Skip to contentSkip to conversation

blk-cgroup: clear blkg->pd[] with WRITE_ONCE() in blkcg_deactivate_policy()

PATCH In review Guopeng Zhang 2 participants 4 messages Jul 7 – Jul 8

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 blk-cgroup: clear blkg->pd[] with WRITE_ONCE() in blkcg_deactivate_policy()

From: Guopeng Zhang <[email protected]>

blkcg_activate_policy() installs blkg->pd[] entries with WRITE_ONCE() and also uses WRITE_ONCE() when clearing them on its error path. blkg_to_pd() is used by RCU readers and reads the same array with READ_ONCE().

blkcg_deactivate_policy() clears the entry with a plain store. Use WRITE_ONCE() there as well.

Fixes 56cc24f59c14 ("blk-cgroup: don't nest queue_lock under rcu in blkcg_print_blkgs()")
---
 block/blk-cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-cgroup.c +1 -1
index d2a1f5903f24..a1dd69f99f5c 100644
@@ -1691,7 +1691,7 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 			if (blkg->pd[pol->plid]->online && pol->pd_offline_fn)
 				pol->pd_offline_fn(blkg->pd[pol->plid]);
 			pol->pd_free_fn(blkg->pd[pol->plid]);
-			blkg->pd[pol->plid] = NULL;
+			WRITE_ONCE(blkg->pd[pol->plid], NULL);
 		}
 		spin_unlock(&blkcg->lock);
 	}
Signature · 1 line
2.43.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 21 lines · Guopeng Zhang

From: Guopeng Zhang <[email protected]>

blkcg_activate_policy() installs blkg->pd[] entries with WRITE_ONCE() and also uses WRITE_ONCE() when clearing them on its error path. blkg_to_pd() is used by RCU readers and reads the same array with READ_ONCE().

blkcg_deactivate_policy() clears the entry with a plain store. Use WRITE_ONCE() there as well.

Fixes 56cc24f59c14 ("blk-cgroup: don't nest queue_lock under rcu in blkcg_print_blkgs()")
---
 block/blk-cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-cgroup.c +0 -0
index d2a1f5903f24..a1dd69f99f5c 100644

block/blk-cgroup.c void blkcg_deactivate_policy(struct gendisk *disk, · line 1697

Quoted 8 lines
block/blk-cgroup.c +1 -1
@@ -1691,7 +1691,7 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 			if (blkg->pd[pol->plid]->online && pol->pd_offline_fn)
 				pol->pd_offline_fn(blkg->pd[pol->plid]);
 			pol->pd_free_fn(blkg->pd[pol->plid]);
-			blkg->pd[pol->plid] = NULL;
+			WRITE_ONCE(blkg->pd[pol->plid], NULL);
 		}
 		spin_unlock(&blkcg->lock);
 	}
#2 lore raw

FYI, this WRITE_ONCE() change is also in Yu Kuai's in-flight series that factors the pd teardown loop into a helper:

https://lore.kernel.org/all/[email protected]/

That series isn't merged yet (v3, 2026-06-25), so this minimal Fixes patch is still useful in the meantime -- just flagging the overlap.

Quoted 35 lines · Tao Cui
Quoted 21 lines · Guopeng Zhang

From: Guopeng Zhang <[email protected]>

blkcg_activate_policy() installs blkg->pd[] entries with WRITE_ONCE() and also uses WRITE_ONCE() when clearing them on its error path. blkg_to_pd() is used by RCU readers and reads the same array with READ_ONCE().

blkcg_deactivate_policy() clears the entry with a plain store. Use WRITE_ONCE() there as well.

Fixes 56cc24f59c14 ("blk-cgroup: don't nest queue_lock under rcu in blkcg_print_blkgs()")
---
 block/blk-cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-cgroup.c +0 -0
index d2a1f5903f24..a1dd69f99f5c 100644

block/blk-cgroup.c void blkcg_deactivate_policy(struct gendisk *disk, · line 1697

Quoted 8 lines
block/blk-cgroup.c +1 -1
@@ -1691,7 +1691,7 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 			if (blkg->pd[pol->plid]->online && pol->pd_offline_fn)
 				pol->pd_offline_fn(blkg->pd[pol->plid]);
 			pol->pd_free_fn(blkg->pd[pol->plid]);
-			blkg->pd[pol->plid] = NULL;
+			WRITE_ONCE(blkg->pd[pol->plid], NULL);
 		}
 		spin_unlock(&blkcg->lock);
 	}
#3 lore raw
Quoted 8 lines · Tao Cui

FYI, this WRITE_ONCE() change is also in Yu Kuai's in-flight series that factors the pd teardown loop into a helper:

https://lore.kernel.org/all/[email protected]/

That series isn't merged yet (v3, 2026-06-25), so this minimal Fixes patch is still useful in the meantime -- just flagging the overlap.

hello,

Thanks for the review, and thanks for pointing out the overlap.

If Yu's series gets merged first, this patch can simply be dropped.

Thanks,
Guopeng

Quoted 37 lines · Tao Cui
Quoted 35 lines · Guopeng Zhang
Quoted 21 lines

From: Guopeng Zhang <[email protected]>

blkcg_activate_policy() installs blkg->pd[] entries with WRITE_ONCE() and also uses WRITE_ONCE() when clearing them on its error path. blkg_to_pd() is used by RCU readers and reads the same array with READ_ONCE().

blkcg_deactivate_policy() clears the entry with a plain store. Use WRITE_ONCE() there as well.

Fixes 56cc24f59c14 ("blk-cgroup: don't nest queue_lock under rcu in blkcg_print_blkgs()")
---
 block/blk-cgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-cgroup.c +0 -0
index d2a1f5903f24..a1dd69f99f5c 100644

block/blk-cgroup.c void blkcg_deactivate_policy(struct gendisk *disk, · line 1697

Quoted 8 lines
block/blk-cgroup.c +1 -1
@@ -1691,7 +1691,7 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 			if (blkg->pd[pol->plid]->online && pol->pd_offline_fn)
 				pol->pd_offline_fn(blkg->pd[pol->plid]);
 			pol->pd_free_fn(blkg->pd[pol->plid]);
-			blkg->pd[pol->plid] = NULL;
+			WRITE_ONCE(blkg->pd[pol->plid], NULL);
 		}
 		spin_unlock(&blkcg->lock);
 	}