Re: [PATCH v2 31/37] drm: rcar-du: Add support for the nomodeset kernel parameter

[Thomas Zimmermann <tzimmermann@suse.de>]

[-- Attachment #1.1: Type: text/plain, Size: 5160 bytes --]

Hi

Am 28.01.22 um 12:36 schrieb Laurent Pinchart:
> Hi Thomas,
> 
> On Fri, Jan 28, 2022 at 12:26:03PM +0100, Thomas Zimmermann wrote:
>> Am 28.01.22 um 12:04 schrieb Laurent Pinchart:
>>> On Fri, Jan 28, 2022 at 11:46:49AM +0100, Thomas Zimmermann wrote:
>>>> Am 28.01.22 um 11:34 schrieb Laurent Pinchart:
>>>>> On Fri, Jan 28, 2022 at 10:33:21AM +0100, Thomas Zimmermann wrote:
>>>>>> Am 28.01.22 um 10:13 schrieb Kieran Bingham:
>>>>>>> Quoting Javier Martinez Canillas (2021-12-17 00:37:46)
>>>>>>>> According to disable Documentation/admin-guide/kernel-parameters.txt, this
>>>>>>>> parameter can be used to disable kernel modesetting.
>>>>>>>>
>>>>>>>> DRM drivers will not perform display-mode changes or accelerated rendering
>>>>>>>> and only the system framebuffer will be available if it was set-up.
>>>>>>>
>>>>>>> What is the 'system framebuffer' in this instance? Reading
>>>>>>> https://www.kernel.org/doc/Documentation/admin-guide/kernel-parameters.txt
>>>>>>> it sounds like that means anything already set up by the bootloader.
>>>>>>
>>>>>> Exactly this.
>>>>>>
>>>>>>>> But only a few DRM drivers currently check for nomodeset, make this driver
>>>>>>>> to also support the command line parameter.
>>>>>>>>
>>>>>>>> Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
>>>>>>>> ---
>>>>>>>>
>>>>>>>> (no changes since v1)
>>>>>>>>
>>>>>>>>      drivers/gpu/drm/rcar-du/rcar_du_drv.c | 3 +++
>>>>>>>>      1 file changed, 3 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/drivers/gpu/drm/rcar-du/rcar_du_drv.c b/drivers/gpu/drm/rcar-du/rcar_du_drv.c
>>>>>>>> index 5a8131ef81d5..982e450233ed 100644
>>>>>>>> --- a/drivers/gpu/drm/rcar-du/rcar_du_drv.c
>>>>>>>> +++ b/drivers/gpu/drm/rcar-du/rcar_du_drv.c
>>>>>>>> @@ -701,6 +701,9 @@ static struct platform_driver rcar_du_platform_driver = {
>>>>>>>>      
>>>>>>>>      static int __init rcar_du_init(void)
>>>>>>>>      {
>>>>>>>> +       if (drm_firmware_drivers_only())
>>>>>>>> +               return -ENODEV;
>>>>>>>> +
>>>>>>>
>>>>>>> This will completely disable all control of the display device when
>>>>>>> nomodeset is enabled.
>>>>>>>
>>>>>>> Is there any requirement for us to support outputting to the display if
>>>>>>> it was previously set up? presumably without setting or changing any
>>>>>>> modes, but simply allowing the existing frame to be updated?
>>>>>>
>>>>>> There's no requirement for your driver. We just want a parameter where
>>>>>> we can conveniently disable most of DRM's drivers and reduce it to a
>>>>>> minimum. Helps distributions to provide a simple fallback mode.  Most
>>>>>> PCI-based drivers already support that. Now we're added it to the other
>>>>>> drivers as well.
>>>>>>
>>>>>>>
>>>>>>> I think the implication is that 'firmware drivers' would mean a display
>>>>>>> could be updated through some firmware interface, which we won't have
>>>>>>> ... so it seems reasonable to accept that this whole driver can be
>>>>>>> disabled in that instance.
>>>>>>
>>>>>> It cannot be 'mode-setted'. We get a pre-configured framebuffer from the
>>>>>> firmware or bootloader. Whatever we draw there shows up on the screen.
>>>>>
>>>>> I doubt that's going to work as you expect, clocks and regulators will
>>>>> get disabled at boot if not used by any driver.
>>>>
>>>> Simpledrm and simplefb attach to these firmware framebuffers. Both
>>>> drivers look at the device tree nodes to acquire the relevant clocks and
>>>> regulators.
>>>
>>> How about clocks and regulators for the ancillary devices, such as
>>> encoders, or in the R-Car case, the external composer handled by the
>>> vsp1 driver (in drivers/media/platform/vsp1) ?
>>>
>>> This approach may work fine on x86 desktop systems, but for ARM-based
>>> devices, the situation is usually more complex.
>>
>> Well, in that case the problem has always been there. We don't make it
>> worse.
> 
> What's the point of adding nomodeset support in those drivers then, if
> it's known not to work ?

But nomodeset is know to work: it disables the driver. The behavior is 
consistent among all hardware-specific drivers.

Javier mentioned that there are kernel parameters to keep unused 
clocks/regulators working.

 From the distribution POV, if we have users with broken graphics 
drivers we want to give them an easy workaround until we find a 
solution. Passing 'nomodeset ...' on the command line gives us that.

Best regards
Thomas

> 
>>>>>>> Reading your mail that brought this thread up in my inbox, I think
>>>>>>> you've already hit merge on this, so don't worry about adding a tag in
>>>>>>> that instance, but I think this is ok.
>>>>>>>
>>>>>>> Reviewed-by: Kieran Bingham <kieran.bingham+renesas@ideasonboard.com>
>>>>>>>
>>>>>>>>             rcar_du_of_init(rcar_du_of_table);
>>>>>>>>      
>>>>>>>>             return platform_driver_register(&rcar_du_platform_driver);
> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]