linux-block.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Ming Lei <ming.lei@redhat.com>
To: Nilay Shroff <nilay@linux.ibm.com>
Cc: "Jens Axboe" <axboe@kernel.dk>,
	linux-block@vger.kernel.org,
	"Shinichiro Kawasaki" <shinichiro.kawasaki@wdc.com>,
	"Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	"Christoph Hellwig" <hch@lst.de>
Subject: Re: [PATCH 12/15] block: move debugfs/sysfs register out of freezing queue
Date: Tue, 15 Apr 2025 20:41:49 +0800	[thread overview]
Message-ID: <Z_5UDa_d5Fubx2_j@fedora> (raw)
In-Reply-To: <43325beb-6147-4f51-8e79-0c31db2ef742@linux.ibm.com>

On Tue, Apr 15, 2025 at 05:51:07PM +0530, Nilay Shroff wrote:
> 
> 
> On 4/15/25 5:24 PM, Ming Lei wrote:
> >>>
> >>> Why is updating nr_requests related with removing hctx attributes?
> >>>
> >>> Can you explain the side effect in details?
> >> Thread 1:
> >> writing-to-blk-mq-sysfs-attribute-nr_requests 
> >>   -> queue_requests_store ==> freezes queue and acquires ->elevator_lock 
> >>     -> blk_mq_update_nr_requests 
> >>       -> blk_mq_tag_update_depth
> >>         -> blk_mq_alloc_map_and_rqs
> >>           -> blk_mq_alloc_rq_map
> >>             -> blk_mq_init_tags ==> updates ->nr_tags and ->nr_reserved_tags 
> >>
> >> Thread2:
> >> blk_mq_update_nr_hw_queues
> >>   -> __blk_mq_update_nr_hw_queues
> >>     -> blk_mq_realloc_tag_set_tags
> >>       -> __blk_mq_alloc_map_and_rqs
> >>         -> blk_mq_alloc_map_and_rqs
> >>           -> blk_mq_alloc_rq_map
> >>             -> blk_mq_init_tags ==> updates ->nr_tags and ->nr_reserved_tags 
> >>
> >> Thread 3:
> >> reading-hctx-sysfs-attribute-nr_tags
> >>   -> blk_mq_hw_sysfs_show ==> acquires ->elevaor_lock
> >>     ->  blk_mq_hw_sysfs_nr_tags_show ==> access nr_tags 
> >>
> >> Thread 4:
> >> reading-hctx-sysfs-attribute-nr_reserved_tags
> >>   -> blk_mq_hw_sysfs_show ==> acquires ->elevaor_lock
> >>     -> blk_mq_hw_sysfs_nr_reserved_tags_show ==> access nr_reserved_tags
> > 
> > `hctx->tags` is guaranteed to be live if above ->show() method, and the
> > elevator lock is actually not needed, which isn't supposed to protect
> > hctx->tags too.
> > 
> I think, the ->elavtor_lock would still be needed for protecting updates to 
> hctx->tags from thread # 1 above and simultaneously reading the hctx->tags from 
> thread #3 and #4 above.

update_nr_requests() can just reduce ->tags.sbitmap size, please see
blk_mq_tag_update_depth(), even it won't change hctx->tags->nr_tags, so there
isn't race.

More importantly ->elevator_lock shouldn't cover ->tags which isn't related
with scheduler, one exception is blk-mq debugfs, in which request may be
allocated from ->sched_tags, even ->tags is visiting.


Thanks, 
Ming


  reply	other threads:[~2025-04-15 12:42 UTC|newest]

Thread overview: 56+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-04-10 13:30 [PATCH 00/15] block: unify elevator changing and fix lockdep warning Ming Lei
2025-04-10 13:30 ` [PATCH 01/15] block: don't call freeze queue in elevator_switch() and elevator_disable() Ming Lei
2025-04-10 13:30 ` [PATCH 02/15] block: add two helpers for registering/un-registering sched debugfs Ming Lei
2025-04-10 14:25   ` Christoph Hellwig
2025-04-10 13:30 ` [PATCH 03/15] block: move sched debugfs register into elvevator_register_queue Ming Lei
2025-04-10 14:27   ` Christoph Hellwig
2025-04-14  0:42     ` Ming Lei
2025-04-10 13:30 ` [PATCH 04/15] block: prevent elevator switch during updating nr_hw_queues Ming Lei
2025-04-10 14:36   ` Christoph Hellwig
2025-04-14  0:54     ` Ming Lei
2025-04-14  6:07       ` Christoph Hellwig
2025-04-15  2:03         ` Ming Lei
2025-04-11 19:13   ` Nilay Shroff
2025-04-14  0:55     ` Ming Lei
2025-04-10 13:30 ` [PATCH 05/15] block: simplify elevator reset for " Ming Lei
2025-04-10 14:40   ` Christoph Hellwig
2025-04-10 15:34   ` Christoph Hellwig
2025-04-14  0:58     ` Ming Lei
2025-04-14  6:09       ` Christoph Hellwig
2025-04-15  2:05         ` Ming Lei
2025-04-10 13:30 ` [PATCH 06/15] block: add helper of elevator_change() Ming Lei
2025-04-10 13:30 ` [PATCH 07/15] block: move blk_unregister_queue() & device_del() after freeze wait Ming Lei
2025-04-14  6:19   ` Christoph Hellwig
2025-04-15  2:26     ` Ming Lei
2025-04-10 13:30 ` [PATCH 08/15] block: add `struct elev_change_ctx` for unifying elevator change Ming Lei
2025-04-14  6:21   ` Christoph Hellwig
2025-04-10 13:30 ` [PATCH 09/15] block: " Ming Lei
2025-04-10 18:37   ` Nilay Shroff
2025-04-14  1:22     ` Ming Lei
2025-04-15 12:30       ` Nilay Shroff
2025-04-16  1:49         ` Ming Lei
2025-04-10 13:30 ` [PATCH 10/15] block: pass elevator_queue to elv_register_queue & unregister_queue Ming Lei
2025-04-14  6:22   ` Christoph Hellwig
2025-04-15  2:31     ` Ming Lei
2025-04-16  4:53       ` Christoph Hellwig
2025-04-10 13:30 ` [PATCH 11/15] block: move elv_register[unregister]_queue out of elevator_lock Ming Lei
2025-04-11 19:20   ` Nilay Shroff
2025-04-14  1:24     ` Ming Lei
2025-04-15  9:39       ` Nilay Shroff
2025-04-15 10:32         ` Ming Lei
2025-04-10 13:30 ` [PATCH 12/15] block: move debugfs/sysfs register out of freezing queue Ming Lei
2025-04-10 18:57   ` Nilay Shroff
2025-04-14  1:42     ` Ming Lei
2025-04-15  9:37       ` Nilay Shroff
2025-04-15 10:06         ` Ming Lei
2025-04-15 11:15           ` Nilay Shroff
2025-04-15 11:54             ` Ming Lei
2025-04-15 12:21               ` Nilay Shroff
2025-04-15 12:41                 ` Ming Lei [this message]
2025-04-10 13:30 ` [PATCH 13/15] block: remove several ->elevator_lock Ming Lei
2025-04-10 19:07   ` Nilay Shroff
2025-04-14  1:46     ` Ming Lei
2025-04-10 13:30 ` [PATCH 14/15] block: move hctx cpuhp add/del out of queue freezing Ming Lei
2025-04-10 13:30 ` [PATCH 15/15] block: move wbt_enable_default() out of queue freezing from scheduler's ->exit() Ming Lei
2025-04-10 19:20   ` Nilay Shroff
2025-04-14  1:55     ` Ming Lei

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Z_5UDa_d5Fubx2_j@fedora \
    --to=ming.lei@redhat.com \
    --cc=axboe@kernel.dk \
    --cc=hch@lst.de \
    --cc=linux-block@vger.kernel.org \
    --cc=nilay@linux.ibm.com \
    --cc=shinichiro.kawasaki@wdc.com \
    --cc=thomas.hellstrom@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).