Re: [stabe-rc 5.9 ] sched: core.c:7270 Illegal context switch in RCU-bh read-side critical section!

[Jakub Kicinski]

On Tue, 15 Dec 2020 06:45:31 -0800 Paul E. McKenney wrote:
> > Crash log:
> > --------------
> > # selftests: bpf: test_tc_edt.sh
> > [  503.796362]
> > [  503.797960] =============================
> > [  503.802131] WARNING: suspicious RCU usage
> > [  503.806232] 5.9.15-rc1 #1 Tainted: G        W
> > [  503.811358] -----------------------------
> > [  503.815444] /usr/src/kernel/kernel/sched/core.c:7270 Illegal
> > context switch in RCU-bh read-side critical section!
> > [  503.825858]
> > [  503.825858] other info that might help us debug this:
> > [  503.825858]
> > [  503.833998]
> > [  503.833998] rcu_scheduler_active = 2, debug_locks = 1
> > [  503.840981] 3 locks held by kworker/u12:1/157:
> > [  503.845514]  #0: ffff0009754ed538
> > ((wq_completion)netns){+.+.}-{0:0}, at: process_one_work+0x208/0x768
> > [  503.855048]  #1: ffff800013e63df0 (net_cleanup_work){+.+.}-{0:0},
> > at: process_one_work+0x208/0x768
> > [  503.864201]  #2: ffff8000129fe3f0 (pernet_ops_rwsem){++++}-{3:3},
> > at: cleanup_net+0x64/0x3b8
> > [  503.872786]
> > [  503.872786] stack backtrace:
> > [  503.877229] CPU: 1 PID: 157 Comm: kworker/u12:1 Tainted: G        W
> >         5.9.15-rc1 #1
> > [  503.885433] Hardware name: ARM Juno development board (r2) (DT)
> > [  503.891382] Workqueue: netns cleanup_net
> > [  503.895324] Call trace:
> > [  503.897786]  dump_backtrace+0x0/0x1f8
> > [  503.901464]  show_stack+0x2c/0x38
> > [  503.904796]  dump_stack+0xec/0x158
> > [  503.908215]  lockdep_rcu_suspicious+0xd4/0xf8
> > [  503.912591]  ___might_sleep+0x1e4/0x208  
> 
> You really are forbidden to invoke ___might_sleep() while in a BH-disable
> region of code, whether due to rcu_read_lock_bh(), local_bh_disable(),
> or whatever else.
> 
> I do see the cond_resched() in inet_twsk_purge(), but I don't immediately
> see a BH-disable region of code.  Maybe someone more familiar with this
> code would have some ideas.
> 
> Or you could place checks for being in a BH-disable further up in
> the code.  Or build with CONFIG_DEBUG_INFO=y to allow more precise
> interpretation of this stack trace.

My money would be on the option that whatever run on this workqueue
before forgot to re-enable BH, but we already have a check for that...
Naresh, do you have the full log? Is there nothing like "BUG: workqueue
leaked lock" above the splat?