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> |
d0ca268b | 32 | |
8fc0abbc JA |
33 | #include "blktrace.h" |
34 | #include "rbtree.h" | |
d0ca268b | 35 | |
2e3e8ded JA |
36 | #define SECONDS(x) ((unsigned long long)(x) / 1000000000) |
37 | #define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000) | |
cfab07eb | 38 | |
d5396421 JA |
39 | static int backwards; |
40 | static unsigned long long genesis_time, last_reported_time; | |
41 | ||
152f6476 JA |
42 | struct io_stats { |
43 | unsigned long qreads, qwrites, creads, cwrites, mreads, mwrites; | |
44 | unsigned long ireads, iwrites; | |
45 | unsigned long long qread_kb, qwrite_kb, cread_kb, cwrite_kb; | |
46 | unsigned long long iread_kb, iwrite_kb; | |
47 | }; | |
48 | ||
d5396421 | 49 | struct per_cpu_info { |
d0ca268b JA |
50 | int cpu; |
51 | int nelems; | |
d0ca268b JA |
52 | |
53 | int fd; | |
87b72777 | 54 | char fname[128]; |
d0ca268b | 55 | |
152f6476 JA |
56 | struct io_stats io_stats; |
57 | }; | |
8fc0abbc | 58 | |
152f6476 JA |
59 | struct per_process_info { |
60 | char name[16]; | |
61 | __u32 pid; | |
62 | struct io_stats io_stats; | |
63 | struct per_process_info *hash_next, *list_next; | |
50adc0ba JA |
64 | |
65 | /* | |
66 | * individual io stats | |
67 | */ | |
b9d40d6f JA |
68 | unsigned long long longest_allocation_wait[2]; |
69 | unsigned long long longest_dispatch_wait[2]; | |
70 | unsigned long long longest_completion_wait[2]; | |
d0ca268b JA |
71 | }; |
72 | ||
152f6476 JA |
73 | #define PPI_HASH_SHIFT (8) |
74 | static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT]; | |
75 | static struct per_process_info *ppi_list; | |
76 | ||
7997c5b0 | 77 | #define S_OPTS "i:o:b:st" |
d5396421 JA |
78 | static struct option l_opts[] = { |
79 | { | |
80 | .name = "input", | |
81 | .has_arg = 1, | |
82 | .flag = NULL, | |
83 | .val = 'i' | |
84 | }, | |
85 | { | |
86 | .name = "output", | |
87 | .has_arg = 1, | |
88 | .flag = NULL, | |
89 | .val = 'o' | |
90 | }, | |
79f19470 JA |
91 | { |
92 | .name = "batch", | |
93 | .has_arg = 1, | |
94 | .flag = NULL, | |
95 | .val = 'b' | |
96 | }, | |
152f6476 JA |
97 | { |
98 | .name = "per program stats", | |
99 | .has_arg = 0, | |
100 | .flag = NULL, | |
101 | .val = 's' | |
102 | }, | |
7997c5b0 JA |
103 | { |
104 | .name = "track ios", | |
105 | .has_arg = 0, | |
106 | .flag = NULL, | |
107 | .val = 't' | |
108 | }, | |
d5396421 JA |
109 | { |
110 | .name = NULL, | |
111 | .has_arg = 0, | |
112 | .flag = NULL, | |
113 | .val = 0 | |
114 | } | |
115 | }; | |
116 | ||
7997c5b0 JA |
117 | static struct rb_root rb_sort_root; |
118 | static struct rb_root rb_track_root; | |
8fc0abbc | 119 | |
7997c5b0 JA |
120 | /* |
121 | * for sorting the displayed output | |
122 | */ | |
8fc0abbc JA |
123 | struct trace { |
124 | struct blk_io_trace *bit; | |
125 | struct rb_node rb_node; | |
126 | }; | |
127 | ||
7997c5b0 JA |
128 | /* |
129 | * for tracking individual ios | |
130 | */ | |
131 | struct io_track { | |
132 | struct rb_node rb_node; | |
133 | ||
134 | __u64 sector; | |
135 | __u32 pid; | |
95c15013 | 136 | unsigned long long allocation_time; |
7997c5b0 JA |
137 | unsigned long long queue_time; |
138 | unsigned long long dispatch_time; | |
139 | unsigned long long completion_time; | |
140 | }; | |
141 | ||
a718bd37 NS |
142 | static int max_cpus; |
143 | static struct per_cpu_info *per_cpu_info; | |
d0ca268b | 144 | |
87b72777 JA |
145 | static unsigned long long events; |
146 | ||
d5396421 | 147 | static char *dev, *output_name; |
152f6476 JA |
148 | static FILE *ofp; |
149 | ||
150 | static int per_process_stats; | |
7997c5b0 | 151 | static int track_ios; |
d0ca268b | 152 | |
79f19470 JA |
153 | #define RB_BATCH_DEFAULT (1024) |
154 | static int rb_batch = RB_BATCH_DEFAULT; | |
155 | ||
412819ce JA |
156 | #define is_done() (*(volatile int *)(&done)) |
157 | static volatile int done; | |
158 | ||
152f6476 JA |
159 | static inline unsigned long hash_long(unsigned long val) |
160 | { | |
16ef714e JA |
161 | #if __WORDSIZE == 32 |
162 | val *= 0x9e370001UL; | |
163 | #elif __WORDSIZE == 64 | |
164 | val *= 0x9e37fffffffc0001UL; | |
165 | #else | |
166 | #error unknown word size | |
167 | #endif | |
168 | ||
169 | return val >> (__WORDSIZE - PPI_HASH_SHIFT); | |
152f6476 JA |
170 | } |
171 | ||
172 | static inline void add_process_to_hash(struct per_process_info *ppi) | |
173 | { | |
174 | const int hash_idx = hash_long(ppi->pid); | |
175 | ||
176 | ppi->hash_next = ppi_hash[hash_idx]; | |
177 | ppi_hash[hash_idx] = ppi; | |
178 | } | |
179 | ||
180 | static inline void add_process_to_list(struct per_process_info *ppi) | |
181 | { | |
182 | ppi->list_next = ppi_list; | |
183 | ppi_list = ppi; | |
184 | } | |
185 | ||
186 | static struct per_process_info *find_process_by_pid(__u32 pid) | |
187 | { | |
188 | const int hash_idx = hash_long(pid); | |
189 | struct per_process_info *ppi; | |
190 | ||
191 | ppi = ppi_hash[hash_idx]; | |
192 | while (ppi) { | |
193 | if (ppi->pid == pid) | |
194 | return ppi; | |
195 | ||
196 | ppi = ppi->hash_next; | |
197 | } | |
198 | ||
199 | return NULL; | |
200 | } | |
201 | ||
7997c5b0 JA |
202 | static inline int trace_rb_insert(struct trace *t) |
203 | { | |
204 | struct rb_node **p = &rb_sort_root.rb_node; | |
205 | struct rb_node *parent = NULL; | |
206 | struct trace *__t; | |
207 | ||
208 | while (*p) { | |
209 | parent = *p; | |
210 | __t = rb_entry(parent, struct trace, rb_node); | |
211 | ||
212 | if (t->bit->sequence < __t->bit->sequence) | |
213 | p = &(*p)->rb_left; | |
214 | else if (t->bit->sequence > __t->bit->sequence) | |
215 | p = &(*p)->rb_right; | |
216 | else { | |
217 | fprintf(stderr, "sequence alias!\n"); | |
218 | return 1; | |
219 | } | |
220 | } | |
221 | ||
222 | rb_link_node(&t->rb_node, parent, p); | |
223 | rb_insert_color(&t->rb_node, &rb_sort_root); | |
224 | return 0; | |
225 | } | |
226 | ||
227 | static inline int track_rb_insert(struct io_track *iot) | |
228 | { | |
229 | struct rb_node **p = &rb_track_root.rb_node; | |
230 | struct rb_node *parent = NULL; | |
231 | struct io_track *__iot; | |
232 | ||
233 | while (*p) { | |
234 | parent = *p; | |
235 | ||
236 | __iot = rb_entry(parent, struct io_track, rb_node); | |
237 | ||
238 | if (iot->sector < __iot->sector) | |
239 | p = &(*p)->rb_left; | |
240 | else if (iot->sector > __iot->sector) | |
241 | p = &(*p)->rb_right; | |
242 | else { | |
243 | fprintf(stderr, "sequence alias!\n"); | |
244 | return 1; | |
245 | } | |
246 | } | |
247 | ||
248 | rb_link_node(&iot->rb_node, parent, p); | |
249 | rb_insert_color(&iot->rb_node, &rb_track_root); | |
250 | return 0; | |
251 | } | |
252 | ||
253 | static struct io_track *__find_track(__u64 sector) | |
254 | { | |
255 | struct rb_node **p = &rb_track_root.rb_node; | |
256 | struct rb_node *parent = NULL; | |
257 | struct io_track *__iot; | |
258 | ||
259 | while (*p) { | |
260 | parent = *p; | |
261 | ||
262 | __iot = rb_entry(parent, struct io_track, rb_node); | |
263 | ||
264 | if (sector < __iot->sector) | |
265 | p = &(*p)->rb_left; | |
266 | else if (sector > __iot->sector) | |
267 | p = &(*p)->rb_right; | |
268 | else | |
269 | return __iot; | |
270 | } | |
271 | ||
272 | return NULL; | |
273 | } | |
274 | ||
50adc0ba | 275 | static struct io_track *find_track(__u32 pid, __u64 sector) |
7997c5b0 | 276 | { |
916b5501 | 277 | struct io_track *iot; |
7997c5b0 JA |
278 | |
279 | iot = __find_track(sector); | |
280 | if (!iot) { | |
281 | iot = malloc(sizeof(*iot)); | |
50adc0ba | 282 | iot->pid = pid; |
7997c5b0 JA |
283 | iot->sector = sector; |
284 | track_rb_insert(iot); | |
285 | } | |
286 | ||
287 | return iot; | |
288 | } | |
289 | ||
2e3e8ded JA |
290 | static void log_track_merge(struct blk_io_trace *t) |
291 | { | |
292 | struct io_track *iot; | |
293 | ||
294 | if (!track_ios) | |
295 | return; | |
296 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
297 | return; | |
298 | ||
299 | iot = __find_track(t->sector - (t->bytes >> 10)); | |
300 | if (!iot) { | |
301 | fprintf(stderr, "Trying to merge on non-existing request\n"); | |
302 | return; | |
303 | } | |
304 | ||
305 | rb_erase(&iot->rb_node, &rb_track_root); | |
306 | iot->sector -= t->bytes >> 10; | |
307 | track_rb_insert(iot); | |
308 | } | |
309 | ||
95c15013 | 310 | static void log_track_getrq(struct blk_io_trace *t) |
2e3e8ded JA |
311 | { |
312 | struct io_track *iot; | |
313 | ||
314 | if (!track_ios) | |
315 | return; | |
316 | ||
50adc0ba | 317 | iot = find_track(t->pid, t->sector); |
95c15013 JA |
318 | iot->allocation_time = t->time; |
319 | } | |
320 | ||
321 | ||
322 | /* | |
323 | * return time between rq allocation and queue | |
324 | */ | |
325 | static unsigned long long log_track_queue(struct blk_io_trace *t) | |
326 | { | |
50adc0ba | 327 | unsigned long long elapsed; |
95c15013 JA |
328 | struct io_track *iot; |
329 | ||
330 | if (!track_ios) | |
331 | return -1; | |
332 | ||
50adc0ba | 333 | iot = find_track(t->pid, t->sector); |
2e3e8ded | 334 | iot->queue_time = t->time; |
50adc0ba JA |
335 | elapsed = iot->queue_time - iot->allocation_time; |
336 | ||
337 | if (per_process_stats) { | |
338 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 339 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 340 | |
b9d40d6f JA |
341 | if (ppi && elapsed > ppi->longest_allocation_wait[w]) |
342 | ppi->longest_allocation_wait[w] = elapsed; | |
50adc0ba JA |
343 | } |
344 | ||
345 | return elapsed; | |
2e3e8ded JA |
346 | } |
347 | ||
348 | /* | |
349 | * return time between queue and issue | |
350 | */ | |
351 | static unsigned long long log_track_issue(struct blk_io_trace *t) | |
352 | { | |
50adc0ba | 353 | unsigned long long elapsed; |
2e3e8ded JA |
354 | struct io_track *iot; |
355 | ||
356 | if (!track_ios) | |
357 | return -1; | |
358 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
359 | return -1; | |
360 | ||
361 | iot = __find_track(t->sector); | |
362 | if (!iot) { | |
363 | fprintf(stderr, "Trying to issue on non-existing request\n"); | |
364 | return -1; | |
365 | } | |
366 | ||
367 | iot->dispatch_time = t->time; | |
50adc0ba JA |
368 | elapsed = iot->dispatch_time - iot->queue_time; |
369 | ||
370 | if (per_process_stats) { | |
371 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 372 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 373 | |
b9d40d6f JA |
374 | if (ppi && elapsed > ppi->longest_dispatch_wait[w]) |
375 | ppi->longest_dispatch_wait[w] = elapsed; | |
50adc0ba JA |
376 | } |
377 | ||
378 | return elapsed; | |
2e3e8ded JA |
379 | } |
380 | ||
381 | /* | |
382 | * return time between dispatch and complete | |
383 | */ | |
384 | static unsigned long long log_track_complete(struct blk_io_trace *t) | |
385 | { | |
386 | unsigned long long elapsed; | |
387 | struct io_track *iot; | |
388 | ||
389 | if (!track_ios) | |
390 | return -1; | |
391 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
392 | return -1; | |
393 | ||
394 | iot = __find_track(t->sector); | |
395 | if (!iot) { | |
396 | fprintf(stderr, "Trying to dispatch on non-existing request\n"); | |
397 | return -1; | |
398 | } | |
399 | ||
400 | iot->completion_time = t->time; | |
401 | elapsed = iot->completion_time - iot->dispatch_time; | |
402 | ||
50adc0ba JA |
403 | if (per_process_stats) { |
404 | struct per_process_info *ppi = find_process_by_pid(iot->pid); | |
b9d40d6f | 405 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 406 | |
b9d40d6f JA |
407 | if (ppi && elapsed > ppi->longest_completion_wait[w]) |
408 | ppi->longest_completion_wait[w] = elapsed; | |
50adc0ba JA |
409 | } |
410 | ||
2e3e8ded JA |
411 | /* |
412 | * kill the trace, we don't need it after completion | |
413 | */ | |
414 | rb_erase(&iot->rb_node, &rb_track_root); | |
415 | free(iot); | |
416 | ||
417 | return elapsed; | |
418 | } | |
419 | ||
420 | ||
152f6476 JA |
421 | static struct io_stats *find_process_io_stats(__u32 pid, char *name) |
422 | { | |
423 | struct per_process_info *ppi = find_process_by_pid(pid); | |
424 | ||
425 | if (!ppi) { | |
426 | ppi = malloc(sizeof(*ppi)); | |
427 | memset(ppi, 0, sizeof(*ppi)); | |
428 | strncpy(ppi->name, name, sizeof(ppi->name)); | |
429 | ppi->pid = pid; | |
430 | add_process_to_hash(ppi); | |
431 | add_process_to_list(ppi); | |
432 | } | |
433 | ||
434 | return &ppi->io_stats; | |
435 | } | |
436 | ||
a718bd37 NS |
437 | static void resize_cpu_info(int cpuid) |
438 | { | |
439 | int new_space, new_max = cpuid + 1; | |
440 | char *new_start; | |
441 | ||
442 | per_cpu_info = realloc(per_cpu_info, new_max * sizeof(*per_cpu_info)); | |
443 | if (!per_cpu_info) { | |
444 | fprintf(stderr, "Cannot allocate CPU info -- %d x %d bytes\n", | |
445 | new_max, (int) sizeof(*per_cpu_info)); | |
446 | exit(1); | |
447 | } | |
448 | ||
449 | new_start = (char *)per_cpu_info + (max_cpus * sizeof(*per_cpu_info)); | |
450 | new_space = (new_max - max_cpus) * sizeof(*per_cpu_info); | |
451 | memset(new_start, 0, new_space); | |
452 | max_cpus = new_max; | |
453 | } | |
454 | ||
455 | static struct per_cpu_info *get_cpu_info(int cpu) | |
456 | { | |
c499bf38 JA |
457 | struct per_cpu_info *pci; |
458 | ||
a718bd37 NS |
459 | if (cpu >= max_cpus) |
460 | resize_cpu_info(cpu); | |
461 | ||
c499bf38 JA |
462 | /* |
463 | * ->cpu might already be set, but just set it unconditionally | |
464 | */ | |
465 | pci = &per_cpu_info[cpu]; | |
466 | pci->cpu = cpu; | |
467 | ||
468 | return pci; | |
a718bd37 NS |
469 | } |
470 | ||
cfab07eb AB |
471 | static inline void check_time(struct blk_io_trace *bit) |
472 | { | |
473 | unsigned long long this = bit->time; | |
474 | unsigned long long last = last_reported_time; | |
475 | ||
476 | backwards = (this < last) ? 'B' : ' '; | |
477 | last_reported_time = this; | |
478 | } | |
479 | ||
152f6476 JA |
480 | static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t, |
481 | int rw) | |
d0ca268b JA |
482 | { |
483 | if (rw) { | |
152f6476 JA |
484 | ios->mwrites++; |
485 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 486 | } else { |
152f6476 JA |
487 | ios->mreads++; |
488 | ios->qread_kb += t->bytes >> 10; | |
489 | } | |
490 | } | |
491 | ||
492 | static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci, | |
493 | int rw) | |
494 | { | |
495 | __account_m(&pci->io_stats, t, rw); | |
496 | ||
497 | if (per_process_stats) { | |
498 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
499 | ||
500 | __account_m(ios, t, rw); | |
d0ca268b JA |
501 | } |
502 | } | |
503 | ||
152f6476 JA |
504 | static inline void __account_q(struct io_stats *ios, struct blk_io_trace *t, |
505 | int rw) | |
d0ca268b JA |
506 | { |
507 | if (rw) { | |
152f6476 JA |
508 | ios->qwrites++; |
509 | ios->qwrite_kb += t->bytes >> 10; | |
d0ca268b | 510 | } else { |
152f6476 JA |
511 | ios->qreads++; |
512 | ios->qread_kb += t->bytes >> 10; | |
513 | } | |
514 | } | |
515 | ||
516 | static inline void account_q(struct blk_io_trace *t, struct per_cpu_info *pci, | |
517 | int rw) | |
518 | { | |
519 | __account_q(&pci->io_stats, t, rw); | |
520 | ||
521 | if (per_process_stats) { | |
522 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
523 | ||
524 | __account_q(ios, t, rw); | |
d0ca268b JA |
525 | } |
526 | } | |
527 | ||
152f6476 | 528 | static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes) |
d0ca268b JA |
529 | { |
530 | if (rw) { | |
152f6476 JA |
531 | ios->cwrites++; |
532 | ios->cwrite_kb += bytes >> 10; | |
d0ca268b | 533 | } else { |
152f6476 JA |
534 | ios->creads++; |
535 | ios->cread_kb += bytes >> 10; | |
536 | } | |
537 | } | |
538 | ||
539 | static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci, | |
540 | int rw, int bytes) | |
541 | { | |
542 | __account_c(&pci->io_stats, rw, bytes); | |
543 | ||
544 | if (per_process_stats) { | |
545 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
546 | ||
547 | __account_c(ios, rw, bytes); | |
d0ca268b JA |
548 | } |
549 | } | |
550 | ||
152f6476 | 551 | static inline void __account_i(struct io_stats *ios, int rw, unsigned int bytes) |
afd2d7ad | 552 | { |
553 | if (rw) { | |
152f6476 JA |
554 | ios->iwrites++; |
555 | ios->iwrite_kb += bytes >> 10; | |
afd2d7ad | 556 | } else { |
152f6476 JA |
557 | ios->ireads++; |
558 | ios->iread_kb += bytes >> 10; | |
afd2d7ad | 559 | } |
560 | } | |
561 | ||
152f6476 JA |
562 | static inline void account_i(struct blk_io_trace *t, struct per_cpu_info *pci, |
563 | int rw) | |
d0ca268b | 564 | { |
152f6476 JA |
565 | __account_i(&pci->io_stats, rw, t->bytes); |
566 | ||
567 | if (per_process_stats) { | |
568 | struct io_stats *ios = find_process_io_stats(t->pid, t->comm); | |
d5396421 | 569 | |
152f6476 JA |
570 | __account_i(ios, rw, t->bytes); |
571 | } | |
572 | } | |
573 | ||
574 | static void output(struct per_cpu_info *pci, char *s) | |
575 | { | |
576 | fprintf(ofp, "%s", s); | |
d0ca268b JA |
577 | } |
578 | ||
3aabcd89 JA |
579 | static char hstring[256]; |
580 | static char tstring[256]; | |
d0ca268b | 581 | |
d5396421 JA |
582 | static inline char *setup_header(struct per_cpu_info *pci, |
583 | struct blk_io_trace *t, char act) | |
d0ca268b JA |
584 | { |
585 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
586 | int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER); | |
587 | int s = t->action & BLK_TC_ACT(BLK_TC_SYNC); | |
588 | char rwbs[4]; | |
589 | int i = 0; | |
590 | ||
591 | if (w) | |
592 | rwbs[i++] = 'W'; | |
593 | else | |
594 | rwbs[i++] = 'R'; | |
595 | if (b) | |
596 | rwbs[i++] = 'B'; | |
597 | if (s) | |
598 | rwbs[i++] = 'S'; | |
599 | ||
600 | rwbs[i] = '\0'; | |
601 | ||
555d3a31 | 602 | sprintf(hstring, "%2d %8ld %5Lu.%09Lu %5u %c %3s", |
d5396421 | 603 | pci->cpu, |
cfab07eb AB |
604 | (unsigned long)t->sequence, SECONDS(t->time), |
605 | NANO_SECONDS(t->time), t->pid, act, rwbs); | |
d0ca268b JA |
606 | |
607 | return hstring; | |
608 | } | |
609 | ||
d5396421 JA |
610 | static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t, |
611 | char act) | |
d0ca268b | 612 | { |
2e3e8ded JA |
613 | unsigned long long elapsed = log_track_complete(t); |
614 | ||
615 | if (elapsed != -1ULL) { | |
b9d40d6f | 616 | unsigned long usec = elapsed / 1000; |
2e3e8ded | 617 | |
b9d40d6f | 618 | sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n", |
2e3e8ded JA |
619 | setup_header(pci, t, act), |
620 | (unsigned long long)t->sector, t->bytes >> 9, | |
621 | usec, t->error); | |
622 | } else { | |
623 | sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act), | |
624 | (unsigned long long)t->sector, t->bytes >> 9, t->error); | |
625 | } | |
626 | ||
d5396421 | 627 | output(pci, tstring); |
d0ca268b JA |
628 | } |
629 | ||
d5396421 JA |
630 | static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t, |
631 | char act) | |
d0ca268b | 632 | { |
95c15013 | 633 | unsigned long long elapsed = log_track_queue(t); |
2e3e8ded | 634 | |
95c15013 | 635 | if (elapsed != -1ULL) { |
b9d40d6f | 636 | unsigned long usec = elapsed / 1000; |
95c15013 | 637 | |
b9d40d6f | 638 | sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n", |
95c15013 JA |
639 | setup_header(pci, t, act), |
640 | (unsigned long long)t->sector, t->bytes >> 9, | |
641 | usec, t->comm); | |
642 | } else { | |
643 | sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act), | |
644 | (unsigned long long)t->sector, t->bytes >> 9, t->comm); | |
645 | } | |
d5396421 | 646 | output(pci, tstring); |
d0ca268b JA |
647 | } |
648 | ||
d5396421 JA |
649 | static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t, |
650 | char act) | |
d0ca268b | 651 | { |
2e3e8ded JA |
652 | unsigned long long elapsed = log_track_issue(t); |
653 | ||
654 | if (elapsed != -1ULL) { | |
655 | double usec = (double) elapsed / 1000; | |
656 | ||
555d3a31 | 657 | sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n", |
2e3e8ded JA |
658 | setup_header(pci, t, act), |
659 | (unsigned long long)t->sector, t->bytes >> 9, | |
660 | usec, t->comm); | |
661 | } else { | |
662 | sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act), | |
663 | (unsigned long long)t->sector, t->bytes >> 9, t->comm); | |
664 | } | |
665 | ||
d5396421 | 666 | output(pci, tstring); |
d0ca268b JA |
667 | } |
668 | ||
d5396421 JA |
669 | static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t, |
670 | char act) | |
d0ca268b | 671 | { |
2e3e8ded JA |
672 | log_track_merge(t); |
673 | ||
984c63b7 | 674 | sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act), |
2955af9d | 675 | (unsigned long long)t->sector, t->bytes >> 9, t->comm); |
d5396421 | 676 | output(pci, tstring); |
d0ca268b JA |
677 | } |
678 | ||
d5396421 JA |
679 | static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t, |
680 | char act) | |
d0ca268b | 681 | { |
2955af9d NS |
682 | sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act), |
683 | (unsigned long long)t->sector, t->bytes >> 9, t->comm); | |
d5396421 | 684 | output(pci, tstring); |
d0ca268b JA |
685 | } |
686 | ||
d5396421 | 687 | static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char act) |
d0ca268b | 688 | { |
87b72777 JA |
689 | unsigned char *buf; |
690 | int i; | |
d0ca268b | 691 | |
d5396421 JA |
692 | sprintf(tstring,"%s ", setup_header(pci, t, act)); |
693 | output(pci, tstring); | |
d0ca268b | 694 | |
87b72777 | 695 | buf = (unsigned char *) t + sizeof(*t); |
d0ca268b JA |
696 | for (i = 0; i < t->pdu_len; i++) { |
697 | sprintf(tstring,"%02x ", buf[i]); | |
d5396421 | 698 | output(pci, tstring); |
d0ca268b JA |
699 | } |
700 | ||
701 | if (act == 'C') { | |
2955af9d NS |
702 | sprintf(tstring,"[%d]\n", t->error); |
703 | output(pci, tstring); | |
704 | } else { | |
705 | sprintf(tstring,"[%s]\n", t->comm); | |
d5396421 | 706 | output(pci, tstring); |
d0ca268b | 707 | } |
87b72777 | 708 | return 0; |
d0ca268b JA |
709 | } |
710 | ||
d5396421 | 711 | static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 712 | { |
87b72777 JA |
713 | int ret = 0; |
714 | ||
d0ca268b JA |
715 | switch (t->action & 0xffff) { |
716 | case __BLK_TA_QUEUE: | |
d5396421 | 717 | log_generic(pci, t, 'Q'); |
d0ca268b JA |
718 | break; |
719 | case __BLK_TA_GETRQ: | |
d5396421 | 720 | log_generic(pci, t, 'G'); |
d0ca268b JA |
721 | break; |
722 | case __BLK_TA_SLEEPRQ: | |
d5396421 | 723 | log_generic(pci, t, 'S'); |
d0ca268b JA |
724 | break; |
725 | case __BLK_TA_REQUEUE: | |
d5396421 | 726 | log_generic(pci, t, 'R'); |
d0ca268b JA |
727 | break; |
728 | case __BLK_TA_ISSUE: | |
d5396421 | 729 | ret = log_pc(pci, t, 'D'); |
d0ca268b JA |
730 | break; |
731 | case __BLK_TA_COMPLETE: | |
d5396421 | 732 | log_pc(pci, t, 'C'); |
d0ca268b JA |
733 | break; |
734 | default: | |
735 | fprintf(stderr, "Bad pc action %x\n", t->action); | |
87b72777 JA |
736 | ret = 1; |
737 | break; | |
d0ca268b JA |
738 | } |
739 | ||
87b72777 | 740 | return ret; |
d0ca268b JA |
741 | } |
742 | ||
d5396421 | 743 | static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b JA |
744 | { |
745 | int w = t->action & BLK_TC_ACT(BLK_TC_WRITE); | |
7997c5b0 | 746 | int act = t->action & 0xffff; |
d0ca268b | 747 | |
7997c5b0 | 748 | switch (act) { |
d0ca268b | 749 | case __BLK_TA_QUEUE: |
152f6476 | 750 | account_q(t, pci, w); |
d5396421 | 751 | log_queue(pci, t, 'Q'); |
d0ca268b JA |
752 | break; |
753 | case __BLK_TA_BACKMERGE: | |
152f6476 | 754 | account_m(t, pci, w); |
d5396421 | 755 | log_merge(pci, t, 'M'); |
d0ca268b JA |
756 | break; |
757 | case __BLK_TA_FRONTMERGE: | |
152f6476 | 758 | account_m(t, pci, w); |
d5396421 | 759 | log_merge(pci, t, 'F'); |
d0ca268b JA |
760 | break; |
761 | case __BLK_TA_GETRQ: | |
95c15013 | 762 | log_track_getrq(t); |
d5396421 | 763 | log_generic(pci, t, 'G'); |
d0ca268b JA |
764 | break; |
765 | case __BLK_TA_SLEEPRQ: | |
d5396421 | 766 | log_generic(pci, t, 'S'); |
d0ca268b JA |
767 | break; |
768 | case __BLK_TA_REQUEUE: | |
152f6476 | 769 | account_c(t, pci, w, -t->bytes); |
d5396421 | 770 | log_queue(pci, t, 'R'); |
d0ca268b JA |
771 | break; |
772 | case __BLK_TA_ISSUE: | |
152f6476 | 773 | account_i(t, pci, w); |
d5396421 | 774 | log_issue(pci, t, 'D'); |
d0ca268b JA |
775 | break; |
776 | case __BLK_TA_COMPLETE: | |
152f6476 | 777 | account_c(t, pci, w, t->bytes); |
d5396421 | 778 | log_complete(pci, t, 'C'); |
d0ca268b JA |
779 | break; |
780 | default: | |
781 | fprintf(stderr, "Bad fs action %x\n", t->action); | |
1f79c4a0 | 782 | break; |
d0ca268b | 783 | } |
d0ca268b JA |
784 | } |
785 | ||
d5396421 | 786 | static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 787 | { |
87b72777 JA |
788 | int ret = 0; |
789 | ||
d0ca268b | 790 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) |
d5396421 | 791 | ret = dump_trace_pc(t, pci); |
d0ca268b | 792 | else |
d5396421 | 793 | dump_trace_fs(t, pci); |
87b72777 JA |
794 | |
795 | events++; | |
796 | return ret; | |
d0ca268b JA |
797 | } |
798 | ||
152f6476 | 799 | static void dump_io_stats(struct io_stats *ios, char *msg) |
5c017e4b | 800 | { |
152f6476 JA |
801 | fprintf(ofp, "%s\n", msg); |
802 | ||
803 | fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb); | |
804 | fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb); | |
0a6b8fc4 | 805 | |
152f6476 JA |
806 | fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb); |
807 | fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb); | |
808 | fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb); | |
809 | fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb); | |
810 | fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' '); | |
0a6b8fc4 | 811 | |
152f6476 | 812 | fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites); |
5c017e4b JA |
813 | } |
814 | ||
50adc0ba JA |
815 | static void dump_wait_stats(struct per_process_info *ppi) |
816 | { | |
b9d40d6f JA |
817 | unsigned long rawait = ppi->longest_allocation_wait[0] / 1000; |
818 | unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000; | |
819 | unsigned long rcwait = ppi->longest_completion_wait[0] / 1000; | |
820 | unsigned long wawait = ppi->longest_allocation_wait[1] / 1000; | |
821 | unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000; | |
822 | unsigned long wcwait = ppi->longest_completion_wait[1] / 1000; | |
823 | ||
824 | fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' '); | |
825 | fprintf(ofp, " Allocation wait: %'8lu\n", wawait); | |
826 | fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' '); | |
827 | fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait); | |
828 | fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' '); | |
829 | fprintf(ofp, " Completion wait: %'8lu\n", wcwait); | |
50adc0ba JA |
830 | } |
831 | ||
152f6476 JA |
832 | static void show_process_stats(void) |
833 | { | |
834 | struct per_process_info *ppi; | |
835 | ||
836 | ppi = ppi_list; | |
837 | while (ppi) { | |
838 | dump_io_stats(&ppi->io_stats, ppi->name); | |
50adc0ba | 839 | dump_wait_stats(ppi); |
152f6476 JA |
840 | ppi = ppi->list_next; |
841 | } | |
842 | ||
843 | fprintf(ofp, "\n"); | |
844 | } | |
845 | ||
846 | static void show_cpu_stats(void) | |
d0ca268b | 847 | { |
d5396421 | 848 | struct per_cpu_info foo, *pci; |
152f6476 | 849 | struct io_stats *ios; |
412819ce | 850 | int i, pci_events = 0; |
5c017e4b JA |
851 | |
852 | memset(&foo, 0, sizeof(foo)); | |
d0ca268b | 853 | |
a718bd37 | 854 | for (i = 0; i < max_cpus; i++) { |
152f6476 JA |
855 | char cpu[8]; |
856 | ||
d5396421 | 857 | pci = &per_cpu_info[i]; |
152f6476 | 858 | ios = &pci->io_stats; |
5c017e4b | 859 | |
d5396421 | 860 | if (!pci->nelems) |
afd2d7ad | 861 | continue; |
862 | ||
152f6476 JA |
863 | foo.io_stats.qreads += ios->qreads; |
864 | foo.io_stats.qwrites += ios->qwrites; | |
865 | foo.io_stats.creads += ios->creads; | |
866 | foo.io_stats.cwrites += ios->cwrites; | |
867 | foo.io_stats.mreads += ios->mreads; | |
868 | foo.io_stats.mwrites += ios->mwrites; | |
11e51068 NS |
869 | foo.io_stats.ireads += ios->ireads; |
870 | foo.io_stats.iwrites += ios->iwrites; | |
152f6476 JA |
871 | foo.io_stats.qread_kb += ios->qread_kb; |
872 | foo.io_stats.qwrite_kb += ios->qwrite_kb; | |
873 | foo.io_stats.cread_kb += ios->cread_kb; | |
874 | foo.io_stats.cwrite_kb += ios->cwrite_kb; | |
11e51068 NS |
875 | foo.io_stats.iread_kb += ios->iread_kb; |
876 | foo.io_stats.iwrite_kb += ios->iwrite_kb; | |
152f6476 JA |
877 | |
878 | snprintf(cpu, sizeof(cpu) - 1, "CPU%d:", i); | |
879 | dump_io_stats(ios, cpu); | |
412819ce | 880 | pci_events++; |
5c017e4b JA |
881 | } |
882 | ||
412819ce | 883 | if (pci_events > 1) { |
152f6476 JA |
884 | fprintf(ofp, "\n"); |
885 | dump_io_stats(&foo.io_stats, "Total:"); | |
5c017e4b | 886 | } |
d0ca268b | 887 | |
152f6476 | 888 | fprintf(ofp, "\nEvents: %'Lu\n", events); |
d0ca268b JA |
889 | } |
890 | ||
2ff323b0 JA |
891 | #define min(a, b) ((a) < (b) ? (a) : (b)) |
892 | ||
893 | static struct blk_io_trace *find_trace(void *p, unsigned long offset, int nr) | |
894 | { | |
895 | unsigned long max_offset = min(offset,nr * sizeof(struct blk_io_trace)); | |
896 | unsigned long off; | |
897 | struct blk_io_trace *bit; | |
898 | __u32 magic; | |
899 | ||
900 | for (off = 0; off < max_offset; off++) { | |
901 | bit = p + off; | |
902 | ||
903 | magic = be32_to_cpu(bit->magic); | |
904 | if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) | |
905 | return bit; | |
906 | } | |
907 | ||
908 | return NULL; | |
909 | } | |
910 | ||
412819ce | 911 | static int sort_entries(void *traces, unsigned long offset, int nr) |
8fc0abbc | 912 | { |
412819ce | 913 | struct per_cpu_info *pci; |
8fc0abbc JA |
914 | struct blk_io_trace *bit; |
915 | struct trace *t; | |
916 | void *start = traces; | |
917 | int nelems = 0; | |
918 | ||
6fe4709e | 919 | while (traces - start <= offset - sizeof(*bit)) { |
412819ce JA |
920 | if (!nr) |
921 | break; | |
922 | ||
2ff323b0 JA |
923 | bit = find_trace(traces, offset - (traces - start), nr); |
924 | if (!bit) | |
925 | break; | |
6fe4709e | 926 | |
8fc0abbc JA |
927 | t = malloc(sizeof(*t)); |
928 | t->bit = bit; | |
929 | memset(&t->rb_node, 0, sizeof(t->rb_node)); | |
930 | ||
6fe4709e JA |
931 | trace_to_cpu(bit); |
932 | ||
66fa7233 JA |
933 | if (verify_trace(bit)) |
934 | break; | |
935 | ||
a718bd37 | 936 | pci = get_cpu_info(bit->cpu); |
412819ce JA |
937 | pci->nelems++; |
938 | ||
8fc0abbc JA |
939 | if (trace_rb_insert(t)) |
940 | return -1; | |
941 | ||
942 | traces += sizeof(*bit) + bit->pdu_len; | |
943 | nelems++; | |
412819ce | 944 | nr--; |
6fe4709e | 945 | } |
8fc0abbc JA |
946 | |
947 | return nelems; | |
948 | } | |
949 | ||
412819ce JA |
950 | static void free_entries_rb(void) |
951 | { | |
952 | struct rb_node *n; | |
953 | ||
7997c5b0 | 954 | while ((n = rb_first(&rb_sort_root)) != NULL) { |
412819ce JA |
955 | struct trace *t = rb_entry(n, struct trace, rb_node); |
956 | ||
7997c5b0 | 957 | rb_erase(&t->rb_node, &rb_sort_root); |
412819ce JA |
958 | free(t); |
959 | } | |
960 | } | |
961 | ||
d5396421 | 962 | static void show_entries_rb(void) |
8fc0abbc | 963 | { |
8fc0abbc | 964 | struct blk_io_trace *bit; |
3aabcd89 | 965 | struct rb_node *n; |
8fc0abbc JA |
966 | struct trace *t; |
967 | int cpu; | |
968 | ||
7997c5b0 | 969 | n = rb_first(&rb_sort_root); |
3aabcd89 JA |
970 | if (!n) |
971 | return; | |
8fc0abbc | 972 | |
3aabcd89 | 973 | do { |
8fc0abbc JA |
974 | t = rb_entry(n, struct trace, rb_node); |
975 | bit = t->bit; | |
976 | ||
d5396421 | 977 | cpu = bit->cpu; |
87b72777 | 978 | if (cpu > max_cpus) { |
8fc0abbc | 979 | fprintf(stderr, "CPU number too large (%d)\n", cpu); |
87b72777 | 980 | break; |
8fc0abbc JA |
981 | } |
982 | ||
cfab07eb AB |
983 | if (genesis_time == 0) |
984 | genesis_time = bit->time; | |
985 | bit->time -= genesis_time; | |
8fc0abbc | 986 | |
cfab07eb | 987 | check_time(bit); |
8fc0abbc | 988 | |
d5396421 | 989 | if (dump_trace(bit, &per_cpu_info[cpu])) |
87b72777 JA |
990 | break; |
991 | ||
8fc0abbc JA |
992 | } while ((n = rb_next(n)) != NULL); |
993 | } | |
994 | ||
1f79c4a0 JA |
995 | static int read_data(int fd, void *buffer, int bytes, int block) |
996 | { | |
997 | int ret, bytes_left, fl; | |
998 | void *p; | |
999 | ||
1000 | fl = fcntl(fd, F_GETFL); | |
1001 | ||
1002 | if (!block) | |
1003 | fcntl(fd, F_SETFL, fl | O_NONBLOCK); | |
1004 | else | |
1005 | fcntl(fd, F_SETFL, fl & ~O_NONBLOCK); | |
1006 | ||
1007 | bytes_left = bytes; | |
1008 | p = buffer; | |
1009 | while (bytes_left > 0) { | |
1010 | ret = read(fd, p, bytes_left); | |
1011 | if (!ret) | |
1012 | return 1; | |
1013 | else if (ret < 0) { | |
1014 | if (errno != EAGAIN) | |
1015 | perror("read"); | |
1016 | return -1; | |
1017 | } else { | |
1018 | p += ret; | |
1019 | bytes_left -= ret; | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | return 0; | |
1024 | } | |
1025 | ||
d5396421 | 1026 | static int do_file(void) |
d0ca268b | 1027 | { |
a718bd37 | 1028 | int i, nfiles; |
d0ca268b | 1029 | |
a718bd37 NS |
1030 | for (i = 0, nfiles = 0;; i++, nfiles++) { |
1031 | struct per_cpu_info *pci; | |
87b72777 JA |
1032 | struct stat st; |
1033 | void *tb; | |
1034 | ||
a718bd37 | 1035 | pci = get_cpu_info(i); |
d0ca268b | 1036 | |
d5396421 JA |
1037 | snprintf(pci->fname, sizeof(pci->fname)-1,"%s_out.%d", dev, i); |
1038 | if (stat(pci->fname, &st) < 0) | |
d0ca268b | 1039 | break; |
afd2d7ad | 1040 | if (!st.st_size) |
1041 | continue; | |
d0ca268b | 1042 | |
d5396421 | 1043 | printf("Processing %s\n", pci->fname); |
8fc0abbc | 1044 | |
87b72777 | 1045 | tb = malloc(st.st_size); |
8fc0abbc | 1046 | |
d5396421 JA |
1047 | pci->fd = open(pci->fname, O_RDONLY); |
1048 | if (pci->fd < 0) { | |
1049 | perror(pci->fname); | |
3aabcd89 | 1050 | break; |
d0ca268b | 1051 | } |
afd2d7ad | 1052 | |
1f79c4a0 | 1053 | if (read_data(pci->fd, tb, st.st_size, 1)) |
3aabcd89 | 1054 | break; |
d0ca268b | 1055 | |
a718bd37 | 1056 | if (sort_entries(tb, st.st_size, ~0U) == -1) |
3aabcd89 | 1057 | break; |
d0ca268b | 1058 | |
d5396421 | 1059 | close(pci->fd); |
d5396421 JA |
1060 | printf("\t%2d %10s %15d\n", i, pci->fname, pci->nelems); |
1061 | ||
1062 | } | |
1063 | ||
1064 | if (!nfiles) { | |
1065 | fprintf(stderr, "No files found\n"); | |
1066 | return 1; | |
1067 | } | |
1068 | ||
1069 | show_entries_rb(); | |
d5396421 JA |
1070 | return 0; |
1071 | } | |
1072 | ||
2ff323b0 | 1073 | static void resize_buffer(void **buffer, long *size, long offset) |
d5396421 | 1074 | { |
2ff323b0 | 1075 | long old_size = *size; |
d5396421 | 1076 | |
51128a28 JA |
1077 | if (*size == 0) |
1078 | *size = 64 * sizeof(struct blk_io_trace); | |
1079 | ||
2ff323b0 JA |
1080 | *size *= 2; |
1081 | *buffer = realloc(*buffer, *size); | |
51128a28 JA |
1082 | |
1083 | if (old_size) | |
1084 | memset(*buffer + offset, 0, *size - old_size); | |
412819ce | 1085 | } |
d5396421 | 1086 | |
0a94916b | 1087 | static int read_sort_events(int fd, void **buffer, long *max_offset) |
412819ce | 1088 | { |
0a94916b | 1089 | long offset; |
412819ce | 1090 | int events; |
d5396421 | 1091 | |
51128a28 | 1092 | events = offset = 0; |
412819ce JA |
1093 | do { |
1094 | struct blk_io_trace *t; | |
1095 | int pdu_len; | |
51128a28 | 1096 | __u32 magic; |
d5396421 | 1097 | |
0a94916b JA |
1098 | if (*max_offset - offset < sizeof(*t)) |
1099 | resize_buffer(buffer, max_offset, offset); | |
d5396421 | 1100 | |
c80c18a7 JA |
1101 | if (read_data(fd, *buffer + offset, sizeof(*t), !events)) |
1102 | break; | |
d5396421 | 1103 | |
412819ce JA |
1104 | t = *buffer + offset; |
1105 | offset += sizeof(*t); | |
d5396421 | 1106 | |
51128a28 JA |
1107 | magic = be32_to_cpu(t->magic); |
1108 | if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { | |
1109 | fprintf(stderr, "Bad magic %x\n", magic); | |
1110 | break; | |
1111 | } | |
1112 | ||
412819ce | 1113 | pdu_len = be16_to_cpu(t->pdu_len); |
2ff323b0 | 1114 | if (pdu_len) { |
51128a28 | 1115 | if (*max_offset - offset <= pdu_len) |
0a94916b | 1116 | resize_buffer(buffer, max_offset, offset); |
d5396421 | 1117 | |
2ff323b0 JA |
1118 | if (read_data(fd, *buffer + offset, pdu_len, 1)) |
1119 | break; | |
d5396421 | 1120 | |
2ff323b0 JA |
1121 | offset += pdu_len; |
1122 | } | |
d5396421 | 1123 | |
412819ce | 1124 | events++; |
79f19470 | 1125 | } while (!is_done() && events < rb_batch); |
d5396421 | 1126 | |
412819ce JA |
1127 | return events; |
1128 | } | |
d5396421 | 1129 | |
412819ce JA |
1130 | static int do_stdin(void) |
1131 | { | |
1132 | int fd; | |
51128a28 | 1133 | void *ptr = NULL; |
0a94916b | 1134 | long max_offset; |
d5396421 | 1135 | |
1f79c4a0 | 1136 | fd = dup(STDIN_FILENO); |
0a94916b | 1137 | max_offset = 0; |
412819ce JA |
1138 | do { |
1139 | int events; | |
d5396421 | 1140 | |
0a94916b | 1141 | events = read_sort_events(fd, &ptr, &max_offset); |
412819ce JA |
1142 | if (!events) |
1143 | break; | |
1144 | ||
2ff323b0 JA |
1145 | if (sort_entries(ptr, ~0UL, events) == -1) |
1146 | break; | |
1147 | ||
412819ce JA |
1148 | show_entries_rb(); |
1149 | free_entries_rb(); | |
d5396421 JA |
1150 | } while (1); |
1151 | ||
51128a28 JA |
1152 | if (ptr) |
1153 | free(ptr); | |
1154 | ||
d5396421 | 1155 | close(fd); |
d5396421 JA |
1156 | return 0; |
1157 | } | |
d0ca268b | 1158 | |
1f79c4a0 | 1159 | static void flush_output(void) |
412819ce | 1160 | { |
152f6476 | 1161 | fflush(ofp); |
412819ce JA |
1162 | } |
1163 | ||
1f79c4a0 | 1164 | static void handle_sigint(int sig) |
412819ce JA |
1165 | { |
1166 | done = 1; | |
1167 | flush_output(); | |
1168 | } | |
1169 | ||
1f79c4a0 JA |
1170 | static void usage(char *prog) |
1171 | { | |
152f6476 | 1172 | fprintf(stderr, "Usage: %s -i <name> [-o <output>][-s]\n", prog); |
1f79c4a0 JA |
1173 | } |
1174 | ||
d5396421 JA |
1175 | int main(int argc, char *argv[]) |
1176 | { | |
152f6476 | 1177 | char *ofp_buffer; |
a66877e6 | 1178 | int c, ret, mode; |
d5396421 JA |
1179 | |
1180 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) { | |
1181 | switch (c) { | |
1182 | case 'i': | |
66efebf8 | 1183 | dev = optarg; |
d5396421 JA |
1184 | break; |
1185 | case 'o': | |
66efebf8 | 1186 | output_name = optarg; |
d5396421 | 1187 | break; |
79f19470 JA |
1188 | case 'b': |
1189 | rb_batch = atoi(optarg); | |
1190 | if (rb_batch <= 0) | |
1191 | rb_batch = RB_BATCH_DEFAULT; | |
1192 | break; | |
152f6476 JA |
1193 | case 's': |
1194 | per_process_stats = 1; | |
1195 | break; | |
7997c5b0 JA |
1196 | case 't': |
1197 | track_ios = 1; | |
1198 | break; | |
d5396421 | 1199 | default: |
1f79c4a0 | 1200 | usage(argv[0]); |
d5396421 JA |
1201 | return 1; |
1202 | } | |
d0ca268b JA |
1203 | } |
1204 | ||
d5396421 | 1205 | if (!dev) { |
1f79c4a0 | 1206 | usage(argv[0]); |
d5396421 JA |
1207 | return 1; |
1208 | } | |
1209 | ||
7997c5b0 JA |
1210 | memset(&rb_sort_root, 0, sizeof(rb_sort_root)); |
1211 | memset(&rb_track_root, 0, sizeof(rb_track_root)); | |
412819ce JA |
1212 | |
1213 | signal(SIGINT, handle_sigint); | |
1214 | signal(SIGHUP, handle_sigint); | |
1215 | signal(SIGTERM, handle_sigint); | |
d5396421 | 1216 | |
d69db225 JA |
1217 | setlocale(LC_NUMERIC, "en_US"); |
1218 | ||
a66877e6 | 1219 | if (!output_name) { |
152f6476 | 1220 | ofp = fdopen(STDOUT_FILENO, "w"); |
a66877e6 JA |
1221 | mode = _IOLBF; |
1222 | } else { | |
152f6476 JA |
1223 | char ofname[128]; |
1224 | ||
1225 | snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name); | |
1226 | ofp = fopen(ofname, "w"); | |
a66877e6 | 1227 | mode = _IOFBF; |
152f6476 JA |
1228 | } |
1229 | ||
1230 | if (!ofp) { | |
1231 | perror("fopen"); | |
1232 | return 1; | |
1233 | } | |
1234 | ||
1235 | ofp_buffer = malloc(4096); | |
a66877e6 | 1236 | if (setvbuf(ofp, ofp_buffer, mode, 4096)) { |
152f6476 JA |
1237 | perror("setvbuf"); |
1238 | return 1; | |
1239 | } | |
1240 | ||
d5396421 JA |
1241 | if (!strcmp(dev, "-")) |
1242 | ret = do_stdin(); | |
1243 | else | |
1244 | ret = do_file(); | |
1245 | ||
152f6476 JA |
1246 | if (per_process_stats) |
1247 | show_process_stats(); | |
1248 | ||
1249 | show_cpu_stats(); | |
1250 | ||
412819ce | 1251 | flush_output(); |
d5396421 | 1252 | return ret; |
d0ca268b | 1253 | } |