On Mon, Nov 7, 2022 at 10:07 AM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote: > What the code does is to not check the return value from > devm_gpiod_get() and then avoid using an erroneous GPIO descriptor > with IS_ERR_OR_NULL(). > > This will miss real errors from the GPIO core that should not be > ignored, such as probe deferral. > > Instead request the GPIO as explicitly optional, which means that > if it doesn't exist, the descriptor returned will be NULL. > > Then we can add error handling and also avoid just doing this on > the device tree path, and simplify the site where the optional > GPIO descriptor is used. > > There were some problems with cleaning up this GPIO descriptor > use in the past, but this is the proper way to deal with it. > > Cc: Rafał Miłecki <rafal@xxxxxxxxxx> > Cc: Chuhong Yuan <hslester96@xxxxxxxxx> > Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx> Actually I think this is something of a regression that needs a Cc: stable tag. On my router (DIR-890L) this is needed for the USB ports to probe and work, I think due to recent refactorings involving device links and whatnot, i.e. probe deferral happens all the time and the deferral error has to be handled. I think many BCM systems are affected. Yours, Linus Walleij
