---------- Forwarded message ---------
From: fengyd <fengy...@gmail.com>
Date: Wed, 8 May 2019 at 12:50
Subject: Re: [Qemu-devel] Fwd: How live migration work for vhost-user
To: Dr. David Alan Gilbert <dgilb...@redhat.com>
Hi,
I checked the DPDK code and found that the function rte_vhost_enqueue_burst
in virtio_net.c is used to receive message.
uint16_t
rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
struct rte_mbuf **pkts, uint16_t count)
{
struct virtio_net *dev = get_device(vid);
if (!dev)
return 0;
if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF))
return virtio_dev_merge_rx(dev, queue_id, pkts, count);
else
return virtio_dev_rx(dev, queue_id, pkts, count);
}
If VIRTIO_NET_F_MRG_RXBUF is supported by guest, virtio_dev_merge_rx is
called, otherwise virtio_dev_rx is called.
And in virtio_dev_merge_rx, last_avail_idx will be updated, and in
virtio_dev_rx, last_avail_idx will not be touched, so it is kept as the
initial value 0.
Our guest virito driver doesn't support VIRTIO_NET_F_MRG_RXBUF, so
last_avail_idx QEMU read from DPDK is always 0?
And The strange thing is the function vhost_user_get_vring_base in
vhost_user.c which returns last_avail_idx to QEMU.
I checked the code in v18.05
<https://git.dpdk.org/dpdk/tree/lib/librte_vhost/vhost_user.c?h=v18.05>,
last_avail_idx is assigned to the return message.
msg->payload.state.num = vq->last_avail_idx;
And in DPDK v17.11
<https://git.dpdk.org/dpdk/tree/lib/librte_vhost/vhost_user.c?h=v17.11>,
last_used_idx is assigned to the return message.
msg->payload.state.num = vq->last_used_idx;
On Tue, 30 Apr 2019 at 00:22, Dr. David Alan Gilbert <dgilb...@redhat.com>
wrote:
> * fengyd (fengy...@gmail.com) wrote:
> > Hi,
> >
> > For vhost, *last_avail_idx* is maintained in vhost_virtqueue
> > but during live migration, *last_avail_idx* is fetched from VirtQueue.
> > Do you know how these two *last_avail_idx *are synchronized?
> >
> > virtio_load related code which is called during live migration:
> >
> > * vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -*
> > * vdev->vq[i].used_idx);*
> > * if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {*
> > * error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "*
> > * "used_idx 0x%x",*
> > * i, vdev->vq[i].vring.num,*
> > * vdev->vq[i].last_avail_idx,*
>
> I don't know that code well; but I think the answer is that since the
> queues themselves are in guest memory, the guest memory is migrated
> by the normal migration code and so the queues version of last_avail_idx
> should be correct. The 'log' mechanism I previously mentioned will
> need to make sure the queue pages are marked dirty to make sure these
> are updated correctly.
>
> Dave
>
> >
> >
> > Thanks
> >
> > On Tue, 23 Apr 2019 at 14:20, fengyd <fengy...@gmail.com> wrote:
> >
> > > Hi,
> > >
> > > I want to add some log to qemu-kvm-ev.
> > > Do you know how to compile qemu-kvm-ev from source code?
> > >
> > > Thanks
> > >
> > > Yafeng
> > >
> > > On Tue, 16 Apr 2019 at 16:47, Dr. David Alan Gilbert <
> dgilb...@redhat.com>
> > > wrote:
> > >
> > >> * fengyd (fengy...@gmail.com) wrote:
> > >> > ---------- Forwarded message ---------
> > >> > From: fengyd <fengy...@gmail.com>
> > >> > Date: Tue, 16 Apr 2019 at 09:17
> > >> > Subject: Re: [Qemu-devel] How live migration work for vhost-user
> > >> > To: Dr. David Alan Gilbert <dgilb...@redhat.com>
> > >> >
> > >> >
> > >> > Hi,
> > >> >
> > >> > Any special feature needs to be supported on guest driver?
> > >> > Because it's OK for standard Linux VM, but not OK for our VM where
> > >> virtio
> > >> > is implemented by ourself.
> > >>
> > >> I'm not sure; you do have to support that 'log' mechanism but I don't
> > >> know what else is needed.
> > >>
> > >> > And with qemu-kvm-ev-2.6, live migration can work with our VM where
> > >> virtio
> > >> > is implemented by ourself.
> > >>
> > >> 2.6 is pretty old, so there's a lot of changes - not sure what's
> > >> relevant.
> > >>
> > >> Dave
> > >>
> > >> > Thanks
> > >> > Yafeng
> > >> >
> > >> > On Mon, 15 Apr 2019 at 22:54, Dr. David Alan Gilbert <
> > >> dgilb...@redhat.com>
> > >> > wrote:
> > >> >
> > >> > > * fengyd (fengy...@gmail.com) wrote:
> > >> > > > Hi,
> > >> > > >
> > >> > > > During live migration, the folloing log can see in
> > >> nova-compute.log in
> > >> > > my
> > >> > > > environment:
> > >> > > > ERROR nova.virt.libvirt.driver
> > >> [req-039a85e1-e7a1-4a63-bc6d-c4b9a044aab6
> > >> > > > 0cdab20dc79f4bc6ae5790e7b4a898ac
> 3363c319773549178acc67f32c78310e -
> > >> > > default
> > >> > > > default] [instance: 5ec719f4-1865-4afe-a207-3d9fae22c410] Live
> > >> Migration
> > >> > > > failure: internal error: qemu unexpectedly closed the monitor:
> > >> > > > 2019-04-15T02:58:22.213897Z qemu-kvm: VQ 0
> > >> > > > size 0x100 < last_avail_idx 0x1e - used_idx 0x23
> > >> > > >
> > >> > > > It's OK for standard Linux VM, but not OK for our VM where
> virtio is
> > >> > > > implemented by ourself.
> > >> > > > KVM version as follow:
> > >> > > > qemu-kvm-common-ev-2.12.0-18.el7_6.3.1.x86_64
> > >> > > > qemu-kvm-ev-2.12.0-18.el7_6.3.1.x86_64
> > >> > > > libvirt-daemon-kvm-3.9.0-14.2.el7.centos.ncir.8.x86_64
> > >> > > >
> > >> > > > Do you know what's the difference between virtio and vhost-user
> > >> during
> > >> > > > migration?
> > >> > > > The function virtio_load in Qemu is called for virtio and
> vhost-user
> > >> > > during
> > >> > > > migration.
> > >> > > > For virtio, last_avail_idx and used_idx are stored in Qemu,
> Qemu
> > >> is
> > >> > > > responsible for updating their values accordingly
> > >> > > > For vhost-user, last_avail_idx and used_idx are stored in
> > >> vhost-user
> > >> > > app,
> > >> > > > eg. DPDK, not in Qemu?
> > >> > > > How does migration work for vhost-user?
> > >> > >
> > >> > > I don't know the details, but my understanding is that vhost-user
> > >> > > tells the vhost-user client about an area of 'log' memory, where
> the
> > >> > > vhost-user client must mark pages as dirty.
> > >> > >
> > >> > > In the qemu source, see docs/interop/vhost-user.txt and see
> > >> > > the VHOST_SET_LOG_BASE and VHOST_USER_SET_LOG_FD calls.
> > >> > >
> > >> > > If the client correctly marks the areas as dirty, then qemu
> > >> > > should resend those pages across.
> > >> > >
> > >> > >
> > >> > > Dave
> > >> > >
> > >> > > > Thanks in advance
> > >> > > > Yafeng
> > >> > > --
> > >> > > Dr. David Alan Gilbert / dgilb...@redhat.com / Manchester, UK
> > >> > >
> > >> --
> > >> Dr. David Alan Gilbert / dgilb...@redhat.com / Manchester, UK
> > >>
> > >
> --
> Dr. David Alan Gilbert / dgilb...@redhat.com / Manchester, UK
>
