Re: [PATCH] of: add dummy of_platform_device_destroy

[Arnd Bergmann]

On Wed, Jan 8, 2020 at 12:04 AM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
>
> On Tue, Jan 7, 2020 at 4:09 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> >
> > The new phy-j721e-wiz driver causes a link failure without CONFIG_OF:
> >
> > drivers/phy/ti/phy-j721e-wiz.o: In function `wiz_remove':
> > phy-j721e-wiz.c:(.text+0x40): undefined reference to `of_platform_device_destroy'
> >
> > Add a dummy version of this function to avoid having to add Kconfig
> > dependencies for the driver.
> >
> > Fixes: 42440de5438a ("phy: ti: j721e-wiz: Add support for WIZ module present in TI J721E SoC")
> > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > ---
> >  include/linux/of_platform.h | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/include/linux/of_platform.h b/include/linux/of_platform.h
> > index 84a966623e78..2551c263e57d 100644
> > --- a/include/linux/of_platform.h
> > +++ b/include/linux/of_platform.h
> > @@ -54,11 +54,16 @@ extern struct platform_device *of_device_alloc(struct device_node *np,
> >                                          struct device *parent);
> >  #ifdef CONFIG_OF
> >  extern struct platform_device *of_find_device_by_node(struct device_node *np);
> > +extern int of_platform_device_destroy(struct device *dev, void *data);
>
> This is already declared, so don't you want to remove the existing one.

Yes, this is what I had intended. I'm surprised there are no warnings when
the compiler sees both an 'extern' and a 'static inline' declaration, but both
clang and gcc seem to accept this, as long as the 'static inline' declaration
comes first.

> >  #else
> >  static inline struct platform_device *of_find_device_by_node(struct device_node *np)
> >  {
> >         return NULL;
> >  }
> > +static inline int of_platform_device_destroy(struct device *dev, void *data)
> > +{
> > +       return 0;
> > +}
>
> I'm curious why this is needed, but of_platform_device_create() is not?

I think what happens in the driver is that this is called from the
probe function
after calling some other interfaces that always return an error without
CONFIG_OF, so the compiler manages to optimize out that call.

I agree it makes sense to treat create the same as remove, I'll send a new
version addressing both comments.

     Arnd