On Fri, Jul 8, 2022 at 7:53 PM Eugenio Perez Martin [off-list ref] wrote:

On Wed, Jun 29, 2022 at 6:16 AM Jason Wang [off-list ref] wrote:

quoted

On Fri, Jun 24, 2022 at 12:08 AM Eugenio Pérez [off-list ref] wrote:

quoted

The ioctl adds support for suspending the device from userspace.

This is a must before getting virtqueue indexes (base) for live migration,
since the device could modify them after userland gets them. There are
individual ways to perform that action for some devices
(VHOST_NET_SET_BACKEND, VHOST_VSOCK_SET_RUNNING, ...) but there was no
way to perform it for any vhost device (and, in particular, vhost-vdpa).

After a successful return of the ioctl call the device must not process
more virtqueue descriptors. The device can answer to read or writes of
config fields as if it were not suspended. In particular, writing to
"queue_enable" with a value of 1 will not make the device start
processing buffers of the virtqueue.

Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
---
 drivers/vhost/vdpa.c       | 19 +++++++++++++++++++
 include/uapi/linux/vhost.h | 14 ++++++++++++++
 2 files changed, 33 insertions(+)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index 3d636e192061..7fa671ac4bdf 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c

@@ -478,6 +478,22 @@ static long vhost_vdpa_get_vqs_count(struct vhost_vdpa *v, u32 __user *argp)
        return 0;
 }

+/* After a successful return of ioctl the device must not process more
+ * virtqueue descriptors. The device can answer to read or writes of config
+ * fields as if it were not suspended. In particular, writing to "queue_enable"
+ * with a value of 1 will not make the device start processing buffers.
+ */
+static long vhost_vdpa_suspend(struct vhost_vdpa *v)
+{
+       struct vdpa_device *vdpa = v->vdpa;
+       const struct vdpa_config_ops *ops = vdpa->config;
+
+       if (!ops->suspend)
+               return -EOPNOTSUPP;
+
+       return ops->suspend(vdpa);
+}
+
 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
                                   void __user *argp)
 {

@@ -654,6 +670,9 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
        case VHOST_VDPA_GET_VQS_COUNT:
                r = vhost_vdpa_get_vqs_count(v, argp);
                break;
+       case VHOST_VDPA_SUSPEND:
+               r = vhost_vdpa_suspend(v);
+               break;
        default:
                r = vhost_dev_ioctl(&v->vdev, cmd, argp);
                if (r == -ENOIOCTLCMD)

diff --git a/include/uapi/linux/vhost.h b/include/uapi/linux/vhost.h
index cab645d4a645..6d9f45163155 100644
--- a/include/uapi/linux/vhost.h
+++ b/include/uapi/linux/vhost.h

@@ -171,4 +171,18 @@
 #define VHOST_VDPA_SET_GROUP_ASID      _IOW(VHOST_VIRTIO, 0x7C, \
                                             struct vhost_vring_state)

+/* Suspend or resume a device so it does not process virtqueue requests anymore
+ *
+ * After the return of ioctl with suspend != 0, the device must finish any
+ * pending operations like in flight requests.

I'm not sure we should mandate the flush here. This probably blocks us
from adding inflight descriptor reporting in the future.

That's right. Maybe we should add a flags argument to allow not to
flush in flight descriptors in the future? Or maybe the right solution
is to discard that requirement and to mandate in_order to be
migratable at the moment?

I think it's better not to limit the device behaviour like flush or
in_order here. This may simplify the work for adding inflight
descriptor support.

For the device that doesn't care about the inflight descriptor, this
patch is sufficient for doing live migration.
For the device that requires an inflight descriptor, this patch is
insufficient, it requires future extension to get those descriptors.
In this case, device has the flexibility to flush or not so:

1) if we don't get any inflight descriptors, the device may do the flush before
2) if we get inflight descriptors, we need to restore them

Thanks

Thanks!

quoted

Thanks

It must also preserve all the

quoted

+ * necessary state (the virtqueue vring base plus the possible device specific
+ * states) that is required for restoring in the future. The device must not
+ * change its configuration after that point.
+ *
+ * After the return of ioctl with suspend == 0, the device can continue
+ * processing buffers as long as typical conditions are met (vq is enabled,
+ * DRIVER_OK status bit is enabled, etc).
+ */
+#define VHOST_VDPA_SUSPEND             _IOW(VHOST_VIRTIO, 0x7D, int)
+
 #endif
--
2.31.1