[PATCHv4,0/7] block: fix lock order and remove redundant locking

Message

Nilay Shroff Feb. 25, 2025, 1:30 p.m. UTC

Hi,

After we modeled the freeze & enter queue as lock for supporting lockdep
under commit f1be1788a32e ("block: model freeze & enter queue as lock
for supporting lockdep"), we received numerous lockdep splats. And one
of those splats[1] reported the potential deadlock due to incorrect lock
ordering issue between q->sysfs-lock and q->q_usage_counter. So some of
the patches in this series are aimed to cut the dependency between q->
sysfs-lock and q->q_usage_counter.

This patchset contains seven patches in the series.

The 1st patch helps acquire q->limits_lock instead of q->sysfs_lock while
reading a set of attributes whose write method is protected with atomic
limit update APIs or updates to these attributes could occur under atomic
limit update APIs such as queue_limit_start_update() and queue_limits_
commit_update(). So all such attributes have been now grouped in queue_
attr_show under ->show_limit() method.

The 2nd patch is preparation to further simplify and group sysfs attributes
in subsequent patches. So in this patch we move acquire/release of q->sysfs
_lock as well as queue freeze/unfreeze under each attributes' respective
->show()/->store() method.

The 3rd patch removes the q->sysfs_lock for all sysfs attributes which
don't need it. We identified all sysfs attributes which don't need any
locking and removed use of q->sysfs_lock for these attributes.

Subsequent patches address remaining attributes individually which require
some form of locking other than q->limits_lock or q->sysfs_lock.

The 4th patch introduce a new dedicated lock for elevator switch/update
and thus eliminates the dependency of sched update on q->sysfs_lock.

The 5th patch protects sysfs attribute nr_requests using q->elevator_lock
instead of q->sysfs_lock as the update to q->nr_requests now happen under
q->elevator_lock.

Similarly, the 6th patch protects sysfs attribute wbt_lat_usec using
q->elevator_lock instead of q->sysfs_lock as the update to wbt state and
latency now happen under q->elevator_lock.

The 7th patch protects read_ahead_kb using q->limits_lock instead of
q->sysfs_lock as update to ->ra_pages is protected by ->limits_lock.
The ->ra_pages is usually calculated from the queue limits by queue_
limits_commit_update.

Please note that above changes were unit tested against blktests and
quick xfstests with lockdep enabled.

[1] https://lore.kernel.org/all/[e-mail]/

---
Changes from v3:

  - Make code comment a proper senteence. Also mention the 
    fields protected by q->elevator_lock (hch)

  - Add a comment describing lock ordering followed for 
    newly introduced ->elevator_lock (Hannes Reinecke)  

Link to v3: https://lore.kernel.org/all/[e-mail]/

Changes from v2:
  - Add a prep patch (second patch in the series) to first move q->sysfs_
    lock and queue freeze from queue_attr_show/queue_attr_store to show/
    store method of the respective attribute; then in subsequent patches
    we remove q->sysfs_lock from attributes that don't require any from of
    locking or replace q->sysfs_lock with appropriate lock for attributes
    that require some form of locking; this way we can get away with using
    ->show_nolock()/->store_nolock() methods (hch)

  - Few other misc updates to commit message and code comments (hch)

  - Rearrange the patchset to improve organization and clarity. Start with
    a patch that groups all attributes requiring protection under ->limits_
    lock. Follow this with a preparatory patch, then introduce a patch that
    consolidates attributes that do not require any form of locking.
    Finally, address the remaining attributes individually in subsequent
    patches, which require some form of locking other than q->limits_lock
    or q->sysfs_lock.

Link to v2: https://lore.kernel.org/all/[e-mail]/

Changes from v1:
  - Audit all sysfs attributes in block layer and find attributes which
    don't need any locking as well as attributes which needs some form of
    locking; then remove locking from queue_attr_store/queue_attr_show and
    move it into the attributes that still need it in some form, followed
    by replacing it with the more suitable locks (hch)

  - Use dedicated lock for elevator switch/update (Ming Lei)

  - Re-arrange patchset to first segregate and group together all
    attributes which don't need locking followed by grouping attributes
    which need some form of locking.

Link to v1: https://lore.kernel.org/all/[e-mail]/

---
Nilay Shroff (7):
  block: acquire q->limits_lock while reading sysfs attributes
  block: move q->sysfs_lock and queue-freeze under show/store method
  block: remove q->sysfs_lock for attributes which don't need it
  block: Introduce a dedicated lock for protecting queue elevator
    updates
  block: protect nr_requests update using q->elevator_lock
  block: protect wbt_lat_usec using q->elevator_lock
  block: protect read_ahead_kb using q->limits_lock

 block/blk-core.c       |   1 +
 block/blk-iocost.c     |   2 +
 block/blk-mq.c         |  15 +-
 block/blk-settings.c   |   2 +-
 block/blk-sysfs.c      | 309 +++++++++++++++++++++++++++--------------
 block/elevator.c       |  43 ++++--
 block/elevator.h       |   2 -
 block/genhd.c          |   9 +-
 include/linux/blkdev.h |   8 ++
 9 files changed, 254 insertions(+), 137 deletions(-)