io_uring: batch reap of dead file registrations
authorJens Axboe <axboe@kernel.dk>
Thu, 14 May 2020 23:21:15 +0000 (17:21 -0600)
committerJens Axboe <axboe@kernel.dk>
Fri, 15 May 2020 17:56:18 +0000 (11:56 -0600)
commit4a38aed2a0a729ccecd84dca5b76d827b9e1294d
tree46d413220c0bfd00f6473b7c514318847558e488
parent0f158b4cf20e7983d5b33878a6aad118cfac4f05
io_uring: batch reap of dead file registrations

We currently embed and queue a work item per fixed_file_ref_node that
we update, but if the workload does a lot of these, then the associated
kworker-events overhead can become quite noticeable.

Since we rarely need to wait on these, batch them at 1 second intervals
instead. If we do need to wait for them, we just flush the pending
delayed work.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c