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

[Peter Zijlstra]

On Tue, Oct 27, 2020 at 12:59:55PM +0100, Peter Zijlstra wrote:
> On Tue, Oct 27, 2020 at 11:29:35AM +0000, Chris Wilson wrote:
> > Quoting tip-bot2 for Peter Zijlstra (2020-10-07 17:20:13)
> > > The following commit has been merged into the locking/core branch of tip:
> > > 
> > > Commit-ID:     24d5a3bffef117ed90685f285c6c9d2faa3a02b4
> > > Gitweb:        https://git.kernel.org/tip/24d5a3bffef117ed90685f285c6c9d2faa3a02b4
> > > Author:        Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > AuthorDate:    Wed, 30 Sep 2020 11:49:37 +02:00
> > > Committer:     Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > CommitterDate: Wed, 07 Oct 2020 18:14:17 +02:00
> > > 
> > > lockdep: Fix usage_traceoverflow
> > > 
> > > Basically print_lock_class_header()'s for loop is out of sync with the
> > > the size of of ->usage_traces[].
> > 
> > We're hitting a problem,
> > 
> > 	$ cat /proc/lockdep_stats
> > 
> > upon boot generates:
> > 
> > [   29.465702] DEBUG_LOCKS_WARN_ON(debug_atomic_read(nr_unused_locks) != nr_unused)
> > [   29.465716] WARNING: CPU: 0 PID: 488 at kernel/locking/lockdep_proc.c:256 lockdep_stats_show+0xa33/0xac0
> > 
> > that bisected to this patch. Only just completed the bisection and
> > thought you would like a heads up.
> 
> Oh hey, that's 'curious'... it does indeed trivially reproduce, let me
> have a poke.

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: