Linux Virtualization - May 2019 - [PATCH 06/10] s390/cio: add basic protected virtualization support

On Wed, 15 May 2019 23:08:17 +0200
Halil Pasic <pasic at linux.ibm.com> wrote:
> On Tue, 14 May 2019 10:47:34 -0400
> "Jason J. Herne" <jjherne at linux.ibm.com> wrote:
> > Are we 
> > worried that virtio data structures are going to be a burden on the
31-bit address space?
> > 
> >   
> 
> That is a good question I can not answer. Since it is currently at least
> a page per queue (because we use dma direct, right Mimu?), I am concerned
> about this.
> 
> Connie, what is your opinion?
Yes, running into problems there was one of my motivations for my
question. I guess it depends on the number of devices and how many
queues they use. The problem is that it affects not only protected virt
guests, but all guests.

On Thu, 16 May 2019 08:32:28 +0200
Cornelia Huck <cohuck at redhat.com> wrote:
> On Wed, 15 May 2019 23:08:17 +0200
> Halil Pasic <pasic at linux.ibm.com> wrote:
> 
> > On Tue, 14 May 2019 10:47:34 -0400
> > "Jason J. Herne" <jjherne at linux.ibm.com> wrote:
> 
> > > Are we 
> > > worried that virtio data structures are going to be a burden on
the 31-bit address space?
> > > 
> > >   
> > 
> > That is a good question I can not answer. Since it is currently at
least
> > a page per queue (because we use dma direct, right Mimu?), I am
concerned
> > about this.
> > 
> > Connie, what is your opinion?
> 
> Yes, running into problems there was one of my motivations for my
> question. I guess it depends on the number of devices and how many
> queues they use. The problem is that it affects not only protected virt
> guests, but all guests.
> 
Unless things are about to change only devices that have
VIRTIO_F_IOMMU_PLATFORM are affected. So it does not necessarily affect
not protected virt guests. (With prot virt we have to use
VIRTIO_F_IOMMU_PLATFORM.)

If it were not like this, I would be much more worried.

@Mimu: Could you please discuss this problem with the team? It might be
worth considering to go back to the design of the RFC (i.e. cio/ccw stuff
allocated from a common cio dma pool which gives you 31 bit addressable
memory, and 64 bit dma mask for a ccw device of a virtio device).

Regards,
Halil

On Thu, 16 May 2019 15:42:45 +0200
Halil Pasic <pasic at linux.ibm.com> wrote:
> On Thu, 16 May 2019 08:32:28 +0200
> Cornelia Huck <cohuck at redhat.com> wrote:
> 
> > On Wed, 15 May 2019 23:08:17 +0200
> > Halil Pasic <pasic at linux.ibm.com> wrote:
> >   
> > > On Tue, 14 May 2019 10:47:34 -0400
> > > "Jason J. Herne" <jjherne at linux.ibm.com>
wrote:
> >   
> > > > Are we 
> > > > worried that virtio data structures are going to be a burden
on the 31-bit address space?
> > > > 
> > > >     
> > > 
> > > That is a good question I can not answer. Since it is currently
at least
> > > a page per queue (because we use dma direct, right Mimu?), I am
concerned
> > > about this.
> > > 
> > > Connie, what is your opinion?  
> > 
> > Yes, running into problems there was one of my motivations for my
> > question. I guess it depends on the number of devices and how many
> > queues they use. The problem is that it affects not only protected
virt
> > guests, but all guests.
> >   
> 
> Unless things are about to change only devices that have
> VIRTIO_F_IOMMU_PLATFORM are affected. So it does not necessarily affect
> not protected virt guests. (With prot virt we have to use
> VIRTIO_F_IOMMU_PLATFORM.)
> 
> If it were not like this, I would be much more worried.
If we go forward with this approach, documenting this side effect of
VIRTIO_F_IOMMU_PLATFORM is something that needs to happen.
> 
> @Mimu: Could you please discuss this problem with the team? It might be
> worth considering to go back to the design of the RFC (i.e. cio/ccw stuff
> allocated from a common cio dma pool which gives you 31 bit addressable
> memory, and 64 bit dma mask for a ccw device of a virtio device).
> 
> Regards,
> Halil
>