Hello Geert, On 8/6/19 9:30 AM, Geert Uytterhoeven wrote: > On Tue, Aug 6, 2019 at 12:48 AM Javier Martinez Canillas > <javierm@xxxxxxxxxx> wrote: >> On 8/1/19 4:17 AM, Masahiro Yamada wrote: >> So I think that we should either: >> >> a) take Kieran's patch or b) remove the i2c_of_match_device_sysfs() fallback >> for OF and require an I2C device table for sysfs instantiation and matching. >> >>> If a driver supports DT and devices are instantiated via DT, >>> in which situation is this useful? >> >> Is useful if you don't have all the I2C devices described in the DT. For example >> a daughterboard with an I2C device is connected to a board through an expansion >> slot or an I2C device connected directly to I2C pins exposed in a machine. >> >> In these cases your I2C devices won't be static so users might want to use the >> sysfs user-space interface to instantiate the I2C devices, i.e: >> >> # echo eeprom 0x50 > /sys/bus/i2c/devices/i2c-3/new_device >> >> as explained in https://github.com/torvalds/linux/blob/master/Documentation/i2c/instantiating-devices#L207 > > Does this actually work with DT names, too? E.g. > > # echo atmel,24c02 > /sys/bus/i2c/devices/i2c-3/new_device > My understanding is that it does. If I'm reading the code correctly the i2c_of_match_device_sysfs() function first attempts to match using both the vendor and device part of the compatible string and if that fails, it strips the vendor part and try to match only using the device part. So you could use any of the following: # echo 24c02 0x50 > /sys/bus/i2c/devices/i2c-3/new_device # echo atmel,24c02 0x50 > /sys/bus/i2c/devices/i2c-3/new_device Best regards, -- Javier Martinez Canillas Software Engineer - Desktop Hardware Enablement Red Hat
