Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 7 Jun 2023 09:15:40 +0530
From:   Viresh Kumar <viresh.kumar@linaro.org>
To:     Mark Brown <broonie@kernel.org>
Cc:     "Rafael J. Wysocki" <rafael@kernel.org>,
        Shuah Khan <shuah@kernel.org>, linux-pm@vger.kernel.org,
        linux-kselftest@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] selftests/cpufreq: Don't enable generic lock debugging
 options
Message-ID: <20230607034540.ldjugl3g6ryyyi5q@vireshk-i7>
References: <20230605-kselftest-cpufreq-options-v1-1-d4621e0c7cbe@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230605-kselftest-cpufreq-options-v1-1-d4621e0c7cbe@kernel.org>
Precedence: bulk

On 06-06-23, 15:11, Mark Brown wrote:
> Currently the the config fragment for cpufreq enables a lot of generic
> lock debugging.  While these options are useful when testing cpufreq
> they aren't actually required to run the tests and are therefore out of
> scope for the cpufreq fragement, they are more of a thing that it's good
> to enable while doing testing than an actual requirement for cpufreq
> testing specifically.  Having these debugging options enabled,
> especially the mutex and spinlock instrumentation, mean that any build
> that includes the cpufreq fragment is both very much larger than a
> standard defconfig (eg, I'm seeing 35% on x86_64) and also slower at
> runtime.
> 
> This is causing real problems for CI systems.  In order to avoid
> building large numbers of kernels they try to group kselftest fragments
> together, frequently just grouping all the kselftest fragments into a
> single block.  The increased size is an issue for memory constrained
> systems and is also problematic for systems with fixed storage
> allocations for kernel images (eg, typical u-boot systems) where it
> frequently causes the kernel to overflow the storage space allocated for
> kernels.  The reduced performance isn't too bad with real hardware but
> can be disruptive on emulated platforms.
> 
> In order to avoid these issues remove these generic instrumentation
> options from the cpufreq fragment, bringing the cpufreq fragment into
> line with other fragments which generally set requirements for testing
> rather than nice to haves.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
 
> Signed-off-by: Mark Brown <broonie@kernel.org>
> ---
>  tools/testing/selftests/cpufreq/config | 8 --------
>  1 file changed, 8 deletions(-)
> 
> diff --git a/tools/testing/selftests/cpufreq/config b/tools/testing/selftests/cpufreq/config
> index 75e900793e8a..ce5068f5a6a2 100644
> --- a/tools/testing/selftests/cpufreq/config
> +++ b/tools/testing/selftests/cpufreq/config
> @@ -5,11 +5,3 @@ CONFIG_CPU_FREQ_GOV_USERSPACE=y
>  CONFIG_CPU_FREQ_GOV_ONDEMAND=y
>  CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y
>  CONFIG_CPU_FREQ_GOV_SCHEDUTIL=y
> -CONFIG_DEBUG_RT_MUTEXES=y
> -CONFIG_DEBUG_PLIST=y
> -CONFIG_DEBUG_SPINLOCK=y
> -CONFIG_DEBUG_MUTEXES=y
> -CONFIG_DEBUG_LOCK_ALLOC=y
> -CONFIG_PROVE_LOCKING=y
> -CONFIG_LOCKDEP=y
> -CONFIG_DEBUG_ATOMIC_SLEEP=y

FWIW, I enabled these earlier as cpufreq core had a history of races
that are normally not caught without these enabled. But I think we
have come a long way from that and these can be removed now.

-- 
viresh