Re: [PATCH 00/60] drm: Make all drivers to honour the nomodeset parameter

[Thomas Zimmermann <tzimmermann@suse.de>]

[-- Attachment #1.1: Type: text/plain, Size: 3080 bytes --]

Hi

Am 15.12.21 um 09:50 schrieb Javier Martinez Canillas:
> Hello Thomas,
> 
> On 12/15/21 09:31, Thomas Zimmermann wrote:
>> Hi Javier,
>>
>> nothing wrong with your patches, but I'd like to propose slightly
>> differnt solution.
>>
>> For many USB drivers, you put the drm_firmware_drivers_only() call into
>> the probe function. For registering, these drivers use
>> module_usb_driver(), which expands to generic device-register functions.
>>
>> I'd like to propose a similar macro for DRM drivers that includes the
>> test for drm_firmware_drivers_only().
>>
>> In drm_drv.h :
>>
>>     #if defined(USB)
>>     static int drm_usb_register(struct usb_driver *usb)
>>     {
>> 	if (drm_firmware_drivers_only())
>> 		return -ENODEV;
>> 	return usb_register_driver(usb);
>>     }
>>     #define drm_module_usb_driver(__usb)
>> 	module_driver(drm_usb_register, usb_deregister)
>>     #endif
>>
>> In each of the USB-based DRM drivers, replace module_usb_driver with
>> drm_module_usb_driver.
>>
>> And then there's PCI [3] and platform drivers, [4] which can be handled
>> similarly. Many PCI drivers open-code the module init and device
> 
> Thanks for the suggestion. I actually thought about this approach as well,
> since seems unnecessary to have all that duplicated logic in every driver.
> 
> But at the end decided to just do the less intrusive change, because can't
> do any testing for most of the drivers.
> 
>> registering with the driver-specific enable parameter. Maybe adding a
>> driver-specific register function would make sense.
>>
> 
> If I understood you correctly, suggested something similar in #dri-devel but
> danvet's suggestion [0] was to keep it simple:
> 
> 16:32 javierm: danvet: yes, I don't think we would be able to do it at this point. I'll add a function pointer
>                         param to the check function so drivers can also define their own check besides nomodeset
> 16:33 danvet: javierm, uh that sounds a bit like overkill? just keep that part of the check in drivers?
> 16:33 javierm: danvet: Ok. Wanted to get rid of the duplicated code but I'm indeed over engineering this :)

An extra pointer sounds indeed like over-engineering. If we take ast for 
example, I'd just do something like this:

   static ast_pci_register_driver(struct pci_driver *pci)
   {
	if (drm_firmware_drivers_only() && ast_modeset == -1)
		return -ENODEV;
	if (ast_modeset == 0)
		return -ENODEV;

	return pci_register_driver(pci);
   }

   module_driver(&ast_pci_driver, ast_pci_register_driver,
		pci_unregister_driver)

That removes some of the boiler-plate module code without changing the 
driver's behavior.

Best regards
Thomas

> 
> [0]: https://people.freedesktop.org/~cbrill/dri-log/?channel=dri-devel&date=2021-11-02
> 
>> Best regards
>> Thomas
>>
> Best regards,
> 

-- 
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 --]