Message-ID: <CAKwvOdkeEqb0HnHoggpwmHiKtDo06VN0N_r9ffv0rzigNLvXUg@mail.gmail.com>
Date:   Tue, 12 Jun 2018 13:19:58 -0700
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     hpa@...or.com
Cc:     sedat.dilek@...il.com, Arnd Bergmann <arnd@...db.de>,
        Andrew Morton <akpm@...ux-foundation.org>, mingo@...hat.com,
        Thomas Gleixner <tglx@...utronix.de>,
        linux-efi@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>,
        x86@...nel.org, virtualization@...ts.linux-foundation.org,
        Alistair Strachan <astrachan@...gle.com>,
        Manoj Gupta <manojgupta@...gle.com>,
        Greg Hackmann <ghackmann@...gle.com>, tstellar@...hat.com,
        Kees Cook <keescook@...gle.com>,
        Masahiro Yamada <yamada.masahiro@...ionext.com>,
        Michal Marek <michal.lkml@...kovi.net>,
        Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>,
        geert@...ux-m68k.org, Will Deacon <will.deacon@....com>,
        mawilcox@...rosoft.com, David Rientjes <rientjes@...gle.com>,
        acme@...hat.com, Philippe Ombredanne <pombredanne@...b.com>,
        Andrey Ryabinin <aryabinin@...tuozzo.com>,
        Kate Stewart <kstewart@...uxfoundation.org>,
        boris.ostrovsky@...cle.com, "J. Kiszka" <jan.kiszka@...mens.com>,
        rostedt@...dmis.org, kirill.shutemov@...ux.intel.com,
        Ard Biesheuvel <ard.biesheuvel@...aro.org>,
        akataria@...are.com, brijesh.singh@....com, Cao.jin@...or.com
Subject: Re: [PATCH v4 1/3] compiler-gcc.h: add gnu_inline to all inline declarations

