SPI and module auto-loading

[Javier Martinez Canillas]

Hello Mark,

We found an issue with module auto-loading on an I2C driver [0] and it turned
out to be a problem on how the I2C subsystem reports the module alias to
user-space. It always report modalias as "i2c:<dev_id>" even when the driver
is probed via DT. The problem with this particular driver is that the I2C
device ID table didn't contain an entry that matched what the I2C core was
reporting as the MODALIAS uevent env var.

I looked at the SPI core and it does the same. It always reports to udev as
modalias "spi:<dev_id>" so the aliases filled in a module from the OF table
are never used.

This can be easily worked around (and probably why it never was an issue) if
the OF and SPI tables are kept in sync but I don't know if that is a hard
requirement for all use-cases (e.g: a SPI driver that is DT only?).

Now, changing this behavior will break module auto-loading for a lot of
drivers relies on the current behavior and don't define a struct of_device_id
or are not using the MODULE_DEVICE_TABLE(of,..) macro but I wonder if that
means that the drivers are broken and should be fixed?

I'm sending an RFC patch [1] to know what you think about it.

Thanks a lot and best regards,
Javier

[0]: https://lkml.org/lkml/2014/9/11/127
[1]