exynos-drm registration: infinite loop

[Inki Dae]

On 2014년 11월 06일 18:29, Thierry Reding wrote:
> On Thu, Nov 06, 2014 at 03:23:27PM +0900, Inki Dae wrote:
>> On 2014년 11월 05일 23:38, Thierry Reding wrote:
>>> On Tue, Nov 04, 2014 at 09:18:46PM +0400, Matwey V. Kornilov wrote:
>>>> Hi,
>>>>
>>>> I run 3.18-rc3 kernel on BeagleBone Black. It doesn't have Exynos DRM
>>>> of course, but I run multi-platform kernel where CONFIG_DRM_EXYNOS is
>>>> set to 'y'.
>>>> The issue here is that the platform probe/init goes to infinite loop
>>>> as the following:
>>>>
>>>> [    5.717343] platform exynos-drm: Driver exynos-drm requests probe deferral
>>>> [    5.726848] exynos-drm-ipp exynos-drm-ipp: drm ipp registered successfully.
>>>> [    5.734700] platform exynos-drm: Driver exynos-drm requests probe deferral
>>>> [    5.744044] exynos-drm-ipp exynos-drm-ipp: drm ipp registered successfully.
>>>> [    5.752010] platform exynos-drm: Driver exynos-drm requests probe deferral
>>>> [    5.761377] exynos-drm-ipp exynos-drm-ipp: drm ipp registered successfully.
>>>> [    5.769291] platform exynos-drm: Driver exynos-drm requests probe deferral
>>>>
>>>> It is quite unexpectable behavior. I would expect that the exynos-drm
>>>> failed to initialize with 'no device' error.
>>>>
>>>> See also for the reference: https://bugzilla.kernel.org/show_bug.cgi?id=87691
>>>
>>> The reason for this seems to be that Exynos DRM instantiates a dummy
>>> device that it can bind to (see exynos_drm_init()). Now this code is
>>> executed unconditonally, so the device will be created whether or not
>>> the kernel/module runs on an Exynos SoC. The driver should really bind
>>> to some real device rather than instantiating a dummy. Or if it must
>>> bind to a dummy then the code needs to at least check that it's running
>>> on an Exynos SoC before instantiating the dummy.
>>>
>>
>> Right, I also think the problem is because we use a dummay device for
>> Exynos drm. As you mentioned, this device can always be bound as long as
>> Exynos drm driver is enabled even without real device. So this issue
>> gives us that we need super device node again. Actually, I had already
>> tried the use  of the super device node like below for Exynos drm but
>> other Samsung guys didn't want to use it.
>> 	http://lists.freedesktop.org/archives/dri-devel/2014-April/056618.html
>>
>> I think now we have the reason that we should use super device node so I
>> will try to support the super device for Exynos drm like other ARM drm
>> drivers did. However, for this, I'd like to listen to other opinions
>> for a while before that.
> 
> Irrespective of any long-term solution you need to fix this as soon as
> possible because this currently causes all sorts of weirdness when
> booting a multi-platform kernel on non-Exynos boards. Probably the
> easiest for now would be to add some soc_is_exynos*() check at the very
> beginning of exynos_drm_init().

Got it.

Thanks,
Inki Dae

> 
> Thierry
>