On Tue, 20 Feb 2024 at 14:56, Julia Lawall <julia.lawall@xxxxxxxx> wrote:
>
>
>
> On Tue, 20 Feb 2024, Dmitry Baryshkov wrote:
>
> > On Tue, 20 Feb 2024 at 13:52, Julia Lawall <julia.lawall@xxxxxxxx> wrote:
> > >
> > >
> > >
> > > On Tue, 20 Feb 2024, Johan Hovold wrote:
> > >
> > > > On Mon, Feb 19, 2024 at 06:48:30PM +0100, Markus Elfring wrote:
> > > > > > The two device node references taken during allocation need to be
> > > > > > dropped when the auxiliary device is freed.
> > > > > …
> > > > > > +++ b/drivers/gpu/drm/bridge/aux-hpd-bridge.c
> > > > > …
> > > > > > @@ -74,6 +75,8 @@ struct device *drm_dp_hpd_bridge_register(struct device *parent,
> > > > > >
> > > > > > ret = auxiliary_device_init(adev);
> > > > > > if (ret) {
> > > > > > + of_node_put(adev->dev.platform_data);
> > > > > > + of_node_put(adev->dev.of_node);
> > > > > > ida_free(&drm_aux_hpd_bridge_ida, adev->id);
> > > > > > kfree(adev);
> > > > > > return ERR_PTR(ret);
> > > > >
> > > > > The last two statements are also used in a previous if branch.
> > > > > https://elixir.bootlin.com/linux/v6.8-rc5/source/drivers/gpu/drm/bridge/aux-hpd-bridge.c#L63
> > > > >
> > > > > How do you think about to avoid such a bit of duplicate source code
> > > > > by adding a label here?
> > > >
> > > > No, the current code is fine and what you are suggesting is in any case
> > > > unrelated to this fix.
> > > >
> > > > If this function ever grows a third error path like that, I too would
> > > > consider it however.
> > >
> > > I guess these of_node_puts can all go away shortly with cleanup anyway?
> >
> > I'm not sure about it. Those are long-living variables, so they are
> > not a subject of cleanup.h, are they?
>
> OK, I didn't look at this code in detail, but cleanup would just call
> of_node_put, not actually free the data.
Yes. The nodes should be put either in case of the failure or (if
everything goes fine) at the device unregistration.
--
With best wishes
Dmitry
