Message ID | 8f4fb91ced02e58ef425189c83214086f1154a0c.1611664710.git.baolin.wang@linux.alibaba.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | blk-cgroup: Use cond_resched() when destroy blkgs | expand |
Hello, Baolin. On Tue, Jan 26, 2021 at 09:33:25PM +0800, Baolin Wang wrote: > On !PREEMPT kernel, we can get below softlockup when doing stress > testing with creating and destroying block cgroup repeatly. The > reason is it may take a long time to acquire the queue's lock in > the loop of blkcg_destroy_blkgs(), thus we can use cond_resched() > instead of cpu_relax() to avoid this issue, since the > blkcg_destroy_blkgs() is not called from atomic contexts. > > [ 4757.010308] watchdog: BUG: soft lockup - CPU#11 stuck for 94s! > [ 4757.010698] Call trace: > [ 4757.010700] blkcg_destroy_blkgs+0x68/0x150 > [ 4757.010701] cgwb_release_workfn+0x104/0x158 > [ 4757.010702] process_one_work+0x1bc/0x3f0 > [ 4757.010704] worker_thread+0x164/0x468 > [ 4757.010705] kthread+0x108/0x138 > > Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com> * Can you please add might_sleep() at the top of the function? * Given that the system can accumulate a huge number of blkgs in pathological cases, I wonder whether a better way to go about it is explicitly testing need_resched() on each loop and release locks and do cond_resched() if true? Thanks.
Hi Tejun, > Hello, Baolin. > > On Tue, Jan 26, 2021 at 09:33:25PM +0800, Baolin Wang wrote: >> On !PREEMPT kernel, we can get below softlockup when doing stress >> testing with creating and destroying block cgroup repeatly. The >> reason is it may take a long time to acquire the queue's lock in >> the loop of blkcg_destroy_blkgs(), thus we can use cond_resched() >> instead of cpu_relax() to avoid this issue, since the >> blkcg_destroy_blkgs() is not called from atomic contexts. >> >> [ 4757.010308] watchdog: BUG: soft lockup - CPU#11 stuck for 94s! >> [ 4757.010698] Call trace: >> [ 4757.010700] blkcg_destroy_blkgs+0x68/0x150 >> [ 4757.010701] cgwb_release_workfn+0x104/0x158 >> [ 4757.010702] process_one_work+0x1bc/0x3f0 >> [ 4757.010704] worker_thread+0x164/0x468 >> [ 4757.010705] kthread+0x108/0x138 >> >> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com> > > * Can you please add might_sleep() at the top of the function? Sure. > > * Given that the system can accumulate a huge number of blkgs in > pathological cases, I wonder whether a better way to go about it is > explicitly testing need_resched() on each loop and release locks and do > cond_resched() if true? Yes, sound better to to me and will update in next version. Thanks for your sugestion.
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c index 3465d6e..af7c0ce 100644 --- a/block/blk-cgroup.c +++ b/block/blk-cgroup.c @@ -1028,7 +1028,7 @@ void blkcg_destroy_blkgs(struct blkcg *blkcg) spin_unlock(&q->queue_lock); } else { spin_unlock_irq(&blkcg->lock); - cpu_relax(); + cond_resched(); spin_lock_irq(&blkcg->lock); } }
On !PREEMPT kernel, we can get below softlockup when doing stress testing with creating and destroying block cgroup repeatly. The reason is it may take a long time to acquire the queue's lock in the loop of blkcg_destroy_blkgs(), thus we can use cond_resched() instead of cpu_relax() to avoid this issue, since the blkcg_destroy_blkgs() is not called from atomic contexts. [ 4757.010308] watchdog: BUG: soft lockup - CPU#11 stuck for 94s! [ 4757.010698] Call trace: [ 4757.010700] blkcg_destroy_blkgs+0x68/0x150 [ 4757.010701] cgwb_release_workfn+0x104/0x158 [ 4757.010702] process_one_work+0x1bc/0x3f0 [ 4757.010704] worker_thread+0x164/0x468 [ 4757.010705] kthread+0x108/0x138 Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com> --- block/blk-cgroup.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)