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
) },
54 static struct tracer_flags blk_tracer_flags
= {
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
;
75 int cpu
= smp_processor_id();
76 bool blk_tracer
= blk_tracer_enabled
;
79 buffer
= blk_tr
->trace_buffer
.buffer
;
81 event
= trace_buffer_lock_reserve(buffer
, TRACE_BLK
,
86 t
= ring_buffer_event_data(event
);
93 t
= relay_reserve(bt
->rchan
, sizeof(*t
) + len
);
95 t
->magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
;
96 t
->time
= ktime_to_ns(ktime_get());
103 memcpy((void *) t
+ sizeof(*t
), data
, len
);
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
114 static void trace_note_tsk(struct task_struct
*tsk
)
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
,
125 spin_unlock_irqrestore(&running_trace_lock
, flags
);
128 static void trace_note_time(struct blk_trace
*bt
)
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
, ...)
150 if (unlikely(bt
->trace_state
!= Blktrace_running
&&
151 !blk_tracer_enabled
))
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
))
161 local_irq_save(flags
);
162 buf
= this_cpu_ptr(bt
->msg_data
);
164 n
= vscnprintf(buf
, BLK_TN_MAX_MSG
, fmt
, 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
,
175 if (((bt
->act_mask
<< BLK_TC_SHIFT
) & what
) == 0)
177 if (sector
&& (sector
< bt
->start_lba
|| sector
> bt
->end_lba
))
179 if (bt
->pid
&& pid
!= bt
->pid
)
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
;
212 bool blk_tracer
= blk_tracer_enabled
;
214 if (unlikely(bt
->trace_state
!= Blktrace_running
&& !blk_tracer
))
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
);
226 if (act_log_check(bt
, what
, sector
, pid
))
228 cpu
= raw_smp_processor_id();
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
,
240 t
= ring_buffer_event_data(event
);
244 if (unlikely(tsk
->btrace_seq
!= blktrace_seq
))
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
);
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());
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
275 t
->pdu_len
= pdu_len
;
278 memcpy((void *) t
+ sizeof(*t
), pdu_data
, pdu_len
);
281 trace_buffer_unlock_commit(buffer
, event
, 0, pc
);
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
);
303 static void blk_trace_cleanup(struct blk_trace
*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
);
318 if (bt
->trace_state
!= Blktrace_running
)
319 blk_trace_cleanup(bt
);
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
;
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
,
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
)
347 struct blk_trace
*bt
;
349 if (count
>= BLK_TN_MAX_MSG
)
352 msg
= kmalloc(count
+ 1, GFP_KERNEL
);
356 if (copy_from_user(msg
, buffer
, count
)) {
362 bt
= filp
->private_data
;
363 __trace_note_message(bt
, "%s", msg
);
369 static const struct file_operations blk_msg_fops
= {
370 .owner
= THIS_MODULE
,
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
))
388 bt
= buf
->chan
->private_data
;
389 atomic_inc(&bt
->dropped
);
393 static int blk_remove_buf_file_callback(struct dentry
*dentry
)
395 debugfs_remove(dentry
);
400 static struct dentry
*blk_create_buf_file_callback(const char *filename
,
401 struct dentry
*parent
,
403 struct rchan_buf
*buf
,
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
;
422 part
= bdev
->bd_part
;
425 bt
->start_lba
= part
->start_sect
;
426 bt
->end_lba
= part
->start_sect
+ part
->nr_sects
;
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
;
444 if (!buts
->buf_size
|| !buts
->buf_nr
)
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 strreplace(buts
->name
, '/', '_');
456 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
461 bt
->sequence
= alloc_percpu(unsigned long);
465 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
471 mutex_lock(&blk_tree_mutex
);
472 if (!blk_tree_root
) {
473 blk_tree_root
= debugfs_create_dir("block", NULL
);
474 if (!blk_tree_root
) {
475 mutex_unlock(&blk_tree_mutex
);
479 mutex_unlock(&blk_tree_mutex
);
481 dir
= debugfs_create_dir(buts
->name
, blk_tree_root
);
488 atomic_set(&bt
->dropped
, 0);
489 INIT_LIST_HEAD(&bt
->running_list
);
492 bt
->dropped_file
= debugfs_create_file("dropped", 0444, dir
, bt
,
494 if (!bt
->dropped_file
)
497 bt
->msg_file
= debugfs_create_file("msg", 0222, dir
, bt
, &blk_msg_fops
);
501 bt
->rchan
= relay_open("trace", dir
, buts
->buf_size
,
502 buts
->buf_nr
, &blk_relay_callbacks
, bt
);
506 bt
->act_mask
= buts
->act_mask
;
508 bt
->act_mask
= (u16
) -1;
510 blk_trace_setup_lba(bt
, bdev
);
512 /* overwrite with user settings */
514 bt
->start_lba
= buts
->start_lba
;
516 bt
->end_lba
= buts
->end_lba
;
519 bt
->trace_state
= Blktrace_setup
;
522 old_bt
= xchg(&q
->blk_trace
, bt
);
524 (void) xchg(&q
->blk_trace
, old_bt
);
528 if (atomic_inc_return(&blk_probes_ref
) == 1)
529 blk_register_tracepoints();
537 int blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
538 struct block_device
*bdev
,
541 struct blk_user_trace_setup buts
;
544 ret
= copy_from_user(&buts
, arg
, sizeof(buts
));
548 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
552 if (copy_to_user(arg
, &buts
, sizeof(buts
))) {
558 EXPORT_SYMBOL_GPL(blk_trace_setup
);
560 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
561 static int compat_blk_trace_setup(struct request_queue
*q
, char *name
,
562 dev_t dev
, struct block_device
*bdev
,
565 struct blk_user_trace_setup buts
;
566 struct compat_blk_user_trace_setup cbuts
;
569 if (copy_from_user(&cbuts
, arg
, sizeof(cbuts
)))
572 buts
= (struct blk_user_trace_setup
) {
573 .act_mask
= cbuts
.act_mask
,
574 .buf_size
= cbuts
.buf_size
,
575 .buf_nr
= cbuts
.buf_nr
,
576 .start_lba
= cbuts
.start_lba
,
577 .end_lba
= cbuts
.end_lba
,
581 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
585 if (copy_to_user(arg
, &buts
.name
, ARRAY_SIZE(buts
.name
))) {
594 int blk_trace_startstop(struct request_queue
*q
, int start
)
597 struct blk_trace
*bt
= q
->blk_trace
;
603 * For starting a trace, we can transition from a setup or stopped
604 * trace. For stopping a trace, the state must be running
608 if (bt
->trace_state
== Blktrace_setup
||
609 bt
->trace_state
== Blktrace_stopped
) {
612 bt
->trace_state
= Blktrace_running
;
613 spin_lock_irq(&running_trace_lock
);
614 list_add(&bt
->running_list
, &running_trace_list
);
615 spin_unlock_irq(&running_trace_lock
);
621 if (bt
->trace_state
== Blktrace_running
) {
622 bt
->trace_state
= Blktrace_stopped
;
623 spin_lock_irq(&running_trace_lock
);
624 list_del_init(&bt
->running_list
);
625 spin_unlock_irq(&running_trace_lock
);
626 relay_flush(bt
->rchan
);
633 EXPORT_SYMBOL_GPL(blk_trace_startstop
);
636 * blk_trace_ioctl: - handle the ioctls associated with tracing
637 * @bdev: the block device
638 * @cmd: the ioctl cmd
639 * @arg: the argument data, if any
642 int blk_trace_ioctl(struct block_device
*bdev
, unsigned cmd
, char __user
*arg
)
644 struct request_queue
*q
;
646 char b
[BDEVNAME_SIZE
];
648 q
= bdev_get_queue(bdev
);
652 mutex_lock(&bdev
->bd_mutex
);
657 ret
= blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
659 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
660 case BLKTRACESETUP32
:
662 ret
= compat_blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
668 ret
= blk_trace_startstop(q
, start
);
670 case BLKTRACETEARDOWN
:
671 ret
= blk_trace_remove(q
);
678 mutex_unlock(&bdev
->bd_mutex
);
683 * blk_trace_shutdown: - stop and cleanup trace structures
684 * @q: the request queue associated with the device
687 void blk_trace_shutdown(struct request_queue
*q
)
690 blk_trace_startstop(q
, 0);
700 * blk_add_trace_rq - Add a trace for a request oriented action
701 * @q: queue the io is for
702 * @rq: the source request
703 * @nr_bytes: number of completed bytes
707 * Records an action against a request. Will log the bio offset + size.
710 static void blk_add_trace_rq(struct request_queue
*q
, struct request
*rq
,
711 unsigned int nr_bytes
, u32 what
)
713 struct blk_trace
*bt
= q
->blk_trace
;
718 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
719 what
|= BLK_TC_ACT(BLK_TC_PC
);
720 __blk_add_trace(bt
, 0, nr_bytes
, rq
->cmd_flags
,
721 what
, rq
->errors
, rq
->cmd_len
, rq
->cmd
);
723 what
|= BLK_TC_ACT(BLK_TC_FS
);
724 __blk_add_trace(bt
, blk_rq_pos(rq
), nr_bytes
,
725 rq
->cmd_flags
, what
, rq
->errors
, 0, NULL
);
729 static void blk_add_trace_rq_abort(void *ignore
,
730 struct request_queue
*q
, struct request
*rq
)
732 blk_add_trace_rq(q
, rq
, blk_rq_bytes(rq
), BLK_TA_ABORT
);
735 static void blk_add_trace_rq_insert(void *ignore
,
736 struct request_queue
*q
, struct request
*rq
)
738 blk_add_trace_rq(q
, rq
, blk_rq_bytes(rq
), BLK_TA_INSERT
);
741 static void blk_add_trace_rq_issue(void *ignore
,
742 struct request_queue
*q
, struct request
*rq
)
744 blk_add_trace_rq(q
, rq
, blk_rq_bytes(rq
), BLK_TA_ISSUE
);
747 static void blk_add_trace_rq_requeue(void *ignore
,
748 struct request_queue
*q
,
751 blk_add_trace_rq(q
, rq
, blk_rq_bytes(rq
), BLK_TA_REQUEUE
);
754 static void blk_add_trace_rq_complete(void *ignore
,
755 struct request_queue
*q
,
757 unsigned int nr_bytes
)
759 blk_add_trace_rq(q
, rq
, nr_bytes
, BLK_TA_COMPLETE
);
763 * blk_add_trace_bio - Add a trace for a bio oriented action
764 * @q: queue the io is for
765 * @bio: the source bio
767 * @error: error, if any
770 * Records an action against a bio. Will log the bio offset + size.
773 static void blk_add_trace_bio(struct request_queue
*q
, struct bio
*bio
,
776 struct blk_trace
*bt
= q
->blk_trace
;
781 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
782 bio
->bi_rw
, what
, error
, 0, NULL
);
785 static void blk_add_trace_bio_bounce(void *ignore
,
786 struct request_queue
*q
, struct bio
*bio
)
788 blk_add_trace_bio(q
, bio
, BLK_TA_BOUNCE
, 0);
791 static void blk_add_trace_bio_complete(void *ignore
,
792 struct request_queue
*q
, struct bio
*bio
,
795 blk_add_trace_bio(q
, bio
, BLK_TA_COMPLETE
, error
);
798 static void blk_add_trace_bio_backmerge(void *ignore
,
799 struct request_queue
*q
,
803 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
, 0);
806 static void blk_add_trace_bio_frontmerge(void *ignore
,
807 struct request_queue
*q
,
811 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
, 0);
814 static void blk_add_trace_bio_queue(void *ignore
,
815 struct request_queue
*q
, struct bio
*bio
)
817 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
, 0);
820 static void blk_add_trace_getrq(void *ignore
,
821 struct request_queue
*q
,
822 struct bio
*bio
, int rw
)
825 blk_add_trace_bio(q
, bio
, BLK_TA_GETRQ
, 0);
827 struct blk_trace
*bt
= q
->blk_trace
;
830 __blk_add_trace(bt
, 0, 0, rw
, BLK_TA_GETRQ
, 0, 0, NULL
);
835 static void blk_add_trace_sleeprq(void *ignore
,
836 struct request_queue
*q
,
837 struct bio
*bio
, int rw
)
840 blk_add_trace_bio(q
, bio
, BLK_TA_SLEEPRQ
, 0);
842 struct blk_trace
*bt
= q
->blk_trace
;
845 __blk_add_trace(bt
, 0, 0, rw
, BLK_TA_SLEEPRQ
,
850 static void blk_add_trace_plug(void *ignore
, struct request_queue
*q
)
852 struct blk_trace
*bt
= q
->blk_trace
;
855 __blk_add_trace(bt
, 0, 0, 0, BLK_TA_PLUG
, 0, 0, NULL
);
858 static void blk_add_trace_unplug(void *ignore
, struct request_queue
*q
,
859 unsigned int depth
, bool explicit)
861 struct blk_trace
*bt
= q
->blk_trace
;
864 __be64 rpdu
= cpu_to_be64(depth
);
868 what
= BLK_TA_UNPLUG_IO
;
870 what
= BLK_TA_UNPLUG_TIMER
;
872 __blk_add_trace(bt
, 0, 0, 0, what
, 0, sizeof(rpdu
), &rpdu
);
876 static void blk_add_trace_split(void *ignore
,
877 struct request_queue
*q
, struct bio
*bio
,
880 struct blk_trace
*bt
= q
->blk_trace
;
883 __be64 rpdu
= cpu_to_be64(pdu
);
885 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
,
886 bio
->bi_iter
.bi_size
, bio
->bi_rw
, BLK_TA_SPLIT
,
887 bio
->bi_error
, sizeof(rpdu
), &rpdu
);
892 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
893 * @ignore: trace callback data parameter (not used)
894 * @q: queue the io is for
895 * @bio: the source bio
896 * @dev: target device
897 * @from: source sector
900 * Device mapper or raid target sometimes need to split a bio because
901 * it spans a stripe (or similar). Add a trace for that action.
904 static void blk_add_trace_bio_remap(void *ignore
,
905 struct request_queue
*q
, struct bio
*bio
,
906 dev_t dev
, sector_t from
)
908 struct blk_trace
*bt
= q
->blk_trace
;
909 struct blk_io_trace_remap r
;
914 r
.device_from
= cpu_to_be32(dev
);
915 r
.device_to
= cpu_to_be32(bio
->bi_bdev
->bd_dev
);
916 r
.sector_from
= cpu_to_be64(from
);
918 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
919 bio
->bi_rw
, BLK_TA_REMAP
, bio
->bi_error
,
924 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
925 * @ignore: trace callback data parameter (not used)
926 * @q: queue the io is for
927 * @rq: the source request
928 * @dev: target device
929 * @from: source sector
932 * Device mapper remaps request to other devices.
933 * Add a trace for that action.
936 static void blk_add_trace_rq_remap(void *ignore
,
937 struct request_queue
*q
,
938 struct request
*rq
, dev_t dev
,
941 struct blk_trace
*bt
= q
->blk_trace
;
942 struct blk_io_trace_remap r
;
947 r
.device_from
= cpu_to_be32(dev
);
948 r
.device_to
= cpu_to_be32(disk_devt(rq
->rq_disk
));
949 r
.sector_from
= cpu_to_be64(from
);
951 __blk_add_trace(bt
, blk_rq_pos(rq
), blk_rq_bytes(rq
),
952 rq_data_dir(rq
), BLK_TA_REMAP
, !!rq
->errors
,
957 * blk_add_driver_data - Add binary message with driver-specific data
958 * @q: queue the io is for
960 * @data: driver-specific data
961 * @len: length of driver-specific data
964 * Some drivers might want to write driver-specific data per request.
967 void blk_add_driver_data(struct request_queue
*q
,
969 void *data
, size_t len
)
971 struct blk_trace
*bt
= q
->blk_trace
;
976 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
)
977 __blk_add_trace(bt
, 0, blk_rq_bytes(rq
), 0,
978 BLK_TA_DRV_DATA
, rq
->errors
, len
, data
);
980 __blk_add_trace(bt
, blk_rq_pos(rq
), blk_rq_bytes(rq
), 0,
981 BLK_TA_DRV_DATA
, rq
->errors
, len
, data
);
983 EXPORT_SYMBOL_GPL(blk_add_driver_data
);
985 static void blk_register_tracepoints(void)
989 ret
= register_trace_block_rq_abort(blk_add_trace_rq_abort
, NULL
);
991 ret
= register_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
993 ret
= register_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
995 ret
= register_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
997 ret
= register_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
999 ret
= register_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1001 ret
= register_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1003 ret
= register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1005 ret
= register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1007 ret
= register_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1009 ret
= register_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1011 ret
= register_trace_block_sleeprq(blk_add_trace_sleeprq
, NULL
);
1013 ret
= register_trace_block_plug(blk_add_trace_plug
, NULL
);
1015 ret
= register_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1017 ret
= register_trace_block_split(blk_add_trace_split
, NULL
);
1019 ret
= register_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1021 ret
= register_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1025 static void blk_unregister_tracepoints(void)
1027 unregister_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1028 unregister_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1029 unregister_trace_block_split(blk_add_trace_split
, NULL
);
1030 unregister_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1031 unregister_trace_block_plug(blk_add_trace_plug
, NULL
);
1032 unregister_trace_block_sleeprq(blk_add_trace_sleeprq
, NULL
);
1033 unregister_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1034 unregister_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1035 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1036 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1037 unregister_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1038 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1039 unregister_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1040 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1041 unregister_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1042 unregister_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1043 unregister_trace_block_rq_abort(blk_add_trace_rq_abort
, NULL
);
1045 tracepoint_synchronize_unregister();
1049 * struct blk_io_tracer formatting routines
1052 static void fill_rwbs(char *rwbs
, const struct blk_io_trace
*t
)
1055 int tc
= t
->action
>> BLK_TC_SHIFT
;
1057 if (t
->action
== BLK_TN_MESSAGE
) {
1062 if (tc
& BLK_TC_FLUSH
)
1065 if (tc
& BLK_TC_DISCARD
)
1067 else if (tc
& BLK_TC_WRITE
)
1074 if (tc
& BLK_TC_FUA
)
1076 if (tc
& BLK_TC_AHEAD
)
1078 if (tc
& BLK_TC_SYNC
)
1080 if (tc
& BLK_TC_META
)
1087 const struct blk_io_trace
*te_blk_io_trace(const struct trace_entry
*ent
)
1089 return (const struct blk_io_trace
*)ent
;
1092 static inline const void *pdu_start(const struct trace_entry
*ent
)
1094 return te_blk_io_trace(ent
) + 1;
1097 static inline u32
t_action(const struct trace_entry
*ent
)
1099 return te_blk_io_trace(ent
)->action
;
1102 static inline u32
t_bytes(const struct trace_entry
*ent
)
1104 return te_blk_io_trace(ent
)->bytes
;
1107 static inline u32
t_sec(const struct trace_entry
*ent
)
1109 return te_blk_io_trace(ent
)->bytes
>> 9;
1112 static inline unsigned long long t_sector(const struct trace_entry
*ent
)
1114 return te_blk_io_trace(ent
)->sector
;
1117 static inline __u16
t_error(const struct trace_entry
*ent
)
1119 return te_blk_io_trace(ent
)->error
;
1122 static __u64
get_pdu_int(const struct trace_entry
*ent
)
1124 const __u64
*val
= pdu_start(ent
);
1125 return be64_to_cpu(*val
);
1128 static void get_pdu_remap(const struct trace_entry
*ent
,
1129 struct blk_io_trace_remap
*r
)
1131 const struct blk_io_trace_remap
*__r
= pdu_start(ent
);
1132 __u64 sector_from
= __r
->sector_from
;
1134 r
->device_from
= be32_to_cpu(__r
->device_from
);
1135 r
->device_to
= be32_to_cpu(__r
->device_to
);
1136 r
->sector_from
= be64_to_cpu(sector_from
);
1139 typedef void (blk_log_action_t
) (struct trace_iterator
*iter
, const char *act
);
1141 static void blk_log_action_classic(struct trace_iterator
*iter
, const char *act
)
1143 char rwbs
[RWBS_LEN
];
1144 unsigned long long ts
= iter
->ts
;
1145 unsigned long nsec_rem
= do_div(ts
, NSEC_PER_SEC
);
1146 unsigned secs
= (unsigned long)ts
;
1147 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1151 trace_seq_printf(&iter
->seq
,
1152 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1153 MAJOR(t
->device
), MINOR(t
->device
), iter
->cpu
,
1154 secs
, nsec_rem
, iter
->ent
->pid
, act
, rwbs
);
1157 static void blk_log_action(struct trace_iterator
*iter
, const char *act
)
1159 char rwbs
[RWBS_LEN
];
1160 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1163 trace_seq_printf(&iter
->seq
, "%3d,%-3d %2s %3s ",
1164 MAJOR(t
->device
), MINOR(t
->device
), act
, rwbs
);
1167 static void blk_log_dump_pdu(struct trace_seq
*s
, const struct trace_entry
*ent
)
1169 const unsigned char *pdu_buf
;
1173 pdu_buf
= pdu_start(ent
);
1174 pdu_len
= te_blk_io_trace(ent
)->pdu_len
;
1179 /* find the last zero that needs to be printed */
1180 for (end
= pdu_len
- 1; end
>= 0; end
--)
1185 trace_seq_putc(s
, '(');
1187 for (i
= 0; i
< pdu_len
; i
++) {
1189 trace_seq_printf(s
, "%s%02x",
1190 i
== 0 ? "" : " ", pdu_buf
[i
]);
1193 * stop when the rest is just zeroes and indicate so
1194 * with a ".." appended
1196 if (i
== end
&& end
!= pdu_len
- 1) {
1197 trace_seq_puts(s
, " ..) ");
1202 trace_seq_puts(s
, ") ");
1205 static void blk_log_generic(struct trace_seq
*s
, const struct trace_entry
*ent
)
1207 char cmd
[TASK_COMM_LEN
];
1209 trace_find_cmdline(ent
->pid
, cmd
);
1211 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1212 trace_seq_printf(s
, "%u ", t_bytes(ent
));
1213 blk_log_dump_pdu(s
, ent
);
1214 trace_seq_printf(s
, "[%s]\n", cmd
);
1217 trace_seq_printf(s
, "%llu + %u [%s]\n",
1218 t_sector(ent
), t_sec(ent
), cmd
);
1220 trace_seq_printf(s
, "[%s]\n", cmd
);
1224 static void blk_log_with_error(struct trace_seq
*s
,
1225 const struct trace_entry
*ent
)
1227 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1228 blk_log_dump_pdu(s
, ent
);
1229 trace_seq_printf(s
, "[%d]\n", t_error(ent
));
1232 trace_seq_printf(s
, "%llu + %u [%d]\n",
1234 t_sec(ent
), t_error(ent
));
1236 trace_seq_printf(s
, "%llu [%d]\n",
1237 t_sector(ent
), t_error(ent
));
1241 static void blk_log_remap(struct trace_seq
*s
, const struct trace_entry
*ent
)
1243 struct blk_io_trace_remap r
= { .device_from
= 0, };
1245 get_pdu_remap(ent
, &r
);
1246 trace_seq_printf(s
, "%llu + %u <- (%d,%d) %llu\n",
1247 t_sector(ent
), t_sec(ent
),
1248 MAJOR(r
.device_from
), MINOR(r
.device_from
),
1249 (unsigned long long)r
.sector_from
);
1252 static void blk_log_plug(struct trace_seq
*s
, const struct trace_entry
*ent
)
1254 char cmd
[TASK_COMM_LEN
];
1256 trace_find_cmdline(ent
->pid
, cmd
);
1258 trace_seq_printf(s
, "[%s]\n", cmd
);
1261 static void blk_log_unplug(struct trace_seq
*s
, const struct trace_entry
*ent
)
1263 char cmd
[TASK_COMM_LEN
];
1265 trace_find_cmdline(ent
->pid
, cmd
);
1267 trace_seq_printf(s
, "[%s] %llu\n", cmd
, get_pdu_int(ent
));
1270 static void blk_log_split(struct trace_seq
*s
, const struct trace_entry
*ent
)
1272 char cmd
[TASK_COMM_LEN
];
1274 trace_find_cmdline(ent
->pid
, cmd
);
1276 trace_seq_printf(s
, "%llu / %llu [%s]\n", t_sector(ent
),
1277 get_pdu_int(ent
), cmd
);
1280 static void blk_log_msg(struct trace_seq
*s
, const struct trace_entry
*ent
)
1282 const struct blk_io_trace
*t
= te_blk_io_trace(ent
);
1284 trace_seq_putmem(s
, t
+ 1, t
->pdu_len
);
1285 trace_seq_putc(s
, '\n');
1289 * struct tracer operations
1292 static void blk_tracer_print_header(struct seq_file
*m
)
1294 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1296 seq_puts(m
, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1300 static void blk_tracer_start(struct trace_array
*tr
)
1302 blk_tracer_enabled
= true;
1305 static int blk_tracer_init(struct trace_array
*tr
)
1308 blk_tracer_start(tr
);
1312 static void blk_tracer_stop(struct trace_array
*tr
)
1314 blk_tracer_enabled
= false;
1317 static void blk_tracer_reset(struct trace_array
*tr
)
1319 blk_tracer_stop(tr
);
1322 static const struct {
1324 void (*print
)(struct trace_seq
*s
, const struct trace_entry
*ent
);
1326 [__BLK_TA_QUEUE
] = {{ "Q", "queue" }, blk_log_generic
},
1327 [__BLK_TA_BACKMERGE
] = {{ "M", "backmerge" }, blk_log_generic
},
1328 [__BLK_TA_FRONTMERGE
] = {{ "F", "frontmerge" }, blk_log_generic
},
1329 [__BLK_TA_GETRQ
] = {{ "G", "getrq" }, blk_log_generic
},
1330 [__BLK_TA_SLEEPRQ
] = {{ "S", "sleeprq" }, blk_log_generic
},
1331 [__BLK_TA_REQUEUE
] = {{ "R", "requeue" }, blk_log_with_error
},
1332 [__BLK_TA_ISSUE
] = {{ "D", "issue" }, blk_log_generic
},
1333 [__BLK_TA_COMPLETE
] = {{ "C", "complete" }, blk_log_with_error
},
1334 [__BLK_TA_PLUG
] = {{ "P", "plug" }, blk_log_plug
},
1335 [__BLK_TA_UNPLUG_IO
] = {{ "U", "unplug_io" }, blk_log_unplug
},
1336 [__BLK_TA_UNPLUG_TIMER
] = {{ "UT", "unplug_timer" }, blk_log_unplug
},
1337 [__BLK_TA_INSERT
] = {{ "I", "insert" }, blk_log_generic
},
1338 [__BLK_TA_SPLIT
] = {{ "X", "split" }, blk_log_split
},
1339 [__BLK_TA_BOUNCE
] = {{ "B", "bounce" }, blk_log_generic
},
1340 [__BLK_TA_REMAP
] = {{ "A", "remap" }, blk_log_remap
},
1343 static enum print_line_t
print_one_line(struct trace_iterator
*iter
,
1346 struct trace_seq
*s
= &iter
->seq
;
1347 const struct blk_io_trace
*t
;
1350 blk_log_action_t
*log_action
;
1352 t
= te_blk_io_trace(iter
->ent
);
1353 what
= t
->action
& ((1 << BLK_TC_SHIFT
) - 1);
1354 long_act
= !!(trace_flags
& TRACE_ITER_VERBOSE
);
1355 log_action
= classic
? &blk_log_action_classic
: &blk_log_action
;
1357 if (t
->action
== BLK_TN_MESSAGE
) {
1358 log_action(iter
, long_act
? "message" : "m");
1359 blk_log_msg(s
, iter
->ent
);
1362 if (unlikely(what
== 0 || what
>= ARRAY_SIZE(what2act
)))
1363 trace_seq_printf(s
, "Unknown action %x\n", what
);
1365 log_action(iter
, what2act
[what
].act
[long_act
]);
1366 what2act
[what
].print(s
, iter
->ent
);
1369 return trace_handle_return(s
);
1372 static enum print_line_t
blk_trace_event_print(struct trace_iterator
*iter
,
1373 int flags
, struct trace_event
*event
)
1375 return print_one_line(iter
, false);
1378 static void blk_trace_synthesize_old_trace(struct trace_iterator
*iter
)
1380 struct trace_seq
*s
= &iter
->seq
;
1381 struct blk_io_trace
*t
= (struct blk_io_trace
*)iter
->ent
;
1382 const int offset
= offsetof(struct blk_io_trace
, sector
);
1383 struct blk_io_trace old
= {
1384 .magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
,
1388 trace_seq_putmem(s
, &old
, offset
);
1389 trace_seq_putmem(s
, &t
->sector
,
1390 sizeof(old
) - offset
+ t
->pdu_len
);
1393 static enum print_line_t
1394 blk_trace_event_print_binary(struct trace_iterator
*iter
, int flags
,
1395 struct trace_event
*event
)
1397 blk_trace_synthesize_old_trace(iter
);
1399 return trace_handle_return(&iter
->seq
);
1402 static enum print_line_t
blk_tracer_print_line(struct trace_iterator
*iter
)
1404 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1405 return TRACE_TYPE_UNHANDLED
;
1407 return print_one_line(iter
, true);
1411 blk_tracer_set_flag(struct trace_array
*tr
, u32 old_flags
, u32 bit
, int set
)
1413 /* don't output context-info for blk_classic output */
1414 if (bit
== TRACE_BLK_OPT_CLASSIC
) {
1416 trace_flags
&= ~TRACE_ITER_CONTEXT_INFO
;
1418 trace_flags
|= TRACE_ITER_CONTEXT_INFO
;
1423 static struct tracer blk_tracer __read_mostly
= {
1425 .init
= blk_tracer_init
,
1426 .reset
= blk_tracer_reset
,
1427 .start
= blk_tracer_start
,
1428 .stop
= blk_tracer_stop
,
1429 .print_header
= blk_tracer_print_header
,
1430 .print_line
= blk_tracer_print_line
,
1431 .flags
= &blk_tracer_flags
,
1432 .set_flag
= blk_tracer_set_flag
,
1435 static struct trace_event_functions trace_blk_event_funcs
= {
1436 .trace
= blk_trace_event_print
,
1437 .binary
= blk_trace_event_print_binary
,
1440 static struct trace_event trace_blk_event
= {
1442 .funcs
= &trace_blk_event_funcs
,
1445 static int __init
init_blk_tracer(void)
1447 if (!register_trace_event(&trace_blk_event
)) {
1448 pr_warning("Warning: could not register block events\n");
1452 if (register_tracer(&blk_tracer
) != 0) {
1453 pr_warning("Warning: could not register the block tracer\n");
1454 unregister_trace_event(&trace_blk_event
);
1461 device_initcall(init_blk_tracer
);
1463 static int blk_trace_remove_queue(struct request_queue
*q
)
1465 struct blk_trace
*bt
;
1467 bt
= xchg(&q
->blk_trace
, NULL
);
1471 if (atomic_dec_and_test(&blk_probes_ref
))
1472 blk_unregister_tracepoints();
1479 * Setup everything required to start tracing
1481 static int blk_trace_setup_queue(struct request_queue
*q
,
1482 struct block_device
*bdev
)
1484 struct blk_trace
*old_bt
, *bt
= NULL
;
1487 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
1491 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
1495 bt
->dev
= bdev
->bd_dev
;
1496 bt
->act_mask
= (u16
)-1;
1498 blk_trace_setup_lba(bt
, bdev
);
1500 old_bt
= xchg(&q
->blk_trace
, bt
);
1501 if (old_bt
!= NULL
) {
1502 (void)xchg(&q
->blk_trace
, old_bt
);
1507 if (atomic_inc_return(&blk_probes_ref
) == 1)
1508 blk_register_tracepoints();
1517 * sysfs interface to enable and configure tracing
1520 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1521 struct device_attribute
*attr
,
1523 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1524 struct device_attribute
*attr
,
1525 const char *buf
, size_t count
);
1526 #define BLK_TRACE_DEVICE_ATTR(_name) \
1527 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1528 sysfs_blk_trace_attr_show, \
1529 sysfs_blk_trace_attr_store)
1531 static BLK_TRACE_DEVICE_ATTR(enable
);
1532 static BLK_TRACE_DEVICE_ATTR(act_mask
);
1533 static BLK_TRACE_DEVICE_ATTR(pid
);
1534 static BLK_TRACE_DEVICE_ATTR(start_lba
);
1535 static BLK_TRACE_DEVICE_ATTR(end_lba
);
1537 static struct attribute
*blk_trace_attrs
[] = {
1538 &dev_attr_enable
.attr
,
1539 &dev_attr_act_mask
.attr
,
1541 &dev_attr_start_lba
.attr
,
1542 &dev_attr_end_lba
.attr
,
1546 struct attribute_group blk_trace_attr_group
= {
1548 .attrs
= blk_trace_attrs
,
1551 static const struct {
1555 { BLK_TC_READ
, "read" },
1556 { BLK_TC_WRITE
, "write" },
1557 { BLK_TC_FLUSH
, "flush" },
1558 { BLK_TC_SYNC
, "sync" },
1559 { BLK_TC_QUEUE
, "queue" },
1560 { BLK_TC_REQUEUE
, "requeue" },
1561 { BLK_TC_ISSUE
, "issue" },
1562 { BLK_TC_COMPLETE
, "complete" },
1563 { BLK_TC_FS
, "fs" },
1564 { BLK_TC_PC
, "pc" },
1565 { BLK_TC_AHEAD
, "ahead" },
1566 { BLK_TC_META
, "meta" },
1567 { BLK_TC_DISCARD
, "discard" },
1568 { BLK_TC_DRV_DATA
, "drv_data" },
1569 { BLK_TC_FUA
, "fua" },
1572 static int blk_trace_str2mask(const char *str
)
1576 char *buf
, *s
, *token
;
1578 buf
= kstrdup(str
, GFP_KERNEL
);
1584 token
= strsep(&s
, ",");
1591 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1592 if (strcasecmp(token
, mask_maps
[i
].str
) == 0) {
1593 mask
|= mask_maps
[i
].mask
;
1597 if (i
== ARRAY_SIZE(mask_maps
)) {
1607 static ssize_t
blk_trace_mask2str(char *buf
, int mask
)
1612 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1613 if (mask
& mask_maps
[i
].mask
) {
1614 p
+= sprintf(p
, "%s%s",
1615 (p
== buf
) ? "" : ",", mask_maps
[i
].str
);
1623 static struct request_queue
*blk_trace_get_queue(struct block_device
*bdev
)
1625 if (bdev
->bd_disk
== NULL
)
1628 return bdev_get_queue(bdev
);
1631 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1632 struct device_attribute
*attr
,
1635 struct hd_struct
*p
= dev_to_part(dev
);
1636 struct request_queue
*q
;
1637 struct block_device
*bdev
;
1638 ssize_t ret
= -ENXIO
;
1640 bdev
= bdget(part_devt(p
));
1644 q
= blk_trace_get_queue(bdev
);
1648 mutex_lock(&bdev
->bd_mutex
);
1650 if (attr
== &dev_attr_enable
) {
1651 ret
= sprintf(buf
, "%u\n", !!q
->blk_trace
);
1652 goto out_unlock_bdev
;
1655 if (q
->blk_trace
== NULL
)
1656 ret
= sprintf(buf
, "disabled\n");
1657 else if (attr
== &dev_attr_act_mask
)
1658 ret
= blk_trace_mask2str(buf
, q
->blk_trace
->act_mask
);
1659 else if (attr
== &dev_attr_pid
)
1660 ret
= sprintf(buf
, "%u\n", q
->blk_trace
->pid
);
1661 else if (attr
== &dev_attr_start_lba
)
1662 ret
= sprintf(buf
, "%llu\n", q
->blk_trace
->start_lba
);
1663 else if (attr
== &dev_attr_end_lba
)
1664 ret
= sprintf(buf
, "%llu\n", q
->blk_trace
->end_lba
);
1667 mutex_unlock(&bdev
->bd_mutex
);
1674 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1675 struct device_attribute
*attr
,
1676 const char *buf
, size_t count
)
1678 struct block_device
*bdev
;
1679 struct request_queue
*q
;
1680 struct hd_struct
*p
;
1682 ssize_t ret
= -EINVAL
;
1687 if (attr
== &dev_attr_act_mask
) {
1688 if (sscanf(buf
, "%llx", &value
) != 1) {
1689 /* Assume it is a list of trace category names */
1690 ret
= blk_trace_str2mask(buf
);
1695 } else if (sscanf(buf
, "%llu", &value
) != 1)
1700 p
= dev_to_part(dev
);
1701 bdev
= bdget(part_devt(p
));
1705 q
= blk_trace_get_queue(bdev
);
1709 mutex_lock(&bdev
->bd_mutex
);
1711 if (attr
== &dev_attr_enable
) {
1713 ret
= blk_trace_setup_queue(q
, bdev
);
1715 ret
= blk_trace_remove_queue(q
);
1716 goto out_unlock_bdev
;
1720 if (q
->blk_trace
== NULL
)
1721 ret
= blk_trace_setup_queue(q
, bdev
);
1724 if (attr
== &dev_attr_act_mask
)
1725 q
->blk_trace
->act_mask
= value
;
1726 else if (attr
== &dev_attr_pid
)
1727 q
->blk_trace
->pid
= value
;
1728 else if (attr
== &dev_attr_start_lba
)
1729 q
->blk_trace
->start_lba
= value
;
1730 else if (attr
== &dev_attr_end_lba
)
1731 q
->blk_trace
->end_lba
= value
;
1735 mutex_unlock(&bdev
->bd_mutex
);
1739 return ret
? ret
: count
;
1742 int blk_trace_init_sysfs(struct device
*dev
)
1744 return sysfs_create_group(&dev
->kobj
, &blk_trace_attr_group
);
1747 void blk_trace_remove_sysfs(struct device
*dev
)
1749 sysfs_remove_group(&dev
->kobj
, &blk_trace_attr_group
);
1752 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1754 #ifdef CONFIG_EVENT_TRACING
1756 void blk_dump_cmd(char *buf
, struct request
*rq
)
1759 int len
= rq
->cmd_len
;
1760 unsigned char *cmd
= rq
->cmd
;
1762 if (rq
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
1767 for (end
= len
- 1; end
>= 0; end
--)
1772 for (i
= 0; i
< len
; i
++) {
1773 buf
+= sprintf(buf
, "%s%02x", i
== 0 ? "" : " ", cmd
[i
]);
1774 if (i
== end
&& end
!= len
- 1) {
1775 sprintf(buf
, " ..");
1781 void blk_fill_rwbs(char *rwbs
, u32 rw
, int bytes
)
1790 else if (rw
& REQ_DISCARD
)
1799 if (rw
& REQ_RAHEAD
)
1805 if (rw
& REQ_SECURE
)
1810 EXPORT_SYMBOL_GPL(blk_fill_rwbs
);
1812 #endif /* CONFIG_EVENT_TRACING */