2 * block queue tracing parse application
4 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 #include <sys/types.h>
37 #define SECONDS(x) ((unsigned long long)(x) / 1000000000)
38 #define NANO_SECONDS(x) ((unsigned long long)(x) % 1000000000)
39 #define DOUBLE_TO_NANO_ULL(d) ((unsigned long long)((d) * 1000000000))
42 #define MINORMASK ((1U << MINORBITS) - 1)
43 #define MAJOR(dev) ((unsigned int) ((dev) >> MINORBITS))
44 #define MINOR(dev) ((unsigned int) ((dev) & MINORMASK))
46 #define min(a, b) ((a) < (b) ? (a) : (b))
49 unsigned long qreads, qwrites, creads, cwrites, mreads, mwrites;
50 unsigned long ireads, iwrites;
51 unsigned long long qread_kb, qwrite_kb, cread_kb, cwrite_kb;
52 unsigned long long iread_kb, iwrite_kb;
53 unsigned long io_unplugs, timer_unplugs;
63 struct io_stats io_stats;
71 unsigned long long events;
72 unsigned long long last_reported_time;
73 struct io_stats io_stats;
76 struct per_cpu_info *cpus;
79 struct per_process_info {
82 struct io_stats io_stats;
83 struct per_process_info *hash_next, *list_next;
88 unsigned long long longest_allocation_wait[2];
89 unsigned long long longest_dispatch_wait[2];
90 unsigned long long longest_completion_wait[2];
93 #define PPI_HASH_SHIFT (8)
94 static struct per_process_info *ppi_hash[1 << PPI_HASH_SHIFT];
95 static struct per_process_info *ppi_list;
97 #define S_OPTS "i:o:b:stqw:"
98 static struct option l_opts[] = {
118 .name = "per program stats",
149 static struct rb_root rb_sort_root;
150 static struct rb_root rb_track_root;
153 * for sorting the displayed output
156 struct blk_io_trace *bit;
157 struct rb_node rb_node;
161 * for tracking individual ios
164 struct rb_node rb_node;
169 unsigned long long allocation_time;
170 unsigned long long queue_time;
171 unsigned long long dispatch_time;
172 unsigned long long completion_time;
176 static struct per_dev_info *devices;
177 static char *get_dev_name(struct per_dev_info *, char *, int);
180 static char *output_name;
182 static unsigned long long genesis_time;
183 static unsigned long long stopwatch_start; /* start from zero by default */
184 static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
186 static int per_process_stats;
187 static int track_ios;
189 #define RB_BATCH_DEFAULT (1024)
190 static int rb_batch = RB_BATCH_DEFAULT;
194 #define is_done() (*(volatile int *)(&done))
195 static volatile int done;
197 static inline unsigned long hash_long(unsigned long val)
201 #elif __WORDSIZE == 64
202 val *= 0x9e37fffffffc0001UL;
204 #error unknown word size
207 return val >> (__WORDSIZE - PPI_HASH_SHIFT);
210 static inline void add_process_to_hash(struct per_process_info *ppi)
212 const int hash_idx = hash_long(ppi->pid);
214 ppi->hash_next = ppi_hash[hash_idx];
215 ppi_hash[hash_idx] = ppi;
218 static inline void add_process_to_list(struct per_process_info *ppi)
220 ppi->list_next = ppi_list;
224 static struct per_process_info *find_process_by_pid(__u32 pid)
226 const int hash_idx = hash_long(pid);
227 struct per_process_info *ppi;
229 ppi = ppi_hash[hash_idx];
234 ppi = ppi->hash_next;
240 static inline int trace_rb_insert(struct trace *t)
242 struct rb_node **p = &rb_sort_root.rb_node;
243 struct rb_node *parent = NULL;
246 if (genesis_time == 0 || t->bit->time < genesis_time)
247 genesis_time = t->bit->time;
251 __t = rb_entry(parent, struct trace, rb_node);
253 if (t->bit->time < __t->bit->time)
255 else if (t->bit->time > __t->bit->time)
257 else if (t->bit->device < __t->bit->device)
259 else if (t->bit->device > __t->bit->device)
261 else if (t->bit->sequence < __t->bit->sequence)
263 else if (t->bit->sequence > __t->bit->sequence)
265 else if (t->bit->device == __t->bit->device) {
267 "sequence alias (%d) on device %d,%d!\n",
269 MAJOR(t->bit->device), MINOR(t->bit->device));
274 rb_link_node(&t->rb_node, parent, p);
275 rb_insert_color(&t->rb_node, &rb_sort_root);
279 static inline int track_rb_insert(struct io_track *iot)
281 struct rb_node **p = &rb_track_root.rb_node;
282 struct rb_node *parent = NULL;
283 struct io_track *__iot;
288 __iot = rb_entry(parent, struct io_track, rb_node);
290 if (iot->device < __iot->device)
292 else if (iot->device > __iot->device)
294 else if (iot->sector < __iot->sector)
296 else if (iot->sector > __iot->sector)
300 "sector alias (%llu) on device %d,%d!\n",
302 MAJOR(iot->device), MINOR(iot->device));
307 rb_link_node(&iot->rb_node, parent, p);
308 rb_insert_color(&iot->rb_node, &rb_track_root);
312 static struct io_track *__find_track(dev_t device, __u64 sector)
314 struct rb_node **p = &rb_track_root.rb_node;
315 struct rb_node *parent = NULL;
316 struct io_track *__iot;
321 __iot = rb_entry(parent, struct io_track, rb_node);
323 if (device < __iot->device)
325 else if (device > __iot->device)
327 else if (sector < __iot->sector)
329 else if (sector > __iot->sector)
338 static struct io_track *find_track(__u32 pid, dev_t device, __u64 sector)
340 struct io_track *iot;
342 iot = __find_track(device, sector);
344 iot = malloc(sizeof(*iot));
346 iot->device = device;
347 iot->sector = sector;
348 track_rb_insert(iot);
354 static void log_track_merge(struct blk_io_trace *t)
356 struct io_track *iot;
360 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
363 iot = __find_track(t->device, t->sector - (t->bytes >> 10));
365 fprintf(stderr, "Trying to merge on non-existing request\n");
369 rb_erase(&iot->rb_node, &rb_track_root);
370 iot->sector -= t->bytes >> 10;
371 track_rb_insert(iot);
374 static void log_track_getrq(struct blk_io_trace *t)
376 struct io_track *iot;
381 iot = find_track(t->pid, t->device, t->sector);
382 iot->allocation_time = t->time;
387 * return time between rq allocation and queue
389 static unsigned long long log_track_queue(struct blk_io_trace *t)
391 unsigned long long elapsed;
392 struct io_track *iot;
397 iot = find_track(t->pid, t->device, t->sector);
398 iot->queue_time = t->time;
399 elapsed = iot->queue_time - iot->allocation_time;
401 if (per_process_stats) {
402 struct per_process_info *ppi = find_process_by_pid(iot->pid);
403 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
405 if (ppi && elapsed > ppi->longest_allocation_wait[w])
406 ppi->longest_allocation_wait[w] = elapsed;
413 * return time between queue and issue
415 static unsigned long long log_track_issue(struct blk_io_trace *t)
417 unsigned long long elapsed;
418 struct io_track *iot;
422 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
425 iot = __find_track(t->device, t->sector);
427 fprintf(stderr, "Trying to issue on non-existing request\n");
431 iot->dispatch_time = t->time;
432 elapsed = iot->dispatch_time - iot->queue_time;
434 if (per_process_stats) {
435 struct per_process_info *ppi = find_process_by_pid(iot->pid);
436 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
438 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
439 ppi->longest_dispatch_wait[w] = elapsed;
446 * return time between dispatch and complete
448 static unsigned long long log_track_complete(struct blk_io_trace *t)
450 unsigned long long elapsed;
451 struct io_track *iot;
455 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
458 iot = __find_track(t->device, t->sector);
460 fprintf(stderr, "Trying to dispatch on non-existing request\n");
464 iot->completion_time = t->time;
465 elapsed = iot->completion_time - iot->dispatch_time;
467 if (per_process_stats) {
