kernel/git/stable/linux.git/block/blk-mq.c, branch linux-rolling-lts

blk-mq: reinsert cached request to the list

2026-08-19T16:18:07+00:00

[ Upstream commit b051bb6bf0a231117036aa607cadf55be8e63910 ] A previous commit removed an optimization out of caution for a scenario that turns out not to be real: all the "queue_exit" goto's are safe to reinsert the request into the cached_rq's plug list as they are either from a non-blocking path, or a successful merge that already holds the queue reference. This optimization is most needed for small sequential workloads that successfully merge into larger requests. Fixes: dc278e9bf2b9 ("blk-mq: pop cached request if it is usable") Suggested-by: Ming Lei Suggested-by: Christoph Hellwig Signed-off-by: Keith Busch Reviewed-by: Chaitanya Kulkarni Link: https://patch.msgid.link/20260526153531.2365935-1-kbusch@meta.com Signed-off-by: Jens Axboe Signed-off-by: Simon Liebold Signed-off-by: Sasha Levin

blk-mq: pop cached request if it is usable

2026-08-19T16:18:07+00:00

[ Upstream commit dc278e9bf2b9513a763353e6b9cc21e0f532954e ] When submitting a bio to blk-mq, if the task should sleep after peeking a cached request, but before it pops it, the plug flushes and calls blk_mq_free_plug_rqs, freeing the cached_rqs. This creates a use-after-free bug. Fix this by popping the cached request before any possible blocking calls if it is suitable for use. Popping this request first holds a queue reference, so avoid any serialization races with queue freezes and can safely proceed with dispatching that request to the driver. This potentially increases a timing window from when a driver wants to freeze its queue to when requests stop being dispatched. That scenario is off the fast path though, and drivers need to appropriately handle requests during a freeze request anyway. The downside is the popped element needs to be individually freed when we performed a bio plug merge. The cached request would have had to be freed later anyway, but this patch does it inline with building the plug list instead of after flushing it. Fixes: b0077e269f6c1 ("blk-mq: make sure active queue usage is held for bio_integrity_prep()") Fixes: 7b4f36cd22a65 ("block: ensure we hold a queue reference when using queue limits") Signed-off-by: Keith Busch Link: https://patch.msgid.link/20260521190253.242065-1-kbusch@meta.com Signed-off-by: Jens Axboe Signed-off-by: Simon Liebold Signed-off-by: Sasha Levin

block: recompute nr_integrity_segments in blk_insert_cloned_request

2026-06-01T15:50:58+00:00

[ Upstream commit 2c6e6a18a37b905cb584eb0dda3ae482162a81ca ] blk_insert_cloned_request() already recomputes nr_phys_segments against the bottom queue, because "the queue settings related to segment counting may differ from the original queue." The exact same reasoning applies to integrity segments: a stacked driver's underlying queue can have tighter virt_boundary_mask, seg_boundary_mask, or max_segment_size than the top queue, in which case blk_rq_count_integrity_sg() against the bottom queue produces a different count than the cached rq->nr_integrity_segments inherited from the source request by blk_rq_prep_clone(). When the cached count is lower than the bottom queue's actual count, blk_rq_map_integrity_sg() trips BUG_ON(segments > rq->nr_integrity_segments); on dispatch. The same families of stacked setups that motivated the existing nr_phys_segments recompute -- dm-multipath fanning out to nvme-rdma in particular -- can produce this. Mirror the nr_phys_segments handling: when the request carries integrity, recompute nr_integrity_segments against the bottom queue and reject the request if it exceeds the bottom queue's max_integrity_segments. blk_rq_count_integrity_sg() and queue_max_integrity_segments() are both already available via , which blk-mq.c includes. This closes a latent gap in the stacking contract and brings the integrity-segment accounting in line with the existing phys-segment accounting. Fixes: 76c313f658d2 ("blk-integrity: improved sg segment mapping") Signed-off-by: Casey Chen Reviewed-by: Christoph Hellwig Link: https://patch.msgid.link/20260511212230.27511-1-cachen@purestorage.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

block: break pcpu_alloc_mutex dependency on freeze_lock

2026-04-02T11:22:59+00:00

[ Upstream commit 539d1b47e935e8384977dd7e5cec370c08b7a644 ] While nr_hw_update allocates tagset tags it acquires ->pcpu_alloc_mutex after ->freeze_lock is acquired or queue is frozen. This potentially creates a circular dependency involving ->fs_reclaim if reclaim is triggered simultaneously in a code path which first acquires ->pcpu_ alloc_mutex. As the queue is already frozen while nr_hw_queue update allocates tagsets, the reclaim can't forward progress and thus it could cause a potential deadlock as reported in lockdep splat[1]. Fix this by pre-allocating tagset tags before we freeze queue during nr_hw_queue update. Later the allocated tagset tags could be safely installed and used after queue is frozen. Reported-by: Yi Zhang Closes: https://lore.kernel.org/all/CAHj4cs8F=OV9s3La2kEQ34YndgfZP-B5PHS4Z8_b9euKG6J4mw@mail.gmail.com/ [1] Signed-off-by: Nilay Shroff Reviewed-by: Ming Lei Tested-by: Yi Zhang Reviewed-by: Yu Kuai [axboe: fix brace style issue] Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

blk-mq: skip CPU offline notify on unmapped hctx

2026-01-08T09:16:52+00:00

[ Upstream commit 10845a105bbcb030647a729f1716c2309da71d33 ] If an hctx has no software ctx mapped, blk_mq_map_swqueue() never allocates tags and leaves hctx->tags NULL. The CPU hotplug offline notifier can still run for that hctx, return early since hctx cannot hold any requests. Signed-off-by: Cong Zhang Fixes: bf0beec0607d ("blk-mq: drain I/O when all CPUs in a hctx are offline") Reviewed-by: Ming Lei Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

block: move elevator tags into struct elevator_resources

2026-01-02T11:56:50+00:00

[ Upstream commit 04728ce90966c54417fd8120a3820104d18ba68d ] This patch introduces a new structure, struct elevator_resources, to group together all elevator-related resources that share the same lifetime. As a first step, this change moves the elevator tag pointer from struct elv_change_ctx into the new struct elevator_resources. Additionally, rename blk_mq_alloc_sched_tags_batch() and blk_mq_free_sched_tags_batch() to blk_mq_alloc_sched_res_batch() and blk_mq_free_sched_res_batch(), respectively. Introduce two new wrapper helpers, blk_mq_alloc_sched_res() and blk_mq_free_sched_res(), around blk_mq_alloc_sched_tags() and blk_mq_free_sched_tags(). These changes pave the way for consolidating the allocation and freeing of elevator-specific resources into common helper functions. This refactoring improves encapsulation and prepares the code for future extensions, allowing additional elevator-specific data to be added to struct elevator_resources without cluttering struct elv_change_ctx. Subsequent patches will extend struct elevator_resources to include other elevator-related data. Reviewed-by: Ming Lei Reviewed-by: Yu Kuai Signed-off-by: Nilay Shroff Signed-off-by: Jens Axboe Stable-dep-of: 9869d3a6fed3 ("block: fix race between wbt_enable_default and IO submission") Signed-off-by: Sasha Levin

block: unify elevator tags and type xarrays into struct elv_change_ctx

2026-01-02T11:56:50+00:00

[ Upstream commit 232143b605387b372dee0ec7830f93b93df5f67d ] Currently, the nr_hw_queues update path manages two disjoint xarrays — one for elevator tags and another for elevator type — both used during elevator switching. Maintaining these two parallel structures for the same purpose adds unnecessary complexity and potential for mismatched state. This patch unifies both xarrays into a single structure, struct elv_change_ctx, which holds all per-queue elevator change context. A single xarray, named elv_tbl, now maps each queue (q->id) in a tagset to its corresponding elv_change_ctx entry, encapsulating the elevator tags, type and name references. This unification simplifies the code, improves maintainability, and clarifies ownership of per-queue elevator state. Reviewed-by: Ming Lei Reviewed-by: Yu Kuai Signed-off-by: Nilay Shroff Signed-off-by: Jens Axboe Stable-dep-of: 9869d3a6fed3 ("block: fix race between wbt_enable_default and IO submission") Signed-off-by: Sasha Levin

block: Use RCU in blk_mq_[un]quiesce_tagset() instead of set->tag_list_lock

2025-12-18T13:03:40+00:00

[ Upstream commit 59e25ef2b413c72da6686d431e7759302cfccafa ] blk_mq_{add,del}_queue_tag_set() functions add and remove queues from tagset, the functions make sure that tagset and queues are marked as shared when two or more queues are attached to the same tagset. Initially a tagset starts as unshared and when the number of added queues reaches two, blk_mq_add_queue_tag_set() marks it as shared along with all the queues attached to it. When the number of attached queues drops to 1 blk_mq_del_queue_tag_set() need to mark both the tagset and the remaining queues as unshared. Both functions need to freeze current queues in tagset before setting on unsetting BLK_MQ_F_TAG_QUEUE_SHARED flag. While doing so, both functions hold set->tag_list_lock mutex, which makes sense as we do not want queues to be added or deleted in the process. This used to work fine until commit 98d81f0df70c ("nvme: use blk_mq_[un]quiesce_tagset") made the nvme driver quiesce tagset instead of quiscing individual queues. blk_mq_quiesce_tagset() does the job and quiesce the queues in set->tag_list while holding set->tag_list_lock also. This results in deadlock between two threads with these stacktraces: __schedule+0x47c/0xbb0 ? timerqueue_add+0x66/0xb0 schedule+0x1c/0xa0 schedule_preempt_disabled+0xa/0x10 __mutex_lock.constprop.0+0x271/0x600 blk_mq_quiesce_tagset+0x25/0xc0 nvme_dev_disable+0x9c/0x250 nvme_timeout+0x1fc/0x520 blk_mq_handle_expired+0x5c/0x90 bt_iter+0x7e/0x90 blk_mq_queue_tag_busy_iter+0x27e/0x550 ? __blk_mq_complete_request_remote+0x10/0x10 ? __blk_mq_complete_request_remote+0x10/0x10 ? __call_rcu_common.constprop.0+0x1c0/0x210 blk_mq_timeout_work+0x12d/0x170 process_one_work+0x12e/0x2d0 worker_thread+0x288/0x3a0 ? rescuer_thread+0x480/0x480 kthread+0xb8/0xe0 ? kthread_park+0x80/0x80 ret_from_fork+0x2d/0x50 ? kthread_park+0x80/0x80 ret_from_fork_asm+0x11/0x20 __schedule+0x47c/0xbb0 ? xas_find+0x161/0x1a0 schedule+0x1c/0xa0 blk_mq_freeze_queue_wait+0x3d/0x70 ? destroy_sched_domains_rcu+0x30/0x30 blk_mq_update_tag_set_shared+0x44/0x80 blk_mq_exit_queue+0x141/0x150 del_gendisk+0x25a/0x2d0 nvme_ns_remove+0xc9/0x170 nvme_remove_namespaces+0xc7/0x100 nvme_remove+0x62/0x150 pci_device_remove+0x23/0x60 device_release_driver_internal+0x159/0x200 unbind_store+0x99/0xa0 kernfs_fop_write_iter+0x112/0x1e0 vfs_write+0x2b1/0x3d0 ksys_write+0x4e/0xb0 do_syscall_64+0x5b/0x160 entry_SYSCALL_64_after_hwframe+0x4b/0x53 The top stacktrace is showing nvme_timeout() called to handle nvme command timeout. timeout handler is trying to disable the controller and as a first step, it needs to blk_mq_quiesce_tagset() to tell blk-mq not to call queue callback handlers. The thread is stuck waiting for set->tag_list_lock as it tries to walk the queues in set->tag_list. The lock is held by the second thread in the bottom stack which is waiting for one of queues to be frozen. The queue usage counter will drop to zero after nvme_timeout() finishes, and this will not happen because the thread will wait for this mutex forever. Given that [un]quiescing queue is an operation that does not need to sleep, update blk_mq_[un]quiesce_tagset() to use RCU instead of taking set->tag_list_lock, update blk_mq_{add,del}_queue_tag_set() to use RCU safe list operations. Also, delete INIT_LIST_HEAD(&q->tag_set_list) in blk_mq_del_queue_tag_set() because we can not re-initialize it while the list is being traversed under RCU. The deleted queue will not be added/deleted to/from a tagset and it will be freed in blk_free_queue() after the end of RCU grace period. Signed-off-by: Mohamed Khalfella Fixes: 98d81f0df70c ("nvme: use blk_mq_[un]quiesce_tagset") Reviewed-by: Ming Lei Reviewed-by: Bart Van Assche Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

blk-mq: Abort suspend when wakeup events are pending

2025-12-18T13:03:36+00:00

[ Upstream commit c196bf43d706592d8801a7513603765080e495fb ] During system suspend, wakeup capable IRQs for block device can be delayed, which can cause blk_mq_hctx_notify_offline() to hang indefinitely while waiting for pending request to complete. Skip the request waiting loop and abort suspend when wakeup events are pending to prevent the deadlock. Fixes: bf0beec0607d ("blk-mq: drain I/O when all CPUs in a hctx are offline") Signed-off-by: Cong Zhang Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

blk-mq: fix stale tag depth for shared sched tags in blk_mq_update_nr_requests()

2025-10-15T13:49:19+00:00

Commit 7f2799c546db ("blk-mq: cleanup shared tags case in blk_mq_update_nr_requests()") moves blk_mq_tag_update_sched_shared_tags() before q->nr_requests is updated, however, it's still using the old q->nr_requests to resize tag depth. Fix this problem by passing in expected new tag depth. Fixes: 7f2799c546db ("blk-mq: cleanup shared tags case in blk_mq_update_nr_requests()") Signed-off-by: Yu Kuai Reviewed-by: Ming Lei Reviewed-by: Nilay Shroff Reported-by: Chris Mason Link: https://lore.kernel.org/linux-block/20251014130507.4187235-2-clm@meta.com/ Signed-off-by: Jens Axboe