[PROPOSED REMOVAL] PV guest superpage mappings

[PROPOSED REMOVAL] PV guest superpage mappings

[Keir Fraser]

The PV superpage mapping feature has been in the hypervisor for a while now,
but I'm not away of any use of this feature by an upstream guest kernel
(e.g., and primarily of interest, pv_ops Linux). Am I mistaken, or is anyone
looking into or interested in this? If the feature is unused we should
remove it from the hypervisor in this development cycle, as it's untested
and is nothing more than a maintenance burden and security risk. If the
feature is only used by vendor patches, the hypervisor-side implementation
belongs in that vendor's patch queues too.

 -- Keir

Re: [PROPOSED REMOVAL] PV guest superpage mappings

[Dave McCracken]

On Wednesday, July 20, 2011, Keir Fraser wrote:
> The PV superpage mapping feature has been in the hypervisor for a while
> now, but I'm not away of any use of this feature by an upstream guest
> kernel (e.g., and primarily of interest, pv_ops Linux). Am I mistaken, or
> is anyone looking into or interested in this? If the feature is unused we
> should remove it from the hypervisor in this development cycle, as it's
> untested and is nothing more than a maintenance burden and security risk.
> If the feature is only used by vendor patches, the hypervisor-side
> implementation belongs in that vendor's patch queues too.

I submitted a patch to use this feature to Jeremy for inclusion in the Linux 
kernel tree back when it first went into the hypervisor.  I was under the 
impression it was in his tree, and would eventually make its way into mainline 
Linux.

I must admit I moved on to other projects and did not follow it carefully.  
What is the current path for Xen-specific patches to be included in Linux 
mainline?

This feature is important for hugepages to work well in a Xen guest.

Dave McCracken
Oracle Corp.
dave.mccracken@oracle.com

Re: Re: [PROPOSED REMOVAL] PV guest superpage mappings

[Ian Campbell]

On Wed, 2011-07-20 at 17:13 +0100, Dave McCracken wrote:
> On Wednesday, July 20, 2011, Keir Fraser wrote:
> > The PV superpage mapping feature has been in the hypervisor for a while
> > now, but I'm not away of any use of this feature by an upstream guest
> > kernel (e.g., and primarily of interest, pv_ops Linux). Am I mistaken, or
> > is anyone looking into or interested in this? If the feature is unused we
> > should remove it from the hypervisor in this development cycle, as it's
> > untested and is nothing more than a maintenance burden and security risk.
> > If the feature is only used by vendor patches, the hypervisor-side
> > implementation belongs in that vendor's patch queues too.
> 
> I submitted a patch to use this feature to Jeremy for inclusion in the Linux 
> kernel tree back when it first went into the hypervisor.  I was under the 
> impression it was in his tree, and would eventually make its way into mainline 
> Linux.

I can't see it in his 2.6.32.x branch, although it may be elsewhere --
there seems to be a hugetlb branch but I don't know if it is merged
anywhere (nor do I know the magic git runes to find out).

In general patches which were against a Xen branch (rather than
upstream) may or may not have gone upstream depending on how core they
were and how much work it would be to make them suitable for upstream
etc (i.e. you couldn't rely 100% on a maintainer doing that work for
you). More recently we've been trying to encourage contributors to
target upstream directly instead of Xen branches for this reason.

> I must admit I moved on to other projects and did not follow it carefully.  
> What is the current path for Xen-specific patches to be included in Linux 
> mainline?

We need patches against a recent upstream kernel to be posted upstream
and reviewed etc. Once upstream review gives the all clear then a Xen
maintainer can queue them for the next merge window.

Hope that helps.

Ian.

> This feature is important for hugepages to work well in a Xen guest.
> 
> Dave McCracken
> Oracle Corp.
> dave.mccracken@oracle.com
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: [PROPOSED REMOVAL] PV guest superpage mappings

[Keir Fraser]

On 20/07/2011 17:13, "Dave McCracken" <dcm@mccr.org> wrote:

> On Wednesday, July 20, 2011, Keir Fraser wrote:
>> The PV superpage mapping feature has been in the hypervisor for a while
>> now, but I'm not away of any use of this feature by an upstream guest
>> kernel (e.g., and primarily of interest, pv_ops Linux). Am I mistaken, or
>> is anyone looking into or interested in this? If the feature is unused we
>> should remove it from the hypervisor in this development cycle, as it's
>> untested and is nothing more than a maintenance burden and security risk.
>> If the feature is only used by vendor patches, the hypervisor-side
>> implementation belongs in that vendor's patch queues too.
> 
> I submitted a patch to use this feature to Jeremy for inclusion in the Linux
> kernel tree back when it first went into the hypervisor.  I was under the
> impression it was in his tree, and would eventually make its way into mainline
> Linux.
> 
> I must admit I moved on to other projects and did not follow it carefully.
> What is the current path for Xen-specific patches to be included in Linux
> mainline?

You could speak to Konrad about it, who's in your own organisation and might
therefore have similarly aligned interests regarding this feature. I'm
afraid that generally the move has been to the patch authors doing more of
the upstreaming legwork. Otherwise the Xen maintainers end up with piles of
impossible to upstream crud to reimplement themselves.

Apart from Konrad there's also Jeremy of course, and also Stefano Stabellini
and Ian Campbell have been pretty active and may be able to advise.

 -- Keir

> This feature is important for hugepages to work well in a Xen guest.
> 
> Dave McCracken
> Oracle Corp.
> dave.mccracken@oracle.com

Re: [PROPOSED REMOVAL] PV guest superpage mappings

[Dave McCracken]

On Wednesday, July 20, 2011, Keir Fraser wrote:
> You could speak to Konrad about it, who's in your own organisation and
> might therefore have similarly aligned interests regarding this feature.
> I'm afraid that generally the move has been to the patch authors doing
> more of the upstreaming legwork. Otherwise the Xen maintainers end up with
> piles of impossible to upstream crud to reimplement themselves.

Thanks.  I was just chatting with Konrad about it.  I'll work up an upstream 
patch.

Dave McCracken
Oracle Corp.