Virtualization - Jan 2022 - [PATCH] virtio_vdpa: Support surprise removal of virtio vdpa device

On Wed, Jan 12, 2022 at 10:23:07AM +0800, Jason Wang
wrote:
> On Tue, Jan 11, 2022 at 7:52 PM Michael S. Tsirkin <mst at
redhat.com> wrote:
> >
> > On Tue, Jan 11, 2022 at 11:36:42AM +0800, ??? wrote:
> > > When virtio vdpa device removed, the abnormal damage of the
device cannot be
> > > perceived normally, which will cause problems similar to:
> > >
> > > 43bb40c5b926
> >
> >
> > Should include the subject of the patch too.
> >
> > > Hence, add the ability to abort the command on surprise removal
> > >
> > > Signed-off-by: dongsheng li <lidongsheng at dayudpu.com>
> >
> > When removing gracefully,
> > I am not sure we should break device unconditionally like this
> > before giving drivers a chance to clean up.
> > Should we just do it for surprise removal?
> 
> That requires a new method to query whether it's a surprise removal.
> 
> Thanks
We can check pci_device_is_present like virtio does.
> >
> > > ---
> > >  drivers/virtio/virtio_vdpa.c | 1 +
> > >  1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/virtio/virtio_vdpa.c
b/drivers/virtio/virtio_vdpa.c
> > > index 4a9ddb44b2a7..fd930409d190 100644
> > > --- a/drivers/virtio/virtio_vdpa.c
> > > +++ b/drivers/virtio/virtio_vdpa.c
> > > @@ -374,6 +374,7 @@ static void virtio_vdpa_remove(struct
vdpa_device *vdpa)
> > >  {
> > >   struct virtio_vdpa_device *vd_dev = vdpa_get_drvdata(vdpa);
> > >
> > > + virtio_break_device(vd_dev->vdev);
> > >   unregister_virtio_device(&vd_dev->vdev);
> > >  }
> > >
> > > --
> > > 2.17.1
> >

On Wed, Jan 12, 2022 at 2:30 PM Michael S. Tsirkin <mst at redhat.com>
wrote:
>
> On Wed, Jan 12, 2022 at 10:23:07AM +0800, Jason Wang wrote:
> > On Tue, Jan 11, 2022 at 7:52 PM Michael S. Tsirkin <mst at
redhat.com> wrote:
> > >
> > > On Tue, Jan 11, 2022 at 11:36:42AM +0800, ??? wrote:
> > > > When virtio vdpa device removed, the abnormal damage of the
device cannot be
> > > > perceived normally, which will cause problems similar to:
> > > >
> > > > 43bb40c5b926
> > >
> > >
> > > Should include the subject of the patch too.
> > >
> > > > Hence, add the ability to abort the command on surprise
removal
> > > >
> > > > Signed-off-by: dongsheng li <lidongsheng at
dayudpu.com>
> > >
> > > When removing gracefully,
> > > I am not sure we should break device unconditionally like this
> > > before giving drivers a chance to clean up.
> > > Should we just do it for surprise removal?
> >
> > That requires a new method to query whether it's a surprise
removal.
> >
> > Thanks
>
> We can check pci_device_is_present like virtio does.
Though the PCI device is the main user, we are vdpa transport that is
unaware of the type of its parent.

So it looks to me we need a new method and PCI parent and do call this
function when needed.

Thanks
>
> > >
> > > > ---
> > > >  drivers/virtio/virtio_vdpa.c | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > >
> > > > diff --git a/drivers/virtio/virtio_vdpa.c
b/drivers/virtio/virtio_vdpa.c
> > > > index 4a9ddb44b2a7..fd930409d190 100644
> > > > --- a/drivers/virtio/virtio_vdpa.c
> > > > +++ b/drivers/virtio/virtio_vdpa.c
> > > > @@ -374,6 +374,7 @@ static void virtio_vdpa_remove(struct
vdpa_device *vdpa)
> > > >  {
> > > >   struct virtio_vdpa_device *vd_dev =
vdpa_get_drvdata(vdpa);
> > > >
> > > > + virtio_break_device(vd_dev->vdev);
> > > >   unregister_virtio_device(&vd_dev->vdev);
> > > >  }
> > > >
> > > > --
> > > > 2.17.1
> > >
>