Re: [PATCH 1/3] ASoC: cs4271: Fix cs4271 I2C and SPI drivers automatic module loading

[Alexander Sverdlin <alexander.sverdlin@gmail.com>]

Hi Mark, Wolfram and all,

On Fri, 2025-10-17 at 16:01 +0100, Mark Brown wrote:
> On Fri, Oct 17, 2025 at 04:41:53PM +0200, Alexander Sverdlin wrote:
> 
> > indeed, that's what I've got for a fake I2C device:
> 
> > # cat /sys/bus/i2c/devices/0-0010/uevent 
> > OF_NAME=cs4271
> > OF_FULLNAME=/soc/i2c@4000000/cs4271@10
> > OF_COMPATIBLE_0=cirrus,cs4271
> > OF_COMPATIBLE_N=1
> > MODALIAS=of:Ncs4271T(null)Ccirrus,cs4271
> 
> > to me it looks like a bug somewhere in I2C core...
> 
> IIRC this has been round the loop a few times and whatever you do
> something breaks.

it turns out it has been changed back in v4.18, with commit af503716ac14
("i2c: core: report OF style module alias for devices registered via OF"), 

but the change is not symmetric with SPI, and SPI core doesn't call 
of_device_uevent_modalias().

Not only is this inconsistent with SPI, but I anticipate yet another problem
if SPI would also adopt the same mechanism:

for instance there are many CODEC drivers in sound/soc, which have
SPI and I2C backends modelled as distinct modules.

In the current situation most of them have MODULE_DEVICE_TABLE(i2c,
and are therefore broken regarding module autoloading.

"Reparing" them as Herve proposed would result in I2C modules being
loaded only via "of:" style modalias and SPI still via "spi:". Which
sounds all but consistent.

If SPI ever adopts the same of_device_uevent_modalias(), both backends
would require "of:" prefixed modalias, and it will not be possible to
load the proper one for the corresponding bus type.

What are your thoughs on this?

-- 
Alexander Sverdlin.