linux/drivers/md/dm-rq.c

855 lines
22 KiB
C
Raw Normal View History

/*
* Copyright (C) 2016 Red Hat, Inc. All rights reserved.
*
* This file is released under the GPL.
*/
#include "dm-core.h"
#include "dm-rq.h"
#include <linux/elevator.h> /* for rq_end_sector() */
#include <linux/blk-mq.h>
#define DM_MSG_PREFIX "core-rq"
#define DM_MQ_NR_HW_QUEUES 1
#define DM_MQ_QUEUE_DEPTH 2048
static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES;
static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH;
/*
* Request-based DM's mempools' reserved IOs set by the user.
*/
#define RESERVED_REQUEST_BASED_IOS 256
static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
static bool use_blk_mq = IS_ENABLED(CONFIG_DM_MQ_DEFAULT);
bool dm_use_blk_mq_default(void)
{
return use_blk_mq;
}
bool dm_use_blk_mq(struct mapped_device *md)
{
return md->use_blk_mq;
}
EXPORT_SYMBOL_GPL(dm_use_blk_mq);
unsigned dm_get_reserved_rq_based_ios(void)
{
return __dm_get_module_param(&reserved_rq_based_ios,
RESERVED_REQUEST_BASED_IOS, DM_RESERVED_MAX_IOS);
}
EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
static unsigned dm_get_blk_mq_nr_hw_queues(void)
{
return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32);
}
static unsigned dm_get_blk_mq_queue_depth(void)
{
return __dm_get_module_param(&dm_mq_queue_depth,
DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH);
}
int dm_request_based(struct mapped_device *md)
{
return queue_is_rq_based(md->queue);
}
static void dm_old_start_queue(struct request_queue *q)
{
unsigned long flags;
spin_lock_irqsave(q->queue_lock, flags);
if (blk_queue_stopped(q))
blk_start_queue(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
static void dm_mq_start_queue(struct request_queue *q)
{
blk_mq_unquiesce_queue(q);
blk_mq_kick_requeue_list(q);
}
void dm_start_queue(struct request_queue *q)
{
if (!q->mq_ops)
dm_old_start_queue(q);
else
dm_mq_start_queue(q);
}
static void dm_old_stop_queue(struct request_queue *q)
{
unsigned long flags;
spin_lock_irqsave(q->queue_lock, flags);
if (!blk_queue_stopped(q))
blk_stop_queue(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
static void dm_mq_stop_queue(struct request_queue *q)
{
if (blk_mq_queue_stopped(q))
return;
dm: Fix a race condition related to stopping and starting queues Ensure that all ongoing dm_mq_queue_rq() and dm_mq_requeue_request() calls have stopped before setting the "queue stopped" flag. This allows to remove the "queue stopped" test from dm_mq_queue_rq() and dm_mq_requeue_request(). This patch fixes a race condition because dm_mq_queue_rq() is called without holding the queue lock and hence BLK_MQ_S_STOPPED can be set at any time while dm_mq_queue_rq() is in progress. This patch prevents that the following hang occurs sporadically when using dm-mq: INFO: task systemd-udevd:10111 blocked for more than 480 seconds. Call Trace: [<ffffffff8161f397>] schedule+0x37/0x90 [<ffffffff816239ef>] schedule_timeout+0x27f/0x470 [<ffffffff8161e76f>] io_schedule_timeout+0x9f/0x110 [<ffffffff8161fb36>] bit_wait_io+0x16/0x60 [<ffffffff8161f929>] __wait_on_bit_lock+0x49/0xa0 [<ffffffff8114fe69>] __lock_page+0xb9/0xc0 [<ffffffff81165d90>] truncate_inode_pages_range+0x3e0/0x760 [<ffffffff81166120>] truncate_inode_pages+0x10/0x20 [<ffffffff81212a20>] kill_bdev+0x30/0x40 [<ffffffff81213d41>] __blkdev_put+0x71/0x360 [<ffffffff81214079>] blkdev_put+0x49/0x170 [<ffffffff812141c0>] blkdev_close+0x20/0x30 [<ffffffff811d48e8>] __fput+0xe8/0x1f0 [<ffffffff811d4a29>] ____fput+0x9/0x10 [<ffffffff810842d3>] task_work_run+0x83/0xb0 [<ffffffff8106606e>] do_exit+0x3ee/0xc40 [<ffffffff8106694b>] do_group_exit+0x4b/0xc0 [<ffffffff81073d9a>] get_signal+0x2ca/0x940 [<ffffffff8101bf43>] do_signal+0x23/0x660 [<ffffffff810022b3>] exit_to_usermode_loop+0x73/0xb0 [<ffffffff81002cb0>] syscall_return_slowpath+0xb0/0xc0 [<ffffffff81624e33>] entry_SYSCALL_64_fastpath+0xa6/0xa8 Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Acked-by: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
2016-10-29 08:22:16 +08:00
blk_mq_quiesce_queue(q);
}
void dm_stop_queue(struct request_queue *q)
{
if (!q->mq_ops)
dm_old_stop_queue(q);
else
dm_mq_stop_queue(q);
}
/*
* Partial completion handling for request-based dm
*/
static void end_clone_bio(struct bio *clone)
{
struct dm_rq_clone_bio_info *info =
container_of(clone, struct dm_rq_clone_bio_info, clone);
struct dm_rq_target_io *tio = info->tio;
unsigned int nr_bytes = info->orig->bi_iter.bi_size;
blk_status_t error = clone->bi_status;
bool is_last = !clone->bi_next;
bio_put(clone);
if (tio->error)
/*
* An error has already been detected on the request.
* Once error occurred, just let clone->end_io() handle
* the remainder.
*/
return;
else if (error) {
/*
* Don't notice the error to the upper layer yet.
* The error handling decision is made by the target driver,
* when the request is completed.
*/
tio->error = error;
goto exit;
}
/*
* I/O for the bio successfully completed.
* Notice the data completion to the upper layer.
*/
tio->completed += nr_bytes;
/*
* Update the original request.
* Do not use blk_end_request() here, because it may complete
* the original request before the clone, and break the ordering.
*/
if (is_last)
exit:
blk_update_request(tio->orig, BLK_STS_OK, tio->completed);
}
static struct dm_rq_target_io *tio_from_request(struct request *rq)
{
return blk_mq_rq_to_pdu(rq);
}
static void rq_end_stats(struct mapped_device *md, struct request *orig)
{
if (unlikely(dm_stats_used(&md->stats))) {
struct dm_rq_target_io *tio = tio_from_request(orig);
tio->duration_jiffies = jiffies - tio->duration_jiffies;
dm_stats_account_io(&md->stats, rq_data_dir(orig),
blk_rq_pos(orig), tio->n_sectors, true,
tio->duration_jiffies, &tio->stats_aux);
}
}
/*
* Don't touch any member of the md after calling this function because
* the md may be freed in dm_put() at the end of this function.
* Or do dm_get() before calling this function and dm_put() later.
*/
static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
{
struct request_queue *q = md->queue;
unsigned long flags;
atomic_dec(&md->pending[rw]);
/* nudge anyone waiting on suspend queue */
if (!md_in_flight(md))
wake_up(&md->wait);
/*
* Run this off this callpath, as drivers could invoke end_io while
* inside their request_fn (and holding the queue lock). Calling
* back into ->request_fn() could deadlock attempting to grab the
* queue lock again.
*/
if (!q->mq_ops && run_queue) {
spin_lock_irqsave(q->queue_lock, flags);
blk_run_queue_async(q);
spin_unlock_irqrestore(q->queue_lock, flags);
}
/*
* dm_put() must be at the end of this function. See the comment above
*/
dm_put(md);
}
/*
* Complete the clone and the original request.
* Must be called without clone's queue lock held,
* see end_clone_request() for more details.
*/
static void dm_end_request(struct request *clone, blk_status_t error)
{
int rw = rq_data_dir(clone);
struct dm_rq_target_io *tio = clone->end_io_data;
struct mapped_device *md = tio->md;
struct request *rq = tio->orig;
blk_rq_unprep_clone(clone);
tio->ti->type->release_clone_rq(clone);
rq_end_stats(md, rq);
if (!rq->q->mq_ops)
blk_end_request_all(rq, error);
else
blk_mq_end_request(rq, error);
rq_completed(md, rw, true);
}
/*
* Requeue the original request of a clone.
*/
static void dm_old_requeue_request(struct request *rq, unsigned long delay_ms)
{
struct request_queue *q = rq->q;
unsigned long flags;
spin_lock_irqsave(q->queue_lock, flags);
blk_requeue_request(q, rq);
blk_delay_queue(q, delay_ms);
spin_unlock_irqrestore(q->queue_lock, flags);
}
static void __dm_mq_kick_requeue_list(struct request_queue *q, unsigned long msecs)
{
blk-mq: Avoid that requeueing starts stopped queues Since blk_mq_requeue_work() starts stopped queues and since execution of this function can be scheduled after a queue has been stopped it is not possible to stop queues without using an additional state variable to track whether or not the queue has been stopped. Hence modify blk_mq_requeue_work() such that it does not start stopped queues. My conclusion after a review of the blk_mq_stop_hw_queues() and blk_mq_{delay_,}kick_requeue_list() callers is as follows: * In the dm driver starting and stopping queues should only happen if __dm_suspend() or __dm_resume() is called and not if the requeue list is processed. * In the SCSI core queue stopping and starting should only be performed by the scsi_internal_device_block() and scsi_internal_device_unblock() functions but not by any other function. Although the blk_mq_stop_hw_queue() call in scsi_queue_rq() may help to reduce CPU load if a LLD queue is full, figuring out whether or not a queue should be restarted when requeueing a command would require to introduce additional locking in scsi_mq_requeue_cmd() to avoid a race with scsi_internal_device_block(). Avoid this complexity by removing the blk_mq_stop_hw_queue() call from scsi_queue_rq(). * In the NVMe core only the functions that call blk_mq_start_stopped_hw_queues() explicitly should start stopped queues. * A blk_mq_start_stopped_hwqueues() call must be added in the xen-blkfront driver in its blkif_recover() function. Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Roger Pau Monné <roger.pau@citrix.com> Cc: Mike Snitzer <snitzer@redhat.com> Cc: James Bottomley <jejb@linux.vnet.ibm.com> Cc: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
2016-10-29 08:20:32 +08:00
blk_mq_delay_kick_requeue_list(q, msecs);
}
void dm_mq_kick_requeue_list(struct mapped_device *md)
{
__dm_mq_kick_requeue_list(dm_get_md_queue(md), 0);
}
EXPORT_SYMBOL(dm_mq_kick_requeue_list);
static void dm_mq_delay_requeue_request(struct request *rq, unsigned long msecs)
{
blk_mq_requeue_request(rq, false);
__dm_mq_kick_requeue_list(rq->q, msecs);
}
static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_requeue)
{
struct mapped_device *md = tio->md;
struct request *rq = tio->orig;
int rw = rq_data_dir(rq);
unsigned long delay_ms = delay_requeue ? 100 : 0;
rq_end_stats(md, rq);
if (tio->clone) {
blk_rq_unprep_clone(tio->clone);
tio->ti->type->release_clone_rq(tio->clone);
}
if (!rq->q->mq_ops)
dm_old_requeue_request(rq, delay_ms);
else
dm_mq_delay_requeue_request(rq, delay_ms);
rq_completed(md, rw, false);
}
static void dm_done(struct request *clone, blk_status_t error, bool mapped)
{
int r = DM_ENDIO_DONE;
struct dm_rq_target_io *tio = clone->end_io_data;
dm_request_endio_fn rq_end_io = NULL;
if (tio->ti) {
rq_end_io = tio->ti->type->rq_end_io;
if (mapped && rq_end_io)
r = rq_end_io(tio->ti, clone, error, &tio->info);
}
if (unlikely(error == BLK_STS_TARGET)) {
if (req_op(clone) == REQ_OP_WRITE_SAME &&
!clone->q->limits.max_write_same_sectors)
disable_write_same(tio->md);
if (req_op(clone) == REQ_OP_WRITE_ZEROES &&
!clone->q->limits.max_write_zeroes_sectors)
disable_write_zeroes(tio->md);
}
switch (r) {
case DM_ENDIO_DONE:
/* The target wants to complete the I/O */
dm_end_request(clone, error);
break;
case DM_ENDIO_INCOMPLETE:
/* The target will handle the I/O */
return;
case DM_ENDIO_REQUEUE:
/* The target wants to requeue the I/O */
dm_requeue_original_request(tio, false);
break;
default:
DMWARN("unimplemented target endio return value: %d", r);
BUG();
}
}
/*
* Request completion handler for request-based dm
*/
static void dm_softirq_done(struct request *rq)
{
bool mapped = true;
struct dm_rq_target_io *tio = tio_from_request(rq);
struct request *clone = tio->clone;
int rw;
if (!clone) {
struct mapped_device *md = tio->md;
rq_end_stats(md, rq);
rw = rq_data_dir(rq);
if (!rq->q->mq_ops)
blk_end_request_all(rq, tio->error);
else
blk_mq_end_request(rq, tio->error);
rq_completed(md, rw, false);
return;
}
if (rq->rq_flags & RQF_FAILED)
mapped = false;
dm_done(clone, tio->error, mapped);
}
/*
* Complete the clone and the original request with the error status
* through softirq context.
*/
static void dm_complete_request(struct request *rq, blk_status_t error)
{
struct dm_rq_target_io *tio = tio_from_request(rq);
tio->error = error;
if (!rq->q->mq_ops)
blk_complete_request(rq);
else
blk_mq_complete_request(rq);
}
/*
* Complete the not-mapped clone and the original request with the error status
* through softirq context.
* Target's rq_end_io() function isn't called.
* This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
*/
static void dm_kill_unmapped_request(struct request *rq, blk_status_t error)
{
rq->rq_flags |= RQF_FAILED;
dm_complete_request(rq, error);
}
/*
* Called with the clone's queue lock held (in the case of .request_fn)
*/
static void end_clone_request(struct request *clone, blk_status_t error)
{
struct dm_rq_target_io *tio = clone->end_io_data;
/*
* Actual request completion is done in a softirq context which doesn't
* hold the clone's queue lock. Otherwise, deadlock could occur because:
* - another request may be submitted by the upper level driver
* of the stacking during the completion
* - the submission which requires queue lock may be done
* against this clone's queue
*/
dm_complete_request(tio->orig, error);
}
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
static blk_status_t dm_dispatch_clone_request(struct request *clone, struct request *rq)
{
blk_status_t r;
if (blk_queue_io_stat(clone->q))
clone->rq_flags |= RQF_IO_STAT;
clone->start_time = jiffies;
r = blk_insert_cloned_request(clone->q, clone);
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
if (r != BLK_STS_OK && r != BLK_STS_RESOURCE)
/* must complete clone in terms of original request */
dm_complete_request(rq, r);
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
return r;
}
static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
void *data)
{
struct dm_rq_target_io *tio = data;
struct dm_rq_clone_bio_info *info =
container_of(bio, struct dm_rq_clone_bio_info, clone);
info->orig = bio_orig;
info->tio = tio;
bio->bi_end_io = end_clone_bio;
return 0;
}
static int setup_clone(struct request *clone, struct request *rq,
struct dm_rq_target_io *tio, gfp_t gfp_mask)
{
int r;
r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
dm_rq_bio_constructor, tio);
if (r)
return r;
clone->end_io = end_clone_request;
clone->end_io_data = tio;
tio->clone = clone;
return 0;
}
static void map_tio_request(struct kthread_work *work);
static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
struct mapped_device *md)
{
tio->md = md;
tio->ti = NULL;
tio->clone = NULL;
tio->orig = rq;
tio->error = 0;
tio->completed = 0;
/*
* Avoid initializing info for blk-mq; it passes
* target-specific data through info.ptr
* (see: dm_mq_init_request)
*/
if (!md->init_tio_pdu)
memset(&tio->info, 0, sizeof(tio->info));
if (md->kworker_task)
kthread: kthread worker API cleanup A good practice is to prefix the names of functions by the name of the subsystem. The kthread worker API is a mix of classic kthreads and workqueues. Each worker has a dedicated kthread. It runs a generic function that process queued works. It is implemented as part of the kthread subsystem. This patch renames the existing kthread worker API to use the corresponding name from the workqueues API prefixed by kthread_: __init_kthread_worker() -> __kthread_init_worker() init_kthread_worker() -> kthread_init_worker() init_kthread_work() -> kthread_init_work() insert_kthread_work() -> kthread_insert_work() queue_kthread_work() -> kthread_queue_work() flush_kthread_work() -> kthread_flush_work() flush_kthread_worker() -> kthread_flush_worker() Note that the names of DEFINE_KTHREAD_WORK*() macros stay as they are. It is common that the "DEFINE_" prefix has precedence over the subsystem names. Note that INIT() macros and init() functions use different naming scheme. There is no good solution. There are several reasons for this solution: + "init" in the function names stands for the verb "initialize" aka "initialize worker". While "INIT" in the macro names stands for the noun "INITIALIZER" aka "worker initializer". + INIT() macros are used only in DEFINE() macros + init() functions are used close to the other kthread() functions. It looks much better if all the functions use the same scheme. + There will be also kthread_destroy_worker() that will be used close to kthread_cancel_work(). It is related to the init() function. Again it looks better if all functions use the same naming scheme. + there are several precedents for such init() function names, e.g. amd_iommu_init_device(), free_area_init_node(), jump_label_init_type(), regmap_init_mmio_clk(), + It is not an argument but it was inconsistent even before. [arnd@arndb.de: fix linux-next merge conflict] Link: http://lkml.kernel.org/r/20160908135724.1311726-1-arnd@arndb.de Link: http://lkml.kernel.org/r/1470754545-17632-3-git-send-email-pmladek@suse.com Suggested-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Petr Mladek <pmladek@suse.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Tejun Heo <tj@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: Josh Triplett <josh@joshtriplett.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Jiri Kosina <jkosina@suse.cz> Cc: Borislav Petkov <bp@suse.de> Cc: Michal Hocko <mhocko@suse.cz> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:55:20 +08:00
kthread_init_work(&tio->work, map_tio_request);
}
/*
* Returns:
* DM_MAPIO_* : the request has been processed as indicated
* DM_MAPIO_REQUEUE : the original request needs to be immediately requeued
* < 0 : the request was completed due to failure
*/
static int map_request(struct dm_rq_target_io *tio)
{
int r;
struct dm_target *ti = tio->ti;
struct mapped_device *md = tio->md;
struct request *rq = tio->orig;
struct request *clone = NULL;
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
blk_status_t ret;
r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
check_again:
switch (r) {
case DM_MAPIO_SUBMITTED:
/* The target has taken the I/O to submit by itself later */
break;
case DM_MAPIO_REMAPPED:
if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
/* -ENOMEM */
ti->type->release_clone_rq(clone);
return DM_MAPIO_REQUEUE;
}
/* The target has remapped the I/O so dispatch it */
trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
blk_rq_pos(rq));
blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback blk_insert_cloned_request() is called in the fast path of a dm-rq driver (e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses blk_mq_request_bypass_insert() to directly append the request to the blk-mq hctx->dispatch_list of the underlying queue. 1) This way isn't efficient enough because the hctx spinlock is always used. 2) With blk_insert_cloned_request(), we completely bypass underlying queue's elevator and depend on the upper-level dm-rq driver's elevator to schedule IO. But dm-rq currently can't get the underlying queue's dispatch feedback at all. Without knowing whether a request was issued or not (e.g. due to underlying queue being busy) the dm-rq elevator will not be able to provide effective IO merging (as a side-effect of dm-rq currently blindly destaging a request from its elevator only to requeue it after a delay, which kills any opportunity for merging). This obviously causes very bad sequential IO performance. Fix this by updating blk_insert_cloned_request() to use blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a request to be issued directly to the underlying queue and returns the dispatch feedback (blk_status_t). If blk_mq_request_direct_issue() returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE to _not_ destage the request. Whereby preserving the opportunity to merge IO. With this, request-based DM's blk-mq sequential IO performance is vastly improved (as much as 3X in mpath/virtio-scsi testing). Signed-off-by: Ming Lei <ming.lei@redhat.com> [blk-mq.c changes heavily influenced by Ming Lei's initial solution, but they were refactored to make them less fragile and easier to read/review] Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18 00:25:57 +08:00
ret = dm_dispatch_clone_request(clone, rq);
if (ret == BLK_STS_RESOURCE) {
blk_rq_unprep_clone(clone);
tio->ti->type->release_clone_rq(clone);
tio->clone = NULL;
if (!rq->q->mq_ops)
r = DM_MAPIO_DELAY_REQUEUE;
else
r = DM_MAPIO_REQUEUE;
goto check_again;
}
break;
case DM_MAPIO_REQUEUE:
/* The target wants to requeue the I/O */
break;
case DM_MAPIO_DELAY_REQUEUE:
/* The target wants to requeue the I/O after a delay */
dm_requeue_original_request(tio, true);
break;
case DM_MAPIO_KILL:
/* The target wants to complete the I/O */
dm_kill_unmapped_request(rq, BLK_STS_IOERR);
break;
default:
DMWARN("unimplemented target map return value: %d", r);
BUG();
}
return r;
}
static void dm_start_request(struct mapped_device *md, struct request *orig)
{
if (!orig->q->mq_ops)
blk_start_request(orig);
else
blk_mq_start_request(orig);
atomic_inc(&md->pending[rq_data_dir(orig)]);
if (md->seq_rq_merge_deadline_usecs) {
md->last_rq_pos = rq_end_sector(orig);
md->last_rq_rw = rq_data_dir(orig);
md->last_rq_start_time = ktime_get();
}
if (unlikely(dm_stats_used(&md->stats))) {
struct dm_rq_target_io *tio = tio_from_request(orig);
tio->duration_jiffies = jiffies;
tio->n_sectors = blk_rq_sectors(orig);
dm_stats_account_io(&md->stats, rq_data_dir(orig),
blk_rq_pos(orig), tio->n_sectors, false, 0,
&tio->stats_aux);
}
/*
* Hold the md reference here for the in-flight I/O.
* We can't rely on the reference count by device opener,
* because the device may be closed during the request completion
* when all bios are completed.
* See the comment in rq_completed() too.
*/
dm_get(md);
}
static int __dm_rq_init_rq(struct mapped_device *md, struct request *rq)
{
struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
/*
* Must initialize md member of tio, otherwise it won't
* be available in dm_mq_queue_rq.
*/
tio->md = md;
if (md->init_tio_pdu) {
/* target-specific per-io data is immediately after the tio */
tio->info.ptr = tio + 1;
}
return 0;
}
static int dm_rq_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp)
{
return __dm_rq_init_rq(q->rq_alloc_data, rq);
}
static void map_tio_request(struct kthread_work *work)
{
struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
if (map_request(tio) == DM_MAPIO_REQUEUE)
dm_requeue_original_request(tio, false);
}
ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
{
return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
}
#define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
const char *buf, size_t count)
{
unsigned deadline;
if (dm_get_md_type(md) != DM_TYPE_REQUEST_BASED)
return count;
if (kstrtouint(buf, 10, &deadline))
return -EINVAL;
if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
md->seq_rq_merge_deadline_usecs = deadline;
return count;
}
static bool dm_old_request_peeked_before_merge_deadline(struct mapped_device *md)
{
ktime_t kt_deadline;
if (!md->seq_rq_merge_deadline_usecs)
return false;
kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
return !ktime_after(ktime_get(), kt_deadline);
}
/*
* q->request_fn for old request-based dm.
* Called with the queue lock held.
*/
static void dm_old_request_fn(struct request_queue *q)
{
struct mapped_device *md = q->queuedata;
struct dm_target *ti = md->immutable_target;
struct request *rq;
struct dm_rq_target_io *tio;
sector_t pos = 0;
if (unlikely(!ti)) {
int srcu_idx;
struct dm_table *map = dm_get_live_table(md, &srcu_idx);
if (unlikely(!map)) {
dm_put_live_table(md, srcu_idx);
return;
}
ti = dm_table_find_target(map, pos);
dm_put_live_table(md, srcu_idx);
}
/*
* For suspend, check blk_queue_stopped() and increment
* ->pending within a single queue_lock not to increment the
* number of in-flight I/Os after the queue is stopped in
* dm_suspend().
*/
while (!blk_queue_stopped(q)) {
rq = blk_peek_request(q);
if (!rq)
return;
/* always use block 0 to find the target for flushes for now */
pos = 0;
if (req_op(rq) != REQ_OP_FLUSH)
pos = blk_rq_pos(rq);
if ((dm_old_request_peeked_before_merge_deadline(md) &&
md_in_flight(md) && rq->bio && !bio_multiple_segments(rq->bio) &&
md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
(ti->type->busy && ti->type->busy(ti))) {
blk_delay_queue(q, 10);
return;
}
dm_start_request(md, rq);
tio = tio_from_request(rq);
init_tio(tio, rq, md);
/* Establish tio->ti before queuing work (map_tio_request) */
tio->ti = ti;
kthread: kthread worker API cleanup A good practice is to prefix the names of functions by the name of the subsystem. The kthread worker API is a mix of classic kthreads and workqueues. Each worker has a dedicated kthread. It runs a generic function that process queued works. It is implemented as part of the kthread subsystem. This patch renames the existing kthread worker API to use the corresponding name from the workqueues API prefixed by kthread_: __init_kthread_worker() -> __kthread_init_worker() init_kthread_worker() -> kthread_init_worker() init_kthread_work() -> kthread_init_work() insert_kthread_work() -> kthread_insert_work() queue_kthread_work() -> kthread_queue_work() flush_kthread_work() -> kthread_flush_work() flush_kthread_worker() -> kthread_flush_worker() Note that the names of DEFINE_KTHREAD_WORK*() macros stay as they are. It is common that the "DEFINE_" prefix has precedence over the subsystem names. Note that INIT() macros and init() functions use different naming scheme. There is no good solution. There are several reasons for this solution: + "init" in the function names stands for the verb "initialize" aka "initialize worker". While "INIT" in the macro names stands for the noun "INITIALIZER" aka "worker initializer". + INIT() macros are used only in DEFINE() macros + init() functions are used close to the other kthread() functions. It looks much better if all the functions use the same scheme. + There will be also kthread_destroy_worker() that will be used close to kthread_cancel_work(). It is related to the init() function. Again it looks better if all functions use the same naming scheme. + there are several precedents for such init() function names, e.g. amd_iommu_init_device(), free_area_init_node(), jump_label_init_type(), regmap_init_mmio_clk(), + It is not an argument but it was inconsistent even before. [arnd@arndb.de: fix linux-next merge conflict] Link: http://lkml.kernel.org/r/20160908135724.1311726-1-arnd@arndb.de Link: http://lkml.kernel.org/r/1470754545-17632-3-git-send-email-pmladek@suse.com Suggested-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Petr Mladek <pmladek@suse.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Tejun Heo <tj@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: Josh Triplett <josh@joshtriplett.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Jiri Kosina <jkosina@suse.cz> Cc: Borislav Petkov <bp@suse.de> Cc: Michal Hocko <mhocko@suse.cz> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:55:20 +08:00
kthread_queue_work(&md->kworker, &tio->work);
BUG_ON(!irqs_disabled());
}
}
/*
* Fully initialize a .request_fn request-based queue.
*/
int dm_old_init_request_queue(struct mapped_device *md, struct dm_table *t)
{
struct dm_target *immutable_tgt;
/* Fully initialize the queue */
md->queue->cmd_size = sizeof(struct dm_rq_target_io);
md->queue->rq_alloc_data = md;
md->queue->request_fn = dm_old_request_fn;
md->queue->init_rq_fn = dm_rq_init_rq;
immutable_tgt = dm_table_get_immutable_target(t);
if (immutable_tgt && immutable_tgt->per_io_data_size) {
/* any target-specific per-io data is immediately after the tio */
md->queue->cmd_size += immutable_tgt->per_io_data_size;
md->init_tio_pdu = true;
}
if (blk_init_allocated_queue(md->queue) < 0)
return -EINVAL;
/* disable dm_old_request_fn's merge heuristic by default */
md->seq_rq_merge_deadline_usecs = 0;
dm_init_normal_md_queue(md);
blk_queue_softirq_done(md->queue, dm_softirq_done);
/* Initialize the request-based DM worker thread */
kthread: kthread worker API cleanup A good practice is to prefix the names of functions by the name of the subsystem. The kthread worker API is a mix of classic kthreads and workqueues. Each worker has a dedicated kthread. It runs a generic function that process queued works. It is implemented as part of the kthread subsystem. This patch renames the existing kthread worker API to use the corresponding name from the workqueues API prefixed by kthread_: __init_kthread_worker() -> __kthread_init_worker() init_kthread_worker() -> kthread_init_worker() init_kthread_work() -> kthread_init_work() insert_kthread_work() -> kthread_insert_work() queue_kthread_work() -> kthread_queue_work() flush_kthread_work() -> kthread_flush_work() flush_kthread_worker() -> kthread_flush_worker() Note that the names of DEFINE_KTHREAD_WORK*() macros stay as they are. It is common that the "DEFINE_" prefix has precedence over the subsystem names. Note that INIT() macros and init() functions use different naming scheme. There is no good solution. There are several reasons for this solution: + "init" in the function names stands for the verb "initialize" aka "initialize worker". While "INIT" in the macro names stands for the noun "INITIALIZER" aka "worker initializer". + INIT() macros are used only in DEFINE() macros + init() functions are used close to the other kthread() functions. It looks much better if all the functions use the same scheme. + There will be also kthread_destroy_worker() that will be used close to kthread_cancel_work(). It is related to the init() function. Again it looks better if all functions use the same naming scheme. + there are several precedents for such init() function names, e.g. amd_iommu_init_device(), free_area_init_node(), jump_label_init_type(), regmap_init_mmio_clk(), + It is not an argument but it was inconsistent even before. [arnd@arndb.de: fix linux-next merge conflict] Link: http://lkml.kernel.org/r/20160908135724.1311726-1-arnd@arndb.de Link: http://lkml.kernel.org/r/1470754545-17632-3-git-send-email-pmladek@suse.com Suggested-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Petr Mladek <pmladek@suse.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Tejun Heo <tj@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: Josh Triplett <josh@joshtriplett.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Jiri Kosina <jkosina@suse.cz> Cc: Borislav Petkov <bp@suse.de> Cc: Michal Hocko <mhocko@suse.cz> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:55:20 +08:00
kthread_init_worker(&md->kworker);
md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
"kdmwork-%s", dm_device_name(md));
if (IS_ERR(md->kworker_task)) {
int error = PTR_ERR(md->kworker_task);
md->kworker_task = NULL;
return error;
}
return 0;
}
static int dm_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
unsigned int hctx_idx, unsigned int numa_node)
{
return __dm_rq_init_rq(set->driver_data, rq);
}
static blk_status_t dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
const struct blk_mq_queue_data *bd)
{
struct request *rq = bd->rq;
struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
struct mapped_device *md = tio->md;
struct dm_target *ti = md->immutable_target;
if (unlikely(!ti)) {
int srcu_idx;
struct dm_table *map = dm_get_live_table(md, &srcu_idx);
ti = dm_table_find_target(map, 0);
dm_put_live_table(md, srcu_idx);
}
if (ti->type->busy && ti->type->busy(ti))
return BLK_STS_RESOURCE;
dm_start_request(md, rq);
/* Init tio using md established in .init_request */
init_tio(tio, rq, md);
/*
* Establish tio->ti before calling map_request().
*/
tio->ti = ti;
/* Direct call is fine since .queue_rq allows allocations */
if (map_request(tio) == DM_MAPIO_REQUEUE) {
/* Undo dm_start_request() before requeuing */
rq_end_stats(md, rq);
rq_completed(md, rq_data_dir(rq), false);
blk_mq_delay_run_hw_queue(hctx, 100/*ms*/);
return BLK_STS_RESOURCE;
}
return BLK_STS_OK;
}
static const struct blk_mq_ops dm_mq_ops = {
.queue_rq = dm_mq_queue_rq,
.complete = dm_softirq_done,
.init_request = dm_mq_init_request,
};
int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t)
{
struct request_queue *q;
struct dm_target *immutable_tgt;
int err;
if (!dm_table_all_blk_mq_devices(t)) {
DMERR("request-based dm-mq may only be stacked on blk-mq device(s)");
return -EINVAL;
}
md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id);
if (!md->tag_set)
return -ENOMEM;
md->tag_set->ops = &dm_mq_ops;
md->tag_set->queue_depth = dm_get_blk_mq_queue_depth();
md->tag_set->numa_node = md->numa_node_id;
md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues();
md->tag_set->driver_data = md;
md->tag_set->cmd_size = sizeof(struct dm_rq_target_io);
immutable_tgt = dm_table_get_immutable_target(t);
if (immutable_tgt && immutable_tgt->per_io_data_size) {
/* any target-specific per-io data is immediately after the tio */
md->tag_set->cmd_size += immutable_tgt->per_io_data_size;
md->init_tio_pdu = true;
}
err = blk_mq_alloc_tag_set(md->tag_set);
if (err)
goto out_kfree_tag_set;
q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
if (IS_ERR(q)) {
err = PTR_ERR(q);
goto out_tag_set;
}
dm_init_md_queue(md);
return 0;
out_tag_set:
blk_mq_free_tag_set(md->tag_set);
out_kfree_tag_set:
kfree(md->tag_set);
return err;
}
void dm_mq_cleanup_mapped_device(struct mapped_device *md)
{
if (md->tag_set) {
blk_mq_free_tag_set(md->tag_set);
kfree(md->tag_set);
}
}
module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices");
module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices");