On Mon, Aug 8, 2022 at 5:12 PM Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote: > > On Mon, Aug 8, 2022 at 1:41 PM Johannes Pointner <h4nn35.work@xxxxxxxxx> wrote: > > > > Hello, > > > > I just tried to upgrade a am335x based board from 5.10 LTS to 5.15 LTS > > and this broke an application for me because the gpiochip numbering > > changed: > > Don't we have somewhere in the documentation that the GPIO chip device s/device// The GPIO chip device name is something different to what you see in the below dumps. > name should be nothing except name. It seems to me you are relying on > the name to be part of ABI, but it's not. The ABI is the real device > behind no matter what name the associated character device has. To > prevent objection, it's the same as for network cards (old) naming > scheme when eth0 and eth1 can be swapped at boot time. > > > <= 5.10: > > gpio0 == gpiochip0 > > gpio1 == gpiochip32 > > gpio2 == gpiochip64 > > gpio3 == gpiochip96 > > > > > 5.10 > > gpio0 == gpiochip96 > > gpio1 == gpiochip0 > > gpio2 == gpiochip32 > > gpio3 == gpiochip64 > > > > This is the same for /dev/gpiochipX > > Sure. > > ... > > > Did I configure something wrong or do I have to adjust the devicetree > > for the board somehow? > > No, you have to write the application based on `libgpiod` or so with > proper handling of the GPIO chips. Note, that if you run `gpiodetect` > you will see the device instance names for your GPIO chips (IIRC it's > called 'label' in the Linux kernel inside APIs) which should be unique > and that is what your application should rely on. -- With Best Regards, Andy Shevchenko
