Re: [tip: locking/core] lockdep: Fix usage_traceoverflow

[Chris Wilson]

Quoting Peter Zijlstra (2020-10-27 12:48:34)
> On Tue, Oct 27, 2020 at 01:30:56PM +0100, Peter Zijlstra wrote:
> > This seems to make it happy. Not quite sure that's the best solution.
> > 
> > diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> > index 3e99dfef8408..81295bc760fe 100644
> > --- a/kernel/locking/lockdep.c
> > +++ b/kernel/locking/lockdep.c
> > @@ -4411,7 +4405,9 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
> >               break;
> >  
> >       case LOCK_USED:
> > -             debug_atomic_dec(nr_unused_locks);
> > +     case LOCK_USED_READ:
> > +             if ((hlock_class(this)->usage_mask & (LOCKF_USED|LOCKF_USED_READ)) == new_mask)
> > +                     debug_atomic_dec(nr_unused_locks);
> >               break;
> >  
> >       default:
> 
> This also works, and I think I likes it better.. anyone?
> 
> diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> index 3e99dfef8408..e603e86c0227 100644
> --- a/kernel/locking/lockdep.c
> +++ b/kernel/locking/lockdep.c
> @@ -4396,6 +4390,9 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
>         if (unlikely(hlock_class(this)->usage_mask & new_mask))
>                 goto unlock;
>  
> +       if (!hlock_class(this)->usage_mask)
> +               debug_atomic_dec(nr_unused_locks);
> +

>From an outside perspective, this is much easier for me to match with
the assertion in lockdep_proc.

Our CI confirms this works, and we are just left with the new issue of

<4> [260.903453] hm#2, depth: 6 [6], eb18a85a2df37d3d != a6ee4649c0022599
<4> [260.903458] WARNING: CPU: 7 PID: 5515 at kernel/locking/lockdep.c:3679 check_chain_key+0x1a4/0x1f0

Thanks,
-Chris