XCP PV driver

XCP PV driver

[Soubir Acharya]

The XCP PV driver does not like the SCSI Op codes for RESERVE/RELEASE 
and possibly RESET.

I am trying to install Microsoft Failover Clustering on a HVM with the 
XCP PV drivers installed and disk arbitration is broken.

Is there a registry setting or otherwise to enable this behavior?

Or can code be changed to return success if these opcodes are seen in 
the CDB?

I am fairly certain that this will make clustering work and serve as a 
valuable tool to play with clustered Windows configurations.

Soubir

RE: XCP PV driver

[Paul Durrant]

No, we don't support I_T nexus registration; the model is a simple non-shared SCSI bus.

  Paul

> -----Original Message-----
> From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-
> bounces@lists.xensource.com] On Behalf Of Soubir Acharya
> Sent: 30 January 2011 01:53
> To: xen-devel@lists.xensource.com
> Subject: [Xen-devel] XCP PV driver
> 
> The XCP PV driver does not like the SCSI Op codes for
> RESERVE/RELEASE
> and possibly RESET.
> 
> I am trying to install Microsoft Failover Clustering on a HVM with
> the
> XCP PV drivers installed and disk arbitration is broken.
> 
> Is there a registry setting or otherwise to enable this behavior?
> 
> Or can code be changed to return success if these opcodes are seen
> in
> the CDB?
> 
> I am fairly certain that this will make clustering work and serve as
> a
> valuable tool to play with clustered Windows configurations.
> 
> Soubir
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

RE: XCP PV driver

[James Harper]

> 
> No, we don't support I_T nexus registration; the model is a simple
non-shared
> SCSI bus.
> 

GPLPV tells windows that these SCSIOP's are supported (just returns
STATUS_SUCCESS), but they don't actually do anything. As Paul implies
there is no mechanism to pass such a message back to the actual device
(which could just be a file on disk anyway). It doesn't block anything
else from accessing the device.

Maybe you could use iSCSI in the DomU instead?

James

>   Paul
> 
> > -----Original Message-----
> > From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-
> > bounces@lists.xensource.com] On Behalf Of Soubir Acharya
> > Sent: 30 January 2011 01:53
> > To: xen-devel@lists.xensource.com
> > Subject: [Xen-devel] XCP PV driver
> >
> > The XCP PV driver does not like the SCSI Op codes for
> > RESERVE/RELEASE
> > and possibly RESET.
> >
> > I am trying to install Microsoft Failover Clustering on a HVM with
> > the
> > XCP PV drivers installed and disk arbitration is broken.
> >
> > Is there a registry setting or otherwise to enable this behavior?
> >
> > Or can code be changed to return success if these opcodes are seen
> > in
> > the CDB?
> >
> > I am fairly certain that this will make clustering work and serve as
> > a
> > valuable tool to play with clustered Windows configurations.
> >
> > Soubir
> >
> >
> > _______________________________________________
> > 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

Re: XCP PV driver

[Soubir Acharya]

On 2/1/2011 8:27 AM, James Harper wrote:
>> No, we don't support I_T nexus registration; the model is a simple
> non-shared
>> SCSI bus.
>>
> GPLPV tells windows that these SCSIOP's are supported (just returns
> STATUS_SUCCESS), but they don't actually do anything. As Paul implies
> there is no mechanism to pass such a message back to the actual device
> (which could just be a file on disk anyway). It doesn't block anything
> else from accessing the device.
>
> Maybe you could use iSCSI in the DomU instead?
 >> True, but not as simple as using files out of dom0 as storage.
> James
>
Thanks. There is no reason the XCP driver could not do the same, right?
Shared disks would still be "unsupported" but it would just work on a 
single physical node.
I think the configuration is useful to train/demo/play around with 
Windows configurations requiring shared disks.
Oracle RAC, Fail-over clustering to name a few.
Even a single node cluster is a valid configuration for learning.

Soubir

RE: XCP PV driver

[Paul Durrant]

I guess that succeeding the ops may affect the slew of WHQL tests that the storage driver is expected to pass.
It'd be kind of nice to make it work 'properly' such that a disk image passed into multiple VMS could be shared via MSCS, and we could still pass WHQL. I'll stick it on the bottom of my 'have a look at this' list.

  Paul

> -----Original Message-----
> From: Soubir Acharya [mailto:acharya@kubisys.com]
> Sent: 01 February 2011 14:42
> To: James Harper
> Cc: Paul Durrant; xen-devel@lists.xensource.com
> Subject: Re: [Xen-devel] XCP PV driver
> 
> On 2/1/2011 8:27 AM, James Harper wrote:
> >> No, we don't support I_T nexus registration; the model is a
> simple
> > non-shared
> >> SCSI bus.
> >>
> > GPLPV tells windows that these SCSIOP's are supported (just
> returns
> > STATUS_SUCCESS), but they don't actually do anything. As Paul
> implies
> > there is no mechanism to pass such a message back to the actual
> device
> > (which could just be a file on disk anyway). It doesn't block
> anything
> > else from accessing the device.
> >
> > Maybe you could use iSCSI in the DomU instead?
>  >> True, but not as simple as using files out of dom0 as storage.
> > James
> >
> Thanks. There is no reason the XCP driver could not do the same,
> right?
> Shared disks would still be "unsupported" but it would just work on
> a
> single physical node.
> I think the configuration is useful to train/demo/play around with
> Windows configurations requiring shared disks.
> Oracle RAC, Fail-over clustering to name a few.
> Even a single node cluster is a valid configuration for learning.
> 
> Soubir
> 

Re: XCP PV driver

[Soubir Acharya]

On 2/1/2011 9:49 AM, Paul Durrant wrote:
> I guess that succeeding the ops may affect the slew of WHQL tests that the storage driver is expected to pass.
> It'd be kind of nice to make it work 'properly' such that a disk image passed into multiple VMS could be shared via MSCS, and we could still pass WHQL. I'll stick it on the bottom of my 'have a look at this' list.

Good point. One option would be to return failure as default (so that we 
are OK with WHQL), and then turn the feature on via a registry setting 
or similar?

I can also research the WHQL requirement for disk arbitration.

Soubir

Re: XCP PV driver

[Soubir Acharya]

Is the latest GP PV driver, signed and will it install on 2008 without 
requiring "bcdedit driver signing off"?

Soubir

On 2/1/2011 8:27 AM, James Harper wrote:
>> No, we don't support I_T nexus registration; the model is a simple
> non-shared
>> SCSI bus.
>>
> GPLPV tells windows that these SCSIOP's are supported (just returns
> STATUS_SUCCESS), but they don't actually do anything. As Paul implies
> there is no mechanism to pass such a message back to the actual device
> (which could just be a file on disk anyway). It doesn't block anything
> else from accessing the device.
>
> Maybe you could use iSCSI in the DomU instead?
>
> James
>
>>    Paul
>>
>>> -----Original Message-----
>>> From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-
>>> bounces@lists.xensource.com] On Behalf Of Soubir Acharya
>>> Sent: 30 January 2011 01:53
>>> To: xen-devel@lists.xensource.com
>>> Subject: [Xen-devel] XCP PV driver
>>>
>>> The XCP PV driver does not like the SCSI Op codes for
>>> RESERVE/RELEASE
>>> and possibly RESET.
>>>
>>> I am trying to install Microsoft Failover Clustering on a HVM with
>>> the
>>> XCP PV drivers installed and disk arbitration is broken.
>>>
>>> Is there a registry setting or otherwise to enable this behavior?
>>>
>>> Or can code be changed to return success if these opcodes are seen
>>> in
>>> the CDB?
>>>
>>> I am fairly certain that this will make clustering work and serve as
>>> a
>>> valuable tool to play with clustered Windows configurations.
>>>
>>> Soubir
>>>
>>>
>>> _______________________________________________
>>> 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
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

RE: XCP PV driver

[James Harper]

> 
> Is the latest GP PV driver, signed and will it install on 2008 without
> requiring "bcdedit driver signing off"?
> 

It's test signed. You have to turn testsigning on for 64 bit windows.

James

RE: XCP PV driver

[Ian Campbell]

On Tue, 2011-02-01 at 22:21 +0000, James Harper wrote: 
> > 
> > Is the latest GP PV driver, signed and will it install on 2008 without
> > requiring "bcdedit driver signing off"?
> > 
> 
> It's test signed. You have to turn testsigning on for 64 bit windows.

Is that process documented somewhere? I was trying to install gplpv
drivers on 2008r2 yesterday to try and repro a backend bug and couldn't
get it to work.

http://wiki.xen.org/xenwiki/XenWindowsGplPv/Installing seems to cover an
older more manual installation process which didn't appear to apply to
the .msi I downloaded. There is a link to an MS document on how to
actually test sign a driver but nothing on how to allow test signing in
the OS. The link is now a redirect on the MS end so perhaps the previous
content of that link was more useful?

Pressing F8 on boot and disabling driver signing didn't help, the
installer still aborted with a message about turning off driver signing.

Ian.

RE: XCP PV driver

[James Harper]

> 
> On Tue, 2011-02-01 at 22:21 +0000, James Harper wrote:
> > >
> > > Is the latest GP PV driver, signed and will it install on 2008
without
> > > requiring "bcdedit driver signing off"?
> > >
> >
> > It's test signed. You have to turn testsigning on for 64 bit
windows.
> 
> Is that process documented somewhere? I was trying to install gplpv
> drivers on 2008r2 yesterday to try and repro a backend bug and
couldn't
> get it to work.

bcdedit /set testsigning on

or something like that

> 
> http://wiki.xen.org/xenwiki/XenWindowsGplPv/Installing seems to cover
an
> older more manual installation process which didn't appear to apply to
> the .msi I downloaded. There is a link to an MS document on how to
> actually test sign a driver but nothing on how to allow test signing
in
> the OS. The link is now a redirect on the MS end so perhaps the
previous
> content of that link was more useful?
> 
> Pressing F8 on boot and disabling driver signing didn't help, the
> installer still aborted with a message about turning off driver
signing.
> 

Yeah. Better to not install than to install and break.

James

RE: XCP PV driver

[Ian Campbell]

On Wed, 2011-02-02 at 08:41 +0000, James Harper wrote:
> bcdedit /set testsigning on
> 
> or something like that

Thanks, precisely that command plus a reboot is what was needed.

Ian.

> 
> > 
> > http://wiki.xen.org/xenwiki/XenWindowsGplPv/Installing seems to cover
> an
> > older more manual installation process which didn't appear to apply to
> > the .msi I downloaded. There is a link to an MS document on how to
> > actually test sign a driver but nothing on how to allow test signing
> in
> > the OS. The link is now a redirect on the MS end so perhaps the
> previous
> > content of that link was more useful?
> > 
> > Pressing F8 on boot and disabling driver signing didn't help, the
> > installer still aborted with a message about turning off driver
> signing.
> > 
> 
> Yeah. Better to not install than to install and break.
> 
> James

Re: XCP PV driver

[Pasi Kärkkäinen]

On Wed, Feb 02, 2011 at 08:34:25AM +0000, Ian Campbell wrote:
> On Tue, 2011-02-01 at 22:21 +0000, James Harper wrote: 
> > > 
> > > Is the latest GP PV driver, signed and will it install on 2008 without
> > > requiring "bcdedit driver signing off"?
> > > 
> > 
> > It's test signed. You have to turn testsigning on for 64 bit windows.
> 
> Is that process documented somewhere? I was trying to install gplpv
> drivers on 2008r2 yesterday to try and repro a backend bug and couldn't
> get it to work.
> 

I've been planning to update Gplpv instructions on the wiki, 
since that's very common source of questions on xen-users and on irc..

-- Pasi

> http://wiki.xen.org/xenwiki/XenWindowsGplPv/Installing seems to cover an
> older more manual installation process which didn't appear to apply to
> the .msi I downloaded. There is a link to an MS document on how to
> actually test sign a driver but nothing on how to allow test signing in
> the OS. The link is now a redirect on the MS end so perhaps the previous
> content of that link was more useful?
> 
> Pressing F8 on boot and disabling driver signing didn't help, the
> installer still aborted with a message about turning off driver signing.
> 
> Ian.
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: XCP PV driver

[Ian Campbell]

On Wed, 2011-02-02 at 09:51 +0000, Pasi Kärkkäinen wrote:
> On Wed, Feb 02, 2011 at 08:34:25AM +0000, Ian Campbell wrote:
> > On Tue, 2011-02-01 at 22:21 +0000, James Harper wrote: 
> > > > 
> > > > Is the latest GP PV driver, signed and will it install on 2008 without
> > > > requiring "bcdedit driver signing off"?
> > > > 
> > > 
> > > It's test signed. You have to turn testsigning on for 64 bit windows.
> > 
> > Is that process documented somewhere? I was trying to install gplpv
> > drivers on 2008r2 yesterday to try and repro a backend bug and couldn't
> > get it to work.
> > 
> 
> I've been planning to update Gplpv instructions on the wiki, 
> since that's very common source of questions on xen-users and on irc..

Thanks. I changed the note about test signing to:
        __'''NOTE:'''__ x64 versions of Vista and Server 2008 require
        signed drivers. The GPLPV drivers are only test signed. In order
        to use them you need to enable test signing by running
        "bcdedit /set testsigning on" and rebooting.
which I hope is somewhat accurate.

Ian.

Re: XCP PV driver

[Pasi Kärkkäinen]

On Wed, Feb 02, 2011 at 10:08:58AM +0000, Ian Campbell wrote:
> On Wed, 2011-02-02 at 09:51 +0000, Pasi Kärkkäinen wrote:
> > On Wed, Feb 02, 2011 at 08:34:25AM +0000, Ian Campbell wrote:
> > > On Tue, 2011-02-01 at 22:21 +0000, James Harper wrote: 
> > > > > 
> > > > > Is the latest GP PV driver, signed and will it install on 2008 without
> > > > > requiring "bcdedit driver signing off"?
> > > > > 
> > > > 
> > > > It's test signed. You have to turn testsigning on for 64 bit windows.
> > > 
> > > Is that process documented somewhere? I was trying to install gplpv
> > > drivers on 2008r2 yesterday to try and repro a backend bug and couldn't
> > > get it to work.
> > > 
> > 
> > I've been planning to update Gplpv instructions on the wiki, 
> > since that's very common source of questions on xen-users and on irc..
> 
> Thanks. I changed the note about test signing to:
>         __'''NOTE:'''__ x64 versions of Vista and Server 2008 require
>         signed drivers. The GPLPV drivers are only test signed. In order
>         to use them you need to enable test signing by running
>         "bcdedit /set testsigning on" and rebooting.
> which I hope is somewhat accurate.
> 

That's better.

The really important point is to reboot *before* installing gplpv.

testsigning needs to be enabled when installing gplpv drivers,
and it gets enabled on reboot.

-- Pasi