| Message ID | 20250402043851.946498-2-ming.lei@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | block: fix lock dependency between freeze and elevator lock | expand |
On Wed, Apr 02, 2025 at 12:38:47PM +0800, Ming Lei wrote: > Add blk_mq_enter_no_io() and blk_mq_exit_no_io() for preventing queue > from handling any FS or passthrough IO, meantime the queue is kept in > non-freeze state. > > The added two APIs are for avoiding many potential lock risk related > with freeze lock. > > Also add two variants of memsave version. > > Signed-off-by: Ming Lei <ming.lei@redhat.com> > --- > block/blk-core.c | 6 ++++-- > block/blk-mq.c | 18 ++++++++++++++++-- > block/blk-mq.h | 19 +++++++++++++++++++ > block/blk.h | 5 +++-- > include/linux/blkdev.h | 8 ++++++++ > 5 files changed, 50 insertions(+), 6 deletions(-) > > diff --git a/block/blk-core.c b/block/blk-core.c > index 4623de79effa..a54a18fada8a 100644 > --- a/block/blk-core.c > +++ b/block/blk-core.c > @@ -319,7 +319,8 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) > smp_rmb(); > wait_event(q->mq_freeze_wq, > (!q->mq_freeze_depth && > - blk_pm_resume_queue(pm, q)) || > + (blk_pm_resume_queue(pm, q) || > + !blk_queue_no_io(q))) || > blk_queue_dying(q)); > if (blk_queue_dying(q)) > return -ENODEV; > @@ -352,7 +353,8 @@ int __bio_queue_enter(struct request_queue *q, struct bio *bio) > smp_rmb(); > wait_event(q->mq_freeze_wq, > (!q->mq_freeze_depth && > - blk_pm_resume_queue(false, q)) || > + (blk_pm_resume_queue(false, q) || Here the above '||' should have been '&&'. > + !blk_queue_no_io(q))) || > test_bit(GD_DEAD, &disk->state)); > if (test_bit(GD_DEAD, &disk->state)) > goto dead; > diff --git a/block/blk-mq.c b/block/blk-mq.c > index ae8494d88897..075ee51066b3 100644 > --- a/block/blk-mq.c > +++ b/block/blk-mq.c > @@ -222,8 +222,7 @@ bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic) > bool unfreeze; > > mutex_lock(&q->mq_freeze_lock); > - if (force_atomic) > - q->q_usage_counter.data->force_atomic = true; > + q->q_usage_counter.data->force_atomic = force_atomic; > q->mq_freeze_depth--; > WARN_ON_ONCE(q->mq_freeze_depth < 0); > if (!q->mq_freeze_depth) { > @@ -278,6 +277,21 @@ void blk_mq_quiesce_queue_nowait(struct request_queue *q) > } > EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue_nowait); > > +void blk_mq_enter_no_io(struct request_queue *q) > +{ > + blk_mq_freeze_queue_nomemsave(q); > + q->no_io = true; > + if (__blk_mq_unfreeze_queue(q, true)) > + blk_unfreeze_release_lock(q); > +} > + > +void blk_mq_exit_no_io(struct request_queue *q) > +{ > + blk_mq_freeze_queue_nomemsave(q); > + q->no_io = false; > + blk_mq_unfreeze_queue_nomemrestore(q); > +} > + > /** > * blk_mq_wait_quiesce_done() - wait until in-progress quiesce is done > * @set: tag_set to wait on > diff --git a/block/blk-mq.h b/block/blk-mq.h > index 3011a78cf16a..f49070c8c05f 100644 > --- a/block/blk-mq.h > +++ b/block/blk-mq.h > @@ -452,4 +452,23 @@ static inline bool blk_mq_can_poll(struct request_queue *q) > q->tag_set->map[HCTX_TYPE_POLL].nr_queues; > } > > +void blk_mq_enter_no_io(struct request_queue *q); > +void blk_mq_exit_no_io(struct request_queue *q); > + > +static inline unsigned int __must_check > +blk_mq_enter_no_io_memsave(struct request_queue *q) > +{ > + unsigned int memflags = memalloc_noio_save(); > + > + blk_mq_enter_no_io(q); > + return memflags; > +} > + > +static inline void > +blk_mq_exit_no_io_memrestore(struct request_queue *q, unsigned int memflags) > +{ > + blk_mq_exit_no_io(q); > + memalloc_noio_restore(memflags); > +} > + > #endif > diff --git a/block/blk.h b/block/blk.h > index 006e3be433d2..7d0994c1d3ad 100644 > --- a/block/blk.h > +++ b/block/blk.h > @@ -56,8 +56,9 @@ static inline bool blk_try_enter_queue(struct request_queue *q, bool pm) > * The code that increments the pm_only counter must ensure that the > * counter is globally visible before the queue is unfrozen. > */ > - if (blk_queue_pm_only(q) && > - (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) > + if ((blk_queue_pm_only(q) && > + (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) || Same with above. Thanks, Ming
On 4/2/25 10:08 AM, Ming Lei wrote: > Add blk_mq_enter_no_io() and blk_mq_exit_no_io() for preventing queue > from handling any FS or passthrough IO, meantime the queue is kept in > non-freeze state. > > The added two APIs are for avoiding many potential lock risk related > with freeze lock. > > Also add two variants of memsave version. > > Signed-off-by: Ming Lei <ming.lei@redhat.com> I hope you will spin another patch replacing '||' with '&&' in blk_queue_enter and __bio_queue_enter as you mentioned in another mail. With that change, this looks good to me: Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
diff --git a/block/blk-core.c b/block/blk-core.c index 4623de79effa..a54a18fada8a 100644 --- a/block/blk-core.c +++ b/block/blk-core.c @@ -319,7 +319,8 @@ int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) smp_rmb(); wait_event(q->mq_freeze_wq, (!q->mq_freeze_depth && - blk_pm_resume_queue(pm, q)) || + (blk_pm_resume_queue(pm, q) || + !blk_queue_no_io(q))) || blk_queue_dying(q)); if (blk_queue_dying(q)) return -ENODEV; @@ -352,7 +353,8 @@ int __bio_queue_enter(struct request_queue *q, struct bio *bio) smp_rmb(); wait_event(q->mq_freeze_wq, (!q->mq_freeze_depth && - blk_pm_resume_queue(false, q)) || + (blk_pm_resume_queue(false, q) || + !blk_queue_no_io(q))) || test_bit(GD_DEAD, &disk->state)); if (test_bit(GD_DEAD, &disk->state)) goto dead; diff --git a/block/blk-mq.c b/block/blk-mq.c index ae8494d88897..075ee51066b3 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -222,8 +222,7 @@ bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic) bool unfreeze; mutex_lock(&q->mq_freeze_lock); - if (force_atomic) - q->q_usage_counter.data->force_atomic = true; + q->q_usage_counter.data->force_atomic = force_atomic; q->mq_freeze_depth--; WARN_ON_ONCE(q->mq_freeze_depth < 0); if (!q->mq_freeze_depth) { @@ -278,6 +277,21 @@ void blk_mq_quiesce_queue_nowait(struct request_queue *q) } EXPORT_SYMBOL_GPL(blk_mq_quiesce_queue_nowait); +void blk_mq_enter_no_io(struct request_queue *q) +{ + blk_mq_freeze_queue_nomemsave(q); + q->no_io = true; + if (__blk_mq_unfreeze_queue(q, true)) + blk_unfreeze_release_lock(q); +} + +void blk_mq_exit_no_io(struct request_queue *q) +{ + blk_mq_freeze_queue_nomemsave(q); + q->no_io = false; + blk_mq_unfreeze_queue_nomemrestore(q); +} + /** * blk_mq_wait_quiesce_done() - wait until in-progress quiesce is done * @set: tag_set to wait on diff --git a/block/blk-mq.h b/block/blk-mq.h index 3011a78cf16a..f49070c8c05f 100644 --- a/block/blk-mq.h +++ b/block/blk-mq.h @@ -452,4 +452,23 @@ static inline bool blk_mq_can_poll(struct request_queue *q) q->tag_set->map[HCTX_TYPE_POLL].nr_queues; } +void blk_mq_enter_no_io(struct request_queue *q); +void blk_mq_exit_no_io(struct request_queue *q); + +static inline unsigned int __must_check +blk_mq_enter_no_io_memsave(struct request_queue *q) +{ + unsigned int memflags = memalloc_noio_save(); + + blk_mq_enter_no_io(q); + return memflags; +} + +static inline void +blk_mq_exit_no_io_memrestore(struct request_queue *q, unsigned int memflags) +{ + blk_mq_exit_no_io(q); + memalloc_noio_restore(memflags); +} + #endif diff --git a/block/blk.h b/block/blk.h index 006e3be433d2..7d0994c1d3ad 100644 --- a/block/blk.h +++ b/block/blk.h @@ -56,8 +56,9 @@ static inline bool blk_try_enter_queue(struct request_queue *q, bool pm) * The code that increments the pm_only counter must ensure that the * counter is globally visible before the queue is unfrozen. */ - if (blk_queue_pm_only(q) && - (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) + if ((blk_queue_pm_only(q) && + (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) || + blk_queue_no_io(q)) goto fail_put; rcu_read_unlock(); diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index e39c45bc0a97..1b8fd63eee80 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -498,6 +498,13 @@ struct request_queue { int quiesce_depth; + /* + * Prevent queue from handling IO + * + * keep it in same cache line with q_usage_counter + */ + bool no_io; + struct gendisk *disk; /* @@ -679,6 +686,7 @@ void blk_queue_flag_clear(unsigned int flag, struct request_queue *q); #define blk_queue_sq_sched(q) test_bit(QUEUE_FLAG_SQ_SCHED, &(q)->queue_flags) #define blk_queue_skip_tagset_quiesce(q) \ ((q)->limits.features & BLK_FEAT_SKIP_TAGSET_QUIESCE) +#define blk_queue_no_io(q) (q->no_io) extern void blk_set_pm_only(struct request_queue *q); extern void blk_clear_pm_only(struct request_queue *q);
Add blk_mq_enter_no_io() and blk_mq_exit_no_io() for preventing queue from handling any FS or passthrough IO, meantime the queue is kept in non-freeze state. The added two APIs are for avoiding many potential lock risk related with freeze lock. Also add two variants of memsave version. Signed-off-by: Ming Lei <ming.lei@redhat.com> --- block/blk-core.c | 6 ++++-- block/blk-mq.c | 18 ++++++++++++++++-- block/blk-mq.h | 19 +++++++++++++++++++ block/blk.h | 5 +++-- include/linux/blkdev.h | 8 ++++++++ 5 files changed, 50 insertions(+), 6 deletions(-)