreads;
1777 total.qwrites += ios->qwrites;
1778 total.creads += ios->creads;
1779 total.cwrites += ios->cwrites;
1780 total.mreads += ios->mreads;
1781 total.mwrites += ios->mwrites;
1782 total.ireads += ios->ireads;
1783 total.iwrites += ios->iwrites;
4054070a
JA
1784 total.rrqueue += ios->rrqueue;
1785 total.wrqueue += ios->wrqueue;
e7c9f3ff
NS
1786 total.qread_kb += ios->qread_kb;
1787 total.qwrite_kb += ios->qwrite_kb;
1788 total.cread_kb += ios->cread_kb;
1789 total.cwrite_kb += ios->cwrite_kb;
1790 total.iread_kb += ios->iread_kb;
1791 total.iwrite_kb += ios->iwrite_kb;
fb2ec796
JA
1792 total.mread_kb += ios->mread_kb;
1793 total.mwrite_kb += ios->mwrite_kb;
801646d6
CS
1794
1795 total.qreads_pc += ios->qreads_pc;
1796 total.qwrites_pc += ios->qwrites_pc;
1797 total.creads_pc += ios->creads_pc;
1798 total.cwrites_pc += ios->cwrites_pc;
1799 total.ireads_pc += ios->ireads_pc;
1800 total.iwrites_pc += ios->iwrites_pc;
1801 total.rrqueue_pc += ios->rrqueue_pc;
1802 total.wrqueue_pc += ios->wrqueue_pc;
1803 total.qread_kb_pc += ios->qread_kb_pc;
1804 total.qwrite_kb_pc += ios->qwrite_kb_pc;
1805 total.iread_kb_pc += ios->iread_kb_pc;
1806 total.iwrite_kb_pc += ios->iwrite_kb_pc;
1807
06639b27
JA
1808 total.timer_unplugs += ios->timer_unplugs;
1809 total.io_unplugs += ios->io_unplugs;
e7c9f3ff
NS
1810
1811 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1812 j, get_dev_name(pdi, name, sizeof(name)));
649c7b66 1813 dump_io_stats(pdi, ios, line);
e7c9f3ff
NS
1814 pci_events++;
1815 }
5c017e4b 1816
e7c9f3ff
NS
1817 if (pci_events > 1) {
1818 fprintf(ofp, "\n");
1819 snprintf(line, sizeof(line) - 1, "Total (%s):",
1820 get_dev_name(pdi, name, sizeof(name)));
649c7b66 1821 dump_io_stats(NULL, &total, line);
e7c9f3ff 1822 }
d0ca268b 1823
20ed6177 1824 wrate = rrate = 0;
20ed6177
JA
1825 msec = (pdi->last_reported_time - pdi->first_reported_time) / 1000000;
1826 if (msec) {
1827 rrate = 1000 * total.cread_kb / msec;
1828 wrate = 1000 * total.cwrite_kb / msec;
1829 }
1830
dce0f678
AB
1831 fprintf(ofp, "\nThroughput (R/W): %'LuKiB/s / %'LuKiB/s\n",
1832 rrate, wrate);
1833 fprintf(ofp, "Events (%s): %'Lu entries\n",
1834 get_dev_name(pdi, line, sizeof(line)), pdi->events);
492da111
AB
1835
1836 collect_pdi_skips(pdi);
8a82e321
MZ
1837 if (!pdi->skips && !pdi->events)
1838 ratio = 0.0;
1839 else
1840 ratio = 100.0 * ((double)pdi->seq_skips /
1841 (double)(pdi->events + pdi->seq_skips));
492da111 1842 fprintf(ofp, "Skips: %'lu forward (%'llu - %5.1lf%%)\n",
8a82e321 1843 pdi->skips, pdi->seq_skips, ratio);
e7c9f3ff 1844 }
d0ca268b
JA
1845}
1846
4f0ae44f
JA
1847static void find_genesis(void)
1848{
1849 struct trace *t = trace_list;
1850
1851 genesis_time = -1ULL;
1852 while (t != NULL) {
1853 if (t->bit->time < genesis_time)
1854 genesis_time = t->bit->time;
1855
1856 t = t->next;
1857 }
7bd4fd0a
OK
1858
1859 /* The time stamp record will usually be the first
1860 * record in the trace, but not always.
1861 */
1862 if (start_timestamp
1863 && start_timestamp != genesis_time) {
1864 long delta = genesis_time - start_timestamp;
1865
1866 abs_start_time.tv_sec += SECONDS(delta);
1867 abs_start_time.tv_nsec += NANO_SECONDS(delta);
1868 if (abs_start_time.tv_nsec < 0) {
1869 abs_start_time.tv_nsec += 1000000000;
1870 abs_start_time.tv_sec -= 1;
1871 } else
1872 if (abs_start_time.tv_nsec > 1000000000) {
1873 abs_start_time.tv_nsec -= 1000000000;
1874 abs_start_time.tv_sec += 1;
1875 }
1876 }
4f0ae44f
JA
1877}
1878
7f4d89e6 1879static inline int check_stopwatch(struct blk_io_trace *bit)
4f0ae44f 1880{
7f4d89e6
JA
1881 if (bit->time < stopwatch_end &&
1882 bit->time >= stopwatch_start)
4f0ae44f
JA
1883 return 0;
1884
1885 return 1;
1886}
1887
53c68c88
JA
1888/*
1889 * return youngest entry read
1890 */
1891static int sort_entries(unsigned long long *youngest)
4f0ae44f 1892{
210824c3
JA
1893 struct per_dev_info *pdi = NULL;
1894 struct per_cpu_info *pci = NULL;
4f0ae44f 1895 struct trace *t;
4f0ae44f
JA
1896
1897 if (!genesis_time)
1898 find_genesis();
1899
d6222db8 1900 *youngest = 0;
4f0ae44f
JA
1901 while ((t = trace_list) != NULL) {
1902 struct blk_io_trace *bit = t->bit;
1903
1904 trace_list = t->next;
1905
7f4d89e6 1906 bit->time -= genesis_time;
4f0ae44f 1907
d6222db8
JA
1908 if (bit->time < *youngest || !*youngest)
1909 *youngest = bit->time;
1910
210824c3
JA
1911 if (!pdi || pdi->dev != bit->device) {
1912 pdi = get_dev_info(bit->device);
1913 pci = NULL;
1914 }
1915
1916 if (!pci || pci->cpu != bit->cpu)
1917 pci = get_cpu_info(pdi, bit->cpu);
1918
1919 if (bit->sequence < pci->smallest_seq_read)
1920 pci->smallest_seq_read = bit->sequence;
774a1a10 1921
7f4d89e6 1922 if (check_stopwatch(bit)) {
4f0ae44f
JA
1923 bit_free(bit);
1924 t_free(t);
1925 continue;
1926 }
1927
2a1b3424 1928 if (trace_rb_insert_sort(t))
53c68c88 1929 return -1;
4f0ae44f
JA
1930 }
1931
53c68c88 1932 return 0;
4f0ae44f
JA
1933}
1934
824c2b39
JA
1935/*
1936 * to continue, we must have traces from all online cpus in the tree
1937 */
1938static int check_cpu_map(struct per_dev_info *pdi)
1939{
1940 unsigned long *cpu_map;
1941 struct rb_node *n;
1942 struct trace *__t;
1943 unsigned int i;
1944 int ret, cpu;
1945
1946 /*
1947 * create a map of the cpus we have traces for
1948 */
1949 cpu_map = malloc(pdi->cpu_map_max / sizeof(long));
1950 n = rb_first(&rb_sort_root);
1951 while (n) {
1952 __t = rb_entry(n, struct trace, rb_node);
1953 cpu = __t->bit->cpu;
1954
1955 cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
1956 n = rb_next(n);
1957 }
1958
1959 /*
b1c8e614
JA
1960 * we can't continue if pdi->cpu_map has entries set that we don't
1961 * have in the sort rbtree. the opposite is not a problem, though
824c2b39
JA
1962 */
1963 ret = 0;
1964 for (i = 0; i < pdi->cpu_map_max / CPUS_PER_LONG; i++) {
1965 if (pdi->cpu_map[i] & ~(cpu_map[i])) {
1966 ret = 1;
1967 break;
1968 }
1969 }
1970
1971 free(cpu_map);
1972 return ret;
1973}
1974
a141a7cd 1975static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force)
2a1b3424 1976{
1ca323a5 1977 struct blk_io_trace *bit = t->bit;
210824c3
JA
1978 unsigned long expected_sequence;
1979 struct per_cpu_info *pci;
1ca323a5 1980 struct trace *__t;
492da111 1981
210824c3
JA
1982 pci = get_cpu_info(pdi, bit->cpu);
1983 expected_sequence = pci->last_sequence + 1;
1984
774a1a10 1985 if (!expected_sequence) {
774a1a10
JA
1986 /*
1987 * 1 should be the first entry, just allow it
1988 */
1989 if (bit->sequence == 1)
1990 return 0;
210824c3 1991 if (bit->sequence == pci->smallest_seq_read)
79ee9704 1992 return 0;
774a1a10 1993
824c2b39 1994 return check_cpu_map(pdi);
774a1a10 1995 }
2a1b3424
JA
1996
1997 if (bit->sequence == expected_sequence)
1998 return 0;
1999
2a1b3424 2000 /*
1c7c54aa
JA
2001 * we may not have seen that sequence yet. if we are not doing
2002 * the final run, break and wait for more entries.
1c24add6 2003 */
210824c3
JA
2004 if (expected_sequence < pci->smallest_seq_read) {
2005 __t = trace_rb_find_last(pdi, pci, expected_sequence);
1ca323a5 2006 if (!__t)
1c7c54aa 2007 goto skip;
2a1b3424 2008
1ca323a5 2009 __put_trace_last(pdi, __t);
2a1b3424 2010 return 0;
a141a7cd
JA
2011 } else if (!force) {
2012 return 1;
0b07f23e 2013 } else {
1c7c54aa 2014skip:
66930177 2015 if (check_current_skips(pci, bit->sequence))
492da111
AB
2016 return 0;
2017
965eca2d 2018 if (expected_sequence < bit->sequence)
66930177 2019 insert_skip(pci, expected_sequence, bit->sequence - 1);
1c7c54aa
JA
2020 return 0;
2021 }
2a1b3424
JA
2022}
2023
a649216c 2024static void show_entries_rb(int force)
8fc0abbc 2025{
1f7afa72
JA
2026 struct per_dev_info *pdi = NULL;
2027 struct per_cpu_info *pci = NULL;
8fc0abbc 2028 struct blk_io_trace *bit;
3aabcd89 2029 struct rb_node *n;
8fc0abbc 2030 struct trace *t;
1f7afa72 2031
7d747d22 2032 while ((n = rb_first(&rb_sort_root)) != NULL) {
dd90748f 2033 if (is_done() && !force && !pipeline)
1f7afa72 2034 break;
8fc0abbc
JA
2035
2036 t = rb_entry(n, struct trace, rb_node);
2037 bit = t->bit;
2038
a43c1c17
JA
2039 if (read_sequence - t->read_sequence < 1 && !force)
2040 break;
2041
210824c3 2042 if (!pdi || pdi->dev != bit->device) {
287fa3d6 2043 pdi = get_dev_info(bit->device);
210824c3
JA
2044 pci = NULL;
2045 }
1f7afa72 2046
e7c9f3ff
NS
2047 if (!pdi) {
2048 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
2049 MAJOR(bit->device), MINOR(bit->device));
2050 break;
2051 }
1f7afa72 2052
a141a7cd
JA
2053 if (check_sequence(pdi, t, force))
2054 break;
cb2a1a62 2055
a141a7cd
JA
2056 if (!force && bit->time > last_allowed_time)
2057 break;
8fc0abbc 2058
4f0ae44f 2059 check_time(pdi, bit);
8fc0abbc 2060
4f0ae44f
JA
2061 if (!pci || pci->cpu != bit->cpu)
2062 pci = get_cpu_info(pdi, bit->cpu);
287fa3d6 2063
210824c3
JA
2064 pci->last_sequence = bit->sequence;
2065
cbc927b6
JA
2066 pci->nelems++;
2067
66930177 2068 if (bit->action & (act_mask << BLK_TC_SHIFT))
98f8386b 2069 dump_trace(bit, pci, pdi);
87b72777 2070
2a1b3424 2071 put_trace(pdi, t);
cb2a1a62 2072 }
8fc0abbc
JA
2073}
2074
c0e0dbc2 2075static int read_data(int fd, void *buffer, int bytes, int block, int *fdblock)
1f79c4a0
JA
2076{
2077 int ret, bytes_left, fl;
2078 void *p;
2079
c0e0dbc2
JA
2080 if (block != *fdblock) {
2081 fl = fcntl(fd, F_GETFL);
1f79c4a0 2082
c0e0dbc2
JA
2083 if (!block) {
2084 *fdblock = 0;
2085 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
2086 } else {
2087 *fdblock = 1;
2088 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
2089 }
2090 }
1f79c4a0
JA
2091
2092 bytes_left = bytes;
2093 p = buffer;
2094 while (bytes_left > 0) {
2095 ret = read(fd, p, bytes_left);
2096 if (!ret)
2097 return 1;
2098 else if (ret < 0) {
db7e0552 2099 if (errno != EAGAIN) {
1f79c4a0 2100 perror("read");
db7e0552
JA
2101 return -1;
2102 }
a649216c 2103
5c0f40f7
JA
2104 /*
2105 * never do partial reads. we can return if we
2106 * didn't read anything and we should not block,
2107 * otherwise wait for data
2108 */
2109 if ((bytes_left == bytes) && !block)
2110 return 1;
2111
2112 usleep(10);
2113 continue;
1f79c4a0
JA
2114 } else {
2115 p += ret;
2116 bytes_left -= ret;
2117 }
2118 }
2119
2120 return 0;
2121}
2122
017d1660
JA
2123static inline __u16 get_pdulen(struct blk_io_trace *bit)
2124{
2125 if (data_is_native)
2126 return bit->pdu_len;
2127
2128 return __bswap_16(bit->pdu_len);
2129}
2130
2131static inline __u32 get_magic(struct blk_io_trace *bit)
2132{
2133 if (data_is_native)
2134 return bit->magic;
2135
2136 return __bswap_32(bit->magic);
2137}
2138
c0e0dbc2 2139static int read_events(int fd, int always_block, int *fdblock)
cb2a1a62 2140{
287fa3d6 2141 struct per_dev_info *pdi = NULL;
e820abd7 2142 unsigned int events = 0;
7d747d22
JA
2143
2144 while (!is_done() && events < rb_batch) {
2145 struct blk_io_trace *bit;
2146 struct trace *t;
db7e0552 2147 int pdu_len, should_block, ret;
7d747d22
JA
2148 __u32 magic;
2149
d36421e4 2150 bit = bit_alloc();
cb2a1a62 2151
c0e0dbc2
JA
2152 should_block = !events || always_block;
2153
db7e0552
JA
