Re: [PATCH] drm/syncobj: make lockdep complain on WAIT_FOR_SUBMIT v2

[Christian König]

Am 19.01.21 um 10:35 schrieb Peter Zijlstra:
> On Mon, Jan 18, 2021 at 07:03:34PM +0100, Christian König wrote:
>
> > diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
> > index 6e74e6745eca..f51458615158 100644
> > --- a/drivers/gpu/drm/drm_syncobj.c
> > +++ b/drivers/gpu/drm/drm_syncobj.c
> > @@ -387,6 +387,13 @@ int drm_syncobj_find_fence(struct drm_file *file_private,
> >   	if (!syncobj)
> >   		return -ENOENT;
> >   
> > +	/* Waiting for userspace with locks help is illegal cause that can
> > +	 * trivial deadlock with page faults for example. Make lockdep complain
> > +	 * about it early on.
> > +	 */
> Egads, the cursed comment style is spreading :/
>
> > +	if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT)
> > +		lockdep_assert_none_held_once();
> > +
> Should this not be part of drm_syncobj_fence_add_wait() instead?

drm_syncobj_fence_add_wait() is only called when the previous try of 
finding the fence wasn't successfully.

If we want to check drivers for stupid behavior for the uncommon wait 
before signal case we need this much earlier.

But I'm going to double check if drm_syncobj_fence_add_wait() isn't used 
elsewhere as well.

> Also, do you want to sprinkle might_sleep() around ?

Good point. Going to add that as well.

Thanks,
Christian.

> >   	*fence = drm_syncobj_fence_get(syncobj);
> >   	drm_syncobj_put(syncobj);
> >   
> > diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h