Date:   Tue, 25 Jun 2019 11:15:57 -0700
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     "Gustavo A. R. Silva" <gustavo@...eddedor.com>,
        Joe Perches <joe@...ches.com>,
        Miguel Ojeda <miguel.ojeda.sandonis@...il.com>,
        Ingo Molnar <mingo@...hat.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Jiri Olsa <jolsa@...hat.com>,
        Namhyung Kim <namhyung@...nel.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Borislav Petkov <bp@...en8.de>,
        "H. Peter Anvin" <hpa@...or.com>,
        "maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT)" <x86@...nel.org>,
        Kan Liang <kan.liang@...ux.intel.com>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Kees Cook <keescook@...omium.org>,
        Shawn Landden <shawn@....icu>,
        Nathan Chancellor <natechancellor@...il.com>,
        Luc Van Oostenryck <luc.vanoostenryck@...il.com>,
        Chandler Carruth <chandlerc@...gle.com>
Subject: Re: [PATCH] perf/x86/intel: Mark expected switch fall-throughs

On Tue, Jun 25, 2019 at 5:47 AM Miguel Ojeda
<miguel.ojeda.sandonis@...il.com> wrote:
>
> On Tue, Jun 25, 2019 at 9:19 AM Peter Zijlstra <peterz@...radead.org> wrote:
> >
> > Can it build a kernel without patches yet? That is, why should I care
> > what LLVM does?

Unreleased versions of Clang built from source can; the latest release
of Clang-8 doesn't have asm goto support required for
CONFIG_JUMP_LABEL.  Things can get complicated based on which kernel
tree/branch (mainline, -next, stable), arch, and configs, but I think
we just have a few long tail bugs left.

We're currently planning multiple output constraint support w/ asm
goto, and have recently implemented things like
__GCC_ASM_FLAG_OUTPUTS__.  If there's other features that we should
start implementing, please let us know.

Give it a shot and let us know what works or doesn't.

For more info, see our
site: https://clangbuiltlinux.github.io/
CI: https://travis-ci.com/ClangBuiltLinux/continuous-integration
Bug tracker: https://github.com/ClangBuiltLinux/linux/issues
wiki: https://github.com/ClangBuiltLinux/linux/wiki

Or reach out to us via
email: clang-built-linux@...glegroups.com
irc: #clangbuiltlinux on chat.freenode.net
or attend our public bi-weekly (once ever 2 weeks, not twice a week) meeting

>
> Having more than a single compiler is always a good idea. You benefit
> from more warnings, more tooling, a second implementation for
> reference/comparison, etc.
>
> As for what is the current state, I think they are close, specially
> for aarch64, but I let Nick, Nathan et. al. answer that! :-) (Cc'd).
> They had a talk in FOSDEM 2019 about it, too.
>
> Also CC'ing Luc since he changed sparse to stop ignoring the attribute
> so that __has_attribute() would work, but I am not sure if there has
> been further work on supporting it properly.
>
> > > Also note that C2x may get [[fallthrough]]. See N2267 and N2335. At
> > > that point, surely tools/IDEs/analyzers will support it :-) The
> > > question is whether we want to wait that long to replace the comments.
> >
> > #define __fallthrough [[fallthrough]]
> >
> > right?
>
> Yes and no. The exact spelling we use does not matter much. My point
> with that paragraph was that since C2x will (maybe) add fallthrough,
> as C++17 did, every compiler/analyzer/IDE/etc. that is still missing
> support for it will have to eventually add it even if they ignore GNU
> attributes. At that point, I would guess most will likely add all
> spellings too.

Regarding __attribute__((fallthrough)), it's currently a work in
progress in Clang:
https://bugs.llvm.org/show_bug.cgi?id=37135
https://reviews.llvm.org/D63260

-- 
Thanks,
~Nick Desaulniers

Powered by blists - more mailing lists