2154 ret = read_data(fd, bit, sizeof(*bit), should_block, fdblock);
2155 if (ret) {
eb9bd4e9 2156 bit_free(bit);
db7e0552
JA
2157 if (!events && ret < 0)
2158 events = ret;
cb2a1a62 2159 break;
eb9bd4e9 2160 }
cb2a1a62 2161
017d1660
JA
2162 /*
2163 * look at first trace to check whether we need to convert
2164 * data in the future
2165 */
9e4cd1b8 2166 if (data_is_native == -1 && check_data_endianness(bit->magic))
017d1660
JA
2167 break;
2168
2169 magic = get_magic(bit);
7d747d22
JA
2170 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2171 fprintf(stderr, "Bad magic %x\n", magic);
2172 break;
2173 }
2174
017d1660 2175 pdu_len = get_pdulen(bit);
7d747d22
JA
2176 if (pdu_len) {
2177 void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2178
c0e0dbc2 2179 if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1, fdblock)) {
eb9bd4e9 2180 bit_free(ptr);
7d747d22 2181 break;
eb9bd4e9 2182 }
7d747d22
JA
2183
2184 bit = ptr;
2185 }
2186
d6222db8
JA
2187 trace_to_cpu(bit);
2188
2189 if (verify_trace(bit)) {
2190 bit_free(bit);
2191 continue;
2192 }
2193
bfc70ad5
JA
2194 /*
2195 * not a real trace, so grab and handle it here
2196 */
1a15f6a8 2197 if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
7bd4fd0a 2198 handle_notify(bit);
a2594911 2199 output_binary(bit, sizeof(*bit) + bit->pdu_len);
bfc70ad5
JA
2200 continue;
2201 }
2202
d36421e4 2203 t = t_alloc();
cb2a1a62
JA
2204 memset(t, 0, sizeof(*t));
2205 t->bit = bit;
a43c1c17 2206 t->read_sequence = read_sequence;
cb2a1a62 2207
7d747d22
JA
2208 t->next = trace_list;
2209 trace_list = t;
1f7afa72 2210
f7bd1a9b 2211 if (!pdi || pdi->dev != bit->device)
287fa3d6
JA
2212 pdi = get_dev_info(bit->device);
2213
2214 if (bit->time > pdi->last_read_time)
2215 pdi->last_read_time = bit->time;
2216
7d747d22 2217 events++;
cb2a1a62
JA
2218 }
2219
7d747d22 2220 return events;
cb2a1a62
JA
2221}
2222
70317a16
AB
2223/*
2224 * Managing input streams
2225 */
2226
2227struct ms_stream {
2228 struct ms_stream *next;
2229 struct trace *first, *last;
73877e12 2230 struct per_dev_info *pdi;
70317a16
AB
2231 unsigned int cpu;
2232};
d0ca268b 2233
70317a16 2234#define MS_HASH(d, c) ((MAJOR(d) & 0xff) ^ (MINOR(d) & 0xff) ^ (cpu & 0xff))
73877e12 2235
70317a16
AB
2236struct ms_stream *ms_head;
2237struct ms_stream *ms_hash[256];
87b72777 2238
70317a16
AB
2239static void ms_sort(struct ms_stream *msp);
2240static int ms_prime(struct ms_stream *msp);
2241
2242static inline struct trace *ms_peek(struct ms_stream *msp)
2243{
2244 return (msp == NULL) ? NULL : msp->first;
2245}
d0ca268b 2246
70317a16
AB
2247static inline __u64 ms_peek_time(struct ms_stream *msp)
2248{
2249 return ms_peek(msp)->bit->time;
2250}
d1d7f15f 2251
70317a16
AB
2252static inline void ms_resort(struct ms_stream *msp)
2253{
2254 if (msp->next && ms_peek_time(msp) > ms_peek_time(msp->next)) {
2255 ms_head = msp->next;
2256 msp->next = NULL;
2257 ms_sort(msp);
2258 }
2259}
e7c9f3ff 2260
70317a16
AB
2261static inline void ms_deq(struct ms_stream *msp)
2262{
2263 msp->first = msp->first->next;
2264 if (!msp->first) {
2265 msp->last = NULL;
2266 if (!ms_prime(msp)) {
2267 ms_head = msp->next;
2268 msp->next = NULL;
2269 return;
d0ca268b 2270 }
d5396421
JA
2271 }
2272
70317a16
AB
2273 ms_resort(msp);
2274}
53c68c88 2275
70317a16
AB
2276static void ms_sort(struct ms_stream *msp)
2277{
2278 __u64 msp_t = ms_peek_time(msp);
2279 struct ms_stream *this_msp = ms_head;
d5396421 2280
70317a16
AB
2281 if (this_msp == NULL)
2282 ms_head = msp;
2283 else if (msp_t < ms_peek_time(this_msp)) {
2284 msp->next = this_msp;
2285 ms_head = msp;
2286 }
2287 else {
2288 while (this_msp->next && ms_peek_time(this_msp->next) < msp_t)
2289 this_msp = this_msp->next;
73877e12 2290
70317a16
AB
2291 msp->next = this_msp->next;
2292 this_msp->next = msp;
2293 }
2294}
d5396421 2295
70317a16
AB
2296static int ms_prime(struct ms_stream *msp)
2297{
2298 __u32 magic;
2299 unsigned int i;
2300 struct trace *t;
2301 struct per_dev_info *pdi = msp->pdi;
2302 struct per_cpu_info *pci = get_cpu_info(pdi, msp->cpu);
2303 struct blk_io_trace *bit = NULL;
2304 int ret, pdu_len, ndone = 0;
d5396421 2305
