Re: [PATCH] xen/virtio: Don't use grant-dma-ops when running as Dom0

[Jürgen Groß]

On 06.01.26 23:47, Teddy Astie wrote:
> Le 06/01/2026 à 20:06, Jürgen Groß a écrit :
> > On 06.01.26 18:36, Teddy Astie wrote:
> > > Dom0 inherit devices from the machine and is usually in PV mode.
> > > If we are running in a virtual that has virtio devices, these devices
> > > would be considered as using grants with Dom0 as backend, while being
> > > the said Dom0 itself, while we want to use these devices like regular
> > > PCI devices.
> > >
> > > Fix this by preventing grant-dma-ops from being used when running as Dom0
> > > (initial domain). We still keep the device-tree logic as-is.
> > >
> > > Signed-off-by: Teddy Astie <teddy.astie@xxxxxxxxxx>
> > > Fixes: 61367688f1fb0 ("xen/virtio: enable grant based virtio on x86")
> > > ---
> > > CC: Juergen Gross <jgross@xxxxxxxx>
> > > CC: Stefano Stabellini <sstabellini@xxxxxxxxxx>
> > > CC: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
> > > CC: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
> > >
> > >    drivers/xen/grant-dma-ops.c | 3 ++-
> > >    1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/xen/grant-dma-ops.c b/drivers/xen/grant-dma-ops.c
> > > index 14077d23f2a1..c2603e700178 100644
> > > --- a/drivers/xen/grant-dma-ops.c
> > > +++ b/drivers/xen/grant-dma-ops.c
> > > @@ -366,7 +366,8 @@ static int xen_grant_init_backend_domid(struct
> > > device *dev,
> > >        if (np) {
> > >            ret = xen_dt_grant_init_backend_domid(dev, np, backend_domid);
> > >            of_node_put(np);
> > > -    } else if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT) ||
> > > xen_pv_domain()) {
> > > +    } else if (!xen_initial_domain() &&
> > > +           (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT) ||
> > > xen_pv_domain())) {
> > >            dev_info(dev, "Using dom0 as backend\n");
> > >            *backend_domid = 0;
> > >            ret = 0;
> >
> > Please make this controllable, e.g. via a boot parameter.
> >
> > It is completely valid to have a virtio device in dom0 with the backend in
> > a domU. You'll need grants in this case.
> Due to
>   > *backend_domid = 0
>
> Dom0 would always be the backend, unless we introduce a new boot
> parameter to select which domain will be the backend.

Ah, right. Okay, for now this is correct, so you can have my

Reviewed-by: Juergen Gross <jgross@xxxxxxxx>

In future things might change, though.

> There is also another issue, as in the xen_initial_domain() case, all
> PCI devices come from hardware. So no virtio-pci device can't come from
> another domain as Linux would pick up pcifront devices only if we are
> not a Dom0 (!xen_initial_domain()).

I think virtio devices should only be handled via pcifront in case they
are NOT implemented by a Xen backend domain.


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature