Commit | Line | Data |
---|---|---|
d956a2cd JA |
1 | /* |
2 | * block queue tracing parse application | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
46e37c55 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
d956a2cd JA |
6 | * |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | */ | |
d0ca268b JA |
22 | #include <sys/types.h> |
23 | #include <sys/stat.h> | |
24 | #include <unistd.h> | |
25 | #include <stdio.h> | |
26 | #include <fcntl.h> | |
27 | #include <stdlib.h> | |
8fc0abbc | 28 | #include <string.h> |
d5396421 | 29 | #include <getopt.h> |
412819ce JA |
30 | #include <errno.h> |
31 | #include <signal.h> | |
d69db225 | 32 | #include <locale.h> |
6e0073ed | 33 | #include <libgen.h> |
d0ca268b | 34 | |
8fc0abbc JA |
35 | #include "blktrace.h" |
36 | #include "rbtree.h" | |
bf0720af | 37 | #include "jhash.h" |
d0ca268b | 38 | |
a3225fed | 39 | static char blkparse_version[] = "0.99.1"; |
52724a0e | 40 | |
492da111 AB |
41 | struct skip_info { |
42 | unsigned long start, end; | |
43 | struct skip_info *prev, *next; | |
44 | }; | |
45 | ||
e7c9f3ff | 46 | struct per_dev_info { |
f7bd1a9b | 47 | dev_t dev; |
e7c9f3ff NS |
48 | char *name; |
49 | ||
50 | int backwards; | |
51 | unsigned long long events; | |
20ed6177 | 52 | unsigned long long first_reported_time; |
e7c9f3ff | 53 | unsigned long long last_reported_time; |
287fa3d6 | 54 | unsigned long long last_read_time; |
e7c9f3ff | 55 | struct io_stats io_stats; |
2990e589 JA |
56 | unsigned long skips; |
57 | unsigned long long seq_skips; | |
649c7b66 JA |
58 | unsigned int max_depth[2]; |
59 | unsigned int cur_depth[2]; | |
e7c9f3ff | 60 | |
f7bd1a9b JA |
61 | struct rb_root rb_track; |
62 | ||
73877e12 | 63 | int nfiles; |
e7c9f3ff | 64 | int ncpus; |
824c2b39 JA |
65 | |
66 | unsigned long *cpu_map; | |
67 | unsigned int cpu_map_max; | |
68 | ||
e7c9f3ff NS |
69 | struct per_cpu_info *cpus; |
70 | }; | |
71 | ||
2990e589 JA |
72 | /* |
73 | * some duplicated effort here, we can unify this hash and the ppi hash later | |
74 | */ | |
75 | struct process_pid_map { | |
76 | pid_t pid; | |
77 | char comm[16]; | |
78 | struct process_pid_map *hash_next, *list_next; | |
79 | }; | |
80 | ||
81 | #define PPM_HASH_SHIFT (8) | |
82 | #define PPM_HASH_SIZE (1 << PPM_HASH_SHIFT) | |
83 | #define PPM_HASH_MASK (PPM_HASH_SIZE - 1) | |
84 | static struct process_pid_map *ppm_hash_table[PPM_HASH_SIZE]; | |
85 | ||
152f6476 | 86 | struct per_process_info { |
2990e589 | 87 | struct process_pid_map *ppm; |
152f6476 JA |
88 | struct io_stats io_stats; |
89 | struct per_process_info *hash_next, *list_next; | |
715d8021 | 90 | int more_than_one; |
50adc0ba JA |
91 | |
92 | /* | |
93 | * individual io stats | |
94 | */ | |
b9d40d6f JA |
95 | unsigned long long longest_allocation_wait[2]; |
96 | unsigned long long longest_dispatch_wait[2]; | |
97 | unsigned long long longest_completion_wait[2]; | |
d0ca268b JA |
98 | }; |
99 | ||
152f6476 | 100 | #define PPI_HASH_SHIFT (8) |
bf0720af JA |
101 | #define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT) |
102 | #define PPI_HASH_MASK (PPI_HASH_SIZE - 1) | |
103 | static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE]; | |
152f6476 | 104 | static struct per_process_info *ppi_list; |
886ecf0e | 105 | static int ppi_list_entries; |
152f6476 | 106 | |
a2594911 | 107 | #define S_OPTS "a:A:i:o:b:stqw:f:F:vVhD:d:" |
d5396421 | 108 | static struct option l_opts[] = { |
98f8386b AB |
109 | { |
110 | .name = "act-mask", | |
111 | .has_arg = required_argument, | |
112 | .flag = NULL, | |
113 | .val = 'a' | |
114 | }, | |
115 | { | |
116 | .name = "set-mask", | |
117 | .has_arg = required_argument, | |
118 | .flag = NULL, | |
119 | .val = 'A' | |
120 | }, | |
d5396421 JA |
121 | { |
122 | .name = "input", | |
428683db | 123 | .has_arg = required_argument, |
d5396421 JA |
124 | .flag = NULL, |
125 | .val = 'i' | |
126 | }, | |
127 | { | |
128 | .name = "output", | |
428683db | 129 | .has_arg = required_argument, |
d5396421 JA |
130 | .flag = NULL, |
131 | .val = 'o' | |
132 | }, | |
79f19470 JA |
133 | { |
134 | .name = "batch", | |
428683db | 135 | .has_arg = required_argument, |
79f19470 JA |
136 | .flag = NULL, |
137 | .val = 'b' | |
138 | }, | |
152f6476 | 139 | { |
3f65c585 | 140 | .name = "per-program-stats", |
428683db | 141 | .has_arg = no_argument, |
152f6476 JA |
142 | .flag = NULL, |
143 | .val = 's' | |
144 | }, | |
7997c5b0 | 145 | { |
3f65c585 | 146 | .name = "track-ios", |
428683db | 147 | .has_arg = no_argument, |
7997c5b0 JA |
148 | .flag = NULL, |
149 | .val = 't' | |
150 | }, | |
1e1c60f1 NS |
151 | { |
152 | .name = "quiet", | |
428683db | 153 | .has_arg = no_argument, |
1e1c60f1 NS |
154 | .flag = NULL, |
155 | .val = 'q' | |
156 | }, | |
46e6968b NS |
157 | { |
158 | .name = "stopwatch", | |
428683db | 159 | .has_arg = required_argument, |
46e6968b NS |
160 | .flag = NULL, |
161 | .val = 'w' | |
162 | }, | |
ab197ca7 AB |
163 | { |
164 | .name = "format", | |
428683db | 165 | .has_arg = required_argument, |
ab197ca7 AB |
166 | .flag = NULL, |
167 | .val = 'f' | |
168 | }, | |
169 | { | |
170 | .name = "format-spec", | |
428683db | 171 | .has_arg = required_argument, |
ab197ca7 AB |
172 | .flag = NULL, |
173 | .val = 'F' | |
174 | }, | |
bf0720af | 175 | { |
3f65c585 | 176 | .name = "hash-by-name", |
bf0720af JA |
177 | .has_arg = no_argument, |
178 | .flag = NULL, | |
d915dee6 | 179 | .val = 'h' |
bf0720af | 180 | }, |
7d1c0411 | 181 | { |
57ea8602 | 182 | .name = "verbose", |
7d1c0411 JA |
183 | .has_arg = no_argument, |
184 | .flag = NULL, | |
57ea8602 | 185 | .val = 'v' |
7d1c0411 | 186 | }, |
52724a0e JA |
187 | { |
188 | .name = "version", | |
189 | .has_arg = no_argument, | |
190 | .flag = NULL, | |
57ea8602 | 191 | .val = 'V' |
52724a0e | 192 | }, |
d1d7f15f | 193 | { |
3f65c585 | 194 | .name = "input-directory", |
d1d7f15f JA |
195 | .has_arg = required_argument, |
196 | .flag = NULL, | |
197 | .val = 'D' | |
198 | }, | |
a2594911 AB |
199 | { |
200 | .name = "dump-binary", | |
201 | .has_arg = required_argument, | |
202 | .flag = NULL, | |
203 | .val = 'd' | |
204 | }, | |
71ef8b7c JA |
205 | { |
206 | .name = NULL, | |
207 | } | |
d5396421 JA |
208 | }; |
209 | ||
7997c5b0 JA |
210 | /* |
211 | * for sorting the displayed output | |
212 | */ | |
8fc0abbc JA |
213 | struct trace { |
214 | struct blk_io_trace *bit; | |
215 | struct rb_node rb_node; | |
cb2a1a62 | 216 | struct trace *next; |
a43c1c17 | 217 | unsigned long read_sequence; |
8fc0abbc JA |
218 | }; |
219 | ||
cb2a1a62 | 220 | static struct rb_root rb_sort_root; |
a649216c JA |
221 | static unsigned long rb_sort_entries; |
222 | ||
cb2a1a62 JA |
223 | static struct trace *trace_list; |
224 | ||
d36421e4 JA |
225 | /* |
226 | * allocation cache | |
227 | */ | |
228 | static struct blk_io_trace *bit_alloc_list; | |
229 | static struct trace *t_alloc_list; | |
230 | ||
7997c5b0 JA |
231 | /* |
232 | * for tracking individual ios | |
233 | */ | |
234 | struct io_track { | |
235 | struct rb_node rb_node; | |
236 | ||
2990e589 | 237 | struct process_pid_map *ppm; |
7997c5b0 | 238 | __u64 sector; |
95c15013 | 239 | unsigned long long allocation_time; |
7997c5b0 JA |
240 | unsigned long long queue_time; |
241 | unsigned long long dispatch_time; | |
242 | unsigned long long completion_time; | |
243 | }; | |
244 | ||
e7c9f3ff NS |
245 | static int ndevices; |
246 | static struct per_dev_info *devices; | |
247 | static char *get_dev_name(struct per_dev_info *, char *, int); | |
210824c3 | 248 | static int trace_rb_insert_last(struct per_dev_info *, struct trace *); |
d0ca268b | 249 | |
71d5d4c9 | 250 | FILE *ofp = NULL; |
e7c9f3ff | 251 | static char *output_name; |
d1d7f15f | 252 | static char *input_dir; |
e7c9f3ff NS |
253 | |
254 | static unsigned long long genesis_time; | |
287fa3d6 | 255 | static unsigned long long last_allowed_time; |
46e6968b | 256 | static unsigned long long stopwatch_start; /* start from zero by default */ |
bc171579 | 257 | static unsigned long long stopwatch_end = -1ULL; /* "infinity" */ |
a43c1c17 | 258 | static unsigned long read_sequence; |
152f6476 JA |
259 | |
260 | static int per_process_stats; | |
cbc927b6 | 261 | static int per_device_and_cpu_stats = 1; |
7997c5b0 | 262 | static int track_ios; |
bf0720af | 263 | static int ppi_hash_by_pid = 1; |
57ea8602 | 264 | static int verbose; |
98f8386b | 265 | static unsigned int act_mask = -1U; |
cbc927b6 | 266 | static int stats_printed; |
86368eb5 | 267 | int data_is_native = -1; |
d0ca268b | 268 | |
a2594911 AB |
269 | static int dump_fd; |
270 | static char *dump_binary; | |
271 | ||
1d24fc14 JA |
272 | static unsigned int t_alloc_cache; |
273 | static unsigned int bit_alloc_cache; | |
274 | ||
7d747d22 | 275 | #define RB_BATCH_DEFAULT (512) |
e820abd7 | 276 | static unsigned int rb_batch = RB_BATCH_DEFAULT; |
79f19470 | 277 | |
e7c9f3ff NS |
278 | static int pipeline; |
279 | ||
412819ce JA |
280 | #define is_done() (*(volatile int *)(&done)) |
281 | static volatile int done; | |
282 | ||
bf0720af JA |
283 | #define JHASH_RANDOM (0x3af5f2ee) |
284 | ||
824c2b39 JA |
285 | #define CPUS_PER_LONG (8 * sizeof(unsigned long)) |
286 | #define CPU_IDX(cpu) ((cpu) / CPUS_PER_LONG) | |
287 | #define CPU_BIT(cpu) ((cpu) & (CPUS_PER_LONG - 1)) | |
288 | ||
a2594911 AB |
289 | static void output_binary(void *buf, int len) |
290 | { | |
291 | if (dump_binary) { | |
292 | int n = write(dump_fd, buf, len); | |
293 | if (n != len) { | |
294 | perror(dump_binary); | |
295 | close(dump_fd); | |
296 | dump_binary = NULL; | |
297 | } | |
298 | } | |
299 | } | |
300 | ||
210824c3 JA |
301 | static void resize_cpu_info(struct per_dev_info *pdi, int cpu) |
302 | { | |
303 | struct per_cpu_info *cpus = pdi->cpus; | |
304 | int ncpus = pdi->ncpus; | |
305 | int new_count = cpu + 1; | |
306 | int new_space, size; | |
307 | char *new_start; | |
308 | ||
309 | size = new_count * sizeof(struct per_cpu_info); | |
310 | cpus = realloc(cpus, size); | |
311 | if (!cpus) { | |
312 | char name[20]; | |
313 | fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n", | |
314 | get_dev_name(pdi, name, sizeof(name)), size); | |
315 | exit(1); | |
316 | } | |
317 | ||
318 | new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info)); | |
319 | new_space = (new_count - ncpus) * sizeof(struct per_cpu_info); | |
320 | memset(new_start, 0, new_space); | |
321 | ||
322 | pdi->ncpus = new_count; | |
323 | pdi->cpus = cpus; | |
324 | ||
325 | for (new_count = 0; new_count < pdi->ncpus; new_count++) { | |
326 | struct per_cpu_info *pci = &pdi->cpus[new_count]; | |
327 | ||
328 | if (!pci->fd) { | |
329 | pci->fd = -1; | |
330 | memset(&pci->rb_last, 0, sizeof(pci->rb_last)); | |
331 | pci->rb_last_entries = 0; | |
332 | pci->last_sequence = -1; | |
333 | } | |
334 | } | |
335 | } | |
336 | ||
337 | static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu) | |
338 | { | |
339 | struct per_cpu_info *pci; | |
340 | ||
341 | if (cpu >= pdi->ncpus) | |
342 | resize_cpu_info(pdi, cpu); | |
343 | ||
344 | pci = &pdi->cpus[cpu]; | |
345 | pci->cpu = cpu; | |
346 | return pci; | |
347 | } | |
348 | ||
349 | ||
350 | static int resize_devices(char *name) | |
351 | { | |
352 | int size = (ndevices + 1) * sizeof(struct per_dev_info); | |
353 | ||
354 | devices = realloc(devices, size); | |
355 | if (!devices) { | |
356 | fprintf(stderr, "Out of memory, device %s (%d)\n", name, size); | |
357 | return 1; | |
358 | } | |
359 | memset(&devices[ndevices], 0, sizeof(struct per_dev_info)); | |
360 | devices[ndevices].name = name; | |
361 | ndevices++; | |
362 | return 0; | |
363 | } | |
364 | ||
365 | static struct per_dev_info *get_dev_info(dev_t dev) | |
366 | { | |
367 | struct per_dev_info *pdi; | |
368 | int i; | |
369 | ||
370 | for (i = 0; i < ndevices; i++) { | |
371 | if (!devices[i].dev) | |
372 | devices[i].dev = dev; | |
373 | if (devices[i].dev == dev) | |
374 | return &devices[i]; | |
375 | } | |
376 | ||
377 | if (resize_devices(NULL)) | |
378 | return NULL; | |
379 | ||
380 | pdi = &devices[ndevices - 1]; | |
381 | pdi->dev = dev; | |
382 | pdi->first_reported_time = 0; | |
383 | pdi->last_read_time = 0; | |
210824c3 JA |
384 | |
385 | return pdi; | |
386 | } | |
387 | ||
66930177 | 388 | static void insert_skip(struct per_cpu_info *pci, unsigned long start, |
492da111 AB |
389 | unsigned long end) |
390 | { | |
391 | struct skip_info *sip; | |
392 | ||
66930177 | 393 | for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) { |
492da111 AB |
394 | if (end == (sip->start - 1)) { |
395 | sip->start = start; | |
396 | return; | |
397 | } else if (start == (sip->end + 1)) { | |
398 | sip->end = end; | |
399 | return; | |
400 | } | |
401 | } | |
402 | ||
403 | sip = malloc(sizeof(struct skip_info)); | |
404 | sip->start = start; | |
405 | sip->end = end; | |
406 | sip->prev = sip->next = NULL; | |
66930177 JA |
407 | if (pci->skips_tail == NULL) |
408 | pci->skips_head = pci->skips_tail = sip; | |
492da111 | 409 | else { |
66930177 JA |
410 | sip->prev = pci->skips_tail; |
411 | pci->skips_tail->next = sip; | |
412 | pci->skips_tail = sip; | |
492da111 AB |
413 | } |
414 | } | |
415 | ||
66930177 | 416 | static void remove_sip(struct per_cpu_info *pci, struct skip_info *sip) |
492da111 AB |
417 | { |
418 | if (sip->prev == NULL) { | |
419 | if (sip->next == NULL) | |
66930177 | 420 | pci->skips_head = pci->skips_tail = NULL; |
492da111 | 421 | else { |
66930177 | 422 | pci->skips_head = sip->next; |
492da111 AB |
423 | sip->next->prev = NULL; |
424 | } | |
425 | } else if (sip->next == NULL) { | |
66930177 | 426 | pci->skips_tail = sip->prev; |
492da111 AB |
427 | sip->prev->next = NULL; |
428 | } else { | |
429 | sip->prev->next = sip->next; | |
430 | sip->next->prev = sip->prev; | |
431 | } | |
432 | ||
433 | sip->prev = sip->next = NULL; | |
434 | free(sip); | |
435 | } | |
436 | ||
437 | #define IN_SKIP(sip,seq) (((sip)->start <= (seq)) && ((seq) <= sip->end)) | |
66930177 | 438 | static int check_current_skips(struct per_cpu_info *pci, unsigned long seq) |
492da111 AB |
439 | { |
440 | struct skip_info *sip; | |
441 | ||
66930177 JA |
442 | for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) { |
443 | if (IN_SKIP(sip, seq)) { | |
492da111 AB |
444 | if (sip->start == seq) { |
445 | if (sip->end == seq) | |
66930177 | 446 | remove_sip(pci, sip); |
492da111 AB |
447 | else |
448 | sip->start += 1; | |
449 | } else if (sip->end == seq) | |
450 | sip->end -= 1; | |
451 | else { | |
452 | sip->end = seq - 1; | |
66930177 | 453 | insert_skip(pci, seq + 1, sip->end); |
492da111 AB |
454 | } |
455 | return 1; | |
456 | } | |
457 | } | |
66930177 | 458 | |
492da111 AB |
459 | return 0; |
460 | } | |
461 | ||
462 | static void collect_pdi_skips(struct per_dev_info *pdi) | |
463 | { | |
464 | struct skip_info *sip; | |
66930177 | 465 | int cpu; |
492da111 AB |
466 | |
467 | pdi->skips = 0; | |
468 | pdi->seq_skips = 0; | |
66930177 JA |
469 | |
470 | for (cpu = 0; cpu < pdi->ncpus; cpu++) { | |
471 | struct per_cpu_info *pci = &pdi->cpus[cpu]; | |
472 | ||
473 | for (sip = pci->skips_head; sip != NULL; sip = sip->next) { | |
474 | pdi->skips++; | |
475 | pdi->seq_skips += (sip->end - sip->start + 1); | |
476 | if (verbose) | |
477 | fprintf(stderr,"(%d,%d): skipping %lu -> %lu\n", | |
478 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
479 | sip->start, sip->end); | |
480 | } | |
492da111 AB |
481 | } |
482 | } | |
483 | ||
824c2b39 JA |
484 | static void cpu_mark_online(struct per_dev_info *pdi, unsigned int cpu) |
485 | { | |
486 | if (cpu >= pdi->cpu_map_max || !pdi->cpu_map) { | |
487 | int new_max = (cpu + CPUS_PER_LONG) & ~(CPUS_PER_LONG - 1); | |
488 | unsigned long *map = malloc(new_max / sizeof(long)); | |
489 | ||
490 | memset(map, 0, new_max / sizeof(long)); | |
491 | ||
492 | if (pdi->cpu_map) { | |
493 | memcpy(map, pdi->cpu_map, pdi->cpu_map_max / sizeof(long)); | |
494 | free(pdi->cpu_map); | |
495 | } | |
496 | ||
497 | pdi->cpu_map = map; | |
498 | pdi->cpu_map_max = new_max; | |
499 | } | |
500 | ||
501 | pdi->cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu)); | |
502 | } | |
503 | ||
504 | static inline void cpu_mark_offline(struct per_dev_info *pdi, int cpu) | |
505 | { | |
506 | pdi->cpu_map[CPU_IDX(cpu)] &= ~(1UL << CPU_BIT(cpu)); | |
507 | } | |
508 | ||
509 | static inline int cpu_is_online(struct per_dev_info *pdi, int cpu) | |
510 | { | |
511 | return (pdi->cpu_map[CPU_IDX(cpu)] & (1UL << CPU_BIT(cpu))) != 0; | |
512 | } | |
513 | ||
bfc70ad5 JA |
514 | static inline int ppm_hash_pid(pid_t pid) |
515 | { | |
516 | return jhash_1word(pid, JHASH_RANDOM) & PPM_HASH_MASK; | |
517 | } | |
518 | ||
519 | static struct process_pid_map *find_ppm(pid_t pid) | |
520 | { | |
521 | const int hash_idx = ppm_hash_pid(pid); | |
522 | struct process_pid_map *ppm; | |
523 | ||
524 | ppm = ppm_hash_table[hash_idx]; | |
525 | while (ppm) { | |
526 | if (ppm->pid == pid) | |
527 | return ppm; | |
528 | ||
529 | ppm = ppm->hash_next; | |
530 | } | |
531 | ||
532 | return NULL; | |
533 | } | |
534 | ||
535 | static void add_ppm_hash(pid_t pid, const char *name) | |
536 | { | |
537 | const int hash_idx = ppm_hash_pid(pid); | |
538 | struct process_pid_map *ppm; | |
539 | ||
540 | ppm = find_ppm(pid); | |
248eac8f JA |
541 | if (!ppm) { |
542 | ppm = malloc(sizeof(*ppm)); | |
543 | memset(ppm, 0, sizeof(*ppm)); | |
544 | ppm->pid = pid; | |
545 | strcpy(ppm->comm, name); | |
546 | ppm->hash_next = ppm_hash_table[hash_idx]; | |
547 | ppm_hash_table[hash_idx] = ppm; | |
bfc70ad5 | 548 | } |
bfc70ad5 JA |
549 | } |
550 | ||
551 | char *find_process_name(pid_t pid) | |
552 | { | |
553 | struct process_pid_map *ppm = find_ppm(pid); | |
554 | ||
555 | if (ppm) | |
556 | return ppm->comm; | |
557 | ||
558 | return NULL; | |
559 | } | |
560 | ||
9e4cd1b8 | 561 | static inline int ppi_hash_pid(pid_t pid) |
bf0720af JA |
562 | { |
563 | return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK; | |
564 | } | |
565 | ||
566 | static inline int ppi_hash_name(const char *name) | |
152f6476 | 567 | { |
bf0720af JA |
568 | return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK; |
569 | } | |
570 | ||
571 | static inline int ppi_hash(struct per_process_info *ppi) | |
572 | { | |
2990e589 JA |
573 | struct process_pid_map *ppm = ppi->ppm; |
574 | ||
bf0720af | 575 | if (ppi_hash_by_pid) |
2990e589 | 576 | return ppi_hash_pid(ppm->pid); |
bf0720af | 577 | |
2990e589 | 578 | return ppi_hash_name(ppm->comm); |
152f6476 JA |
579 | } |
580 | ||
bfc70ad5 | 581 | static inline void add_ppi_to_hash(struct per_process_info *ppi) |
152f6476 | 582 | { |
bf0720af | 583 | const int hash_idx = ppi_hash(ppi); |
152f6476 | 584 | |
bf0720af JA |
585 | ppi->hash_next = ppi_hash_table[hash_idx]; |
586 | ppi_hash_table[hash_idx] = ppi; | |
152f6476 JA |
587 | } |
588 | ||
bfc70ad5 | 589 | static inline void add_ppi_to_list(struct per_process_info *ppi) |
152f6476 JA |
590 | { |
591 | ppi->list_next = ppi_list; | |
592 | ppi_list = ppi; | |
886ecf0e | 593 | ppi_list_entries++; |
152f6476 JA |
594 | } |
595 | ||
bfc70ad5 | 596 | static struct per_process_info *find_ppi_by_name(char *name) |
bf0720af JA |
597 | { |
598 | const int hash_idx = ppi_hash_name(name); | |
599 | struct per_process_info *ppi; | |
600 | ||
601 | ppi = ppi_hash_table[hash_idx]; | |
602 | while (ppi) { | |
2990e589 JA |
603 | struct process_pid_map *ppm = ppi->ppm; |
604 | ||
605 | if (!strcmp(ppm->comm, name)) | |
bf0720af JA |
606 | return ppi; |
607 | ||
608 | ppi = ppi->hash_next; | |
609 | } | |
610 | ||
611 | return NULL; | |
612 | } | |
613 | ||
9e4cd1b8 | 614 | static struct per_process_info *find_ppi_by_pid(pid_t pid) |
152f6476 | 615 | { |
bf0720af | 616 | const int hash_idx = ppi_hash_pid(pid); |
152f6476 JA |
617 | struct per_process_info *ppi; |
618 | ||
bf0720af | 619 | ppi = ppi_hash_table[hash_idx]; |
152f6476 | 620 | while (ppi) { |
2990e589 JA |
621 | struct process_pid_map *ppm = ppi->ppm; |
622 | ||
623 | if (ppm->pid == pid) | |
152f6476 JA |
624 | return ppi; |
625 | ||
626 | ppi = ppi->hash_next; | |
627 | } | |
628 | ||
629 | return NULL; | |
630 | } | |
631 | ||
9e4cd1b8 | 632 | static struct per_process_info *find_ppi(pid_t pid) |
bf0720af | 633 | { |
715d8021 | 634 | struct per_process_info *ppi; |
bfc70ad5 | 635 | char *name; |
715d8021 | 636 | |
bf0720af | 637 | if (ppi_hash_by_pid) |
bfc70ad5 JA |
638 | return find_ppi_by_pid(pid); |
639 | ||
640 | name = find_process_name(pid); | |
641 | if (!name) | |
642 | return NULL; | |
bf0720af | 643 | |
bfc70ad5 | 644 | ppi = find_ppi_by_name(name); |
2990e589 | 645 | if (ppi && ppi->ppm->pid != pid) |
715d8021 JA |
646 | ppi->more_than_one = 1; |
647 | ||
648 | return ppi; | |
bf0720af JA |
649 | } |
650 | ||
210824c3 JA |
651 | /* |
652 | * struct trace and blktrace allocation cache, we do potentially | |
653 | * millions of mallocs for these structures while only using at most | |
654 | * a few thousand at the time | |
655 | */ | |
656 | static inline void t_free(struct trace *t) | |
657 | { | |
658 | if (t_alloc_cache < 1024) { | |
659 | t->next = t_alloc_list; | |
660 | t_alloc_list = t; | |
661 | t_alloc_cache++; | |
662 | } else | |
663 | free(t); | |
664 | } | |
665 | ||
666 | static inline struct trace *t_alloc(void) | |
667 | { | |
668 | struct trace *t = t_alloc_list; | |
669 | ||
670 | if (t) { | |
671 | t_alloc_list = t->next; | |
672 | t_alloc_cache--; | |
673 | return t; | |
674 | } | |
675 | ||
676 | return malloc(sizeof(*t)); | |
677 | } | |
678 | ||
679 | static inline void bit_free(struct blk_io_trace *bit) | |
680 | { | |
681 | if (bit_alloc_cache < 1024 && !bit->pdu_len) { | |
682 | /* | |
683 | * abuse a 64-bit field for a next pointer for the free item | |
684 | */ | |
685 | bit->time = (__u64) (unsigned long) bit_alloc_list; | |
686 | bit_alloc_list = (struct blk_io_trace *) bit; | |
687 | bit_alloc_cache++; | |
688 | } else | |
689 | free(bit); | |
690 | } | |
691 | ||
692 | static inline struct blk_io_trace *bit_alloc(void) | |
693 | { | |
694 | struct blk_io_trace *bit = bit_alloc_list; | |
695 | ||
696 | if (bit) { | |
697 | bit_alloc_list = (struct blk_io_trace *) (unsigned long) \ | |
698 | bit->time; | |
699 | bit_alloc_cache--; | |
700 | return bit; | |
701 | } | |
702 | ||
703 | return malloc(sizeof(*bit)); | |
704 | } | |
705 | ||
706 | static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t) | |
707 | { | |
708 | struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu); | |
709 | ||
710 | rb_erase(&t->rb_node, &pci->rb_last); | |
711 | pci->rb_last_entries--; | |
712 | ||
713 | bit_free(t->bit); | |
714 | t_free(t); | |
715 | } | |
716 | ||
717 | static void put_trace(struct per_dev_info *pdi, struct trace *t) | |
718 | { | |
719 | rb_erase(&t->rb_node, &rb_sort_root); | |
720 | rb_sort_entries--; | |
721 | ||
722 | trace_rb_insert_last(pdi, t); | |
723 | } | |
724 | ||
89482da6 | 725 | static inline int trace_rb_insert(struct trace *t, struct rb_root *root) |
7997c5b0 | 726 | { |
2a1b3424 | 727 | struct rb_node **p = &root->rb_node; |
7997c5b0 JA |
728 | struct rb_node *parent = NULL; |
729 | struct trace *__t; | |
730 | ||
731 | while (*p) { | |
732 | parent = *p; | |
2a1b3424 | 733 | |
7997c5b0 JA |
734 | __t = rb_entry(parent, struct trace, rb_node); |
735 | ||
89482da6 JA |
736 | if (t->bit->time < __t->bit->time) |
737 | p = &(*p)->rb_left; | |
738 | else if (t->bit->time > __t->bit->time) | |
739 | p = &(*p)->rb_right; | |
740 | else if (t->bit->device < __t->bit->device) | |
e7c9f3ff NS |
741 | p = &(*p)->rb_left; |
742 | else if (t->bit->device > __t->bit->device) | |
743 | p = &(*p)->rb_right; | |
dcf0f7ed JA |
744 | else if (t->bit->sequence < __t->bit->sequence) |
745 | p = &(*p)->rb_left; | |
0b07f23e | 746 | else /* >= sequence */ |
dcf0f7ed | 747 | p = &(*p)->rb_right; |
7997c5b0 JA |
748 | } |
749 | ||
750 | rb_link_node(&t->rb_node, parent, p); | |
2a1b3424 | 751 | rb_insert_color(&t->rb_node, root); |
7997c5b0 JA |
752 | return 0; |
753 | } | |
754 | ||
2a1b3424 | 755 | static inline int trace_rb_insert_sort(struct trace *t) |
e3556946 | 756 | { |
89482da6 | 757 | if (!trace_rb_insert(t, &rb_sort_root)) { |
2a1b3424 JA |
758 | rb_sort_entries++; |
759 | return 0; | |
760 | } | |
761 | ||
762 | return 1; | |
763 | } | |
764 | ||
210824c3 | 765 | static int trace_rb_insert_last(struct per_dev_info *pdi, struct trace *t) |
2a1b3424 | 766 | { |
210824c3 JA |
767 | struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu); |
768 | ||
769 | if (trace_rb_insert(t, &pci->rb_last)) | |
770 | return 1; | |
771 | ||
772 | pci->rb_last_entries++; | |
773 | ||
774 | if (pci->rb_last_entries > rb_batch * pdi->nfiles) { | |
775 | struct rb_node *n = rb_first(&pci->rb_last); | |
776 | ||
777 | t = rb_entry(n, struct trace, rb_node); | |
778 | __put_trace_last(pdi, t); | |
2a1b3424 JA |
779 | } |
780 | ||
210824c3 | 781 | return 0; |
2a1b3424 JA |
782 | } |
783 | ||
784 | static struct trace *trace_rb_find(dev_t device, unsigned long sequence, | |
785 | struct rb_root *root, int order) | |
786 | { | |
787 | struct rb_node *n = root->rb_node; | |
788 | struct rb_node *prev = NULL; | |
e3556946 JA |
789 | struct trace *__t; |
790 | ||
2a1b3424 JA |
791 | while (n) { |
792 | __t = rb_entry(n, struct trace, rb_node); | |
793 | prev = n; | |
e3556946 | 794 | |
0583b6a2 | 795 | if (device < __t->bit->device) |
2a1b3424 | 796 | n = n->rb_left; |
0583b6a2 | 797 | else if (device > __t->bit->device) |
2a1b3424 | 798 | n = n->rb_right; |
0583b6a2 | 799 | else if (sequence < __t->bit->sequence) |
2a1b3424 | 800 | n = n->rb_left; |
e3556946 | 801 | else if (sequence > __t->bit->sequence) |
2a1b3424 | 802 | n = n->rb_right; |
e3556946 JA |
803 | else |
804 | return __t; | |
805 | } | |
806 | ||
2a1b3424 JA |
807 | /* |
808 | * hack - the list may not be sequence ordered because some | |
809 | * events don't have sequence and time matched. so we end up | |
810 | * being a little off in the rb lookup here, because we don't | |
811 | * know the time we are looking for. compensate by browsing | |
812 | * a little ahead from the last entry to find the match | |
813 | */ | |
814 | if (order && prev) { | |
815 | int max = 5; | |
816 | ||
817 | while (((n = rb_next(prev)) != NULL) && max--) { | |
818 | __t = rb_entry(n, struct trace, rb_node); | |
492da111 | 819 | |
2a1b3424 JA |
820 | if (__t->bit->device == device && |
821 | __t->bit->sequence == sequence) | |
822 | return __t; | |
823 | ||
824 | prev = n; | |
825 | } | |
826 | } | |
492da111 | 827 | |
e3556946 JA |
828 | return NULL; |
829 | } | |
830 | ||
2a1b3424 | 831 | static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi, |
210824c3 | 832 | struct per_cpu_info *pci, |
2a1b3424 JA |
833 | unsigned long seq) |
834 | { | |
210824c3 | 835 | return trace_rb_find(pdi->dev, seq, &pci->rb_last, 0); |
2a1b3424 JA |
836 | } |
837 | ||
f7bd1a9b | 838 | static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot) |
7997c5b0 | 839 | { |
f7bd1a9b | 840 | struct rb_node **p = &pdi->rb_track.rb_node; |
7997c5b0 JA |
841 | struct rb_node *parent = NULL; |
842 | struct io_track *__iot; | |
843 | ||
844 | while (*p) { | |
845 | parent = *p; | |
7997c5b0 JA |
846 | __iot = rb_entry(parent, struct io_track, rb_node); |
847 | ||
f7bd1a9b | 848 | if (iot->sector < __iot->sector) |
7997c5b0 JA |
849 | p = &(*p)->rb_left; |
850 | else if (iot->sector > __iot->sector) | |
851 | p = &(*p)->rb_right; | |
852 | else { | |
e7c9f3ff | 853 | fprintf(stderr, |
ab197ca7 AB |
854 | "sector alias (%Lu) on device %d,%d!\n", |
855 | (unsigned long long) iot->sector, | |
f7bd1a9b | 856 | MAJOR(pdi->dev), MINOR(pdi->dev)); |
7997c5b0 JA |
857 | return 1; |
858 | } | |
859 | } | |
860 | ||
861 | rb_link_node(&iot->rb_node, parent, p); | |
f7bd1a9b | 862 | rb_insert_color(&iot->rb_node, &pdi->rb_track); |
7997c5b0 JA |
863 | return 0; |
864 | } | |
865 | ||
f7bd1a9b | 866 | static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector) |
7997c5b0 | 867 | { |
f7bd1a9b | 868 | struct rb_node *n = pdi->rb_track.rb_node; |
7997c5b0 JA |
869 | struct io_track *__iot; |
870 | ||
2a1b3424 JA |
871 | while (n) { |
872 | __iot = rb_entry(n, struct io_track, rb_node); | |
7997c5b0 | 873 | |
f7bd1a9b | 874 | if (sector < __iot->sector) |
2a1b3424 | 875 | n = n->rb_left; |
7997c5b0 | 876 | else if (sector > __iot->sector) |
2a1b3424 | 877 | n = n->rb_right; |
7997c5b0 JA |
878 | else |
879 | return __iot; | |
880 | } | |
881 | ||
882 | return NULL; | |
883 | } | |
884 | ||
9e4cd1b8 | 885 | static struct io_track *find_track(struct per_dev_info *pdi, pid_t pid, |
bfc70ad5 | 886 | __u64 sector) |
7997c5b0 | 887 | { |
916b5501 | 888 | struct io_track *iot; |
7997c5b0 | 889 | |
f7bd1a9b | 890 | iot = __find_track(pdi, sector); |
7997c5b0 JA |
891 | if (!iot) { |
892 | iot = malloc(sizeof(*iot)); | |
2990e589 | 893 | iot->ppm = find_ppm(pid); |
7997c5b0 | 894 | iot->sector = sector; |
f7bd1a9b | 895 | track_rb_insert(pdi, iot); |
7997c5b0 JA |
896 | } |
897 | ||
898 | return iot; | |
899 | } | |
900 | ||
f7bd1a9b JA |
901 | static void log_track_frontmerge(struct per_dev_info *pdi, |
902 | struct blk_io_trace *t) | |
2e3e8ded JA |
903 | { |
904 | struct io_track *iot; | |
905 | ||
906 | if (!track_ios) | |
907 | return; | |
2e3e8ded | 908 | |
ae957cbc | 909 | iot = __find_track(pdi, t->sector + t_sec(t)); |
cb2a1a62 | 910 | if (!iot) { |
57ea8602 JA |
911 | if (verbose) |
912 | fprintf(stderr, "merge not found for (%d,%d): %llu\n", | |
913 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
914 | (unsigned long long) t->sector + t_sec(t)); | |
cb2a1a62 | 915 | return; |
2e3e8ded | 916 | } |
cb2a1a62 | 917 | |
f7bd1a9b | 918 | rb_erase(&iot->rb_node, &pdi->rb_track); |
ae957cbc | 919 | iot->sector -= t_sec(t); |
f7bd1a9b | 920 | track_rb_insert(pdi, iot); |
2e3e8ded JA |
921 | } |
922 | ||
f7bd1a9b | 923 | static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t) |
2e3e8ded JA |
924 | { |
925 | struct io_track *iot; | |
926 | ||
927 | if (!track_ios) | |
928 | return; | |
929 | ||
bfc70ad5 | 930 | iot = find_track(pdi, t->pid, t->sector); |
95c15013 JA |
931 | iot->allocation_time = t->time; |
932 | } | |
933 | ||
753f9091 JA |
934 | static inline int is_remapper(struct per_dev_info *pdi) |
935 | { | |
936 | int major = MAJOR(pdi->dev); | |
937 | ||
938 | return (major == 253 || major == 9); | |
939 | } | |
940 | ||
941 | /* | |
942 | * for md/dm setups, the interesting cycle is Q -> C. So track queueing | |
943 | * time here, as dispatch time | |
944 | */ | |
945 | static void log_track_queue(struct per_dev_info *pdi, struct blk_io_trace *t) | |
946 | { | |
947 | struct io_track *iot; | |
948 | ||
949 | if (!track_ios) | |
950 | return; | |
951 | if (!is_remapper(pdi)) | |
952 | return; | |
953 | ||
bfc70ad5 | 954 | iot = find_track(pdi, t->pid, t->sector); |
753f9091 JA |
955 | iot->dispatch_time = t->time; |
956 | } | |
957 | ||
95c15013 | 958 | /* |
b6076a9b | 959 | * return time between rq allocation and insertion |
95c15013 | 960 | */ |
f7bd1a9b JA |
961 | static unsigned long long log_track_insert(struct per_dev_info *pdi, |
962 | struct blk_io_trace *t) | |
95c15013 | 963 | { |
50adc0ba | 964 | unsigned long long elapsed; |
95c15013 JA |
965 | struct io_track *iot; |
966 | ||
967 | if (!track_ios) | |
968 | return -1; | |
969 | ||
bfc70ad5 | 970 | iot = find_track(pdi, t->pid, t->sector); |
2e3e8ded | 971 | iot->queue_time = t->time; |
acd70d21 JA |
972 | |
973 | if (!iot->allocation_time) | |
974 | return -1; | |
975 | ||
50adc0ba JA |
976 | elapsed = iot->queue_time - iot->allocation_time; |
977 | ||
978 | if (per_process_stats) { | |
2990e589 | 979 | struct per_process_info *ppi = find_ppi(iot->ppm->pid); |
b9d40d6f | 980 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 981 | |
b9d40d6f JA |
982 | if (ppi && elapsed > ppi->longest_allocation_wait[w]) |
983 | ppi->longest_allocation_wait[w] = elapsed; | |
50adc0ba JA |
984 | } |
985 | ||
986 | return elapsed; | |
2e3e8ded JA |
987 | } |
988 | ||
989 | /* | |
990 | * return time between queue and issue | |
991 | */ | |
f7bd1a9b JA |
992 | static unsigned long long log_track_issue(struct per_dev_info *pdi, |
993 | struct blk_io_trace *t) | |
2e3e8ded | 994 | { |
50adc0ba | 995 | unsigned long long elapsed; |
2e3e8ded JA |
996 | struct io_track *iot; |
997 | ||
998 | if (!track_ios) | |
999 | return -1; | |
1000 | if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0) | |
1001 | return -1; | |
1002 | ||
f7bd1a9b | 1003 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 1004 | if (!iot) { |
57ea8602 JA |
1005 | if (verbose) |
1006 | fprintf(stderr, "issue not found for (%d,%d): %llu\n", | |
1007 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
1008 | (unsigned long long) t->sector); | |
2e3e8ded | 1009 | return -1; |
cb2a1a62 | 1010 | } |
2e3e8ded JA |
1011 | |
1012 | iot->dispatch_time = t->time; | |
50adc0ba JA |
1013 | elapsed = iot->dispatch_time - iot->queue_time; |
1014 | ||
1015 | if (per_process_stats) { | |
2990e589 | 1016 | struct per_process_info *ppi = find_ppi(iot->ppm->pid); |
b9d40d6f | 1017 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 1018 | |
b9d40d6f JA |
1019 | if (ppi && elapsed > ppi->longest_dispatch_wait[w]) |
1020 | ppi->longest_dispatch_wait[w] = elapsed; | |
50adc0ba JA |
1021 | } |
1022 | ||
1023 | return elapsed; | |
2e3e8ded JA |
1024 | } |
1025 | ||
1026 | /* | |
1027 | * return time between dispatch and complete | |
1028 | */ | |
f7bd1a9b JA |
1029 | static unsigned long long log_track_complete(struct per_dev_info *pdi, |
1030 | struct blk_io_trace *t) | |
2e3e8ded JA |
1031 | { |
1032 | unsigned long long elapsed; | |
1033 | struct io_track *iot; | |
1034 | ||
1035 | if (!track_ios) | |
1036 | return -1; | |
2e3e8ded | 1037 | |
f7bd1a9b | 1038 | iot = __find_track(pdi, t->sector); |
cb2a1a62 | 1039 | if (!iot) { |
57ea8602 JA |
1040 | if (verbose) |
1041 | fprintf(stderr,"complete not found for (%d,%d): %llu\n", | |
1042 | MAJOR(pdi->dev), MINOR(pdi->dev), | |
1043 | (unsigned long long) t->sector); | |
2e3e8ded | 1044 | return -1; |
cb2a1a62 | 1045 | } |
2e3e8ded JA |
1046 | |
1047 | iot->completion_time = t->time; | |
1048 | elapsed = iot->completion_time - iot->dispatch_time; | |
1049 | ||
50adc0ba | 1050 | if (per_process_stats) { |
2990e589 | 1051 | struct per_process_info *ppi = find_ppi(iot->ppm->pid); |
b9d40d6f | 1052 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
50adc0ba | 1053 | |
b9d40d6f JA |
1054 | if (ppi && elapsed > ppi->longest_completion_wait[w]) |
1055 | ppi->longest_completion_wait[w] = elapsed; | |
50adc0ba JA |
1056 | } |
1057 | ||
2e3e8ded JA |
1058 | /* |
1059 | * kill the trace, we don't need it after completion | |
1060 | */ | |
f7bd1a9b | 1061 | rb_erase(&iot->rb_node, &pdi->rb_track); |
2e3e8ded JA |
1062 | free(iot); |
1063 | ||
1064 | return elapsed; | |
1065 | } | |
1066 | ||
1067 | ||
9e4cd1b8 | 1068 | static struct io_stats *find_process_io_stats(pid_t pid) |
152f6476 | 1069 | { |
bfc70ad5 | 1070 | struct per_process_info *ppi = find_ppi(pid); |
152f6476 JA |
1071 | |
1072 | if (!ppi) { | |
1073 | ppi = malloc(sizeof(*ppi)); | |
1074 | memset(ppi, 0, sizeof(*ppi)); | |
2990e589 | 1075 | ppi->ppm = find_ppm(pid); |
bfc70ad5 JA |
1076 | add_ppi_to_hash(ppi); |
1077 | add_ppi_to_list(ppi); | |
152f6476 JA |
1078 | } |
1079 | ||
1080 | return &ppi->io_stats; | |
1081 | } | |
1082 | ||
e7c9f3ff NS |
1083 | static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size) |
1084 | { | |
1085 | if (pdi->name) | |
1086 | snprintf(buffer, size, "%s", pdi->name); | |
1087 | else | |
f7bd1a9b | 1088 | snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev)); |
e7c9f3ff NS |
1089 | return buffer; |
1090 | } | |
1091 | ||
e7c9f3ff | 1092 | static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit) |
cfab07eb AB |
1093 | { |
1094 | unsigned long long this = bit->time; | |
e7c9f3ff | 1095 | unsigned long long last = pdi->last_reported_time; |
cfab07eb | 1096 | |
e7c9f3ff NS |
1097 | pdi->backwards = (this < last) ? 'B' : ' '; |
1098 | pdi->last_reported_time = this; | |
cfab07eb AB |
1099 | } |
1100 | ||
152f6476 JA |
1101 | static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t, |
1102 | int rw) | |
d0ca268b JA |
1103 | { |
1104 | if (rw) { | |
152f6476 | 1105 | ios->mwrites++; |
ae957cbc | 1106 | ios->qwrite_kb += t_kb(t); |
d0ca268b | 1107 | } else { |
152f6476 | 1108 | ios->mreads++; |
ae957cbc | 1109 | ios->qread_kb += t_kb(t); |
152f6476 JA |
1110 | } |
1111 | } | |
1112 | ||
1113 | static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci, | |
1114 | int rw) | |
1115 | { | |
1116 | __account_m(&pci->io_stats, t, rw); | |
1117 | ||
1118 | if (per_process_stats) { | |
bfc70ad5 | 1119 | struct io_stats *ios = find_process_io_stats(t->pid); |
152f6476 JA |
1120 | |
1121 | __account_m(ios, t, rw); | |
d0ca268b JA |
1122 | } |
1123 | } | |
1124 | ||
b6076a9b JA |
1125 | static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t, |
1126 | int rw) | |
d0ca268b JA |
1127 | { |
1128 | if (rw) { | |
152f6476 | 1129 | ios->qwrites++; |
ae957cbc | 1130 | ios->qwrite_kb += t_kb(t); |
d0ca268b | 1131 | } else { |
152f6476 | 1132 | ios->qreads++; |
ae957cbc | 1133 | ios->qread_kb += t_kb(t); |
152f6476 JA |
1134 | } |
1135 | } | |
1136 | ||
b6076a9b JA |
1137 | static inline void account_queue(struct blk_io_trace *t, |
1138 | struct per_cpu_info *pci, int rw) | |
152f6476 | 1139 | { |
b6076a9b | 1140 | __account_queue(&pci->io_stats, t, rw); |
152f6476 JA |
1141 | |
1142 | if (per_process_stats) { | |
bfc70ad5 | 1143 | struct io_stats *ios = find_process_io_stats(t->pid); |
152f6476 | 1144 | |
b6076a9b | 1145 | __account_queue(ios, t, rw); |
d0ca268b JA |
1146 | } |
1147 | } | |
1148 | ||
e21dc4dd | 1149 | static inline void __account_c(struct io_stats *ios, int rw, int bytes) |
d0ca268b JA |
1150 | { |
1151 | if (rw) { | |
152f6476 JA |
1152 | ios->cwrites++; |
1153 | ios->cwrite_kb += bytes >> 10; | |
d0ca268b | 1154 | } else { |
152f6476 JA |
1155 | ios->creads++; |
1156 | ios->cread_kb += bytes >> 10; | |
1157 | } | |
1158 | } | |
1159 | ||
1160 | static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci, | |
1161 | int rw, int bytes) | |
1162 | { | |
1163 | __account_c(&pci->io_stats, rw, bytes); | |
1164 | ||
1165 | if (per_process_stats) { | |
bfc70ad5 | 1166 | struct io_stats *ios = find_process_io_stats(t->pid); |
152f6476 JA |
1167 | |
1168 | __account_c(ios, rw, bytes); | |
d0ca268b JA |
1169 | } |
1170 | } | |
1171 | ||
b6076a9b JA |
1172 | static inline void __account_issue(struct io_stats *ios, int rw, |
1173 | unsigned int bytes) | |
afd2d7ad | 1174 | { |
1175 | if (rw) { | |
152f6476 JA |
1176 | ios->iwrites++; |
1177 | ios->iwrite_kb += bytes >> 10; | |
afd2d7ad | 1178 | } else { |
152f6476 JA |
1179 | ios->ireads++; |
1180 | ios->iread_kb += bytes >> 10; | |
afd2d7ad | 1181 | } |
1182 | } | |
1183 | ||
b6076a9b JA |
1184 | static inline void account_issue(struct blk_io_trace *t, |
1185 | struct per_cpu_info *pci, int rw) | |
d0ca268b | 1186 | { |
b6076a9b | 1187 | __account_issue(&pci->io_stats, rw, t->bytes); |
152f6476 JA |
1188 | |
1189 | if (per_process_stats) { | |
bfc70ad5 | 1190 | struct io_stats *ios = find_process_io_stats(t->pid); |
d5396421 | 1191 | |
b6076a9b | 1192 | __account_issue(ios, rw, t->bytes); |
152f6476 JA |
1193 | } |
1194 | } | |
1195 | ||
06639b27 JA |
1196 | static inline void __account_unplug(struct io_stats *ios, int timer) |
1197 | { | |
1198 | if (timer) | |
1199 | ios->timer_unplugs++; | |
1200 | else | |
1201 | ios->io_unplugs++; | |
1202 | } | |
1203 | ||
1204 | static inline void account_unplug(struct blk_io_trace *t, | |
1205 | struct per_cpu_info *pci, int timer) | |
1206 | { | |
1207 | __account_unplug(&pci->io_stats, timer); | |
1208 | ||
1209 | if (per_process_stats) { | |
bfc70ad5 | 1210 | struct io_stats *ios = find_process_io_stats(t->pid); |
06639b27 JA |
1211 | |
1212 | __account_unplug(ios, timer); | |
1213 | } | |
1214 | } | |
1215 | ||
4054070a JA |
1216 | static inline void __account_requeue(struct io_stats *ios, |
1217 | struct blk_io_trace *t, int rw) | |
1218 | { | |
1219 | if (rw) { | |
1220 | ios->wrqueue++; | |
1221 | ios->iwrite_kb -= t_kb(t); | |
1222 | } else { | |
1223 | ios->rrqueue++; | |
1224 | ios->iread_kb -= t_kb(t); | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | static inline void account_requeue(struct blk_io_trace *t, | |
1229 | struct per_cpu_info *pci, int rw) | |
1230 | { | |
1231 | __account_requeue(&pci->io_stats, t, rw); | |
1232 | ||
1233 | if (per_process_stats) { | |
bfc70ad5 | 1234 | struct io_stats *ios = find_process_io_stats(t->pid); |
4054070a JA |
1235 | |
1236 | __account_requeue(ios, t, rw); | |
1237 | } | |
1238 | } | |
1239 | ||
f7bd1a9b JA |
1240 | static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci, |
1241 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 1242 | { |
f7bd1a9b | 1243 | process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL); |
ab197ca7 AB |
1244 | } |
1245 | ||
f7bd1a9b JA |
1246 | static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci, |
1247 | struct blk_io_trace *t, char *act) | |
b6076a9b | 1248 | { |
f7bd1a9b | 1249 | process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL); |
b6076a9b JA |
1250 | } |
1251 | ||
ab197ca7 AB |
1252 | static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t, |
1253 | char *act) | |
1254 | { | |
b6076a9b | 1255 | process_fmt(act, pci, t, -1, 0, NULL); |
ab197ca7 | 1256 | } |
2e3e8ded | 1257 | |
f7bd1a9b JA |
1258 | static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci, |
1259 | struct blk_io_trace *t, char *act) | |
ab197ca7 | 1260 | { |
f7bd1a9b | 1261 | process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL); |
d0ca268b JA |
1262 | } |
1263 | ||
f7bd1a9b JA |
1264 | static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci, |
1265 | struct blk_io_trace *t, char *act) | |
d0ca268b | 1266 | { |
a01516de | 1267 | if (act[0] == 'F') |
f7bd1a9b | 1268 | log_track_frontmerge(pdi, t); |
2e3e8ded | 1269 | |
ab197ca7 | 1270 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
1271 | } |
1272 | ||
dfe34da1 | 1273 | static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1274 | char *act) |
dfe34da1 | 1275 | { |
ab197ca7 | 1276 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
dfe34da1 JA |
1277 | } |
1278 | ||
d5396421 | 1279 | static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1280 | char *act) |
d0ca268b | 1281 | { |
ab197ca7 | 1282 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
d0ca268b JA |
1283 | } |
1284 | ||
ab197ca7 | 1285 | static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t, |
3639a11e | 1286 | char *act) |
67e14fdc | 1287 | { |
ab197ca7 | 1288 | process_fmt(act, pci, t, -1ULL, 0, NULL); |
67e14fdc JA |
1289 | } |
1290 | ||
93f1c611 JA |
1291 | static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t, |
1292 | char *act) | |
1293 | { | |
1294 | process_fmt(act, pci, t, -1ULL, 0, NULL); | |
1295 | } | |
1296 | ||
ab197ca7 | 1297 | static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act) |
d0ca268b | 1298 | { |
ab197ca7 | 1299 | unsigned char *buf = (unsigned char *) t + sizeof(*t); |
d0ca268b | 1300 | |
ab197ca7 | 1301 | process_fmt(act, pci, t, -1ULL, t->pdu_len, buf); |
d0ca268b JA |
1302 | } |
1303 | ||
ff3a732c | 1304 | static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci) |
d0ca268b | 1305 | { |
56f2af81 JA |
1306 | int act = t->action & 0xffff; |
1307 | ||
1308 | switch (act) { | |
d0ca268b | 1309 | case __BLK_TA_QUEUE: |
3639a11e | 1310 | log_generic(pci, t, "Q"); |
d0ca268b JA |
1311 | break; |
1312 | case __BLK_TA_GETRQ: | |
3639a11e | 1313 | log_generic(pci, t, "G"); |
d0ca268b JA |
1314 | break; |
1315 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 1316 | log_generic(pci, t, "S"); |
d0ca268b JA |
1317 | break; |
1318 | case __BLK_TA_REQUEUE: | |
3639a11e | 1319 | log_generic(pci, t, "R"); |
d0ca268b JA |
1320 | break; |
1321 | case __BLK_TA_ISSUE: | |
ab197ca7 | 1322 | log_pc(pci, t, "D"); |
d0ca268b JA |
1323 | break; |
1324 | case __BLK_TA_COMPLETE: | |
3639a11e | 1325 | log_pc(pci, t, "C"); |
d0ca268b | 1326 | break; |
56f2af81 JA |
1327 | case __BLK_TA_INSERT: |
1328 | log_pc(pci, t, "I"); | |
1329 | break; | |
d0ca268b | 1330 | default: |
56f2af81 | 1331 | fprintf(stderr, "Bad pc action %x\n", act); |
87b72777 | 1332 | break; |
d0ca268b | 1333 | } |
d0ca268b JA |
1334 | } |
1335 | ||
f7bd1a9b JA |
1336 | static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi, |
1337 | struct per_cpu_info *pci) | |
d0ca268b | 1338 | { |
649c7b66 | 1339 | int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; |
7997c5b0 | 1340 | int act = t->action & 0xffff; |
d0ca268b | 1341 | |
7997c5b0 | 1342 | switch (act) { |
d0ca268b | 1343 | case __BLK_TA_QUEUE: |
753f9091 | 1344 | log_track_queue(pdi, t); |
b6076a9b | 1345 | account_queue(t, pci, w); |
3639a11e | 1346 | log_queue(pci, t, "Q"); |
d0ca268b | 1347 | break; |
b6076a9b | 1348 | case __BLK_TA_INSERT: |
f7bd1a9b | 1349 | log_insert(pdi, pci, t, "I"); |
b6076a9b | 1350 | break; |
d0ca268b | 1351 | case __BLK_TA_BACKMERGE: |
152f6476 | 1352 | account_m(t, pci, w); |
f7bd1a9b | 1353 | log_merge(pdi, pci, t, "M"); |
d0ca268b JA |
1354 | break; |
1355 | case __BLK_TA_FRONTMERGE: | |
152f6476 | 1356 | account_m(t, pci, w); |
f7bd1a9b | 1357 | log_merge(pdi, pci, t, "F"); |
d0ca268b JA |
1358 | break; |
1359 | case __BLK_TA_GETRQ: | |
f7bd1a9b | 1360 | log_track_getrq(pdi, t); |
3639a11e | 1361 | log_generic(pci, t, "G"); |
d0ca268b JA |
1362 | break; |
1363 | case __BLK_TA_SLEEPRQ: | |
3639a11e | 1364 | log_generic(pci, t, "S"); |
d0ca268b JA |
1365 | break; |
1366 | case __BLK_TA_REQUEUE: | |
65f2deb5 JA |
1367 | /* |
1368 | * can happen if we miss traces, don't let it go | |
1369 | * below zero | |
1370 | */ | |
1371 | if (pdi->cur_depth[w]) | |
1372 | pdi->cur_depth[w]--; | |
4054070a | 1373 | account_requeue(t, pci, w); |
3639a11e | 1374 | log_queue(pci, t, "R"); |
d0ca268b JA |
1375 | break; |
1376 | case __BLK_TA_ISSUE: | |
b6076a9b | 1377 | account_issue(t, pci, w); |
649c7b66 JA |
1378 | pdi->cur_depth[w]++; |
1379 | if (pdi->cur_depth[w] > pdi->max_depth[w]) | |
1380 | pdi->max_depth[w] = pdi->cur_depth[w]; | |
f7bd1a9b | 1381 | log_issue(pdi, pci, t, "D"); |
d0ca268b JA |
1382 | break; |
1383 | case __BLK_TA_COMPLETE: | |
65f2deb5 JA |
1384 | if (pdi->cur_depth[w]) |
1385 | pdi->cur_depth[w]--; | |
152f6476 | 1386 | account_c(t, pci, w, t->bytes); |
f7bd1a9b | 1387 | log_complete(pdi, pci, t, "C"); |
d0ca268b | 1388 | break; |
88b1a526 | 1389 | case __BLK_TA_PLUG: |
3639a11e | 1390 | log_action(pci, t, "P"); |
88b1a526 | 1391 | break; |
3639a11e | 1392 | case __BLK_TA_UNPLUG_IO: |
06639b27 | 1393 | account_unplug(t, pci, 0); |
3639a11e JA |
1394 | log_unplug(pci, t, "U"); |
1395 | break; | |
1396 | case __BLK_TA_UNPLUG_TIMER: | |
06639b27 | 1397 | account_unplug(t, pci, 1); |
3639a11e | 1398 | log_unplug(pci, t, "UT"); |
88b1a526 | 1399 | break; |
93f1c611 JA |
1400 | case __BLK_TA_SPLIT: |
1401 | log_split(pci, t, "X"); | |
1402 | break; | |
1403 | case __BLK_TA_BOUNCE: | |
1404 | log_generic(pci, t, "B"); | |
1405 | break; | |
a8f30e64 JA |
1406 | case __BLK_TA_REMAP: |
1407 | log_generic(pci, t, "A"); | |
1408 | break; | |
d0ca268b JA |
1409 | default: |
1410 | fprintf(stderr, "Bad fs action %x\n", t->action); | |
1f79c4a0 | 1411 | break; |
d0ca268b | 1412 | } |
d0ca268b JA |
1413 | } |
1414 | ||
ff3a732c JA |
1415 | static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci, |
1416 | struct per_dev_info *pdi) | |
d0ca268b JA |
1417 | { |
1418 | if (t->action & BLK_TC_ACT(BLK_TC_PC)) | |
ff3a732c | 1419 | dump_trace_pc(t, pci); |
d0ca268b | 1420 | else |
f7bd1a9b | 1421 | dump_trace_fs(t, pdi, pci); |
87b72777 | 1422 | |
20ed6177 JA |
1423 | if (!pdi->events) |
1424 | pdi->first_reported_time = t->time; | |
1425 | ||
e7c9f3ff | 1426 | pdi->events++; |
a2594911 AB |
1427 | |
1428 | output_binary(t, sizeof(*t) + t->pdu_len); | |
d0ca268b JA |
1429 | } |
1430 | ||
4c523165 JA |
1431 | /* |
1432 | * print in a proper way, not too small and not too big. if more than | |
1433 | * 1000,000K, turn into M and so on | |
1434 | */ | |
1435 | static char *size_cnv(char *dst, unsigned long long num, int in_kb) | |
1436 | { | |
da19e768 | 1437 | char suff[] = { '\0', 'K', 'M', 'G', 'P' }; |
0dc3602c | 1438 | unsigned int i = 0; |
4c523165 JA |
1439 | |
1440 | if (in_kb) | |
1441 | i++; | |
1442 | ||
0dc3602c | 1443 | while (num > 1000 * 1000ULL && (i < sizeof(suff) - 1)) { |
4c523165 JA |
1444 | i++; |
1445 | num /= 1000; | |
1446 | } | |
1447 | ||
1448 | sprintf(dst, "%'8Lu%c", num, suff[i]); | |
1449 | return dst; | |
1450 | } | |
1451 | ||
649c7b66 JA |
1452 | static void dump_io_stats(struct per_dev_info *pdi, struct io_stats *ios, |
1453 | char *msg) | |
5c017e4b | 1454 | { |
4c523165 JA |
1455 | static char x[256], y[256]; |
1456 | ||
152f6476 JA |
1457 | fprintf(ofp, "%s\n", msg); |
1458 | ||
4c523165 JA |
1459 | fprintf(ofp, " Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads, 0), size_cnv(y, ios->qread_kb, 1)); |
1460 | fprintf(ofp, " Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites, 0), size_cnv(y, ios->qwrite_kb, 1)); | |
0a6b8fc4 | 1461 | |
4c523165 JA |
1462 | fprintf(ofp, " Read Dispatches: %s, %siB\t", size_cnv(x, ios->ireads, 0), size_cnv(y, ios->iread_kb, 1)); |
1463 | fprintf(ofp, " Write Dispatches: %s, %siB\n", size_cnv(x, ios->iwrites, 0), size_cnv(y, ios->iwrite_kb, 1)); | |
4054070a JA |
1464 | fprintf(ofp, " Reads Requeued: %s\t\t", size_cnv(x, ios->rrqueue, 0)); |
1465 | fprintf(ofp, " Writes Requeued: %s\n", size_cnv(x, ios->wrqueue, 0)); | |
4c523165 JA |
1466 | fprintf(ofp, " Reads Completed: %s, %siB\t", size_cnv(x, ios->creads, 0), size_cnv(y, ios->cread_kb, 1)); |
1467 | fprintf(ofp, " Writes Completed: %s, %siB\n", size_cnv(x, ios->cwrites, 0), size_cnv(y, ios->cwrite_kb, 1)); | |
152f6476 | 1468 | fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' '); |
152f6476 | 1469 | fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites); |
649c7b66 JA |
1470 | if (pdi) { |
1471 | fprintf(ofp, " Read depth: %'8u%8c\t", pdi->max_depth[0], ' '); | |
1472 | fprintf(ofp, " Write depth: %'8u\n", pdi->max_depth[1]); | |
1473 | } | |
06639b27 JA |
1474 | fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' '); |
1475 | fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs); | |
5c017e4b JA |
1476 | } |
1477 | ||
50adc0ba JA |
1478 | static void dump_wait_stats(struct per_process_info *ppi) |
1479 | { | |
b9d40d6f JA |
1480 | unsigned long rawait = ppi->longest_allocation_wait[0] / 1000; |
1481 | unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000; | |
1482 | unsigned long rcwait = ppi->longest_completion_wait[0] / 1000; | |
1483 | unsigned long wawait = ppi->longest_allocation_wait[1] / 1000; | |
1484 | unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000; | |
1485 | unsigned long wcwait = ppi->longest_completion_wait[1] / 1000; | |
1486 | ||
1487 | fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' '); | |
1488 | fprintf(ofp, " Allocation wait: %'8lu\n", wawait); | |
1489 | fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' '); | |
1490 | fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait); | |
1491 | fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' '); | |
1492 | fprintf(ofp, " Completion wait: %'8lu\n", wcwait); | |
50adc0ba JA |
1493 | } |
1494 | ||
886ecf0e JA |
1495 | static int ppi_name_compare(const void *p1, const void *p2) |
1496 | { | |
1497 | struct per_process_info *ppi1 = *((struct per_process_info **) p1); | |
1498 | struct per_process_info *ppi2 = *((struct per_process_info **) p2); | |
1499 | int res; | |
1500 | ||
2990e589 | 1501 | res = strverscmp(ppi1->ppm->comm, ppi2->ppm->comm); |
886ecf0e | 1502 | if (!res) |
2990e589 | 1503 | res = ppi1->ppm->pid > ppi2->ppm->pid; |
886ecf0e JA |
1504 | |
1505 | return res; | |
1506 | } | |
1507 | ||
1508 | static void sort_process_list(void) | |
1509 | { | |
1510 | struct per_process_info **ppis; | |
1511 | struct per_process_info *ppi; | |
1512 | int i = 0; | |
1513 | ||
1514 | ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *)); | |
1515 | ||
1516 | ppi = ppi_list; | |
1517 | while (ppi) { | |
06e6f286 | 1518 | ppis[i++] = ppi; |
886ecf0e JA |
1519 | ppi = ppi->list_next; |
1520 | } | |
1521 | ||
06e6f286 | 1522 | qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare); |
886ecf0e JA |
1523 | |
1524 | i = ppi_list_entries - 1; | |
1525 | ppi_list = NULL; | |
1526 | while (i >= 0) { | |
1527 | ppi = ppis[i]; | |
1528 | ||
1529 | ppi->list_next = ppi_list; | |
1530 | ppi_list = ppi; | |
1531 | i--; | |
1532 | } | |
50c38702 JA |
1533 | |
1534 | free(ppis); | |
886ecf0e JA |
1535 | } |
1536 | ||
152f6476 JA |
1537 | static void show_process_stats(void) |
1538 | { | |
1539 | struct per_process_info *ppi; | |
1540 | ||
886ecf0e JA |
1541 | sort_process_list(); |
1542 | ||
152f6476 JA |
1543 | ppi = ppi_list; |
1544 | while (ppi) { | |
2990e589 | 1545 | struct process_pid_map *ppm = ppi->ppm; |
ce8b6b4f JA |
1546 | char name[64]; |
1547 | ||
715d8021 | 1548 | if (ppi->more_than_one) |
2990e589 | 1549 | sprintf(name, "%s (%u, ...)", ppm->comm, ppm->pid); |
715d8021 | 1550 | else |
2990e589 | 1551 | sprintf(name, "%s (%u)", ppm->comm, ppm->pid); |
bf0720af | 1552 | |
649c7b66 | 1553 | dump_io_stats(NULL, &ppi->io_stats, name); |
50adc0ba | 1554 | dump_wait_stats(ppi); |
152f6476 JA |
1555 | ppi = ppi->list_next; |
1556 | } | |
1557 | ||
1558 | fprintf(ofp, "\n"); | |
1559 | } | |
1560 | ||
e7c9f3ff | 1561 | static void show_device_and_cpu_stats(void) |
d0ca268b | 1562 | { |
e7c9f3ff NS |
1563 | struct per_dev_info *pdi; |
1564 | struct per_cpu_info *pci; | |
1565 | struct io_stats total, *ios; | |
20ed6177 | 1566 | unsigned long long rrate, wrate, msec; |
e7c9f3ff NS |
1567 | int i, j, pci_events; |
1568 | char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3]; | |
1569 | char name[32]; | |
1570 | ||
1571 | for (pdi = devices, i = 0; i < ndevices; i++, pdi++) { | |
1572 | ||
1573 | memset(&total, 0, sizeof(total)); | |
1574 | pci_events = 0; | |
1575 | ||
1576 | if (i > 0) | |
1577 | fprintf(ofp, "\n"); | |
1578 | ||
1579 | for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) { | |
1580 | if (!pci->nelems) | |
1581 | continue; | |
1582 | ||
1583 | ios = &pci->io_stats; | |
1584 | total.qreads += ios->qreads; | |
1585 | total.qwrites += ios->qwrites; | |
1586 | total.creads += ios->creads; | |
1587 | total.cwrites += ios->cwrites; | |
1588 | total.mreads += ios->mreads; | |
1589 | total.mwrites += ios->mwrites; | |
1590 | total.ireads += ios->ireads; | |
1591 | total.iwrites += ios->iwrites; | |
4054070a JA |
1592 | total.rrqueue += ios->rrqueue; |
1593 | total.wrqueue += ios->wrqueue; | |
e7c9f3ff NS |
1594 | total.qread_kb += ios->qread_kb; |
1595 | total.qwrite_kb += ios->qwrite_kb; | |
1596 | total.cread_kb += ios->cread_kb; | |
1597 | total.cwrite_kb += ios->cwrite_kb; | |
1598 | total.iread_kb += ios->iread_kb; | |
1599 | total.iwrite_kb += ios->iwrite_kb; | |
06639b27 JA |
1600 | total.timer_unplugs += ios->timer_unplugs; |
1601 | total.io_unplugs += ios->io_unplugs; | |
e7c9f3ff NS |
1602 | |
1603 | snprintf(line, sizeof(line) - 1, "CPU%d (%s):", | |
1604 | j, get_dev_name(pdi, name, sizeof(name))); | |
649c7b66 | 1605 | dump_io_stats(pdi, ios, line); |
e7c9f3ff NS |
1606 | pci_events++; |
1607 | } | |
5c017e4b | 1608 | |
e7c9f3ff NS |
1609 | if (pci_events > 1) { |
1610 | fprintf(ofp, "\n"); | |
1611 | snprintf(line, sizeof(line) - 1, "Total (%s):", | |
1612 | get_dev_name(pdi, name, sizeof(name))); | |
649c7b66 | 1613 | dump_io_stats(NULL, &total, line); |
e7c9f3ff | 1614 | } |
d0ca268b | 1615 | |
20ed6177 | 1616 | wrate = rrate = 0; |
20ed6177 JA |
1617 | msec = (pdi->last_reported_time - pdi->first_reported_time) / 1000000; |
1618 | if (msec) { | |
1619 | rrate = 1000 * total.cread_kb / msec; | |
1620 | wrate = 1000 * total.cwrite_kb / msec; | |
1621 | } | |
1622 | ||
dce0f678 AB |
1623 | fprintf(ofp, "\nThroughput (R/W): %'LuKiB/s / %'LuKiB/s\n", |
1624 | rrate, wrate); | |
1625 | fprintf(ofp, "Events (%s): %'Lu entries\n", | |
1626 | get_dev_name(pdi, line, sizeof(line)), pdi->events); | |
492da111 AB |
1627 | |
1628 | collect_pdi_skips(pdi); | |
1629 | fprintf(ofp, "Skips: %'lu forward (%'llu - %5.1lf%%)\n", | |
dce0f678 AB |
1630 | pdi->skips,pdi->seq_skips, |
1631 | 100.0 * ((double)pdi->seq_skips / | |
dce0f678 | 1632 | (double)(pdi->events + pdi->seq_skips))); |
e7c9f3ff | 1633 | } |
d0ca268b JA |
1634 | } |
1635 | ||
4f0ae44f JA |
1636 | static void find_genesis(void) |
1637 | { | |
1638 | struct trace *t = trace_list; | |
1639 | ||
1640 | genesis_time = -1ULL; | |
1641 | while (t != NULL) { | |
1642 | if (t->bit->time < genesis_time) | |
1643 | genesis_time = t->bit->time; | |
1644 | ||
1645 | t = t->next; | |
1646 | } | |
1647 | } | |
1648 | ||
7f4d89e6 | 1649 | static inline int check_stopwatch(struct blk_io_trace *bit) |
4f0ae44f | 1650 | { |
7f4d89e6 JA |
1651 | if (bit->time < stopwatch_end && |
1652 | bit->time >= stopwatch_start) | |
4f0ae44f JA |
1653 | return 0; |
1654 | ||
1655 | return 1; | |
1656 | } | |
1657 | ||
53c68c88 JA |
1658 | /* |
1659 | * return youngest entry read | |
1660 | */ | |
1661 | static int sort_entries(unsigned long long *youngest) | |
4f0ae44f | 1662 | { |
210824c3 JA |
1663 | struct per_dev_info *pdi = NULL; |
1664 | struct per_cpu_info *pci = NULL; | |
4f0ae44f | 1665 | struct trace *t; |
4f0ae44f JA |
1666 | |
1667 | if (!genesis_time) | |
1668 | find_genesis(); | |
1669 | ||
d6222db8 | 1670 | *youngest = 0; |
4f0ae44f JA |
1671 | while ((t = trace_list) != NULL) { |
1672 | struct blk_io_trace *bit = t->bit; | |
1673 | ||
1674 | trace_list = t->next; | |
1675 | ||
7f4d89e6 | 1676 | bit->time -= genesis_time; |
4f0ae44f | 1677 | |
d6222db8 JA |
1678 | if (bit->time < *youngest || !*youngest) |
1679 | *youngest = bit->time; | |
1680 | ||
210824c3 JA |
1681 | if (!pdi || pdi->dev != bit->device) { |
1682 | pdi = get_dev_info(bit->device); | |
1683 | pci = NULL; | |
1684 | } | |
1685 | ||
1686 | if (!pci || pci->cpu != bit->cpu) | |
1687 | pci = get_cpu_info(pdi, bit->cpu); | |
1688 | ||
1689 | if (bit->sequence < pci->smallest_seq_read) | |
1690 | pci->smallest_seq_read = bit->sequence; | |
774a1a10 | 1691 | |
7f4d89e6 | 1692 | if (check_stopwatch(bit)) { |
4f0ae44f JA |
1693 | bit_free(bit); |
1694 | t_free(t); | |
1695 | continue; | |
1696 | } | |
1697 | ||
2a1b3424 | 1698 | if (trace_rb_insert_sort(t)) |
53c68c88 | 1699 | return -1; |
4f0ae44f JA |
1700 | } |
1701 | ||
53c68c88 | 1702 | return 0; |
4f0ae44f JA |
1703 | } |
1704 | ||
824c2b39 JA |
1705 | /* |
1706 | * to continue, we must have traces from all online cpus in the tree | |
1707 | */ | |
1708 | static int check_cpu_map(struct per_dev_info *pdi) | |
1709 | { | |
1710 | unsigned long *cpu_map; | |
1711 | struct rb_node *n; | |
1712 | struct trace *__t; | |
1713 | unsigned int i; | |
1714 | int ret, cpu; | |
1715 | ||
1716 | /* | |
1717 | * create a map of the cpus we have traces for | |
1718 | */ | |
1719 | cpu_map = malloc(pdi->cpu_map_max / sizeof(long)); | |
1720 | n = rb_first(&rb_sort_root); | |
1721 | while (n) { | |
1722 | __t = rb_entry(n, struct trace, rb_node); | |
1723 | cpu = __t->bit->cpu; | |
1724 | ||
1725 | cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu)); | |
1726 | n = rb_next(n); | |
1727 | } | |
1728 | ||
1729 | /* | |
b1c8e614 JA |
1730 | * we can't continue if pdi->cpu_map has entries set that we don't |
1731 | * have in the sort rbtree. the opposite is not a problem, though | |
824c2b39 JA |
1732 | */ |
1733 | ret = 0; | |
1734 | for (i = 0; i < pdi->cpu_map_max / CPUS_PER_LONG; i++) { | |
1735 | if (pdi->cpu_map[i] & ~(cpu_map[i])) { | |
1736 | ret = 1; | |
1737 | break; | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | free(cpu_map); | |
1742 | return ret; | |
1743 | } | |
1744 | ||
a141a7cd | 1745 | static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force) |
2a1b3424 | 1746 | { |
1ca323a5 | 1747 | struct blk_io_trace *bit = t->bit; |
210824c3 JA |
1748 | unsigned long expected_sequence; |
1749 | struct per_cpu_info *pci; | |
1ca323a5 | 1750 | struct trace *__t; |
492da111 | 1751 | |
210824c3 JA |
1752 | pci = get_cpu_info(pdi, bit->cpu); |
1753 | expected_sequence = pci->last_sequence + 1; | |
1754 | ||
774a1a10 | 1755 | if (!expected_sequence) { |
774a1a10 JA |
1756 | /* |
1757 | * 1 should be the first entry, just allow it | |
1758 | */ | |
1759 | if (bit->sequence == 1) | |
1760 | return 0; | |
210824c3 | 1761 | if (bit->sequence == pci->smallest_seq_read) |
79ee9704 | 1762 | return 0; |
774a1a10 | 1763 | |
824c2b39 | 1764 | return check_cpu_map(pdi); |
774a1a10 | 1765 | } |
2a1b3424 JA |
1766 | |
1767 | if (bit->sequence == expected_sequence) | |
1768 | return 0; | |
1769 | ||
2a1b3424 | 1770 | /* |
1c7c54aa JA |
1771 | * we may not have seen that sequence yet. if we are not doing |
1772 | * the final run, break and wait for more entries. | |
1c24add6 | 1773 | */ |
210824c3 JA |
1774 | if (expected_sequence < pci->smallest_seq_read) { |
1775 | __t = trace_rb_find_last(pdi, pci, expected_sequence); | |
1ca323a5 | 1776 | if (!__t) |
1c7c54aa | 1777 | goto skip; |
2a1b3424 | 1778 | |
1ca323a5 | 1779 | __put_trace_last(pdi, __t); |
2a1b3424 | 1780 | return 0; |
a141a7cd JA |
1781 | } else if (!force) { |
1782 | return 1; | |
0b07f23e | 1783 | } else { |
1c7c54aa | 1784 | skip: |
66930177 | 1785 | if (check_current_skips(pci, bit->sequence)) |
492da111 AB |
1786 | return 0; |
1787 | ||
965eca2d | 1788 | if (expected_sequence < bit->sequence) |
66930177 | 1789 | insert_skip(pci, expected_sequence, bit->sequence - 1); |
1c7c54aa JA |
1790 | return 0; |
1791 | } | |
2a1b3424 JA |
1792 | } |
1793 | ||
a649216c | 1794 | static void show_entries_rb(int force) |
8fc0abbc | 1795 | { |
1f7afa72 JA |
1796 | struct per_dev_info *pdi = NULL; |
1797 | struct per_cpu_info *pci = NULL; | |
8fc0abbc | 1798 | struct blk_io_trace *bit; |
3aabcd89 | 1799 | struct rb_node *n; |
8fc0abbc | 1800 | struct trace *t; |
1f7afa72 | 1801 | |
7d747d22 | 1802 | while ((n = rb_first(&rb_sort_root)) != NULL) { |
dd90748f | 1803 | if (is_done() && !force && !pipeline) |
1f7afa72 | 1804 | break; |
8fc0abbc JA |
1805 | |
1806 | t = rb_entry(n, struct trace, rb_node); | |
1807 | bit = t->bit; | |
1808 | ||
a43c1c17 JA |
1809 | if (read_sequence - t->read_sequence < 1 && !force) |
1810 | break; | |
1811 | ||
210824c3 | 1812 | if (!pdi || pdi->dev != bit->device) { |
287fa3d6 | 1813 | pdi = get_dev_info(bit->device); |
210824c3 JA |
1814 | pci = NULL; |
1815 | } | |
1f7afa72 | 1816 | |
e7c9f3ff NS |
1817 | if (!pdi) { |
1818 | fprintf(stderr, "Unknown device ID? (%d,%d)\n", | |
1819 | MAJOR(bit->device), MINOR(bit->device)); | |
1820 | break; | |
1821 | } | |
1f7afa72 | 1822 | |
a141a7cd JA |
1823 | if (check_sequence(pdi, t, force)) |
1824 | break; | |
cb2a1a62 | 1825 | |
a141a7cd JA |
1826 | if (!force && bit->time > last_allowed_time) |
1827 | break; | |
8fc0abbc | 1828 | |
4f0ae44f | 1829 | check_time(pdi, bit); |
8fc0abbc | 1830 | |
4f0ae44f JA |
1831 | if (!pci || pci->cpu != bit->cpu) |
1832 | pci = get_cpu_info(pdi, bit->cpu); | |
287fa3d6 | 1833 | |
210824c3 JA |
1834 | pci->last_sequence = bit->sequence; |
1835 | ||
cbc927b6 JA |
1836 | pci->nelems++; |
1837 | ||
66930177 | 1838 | if (bit->action & (act_mask << BLK_TC_SHIFT)) |
98f8386b | 1839 | dump_trace(bit, pci, pdi); |
87b72777 | 1840 | |
2a1b3424 | 1841 | put_trace(pdi, t); |
cb2a1a62 | 1842 | } |
8fc0abbc JA |
1843 | } |
1844 | ||
c0e0dbc2 | 1845 | static int read_data(int fd, void *buffer, int bytes, int block, int *fdblock) |
1f79c4a0 JA |
1846 | { |
1847 | int ret, bytes_left, fl; | |
1848 | void *p; | |
1849 | ||
c0e0dbc2 JA |
1850 | if (block != *fdblock) { |
1851 | fl = fcntl(fd, F_GETFL); | |
1f79c4a0 | 1852 | |
c0e0dbc2 JA |
1853 | if (!block) { |
1854 | *fdblock = 0; | |
1855 | fcntl(fd, F_SETFL, fl | O_NONBLOCK); | |
1856 | } else { | |
1857 | *fdblock = 1; | |
1858 | fcntl(fd, F_SETFL, fl & ~O_NONBLOCK); | |
1859 | } | |
1860 | } | |
1f79c4a0 JA |
1861 | |
1862 | bytes_left = bytes; | |
1863 | p = buffer; | |
1864 | while (bytes_left > 0) { | |
1865 | ret = read(fd, p, bytes_left); | |
1866 | if (!ret) | |
1867 | return 1; | |
1868 | else if (ret < 0) { | |
db7e0552 | 1869 | if (errno != EAGAIN) { |
1f79c4a0 | 1870 | perror("read"); |
db7e0552 JA |
1871 | return -1; |
1872 | } | |
a649216c | 1873 | |
5c0f40f7 JA |
1874 | /* |
1875 | * never do partial reads. we can return if we | |
1876 | * didn't read anything and we should not block, | |
1877 | * otherwise wait for data | |
1878 | */ | |
1879 | if ((bytes_left == bytes) && !block) | |
1880 | return 1; | |
1881 | ||
1882 | usleep(10); | |
1883 | continue; | |
1f79c4a0 JA |
1884 | } else { |
1885 | p += ret; | |
1886 | bytes_left -= ret; | |
1887 | } | |
1888 | } | |
1889 | ||
1890 | return 0; | |
1891 | } | |
1892 | ||
017d1660 JA |
1893 | static inline __u16 get_pdulen(struct blk_io_trace *bit) |
1894 | { | |
1895 | if (data_is_native) | |
1896 | return bit->pdu_len; | |
1897 | ||
1898 | return __bswap_16(bit->pdu_len); | |
1899 | } | |
1900 | ||
1901 | static inline __u32 get_magic(struct blk_io_trace *bit) | |
1902 | { | |
1903 | if (data_is_native) | |
1904 | return bit->magic; | |
1905 | ||
1906 | return __bswap_32(bit->magic); | |
1907 | } | |
1908 | ||
c0e0dbc2 | 1909 | static int read_events(int fd, int always_block, int *fdblock) |
cb2a1a62 | 1910 | { |
287fa3d6 | 1911 | struct per_dev_info *pdi = NULL; |
e820abd7 | 1912 | unsigned int events = 0; |
7d747d22 JA |
1913 | |
1914 | while (!is_done() && events < rb_batch) { | |
1915 | struct blk_io_trace *bit; | |
1916 | struct trace *t; | |
db7e0552 | 1917 | int pdu_len, should_block, ret; |
7d747d22 JA |
1918 | __u32 magic; |
1919 | ||
d36421e4 | 1920 | bit = bit_alloc(); |
cb2a1a62 | 1921 | |
c0e0dbc2 JA |
1922 | should_block = !events || always_block; |
1923 | ||
db7e0552 JA |
1924 | ret = read_data(fd, bit, sizeof(*bit), should_block, fdblock); |
1925 | if (ret) { | |
eb9bd4e9 | 1926 | bit_free(bit); |
db7e0552 JA |
1927 | if (!events && ret < 0) |
1928 | events = ret; | |
cb2a1a62 | 1929 | break; |
eb9bd4e9 | 1930 | } |
cb2a1a62 | 1931 | |
017d1660 JA |
1932 | /* |
1933 | * look at first trace to check whether we need to convert | |
1934 | * data in the future | |
1935 | */ | |
9e4cd1b8 | 1936 | if (data_is_native == -1 && check_data_endianness(bit->magic)) |
017d1660 JA |
1937 | break; |
1938 | ||
1939 | magic = get_magic(bit); | |
7d747d22 JA |
1940 | if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { |
1941 | fprintf(stderr, "Bad magic %x\n", magic); | |
1942 | break; | |
1943 | } | |
1944 | ||
017d1660 | 1945 | pdu_len = get_pdulen(bit); |
7d747d22 JA |
1946 | if (pdu_len) { |
1947 | void *ptr = realloc(bit, sizeof(*bit) + pdu_len); | |
1948 | ||
c0e0dbc2 | 1949 | if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1, fdblock)) { |
eb9bd4e9 | 1950 | bit_free(ptr); |
7d747d22 | 1951 | break; |
eb9bd4e9 | 1952 | } |
7d747d22 JA |
1953 | |
1954 | bit = ptr; | |
1955 | } | |
1956 | ||
d6222db8 JA |
1957 | trace_to_cpu(bit); |
1958 | ||
1959 | if (verify_trace(bit)) { | |
1960 | bit_free(bit); | |
1961 | continue; | |
1962 | } | |
1963 | ||
bfc70ad5 JA |
1964 | /* |
1965 | * not a real trace, so grab and handle it here | |
1966 | */ | |
1967 | if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY)) { | |
1968 | add_ppm_hash(bit->pid, (char *) bit + sizeof(*bit)); | |
a2594911 | 1969 | output_binary(bit, sizeof(*bit) + bit->pdu_len); |
bfc70ad5 JA |
1970 | continue; |
1971 | } | |
1972 | ||
d36421e4 | 1973 | t = t_alloc(); |
cb2a1a62 JA |
1974 | memset(t, 0, sizeof(*t)); |
1975 | t->bit = bit; | |
a43c1c17 | 1976 | t->read_sequence = read_sequence; |
cb2a1a62 | 1977 | |
7d747d22 JA |
1978 | t->next = trace_list; |
1979 | trace_list = t; | |
1f7afa72 | 1980 | |
f7bd1a9b | 1981 | if (!pdi || pdi->dev != bit->device) |
287fa3d6 JA |
1982 | pdi = get_dev_info(bit->device); |
1983 | ||
1984 | if (bit->time > pdi->last_read_time) | |
1985 | pdi->last_read_time = bit->time; | |
1986 | ||
7d747d22 | 1987 | events++; |
cb2a1a62 JA |
1988 | } |
1989 | ||
7d747d22 | 1990 | return events; |
cb2a1a62 JA |
1991 | } |
1992 | ||
d5396421 | 1993 | static int do_file(void) |
d0ca268b | 1994 | { |
7d747d22 | 1995 | struct per_cpu_info *pci; |
73877e12 NS |
1996 | struct per_dev_info *pdi; |
1997 | int i, j, events, events_added; | |
d0ca268b | 1998 | |
7d747d22 JA |
1999 | /* |
2000 | * first prepare all files for reading | |
2001 | */ | |
e8741a4a | 2002 | for (i = 0; i < ndevices; i++) { |
73877e12 NS |
2003 | pdi = &devices[i]; |
2004 | pdi->nfiles = 0; | |
73877e12 | 2005 | |
74feaeb5 | 2006 | for (j = 0;; j++) { |
e7c9f3ff | 2007 | struct stat st; |
d1d7f15f | 2008 | int len = 0; |
6e0073ed | 2009 | char *p, *dname; |
87b72777 | 2010 | |
e7c9f3ff NS |
2011 | pci = get_cpu_info(pdi, j); |
2012 | pci->cpu = j; | |
7d747d22 | 2013 | pci->fd = -1; |
c0e0dbc2 | 2014 | pci->fdblock = -1; |
6e0073ed JA |
2015 | |
2016 | p = strdup(pdi->name); | |
2017 | dname = dirname(p); | |
2018 | if (strcmp(dname, ".")) { | |
2019 | input_dir = dname; | |
2020 | p = strdup(pdi->name); | |
2021 | strcpy(pdi->name, basename(p)); | |
2022 | } | |
2023 | free(p); | |
d0ca268b | 2024 | |
d1d7f15f JA |
2025 | if (input_dir) |
2026 | len = sprintf(pci->fname, "%s/", input_dir); | |
2027 | ||
2028 | snprintf(pci->fname + len, sizeof(pci->fname)-1-len, | |
7d747d22 | 2029 | "%s.blktrace.%d", pdi->name, pci->cpu); |
e7c9f3ff NS |
2030 | if (stat(pci->fname, &st) < 0) |
2031 | break; | |
74feaeb5 AB |
2032 | if (st.st_size) { |
2033 | pci->fd = open(pci->fname, O_RDONLY); | |
2034 | if (pci->fd < 0) { | |
2035 | perror(pci->fname); | |
2036 | continue; | |
2037 | } | |
e7c9f3ff NS |
2038 | } |
2039 | ||
7d747d22 | 2040 | printf("Input file %s added\n", pci->fname); |
73877e12 | 2041 | pdi->nfiles++; |
824c2b39 | 2042 | cpu_mark_online(pdi, pci->cpu); |
d0ca268b | 2043 | } |
d5396421 JA |
2044 | } |
2045 | ||
7d747d22 JA |
2046 | /* |
2047 | * now loop over the files reading in the data | |
2048 | */ | |
412819ce | 2049 | do { |
53c68c88 JA |
2050 | unsigned long long youngest; |
2051 | ||
7d747d22 | 2052 | events_added = 0; |
287fa3d6 | 2053 | last_allowed_time = -1ULL; |
a43c1c17 | 2054 | read_sequence++; |
d5396421 | 2055 | |
7d747d22 | 2056 | for (i = 0; i < ndevices; i++) { |
73877e12 | 2057 | pdi = &devices[i]; |
66930177 | 2058 | pdi->last_read_time = -1ULL; |
73877e12 NS |
2059 | |
2060 | for (j = 0; j < pdi->nfiles; j++) { | |
d5396421 | 2061 | |
73877e12 | 2062 | pci = get_cpu_info(pdi, j); |
d5396421 | 2063 | |
7d747d22 JA |
2064 | if (pci->fd == -1) |
2065 | continue; | |
51128a28 | 2066 | |
210824c3 JA |
2067 | pci->smallest_seq_read = -1; |
2068 | ||
c0e0dbc2 | 2069 | events = read_events(pci->fd, 1, &pci->fdblock); |
db7e0552 | 2070 | if (events <= 0) { |
824c2b39 | 2071 | cpu_mark_offline(pdi, pci->cpu); |
7d747d22 JA |
2072 | close(pci->fd); |
2073 | pci->fd = -1; | |
2074 | continue; | |
2075 | } | |
d5396421 | 2076 | |
287fa3d6 JA |
2077 | if (pdi->last_read_time < last_allowed_time) |
2078 | last_allowed_time = pdi->last_read_time; | |
d5396421 | 2079 | |
7d747d22 JA |
2080 | events_added += events; |
2081 | } | |
2ff323b0 | 2082 | } |
d5396421 | 2083 | |
53c68c88 JA |
2084 | if (sort_entries(&youngest)) |
2085 | break; | |
2086 | ||
2087 | if (youngest > stopwatch_end) | |
287fa3d6 JA |
2088 | break; |
2089 | ||
a649216c | 2090 | show_entries_rb(0); |
cb2a1a62 | 2091 | |
7d747d22 | 2092 | } while (events_added); |
d5396421 | 2093 | |
a649216c JA |
2094 | if (rb_sort_entries) |
2095 | show_entries_rb(1); | |
2096 | ||
7d747d22 | 2097 | return 0; |
412819ce | 2098 | } |
d5396421 | 2099 | |
412819ce JA |
2100 | static int do_stdin(void) |
2101 | { | |
53c68c88 | 2102 | unsigned long long youngest; |
c0e0dbc2 | 2103 | int fd, events, fdblock; |
d5396421 | 2104 | |
be925321 | 2105 | last_allowed_time = -1ULL; |
1f79c4a0 | 2106 | fd = dup(STDIN_FILENO); |
0b07f23e JA |
2107 | if (fd == -1) { |
2108 | perror("dup stdin"); | |
2109 | return -1; | |
2110 | } | |
d5396421 | 2111 | |
c0e0dbc2 | 2112 | fdblock = -1; |
db7e0552 | 2113 | while ((events = read_events(fd, 0, &fdblock)) > 0) { |
4ab42801 | 2114 | read_sequence++; |
412819ce | 2115 | |
210824c3 | 2116 | #if 0 |
0b07f23e | 2117 | smallest_seq_read = -1U; |
210824c3 | 2118 | #endif |
0b07f23e | 2119 | |
53c68c88 JA |
2120 | if (sort_entries(&youngest)) |
2121 | break; | |
2122 | ||
2123 | if (youngest > stopwatch_end) | |
2ff323b0 JA |
2124 | break; |
2125 | ||
763d936e | 2126 | show_entries_rb(0); |
0b07f23e | 2127 | } |
d5396421 | 2128 | |
a649216c JA |
2129 | if (rb_sort_entries) |
2130 | show_entries_rb(1); | |
2131 | ||
d5396421 | 2132 | close(fd); |
d5396421 JA |
2133 | return 0; |
2134 | } | |
d0ca268b | 2135 | |
cbc927b6 | 2136 | static void show_stats(void) |
412819ce | 2137 | { |
cbc927b6 JA |
2138 | if (!ofp) |
2139 | return; | |
2140 | if (stats_printed) | |
2141 | return; | |
2142 | ||
2143 | stats_printed = 1; | |
2144 | ||
2145 | if (per_process_stats) | |
2146 | show_process_stats(); | |
2147 | ||
2148 | if (per_device_and_cpu_stats) | |
2149 | show_device_and_cpu_stats(); | |
2150 | ||
152f6476 | 2151 | fflush(ofp); |
412819ce JA |
2152 | } |
2153 | ||
e820abd7 | 2154 | static void handle_sigint(__attribute__((__unused__)) int sig) |
412819ce JA |
2155 | { |
2156 | done = 1; | |
412819ce JA |
2157 | } |
2158 | ||
46e6968b NS |
2159 | /* |
2160 | * Extract start and duration times from a string, allowing | |
2161 | * us to specify a time interval of interest within a trace. | |
2162 | * Format: "duration" (start is zero) or "start:duration". | |
2163 | */ | |
2164 | static int find_stopwatch_interval(char *string) | |
2165 | { | |
2166 | double value; | |
2167 | char *sp; | |
2168 | ||
2169 | value = strtod(string, &sp); | |
2170 | if (sp == string) { | |
2171 | fprintf(stderr,"Invalid stopwatch timer: %s\n", string); | |
2172 | return 1; | |
2173 | } | |
2174 | if (*sp == ':') { | |
2175 | stopwatch_start = DOUBLE_TO_NANO_ULL(value); | |
2176 | string = sp + 1; | |
2177 | value = strtod(string, &sp); | |
2178 | if (sp == string || *sp != '\0') { | |
2179 | fprintf(stderr,"Invalid stopwatch duration time: %s\n", | |
2180 | string); | |
2181 | return 1; | |
2182 | } | |
2183 | } else if (*sp != '\0') { | |
2184 | fprintf(stderr,"Invalid stopwatch start timer: %s\n", string); | |
2185 | return 1; | |
2186 | } | |
1b928247 JA |
2187 | stopwatch_end = DOUBLE_TO_NANO_ULL(value); |
2188 | if (stopwatch_end <= stopwatch_start) { | |
2189 | fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n", | |
2190 | stopwatch_start, stopwatch_end); | |
2191 | return 1; | |
2192 | } | |
2193 | ||
46e6968b NS |
2194 | return 0; |
2195 | } | |
2196 | ||
52724a0e JA |
2197 | static char usage_str[] = \ |
2198 | "[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \ | |
2199 | "[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \ | |
2200 | "\t-i Input file containing trace data, or '-' for stdin\n" \ | |
d1d7f15f | 2201 | "\t-D Directory to prepend to input file names\n" \ |
52724a0e | 2202 | "\t-o Output file. If not given, output is stdout\n" \ |
a2594911 | 2203 | "\t-d Output file. If specified, binary data is written to file\n" \ |
52724a0e JA |
2204 | "\t-b stdin read batching\n" \ |
2205 | "\t-s Show per-program io statistics\n" \ | |
ac740d98 | 2206 | "\t-h Hash processes by name, not pid\n" \ |
52724a0e JA |
2207 | "\t-t Track individual ios. Will tell you the time a request took\n" \ |
2208 | "\t to get queued, to get dispatched, and to get completed\n" \ | |
2209 | "\t-q Quiet. Don't display any stats at the end of the trace\n" \ | |
2210 | "\t-w Only parse data between the given time interval in seconds.\n" \ | |
2211 | "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \ | |
d915dee6 JA |
2212 | "\t-f Output format. Customize the output format. The format field\n" \ |
2213 | "\t identifies can be found in the documentation\n" \ | |
52724a0e | 2214 | "\t-F Format specification. Can be found in the documentation\n" \ |
57ea8602 JA |
2215 | "\t-v More verbose for marginal errors\n" \ |
2216 | "\t-V Print program version info\n\n"; | |
52724a0e | 2217 | |
1f79c4a0 JA |
2218 | static void usage(char *prog) |
2219 | { | |
52724a0e | 2220 | fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str); |
1f79c4a0 JA |
2221 | } |
2222 | ||
d5396421 JA |
2223 | int main(int argc, char *argv[]) |
2224 | { | |
152f6476 | 2225 | char *ofp_buffer; |
98f8386b | 2226 | int i, c, ret, mode; |
98f8386b | 2227 | int act_mask_tmp = 0; |
d5396421 JA |
2228 | |
2229 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) { | |
2230 | switch (c) { | |
98f8386b AB |
2231 | case 'a': |
2232 | i = find_mask_map(optarg); | |
2233 | if (i < 0) { | |
2234 | fprintf(stderr,"Invalid action mask %s\n", | |
2235 | optarg); | |
2236 | return 1; | |
2237 | } | |
2238 | act_mask_tmp |= i; | |
2239 | break; | |
2240 | ||
2241 | case 'A': | |
2242 | if ((sscanf(optarg, "%x", &i) != 1) || | |
2243 | !valid_act_opt(i)) { | |
2244 | fprintf(stderr, | |
2245 | "Invalid set action mask %s/0x%x\n", | |
2246 | optarg, i); | |
2247 | return 1; | |
2248 | } | |
2249 | act_mask_tmp = i; | |
2250 | break; | |
d5396421 | 2251 | case 'i': |
e7c9f3ff NS |
2252 | if (!strcmp(optarg, "-") && !pipeline) |
2253 | pipeline = 1; | |
2254 | else if (resize_devices(optarg) != 0) | |
2255 | return 1; | |
d5396421 | 2256 | break; |
d1d7f15f JA |
2257 | case 'D': |
2258 | input_dir = optarg; | |
2259 | break; | |
d5396421 | 2260 | case 'o': |
66efebf8 | 2261 | output_name = optarg; |
d5396421 | 2262 | break; |
79f19470 JA |
2263 | case 'b': |
2264 | rb_batch = atoi(optarg); | |
2265 | if (rb_batch <= 0) | |
2266 | rb_batch = RB_BATCH_DEFAULT; | |
2267 | break; | |
152f6476 JA |
2268 | case 's': |
2269 | per_process_stats = 1; | |
2270 | break; | |
7997c5b0 JA |
2271 | case 't': |
2272 | track_ios = 1; | |
2273 | break; | |
1e1c60f1 NS |
2274 | case 'q': |
2275 | per_device_and_cpu_stats = 0; | |
2276 | break; | |
46e6968b NS |
2277 | case 'w': |
2278 | if (find_stopwatch_interval(optarg) != 0) | |
2279 | return 1; | |
2280 | break; | |
ab197ca7 AB |
2281 | case 'f': |
2282 | set_all_format_specs(optarg); | |
2283 | break; | |
2284 | case 'F': | |
2285 | if (add_format_spec(optarg) != 0) | |
2286 | return 1; | |
2287 | break; | |
d915dee6 | 2288 | case 'h': |
715d8021 | 2289 | ppi_hash_by_pid = 0; |
bf0720af | 2290 | break; |
52724a0e | 2291 | case 'v': |
57ea8602 JA |
2292 | verbose++; |
2293 | break; | |
2294 | case 'V': | |
52724a0e JA |
2295 | printf("%s version %s\n", argv[0], blkparse_version); |
2296 | return 0; | |
a2594911 AB |
2297 | case 'd': |
2298 | dump_binary = optarg; | |
2299 | break; | |
d5396421 | 2300 | default: |
1f79c4a0 | 2301 | usage(argv[0]); |
d5396421 JA |
2302 | return 1; |
2303 | } | |
d0ca268b JA |
2304 | } |
2305 | ||
e7c9f3ff NS |
2306 | while (optind < argc) { |
2307 | if (!strcmp(argv[optind], "-") && !pipeline) | |
2308 | pipeline = 1; | |
2309 | else if (resize_devices(argv[optind]) != 0) | |
2310 | return 1; | |
2311 | optind++; | |
2312 | } | |
2313 | ||
2314 | if (!pipeline && !ndevices) { | |
1f79c4a0 | 2315 | usage(argv[0]); |
d5396421 JA |
2316 | return 1; |
2317 | } | |
2318 | ||
98f8386b AB |
2319 | if (act_mask_tmp != 0) |
2320 | act_mask = act_mask_tmp; | |
2321 | ||
7997c5b0 | 2322 | memset(&rb_sort_root, 0, sizeof(rb_sort_root)); |
412819ce JA |
2323 | |
2324 | signal(SIGINT, handle_sigint); | |
2325 | signal(SIGHUP, handle_sigint); | |
2326 | signal(SIGTERM, handle_sigint); | |
d5396421 | 2327 | |
d69db225 JA |
2328 | setlocale(LC_NUMERIC, "en_US"); |
2329 | ||
a66877e6 | 2330 | if (!output_name) { |
152f6476 | 2331 | ofp = fdopen(STDOUT_FILENO, "w"); |
a66877e6 JA |
2332 | mode = _IOLBF; |
2333 | } else { | |
152f6476 JA |
2334 | char ofname[128]; |
2335 | ||
7a9690c0 | 2336 | snprintf(ofname, sizeof(ofname) - 1, "%s", output_name); |
152f6476 | 2337 | ofp = fopen(ofname, "w"); |
a66877e6 | 2338 | mode = _IOFBF; |
152f6476 JA |
2339 | } |
2340 | ||
2341 | if (!ofp) { | |
2342 | perror("fopen"); | |
2343 | return 1; | |
2344 | } | |
2345 | ||
2346 | ofp_buffer = malloc(4096); | |
a66877e6 | 2347 | if (setvbuf(ofp, ofp_buffer, mode, 4096)) { |
152f6476 JA |
2348 | perror("setvbuf"); |
2349 | return 1; | |
2350 | } | |
2351 | ||
a2594911 AB |
2352 | if (dump_binary) { |
2353 | dump_fd = creat(dump_binary, 0666); | |
2354 | if (dump_fd < 0) { | |
2355 | perror(dump_binary); | |
2356 | dump_binary = NULL; | |
2357 | return 1; | |
2358 | } | |
2359 | } | |
2360 | ||
e7c9f3ff | 2361 | if (pipeline) |
d5396421 JA |
2362 | ret = do_stdin(); |
2363 | else | |
2364 | ret = do_file(); | |
2365 | ||
cbc927b6 | 2366 | show_stats(); |
eb9bd4e9 | 2367 | free(ofp_buffer); |
d5396421 | 2368 | return ret; |
d0ca268b | 2369 | } |