[PATCH] drm/bridge: adv7511: Do not merge adv7511_mode_set() with atomic_enable()

[Tommaso Merciai]

Hi All,

On 26/05/25 16:28, Laurent Pinchart wrote:
> On Mon, May 26, 2025 at 04:13:23PM +0200, Tommaso Merciai wrote:
>> On 26/05/25 16:02, Tommaso Merciai wrote:
>>> On 26/05/25 15:18, Dmitry Baryshkov wrote:
>>>> On 26/05/2025 14:40, Maxime Ripard wrote:
>>>>> On Mon, May 26, 2025 at 01:19:23PM +0200, Tommaso Merciai wrote:
>>>>>> On 26/05/25 12:49, Laurent Pinchart wrote:
>>>>>>> On Mon, May 26, 2025 at 11:58:37AM +0200, Tommaso Merciai wrote:
>>>>>>>> On 26/05/25 11:26, Maxime Ripard wrote:
>>>>>>>>> On Mon, May 26, 2025 at 10:54:52AM +0200, Tommaso Merciai wrote:
>>>>>>>>>> After adv7511_mode_set() was merged into .atomic_enable(), only the
>>>>>>>>>> native resolution is working when using modetest.
>>>>>>>>>>
>>>>>>>>>> This is caused by incorrect timings: adv7511_mode_set() must not be
>>>>>>>>>> merged into .atomic_enable().
>>>>>>>>>>
>>>>>>>>>> Move adv7511_mode_set() back to the .mode_set() callback in
>>>>>>>>>> drm_bridge_funcs to restore correct behavior.
>>>>>>>>>>
>>>>>>>>>> Fixes: 0a9e2f0a6466 ("drm/bridge: adv7511: switch to the HDMI
>>>>>>>>>> connector helpers")
>>>>>>>>>> Reported-by: Biju Das <biju.das.jz at bp.renesas.com>
>>>>>>>>>> Closes: https://lore.kernel.org/all/aDB8bD6cF7qiSpKd@tom-desktop/
>>>>>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai.xr at bp.renesas.com>
>>>>>>>>>
>>>>>>>>> Explaining why, both in the commit log and the comments, would be
>>>>>>>>> nice.
>>>>>>>>> Because I can't think of any good reason it just can't work for that
>>>>>>>>> bridge.
>>>>>>>>
>>>>>>>> Sorry, let me clarify and share with you some details:
>>>>>>>>
>>>>>>>> adv7511_mode_set:
>>>>>>>>      - Is setting up timings registers for the DSI2HDMI bridge in
>>>>>>>> our case
>>>>>>>>        we are using ADV7535 bridge.
>>>>>>>>
>>>>>>>> rzg2l_mipi_dsi_atomic_enable:
>>>>>>>>      - Is setting up the vclock for the DSI ip
>>>>>>>>
>>>>>>>> Testing new/old implementation a bit we found the following:
>>>>>>>>
>>>>>>>> root at smarc-rzg3e:~# modetest -M rzg2l-du -d -s HDMI-
>>>>>>>> A-1:800x600-56.25 at XR24
>>>>>>>> setting mode 800x600-56.25Hz on connectors HDMI-A-1, crtc 62
>>>>>>>> [   49.273134] adv7511_mode_set_old: drm_mode_vrefresh(mode) = 56
>>>>>>>> [   49.281006] rzg2l_mipi_dsi_atomic_enable: mode->clock: 36000
>>>>>>>>
>>>>>>>> root at smarc-rzg3e:~# modetest -M rzg2l-du -d -s HDMI-
>>>>>>>> A-1:800x600-56.25 at XR24
>>>>>>>> setting mode 800x600-56.25Hz on connectors HDMI-A-1, crtc 62
>>>>>>>> [   74.076881] rzg2l_mipi_dsi_atomic_enable: mode->clock: 36000
>>>>>>>> [   74.092130] adv7511_mode_set: drm_mode_vrefresh(adj_mode) = 56
>>>>>>>>
>>>>>>>> Same result but different timing (in function call perspective):
>>>>>>>>
>>>>>>>>      - old: adv7511_mode_set() is call before
>>>>>>>> rzg2l_mipi_dsi_atomic_enable()
>>>>>>>>      - new: adv7511_mode_set() is call after
>>>>>>>> rzg2l_mipi_dsi_atomic_enable()
>>>>>>>
>>>>>>> What is "old" and "new" here ? Is it before and after Dmitry's
>>>>>>> patch, or
>>>>>>> before and after yours ? Please be precise when describing problems.
>>>>>>
>>>>>> Sorry, you are completely right:
>>>>>>
>>>>>>    - old --> before Dmitry's patch
>>>>>>    - new --> after Dmitry's patch
>>>>>>
>>>>>>>
>>>>>>>> What do you think? Thanks in advance.
>>>>>>>
>>>>>>> You're only explaining above what the "old" and "new" behaviours are,
>>>>>>> and claiming one of them is causing an issue, but you're not
>>>>>>> explaining
>>>>>>> *why* it causes an issue. That's what your commit message is
>>>>>>> expected to
>>>>>>> detail.
>>>>>>>
>>>>>>
>>>>>> Thanks for the clarification! :)
>>>>>> I will send v2 explaining better this.
>>>>>
>>>>> In particular, if the driver needs to have mode_set called before
>>>>> atomic_enable, you should say why moving the call to mode_set earlier in
>>>>> the function wouldn't work.
>>>>
>>>> It might be the same thing as we had on PS8640: it had to be brought
>>>> up before the host starts the DSI link, so that there is no clock
>>>> input on the DSI clock lane.
>>>>
>>>
>>> Some updates on my side:
>>>
>>> I'm not seeing any differences from a regs perspective when using the
>>> old driver version (before Dmitry's patch) and the new driver version
>>> (after Dmitry's patch).
>>>
>>> In particular, i2cdump -f -y 7 0x4c shows me the same result.
>>
>> Please ignore this (wrong address)
>>
>> The right test is: i2cdump -f -y 7 0x3d
>>
>> And I'm seeing the following differences:
>>
>> # WORK:
>> reg | val
>> 0x3d → 0x00
>> 0x3e → 0x00
>>
>> # DON't WORK
>> reg | val
>> 0x3d → 0x10
>> 0x3e → 0x40
>>
>>> Unfortunately, since I don't have the ADV7535 datasheet, I believe this
>>> issue may be related to the functions call sequence.
> 
> You could try to get the documentation from Analog Devices.
> 
> This being said, the above registers are documented in the ADV7511
> programming guide, which is publicly available. They may differ in the
> ADV7535 though.
> 
>>> I agree with Dmitry's theory.
>>>
>>> Let me gently know if you need some more test on my side. Thanks in
>>> advance.
> 

FYI, I've tested the following pipeline:

DU1 (RGB Output) -> adv7513 -> HDMI Panel

All working fine on my side with the Dmitry's patch.
Same driver, But broken on DSI interface(ADV7535)

Thanks & Regards,
Tommaso