[0/5] Bug fix for recent batching change

Message

Logan Gunthorpe July 27, 2022, 9:05 p.m. UTC

Hey,

We hit another bug on my recent batching patch. In this case the
bug has never been hit with the current md/md-next branch but
some other patches we were working on changed the timing such
that we hit this bug. It is theoretically possible to hit in
the md/md-next batch so this patchset contains a fix.

The fix is the last commit. The first four commits are some
basic refactoring that makes the final commit a bit easier.

A git repo is here and is based on current md/md-next (7a6f9e9cf1):

   https://github.com/sbates130272/linux-p2pmem raid5_batch_quiesce

Thanks,

Logan

--

Logan Gunthorpe (5):
  md/raid5: Refactor raid5_get_active_stripe()
  md/raid5: Make is_inactive_blocked() helper
  md/raid5: Drop unnecessary call to r5c_check_stripe_cache_usage()
  md/raid5: Move stripe_request_ctx up
  md/raid5: Ensure batch_last is released before sleeping for quiesce

 drivers/md/raid5.c | 162 ++++++++++++++++++++++++++++-----------------
 drivers/md/raid5.h |   2 +-
 2 files changed, 101 insertions(+), 63 deletions(-)


base-commit: 7a6f9e9cf1befa0a1578501966d3c9b0cae46727
--
2.30.2

Comments

Song Liu July 28, 2022, 5:55 a.m. UTC | #1

On Wed, Jul 27, 2022 at 2:06 PM Logan Gunthorpe <[e-mail]> wrote:
>
> Hey,
>
> We hit another bug on my recent batching patch. In this case the
> bug has never been hit with the current md/md-next branch but
> some other patches we were working on changed the timing such
> that we hit this bug. It is theoretically possible to hit in
> the md/md-next batch so this patchset contains a fix.
>
> The fix is the last commit. The first four commits are some
> basic refactoring that makes the final commit a bit easier.
>
> A git repo is here and is based on current md/md-next (7a6f9e9cf1):
>
>    https://github.com/sbates130272/linux-p2pmem raid5_batch_quiesce

Applied to md-next. Thanks!

Song