468 struct per_process_info *ppi = find_process_by_pid(iot->pid);
469 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
471 if (ppi && elapsed > ppi->longest_completion_wait[w])
472 ppi->longest_completion_wait[w] = elapsed;
476 * kill the trace, we don't need it after completion
478 rb_erase(&iot->rb_node, &rb_track_root);
485 static struct io_stats *find_process_io_stats(__u32 pid, char *name)
487 struct per_process_info *ppi = find_process_by_pid(pid);
490 ppi = malloc(sizeof(*ppi));
491 memset(ppi, 0, sizeof(*ppi));
492 strncpy(ppi->name, name, sizeof(ppi->name));
494 add_process_to_hash(ppi);
495 add_process_to_list(ppi);
498 return &ppi->io_stats;
502 static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
504 struct per_cpu_info *cpus = pdi->cpus;
505 int ncpus = pdi->ncpus;
506 int new_count = cpu + 1;
510 size = new_count * sizeof(struct per_cpu_info);
511 cpus = realloc(cpus, size);
514 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
515 get_dev_name(pdi, name, sizeof(name)), size);
519 new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
520 new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
521 memset(new_start, 0, new_space);
523 pdi->ncpus = new_count;
527 static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
529 if (cpu >= pdi->ncpus)
530 resize_cpu_info(pdi, cpu);
531 return &pdi->cpus[cpu];
535 static int resize_devices(char *name)
537 int size = (ndevices + 1) * sizeof(struct per_dev_info);
539 devices = realloc(devices, size);
541 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
544 memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
545 devices[ndevices].name = name;
550 static struct per_dev_info *get_dev_info(dev_t id, int create)
554 for (i = 0; i < ndevices; i++)
555 if (devices[i].id == id)
559 if (resize_devices(NULL) != 0)
561 return &devices[ndevices-1];
564 static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
567 snprintf(buffer, size, "%s", pdi->name);
569 snprintf(buffer, size, "%d,%d", MAJOR(pdi->id), MINOR(pdi->id));
574 static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
576 unsigned long long this = bit->time;
577 unsigned long long last = pdi->last_reported_time;
579 pdi->backwards = (this < last) ? 'B' : ' ';
580 pdi->last_reported_time = this;
584 static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
589 ios->qwrite_kb += t->bytes >> 10;
592 ios->qread_kb += t->bytes >> 10;
596 static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
599 __account_m(&pci->io_stats, t, rw);
601 if (per_process_stats) {
602 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
604 __account_m(ios, t, rw);
608 static inline void __account_q(struct io_stats *ios, struct blk_io_trace *t,
613 ios->qwrite_kb += t->bytes >> 10;
616 ios->qread_kb += t->bytes >> 10;
620 static inline void account_q(struct blk_io_trace *t, struct per_cpu_info *pci,
623 __account_q(&pci->io_stats, t, rw);
625 if (per_process_stats) {
626 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
628 __account_q(ios, t, rw);
632 static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes)
636 ios->cwrite_kb += bytes >> 10;
639 ios->cread_kb += bytes >> 10;
643 static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
646 __account_c(&pci->io_stats, rw, bytes);
648 if (per_process_stats) {
649 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
651 __account_c(ios, rw, bytes);
655 static inline void __account_i(struct io_stats *ios, int rw, unsigned int bytes)
659 ios->iwrite_kb += bytes >> 10;
662 ios->iread_kb += bytes >> 10;
666 static inline void account_i(struct blk_io_trace *t, struct per_cpu_info *pci,
669 __account_i(&pci->io_stats, rw, t->bytes);
671 if (per_process_stats) {
