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;
Message-ID: <4dff11851bf3fd306ff623bbcc29fca4b2e690e1.camel@suse.cz>
Subject: Re: [PATCH] sched/core: Fix arch_scale_freq_tick() on tickless
 systems
From:   Giovanni Gherdovich <ggherdovich@suse.cz>
To:     Yair Podemsky <ypodemsk@redhat.com>, peterz@infradead.org,
        mingo@redhat.com, juri.lelli@redhat.com,
        vincent.guittot@linaro.org, dietmar.eggemann@arm.com,
        rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
        bristot@redhat.com, vschneid@redhat.com,
        rafael.j.wysocki@intel.com, jlelli@redhat.com, mtosatti@redhat.com,
        nsaenz@kernel.org, linux-kernel@vger.kernel.org
Date:   Wed, 14 Dec 2022 16:35:42 +0100
In-Reply-To: <20221130125121.34407-1-ypodemsk@redhat.com>
References: <20221130125121.34407-1-ypodemsk@redhat.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.34.4 
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Wed, 2022-11-30 at 14:51 +0200, Yair Podemsky wrote:
> In order for the scheduler to be frequency invariant we measure the
> ratio between the maximum cpu frequency and the actual cpu frequency.
> During long tickless periods of time the calculations that keep track
> of that might overflow, in the function scale_freq_tick():
> 
> if (check_shl_overflow(acnt, 2*SCHED_CAPACITY_SHIFT, &acnt))
> »       goto error;
> 
> eventually forcing the kernel to disable the feature for all cpus,
> and show the warning message
> "Scheduler frequency invariance went wobbly, disabling!".
> Let's avoid that by limiting the frequency invariant calculations
> to cpus with regular tick.
> 
> Fixes: e2b0d619b400 ("x86, sched: check for counters overflow in frequency invariant accounting")
> Signed-off-by: Yair Podemsky <ypodemsk@redhat.com>
> Suggested-by: "Peter Zijlstra (Intel)" <peterz@infradead.org>

Acked-by: Giovanni Gherdovich <ggherdovich@suse.cz>

> ---
> V1 -> V2: solution approach was changed from detecting long tickless periods
> to frequency invariant measurements on housekeeping cpus only.
> Link: ee89073a1e9de11c7bd7726eb5da71a0e8795099.camel@redhat.com
> ---
>  kernel/sched/core.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index daff72f00385..1bb0a840c817 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5469,7 +5469,9 @@ void scheduler_tick(void)
>  	unsigned long thermal_pressure;
>  	u64 resched_latency;
>  
> -	arch_scale_freq_tick();
> +	if (housekeeping_cpu(cpu, HK_TYPE_TICK))
> +		arch_scale_freq_tick();
> +
>  	sched_clock_tick();
>  
>  	rq_lock(rq, &rf);