Re: [PATCH] [media] v4l2-async: Always unregister the subdev on failure

[Alban Bedel]

On Fri, 1 Jul 2016 13:55:44 +0200
Hans Verkuil <hverk...@xs4all.nl> wrote:

> On 05/11/2016 06:32 PM, Alban Bedel wrote:
> > On Wed, 11 May 2016 12:22:44 -0400
> > Javier Martinez Canillas <jav...@osg.samsung.com> wrote:
> >   
> >> Hello Alban,
> >>
> >> On 05/11/2016 11:40 AM, Alban Bedel wrote:  
> >>> In v4l2_async_test_notify() if the registered_async callback or the
> >>> complete notifier returns an error the subdev is not unregistered.
> >>> This leave paths where v4l2_async_register_subdev() can fail but
> >>> leave the subdev still registered.
> >>>
> >>> Add the required calls to v4l2_device_unregister_subdev() to plug
> >>> these holes.
> >>>
> >>> Signed-off-by: Alban Bedel <alban.be...@avionic-design.de>
> >>> ---
> >>>  drivers/media/v4l2-core/v4l2-async.c | 10 ++++++++--
> >>>  1 file changed, 8 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/media/v4l2-core/v4l2-async.c 
> >>> b/drivers/media/v4l2-core/v4l2-async.c
> >>> index ceb28d4..43393f8 100644
> >>> --- a/drivers/media/v4l2-core/v4l2-async.c
> >>> +++ b/drivers/media/v4l2-core/v4l2-async.c
> >>> @@ -121,13 +121,19 @@ static int v4l2_async_test_notify(struct 
> >>> v4l2_async_notifier *notifier,
> >>>  
> >>>   ret = v4l2_subdev_call(sd, core, registered_async);
> >>>   if (ret < 0 && ret != -ENOIOCTLCMD) {
> >>> +         v4l2_device_unregister_subdev(sd);
> >>>           if (notifier->unbind)
> >>>                   notifier->unbind(notifier, sd, asd);
> >>>           return ret;
> >>>   }
> >>>  
> >>> - if (list_empty(&notifier->waiting) && notifier->complete)
> >>> -         return notifier->complete(notifier);
> >>> + if (list_empty(&notifier->waiting) && notifier->complete) {
> >>> +         ret = notifier->complete(notifier);
> >>> +         if (ret < 0) {
> >>> +                 v4l2_device_unregister_subdev(sd);  
> >>
> >> Isn't a call to notifier->unbind() missing here as well?
> >>
> >> Also, I think the error path is becoming too duplicated and complex, so
> >> maybe we can have a single error path and use goto labels as is common
> >> in Linux? For example something like the following (not tested) can be
> >> squashed on top of your change:  
> > 
> > Yes, that look better. I'll test it and report tomorrow.  
> 
> I haven't heard anything back about this. Did you manage to test it?

Yes, that's working fine. Sorry for the delay, I'm sending the v2 patch.

Alban

pgpigT3AWV8I_.pgp
Description: OpenPGP digital signature