672 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
674 __account_i(ios, rw, t->bytes);
678 static inline void __account_unplug(struct io_stats *ios, int timer)
681 ios->timer_unplugs++;
686 static inline void account_unplug(struct blk_io_trace *t,
687 struct per_cpu_info *pci, int timer)
689 __account_unplug(&pci->io_stats, timer);
691 if (per_process_stats) {
692 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
694 __account_unplug(ios, timer);
698 static void output(struct per_cpu_info *pci, char *s)
700 fprintf(ofp, "%s", s);
703 static char hstring[256];
704 static char tstring[256];
706 static inline char *setup_header(struct per_cpu_info *pci,
707 struct blk_io_trace *t, char *act)
709 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
710 int b = t->action & BLK_TC_ACT(BLK_TC_BARRIER);
711 int s = t->action & BLK_TC_ACT(BLK_TC_SYNC);
726 sprintf(hstring, "%3d,%-3d %2d %8ld %5Lu.%09Lu %5u %2s %3s",
727 MAJOR(t->device), MINOR(t->device), pci->cpu,
728 (unsigned long)t->sequence, SECONDS(t->time),
729 NANO_SECONDS(t->time), t->pid, act, rwbs);
734 static void log_complete(struct per_cpu_info *pci, struct blk_io_trace *t,
737 unsigned long long elapsed = log_track_complete(t);
739 if (elapsed != -1ULL) {
740 unsigned long usec = elapsed / 1000;
742 sprintf(tstring,"%s %Lu + %u (%8lu) [%d]\n",
743 setup_header(pci, t, act),
744 (unsigned long long)t->sector, t->bytes >> 9,
747 sprintf(tstring,"%s %Lu + %u [%d]\n", setup_header(pci, t, act),
748 (unsigned long long)t->sector, t->bytes >> 9, t->error);
751 output(pci, tstring);
754 static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
757 unsigned long long elapsed = log_track_queue(t);
759 if (elapsed != -1ULL) {
760 unsigned long usec = elapsed / 1000;
762 sprintf(tstring,"%s %Lu + %u (%8lu) [%s]\n",
763 setup_header(pci, t, act),
764 (unsigned long long)t->sector, t->bytes >> 9,
767 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
768 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
770 output(pci, tstring);
773 static void log_issue(struct per_cpu_info *pci, struct blk_io_trace *t,
776 unsigned long long elapsed = log_track_issue(t);
778 if (elapsed != -1ULL) {
779 double usec = (double) elapsed / 1000;
781 sprintf(tstring,"%s %Lu + %u (%8.2f) [%s]\n",
782 setup_header(pci, t, act),
783 (unsigned long long)t->sector, t->bytes >> 9,
786 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
787 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
790 output(pci, tstring);
793 static void log_merge(struct per_cpu_info *pci, struct blk_io_trace *t,
798 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
799 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
800 output(pci, tstring);
803 static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
806 sprintf(tstring,"%s [%s]\n", setup_header(pci, t, act), t->comm);
807 output(pci, tstring);
810 static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
813 sprintf(tstring,"%s %Lu + %u [%s]\n", setup_header(pci, t, act),
814 (unsigned long long)t->sector, t->bytes >> 9, t->comm);
815 output(pci, tstring);
818 static int log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
824 len = sprintf(tstring,"%s [%s] ", setup_header(pci, t, act), t->comm);
825 depth = (__u64 *) ((char *) t + sizeof(*t));
826 sprintf(tstring + len, "%u\n", (unsigned int) be64_to_cpu(*depth));
827 output(pci, tstring);
832 static int log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
837 sprintf(tstring,"%s ", setup_header(pci, t, act));
838 output(pci, tstring);
840 buf = (unsigned char *) t + sizeof(*t);
841 for (i = 0; i < t->pdu_len; i++) {
842 sprintf(tstring,"%02x ", buf[i]);
843 output(pci, tstring);
847 sprintf(tstring,"[%d]\n", t->error);