On Tue, Jun 12, 2018 at 11:51 AM H. Peter Anvin <hpa@...or.com> wrote:
>
> <caoj.fnst@...fujitsu.com>,Greg KH <gregkh@...uxfoundation.org>,jarkko.sakkinen@...ux.intel.com,jgross@...e.com,Josh Poimboeuf <jpoimboe@...hat.com>,Matthias Kaehlcke <mka@...omium.org>,thomas.lendacky@....com,Thiebaud Weksteen <tweek@...gle.com>,mjg59@...gle.com,joe@...ches.com
> From: hpa@...or.com
> Message-ID: <191E4EBE-4CB2-4C8B-AB61-689A91FFE7A8@...or.com>
>
> On June 12, 2018 11:33:14 AM PDT, Nick Desaulniers <ndesaulniers@...gle.com> wrote:
> >On Fri, Jun 8, 2018 at 3:04 AM Sedat Dilek <sedat.dilek@...il.com>
> >wrote:
> >>
> >> On Fri, Jun 8, 2018 at 9:59 AM, Arnd Bergmann <arnd@...db.de> wrote:
> >> > On Thu, Jun 7, 2018 at 10:49 PM, Nick Desaulniers
> >> > <ndesaulniers@...gle.com> wrote:
> >> >> Functions marked extern inline do not emit an externally visible
> >> >> function when the gnu89 C standard is used. Some KBUILD Makefiles
> >> >> overwrite KBUILD_CFLAGS. This is an issue for GCC 5.1+ users as
> >without
> >> >> an explicit C standard specified, the default is gnu11. Since c99,
> >the
> >> >> semantics of extern inline have changed such that an externally
> >visible
> >> >> function is always emitted. This can lead to multiple definition
> >errors
> >> >> of extern inline functions at link time of compilation units whose
> >build
> >> >> files have removed an explicit C standard compiler flag for users
> >of GCC
> >> >> 5.1+ or Clang.
> >> >>
> >> >> Signed-off-by: Nick Desaulniers <ndesaulniers@...gle.com>
> >> >> Suggested-by: H. Peter Anvin <hpa@...or.com>
> >> >> Suggested-by: Joe Perches <joe@...ches.com>
> >> >
> >> > I suspect this will break Geert's gcc-4.1.2, which I think doesn't
> >have that
> >> > attribute yet (4.1.3 or higher have it according to the
> >documentation.
> >> >
> >> > It wouldn't be hard to work around that if we want to keep that
> >version
> >> > working, or we could decide that it's time to officially stop
> >supporting
> >> > that version, but we should probably decide on one or the other.
> >
> >Heh, so earlier we decided against compiler flags (-std=gnu89 or
> >-fgnu89-inline) in preference to function attributes.  The function
> >attribute is preferable as some of the Makefiles [accidentally?]
> >overwrite KBUILD_CFLAGS, which is problematic for gcc 5.1 users as the
> >implicit c standard used was changed to gnu11 from gnu89.  What's nice
> >is that to support gcc 4.1 users, we simply don't need to add any
> >attribute, as their implicit c standard is gnu89 which has the
> >semantics for extern inline that we want.  I have a simple change to
> >this patch that can support users of various gcc versions, see below:
> >
> >> Good point.
> >> What is the minimum requirement of GCC version currently?
> >> AFAICS x86/asm-goto support requires GCC >= 4.5?
> >
> >Yes, but that's only for x86, IIUC.  It seems the kernel may have
> >different minimum required versions of GCC based on arch then?  That
> >may be ok, but I'm not sure that's easy to keep track of without
> >having it explicitly stated somewhere like the docs perhaps?
> >
> >> Just FYI...
> >> ...saw the last days in upstream commits that kbuild/kconfig for
> >> 4.18-rc1 offers possibilities to check for cc-version dependencies.
> >
> >Those will be helpful.  If we want to pursue compiler flags, which get
> >set some Makefiles, then yes.  But I think a simpler change to my
> >patch would be as below.
> >
> >It seems gcc did not get __has_attribute [0] until 5.1, but will
> >define __GNUC_GNU_INLINE__ if supported. [1]  Unfortunately, Clang
> >does not define __GNUC_GNU_INLINE__ [2].  So a proper feature test
> >might look like:
> >
> >```
> >#ifndef __has_attribute
> >#define __has_attribute(x) 0
> >#endif
> >
> >#if defined(__GNUC_GNU_INLINE__) || __has_attribute(gnu_inline)
> >#define __gnu_inline __attribute__(gnu_inline)
> >#endif
> >
> >#define inline inline __attribute__((always_inline, unused)) notrace
> >__gnu_inline
> >```
> >
> >Thoughts on this approach? I can send a v5 tomorrow if there's no
> >major issues.  Feedback appreciated, as always.
> >
> >[0] https://clang.llvm.org/docs/LanguageExtensions.html#has-attribute
> >[1]
> >https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#Common-Function-Attributes
> >[2] https://bugs.llvm.org/show_bug.cgi?id=37784
>
> Please fix clang. It isn't all that hard to fix.
>
> However, __GCC_GNU_INLINE__ means you are in GNU mode by default, on gcc's new enough to have multiple misses.
>
> The right thing to look for is __GCC_STDC_INLINE__ in which case you need the attribute.

Oh, by [0] __GCC_STDC_INLINE__ is indeed actually the correct symbol
to check for.  Clang does support that, so nothing to fix there.

> By the way, you should check clang against gcc's predefined macros by doing:
>
> gcc [options] -x c -Wp,-dM -E /dev/null | sort
>
> Options can change the predefined macros substantially, especially the, -std=, arch and -O options. -x c can be replaced with e.g. -x c++, objective-c, assembler-with-cpp etc.

Neat, I'll have to bookmark that incantation.  I can s/gcc/clang/ to
get a similar list (which is how I know it supports
__GCC_STDC_INLINE__).

Patch now becomes something like:

#ifdef __GNUC_GNU_INLINE__
#define __gnu_inline __attribute__((gnu_inline))
#else
#define __gnu_inline
#endif

#define inline inline __attribute__((always_inline, unused)) notrace
__gnu_inline
...

Issues with that approach?

[0] https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html
-- 
Thanks,
~Nick Desaulniers

Powered by blists - more mailing lists