lkml.org 
[lkml]   [2015]   [Aug]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRE: [tip:x86/platform] x86/hyperv: Mark the Hyper-V TSC as unstable
Date


> -----Original Message-----
> From: Peter Zijlstra [mailto:peterz@infradead.org]
> Sent: Friday, August 21, 2015 7:34 AM
> To: mingo@kernel.org; Haiyang Zhang <haiyangz@microsoft.com>;
> tglx@linutronix.de; hpa@zytor.com; vkuznets@redhat.com; KY Srinivasan
> <kys@microsoft.com>; torvalds@linux-foundation.org; linux-
> kernel@vger.kernel.org
> Cc: linux-tip-commits@vger.kernel.org
> Subject: Re: [tip:x86/platform] x86/hyperv: Mark the Hyper-V TSC as
> unstable
>
> On Fri, Aug 21, 2015 at 12:45:50AM -0700, tip-bot for Vitaly Kuznetsov wrote:
> > Commit-ID: 88c9281a9fba67636ab26c1fd6afbc78a632374f
> > Gitweb:
> https://na01.safelinks.protection.outlook.com/?url=http%3a%2f%2fgit.kern
> el.org%2ftip%2f88c9281a9fba67636ab26c1fd6afbc78a632374f&data=01%7c01
> %7ckys%40microsoft.com%7cca840f9ebba34958dc9a08d2aa35a6c4%7c72f988
> bf86f141af91ab2d7cd011db47%7c1&sdata=hzE233wXaLAYShWPE18SQJQvQe
> JguAdB6oj%2bF%2blx7Ok%3d
> > Author: Vitaly Kuznetsov <vkuznets@redhat.com>
> > AuthorDate: Wed, 19 Aug 2015 09:54:24 -0700
> > Committer: Ingo Molnar <mingo@kernel.org>
> > CommitDate: Fri, 21 Aug 2015 08:44:38 +0200
> >
> > x86/hyperv: Mark the Hyper-V TSC as unstable
> >
> > The Hyper-V top-level functional specification states, that
> > "algorithms should be resilient to sudden jumps forward or
> > backward in the TSC value", this means that we should consider
> > TSC as unstable. In some cases tsc tests are able to detect the
> > instability, it was detected in 543 out of 646 boots in my
> > testing:
> >
> > Measured 6277 cycles TSC warp between CPUs, turning off TSC clock.
> > tsc: Marking TSC unstable due to check_tsc_sync_source failed
> >
> > This is, however, just a heuristic. On Hyper-V platform there
> > are two good clocksources: MSR-based hyperv_clocksource and
> > recently introduced TSC page.
>
> *groan*.. and where are the paravirt ops like pv_time_ops for hyperv to
> fix up this mess?

We register external clocksource and clockevents devices based on a hypervisor based time source and
timers to deal with this issue.

Regards,

K. Y


\
 
 \ /
  Last update: 2015-08-21 18:01    [W:0.046 / U:0.112 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site