diff options
| author | Kent Overstreet <kent.overstreet@linux.dev> | 2022-10-15 00:34:38 -0400 |
|---|---|---|
| committer | Kent Overstreet <kent.overstreet@linux.dev> | 2023-10-22 17:09:41 -0400 |
| commit | e4b7254c754b676a6f4d607fd92cd71d221ff130 (patch) | |
| tree | fb1316ac3a0b89173514151e27580cd823366c38 /fs/bcachefs/six.c | |
| parent | 5b254da5733d9b8c6a13073fecc506c2861aaeb2 (diff) | |
six locks: Fix a lost wakeup
There was a lost wakeup between a read unlock in percpu mode and a write
lock. The unlock path unlocks, then executes a barrier, then checks for
waiters; correspondingly, the lock side should set the wait bit and
execute a barrier, then attempt to take the lock.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Diffstat (limited to 'fs/bcachefs/six.c')
| -rw-r--r-- | fs/bcachefs/six.c | 11 |
1 files changed, 8 insertions, 3 deletions
diff --git a/fs/bcachefs/six.c b/fs/bcachefs/six.c index 01ff210ff18c..abdc2414f58b 100644 --- a/fs/bcachefs/six.c +++ b/fs/bcachefs/six.c @@ -198,6 +198,14 @@ retry: atomic64_add(__SIX_VAL(write_locking, 1), &lock->state.counter); smp_mb__after_atomic(); + } else if (!(lock->state.waiters & (1 << SIX_LOCK_write))) { + atomic64_add(__SIX_VAL(waiters, 1 << SIX_LOCK_write), + &lock->state.counter); + /* + * pairs with barrier after unlock and before checking + * for readers in unlock path + */ + smp_mb__after_atomic(); } ret = !pcpu_read_count(lock); @@ -212,9 +220,6 @@ retry: if (ret || try) v -= __SIX_VAL(write_locking, 1); - if (!ret && !try && !(lock->state.waiters & (1 << SIX_LOCK_write))) - v += __SIX_VAL(waiters, 1 << SIX_LOCK_write); - if (try && !ret) { old.v = atomic64_add_return(v, &lock->state.counter); six_lock_wakeup(lock, old, SIX_LOCK_read); |