* Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx> [230413 06:24]: > Hello Tony, > > On Thu, Apr 13, 2023 at 08:12:22AM +0300, Tony Lindgren wrote: > > * Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx> [230406 08:23]: > > > --- a/drivers/i2c/busses/i2c-omap.c > > > +++ b/drivers/i2c/busses/i2c-omap.c > > > @@ -1525,14 +1525,17 @@ static int omap_i2c_remove(struct platform_device *pdev) > > > int ret; > > > > > > i2c_del_adapter(&omap->adapter); > > > - ret = pm_runtime_resume_and_get(&pdev->dev); > > > + > > > + ret = pm_runtime_get_sync(&pdev->dev); > > > > It's better to use pm_runtime_resume_and_get() nowadays in general as > > it does pm_runtime_put_noidle() on errors. > > Sticking to pm_runtime_resume_and_get() complicates the change however, > because the function calls pm_runtime_put_sync() already. So with > pm_runtime_resume_and_get() I'd need > > if (ret >= 0) > pm_runtime_put_sync(&pdev->dev); > > instead of a plain > > pm_runtime_put_sync(&pdev->dev); In that case you still need to do pm_runtime_put_noidle() on errors, so not sure what's the best way here. > > Not sure if there are changes needed here, maybe warn and return early > > if needed? > > The idea of "return early" in a remove callback is exactly what I want > to get rid of. > > See > https://lore.kernel.org/linux-spi/20230317084232.142257-3-u.kleine-koenig@xxxxxxxxxxxxxx > for an example. Oh OK. Care to clarify a bit why we are not allowed to return errors on remove though? Are we getting rid of the return value for remove? Sorry if I'm not following the cunning plan here :) Regards, Tony
