Re: Re: [PATCH] xen: mask XSAVE in cpuid since we don't allow guests to use it

[Jeremy Fitzhardinge <jeremy@goop.org>]

Andrew Lyon wrote:
> On Wed, Mar 11, 2009 at 4:27 PM, Boris Derzhavets <bderzhavets@yahoo.com> wrote:
>   
>> It's not so important for me. The important thing is:-
>>
>> CentOS 5.2 PV DomU may be loaded only the very first time under Suse's
>> 2.6.27 xen-ified kernel & Xen Unstable ( the most recent). Attempt to
>> shutdown and start it again gives VBD cannot be connected . Hotplug scripts
>> not working. That's a core issue, same behavior as under 2.6.29-rc7 (with
>> XSAVE patch ). Kernel doesn't seem to be a root cause.
>> I believe Xen unstable is broken in some way.
>>
>> Boris.
>>     
>
> This sounds very much like the problem I described in thread subject
> "domain id number on xen unstable", I maintain my own 2.6.27 dom0
> kernel using opensuse Xen patches which I rebase to apply to vanilla
> without the many other patches opensuse usually applies to the kernel
> tree, so from a Xen point of view I am using a very similar kernel to
> yours, I found that I could start a hvm but after shutting it down
> attempting to restart it would fail or hang, sometimes I got hotplug
> error that vbd could not be connected.
>
> The script that has problems on my system is xen-hotplug-cleanup , the
> first time it is run it puts a lock in /var/run/xen-hotplug and never
> removes it, so the next time the script runs it blocks waiting for the
> lock and eventually times out.
>
> The offending line in the script is:
>
> vm=$(xenstore-read "/local/domain/${path_array[2]}/vm")
>
> putting a echo immediately after that line shows that nothing after it
> is executed, which is why the lock is not released.
>
> replacing xen-hotplug-cleanup and xen-hotplug-common with the ones
> from 3.3.1 seems to help, but after starting and stopping a few vm's
> the entire system reboots, so I think some objects are not being
> cleaned up, not surprising really, can hardly expect scripts to work
> with the wrong version of Xen.
>
> This problem has got me stuck into a nasty corner, only Xen unstable
> can fit our virtualization requirements (need viridian for stable
> windows smp), but the Xensource kernel is too old for our hardware.
>
> I am going to put some serious effort into debugging this in the next few days.
>   

Change 19242:226031d62fc5 is the last one which works for me.  I haven't 
managed to bisect which specific change the problem starts occuring in, 
but 19250:b3774712e654 seems like a good starting point.

    J
> Andy
>
>   
>>
>> --- On Wed, 3/11/09, Jeremy Fitzhardinge <jeremy@goop.org> wrote:
>>
>> From: Jeremy Fitzhardinge <jeremy@goop.org>
>> Subject: Re: [Xen-devel] Re: [PATCH] xen: mask XSAVE in cpuid since we don't
>> allow guests to use it
>> To: bderzhavets@yahoo.com
>> Cc: "Xen-devel" <xen-devel@lists.xensource.com>, "Keir Fraser"
>> <keir.fraser@eu.citrix.com>
>> Date: Wednesday, March 11, 2009, 12:09 PM
>>
>> Boris Derzhavets wrote:
>>     
>>> Blktap helps out for multiple CentOS PV DomU restarts (with image on FS)
>>>       
>> under  Suse's 2.6.27 xen-ified kernel & Xen Unstable ( the most recent).
>>     
>>> But it seems not implemented yet for 2.6.29-rc7
>>>
>>>       
>> So your conclusion is that there's a regression in the tools stack when
>> using blkback rather than blktab?
>>
>>    J
>>
>>
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xensource.com
>> http://lists.xensource.com/xen-devel
>>
>>
>>     
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>