Re: linux-next: build warning after merge of the akpm-current tree

[Marco Elver <elver@google.com>]

On Tue, 15 Sep 2020 at 06:03, 'David Gow' via kasan-dev
<kasan-dev@googlegroups.com> wrote:
>
> [+kasan-dev, +kunit-dev]
>
> On Mon, Sep 14, 2020 at 3:01 PM Stephen Rothwell <sfr@canb.auug.org.au> wrote:
> >
> > Hi all,
> >
> > After merging the akpm-current tree, today's linux-next build (x86_64
> > allmodconfig) produced this warning:
> >
> > In file included from lib/test_kasan_module.c:16:
> > lib/../mm/kasan/kasan.h:232:6: warning: conflicting types for built-in function '__asan_register_globals'; expected 'void(void *, long int)' [-Wbuiltin-declaration-mismatch]
> >   232 | void __asan_register_globals(struct kasan_global *globals, size_t size);
> >       |      ^~~~~~~~~~~~~~~~~~~~~~~
> > lib/../mm/kasan/kasan.h:233:6: warning: conflicting types for built-in function '__asan_unregister_globals'; expected 'void(void *, long int)' [-Wbuiltin-declaration-mismatch]
> >   233 | void __asan_unregister_globals(struct kasan_global *globals, size_t size);
> >       |      ^~~~~~~~~~~~~~~~~~~~~~~~~
> > lib/../mm/kasan/kasan.h:235:6: warning: conflicting types for built-in function '__asan_alloca_poison'; expected 'void(void *, long int)' [-Wbuiltin-declaration-mismatch]
> >   235 | void __asan_alloca_poison(unsigned long addr, size_t size);
> >       |      ^~~~~~~~~~~~~~~~~~~~
> > lib/../mm/kasan/kasan.h:236:6: warning: conflicting types for built-in function '__asan_allocas_unpoison'; expected 'void(void *, long int)' [-Wbuiltin-declaration-mismatch]
> >   236 | void __asan_allocas_unpoison(const void *stack_top, const void *stack_bottom);
> >       |      ^~~~~~~~~~~~~~~~~~~~~~~
> > lib/../mm/kasan/kasan.h:238:6: warning: conflicting types for built-in function '__asan_load1'; expected 'void(void *)' [-Wbuiltin-declaration-mismatch]
> >   238 | void __asan_load1(unsigned long addr);
> >       |      ^~~~~~~~~~~~
> [...some more similar warnings truncated...]
>
> Whoops -- these are an issue with the patch: the test_kasan_module.c
> file should be built with -fno-builtin. I've out a new version of the
> series which fixes this:
> https://lore.kernel.org/linux-mm/20200915035828.570483-1-davidgow@google.com/T/#t
>
> Basically, the fix is just:
>
> diff --git a/lib/Makefile b/lib/Makefile
> index 8c94cad26db7..d4af75136c54 100644
> --- a/lib/Makefile
> +++ b/lib/Makefile
> @@ -69,6 +69,7 @@ obj-$(CONFIG_KASAN_KUNIT_TEST) += test_kasan.o
>  CFLAGS_test_kasan.o += -fno-builtin
>  CFLAGS_test_kasan.o += $(call cc-disable-warning, vla)
>  obj-$(CONFIG_TEST_KASAN_MODULE) += test_kasan_module.o
> +CFLAGS_test_kasan_module.o += -fno-builtin
>  obj-$(CONFIG_TEST_UBSAN) += test_ubsan.o
>  CFLAGS_test_ubsan.o += $(call cc-disable-warning, vla)
>  UBSAN_SANITIZE_test_ubsan.o := y

That's reasonable, given it's already done for test_kasan.o.

Although the warnings only occur because it's including
"../mm/kasan/kasan.h", which include declarations for the
instrumentation functions. AFAIK, those declarations only exist to
avoid missing-declaration warnings; in which case all of them could
just be moved above their definitions in generic.c (which would also
avoid some repetition for the ones defined with macros). But given the
various other KASAN patches in-flight, to avoid conflicts let's leave
this as-is, but it's something to improve in case we wanted to get rid
of the fno-builtin.

Thanks,
-- Marco