Move scheduler tags (sched_tags) allocation and deallocation outside
both the ->elevator_lock and ->freeze_lock when updating nr_hw_queues.
This change breaks the dependency chain from the percpu allocator lock
to the elevator lock, helping to prevent potential deadlocks, as
observed in the reported lockdep splat[1].
This commit introduces batch allocation and deallocation helpers for
sched_tags, which are now used from within __blk_mq_update_nr_hw_queues
routine while iterating through the tagset.
With this change, all sched_tags memory management is handled entirely
outside the ->elevator_lock and the ->freeze_lock context, thereby
eliminating the lock dependency that could otherwise manifest during
nr_hw_queues updates.
[1] https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reported-by: Stefan Haberland <sth@linux.ibm.com>
Closes: https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250730074614.2537382-4-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Recent lockdep reports [1] have revealed a potential deadlock caused by a
lock dependency between the percpu allocator lock and the elevator lock.
This issue can be avoided by ensuring that the allocation and release of
scheduler tags (sched_tags) are performed outside the elevator lock.
Furthermore, the queue does not need to be remain frozen during these
operations.
To address this, move all sched_tags allocations and deallocations outside
of both the ->elevator_lock and the ->freeze_lock. Since the lifetime of
the elevator queue and its associated sched_tags is closely tied, the
allocated sched_tags are now stored in the elevator queue structure. Then,
during the actual elevator switch (which runs under ->freeze_lock and
->elevator_lock), the pre-allocated sched_tags are assigned to the
appropriate q->hctx. Once the elevator switch is complete and the locks
are released, the old elevator queue and its associated sched_tags are
freed.
This commit specifically addresses the allocation/deallocation of sched_
tags during elevator switching. Note that sched_tags may also be allocated
in other contexts, such as during nr_hw_queues updates. Supporting that
use case will require batch allocation/deallocation, which will be handled
in a follow-up patch.
This restructuring ensures that sched_tags memory management occurs
entirely outside of the ->elevator_lock and ->freeze_lock context,
eliminating the lock dependency problem seen during scheduler updates.
[1] https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reported-by: Stefan Haberland <sth@linux.ibm.com>
Closes: https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250730074614.2537382-3-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for allocating sched_tags before freezing the request
queue and acquiring ->elevator_lock, move the elevator queue allocation
logic from the elevator ops ->init_sched callback into blk_mq_init_sched.
As elevator_alloc is now only invoked from block layer core, we don't
need to export it, so unexport elevator_alloc function.
This refactoring provides a centralized location for elevator queue
initialization, which makes it easier to store pre-allocated sched_tags
in the struct elevator_queue during later changes.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250730074614.2537382-2-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Prepare for moving elv_register[unregister]_queue out of elevator_lock
& queue freezing, so we may have to call elv_unregister_queue() after
elevator ->exit() is called, then there is small window for user to
call into ->show()/store(), and user-after-free can be caused.
Fail to show/store elevator sysfs attribute if elevator is dying by
adding one new flag of ELEVATOR_FLAG_DYNG, which is protected by
elevator ->sysfs_lock.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://lore.kernel.org/r/20250505141805.2751237-20-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
sched debugfs shares same lifetime with scheduler's kobject, and same
lock(elevator lock), so move sched debugfs register/unregister into
elevator_register_queue() and elevator_unregister_queue().
Then we needn't blk_mq_debugfs_register() for us to register sched
debugfs any more.
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add blk_mq_sched_reg_debugfs()/blk_mq_sched_unreg_debugfs() to clean up
sched init/exit code a bit.
Register & unregister debugfs for sched & sched_hctx order is changed a
bit, but it is safe because sched & sched_hctx is guaranteed to be ready
when exporting via debugfs.
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The 'nr_budgets' argument of blk_mq_dispatch_rq_list() is either the
number of elements in the 'list' argument or zero. Instead of passing
the number of list elements to blk_mq_dispatch_rq_list(), pass a boolean
argument that indicates whether or not blk_mq_dispatch_rq_list() should
request the block driver for a budget for each request in 'list'.
