Message-ID: <a1141faf8c6a0a924d87132fb4a297cd6d47e09d.camel@redhat.com>
Date:   Wed, 12 Aug 2020 11:44:56 -0400
From:   Lyude Paul <lyude@...hat.com>
To:     Daniel Vetter <daniel@...ll.ch>, Zwane Mwaikambo <zwanem@...il.com>
Cc:     tcamuso@...hat.com, dkwon@...hat.com,
        Linux Kernel <linux-kernel@...r.kernel.org>,
        dri-devel <dri-devel@...ts.freedesktop.org>
Subject: Re: [PATCH] drm: assure aux_dev is nonzero before using it

On Wed, 2020-08-12 at 16:10 +0200, Daniel Vetter wrote:
> On Wed, Aug 12, 2020 at 12:16 AM Zwane Mwaikambo <zwanem@...il.com> wrote:
> > On Tue, 11 Aug 2020, Daniel Vetter wrote:
> > 
> > > On Mon, Aug 10, 2020 at 10:11:50AM -0700, Zwane Mwaikambo wrote:
> > > > Hi Folks,
> > > >     I know this thread eventually dropped off due to not identifying
> > > > the underlying issue. It's still occuring on 5.8 and in my case it
> > > > happened because the udev device nodes for the DP aux devices were not
> > > > cleaned up whereas the kernel had no association with them. I can
> > > > reproduce the bug just by creating a device node for a non-existent
> > > > minor
> > > > device and calling open().
> > > 
> > > Hm I don't have that thread anymore, but generally these bugs are solved
> > > by not registering the device before it's ready for use. We do have
> > > drm_connector->late_register for that stuff. Just a guess since I'm not
> > > seeing full details here.
> > 
> > In this particular case, the physical device disappeared before the nodes
> > were cleaned up. It involves putting a computer to sleep with a monitor
> > plugged in and then waking it up with the monitor unplugged.
> 
> We also have early_unregister for the reverse, but yes this sounds
> more tricky ... Adding Lyude who's been working on way too much
> lifetime fun around dp recently.
> -Daniel
> 
Hi-I think just checking whether the auxdev is NULL or not is a reasonable
fix, although I am curious as to how exactly the aux dev's parent is getting
destroyed before it's child, which I would have thought would be the only way
you could hit this?

> > 
> > > > To me it still makes sense to just check aux_dev because the chardev
> > > > has
> > > > no way to check before calling.
> > > > 
> > > > (gdb) list *drm_dp_aux_dev_get_by_minor+0x29
> > > > 0x17b39 is in drm_dp_aux_dev_get_by_minor
> > > > (drivers/gpu/drm/drm_dp_aux_dev.c:65).
> > > > 60      static struct drm_dp_aux_dev
> > > > *drm_dp_aux_dev_get_by_minor(unsigned index)
> > > > 61      {
> > > > 62              struct drm_dp_aux_dev *aux_dev = NULL;
> > > > 63
> > > > 64              mutex_lock(&aux_idr_mutex);
> > > > 65              aux_dev = idr_find(&aux_idr, index);
> > > > 66              if (!kref_get_unless_zero(&aux_dev->refcount))
> > > > 67                      aux_dev = NULL;
> > > > 68              mutex_unlock(&aux_idr_mutex);
> > > > 69
> > > > (gdb) p/x &((struct drm_dp_aux_dev *)(0x0))->refcount
> > > > $8 = 0x18
> > > > 
> > > > static int auxdev_open(struct inode *inode, struct file *file)
> > > > {
> > > >     unsigned int minor = iminor(inode);
> > > >     struct drm_dp_aux_dev *aux_dev;
> > > > 
> > > >     aux_dev = drm_dp_aux_dev_get_by_minor(minor);
> > > >     if (!aux_dev)
> > > >         return -ENODEV;
> > > > 
> > > >     file->private_data = aux_dev;
> > > >     return 0;
> > > > }
> > > > 
> > > > 
> > > > _______________________________________________
> > > > dri-devel mailing list
> > > > dri-devel@...ts.freedesktop.org
> > > > https://lists.freedesktop.org/mailman/listinfo/dri-devel
> 
> 
-- 
Cheers,
	Lyude Paul (she/her)
	Software Engineer at Red Hat

Powered by blists - more mailing lists