On Tue, Feb 2, 2021 at 9:06 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> On Tue, Feb 2, 2021 at 8:20 PM Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
> > This reverts commit a06cd8cf97a3 ("ASoC: da7218: skip of_device_id table
> > when !CONFIG_OF") because we want to make of_match_device() stop using
> > of_match_ptr() internally, confusing compilers and causing ifdef
a> > pollution.
> >
> > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > Cc: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> > Acked-by: Arnd Bergmann <arnd@xxxxxxxx>
> > Cc: Mark Brown <broonie@xxxxxxxxxx>
> > Signed-off-by: Stephen Boyd <swboyd@xxxxxxxxxxxx>
>
> Thanks for your patch!
>
> > ---
> >
> > Changes from v1:
> > * Dropped of_match_ptr() in driver too
> >
> > sound/soc/codecs/da7218.c | 4 +---
> > 1 file changed, 1 insertion(+), 3 deletions(-)
> >
> > diff --git a/sound/soc/codecs/da7218.c b/sound/soc/codecs/da7218.c
> > index 2bfafbe9e3dc..83cace9d0139 100644
> > --- a/sound/soc/codecs/da7218.c
> > +++ b/sound/soc/codecs/da7218.c
> > @@ -2278,14 +2278,12 @@ static irqreturn_t da7218_irq_thread(int irq, void *data)
> > * DT
> > */
> >
> > -#ifdef CONFIG_OF
> > static const struct of_device_id da7218_of_match[] = {
> > { .compatible = "dlg,da7217", .data = (void *) DA7217_DEV_ID },
> > { .compatible = "dlg,da7218", .data = (void *) DA7218_DEV_ID },
> > { }
> > };
> > MODULE_DEVICE_TABLE(of, da7218_of_match);
> > -#endif
> >
> > static inline int da7218_of_get_id(struct device *dev)
> > {
> > @@ -3311,7 +3309,7 @@ MODULE_DEVICE_TABLE(i2c, da7218_i2c_id);
> > static struct i2c_driver da7218_i2c_driver = {
> > .driver = {
> > .name = "da7218",
> > - .of_match_table = of_match_ptr(da7218_of_match),
> > + .of_match_table = da7218_of_match,
>
> This does mean the compiler can no longer optimize the table away
> in the CONFIG_OF=n case. Is that intentional?
It already couldn't do that because of the other reference.
In general, I don't see much value in optimizing the match table
away regardless, since most drivers that have one do not work
without a devicetree anyway. It seems that this could in theory
work using a locally defined platform device, but I don't see any
such users in the mainline kernel.
Arnd
