Skip to contentSkip to conversation

srcu: Fix WARN_ON() for timer_delete_sync() in cleanup_srcu_struct()

PATCH In review Zqiang 4 participants 5 messages Aug 11 – Aug 12

Subsystems (2)
lore
Share & tools
lore

Zqiang

author reviewer
#0 lore raw

PATCH v2 1/1 srcu: Fix WARN_ON() for timer_delete_sync() in cleanup_srcu_struct()

The WARN_ON() for timer_delete_sync() means that the caller forgot to call srcu_barrier() before cleanup. however, it still can trigger even when srcu_barrier() was properly called.

When a SRCU grace period ends and find that the allocation of srcu_node tree has been completed, but the ss_state is still less than SRCU_SIZE_BIG, the mask=~0 causes delay_work timer be queued on every CPU which has been online regardless of whether it's sdp->srcu_cblist has callbacks. the srcu_barrier() only wait srcu_barrier_head's callbacks to complete, however, the srcu_barrier_head will not be inserted into an empty sdp->srcu_cblist. after that, when cleanup_srcu_struct() finds a pending delay_work timer on a CPU with no callbacks, triggering a false positive.

This commit therefore add rcu_segcblist_n_cbs() to WARN_ON(), make the warning triggers only when the timer is still pending and there are actual outstanding callbacks.

Fixes 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is successfully deleted")

---
v1->v2:
Update commit log and modify indentation.

 kernel/rcu/srcutree.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/rcu/srcutree.c +3 -2
index 35fface51d50..be420ecde851 100644
@@ -753,8 +753,9 @@ void cleanup_srcu_struct(struct srcu_struct *ssp)
 
 		// Call srcu_barrier() before this cleanup_srcu_struct()
 		// to avoid triggering this WARN_ON().
-		if (WARN_ON(timer_delete_sync(&sdp->delay_work)) &&
-					rcu_cpu_beenfullyonline(sdp->cpu))
+		if (WARN_ON(timer_delete_sync(&sdp->delay_work) &&
+			    rcu_segcblist_n_cbs(&sdp->srcu_cblist)) &&
+		    rcu_cpu_beenfullyonline(sdp->cpu))
 			queue_work_on(sdp->cpu, rcu_gp_wq, &sdp->work);
 		flush_work(&sdp->work);
 		if (WARN_ON(rcu_segcblist_n_cbs(&sdp->srcu_cblist)))
Signature · 1 line
2.48.1
maintainer Reviewed-by on v2 1/1
#1 lore raw
Quoted 17 lines · Zqiang

The WARN_ON() for timer_delete_sync() means that the caller forgot to call srcu_barrier() before cleanup. however, it still can trigger even when srcu_barrier() was properly called.

When a SRCU grace period ends and find that the allocation of srcu_node tree has been completed, but the ss_state is still less than SRCU_SIZE_BIG, the mask=~0 causes delay_work timer be queued on every CPU which has been online regardless of whether it's sdp->srcu_cblist has callbacks. the srcu_barrier() only wait srcu_barrier_head's callbacks to complete, however, the srcu_barrier_head will not be inserted into an empty sdp->srcu_cblist. after that, when cleanup_srcu_struct() finds a pending delay_work timer on a CPU with no callbacks, triggering a false positive.

This commit therefore add rcu_segcblist_n_cbs() to WARN_ON(), make the warning triggers only when the timer is still pending and there are actual outstanding callbacks.

Quoted 6 lines
Fixes 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is successfully deleted")
Signature · 2 lines
Frederic Weisbecker
SUSE Labs
maintainer
#2 lore raw
Quoted 18 lines · Frederic Weisbecker
Quoted 17 lines · Zqiang

The WARN_ON() for timer_delete_sync() means that the caller forgot to call srcu_barrier() before cleanup. however, it still can trigger even when srcu_barrier() was properly called.

When a SRCU grace period ends and find that the allocation of srcu_node tree has been completed, but the ss_state is still less than SRCU_SIZE_BIG, the mask=~0 causes delay_work timer be queued on every CPU which has been online regardless of whether it's sdp->srcu_cblist has callbacks. the srcu_barrier() only wait srcu_barrier_head's callbacks to complete, however, the srcu_barrier_head will not be inserted into an empty sdp->srcu_cblist. after that, when cleanup_srcu_struct() finds a pending delay_work timer on a CPU with no callbacks, triggering a false positive.

This commit therefore add rcu_segcblist_n_cbs() to WARN_ON(), make the warning triggers only when the timer is still pending and there are actual outstanding callbacks.

Quoted 8 lines
Quoted 6 lines
Fixes 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is successfully deleted")

I pulled this in for further review and testing. My plan is to fold it in to the commit that it fixes, of course with attribution.

Thank you both!

Thanx, Paul

Tested-by on v2 1/1
#3 lore raw
In reply to Zqiang
Quoted 17 lines · Zqiang

The WARN_ON() for timer_delete_sync() means that the caller forgot to call srcu_barrier() before cleanup. however, it still can trigger even when srcu_barrier() was properly called.

When a SRCU grace period ends and find that the allocation of srcu_node tree has been completed, but the ss_state is still less than SRCU_SIZE_BIG, the mask=~0 causes delay_work timer be queued on every CPU which has been online regardless of whether it's sdp->srcu_cblist has callbacks. the srcu_barrier() only wait srcu_barrier_head's callbacks to complete, however, the srcu_barrier_head will not be inserted into an empty sdp->srcu_cblist. after that, when cleanup_srcu_struct() finds a pending delay_work timer on a CPU with no callbacks, triggering a false positive.

This commit therefore add rcu_segcblist_n_cbs() to WARN_ON(), make the warning triggers only when the timer is still pending and there are actual outstanding callbacks.

Quoted 6 lines
Fixes 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is successfully deleted")
maintainer
#4 lore raw
Quoted 18 lines · Breno Leitao
Quoted 17 lines · Zqiang

The WARN_ON() for timer_delete_sync() means that the caller forgot to call srcu_barrier() before cleanup. however, it still can trigger even when srcu_barrier() was properly called.

When a SRCU grace period ends and find that the allocation of srcu_node tree has been completed, but the ss_state is still less than SRCU_SIZE_BIG, the mask=~0 causes delay_work timer be queued on every CPU which has been online regardless of whether it's sdp->srcu_cblist has callbacks. the srcu_barrier() only wait srcu_barrier_head's callbacks to complete, however, the srcu_barrier_head will not be inserted into an empty sdp->srcu_cblist. after that, when cleanup_srcu_struct() finds a pending delay_work timer on a CPU with no callbacks, triggering a false positive.

This commit therefore add rcu_segcblist_n_cbs() to WARN_ON(), make the warning triggers only when the timer is still pending and there are actual outstanding callbacks.

Quoted 8 lines
Quoted 6 lines
Fixes 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is successfully deleted")

Thank you, Breno! I will add this to the original, which I merged this fix into.

Thanx, Paul