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; |
e7c9f3ff NS |
58 | int ncpus; |
59 | struct per_cpu_info *cpus; | |
60 | }; | |
61 | ||
152f6476 JA |
62 | struct per_process_info { |
63 | char name[16]; | |
64 | __u32 pid; | |
65 | struct io_stats io_stats; | |
66 | struct per_process_info *hash_next, *list_next; | |
715d8021 | 67 | int more_than_one; |
50adc0ba JA |
68 | |
69 | /* | |
70 | * individual io stats | |
71 | */ | |
b9d40d6f JA |
72 | unsigned long long longest_allocation_wait[2]; |
73 | unsigned long long longest_dispatch_wait[2]; | |
74 | unsigned long long longest_completion_wait[2]; | |
d0ca268b JA |
75 | }; |
76 | ||
152f6476 | 77 | #define PPI_HASH_SHIFT (8) |
bf0720af JA |
78 | #define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT) |
79 | #define PPI_HASH_MASK (PPI_HASH_SIZE - 1) | |
80 | static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE]; | |
152f6476 | 81 | static struct per_process_info *ppi_list; |
886ecf0e | 82 | static int ppi_list_entries; |
152f6476 | 83 | |
7d1c0411 | 84 | #define S_OPTS "i:o:b:stqw:f:F:vnm" |
d5396421 JA |
85 | static struct option l_opts[] = { |
86 | { | |
87 | .name = "input", | |
428683db | 88 | .has_arg = required_argument, |
d5396421 JA |
89 | .flag = NULL, |
90 | .val = 'i' | |
91 | }, | |
92 | { | |
93 | .name = "output", | |
428683db | 94 | .has_arg = required_argument, |
d5396421 JA |
95 | .flag = NULL, |
96 | .val = 'o' | |
97 | }, | |
79f19470 JA |
98 | { |
99 | .name = "batch", | |
428683db | 100 | .has_arg = required_argument, |
79f19470 JA |
101 | .flag = NULL, |
102 | .val = 'b' | |
103 | }, | |
152f6476 JA |
104 | { |
105 | .name = "per program stats", | |
428683db | 106 | .has_arg = no_argument, |
152f6476 JA |
107 | .flag = NULL, |
108 | .val = 's' | |
109 | }, | |
7997c5b0 JA |
110 | { |
111 | .name = "track ios", | |
428683db | 112 | .has_arg = no_argument, |
7997c5b0 JA |
113 | .flag = NULL, |
114 | .val = 't' | |
115 | }, | |
1e1c60f1 NS |
116 | { |
117 | .name = "quiet", | |
428683db | 118 | .has_arg = no_argument, |
1e1c60f1 NS |
119 | .flag = NULL, |
120 | .val = 'q' | |
121 | }, | |
46e6968b NS |
122 | { |
123 | .name = "stopwatch", | |
428683db | 124 | .has_arg = required_argument, |
46e6968b NS |
125 | .flag = NULL, |
126 | .val = 'w' | |
127 | }, | |
ab197ca7 AB |
128 | { |
129 | .name = "format", | |
428683db | 130 | .has_arg = required_argument, |
ab197ca7 AB |
131 | .flag = NULL, |
132 | .val = 'f' | |
133 | }, | |
134 | { | |
135 | .name = "format-spec", | |
428683db | 136 | .has_arg = required_argument, |
ab197ca7 AB |
137 | .flag = NULL, |
138 | .val = 'F' | |
139 | }, | |
bf0720af JA |
140 | { |
141 | .name = "hash by name", | |
142 | .has_arg = no_argument, | |
143 | .flag = NULL, | |
144 | .val = 'n' | |
145 | }, | |
7d1c0411 JA |
146 | { |
147 | .name = "missing", | |
148 | .has_arg = no_argument, | |
149 | .flag = NULL, | |
150 | .val = 'm' | |
151 | }, | |
52724a0e JA |
152 | { |
153 | .name = "version", | |
154 | .has_arg = no_argument, | |
155 | .flag = NULL, | |
156 | .val = 'v' | |
157 | }, | |
d5396421 JA |
158 | }; |
159 | ||
7997c5b0 JA |
160 | /* |
161 | * for sorting the displayed output | |
162 | */ | |
8fc0abbc JA |
163 | struct trace { |
164 | struct blk_io_trace *bit; | |
165 | struct rb_node rb_node; | |
cb2a1a62 | 166 | struct trace *next; |
8fc0abbc JA |
167 | }; |
168 | ||
cb2a1a62 | 169 | static struct rb_root rb_sort_root; |
a649216c JA |
170 | static unsigned long rb_sort_entries; |
171 | ||
cb2a1a62 JA |
172 | static struct trace *trace_list; |
173 | ||
d36421e4 JA |
174 | /* |
175 | * allocation cache | |
176 | */ | |
177 | static struct blk_io_trace *bit_alloc_list; | |
178 | static struct trace *t_alloc_list; | |
179 | ||
7997c5b0 JA |
180 | /* |
181 | * for tracking individual ios | |
182 | */ | |
183 | struct io_track { | |
184 | struct rb_node rb_node; | |
185 | ||
186 | __u64 sector; | |
187 | __u32 pid; | |
bf0720af | 188 | char comm[16]; |
95c15013 | 189 | unsigned long long allocation_time; |
7997c5b0 JA |
190 | unsigned long long queue_time; |
191 | unsigned long long dispatch_time; | |
192 | unsigned long long completion_time; | |
193 | }; | |
194 | ||
e7c9f3ff NS |
195 | static int ndevices; |
196 | static struct per_dev_info *devices; | |
197 | static char *get_dev_name(struct per_dev_info *, char *, int); | |
d0ca268b | 198 | |
71d5d4c9 | 199 | FILE *ofp = NULL; |
e7c9f3ff NS |
200 | static char *output_name; |
201 | ||
202 | static unsigned long long genesis_time; | |
287fa3d6 | 203 | static unsigned long long last_allowed_time; |
1c7c54aa | 204 | static unsigned int smallest_seq_read; |
46e6968b NS |
205 | static unsigned long long stopwatch_start; /* start from zero by default */ |
206 | static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */ | |
152f6476 JA |
207 | |
208 | static int per_process_stats; | |
7997c5b0 | 209 | static int track_ios; |
bf0720af | 210 | static int ppi_hash_by_pid = 1; |
7d1c0411 | 211 | static int print_missing; |
d0ca268b | 212 | |
1d24fc14 JA |
213 | static unsigned int t_alloc_cache; |
214 | static unsigned int bit_alloc_cache; | |
215 | ||
7d747d22 | 216 | #define RB_BATCH_DEFAULT (512) |
e820abd7 | 217 | static unsigned int rb_batch = RB_BATCH_DEFAULT; |
79f19470 | 218 | |
e7c9f3ff NS |
219 | static int pipeline; |
220 | ||
412819ce JA |
221 | #define is_done() (*(volatile int *)(&done)) |
222 | static volatile int done; | |
223 | ||
bf0720af JA |
224 | #define JHASH_RANDOM (0x3af5f2ee) |
225 | ||
226 | static inline int ppi_hash_pid(__u32 pid) | |
227 | { | |
228 | return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK; | |
229 | } | |
230 | ||
231 | static inline int ppi_hash_name(const char *name) | |
152f6476 | 232 | { |
bf0720af JA |
233 | return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK; |
234 | } | |
235 | ||
236 | static inline int ppi_hash(struct per_process_info *ppi) | |
237 | { | |
238 | if (ppi_hash_by_pid) | |
239 | return ppi_hash_pid(ppi->pid); | |
240 | ||
bf0720af | 241 | return ppi_hash_name(ppi->name); |
152f6476 JA |
242 | } |
243 | ||
244 | static inline void add_process_to_hash(struct per_process_info *ppi) | |
245 | { | |
bf0720af | 246 | const int hash_idx = ppi_hash(ppi); |
152f6476 | 247 | |
bf0720af JA |
248 | ppi->hash_next = ppi_hash_table[hash_idx]; |
249 | ppi_hash_table[hash_idx] = ppi; | |
152f6476 JA |
250 | } |
251 | ||
252 | static inline void add_process_to_list(struct per_process_info *ppi) | |
253 | { | |
254 | ppi->list_next = ppi_list; | |
255 | ppi_list = ppi; | |
886ecf0e | 256 | ppi_list_entries++; |
152f6476 JA |
257 | } |
258 | ||
bf0720af JA |
259 | static struct per_process_info *find_process_by_name(char *name) |
260 | { | |
261 | const int hash_idx = ppi_hash_name(name); | |
262 | struct per_process_info *ppi; | |
263 | ||
264 | ppi = ppi_hash_table[hash_idx]; | |
265 | while (ppi) { | |
266 | if (!strcmp(ppi->name, name)) | |
267 | return ppi; | |
268 | ||
269 | ppi = ppi->hash_next; | |
270 | } | |
271 | ||
272 | return NULL; | |
273 | } | |
274 | ||
152f6476 JA |
275 | static struct per_process_info *find_process_by_pid(__u32 pid) |
276 | { | |
bf0720af | 277 | const int hash_idx = ppi_hash_pid(pid); |
152f6476 JA |
278 | struct per_process_info *ppi; |
279 | ||
bf0720af | 280 | ppi = ppi_hash_table[hash_idx]; |
152f6476 JA |
281 | while (ppi) { |
282 | if (ppi->pid == pid) | |
283 | return ppi; | |
284 | ||
285 | ppi = ppi->hash_next; | |
286 | } | |
287 | ||
288 | return NULL; | |
289 | } | |
290 | ||
bf0720af JA |
291 | static struct per_process_info *find_process(__u32 pid, char *name) |
292 | { | |
715d8021 JA |
293 | struct per_process_info *ppi; |
294 | ||
bf0720af JA |
295 | if (ppi_hash_by_pid) |
296 | return find_process_by_pid(pid); | |
297 | ||
715d8021 JA |
298 | ppi = find_process_by_name(name); |
299 | if (ppi && ppi->pid != pid) | |
300 | ppi->more_than_one = 1; | |
301 | ||
302 | return ppi; | |
bf0720af JA |
303 | } |
304 | ||
23211fdf JA |
305 | static inline int trace_rb_insert(struct trace *t, struct rb_root *root, |
306 | int check_time) | |
7997c5b0 | 307 | { |
2a1b3424 | 308 | struct rb_node **p = &root->rb_node; |
7997c5b0 JA |
309 | struct rb_node *parent = NULL; |
310 | struct trace *__t; | |
311 | ||
312 | while (*p) { | |
313 | parent = *p; | |
2a1b3424 | 314 | |
7997c5b0 JA |
315 | __t = rb_entry(parent, struct trace, rb_node); |
316 | ||
23211fdf JA |
317 | if (check_time) { |
318 | if (t->bit->time < __t->bit->time) { | |
319 | p = &(*p)->rb_left; | |
320 | continue; | |
321 | } else if (t->bit->time > __t->bit->time) { | |
322 | p = &(*p)->rb_right; | |
323 | continue; | |
324 | } | |
325 | } | |
326 | if (t->bit->device < __t->bit->device) | |
e7c9f3ff NS |
327 | p = &(*p)->rb_left; |
328 | else if (t->bit->device > __t->bit->device) | |
329 | p = &(*p)->rb_right; | |
dcf0f7ed JA |
330 | else if (t->bit->sequence < __t->bit->sequence) |
331 | p = &(*p)->rb_left; | |
332 | else if (t->bit->sequence > __t->bit->sequence) | |
333 | p = &(*p)->rb_right; | |
e7c9f3ff NS |
334 | else if (t->bit->device == __t->bit->device) { |
335 | fprintf(stderr, | |
336 | "sequence alias (%d) on device %d,%d!\n", | |
337 | t->bit->sequence, | |
338 | MAJOR(t->bit->device), MINOR(t->bit->device)); | |
7997c5b0 JA |
339 | return 1; |
340 | } | |
341 | } | |
342 | ||
343 | rb_link_node(&t->rb_node, parent, p); | |
2a1b3424 | 344 | rb_insert_color(&t->rb_node, root); |
7997c5b0 JA |
345 | return 0; |
346 | } | |
347 | ||
2a1b3424 | 348 | static inline int trace_rb_insert_sort(struct trace *t) |
e3556946 | 349 | { |
23211fdf | 350 | if (!trace_rb_insert(t, &rb_sort_root, 1)) { |
2a1b3424 JA |
351 | rb_sort_entries++; |
352 | return 0; | |
353 | } | |
354 | ||
355 | return 1; | |
356 | } | |
357 | ||
358 | static inline int trace_rb_insert_last(struct per_dev_info *pdi,struct trace *t) | |
359 | { | |
23211fdf | 360 | if (!trace_rb_insert(t, &pdi->rb_last, 1)) { |
2a1b3424 JA |
361 | pdi->rb_last_entries++; |
362 | return 0; | |
363 | } | |
364 | ||
365 | return 1; | |
366 | } | |
367 | ||
368 | static struct trace *trace_rb_find(dev_t device, unsigned long sequence, | |
369 | struct rb_root *root, int order) | |
370 | { | |
371 | struct rb_node *n = root->rb_node; | |
372 | struct rb_node *prev = NULL; | |
e3556946 JA |
373 | struct trace *__t; |
374 | ||
2a1b3424 JA |
375 | while (n) { |
376 | __t = rb_entry(n, struct trace, rb_node); | |
377 | prev = n; | |
e3556946 | 378 | |
0583b6a2 | 379 | if (device < __t->bit->device) |
2a1b3424 | 380 | n = n->rb_left; |
0583b6a2 | 381 | else if (device > __t->bit->device) |
2a1b3424 | 382 | n = n->rb_right; |
0583b6a2 | 383 | else if (sequence < __t->bit->sequence) |
2a1b3424 | 384 | n = n->rb_left; |
e3556946 | 385 | else if (sequence > __t->bit->sequence) |
2a1b3424 | 386 | n = n->rb_right; |
e3556946 JA |
387 | else |
388 | return __t; | |
389 | } | |
390 | ||
2a1b3424 JA |
391 | /* |
392 | * hack - the list may not be sequence ordered because some | |
393 | * events don't have sequence and time matched. so we end up | |
394 | * being a little off in the rb lookup here, because we don't | |
395 | * know the time we are looking for. compensate by browsing | |
396 | * a little ahead from the last entry to find the match | |
397 | */ | |
398 | if (order && prev) { | |
399 | int max = 5; | |
400 | ||
401 | while (((n = rb_next(prev)) != NULL) && max--) { | |
402 | __t = rb_entry(n, struct trace, rb_node); | |
403 | ||
404 | if (__t->bit->device == device && | |
405 | __t->bit->sequence == sequence) | |
406 | return __t; | |
407 | ||
408 | prev = n; | |
409 | } | |
410 | } | |
411 | ||
e3556946 JA |
412 | return NULL; |
413 | } | |
414 | ||
2a1b3424 JA |
415 | static inline struct trace *trace_rb_find_sort(dev_t dev, unsigned long seq) |
416 | { | |
417 | return trace_rb_find(dev, seq, &rb_sort_root, 1); | |
418 | } | |
419 | ||
420 | static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi, | |
421 | unsigned long seq) | |
422 | { | |
f7bd1a9b | 423 | return trace_rb_find(pdi->dev, seq, &pdi->rb_last, 0); |
2a1b3424 JA |
424 | } |
425 | ||
f7bd1a9b | 426 | static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot) |
7997c5b0 | 427 | { |
f7bd1a9b | 428 | struct rb_node **p = &pdi->rb_track.rb_node; |
7997c5b0 JA |
429 | struct rb_node *parent = NULL; |
430 | struct io_track *__iot; | |
431 | ||
432 | while (*p) { | |
433 | parent = *p; | |
7997c5b0 JA |
434 | __iot = rb_entry(parent, struct io_track, rb_node); |
435 | ||
f7bd1a9b | 436 | if (iot->sector < __iot->sector) |
7997c5b0 JA |
437 | p = &(*p)->rb_left; |
438 | else if (iot->sector > __iot->sector) | |
439 | p = &(*p)->rb_right; | |
440 | else { | |
e7c9f3ff | 441 | fprintf(stderr, |
ab197ca7 AB |
442 | "sector alias (%Lu) on device %d,%d!\n", |
443 | (unsigned long long) iot->sector, | |
f7bd1a9b | 444 | MAJOR(pdi->dev), MINOR(pdi->dev)); |
7997c5b0 JA |
445 | return 1; |
446 | } | |
447 | } | |
448 | ||
449 | rb_link_node(&iot->rb_node, parent, p); | |
f7bd1a9b | 450 | rb_insert_color(&iot->rb_node, &pdi->rb_track); |
7997c5b0 JA |
451 | return 0; |
452 | } | |
453 | ||
f7bd1a9b | 454 | static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector) |
7997c5b0 | 455 | { |
f7bd1a9b | 456 | struct rb_node *n = pdi->rb_track.rb_node; |
7997c5b0 JA |
457 | struct io_track *__iot; |
458 | ||
2a1b3424 JA |
459 | while (n) { |
460 | __iot = rb_entry(n, struct io_track, rb_node); | |
7997c5b0 | 461 | |
f7bd1a9b | 462 | if (sector < __iot->sector) |
2a1b3424 | 463 | n = n->rb_left; |
7997c5b0 | 464 | else if (sector > __iot->sector) |
2a1b3424 | 465 | n = n->rb_right; |
7997c5b0 JA |
466 | else |
467 | return __iot; | |
468 | } | |
469 | ||
470 | return NULL; | |
471 | } | |
472 | ||
f7bd1a9b JA |
473 | static struct io_track *find_track(struct per_dev_info *pdi, __u32 pid, |
474 | char *comm, __u64 sector) | |
7997c5b0 | 475 | { |
916b5501 | 476 | struct io_track *iot; |
7997c5b0 | 477 | |
f7bd1a9b | 478 | iot = __find_track(pdi, sector); |
7997c5b0 JA |
479 | if (!iot) { |
480 | iot = malloc(sizeof(*iot)); | |
50adc0ba | 481 | iot->pid = pid; |
bf0720af | 482 | memcpy(iot->comm, comm, sizeof(iot->comm)); |
7997c5b0 | 483 | iot->sector = sector; |
f7bd1a9b | 484 | track_rb_insert(pdi, iot); |
7997c5b0 JA |
485 | } |
486 | ||
487 | return iot; | |
488 | } | |
489 | ||
f7bd1a9b JA |
490 | static void log_track_frontmerge(struct per_dev_info *pdi, |
491 | struct blk_io_trace *t) | |
2e3e8ded JA |
492 | { |
493 | struct io_track *iot; | |
494 | ||
495 | if (!track_ios) | |
496 | return; | |
2e3e8ded | 497 | |
f7bd1a9b | 498 | iot = __find_track(pdi, t->sector + (t->bytes >> 9)); |
cb2a1a62 | 499 | if (!iot) { |
acd70d21 | 500 | fprintf(stderr, "merge not found for (%d,%d): %llu\n", |
f7bd1a9b | 501 | MAJOR(pdi->dev), MINOR(pdi->dev), |
85e84946 | 502 | (unsigned long long) t->sector + (t->bytes >> 9)); |
cb2a1a62 | 503 | return; |
2e3e8ded | 504 | } |
cb2a1a62 | 505 | |
f7bd1a9b | 506 | rb_erase(&iot->rb_node, &pdi->rb_track); |
a01516de | 507 | iot->sector -= t->bytes >> 9; |
f7bd1a9b | 508 | track_rb_insert(pdi, iot); |
2e3e8ded JA |
509 | } |
510 | ||
f7bd1a9b | 511 | static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t) |
2e3e8ded JA |
512 | { |
513 | struct io_track *iot; | |
514 | ||
515 | if (!track_ios) | |
516 | return; | |
517 | ||
f7bd1a9b | 518 | iot = find_track(pdi, t->pid, t->comm, t->sector); |
95c15013 JA |
519 | iot->allocation_time = t->time; |
520 | } | |
521 | ||
95c15013 | 522 | /* |
b6076a9b | 523 | * return time between rq allocation and insertion |
95c15013 | 524 | */ |
f7bd1a9b JA |
525 | static unsigned long long log_track_insert(struct per_dev_info *pdi, |
526 | struct blk_io_trace *t) | |
95c15013 | 527 | { |
50adc0ba | 528 | unsigned long long elapsed; |
95c15013 JA |
529 | struct io_track *iot; |
530 | ||
531 | if (!track_ios) | |
532 | return -1; | |
533 | ||
f7bd1a9b | 534 | iot = find_track(pdi, t->pid, t->comm, t->sector); |
2e3e8ded | 535 | iot->queue_time = t->time; |
acd70d21 JA |
536 | |
537 | if (!iot->allocation_time) | |
538 | return -1; | |
539 | ||
50adc0ba JA |
540 | elapsed = iot->queue_time - iot->allocation_time; |
541 | ||
542 | if (per_process_stats) { | |
bf0720af | 543 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 544 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 545 | |
b9d40d6f JA |
546 | if (ppi && elapsed > ppi->longest_allocation_wait[w]) |
547 | ppi->longest_allocation_wait[w] = elapsed; | |
50adc0ba JA |
548 | } |
549 | ||
550 | return elapsed; | |
2e3e8ded JA |
551 | } |
552 | ||
553 | /* | |
554 | * return time between queue and issue | |
555 | */ | |
f7bd1a9b JA |
556 | static unsigned long long log_track_issue(struct per_dev_info *pdi, |
557 | struct blk_io_trace *t) | |
2e3e8ded | 558 | { |
50adc0ba | 559 | unsigned long long elapsed; |
2e3e8ded JA |
560 | struct io_track *iot; |
561 | ||
562 | if (!track_ios) | |
563 | return -1; | |
564 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
565 | return -1; | |
566 | ||
f7bd1a9b | 567 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 568 | if (!iot) { |
acd70d21 | 569 | fprintf(stderr, "issue not found for (%d,%d): %llu\n", |
85e84946 NS |
570 | MAJOR(pdi->dev), MINOR(pdi->dev), |
571 | (unsigned long long) t->sector); | |
2e3e8ded | 572 | return -1; |
cb2a1a62 | 573 | } |
2e3e8ded JA |
574 | |
575 | iot->dispatch_time = t->time; | |
50adc0ba JA |
576 | elapsed = iot->dispatch_time - iot->queue_time; |
577 | ||
578 | if (per_process_stats) { | |
bf0720af | 579 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 580 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 581 | |
b9d40d6f JA |
582 | if (ppi && elapsed > ppi->longest_dispatch_wait[w]) |
583 | ppi->longest_dispatch_wait[w] = elapsed; | |
50adc0ba JA |
584 | } |
585 | ||
586 | return elapsed; | |
2e3e8ded JA |
587 | } |
588 | ||
589 | /* | |
590 | * return time between dispatch and complete | |
591 | */ | |
f7bd1a9b JA |
592 | static unsigned long long log_track_complete(struct per_dev_info *pdi, |
593 | struct blk_io_trace *t) | |
2e3e8ded JA |
594 | { |
595 | unsigned long long elapsed; | |
596 | struct io_track *iot; | |
597 | ||
598 | if (!track_ios) | |
599 | return -1; | |
600 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
601 | return -1; | |
602 | ||
f7bd1a9b | 603 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 604 | if (!iot) { |
acd70d21 | 605 | fprintf(stderr, "complete not found for (%d,%d): %llu\n", |
85e84946 NS |
606 | MAJOR(pdi->dev), MINOR(pdi->dev), |
607 | (unsigned long long) t->sector); | |
2e3e8ded | 608 | return -1; |
cb2a1a62 | 609 | } |
2e3e8ded JA |
610 | |
611 | iot->completion_time = t->time; | |
612 | elapsed = iot->completion_time - iot->dispatch_time; | |
613 | ||
50adc0ba | 614 | if (per_process_stats) { |
bf0720af | 615 | struct per_process_info *ppi = find_process(iot->pid,iot->comm); |
b9d40d6f | 616 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 617 | |
b9d40d6f JA |
618 | if (ppi && elapsed > ppi->longest_completion_wait[w]) |
619 | ppi->longest_completion_wait[w] = elapsed; | |
50adc0ba JA |
620 | } |
621 | ||
2e3e8ded JA |
622 | /* |
623 | * kill the trace, we don't need it after completion | |
624 | */ | |
f7bd1a9b | 625 | rb_erase(&iot->rb_node, &pdi->rb_track); |
2e3e8ded JA |
626 | free(iot); |
627 | ||
628 | return elapsed; | |
629 | } | |
630 | ||
631 | ||
152f6476 JA |
632 | static struct io_stats *find_process_io_stats(__u32 pid, char *name) |
633 | { | |
bf0720af | 634 | struct per_process_info *ppi = find_process(pid, name); |
152f6476 JA |
635 | |
636 | if (!ppi) { | |
637 | ppi = malloc(sizeof(*ppi)); | |
638 | memset(ppi, 0, sizeof(*ppi)); | |
bf0720af | 639 | memcpy(ppi->name, name, 16); |
152f6476 JA |
640 | ppi->pid = pid; |
641 | add_process_to_hash(ppi); | |
642 | add_process_to_list(ppi); | |
643 | } | |
644 | ||
645 | return &ppi->io_stats; | |
646 | } | |
647 | ||
e7c9f3ff | 648 | static void resize_cpu_info(struct per_dev_info *pdi, int cpu) |
a718bd37 | 649 | { |
e7c9f3ff NS |
650 | struct per_cpu_info *cpus = pdi->cpus; |
651 | int ncpus = pdi->ncpus; | |
652 | int new_count = cpu + 1; | |
653 | int new_space, size; | |
a718bd37 NS |
654 | char *new_start; |
655 | ||
e7c9f3ff NS |
656 | size = new_count * sizeof(struct per_cpu_info); |
657 | cpus = realloc(cpus, size); | |
658 | if (!cpus) { | |
659 | char name[20]; | |
660 | fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n", | |
661 | get_dev_name(pdi, name, sizeof(name)), size); | |
a718bd37 NS |
662 | exit(1); |
663 | } | |
664 | ||
e7c9f3ff NS |
665 | new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info)); |
666 | new_space = (new_count - ncpus) * sizeof(struct per_cpu_info); | |
a718bd37 | 667 | memset(new_start, 0, new_space); |
e7c9f3ff NS |
668 | |
669 | pdi->ncpus = new_count; | |
670 | pdi->cpus = cpus; | |
671 | } | |
cb2a1a62 | 672 | |
e7c9f3ff NS |
673 | static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu) |
674 | { | |
cb2a1a62 JA |
675 | struct per_cpu_info *pci; |
676 | ||
e7c9f3ff NS |
677 | if (cpu >= pdi->ncpus) |
678 | resize_cpu_info(pdi, cpu); | |
cb2a1a62 JA |
679 | |
680 | pci = &pdi->cpus[cpu]; | |
681 | pci->cpu = cpu; | |
682 | return pci; | |
a718bd37 NS |
683 | } |
684 | ||
e7c9f3ff NS |
685 | |
686 | static int resize_devices(char *name) | |
a718bd37 | 687 | { |
e7c9f3ff | 688 | int size = (ndevices + 1) * sizeof(struct per_dev_info); |
c499bf38 | 689 | |
e7c9f3ff NS |
690 | devices = realloc(devices, size); |
691 | if (!devices) { | |
692 | fprintf(stderr, "Out of memory, device %s (%d)\n", name, size); | |
693 | return 1; | |
694 | } | |
695 | memset(&devices[ndevices], 0, sizeof(struct per_dev_info)); | |
696 | devices[ndevices].name = name; | |
697 | ndevices++; | |
698 | return 0; | |
699 | } | |
a718bd37 | 700 | |
f7bd1a9b | 701 | static struct per_dev_info *get_dev_info(dev_t dev) |
e7c9f3ff | 702 | { |
cb2a1a62 | 703 | struct per_dev_info *pdi; |
e7c9f3ff | 704 | int i; |
c499bf38 | 705 | |
7d747d22 | 706 | for (i = 0; i < ndevices; i++) { |
f7bd1a9b JA |
707 | if (!devices[i].dev) |
708 | devices[i].dev = dev; | |
709 | if (devices[i].dev == dev) | |
e7c9f3ff | 710 | return &devices[i]; |
7d747d22 | 711 | } |
cb2a1a62 | 712 | |
2a1b3424 | 713 | if (resize_devices(NULL)) |
e7c9f3ff | 714 | return NULL; |
cb2a1a62 JA |
715 | |
716 | pdi = &devices[ndevices - 1]; | |
f7bd1a9b | 717 | pdi->dev = dev; |
d6222db8 | 718 | pdi->last_sequence = -1; |
287fa3d6 | 719 | pdi->last_read_time = 0; |
2a1b3424 JA |
720 | memset(&pdi->rb_last, 0, sizeof(pdi->rb_last)); |
721 | pdi->rb_last_entries = 0; | |
cb2a1a62 | 722 | return pdi; |
a718bd37 NS |
723 | } |
724 | ||
e7c9f3ff NS |
725 | static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size) |
726 | { | |
727 | if (pdi->name) | |
728 | snprintf(buffer, size, "%s", pdi->name); | |
729 | else | |
f7bd1a9b | 730 | snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev)); |
e7c9f3ff NS |
731 | return buffer; |
732 | } | |
733 | ||
e7c9f3ff | 734 | static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit) |
cfab07eb AB |
735 | { |
736 | unsigned long long this = bit->time; | |
e7c9f3ff | 737 | unsigned long long last = pdi->last_reported_time; |
cfab07eb | 738 | |
e7c9f3ff NS |
739 | pdi->backwards = (this < last) ? 'B' : ' '; |
740 | pdi->last_reported_time = this; | |
cfab07eb AB |
741 | } |
742 | ||
152f6476 JA |
743 | static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t, |
744 | int rw) | |
d0ca268b JA |
745 | { |
746 | if (rw) { | |
152f6476 JA |
747 | ios->mwrites++; |
748 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 749 | } else { |
152f6476 JA |
750 | ios->mreads++; |
751 | ios->qread_kb += t->bytes >> 10; | |
752 | } | |
753 | } | |
754 | ||
755 | static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci, | |
756 | int rw) | |
757 | { | |
758 | __account_m(&pci->io_stats, t, rw); | |
759 | ||
760 | if (per_process_stats) { | |
761 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
762 | ||
763 | __account_m(ios, t, rw); | |
d0ca268b JA |
764 | } |
765 | } | |
766 | ||
b6076a9b JA |
767 | static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t, |
768 | int rw) | |
d0ca268b JA |
769 | { |
770 | if (rw) { | |
152f6476 JA |
771 | ios->qwrites++; |
772 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 773 | } else { |
152f6476 JA |
774 | ios->qreads++; |
775 | ios->qread_kb += t->bytes >> 10; | |
776 | } | |
777 | } | |
778 | ||
b6076a9b JA |
779 | static inline void account_queue(struct blk_io_trace *t, |
780 | struct per_cpu_info *pci, int rw) | |
152f6476 | 781 | { |
b6076a9b | 782 | __account_queue(&pci->io_stats, t, rw); |
152f6476 JA |
783 | |
784 | if (per_process_stats) { | |
785 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
786 | ||
b6076a9b | 787 | __account_queue(ios, t, rw); |
d0ca268b JA |
788 | } |
789 | } | |
790 | ||
152f6476 | 791 | static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes) |
d0ca268b JA |
792 | { |
793 | if (rw) { | |
152f6476 JA |
794 | ios->cwrites++; |
795 | ios->cwrite_kb += bytes >> 10; | |
d0ca268b | 796 | } else { |
152f6476 JA |
797 | ios->creads++; |
798 | ios->cread_kb += bytes >> 10; | |
799 | } | |
800 | } | |
801 | ||
802 | static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci, | |
803 | int rw, int bytes) | |
804 | { | |
805 | __account_c(&pci->io_stats, rw, bytes); | |
806 | ||
807 | if (per_process_stats) { | |
808 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
809 | ||
810 | __account_c(ios, rw, bytes); | |
d0ca268b JA |
811 | } |
812 | } | |
813 | ||
b6076a9b JA |
814 | static inline void __account_issue(struct io_stats *ios, int rw, |
815 | unsigned int bytes) | |
afd2d7ad | 816 | { |
817 | if (rw) { | |
152f6476 JA |
818 | ios->iwrites++; |
819 | ios->iwrite_kb += bytes >> 10; | |
afd2d7ad | 820 | } else { |
152f6476 JA |
821 | ios->ireads++; |
822 | ios->iread_kb += bytes >> 10; | |
afd2d7ad | 823 | } |
824 | } | |
825 | ||
b6076a9b JA |
826 | static inline void account_issue(struct blk_io_trace *t, |
827 | struct per_cpu_info *pci, int rw) | |
d0ca268b | 828 | { |
b6076a9b | 829 | __account_issue(&pci->io_stats, rw, t->bytes); |
152f6476 JA |
830 | |
831 | if (per_process_stats) { | |
832 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
d5396421 | 833 | |
b6076a9b | 834 | __account_issue(ios, rw, t->bytes); |
152f6476 JA |
835 | } |
836 | } | |
837 | ||
06639b27 JA |
838 | static inline void __account_unplug(struct io_stats *ios, int timer) |
839 | { | |
840 | if (timer) | |
841 | ios->timer_unplugs++; | |
842 | else | |
843 | ios->io_unplugs++; | |
844 | } | |
845 | ||
846 | static inline void account_unplug(struct blk_io_trace *t, | |
847 | struct per_cpu_info *pci, int timer) | |
848 | { | |
849 | __account_unplug(&pci->io_stats, timer); | |
850 | ||
851 | if (per_process_stats) { | |
852 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
853 | ||
854 | __account_unplug(ios, timer); | |
855 | } | |
856 | } | |
857 | ||
f7bd1a9b JA |
858 | static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci, |
859 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 860 | { |
f7bd1a9b | 861 | process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL); |
ab197ca7 AB |
862 | } |
863 | ||
f7bd1a9b JA |
864 | static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci, |
865 | struct blk_io_trace *t, char *act) | |
b6076a9b | 866 | { |
f7bd1a9b | 867 | process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL); |
b6076a9b JA |
868 | } |
869 | ||
ab197ca7 AB |
870 | static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t, |
871 | char *act) | |
872 | { | |
b6076a9b | 873 | process_fmt(act, pci, t, -1, 0, NULL); |
ab197ca7 | 874 | } |
2e3e8ded | 875 | |
f7bd1a9b JA |
876 | static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci, |
877 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 878 | { |
f7bd1a9b | 879 | process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL); |
d0ca268b JA |
880 | } |
881 | ||
f7bd1a9b JA |
882 | static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci, |
883 | struct blk_io_trace *t, char *act) | |
d0ca268b | 884 | { |
a01516de | 885 | if (act[0] == 'F') |
f7bd1a9b | 886 | log_track_frontmerge(pdi, t); |
2e3e8ded | 887 | |
ab197ca7 | 888 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
889 | } |
890 | ||
dfe34da1 | 891 | static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 892 | char *act) |
dfe34da1 | 893 | { |
ab197ca7 | 894 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
dfe34da1 JA |
895 | } |
896 | ||
d5396421 | 897 | static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 898 | char *act) |
d0ca268b | 899 | { |
ab197ca7 | 900 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
901 | } |
902 | ||
ab197ca7 | 903 | static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 904 | char *act) |
67e14fdc | 905 | { |
ab197ca7 | 906 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
67e14fdc JA |
907 | } |
908 | ||
93f1c611 JA |
909 | static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t, |
910 | char *act) | |
911 | { | |
912 | process_fmt(act, pci, t, -1ULL, 0, NULL); | |
913 | } | |
914 | ||
ab197ca7 | 915 | static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act) |
d0ca268b | 916 | { |
ab197ca7 | 917 | unsigned char *buf = (unsigned char *) t + sizeof(*t); |
d0ca268b | 918 | |
ab197ca7 | 919 | process_fmt(act, pci, t, -1ULL, t->pdu_len, buf); |
d0ca268b JA |
920 | } |
921 | ||
ff3a732c | 922 | static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 923 | { |
56f2af81 JA |
924 | int act = t->action & 0xffff; |
925 | ||
926 | switch (act) { | |
d0ca268b | 927 | case __BLK_TA_QUEUE: |
3639a11e | 928 | log_generic(pci, t, "Q"); |
d0ca268b JA |
929 | break; |
930 | case __BLK_TA_GETRQ: | |
3639a11e | 931 | log_generic(pci, t, "G"); |
d0ca268b JA |
932 | break; |
933 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 934 | log_generic(pci, t, "S"); |
d0ca268b JA |
935 | break; |
936 | case __BLK_TA_REQUEUE: | |
3639a11e | 937 | log_generic(pci, t, "R"); |
d0ca268b JA |
938 | break; |
939 | case __BLK_TA_ISSUE: | |
ab197ca7 | 940 | log_pc(pci, t, "D"); |
d0ca268b JA |
941 | break; |
942 | case __BLK_TA_COMPLETE: | |
3639a11e | 943 | log_pc(pci, t, "C"); |
d0ca268b | 944 | break; |
56f2af81 JA |
945 | case __BLK_TA_INSERT: |
946 | log_pc(pci, t, "I"); | |
947 | break; | |
d0ca268b | 948 | default: |
56f2af81 | 949 | fprintf(stderr, "Bad pc action %x\n", act); |
87b72777 | 950 | break; |
d0ca268b | 951 | } |
d0ca268b JA |
952 | } |
953 | ||
f7bd1a9b JA |
954 | static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi, |
955 | struct per_cpu_info *pci) | |
d0ca268b JA |
956 | { |
957 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
7997c5b0 | 958 | int act = t->action & 0xffff; |
d0ca268b | 959 | |
7997c5b0 | 960 | switch (act) { |
d0ca268b | 961 | case __BLK_TA_QUEUE: |
b6076a9b | 962 | account_queue(t, pci, w); |
3639a11e | 963 | log_queue(pci, t, "Q"); |
d0ca268b | 964 | break; |
b6076a9b | 965 | case __BLK_TA_INSERT: |
f7bd1a9b | 966 | log_insert(pdi, pci, t, "I"); |
b6076a9b | 967 | break; |
d0ca268b | 968 | case __BLK_TA_BACKMERGE: |
152f6476 | 969 | account_m(t, pci, w); |
f7bd1a9b | 970 | log_merge(pdi, pci, t, "M"); |
d0ca268b JA |
971 | break; |
972 | case __BLK_TA_FRONTMERGE: | |
152f6476 | 973 | account_m(t, pci, w); |
f7bd1a9b | 974 | log_merge(pdi, pci, t, "F"); |
d0ca268b JA |
975 | break; |
976 | case __BLK_TA_GETRQ: | |
f7bd1a9b | 977 | log_track_getrq(pdi, t); |
3639a11e | 978 | log_generic(pci, t, "G"); |
d0ca268b JA |
979 | break; |
980 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 981 | log_generic(pci, t, "S"); |
d0ca268b JA |
982 | break; |
983 | case __BLK_TA_REQUEUE: | |
152f6476 | 984 | account_c(t, pci, w, -t->bytes); |
3639a11e | 985 | log_queue(pci, t, "R"); |
d0ca268b JA |
986 | break; |
987 | case __BLK_TA_ISSUE: | |
b6076a9b | 988 | account_issue(t, pci, w); |
f7bd1a9b | 989 | log_issue(pdi, pci, t, "D"); |
d0ca268b JA |
990 | break; |
991 | case __BLK_TA_COMPLETE: | |
152f6476 | 992 | account_c(t, pci, w, t->bytes); |
f7bd1a9b | 993 | log_complete(pdi, pci, t, "C"); |
d0ca268b | 994 | break; |
88b1a526 | 995 | case __BLK_TA_PLUG: |
3639a11e | 996 | log_action(pci, t, "P"); |
88b1a526 | 997 | break; |
3639a11e | 998 | case __BLK_TA_UNPLUG_IO: |
06639b27 | 999 | account_unplug(t, pci, 0); |
3639a11e JA |
1000 | log_unplug(pci, t, "U"); |
1001 | break; | |
1002 | case __BLK_TA_UNPLUG_TIMER: | |
06639b27 | 1003 | account_unplug(t, pci, 1); |
3639a11e | 1004 | log_unplug(pci, t, "UT"); |
88b1a526 | 1005 | break; |
93f1c611 JA |
1006 | case __BLK_TA_SPLIT: |
1007 | log_split(pci, t, "X"); | |
1008 | break; | |
1009 | case __BLK_TA_BOUNCE: | |
1010 | log_generic(pci, t, "B"); | |
1011 | break; | |
d0ca268b JA |
1012 | default: |
1013 | fprintf(stderr, "Bad fs action %x\n", t->action); | |
1f79c4a0 | 1014 | break; |
d0ca268b | 1015 | } |
d0ca268b JA |
1016 | } |
1017 | ||
ff3a732c JA |
1018 | static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci, |
1019 | struct per_dev_info *pdi) | |
d0ca268b JA |
1020 | { |
1021 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) | |
ff3a732c | 1022 | dump_trace_pc(t, pci); |
d0ca268b | 1023 | else |
f7bd1a9b | 1024 | dump_trace_fs(t, pdi, pci); |
87b72777 | 1025 | |
e7c9f3ff | 1026 | pdi->events++; |
d0ca268b JA |
1027 | } |
1028 | ||
152f6476 | 1029 | static void dump_io_stats(struct io_stats *ios, char *msg) |
5c017e4b | 1030 | { |
152f6476 JA |
1031 | fprintf(ofp, "%s\n", msg); |
1032 | ||
1033 | fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb); | |
1034 | fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb); | |
0a6b8fc4 | 1035 | |
152f6476 JA |
1036 | fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb); |
1037 | fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb); | |
1038 | fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb); | |
1039 | fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb); | |
1040 | fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' '); | |
152f6476 | 1041 | fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites); |
06639b27 JA |
1042 | fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' '); |
1043 | fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs); | |
5c017e4b JA |
1044 | } |
1045 | ||
50adc0ba JA |
1046 | static void dump_wait_stats(struct per_process_info *ppi) |
1047 | { | |
b9d40d6f JA |
1048 | unsigned long rawait = ppi->longest_allocation_wait[0] / 1000; |
1049 | unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000; | |
1050 | unsigned long rcwait = ppi->longest_completion_wait[0] / 1000; | |
1051 | unsigned long wawait = ppi->longest_allocation_wait[1] / 1000; | |
1052 | unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000; | |
1053 | unsigned long wcwait = ppi->longest_completion_wait[1] / 1000; | |
1054 | ||
1055 | fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' '); | |
1056 | fprintf(ofp, " Allocation wait: %'8lu\n", wawait); | |
1057 | fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' '); | |
1058 | fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait); | |
1059 | fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' '); | |
1060 | fprintf(ofp, " Completion wait: %'8lu\n", wcwait); | |
50adc0ba JA |
1061 | } |
1062 | ||
886ecf0e JA |
1063 | static int ppi_name_compare(const void *p1, const void *p2) |
1064 | { | |
1065 | struct per_process_info *ppi1 = *((struct per_process_info **) p1); | |
1066 | struct per_process_info *ppi2 = *((struct per_process_info **) p2); | |
1067 | int res; | |
1068 | ||
1069 | res = strverscmp(ppi1->name, ppi2->name); | |
1070 | if (!res) | |
06e6f286 | 1071 | res = ppi1->pid > ppi2->pid; |
886ecf0e JA |
1072 | |
1073 | return res; | |
1074 | } | |
1075 | ||
1076 | static void sort_process_list(void) | |
1077 | { | |
1078 | struct per_process_info **ppis; | |
1079 | struct per_process_info *ppi; | |
1080 | int i = 0; | |
1081 | ||
1082 | ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *)); | |
1083 | ||
1084 | ppi = ppi_list; | |
1085 | while (ppi) { | |
06e6f286 | 1086 | ppis[i++] = ppi; |
886ecf0e JA |
1087 | ppi = ppi->list_next; |
1088 | } | |
1089 | ||
06e6f286 | 1090 | qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare); |
886ecf0e JA |
1091 | |
1092 | i = ppi_list_entries - 1; | |
1093 | ppi_list = NULL; | |
1094 | while (i >= 0) { | |
1095 | ppi = ppis[i]; | |
1096 | ||
1097 | ppi->list_next = ppi_list; | |
1098 | ppi_list = ppi; | |
1099 | i--; | |
1100 | } | |
50c38702 JA |
1101 | |
1102 | free(ppis); | |
886ecf0e JA |
1103 | } |
1104 | ||
152f6476 JA |
1105 | static void show_process_stats(void) |
1106 | { | |
1107 | struct per_process_info *ppi; | |
1108 | ||
886ecf0e JA |
1109 | sort_process_list(); |
1110 | ||
152f6476 JA |
1111 | ppi = ppi_list; |
1112 | while (ppi) { | |
ce8b6b4f JA |
1113 | char name[64]; |
1114 | ||
715d8021 | 1115 | if (ppi->more_than_one) |
bf0720af | 1116 | sprintf(name, "%s (%u, ...)", ppi->name, ppi->pid); |
715d8021 JA |
1117 | else |
1118 | sprintf(name, "%s (%u)", ppi->name, ppi->pid); | |
bf0720af | 1119 | |
ce8b6b4f | 1120 | dump_io_stats(&ppi->io_stats, name); |
50adc0ba | 1121 | dump_wait_stats(ppi); |
152f6476 JA |
1122 | ppi = ppi->list_next; |
1123 | } | |
1124 | ||
1125 | fprintf(ofp, "\n"); | |
1126 | } | |
1127 | ||
e7c9f3ff | 1128 | static void show_device_and_cpu_stats(void) |
d0ca268b | 1129 | { |
e7c9f3ff NS |
1130 | struct per_dev_info *pdi; |
1131 | struct per_cpu_info *pci; | |
1132 | struct io_stats total, *ios; | |
1133 | int i, j, pci_events; | |
1134 | char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3]; | |
1135 | char name[32]; | |
1136 | ||
1137 | for (pdi = devices, i = 0; i < ndevices; i++, pdi++) { | |
1138 | ||
1139 | memset(&total, 0, sizeof(total)); | |
1140 | pci_events = 0; | |
1141 | ||
1142 | if (i > 0) | |
1143 | fprintf(ofp, "\n"); | |
1144 | ||
1145 | for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) { | |
1146 | if (!pci->nelems) | |
1147 | continue; | |
1148 | ||
1149 | ios = &pci->io_stats; | |
1150 | total.qreads += ios->qreads; | |
1151 | total.qwrites += ios->qwrites; | |
1152 | total.creads += ios->creads; | |
1153 | total.cwrites += ios->cwrites; | |
1154 | total.mreads += ios->mreads; | |
1155 | total.mwrites += ios->mwrites; | |
1156 | total.ireads += ios->ireads; | |
1157 | total.iwrites += ios->iwrites; | |
1158 | total.qread_kb += ios->qread_kb; | |
1159 | total.qwrite_kb += ios->qwrite_kb; | |
1160 | total.cread_kb += ios->cread_kb; | |
1161 | total.cwrite_kb += ios->cwrite_kb; | |
1162 | total.iread_kb += ios->iread_kb; | |
1163 | total.iwrite_kb += ios->iwrite_kb; | |
06639b27 JA |
1164 | total.timer_unplugs += ios->timer_unplugs; |
1165 | total.io_unplugs += ios->io_unplugs; | |
e7c9f3ff NS |
1166 | |
1167 | snprintf(line, sizeof(line) - 1, "CPU%d (%s):", | |
1168 | j, get_dev_name(pdi, name, sizeof(name))); | |
1169 | dump_io_stats(ios, line); | |
1170 | pci_events++; | |
1171 | } | |
5c017e4b | 1172 | |
e7c9f3ff NS |
1173 | if (pci_events > 1) { |
1174 | fprintf(ofp, "\n"); | |
1175 | snprintf(line, sizeof(line) - 1, "Total (%s):", | |
1176 | get_dev_name(pdi, name, sizeof(name))); | |
1177 | dump_io_stats(&total, line); | |
1178 | } | |
d0ca268b | 1179 | |
dcf0f7ed | 1180 | fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n", |
0aa8a62d | 1181 | get_dev_name(pdi, line, sizeof(line)), pdi->events, |
dcf0f7ed | 1182 | pdi->skips); |
e7c9f3ff | 1183 | } |
d0ca268b JA |
1184 | } |
1185 | ||
d36421e4 JA |
1186 | /* |
1187 | * struct trace and blktrace allocation cache, we do potentially | |
1188 | * millions of mallocs for these structures while only using at most | |
1189 | * a few thousand at the time | |
1190 | */ | |
1191 | static inline void t_free(struct trace *t) | |
1192 | { | |
3c8413cb | 1193 | if (t_alloc_cache < 1024) { |
1d24fc14 JA |
1194 | t->next = t_alloc_list; |
1195 | t_alloc_list = t; | |
3c8413cb JA |
1196 | t_alloc_cache++; |
1197 | } else | |
1198 | free(t); | |
d36421e4 JA |
1199 | } |
1200 | ||
1201 | static inline struct trace *t_alloc(void) | |
1202 | { | |
1203 | struct trace *t = t_alloc_list; | |
1204 | ||
1205 | if (t) { | |
1206 | t_alloc_list = t->next; | |
3c8413cb | 1207 | t_alloc_cache--; |
d36421e4 JA |
1208 | return t; |
1209 | } | |
1210 | ||
1211 | return malloc(sizeof(*t)); | |
1212 | } | |
1213 | ||
1214 | static inline void bit_free(struct blk_io_trace *bit) | |
1215 | { | |
3c8413cb | 1216 | if (bit_alloc_cache < 1024) { |
1d24fc14 JA |
1217 | /* |
1218 | * abuse a 64-bit field for a next pointer for the free item | |
1219 | */ | |
1220 | bit->time = (__u64) (unsigned long) bit_alloc_list; | |
1221 | bit_alloc_list = (struct blk_io_trace *) bit; | |
3c8413cb JA |
1222 | bit_alloc_cache++; |
1223 | } else | |
1224 | free(bit); | |
d36421e4 JA |
1225 | } |
1226 | ||
1227 | static inline struct blk_io_trace *bit_alloc(void) | |
1228 | { | |
1229 | struct blk_io_trace *bit = bit_alloc_list; | |
1230 | ||
1231 | if (bit) { | |
0c39425c NS |
1232 | bit_alloc_list = (struct blk_io_trace *) (unsigned long) \ |
1233 | bit->time; | |
3c8413cb | 1234 | bit_alloc_cache--; |
d36421e4 JA |
1235 | return bit; |
1236 | } | |
1237 | ||
1238 | return malloc(sizeof(*bit)); | |
1239 | } | |
1240 | ||
4f0ae44f JA |
1241 | static void find_genesis(void) |
1242 | { | |
1243 | struct trace *t = trace_list; | |
1244 | ||
1245 | genesis_time = -1ULL; | |
1246 | while (t != NULL) { | |
1247 | if (t->bit->time < genesis_time) | |
1248 | genesis_time = t->bit->time; | |
1249 | ||
1250 | t = t->next; | |
1251 | } | |
1252 | } | |
1253 | ||
7f4d89e6 | 1254 | static inline int check_stopwatch(struct blk_io_trace *bit) |
4f0ae44f | 1255 | { |
7f4d89e6 JA |
1256 | if (bit->time < stopwatch_end && |
1257 | bit->time >= stopwatch_start) | |
4f0ae44f JA |
1258 | return 0; |
1259 | ||
1260 | return 1; | |
1261 | } | |
1262 | ||
53c68c88 JA |
1263 | /* |
1264 | * return youngest entry read | |
1265 | */ | |
1266 | static int sort_entries(unsigned long long *youngest) | |
4f0ae44f JA |
1267 | { |
1268 | struct trace *t; | |
4f0ae44f JA |
1269 | |
1270 | if (!genesis_time) | |
1271 | find_genesis(); | |
1272 | ||
d6222db8 | 1273 | *youngest = 0; |
4f0ae44f JA |
1274 | while ((t = trace_list) != NULL) { |
1275 | struct blk_io_trace *bit = t->bit; | |
1276 | ||
1277 | trace_list = t->next; | |
1278 | ||
7f4d89e6 | 1279 | bit->time -= genesis_time; |
4f0ae44f | 1280 | |
d6222db8 JA |
1281 | if (bit->time < *youngest || !*youngest) |
1282 | *youngest = bit->time; | |
1283 | ||
7f4d89e6 | 1284 | if (check_stopwatch(bit)) { |
4f0ae44f JA |
1285 | bit_free(bit); |
1286 | t_free(t); | |
1287 | continue; | |
1288 | } | |
1289 | ||
2a1b3424 | 1290 | if (trace_rb_insert_sort(t)) |
53c68c88 | 1291 | return -1; |
1c7c54aa JA |
1292 | |
1293 | if (bit->sequence < smallest_seq_read) | |
1294 | smallest_seq_read = bit->sequence; | |
4f0ae44f JA |
1295 | } |
1296 | ||
53c68c88 | 1297 | return 0; |
4f0ae44f JA |
1298 | } |
1299 | ||
1c24add6 JA |
1300 | static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t) |
1301 | { | |
1302 | rb_erase(&t->rb_node, &pdi->rb_last); | |
1303 | pdi->rb_last_entries--; | |
1304 | ||
1305 | bit_free(t->bit); | |
1306 | t_free(t); | |
1307 | } | |
1308 | ||
1309 | static void put_trace(struct per_dev_info *pdi, struct trace *t) | |
2a1b3424 JA |
1310 | { |
1311 | rb_erase(&t->rb_node, &rb_sort_root); | |
1312 | rb_sort_entries--; | |
1313 | ||
1314 | trace_rb_insert_last(pdi, t); | |
1315 | ||
1c24add6 | 1316 | if (pdi->rb_last_entries > rb_batch * pdi->nfiles) { |
2a1b3424 JA |
1317 | struct rb_node *n = rb_first(&pdi->rb_last); |
1318 | ||
1319 | t = rb_entry(n, struct trace, rb_node); | |
1c24add6 | 1320 | __put_trace_last(pdi, t); |
2a1b3424 JA |
1321 | } |
1322 | } | |
1323 | ||
1c7c54aa JA |
1324 | static int check_sequence(struct per_dev_info *pdi, struct blk_io_trace *bit, |
1325 | int force) | |
2a1b3424 JA |
1326 | { |
1327 | unsigned long expected_sequence = pdi->last_sequence + 1; | |
1c7c54aa | 1328 | struct trace *t; |
d6222db8 JA |
1329 | |
1330 | /* | |
1331 | * first entry, always ok | |
1332 | */ | |
1333 | if (!expected_sequence) | |
1334 | return 0; | |
2a1b3424 JA |
1335 | |
1336 | if (bit->sequence == expected_sequence) | |
1337 | return 0; | |
1338 | ||
2a1b3424 | 1339 | /* |
1c7c54aa JA |
1340 | * we may not have seen that sequence yet. if we are not doing |
1341 | * the final run, break and wait for more entries. | |
1c24add6 | 1342 | */ |
1c7c54aa JA |
1343 | if (expected_sequence < smallest_seq_read) { |
1344 | t = trace_rb_find_last(pdi, expected_sequence); | |
1345 | if (!t) | |
1346 | goto skip; | |
2a1b3424 | 1347 | |
7d1c0411 | 1348 | __put_trace_last(pdi, t); |
2a1b3424 | 1349 | return 0; |
1c7c54aa | 1350 | } else if (!force) |
2a1b3424 | 1351 | return 1; |
1c7c54aa JA |
1352 | else { |
1353 | skip: | |
7d1c0411 JA |
1354 | if (print_missing) { |
1355 | fprintf(stderr, "(%d,%d): skipping %lu -> %u\n", | |
1356 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
1357 | pdi->last_sequence, bit->sequence); | |
1358 | } | |
1c7c54aa JA |
1359 | pdi->skips++; |
1360 | return 0; | |
1361 | } | |
2a1b3424 JA |
1362 | } |
1363 | ||
a649216c | 1364 | static void show_entries_rb(int force) |
8fc0abbc | 1365 | { |
1f7afa72 JA |
1366 | struct per_dev_info *pdi = NULL; |
1367 | struct per_cpu_info *pci = NULL; | |
8fc0abbc | 1368 | struct blk_io_trace *bit; |
3aabcd89 | 1369 | struct rb_node *n; |
8fc0abbc | 1370 | struct trace *t; |
1f7afa72 | 1371 | |
7d747d22 | 1372 | while ((n = rb_first(&rb_sort_root)) != NULL) { |
1f7afa72 JA |
1373 | if (done) |
1374 | break; | |
8fc0abbc JA |
1375 | |
1376 | t = rb_entry(n, struct trace, rb_node); | |
1377 | bit = t->bit; | |
1378 | ||
f7bd1a9b | 1379 | if (!pdi || pdi->dev != bit->device) |
287fa3d6 | 1380 | pdi = get_dev_info(bit->device); |
1f7afa72 | 1381 | |
e7c9f3ff NS |
1382 | if (!pdi) { |
1383 | fprintf(stderr, "Unknown device ID? (%d,%d)\n", | |
1384 | MAJOR(bit->device), MINOR(bit->device)); | |
1385 | break; | |
1386 | } | |
1f7afa72 | 1387 | |
1c7c54aa | 1388 | if (check_sequence(pdi, bit, force)) |
2a1b3424 | 1389 | break; |
cb2a1a62 | 1390 | |
a649216c | 1391 | if (!force && bit->time > last_allowed_time) |
46e6968b | 1392 | break; |
8fc0abbc | 1393 | |
be925321 JA |
1394 | pdi->last_sequence = bit->sequence; |
1395 | ||
4f0ae44f | 1396 | check_time(pdi, bit); |
8fc0abbc | 1397 | |
4f0ae44f JA |
1398 | if (!pci || pci->cpu != bit->cpu) |
1399 | pci = get_cpu_info(pdi, bit->cpu); | |
287fa3d6 | 1400 | |
4f0ae44f | 1401 | dump_trace(bit, pci, pdi); |
87b72777 | 1402 | |
2a1b3424 | 1403 | put_trace(pdi, t); |
cb2a1a62 | 1404 | } |
8fc0abbc JA |
1405 | } |
1406 | ||
1f79c4a0 JA |
1407 | static int read_data(int fd, void *buffer, int bytes, int block) |
1408 | { | |
1409 | int ret, bytes_left, fl; | |
1410 | void *p; | |
1411 | ||
1412 | fl = fcntl(fd, F_GETFL); | |
1413 | ||
1414 | if (!block) | |
1415 | fcntl(fd, F_SETFL, fl | O_NONBLOCK); | |
1416 | else | |
1417 | fcntl(fd, F_SETFL, fl & ~O_NONBLOCK); | |
1418 | ||
1419 | bytes_left = bytes; | |
1420 | p = buffer; | |
1421 | while (bytes_left > 0) { | |
1422 | ret = read(fd, p, bytes_left); | |
1423 | if (!ret) | |
1424 | return 1; | |
1425 | else if (ret < 0) { | |
1426 | if (errno != EAGAIN) | |
1427 | perror("read"); | |
a649216c | 1428 | |
1f79c4a0 JA |
1429 | return -1; |
1430 | } else { | |
1431 | p += ret; | |
1432 | bytes_left -= ret; | |
1433 | } | |
1434 | } | |
1435 | ||
1436 | return 0; | |
1437 | } | |
1438 | ||
a649216c | 1439 | static int read_events(int fd, int always_block) |
cb2a1a62 | 1440 | { |
287fa3d6 | 1441 | struct per_dev_info *pdi = NULL; |
e820abd7 | 1442 | unsigned int events = 0; |
7d747d22 JA |
1443 | |
1444 | while (!is_done() && events < rb_batch) { | |
1445 | struct blk_io_trace *bit; | |
1446 | struct trace *t; | |
1447 | int pdu_len; | |
1448 | __u32 magic; | |
1449 | ||
d36421e4 | 1450 | bit = bit_alloc(); |
cb2a1a62 | 1451 | |
a649216c | 1452 | if (read_data(fd, bit, sizeof(*bit), !events || always_block)) |
cb2a1a62 JA |
1453 | break; |
1454 | ||
7d747d22 JA |
1455 | magic = be32_to_cpu(bit->magic); |
1456 | if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
1457 | fprintf(stderr, "Bad magic %x\n", magic); | |
1458 | break; | |
1459 | } | |
1460 | ||
1461 | pdu_len = be16_to_cpu(bit->pdu_len); | |
1462 | if (pdu_len) { | |
1463 | void *ptr = realloc(bit, sizeof(*bit) + pdu_len); | |
1464 | ||
1465 | if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1)) | |
1466 | break; | |
1467 | ||
1468 | bit = ptr; | |
1469 | } | |
1470 | ||
d6222db8 JA |
1471 | trace_to_cpu(bit); |
1472 | ||
1473 | if (verify_trace(bit)) { | |
1474 | bit_free(bit); | |
1475 | continue; | |
1476 | } | |
1477 | ||
d36421e4 | 1478 | t = t_alloc(); |
cb2a1a62 JA |
1479 | memset(t, 0, sizeof(*t)); |
1480 | t->bit = bit; | |
1481 | ||
7d747d22 JA |
1482 | t->next = trace_list; |
1483 | trace_list = t; | |
1f7afa72 | 1484 | |
f7bd1a9b | 1485 | if (!pdi || pdi->dev != bit->device) |
287fa3d6 JA |
1486 | pdi = get_dev_info(bit->device); |
1487 | ||
1488 | if (bit->time > pdi->last_read_time) | |
1489 | pdi->last_read_time = bit->time; | |
1490 | ||
7d747d22 | 1491 | events++; |
cb2a1a62 JA |
1492 | } |
1493 | ||
7d747d22 | 1494 | return events; |
cb2a1a62 JA |
1495 | } |
1496 | ||
d5396421 | 1497 | static int do_file(void) |
d0ca268b | 1498 | { |
7d747d22 | 1499 | struct per_cpu_info *pci; |
73877e12 NS |
1500 | struct per_dev_info *pdi; |
1501 | int i, j, events, events_added; | |
d0ca268b | 1502 | |
7d747d22 JA |
1503 | /* |
1504 | * first prepare all files for reading | |
1505 | */ | |
e8741a4a | 1506 | for (i = 0; i < ndevices; i++) { |
73877e12 NS |
1507 | pdi = &devices[i]; |
1508 | pdi->nfiles = 0; | |
d6222db8 | 1509 | pdi->last_sequence = -1; |
73877e12 | 1510 | |
74feaeb5 | 1511 | for (j = 0;; j++) { |
e7c9f3ff | 1512 | struct stat st; |
87b72777 | 1513 | |
e7c9f3ff NS |
1514 | pci = get_cpu_info(pdi, j); |
1515 | pci->cpu = j; | |
7d747d22 | 1516 | pci->fd = -1; |
d0ca268b | 1517 | |
e7c9f3ff | 1518 | snprintf(pci->fname, sizeof(pci->fname)-1, |
7d747d22 | 1519 | "%s.blktrace.%d", pdi->name, pci->cpu); |
e7c9f3ff NS |
1520 | if (stat(pci->fname, &st) < 0) |
1521 | break; | |
74feaeb5 AB |
1522 | if (st.st_size) { |
1523 | pci->fd = open(pci->fname, O_RDONLY); | |
1524 | if (pci->fd < 0) { | |
1525 | perror(pci->fname); | |
1526 | continue; | |
1527 | } | |
e7c9f3ff NS |
1528 | } |
1529 | ||
7d747d22 | 1530 | printf("Input file %s added\n", pci->fname); |
73877e12 | 1531 | pdi->nfiles++; |
d0ca268b | 1532 | } |
d5396421 JA |
1533 | } |
1534 | ||
7d747d22 JA |
1535 | /* |
1536 | * now loop over the files reading in the data | |
1537 | */ | |
412819ce | 1538 | do { |
53c68c88 JA |
1539 | unsigned long long youngest; |
1540 | ||
7d747d22 | 1541 | events_added = 0; |
287fa3d6 | 1542 | last_allowed_time = -1ULL; |
1c7c54aa | 1543 | smallest_seq_read = -1U; |
d5396421 | 1544 | |
7d747d22 | 1545 | for (i = 0; i < ndevices; i++) { |
73877e12 NS |
1546 | pdi = &devices[i]; |
1547 | ||
1548 | for (j = 0; j < pdi->nfiles; j++) { | |
d5396421 | 1549 | |
73877e12 | 1550 | pci = get_cpu_info(pdi, j); |
d5396421 | 1551 | |
7d747d22 JA |
1552 | if (pci->fd == -1) |
1553 | continue; | |
51128a28 | 1554 | |
a649216c | 1555 | events = read_events(pci->fd, 1); |
7d747d22 JA |
1556 | if (!events) { |
1557 | close(pci->fd); | |
1558 | pci->fd = -1; | |
1559 | continue; | |
1560 | } | |
d5396421 | 1561 | |
287fa3d6 JA |
1562 | if (pdi->last_read_time < last_allowed_time) |
1563 | last_allowed_time = pdi->last_read_time; | |
d5396421 | 1564 | |
7d747d22 JA |
1565 | events_added += events; |
1566 | } | |
2ff323b0 | 1567 | } |
d5396421 | 1568 | |
53c68c88 JA |
1569 | if (sort_entries(&youngest)) |
1570 | break; | |
1571 | ||
1572 | if (youngest > stopwatch_end) | |
287fa3d6 JA |
1573 | break; |
1574 | ||
a649216c | 1575 | show_entries_rb(0); |
cb2a1a62 | 1576 | |
7d747d22 | 1577 | } while (events_added); |
d5396421 | 1578 | |
a649216c JA |
1579 | if (rb_sort_entries) |
1580 | show_entries_rb(1); | |
1581 | ||
7d747d22 | 1582 | return 0; |
412819ce | 1583 | } |
d5396421 | 1584 | |
412819ce JA |
1585 | static int do_stdin(void) |
1586 | { | |
53c68c88 | 1587 | unsigned long long youngest; |
412819ce | 1588 | int fd; |
d5396421 | 1589 | |
be925321 | 1590 | last_allowed_time = -1ULL; |
1f79c4a0 | 1591 | fd = dup(STDIN_FILENO); |
412819ce JA |
1592 | do { |
1593 | int events; | |
d5396421 | 1594 | |
a649216c | 1595 | events = read_events(fd, 0); |
412819ce JA |
1596 | if (!events) |
1597 | break; | |
1598 | ||
53c68c88 JA |
1599 | if (sort_entries(&youngest)) |
1600 | break; | |
1601 | ||
1602 | if (youngest > stopwatch_end) | |
2ff323b0 JA |
1603 | break; |
1604 | ||
a649216c | 1605 | show_entries_rb(0); |
d5396421 JA |
1606 | } while (1); |
1607 | ||
a649216c JA |
1608 | if (rb_sort_entries) |
1609 | show_entries_rb(1); | |
1610 | ||
d5396421 | 1611 | close(fd); |
d5396421 JA |
1612 | return 0; |
1613 | } | |
d0ca268b | 1614 | |
1f79c4a0 | 1615 | static void flush_output(void) |
412819ce | 1616 | { |
152f6476 | 1617 | fflush(ofp); |
412819ce JA |
1618 | } |
1619 | ||
e820abd7 | 1620 | static void handle_sigint(__attribute__((__unused__)) int sig) |
412819ce JA |
1621 | { |
1622 | done = 1; | |
1623 | flush_output(); | |
1624 | } | |
1625 | ||
46e6968b NS |
1626 | /* |
1627 | * Extract start and duration times from a string, allowing | |
1628 | * us to specify a time interval of interest within a trace. | |
1629 | * Format: "duration" (start is zero) or "start:duration". | |
1630 | */ | |
1631 | static int find_stopwatch_interval(char *string) | |
1632 | { | |
1633 | double value; | |
1634 | char *sp; | |
1635 | ||
1636 | value = strtod(string, &sp); | |
1637 | if (sp == string) { | |
1638 | fprintf(stderr,"Invalid stopwatch timer: %s\n", string); | |
1639 | return 1; | |
1640 | } | |
1641 | if (*sp == ':') { | |
1642 | stopwatch_start = DOUBLE_TO_NANO_ULL(value); | |
1643 | string = sp + 1; | |
1644 | value = strtod(string, &sp); | |
1645 | if (sp == string || *sp != '\0') { | |
1646 | fprintf(stderr,"Invalid stopwatch duration time: %s\n", | |
1647 | string); | |
1648 | return 1; | |
1649 | } | |
1650 | } else if (*sp != '\0') { | |
1651 | fprintf(stderr,"Invalid stopwatch start timer: %s\n", string); | |
1652 | return 1; | |
1653 | } | |
1b928247 JA |
1654 | stopwatch_end = DOUBLE_TO_NANO_ULL(value); |
1655 | if (stopwatch_end <= stopwatch_start) { | |
1656 | fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n", | |
1657 | stopwatch_start, stopwatch_end); | |
1658 | return 1; | |
1659 | } | |
1660 | ||
46e6968b NS |
1661 | return 0; |
1662 | } | |
1663 | ||
52724a0e JA |
1664 | static char usage_str[] = \ |
1665 | "[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \ | |
1666 | "[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \ | |
1667 | "\t-i Input file containing trace data, or '-' for stdin\n" \ | |
1668 | "\t-o Output file. If not given, output is stdout\n" \ | |
1669 | "\t-b stdin read batching\n" \ | |
1670 | "\t-s Show per-program io statistics\n" \ | |
bf0720af | 1671 | "\t-n Hash processes by name, not pid\n" \ |
52724a0e JA |
1672 | "\t-t Track individual ios. Will tell you the time a request took\n" \ |
1673 | "\t to get queued, to get dispatched, and to get completed\n" \ | |
1674 | "\t-q Quiet. Don't display any stats at the end of the trace\n" \ | |
1675 | "\t-w Only parse data between the given time interval in seconds.\n" \ | |
1676 | "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \ | |
1677 | "\t -f Output format. Customize the output format. The format field\n" \ | |
1678 | "\t identifies can be found in the documentation\n" \ | |
1679 | "\t-F Format specification. Can be found in the documentation\n" \ | |
7d1c0411 | 1680 | "\t-m Print missing entries\n" \ |
52724a0e JA |
1681 | "\t-v Print program version info\n\n"; |
1682 | ||
1f79c4a0 JA |
1683 | static void usage(char *prog) |
1684 | { | |
52724a0e | 1685 | fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str); |
1f79c4a0 JA |
1686 | } |
1687 | ||
d5396421 JA |
1688 | int main(int argc, char *argv[]) |
1689 | { | |
152f6476 | 1690 | char *ofp_buffer; |
a66877e6 | 1691 | int c, ret, mode; |
1e1c60f1 | 1692 | int per_device_and_cpu_stats = 1; |
d5396421 JA |
1693 | |
1694 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) { | |
1695 | switch (c) { | |
1696 | case 'i': | |
e7c9f3ff NS |
1697 | if (!strcmp(optarg, "-") && !pipeline) |
1698 | pipeline = 1; | |
1699 | else if (resize_devices(optarg) != 0) | |
1700 | return 1; | |
d5396421 JA |
1701 | break; |
1702 | case 'o': | |
66efebf8 | 1703 | output_name = optarg; |
d5396421 | 1704 | break; |
79f19470 JA |
1705 | case 'b': |
1706 | rb_batch = atoi(optarg); | |
1707 | if (rb_batch <= 0) | |
1708 | rb_batch = RB_BATCH_DEFAULT; | |
1709 | break; | |
152f6476 JA |
1710 | case 's': |
1711 | per_process_stats = 1; | |
1712 | break; | |
7997c5b0 JA |
1713 | case 't': |
1714 | track_ios = 1; | |
1715 | break; | |
1e1c60f1 NS |
1716 | case 'q': |
1717 | per_device_and_cpu_stats = 0; | |
1718 | break; | |
46e6968b NS |
1719 | case 'w': |
1720 | if (find_stopwatch_interval(optarg) != 0) | |
1721 | return 1; | |
1722 | break; | |
ab197ca7 AB |
1723 | case 'f': |
1724 | set_all_format_specs(optarg); | |
1725 | break; | |
1726 | case 'F': | |
1727 | if (add_format_spec(optarg) != 0) | |
1728 | return 1; | |
1729 | break; | |
bf0720af | 1730 | case 'n': |
715d8021 | 1731 | ppi_hash_by_pid = 0; |
bf0720af | 1732 | break; |
7d1c0411 JA |
1733 | case 'm': |
1734 | print_missing = 1; | |
1735 | break; | |
52724a0e JA |
1736 | case 'v': |
1737 | printf("%s version %s\n", argv[0], blkparse_version); | |
1738 | return 0; | |
d5396421 | 1739 | default: |
1f79c4a0 | 1740 | usage(argv[0]); |
d5396421 JA |
1741 | return 1; |
1742 | } | |
d0ca268b JA |
1743 | } |
1744 | ||
e7c9f3ff NS |
1745 | while (optind < argc) { |
1746 | if (!strcmp(argv[optind], "-") && !pipeline) | |
1747 | pipeline = 1; | |
1748 | else if (resize_devices(argv[optind]) != 0) | |
1749 | return 1; | |
1750 | optind++; | |
1751 | } | |
1752 | ||
1753 | if (!pipeline && !ndevices) { | |
1f79c4a0 | 1754 | usage(argv[0]); |
d5396421 JA |
1755 | return 1; |
1756 | } | |
1757 | ||
7997c5b0 | 1758 | memset(&rb_sort_root, 0, sizeof(rb_sort_root)); |
412819ce JA |
1759 | |
1760 | signal(SIGINT, handle_sigint); | |
1761 | signal(SIGHUP, handle_sigint); | |
1762 | signal(SIGTERM, handle_sigint); | |
d5396421 | 1763 | |
d69db225 JA |
1764 | setlocale(LC_NUMERIC, "en_US"); |
1765 | ||
a66877e6 | 1766 | if (!output_name) { |
152f6476 | 1767 | ofp = fdopen(STDOUT_FILENO, "w"); |
a66877e6 JA |
1768 | mode = _IOLBF; |
1769 | } else { | |
152f6476 JA |
1770 | char ofname[128]; |
1771 | ||
7a9690c0 | 1772 | snprintf(ofname, sizeof(ofname) - 1, "%s", output_name); |
152f6476 | 1773 | ofp = fopen(ofname, "w"); |
a66877e6 | 1774 | mode = _IOFBF; |
152f6476 JA |
1775 | } |
1776 | ||
1777 | if (!ofp) { | |
1778 | perror("fopen"); | |
1779 | return 1; | |
1780 | } | |
1781 | ||
1782 | ofp_buffer = malloc(4096); | |
a66877e6 | 1783 | if (setvbuf(ofp, ofp_buffer, mode, 4096)) { |
152f6476 JA |
1784 | perror("setvbuf"); |
1785 | return 1; | |
1786 | } | |
1787 | ||
e7c9f3ff | 1788 | if (pipeline) |
d5396421 JA |
1789 | ret = do_stdin(); |
1790 | else | |
1791 | ret = do_file(); | |
1792 | ||
152f6476 JA |
1793 | if (per_process_stats) |
1794 | show_process_stats(); | |
1795 | ||
1e1c60f1 NS |
1796 | if (per_device_and_cpu_stats) |
1797 | show_device_and_cpu_stats(); | |
152f6476 | 1798 | |
412819ce | 1799 | flush_output(); |
d5396421 | 1800 | return ret; |
d0ca268b | 1801 | } |