Re: [patch 2/2] x86/tsc: Force TSC_ADJUST register to value >= zero

From: Thomas Gleixner <tglx@linutronix.de>
To: LKML <linux-kernel@vger.kernel.org>
Cc: x86@kernel.org, Peter Zijlstra <peterz@infradead.org>,
	Borislav Petkov <bp@alien8.de>,
	Bruce Schlobohm <bruce.schlobohm@intel.com>,
	Roland Scheidegger <rscheidegger_lists@hispeed.ch>,
	Kevin Stanton <kevin.b.stanton@intel.com>,
	Allen Hung <allen_hung@dell.com>,
	stable@vger.kernel.org
Subject: Re: [patch 2/2] x86/tsc: Force TSC_ADJUST register to value >= zero
Date: Fri, 16 Dec 2016 14:33:39 +0100 (CET)	[thread overview]
Message-ID: <alpine.DEB.2.20.1612161410040.3470@nanos> (raw)
In-Reply-To: <alpine.DEB.2.20.1612161121140.3470@nanos>

On Fri, 16 Dec 2016, Thomas Gleixner wrote:
> On Tue, 13 Dec 2016, Thomas Gleixner wrote:
> > Roland reported that his DELL T5810 sports a value add BIOS which
> > completely wreckages the TSC. The squirmware [(TM) Ingo Molnar] boots with
> > random negative TSC_ADJUST values, different on all CPUs. That renders the
> > TSC useless because the sycnchronization check fails.
> 
> While everyone assumed that this is the usual DELL squirmware problem, I
> have to say it's not.
> 
> Just got my hands on a Skylake based Lenovo S510 box and it shows the same
> feature:
> 
> TSC ADJUST: CPU0: -10123656703215
>     	    CPU1: -10123656796701
> 	    CPU2: -10123656797460
> 	    CPU3: -10123656798366
> 
> Which causes the TSC to be out of sync on a stock upstream kernel and the
> TSC deadline timer wreckage is happening on that machine as well.
> 
> I'm pretty sure, that this well thought out feature to 'hide power on time'
> from TSC has not been independently 'invented' by DELL and Lenovo BIOS
> tinkerers.
> 
> I rather have the impression that this is an advisory or feature kit from
> some other entity. Whoever came up with this misfeature at Intel and/or
> Microsoft (sorry, I could not come up with any other suspects) should be
> promoted to run the 'Linux on feature-plagued systems' hot line.

Just to add another data point here.

On cold boot the TSC_ADJUST value on that LENOVO machine is: -24534293,
which is about 9ms.

So assumed that the SDM is correct in this point and the counter starts at
0 after power on, then 9ms later might be right in that magic blob which
does the low level bringup of CPUs. That comes from the CPU vendor and runs
_BEFORE_ the system vendor BIOS can create havoc.

Dealing with timers on x86 feels like a Sisyphean task.

Thanks,

	tglx