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:   Tue, 15 Mar 2022 09:33:08 +0800
From:   Feng Tang <feng.tang@intel.com>
To:     Nicolas Saenz Julienne <nsaenzju@redhat.com>
Cc:     tglx@linutronix.de, andi.kleen@intel.com, bp@alien8.de,
        dave.hansen@intel.com, hpa@zytor.com, len.brown@intel.com,
        linux-kernel@vger.kernel.org, mingo@redhat.com, paulmck@kernel.org,
        peterz@infradead.org, rui.zhang@intel.com, tim.c.chen@intel.com,
        x86@kernel.org, mtosatti@redhat.com, frederic@kernel.org
Subject: Re: [PATCH v3 1/2] x86/tsc: add a timer to make sure tsc_adjust is
 always checked
Message-ID: <20220315013308.GA36349@shbuild999.sh.intel.com>
References: <20211117023751.24190-1-feng.tang@intel.com>
 <20220314175207.274870-1-nsaenzju@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220314175207.274870-1-nsaenzju@redhat.com>
Precedence: bulk

Hi Nicolas,

Thanks for raising it.

On Mon, Mar 14, 2022 at 06:52:07PM +0100, Nicolas Saenz Julienne wrote:
> Hi Feng, Thomas,
> 
> > On Wed, Nov 17, 2021 at 10:37:51AM +0800, Feng Tang wrote:
> > Normally the tsc_sync will get checked every time system enters idle state,
> > but Thomas Gleixner mentioned there is still a caveat that a system won't
> > enter idle [1], either because it's too busy or configured purposely to not
> > enter idle. Setup a periodic timer (every 10 minitues) to make sure the
> > check is always on.
> > 
> > [1]. https://lore.kernel.org/lkml/875z286xtk.fsf@nanos.tec.linutronix.de/
> > Signed-off-by: Feng Tang <feng.tang@intel.com>
> 
> I can see this timer interrupting my system's nohz_full CPUs. It'd be nice to
> be able to avoid the noise. A solution is using 'tsc=reliable', but IIUC this
> is not what the flag was created for. Any ideas/suggestions?
 
This patch is about correctness. And yes, as you said, the 'tsc=reliable'
works. Another thought is to leverage the 'housekeeping_mask' to exclude
the isolated nohz_full CPUs, but the 'caveat' what the patch try to solve
remains.

Also before this patchset, the clocksource watchdog timer is fired every
500ms, which should interrupt nohz_full CUPS more? 

Thanks,
Feng

> Regards,
> Nicolas