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;
Date:   Tue, 13 Jul 2021 15:17:56 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Wei Liu <wei.liu@kernel.org>
Cc:     Ani Sinha <ani@anisinha.ca>, linux-kernel@vger.kernel.org,
        anirban.sinha@nokia.com, mikelley@microsoft.com,
        "K. Y. Srinivasan" <kys@microsoft.com>,
        Haiyang Zhang <haiyangz@microsoft.com>,
        Stephen Hemminger <sthemmin@microsoft.com>,
        Dexuan Cui <decui@microsoft.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
        linux-hyperv@vger.kernel.org
Subject: Re: [PATCH] Hyper-V: fix for unwanted manipulation of sched_clock
 when TSC marked unstable
Message-ID: <20210713131756.GD4170@worktop.programming.kicks-ass.net>
References: <20210713030522.1714803-1-ani@anisinha.ca>
 <20210713130446.gt7k3cwlmhsxtltw@liuwe-devbox-debian-v2>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210713130446.gt7k3cwlmhsxtltw@liuwe-devbox-debian-v2>
Precedence: bulk

On Tue, Jul 13, 2021 at 01:04:46PM +0000, Wei Liu wrote:
> On Tue, Jul 13, 2021 at 08:35:21AM +0530, Ani Sinha wrote:
> > Marking TSC as unstable has a side effect of marking sched_clock as
> > unstable when TSC is still being used as the sched_clock. This is not
> > desirable. Hyper-V ultimately uses a paravirtualized clock source that
> > provides a stable scheduler clock even on systems without TscInvariant
> > CPU capability. Hence, mark_tsc_unstable() call should be called _after_
> > scheduler clock has been changed to the paravirtualized clocksource. This
> > will prevent any unwanted manipulation of the sched_clock. Only TSC will
> > be correctly marked as unstable.
> 
> Correct me if I'm wrong, what you're trying to address is that
> sched_clock remains marked as unstable even after Linux has switched to
> a stable clock source.
> 
> I think a better approach will be to mark the sched_clock as stable when
> we switch to the paravirtualized clock source.

No.. unstable->stable transitions are unsound. You get to switch to your
paravirt clock earlier.