x86/xen: resume timer irqs early
[linux/fpc-iii.git] / kernel / trace / blktrace.c
blobe0e5f73db5a49ad0edb3ad0e382a719d5686ba72
1 /*
2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/slab.h>
25 #include <linux/debugfs.h>
26 #include <linux/export.h>
27 #include <linux/time.h>
28 #include <linux/uaccess.h>
29 #include <linux/list.h>
31 #include <trace/events/block.h>
33 #include "trace_output.h"
35 #ifdef CONFIG_BLK_DEV_IO_TRACE
37 static unsigned int blktrace_seq __read_mostly = 1;
39 static struct trace_array *blk_tr;
40 static bool blk_tracer_enabled __read_mostly;
42 static LIST_HEAD(running_trace_list);
43 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
45 /* Select an alternative, minimalistic output than the original one */
46 #define TRACE_BLK_OPT_CLASSIC 0x1
48 static struct tracer_opt blk_tracer_opts[] = {
49 /* Default disable the minimalistic output */
50 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
51 { }
54 static struct tracer_flags blk_tracer_flags = {
55 .val = 0,
56 .opts = blk_tracer_opts,
59 /* Global reference count of probes */
60 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
62 static void blk_register_tracepoints(void);
63 static void blk_unregister_tracepoints(void);
66 * Send out a notify message.
68 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
69 const void *data, size_t len)
71 struct blk_io_trace *t;
72 struct ring_buffer_event *event = NULL;
73 struct ring_buffer *buffer = NULL;
74 int pc = 0;
75 int cpu = smp_processor_id();
76 bool blk_tracer = blk_tracer_enabled;
78 if (blk_tracer) {
79 buffer = blk_tr->trace_buffer.buffer;
80 pc = preempt_count();
81 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
82 sizeof(*t) + len,
83 0, pc);
84 if (!event)
85 return;
86 t = ring_buffer_event_data(event);
87 goto record_it;
90 if (!bt->rchan)
91 return;
93 t = relay_reserve(bt->rchan, sizeof(*t) + len);
94 if (t) {
95 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
96 t->time = ktime_to_ns(ktime_get());
97 record_it:
98 t->device = bt->dev;
99 t->action = action;
100 t->pid = pid;
101 t->cpu = cpu;
102 t->pdu_len = len;
103 memcpy((void *) t + sizeof(*t), data, len);
105 if (blk_tracer)
106 trace_buffer_unlock_commit(buffer, event, 0, pc);
111 * Send out a notify for this process, if we haven't done so since a trace
112 * started
114 static void trace_note_tsk(struct task_struct *tsk)
116 unsigned long flags;
117 struct blk_trace *bt;
119 tsk->btrace_seq = blktrace_seq;
120 spin_lock_irqsave(&running_trace_lock, flags);
121 list_for_each_entry(bt, &running_trace_list, running_list) {
122 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
123 sizeof(tsk->comm));
125 spin_unlock_irqrestore(&running_trace_lock, flags);
128 static void trace_note_time(struct blk_trace *bt)
130 struct timespec now;
131 unsigned long flags;
132 u32 words[2];
134 getnstimeofday(&now);
135 words[0] = now.tv_sec;
136 words[1] = now.tv_nsec;
138 local_irq_save(flags);
139 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
140 local_irq_restore(flags);
143 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
145 int n;
146 va_list args;
147 unsigned long flags;
148 char *buf;
150 if (unlikely(bt->trace_state != Blktrace_running &&
151 !blk_tracer_enabled))
152 return;
155 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
156 * message to the trace.
158 if (!(bt->act_mask & BLK_TC_NOTIFY))
159 return;
161 local_irq_save(flags);
162 buf = this_cpu_ptr(bt->msg_data);
163 va_start(args, fmt);
164 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
165 va_end(args);
167 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
168 local_irq_restore(flags);
170 EXPORT_SYMBOL_GPL(__trace_note_message);
172 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
173 pid_t pid)
175 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
176 return 1;
177 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
178 return 1;
179 if (bt->pid && pid != bt->pid)
180 return 1;
182 return 0;
186 * Data direction bit lookup
188 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
189 BLK_TC_ACT(BLK_TC_WRITE) };
191 #define BLK_TC_RAHEAD BLK_TC_AHEAD
193 /* The ilog2() calls fall out because they're constant */
194 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
195 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
198 * The worker for the various blk_add_trace*() types. Fills out a
199 * blk_io_trace structure and places it in a per-cpu subbuffer.
201 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
202 int rw, u32 what, int error, int pdu_len, void *pdu_data)
204 struct task_struct *tsk = current;
205 struct ring_buffer_event *event = NULL;
206 struct ring_buffer *buffer = NULL;
207 struct blk_io_trace *t;
208 unsigned long flags = 0;
209 unsigned long *sequence;
210 pid_t pid;
211 int cpu, pc = 0;
212 bool blk_tracer = blk_tracer_enabled;
214 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
215 return;
217 what |= ddir_act[rw & WRITE];
218 what |= MASK_TC_BIT(rw, SYNC);
219 what |= MASK_TC_BIT(rw, RAHEAD);
220 what |= MASK_TC_BIT(rw, META);
221 what |= MASK_TC_BIT(rw, DISCARD);
222 what |= MASK_TC_BIT(rw, FLUSH);
223 what |= MASK_TC_BIT(rw, FUA);
225 pid = tsk->pid;
226 if (act_log_check(bt, what, sector, pid))
227 return;
228 cpu = raw_smp_processor_id();
230 if (blk_tracer) {
231 tracing_record_cmdline(current);
233 buffer = blk_tr->trace_buffer.buffer;
234 pc = preempt_count();
235 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
236 sizeof(*t) + pdu_len,
237 0, pc);
238 if (!event)
239 return;
240 t = ring_buffer_event_data(event);
241 goto record_it;
244 if (unlikely(tsk->btrace_seq != blktrace_seq))
245 trace_note_tsk(tsk);
248 * A word about the locking here - we disable interrupts to reserve
249 * some space in the relay per-cpu buffer, to prevent an irq
250 * from coming in and stepping on our toes.
252 local_irq_save(flags);
253 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
254 if (t) {
255 sequence = per_cpu_ptr(bt->sequence, cpu);
257 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
258 t->sequence = ++(*sequence);
259 t->time = ktime_to_ns(ktime_get());
260 record_it:
262 * These two are not needed in ftrace as they are in the
263 * generic trace_entry, filled by tracing_generic_entry_update,
264 * but for the trace_event->bin() synthesizer benefit we do it
265 * here too.
267 t->cpu = cpu;
268 t->pid = pid;
270 t->sector = sector;
271 t->bytes = bytes;
272 t->action = what;
273 t->device = bt->dev;
274 t->error = error;
275 t->pdu_len = pdu_len;
277 if (pdu_len)
278 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
280 if (blk_tracer) {
281 trace_buffer_unlock_commit(buffer, event, 0, pc);
282 return;
286 local_irq_restore(flags);
289 static struct dentry *blk_tree_root;
290 static DEFINE_MUTEX(blk_tree_mutex);
292 static void blk_trace_free(struct blk_trace *bt)
294 debugfs_remove(bt->msg_file);
295 debugfs_remove(bt->dropped_file);
296 relay_close(bt->rchan);
297 debugfs_remove(bt->dir);
298 free_percpu(bt->sequence);
299 free_percpu(bt->msg_data);
300 kfree(bt);
303 static void blk_trace_cleanup(struct blk_trace *bt)
305 blk_trace_free(bt);
306 if (atomic_dec_and_test(&blk_probes_ref))
307 blk_unregister_tracepoints();
310 int blk_trace_remove(struct request_queue *q)
312 struct blk_trace *bt;
314 bt = xchg(&q->blk_trace, NULL);
315 if (!bt)
316 return -EINVAL;
318 if (bt->trace_state != Blktrace_running)
319 blk_trace_cleanup(bt);
321 return 0;
323 EXPORT_SYMBOL_GPL(blk_trace_remove);
325 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
326 size_t count, loff_t *ppos)
328 struct blk_trace *bt = filp->private_data;
329 char buf[16];
331 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
333 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
336 static const struct file_operations blk_dropped_fops = {
337 .owner = THIS_MODULE,
338 .open = simple_open,
339 .read = blk_dropped_read,
340 .llseek = default_llseek,
343 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
344 size_t count, loff_t *ppos)
346 char *msg;
347 struct blk_trace *bt;
349 if (count >= BLK_TN_MAX_MSG)
350 return -EINVAL;
352 msg = kmalloc(count + 1, GFP_KERNEL);
353 if (msg == NULL)
354 return -ENOMEM;
356 if (copy_from_user(msg, buffer, count)) {
357 kfree(msg);
358 return -EFAULT;
361 msg[count] = '\0';
362 bt = filp->private_data;
363 __trace_note_message(bt, "%s", msg);
364 kfree(msg);
366 return count;
369 static const struct file_operations blk_msg_fops = {
370 .owner = THIS_MODULE,
371 .open = simple_open,
372 .write = blk_msg_write,
373 .llseek = noop_llseek,
377 * Keep track of how many times we encountered a full subbuffer, to aid
378 * the user space app in telling how many lost events there were.
380 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
381 void *prev_subbuf, size_t prev_padding)
383 struct blk_trace *bt;
385 if (!relay_buf_full(buf))
386 return 1;
388 bt = buf->chan->private_data;
389 atomic_inc(&bt->dropped);
390 return 0;
393 static int blk_remove_buf_file_callback(struct dentry *dentry)
395 debugfs_remove(dentry);
397 return 0;
400 static struct dentry *blk_create_buf_file_callback(const char *filename,
401 struct dentry *parent,
402 umode_t mode,
403 struct rchan_buf *buf,
404 int *is_global)
406 return debugfs_create_file(filename, mode, parent, buf,
407 &relay_file_operations);
410 static struct rchan_callbacks blk_relay_callbacks = {
411 .subbuf_start = blk_subbuf_start_callback,
412 .create_buf_file = blk_create_buf_file_callback,
413 .remove_buf_file = blk_remove_buf_file_callback,
416 static void blk_trace_setup_lba(struct blk_trace *bt,
417 struct block_device *bdev)
419 struct hd_struct *part = NULL;
421 if (bdev)
422 part = bdev->bd_part;
424 if (part) {
425 bt->start_lba = part->start_sect;
426 bt->end_lba = part->start_sect + part->nr_sects;
427 } else {
428 bt->start_lba = 0;
429 bt->end_lba = -1ULL;
434 * Setup everything required to start tracing
436 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
437 struct block_device *bdev,
438 struct blk_user_trace_setup *buts)
440 struct blk_trace *old_bt, *bt = NULL;
441 struct dentry *dir = NULL;
442 int ret, i;
444 if (!buts->buf_size || !buts->buf_nr)
445 return -EINVAL;
447 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
448 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
451 * some device names have larger paths - convert the slashes
452 * to underscores for this to work as expected
454 for (i = 0; i < strlen(buts->name); i++)
455 if (buts->name[i] == '/')
456 buts->name[i] = '_';
458 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
459 if (!bt)
460 return -ENOMEM;
462 ret = -ENOMEM;
463 bt->sequence = alloc_percpu(unsigned long);
464 if (!bt->sequence)
465 goto err;
467 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
468 if (!bt->msg_data)
469 goto err;
471 ret = -ENOENT;
473 mutex_lock(&blk_tree_mutex);
474 if (!blk_tree_root) {
475 blk_tree_root = debugfs_create_dir("block", NULL);
476 if (!blk_tree_root) {
477 mutex_unlock(&blk_tree_mutex);
478 goto err;
481 mutex_unlock(&blk_tree_mutex);
483 dir = debugfs_create_dir(buts->name, blk_tree_root);
485 if (!dir)
486 goto err;
488 bt->dir = dir;
489 bt->dev = dev;
490 atomic_set(&bt->dropped, 0);
491 INIT_LIST_HEAD(&bt->running_list);
493 ret = -EIO;
494 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
495 &blk_dropped_fops);
496 if (!bt->dropped_file)
497 goto err;
499 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
500 if (!bt->msg_file)
501 goto err;
503 bt->rchan = relay_open("trace", dir, buts->buf_size,
504 buts->buf_nr, &blk_relay_callbacks, bt);
505 if (!bt->rchan)
506 goto err;
508 bt->act_mask = buts->act_mask;
509 if (!bt->act_mask)
510 bt->act_mask = (u16) -1;
512 blk_trace_setup_lba(bt, bdev);
514 /* overwrite with user settings */
515 if (buts->start_lba)
516 bt->start_lba = buts->start_lba;
517 if (buts->end_lba)
518 bt->end_lba = buts->end_lba;
520 bt->pid = buts->pid;
521 bt->trace_state = Blktrace_setup;
523 ret = -EBUSY;
524 old_bt = xchg(&q->blk_trace, bt);
525 if (old_bt) {
526 (void) xchg(&q->blk_trace, old_bt);
527 goto err;
530 if (atomic_inc_return(&blk_probes_ref) == 1)
531 blk_register_tracepoints();
533 return 0;
534 err:
535 blk_trace_free(bt);
536 return ret;
539 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
540 struct block_device *bdev,
541 char __user *arg)
543 struct blk_user_trace_setup buts;
544 int ret;
546 ret = copy_from_user(&buts, arg, sizeof(buts));
547 if (ret)
548 return -EFAULT;
550 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
551 if (ret)
552 return ret;
554 if (copy_to_user(arg, &buts, sizeof(buts))) {
555 blk_trace_remove(q);
556 return -EFAULT;
558 return 0;
560 EXPORT_SYMBOL_GPL(blk_trace_setup);
562 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
563 static int compat_blk_trace_setup(struct request_queue *q, char *name,
564 dev_t dev, struct block_device *bdev,
565 char __user *arg)
567 struct blk_user_trace_setup buts;
568 struct compat_blk_user_trace_setup cbuts;
569 int ret;
571 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
572 return -EFAULT;
574 buts = (struct blk_user_trace_setup) {
575 .act_mask = cbuts.act_mask,
576 .buf_size = cbuts.buf_size,
577 .buf_nr = cbuts.buf_nr,
578 .start_lba = cbuts.start_lba,
579 .end_lba = cbuts.end_lba,
580 .pid = cbuts.pid,
582 memcpy(&buts.name, &cbuts.name, 32);
584 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
585 if (ret)
586 return ret;
588 if (copy_to_user(arg, &buts.name, 32)) {
589 blk_trace_remove(q);
590 return -EFAULT;
593 return 0;
595 #endif
597 int blk_trace_startstop(struct request_queue *q, int start)
599 int ret;
600 struct blk_trace *bt = q->blk_trace;
602 if (bt == NULL)
603 return -EINVAL;
606 * For starting a trace, we can transition from a setup or stopped
607 * trace. For stopping a trace, the state must be running
609 ret = -EINVAL;
610 if (start) {
611 if (bt->trace_state == Blktrace_setup ||
612 bt->trace_state == Blktrace_stopped) {
613 blktrace_seq++;
614 smp_mb();
615 bt->trace_state = Blktrace_running;
616 spin_lock_irq(&running_trace_lock);
617 list_add(&bt->running_list, &running_trace_list);
618 spin_unlock_irq(&running_trace_lock);
620 trace_note_time(bt);
621 ret = 0;
623 } else {
624 if (bt->trace_state == Blktrace_running) {
625 bt->trace_state = Blktrace_stopped;
626 spin_lock_irq(&running_trace_lock);
627 list_del_init(&bt->running_list);
628 spin_unlock_irq(&running_trace_lock);
629 relay_flush(bt->rchan);
630 ret = 0;
634 return ret;
636 EXPORT_SYMBOL_GPL(blk_trace_startstop);
639 * blk_trace_ioctl: - handle the ioctls associated with tracing
640 * @bdev: the block device
641 * @cmd: the ioctl cmd
642 * @arg: the argument data, if any
645 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
647 struct request_queue *q;
648 int ret, start = 0;
649 char b[BDEVNAME_SIZE];
651 q = bdev_get_queue(bdev);
652 if (!q)
653 return -ENXIO;
655 mutex_lock(&bdev->bd_mutex);
657 switch (cmd) {
658 case BLKTRACESETUP:
659 bdevname(bdev, b);
660 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
661 break;
662 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
663 case BLKTRACESETUP32:
664 bdevname(bdev, b);
665 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
666 break;
667 #endif
668 case BLKTRACESTART:
669 start = 1;
670 case BLKTRACESTOP:
671 ret = blk_trace_startstop(q, start);
672 break;
673 case BLKTRACETEARDOWN:
674 ret = blk_trace_remove(q);
675 break;
676 default:
677 ret = -ENOTTY;
678 break;
681 mutex_unlock(&bdev->bd_mutex);
682 return ret;
686 * blk_trace_shutdown: - stop and cleanup trace structures
687 * @q: the request queue associated with the device
690 void blk_trace_shutdown(struct request_queue *q)
692 if (q->blk_trace) {
693 blk_trace_startstop(q, 0);
694 blk_trace_remove(q);
699 * blktrace probes
703 * blk_add_trace_rq - Add a trace for a request oriented action
704 * @q: queue the io is for
705 * @rq: the source request
706 * @nr_bytes: number of completed bytes
707 * @what: the action
709 * Description:
710 * Records an action against a request. Will log the bio offset + size.
