Skip to contentSkip to conversation

kernfs: skip lockdep annotation always if ignore_lockdep is set

PATCH In review Imre Deak 3 participants 3 messages Feb 7, 2014 kernfs

lore
Share & tools
lore

Imre Deak

author
#0 lore raw

PATCH 1/1 kernfs: skip lockdep annotation always if ignore_lockdep is set

We skip lockdep annotations for sysfs attributes with ignore_lockdep set. An exception is kernfs_deactivate where we annotate even in this case. Since the lockdep map key needed for the annotation is not initialized whenever ignore_lockdep is set, we'll get a warning for the uninitialized key.

Note that this happens since,

commit 517e64f57883bd63c5a4ab8b3d0d3ed68c55d0cf

Author: Tejun Heo <[email protected]>
Date:   Thu Nov 28 14:54:29 2013 -0500

Before this change we had a valid key even in case of ignore_lockdep, so the annotation in kernfs_deactivate worked, or at least didn't produce the above warning. Fix this by skipping the annotation whenever ignore_lockdep is set.

The actual attribute triggering the bug was 'delete_device' in drivers/i2c/i2c-core.c, for the backtrace see the reference below.

Reference: http://lists.freedesktop.org/archives/intel-gfx/2014-February/039663.html

---
 fs/kernfs/dir.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/kernfs/dir.c +8 -4
index 5104cf5..bd6e18b 100644
@@ -187,19 +187,23 @@ static void kernfs_deactivate(struct kernfs_node *kn)
 
 	kn->u.completion = (void *)&wait;
 
-	rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
+	if (kn->flags & KERNFS_LOCKDEP)
+		rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
 	/* atomic_add_return() is a mb(), put_active() will always see
 	 * the updated kn->u.completion.
 	 */
 	v = atomic_add_return(KN_DEACTIVATED_BIAS, &kn->active);
 
 	if (v != KN_DEACTIVATED_BIAS) {
-		lock_contended(&kn->dep_map, _RET_IP_);
+		if (kn->flags & KERNFS_LOCKDEP)
+			lock_contended(&kn->dep_map, _RET_IP_);
 		wait_for_completion(&wait);
 	}
 
-	lock_acquired(&kn->dep_map, _RET_IP_);
-	rwsem_release(&kn->dep_map, 1, _RET_IP_);
+	if (kn->flags & KERNFS_LOCKDEP) {
+		lock_acquired(&kn->dep_map, _RET_IP_);
+		rwsem_release(&kn->dep_map, 1, _RET_IP_);
+	}
 }
 
 /**
Signature · 1 line
1.8.1.2
maintainer
#1 lore raw
Quoted 12 lines · Imre Deak

We skip lockdep annotations for sysfs attributes with ignore_lockdep set. An exception is kernfs_deactivate where we annotate even in this case. Since the lockdep map key needed for the annotation is not initialized whenever ignore_lockdep is set, we'll get a warning for the uninitialized key.

Note that this happens since,

commit 517e64f57883bd63c5a4ab8b3d0d3ed68c55d0cf

Author: Tejun Heo <[email protected]>
Date:   Thu Nov 28 14:54:29 2013 -0500
Quoted 7 lines

Before this change we had a valid key even in case of ignore_lockdep, so the annotation in kernfs_deactivate worked, or at least didn't produce the above warning. Fix this by skipping the annotation whenever ignore_lockdep is set.

The actual attribute triggering the bug was 'delete_device' in drivers/i2c/i2c-core.c, for the backtrace see the reference below.

I think Greg already has about the same patch queued.

http://lkml.kernel.org/g/[email protected]

Thanks.

Signature · 1 line
tejun
#2 lore raw
Quoted 18 lines · Tejun Heo
Quoted 17 lines · Imre Deak

We skip lockdep annotations for sysfs attributes with ignore_lockdep set. An exception is kernfs_deactivate where we annotate even in this case. Since the lockdep map key needed for the annotation is not initialized whenever ignore_lockdep is set, we'll get a warning for the uninitialized key.

Note that this happens since,

commit 517e64f57883bd63c5a4ab8b3d0d3ed68c55d0cf

Author: Tejun Heo <[email protected]>
Date:   Thu Nov 28 14:54:29 2013 -0500

Before this change we had a valid key even in case of ignore_lockdep, so the annotation in kernfs_deactivate worked, or at least didn't produce the above warning. Fix this by skipping the annotation whenever ignore_lockdep is set.

Quoted 6 lines
Quoted 2 lines

The actual attribute triggering the bug was 'delete_device' in drivers/i2c/i2c-core.c, for the backtrace see the reference below.

I think Greg already has about the same patch queued.

http://lkml.kernel.org/g/[email protected]

Yes, I just sent it to Linus a few minutes ago.

thanks,

greg k-h