Remove the code for counting list elements from blk_mq_dispatch_rq_list()
callers where possible. Remove the code that decrements nr_budgets from
blk_mq_dispatch_rq_list() because it is superfluous. Each request that
is processed by blk_mq_dispatch_rq_list() is in one of these two states
if 'get_budget' is false:
* Either the request is on 'list' and the budget for the request has to
be released from the error path.
* Or the request is not on 'list' and q->mq_ops->queue_rq() has already
released the budget (ret != BLK_STS_OK) or q->mq_ops->queue_rq() will
release the budget asynchronously (ret == BLK_STS_OK).
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: John Garry <john.g.garry@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20250415205134.3650042-1-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
BLK_MQ_F_SHOULD_MERGE is set for all tag_sets except those that purely
process passthrough commands (bsg-lib, ufs tmf, various nvme admin
queues) and thus don't even check the flag. Remove it to simplify the
driver interface.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241219060214.1928848-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Nobody uses the debugfs hctx 'run' attribute. Hence remove this
attribute and also the code that updates the corresponding member
variable.
Suggested-by: Jens Axboe <axboe@kernel.dk>
Cc: Gabriel Ryan <gabe@cs.columbia.edu>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240117203609.4122520-1-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__blk_mq_sched_dispatch_requests currently has duplicated logic
for the cases where requests are on the hctx dispatch list or not.
Merge the two with a new need_dispatch variable and remove a few
pointless local variables.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413060651.694656-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_sched_insert_request is the main request insert helper and not
directly I/O scheduler related. Move blk_mq_sched_insert_request to
blk-mq.c, rename it to blk_mq_insert_request and mark it static.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413064057.707578-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_dispatch_plug_list is the only caller of
blk_mq_sched_insert_requests, and it makes sense to just fold it there
as blk_mq_sched_insert_requests isn't specific to I/O schedulers despite
the name.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413064057.707578-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move all logic related to the direct insert (including the call to
blk_mq_run_hw_queue) into blk_mq_insert_requests to streamline the code
flow up a bit, and to allow marking blk_mq_try_issue_list_directly
static.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413064057.707578-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-mq.h needs various definitions from <linux/blk-mq.h>,
include it there instead of relying on the source files to include
both.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413064057.707578-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk-mq-tag.h is always included by blk-mq.h, and causes recursive
inclusion hell with further changes. Just merge it into blk-mq.h
instead.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20230413064057.707578-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 88022d7201 ("blk-mq: don't handle failure in .get_budget")
remove BLK_STS_RESOURCE return value and we only check if we can get
the budget from .get_budget() now.
Correct stale comment that ".get_budget() returns BLK_STS_NO_RESOURCE"
to ".get_budget() fails to get the budget".
Fixes: 88022d7201 ("blk-mq: don't handle failure in .get_budget")
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 97889f9ac2 ("blk-mq: remove synchronize_rcu() from
blk_mq_del_queue_tag_set()") remove handle of TAG_SHARED in restart,
then shared_hctx_restart counted for how many hardware queues are marked
for restart is removed too.
Remove the stale comment that we still count hardware queues need restart.
Fixes: 97889f9ac2 ("blk-mq: remove synchronize_rcu() from blk_mq_del_queue_tag_set()")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split an elevator_disable helper from elevator_switch for the case where
we want to switch to no scheduler at all. This includes removing the
pointless elevator_switch_mq helper and removing the switch to no
schedule logic from blk_mq_init_sched.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221030100714.876891-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The current reference management logic of io scheduler modules contains
refcnt problems. For example, blk_mq_init_sched may fail before or after
the calling of e->ops.init_sched. If it fails before the calling, it does
nothing to the reference to the io scheduler module. But if it fails after
the calling, it releases the reference by calling kobject_put(&eq->kobj).
