projects / blktrace.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
iowatcher: Add possibility to limit seconds from below
[blktrace.git] / iowatcher / blkparse.c
CommitLineData
9e066e23
CM		 1/*
2 * Copyright (C) 2012 Fusion-io
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16 *
17 * Parts of this file were imported from Jens Axboe's blktrace sources (also GPL)
18 */
19#include <sys/types.h>
20#include <sys/stat.h>
21#include <fcntl.h>
22#include <unistd.h>
23#include <stdlib.h>
24#include <stdio.h>
25#include <math.h>
26#include <inttypes.h>
27#include <string.h>
28#include <asm/types.h>
29#include <errno.h>
30#include <sys/mman.h>
31#include <time.h>
32#include <math.h>
33
34#include "plot.h"
35#include "blkparse.h"
36#include "list.h"
37#include "tracers.h"
38
39#define IO_HASH_TABLE_BITS 11
40#define IO_HASH_TABLE_SIZE (1 << IO_HASH_TABLE_BITS)
41static struct list_head io_hash_table[IO_HASH_TABLE_SIZE];
42static u64 ios_in_flight = 0;
43
44
45/*
46 * Trace categories
47 */
48enum {
49 BLK_TC_READ = 1 << 0, /* reads */
50 BLK_TC_WRITE = 1 << 1, /* writes */
51 BLK_TC_FLUSH = 1 << 2, /* flush */
52 BLK_TC_SYNC = 1 << 3, /* sync */
53 BLK_TC_QUEUE = 1 << 4, /* queueing/merging */
54 BLK_TC_REQUEUE = 1 << 5, /* requeueing */
55 BLK_TC_ISSUE = 1 << 6, /* issue */
56 BLK_TC_COMPLETE = 1 << 7, /* completions */
57 BLK_TC_FS = 1 << 8, /* fs requests */
58 BLK_TC_PC = 1 << 9, /* pc requests */
59 BLK_TC_NOTIFY = 1 << 10, /* special message */
60 BLK_TC_AHEAD = 1 << 11, /* readahead */
61 BLK_TC_META = 1 << 12, /* metadata */
62 BLK_TC_DISCARD = 1 << 13, /* discard requests */
63 BLK_TC_DRV_DATA = 1 << 14, /* binary driver data */
64 BLK_TC_FUA = 1 << 15, /* fua requests */
65
66 BLK_TC_END = 1 << 15, /* we've run out of bits! */
67};
68
69#define BLK_TC_SHIFT (16)
70#define BLK_TC_ACT(act) ((act) << BLK_TC_SHIFT)
71#define BLK_DATADIR(a) (((a) >> BLK_TC_SHIFT) & (BLK_TC_READ | BLK_TC_WRITE))
72
73/*
74 * Basic trace actions
75 */
76enum {
77 __BLK_TA_QUEUE = 1, /* queued */
78 __BLK_TA_BACKMERGE, /* back merged to existing rq */
79 __BLK_TA_FRONTMERGE, /* front merge to existing rq */
80 __BLK_TA_GETRQ, /* allocated new request */
81 __BLK_TA_SLEEPRQ, /* sleeping on rq allocation */
82 __BLK_TA_REQUEUE, /* request requeued */
83 __BLK_TA_ISSUE, /* sent to driver */
84 __BLK_TA_COMPLETE, /* completed by driver */
85 __BLK_TA_PLUG, /* queue was plugged */
86 __BLK_TA_UNPLUG_IO, /* queue was unplugged by io */
87 __BLK_TA_UNPLUG_TIMER, /* queue was unplugged by timer */
88 __BLK_TA_INSERT, /* insert request */
89 __BLK_TA_SPLIT, /* bio was split */
90 __BLK_TA_BOUNCE, /* bio was bounced */
91 __BLK_TA_REMAP, /* bio was remapped */
92 __BLK_TA_ABORT, /* request aborted */
93 __BLK_TA_DRV_DATA, /* binary driver data */
94};
95
1582ecc9
JK		 96#define BLK_TA_MASK ((1 << BLK_TC_SHIFT) - 1)
97
9e066e23
CM		 98/*
99 * Notify events.
100 */
101enum blktrace_notify {
102 __BLK_TN_PROCESS = 0, /* establish pid/name mapping */
103 __BLK_TN_TIMESTAMP, /* include system clock */
104 __BLK_TN_MESSAGE, /* Character string message */
105};
106
107/*
108 * Trace actions in full. Additionally, read or write is masked
109 */
110#define BLK_TA_QUEUE (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
111#define BLK_TA_BACKMERGE (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
112#define BLK_TA_FRONTMERGE (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
113#define BLK_TA_GETRQ (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
114#define BLK_TA_SLEEPRQ (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
115#define BLK_TA_REQUEUE (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
116#define BLK_TA_ISSUE (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
117#define BLK_TA_COMPLETE (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
118#define BLK_TA_PLUG (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
119#define BLK_TA_UNPLUG_IO (__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
120#define BLK_TA_UNPLUG_TIMER (__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
121#define BLK_TA_INSERT (__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
122#define BLK_TA_SPLIT (__BLK_TA_SPLIT)
123#define BLK_TA_BOUNCE (__BLK_TA_BOUNCE)
124#define BLK_TA_REMAP (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
125#define BLK_TA_ABORT (__BLK_TA_ABORT | BLK_TC_ACT(BLK_TC_QUEUE))
126#define BLK_TA_DRV_DATA (__BLK_TA_DRV_DATA | BLK_TC_ACT(BLK_TC_DRV_DATA))
127
128#define BLK_TN_PROCESS (__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
129#define BLK_TN_TIMESTAMP (__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
130#define BLK_TN_MESSAGE (__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
131
132#define BLK_IO_TRACE_MAGIC 0x65617400
133#define BLK_IO_TRACE_VERSION 0x07
134/*
135 * The trace itself
136 */
137struct blk_io_trace {
138 __u32 magic; /* MAGIC << 8 | version */
139 __u32 sequence; /* event number */
140 __u64 time; /* in nanoseconds */
141 __u64 sector; /* disk offset */
142 __u32 bytes; /* transfer length */
143 __u32 action; /* what happened */
144 __u32 pid; /* who did it */
145 __u32 device; /* device identifier (dev_t) */
146 __u32 cpu; /* on what cpu did it happen */
147 __u16 error; /* completion error */
148 __u16 pdu_len; /* length of data after this trace */
149};
150
151struct pending_io {
152 /* sector offset of this IO */
153 u64 sector;
154
155 /* time this IO was dispatched */
156 u64 dispatch_time;
157 /* time this IO was finished */
158 u64 completion_time;
159 struct list_head hash_list;
160};
161
162#define MINORBITS 20
163#define MINORMASK ((1 << MINORBITS) - 1)
164#define SECONDS(x) ((unsigned long long)(x) / 1000000000)
165#define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000)
166#define DOUBLE_TO_NANO_ULL(d) ((unsigned long long)((d) * 1000000000))
167#define CHECK_MAGIC(t) (((t)->magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
168
169void init_io_hash_table(void)
170{
171 int i;
172 struct list_head *head;
173
174 for (i = 0; i < IO_HASH_TABLE_SIZE; i++) {
175 head = io_hash_table + i;
176 INIT_LIST_HEAD(head);
177 }
178}
179
180/* taken from the kernel hash.h */
181static inline u64 hash_sector(u64 val)
182{
183 u64 hash = val;
184
185 /* Sigh, gcc can't optimise this alone like it does for 32 bits. */
186 u64 n = hash;
187 n <<= 18;
188 hash -= n;
189 n <<= 33;
190 hash -= n;
191 n <<= 3;
192 hash += n;
193 n <<= 3;
194 hash -= n;
195 n <<= 4;
196 hash += n;
197 n <<= 2;
198 hash += n;
199
200 /* High bits are more random, so use them. */
201 return hash >> (64 - IO_HASH_TABLE_BITS);
202}
203
204static int hash_table_insert(struct pending_io *ins_pio)
205{
206 u64 sector = ins_pio->sector;
207 int slot = hash_sector(sector);
208 struct list_head *head;
209 struct pending_io *pio;
210
211 head = io_hash_table + slot;
212 list_for_each_entry(pio, head, hash_list) {
213 if (pio->sector == sector)
214 return -EEXIST;
215 }
216 list_add_tail(&ins_pio->hash_list, head);
217 return 0;
218}
219
220static struct pending_io *hash_table_search(u64 sector)
221{
222 int slot = hash_sector(sector);
223 struct list_head *head;
224 struct pending_io *pio;
225
226 head = io_hash_table + slot;
227 list_for_each_entry(pio, head, hash_list) {
228 if (pio->sector == sector)
229 return pio;
230 }
231 return NULL;
232}
233
234static int hash_dispatched_io(struct blk_io_trace *io)
235{
236 struct pending_io *pio;
237 int ret;
238
239 pio = calloc(1, sizeof(*pio));
240 pio->sector = io->sector;
241 pio->dispatch_time = io->time;
242
243 ret = hash_table_insert(pio);
244 if (ret == -EEXIST) {
245 /* crud, the IO isn't here */
246 free(pio);
247 }
248 return ret;
249}
250
251static struct pending_io *hash_completed_io(struct blk_io_trace *io)
252{
253 struct pending_io *pio;
254
255 pio = hash_table_search(io->sector);
256
257 if (!pio)
258 return NULL;
259 return pio;
260}
261
262static void handle_notify(struct trace *trace)
263{
264 struct blk_io_trace *io = trace->io;
265 void *payload = (char *)io + sizeof(*io);
266 u32 two32[2];
267
268
269 if (io->action != BLK_TN_TIMESTAMP)
270 return;
271
272 if (io->pdu_len != sizeof(two32))
273 return;
274
275 memcpy(two32, payload, sizeof(two32));
276 trace->start_timestamp = io->time;
277 trace->abs_start_time.tv_sec = two32[0];
278 trace->abs_start_time.tv_nsec = two32[1];
279 if (trace->abs_start_time.tv_nsec < 0) {
280 trace->abs_start_time.tv_sec--;
281 trace->abs_start_time.tv_nsec += 1000000000;
282 }
283}
284
285int next_record(struct trace *trace)
286{
287 int skip = trace->io->pdu_len;
288 u64 offset;
289
290 trace->cur += sizeof(*trace->io) + skip;
291 offset = trace->cur - trace->start;
292 if (offset >= trace->len)
293 return 1;
294
295 trace->io = (struct blk_io_trace *)trace->cur;
296 return 0;
297}
298
299void first_record(struct trace *trace)
300{
301 trace->cur = trace->start;
302 trace->io = (struct blk_io_trace *)trace->cur;
303}
304
bfb0e441
CM		 305int is_io_event(struct blk_io_trace *test)
306{
307 char *message;
308 if (!(test->action & BLK_TC_ACT(BLK_TC_NOTIFY)))
309 return 1;
310 if (test->action == BLK_TN_MESSAGE) {
311 int len = test->pdu_len;
312 if (len < 3)
313 return 0;
314 message = (char *)(test + 1);
315 if (strncmp(message, "fio ", 4) == 0) {
316 return 1;
317 }
318 }
319 return 0;
320}
321
9e066e23
CM		 322u64 find_last_time(struct trace *trace)
323{
324 char *p = trace->start + trace->len;
325 struct blk_io_trace *test;
326 int search_len = 0;
327 u64 found = 0;
328
329 if (trace->len < sizeof(*trace->io))
330 return 0;
331 p -= sizeof(*trace->io);
332 while (p >= trace->start) {
333 test = (struct blk_io_trace *)p;
bfb0e441 334 if (CHECK_MAGIC(test) && is_io_event(test)) {
9e066e23
CM		 335 u64 offset = p - trace->start;
336 if (offset + sizeof(*test) + test->pdu_len == trace->len) {
337 return test->time;
338 }
339 }
340 p--;
341 search_len++;
342 if (search_len > 8192) {
343 break;
344 }
345 }
346
347 /* searching backwards didn't work out, we'll have to scan the file */
348 first_record(trace);
349 while (1) {
bfb0e441 350 if (is_io_event(trace->io))
9e066e23
CM		 351 found = trace->io->time;
352 if (next_record(trace))
353 break;
354 }
355 first_record(trace);
356 return found;
357}
358
bfb0e441
CM		 359int parse_fio_bank_message(struct trace *trace, u64 *bank_ret, u64 *offset_ret,
360 u64 *num_banks_ret)
361{
362 char *s;
363 char *next;
364 char *message;
365 struct blk_io_trace *test = trace->io;
366 int len = test->pdu_len;
367 u64 bank;
368 u64 offset;
369 u64 num_banks;
370
371 if (!(test->action & BLK_TC_ACT(BLK_TC_NOTIFY)))
372 return -1;
373 if (test->action != BLK_TN_MESSAGE)
374 return -1;
375
376 /* the message is fio rw bank offset num_banks */
377 if (len < 3)
378 return -1;
379 message = (char *)(test + 1);
380 if (strncmp(message, "fio r ", 6) != 0)
381 return -1;
382
383 message = strndup(message, len);
384 s = strchr(message, ' ');
385 if (!s)
386 goto out;
387 s++;
388 s = strchr(s, ' ');
389 if (!s)
390 goto out;
391
392 bank = strtoll(s, &next, 10);
393 if (s == next)
394 goto out;
395 s = next;
396
397 offset = strtoll(s, &next, 10);
398 if (s == next)
399 goto out;
400 s = next;
401
402 num_banks = strtoll(s, &next, 10);
403 if (s == next)
404 goto out;
405
406 *bank_ret = bank;
407 *offset_ret = offset;
408 *num_banks_ret = num_banks;
409
410 return 0;
411out:
412 free(message);
413 return -1;
414}
415
9b9fa04b
JK		 416void find_extreme_offsets(struct trace *trace, u64 *min_ret, u64 *max_ret, u64 *max_bank_ret,
417 u64 *max_offset_ret)
9e066e23
CM		 418{
419 u64 found = 0;
9b9fa04b 420 u64 max = 0, min = ~(u64)0;
bfb0e441
CM		 421 u64 max_bank = 0;
422 u64 max_bank_offset = 0;
423 u64 num_banks = 0;
9e066e23
CM		 424 first_record(trace);
425 while (1) {
426 if (!(trace->io->action & BLK_TC_ACT(BLK_TC_NOTIFY))) {
427 found = trace->io->sector << 9;
9b9fa04b
JK		 428 if (found < min)
429 min = found;
9e066e23 430
9b9fa04b
JK		 431 found += trace->io->bytes;
432 if (max < found)
9e066e23 433 max = found;
bfb0e441
CM		 434 } else {
435 u64 bank;
436 u64 offset;
437 if (!parse_fio_bank_message(trace, &bank,
438 &offset, &num_banks)) {
439 if (bank > max_bank)
440 max_bank = bank;
441 if (offset > max_bank_offset)
442 max_bank_offset = offset;
443 }
9e066e23
CM		 444 }
445 if (next_record(trace))
446 break;
447 }
448 first_record(trace);
9b9fa04b 449 *min_ret = min;
bfb0e441
CM		 450 *max_ret = max;
451 *max_bank_ret = max_bank;
452 *max_offset_ret = max_bank_offset;
9e066e23
CM		 453}
454
9b9fa04b 455int filter_outliers(struct trace *trace, u64 min_offset, u64 max_offset,
9e066e23
CM		 456 u64 *yzoom_min, u64 *yzoom_max)
457{
458 int hits[11];
459 u64 max_per_bucket[11];
9b9fa04b
JK		 460 u64 min_per_bucket[11];
461 u64 bytes_per_bucket = (max_offset - min_offset + 1) / 10;
9e066e23
CM		 462 int slot;
463 int fat_count = 0;
464
465 memset(hits, 0, sizeof(int) * 11);
466 memset(max_per_bucket, 0, sizeof(u64) * 11);
9b9fa04b 467 memset(min_per_bucket, 0xff, sizeof(u64) * 11);
9e066e23
CM		 468 first_record(trace);
469 while (1) {
41fdf407
JK		 470 if (!(trace->io->action & BLK_TC_ACT(BLK_TC_NOTIFY)) &&
471 (trace->io->action & BLK_TA_MASK) == __BLK_TA_QUEUE) {
9b9fa04b
JK		 472 u64 off = (trace->io->sector << 9) - min_offset;
473
474 slot = (int)(off / bytes_per_bucket);
475 hits[slot]++;
476 if (off < min_per_bucket[slot])
477 min_per_bucket[slot] = off;
478
479 off += trace->io->bytes;
480 slot = (int)(off / bytes_per_bucket);
9e066e23 481 hits[slot]++;
9b9fa04b
JK		 482 if (off > max_per_bucket[slot])
483 max_per_bucket[slot] = off;
9e066e23
CM		 484 }
485 if (next_record(trace))
486 break;
487 }
488 first_record(trace);
489 for (slot = 0; slot < 11; slot++) {
490 if (hits[slot] > fat_count) {
491 fat_count = hits[slot];
492 }
493 }
494
495 *yzoom_max = max_offset;
496 for (slot = 10; slot >= 0; slot--) {
497 double d = hits[slot];
498
499 if (d >= (double)fat_count * .05) {
9b9fa04b 500 *yzoom_max = max_per_bucket[slot] + min_offset;
9e066e23
CM		 501 break;
502 }
503 }
504
9b9fa04b 505 *yzoom_min = min_offset;
9e066e23
CM		 506 for (slot = 0; slot < 10; slot++) {
507 double d = hits[slot];
508
509 if (d >= (double)fat_count * .05) {
9b9fa04b 510 *yzoom_min = min_per_bucket[slot] + min_offset;
9e066e23
CM		 511 break;
512 }
513 }
514 return 0;
515}
516
517static char *find_trace_file(char *filename)
518{
519 int ret;
520 struct stat st;
521 char line[1024];
522 char *dot;
523 char *try;
524
525 ret = stat(filename, &st);
526 if (ret == 0)
527 return strdup(filename);
528
529 snprintf(line, 1024, "%s.%s", filename, "dump");
e199d546 530 ret = stat(line, &st);
9e066e23
CM		 531 if (ret == 0)
532 return strdup(line);
533
534 try = strdup(filename);
535 dot = strrchr(try, '.');
536 if (!dot || strcmp(".dump", dot) != 0) {
537 if (dot)
538 *dot = '\0';
539 snprintf(line, 1024, "%s%s", try, ".blktrace.0");
540 ret = stat(line, &st);
541 if (ret == 0) {
542 blktrace_to_dump(try);
543 snprintf(line, 1024, "%s.%s", try, "dump");
544 ret = stat(line, &st);
545 if (ret == 0) {
546 free(try);
547 return strdup(line);
548 }
549 }
550 }
551 free(try);
552 return NULL;
553}
554struct trace *open_trace(char *filename)
555{
556 int fd;
557 char *p;
558 struct stat st;
559 int ret;
560 struct trace *trace;
561 char *found_filename;
562
563 trace = calloc(1, sizeof(*trace));
564 if (!trace) {
565 fprintf(stderr, "unable to allocate memory for trace\n");
566 return NULL;
567 }
568
569 found_filename = find_trace_file(filename);
570 if (!found_filename) {
571 fprintf(stderr, "Unable to find trace file %s\n", filename);
572 goto fail;
573 }
9e066e23
CM		 574 filename = found_filename;
575
576 fd = open(filename, O_RDONLY);
577 if (fd < 0) {
578 fprintf(stderr, "Unable to open trace file %s err %s\n", filename, strerror(errno));
579 goto fail;
580 }
581 ret = fstat(fd, &st);
582 if (ret < 0) {
583 fprintf(stderr, "stat failed on %s err %s\n", filename, strerror(errno));
584 goto fail_fd;
585 }
586 p = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
587 if (p == MAP_FAILED) {
588 fprintf(stderr, "Unable to mmap trace file %s, err %s\n", filename, strerror(errno));
589 goto fail_fd;
590 }
591 trace->fd = fd;
592 trace->len = st.st_size;
593 trace->start = p;
594 trace->cur = p;
595 trace->io = (struct blk_io_trace *)p;
596 return trace;
597
598fail_fd:
599 close(fd);
600fail:
601 free(trace);
602 return NULL;
603}
604static inline int tput_event(struct trace *trace)
605{
606 if (trace->found_completion)
607 return __BLK_TA_COMPLETE;
608 if (trace->found_issue)
609 return __BLK_TA_ISSUE;
610 if (trace->found_queue)
611 return __BLK_TA_QUEUE;
612
613 return __BLK_TA_COMPLETE;
614}
615
616static inline int io_event(struct trace *trace)
617{
618 if (trace->found_queue)
619 return __BLK_TA_QUEUE;
620 if (trace->found_issue)
621 return __BLK_TA_ISSUE;
622 if (trace->found_completion)
623 return __BLK_TA_COMPLETE;
624
625 return __BLK_TA_COMPLETE;
626}
627
628void add_tput(struct trace *trace, struct graph_line_data *gld)
629{
630 struct blk_io_trace *io = trace->io;
1582ecc9 631 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 632 int seconds;
633
634 if (io->action & BLK_TC_ACT(BLK_TC_NOTIFY))
635 return;
636
637 if (action != tput_event(trace))
638 return;
639
640 seconds = SECONDS(io->time);
f752a6eb
JK		 641 if (seconds > gld->max_seconds) {
642 fprintf(stderr, "Bad record %d %d %d\n", seconds, gld->max_seconds, action);
9e066e23
CM		 643 abort();
644 }
645
646 gld->data[seconds].sum += io->bytes;
647 gld->data[seconds].count = 1;
648 if (gld->data[seconds].sum > gld->max)
649 gld->max = gld->data[seconds].sum;
650}
651
652void add_io(struct trace *trace, struct graph_dot_data *gdd_writes,
653 struct graph_dot_data *gdd_reads)
654{
655 struct blk_io_trace *io = trace->io;
1582ecc9 656 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 657 u64 offset;
658
659 if (io->action & BLK_TC_ACT(BLK_TC_NOTIFY))
660 return;
661
662 if (action != io_event(trace))
663 return;
664
665 offset = io->sector << 9;
666
667 if (BLK_DATADIR(io->action) & BLK_TC_READ)
668 set_gdd_bit(gdd_reads, offset, io->bytes, io->time);
669 else if (BLK_DATADIR(io->action) & BLK_TC_WRITE)
670 set_gdd_bit(gdd_writes, offset, io->bytes, io->time);
671}
672
673void add_pending_io(struct trace *trace, struct graph_line_data *gld)
674{
675 int ret;
676 int seconds;
677 struct blk_io_trace *io = trace->io;
1582ecc9 678 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 679 double avg;
680
681 if (io->action & BLK_TC_ACT(BLK_TC_NOTIFY))
682 return;
683
684 if (action != __BLK_TA_ISSUE)
685 return;
686
687 seconds = SECONDS(io->time);
f752a6eb
JK		 688 if (seconds > gld->max_seconds) {
689 fprintf(stderr, "Bad record %d %d\n", seconds, gld->max_seconds);
9e066e23
CM		 690 abort();
691 }
692
693 ret = hash_dispatched_io(trace->io);
694 if (ret)
695 return;
696
697 ios_in_flight++;
698
699 gld->data[seconds].sum += ios_in_flight;
700 gld->data[seconds].count++;
701
702 avg = (double)gld->data[seconds].sum / gld->data[seconds].count;
703 if (gld->max < (u64)avg) {
704 gld->max = avg;
705 }
706}
707
708void add_completed_io(struct trace *trace,
709 struct graph_line_data *latency_gld)
710{
711 struct blk_io_trace *io = trace->io;
712 int seconds;
1582ecc9 713 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 714 struct pending_io *pio;
715 double avg;
716 u64 latency;
717
718 if (io->action & BLK_TC_ACT(BLK_TC_NOTIFY))
719 return;
720
721 if (action != __BLK_TA_COMPLETE)
722 return;
723
724 seconds = SECONDS(io->time);
725
726 pio = hash_completed_io(trace->io);
727 if (!pio)
728 return;
729
730 if (ios_in_flight > 0)
731 ios_in_flight--;
732 if (io->time >= pio->dispatch_time) {
733 latency = io->time - pio->dispatch_time;
734 latency_gld->data[seconds].sum += latency;
735 latency_gld->data[seconds].count++;
736 }
737
738 list_del(&pio->hash_list);
739 free(pio);
740
741 avg = (double)latency_gld->data[seconds].sum /
742 latency_gld->data[seconds].count;
743 if (latency_gld->max < (u64)avg) {
744 latency_gld->max = avg;
745 }
746}
747
748void add_iop(struct trace *trace, struct graph_line_data *gld)
749{
750 struct blk_io_trace *io = trace->io;
1582ecc9 751 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 752 int seconds;
753
754 if (io->action & BLK_TC_ACT(BLK_TC_NOTIFY))
755 return;
756
757 /* iops and tput use the same events */
758 if (action != tput_event(trace))
759 return;
760
761 seconds = SECONDS(io->time);
f752a6eb
JK		 762 if (seconds > gld->max_seconds) {
763 fprintf(stderr, "Bad record %d %d\n", seconds, gld->max_seconds);
9e066e23
CM		 764 abort();
765 }
766
767 gld->data[seconds].sum += 1;
768 gld->data[seconds].count = 1;
769 if (gld->data[seconds].sum > gld->max)
770 gld->max = gld->data[seconds].sum;
771}
772
773void check_record(struct trace *trace)
774{
775 struct blk_io_trace *io = trace->io;
1582ecc9 776 int action = io->action & BLK_TA_MASK;
9e066e23
CM		 777
778 if (!(io->action & BLK_TC_ACT(BLK_TC_NOTIFY))) {
779 switch (action) {
780 case __BLK_TA_COMPLETE:
781 trace->found_completion = 1;
782 break;
783 case __BLK_TA_ISSUE:
784 trace->found_issue = 1;
785 break;
786 case __BLK_TA_QUEUE:
787 trace->found_queue = 1;
788 break;
789 };
790 }
791 handle_notify(trace);
792}
Block queue io tracer