commit ccfc7bf1f0 points out bios from conf->bio_end_io_list also
contribute to conf->nr_queued counter, that's right. But the fix
could be improved. The original fix replaces list_add() by a while()
loop to iterate every bio on conf->bio_end_io_list, and decrease
conf->nr_queued. If we look at the code several lines after, we may
find there is another while() loop to iterate every node from tmp
list which contains the original content of conf->bio_end_io_list.
Yes, we can decrease conf->nr_queued here, then we can avoid the
previous extra while() loop, which consumes more CPU cycles and hold
a spin lock longer time when conf->bio_end_io_list is not tiny.
This patch changes to decrease conf->nr_queued in loop of
while(!list_empty(&tmp)){}, it avoids an extra loop execution, and
avoids holding conf->device_lock for too long time.
Signed-off-by: Coly Li <redacted>
Cc: Shaohua Li <redacted>
---
drivers/md/raid1.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 29e2df5..ea98a73 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -2388,16 +2388,17 @@ static void raid1d(struct md_thread *thread)
LIST_HEAD(tmp);
spin_lock_irqsave(&conf->device_lock, flags);
if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
- while (!list_empty(&conf->bio_end_io_list)) {
- list_move(conf->bio_end_io_list.prev, &tmp);
- conf->nr_queued--;
- }
+ list_add(&tmp, &conf->bio_end_io_list);
+ list_del_init(&conf->bio_end_io_list);
}
spin_unlock_irqrestore(&conf->device_lock, flags);
while (!list_empty(&tmp)) {
r1_bio = list_first_entry(&tmp, struct r1bio,
retry_list);
list_del(&r1_bio->retry_list);
+ spin_lock_irqsave(&conf->device_lock, flags);
+ conf->nr_queued--;
+ spin_unlock_irqrestore(&conf->device_lock, flags);
if (mddev->degraded)
set_bit(R1BIO_Degraded, &r1_bio->state);
if (test_bit(R1BIO_WriteError, &r1_bio->state))--
2.6.6