summaryrefslogtreecommitdiff
path: root/fs/bcachefs/six.c
diff options
context:
space:
mode:
authorKent Overstreet <kent.overstreet@linux.dev>2023-05-21 23:41:56 -0400
committerKent Overstreet <kent.overstreet@linux.dev>2023-10-22 17:10:02 -0400
commit37f612bea5bd921e71537df3559a117dffb0956d (patch)
tree7abdce668d12860c2680600e13a5e83f9b6845a8 /fs/bcachefs/six.c
parent91d16f16d0fd4b6eb8503068ea7f6ad8305e32db (diff)
six locks: Improve spurious wakeup handling in pcpu reader mode
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Diffstat (limited to 'fs/bcachefs/six.c')
-rw-r--r--fs/bcachefs/six.c41
1 files changed, 27 insertions, 14 deletions
diff --git a/fs/bcachefs/six.c b/fs/bcachefs/six.c
index a1f007095ec9..fcc74e626db0 100644
--- a/fs/bcachefs/six.c
+++ b/fs/bcachefs/six.c
@@ -193,6 +193,15 @@ static inline unsigned pcpu_read_count(struct six_lock *lock)
return read_count;
}
+/*
+ * __do_six_trylock() - main trylock routine
+ *
+ * Returns 1 on success, 0 on failure
+ *
+ * In percpu reader mode, a failed trylock may cause a spurious trylock failure
+ * for anoter thread taking the competing lock type, and we may havve to do a
+ * wakeup: when a wakeup is required, we return -1 - wakeup_type.
+ */
static int __do_six_trylock(struct six_lock *lock, enum six_lock_type type,
struct task_struct *task, bool try)
{
@@ -219,8 +228,20 @@ static int __do_six_trylock(struct six_lock *lock, enum six_lock_type type,
* the lock, then issues a full memory barrier, then reads from the
* other thread's variable to check if the other thread thinks it has
* the lock. If we raced, we backoff and retry/sleep.
+ *
+ * Failure to take the lock may cause a spurious trylock failure in
+ * another thread, because we temporarily set the lock to indicate that
+ * we held it. This would be a problem for a thread in six_lock(), when
+ * they are calling trylock after adding themself to the waitlist and
+ * prior to sleeping.
+ *
+ * Therefore, if we fail to get the lock, and there were waiters of the
+ * type we conflict with, we will have to issue a wakeup.
+ *
+ * Since we may be called under wait_lock (and by the wakeup code
+ * itself), we return that the wakeup has to be done instead of doing it
+ * here.
*/
-
if (type == SIX_LOCK_read && lock->readers) {
preempt_disable();
this_cpu_inc(*lock->readers); /* signal that we own lock */
@@ -233,17 +254,11 @@ static int __do_six_trylock(struct six_lock *lock, enum six_lock_type type,
this_cpu_sub(*lock->readers, !ret);
preempt_enable();
- /*
- * If we failed because a writer was trying to take the
- * lock, issue a wakeup because we might have caused a
- * spurious trylock failure:
- */
- if (old & SIX_STATE_WRITE_LOCKING)
+ if (!ret && (old & SIX_STATE_WAITING_WRITE))
ret = -1 - SIX_LOCK_write;
} else if (type == SIX_LOCK_write && lock->readers) {
if (try) {
- atomic64_add(SIX_STATE_WRITE_LOCKING,
- &lock->state);
+ atomic64_add(SIX_STATE_WRITE_LOCKING, &lock->state);
smp_mb__after_atomic();
}
@@ -259,12 +274,10 @@ static int __do_six_trylock(struct six_lock *lock, enum six_lock_type type,
if (ret || try)
v -= SIX_STATE_WRITE_LOCKING;
- if (try && !ret) {
+ if (v) {
old = atomic64_add_return(v, &lock->state);
- if (old & SIX_STATE_WAITING_READ)
+ if (!ret && try && (old & SIX_STATE_WAITING_READ))
ret = -1 - SIX_LOCK_read;
- } else {
- atomic64_add(v, &lock->state);
}
} else {
v = atomic64_read(&lock->state);
@@ -422,7 +435,7 @@ bool six_relock_ip(struct six_lock *lock, enum six_lock_type type,
*/
if (ret)
six_acquire(&lock->dep_map, 1, type == SIX_LOCK_read, ip);
- else if (old & SIX_STATE_WRITE_LOCKING)
+ else if (old & SIX_STATE_WAITING_WRITE)
six_lock_wakeup(lock, old, SIX_LOCK_write);
return ret;