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> |
d0ca268b | 33 | |
8fc0abbc JA |
34 | #include "blktrace.h" |
35 | #include "rbtree.h" | |
bf0720af | 36 | #include "jhash.h" |
d0ca268b | 37 | |
52724a0e JA |
38 | static char blkparse_version[] = "0.90"; |
39 | ||
e7c9f3ff | 40 | struct per_dev_info { |
f7bd1a9b | 41 | dev_t dev; |
e7c9f3ff NS |
42 | char *name; |
43 | ||
44 | int backwards; | |
45 | unsigned long long events; | |
46 | unsigned long long last_reported_time; | |
287fa3d6 | 47 | unsigned long long last_read_time; |
e7c9f3ff | 48 | struct io_stats io_stats; |
cb2a1a62 | 49 | unsigned long last_sequence; |
dcf0f7ed | 50 | unsigned long skips; |
e7c9f3ff | 51 | |
2a1b3424 JA |
52 | struct rb_root rb_last; |
53 | unsigned long rb_last_entries; | |
54 | ||
f7bd1a9b JA |
55 | struct rb_root rb_track; |
56 | ||
73877e12 | 57 | int nfiles; |
774a1a10 | 58 | int nopenfiles; |
e7c9f3ff NS |
59 | int ncpus; |
60 | struct per_cpu_info *cpus; | |
61 | }; | |
62 | ||
152f6476 JA |
63 | struct per_process_info { |
64 | char name[16]; | |
65 | __u32 pid; | |
66 | struct io_stats io_stats; | |
67 | struct per_process_info *hash_next, *list_next; | |
715d8021 | 68 | int more_than_one; |
50adc0ba JA |
69 | |
70 | /* | |
71 | * individual io stats | |
72 | */ | |
b9d40d6f JA |
73 | unsigned long long longest_allocation_wait[2]; |
74 | unsigned long long longest_dispatch_wait[2]; | |
75 | unsigned long long longest_completion_wait[2]; | |
d0ca268b JA |
76 | }; |
77 | ||
152f6476 | 78 | #define PPI_HASH_SHIFT (8) |
bf0720af JA |
79 | #define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT) |
80 | #define PPI_HASH_MASK (PPI_HASH_SIZE - 1) | |
81 | static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE]; | |
152f6476 | 82 | static struct per_process_info *ppi_list; |
886ecf0e | 83 | static int ppi_list_entries; |
152f6476 | 84 | |
d915dee6 | 85 | #define S_OPTS "a:A:i:o:b:stqw:f:F:vVhD:" |
d5396421 | 86 | static struct option l_opts[] = { |
98f8386b AB |
87 | { |
88 | .name = "act-mask", | |
89 | .has_arg = required_argument, | |
90 | .flag = NULL, | |
91 | .val = 'a' | |
92 | }, | |
93 | { | |
94 | .name = "set-mask", | |
95 | .has_arg = required_argument, | |
96 | .flag = NULL, | |
97 | .val = 'A' | |
98 | }, | |
d5396421 JA |
99 | { |
100 | .name = "input", | |
428683db | 101 | .has_arg = required_argument, |
d5396421 JA |
102 | .flag = NULL, |
103 | .val = 'i' | |
104 | }, | |
105 | { | |
106 | .name = "output", | |
428683db | 107 | .has_arg = required_argument, |
d5396421 JA |
108 | .flag = NULL, |
109 | .val = 'o' | |
110 | }, | |
79f19470 JA |
111 | { |
112 | .name = "batch", | |
428683db | 113 | .has_arg = required_argument, |
79f19470 JA |
114 | .flag = NULL, |
115 | .val = 'b' | |
116 | }, | |
152f6476 | 117 | { |
3f65c585 | 118 | .name = "per-program-stats", |
428683db | 119 | .has_arg = no_argument, |
152f6476 JA |
120 | .flag = NULL, |
121 | .val = 's' | |
122 | }, | |
7997c5b0 | 123 | { |
3f65c585 | 124 | .name = "track-ios", |
428683db | 125 | .has_arg = no_argument, |
7997c5b0 JA |
126 | .flag = NULL, |
127 | .val = 't' | |
128 | }, | |
1e1c60f1 NS |
129 | { |
130 | .name = "quiet", | |
428683db | 131 | .has_arg = no_argument, |
1e1c60f1 NS |
132 | .flag = NULL, |
133 | .val = 'q' | |
134 | }, | |
46e6968b NS |
135 | { |
136 | .name = "stopwatch", | |
428683db | 137 | .has_arg = required_argument, |
46e6968b NS |
138 | .flag = NULL, |
139 | .val = 'w' | |
140 | }, | |
ab197ca7 AB |
141 | { |
142 | .name = "format", | |
428683db | 143 | .has_arg = required_argument, |
ab197ca7 AB |
144 | .flag = NULL, |
145 | .val = 'f' | |
146 | }, | |
147 | { | |
148 | .name = "format-spec", | |
428683db | 149 | .has_arg = required_argument, |
ab197ca7 AB |
150 | .flag = NULL, |
151 | .val = 'F' | |
152 | }, | |
bf0720af | 153 | { |
3f65c585 | 154 | .name = "hash-by-name", |
bf0720af JA |
155 | .has_arg = no_argument, |
156 | .flag = NULL, | |
d915dee6 | 157 | .val = 'h' |
bf0720af | 158 | }, |
7d1c0411 | 159 | { |
57ea8602 | 160 | .name = "verbose", |
7d1c0411 JA |
161 | .has_arg = no_argument, |
162 | .flag = NULL, | |
57ea8602 | 163 | .val = 'v' |
7d1c0411 | 164 | }, |
52724a0e JA |
165 | { |
166 | .name = "version", | |
167 | .has_arg = no_argument, | |
168 | .flag = NULL, | |
57ea8602 | 169 | .val = 'V' |
52724a0e | 170 | }, |
d1d7f15f | 171 | { |
3f65c585 | 172 | .name = "input-directory", |
d1d7f15f JA |
173 | .has_arg = required_argument, |
174 | .flag = NULL, | |
175 | .val = 'D' | |
176 | }, | |
71ef8b7c JA |
177 | { |
178 | .name = NULL, | |
179 | } | |
d5396421 JA |
180 | }; |
181 | ||
7997c5b0 JA |
182 | /* |
183 | * for sorting the displayed output | |
184 | */ | |
8fc0abbc JA |
185 | struct trace { |
186 | struct blk_io_trace *bit; | |
187 | struct rb_node rb_node; | |
cb2a1a62 | 188 | struct trace *next; |
8fc0abbc JA |
189 | }; |
190 | ||
cb2a1a62 | 191 | static struct rb_root rb_sort_root; |
a649216c JA |
192 | static unsigned long rb_sort_entries; |
193 | ||
cb2a1a62 JA |
194 | static struct trace *trace_list; |
195 | ||
d36421e4 JA |
196 | /* |
197 | * allocation cache | |
198 | */ | |
199 | static struct blk_io_trace *bit_alloc_list; | |
200 | static struct trace *t_alloc_list; | |
201 | ||
7997c5b0 JA |
202 | /* |
203 | * for tracking individual ios | |
204 | */ | |
205 | struct io_track { | |
206 | struct rb_node rb_node; | |
207 | ||
208 | __u64 sector; | |
209 | __u32 pid; | |
bf0720af | 210 | char comm[16]; |
95c15013 | 211 | unsigned long long allocation_time; |
7997c5b0 JA |
212 | unsigned long long queue_time; |
213 | unsigned long long dispatch_time; | |
214 | unsigned long long completion_time; | |
215 | }; | |
216 | ||
e7c9f3ff NS |
217 | static int ndevices; |
218 | static struct per_dev_info *devices; | |
219 | static char *get_dev_name(struct per_dev_info *, char *, int); | |
d0ca268b | 220 | |
71d5d4c9 | 221 | FILE *ofp = NULL; |
e7c9f3ff | 222 | static char *output_name; |
d1d7f15f | 223 | static char *input_dir; |
e7c9f3ff NS |
224 | |
225 | static unsigned long long genesis_time; | |
287fa3d6 | 226 | static unsigned long long last_allowed_time; |
1c7c54aa | 227 | static unsigned int smallest_seq_read; |
46e6968b NS |
228 | static unsigned long long stopwatch_start; /* start from zero by default */ |
229 | static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */ | |
152f6476 JA |
230 | |
231 | static int per_process_stats; | |
cbc927b6 | 232 | static int per_device_and_cpu_stats = 1; |
7997c5b0 | 233 | static int track_ios; |
bf0720af | 234 | static int ppi_hash_by_pid = 1; |
57ea8602 | 235 | static int verbose; |
98f8386b | 236 | static unsigned int act_mask = -1U; |
cbc927b6 | 237 | static int stats_printed; |
d0ca268b | 238 | |
1d24fc14 JA |
239 | static unsigned int t_alloc_cache; |
240 | static unsigned int bit_alloc_cache; | |
241 | ||
7d747d22 | 242 | #define RB_BATCH_DEFAULT (512) |
e820abd7 | 243 | static unsigned int rb_batch = RB_BATCH_DEFAULT; |
79f19470 | 244 | |
e7c9f3ff NS |
245 | static int pipeline; |
246 | ||
412819ce JA |
247 | #define is_done() (*(volatile int *)(&done)) |
248 | static volatile int done; | |
249 | ||
bf0720af JA |
250 | #define JHASH_RANDOM (0x3af5f2ee) |
251 | ||
252 | static inline int ppi_hash_pid(__u32 pid) | |
253 | { | |
254 | return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK; | |
255 | } | |
256 | ||
257 | static inline int ppi_hash_name(const char *name) | |
152f6476 | 258 | { |
bf0720af JA |
259 | return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK; |
260 | } | |
261 | ||
262 | static inline int ppi_hash(struct per_process_info *ppi) | |
263 | { | |
264 | if (ppi_hash_by_pid) | |
265 | return ppi_hash_pid(ppi->pid); | |
266 | ||
bf0720af | 267 | return ppi_hash_name(ppi->name); |
152f6476 JA |
268 | } |
269 | ||
270 | static inline void add_process_to_hash(struct per_process_info *ppi) | |
271 | { | |
bf0720af | 272 | const int hash_idx = ppi_hash(ppi); |
152f6476 | 273 | |
