[RFC,0/1] A rare failure to wake up sync tasks

Message

Benjamin Coddington July 10, 2024, 8:32 p.m. UTC

Hey all - RFC patch here because this race is very hard to reproduce so I
will be testing this for a few days before feeling certain that this fix
does the job.

Meantime, I'd love some feedback from any one that has mastered load/store
races.  Do I really need a full smp_mb() here?  I think I do because the
race is not to the tk_runstate's ACTIVE bit which is set under the queue
lock, but rather to the wait_queue_head..  I'd love to be corrected.

Comments much appreciated.. this problem has been very elusive to catch.

Benjamin Coddington (1):
  SUNRPC: Fix a race to wake a sync task

 net/sunrpc/sched.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)