[PATCH] ARM: dts: remove display power domain for exynos5420

[Kukjin Kim]

On 07/14/14 18:09, Rahul Sharma wrote:
> Hi Kukjin,
>
Hi Rahul,

> Please consider this patch for your branch. This patch is important
> for Display for Exynos5 boards.
>
OK, I will take this for exynos5420 display for 3.17...BTW I'm not sure 
this is a regression for 3.16 or not...if so, please let me know.

Thanks,
Kukjin

> Regards,
> Rahul Sharma.
>
> On 9 July 2014 17:00, Rahul Sharma<rahul.sharma at samsung.com>  wrote:
>> Hi Tomasz,
>>
>> On 8 July 2014 21:04, Tomasz Figa<t.figa at samsung.com>  wrote:
>>> Hi Rahul,
>>>
>>> On 07.07.2014 15:37, Rahul Sharma wrote:
>>>> Hi Andrej, Inki,
>>>>
>>>> On 18 June 2014 12:06, Rahul Sharma<rahul.sharma at samsung.com>  wrote:
>>>>> Hi Andrej,
>>>>>
>>>>> On 18 June 2014 11:46, Andrzej Hajda<a.hajda at samsung.com>  wrote:
>>>>>> On 06/17/2014 07:49 AM, Rahul Sharma wrote:
>>>>>>> Hi All,
>>>>>>>
>>>>>>> Please review this patch.
>>>>>>>
>>>>>>> Regards,
>>>>>>> Rahul Sharma
>>>>>>>
>>>>>>> On 9 June 2014 16:58, Rahul Sharma<rahul.sharma at samsung.com>  wrote:
>>>>>>>> Display domain is removed due to instability issues. Explaining
>>>>>>>> the problem below:
>>>>>>>>
>>>>>>>> exynos_init_late triggers the pm_genpd_poweroff_unused which
>>>>>>>> powers off the unused power domains. This call hits before
>>>>>>>> the trigger to deferred probes.
>>>>>>>>
>>>>>>>> DRM DP Panel defers the probe due to supply get failure. By the
>>>>>>>> time, deferred probe is scheduled again, Display Power Domain is
>>>>>>>> powered off by pm_genpd_poweroff_unused.
>>>>>>>>
>>>>>>>> FIMD and DP drivers are accessing registers during Probe and Bind
>>>>>>>> callbacks. If display domain is enabled/disabled around register
>>>>>>>> accesses, display domain gets unstable and we are getting Power
>>>>>>>> Domain Disable fail notification. Increasing the Timeout also
>>>>>>>> didn't help.
>>>>>>
>>>>>> As I understand the problem is that fimd and dp drivers access hw
>>>>>> registers without enabling power domain. So the proper solution is to
>>>>>> fix these drivers.
>>>>>
>>>>> That is also a problem but I fixed those accesses in my local kernel before
>>>>> hitting this issue. If we do register accesses in FIMD/DP probe/bind we
>>>>> observes "Prefetch abort" exception. But here the problem is that 'DP
>>>>> domain disable' starts failing if we enable/disable multiple times.
>>>>>
>>>>>>
>>>>>> Btw. there are already patches removing hw access from probe/bind of
>>>>>> fimd. I guess removing also hw access from dp probe/bind could be a good
>>>>>> solution.
>>>>>
>>>>> Please let me know the links for posted patches. I will test with those patches.
>>>>
>>>> Is there any update on this? Please share the patches which fixes the
>>>> above issue or avoid the above scenario of multiple PM Domain enable/disable.
>>>> I will test them for exynos5 based boards. Otherwise we should get this change
>>>> merged else display will remain broken for exynos5 based boards.
>>>
>>> Andrzej is on holidays right now, so he won't be able to reply in this
>>> thread until he's back. Here are two patches I was able to find on the
>>> related MLs that might be fixing the cause of your issues:
>>
>> Ok. thanks for the update.
>>
>> We should test with Ajay's patches which includes DP probe deferring
>> based on availability of bridge chip. +
>>
>> few patches which cleanup register access in FIMD and DP probe/bind
>> OR
>> few patches to enable Display power domain and clocks just before the
>> register access. As done in https://lkml.org/lkml/2014/7/4/188.
>>
>> Later solution results into display power domain enable failure in case of DRM
>> probe defer.
>>
>> I am just curious if Andrej has some solution for the first approach which can
>> be tested for defer probe and S2R scenarios.
>>
>> Regards,
>> Rahul Sharma.
>>
>>>
>>> [PATCH] drm/exynos: remove hardware overlays disable from fimd probe
>>> (https://www.mail-archive.com/linux-samsung-soc@vger.kernel.org/msg31629.html)
>>>
>>> [PATCH] drm/exynos: fimd: Keep power enabled during fimd_bind
>>> (https://lkml.org/lkml/2014/7/4/188)