PCI hole resize support for 4.5? If not, workaround?

PCI hole resize support for 4.5? If not, workaround?

[Steve Freitas]

Hi all,

I'm suffering from bug 28 [1], unable to assign more than 3.5 gigs of 
RAM to my Windows VM while using VGA passthrough. This is a showstopper 
for me.

Since #28 is still listed as up for grabs for 4.5 [2], I assume a fix is 
not likely to make it in. However, a message from Ian [3] indicated that 
a workaround was available at some point. Will a workaround be offered 
for 4.4 or 4.5?

I'm using Xen 4.4.1 on Debian Jessie (official distro packages), kernel 
3.16, Xeon 5660, Intel 5500 chipset (rev 22), Nvidia Quadro FX 4800 GPU.

Thanks,

Steve

[1] http://bugs.xenproject.org/xen/bug/28
[2] http://marc.info/?l=xen-devel&m=141356622719233
[3] http://marc.info/?l=xen-devel&m=138918385728185

Re: PCI hole resize support for 4.5? If not, workaround?

[Don Slutz]

On 10/18/14 02:17, Steve Freitas wrote:
> Hi all,
>
> I'm suffering from bug 28 [1], unable to assign more than 3.5 gigs of 
> RAM to my Windows VM while using VGA passthrough. This is a 
> showstopper for me.
>
> Since #28 is still listed as up for grabs for 4.5 [2], I assume a fix 
> is not likely to make it in. However, a message from Ian [3] indicated 
> that a workaround was available at some point. Will a workaround be 
> offered for 4.4 or 4.5?
>

Adding the patch:

http://lists.xenproject.org/archives/html/xen-devel/2014-10/msg01447.html

to 4.5-unstable

and adding

mmio_hole_size = 536870912

to your .xfg should allow you to use more memory (the number above
is .5 gig, which should work based on 3.5 gigs working).

    -Don Slutz

> I'm using Xen 4.4.1 on Debian Jessie (official distro packages), 
> kernel 3.16, Xeon 5660, Intel 5500 chipset (rev 22), Nvidia Quadro FX 
> 4800 GPU.
>
> Thanks,
>
> Steve
>
> [1] http://bugs.xenproject.org/xen/bug/28
> [2] http://marc.info/?l=xen-devel&m=141356622719233
> [3] http://marc.info/?l=xen-devel&m=138918385728185
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

Re: PCI hole resize support for 4.5? If not, workaround?

[Steve Freitas]

On 10/20/2014 6:06, Don Slutz wrote:
> On 10/18/14 02:17, Steve Freitas wrote:
>> Hi all,
>>
>> I'm suffering from bug 28 [1], unable to assign more than 3.5 gigs of 
>> RAM to my Windows VM while using VGA passthrough. This is a 
>> showstopper for me.
>>
>> Since #28 is still listed as up for grabs for 4.5 [2], I assume a fix 
>> is not likely to make it in. However, a message from Ian [3] 
>> indicated that a workaround was available at some point. Will a 
>> workaround be offered for 4.4 or 4.5?
>>
>
> Adding the patch:
>
> http://lists.xenproject.org/archives/html/xen-devel/2014-10/msg01447.html
>
> to 4.5-unstable
>
> and adding
>
> mmio_hole_size = 536870912
>
> to your .xfg should allow you to use more memory (the number above
> is .5 gig, which should work based on 3.5 gigs working).

Hi Don!

Thanks for your reply and the patch, my searches didn't turn up your 
post. Any idea if I'll run into trouble trying to apply this to 4.4.1? 
I'd like to stick with (minimally modified) distro packages if possible, 
and Debian Jessie is on 4.4.1. And any feeling for whether this will 
make it into 4.5? Sure would be nice to see this problem fixed for good.

Thanks again.

Steve

Re: PCI hole resize support for 4.5? If not, workaround?

[Don Slutz]

On 10/20/14 13:14, Steve Freitas wrote:
> On 10/20/2014 6:06, Don Slutz wrote:
>> On 10/18/14 02:17, Steve Freitas wrote:
>>> Hi all,
>>>
>>> I'm suffering from bug 28 [1], unable to assign more than 3.5 gigs 
>>> of RAM to my Windows VM while using VGA passthrough. This is a 
>>> showstopper for me.
>>>
>>> Since #28 is still listed as up for grabs for 4.5 [2], I assume a 
>>> fix is not likely to make it in. However, a message from Ian [3] 
>>> indicated that a workaround was available at some point. Will a 
>>> workaround be offered for 4.4 or 4.5?
>>>
>>
>> Adding the patch:
>>
>> http://lists.xenproject.org/archives/html/xen-devel/2014-10/msg01447.html 
>>
>>
>> to 4.5-unstable
>>
>> and adding
>>
>> mmio_hole_size = 536870912
>>
>> to your .xfg should allow you to use more memory (the number above
>> is .5 gig, which should work based on 3.5 gigs working).
>
> Hi Don!
>
> Thanks for your reply and the patch, my searches didn't turn up your 
> post. Any idea if I'll run into trouble trying to apply this to 4.4.1?

This does require a changed QEMU.  QEMU 2.1 (2.1.2 is the current 
released version)
or the QEMU for Xen 
(http://xenbits.xen.org/git-http/qemu-upstream-unstable.git) in
4.5 will work.

> I'd like to stick with (minimally modified) distro packages if 
> possible, and Debian Jessie is on 4.4.1. And any feeling for whether 
> this will make it into 4.5?

I think it still has a good chance.  Just posted v8, name changed from 
mmio_hole_size
to mmio_hole and the units changed from bytes to MBYTES.  So with v8 you
say:

mmio_hole = 512


> Sure would be nice to see this problem fixed for good.

The change to 4.5 is not a complete fix, but a good 1st step.

    -Don Slutz

>
> Thanks again.
>
> Steve
>

Re: PCI hole resize support for 4.5? If not, workaround?

[Steve Freitas]

On 10/21/2014 6:14, Don Slutz wrote:
> On 10/20/14 13:14, Steve Freitas wrote:
>> Sure would be nice to see this problem fixed for good.
>
> The change to 4.5 is not a complete fix, but a good 1st step.

If you don't mind, could you describe the next steps? Might be worth 
adding them as a suggested roadmap to bug #28.

Steve

Re: PCI hole resize support for 4.5? If not, workaround?

[Slutz, Donald Christopher]

On 10/21/14 13:33, Steve Freitas wrote:
> On 10/21/2014 6:14, Don Slutz wrote:
>> On 10/20/14 13:14, Steve Freitas wrote:
>>> Sure would be nice to see this problem fixed for good.
>>
>> The change to 4.5 is not a complete fix, but a good 1st step.
>
> If you don't mind, could you describe the next steps? Might be worth 
> adding them as a suggested roadmap to bug #28.
>

I am not sure of the next steps.  I do know that the code in QEMU 
traditional
would need to be understood.  Also looking through the mailing list
archives for details on why this was not up-stream can help.

    -Don Slutz

> Steve

Re: PCI hole resize support for 4.5? If not, workaround?

[Konrad Rzeszutek Wilk]

On Wed, Oct 29, 2014 at 01:33:56PM +0000, Slutz, Donald Christopher wrote:
> On 10/21/14 13:33, Steve Freitas wrote:
> > On 10/21/2014 6:14, Don Slutz wrote:
> >> On 10/20/14 13:14, Steve Freitas wrote:
> >>> Sure would be nice to see this problem fixed for good.
> >>
> >> The change to 4.5 is not a complete fix, but a good 1st step.
> >
> > If you don't mind, could you describe the next steps? Might be worth 
> > adding them as a suggested roadmap to bug #28.
> >
> 
> I am not sure of the next steps.  I do know that the code in QEMU 
> traditional
> would need to be understood.  Also looking through the mailing list
> archives for details on why this was not up-stream can help.

Perhaps making e820_hole work with HVM guests? That way your guests'
E820 looks like the host - which can help quite much with
finicky PCI devices.

> 
>     -Don Slutz
> 
> > Steve
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel