Re: [tip: sched/urgent] sched/core: Avoid spurious lock dependencies

From: Sebastian Andrzej Siewior
Date: Fri Nov 22 2019 - 15:01:27 EST

On 2019-11-13 10:06:28 [-0000], tip-bot2 for Peter Zijlstra wrote:
> sched/core: Avoid spurious lock dependencies
> While seemingly harmless, __sched_fork() does hrtimer_init(), which,
> when DEBUG_OBJETS, can end up doing allocations.
> This then results in the following lock order:
> rq->lock
> zone->lock.rlock
> batched_entropy_u64.lock
> Which in turn causes deadlocks when we do wakeups while holding that
> batched_entropy lock -- as the random code does.

Peter, can it _really_ cause deadlocks? My understanding was that the
batched_entropy_u64.lock is a per-CPU lock and can _not_ cause a
deadlock because it can be always acquired on multiple CPUs
simultaneously (and it is never acquired cross-CPU).
Lockdep is simply not smart enough to see that and complains about it
like it would complain about a regular lock in this case.