70317a16
AB
2306 for (i = 0; !is_done() && pci->fd >= 0 && i < rb_batch; i++) {
2307 bit = bit_alloc();
2308 ret = read_data(pci->fd, bit, sizeof(*bit), 1, &pci->fdblock);
2309 if (ret)
2310 goto err;
51128a28 2311
70317a16
AB
2312 if (data_is_native == -1 && check_data_endianness(bit->magic))
2313 goto err;
210824c3 2314
70317a16
AB
2315 magic = get_magic(bit);
2316 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2317 fprintf(stderr, "Bad magic %x\n", magic);
2318 goto err;
d5396421 2319
70317a16 2320 }
d5396421 2321
70317a16
AB
2322 pdu_len = get_pdulen(bit);
2323 if (pdu_len) {
2324 void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2325 ret = read_data(pci->fd, ptr + sizeof(*bit), pdu_len,
2326 1, &pci->fdblock);
2327 if (ret) {
2328 free(ptr);
4eb899a6 2329 bit = NULL;
70317a16 2330 goto err;
7d747d22 2331 }
70317a16
AB
2332
2333 bit = ptr;
2ff323b0 2334 }
d5396421 2335
70317a16
AB
2336 trace_to_cpu(bit);
2337 if (verify_trace(bit))
2338 goto err;
53c68c88 2339
1a15f6a8 2340 if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
7bd4fd0a 2341 handle_notify(bit);
70317a16
AB
2342 output_binary(bit, sizeof(*bit) + bit->pdu_len);
2343 bit_free(bit);
287fa3d6 2344
70317a16
AB
2345 i -= 1;
2346 continue;
2347 }
cb2a1a62 2348
70317a16
AB
2349 if (bit->time > pdi->last_read_time)
2350 pdi->last_read_time = bit->time;
d5396421 2351
70317a16
AB
2352 t = t_alloc();
2353 memset(t, 0, sizeof(*t));
2354 t->bit = bit;
2355
2356 if (msp->first == NULL)
2357 msp->first = msp->last = t;
2358 else {
2359 msp->last->next = t;
2360 msp->last = t;
2361 }
2362
2363 ndone++;
2364 }
2365
2366 return ndone;
2367
2368err:
2369 if (bit) bit_free(bit);
2370
2371 cpu_mark_offline(pdi, pci->cpu);
2372 close(pci->fd);
2373 pci->fd = -1;
2374
2375 return ndone;
2376}
2377
2378static struct ms_stream *ms_alloc(struct per_dev_info *pdi, int cpu)
2379{
2380 struct ms_stream *msp = malloc(sizeof(*msp));
2381
2382 msp->next = NULL;
2383 msp->first = msp->last = NULL;
2384 msp->pdi = pdi;
2385 msp->cpu = cpu;
2386
2387 if (ms_prime(msp))
2388 ms_sort(msp);
2389
2390 return msp;
2391}
2392
2393static int setup_file(struct per_dev_info *pdi, int cpu)
2394{
2395 int len = 0;
2396 struct stat st;
2397 char *p, *dname;
2398 struct per_cpu_info *pci = get_cpu_info(pdi, cpu);
2399
2400 pci->cpu = cpu;
2401 pci->fdblock = -1;
2402
2403 p = strdup(pdi->name);
2404 dname = dirname(p);
2405 if (strcmp(dname, ".")) {
2406 input_dir = dname;
2407 p = strdup(pdi->name);
2408 strcpy(pdi->name, basename(p));
2409 }
2410 free(p);
2411
2412 if (input_dir)
2413 len = sprintf(pci->fname, "%s/", input_dir);
2414
2415 snprintf(pci->fname + len, sizeof(pci->fname)-1-len,
2416 "%s.blktrace.%d", pdi->name, pci->cpu);
8afe3d7d 2417 if (stat(pci->fname, &st) < 0)
70317a16 2418 return 0;
8afe3d7d
AB
2419 if (!st.st_size)
2420 return 1;
70317a16
AB
2421
2422 pci->fd = open(pci->fname, O_RDONLY);
2423 if (pci->fd < 0) {
2424 perror(pci->fname);
2425 return 0;
2426 }
2427
2428 printf("Input file %s added\n", pci->fname);
2429 cpu_mark_online(pdi, pci->cpu);
2430
2431 pdi->nfiles++;
2432 ms_alloc(pdi, pci->cpu);
2433
2434 return 1;
2435}
2436
2437static int handle(struct ms_stream *msp)
2438{
2439 struct trace *t;
2440 struct per_dev_info *pdi;
2441 struct per_cpu_info *pci;
2442 struct blk_io_trace *bit;
2443
2444 t = ms_peek(msp);
70317a16
AB
2445
2446 bit = t->bit;
2447 pdi = msp->pdi;
2448 pci = get_cpu_info(pdi, msp->cpu);
2449 pci->nelems++;
8091de93 2450 bit->time -= genesis_time;
7072ee3f
LU
2451
2452 if (t->bit->time > stopwatch_end)
2453 return 0;
2454
8091de93 2455 pdi->last_reported_time = bit->time;
7072ee3f
LU
2456 if ((bit->action & (act_mask << BLK_TC_SHIFT))&&
2457 t->bit->time >= stopwatch_start)
70317a16
AB
2458 dump_trace(bit, pci, pdi);
2459
2460 ms_deq(msp);
2461
2462 if (text_output)
2463 trace_rb_insert_last(pdi, t);
2464 else {
2465 bit_free(t->bit);
2466 t_free(t);
2467 }
2468
2469 return 1;
2470}
2471
7d340756
MZ
2472/*
2473 * Check if we need to sanitize the name. We allow 'foo', or if foo.blktrace.X
2474 * is given, then strip back down to 'foo' to avoid missing files.
2475 */
2476static int name_fixup(char *name)
2477{
2478 char *b;
2479
2480 if (!name)
2481 return 1;
2482
2483 b = strstr(name, ".blktrace.");
2484 if (b)
2485 *b = '\0';
2486
2487 return 0;
2488}
2489
70317a16
AB
2490static int do_file(void)
2491{
7d340756 2492 int i, cpu, ret;
70317a16
AB
2493 struct per_dev_info *pdi;
2494
2495 /*
2496 * first prepare all files for reading
2497 */
2498 for (i = 0; i < ndevices; i++) {
2499 pdi = &devices[i];
7d340756
MZ
2500 ret = name_fixup(pdi->name);
2501 if (ret)
2502 return ret;
2503
70317a16
AB
2504 for (cpu = 0; setup_file(pdi, cpu); cpu++)
2505 ;
2506 }
2507
8091de93
AB
2508 /*
2509 * Get the initial time stamp
2510 */
2511 if (ms_head)
2512 genesis_time = ms_peek_time(ms_head);
2513
70317a16
AB
2514 /*
2515 * Keep processing traces while any are left
2516 */
2517 while (!is_done() && ms_head && handle(ms_head))
2518 ;
a649216c 2519
7d747d22 2520 return 0;
412819ce 2521}
d5396421 2522
67076cbc 2523static void do_pipe(int fd)
412819ce 2524{
53c68c88 2525 unsigned long long youngest;
67076cbc 2526 int events, fdblock;
d5396421 2527
be925321 2528 last_allowed_time = -1ULL;
c0e0dbc2 2529 fdblock = -1;
db7e0552 2530 while ((events = read_events(fd, 0, &fdblock)) > 0) {
4ab42801 2531 read_sequence++;
412819ce 2532
210824c3 2533#if 0
0b07f23e 2534 smallest_seq_read = -1U;
210824c3 2535#endif
0b07f23e 2536
53c68c88
JA
2537 if (sort_entries(&youngest))
2538 break;
2539
2540 if (youngest > stopwatch_end)
2ff323b0
JA
2541 break;
2542
763d936e 2543 show_entries_rb(0);
0b07f23e 2544 }
d5396421 2545
a649216c
JA
2546 if (rb_sort_entries)
2547 show_entries_rb(1);
67076cbc
JA
2548}
2549
2550static int do_fifo(void)
2551{
2552 int fd;
2553
2554 if (!strcmp(pipename, "-"))
2555 fd = dup(STDIN_FILENO);
2556 else
2557 fd = open(pipename, O_RDONLY);
2558
2559 if (fd == -1) {
2560 perror("dup stdin");
2561 return -1;
2562 }
a649216c 2563
67076cbc 2564 do_pipe(fd);
d5396421 2565 close(fd);
d5396421
JA
2566 return 0;
2567}
d0ca268b 2568
cbc927b6 2569static void show_stats(void)
412819ce 2570{
cbc927b6
JA
2571 if (!ofp)
2572 return;
2573 if (stats_printed)
2574 return;
2575
2576 stats_printed = 1;
2577
2578 if (per_process_stats)
2579 show_process_stats();
2580
2581 if (per_device_and_cpu_stats)
2582 show_device_and_cpu_stats();
2583
152f6476 2584 fflush(ofp);
412819ce
JA
2585}
2586
e820abd7 2587static void handle_sigint(__attribute__((__unused__)) int sig)
412819ce
JA
2588{
2589 done = 1;
412819ce
JA
2590}
2591
46e6968b
NS
2592/*
2593 * Extract start and duration times from a string, allowing
2594 * us to specify a time interval of interest within a trace.
2595 * Format: "duration" (start is zero) or "start:duration".
2596 */
2597static int find_stopwatch_interval(char *string)
2598{
2599 double value;
2600 char *sp;
2601
2602 value = strtod(string, &sp);
2603 if (sp == string) {
2604 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
2605 return 1;
2606 }
2607 if (*sp == ':') {
2608 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
2609 string = sp + 1;
2610 value = strtod(string, &sp);
2611 if (sp == string || *sp != '\0') {
2612 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
2613 string);
2614 return 1;
2615 }
2616 } else if (*sp != '\0') {
2617 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
2618 return 1;
2619 }
1b928247
JA
2620 stopwatch_end = DOUBLE_TO_NANO_ULL(value);
2621 if (stopwatch_end <= stopwatch_start) {
2622 fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n",
2623 stopwatch_start, stopwatch_end);
2624 return 1;
2625 }
2626
46e6968b
NS
2627 return 0;
2628}
2629
67076cbc
JA
2630static int is_pipe(const char *str)
2631{
2632 struct stat st;
2633
2634 if (!strcmp(str, "-"))
2635 return 1;
2636 if (!stat(str, &st) && S_ISFIFO(st.st_mode))
2637 return 1;
2638
2639 return 0;
2640}
2641
234db09d
AB
2642#define S_OPTS "a:A:b:D:d:f:F:hi:o:Oqstw:vV"
2643static char usage_str[] = "\n\n" \
2644 "-i <file> | --input=<file>\n" \
2645 "[ -a <action field> | --act-mask=<action field> ]\n" \
2646 "[ -A <action mask> | --set-mask=<action mask> ]\n" \
2647 "[ -b <traces> | --batch=<traces> ]\n" \
2648 "[ -d <file> | --dump-binary=<file> ]\n" \
2649 "[ -D <dir> | --input-directory=<dir> ]\n" \
2650 "[ -f <format> | --format=<format> ]\n" \
2651 "[ -F <spec> | --format-spec=<spec> ]\n" \
2652 "[ -h | --hash-by-name ]\n" \
2653 "[ -o <file> | --output=<file> ]\n" \
2654 "[ -O | --no-text-output ]\n" \
2655 "[ -q | --quiet ]\n" \
2656 "[ -s | --per-program-stats ]\n" \
2657 "[ -t | --track-ios ]\n" \
2658 "[ -w <time> | --stopwatch=<time> ]\n" \
2659 "[ -v | --verbose ]\n" \
2660 "[ -V | --version ]\n\n" \
2661 "\t-b stdin read batching\n" \
2662 "\t-d Output file. If specified, binary data is written to file\n" \
d1d7f15f 2663 "\t-D Directory to prepend to input file names\n" \
234db09d
AB
2664 "\t-f Output format. Customize the output format. The format field\n" \
2665 "\t identifies can be found in the documentation\n" \
2666 "\t-F Format specification. Can be found in the documentation\n" \
2667 "\t-h Hash processes by name, not pid\n" \
2668 "\t-i Input file containing trace data, or '-' for stdin\n" \
52724a0e 2669 "\t-o Output file. If not given, output is stdout\n" \
234db09d
AB
2670 "\t-O Do NOT output text data\n" \
2671 "\t-q Quiet. Don't display any stats at the end of the trace\n" \
52724a0e
JA
2672 "\t-s Show per-program io statistics\n" \
2673 "\t-t Track individual ios. Will tell you the time a request took\n" \
2674 "\t to get queued, to get dispatched, and to get completed\n" \
52724a0e
JA
2675 "\t-w Only parse data between the given time interval in seconds.\n" \
2676 "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \
57ea8602
JA
2677 "\t-v More verbose for marginal errors\n" \
2678 "\t-V Print program version info\n\n";
52724a0e 2679
1f79c4a0
JA
2680static void usage(char *prog)
2681{
52724a0e 2682 fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str);
1f79c4a0
JA
2683}
2684
d5396421
JA
2685int main(int argc, char *argv[])
2686{
98f8386b 2687 int i, c, ret, mode;
98f8386b 2688 int act_mask_tmp = 0;
234db09d 2689 char *ofp_buffer = NULL;
346d8a74 2690 char *bin_ofp_buffer = NULL;
d5396421
JA
2691
2692 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
2693 switch (c) {
98f8386b
AB
2694 case 'a':
2695 i = find_mask_map(optarg);
2696 if (i < 0) {
2697 fprintf(stderr,"Invalid action mask %s\n",
2698 optarg);
2699 return 1;
2700 }
2701 act_mask_tmp |= i;
2702 break;
2703
2704 case 'A':
2705 if ((sscanf(optarg, "%x", &i) != 1) ||
2706 !valid_act_opt(i)) {
2707 fprintf(stderr,
2708 "Invalid set action mask %s/0x%x\n",
2709 optarg, i);
2710 return 1;
2711 }
2712 act_mask_tmp = i;
2713 break;
d5396421 2714 case 'i':
67076cbc 2715 if (is_pipe(optarg) && !pipeline) {
e7c9f3ff 2716 pipeline = 1;
67076cbc
JA
2717 pipename = strdup(optarg);
2718 } else if (resize_devices(optarg) != 0)
e7c9f3ff 2719 return 1;
d5396421 2720 break;
d1d7f15f
JA
2721 case 'D':
2722 input_dir = optarg;
2723 break;
d5396421 2724 case 'o':
66efebf8 2725 output_name = optarg;
d5396421 2726 break;
234db09d
AB
2727 case 'O':
2728 text_output = 0;
2729 break;
79f19470
JA
2730 case 'b':
2731 rb_batch = atoi(optarg);
2732 if (rb_batch <= 0)
2733 rb_batch = RB_BATCH_DEFAULT;
2734 break;
152f6476
JA
2735 case 's':
2736 per_process_stats = 1;
2737 break;
7997c5b0
JA
2738 case 't':
2739 track_ios = 1;
2740 break;
1e1c60f1
NS
2741 case 'q':
2742 per_device_and_cpu_stats = 0;
2743 break;
46e6968b
NS
2744 case 'w':
2745 if (find_stopwatch_interval(optarg) != 0)
2746 return 1;
2747 break;
ab197ca7
AB
2748 case 'f':
2749 set_all_format_specs(optarg);
2750 break;
2751 case 'F':
2752 if (add_format_spec(optarg) != 0)
2753 return 1;
2754 break;
d915dee6 2755 case 'h':
715d8021 2756 ppi_hash_by_pid = 0;
bf0720af 2757 break;
52724a0e 2758 case 'v':
57ea8602
JA
2759 verbose++;
2760 break;
2761 case 'V':
52724a0e
JA
2762 printf("%s version %s\n", argv[0], blkparse_version);
2763 return 0;
a2594911
AB
2764 case 'd':
2765 dump_binary = optarg;
2766 break;
d5396421 2767 default:
1f79c4a0 2768 usage(argv[0]);
d5396421
JA
2769 return 1;
2770 }
d0ca268b
JA
2771 }
2772
e7c9f3ff 2773 while (optind < argc) {
67076cbc 2774 if (is_pipe(argv[optind]) && !pipeline) {
e7c9f3ff 2775 pipeline = 1;
00cd3044 2776 pipename = strdup(argv[optind]);
67076cbc 2777 } else if (resize_devices(argv[optind]) != 0)
e7c9f3ff
NS
2778 return 1;
2779 optind++;
2780 }
2781
2782 if (!pipeline && !ndevices) {
1f79c4a0 2783 usage(argv[0]);
d5396421
JA
2784 return 1;
2785 }
2786
98f8386b
AB
2787 if (act_mask_tmp != 0)
2788 act_mask = act_mask_tmp;
2789
7997c5b0 2790 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
412819ce
JA
2791
2792 signal(SIGINT, handle_sigint);
2793 signal(SIGHUP, handle_sigint);
2794 signal(SIGTERM, handle_sigint);
d5396421 2795
d69db225
JA
2796 setlocale(LC_NUMERIC, "en_US");
2797
234db09d
AB
2798 if (text_output) {
2799 if (!output_name) {
2800 ofp = fdopen(STDOUT_FILENO, "w");
2801 mode = _IOLBF;
2802 } else {
2803 char ofname[128];
152f6476 2804
234db09d
AB
2805 snprintf(ofname, sizeof(ofname) - 1, "%s", output_name);
2806 ofp = fopen(ofname, "w");
2807 mode = _IOFBF;
2808 }
152f6476 2809
234db09d
AB
2810 if (!ofp) {
2811 perror("fopen");
2812 return 1;
2813 }
152f6476 2814
234db09d
AB
2815 ofp_buffer = malloc(4096);
2816 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
2817 perror("setvbuf");
2818 return 1;
2819 }
152f6476
JA
2820 }
2821
a2594911 2822 if (dump_binary) {
346d8a74
AB
2823 dump_fp = fopen(dump_binary, "w");
2824 if (!dump_fp) {
a2594911
AB
2825 perror(dump_binary);
2826 dump_binary = NULL;
2827 return 1;
2828 }
346d8a74
AB
2829 bin_ofp_buffer = malloc(128 * 1024);
2830 if (setvbuf(dump_fp, bin_ofp_buffer, _IOFBF, 128 * 1024)) {
2831 perror("setvbuf binary");
2832 return 1;
2833 }
a2594911
AB
2834 }
2835
e7c9f3ff 2836 if (pipeline)
67076cbc 2837 ret = do_fifo();
d5396421
JA
2838 else
2839 ret = do_file();
2840
fb863d7c
MZ
2841 if (!ret)
2842 show_stats();
2843
8091de93
AB
2844 if (ofp_buffer) {
2845 fflush(ofp);
234db09d 2846 free(ofp_buffer);
8091de93
AB
2847 }
2848 if (bin_ofp_buffer) {
2849 fflush(dump_fp);
346d8a74 2850 free(bin_ofp_buffer);
8091de93 2851 }
d5396421 2852 return ret;
d0ca268b 2853}