Re: [PATCH V2] vhost_vdpa: no need to fetch vring base when poweroff

[Jason Wang <jasowang@redhat.com>]

On Tue, Jul 11, 2023 at 12:09 PM Zhu, Lingshan <lingshan.zhu@intel.com> wrote:
>
>
>
> On 7/11/2023 10:50 AM, Jason Wang wrote:
> > On Mon, Jul 10, 2023 at 4:53 PM Zhu Lingshan <lingshan.zhu@intel.com> wrote:
> >> In the poweroff routine, no need to fetch last available index.
> >>
> > This is because there's no concept of shutdown in the vhost layer, it
> > only knows start and stop.
> >
> >> This commit also provides a better debug message in the vhost
> >> caller vhost_virtqueue_stop,
> > A separate patch is better.
> OK
> >
> >> because if vhost does not fetch
> >> the last avail idx successfully, maybe the device does not
> >> suspend, vhost will sync last avail idx to vring used idx as a
> >> work around, not a failure.
> > This only happens if we return a negative value?
> Yes
> >
> >> Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
> >> ---
> >>   hw/virtio/vhost-vdpa.c | 10 ++++++++++
> >>   hw/virtio/vhost.c      |  2 +-
> >>   2 files changed, 11 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> >> index 3c575a9a6e..10b445f64e 100644
> >> --- a/hw/virtio/vhost-vdpa.c
> >> +++ b/hw/virtio/vhost-vdpa.c
> >> @@ -26,6 +26,7 @@
> >>   #include "cpu.h"
> >>   #include "trace.h"
> >>   #include "qapi/error.h"
> >> +#include "sysemu/runstate.h"
> >>
> >>   /*
> >>    * Return one past the end of the end of section. Be careful with uint64_t
> >> @@ -1391,6 +1392,15 @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,
> >>       struct vhost_vdpa *v = dev->opaque;
> >>       int ret;
> >>
> >> +    if (runstate_check(RUN_STATE_SHUTDOWN)) {
> >> +        /*
> >> +         * Some devices do not support this call properly,
> >> +         * and we don't need to retrieve the indexes
> >> +         * if it is shutting down
> >> +         */
> >> +        return 0;
> > Checking runstate in the vhost code seems like a layer violation.
> >
> > What happens without this patch?
> vhost tries to fetch vring base,
> vhost_vdpa needs suspend the device before retrieving last_avail_idx.
> However not all devices can support .suspend properly so this call
> may fail.

I think this is where I'm lost. If the device doesn't support
suspending, any reason we only try to fix the case of shutdown?

Btw, the fail is intended:

    if (!v->suspended) {
        /*
         * Cannot trust in value returned by device, let vhost recover used
         * idx from guest.
         */
        return -1;
    }

And if we return to success here, will we go to set an uninitialized
last avail idx?

    r = dev->vhost_ops->vhost_get_vring_base(dev, &state);
    if (r < 0) {
    ...
    }.else {
        virtio_queue_set_last_avail_idx(vdev, idx, state.num);
    }

Thanks

> Then vhost will print an error shows something failed.
>
> The error msg is confused, as stated in the commit log, restoring
> last_avail_idx with guest used idx
> is a workaround rather than a failure. And no needs to fetch last_avail_idx
> when power off.
>
> Thanks
> >
> > Thanks
> >
> >> +    }
> >> +
> >>       if (v->shadow_vqs_enabled) {
> >>           ring->num = virtio_queue_get_last_avail_idx(dev->vdev, ring->index);
> >>           return 0;
> >> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> >> index 82394331bf..7dd90cff3a 100644
> >> --- a/hw/virtio/vhost.c
> >> +++ b/hw/virtio/vhost.c
> >> @@ -1262,7 +1262,7 @@ void vhost_virtqueue_stop(struct vhost_dev *dev,
> >>
> >>       r = dev->vhost_ops->vhost_get_vring_base(dev, &state);
> >>       if (r < 0) {
> >> -        VHOST_OPS_DEBUG(r, "vhost VQ %u ring restore failed: %d", idx, r);
> >> +        VHOST_OPS_DEBUG(r, "sync last avail idx to the guest used idx for vhost VQ %u", idx);
> >>           /* Connection to the backend is broken, so let's sync internal
> >>            * last avail idx to the device used idx.
> >>            */
> >> --
> >> 2.39.3
> >>
>