713 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
714 unsigned int nr_bytes, u32 what)
716 struct blk_trace *bt = q->blk_trace;
718 if (likely(!bt))
719 return;
721 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
722 what |= BLK_TC_ACT(BLK_TC_PC);
723 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
724 what, rq->errors, rq->cmd_len, rq->cmd);
725 } else {
726 what |= BLK_TC_ACT(BLK_TC_FS);
727 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
728 rq->cmd_flags, what, rq->errors, 0, NULL);
732 static void blk_add_trace_rq_abort(void *ignore,
733 struct request_queue *q, struct request *rq)
735 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
738 static void blk_add_trace_rq_insert(void *ignore,
739 struct request_queue *q, struct request *rq)
741 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
744 static void blk_add_trace_rq_issue(void *ignore,
745 struct request_queue *q, struct request *rq)
747 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
750 static void blk_add_trace_rq_requeue(void *ignore,
751 struct request_queue *q,
752 struct request *rq)
754 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
757 static void blk_add_trace_rq_complete(void *ignore,
758 struct request_queue *q,
759 struct request *rq,
760 unsigned int nr_bytes)
762 blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
766 * blk_add_trace_bio - Add a trace for a bio oriented action
767 * @q: queue the io is for
768 * @bio: the source bio
769 * @what: the action
770 * @error: error, if any
772 * Description:
773 * Records an action against a bio. Will log the bio offset + size.
776 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
777 u32 what, int error)
779 struct blk_trace *bt = q->blk_trace;
781 if (likely(!bt))
782 return;
784 if (!error && !bio_flagged(bio, BIO_UPTODATE))
785 error = EIO;
787 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
788 error, 0, NULL);
791 static void blk_add_trace_bio_bounce(void *ignore,
792 struct request_queue *q, struct bio *bio)
794 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
797 static void blk_add_trace_bio_complete(void *ignore,
798 struct request_queue *q, struct bio *bio,
799 int error)
801 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
804 static void blk_add_trace_bio_backmerge(void *ignore,
805 struct request_queue *q,
806 struct request *rq,
807 struct bio *bio)
809 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
812 static void blk_add_trace_bio_frontmerge(void *ignore,
813 struct request_queue *q,
814 struct request *rq,
815 struct bio *bio)
817 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
820 static void blk_add_trace_bio_queue(void *ignore,
821 struct request_queue *q, struct bio *bio)
823 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
826 static void blk_add_trace_getrq(void *ignore,
827 struct request_queue *q,
828 struct bio *bio, int rw)
830 if (bio)
831 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
832 else {
833 struct blk_trace *bt = q->blk_trace;
835 if (bt)
836 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
841 static void blk_add_trace_sleeprq(void *ignore,
842 struct request_queue *q,
843 struct bio *bio, int rw)
845 if (bio)
846 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
847 else {
848 struct blk_trace *bt = q->blk_trace;
850 if (bt)
851 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
852 0, 0, NULL);
856 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
858 struct blk_trace *bt = q->blk_trace;
860 if (bt)
861 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
864 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
865 unsigned int depth, bool explicit)
867 struct blk_trace *bt = q->blk_trace;
869 if (bt) {
870 __be64 rpdu = cpu_to_be64(depth);
871 u32 what;
873 if (explicit)
874 what = BLK_TA_UNPLUG_IO;
875 else
876 what = BLK_TA_UNPLUG_TIMER;
878 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
882 static void blk_add_trace_split(void *ignore,
883 struct request_queue *q, struct bio *bio,
884 unsigned int pdu)
886 struct blk_trace *bt = q->blk_trace;
888 if (bt) {
889 __be64 rpdu = cpu_to_be64(pdu);
891 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
892 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
893 sizeof(rpdu), &rpdu);
898 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
899 * @ignore: trace callback data parameter (not used)
900 * @q: queue the io is for
901 * @bio: the source bio
902 * @dev: target device
903 * @from: source sector
905 * Description:
906 * Device mapper or raid target sometimes need to split a bio because
907 * it spans a stripe (or similar). Add a trace for that action.
910 static void blk_add_trace_bio_remap(void *ignore,
911 struct request_queue *q, struct bio *bio,
912 dev_t dev, sector_t from)
914 struct blk_trace *bt = q->blk_trace;
915 struct blk_io_trace_remap r;
917 if (likely(!bt))
918 return;
920 r.device_from = cpu_to_be32(dev);
921 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
922 r.sector_from = cpu_to_be64(from);
924 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
925 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
926 sizeof(r), &r);
930 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
931 * @ignore: trace callback data parameter (not used)
932 * @q: queue the io is for
933 * @rq: the source request
934 * @dev: target device
935 * @from: source sector
937 * Description:
938 * Device mapper remaps request to other devices.
939 * Add a trace for that action.
942 static void blk_add_trace_rq_remap(void *ignore,
943 struct request_queue *q,
944 struct request *rq, dev_t dev,
945 sector_t from)
947 struct blk_trace *bt = q->blk_trace;
948 struct blk_io_trace_remap r;
950 if (likely(!bt))
951 return;
953 r.device_from = cpu_to_be32(dev);
954 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
955 r.sector_from = cpu_to_be64(from);
957 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
958 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
959 sizeof(r), &r);
963 * blk_add_driver_data - Add binary message with driver-specific data
964 * @q: queue the io is for
965 * @rq: io request
966 * @data: driver-specific data
967 * @len: length of driver-specific data
969 * Description:
970 * Some drivers might want to write driver-specific data per request.
973 void blk_add_driver_data(struct request_queue *q,
974 struct request *rq,
975 void *data, size_t len)
977 struct blk_trace *bt = q->blk_trace;
979 if (likely(!bt))
980 return;
982 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
983 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
984 BLK_TA_DRV_DATA, rq->errors, len, data);
985 else
986 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
987 BLK_TA_DRV_DATA, rq->errors, len, data);
989 EXPORT_SYMBOL_GPL(blk_add_driver_data);
991 static void blk_register_tracepoints(void)
993 int ret;
995 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
996 WARN_ON(ret);
997 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
998 WARN_ON(ret);
999 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1000 WARN_ON(ret);
1001 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1002 WARN_ON(ret);
1003 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1004 WARN_ON(ret);
1005 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1006 WARN_ON(ret);
1007 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1008 WARN_ON(ret);
1009 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1010 WARN_ON(ret);
1011 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1012 WARN_ON(ret);
1013 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1014 WARN_ON(ret);
1015 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1016 WARN_ON(ret);
1017 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1018 WARN_ON(ret);
1019 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1020 WARN_ON(ret);
1021 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1022 WARN_ON(ret);
1023 ret = register_trace_block_split(blk_add_trace_split, NULL);
1024 WARN_ON(ret);
1025 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1026 WARN_ON(ret);
1027 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1028 WARN_ON(ret);
1031 static void blk_unregister_tracepoints(void)
1033 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1034 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1035 unregister_trace_block_split(blk_add_trace_split, NULL);
1036 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1037 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1038 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1039 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1040 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1041 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1042 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1043 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1044 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1045 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1046 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1047 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1048 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1049 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1051 tracepoint_synchronize_unregister();
1055 * struct blk_io_tracer formatting routines
1058 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1060 int i = 0;
1061 int tc = t->action >> BLK_TC_SHIFT;
1063 if (t->action == BLK_TN_MESSAGE) {
1064 rwbs[i++] = 'N';
1065 goto out;
1068 if (tc & BLK_TC_FLUSH)
1069 rwbs[i++] = 'F';
1071 if (tc & BLK_TC_DISCARD)
1072 rwbs[i++] = 'D';
1073 else if (tc & BLK_TC_WRITE)
1074 rwbs[i++] = 'W';
1075 else if (t->bytes)
1076 rwbs[i++] = 'R';
1077 else
1078 rwbs[i++] = 'N';
1080 if (tc & BLK_TC_FUA)
1081 rwbs[i++] = 'F';
1082 if (tc & BLK_TC_AHEAD)
1083 rwbs[i++] = 'A';
1084 if (tc & BLK_TC_SYNC)
1085 rwbs[i++] = 'S';
1086 if (tc & BLK_TC_META)
1087 rwbs[i++] = 'M';
1088 out:
1089 rwbs[i] = '\0';
1092 static inline
1093 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1095 return (const struct blk_io_trace *)ent;
1098 static inline const void *pdu_start(const struct trace_entry *ent)
1100 return te_blk_io_trace(ent) + 1;
1103 static inline u32 t_action(const struct trace_entry *ent)
1105 return te_blk_io_trace(ent)->action;
1108 static inline u32 t_bytes(const struct trace_entry *ent)
1110 return te_blk_io_trace(ent)->bytes;
1113 static inline u32 t_sec(const struct trace_entry *ent)
1115 return te_blk_io_trace(ent)->bytes >> 9;
1118 static inline unsigned long long t_sector(const struct trace_entry *ent)
1120 return te_blk_io_trace(ent)->sector;
1123 static inline __u16 t_error(const struct trace_entry *ent)
1125 return te_blk_io_trace(ent)->error;
1128 static __u64 get_pdu_int(const struct trace_entry *ent)
1130 const __u64 *val = pdu_start(ent);
1131 return be64_to_cpu(*val);
1134 static void get_pdu_remap(const struct trace_entry *ent,
1135 struct blk_io_trace_remap *r)
1137 const struct blk_io_trace_remap *__r = pdu_start(ent);
1138 __u64 sector_from = __r->sector_from;
1140 r->device_from = be32_to_cpu(__r->device_from);
1141 r->device_to = be32_to_cpu(__r->device_to);
1142 r->sector_from = be64_to_cpu(sector_from);
1145 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1147 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1149 char rwbs[RWBS_LEN];
1150 unsigned long long ts = iter->ts;
1151 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1152 unsigned secs = (unsigned long)ts;
1153 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1155 fill_rwbs(rwbs, t);
1157 return trace_seq_printf(&iter->seq,
1158 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1159 MAJOR(t->device), MINOR(t->device), iter->cpu,
1160 secs, nsec_rem, iter->ent->pid, act, rwbs);
1163 static int blk_log_action(struct trace_iterator *iter, const char *act)
1165 char rwbs[RWBS_LEN];
1166 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1168 fill_rwbs(rwbs, t);
1169 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1170 MAJOR(t->device), MINOR(t->device), act, rwbs);
1173 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1175 const unsigned char *pdu_buf;
1176 int pdu_len;
1177 int i, end, ret;
1179 pdu_buf = pdu_start(ent);
1180 pdu_len = te_blk_io_trace(ent)->pdu_len;
1182 if (!pdu_len)
1183 return 1;
1185 /* find the last zero that needs to be printed */
1186 for (end = pdu_len - 1; end >= 0; end--)
1187 if (pdu_buf[end])
1188 break;
1189 end++;
1191 if (!trace_seq_putc(s, '('))
1192 return 0;
1194 for (i = 0; i < pdu_len; i++) {
1196 ret = trace_seq_printf(s, "%s%02x",
1197 i == 0 ? "" : " ", pdu_buf[i]);
1198 if (!ret)
1199 return ret;
1202 * stop when the rest is just zeroes and indicate so
1203 * with a ".." appended
1205 if (i == end && end != pdu_len - 1)
1206 return trace_seq_puts(s, " ..) ");
1209 return trace_seq_puts(s, ") ");
1212 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1214 char cmd[TASK_COMM_LEN];
1216 trace_find_cmdline(ent->pid, cmd);
1218 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1219 int ret;
1221 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1222 if (!ret)
1223 return 0;
1224 ret = blk_log_dump_pdu(s, ent);
1225 if (!ret)
1226 return 0;
1227 return trace_seq_printf(s, "[%s]\n", cmd);
1228 } else {
1229 if (t_sec(ent))
1230 return trace_seq_printf(s, "%llu + %u [%s]\n",
1231 t_sector(ent), t_sec(ent), cmd);
1232 return trace_seq_printf(s, "[%s]\n", cmd);
1236 static int blk_log_with_error(struct trace_seq *s,
1237 const struct trace_entry *ent)
1239 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1240 int ret;
1242 ret = blk_log_dump_pdu(s, ent);
1243 if (ret)
1244 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1245 return 0;
1246 } else {
1247 if (t_sec(ent))
1248 return trace_seq_printf(s, "%llu + %u [%d]\n",
1249 t_sector(ent),
1250 t_sec(ent), t_error(ent));
1251 return trace_seq_printf(s, "%llu [%d]\n",
1252 t_sector(ent), t_error(ent));
1256 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1258 struct blk_io_trace_remap r = { .device_from = 0, };
1260 get_pdu_remap(ent, &r);
1261 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1262 t_sector(ent), t_sec(ent),
1263 MAJOR(r.device_from), MINOR(r.device_from),
1264 (unsigned long long)r.sector_from);
1267 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1269 char cmd[TASK_COMM_LEN];
1271 trace_find_cmdline(ent->pid, cmd);
1273 return trace_seq_printf(s, "[%s]\n", cmd);
1276 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1278 char cmd[TASK_COMM_LEN];
1280 trace_find_cmdline(ent->pid, cmd);
1282 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1285 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1287 char cmd[TASK_COMM_LEN];
1289 trace_find_cmdline(ent->pid, cmd);
1291 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1292 get_pdu_int(ent), cmd);
1295 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1297 int ret;
1298 const struct blk_io_trace *t = te_blk_io_trace(ent);
1300 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1301 if (ret)
1302 return trace_seq_putc(s, '\n');
1303 return ret;
1307 * struct tracer operations
1310 static void blk_tracer_print_header(struct seq_file *m)
1312 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1313 return;
1314 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1315 "# | | | | | |\n");
1318 static void blk_tracer_start(struct trace_array *tr)
1320 blk_tracer_enabled = true;
1323 static int blk_tracer_init(struct trace_array *tr)
1325 blk_tr = tr;
1326 blk_tracer_start(tr);
1327 return 0;
1330 static void blk_tracer_stop(struct trace_array *tr)
1332 blk_tracer_enabled = false;
1335 static void blk_tracer_reset(struct trace_array *tr)
1337 blk_tracer_stop(tr);
1340 static const struct {
1341 const char *act[2];
1342 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1343 } what2act[] = {
1344 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1345 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1346 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1347 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1348 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1349 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1350 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1351 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1352 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1353 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1354 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1355 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1356 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1357 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1358 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1361 static enum print_line_t print_one_line(struct trace_iterator *iter,
1362 bool classic)
1364 struct trace_seq *s = &iter->seq;
1365 const struct blk_io_trace *t;
1366 u16 what;
1367 int ret;
1368 bool long_act;
1369 blk_log_action_t *log_action;
1371 t = te_blk_io_trace(iter->ent);
1372 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1373 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1374 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1376 if (t->action == BLK_TN_MESSAGE) {
1377 ret = log_action(iter, long_act ? "message" : "m");
1378 if (ret)
1379 ret = blk_log_msg(s, iter->ent);
1380 goto out;
1383 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1384 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1385 else {
1386 ret = log_action(iter, what2act[what].act[long_act]);
1387 if (ret)
1388 ret = what2act[what].print(s, iter->ent);
1390 out:
1391 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1394 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1395 int flags, struct trace_event *event)
1397 return print_one_line(iter, false);
1400 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1402 struct trace_seq *s = &iter->seq;
1403 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1404 const int offset = offsetof(struct blk_io_trace, sector);
1405 struct blk_io_trace old = {
1406 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1407 .time = iter->ts,
1410 if (!trace_seq_putmem(s, &old, offset))
1411 return 0;
1412 return trace_seq_putmem(s, &t->sector,
1413 sizeof(old) - offset + t->pdu_len);
1416 static enum print_line_t
1417 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1418 struct trace_event *event)
1420 return blk_trace_synthesize_old_trace(iter) ?
1421 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1424 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1426 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1427 return TRACE_TYPE_UNHANDLED;
1429 return print_one_line(iter, true);
1432 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1434 /* don't output context-info for blk_classic output */
1435 if (bit == TRACE_BLK_OPT_CLASSIC) {
1436 if (set)
1437 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1438 else
1439 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1441 return 0;
1444 static struct tracer blk_tracer __read_mostly = {
1445 .name = "blk",
1446 .init = blk_tracer_init,
1447 .reset = blk_tracer_reset,
1448 .start = blk_tracer_start,
1449 .stop = blk_tracer_stop,
1450 .print_header = blk_tracer_print_header,
1451 .print_line = blk_tracer_print_line,
1452 .flags = &blk_tracer_flags,
1453 .set_flag = blk_tracer_set_flag,
1456 static struct trace_event_functions trace_blk_event_funcs = {
1457 .trace = blk_trace_event_print,
1458 .binary = blk_trace_event_print_binary,
1461 static struct trace_event trace_blk_event = {
1462 .type = TRACE_BLK,
1463 .funcs = &trace_blk_event_funcs,
1466 static int __init init_blk_tracer(void)
1468 if (!register_ftrace_event(&trace_blk_event)) {
1469 pr_warning("Warning: could not register block events\n");
1470 return 1;
1473 if (register_tracer(&blk_tracer) != 0) {
1474 pr_warning("Warning: could not register the block tracer\n");
1475 unregister_ftrace_event(&trace_blk_event);
1476 return 1;
1479 return 0;
1482 device_initcall(init_blk_tracer);
1484 static int blk_trace_remove_queue(struct request_queue *q)
1486 struct blk_trace *bt;
1488 bt = xchg(&q->blk_trace, NULL);
1489 if (bt == NULL)
1490 return -EINVAL;
1492 if (atomic_dec_and_test(&blk_probes_ref))
1493 blk_unregister_tracepoints();
1495 spin_lock_irq(&running_trace_lock);
1496 list_del(&bt->running_list);
1497 spin_unlock_irq(&running_trace_lock);
1498 blk_trace_free(bt);
1499 return 0;
1503 * Setup everything required to start tracing
1505 static int blk_trace_setup_queue(struct request_queue *q,
1506 struct block_device *bdev)
1508 struct blk_trace *old_bt, *bt = NULL;
1509 int ret = -ENOMEM;
1511 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1512 if (!bt)
1513 return -ENOMEM;
1515 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1516 if (!bt->msg_data)
1517 goto free_bt;
1519 bt->dev = bdev->bd_dev;
1520 bt->act_mask = (u16)-1;
1522 blk_trace_setup_lba(bt, bdev);
1524 old_bt = xchg(&q->blk_trace, bt);
1525 if (old_bt != NULL) {
1526 (void)xchg(&q->blk_trace, old_bt);
1527 ret = -EBUSY;
1528 goto free_bt;
1531 if (atomic_inc_return(&blk_probes_ref) == 1)
1532 blk_register_tracepoints();
1533 return 0;
1535 free_bt:
1536 blk_trace_free(bt);
1537 return ret;
1541 * sysfs interface to enable and configure tracing
1544 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1545 struct device_attribute *attr,
1546 char *buf);
1547 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1548 struct device_attribute *attr,
1549 const char *buf, size_t count);
1550 #define BLK_TRACE_DEVICE_ATTR(_name) \
1551 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1552 sysfs_blk_trace_attr_show, \
1553 sysfs_blk_trace_attr_store)
1555 static BLK_TRACE_DEVICE_ATTR(enable);
1556 static BLK_TRACE_DEVICE_ATTR(act_mask);
1557 static BLK_TRACE_DEVICE_ATTR(pid);
1558 static BLK_TRACE_DEVICE_ATTR(start_lba);
1559 static BLK_TRACE_DEVICE_ATTR(end_lba);
1561 static struct attribute *blk_trace_attrs[] = {
1562 &dev_attr_enable.attr,
1563 &dev_attr_act_mask.attr,
1564 &dev_attr_pid.attr,
1565 &dev_attr_start_lba.attr,
1566 &dev_attr_end_lba.attr,
1567 NULL
1570 struct attribute_group blk_trace_attr_group = {
1571 .name = "trace",
1572 .attrs = blk_trace_attrs,
1575 static const struct {
1576 int mask;
1577 const char *str;
1578 } mask_maps[] = {
1579 { BLK_TC_READ, "read" },
1580 { BLK_TC_WRITE, "write" },
1581 { BLK_TC_FLUSH, "flush" },
1582 { BLK_TC_SYNC, "sync" },
1583 { BLK_TC_QUEUE, "queue" },
1584 { BLK_TC_REQUEUE, "requeue" },
1585 { BLK_TC_ISSUE, "issue" },
1586 { BLK_TC_COMPLETE, "complete" },
1587 { BLK_TC_FS, "fs" },
1588 { BLK_TC_PC, "pc" },
1589 { BLK_TC_AHEAD, "ahead" },
1590 { BLK_TC_META, "meta" },
1591 { BLK_TC_DISCARD, "discard" },
1592 { BLK_TC_DRV_DATA, "drv_data" },
1593 { BLK_TC_FUA, "fua" },
1596 static int blk_trace_str2mask(const char *str)
1598 int i;
1599 int mask = 0;
1600 char *buf, *s, *token;
1602 buf = kstrdup(str, GFP_KERNEL);
1603 if (buf == NULL)
1604 return -ENOMEM;
1605 s = strstrip(buf);
1607 while (1) {
1608 token = strsep(&s, ",");
1609 if (token == NULL)
1610 break;
1612 if (*token == '\0')
1613 continue;
1615 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1616 if (strcasecmp(token, mask_maps[i].str) == 0) {
1617 mask |= mask_maps[i].mask;
1618 break;
1621 if (i == ARRAY_SIZE(mask_maps)) {
1622 mask = -EINVAL;
1623 break;
1626 kfree(buf);
1628 return mask;
1631 static ssize_t blk_trace_mask2str(char *buf, int mask)
1633 int i;
1634 char *p = buf;
1636 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1637 if (mask & mask_maps[i].mask) {
1638 p += sprintf(p, "%s%s",
1639 (p == buf) ? "" : ",", mask_maps[i].str);
1642 *p++ = '\n';
1644 return p - buf;
1647 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1649 if (bdev->bd_disk == NULL)
1650 return NULL;
1652 return bdev_get_queue(bdev);
1655 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1656 struct device_attribute *attr,
1657 char *buf)
1659 struct hd_struct *p = dev_to_part(dev);
1660 struct request_queue *q;
1661 struct block_device *bdev;
1662 ssize_t ret = -ENXIO;
1664 bdev = bdget(part_devt(p));
1665 if (bdev == NULL)
1666 goto out;
1668 q = blk_trace_get_queue(bdev);
1669 if (q == NULL)
1670 goto out_bdput;
1672 mutex_lock(&bdev->bd_mutex);
1674 if (attr == &dev_attr_enable) {
1675 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1676 goto out_unlock_bdev;
1679 if (q->blk_trace == NULL)
1680 ret = sprintf(buf, "disabled\n");
1681 else if (attr == &dev_attr_act_mask)
1682 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1683 else if (attr == &dev_attr_pid)
1684 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1685 else if (attr == &dev_attr_start_lba)
1686 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1687 else if (attr == &dev_attr_end_lba)
1688 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1690 out_unlock_bdev:
1691 mutex_unlock(&bdev->bd_mutex);
1692 out_bdput:
1693 bdput(bdev);
1694 out:
1695 return ret;
1698 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1699 struct device_attribute *attr,
1700 const char *buf, size_t count)
1702 struct block_device *bdev;
1703 struct request_queue *q;
1704 struct hd_struct *p;
1705 u64 value;
1706 ssize_t ret = -EINVAL;
1708 if (count == 0)
1709 goto out;
1711 if (attr == &dev_attr_act_mask) {
1712 if (sscanf(buf, "%llx", &value) != 1) {
1713 /* Assume it is a list of trace category names */
1714 ret = blk_trace_str2mask(buf);
1715 if (ret < 0)
1716 goto out;
1717 value = ret;
1719 } else if (sscanf(buf, "%llu", &value) != 1)
1720 goto out;
1722 ret = -ENXIO;
1724 p = dev_to_part(dev);
1725 bdev = bdget(part_devt(p));
1726 if (bdev == NULL)
1727 goto out;
1729 q = blk_trace_get_queue(bdev);
1730 if (q == NULL)
1731 goto out_bdput;
1733 mutex_lock(&bdev->bd_mutex);
1735 if (attr == &dev_attr_enable) {
1736 if (value)
1737 ret = blk_trace_setup_queue(q, bdev);
1738 else
1739 ret = blk_trace_remove_queue(q);
1740 goto out_unlock_bdev;
1743 ret = 0;
1744 if (q->blk_trace == NULL)
1745 ret = blk_trace_setup_queue(q, bdev);
1747 if (ret == 0) {
1748 if (attr == &dev_attr_act_mask)
1749 q->blk_trace->act_mask = value;
1750 else if (attr == &dev_attr_pid)
1751 q->blk_trace->pid = value;
1752 else if (attr == &dev_attr_start_lba)
1753 q->blk_trace->start_lba = value;
1754 else if (attr == &dev_attr_end_lba)
1755 q->blk_trace->end_lba = value;
1758 out_unlock_bdev:
1759 mutex_unlock(&bdev->bd_mutex);
1760 out_bdput:
1761 bdput(bdev);
1762 out:
1763 return ret ? ret : count;
1766 int blk_trace_init_sysfs(struct device *dev)
1768 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1771 void blk_trace_remove_sysfs(struct device *dev)
1773 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1776 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1778 #ifdef CONFIG_EVENT_TRACING
1780 void blk_dump_cmd(char *buf, struct request *rq)
1782 int i, end;
1783 int len = rq->cmd_len;
1784 unsigned char *cmd = rq->cmd;
1786 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1787 buf[0] = '\0';
1788 return;
1791 for (end = len - 1; end >= 0; end--)
1792 if (cmd[end])
1793 break;
1794 end++;
1796 for (i = 0; i < len; i++) {
1797 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1798 if (i == end && end != len - 1) {
1799 sprintf(buf, " ..");
1800 break;
1805 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1807 int i = 0;
1809 if (rw & REQ_FLUSH)
1810 rwbs[i++] = 'F';
1812 if (rw & WRITE)
1813 rwbs[i++] = 'W';
1814 else if (rw & REQ_DISCARD)
1815 rwbs[i++] = 'D';
1816 else if (bytes)
1817 rwbs[i++] = 'R';
1818 else
1819 rwbs[i++] = 'N';
1821 if (rw & REQ_FUA)
1822 rwbs[i++] = 'F';
1823 if (rw & REQ_RAHEAD)
1824 rwbs[i++] = 'A';
1825 if (rw & REQ_SYNC)
1826 rwbs[i++] = 'S';
1827 if (rw & REQ_META)
1828 rwbs[i++] = 'M';
1829 if (rw & REQ_SECURE)
1830 rwbs[i++] = 'E';
1832 rwbs[i] = '\0';
1834 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1836 #endif /* CONFIG_EVENT_TRACING */