Date:   Thu, 9 Dec 2021 08:44:00 -0800
From:   "Paul E. McKenney" <paulmck@kernel.org>
To:     Gang Li <ligang.bdlg@bytedance.com>
Cc:     john.stultz@linaro.org, linux-kernel@vger.kernel.org,
        sboyd@kernel.org, tglx@linutronix.de, yanghui.def@bytedance.com,
        brookxu.cn@gmail.com, luming.yu@gmail.com,
        songmuchun@bytedance.com, zhengqi.arch@bytedance.com
Subject: Re: [PATCH] Clocksource: Avoid misjudgment of clocksource
Message-ID: <20211209164400.GC641268@paulmck-ThinkPad-P17-Gen-1>
Reply-To: paulmck@kernel.org
References: <CALAqxLWUNFozhfhuVFAPo9xGgO+xsXPQ=i5w1Y0E9-w-PdHXgw@mail.gmail.com>
 <b32b91a9-7cd8-e24b-b608-8bf55a8abe16@bytedance.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b32b91a9-7cd8-e24b-b608-8bf55a8abe16@bytedance.com>
Precedence: bulk

On Thu, Dec 09, 2021 at 09:14:05PM +0800, Gang Li wrote:
> Hi all,
> 
> How about changing clocksource_watchdog from classical timer to hrtimer?
> hrtimer is more accurate and stable, and makes it much easier for us to deal
> with misjudgments.

To make hrtimer more accurate in the busy-system scenarios that are
under test, we would need to specify HRTIMER_MODE_HARD.  Otherwise,
hrtimer handlers execute in softirq, just like timers do.  So without
the HRTIMER_MODE_HARD, the hrtimers would be delayed just as badly as
we are seeing for the normal timers.

Unfortunately, use of HRTIMER_MODE_HARD has response-time consequences,
especially when the timer reads need to be retried.  So my guess is
that hrtimers are not a helpful path forward here.

							Thanx, Paul