Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 10 Apr 2015 23:41:52 -0700
From:      Paul Vixie <paul@redbarn.org>
To:        Neel Natu <neelnatu@gmail.com>
Cc:        "freebsd-virtualization@freebsd.org" <freebsd-virtualization@freebsd.org>
Subject:   Re: bhyve clock problem, solved by kern.timecounter.hardware="TSC-low" in /etc/sysctl.conf
Message-ID:  <5528C230.8070406@redbarn.org>
In-Reply-To: <CAFgRE9EqgP%2B4WZdY3%2BJU3dGDTujgwpK5VD9Xd=1y2T-9jssyug@mail.gmail.com>
References:  <552809F4.6070206@redbarn.org> <CAFgRE9EqgP%2B4WZdY3%2BJU3dGDTujgwpK5VD9Xd=1y2T-9jssyug@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help


Neel Natu wrote:
> Hi Paul,
>
> On Fri, Apr 10, 2015 at 10:35 AM, Paul Vixie <paul@redbarn.org> wrote:
>> ...
>>
>> can we make TSC-low the default?
>>
>
> The choice of using the TSC is not without issues:
>
> - As rstone@ points out the TSCs need to be synchronized across physical cpus.

ok, then the reason i'm not seeing this is i have a single 6-core CPU.
(ntpd would complain about clocks going backward.)


> - Depending on system load the guest's estimate of the TSC frequency
> might be way off the mark.

i see what you mean.

> So, in a way the HPET or the ACPI time counters are better since both
> the host and guest agree on the frequency. But it seems there is
> either an issue with the emulation or an artifact due to the
> guest-to-host round trip time.

to be fair, some of my bhyve's ran fast (so, positive adjustments by
ntpd), some ran slow (negative adjustments). i did not try every
possible timecounter hardware; TSC-low fixed it, so i stopped.

> I have been doing experiments but don't have anything conclusive yet.

thanks for looking into it. paravirtualization (kvm-clock or some
equivalent) seems desirable.

-- 
Paul Vixie



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5528C230.8070406>