[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [Xen-users] Xen 4.3.1 / Linux 3.12 panic

  • To: Jan Beulich <JBeulich@xxxxxxxx>
  • From: Wouter de Geus <benv-xensource.com@xxxxxxxxxxxxx>
  • Date: Thu, 7 Nov 2013 14:10:23 +0100
  • Cc: Jinsong Liu <jinsong.liu@xxxxxxxxx>, xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>
  • Comment: DKIM? See http://www.dkim.org
  • Comment: DomainKeys? See http://antispam.yahoo.com/domainkeys
  • Delivery-date: Thu, 07 Nov 2013 13:10:45 +0000
  • Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=default; d=junerules.com; h=Received:Received:Date:From:To:Cc:Subject:Message-ID:Mail-Followup-To:References:MIME-Version:Content-Type:Content-Disposition:In-Reply-To:Organization:X-URL:User-Agent; b=RoUIfF77t6QOcAmMspOUAIix0liJ7LH+fF7KeFMvT7YIZicZl6WUGnJMQi/tBl i7mvvSJvpAOjv7Uu7gpZeip3lQaNzdzJkEjzUkucjJMaoV1p8eVKaDx47JNJxVjx RL04aU7W7KwUu6UDpk/mUVWL09ZUIrGaZ9W2sepKVMN/Q=;
  • List-id: Xen developer discussion <xen-devel.lists.xen.org>
  • Mail-followup-to: Jan Beulich <JBeulich@xxxxxxxx>, Jinsong Liu <jinsong.liu@xxxxxxxxx>, xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>, konrad.wilk@xxxxxxxxxx

* Jan Beulich <JBeulich@xxxxxxxx> [2013-11-07 11:57:17 +0000]:

> > - PowerCap     [P-state 0] (P-state 0 through 4)
> Now we'd need to know what HPC actually means (it means nothing
> to me in this context) - I'd have expected the PowerCap (as referring
> to P-states) to be the interesting one.

Would you like me to test the PowerCap setting? If so, in combination with
the other settings set to what? Note that the PowerCap setting can't be
disabled by itself. (unless P-state 4 counts as disabled?)

According to a faq on supermicro.com (this is a Supermicro board after all)
Q: I noticed that the newer BIOS supporting AMD 6200 series CPUs have a
   P-state HPC Mode option. Can you provide some info on this mode?
A: HPC mode only keeps maximum and minimum states. In system idle mode CPU
   will stay at P4 state for power saving. Once CPU detects higher
   activities, CPU will jump up to P0 or boost state to reduce clock ramp
   up latency.

> In any event - with cpufreq=dom0 and no cpufreq drivers loaded
> in dom0 (which as Konrad says should be the default), there
> shouldn't be any P-state management.

I thought cpufreq=xen was the default - at least according to

> But you being able tom suppress the problem with cpufreq=none
> also suggests that quite likely there's either a problem with the
> silicon, or the PowerNow driver in Xen went sufficiently much out
> of date wrt newer CPUs that it's not usable anymore (it certainly
> hasn't been touched in meaningful ways for quite a while). You
> may have said so before, but can you confirm that under native
> Linux with acpi-cpufreq (or the powernow driver) loaded, you
> don't have this kind of problem? If so, could you please provide
> contents of the respective sysfs nodes?

I started tinkering on this new machine with (Slackware's) linux 3.10.17 kernel
and had no problems whatsoever. The problems only started after booting Xen
with my new custom 3.12 kernel.

I just booted the machine with the 3.12-dom0 kernel without Xen.  And rebooted
since I guess you're interested in the contents with HPC Mode enabled ;)
I've attached the output of dmesg to this email.

Not sure which sysfs nodes you're interested in though, there's:
/sys/module/acpi_cpufreq/parameters/acpi_pstate_strict (contents: 0)

Then per CPU we have /sys/devices/system/cpu/cpu0/cpufreq
with (for CPU 0):
affected_cpus -> 0
bios_limit -> 2600000
cpb -> 1
cpuinfo_cur_freq -> 2600000
cpuinfo_max_freq -> 2600000
cpuinfo_min_freq -> 1400000
cpuinfo_transition_latency -> 5000
freqdomain_cpus -> 0 1
related_cpus -> 0
scaling_available_frequencies -> 2600000 1400000
scaling_available_governors -> conservative ondemand userspace powersave 
scaling_cur_freq -> 2600000
scaling_driver -> acpi-cpufreq
scaling_governor -> performance
scaling_max_freq -> 2600000
scaling_min_freq -> 1400000
scaling_setspeed -> <unsupported>

If there's any other entry you would like to hear please let me know :)
Meanwhile the machine is still stable after going through several kernel
compilations and some heavy I/O (just for testing).



Attachment: dmesg-3.12-native-all-enabled
Description: Text document

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.