Re: [PATCH rcu 02/18] rcu: Fix stall-warning deadlock due to non-release of rcu_node ->lock

[Qais Yousef]

On 08/03/21 09:28, Paul E. McKenney wrote:
> On Tue, Aug 03, 2021 at 05:12:21PM +0100, Qais Yousef wrote:
> > On 08/03/21 08:52, Paul E. McKenney wrote:
> > > On Tue, Aug 03, 2021 at 03:24:58PM +0100, Qais Yousef wrote:
> > > > Hi
> > > > 
> > > > On 07/21/21 13:21, Paul E. McKenney wrote:
> > > > > From: Yanfei Xu <yanfei.xu@xxxxxxxxxxxxx>
> > > > > 
> > > > > If rcu_print_task_stall() is invoked on an rcu_node structure that does
> > > > > not contain any tasks blocking the current grace period, it takes an
> > > > > early exit that fails to release that rcu_node structure's lock.  This
> > > > > results in a self-deadlock, which is detected by lockdep.
> > > > > 
> > > > > To reproduce this bug:
> > > > > 
> > > > > tools/testing/selftests/rcutorture/bin/kvm.sh --allcpus --duration 3 --trust-make --configs "TREE03" --kconfig "CONFIG_PROVE_LOCKING=y" --bootargs "rcutorture.stall_cpu=30 rcutorture.stall_cpu_block=1 rcutorture.fwd_progress=0 rcutorture.test_boost=0"
> > > > > 
> > > > > This will also result in other complaints, including RCU's scheduler
> > > > > hook complaining about blocking rather than preemption and an rcutorture
> > > > > writer stall.
> > > > > 
> > > > > Only a partial RCU CPU stall warning message will be printed because of
> > > > > the self-deadlock.
> > > > > 
> > > > > This commit therefore releases the lock on the rcu_print_task_stall()
> > > > > function's early exit path.
> > > > > 
> > > > > Fixes: c583bcb8f5ed ("rcu: Don't invoke try_invoke_on_locked_down_task() with irqs disabled")
> > > > > Signed-off-by: Yanfei Xu <yanfei.xu@xxxxxxxxxxxxx>
> > > > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > > > ---
> > > > 
> > > > We are seeing similar stall/deadlock issue on android 5.10 kernel, is the fix
> > > > relevant here? Trying to apply the patches and test, but the problem is tricky
> > > > to reproduce so thought worth asking first.
> > > 
> > > Looks like the relevant symptoms to me, so I suggest trying this series
> > > from -rcu:
> > > 
> > > 8baded711edc ("rcu: Fix to include first blocked task in stall warning")
> > > f6b3995a8b56 ("rcu: Fix stall-warning deadlock due to non-release of rcu_node ->lock")
> > 
> > Great thanks. These are the ones we picked as the rest was a bit tricky to
> > apply on 5.10.
> > 
> > While at it, we see these errors too though they look harmless. They happen
> > all the time
> > 
> > 	[  595.292685] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #02!!!"}
> > 	[  595.301467] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 	[  595.389353] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 	[  595.397454] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 	[  595.417112] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 	[  595.425215] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 	[  595.438807] NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!"}
> > 
> > I used to see them on mainline a while back but seem to have been fixed.
> > Something didn't get backported to 5.10 perhaps?
> 
> I believe that you need at least this one:
> 
> 47c218dcae65 ("tick/sched: Prevent false positive softirq pending warnings on RT")

After looking at the content of the patch, it's not related. We don't run with
PREEMPT_RT.

I think we're hitting a genuine issue, most likely due to out-of-tree changes
done by Android to fix RT latency problems against softirq (surprise surprise).

Thanks for your help and sorry for the noise.

Cheers

--
Qais Yousef