Re: [PATCH] IOMMU and ATS not supported by vhost-user filesystem.

[Stefan Hajnoczi]

On Thu, Jan 28, 2021 at 12:41:15PM -0300, Leonardo Augusto Guimarães Garcia 
wrote:
> On 1/27/21 8:19 AM, Stefan Hajnoczi wrote:
> > On Tue, Jan 26, 2021 at 03:23:38PM -0300, lagarcia@linux.ibm.com wrote:
> > +
> > +   if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
> > +       !(fs->vhost_dev.hdev_features & (1ull << VIRTIO_F_IOMMU_PLATFORM))) 
> > {
> > +       error_setg(errp, "IOMMU is not supported by the vhost-user device 
> > backend");
> > +       goto err_iommu_needed;
> > +   }
> > 
> > Also, can this logic be made generic for all vhost-user devices? It's
> > not really specific to vhost-user-fs.
> 
> 
> I am afraid I will need some additional guidance on how to do that. If I am
> reading the code correctly, the vhost-user devices don't follow any specific
> pattern. Looking at the vhost-user-fs code path, we have:
> 
> vuf_device_realize -> vhost_dev_init -> vhost_user_backend_init
> 
> So, I thought that naturally, if the check was in vuf_device_realize on my
> previous patch, I should move it to vhost_user_backend_init. However, in
> order to check if the VirtIODevice has been specified with the
> VIRTIO_F_IOMMU_PLATFORM option, I would need to have access to the
> VirtIODevice inside vhost_user_backend_init, which currently is not
> available and not one of the arguments of vhost_backend_init virtual
> function it implements. vhost_dev_init doesn't have access to VirIODevices
> as well. Looking at other device types that call vhost_dev_init, not all of
> them initialized the VirtIODevice by the time vhost_dev_init is called (e.g.
> cryptodev-host). Hence, adding a VirtIODevice as parameter to vhost_dev_init
> and vhost_backedn_init is not a feasible solution. vhost_dev_init does
> receive structu vhost_dev which has a field VirtIODevice vdev. But as the
> VirtIODevice hasn't been initialized by the time vhost_dev_init is called on
> all vhost-user devices today also makes this not a solution.
> 
> Any ideas on this? Is it correct for a virtio-user devices to call
> vhost_dev_init before having VirtIODevice ready?

Maybe Michael Tsirkin has an idea. Otherwise let's go with a
vhost-user-fs fix.

Stefan

signature.asc
Description: PGP signature