848 output(pci, tstring);
850 sprintf(tstring,"[%s]\n", t->comm);
851 output(pci, tstring);
856 static int dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
860 switch (t->action & 0xffff) {
862 log_generic(pci, t, "Q");
865 log_generic(pci, t, "G");
867 case __BLK_TA_SLEEPRQ:
868 log_generic(pci, t, "S");
870 case __BLK_TA_REQUEUE:
871 log_generic(pci, t, "R");
874 ret = log_pc(pci, t, "D");
876 case __BLK_TA_COMPLETE:
880 fprintf(stderr, "Bad pc action %x\n", t->action);
888 static void dump_trace_fs(struct blk_io_trace *t, struct per_cpu_info *pci)
890 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
891 int act = t->action & 0xffff;
895 account_q(t, pci, w);
896 log_queue(pci, t, "Q");
898 case __BLK_TA_BACKMERGE:
899 account_m(t, pci, w);
900 log_merge(pci, t, "M");
902 case __BLK_TA_FRONTMERGE:
903 account_m(t, pci, w);
904 log_merge(pci, t, "F");
908 log_generic(pci, t, "G");
910 case __BLK_TA_SLEEPRQ:
911 log_generic(pci, t, "S");
913 case __BLK_TA_REQUEUE:
914 account_c(t, pci, w, -t->bytes);
915 log_queue(pci, t, "R");
918 account_i(t, pci, w);
919 log_issue(pci, t, "D");
921 case __BLK_TA_COMPLETE:
922 account_c(t, pci, w, t->bytes);
923 log_complete(pci, t, "C");
926 log_action(pci, t, "P");
928 case __BLK_TA_UNPLUG_IO:
929 account_unplug(t, pci, 0);
930 log_unplug(pci, t, "U");
932 case __BLK_TA_UNPLUG_TIMER:
933 account_unplug(t, pci, 1);
934 log_unplug(pci, t, "UT");
937 fprintf(stderr, "Bad fs action %x\n", t->action);
942 static int dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
943 struct per_dev_info *pdi)
947 if (t->action & BLK_TC_ACT(BLK_TC_PC))
948 ret = dump_trace_pc(t, pci);
950 dump_trace_fs(t, pci);
956 static void dump_io_stats(struct io_stats *ios, char *msg)
958 fprintf(ofp, "%s\n", msg);
960 fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
961 fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
963 fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
964 fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
965 fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
966 fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
967 fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
968 fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
969 fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
970 fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
973 static void dump_wait_stats(struct per_process_info *ppi)
975 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
976 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
977 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
978 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
979 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
980 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
982 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
983 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
984 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
985 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
986 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
987 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
990 static void show_process_stats(void)
992 struct per_process_info *ppi;
996 dump_io_stats(&ppi->io_stats, ppi->name);
997 dump_wait_stats(ppi);
998 ppi = ppi->list_next;
1004 static void show_device_and_cpu_stats(void)
1006 struct per_dev_info *pdi;
1007 struct per_cpu_info *pci;
1008 struct io_stats total, *ios;
1009 int i, j, pci_events;
1010 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1013 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1015 memset(&total, 0, sizeof(total));
1021 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1025 ios = &pci->io_stats;
1026 total.qreads += ios->qreads;
1027 total.qwrites += ios->qwrites;
1028 total.creads += ios->creads;
1029 total.cwrites += ios->cwrites;
1030 total.mreads += ios->mreads;
1031 total.mwrites += ios->mwrites;
1032 total.ireads += ios->ireads;
1033 total.iwrites += ios->iwrites;
1034 total.qread_kb += ios->qread_kb;
1035 total.qwrite_kb += ios->qwrite_kb;
1036 total.cread_kb += ios->cread_kb;
1037 total.cwrite_kb += ios->cwrite_kb;
1038 total.iread_kb += ios->iread_kb;
1039 total.iwrite_kb += ios->iwrite_kb;
1040 total.timer_unplugs += ios->timer_unplugs;
1041 total.io_unplugs += ios->io_unplugs;
1043 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1044 j, get_dev_name(pdi, name, sizeof(name)));
1045 dump_io_stats(ios, line);
1049 if (pci_events > 1) {
1051 snprintf(line, sizeof(line) - 1, "Total (%s):",
1052 get_dev_name(pdi, name, sizeof(name)));
1053 dump_io_stats(&total, line);
1056 fprintf(ofp, "Events (%s): %'Lu\n",
1057 get_dev_name(pdi, line, sizeof(line)), pdi->events);
1061 static struct blk_io_trace *find_trace(void *p, unsigned long offset, int nr)
1063 unsigned long max_offset = min(offset,nr * sizeof(struct blk_io_trace));
1065 struct blk_io_trace *bit;
1068 for (off = 0; off < max_offset; off++) {
1071 magic = be32_to_cpu(bit->magic);
1072 if ((magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
1079 static int sort_entries(void *traces, unsigned long offset, int nr,
1080 struct per_dev_info *fpdi, struct per_cpu_info *fpci)
1082 struct per_dev_info *pdi;
1083 struct per_cpu_info *pci;
1084 struct blk_io_trace *bit;
1086 void *start = traces;
1088 while (traces - start <= offset - sizeof(*bit)) {
1092 bit = find_trace(traces, offset - (traces - start), nr);
1096 t = malloc(sizeof(*t));
1098 fprintf(stderr, "Out of memory, seq %d on dev %d,%d\n",
1100 MAJOR(bit->device), MINOR(bit->device));
1104 memset(&t->rb_node, 0, sizeof(t->rb_node));
1108 if (verify_trace(bit)) {
1113 pdi = fpdi ? fpdi : get_dev_info(bit->device, 1);
1114 pdi->id = bit->device;
1115 pci = fpci ? fpci : get_cpu_info(pdi, bit->cpu);
1116 pci->cpu = bit->cpu;
1119 if (trace_rb_insert(t)) {
1124 traces += sizeof(*bit) + bit->pdu_len;
1131 static void free_entries_rb(void)
1135 while ((n = rb_first(&rb_sort_root)) != NULL) {
1136 struct trace *t = rb_entry(n, struct trace, rb_node);
1138 rb_erase(&t->rb_node, &rb_sort_root);
1143 static void show_entries_rb(void)
1145 struct per_dev_info *pdi;
1146 struct blk_io_trace *bit;
1151 n = rb_first(&rb_sort_root);
1156 t = rb_entry(n, struct trace, rb_node);
1159 pdi = get_dev_info(bit->device, 0);
1161 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
1162 MAJOR(bit->device), MINOR(bit->device));
1166 if (cpu > pdi->ncpus) {
1167 fprintf(stderr, "Unknown CPU ID? (%d, device %d,%d)\n",
1168 cpu, MAJOR(bit->device), MINOR(bit->device));
1172 bit->time -= genesis_time;
1173 if (bit->time < stopwatch_start)
1175 if (bit->time >= stopwatch_end)
1178 check_time(pdi, bit);
1180 if (dump_trace(bit, &pdi->cpus[cpu], pdi))
1183 } while ((n = rb_next(n)) != NULL);
1186 static int read_data(int fd, void *buffer, int bytes, int block)
1188 int ret, bytes_left, fl;
1191 fl = fcntl(fd, F_GETFL);
1194 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
1196 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
1200 while (bytes_left > 0) {
1201 ret = read(fd, p, bytes_left);
1205 if (errno != EAGAIN)
1217 static int do_file(void)
1219 struct per_dev_info *pdi;
1220 int i, j, nfiles = 0;
1222 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1223 for (j = 0;; j++, nfiles++) {
1224 struct per_cpu_info *pci;
1228 pci = get_cpu_info(pdi, j);
1231 snprintf(pci->fname, sizeof(pci->fname)-1,
1232 "%s_out.%d", pdi->name, j);
1233 if (stat(pci->fname, &st) < 0)
1238 printf("Processing %s\n", pci->fname);
1240 tb = malloc(st.st_size);
1242 fprintf(stderr, "Out of memory, skip file %s\n",
1247 pci->fd = open(pci->fname, O_RDONLY);
1254 if (read_data(pci->fd, tb, st.st_size, 1)) {
1260 if (sort_entries(tb, st.st_size, ~0U, pdi, pci) == -1) {
1266 printf("Completed %s (CPU%d %d, entries)\n",
1267 pci->fname, j, pci->nelems);
1273 fprintf(stderr, "No files found\n");
1281 static void resize_buffer(void **buffer, long *size, long offset)
1283 long old_size = *size;
1286 *size = 64 * sizeof(struct blk_io_trace);
1289 *buffer = realloc(*buffer, *size);
1292 memset(*buffer + offset, 0, *size - old_size);
1295 static int read_sort_events(int fd, void **buffer, long *max_offset)
1300 events = offset = 0;
1302 struct blk_io_trace *t;
1306 if (*max_offset - offset < sizeof(*t))
1307 resize_buffer(buffer, max_offset, offset);
1309 if (read_data(fd, *buffer + offset, sizeof(*t), !events))
1312 t = *buffer + offset;
1313 offset += sizeof(*t);
1315 magic = be32_to_cpu(t->magic);
1316 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1317 fprintf(stderr, "Bad magic %x\n", magic);
1321 pdu_len = be16_to_cpu(t->pdu_len);
1323 if (*max_offset - offset <= pdu_len)
1324 resize_buffer(buffer, max_offset, offset);
1326 if (read_data(fd, *buffer + offset, pdu_len, 1))
1333 } while (!is_done() && events < rb_batch);
1338 static int do_stdin(void)
1344 fd = dup(STDIN_FILENO);
1349 events = read_sort_events(fd, &ptr, &max_offset);
1353 if (sort_entries(ptr, ~0UL, events, NULL, NULL) == -1)
1367 static void flush_output(void)
1372 static void handle_sigint(int sig)
1379 * Extract start and duration times from a string, allowing
1380 * us to specify a time interval of interest within a trace.
1381 * Format: "duration" (start is zero) or "start:duration".
1383 static int find_stopwatch_interval(char *string)
1388 value = strtod(string, &sp);
1390 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
1394 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
1396 value = strtod(string, &sp);
1397 if (sp == string || *sp != '\0') {
1398 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
1402 } else if (*sp != '\0') {
1403 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
1406 stopwatch_end = stopwatch_start + DOUBLE_TO_NANO_ULL(value);
1410 static void usage(char *prog)
1412 fprintf(stderr, "Usage: %s "
1413 "[-i <name>] [-o <output>] [-s] [-w N[:n]] <name>...\n",
1417 int main(int argc, char *argv[])
1421 int per_device_and_cpu_stats = 1;
1423 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1426 if (!strcmp(optarg, "-") && !pipeline)
1428 else if (resize_devices(optarg) != 0)
1432 output_name = optarg;
1435 rb_batch = atoi(optarg);
1437 rb_batch = RB_BATCH_DEFAULT;
1440 per_process_stats = 1;
1446 per_device_and_cpu_stats = 0;
1449 if (find_stopwatch_interval(optarg) != 0)
1458 while (optind < argc) {
1459 if (!strcmp(argv[optind], "-") && !pipeline)
1461 else if (resize_devices(argv[optind]) != 0)
1466 if (!pipeline && !ndevices) {
1471 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
1472 memset(&rb_track_root, 0, sizeof(rb_track_root));
1474 signal(SIGINT, handle_sigint);
1475 signal(SIGHUP, handle_sigint);
1476 signal(SIGTERM, handle_sigint);
1478 setlocale(LC_NUMERIC, "en_US");
1481 ofp = fdopen(STDOUT_FILENO, "w");
1486 snprintf(ofname, sizeof(ofname) - 1, "%s.log", output_name);
1487 ofp = fopen(ofname, "w");
1496 ofp_buffer = malloc(4096);
1497 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
1507 if (per_process_stats)
1508 show_process_stats();
1510 if (per_device_and_cpu_stats)
1511 show_device_and_cpu_stats();