Re: [PATCH] virtio_ring: use DMA when memory encryption is active

[Sebastian Hofmann <sebastian@kaemmelot.de>]

> Michael S. Tsirkin <mst@redhat.com> hat am 23.10.2020 17:49 geschrieben:
> 
>  
> On Fri, Oct 23, 2020 at 11:00:54AM +0200, Sebastian Hofmann wrote:
> > > Michael S. Tsirkin <mst@redhat.com> hat am 22.10.2020 13:39 geschrieben:
> > > 
> > >  
> > > On Wed, Oct 21, 2020 at 05:14:25PM +0200, Sebastian Hofmann wrote:
> > > > virtio_ring does not work with active memory encryption because the host cannot read it. Fix this by enforcing the use of DMA which uses shared (unencrypted) memory pages.
> > > > 
> > > > Signed-off-by: Sebastian Hofmann <sebastian@kaemmelot.de>
> > > 
> > > 
> > > Sorry, no.
> > > host which can not access all of driver memory must set VIRTIO_F_ACCESS_PLATFORM.
> > > 
> > > Not worth it to work around broken hosts.
> > > 
> > > Xen is an exception we carry around since it predates the
> > > introduction of VIRTIO_F_ACCESS_PLATFORM.
> > > 
> > > 
> > 
> > Thanks for pointing out VIRTIO_F_ACCESS_PLATFORM which I was not aware of. Maybe that patch was a bit naïve.
> > 
> > Basically I'm looking for a way to use vsock with qemu on AMD SEV. When I try to use IOMMU for vsock I get an EOPNOTSUPP out of vhost_vsock_set_features.
> > 
> > Is there a reason why vhost_vsock_set_features doesn't use vhost_init_device_iotlb as done in the net device? Because that would have been my next attempt.
> > I would appreciate a short comment on this idea or a recommendation for another solution that is better than the patch below.
> 
> Not sure I understand the problem. Are you using qemu? If so just add
> iommu_platform=on and you are done.
> 

That would be nice, but once I set iommu_platform=on (using Linux 5.4 as host and guest, qemu 5.1.0):
qemu-system-x86_64 -enable-kvm -cpu host -machine q35 -nographic -no-user-config -nodefaults -serial stdio \
-global virtio-mmio.force-legacy=off \
-device vhost-vsock-pci,guest-cid=3,disable-legacy=on,iommu_platform=on \
-object sev-guest,id=sev0,cbitpos=47,reduced-phys-bits=5 \
-machine dump-guest-core=off,memory-encryption=sev0 \
[some more arguments...]

...
qemu-system-x86_64: vhost_set_features failed: Operation not supported (95)
qemu-system-x86_64: Error starting vhost: 95
...

Therefore my question if it would be enough to use vhost_init_device_iotlb instead of returning EOPNOTSUPP in vhost_vsock_set_features when VIRTIO_F_ACCESS_PLATFORM is passed. Equivalent to what I see in vhost_net_set_features.
Or maybe I'm missing something important?

> > > > ---
> > > >  drivers/virtio/virtio_ring.c | 5 +++++
> > > >  1 file changed, 5 insertions(+)
> > > > 
> > > > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> > > > index becc77697960..8c68c475ec21 100644
> > > > --- a/drivers/virtio/virtio_ring.c
> > > > +++ b/drivers/virtio/virtio_ring.c
> > > > @@ -12,6 +12,7 @@
> > > >  #include <linux/hrtimer.h>
> > > >  #include <linux/dma-mapping.h>
> > > >  #include <xen/xen.h>
> > > > +#include <linux/mem_encrypt.h>
> > > >  
> > > >  #ifdef DEBUG
> > > >  /* For development, we want to crash whenever the ring is screwed. */
> > > > @@ -255,6 +256,10 @@ static bool vring_use_dma_api(struct virtio_device *vdev)
> > > >  	if (xen_domain())
> > > >  		return true;
> > > >  
> > > > +	/* Memory encryption requires DMA */
> > > > +	if (mem_encrypt_active())
> > > > +		return true;
> > > > +
> > > >  	return false;
> > > >  }
> > > >  
> > > > -- 
> > > > 2.25.1
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization