Re: [PATCH] md: optimize barrier usage for Rmw atomic bitops

From: Guoqing Jiang
Date: Fri Jan 31 2020 - 11:43:59 EST




On 1/29/20 7:14 PM, Davidlohr Bueso wrote:
For both set and clear_bit, we can avoid the unnecessary barrier
on non LL/SC architectures, such as x86. Instead, use the
smp_mb__{before,after}_atomic() calls.

Signed-off-by: Davidlohr Bueso <dbueso@xxxxxxx>
---
drivers/md/md.c | 2 +-
drivers/md/raid10.c | 7 ++++---
drivers/md/raid5.c | 9 +++++----
3 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 4824d50526fa..4ed2eb6933f7 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -2561,7 +2561,7 @@ static bool set_in_sync(struct mddev *mddev)
* Ensure ->in_sync is visible before we clear
* ->sync_checkers.
*/
- smp_mb();
+ smp_mb__before_atomic();
set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
sysfs_notify_dirent_safe(mddev->sysfs_state);
}
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index ec136e44aef7..1993a1958c75 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -1865,9 +1865,10 @@ static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
/* We must have just cleared 'rdev' */
p->rdev = p->replacement;
clear_bit(Replacement, &p->replacement->flags);
- smp_mb(); /* Make sure other CPUs may see both as identical
- * but will never see neither -- if they are careful.
- */
+ /* Make sure other CPUs may see both as identical
+ * but will never see neither -- if they are careful.
+ */

Since we are here, it is better to change the comment style to

/*
* ...
*/

+ smp_mb__after_atomic();
p->replacement = NULL;
}
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index ba00e9877f02..3ad6209287cf 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -364,7 +364,7 @@ static int release_stripe_list(struct r5conf *conf,
int hash;
/* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
- smp_mb();
+ smp_mb__before_atomic();
clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state);
/*
* Don't worry the bit is set here, because if the bit is set
@@ -7654,9 +7654,10 @@ static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
/* We must have just cleared 'rdev' */
p->rdev = p->replacement;
clear_bit(Replacement, &p->replacement->flags);
- smp_mb(); /* Make sure other CPUs may see both as identical
- * but will never see neither - if they are careful
- */

Ditto.

Acked-by: Guoqing Jiang <guoqing.jiang@xxxxxxxxxxxxxxx>

Thanks,
Guoqing