Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH] gcov,x86: Mark GCOV broken for x86
To:     Peter Zijlstra <peterz@infradead.org>, x86@kernel.org,
        linux-kernel@vger.kernel.org
Cc:     johannes.berg@intel.com, ndesaulniers@google.com,
        nathan@kernel.org, keescook@chromium.org, elver@google.com,
        mark.rutland@arm.com
References: <YMcssV/n5IBGv4f0@hirez.programming.kicks-ass.net>
From:   Peter Oberparleiter <oberpar@linux.ibm.com>
Message-ID: <2f8a4e21-a09b-8c8d-54ce-45cf2f0e83ff@linux.ibm.com>
Date:   Mon, 14 Jun 2021 16:43:27 +0200
MIME-Version: 1.0
In-Reply-To: <YMcssV/n5IBGv4f0@hirez.programming.kicks-ass.net>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 14.06.2021 12:17, Peter Zijlstra wrote:
> As recently discovered, there is no function attribute to disable the
> -fprofile-generate instrumentation. As such, GCOV is fundamentally
> incompatible with architectures that rely on 'noinstr' for correctness.

Does this problem affect all code or just those pieces that use
'noinstr'? Doing a quick grep over kernel source shows me ~40 source
files containing 'noinstr' vs. ~30000 that don't.

It seems to me like an extreme measure to disable gcov-based profiling
for all files on an architecture when only a small fraction of code
would actually be affected.

I'll gladly admit that I haven't followed the full discussion that lead
to your patch, so maybe some of the following suggestions may already
have been proposed.

What about marking source files that contain 'noinstr' using the

  GCOV_PROFILE_<filename.o> := n

directive that gcov-kernel profiling provides to exclude those files
from being compiled with the corresponding profiling flags? If that's
too much effort there's also a directive for excluding all files in a
directory.

If there was a way to automatically identify 'noinstr'-afflicted source
files (e.g. by grepping the pre-processed source files), one could also
automate this process by adjusting the kbuild-code that adds profiling
flags to automatically exclude such files.

> 
> Until such time as that compilers have added a function attribute to
> disable this instrumentation, mark GCOV as broken.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  arch/x86/Kconfig    | 2 +-
>  kernel/gcov/Kconfig | 4 ++++
>  2 files changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index 86dae426798b..c0f8c9d4c31a 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -75,7 +75,7 @@ config X86
>  	select ARCH_HAS_FAST_MULTIPLIER
>  	select ARCH_HAS_FILTER_PGPROT
>  	select ARCH_HAS_FORTIFY_SOURCE
> -	select ARCH_HAS_GCOV_PROFILE_ALL
> +	select ARCH_HAS_GCOV_BROKEN

Assuming none of the above mentioned alternatives are viable, removing
ARCH_HAS_GCOV_PROFILE_ALL should be enough for your purpose. This way
you are already excluding all source files from automatic profiling on x86.

Users that are absolutely sure that their code can work with
gcov-profiling can manually edit their sub-Makefiles to list those files
that should be instrumented. In my opinion your introduction of
ARCH_HAS_GCOV_BROKEN unnecessarily takes away this capability.


Regards,
  Peter Oberparleiter

-- 
Peter Oberparleiter
Linux on Z Development - IBM Germany