On Wed, Sep 20, 2023 at 08:51:57AM +0200, Takashi Iwai wrote:
> On Tue, 19 Sep 2023 22:44:28 +0200,
> Nick Desaulniers wrote:
> >
> > On Mon, Sep 18, 2023 at 10:51:29AM +0100, Richard Fitzgerald wrote:
> (snip)
> > > +static void cirrus_scodec_test_set_gpio_ref_arg(struct software_node_ref_args *arg,
> > > + int gpio_num)
> > > +{
> > > + struct software_node_ref_args template =
> > > + SOFTWARE_NODE_REFERENCE(&cirrus_scodec_test_gpio_swnode, gpio_num, 0);
> >
> > I'm observing the following error when building with:
> >
> > $ make LLVM=1 -j128 allmodconfig sound/pci/hda/cirrus_scodec_test.o
> >
> > sound/pci/hda/cirrus_scodec_test.c:151:60: error: initializer element is not a compile-time constant
> > 151 | SOFTWARE_NODE_REFERENCE(&cirrus_scodec_test_gpio_swnode, gpio_num, 0);
> > | ^~~~~~~~
> > /builds/linux/include/linux/property.h:291:37: note: expanded from macro 'SOFTWARE_NODE_REFERENCE'
> > 291 | .nargs = ARRAY_SIZE(((u64[]){ 0, ##__VA_ARGS__ })) - 1, \
> > | ^~~~~~~~~~~
> > /builds/linux/include/linux/kernel.h:57:75: note: expanded from macro 'ARRAY_SIZE'
> > 57 | #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) + __must_be_array(arr))
> > | ^~~
> > /builds/linux/include/linux/compiler.h:228:59: note: expanded from macro '__must_be_array'
> > 228 | #define __must_be_array(a) BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0]))
> > | ^
> > /builds/linux/include/linux/compiler_types.h:366:63: note: expanded from macro '__same_type'
> > 366 | #define __same_type(a, b) __builtin_types_compatible_p(typeof(a), typeof(b))
> > | ^
> > /builds/linux/include/linux/build_bug.h:16:62: note: expanded from macro 'BUILD_BUG_ON_ZERO'
> > 16 | #define BUILD_BUG_ON_ZERO(e) ((int)(sizeof(struct { int:(-!!(e)); })))
> > | ^
>
> Hm, this looks like some inconsistent handling of the temporary array
> passed to ARRAY_SIZE() in the SOFTWARE_NODE_REFERENCE macro. LLVM
> can't treat it if it contains a variable in the given array, while GCC
> doesn't care.
>
> A hackish workaround would be the patch like below, but it's really
> ugly. Ideally speaking, it should be fixed in linux/properties.h, but
> I have no idea how to fix there for LLVM.
>
> Adding more relevant people to Cc.
Thank you, I think it's quite easy to fix. Lemme cook the patch...
--
With Best Regards,
Andy Shevchenko
