[PATCH v9 20/23] drm/rockchip: Make VOP driver optional

[Andy Yan]

Hi Sacha:

On 4/5/22 17:05, Sascha Hauer wrote:
> On Sat, Apr 02, 2022 at 09:25:33AM +0800, Andy Yan wrote:
>> Hi Sascha:
>>
>> On 4/1/22 20:55, Sascha Hauer wrote:
>>> On Thu, Mar 31, 2022 at 07:00:34PM +0800, Andy Yan wrote:
>>>> Hi:
>>>>
>>>> On 3/31/22 16:18, Sascha Hauer wrote:
>>>>> On Thu, Mar 31, 2022 at 03:20:37PM +0800, Andy Yan wrote:
>>>>>> Hi Sascha:
>>>>>>
>>>>>> On 3/31/22 15:06, Sascha Hauer wrote:
>>>>>>> On Wed, Mar 30, 2022 at 08:50:09PM +0800, Andy Yan wrote:
>>>>>>>> Hi Sascha:
>>>>>>>>
>>>>>>>> On 3/30/22 14:39, Sascha Hauer wrote:
>>>>>>>>> Hi Andy,
>>>>>>>>>
>>>>>>>>> On Tue, Mar 29, 2022 at 07:56:27PM +0800, Andy Yan wrote:
>>>>>>>>>> Hi Sascha:
>>>>>>>>>>
>>>>>>>>>> On 3/28/22 23:11, Sascha Hauer wrote:
>>>>>>>>>>> With upcoming VOP2 support VOP won't be the only choice anymore, so make
>>>>>>>>>>> the VOP driver optional.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
>>>>>>>>>>> ---
>>>>>>>>>>>        drivers/gpu/drm/rockchip/Kconfig            | 8 ++++++++
>>>>>>>>>>>        drivers/gpu/drm/rockchip/Makefile           | 3 ++-
>>>>>>>>>>>        drivers/gpu/drm/rockchip/rockchip_drm_drv.c | 2 +-
>>>>>>>>>>>        3 files changed, 11 insertions(+), 2 deletions(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/gpu/drm/rockchip/Kconfig b/drivers/gpu/drm/rockchip/Kconfig
>>>>>>>>>>> index fa5cfda4e90e3..7d22e2997a571 100644
>>>>>>>>>>> --- a/drivers/gpu/drm/rockchip/Kconfig
>>>>>>>>>>> +++ b/drivers/gpu/drm/rockchip/Kconfig
>>>>>>>>>>> @@ -23,8 +23,16 @@ config DRM_ROCKCHIP
>>>>>>>>>>>        if DRM_ROCKCHIP
>>>>>>>>>>> +config ROCKCHIP_VOP
>>>>>>>>>>> +	bool "Rockchip VOP driver"
>>>>>>>>>>> +	default y
>>>>>>>>>>> +	help
>>>>>>>>>>> +	  This selects support for the VOP driver. You should enable it
>>>>>>>>>>> +	  on all older SoCs up to RK3399.
>>>>>>>>> That reminds me that I wanted to rephrase this. Will change in next
>>>>>>>>> round.
>>>>>>>>>
>>>>>>>>>>> +
>>>>>>>>>>>        config ROCKCHIP_ANALOGIX_DP
>>>>>>>>>>>        	bool "Rockchip specific extensions for Analogix DP driver"
>>>>>>>>>>> +	depends on ROCKCHIP_VOP
>>>>>>>>>> Aanlogix dp is also on vop2 base soc such as  rk356x and rk3588.
>>>>>>> BTW I just looked at the downstream driver. Here we have the same
>>>>>>> situation that the analogix dp driver calls rockchip_drm_wait_vact_end()
>>>>>>> which is implemented in the VOP driver, so when the analogix dp driver
>>>>>>> is actually used on a VOP2 SoC then it is either used in a way that
>>>>>>> rockchip_drm_wait_vact_end() will never be called or it explodes in all
>>>>>>> colours.
>>>>>>>
>>>>>>>>> I added the dependency because analogix_dp-rockchip.c calls
>>>>>>>>> rockchip_drm_wait_vact_end() which is implemented in the VOP driver,
>>>>>>>>> so this driver currenty can't work with the VOP2 driver and can't
>>>>>>>>> be linked without the VOP driver being present.
>>>>>>>>> I'll add a few words to the commit message.
>>>>>>>> Maybe a better direction is move rockchip_drm_wait_vact_end from the VOP
>>>>>>>> driver to rockchip_drm_drv.c
>>>>>>> I am not sure if that's really worth it. Yes, the direction might be the
>>>>>>> right one, but I would really prefer when somebody does the change who
>>>>>>> can test and confirm that the analogix dp really works with VOP2 in the
>>>>>>> end.
>>>>>> If follow this point, the current DW_MIPI also has not been tested for
>>>>>> confirm that it
>>>>>>
>>>>>> can really work with VOP2, so you should also make it depends on
>>>>>> ROCKCHIP_VOP.
> Here you are suggesting to add even more Kconfig dependencies.
>
>>>>> Well at least I have patches here which make DW_MIPI work with VOP2 ;)
>>>> But you DW_MIPI patches for rk356x didn't come. So this is not keep
>>>> consistency with this point.
>>>>
>>>>> What about the others, like LVDS and RGB?
>>>> Yes, we also have other interface , RK356X has LVDS/RGB/BT1120/BT656, RK3588
>>>> has BT1120/BT656, no LVDS or RGB.
>>>>
>>>>>> I think the current solution is just a workaround to make your patch pass
>>>>>> the kernel compile
>>>>> Indeed.
>>>>>
>>>>> I agree that it would be good to add a note somewhere which outputs
>>>>> work with the VOP2 driver (currently only HDMI), but I wonder if Kconfig
>>>>> dependencies is the right place for it, because only people who deliberately
>>>>> disable VOP support will see this information.
>>>>> Maybe we should rather add it to the Kconfig help text?
>>>> If a device is supported for this soc, we will add dt node at the dtsi file.
>>>>
>>>> A Kconfig dependencies don't seems a good idea.
> Here you say Kconfig dependencies are no good idea.