bf0720af JA |
274 | ppi->hash_next = ppi_hash_table[hash_idx]; |
275 | ppi_hash_table[hash_idx] = ppi; | |
152f6476 JA |
276 | } |
277 | ||
278 | static inline void add_process_to_list(struct per_process_info *ppi) | |
279 | { | |
280 | ppi->list_next = ppi_list; | |
281 | ppi_list = ppi; | |
886ecf0e | 282 | ppi_list_entries++; |
152f6476 JA |
283 | } |
284 | ||
bf0720af JA |
285 | static struct per_process_info *find_process_by_name(char *name) |
286 | { | |
287 | const int hash_idx = ppi_hash_name(name); | |
288 | struct per_process_info *ppi; | |
289 | ||
290 | ppi = ppi_hash_table[hash_idx]; | |
291 | while (ppi) { | |
292 | if (!strcmp(ppi->name, name)) | |
293 | return ppi; | |
294 | ||
295 | ppi = ppi->hash_next; | |
296 | } | |
297 | ||
298 | return NULL; | |
299 | } | |
300 | ||
152f6476 JA |
301 | static struct per_process_info *find_process_by_pid(__u32 pid) |
302 | { | |
bf0720af | 303 | const int hash_idx = ppi_hash_pid(pid); |
152f6476 JA |
304 | struct per_process_info *ppi; |
305 | ||
bf0720af | 306 | ppi = ppi_hash_table[hash_idx]; |
152f6476 JA |
307 | while (ppi) { |
308 | if (ppi->pid == pid) | |
309 | return ppi; | |
310 | ||
311 | ppi = ppi->hash_next; | |
312 | } | |
313 | ||
314 | return NULL; | |
315 | } | |
316 | ||
bf0720af JA |
317 | static struct per_process_info *find_process(__u32 pid, char *name) |
318 | { | |
715d8021 JA |
319 | struct per_process_info *ppi; |
320 | ||
bf0720af JA |
321 | if (ppi_hash_by_pid) |
322 | return find_process_by_pid(pid); | |
323 | ||
715d8021 JA |
324 | ppi = find_process_by_name(name); |
325 | if (ppi && ppi->pid != pid) | |
326 | ppi->more_than_one = 1; | |
327 | ||
328 | return ppi; | |
bf0720af JA |
329 | } |
330 | ||
23211fdf JA |
331 | static inline int trace_rb_insert(struct trace *t, struct rb_root *root, |
332 | int check_time) | |
7997c5b0 | 333 | { |
2a1b3424 | 334 | struct rb_node **p = &root->rb_node; |
7997c5b0 JA |
335 | struct rb_node *parent = NULL; |
336 | struct trace *__t; | |
337 | ||
338 | while (*p) { | |
339 | parent = *p; | |
2a1b3424 | 340 | |
7997c5b0 JA |
341 | __t = rb_entry(parent, struct trace, rb_node); |
342 | ||
23211fdf JA |
343 | if (check_time) { |
344 | if (t->bit->time < __t->bit->time) { | |
345 | p = &(*p)->rb_left; | |
346 | continue; | |
347 | } else if (t->bit->time > __t->bit->time) { | |
348 | p = &(*p)->rb_right; | |
349 | continue; | |
350 | } | |
351 | } | |
352 | if (t->bit->device < __t->bit->device) | |
e7c9f3ff NS |
353 | p = &(*p)->rb_left; |
354 | else if (t->bit->device > __t->bit->device) | |
355 | p = &(*p)->rb_right; | |
dcf0f7ed JA |
356 | else if (t->bit->sequence < __t->bit->sequence) |
357 | p = &(*p)->rb_left; | |
0b07f23e | 358 | else /* >= sequence */ |
dcf0f7ed | 359 | p = &(*p)->rb_right; |
7997c5b0 JA |
360 | } |
361 | ||
362 | rb_link_node(&t->rb_node, parent, p); | |
2a1b3424 | 363 | rb_insert_color(&t->rb_node, root); |
7997c5b0 JA |
364 | return 0; |
365 | } | |
366 | ||
2a1b3424 | 367 | static inline int trace_rb_insert_sort(struct trace *t) |
e3556946 | 368 | { |
23211fdf | 369 | if (!trace_rb_insert(t, &rb_sort_root, 1)) { |
2a1b3424 JA |
370 | rb_sort_entries++; |
371 | return 0; | |
372 | } | |
373 | ||
374 | return 1; | |
375 | } | |
376 | ||
377 | static inline int trace_rb_insert_last(struct per_dev_info *pdi,struct trace *t) | |
378 | { | |
23211fdf | 379 | if (!trace_rb_insert(t, &pdi->rb_last, 1)) { |
2a1b3424 JA |
380 | pdi->rb_last_entries++; |
381 | return 0; | |
382 | } | |
383 | ||
384 | return 1; | |
385 | } | |
386 | ||
387 | static struct trace *trace_rb_find(dev_t device, unsigned long sequence, | |
388 | struct rb_root *root, int order) | |
389 | { | |
390 | struct rb_node *n = root->rb_node; | |
391 | struct rb_node *prev = NULL; | |
e3556946 JA |
392 | struct trace *__t; |
393 | ||
2a1b3424 JA |
394 | while (n) { |
395 | __t = rb_entry(n, struct trace, rb_node); | |
396 | prev = n; | |
e3556946 | 397 | |
0583b6a2 | 398 | if (device < __t->bit->device) |
2a1b3424 | 399 | n = n->rb_left; |
0583b6a2 | 400 | else if (device > __t->bit->device) |
2a1b3424 | 401 | n = n->rb_right; |
0583b6a2 | 402 | else if (sequence < __t->bit->sequence) |
2a1b3424 | 403 | n = n->rb_left; |
e3556946 | 404 | else if (sequence > __t->bit->sequence) |
2a1b3424 | 405 | n = n->rb_right; |
e3556946 JA |
406 | else |
407 | return __t; | |
408 | } | |
409 | ||
2a1b3424 JA |
410 | /* |
411 | * hack - the list may not be sequence ordered because some | |
412 | * events don't have sequence and time matched. so we end up | |
413 | * being a little off in the rb lookup here, because we don't | |
414 | * know the time we are looking for. compensate by browsing | |
415 | * a little ahead from the last entry to find the match | |
416 | */ | |
417 | if (order && prev) { | |
418 | int max = 5; | |
419 | ||
420 | while (((n = rb_next(prev)) != NULL) && max--) { | |
421 | __t = rb_entry(n, struct trace, rb_node); | |
422 | ||
423 | if (__t->bit->device == device && | |
424 | __t->bit->sequence == sequence) | |
425 | return __t; | |
426 | ||
427 | prev = n; | |
428 | } | |
429 | } | |
430 | ||
e3556946 JA |
431 | return NULL; |
432 | } | |
433 | ||
2a1b3424 JA |
434 | static inline struct trace *trace_rb_find_sort(dev_t dev, unsigned long seq) |
435 | { | |
436 | return trace_rb_find(dev, seq, &rb_sort_root, 1); | |
437 | } | |
438 | ||
439 | static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi, | |
440 | unsigned long seq) | |
441 | { | |
f7bd1a9b | 442 | return trace_rb_find(pdi->dev, seq, &pdi->rb_last, 0); |
2a1b3424 JA |
443 | } |
444 | ||
f7bd1a9b | 445 | static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot) |
7997c5b0 | 446 | { |
f7bd1a9b | 447 | struct rb_node **p = &pdi->rb_track.rb_node; |
7997c5b0 JA |
448 | struct rb_node *parent = NULL; |
449 | struct io_track *__iot; | |
450 | ||
451 | while (*p) { | |
452 | parent = *p; | |
7997c5b0 JA |
453 | __iot = rb_entry(parent, struct io_track, rb_node); |
454 | ||
f7bd1a9b | 455 | if (iot->sector < __iot->sector) |
7997c5b0 JA |
456 | p = &(*p)->rb_left; |
457 | else if (iot->sector > __iot->sector) | |
458 | p = &(*p)->rb_right; | |
459 | else { | |
e7c9f3ff | 460 | fprintf(stderr, |
ab197ca7 AB |
461 | "sector alias (%Lu) on device %d,%d!\n", |
462 | (unsigned long long) iot->sector, | |
f7bd1a9b | 463 | MAJOR(pdi->dev), MINOR(pdi->dev)); |
7997c5b0 JA |
464 | return 1; |
465 | } | |
466 | } | |
467 | ||
468 | rb_link_node(&iot->rb_node, parent, p); | |
f7bd1a9b | 469 | rb_insert_color(&iot->rb_node, &pdi->rb_track); |
7997c5b0 JA |
470 | return 0; |
471 | } | |
472 | ||
f7bd1a9b | 473 | static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector) |
7997c5b0 | 474 | { |
f7bd1a9b | 475 | struct rb_node *n = pdi->rb_track.rb_node; |
7997c5b0 JA |
476 | struct io_track *__iot; |
477 | ||
2a1b3424 JA |
478 | while (n) { |
479 | __iot = rb_entry(n, struct io_track, rb_node); | |
7997c5b0 | 480 | |
f7bd1a9b | 481 | if (sector < __iot->sector) |
2a1b3424 | 482 | n = n->rb_left; |
7997c5b0 | 483 | else if (sector > __iot->sector) |
2a1b3424 | 484 | n = n->rb_right; |
7997c5b0 JA |
485 | else |
486 | return __iot; | |
487 | } | |
488 | ||
489 | return NULL; | |
490 | } | |
491 | ||
f7bd1a9b JA |
492 | static struct io_track *find_track(struct per_dev_info *pdi, __u32 pid, |
493 | char *comm, __u64 sector) | |
7997c5b0 | 494 | { |
916b5501 | 495 | struct io_track *iot; |
7997c5b0 | 496 | |
f7bd1a9b | 497 | iot = __find_track(pdi, sector); |
7997c5b0 JA |
498 | if (!iot) { |
499 | iot = malloc(sizeof(*iot)); | |
50adc0ba | 500 | iot->pid = pid; |
bf0720af | 501 | memcpy(iot->comm, comm, sizeof(iot->comm)); |
7997c5b0 | 502 | iot->sector = sector; |
f7bd1a9b | 503 | track_rb_insert(pdi, iot); |
7997c5b0 JA |
504 | } |
505 | ||
506 | return iot; | |
507 | } | |
508 | ||
f7bd1a9b JA |
509 | static void log_track_frontmerge(struct per_dev_info *pdi, |
510 | struct blk_io_trace *t) | |
2e3e8ded JA |
511 | { |
512 | struct io_track *iot; | |
513 | ||
514 | if (!track_ios) | |
515 | return; | |
2e3e8ded | 516 | |
ae957cbc | 517 | iot = __find_track(pdi, t->sector + t_sec(t)); |
cb2a1a62 | 518 | if (!iot) { |
57ea8602 JA |
519 | if (verbose) |
520 | fprintf(stderr, "merge not found for (%d,%d): %llu\n", | |
521 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
522 | (unsigned long long) t->sector + t_sec(t)); | |
cb2a1a62 | 523 | return; |
2e3e8ded | 524 | } |
cb2a1a62 | 525 | |
f7bd1a9b | 526 | rb_erase(&iot->rb_node, &pdi->rb_track); |
ae957cbc | 527 | iot->sector -= t_sec(t); |
f7bd1a9b | 528 | track_rb_insert(pdi, iot); |
2e3e8ded JA |
529 | } |
530 | ||
f7bd1a9b | 531 | static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t) |
2e3e8ded JA |
532 | { |
533 | struct io_track *iot; | |
534 | ||
535 | if (!track_ios) | |
536 | return; | |
537 | ||
f7bd1a9b | 538 | iot = find_track(pdi, t->pid, t->comm, t->sector); |
95c15013 JA |
539 | iot->allocation_time = t->time; |
540 | } | |
541 | ||
95c15013 | 542 | /* |
b6076a9b | 543 | * return time between rq allocation and insertion |
95c15013 | 544 | */ |
f7bd1a9b JA |
545 | static unsigned long long log_track_insert(struct per_dev_info *pdi, |
546 | struct blk_io_trace *t) | |
95c15013 | 547 | { |
50adc0ba | 548 | unsigned long long elapsed; |
95c15013 JA |
549 | struct io_track *iot; |
550 | ||
551 | if (!track_ios) | |
552 | return -1; | |
553 | ||
f7bd1a9b | 554 | iot = find_track(pdi, t->pid, t->comm, t->sector); |
2e3e8ded | 555 | iot->queue_time = t->time; |
acd70d21 JA |
556 | |
557 | if (!iot->allocation_time) | |
558 | return -1; | |
559 | ||
50adc0ba JA |
560 | elapsed = iot->queue_time - iot->allocation_time; |
561 | ||
562 | if (per_process_stats) { | |
bf0720af | 563 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 564 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 565 | |
b9d40d6f JA |
566 | if (ppi && elapsed > ppi->longest_allocation_wait[w]) |
567 | ppi->longest_allocation_wait[w] = elapsed; | |
50adc0ba JA |
568 | } |
569 | ||
570 | return elapsed; | |
2e3e8ded JA |
571 | } |
572 | ||
573 | /* | |
574 | * return time between queue and issue | |
575 | */ | |
f7bd1a9b JA |
576 | static unsigned long long log_track_issue(struct per_dev_info *pdi, |
577 | struct blk_io_trace *t) | |
2e3e8ded | 578 | { |
50adc0ba | 579 | unsigned long long elapsed; |
2e3e8ded JA |
580 | struct io_track *iot; |
581 | ||
582 | if (!track_ios) | |
583 | return -1; | |
584 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
585 | return -1; | |
586 | ||
f7bd1a9b | 587 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 588 | if (!iot) { |
57ea8602 JA |
589 | if (verbose) |
590 | fprintf(stderr, "issue not found for (%d,%d): %llu\n", | |
591 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
592 | (unsigned long long) t->sector); | |
2e3e8ded | 593 | return -1; |
cb2a1a62 | 594 | } |
2e3e8ded JA |
595 | |
596 | iot->dispatch_time = t->time; | |
50adc0ba JA |
597 | elapsed = iot->dispatch_time - iot->queue_time; |
598 | ||
599 | if (per_process_stats) { | |
bf0720af | 600 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 601 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 602 | |
b9d40d6f JA |
603 | if (ppi && elapsed > ppi->longest_dispatch_wait[w]) |
604 | ppi->longest_dispatch_wait[w] = elapsed; | |
50adc0ba JA |
605 | } |
606 | ||
607 | return elapsed; | |
2e3e8ded JA |
608 | } |
609 | ||
610 | /* | |
611 | * return time between dispatch and complete | |
612 | */ | |
f7bd1a9b JA |
613 | static unsigned long long log_track_complete(struct per_dev_info *pdi, |
614 | struct blk_io_trace *t) | |
2e3e8ded JA |
615 | { |
616 | unsigned long long elapsed; | |
617 | struct io_track *iot; | |
618 | ||
619 | if (!track_ios) | |
620 | return -1; | |
621 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
622 | return -1; | |
623 | ||
f7bd1a9b | 624 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 625 | if (!iot) { |
57ea8602 JA |
626 | if (verbose) |
627 | fprintf(stderr,"complete not found for (%d,%d): %llu\n", | |
628 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
629 | (unsigned long long) t->sector); | |
2e3e8ded | 630 | return -1; |
cb2a1a62 | 631 | } |
2e3e8ded JA |
632 | |
633 | iot->completion_time = t->time; | |
634 | elapsed = iot->completion_time - iot->dispatch_time; | |
635 | ||
50adc0ba | 636 | if (per_process_stats) { |
bf0720af | 637 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 638 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 639 | |
b9d40d6f JA |
640 | if (ppi && elapsed > ppi->longest_completion_wait[w]) |
641 | ppi->longest_completion_wait[w] = elapsed; | |
50adc0ba JA |
642 | } |
643 | ||
2e3e8ded JA |
644 | /* |
645 | * kill the trace, we don't need it after completion | |
646 | */ | |
f7bd1a9b | 647 | rb_erase(&iot->rb_node, &pdi->rb_track); |
2e3e8ded JA |
648 | free(iot); |
649 | ||
650 | return elapsed; | |
651 | } | |
652 | ||
653 | ||
152f6476 JA |
654 | static struct io_stats *find_process_io_stats(__u32 pid, char *name) |
655 | { | |
bf0720af | 656 | struct per_process_info *ppi = find_process(pid, name); |
152f6476 JA |
657 | |
658 | if (!ppi) { | |
659 | ppi = malloc(sizeof(*ppi)); | |
660 | memset(ppi, 0, sizeof(*ppi)); | |
bf0720af | 661 | memcpy(ppi->name, name, 16); |
152f6476 JA |
662 | ppi->pid = pid; |
663 | add_process_to_hash(ppi); | |
664 | add_process_to_list(ppi); | |
665 | } | |
666 | ||
667 | return &ppi->io_stats; | |
668 | } | |
669 | ||
e7c9f3ff | 670 | static void resize_cpu_info(struct per_dev_info *pdi, int cpu) |
a718bd37 | 671 | { |
e7c9f3ff NS |
672 | struct per_cpu_info *cpus = pdi->cpus; |
673 | int ncpus = pdi->ncpus; | |
674 | int new_count = cpu + 1; | |
675 | int new_space, size; | |
a718bd37 NS |
676 | char *new_start; |
677 | ||
e7c9f3ff NS |
678 | size = new_count * sizeof(struct per_cpu_info); |
679 | cpus = realloc(cpus, size); | |
680 | if (!cpus) { | |
681 | char name[20]; | |
682 | fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n", | |
683 | get_dev_name(pdi, name, sizeof(name)), size); | |
a718bd37 NS |
684 | exit(1); |
685 | } | |
686 | ||
e7c9f3ff NS |
687 | new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info)); |
688 | new_space = (new_count - ncpus) * sizeof(struct per_cpu_info); | |
a718bd37 | 689 | memset(new_start, 0, new_space); |
e7c9f3ff NS |
690 | |
691 | pdi->ncpus = new_count; | |
692 | pdi->cpus = cpus; | |
693 | } | |
cb2a1a62 | 694 | |
e7c9f3ff NS |
695 | static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu) |
696 | { | |
cb2a1a62 JA |
697 | struct per_cpu_info *pci; |
698 | ||
e7c9f3ff NS |
699 | if (cpu >= pdi->ncpus) |
700 | resize_cpu_info(pdi, cpu); | |
cb2a1a62 JA |
701 | |
702 | pci = &pdi->cpus[cpu]; | |
703 | pci->cpu = cpu; | |
704 | return pci; | |
a718bd37 NS |
705 | } |
706 | ||
e7c9f3ff NS |
707 | |
708 | static int resize_devices(char *name) | |
a718bd37 | 709 | { |
e7c9f3ff | 710 | int size = (ndevices + 1) * sizeof(struct per_dev_info); |
c499bf38 | 711 | |
e7c9f3ff NS |
712 | devices = realloc(devices, size); |
713 | if (!devices) { | |
714 | fprintf(stderr, "Out of memory, device %s (%d)\n", name, size); | |
715 | return 1; | |
716 | } | |
717 | memset(&devices[ndevices], 0, sizeof(struct per_dev_info)); | |
718 | devices[ndevices].name = name; | |
719 | ndevices++; | |
720 | return 0; | |
721 | } | |
a718bd37 | 722 | |
f7bd1a9b | 723 | static struct per_dev_info *get_dev_info(dev_t dev) |
e7c9f3ff | 724 | { |
cb2a1a62 | 725 | struct per_dev_info *pdi; |
e7c9f3ff | 726 | int i; |
c499bf38 | 727 | |
7d747d22 | 728 | for (i = 0; i < ndevices; i++) { |
f7bd1a9b JA |
729 | if (!devices[i].dev) |
730 | devices[i].dev = dev; | |
731 | if (devices[i].dev == dev) | |
e7c9f3ff | 732 | return &devices[i]; |
7d747d22 | 733 | } |
cb2a1a62 | 734 | |
2a1b3424 | 735 | if (resize_devices(NULL)) |
e7c9f3ff | 736 | return NULL; |
cb2a1a62 JA |
737 | |
738 | pdi = &devices[ndevices - 1]; | |
f7bd1a9b | 739 | pdi->dev = dev; |
d6222db8 | 740 | pdi->last_sequence = -1; |
287fa3d6 | 741 | pdi->last_read_time = 0; |
2a1b3424 JA |
742 | memset(&pdi->rb_last, 0, sizeof(pdi->rb_last)); |
743 | pdi->rb_last_entries = 0; | |
cb2a1a62 | 744 | return pdi; |
a718bd37 NS |
745 | } |
746 | ||
e7c9f3ff NS |
747 | static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size) |
748 | { | |
749 | if (pdi->name) | |
750 | snprintf(buffer, size, "%s", pdi->name); | |
751 | else | |
f7bd1a9b | 752 | snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev)); |
e7c9f3ff NS |
753 | return buffer; |
754 | } | |
755 | ||
e7c9f3ff | 756 | static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit) |
cfab07eb AB |
757 | { |
758 | unsigned long long this = bit->time; | |
e7c9f3ff | 759 | unsigned long long last = pdi->last_reported_time; |
cfab07eb | 760 | |
e7c9f3ff NS |
761 | pdi->backwards = (this < last) ? 'B' : ' '; |
762 | pdi->last_reported_time = this; | |
cfab07eb AB |
763 | } |
764 | ||
152f6476 JA |
765 | static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t, |
766 | int rw) | |
d0ca268b JA |
767 | { |
768 | if (rw) { | |
152f6476 | 769 | ios->mwrites++; |
ae957cbc | 770 | ios->qwrite_kb += t_kb(t); |
d0ca268b | 771 | } else { |
152f6476 | 772 | ios->mreads++; |
ae957cbc | 773 | ios->qread_kb += t_kb(t); |
152f6476 JA |
774 | } |
775 | } | |
776 | ||
777 | static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci, | |
778 | int rw) | |
779 | { | |
780 | __account_m(&pci->io_stats, t, rw); | |
781 | ||
782 | if (per_process_stats) { | |
783 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
784 | ||
785 | __account_m(ios, t, rw); | |
d0ca268b JA |
786 | } |
787 | } | |
788 | ||
b6076a9b JA |
789 | static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t, |
790 | int rw) | |
d0ca268b JA |
791 | { |
792 | if (rw) { | |
152f6476 | 793 | ios->qwrites++; |
ae957cbc | 794 | ios->qwrite_kb += t_kb(t); |
d0ca268b | 795 | } else { |
152f6476 | 796 | ios->qreads++; |
ae957cbc | 797 | ios->qread_kb += t_kb(t); |
152f6476 JA |
798 | } |
799 | } | |
800 | ||
b6076a9b JA |
801 | static inline void account_queue(struct blk_io_trace *t, |
802 | struct per_cpu_info *pci, int rw) | |
152f6476 | 803 | { |
b6076a9b | 804 | __account_queue(&pci->io_stats, t, rw); |
152f6476 JA |
805 | |
806 | if (per_process_stats) { | |
807 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
808 | ||
b6076a9b | 809 | __account_queue(ios, t, rw); |
d0ca268b JA |
810 | } |
811 | } | |
812 | ||
152f6476 | 813 | static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes) |
d0ca268b JA |
814 | { |
815 | if (rw) { | |
152f6476 JA |
816 | ios->cwrites++; |
817 | ios->cwrite_kb += bytes >> 10; | |
d0ca268b | 818 | } else { |
152f6476 JA |
819 | ios->creads++; |
820 | ios->cread_kb += bytes >> 10; | |
821 | } | |
822 | } | |
823 | ||
824 | static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci, | |
825 | int rw, int bytes) | |
826 | { | |
827 | __account_c(&pci->io_stats, rw, bytes); | |
828 | ||
829 | if (per_process_stats) { | |
830 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
831 | ||
832 | __account_c(ios, rw, bytes); | |
d0ca268b JA |
833 | } |
834 | } | |
835 | ||
b6076a9b JA |
836 | static inline void __account_issue(struct io_stats *ios, int rw, |
837 | unsigned int bytes) | |
afd2d7ad | 838 | { |
839 | if (rw) { | |
152f6476 JA |
840 | ios->iwrites++; |
841 | ios->iwrite_kb += bytes >> 10; | |
afd2d7ad | 842 | } else { |
152f6476 JA |
843 | ios->ireads++; |
844 | ios->iread_kb += bytes >> 10; | |
afd2d7ad | 845 | } |
846 | } | |
847 | ||
b6076a9b JA |
848 | static inline void account_issue(struct blk_io_trace *t, |
849 | struct per_cpu_info *pci, int rw) | |
d0ca268b | 850 | { |
b6076a9b | 851 | __account_issue(&pci->io_stats, rw, t->bytes); |
152f6476 JA |
852 | |
853 | if (per_process_stats) { | |
854 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
d5396421 | 855 | |
b6076a9b | 856 | __account_issue(ios, rw, t->bytes); |
152f6476 JA |
857 | } |
858 | } | |
859 | ||
06639b27 JA |
860 | static inline void __account_unplug(struct io_stats *ios, int timer) |
861 | { | |
862 | if (timer) | |
863 | ios->timer_unplugs++; | |
864 | else | |
865 | ios->io_unplugs++; | |
866 | } | |
867 | ||
868 | static inline void account_unplug(struct blk_io_trace *t, | |
869 | struct per_cpu_info *pci, int timer) | |
870 | { | |
871 | __account_unplug(&pci->io_stats, timer); | |
872 | ||
873 | if (per_process_stats) { | |
874 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
875 | ||
876 | __account_unplug(ios, timer); | |
877 | } | |
878 | } | |
879 | ||
f7bd1a9b JA |
880 | static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci, |
881 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 882 | { |
f7bd1a9b | 883 | process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL); |
ab197ca7 AB |
884 | } |
885 | ||
f7bd1a9b JA |
886 | static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci, |
887 | struct blk_io_trace *t, char *act) | |
b6076a9b | 888 | { |
f7bd1a9b | 889 | process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL); |
b6076a9b JA |
890 | } |
891 | ||
ab197ca7 AB |
892 | static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t, |
893 | char *act) | |
894 | { | |
b6076a9b | 895 | process_fmt(act, pci, t, -1, 0, NULL); |
ab197ca7 | 896 | } |
2e3e8ded | 897 | |
f7bd1a9b JA |
898 | static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci, |
899 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 900 | { |
f7bd1a9b | 901 | process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL); |
d0ca268b JA |
902 | } |
903 | ||
f7bd1a9b JA |
904 | static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci, |
905 | struct blk_io_trace *t, char *act) | |
d0ca268b | 906 | { |
a01516de | 907 | if (act[0] == 'F') |
f7bd1a9b | 908 | log_track_frontmerge(pdi, t); |
2e3e8ded | 909 | |
ab197ca7 | 910 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
911 | } |
912 | ||
dfe34da1 | 913 | static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 914 | char *act) |
dfe34da1 | 915 | { |
ab197ca7 | 916 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
dfe34da1 JA |
917 | } |
918 | ||
d5396421 | 919 | static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 920 | char *act) |
d0ca268b | 921 | { |
ab197ca7 | 922 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
923 | } |
924 | ||
ab197ca7 | 925 | static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 926 | char *act) |
67e14fdc | 927 | { |
ab197ca7 | 928 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
67e14fdc JA |
929 | } |
930 | ||
93f1c611 JA |
931 | static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t, |
932 | char *act) | |
933 | { | |
934 | process_fmt(act, pci, t, -1ULL, 0, NULL); | |
935 | } | |
936 | ||
ab197ca7 | 937 | static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act) |
d0ca268b | 938 | { |
ab197ca7 | 939 | unsigned char *buf = (unsigned char *) t + sizeof(*t); |
d0ca268b | 940 | |
ab197ca7 | 941 | process_fmt(act, pci, t, -1ULL, t->pdu_len, buf); |
d0ca268b JA |
942 | } |
943 | ||
ff3a732c | 944 | static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 945 | { |
56f2af81 JA |
946 | int act = t->action & 0xffff; |
947 | ||
948 | switch (act) { | |
d0ca268b | 949 | case __BLK_TA_QUEUE: |
3639a11e | 950 | log_generic(pci, t, "Q"); |
d0ca268b JA |
951 | break; |
952 | case __BLK_TA_GETRQ: | |
3639a11e | 953 | log_generic(pci, t, "G"); |
d0ca268b JA |
954 | break; |
955 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 956 | log_generic(pci, t, "S"); |
d0ca268b JA |
957 | break; |
958 | case __BLK_TA_REQUEUE: | |
3639a11e | 959 | log_generic(pci, t, "R"); |
d0ca268b JA |
960 | break; |
961 | case __BLK_TA_ISSUE: | |
ab197ca7 | 962 | log_pc(pci, t, "D"); |
d0ca268b JA |
963 | break; |
964 | case __BLK_TA_COMPLETE: | |
3639a11e | 965 | log_pc(pci, t, "C"); |
d0ca268b | 966 | break; |
56f2af81 JA |
967 | case __BLK_TA_INSERT: |
968 | log_pc(pci, t, "I"); | |
969 | break; | |
d0ca268b | 970 | default: |
56f2af81 | 971 | fprintf(stderr, "Bad pc action %x\n", act); |
87b72777 | 972 | break; |
d0ca268b | 973 | } |
d0ca268b JA |
974 | } |
975 | ||
f7bd1a9b JA |
976 | static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi, |
977 | struct per_cpu_info *pci) | |
d0ca268b JA |
978 | { |
979 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
7997c5b0 | 980 | int act = t->action & 0xffff; |
d0ca268b | 981 | |
7997c5b0 | 982 | switch (act) { |
d0ca268b | 983 | case __BLK_TA_QUEUE: |
b6076a9b | 984 | account_queue(t, pci, w); |
3639a11e | 985 | log_queue(pci, t, "Q"); |
d0ca268b | 986 | break; |
b6076a9b | 987 | case __BLK_TA_INSERT: |
f7bd1a9b | 988 | log_insert(pdi, pci, t, "I"); |
b6076a9b | 989 | break; |
d0ca268b | 990 | case __BLK_TA_BACKMERGE: |
152f6476 | 991 | account_m(t, pci, w); |
f7bd1a9b | 992 | log_merge(pdi, pci, t, "M"); |
d0ca268b JA |
993 | break; |
994 | case __BLK_TA_FRONTMERGE: | |
152f6476 | 995 | account_m(t, pci, w); |
f7bd1a9b | 996 | log_merge(pdi, pci, t, "F"); |
d0ca268b JA |
997 | break; |
998 | case __BLK_TA_GETRQ: | |
f7bd1a9b | 999 | log_track_getrq(pdi, t); |
3639a11e | 1000 | log_generic(pci, t, "G"); |
d0ca268b JA |
1001 | break; |
1002 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 1003 | log_generic(pci, t, "S"); |
d0ca268b JA |
1004 | break; |
1005 | case __BLK_TA_REQUEUE: | |
152f6476 | 1006 | account_c(t, pci, w, -t->bytes); |
3639a11e | 1007 | log_queue(pci, t, "R"); |
d0ca268b JA |
1008 | break; |
1009 | case __BLK_TA_ISSUE: | |
b6076a9b | 1010 | account_issue(t, pci, w); |
f7bd1a9b | 1011 | log_issue(pdi, pci, t, "D"); |
d0ca268b JA |
1012 | break; |
1013 | case __BLK_TA_COMPLETE: | |
152f6476 | 1014 | account_c(t, pci, w, t->bytes); |
f7bd1a9b | 1015 | log_complete(pdi, pci, t, "C"); |
d0ca268b | 1016 | break; |
88b1a526 | 1017 | case __BLK_TA_PLUG: |
3639a11e | 1018 | log_action(pci, t, "P"); |
88b1a526 | 1019 | break; |
3639a11e | 1020 | case __BLK_TA_UNPLUG_IO: |
06639b27 | 1021 | account_unplug(t, pci, 0); |
3639a11e JA |
1022 | log_unplug(pci, t, "U"); |
1023 | break; | |
1024 | case __BLK_TA_UNPLUG_TIMER: | |
06639b27 | 1025 | account_unplug(t, pci, 1); |
3639a11e | 1026 | log_unplug(pci, t, "UT"); |
88b1a526 | 1027 | break; |
93f1c611 JA |
1028 | case __BLK_TA_SPLIT: |
1029 | log_split(pci, t, "X"); | |
1030 | break; | |
1031 | case __BLK_TA_BOUNCE: | |
1032 | log_generic(pci, t, "B"); | |
1033 | break; | |
a8f30e64 JA |
1034 | case __BLK_TA_REMAP: |
1035 | log_generic(pci, t, "A"); | |
1036 | break; | |
d0ca268b JA |
1037 | default: |
1038 | fprintf(stderr, "Bad fs action %x\n", t->action); | |
1f79c4a0 | 1039 | break; |
d0ca268b | 1040 | } |
d0ca268b JA |
1041 | } |
1042 | ||
ff3a732c JA |
1043 | static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci, |
1044 | struct per_dev_info *pdi) | |
d0ca268b JA |
1045 | { |
1046 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) | |
ff3a732c | 1047 | dump_trace_pc(t, pci); |
d0ca268b | 1048 | else |
f7bd1a9b | 1049 | dump_trace_fs(t, pdi, pci); |
87b72777 | 1050 | |
e7c9f3ff | 1051 | pdi->events++; |
d0ca268b JA |
1052 | } |
1053 | ||
152f6476 | 1054 | static void dump_io_stats(struct io_stats *ios, char *msg) |
5c017e4b | 1055 | { |
152f6476 JA |
1056 | fprintf(ofp, "%s\n", msg); |
1057 | ||
1058 | fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb); | |
1059 | fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb); | |
0a6b8fc4 | 1060 | |
152f6476 JA |
1061 | fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb); |
1062 | fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb); | |
1063 | fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb); | |
1064 | fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb); | |
1065 | fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' '); | |
152f6476 | 1066 | fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites); |
06639b27 JA |
1067 | fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' '); |
1068 | fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs); | |
5c017e4b JA |
1069 | } |
1070 | ||
50adc0ba JA |
1071 | static void dump_wait_stats(struct per_process_info *ppi) |
1072 | { | |
b9d40d6f JA |
1073 | unsigned long rawait = ppi->longest_allocation_wait[0] / 1000; |
1074 | unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000; | |
1075 | unsigned long rcwait = ppi->longest_completion_wait[0] / 1000; | |
1076 | unsigned long wawait = ppi->longest_allocation_wait[1] / 1000; | |
1077 | unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000; | |
1078 | unsigned long wcwait = ppi->longest_completion_wait[1] / 1000; | |
1079 | ||
1080 | fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' '); | |
1081 | fprintf(ofp, " Allocation wait: %'8lu\n", wawait); | |
1082 | fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' '); | |
1083 | fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait); | |
1084 | fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' '); | |
1085 | fprintf(ofp, " Completion wait: %'8lu\n", wcwait); | |
50adc0ba JA |
1086 | } |
1087 | ||
886ecf0e JA |
1088 | static int ppi_name_compare(const void *p1, const void *p2) |
1089 | { | |
1090 | struct per_process_info *ppi1 = *((struct per_process_info **) p1); | |
1091 | struct per_process_info *ppi2 = *((struct per_process_info **) p2); | |
1092 | int res; | |
1093 | ||
1094 | res = strverscmp(ppi1->name, ppi2->name); | |
1095 | if (!res) | |
06e6f286 | 1096 | res = ppi1->pid > ppi2->pid; |
886ecf0e JA |
1097 | |
1098 | return res; | |
1099 | } | |
1100 | ||
1101 | static void sort_process_list(void) | |
1102 | { | |
1103 | struct per_process_info **ppis; | |
1104 | struct per_process_info *ppi; | |
1105 | int i = 0; | |
1106 | ||
1107 | ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *)); | |
1108 | ||
1109 | ppi = ppi_list; | |
1110 | while (ppi) { | |
06e6f286 | 1111 | ppis[i++] = ppi; |
886ecf0e JA |
1112 | ppi = ppi->list_next; |
1113 | } | |
1114 | ||
06e6f286 | 1115 | qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare); |
886ecf0e JA |
1116 | |
1117 | i = ppi_list_entries - 1; | |
1118 | ppi_list = NULL; | |
1119 | while (i >= 0) { | |
1120 | ppi = ppis[i]; | |
1121 | ||
1122 | ppi->list_next = ppi_list; | |
1123 | ppi_list = ppi; | |
1124 | i--; | |
1125 | } | |
50c38702 JA |
1126 | |
1127 | free(ppis); | |
886ecf0e JA |
1128 | } |
1129 | ||
152f6476 JA |
1130 | static void show_process_stats(void) |
1131 | { | |
1132 | struct per_process_info *ppi; | |
1133 | ||
886ecf0e JA |
1134 | sort_process_list(); |
1135 | ||
152f6476 JA |
1136 | ppi = ppi_list; |
1137 | while (ppi) { | |
ce8b6b4f JA |
1138 | char name[64]; |
1139 | ||
715d8021 | 1140 | if (ppi->more_than_one) |
bf0720af | 1141 | sprintf(name, "%s (%u, ...)", ppi->name, ppi->pid); |
715d8021 JA |
1142 | else |
1143 | sprintf(name, "%s (%u)", ppi->name, ppi->pid); | |
bf0720af | 1144 | |
ce8b6b4f | 1145 | dump_io_stats(&ppi->io_stats, name); |
50adc0ba | 1146 | dump_wait_stats(ppi); |
152f6476 JA |
1147 | ppi = ppi->list_next; |
1148 | } | |
1149 | ||
1150 | fprintf(ofp, "\n"); | |
1151 | } | |
1152 | ||
e7c9f3ff | 1153 | static void show_device_and_cpu_stats(void) |
d0ca268b | 1154 | { |
e7c9f3ff NS |
1155 | struct per_dev_info *pdi; |
1156 | struct per_cpu_info *pci; | |
1157 | struct io_stats total, *ios; | |
1158 | int i, j, pci_events; | |
1159 | char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3]; | |
1160 | char name[32]; | |
1161 | ||
1162 | for (pdi = devices, i = 0; i < ndevices; i++, pdi++) { | |
1163 | ||
1164 | memset(&total, 0, sizeof(total)); | |
1165 | pci_events = 0; | |
1166 | ||
1167 | if (i > 0) | |
1168 | fprintf(ofp, "\n"); | |
1169 | ||
1170 | for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) { | |
1171 | if (!pci->nelems) | |
1172 | continue; | |
1173 | ||
1174 | ios = &pci->io_stats; | |
1175 | total.qreads += ios->qreads; | |
1176 | total.qwrites += ios->qwrites; | |
1177 | total.creads += ios->creads; | |
1178 | total.cwrites += ios->cwrites; | |
1179 | total.mreads += ios->mreads; | |
1180 | total.mwrites += ios->mwrites; | |
1181 | total.ireads += ios->ireads; | |
1182 | total.iwrites += ios->iwrites; | |
1183 | total.qread_kb += ios->qread_kb; | |
1184 | total.qwrite_kb += ios->qwrite_kb; | |
1185 | total.cread_kb += ios->cread_kb; | |
1186 | total.cwrite_kb += ios->cwrite_kb; | |
1187 | total.iread_kb += ios->iread_kb; | |
1188 | total.iwrite_kb += ios->iwrite_kb; | |
06639b27 JA |
1189 | total.timer_unplugs += ios->timer_unplugs; |
1190 | total.io_unplugs += ios->io_unplugs; | |
e7c9f3ff NS |
1191 | |
1192 | snprintf(line, sizeof(line) - 1, "CPU%d (%s):", | |
1193 | j, get_dev_name(pdi, name, sizeof(name))); | |
1194 | dump_io_stats(ios, line); | |
1195 | pci_events++; | |
1196 | } | |
5c017e4b | 1197 | |
e7c9f3ff NS |
1198 | if (pci_events > 1) { |
1199 | fprintf(ofp, "\n"); | |
1200 | snprintf(line, sizeof(line) - 1, "Total (%s):", | |
1201 | get_dev_name(pdi, name, sizeof(name))); | |
1202 | dump_io_stats(&total, line); | |
1203 | } | |
d0ca268b | 1204 | |
dcf0f7ed | 1205 | fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n", |
0aa8a62d | 1206 | get_dev_name(pdi, line, sizeof(line)), pdi->events, |
dcf0f7ed | 1207 | pdi->skips); |
e7c9f3ff | 1208 | } |
d0ca268b JA |
1209 | } |
1210 | ||
d36421e4 JA |
1211 | /* |
1212 | * struct trace and blktrace allocation cache, we do potentially | |
1213 | * millions of mallocs for these structures while only using at most | |
1214 | * a few thousand at the time | |
1215 | */ | |
1216 | static inline void t_free(struct trace *t) | |
1217 | { | |
3c8413cb | 1218 | if (t_alloc_cache < 1024) { |
1d24fc14 JA |
1219 | t->next = t_alloc_list; |
1220 | t_alloc_list = t; | |
3c8413cb JA |
1221 | t_alloc_cache++; |
1222 | } else | |
1223 | free(t); | |
d36421e4 JA |
1224 | } |
1225 | ||
1226 | static inline struct trace *t_alloc(void) | |
1227 | { | |
1228 | struct trace *t = t_alloc_list; | |
1229 | ||
1230 | if (t) { | |
1231 | t_alloc_list = t->next; | |
3c8413cb | 1232 | t_alloc_cache--; |
d36421e4 JA |
1233 | return t; |
1234 | } | |
1235 | ||
1236 | return malloc(sizeof(*t)); | |
1237 | } | |
1238 | ||
1239 | static inline void bit_free(struct blk_io_trace *bit) | |
1240 | { | |
3c8413cb | 1241 | if (bit_alloc_cache < 1024) { |
1d24fc14 JA |
1242 | /* |
1243 | * abuse a 64-bit field for a next pointer for the free item | |
1244 | */ | |
1245 | bit->time = (__u64) (unsigned long) bit_alloc_list; | |
1246 | bit_alloc_list = (struct blk_io_trace *) bit; | |
3c8413cb JA |
1247 | bit_alloc_cache++; |
1248 | } else | |
1249 | free(bit); | |
d36421e4 JA |
1250 | } |
1251 | ||
1252 | static inline struct blk_io_trace *bit_alloc(void) | |
1253 | { | |
1254 | struct blk_io_trace *bit = bit_alloc_list; | |
1255 | ||
1256 | if (bit) { | |
0c39425c NS |
1257 | bit_alloc_list = (struct blk_io_trace *) (unsigned long) \ |
1258 | bit->time; | |
3c8413cb | 1259 | bit_alloc_cache--; |
d36421e4 JA |
1260 | return bit; |
1261 | } | |
1262 | ||
1263 | return malloc(sizeof(*bit)); | |
1264 | } | |
1265 | ||
4f0ae44f JA |
1266 | static void find_genesis(void) |
1267 | { | |
1268 | struct trace *t = trace_list; | |
1269 | ||
1270 | genesis_time = -1ULL; | |
1271 | while (t != NULL) { | |
1272 | if (t->bit->time < genesis_time) | |
1273 | genesis_time = t->bit->time; | |
1274 | ||
1275 | t = t->next; | |
1276 | } | |
1277 | } | |
1278 | ||
7f4d89e6 | 1279 | static inline int check_stopwatch(struct blk_io_trace *bit) |
4f0ae44f | 1280 | { |
7f4d89e6 JA |
1281 | if (bit->time < stopwatch_end && |
1282 | bit->time >= stopwatch_start) | |
4f0ae44f JA |
1283 | return 0; |
1284 | ||
1285 | return 1; | |
1286 | } | |
1287 | ||
53c68c88 JA |
1288 | /* |
1289 | * return youngest entry read | |
1290 | */ | |
1291 | static int sort_entries(unsigned long long *youngest) | |
4f0ae44f JA |
1292 | { |
1293 | struct trace *t; | |
4f0ae44f JA |
1294 | |
1295 | if (!genesis_time) | |
1296 | find_genesis(); | |
1297 | ||
d6222db8 | 1298 | *youngest = 0; |
4f0ae44f JA |
1299 | while ((t = trace_list) != NULL) { |
1300 | struct blk_io_trace *bit = t->bit; | |
1301 | ||
1302 | trace_list = t->next; | |
1303 | ||
7f4d89e6 | 1304 | bit->time -= genesis_time; |
4f0ae44f | 1305 | |
d6222db8 JA |
1306 | if (bit->time < *youngest || !*youngest) |
1307 | *youngest = bit->time; | |
1308 | ||
774a1a10 JA |
1309 | if (bit->sequence < smallest_seq_read) |
1310 | smallest_seq_read = bit->sequence; | |
1311 | ||
7f4d89e6 | 1312 | if (check_stopwatch(bit)) { |
4f0ae44f JA |
1313 | bit_free(bit); |
1314 | t_free(t); | |
1315 | continue; | |
1316 | } | |
1317 | ||
2a1b3424 | 1318 | if (trace_rb_insert_sort(t)) |
53c68c88 | 1319 | return -1; |
4f0ae44f JA |
1320 | } |
1321 | ||
53c68c88 | 1322 | return 0; |
4f0ae44f JA |
1323 | } |
1324 | ||
1c24add6 JA |
1325 | static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t) |
1326 | { | |
1327 | rb_erase(&t->rb_node, &pdi->rb_last); | |
1328 | pdi->rb_last_entries--; | |
1329 | ||
1330 | bit_free(t->bit); | |
1331 | t_free(t); | |
1332 | } | |
1333 | ||
1334 | static void put_trace(struct per_dev_info *pdi, struct trace *t) | |
2a1b3424 JA |
1335 | { |
1336 | rb_erase(&t->rb_node, &rb_sort_root); | |
1337 | rb_sort_entries--; | |
1338 | ||
1339 | trace_rb_insert_last(pdi, t); | |
1340 | ||
1c24add6 | 1341 | if (pdi->rb_last_entries > rb_batch * pdi->nfiles) { |
2a1b3424 JA |
1342 | struct rb_node *n = rb_first(&pdi->rb_last); |
1343 | ||
1344 | t = rb_entry(n, struct trace, rb_node); | |
1c24add6 | 1345 | __put_trace_last(pdi, t); |
2a1b3424 JA |
1346 | } |
1347 | } | |
1348 | ||
a141a7cd | 1349 | static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force) |
2a1b3424 JA |
1350 | { |
1351 | unsigned long expected_sequence = pdi->last_sequence + 1; | |
1ca323a5 JA |
1352 | struct blk_io_trace *bit = t->bit; |
1353 | struct trace *__t; | |
d6222db8 | 1354 | |
774a1a10 JA |
1355 | if (!expected_sequence) { |
1356 | struct rb_node *n; | |
1357 | char *cpu_seen; | |
1358 | int cpus; | |
1359 | ||
1360 | /* | |
1361 | * 1 should be the first entry, just allow it | |
1362 | */ | |
1363 | if (bit->sequence == 1) | |
1364 | return 0; | |
1365 | ||
1366 | /* | |
1367 | * if we are starting somewhere else, check that we have | |
1368 | * entries from all cpus in the tree before dumping one | |
1369 | */ | |
1370 | cpu_seen = malloc(pdi->nopenfiles); | |
1371 | n = rb_first(&rb_sort_root); | |
1372 | cpus = 0; | |
1373 | while (n) { | |
1374 | __t = rb_entry(n, struct trace, rb_node); | |
1375 | ||
1376 | if (!cpu_seen[__t->bit->cpu]) { | |
1377 | cpu_seen[__t->bit->cpu] = 1; | |
1378 | cpus++; | |
1379 | } | |
1380 | n = rb_next(n); | |
1381 | } | |
1382 | ||
1383 | free(cpu_seen); | |
1384 | return cpus != pdi->nopenfiles; | |
1385 | } | |
2a1b3424 JA |
1386 | |
1387 | if (bit->sequence == expected_sequence) | |
1388 | return 0; | |
1389 | ||
2a1b3424 | 1390 | /* |
1c7c54aa JA |
1391 | * we may not have seen that sequence yet. if we are not doing |
1392 | * the final run, break and wait for more entries. | |
1c24add6 | 1393 | */ |
1c7c54aa | 1394 | if (expected_sequence < smallest_seq_read) { |
1ca323a5 JA |
1395 | __t = trace_rb_find_last(pdi, expected_sequence); |
1396 | if (!__t) | |
1c7c54aa | 1397 | goto skip; |
2a1b3424 | 1398 | |
1ca323a5 | 1399 | __put_trace_last(pdi, __t); |
2a1b3424 | 1400 | return 0; |
a141a7cd JA |
1401 | } else if (!force) { |
1402 | return 1; | |
0b07f23e | 1403 | } else { |
1c7c54aa | 1404 | skip: |
57ea8602 | 1405 | if (verbose) { |
7d1c0411 JA |
1406 | fprintf(stderr, "(%d,%d): skipping %lu -> %u\n", |
1407 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
1408 | pdi->last_sequence, bit->sequence); | |
1409 | } | |
1c7c54aa JA |
1410 | pdi->skips++; |
1411 | return 0; | |
1412 | } | |
2a1b3424 JA |
1413 | } |
1414 | ||
a649216c | 1415 | static void show_entries_rb(int force) |
8fc0abbc | 1416 | { |
1f7afa72 JA |
1417 | struct per_dev_info *pdi = NULL; |
1418 | struct per_cpu_info *pci = NULL; | |
8fc0abbc | 1419 | struct blk_io_trace *bit; |
3aabcd89 | 1420 | struct rb_node *n; |
8fc0abbc | 1421 | struct trace *t; |
1f7afa72 | 1422 | |
7d747d22 | 1423 | while ((n = rb_first(&rb_sort_root)) != NULL) { |
dd90748f | 1424 | if (is_done() && !force && !pipeline) |
1f7afa72 | 1425 | break; |
8fc0abbc JA |
1426 | |
1427 | t = rb_entry(n, struct trace, rb_node); | |
1428 | bit = t->bit; | |
1429 | ||
f7bd1a9b | 1430 | if (!pdi || pdi->dev != bit->device) |
287fa3d6 | 1431 | pdi = get_dev_info(bit->device); |
1f7afa72 | 1432 | |
e7c9f3ff NS |
1433 | if (!pdi) { |
1434 | fprintf(stderr, "Unknown device ID? (%d,%d)\n", | |
1435 | MAJOR(bit->device), MINOR(bit->device)); | |
1436 | break; | |
1437 | } | |
1f7afa72 | 1438 | |
a141a7cd JA |
1439 | if (check_sequence(pdi, t, force)) |
1440 | break; | |
cb2a1a62 | 1441 | |
a141a7cd JA |
1442 | if (!force && bit->time > last_allowed_time) |
1443 | break; | |
8fc0abbc | 1444 | |
be925321 JA |
1445 | pdi->last_sequence = bit->sequence; |
1446 | ||
4f0ae44f | 1447 | check_time(pdi, bit); |
8fc0abbc | 1448 | |
4f0ae44f JA |
1449 | if (!pci || pci->cpu != bit->cpu) |
1450 | pci = get_cpu_info(pdi, bit->cpu); | |
287fa3d6 | 1451 | |
cbc927b6 JA |
1452 | pci->nelems++; |
1453 | ||
98f8386b AB |
1454 | if (bit->action & (act_mask << BLK_TC_SHIFT)) |
1455 | dump_trace(bit, pci, pdi); | |
87b72777 | 1456 | |
2a1b3424 | 1457 | put_trace(pdi, t); |
cb2a1a62 | 1458 | } |
8fc0abbc JA |
1459 | } |
1460 | ||
1f79c4a0 JA |
1461 | static int read_data(int fd, void *buffer, int bytes, int block) |
1462 | { | |
1463 | int ret, bytes_left, fl; | |
1464 | void *p; | |
1465 | ||
1466 | fl = fcntl(fd, F_GETFL); | |
1467 | ||
1468 | if (!block) | |
1469 | fcntl(fd, F_SETFL, fl | O_NONBLOCK); | |
1470 | else | |
1471 | fcntl(fd, F_SETFL, fl & ~O_NONBLOCK); | |
1472 | ||
1473 | bytes_left = bytes; | |
1474 | p = buffer; | |
1475 | while (bytes_left > 0) { | |
1476 | ret = read(fd, p, bytes_left); | |
1477 | if (!ret) | |
1478 | return 1; | |
1479 | else if (ret < 0) { | |
1480 | if (errno != EAGAIN) | |
1481 | perror("read"); | |
a649216c | 1482 | |
1f79c4a0 JA |
1483 | return -1; |
1484 | } else { | |
1485 | p += ret; | |
1486 | bytes_left -= ret; | |
1487 | } | |
1488 | } | |
1489 | ||
1490 | return 0; | |
1491 | } | |
1492 | ||
a649216c | 1493 | static int read_events(int fd, int always_block) |
cb2a1a62 | 1494 | { |
287fa3d6 | 1495 | struct per_dev_info *pdi = NULL; |
e820abd7 | 1496 | unsigned int events = 0; |
7d747d22 JA |
1497 | |
1498 | while (!is_done() && events < rb_batch) { | |
1499 | struct blk_io_trace *bit; | |
1500 | struct trace *t; | |
1501 | int pdu_len; | |
1502 | __u32 magic; | |
1503 | ||
d36421e4 | 1504 | bit = bit_alloc(); |
cb2a1a62 | 1505 | |
a649216c | 1506 | if (read_data(fd, bit, sizeof(*bit), !events || always_block)) |
cb2a1a62 JA |
1507 | break; |
1508 | ||
7d747d22 JA |
1509 | magic = be32_to_cpu(bit->magic); |
1510 | if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
1511 | fprintf(stderr, "Bad magic %x\n", magic); | |
1512 | break; | |
1513 | } | |
1514 | ||
1515 | pdu_len = be16_to_cpu(bit->pdu_len); | |
1516 | if (pdu_len) { | |
1517 | void *ptr = realloc(bit, sizeof(*bit) + pdu_len); | |
1518 | ||
1519 | if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1)) | |
1520 | break; | |
1521 | ||
1522 | bit = ptr; | |
1523 | } | |
1524 | ||
d6222db8 JA |
1525 | trace_to_cpu(bit); |
1526 | ||
1527 | if (verify_trace(bit)) { | |
1528 | bit_free(bit); | |
1529 | continue; | |
1530 | } | |
1531 | ||
d36421e4 | 1532 | t = t_alloc(); |
cb2a1a62 JA |
1533 | memset(t, 0, sizeof(*t)); |
1534 | t->bit = bit; | |
1535 | ||
7d747d22 JA |
1536 | t->next = trace_list; |
1537 | trace_list = t; | |
1f7afa72 | 1538 | |
f7bd1a9b | 1539 | if (!pdi || pdi->dev != bit->device) |
287fa3d6 JA |
1540 | pdi = get_dev_info(bit->device); |
1541 | ||
1542 | if (bit->time > pdi->last_read_time) | |
1543 | pdi->last_read_time = bit->time; | |
1544 | ||
7d747d22 | 1545 | events++; |
cb2a1a62 JA |
1546 | } |
1547 | ||
7d747d22 | 1548 | return events; |
cb2a1a62 JA |
1549 | } |
1550 | ||
d5396421 | 1551 | static int do_file(void) |
d0ca268b | 1552 | { |
7d747d22 | 1553 | struct per_cpu_info *pci; |
73877e12 NS |
1554 | struct per_dev_info *pdi; |
1555 | int i, j, events, events_added; | |
d0ca268b | 1556 | |
7d747d22 JA |
1557 | /* |
1558 | * first prepare all files for reading | |
1559 | */ | |
e8741a4a | 1560 | for (i = 0; i < ndevices; i++) { |
73877e12 NS |
1561 | pdi = &devices[i]; |
1562 | pdi->nfiles = 0; | |
d6222db8 | 1563 | pdi->last_sequence = -1; |
73877e12 | 1564 | |
74feaeb5 | 1565 | for (j = 0;; j++) { |
e7c9f3ff | 1566 | struct stat st; |
d1d7f15f | 1567 | int len = 0; |
87b72777 | 1568 | |
e7c9f3ff NS |
1569 | pci = get_cpu_info(pdi, j); |
1570 | pci->cpu = j; | |
7d747d22 | 1571 | pci->fd = -1; |
d0ca268b | 1572 | |
d1d7f15f JA |
1573 | if (input_dir) |
1574 | len = sprintf(pci->fname, "%s/", input_dir); | |
1575 | ||
1576 | snprintf(pci->fname + len, sizeof(pci->fname)-1-len, | |
7d747d22 | 1577 | "%s.blktrace.%d", pdi->name, pci->cpu); |
e7c9f3ff NS |
1578 | if (stat(pci->fname, &st) < 0) |
1579 | break; | |
74feaeb5 AB |
1580 | if (st.st_size) { |
1581 | pci->fd = open(pci->fname, O_RDONLY); | |
1582 | if (pci->fd < 0) { | |
1583 | perror(pci->fname); | |
1584 | continue; | |
1585 | } | |
e7c9f3ff NS |
1586 | } |
1587 | ||
7d747d22 | 1588 | printf("Input file %s added\n", pci->fname); |
73877e12 | 1589 | pdi->nfiles++; |
774a1a10 | 1590 | pdi->nopenfiles++; |
d0ca268b | 1591 | } |
d5396421 JA |
1592 | } |
1593 | ||
7d747d22 JA |
1594 | /* |
1595 | * now loop over the files reading in the data | |
1596 | */ | |
412819ce | 1597 | do { |
53c68c88 JA |
1598 | unsigned long long youngest; |
1599 | ||
7d747d22 | 1600 | events_added = 0; |
287fa3d6 | 1601 | last_allowed_time = -1ULL; |
1c7c54aa | 1602 | smallest_seq_read = -1U; |
d5396421 | 1603 | |
7d747d22 | 1604 | for (i = 0; i < ndevices; i++) { |
73877e12 NS |
1605 | pdi = &devices[i]; |
1606 | ||
1607 | for (j = 0; j < pdi->nfiles; j++) { | |
d5396421 | 1608 | |
73877e12 | 1609 | pci = get_cpu_info(pdi, j); |
d5396421 | 1610 | |
7d747d22 JA |
1611 | if (pci->fd == -1) |
1612 | continue; | |
51128a28 | 1613 | |
a649216c | 1614 | events = read_events(pci->fd, 1); |
7d747d22 JA |
1615 | if (!events) { |
1616 | close(pci->fd); | |
1617 | pci->fd = -1; | |
774a1a10 | 1618 | pdi->nopenfiles--; |
7d747d22 JA |
1619 | continue; |
1620 | } | |
d5396421 | 1621 | |
287fa3d6 JA |
1622 | if (pdi->last_read_time < last_allowed_time) |
1623 | last_allowed_time = pdi->last_read_time; | |
d5396421 | 1624 | |
7d747d22 JA |
1625 | events_added += events; |
1626 | } | |
2ff323b0 | 1627 | } |
d5396421 | 1628 | |
53c68c88 JA |
1629 | if (sort_entries(&youngest)) |
1630 | break; | |
1631 | ||
1632 | if (youngest > stopwatch_end) | |
287fa3d6 JA |
1633 | break; |
1634 | ||
a649216c | 1635 | show_entries_rb(0); |
cb2a1a62 | 1636 | |
7d747d22 | 1637 | } while (events_added); |
d5396421 | 1638 | |
a649216c JA |
1639 | if (rb_sort_entries) |
1640 | show_entries_rb(1); | |
1641 | ||
7d747d22 | 1642 | return 0; |
412819ce | 1643 | } |
d5396421 | 1644 | |
412819ce JA |
1645 | static int do_stdin(void) |
1646 | { | |
53c68c88 | 1647 | unsigned long long youngest; |
763d936e | 1648 | int fd, events; |
d5396421 | 1649 | |
be925321 | 1650 | last_allowed_time = -1ULL; |
1f79c4a0 | 1651 | fd = dup(STDIN_FILENO); |
0b07f23e JA |
1652 | if (fd == -1) { |
1653 | perror("dup stdin"); | |
1654 | return -1; | |
1655 | } | |
d5396421 | 1656 | |
0b07f23e | 1657 | while ((events = read_events(fd, 0)) != 0) { |
412819ce | 1658 | |
0b07f23e JA |
1659 | smallest_seq_read = -1U; |
1660 | ||
53c68c88 JA |
1661 | if (sort_entries(&youngest)) |
1662 | break; | |
1663 | ||
1664 | if (youngest > stopwatch_end) | |
2ff323b0 JA |
1665 | break; |
1666 | ||
763d936e | 1667 | show_entries_rb(0); |
0b07f23e | 1668 | } |
d5396421 | 1669 | |
a649216c JA |
1670 | if (rb_sort_entries) |
1671 | show_entries_rb(1); | |
1672 | ||
d5396421 | 1673 | close(fd); |
d5396421 JA |
1674 | return 0; |
1675 | } | |
d0ca268b | 1676 | |
cbc927b6 | 1677 | static void show_stats(void) |
412819ce | 1678 | { |
cbc927b6 JA |
1679 | if (!ofp) |
1680 | return; | |
1681 | if (stats_printed) | |
1682 | return; | |
1683 | ||
1684 | stats_printed = 1; | |
1685 | ||
1686 | if (per_process_stats) | |
1687 | show_process_stats(); | |
1688 | ||
1689 | if (per_device_and_cpu_stats) | |
1690 | show_device_and_cpu_stats(); | |
1691 | ||
152f6476 | 1692 | fflush(ofp); |
412819ce JA |
1693 | } |
1694 | ||
e820abd7 | 1695 | static void handle_sigint(__attribute__((__unused__)) int sig) |
412819ce JA |
1696 | { |
1697 | done = 1; | |
cbc927b6 | 1698 | show_stats(); |
412819ce JA |
1699 | } |
1700 | ||
46e6968b NS |
1701 | /* |
1702 | * Extract start and duration times from a string, allowing | |
1703 | * us to specify a time interval of interest within a trace. | |
1704 | * Format: "duration" (start is zero) or "start:duration". | |
1705 | */ | |
1706 | static int find_stopwatch_interval(char *string) | |
1707 | { | |
1708 | double value; | |
1709 | char *sp; | |
1710 | ||
1711 | value = strtod(string, &sp); | |
1712 | if (sp == string) { | |
1713 | fprintf(stderr,"Invalid stopwatch timer: %s\n", string); | |
1714 | return 1; | |
1715 | } | |
1716 | if (*sp == ':') { | |
1717 | stopwatch_start = DOUBLE_TO_NANO_ULL(value); | |
1718 | string = sp + 1; | |
1719 | value = strtod(string, &sp); | |
1720 | if (sp == string || *sp != '\0') { | |
1721 | fprintf(stderr,"Invalid stopwatch duration time: %s\n", | |
1722 | string); | |
1723 | return 1; | |
1724 | } | |
1725 | } else if (*sp != '\0') { | |
1726 | fprintf(stderr,"Invalid stopwatch start timer: %s\n", string); | |
1727 | return 1; | |
1728 | } | |
1b928247 JA |
1729 | stopwatch_end = DOUBLE_TO_NANO_ULL(value); |
1730 | if (stopwatch_end <= stopwatch_start) { | |
1731 | fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n", | |
1732 | stopwatch_start, stopwatch_end); | |
1733 | return 1; | |
1734 | } | |
1735 | ||
46e6968b NS |
1736 | return 0; |
1737 | } | |
1738 | ||
52724a0e JA |
1739 | static char usage_str[] = \ |
1740 | "[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \ | |
1741 | "[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \ | |
1742 | "\t-i Input file containing trace data, or '-' for stdin\n" \ | |
d1d7f15f | 1743 | "\t-D Directory to prepend to input file names\n" \ |
52724a0e JA |
1744 | "\t-o Output file. If not given, output is stdout\n" \ |
1745 | "\t-b stdin read batching\n" \ | |
1746 | "\t-s Show per-program io statistics\n" \ | |
bf0720af | 1747 | "\t-n Hash processes by name, not pid\n" \ |
52724a0e JA |
1748 | "\t-t Track individual ios. Will tell you the time a request took\n" \ |
1749 | "\t to get queued, to get dispatched, and to get completed\n" \ | |
1750 | "\t-q Quiet. Don't display any stats at the end of the trace\n" \ | |
1751 | "\t-w Only parse data between the given time interval in seconds.\n" \ | |
1752 | "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \ | |
d915dee6 JA |
1753 | "\t-f Output format. Customize the output format. The format field\n" \ |
1754 | "\t identifies can be found in the documentation\n" \ | |
52724a0e | 1755 | "\t-F Format specification. Can be found in the documentation\n" \ |
57ea8602 JA |
1756 | "\t-v More verbose for marginal errors\n" \ |
1757 | "\t-V Print program version info\n\n"; | |
52724a0e | 1758 | |
1f79c4a0 JA |
1759 | static void usage(char *prog) |
1760 | { | |
52724a0e | 1761 | fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str); |
1f79c4a0 JA |
1762 | } |
1763 | ||
d5396421 JA |
1764 | int main(int argc, char *argv[]) |
1765 | { | |
152f6476 | 1766 | char *ofp_buffer; |
98f8386b | 1767 | int i, c, ret, mode; |
98f8386b | 1768 | int act_mask_tmp = 0; |
d5396421 JA |
1769 | |
1770 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) { | |
1771 | switch (c) { | |
98f8386b AB |
1772 | case 'a': |
1773 | i = find_mask_map(optarg); | |
1774 | if (i < 0) { | |
1775 | fprintf(stderr,"Invalid action mask %s\n", | |
1776 | optarg); | |
1777 | return 1; | |
1778 | } | |
1779 | act_mask_tmp |= i; | |
1780 | break; | |
1781 | ||
1782 | case 'A': | |
1783 | if ((sscanf(optarg, "%x", &i) != 1) || | |
1784 | !valid_act_opt(i)) { | |
1785 | fprintf(stderr, | |
1786 | "Invalid set action mask %s/0x%x\n", | |
1787 | optarg, i); | |
1788 | return 1; | |
1789 | } | |
1790 | act_mask_tmp = i; | |
1791 | break; | |
d5396421 | 1792 | case 'i': |
e7c9f3ff NS |
1793 | if (!strcmp(optarg, "-") && !pipeline) |
1794 | pipeline = 1; | |
1795 | else if (resize_devices(optarg) != 0) | |
1796 | return 1; | |
d5396421 | 1797 | break; |
d1d7f15f JA |
1798 | case 'D': |
1799 | input_dir = optarg; | |
1800 | break; | |
d5396421 | 1801 | case 'o': |
66efebf8 | 1802 | output_name = optarg; |
d5396421 | 1803 | break; |
79f19470 JA |
1804 | case 'b': |
1805 | rb_batch = atoi(optarg); | |
1806 | if (rb_batch <= 0) | |
1807 | rb_batch = RB_BATCH_DEFAULT; | |
1808 | break; | |
152f6476 JA |
1809 | case 's': |
1810 | per_process_stats = 1; | |
1811 | break; | |
7997c5b0 JA |
1812 | case 't': |
1813 | track_ios = 1; | |
1814 | break; | |
1e1c60f1 NS |
1815 | case 'q': |
1816 | per_device_and_cpu_stats = 0; | |
1817 | break; | |
46e6968b NS |
1818 | case 'w': |
1819 | if (find_stopwatch_interval(optarg) != 0) | |
1820 | return 1; | |
1821 | break; | |
ab197ca7 AB |
1822 | case 'f': |
1823 | set_all_format_specs(optarg); | |
1824 | break; | |
1825 | case 'F': | |
1826 | if (add_format_spec(optarg) != 0) | |
1827 | return 1; | |
1828 | break; | |
d915dee6 | 1829 | case 'h': |
715d8021 | 1830 | ppi_hash_by_pid = 0; |
bf0720af | 1831 | break; |
52724a0e | 1832 | case 'v': |
57ea8602 JA |
1833 | verbose++; |
1834 | break; | |
1835 | case 'V': | |
52724a0e JA |
1836 | printf("%s version %s\n", argv[0], blkparse_version); |
1837 | return 0; | |
d5396421 | 1838 | default: |
1f79c4a0 | 1839 | usage(argv[0]); |
d5396421 JA |
1840 | return 1; |
1841 | } | |
d0ca268b JA |
1842 | } |
1843 | ||
e7c9f3ff NS |
1844 | while (optind < argc) { |
1845 | if (!strcmp(argv[optind], "-") && !pipeline) | |
1846 | pipeline = 1; | |
1847 | else if (resize_devices(argv[optind]) != 0) | |
1848 | return 1; | |
1849 | optind++; | |
1850 | } | |
1851 | ||
1852 | if (!pipeline && !ndevices) { | |
1f79c4a0 | 1853 | usage(argv[0]); |
d5396421 JA |
1854 | return 1; |
1855 | } | |
1856 | ||
98f8386b AB |
1857 | if (act_mask_tmp != 0) |
1858 | act_mask = act_mask_tmp; | |
1859 | ||
7997c5b0 | 1860 | memset(&rb_sort_root, 0, sizeof(rb_sort_root)); |
412819ce JA |
1861 | |
1862 | signal(SIGINT, handle_sigint); | |
1863 | signal(SIGHUP, handle_sigint); | |
1864 | signal(SIGTERM, handle_sigint); | |
d5396421 | 1865 | |
d69db225 JA |
1866 | setlocale(LC_NUMERIC, "en_US"); |
1867 | ||
a66877e6 | 1868 | if (!output_name) { |
152f6476 | 1869 | ofp = fdopen(STDOUT_FILENO, "w"); |
a66877e6 JA |
1870 | mode = _IOLBF; |
1871 | } else { | |
152f6476 JA |
1872 | char ofname[128]; |
1873 | ||
7a9690c0 | 1874 | snprintf(ofname, sizeof(ofname) - 1, "%s", output_name); |
152f6476 | 1875 | ofp = fopen(ofname, "w"); |
a66877e6 | 1876 | mode = _IOFBF; |
152f6476 JA |
1877 | } |
1878 | ||
1879 | if (!ofp) { | |
1880 | perror("fopen"); | |
1881 | return 1; | |
1882 | } | |
1883 | ||
1884 | ofp_buffer = malloc(4096); | |
a66877e6 | 1885 | if (setvbuf(ofp, ofp_buffer, mode, 4096)) { |
152f6476 JA |
1886 | perror("setvbuf"); |
1887 | return 1; | |
1888 | } | |
1889 | ||
e7c9f3ff | 1890 | if (pipeline) |
d5396421 JA |
1891 | ret = do_stdin(); |
1892 | else | |
1893 | ret = do_file(); | |
1894 | ||
cbc927b6 | 1895 | show_stats(); |
d5396421 | 1896 | return ret; |
d0ca268b | 1897 | } |