Commit | Line | Data |
---|---|---|
d956a2cd JA |
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 | */ | |
d0ca268b JA |
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> | |
8fc0abbc | 27 | #include <string.h> |
d5396421 | 28 | #include <getopt.h> |
412819ce JA |
29 | #include <errno.h> |
30 | #include <signal.h> | |
d69db225 | 31 | #include <locale.h> |
46e6968b | 32 | #include <limits.h> |
ab197ca7 | 33 | #include <ctype.h> |
d0ca268b | 34 | |
8fc0abbc JA |
35 | #include "blktrace.h" |
36 | #include "rbtree.h" | |
d0ca268b | 37 | |
52724a0e JA |
38 | static char blkparse_version[] = "0.90"; |
39 | ||
2e3e8ded JA |
40 | #define SECONDS(x) ((unsigned long long)(x) / 1000000000) |
41 | #define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000) | |
46e6968b | 42 | #define DOUBLE_TO_NANO_ULL(d) ((unsigned long long)((d) * 1000000000)) |
cfab07eb | 43 | |
e7c9f3ff NS |
44 | #define MINORBITS 20 |
45 | #define MINORMASK ((1U << MINORBITS) - 1) | |
46 | #define MAJOR(dev) ((unsigned int) ((dev) >> MINORBITS)) | |
47 | #define MINOR(dev) ((unsigned int) ((dev) & MINORMASK)) | |
48 | ||
49 | #define min(a, b) ((a) < (b) ? (a) : (b)) | |
d5396421 | 50 | |
ab197ca7 AB |
51 | #define HEADER "%D %2c %8s %5T.%9t %5p %2a %3d " |
52 | ||
152f6476 JA |
53 | struct io_stats { |
54 | unsigned long qreads, qwrites, creads, cwrites, mreads, mwrites; | |
55 | unsigned long ireads, iwrites; | |
56 | unsigned long long qread_kb, qwrite_kb, cread_kb, cwrite_kb; | |
57 | unsigned long long iread_kb, iwrite_kb; | |
06639b27 | 58 | unsigned long io_unplugs, timer_unplugs; |
152f6476 JA |
59 | }; |
60 | ||
d5396421 | 61 | struct per_cpu_info { |
d0ca268b JA |
62 | int cpu; |
63 | int nelems; | |
d0ca268b JA |
64 | |
65 | int fd; | |
87b72777 | 66 | char fname[128]; |
d0ca268b | 67 | |
152f6476 JA |
68 | struct io_stats io_stats; |
69 | }; | |
8fc0abbc | 70 | |
e7c9f3ff NS |
71 | struct per_dev_info { |
72 | dev_t id; | |
73 | char *name; | |
74 | ||
75 | int backwards; | |
76 | unsigned long long events; | |
77 | unsigned long long last_reported_time; | |
78 | struct io_stats io_stats; | |
cb2a1a62 | 79 | unsigned long last_sequence; |
0aa8a62d | 80 | unsigned long skips; |
e7c9f3ff NS |
81 | |
82 | int ncpus; | |
83 | struct per_cpu_info *cpus; | |
84 | }; | |
85 | ||
152f6476 JA |
86 | struct per_process_info { |
87 | char name[16]; | |
88 | __u32 pid; | |
89 | struct io_stats io_stats; | |
90 | struct per_process_info *hash_next, *list_next; | |
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 JA |
100 | #define PPI_HASH_SHIFT (8) |
101 | static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT]; | |
102 | static struct per_process_info *ppi_list; | |
886ecf0e | 103 | static int ppi_list_entries; |
152f6476 | 104 | |
52724a0e | 105 | #define S_OPTS "i:o:b:stqw:f:F:v" |
d5396421 JA |
106 | static struct option l_opts[] = { |
107 | { | |
108 | .name = "input", | |
428683db | 109 | .has_arg = required_argument, |
d5396421 JA |
110 | .flag = NULL, |
111 | .val = 'i' | |
112 | }, | |
113 | { | |
114 | .name = "output", | |
428683db | 115 | .has_arg = required_argument, |
d5396421 JA |
116 | .flag = NULL, |
117 | .val = 'o' | |
118 | }, | |
79f19470 JA |
119 | { |
120 | .name = "batch", | |
428683db | 121 | .has_arg = required_argument, |
79f19470 JA |
122 | .flag = NULL, |
123 | .val = 'b' | |
124 | }, | |
152f6476 JA |
125 | { |
126 | .name = "per program stats", | |
428683db | 127 | .has_arg = no_argument, |
152f6476 JA |
128 | .flag = NULL, |
129 | .val = 's' | |
130 | }, | |
7997c5b0 JA |
131 | { |
132 | .name = "track ios", | |
428683db | 133 | .has_arg = no_argument, |
7997c5b0 JA |
134 | .flag = NULL, |
135 | .val = 't' | |
136 | }, | |
1e1c60f1 NS |
137 | { |
138 | .name = "quiet", | |
428683db | 139 | .has_arg = no_argument, |
1e1c60f1 NS |
140 | .flag = NULL, |
141 | .val = 'q' | |
142 | }, | |
46e6968b NS |
143 | { |
144 | .name = "stopwatch", | |
428683db | 145 | .has_arg = required_argument, |
46e6968b NS |
146 | .flag = NULL, |
147 | .val = 'w' | |
148 | }, | |
ab197ca7 AB |
149 | { |
150 | .name = "format", | |
428683db | 151 | .has_arg = required_argument, |
ab197ca7 AB |
152 | .flag = NULL, |
153 | .val = 'f' | |
154 | }, | |
155 | { | |
156 | .name = "format-spec", | |
428683db | 157 | .has_arg = required_argument, |
ab197ca7 AB |
158 | .flag = NULL, |
159 | .val = 'F' | |
160 | }, | |
52724a0e JA |
161 | { |
162 | .name = "version", | |
163 | .has_arg = no_argument, | |
164 | .flag = NULL, | |
165 | .val = 'v' | |
166 | }, | |
d5396421 JA |
167 | }; |
168 | ||
7997c5b0 JA |
169 | /* |
170 | * for sorting the displayed output | |
171 | */ | |
8fc0abbc JA |
172 | struct trace { |
173 | struct blk_io_trace *bit; | |
174 | struct rb_node rb_node; | |
cb2a1a62 JA |
175 | struct trace *next; |
176 | int skipped; | |
8fc0abbc JA |
177 | }; |
178 | ||
cb2a1a62 JA |
179 | static struct rb_root rb_sort_root; |
180 | static struct rb_root rb_track_root; | |
181 | ||
182 | static struct trace *trace_list; | |
183 | ||
7997c5b0 JA |
184 | /* |
185 | * for tracking individual ios | |
186 | */ | |
187 | struct io_track { | |
188 | struct rb_node rb_node; | |
189 | ||
e7c9f3ff | 190 | dev_t device; |
7997c5b0 JA |
191 | __u64 sector; |
192 | __u32 pid; | |
95c15013 | 193 | unsigned long long allocation_time; |
7997c5b0 JA |
194 | unsigned long long queue_time; |
195 | unsigned long long dispatch_time; | |
196 | unsigned long long completion_time; | |
197 | }; | |
198 | ||
e7c9f3ff NS |
199 | static int ndevices; |
200 | static struct per_dev_info *devices; | |
201 | static char *get_dev_name(struct per_dev_info *, char *, int); | |
d0ca268b | 202 | |
152f6476 | 203 | static FILE *ofp; |
e7c9f3ff NS |
204 | static char *output_name; |
205 | ||
206 | static unsigned long long genesis_time; | |
46e6968b NS |
207 | static unsigned long long stopwatch_start; /* start from zero by default */ |
208 | static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */ | |
152f6476 JA |
209 | |
210 | static int per_process_stats; | |
7997c5b0 | 211 | static int track_ios; |
d0ca268b | 212 | |
7d747d22 | 213 | #define RB_BATCH_DEFAULT (512) |
79f19470 JA |
214 | static int rb_batch = RB_BATCH_DEFAULT; |
215 | ||
e7c9f3ff NS |
216 | static int pipeline; |
217 | ||
412819ce JA |
218 | #define is_done() (*(volatile int *)(&done)) |
219 | static volatile int done; | |
220 | ||
152f6476 JA |
221 | static inline unsigned long hash_long(unsigned long val) |
222 | { | |
16ef714e JA |
223 | #if __WORDSIZE == 32 |
224 | val *= 0x9e370001UL; | |
225 | #elif __WORDSIZE == 64 | |
226 | val *= 0x9e37fffffffc0001UL; | |
227 | #else | |
228 | #error unknown word size | |
229 | #endif | |
230 | ||
231 | return val >> (__WORDSIZE - PPI_HASH_SHIFT); | |
152f6476 JA |
232 | } |
233 | ||
234 | static inline void add_process_to_hash(struct per_process_info *ppi) | |
235 | { | |
236 | const int hash_idx = hash_long(ppi->pid); | |
237 | ||
238 | ppi->hash_next = ppi_hash[hash_idx]; | |
239 | ppi_hash[hash_idx] = ppi; | |
240 | } | |
241 | ||
242 | static inline void add_process_to_list(struct per_process_info *ppi) | |
243 | { | |
244 | ppi->list_next = ppi_list; | |
245 | ppi_list = ppi; | |
886ecf0e | 246 | ppi_list_entries++; |
152f6476 JA |
247 | } |
248 | ||
249 | static struct per_process_info *find_process_by_pid(__u32 pid) | |
250 | { | |
251 | const int hash_idx = hash_long(pid); | |
252 | struct per_process_info *ppi; | |
253 | ||
254 | ppi = ppi_hash[hash_idx]; | |
255 | while (ppi) { | |
256 | if (ppi->pid == pid) | |