As the callers of blk_mq_init_sched can't know exactly where the failure
happens, they can't handle the reference to the io scheduler module
properly: releasing the reference on failure results in double-release if
blk_mq_init_sched has released it, and not releasing the reference results
in ghost reference if blk_mq_init_sched did not release it either.
The same problem also exists in io schedulers' init_sched implementations.
We can address the problem by adding releasing statements to the error
handling procedures of blk_mq_init_sched and init_sched implementations.
But that is counterintuitive and requires modifications to existing io
schedulers.
Instead, We make elevator_alloc get the io scheduler module references
that will be released by elevator_release. And then, we match each
elevator_get with an elevator_put. Therefore, each reference to an io
scheduler module explicitly has its own getter and releaser, and we no
longer need to worry about the refcnt problems.
The bugs and the patch can be validated with tools here:
https://github.com/nickyc975/linux_elv_refcnt_bug.git
[hch: split out a few bits into separate patches, use a non-try
module_get in elevator_alloc]
Signed-off-by: Jinlong Chen <nickyc975@zju.edu.cn>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221020064819.1469928-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Various places like I/O schedulers or the QOS infrastructure try to
register debugfs files on demans, which can race with creating and
removing the main queue debugfs directory. Use the existing
debugfs_mutex to serialize all debugfs operations that rely on
q->debugfs_dir or the directories hanging off it.
To make the teardown code a little simpler declare all debugfs dentry
pointers and not just the main one uncoditionally in blkdev.h.
Move debugfs_mutex next to the dentries that it protects and document
what it is used for.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220614074827.458955-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
q->elevator is referred in blk_mq_has_sqsched() without any protection,
no .q_usage_counter is held, no queue srcu and rcu read lock is held,
so potential use-after-free may be triggered.
Fix the issue by adding one queue flag for checking if the elevator
uses single queue style dispatch. Meantime the elevator feature flag
of ELEVATOR_F_MQ_AWARE isn't needed any more.
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220616014401.817001-3-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When IO requests are made continuously and the target block device
handles requests faster than request arrival, the request dispatch loop
keeps on repeating to dispatch the arriving requests very long time,
more than a minute. Since the loop runs as a workqueue worker task, the
very long loop duration triggers workqueue watchdog timeout and BUG [1].
To avoid the very long loop duration, break the loop periodically. When
opportunity to dispatch requests still exists, check need_resched(). If
need_resched() returns true, the dispatch loop already consumed its time
slice, then reschedule the dispatch work and break the loop. With heavy
IO load, need_resched() does not return true for 20~30 seconds. To cover
such case, check time spent in the dispatch loop with jiffies. If more
than 1 second is spent, reschedule the dispatch work and break the loop.
[1]
[ 609.691437] BUG: workqueue lockup - pool cpus=10 node=1 flags=0x0 nice=-20 stuck for 35s!
[ 609.701820] Showing busy workqueues and worker pools:
[ 609.707915] workqueue events: flags=0x0
[ 609.712615] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 refcnt=2
[ 609.712626] pending: drm_fb_helper_damage_work [drm_kms_helper]
[ 609.712687] workqueue events_freezable: flags=0x4
[ 609.732943] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 refcnt=2
[ 609.732952] pending: pci_pme_list_scan
[ 609.732968] workqueue events_power_efficient: flags=0x80
[ 609.751947] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 refcnt=2
[ 609.751955] pending: neigh_managed_work
[ 609.752018] workqueue kblockd: flags=0x18
[ 609.769480] pwq 21: cpus=10 node=1 flags=0x0 nice=-20 active=3/256 refcnt=4
[ 609.769488] in-flight: 1020:blk_mq_run_work_fn
[ 609.769498] pending: blk_mq_timeout_work, blk_mq_run_work_fn
[ 609.769744] pool 21: cpus=10 node=1 flags=0x0 nice=-20 hung=35s workers=2 idle: 67
[ 639.899730] BUG: workqueue lockup - pool cpus=10 node=1 flags=0x0 nice=-20 stuck for 66s!
[ 639.909513] Showing busy workqueues and worker pools:
[ 639.915404] workqueue events: flags=0x0
[ 639.920197] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 refcnt=2
[ 639.920215] pending: drm_fb_helper_damage_work [drm_kms_helper]
[ 639.920365] workqueue kblockd: flags=0x18
[ 639.939932] pwq 21: cpus=10 node=1 flags=0x0 nice=-20 active=3/256 refcnt=4
[ 639.939942] in-flight: 1020:blk_mq_run_work_fn
[ 639.939955] pending: blk_mq_timeout_work, blk_mq_run_work_fn
[ 639.940212] pool 21: cpus=10 node=1 flags=0x0 nice=-20 hung=66s workers=2 idle: 67
Fixes: 6e6fcbc27e ("blk-mq: support batching dispatch in case of io")
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Cc: stable@vger.kernel.org # v5.10+
Link: https://lore.kernel.org/linux-block/20220310091649.zypaem5lkyfadymg@shindev/
Link: https://lore.kernel.org/r/20220318022641.133484-1-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It is inevitable to cause use-after-free on q->queue_hw_ctx between
queue_for_each_hw_ctx() and blk_mq_update_nr_hw_queues(). And converting
to xarray can fix the uaf, meantime code gets cleaner.
Prepare for converting q->queue_hctx_ctx into xarray, one thing is that
xa_for_each() can only accept 'unsigned long' as index, so changes type
of hctx index of queue_for_each_hw_ctx() into 'unsigned long'.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220308073219.91173-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It isn't necessary to call blk_mq_run_dispatch_ops() once for issuing
single request directly, and enough to do it one time when issuing from
whole list.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20211203131534.3668411-5-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move blk_mq_sched_assign_ioc so that many interfaces from the file can
be marked static. Rename the function to ioc_find_get_icq as well and
return the icq to simplify the interface.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20211126115817.2087431-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This reverts commit 4896c4e64ba5d5d5acdbcf68c5910dd4f6d8fa62.
The helper is not needed any more.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20211126115817.2087431-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently we lookup ICQ only after the request is allocated. However BFQ
will want to decide how many scheduler tags it allows a given bfq queue
(effectively a process) to consume based on cgroup weight. So provide a
function blk_mq_sched_get_icq() so that BFQ can lookup ICQ earlier.
Acked-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20211125133645.27483-1-jack@suse.cz
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The only user of the io_context for IO is BFQ, yet we put the checking
and logic of it into the normal IO path.
Put the creation into blk_mq_sched_assign_ioc(), and have BFQ use that
helper.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_sched_bio_merge is only called from blk-mq.c:blk_attempt_bio_merge(),
which is called when queue usage counter is grabbed already:
1) blk_mq_get_new_requests()
2) blk_mq_get_request()
- cached request in current plug owns one queue usage counter
So don't grab ->q_usage_counter in blk_mq_sched_bio_merge(), and more
importantly this nest way causes hang in blk_mq_freeze_queue_wait().
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20211111085134.345235-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Retain the old logic for the fops based submit, but for our internal
blk_mq_submit_bio(), move the queue entering logic into the core
function itself.
We need to be a bit careful if going into the scheduler, as a scheduler
or queue mappings can arbitrarily change before we have entered the queue.
Have the bio scheduler mapping do that separately, it's a very cheap
operation compared to actually doing merging locking and lookups.
Reviewed-by: Christoph Hellwig <hch@lst.de>
[axboe: update to check merge post submit_bio_checks() doing remap...]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In the if branch, e is checked. In the else branch, ->dispatch_busy is
merely a number and has no effect on !e. We should remove the check of
!e since it is always true.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Link: https://lore.kernel.org/r/20211029202945.3052-1-sakiwit@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We should not reference the queue tagset in blk_mq_sched_tags_teardown()
(see function comment) for the blk-mq flags, so use the passed flags
instead.
This solves a use-after-free, similarly fixed earlier (and since broken
again) in commit f0c1c4d286 ("blk-mq: fix use-after-free in
blk_mq_exit_sched").
Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Tested-by: Anders Roxell <anders.roxell@linaro.org>
Fixes: e155b0c238 ("blk-mq: Use shared tags for shared sbitmap support")
Signed-off-by: John Garry <john.garry@huawei.com>
Link: https://lore.kernel.org/r/1634890340-15432-1-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
These were added as part of early days debugging for blk-mq, and they
are not really useful anymore. Rather than spend cycles updating them,
just get rid of them.
As a bonus, this shrinks the per-cpu software queue size from 256b
to 192b. That's a whole cacheline less.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently we use separate sbitmap pairs and active_queues atomic_t for
shared sbitmap support.
However a full sets of static requests are used per HW queue, which is
quite wasteful, considering that the total number of requests usable at
any given time across all HW queues is limited by the shared sbitmap depth.
As such, it is considerably more memory efficient in the case of shared
sbitmap to allocate a set of static rqs per tag set or request queue, and
not per HW queue.
So replace the sbitmap pairs and active_queues atomic_t with a shared
tags per tagset and request queue, which will hold a set of shared static
rqs.
Since there is now no valid HW queue index to be passed to the blk_mq_ops
.init and .exit_request callbacks, pass an invalid index token. This
changes the semantics of the APIs, such that the callback would need to
validate the HW queue index before using it. Currently no user of shared
sbitmap actually uses the HW queue index (as would be expected).
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1633429419-228500-13-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a function to combine allocating tags and the associated requests,
and factor out common patterns to use this new function.
Some function only call blk_mq_alloc_map_and_rqs() now, but more
functionality will be added later.
Also make blk_mq_alloc_rq_map() and blk_mq_alloc_rqs() static since they
are only used in blk-mq.c, and finally rename some functions for
conciseness and consistency with other function names:
- __blk_mq_alloc_map_and_{request -> rqs}()
- blk_mq_alloc_{map_and_requests -> set_map_and_rqs}()
Suggested-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1633429419-228500-11-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Put the functionality to update the sched shared sbitmap size in a common
function.
Since the same formula is always used to resize, and it can be got from
the request queue argument, so just pass the request queue pointer.
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://lore.kernel.org/r/1633429419-228500-10-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
To be more concise and consistent in naming, rename
blk_mq_sched_free_requests() -> blk_mq_sched_free_rqs().
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1633429419-228500-7-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Function blk_mq_sched_alloc_tags() does same as
__blk_mq_alloc_map_and_request(), so give a similar name to be consistent.
Similarly rename label err_free_tags -> err_free_map_and_rqs.
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://lore.kernel.org/r/1633429419-228500-6-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It is a bit confusing that there is BLKDEV_MAX_RQ and MAX_SCHED_RQ, as
the name BLKDEV_MAX_RQ would imply the max requests always, which it is
not.
Rename to BLKDEV_MAX_RQ to BLKDEV_DEFAULT_RQ, matching its usage - that being
the default number of requests assigned when allocating a request queue.
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://lore.kernel.org/r/1633429419-228500-3-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If the blk_mq_sched_alloc_tags() -> blk_mq_alloc_rqs() call fails, then we
call blk_mq_sched_free_tags() -> blk_mq_free_rqs().
It is incorrect to do so, as any rqs would have already been freed in the
blk_mq_alloc_rqs() call.
Fix by calling blk_mq_free_rq_map() only directly.
Fixes: 6917ff0b5b ("blk-mq-sched: refactor scheduler initialization")
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1627378373-148090-1-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Lockdep complains about lock inversion between ioc->lock and bfqd->lock:
bfqd -> ioc:
put_io_context+0x33/0x90 -> ioc->lock grabbed
blk_mq_free_request+0x51/0x140
blk_put_request+0xe/0x10
blk_attempt_req_merge+0x1d/0x30
elv_attempt_insert_merge+0x56/0xa0
blk_mq_sched_try_insert_merge+0x4b/0x60
bfq_insert_requests+0x9e/0x18c0 -> bfqd->lock grabbed
blk_mq_sched_insert_requests+0xd6/0x2b0
blk_mq_flush_plug_list+0x154/0x280
blk_finish_plug+0x40/0x60
ext4_writepages+0x696/0x1320
do_writepages+0x1c/0x80
__filemap_fdatawrite_range+0xd7/0x120
sync_file_range+0xac/0xf0
ioc->bfqd:
bfq_exit_icq+0xa3/0xe0 -> bfqd->lock grabbed
put_io_context_active+0x78/0xb0 -> ioc->lock grabbed
exit_io_context+0x48/0x50
do_exit+0x7e9/0xdd0
do_group_exit+0x54/0xc0
To avoid this inversion we change blk_mq_sched_try_insert_merge() to not
free the merged request but rather leave that upto the caller similarly
to blk_mq_sched_try_merge(). And in bfq_insert_requests() we make sure
to free all the merged requests after dropping bfqd->lock.
Fixes: aee69d78de ("block, bfq: introduce the BFQ-v0 I/O scheduler as an extra scheduler")
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210623093634.27879-3-jack@suse.cz
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The insert_requests and dispatch_request elevator operations are
mandatory for the correct execution of an elevator, and all implemented
elevators (bfq, kyber and mq-deadline) implement them. As a result,
there is no need to check for these operations before calling them when
a queue has an elevator set. This simplifies the code in
__blk_mq_sched_dispatch_requests() and blk_mq_sched_insert_request().
To avoid out-of-tree elevators to crash the kernel in case of bad
implementation, add a check in elv_register() to verify that these
operations are implemented.
A small, probably not significant, IOPS improvement of 0.1% is observed
with this patch applied (4.117 MIOPS to 4.123 MIOPS, average of 20 fio
runs doing 4K random direct reads with psync and 32 jobs).
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20210618015922.713999-1-damien.lemoal@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
tagset can't be used after blk_cleanup_queue() is returned because
freeing tagset usually follows blk_clenup_queue(). Commit d97e594c51
("blk-mq: Use request queue-wide tags for tagset-wide sbitmap") adds
check on q->tag_set->flags in blk_mq_exit_sched(), and causes
use-after-free.
Fixes it by using hctx->flags.
Reported-by: syzbot+77ba3d171a25c56756ea@syzkaller.appspotmail.com
Fixes: d97e594c51 ("blk-mq: Use request queue-wide tags for tagset-wide sbitmap")
Cc: John Garry <john.garry@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: John Garry <john.garry@huawei.com>
Reviewed-by: John Garry <john.garry@huawei.com>
Link: https://lore.kernel.org/r/20210609063046.122843-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Provided the device driver does not implement dispatch budget accounting
(which only SCSI does) the loop in __blk_mq_do_dispatch_sched() pulls
requests from the IO scheduler as long as it is willing to give out any.
That defeats scheduling heuristics inside the scheduler by creating
false impression that the device can take more IO when it in fact
cannot.
For example with BFQ IO scheduler on top of virtio-blk device setting
blkio cgroup weight has barely any impact on observed throughput of
async IO because __blk_mq_do_dispatch_sched() always sucks out all the
IO queued in BFQ. BFQ first submits IO from higher weight cgroups but
when that is all dispatched, it will give out IO of lower weight cgroups
as well. And then we have to wait for all this IO to be dispatched to
the disk (which means lot of it actually has to complete) before the
IO scheduler is queried again for dispatching more requests. This
completely destroys any service differentiation.
So grab request tag for a request pulled out of the IO scheduler already
in __blk_mq_do_dispatch_sched() and do not pull any more requests if we
cannot get it because we are unlikely to be able to dispatch it. That
way only single request is going to wait in the dispatch list for some
tag to free.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210603104721.6309-1-jack@suse.cz
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The tags used for an IO scheduler are currently per hctx.
As such, when q->nr_hw_queues grows, so does the request queue total IO
scheduler tag depth.
This may cause problems for SCSI MQ HBAs whose total driver depth is
fixed.
Ming and Yanhui report higher CPU usage and lower throughput in scenarios
where the fixed total driver tag depth is appreciably lower than the total
scheduler tag depth:
https://lore.kernel.org/linux-block/440dfcfc-1a2c-bd98-1161-cec4d78c6dfc@huawei.com/T/#mc0d6d4f95275a2743d1c8c3e4dc9ff6c9aa3a76b
In that scenario, since the scheduler tag is got first, much contention
is introduced since a driver tag may not be available after we have got
the sched tag.
Improve this scenario by introducing request queue-wide tags for when
a tagset-wide sbitmap is used. The static sched requests are still
allocated per hctx, as requests are initialised per hctx, as in
blk_mq_init_request(..., hctx_idx, ...) ->
set->ops->init_request(.., hctx_idx, ...).
For simplicity of resizing the request queue sbitmap when updating the
request queue depth, just init at the max possible size, so we don't need
to deal with the possibly with swapping out a new sbitmap for old if
we need to grow.
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/1620907258-30910-3-git-send-email-john.garry@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__blk_mq_sched_bio_merge() gets the ctx and hctx for the current CPU and
passes the hctx to ->bio_merge(). kyber_bio_merge() then gets the ctx
for the current CPU again and uses that to get the corresponding Kyber
context in the passed hctx. However, the thread may be preempted between
the two calls to blk_mq_get_ctx(), and the ctx returned the second time
may no longer correspond to the passed hctx. This "works" accidentally
most of the time, but it can cause us to read garbage if the second ctx
came from an hctx with more ctx's than the first one (i.e., if
ctx->index_hw[hctx->type] > hctx->nr_ctx).
This manifested as this UBSAN array index out of bounds error reported
by Jakub:
UBSAN: array-index-out-of-bounds in ../kernel/locking/qspinlock.c:130:9
index 13106 is out of range for type 'long unsigned int [128]'
Call Trace:
dump_stack+0xa4/0xe5
ubsan_epilogue+0x5/0x40
__ubsan_handle_out_of_bounds.cold.13+0x2a/0x34
queued_spin_lock_slowpath+0x476/0x480
do_raw_spin_lock+0x1c2/0x1d0
kyber_bio_merge+0x112/0x180
blk_mq_submit_bio+0x1f5/0x1100
submit_bio_noacct+0x7b0/0x870
submit_bio+0xc2/0x3a0
btrfs_map_bio+0x4f0/0x9d0
btrfs_submit_data_bio+0x24e/0x310
submit_one_bio+0x7f/0xb0
submit_extent_page+0xc4/0x440
__extent_writepage_io+0x2b8/0x5e0
__extent_writepage+0x28d/0x6e0
extent_write_cache_pages+0x4d7/0x7a0
extent_writepages+0xa2/0x110
do_writepages+0x8f/0x180
__writeback_single_inode+0x99/0x7f0
writeback_sb_inodes+0x34e/0x790
__writeback_inodes_wb+0x9e/0x120
wb_writeback+0x4d2/0x660
wb_workfn+0x64d/0xa10
process_one_work+0x53a/0xa80
worker_thread+0x69/0x5b0
kthread+0x20b/0x240
ret_from_fork+0x1f/0x30
Only Kyber uses the hctx, so fix it by passing the request_queue to
->bio_merge() instead. BFQ and mq-deadline just use that, and Kyber can
map the queues itself to avoid the mismatch.
Fixes: a6088845c2 ("block: kyber: make kyber more friendly with merging")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Link: https://lore.kernel.org/r/c7598605401a48d5cfeadebb678abd10af22b83f.1620691329.git.osandov@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
