Re: [RFC v6] virtio/vsock: add two more queues for datagram types

[Stefano Garzarella]

On Wed, Sep 15, 2021 at 08:59:17PM -0700, Jiang Wang . wrote:
> On Tue, Sep 14, 2021 at 5:46 AM Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > On Mon, Sep 13, 2021 at 10:18:43PM +0000, Jiang Wang wrote:
> > > Datagram sockets are connectionless and unreliable.
> > > The sender does not know the capacity of the receiver
> > > and may send more packets than the receiver can handle.
> > >
> > > Add two more dedicate virtqueues for datagram sockets,
> > > so that it will not unfairly steal resources from
> > > stream and future connection-oriented sockets.
> > >
> > > Signed-off-by: Jiang Wang <jiang.wang@bytedance.com>
> > > ---
> > > v1 -> v2: use qemu cmd option to control number of queues,
> > >         removed configuration settings for dgram.
> > > v2 -> v3: use ioctl to get features and decide number of
> > >         virt queues, instead of qemu cmd option.
> > > v3 -> v4: change DGRAM feature bit value to 2. Add an argument
> > >         in vhost_vsock_common_realize to indicate dgram is supported or not.
> > > v4 -> v5: don't open dev to get vhostfd. Removed leftover definition of
> > >         enable_dgram
> > > v5 -> v6: fix style errors. Imporve error handling of
> > >         vhost_vsock_dgram_supported. Rename MAX_VQS_WITH_DGRAM and another 
> > one.
> > >
> > >  hw/virtio/vhost-user-vsock.c                  |  2 +-
> > >  hw/virtio/vhost-vsock-common.c                | 25 ++++++++++++--
> > >  hw/virtio/vhost-vsock.c                       | 34 ++++++++++++++++++-
> > >  include/hw/virtio/vhost-vsock-common.h        |  6 ++--
> > >  include/hw/virtio/vhost-vsock.h               |  3 ++
> > >  include/standard-headers/linux/virtio_vsock.h |  1 +
> > >  6 files changed, 64 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/hw/virtio/vhost-user-vsock.c b/hw/virtio/vhost-user-vsock.c
> > > index 6095ed7349..e9ec0e1c00 100644
> > > --- a/hw/virtio/vhost-user-vsock.c
> > > +++ b/hw/virtio/vhost-user-vsock.c
> > > @@ -105,7 +105,7 @@ static void vuv_device_realize(DeviceState *dev, Error 
> > **errp)
> > >          return;
> > >      }
> > >
> > > -    vhost_vsock_common_realize(vdev, "vhost-user-vsock");
> > > +    vhost_vsock_common_realize(vdev, "vhost-user-vsock", false);
> >
> > VIRTIO_VSOCK_F_DGRAM support should work equally well for
> > vhost-user-vsock. I don't think there is a need to disable it here.
> Stefano mentioned in previous email ( V3 ) that I can disable dgram
> for vhost-user-vsock for now. I think we need some extra changes to
> fully support vhost-vsock-user, like feature discovery?

I think Stefan is suggesting something similar of what we discussed 
here:
https://lists.nongnu.org/archive/html/qemu-devel/2021-06/msg06689.html

We can allocate all the queues, and choose at runtime which queue to use 
for events, that is the only queue used by QEMU.

We can check that in vhost_vsock_common_start(), just before starting 
the device, where we know the features acked by the guest 
(vdev->guest_features).

And I agree that would be the best approach, since we don't need 
discovery anymore, and the same code works also for vhost-user-vsock as 
is.

> > >
> > >      vhost_dev_set_config_notifier(&vvc->vhost_dev, &vsock_ops);
> > >
> > > diff --git a/hw/virtio/vhost-vsock-common.c 
> > > b/hw/virtio/vhost-vsock-common.c
> > > index 4ad6e234ad..d94636e04e 100644
> > > --- a/hw/virtio/vhost-vsock-common.c
> > > +++ b/hw/virtio/vhost-vsock-common.c
> > > @@ -17,6 +17,8 @@
> > >  #include "hw/virtio/vhost-vsock.h"
> > >  #include "qemu/iov.h"
> > >  #include "monitor/monitor.h"
> > > +#include <sys/ioctl.h>
> > > +#include <linux/vhost.h>
> > >
> > >  int vhost_vsock_common_start(VirtIODevice *vdev)
> > >  {
> > > @@ -196,9 +198,11 @@ int vhost_vsock_common_post_load(void *opaque, 
> > > int version_id)
> > >      return 0;
> > >  }
> > >
> > > -void vhost_vsock_common_realize(VirtIODevice *vdev, const char *name)
> > > +void vhost_vsock_common_realize(VirtIODevice *vdev, const char *name,
> > > +                               bool enable_dgram)
> > >  {
> > >      VHostVSockCommon *vvc = VHOST_VSOCK_COMMON(vdev);
> > > +    int nvqs = VHOST_VSOCK_NVQS;
> > >
> > >      virtio_init(vdev, name, VIRTIO_ID_VSOCK,
> > >                  sizeof(struct virtio_vsock_config));
> > > @@ -209,12 +213,20 @@ void vhost_vsock_common_realize(VirtIODevice *vdev, 
> > const char *name)
> > >      vvc->trans_vq = virtio_add_queue(vdev, VHOST_VSOCK_QUEUE_SIZE,
> > >                                         vhost_vsock_common_handle_output);
> > >
> > > +    if (enable_dgram) {
> > > +        nvqs = VHOST_VSOCK_NVQS_DGRAM;
> > > +        vvc->dgram_recv_vq = virtio_add_queue(vdev, 
> > > VHOST_VSOCK_QUEUE_SIZE,
> > > +                                              
> > vhost_vsock_common_handle_output);
> > > +        vvc->dgram_trans_vq = virtio_add_queue(vdev, VHOST_VSOCK_QUEUE_SIZE,
> > > +                                              
> > vhost_vsock_common_handle_output);
> > > +    }
> >
> > I think the virtqueues can be added unconditionally.
> OK.
> > > +
> > >      /* The event queue belongs to QEMU */
> > >      vvc->event_vq = virtio_add_queue(vdev, VHOST_VSOCK_QUEUE_SIZE,
> > >                                         vhost_vsock_common_handle_output);
> > >
> > > -    vvc->vhost_dev.nvqs = ARRAY_SIZE(vvc->vhost_vqs);
> > > -    vvc->vhost_dev.vqs = vvc->vhost_vqs;
> > > +    vvc->vhost_dev.nvqs = nvqs;
> > > +    vvc->vhost_dev.vqs = g_new0(struct vhost_virtqueue, vvc->vhost_dev.nvqs);
> >
> > IIUC the number of virtqueues needs to be the maximum supported number.
> > It's okay to have more virtqueues than needed. The feature bit is used
> > by the driver to detect the device's support for dgrams, not the number
> > of virtqueues.
> OK. I can just make these changes. But I am not able to test 
> vhost-user-vsock
> as of now. I tried to follow the steps on here:
> https://patchew.org/QEMU/20200515152110.202917-1-sgarzare@redhat.com/
> But the git repo for the vhost-user-vsock is kind of broken. I tried to
> fix it but I am new to rust so it will take some time. Is there any updated
> or an easier way to test vhost-user-vsock?

Yep, it was a PoC, please use the new vhost-user-vsock that Harsha is 
developing here:
https://github.com/rust-vmm/vhost-device/pull/7

Feel free to reach me or Harsha if you have any issue.

> > >
> > >      vvc->post_load_timer = NULL;
> > >  }
> > > @@ -227,6 +239,13 @@ void vhost_vsock_common_unrealize(VirtIODevice *vdev)
> > >
> > >      virtio_delete_queue(vvc->recv_vq);
> > >      virtio_delete_queue(vvc->trans_vq);
> > > +    if (vvc->vhost_dev.nvqs == VHOST_VSOCK_NVQS_DGRAM) {
> > > +        virtio_delete_queue(vvc->dgram_recv_vq);
> > > +        virtio_delete_queue(vvc->dgram_trans_vq);
> > > +    }
> > > +
> > > +    g_free(vvc->vhost_dev.vqs);
> > > +
> > >      virtio_delete_queue(vvc->event_vq);
> > >      virtio_cleanup(vdev);
> > >  }
> > > diff --git a/hw/virtio/vhost-vsock.c b/hw/virtio/vhost-vsock.c
> > > index 1b1a5c70ed..891d38e226 100644
> > > --- a/hw/virtio/vhost-vsock.c
> > > +++ b/hw/virtio/vhost-vsock.c
> > > @@ -20,9 +20,12 @@
> > >  #include "hw/qdev-properties.h"
> > >  #include "hw/virtio/vhost-vsock.h"
> > >  #include "monitor/monitor.h"
> > > +#include <sys/ioctl.h>
> > > +#include <linux/vhost.h>
> > >
> > >  const int feature_bits[] = {
> > >      VIRTIO_VSOCK_F_SEQPACKET,
> > > +    VIRTIO_VSOCK_F_DGRAM,
> >
> > Stefano is currently working on a way to control live migration
> > compatibility when features like seqpacket or dgram aren't available. He
> > can suggest how to do this for dgram.
> Yes. I watched that email thread. I can make similar changes to
> DGRAM. I will do it for the next version.

I'll send a new patch for seqpacket (using on,off,auto prop) hopefully 
tomorrow or early next week, then you can rebase on it. It should be 
simple to add the dgram feature too.

I'll keep you in CC.

Thanks,
Stefano