257 | return ppi; | |
258 | ||
259 | ppi = ppi->hash_next; | |
260 | } | |
261 | ||
262 | return NULL; | |
263 | } | |
264 | ||
7997c5b0 JA |
265 | static inline int trace_rb_insert(struct trace *t) |
266 | { | |
267 | struct rb_node **p = &rb_sort_root.rb_node; | |
268 | struct rb_node *parent = NULL; | |
269 | struct trace *__t; | |
270 | ||
e7c9f3ff NS |
271 | if (genesis_time == 0 || t->bit->time < genesis_time) |
272 | genesis_time = t->bit->time; | |
273 | ||
7997c5b0 JA |
274 | while (*p) { |
275 | parent = *p; | |
276 | __t = rb_entry(parent, struct trace, rb_node); | |
277 | ||
e7c9f3ff NS |
278 | if (t->bit->time < __t->bit->time) |
279 | p = &(*p)->rb_left; | |
280 | else if (t->bit->time > __t->bit->time) | |
281 | p = &(*p)->rb_right; | |
282 | else if (t->bit->device < __t->bit->device) | |
283 | p = &(*p)->rb_left; | |
284 | else if (t->bit->device > __t->bit->device) | |
285 | p = &(*p)->rb_right; | |
286 | else if (t->bit->sequence < __t->bit->sequence) | |
7997c5b0 JA |
287 | p = &(*p)->rb_left; |
288 | else if (t->bit->sequence > __t->bit->sequence) | |
289 | p = &(*p)->rb_right; | |
e7c9f3ff NS |
290 | else if (t->bit->device == __t->bit->device) { |
291 | fprintf(stderr, | |
292 | "sequence alias (%d) on device %d,%d!\n", | |
293 | t->bit->sequence, | |
294 | MAJOR(t->bit->device), MINOR(t->bit->device)); | |
7997c5b0 JA |
295 | return 1; |
296 | } | |
297 | } | |
298 | ||
299 | rb_link_node(&t->rb_node, parent, p); | |
300 | rb_insert_color(&t->rb_node, &rb_sort_root); | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static inline int track_rb_insert(struct io_track *iot) | |
305 | { | |
306 | struct rb_node **p = &rb_track_root.rb_node; | |
307 | struct rb_node *parent = NULL; | |
308 | struct io_track *__iot; | |
309 | ||
310 | while (*p) { | |
311 | parent = *p; | |
e7c9f3ff | 312 | |
7997c5b0 JA |
313 | __iot = rb_entry(parent, struct io_track, rb_node); |
314 | ||
e7c9f3ff NS |
315 | if (iot->device < __iot->device) |
316 | p = &(*p)->rb_left; | |
317 | else if (iot->device > __iot->device) | |
318 | p = &(*p)->rb_right; | |
319 | else if (iot->sector < __iot->sector) | |
7997c5b0 JA |
320 | p = &(*p)->rb_left; |
321 | else if (iot->sector > __iot->sector) | |
322 | p = &(*p)->rb_right; | |
323 | else { | |
e7c9f3ff | 324 | fprintf(stderr, |
ab197ca7 AB |
325 | "sector alias (%Lu) on device %d,%d!\n", |
326 | (unsigned long long) iot->sector, | |
e7c9f3ff | 327 | MAJOR(iot->device), MINOR(iot->device)); |
7997c5b0 JA |
328 | return 1; |
329 | } | |
330 | } | |
331 | ||
332 | rb_link_node(&iot->rb_node, parent, p); | |
333 | rb_insert_color(&iot->rb_node, &rb_track_root); | |
334 | return 0; | |
335 | } | |
336 | ||
e7c9f3ff | 337 | static struct io_track *__find_track(dev_t device, __u64 sector) |
7997c5b0 JA |
338 | { |
339 | struct rb_node **p = &rb_track_root.rb_node; | |
340 | struct rb_node *parent = NULL; | |
341 | struct io_track *__iot; | |
342 | ||
343 | while (*p) { | |
344 | parent = *p; | |
345 | ||
346 | __iot = rb_entry(parent, struct io_track, rb_node); | |
347 | ||
e7c9f3ff NS |
348 | if (device < __iot->device) |
349 | p = &(*p)->rb_left; | |
350 | else if (device > __iot->device) | |
351 | p = &(*p)->rb_right; | |
352 | else if (sector < __iot->sector) | |
7997c5b0 JA |
353 | p = &(*p)->rb_left; |
354 | else if (sector > __iot->sector) | |
355 | p = &(*p)->rb_right; | |
356 | else | |
357 | return __iot; | |
358 | } | |
359 | ||
360 | return NULL; | |
361 | } | |
362 | ||
e7c9f3ff | 363 | static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector) |
7997c5b0 | 364 | { |
916b5501 | 365 | struct io_track *iot; |
7997c5b0 | 366 | |
e7c9f3ff | 367 | iot = __find_track(device, sector); |
7997c5b0 JA |
368 | if (!iot) { |
369 | iot = malloc(sizeof(*iot)); | |
50adc0ba | 370 | iot->pid = pid; |
e7c9f3ff | 371 | iot->device = device; |
7997c5b0 JA |
372 | iot->sector = sector; |
373 | track_rb_insert(iot); | |
374 | } | |
375 | ||
376 | return iot; | |
377 | } | |
378 | ||
a01516de | 379 | static void log_track_frontmerge(struct blk_io_trace *t) |
2e3e8ded JA |
380 | { |
381 | struct io_track *iot; | |
382 | ||
383 | if (!track_ios) | |
384 | return; | |
2e3e8ded | 385 | |
a01516de | 386 | iot = __find_track(t->device, t->sector + (t->bytes >> 9)); |
cb2a1a62 JA |
387 | if (!iot) { |
388 | fprintf(stderr, "failed to find mergeable event\n"); | |
389 | return; | |
2e3e8ded | 390 | } |
cb2a1a62 JA |
391 | |
392 | rb_erase(&iot->rb_node, &rb_track_root); | |
a01516de | 393 | iot->sector -= t->bytes >> 9; |
cb2a1a62 | 394 | track_rb_insert(iot); |
2e3e8ded JA |
395 | } |
396 | ||
95c15013 | 397 | static void log_track_getrq(struct blk_io_trace *t) |
2e3e8ded JA |
398 | { |
399 | struct io_track *iot; | |
400 | ||
401 | if (!track_ios) | |
402 | return; | |
403 | ||
e7c9f3ff | 404 | iot = find_track(t->pid, t->device, t->sector); |
95c15013 JA |
405 | iot->allocation_time = t->time; |
406 | } | |
407 | ||
408 | ||
409 | /* | |
b6076a9b | 410 | * return time between rq allocation and insertion |
95c15013 | 411 | */ |
b6076a9b | 412 | static unsigned long long log_track_insert(struct blk_io_trace *t) |
95c15013 | 413 | { |
50adc0ba | 414 | unsigned long long elapsed; |
95c15013 JA |
415 | struct io_track *iot; |
416 | ||
417 | if (!track_ios) | |
418 | return -1; | |
419 | ||
e7c9f3ff | 420 | iot = find_track(t->pid, t->device, t->sector); |
2e3e8ded | 421 | iot->queue_time = t->time; |
50adc0ba JA |
422 | elapsed = iot->queue_time - iot->allocation_time; |
423 | ||
424 | if (per_process_stats) { | |
425 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 426 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 427 | |
b9d40d6f JA |
428 | if (ppi && elapsed > ppi->longest_allocation_wait[w]) |
429 | ppi->longest_allocation_wait[w] = elapsed; | |
50adc0ba JA |
430 | } |
431 | ||
432 | return elapsed; | |
2e3e8ded JA |
433 | } |
434 | ||
435 | /* | |
436 | * return time between queue and issue | |
437 | */ | |
438 | static unsigned long long log_track_issue(struct blk_io_trace *t) | |
439 | { | |
50adc0ba | 440 | unsigned long long elapsed; |
2e3e8ded JA |
441 | struct io_track *iot; |
442 | ||
443 | if (!track_ios) | |
444 | return -1; | |
445 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
446 | return -1; | |
447 | ||
e7c9f3ff | 448 | iot = __find_track(t->device, t->sector); |
cb2a1a62 JA |
449 | if (!iot) { |
450 | fprintf(stderr, "failed to find issue event\n"); | |
2e3e8ded | 451 | return -1; |
cb2a1a62 | 452 | } |
2e3e8ded JA |
453 | |
454 | iot->dispatch_time = t->time; | |
50adc0ba JA |
455 | elapsed = iot->dispatch_time - iot->queue_time; |
456 | ||
457 | if (per_process_stats) { | |
458 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 459 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 460 | |
b9d40d6f JA |
461 | if (ppi && elapsed > ppi->longest_dispatch_wait[w]) |
462 | ppi->longest_dispatch_wait[w] = elapsed; | |
50adc0ba JA |
463 | } |
464 | ||
465 | return elapsed; | |
2e3e8ded JA |
466 | } |
467 | ||
468 | /* | |
469 | * return time between dispatch and complete | |
470 | */ | |
471 | static unsigned long long log_track_complete(struct blk_io_trace *t) | |
472 | { | |
473 | unsigned long long elapsed; | |
474 | struct io_track *iot; | |
475 | ||
476 | if (!track_ios) | |
477 | return -1; | |
478 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
479 | return -1; | |
480 | ||
e7c9f3ff | 481 | iot = __find_track(t->device, t->sector); |
cb2a1a62 JA |
482 | if (!iot) { |
483 | fprintf(stderr, "failed to find complete event\n"); | |
2e3e8ded | 484 | return -1; |
cb2a1a62 | 485 | } |
2e3e8ded JA |
486 | |
487 | iot->completion_time = t->time; | |
488 | elapsed = iot->completion_time - iot->dispatch_time; | |
489 | ||
50adc0ba JA |
490 | if (per_process_stats) { |
491 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 492 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 493 | |
b9d40d6f JA |
494 | if (ppi && elapsed > ppi->longest_completion_wait[w]) |
495 | ppi->longest_completion_wait[w] = elapsed; | |
50adc0ba JA |
496 | } |
497 | ||
2e3e8ded JA |
498 | /* |
499 | * kill the trace, we don't need it after completion | |
500 | */ | |
501 | rb_erase(&iot->rb_node, &rb_track_root); | |
502 | free(iot); | |
503 | ||
504 | return elapsed; | |
505 | } | |
506 | ||
507 | ||
152f6476 JA |
508 | static struct io_stats *find_process_io_stats(__u32 pid, char *name) |
509 | { | |
510 | struct per_process_info *ppi = find_process_by_pid(pid); | |
511 | ||
512 | if (!ppi) { | |
513 | ppi = malloc(sizeof(*ppi)); | |
514 | memset(ppi, 0, sizeof(*ppi)); | |
515 | strncpy(ppi->name, name, sizeof(ppi->name)); | |
516 | ppi->pid = pid; | |
517 | add_process_to_hash(ppi); | |
518 | add_process_to_list(ppi); | |
519 | } | |
520 | ||
521 | return &ppi->io_stats; | |
522 | } | |
523 | ||
e7c9f3ff NS |
524 | |
525 | static void resize_cpu_info(struct per_dev_info *pdi, int cpu) | |
a718bd37 | 526 | { |
e7c9f3ff NS |
527 | struct per_cpu_info *cpus = pdi->cpus; |
528 | int ncpus = pdi->ncpus; | |
529 | int new_count = cpu + 1; | |
530 | int new_space, size; | |
a718bd37 NS |
531 | char *new_start; |
532 | ||
e7c9f3ff NS |
533 | size = new_count * sizeof(struct per_cpu_info); |
534 | cpus = realloc(cpus, size); | |
535 | if (!cpus) { | |
536 | char name[20]; | |
537 | fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n", | |
538 | get_dev_name(pdi, name, sizeof(name)), size); | |
a718bd37 NS |
539 | exit(1); |
540 | } | |
541 | ||
e7c9f3ff NS |
542 | new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info)); |
543 | new_space = (new_count - ncpus) * sizeof(struct per_cpu_info); | |
a718bd37 | 544 | memset(new_start, 0, new_space); |
e7c9f3ff NS |
545 | |
546 | pdi->ncpus = new_count; | |
547 | pdi->cpus = cpus; | |
548 | } | |
cb2a1a62 | 549 | |
e7c9f3ff NS |
550 | static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu) |
551 | { | |
cb2a1a62 JA |
552 | struct per_cpu_info *pci; |
553 | ||
e7c9f3ff NS |
554 | if (cpu >= pdi->ncpus) |
555 | resize_cpu_info(pdi, cpu); | |
cb2a1a62 JA |
556 | |
557 | pci = &pdi->cpus[cpu]; | |
558 | pci->cpu = cpu; | |
559 | return pci; | |
a718bd37 NS |
560 | } |
561 | ||
e7c9f3ff NS |
562 | |
563 | static int resize_devices(char *name) | |
a718bd37 | 564 | { |
e7c9f3ff | 565 | int size = (ndevices + 1) * sizeof(struct per_dev_info); |
c499bf38 | 566 | |
e7c9f3ff NS |
567 | devices = realloc(devices, size); |
568 | if (!devices) { | |
569 | fprintf(stderr, "Out of memory, device %s (%d)\n", name, size); | |
570 | return 1; | |
571 | } | |
572 | memset(&devices[ndevices], 0, sizeof(struct per_dev_info)); | |
573 | devices[ndevices].name = name; | |
574 | ndevices++; | |
575 | return 0; | |
576 | } | |
a718bd37 | 577 | |
cb2a1a62 | 578 | static struct per_dev_info *get_dev_info(dev_t id) |
e7c9f3ff | 579 | { |
cb2a1a62 | 580 | struct per_dev_info *pdi; |
e7c9f3ff | 581 | int i; |
c499bf38 | 582 | |
7d747d22 JA |
583 | for (i = 0; i < ndevices; i++) { |
584 | if (!devices[i].id) | |
585 | devices[i].id = id; | |
e7c9f3ff NS |
586 | if (devices[i].id == id) |
587 | return &devices[i]; | |
7d747d22 | 588 | } |
cb2a1a62 | 589 | |
e7c9f3ff NS |
590 | if (resize_devices(NULL) != 0) |
591 | return NULL; | |
cb2a1a62 JA |
592 | |
593 | pdi = &devices[ndevices - 1]; | |
594 | pdi->id = id; | |
1b928247 | 595 | pdi->last_sequence = -1; |
cb2a1a62 | 596 | return pdi; |
a718bd37 NS |
597 | } |
598 | ||
e7c9f3ff NS |
599 | static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size) |
600 | { | |
601 | if (pdi->name) | |
602 | snprintf(buffer, size, "%s", pdi->name); | |
603 | else | |
604 | snprintf(buffer, size, "%d,%d", MAJOR(pdi->id), MINOR(pdi->id)); | |
605 | return buffer; | |
606 | } | |
607 | ||
e7c9f3ff | 608 | static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit) |
cfab07eb AB |
609 | { |
610 | unsigned long long this = bit->time; | |
e7c9f3ff | 611 | unsigned long long last = pdi->last_reported_time; |
cfab07eb | 612 | |
e7c9f3ff NS |
613 | pdi->backwards = (this < last) ? 'B' : ' '; |
614 | pdi->last_reported_time = this; | |
cfab07eb AB |
615 | } |
616 | ||
152f6476 JA |
617 | static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t, |
618 | int rw) | |
d0ca268b JA |
619 | { |
620 | if (rw) { | |
152f6476 JA |
621 | ios->mwrites++; |
622 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 623 | } else { |
152f6476 JA |
624 | ios->mreads++; |
625 | ios->qread_kb += t->bytes >> 10; | |
626 | } | |
627 | } | |
628 | ||
629 | static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci, | |
630 | int rw) | |
631 | { | |
632 | __account_m(&pci->io_stats, t, rw); | |
633 | ||
634 | if (per_process_stats) { | |
635 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
636 | ||
637 | __account_m(ios, t, rw); | |
d0ca268b JA |
638 | } |
639 | } | |
640 | ||
b6076a9b JA |
641 | static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t, |
642 | int rw) | |
d0ca268b JA |
643 | { |
644 | if (rw) { | |
152f6476 JA |
645 | ios->qwrites++; |
646 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 647 | } else { |
152f6476 JA |
648 | ios->qreads++; |
649 | ios->qread_kb += t->bytes >> 10; | |
650 | } | |
651 | } | |
652 | ||
b6076a9b JA |
653 | static inline void account_queue(struct blk_io_trace *t, |
654 | struct per_cpu_info *pci, int rw) | |
152f6476 | 655 | { |
b6076a9b | 656 | __account_queue(&pci->io_stats, t, rw); |
152f6476 JA |
657 | |
658 | if (per_process_stats) { | |
659 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
660 | ||
b6076a9b | 661 | __account_queue(ios, t, rw); |
d0ca268b JA |
662 | } |
663 | } | |
664 | ||
152f6476 | 665 | static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes) |
d0ca268b JA |
666 | { |
667 | if (rw) { | |
152f6476 JA |
668 | ios->cwrites++; |
669 | ios->cwrite_kb += bytes >> 10; | |
d0ca268b | 670 | } else { |
152f6476 JA |
671 | ios->creads++; |
672 | ios->cread_kb += bytes >> 10; | |
673 | } | |
674 | } | |
675 | ||
676 | static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci, | |
677 | int rw, int bytes) | |
678 | { | |
679 | __account_c(&pci->io_stats, rw, bytes); | |
680 | ||
681 | if (per_process_stats) { | |
682 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
683 | ||
684 | __account_c(ios, rw, bytes); | |
d0ca268b JA |
685 | } |
686 | } | |
687 | ||
b6076a9b JA |
688 | static inline void __account_issue(struct io_stats *ios, int rw, |
689 | unsigned int bytes) | |
afd2d7ad | 690 | { |
691 | if (rw) { | |
152f6476 JA |
692 | ios->iwrites++; |
693 | ios->iwrite_kb += bytes >> 10; | |
afd2d7ad | 694 | } else { |
152f6476 JA |
695 | ios->ireads++; |
696 | ios->iread_kb += bytes >> 10; | |
afd2d7ad | 697 | } |
698 | } | |
699 | ||
b6076a9b JA |
700 | static inline void account_issue(struct blk_io_trace *t, |
701 | struct per_cpu_info *pci, int rw) | |
d0ca268b | 702 | { |
b6076a9b | 703 | __account_issue(&pci->io_stats, rw, t->bytes); |
152f6476 JA |
704 | |
705 | if (per_process_stats) { | |
706 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
d5396421 | 707 | |
b6076a9b | 708 | __account_issue(ios, rw, t->bytes); |
152f6476 JA |
709 | } |
710 | } | |
711 | ||
06639b27 JA |
712 | static inline void __account_unplug(struct io_stats *ios, int timer) |
713 | { | |
714 | if (timer) | |
715 | ios->timer_unplugs++; | |
716 | else | |
717 | ios->io_unplugs++; | |
718 | } | |
719 | ||
720 | static inline void account_unplug(struct blk_io_trace *t, | |
721 | struct per_cpu_info *pci, int timer) | |
722 | { | |
723 | __account_unplug(&pci->io_stats, timer); | |
724 | ||
725 | if (per_process_stats) { | |
726 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
727 | ||
728 | __account_unplug(ios, timer); | |
729 | } | |
730 | } | |
731 | ||
ab197ca7 AB |
732 | #define VALID_SPECS "BCDFGMPQRSTU" |
733 | static char *override_format[256]; | |
734 | static inline int valid_spec(int spec) | |
152f6476 | 735 | { |
ab197ca7 | 736 | return strchr(VALID_SPECS, spec) != NULL; |
d0ca268b JA |
737 | } |
738 | ||
ab197ca7 | 739 | static void set_all_format_specs(char *optarg) |
d0ca268b | 740 | { |
ab197ca7 | 741 | char *p; |
d0ca268b | 742 | |
ab197ca7 AB |
743 | for (p = VALID_SPECS; *p; p++) |
744 | if (override_format[(int)(*p)] == NULL) | |
745 | override_format[(int)(*p)] = strdup(optarg); | |
746 | } | |
d0ca268b | 747 | |
ab197ca7 AB |
748 | static int add_format_spec(char *optarg) |
749 | { | |
750 | int spec = optarg[0]; | |
d0ca268b | 751 | |
ab197ca7 AB |
752 | if (!valid_spec(spec)) { |
753 | fprintf(stderr,"Bad format specifier %c\n", spec); | |
754 | return 1; | |
755 | } | |
756 | if (optarg[1] != ',') { | |
757 | fprintf(stderr,"Bad format specifier - need ',' %s\n", optarg); | |
758 | return 1; | |
759 | } | |
760 | optarg += 2; | |
761 | if (*optarg == '\0') { | |
762 | fprintf(stderr,"Bad format specifier - need fmt %s\n", optarg); | |
763 | return 1; | |
764 | } | |
765 | ||
766 | /* | |
767 | * Set both merges (front and back) | |
768 | */ | |
769 | if (spec == 'M') { | |
770 | override_format['B'] = strdup(optarg); | |
771 | override_format['M'] = strdup(optarg); | |
772 | } else | |
773 | override_format[spec] = strdup(optarg); | |
d0ca268b | 774 | |
ab197ca7 | 775 | return 0; |
d0ca268b JA |
776 | } |
777 | ||
ab197ca7 AB |
778 | static void print_field(char *act, struct per_cpu_info *pci, |
779 | struct blk_io_trace *t, unsigned long long elapsed, | |
780 | int pdu_len, unsigned char *pdu_buf, char field, | |
781 | int minus, int has_w, int width) | |
d0ca268b | 782 | { |
ab197ca7 | 783 | char format[64]; |
2e3e8ded | 784 | |
ab197ca7 AB |
785 | if (has_w) { |
786 | if (minus) | |
787 | sprintf(format, "%%-%d", width); | |
788 | else | |
789 | sprintf(format, "%%%d", width); | |
790 | } else | |
791 | sprintf(format, "%%"); | |
792 | ||
793 | switch (field) { | |
794 | case 'a': | |
795 | fprintf(ofp, strcat(format, "s"), act); | |
796 | break; | |
797 | case 'c': | |
798 | fprintf(ofp, strcat(format, "d"), pci->cpu); | |
799 | break; | |
800 | case 'C': | |
801 | fprintf(ofp, strcat(format, "s"), t->comm); | |
802 | break; | |
803 | case 'd': { | |
804 | char rwbs[4]; | |
805 | int i = 0; | |
806 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
807 | int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER); | |
808 | int s = t->action & BLK_TC_ACT(BLK_TC_SYNC); | |
809 | if (w) | |
810 | rwbs[i++] = 'W'; | |
811 | else | |
812 | rwbs[i++] = 'R'; | |
813 | if (b) | |
814 | rwbs[i++] = 'B'; | |
815 | if (s) | |
816 | rwbs[i++] = 'S'; | |
817 | rwbs[i] = '\0'; | |
818 | fprintf(ofp, strcat(format, "s"), rwbs); | |
819 | break; | |
820 | } | |
821 | case 'D': /* format width ignored */ | |
822 | fprintf(ofp,"%3d,%-3d", MAJOR(t->device), MINOR(t->device)); | |
823 | break; | |
824 | case 'e': | |
825 | fprintf(ofp, strcat(format, "d"), t->error); | |
826 | break; | |
827 | case 'M': | |
828 | fprintf(ofp, strcat(format, "d"), MAJOR(t->device)); | |
829 | break; | |
830 | case 'm': | |
831 | fprintf(ofp, strcat(format, "d"), MINOR(t->device)); | |
832 | break; | |
833 | case 'n': | |
834 | fprintf(ofp, strcat(format, "u"), t->bytes >> 9); | |
835 | break; | |
836 | case 'p': | |
837 | fprintf(ofp, strcat(format, "u"), t->pid); | |
838 | break; | |
839 | case 'P': /* format width ignored */ | |
840 | if ((pdu_len > 0) && (pdu_buf != NULL)) { | |
841 | int i; | |
842 | unsigned char *p = pdu_buf; | |
182f2fcc JA |
843 | for (i = 0; i < pdu_len; i++) { |
844 | if (i) | |
845 | fprintf(ofp, " "); | |
846 | ||
847 | fprintf(ofp, "%02x", *p++); | |
848 | } | |
ab197ca7 AB |
849 | } |
850 | break; | |
851 | case 's': | |
852 | fprintf(ofp, strcat(format, "ld"), t->sequence); | |
853 | break; | |
854 | case 'S': | |
855 | fprintf(ofp, strcat(format, "lu"), t->sector); | |
856 | break; | |
857 | case 't': | |
858 | sprintf(format, "%%0%dlu", has_w ? width : 9); | |
859 | fprintf(ofp, format, NANO_SECONDS(t->time)); | |
860 | break; | |
861 | case 'T': | |
862 | fprintf(ofp, strcat(format, "d"), SECONDS(t->time)); | |
863 | break; | |
864 | case 'u': | |
865 | if (elapsed == -1ULL) { | |
866 | fprintf(stderr, "Expecting elapsed value\n"); | |
867 | exit(1); | |
868 | } | |
869 | fprintf(ofp, strcat(format, "llu"), elapsed / 1000); | |
870 | break; | |
871 | case 'U': { | |
872 | __u64 *depth = (__u64 *) ((char *) t + sizeof(*t)); | |
873 | fprintf(ofp, strcat(format, "u"), | |
874 | (unsigned int) be64_to_cpu(*depth)); | |
875 | break; | |
876 | } | |
877 | default: | |
878 | fprintf(ofp,strcat(format, "c"), field); | |
879 | break; | |
880 | } | |
881 | } | |
2e3e8ded | 882 | |
ab197ca7 AB |
883 | static char *parse_field(char *act, struct per_cpu_info *pci, |
884 | struct blk_io_trace *t, unsigned long long elapsed, | |
885 | int pdu_len, unsigned char *pdu_buf, | |
886 | char *master_format) | |
887 | { | |
888 | int minus = 0; | |
889 | int has_w = 0; | |
890 | int width = 0; | |
891 | char *p = master_format; | |
892 | ||
893 | if (*p == '-') { | |
894 | minus = 1; | |
895 | p++; | |
2e3e8ded | 896 | } |
ab197ca7 AB |
897 | if (isdigit(*p)) { |
898 | has_w = 1; | |
899 | do { | |
900 | width = (width * 10) + (*p++ - '0'); | |
901 | } while ((*p) && (isdigit(*p))); | |
902 | } | |
903 | if (*p) { | |
904 | print_field(act, pci, t, elapsed, pdu_len, pdu_buf, *p++, | |
905 | minus, has_w, width); | |
906 | } | |
907 | return p; | |
d0ca268b JA |
908 | } |
909 | ||
ab197ca7 AB |
910 | static char *fmt_select(int fmt_spec, struct blk_io_trace *t, |
911 | unsigned long long elapsed) | |
d0ca268b | 912 | { |
ab197ca7 AB |
913 | char *fmt; |
914 | char scratch_format[1024]; | |
2e3e8ded | 915 | |
ab197ca7 AB |
916 | if (override_format[fmt_spec] != NULL) |
917 | return override_format[fmt_spec]; | |
95c15013 | 918 | |
ab197ca7 AB |
919 | switch (fmt_spec) { |
920 | case 'C': /* Complete */ | |
921 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) { | |
922 | strcpy(scratch_format, HEADER); | |
182f2fcc | 923 | strcat(scratch_format, "(%P) "); |
ab197ca7 AB |
924 | } else { |
925 | strcpy(scratch_format, HEADER "%S + %n "); | |
926 | if (elapsed != -1ULL) | |
927 | strcat(scratch_format, "(%8u) "); | |
928 | } | |
929 | strcat(scratch_format, "[%e]\n"); | |
930 | fmt = scratch_format; | |
931 | break; | |
932 | ||
933 | case 'D': /* Issue */ | |
934 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) { | |
935 | strcpy(scratch_format, HEADER); | |
182f2fcc | 936 | strcat(scratch_format, "%n (%P) "); |
ab197ca7 AB |
937 | } else { |
938 | strcpy(scratch_format, HEADER "%S + %n "); | |
939 | if (elapsed != -1ULL) | |
940 | strcat(scratch_format, "(%8u) "); | |
941 | } | |
942 | strcat(scratch_format,"[%C]\n"); | |
943 | fmt = scratch_format; | |
944 | break; | |
945 | ||
b6076a9b JA |
946 | case 'I': /* Insert */ |
947 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) { | |
948 | strcpy(scratch_format, HEADER); | |
182f2fcc | 949 | strcat(scratch_format, "%n (%P) "); |
b6076a9b JA |
950 | } else { |
951 | strcpy(scratch_format, HEADER "%S + %n "); | |
952 | if (elapsed != -1ULL) | |
953 | strcat(scratch_format, "(%8u) "); | |
954 | } | |
955 | strcat(scratch_format,"[%C]\n"); | |
956 | fmt = scratch_format; | |
957 | break; | |
958 | ||
ab197ca7 | 959 | case 'Q': /* Queue */ |
93f1c611 | 960 | case 'W': /* Bounce */ |
ab197ca7 AB |
961 | strcpy(scratch_format, HEADER "%S + %n "); |
962 | if (elapsed != -1ULL) | |
963 | strcat(scratch_format, "(%8u) "); | |
964 | strcat(scratch_format,"[%C]\n"); | |
965 | fmt = scratch_format; | |
966 | break; | |
967 | ||
968 | case 'B': /* Back merge */ | |
969 | case 'F': /* Front merge */ | |
970 | case 'M': /* Front or back merge */ | |
971 | fmt = HEADER "%S + %n [%C]\n"; | |
972 | break; | |
973 | ||
974 | case 'P': /* Plug */ | |
975 | fmt = HEADER "[%C]\n"; | |
976 | break; | |
977 | ||
978 | case 'G': /* Get request */ | |
979 | case 'S': /* Sleep request */ | |
980 | fmt = HEADER "%S + %n [%C]\n"; | |
981 | break; | |
982 | ||
983 | case 'U': /* Unplug IO */ | |
984 | case 'T': /* Unplug timer */ | |
985 | fmt = HEADER "[%C] %U\n"; | |
986 | break; | |
987 | ||
93f1c611 JA |
988 | case 'X': /* Split */ |
989 | strcpy(scratch_format, HEADER "%S / %U "); | |
990 | strcat(scratch_format,"[%C]\n"); | |
991 | fmt = scratch_format; | |
992 | break; | |
993 | ||
ab197ca7 AB |
994 | default: |
995 | fprintf(stderr,"FATAL: Invalid format spec %c\n", fmt_spec); | |
996 | exit(1); | |
997 | /*NOTREACHED*/ | |
95c15013 | 998 | } |
ab197ca7 AB |
999 | |
1000 | return fmt; | |
d0ca268b JA |
1001 | } |
1002 | ||
ab197ca7 AB |
1003 | static void process_fmt(char *act, struct per_cpu_info *pci, |
1004 | struct blk_io_trace *t, unsigned long long elapsed, | |
1005 | int pdu_len, unsigned char *pdu_buf) | |
d0ca268b | 1006 | { |
ab197ca7 | 1007 | char *p = fmt_select(act[0], t, elapsed); |
2e3e8ded | 1008 | |
ab197ca7 AB |
1009 | while (*p) { |
1010 | switch (*p) { | |
1011 | case '%': /* Field specifier */ | |
1012 | p++; | |
1013 | if (*p == '%') | |
1014 | fprintf(ofp, "%c", *p++); | |
1015 | else if (!*p) | |
1016 | fprintf(ofp, "%c", '%'); | |
1017 | else | |
1018 | p = parse_field(act, pci, t, elapsed, | |
1019 | pdu_len, pdu_buf, p); | |
1020 | break; | |
1021 | case '\\': { /* escape */ | |
1022 | switch (p[1]) { | |
1023 | case 'b': fprintf(ofp, "\b"); break; | |
1024 | case 'n': fprintf(ofp, "\n"); break; | |
1025 | case 'r': fprintf(ofp, "\r"); break; | |
1026 | case 't': fprintf(ofp, "\t"); break; | |
1027 | default: | |
1028 | fprintf(stderr, | |
1029 | "Invalid escape char in format %c\n", | |
1030 | p[1]); | |
1031 | exit(1); | |
1032 | /*NOTREACHED*/ | |
1033 | } | |
1034 | p += 2; | |
1035 | break; | |
1036 | } | |
1037 | default: | |
1038 | fprintf(ofp, "%c", *p++); | |
1039 | break; | |
1040 | } | |
2e3e8ded | 1041 | } |
ab197ca7 AB |
1042 | } |
1043 | ||
1044 | static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t, | |
1045 | char *act) | |
1046 | { | |
1047 | process_fmt(act, pci, t, log_track_complete(t), 0, NULL); | |
1048 | } | |
1049 | ||
b6076a9b JA |
1050 | static void log_insert(struct per_cpu_info *pci, struct blk_io_trace *t, |
1051 | char *act) | |
1052 | { | |
1053 | process_fmt(act, pci, t, log_track_insert(t), 0, NULL); | |
1054 | } | |
1055 | ||
ab197ca7 AB |
1056 | static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t, |
1057 | char *act) | |
1058 | { | |
b6076a9b | 1059 | process_fmt(act, pci, t, -1, 0, NULL); |
ab197ca7 | 1060 | } |
2e3e8ded | 1061 | |
ab197ca7 AB |
1062 | static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t, |
1063 | char *act) | |
1064 | { | |
1065 | process_fmt(act, pci, t, log_track_issue(t), 0, NULL); | |
d0ca268b JA |
1066 | } |
1067 | ||
d5396421 | 1068 | static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1069 | char *act) |
d0ca268b | 1070 | { |
a01516de JA |
1071 | if (act[0] == 'F') |
1072 | log_track_frontmerge(t); | |
2e3e8ded | 1073 | |
ab197ca7 | 1074 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
1075 | } |
1076 | ||
dfe34da1 | 1077 | static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1078 | char *act) |
dfe34da1 | 1079 | { |
ab197ca7 | 1080 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
dfe34da1 JA |
1081 | } |
1082 | ||
d5396421 | 1083 | static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1084 | char *act) |
d0ca268b | 1085 | { |
ab197ca7 | 1086 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
1087 | } |
1088 | ||
ab197ca7 | 1089 | static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1090 | char *act) |
67e14fdc | 1091 | { |
ab197ca7 | 1092 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
67e14fdc JA |
1093 | } |
1094 | ||
93f1c611 JA |
1095 | static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t, |
1096 | char *act) | |
1097 | { | |
1098 | process_fmt(act, pci, t, -1ULL, 0, NULL); | |
1099 | } | |
1100 | ||
ab197ca7 | 1101 | static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act) |
d0ca268b | 1102 | { |
ab197ca7 | 1103 | unsigned char *buf = (unsigned char *) t + sizeof(*t); |
d0ca268b | 1104 | |
ab197ca7 | 1105 | process_fmt(act, pci, t, -1ULL, t->pdu_len, buf); |
d0ca268b JA |
1106 | } |
1107 | ||
ff3a732c | 1108 | static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 1109 | { |
56f2af81 JA |
1110 | int act = t->action & 0xffff; |
1111 | ||
1112 | switch (act) { | |
d0ca268b | 1113 | case __BLK_TA_QUEUE: |
3639a11e | 1114 | log_generic(pci, t, "Q"); |
d0ca268b JA |
1115 | break; |
1116 | case __BLK_TA_GETRQ: | |
3639a11e | 1117 | log_generic(pci, t, "G"); |
d0ca268b JA |
1118 | break; |
1119 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 1120 | log_generic(pci, t, "S"); |
d0ca268b JA |
1121 | break; |
1122 | case __BLK_TA_REQUEUE: | |
3639a11e | 1123 | log_generic(pci, t, "R"); |
d0ca268b JA |
1124 | break; |
1125 | case __BLK_TA_ISSUE: | |
ab197ca7 | 1126 | log_pc(pci, t, "D"); |
d0ca268b JA |
1127 | break; |
1128 | case __BLK_TA_COMPLETE: | |
3639a11e | 1129 | log_pc(pci, t, "C"); |
d0ca268b | 1130 | break; |
56f2af81 JA |
1131 | case __BLK_TA_INSERT: |
1132 | log_pc(pci, t, "I"); | |
1133 | break; | |
d0ca268b | 1134 | default: |
56f2af81 | 1135 | fprintf(stderr, "Bad pc action %x\n", act); |
87b72777 | 1136 | break; |
d0ca268b | 1137 | } |
d0ca268b JA |
1138 | } |
1139 | ||
d5396421 | 1140 | static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b JA |
1141 | { |
1142 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
7997c5b0 | 1143 | int act = t->action & 0xffff; |
d0ca268b | 1144 | |
7997c5b0 | 1145 | switch (act) { |
d0ca268b | 1146 | case __BLK_TA_QUEUE: |
b6076a9b | 1147 | account_queue(t, pci, w); |
3639a11e | 1148 | log_queue(pci, t, "Q"); |
d0ca268b | 1149 | break; |
b6076a9b JA |
1150 | case __BLK_TA_INSERT: |
1151 | log_insert(pci, t, "I"); | |
1152 | break; | |
d0ca268b | 1153 | case __BLK_TA_BACKMERGE: |
152f6476 | 1154 | account_m(t, pci, w); |
3639a11e | 1155 | log_merge(pci, t, "M"); |
d0ca268b JA |
1156 | break; |
1157 | case __BLK_TA_FRONTMERGE: | |
152f6476 | 1158 | account_m(t, pci, w); |
3639a11e | 1159 | log_merge(pci, t, "F"); |
d0ca268b JA |
1160 | break; |
1161 | case __BLK_TA_GETRQ: | |
95c15013 | 1162 | log_track_getrq(t); |
3639a11e | 1163 | log_generic(pci, t, "G"); |
d0ca268b JA |
1164 | break; |
1165 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 1166 | log_generic(pci, t, "S"); |
d0ca268b JA |
1167 | break; |
1168 | case __BLK_TA_REQUEUE: | |
152f6476 | 1169 | account_c(t, pci, w, -t->bytes); |
3639a11e | 1170 | log_queue(pci, t, "R"); |
d0ca268b JA |
1171 | break; |
1172 | case __BLK_TA_ISSUE: | |
b6076a9b | 1173 | account_issue(t, pci, w); |
3639a11e | 1174 | log_issue(pci, t, "D"); |
d0ca268b JA |
1175 | break; |
1176 | case __BLK_TA_COMPLETE: | |
152f6476 | 1177 | account_c(t, pci, w, t->bytes); |
3639a11e | 1178 | log_complete(pci, t, "C"); |
d0ca268b | 1179 | break; |
88b1a526 | 1180 | case __BLK_TA_PLUG: |
3639a11e | 1181 | log_action(pci, t, "P"); |
88b1a526 | 1182 | break; |
3639a11e | 1183 | case __BLK_TA_UNPLUG_IO: |
06639b27 | 1184 | account_unplug(t, pci, 0); |
3639a11e JA |
1185 | log_unplug(pci, t, "U"); |
1186 | break; | |
1187 | case __BLK_TA_UNPLUG_TIMER: | |
06639b27 | 1188 | account_unplug(t, pci, 1); |
3639a11e | 1189 | log_unplug(pci, t, "UT"); |
88b1a526 | 1190 | break; |
93f1c611 JA |
1191 | case __BLK_TA_SPLIT: |
1192 | log_split(pci, t, "X"); | |
1193 | break; | |
1194 | case __BLK_TA_BOUNCE: | |
1195 | log_generic(pci, t, "B"); | |
1196 | break; | |
d0ca268b JA |
1197 | default: |
1198 | fprintf(stderr, "Bad fs action %x\n", t->action); | |
1f79c4a0 | 1199 | break; |
d0ca268b | 1200 | } |
d0ca268b JA |
1201 | } |
1202 | ||
ff3a732c JA |
1203 | static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci, |
1204 | struct per_dev_info *pdi) | |
d0ca268b JA |
1205 | { |
1206 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) | |
ff3a732c | 1207 | dump_trace_pc(t, pci); |
d0ca268b | 1208 | else |
d5396421 | 1209 | dump_trace_fs(t, pci); |
87b72777 | 1210 | |
e7c9f3ff | 1211 | pdi->events++; |
d0ca268b JA |
1212 | } |
1213 | ||
152f6476 | 1214 | static void dump_io_stats(struct io_stats *ios, char *msg) |
5c017e4b | 1215 | { |
152f6476 JA |
1216 | fprintf(ofp, "%s\n", msg); |
1217 | ||
1218 | fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb); | |
1219 | fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb); | |
0a6b8fc4 | 1220 | |
152f6476 JA |
1221 | fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb); |
1222 | fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb); | |
1223 | fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb); | |
1224 | fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb); | |
1225 | fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' '); | |
152f6476 | 1226 | fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites); |
06639b27 JA |
1227 | fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' '); |
1228 | fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs); | |
5c017e4b JA |
1229 | } |
1230 | ||
50adc0ba JA |
1231 | static void dump_wait_stats(struct per_process_info *ppi) |
1232 | { | |
b9d40d6f JA |
1233 | unsigned long rawait = ppi->longest_allocation_wait[0] / 1000; |
1234 | unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000; | |
1235 | unsigned long rcwait = ppi->longest_completion_wait[0] / 1000; | |
1236 | unsigned long wawait = ppi->longest_allocation_wait[1] / 1000; | |
1237 | unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000; | |
1238 | unsigned long wcwait = ppi->longest_completion_wait[1] / 1000; | |
1239 | ||
1240 | fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' '); | |
1241 | fprintf(ofp, " Allocation wait: %'8lu\n", wawait); | |
1242 | fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' '); | |
1243 | fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait); | |
1244 | fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' '); | |
1245 | fprintf(ofp, " Completion wait: %'8lu\n", wcwait); | |
50adc0ba JA |
1246 | } |
1247 | ||
886ecf0e JA |
1248 | static int ppi_name_compare(const void *p1, const void *p2) |
1249 | { | |
1250 | struct per_process_info *ppi1 = *((struct per_process_info **) p1); | |
1251 | struct per_process_info *ppi2 = *((struct per_process_info **) p2); | |
1252 | int res; | |
1253 | ||
1254 | res = strverscmp(ppi1->name, ppi2->name); | |
1255 | if (!res) | |
06e6f286 | 1256 | res = ppi1->pid > ppi2->pid; |
886ecf0e JA |
1257 | |
1258 | return res; | |
1259 | } | |
1260 | ||
1261 | static void sort_process_list(void) | |
1262 | { | |
1263 | struct per_process_info **ppis; | |
1264 | struct per_process_info *ppi; | |
1265 | int i = 0; | |
1266 | ||
1267 | ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *)); | |
1268 | ||
1269 | ppi = ppi_list; | |
1270 | while (ppi) { | |
06e6f286 | 1271 | ppis[i++] = ppi; |
886ecf0e JA |
1272 | ppi = ppi->list_next; |
1273 | } | |
1274 | ||
06e6f286 | 1275 | qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare); |
886ecf0e JA |
1276 | |
1277 | i = ppi_list_entries - 1; | |
1278 | ppi_list = NULL; | |
1279 | while (i >= 0) { | |
1280 | ppi = ppis[i]; | |
1281 | ||
1282 | ppi->list_next = ppi_list; | |
1283 | ppi_list = ppi; | |
1284 | i--; | |
1285 | } | |
50c38702 JA |
1286 | |
1287 | free(ppis); | |
886ecf0e JA |
1288 | } |
1289 | ||
152f6476 JA |
1290 | static void show_process_stats(void) |
1291 | { | |
1292 | struct per_process_info *ppi; | |
1293 | ||
886ecf0e JA |
1294 | sort_process_list(); |
1295 | ||
152f6476 JA |
1296 | ppi = ppi_list; |
1297 | while (ppi) { | |
ce8b6b4f JA |
1298 | char name[64]; |
1299 | ||
1300 | snprintf(name, sizeof(name)-1, "%s (%u)", ppi->name, ppi->pid); | |
1301 | dump_io_stats(&ppi->io_stats, name); | |
50adc0ba | 1302 | dump_wait_stats(ppi); |
152f6476 JA |
1303 | ppi = ppi->list_next; |
1304 | } | |
1305 | ||
1306 | fprintf(ofp, "\n"); | |
1307 | } | |
1308 | ||
e7c9f3ff | 1309 | static void show_device_and_cpu_stats(void) |
d0ca268b | 1310 | { |
e7c9f3ff NS |
1311 | struct per_dev_info *pdi; |
1312 | struct per_cpu_info *pci; | |
1313 | struct io_stats total, *ios; | |
1314 | int i, j, pci_events; | |
1315 | char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3]; | |
1316 | char name[32]; | |
1317 | ||
1318 | for (pdi = devices, i = 0; i < ndevices; i++, pdi++) { | |
1319 | ||
1320 | memset(&total, 0, sizeof(total)); | |
1321 | pci_events = 0; | |
1322 | ||
1323 | if (i > 0) | |
1324 | fprintf(ofp, "\n"); | |
1325 | ||
1326 | for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) { | |
1327 | if (!pci->nelems) | |
1328 | continue; | |
1329 | ||
1330 | ios = &pci->io_stats; | |
1331 | total.qreads += ios->qreads; | |
1332 | total.qwrites += ios->qwrites; | |
1333 | total.creads += ios->creads; | |
1334 | total.cwrites += ios->cwrites; | |
1335 | total.mreads += ios->mreads; | |
1336 | total.mwrites += ios->mwrites; | |
1337 | total.ireads += ios->ireads; | |
1338 | total.iwrites += ios->iwrites; | |
1339 | total.qread_kb += ios->qread_kb; | |
1340 | total.qwrite_kb += ios->qwrite_kb; | |
1341 | total.cread_kb += ios->cread_kb; | |
1342 | total.cwrite_kb += ios->cwrite_kb; | |
1343 | total.iread_kb += ios->iread_kb; | |
1344 | total.iwrite_kb += ios->iwrite_kb; | |
06639b27 JA |
1345 | total.timer_unplugs += ios->timer_unplugs; |
1346 | total.io_unplugs += ios->io_unplugs; | |
e7c9f3ff NS |
1347 | |
1348 | snprintf(line, sizeof(line) - 1, "CPU%d (%s):", | |
1349 | j, get_dev_name(pdi, name, sizeof(name))); | |
1350 | dump_io_stats(ios, line); | |
1351 | pci_events++; | |
1352 | } | |
5c017e4b | 1353 | |
e7c9f3ff NS |
1354 | if (pci_events > 1) { |
1355 | fprintf(ofp, "\n"); | |
1356 | snprintf(line, sizeof(line) - 1, "Total (%s):", | |
1357 | get_dev_name(pdi, name, sizeof(name))); | |
1358 | dump_io_stats(&total, line); | |
1359 | } | |
d0ca268b | 1360 | |
0aa8a62d JA |
1361 | fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n", |
1362 | get_dev_name(pdi, line, sizeof(line)), pdi->events, | |
1363 | pdi->skips); | |
e7c9f3ff | 1364 | } |
d0ca268b JA |
1365 | } |
1366 | ||
1f7afa72 JA |
1367 | static inline int verify_and_add_trace(struct trace *t) |
1368 | { | |
1369 | if (verify_trace(t->bit)) | |
1370 | return 1; | |
1371 | if (trace_rb_insert(t)) | |
1372 | return 1; | |
1373 | ||
1374 | return 0; | |
1375 | } | |
1376 | ||
cb2a1a62 | 1377 | static int sort_entries(void) |
8fc0abbc | 1378 | { |
8fc0abbc | 1379 | struct trace *t; |
cb2a1a62 | 1380 | int nr = 0; |
8fc0abbc | 1381 | |
cb2a1a62 | 1382 | while ((t = trace_list) != NULL) { |
cb2a1a62 | 1383 | trace_list = t->next; |
7d747d22 JA |
1384 | if (verify_and_add_trace(t)) |
1385 | printf("oops\n"); | |
cb2a1a62 | 1386 | nr++; |
6fe4709e | 1387 | } |
8fc0abbc | 1388 | |
cb2a1a62 | 1389 | return nr; |
412819ce JA |
1390 | } |
1391 | ||
7d747d22 | 1392 | static void show_entries_rb(void) |
8fc0abbc | 1393 | { |
1f7afa72 JA |
1394 | struct per_dev_info *pdi = NULL; |
1395 | struct per_cpu_info *pci = NULL; | |
8fc0abbc | 1396 | struct blk_io_trace *bit; |
3aabcd89 | 1397 | struct rb_node *n; |
8fc0abbc | 1398 | struct trace *t; |
1f7afa72 JA |
1399 | __u32 device = 0; |
1400 | int cpu = 0; | |
1401 | ||
7d747d22 | 1402 | while ((n = rb_first(&rb_sort_root)) != NULL) { |
8fc0abbc | 1403 | |
1f7afa72 JA |
1404 | if (done) |
1405 | break; | |
8fc0abbc JA |
1406 | |
1407 | t = rb_entry(n, struct trace, rb_node); | |
1408 | bit = t->bit; | |
1409 | ||
1f7afa72 JA |
1410 | if (!pdi || device != bit->device) { |
1411 | device = bit->device; | |
1412 | pdi = get_dev_info(device); | |
1413 | } | |
1414 | ||
e7c9f3ff NS |
1415 | if (!pdi) { |
1416 | fprintf(stderr, "Unknown device ID? (%d,%d)\n", | |
1417 | MAJOR(bit->device), MINOR(bit->device)); | |
1418 | break; | |
1419 | } | |
1f7afa72 JA |
1420 | |
1421 | if (bit->cpu > pdi->ncpus) { | |
e7c9f3ff | 1422 | fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n", |
7d747d22 JA |
1423 | bit->cpu, MAJOR(bit->device), |
1424 | MINOR(bit->device)); | |
87b72777 | 1425 | break; |
8fc0abbc JA |
1426 | } |
1427 | ||
1f7afa72 JA |
1428 | if (!pci || cpu != bit->cpu) { |
1429 | cpu = bit->cpu; | |
1430 | pci = get_cpu_info(pdi, cpu); | |
1431 | } | |
1432 | ||
cb2a1a62 JA |
1433 | /* |
1434 | * back off displaying more info if we are out of sync | |
1435 | * on SMP systems. to prevent stalling on lost events, | |
0aa8a62d | 1436 | * only allow an event to us a few times |
cb2a1a62 | 1437 | */ |
1b928247 JA |
1438 | if (bit->sequence != (pdi->last_sequence + 1) |
1439 | && pdi->last_sequence != -1) { | |
7d747d22 | 1440 | if (t->skipped < 5) { |
0aa8a62d | 1441 | t->skipped++; |
cb2a1a62 | 1442 | break; |
0aa8a62d JA |
1443 | } else { |
1444 | fprintf(stderr, "skipping from %lu to %u\n", pdi->last_sequence, bit->sequence); | |
1445 | pdi->skips++; | |
cb2a1a62 JA |
1446 | } |
1447 | } | |
1448 | ||
1449 | pdi->last_sequence = bit->sequence; | |
1450 | ||
cfab07eb | 1451 | bit->time -= genesis_time; |
46e6968b NS |
1452 | if (bit->time >= stopwatch_end) |
1453 | break; | |
8fc0abbc | 1454 | |
1b928247 JA |
1455 | if (bit->time >= stopwatch_start) { |
1456 | check_time(pdi, bit); | |
8fc0abbc | 1457 | |
1f7afa72 | 1458 | dump_trace(bit, pci, pdi); |
1b928247 | 1459 | } |
87b72777 | 1460 | |
7d747d22 JA |
1461 | rb_erase(&t->rb_node, &rb_sort_root); |
1462 | free(bit); | |
1463 | free(t); | |
cb2a1a62 | 1464 | } |
8fc0abbc JA |
1465 | } |
1466 | ||
1f79c4a0 JA |
1467 | static int read_data(int fd, void *buffer, int bytes, int block) |
1468 | { | |
1469 | int ret, bytes_left, fl; | |
1470 | void *p; | |
1471 | ||
1472 | fl = fcntl(fd, F_GETFL); | |
1473 | ||
1474 | if (!block) | |
1475 | fcntl(fd, F_SETFL, fl | O_NONBLOCK); | |
1476 | else | |
1477 | fcntl(fd, F_SETFL, fl & ~O_NONBLOCK); | |
1478 | ||
1479 | bytes_left = bytes; | |
1480 | p = buffer; | |
1481 | while (bytes_left > 0) { | |
1482 | ret = read(fd, p, bytes_left); | |
1483 | if (!ret) | |
1484 | return 1; | |
1485 | else if (ret < 0) { | |
1486 | if (errno != EAGAIN) | |
1487 | perror("read"); | |
1488 | return -1; | |
1489 | } else { | |
1490 | p += ret; | |
1491 | bytes_left -= ret; | |
1492 | } | |
1493 | } | |
1494 | ||
1495 | return 0; | |
1496 | } | |
1497 | ||
7d747d22 | 1498 | static int read_sort_events(int fd) |
cb2a1a62 | 1499 | { |
7d747d22 JA |
1500 | int events = 0; |
1501 | ||
1502 | while (!is_done() && events < rb_batch) { | |
1503 | struct blk_io_trace *bit; | |
1504 | struct trace *t; | |
1505 | int pdu_len; | |
1506 | __u32 magic; | |
1507 | ||
1508 | bit = malloc(sizeof(*bit)); | |
cb2a1a62 | 1509 | |
7d747d22 | 1510 | if (read_data(fd, bit, sizeof(*bit), !events)) |
cb2a1a62 JA |
1511 | break; |
1512 | ||
7d747d22 JA |
1513 | magic = be32_to_cpu(bit->magic); |
1514 | if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
1515 | fprintf(stderr, "Bad magic %x\n", magic); | |
1516 | break; | |
1517 | } | |
1518 | ||
1519 | pdu_len = be16_to_cpu(bit->pdu_len); | |
1520 | if (pdu_len) { | |
1521 | void *ptr = realloc(bit, sizeof(*bit) + pdu_len); | |
1522 | ||
1523 | if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1)) | |
1524 | break; | |
1525 | ||
1526 | bit = ptr; | |
1527 | } | |
1528 | ||
cb2a1a62 JA |
1529 | t = malloc(sizeof(*t)); |
1530 | memset(t, 0, sizeof(*t)); | |
1531 | t->bit = bit; | |
1532 | ||
f1327cc5 | 1533 | trace_to_cpu(bit); |
7d747d22 JA |
1534 | t->next = trace_list; |
1535 | trace_list = t; | |
1f7afa72 | 1536 | |
7d747d22 | 1537 | events++; |
cb2a1a62 JA |
1538 | } |
1539 | ||
7d747d22 | 1540 | return events; |
cb2a1a62 JA |
1541 | } |
1542 | ||
d5396421 | 1543 | static int do_file(void) |
d0ca268b | 1544 | { |
7d747d22 JA |
1545 | struct per_cpu_info *pci; |
1546 | int i, j, nfiles = 0, events, events_added; | |
d0ca268b | 1547 | |
7d747d22 JA |
1548 | /* |
1549 | * first prepare all files for reading | |
1550 | */ | |
e8741a4a | 1551 | for (i = 0; i < ndevices; i++) { |
e7c9f3ff | 1552 | for (j = 0;; j++, nfiles++) { |
e8741a4a | 1553 | struct per_dev_info *pdi; |
e7c9f3ff | 1554 | struct stat st; |
87b72777 | 1555 | |
e8741a4a | 1556 | pdi = &devices[i]; |
1b928247 | 1557 | pdi->last_sequence = -1; |
e7c9f3ff NS |
1558 | pci = get_cpu_info(pdi, j); |
1559 | pci->cpu = j; | |
7d747d22 | 1560 | pci->fd = -1; |
d0ca268b | 1561 | |
e7c9f3ff | 1562 | snprintf(pci->fname, sizeof(pci->fname)-1, |
7d747d22 | 1563 | "%s.blktrace.%d", pdi->name, pci->cpu); |
e7c9f3ff NS |
1564 | if (stat(pci->fname, &st) < 0) |
1565 | break; | |
1566 | if (!st.st_size) | |
1567 | continue; | |
1568 | ||
e7c9f3ff NS |
1569 | pci->fd = open(pci->fname, O_RDONLY); |
1570 | if (pci->fd < 0) { | |
1571 | perror(pci->fname); | |
e7c9f3ff NS |
1572 | continue; |
1573 | } | |
1574 | ||
7d747d22 | 1575 | printf("Input file %s added\n", pci->fname); |
d0ca268b | 1576 | } |
d5396421 JA |
1577 | } |
1578 | ||
7d747d22 JA |
1579 | /* |
1580 | * now loop over the files reading in the data | |
1581 | */ | |
412819ce | 1582 | do { |
7d747d22 | 1583 | events_added = 0; |
d5396421 | 1584 | |
7d747d22 JA |
1585 | for (i = 0; i < ndevices; i++) { |
1586 | for (j = 0; j < nfiles; j++) { | |
d5396421 | 1587 | |
7d747d22 | 1588 | pci = get_cpu_info(&devices[i], j); |
d5396421 | 1589 | |
7d747d22 JA |
1590 | if (pci->fd == -1) |
1591 | continue; | |
51128a28 | 1592 | |
7d747d22 JA |
1593 | events = read_sort_events(pci->fd); |
1594 | if (!events) { | |
1595 | close(pci->fd); | |
1596 | pci->fd = -1; | |
1597 | continue; | |
1598 | } | |
d5396421 | 1599 | |
7d747d22 JA |
1600 | if (sort_entries() == -1) |
1601 | continue; | |
d5396421 | 1602 | |
7d747d22 JA |
1603 | events_added += events; |
1604 | } | |
2ff323b0 | 1605 | } |
d5396421 | 1606 | |
7d747d22 | 1607 | show_entries_rb(); |
cb2a1a62 | 1608 | |
7d747d22 | 1609 | } while (events_added); |
d5396421 | 1610 | |
7d747d22 | 1611 | return 0; |
412819ce | 1612 | } |
d5396421 | 1613 | |
412819ce JA |
1614 | static int do_stdin(void) |
1615 | { | |
1616 | int fd; | |
d5396421 | 1617 | |
1f79c4a0 | 1618 | fd = dup(STDIN_FILENO); |
412819ce JA |
1619 | do { |
1620 | int events; | |
d5396421 | 1621 | |
cb2a1a62 | 1622 | events = read_sort_events(fd); |
412819ce JA |
1623 | if (!events) |
1624 | break; | |
1625 | ||
cb2a1a62 | 1626 | if (sort_entries() == -1) |
2ff323b0 JA |
1627 | break; |
1628 | ||
7d747d22 | 1629 | show_entries_rb(); |
d5396421 JA |
1630 | } while (1); |
1631 | ||
1632 | close(fd); | |
d5396421 JA |
1633 | return 0; |
1634 | } | |
d0ca268b | 1635 | |
1f79c4a0 | 1636 | static void flush_output(void) |
412819ce | 1637 | { |
152f6476 | 1638 | fflush(ofp); |
412819ce JA |
1639 | } |
1640 | ||
1f79c4a0 | 1641 | static void handle_sigint(int sig) |
412819ce JA |
1642 | { |
1643 | done = 1; | |
1644 | flush_output(); | |
1645 | } | |
1646 | ||
46e6968b NS |
1647 | /* |
1648 | * Extract start and duration times from a string, allowing | |
1649 | * us to specify a time interval of interest within a trace. | |
1650 | * Format: "duration" (start is zero) or "start:duration". | |
1651 | */ | |
1652 | static int find_stopwatch_interval(char *string) | |
1653 | { | |
1654 | double value; | |
1655 | char *sp; | |
1656 | ||
1657 | value = strtod(string, &sp); | |
1658 | if (sp == string) { | |
1659 | fprintf(stderr,"Invalid stopwatch timer: %s\n", string); | |
1660 | return 1; | |
1661 | } | |
1662 | if (*sp == ':') { | |
1663 | stopwatch_start = DOUBLE_TO_NANO_ULL(value); | |
1664 | string = sp + 1; | |
1665 | value = strtod(string, &sp); | |
1666 | if (sp == string || *sp != '\0') { | |
1667 | fprintf(stderr,"Invalid stopwatch duration time: %s\n", | |
1668 | string); | |
1669 | return 1; | |
1670 | } | |
1671 | } else if (*sp != '\0') { | |
1672 | fprintf(stderr,"Invalid stopwatch start timer: %s\n", string); | |
1673 | return 1; | |
1674 | } | |
1b928247 JA |
1675 | stopwatch_end = DOUBLE_TO_NANO_ULL(value); |
1676 | if (stopwatch_end <= stopwatch_start) { | |
1677 | fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n", | |
1678 | stopwatch_start, stopwatch_end); | |
1679 | return 1; | |
1680 | } | |
1681 | ||
46e6968b NS |
1682 | return 0; |
1683 | } | |
1684 | ||
52724a0e JA |
1685 | static char usage_str[] = \ |
1686 | "[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \ | |
1687 | "[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \ | |
1688 | "\t-i Input file containing trace data, or '-' for stdin\n" \ | |
1689 | "\t-o Output file. If not given, output is stdout\n" \ | |
1690 | "\t-b stdin read batching\n" \ | |
1691 | "\t-s Show per-program io statistics\n" \ | |
1692 | "\t-t Track individual ios. Will tell you the time a request took\n" \ | |
1693 | "\t to get queued, to get dispatched, and to get completed\n" \ | |
1694 | "\t-q Quiet. Don't display any stats at the end of the trace\n" \ | |
1695 | "\t-w Only parse data between the given time interval in seconds.\n" \ | |
1696 | "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \ | |
1697 | "\t -f Output format. Customize the output format. The format field\n" \ | |
1698 | "\t identifies can be found in the documentation\n" \ | |
1699 | "\t-F Format specification. Can be found in the documentation\n" \ | |
1700 | "\t-v Print program version info\n\n"; | |
1701 | ||
1f79c4a0 JA |
1702 | static void usage(char *prog) |
1703 | { | |
52724a0e | 1704 | fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str); |
1f79c4a0 JA |
1705 | } |
1706 | ||
d5396421 JA |
1707 | int main(int argc, char *argv[]) |
1708 | { | |
152f6476 | 1709 | char *ofp_buffer; |
a66877e6 | 1710 | int c, ret, mode; |
1e1c60f1 | 1711 | int per_device_and_cpu_stats = 1; |
d5396421 JA |
1712 | |
1713 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) { | |
1714 | switch (c) { | |
1715 | case 'i': | |
e7c9f3ff NS |
1716 | if (!strcmp(optarg, "-") && !pipeline) |
1717 | pipeline = 1; | |
1718 | else if (resize_devices(optarg) != 0) | |
1719 | return 1; | |
d5396421 JA |
1720 | break; |
1721 | case 'o': | |
66efebf8 | 1722 | output_name = optarg; |
d5396421 | 1723 | break; |
79f19470 JA |
1724 | case 'b': |
1725 | rb_batch = atoi(optarg); | |
1726 | if (rb_batch <= 0) | |
1727 | rb_batch = RB_BATCH_DEFAULT; | |
1728 | break; | |
152f6476 JA |
1729 | case 's': |
1730 | per_process_stats = 1; | |
1731 | break; | |
7997c5b0 JA |
1732 | case 't': |
1733 | track_ios = 1; | |
1734 | break; | |
1e1c60f1 NS |
1735 | case 'q': |
1736 | per_device_and_cpu_stats = 0; | |
1737 | break; | |
46e6968b NS |
1738 | case 'w': |
1739 | if (find_stopwatch_interval(optarg) != 0) | |
1740 | return 1; | |
1741 | break; | |
ab197ca7 AB |
1742 | case 'f': |
1743 | set_all_format_specs(optarg); | |
1744 | break; | |
1745 | case 'F': | |
1746 | if (add_format_spec(optarg) != 0) | |
1747 | return 1; | |
1748 | break; | |
52724a0e JA |
1749 | case 'v': |
1750 | printf("%s version %s\n", argv[0], blkparse_version); | |
1751 | return 0; | |
d5396421 | 1752 | default: |
1f79c4a0 | 1753 | usage(argv[0]); |
d5396421 JA |
1754 | return 1; |
1755 | } | |
d0ca268b JA |
1756 | } |
1757 | ||
e7c9f3ff NS |
1758 | while (optind < argc) { |
1759 | if (!strcmp(argv[optind], "-") && !pipeline) | |
1760 | pipeline = 1; | |
1761 | else if (resize_devices(argv[optind]) != 0) | |
1762 | return 1; | |
1763 | optind++; | |
1764 | } | |
1765 | ||
1766 | if (!pipeline && !ndevices) { | |
1f79c4a0 | 1767 | usage(argv[0]); |
d5396421 JA |
1768 | return 1; |
1769 | } | |
1770 | ||
7997c5b0 JA |
1771 | memset(&rb_sort_root, 0, sizeof(rb_sort_root)); |
1772 | memset(&rb_track_root, 0, sizeof(rb_track_root)); | |
412819ce JA |
1773 | |
1774 | signal(SIGINT, handle_sigint); | |
1775 | signal(SIGHUP, handle_sigint); | |
1776 | signal(SIGTERM, handle_sigint); | |
d5396421 | 1777 | |
d69db225 JA |
1778 | setlocale(LC_NUMERIC, "en_US"); |
1779 | ||
a66877e6 | 1780 | if (!output_name) { |
152f6476 | 1781 | ofp = fdopen(STDOUT_FILENO, "w"); |
a66877e6 JA |
1782 | mode = _IOLBF; |
1783 | } else { | |
152f6476 JA |
1784 | char ofname[128]; |
1785 | ||
1786 | snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name); | |
1787 | ofp = fopen(ofname, "w"); | |
a66877e6 | 1788 | mode = _IOFBF; |
152f6476 JA |
1789 | } |
1790 | ||
1791 | if (!ofp) { | |
1792 | perror("fopen"); | |
1793 | return 1; | |
1794 | } | |
1795 | ||
1796 | ofp_buffer = malloc(4096); | |
a66877e6 | 1797 | if (setvbuf(ofp, ofp_buffer, mode, 4096)) { |
152f6476 JA |
1798 | perror("setvbuf"); |
1799 | return 1; | |
1800 | } | |
1801 | ||
e7c9f3ff | 1802 | if (pipeline) |
d5396421 JA |
1803 | ret = do_stdin(); |
1804 | else | |
1805 | ret = do_file(); | |
1806 | ||
152f6476 JA |
1807 | if (per_process_stats) |
1808 | show_process_stats(); | |
1809 | ||
1e1c60f1 NS |
1810 | if (per_device_and_cpu_stats) |
1811 | show_device_and_cpu_stats(); | |
152f6476 | 1812 | |
412819ce | 1813 | flush_output(); |
d5396421 | 1814 | return ret; |
d0ca268b | 1815 | } |