Yes. It's not a good idea. So I don't want to see you use a Kcofig 
dependence

to disable a module to avoid compile which introduced by your patch.

>
>>> Ok, this means we can keep my current approach with just letting
>>> ROCKCHIP_ANALOGIX_DP depend on ROCKCHIP_VOP to avoid having a non
>> Excuse me? How do you get this conclusion ?
> Given that you say that you want to have both more and less Kconfig
> dependencies I came to the conclusion that I only add one where it's
> necessary to compile the driver.
>
>> I said before,  vop and vop2 based platforms both have ROCKCHIP_ANALOGIX_DP.
> Maybe, but vop2 with ROCKCHIP_ANALOGIX_DP doesn't even work in the
> Rockchip downstream kernel, so I wonder how relevant this usecase really
> is.


No, this is not the truth. Rockchip_ANALOGIX_DP of course work with the

vendor kernel. We have many rk356x based products shipped with edp.

Even the VGA output interface on RK3568_EVB1 is drived by

ROCKCHIP_ANALOGIX_DP with a RTD2166 eDP to VGA convert

chip.


So how do you get conclusion that ROCKCHIP_ANALOGIX_DP can't work with

the Rockchip downstream kernel? Is it because you can't make the DP work on

your board? If it is, please contact the supplier who gave you the board.


Do you have a RK3568_EVB1 that has a VGA output interface on board?

If you have it, I can offer you image to verify the DP.


>> If this patch will cause the compile error, please do a real fix, not a
> I can't, because I don't have any hardware to test the Analogix DP on a
> VOP hardware, and given that Analogix DP in conjunction with VOP2 hardware is
> not even supported in the downstream Kernel I am not sure if it's really
> worth doing that.


Again, this is not the truth, see above.

I am not ask you support the ROCKCHIP_ANALOGIX_DP on upstream, I just

want you can give a better solution when you patch cause the compile error.

Disable a module when it conflict with your patch is too rough.

>
> Moving rockchip_drm_wait_vact_end() to rockchip_drm_drv.c doesn't work
> with mainline currently, we first would have to add a struct crtc_funcs
> to struct rockchip_drm_private. Yes, that could be done.
>
>> workaround that may deliver misleading information.
> The Kconfig dependency quite clearly says that the Analogix DP currently
> doesn't work with the VOP2. Anyone who wants to change that can use that
> information as a starting point and implement whatever is necessary and
> likely has the hardware to verify the work. I don't want to solve
> problems that *might* arise in the future, and in this case it's not a
> direction decision that we might regret in the future.
>
> Sascha
>
>
>