Re: [PATCH] [v2] Kbuild: move to -std=gnu11

[Fangrui Song <maskray@google.com>]

Thanks for the patch!

(Was always wondering which of binutils and kernel would migrate to C99+
earlier... binutils won)

On 2022-02-28, Nick Desaulniers wrote:
>On Mon, Feb 28, 2022 at 2:32 AM Arnd Bergmann <arnd@kernel.org> wrote:
>>
>> From: Arnd Bergmann <arnd@arndb.de>
>>
>> During a patch discussion, Linus brought up the option of changing
>> the C standard version from gnu89 to gnu99, which allows using variable
>> declaration inside of a for() loop. While the C99, C11 and later standards
>> introduce many other features, most of these are already available in
>> gnu89 as GNU extensions as well.
>>
>> An earlier attempt to do this when gcc-5 started defaulting to
>> -std=gnu11 failed because at the time that caused warnings about
>> designated initializers with older compilers. Now that gcc-5.1 is the
>> minimum compiler version used for building kernels, that is no longer a
>> concern. Similarly, the behavior of 'inline' functions changes between
>
>More precisely, the semantics of "extern inline" functions changed
>between ISO C90 and ISO C99.

Perhaps a clearer explanation to readers is: "extern inline" and "inline" swap
semantics with gnu_inline (-fgnu89-inline or __attribute__((__gnu_inline__))).

>That's the only concern I have, which I doubt is an issue. The kernel
>is already covered by the function attribute as you note.
>
>Just to have some measure:
>$ git grep -rn "extern inline" | wc -l
>116

"^inline" behaves like C99+ "extern inline"

Agree this is handled by

     #define inline inline __gnu_inline __inline_maybe_unused notrace

>Most of those are in arch/alpha/ which is curious; I wonder if those
>were intentional.
>
>(I do worry about Makefiles that completely reset KBUILD_CFLAGS
>though; the function attributes still take precedence).
>
>> gnu89 and gnu11, but this was taken care of by defining 'inline' to
>> include __attribute__((gnu_inline)) in order to allow building with
>> clang a while ago.
>>
>> One minor issue that remains is an added gcc warning for shifts of
>> negative integers when building with -Werror, which happens with the
>> 'make W=1' option, as well as for three drivers in the kernel that always
>> enable -Werror, but it was only observed with the i915 driver so far.
>> To be on the safe side, add -Wno-shift-negative-value to any -Wextra
>> in a Makefile.
>>
>> Nathan Chancellor reported an additional -Wdeclaration-after-statement
>> warning that appears in a system header on arm, this still needs a
>> workaround.
>
>Ack; I think we can just fix this in clang.
>
>>
>> The differences between gnu99, gnu11, gnu1x and gnu17 are fairly
>> minimal and mainly impact warnings at the -Wpedantic level that the
>> kernel never enables. Between these, gnu11 is the newest version
>> that is supported by all supported compiler versions, though it is
>> only the default on gcc-5, while all other supported versions of
>> gcc or clang default to gnu1x/gnu17.
>
>I agree. With the fixup to s/Werror/Wextra.
>
>Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
>
>>
>> Link: https://lore.kernel.org/lkml/CAHk-=wiyCH7xeHcmiFJ-YgXUy2Jaj7pnkdKpcovt8fYbVFW3TA@mail.gmail.com/
>> Link: https://github.com/ClangBuiltLinux/linux/issues/1603
>> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
>> Cc: Masahiro Yamada <masahiroy@kernel.org>
>> Cc: linux-kbuild@vger.kernel.org
>> Cc: llvm@lists.linux.dev
>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>
>-- 
>Thanks,
>~Nick Desaulniers
>