Hans de Goede <hdegoede@xxxxxxxxxx> writes: [...] >>> Exposing modaliases for a non supported way of binding the driver does not really seem useful ? >> >> However binding the device and module loading (uevent) uses sometimes >> different pieces. Maybe something changed in kernel, but sometime ago >> certain buses where sending uevent for module loading with one ID (e.g. >> platform or spi bus) but device matching would be according to OF. Thus >> if you did not have entries in spi_device_id, the module would not autoload. >> >> It was exactly the case for example here: >> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c46ed2281bbe4b84e6f3d4bdfb0e4e9ab813fa9d&context=30&ignorews=0&dt=0 >> >> You needed spi_device_id for proper module autoloading. >> >> Unless something change in between in the kernel? > > Looks like your right, the spi_uevent() code always emits "spi:xxxxxxx" style modalias even for dt/of enumerated devices: > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/spi/spi.c#n398 > > So the table needs to stay. > Yeah, and in fact dropping the spi_device_id table will cause the kernel to warn that an spi_device_id entry doesn't exist for a given of_device_id entry since commit 5fa6863ba692 ("spi: Check we have a spi_device_id for each DT compatible"). Fixing that is not trivial because a lot of drivers are rely on current behaviour of the SPI core always returning a spi:<dev> modalias. So don't even have an OF table, even when the SPI devices are instantiated by DT. -- Best regards, Javier Martinez Canillas Core Platforms Red Hat
