[dpdk-dev] [PATCH] vhost: support non multiqueue guests

[Maxime Coquelin]

Hi Alan,

On 01/31/2018 03:42 PM, alangordondewar at gmail.com wrote:
> From: Alan Dewar<alan.dewar at att.com>
> 
> Performance of vhost interfaces can be improved by having multiple
> TX/RX queue-pairs.  QEMU can be told to use multiple queue-pairs for
> a vhost interface when starting the guest VM.  The DPDK will also
> configure multiple queue-pairs in response to requests from QEMU.
> 
> Later when the guest VM reaches the running state, it can decide to not
> support the multiqueue option.   This information is passed down from
> the guest VM to QEMU, and from QEMU to the DPDK, but the DPDK ignores
> it.
> 
> Because the guest VM doesn't support the multiqueue option it will only
> initialise the first queue-pair, and in turn the DPDK will not signal
> that the vhost interface is up.
> 
> This change allows the DPDK to signal that the vhost interface is up
> after only the first queue-pair is fully initialised if the guest VM
> does not support the multiqueue option.
> 
> Signed-off-by: Alan Dewar<alan.dewar at att.com>
> ---
>   lib/librte_vhost/vhost.c      |  6 +++---
>   lib/librte_vhost/vhost_user.c |  8 +++++++-
>   lib/librte_vhost/virtio_net.c | 12 +++++++++---
>   3 files changed, 19 insertions(+), 7 deletions(-)

We already implemented a workaround to fix this issue:
  commit e29109323595beb3884da58126ebb3b878cb66f5
Author: Maxime Coquelin <maxime.coquelin at redhat.com>
Date:   Wed Dec 13 09:51:09 2017 +0100

     vhost: destroy unused virtqueues when multiqueue not negotiated

     QEMU sends VHOST_USER_SET_VRING_CALL requests for all queues
     declared in QEMU command line before the guest is started.
     It has the effect in DPDK vhost-user backend to allocate vrings
     for all queues declared by QEMU.

     If the first driver being used does not support multiqueue,
     the device never changes to VIRTIO_DEV_RUNNING state as only
     the first queue pair is initialized. One driver impacted by
     this bug is virtio-net's iPXE driver which does not support
     VIRTIO_NET_F_MQ feature.

     It is safe to destroy unused virtqueues in SET_FEATURES request
     handler, as it is ensured the device is not in running state
     at this stage, so virtqueues aren't being processed.

     Signed-off-by: Maxime Coquelin <maxime.coquelin at redhat.com>
     Acked-by: Laszlo Ersek <lersek at redhat.com>
     Acked-by: Yuanhan Liu <yliu at fridaylinux.org>


Could you try with latest master and confirm it solves the issue on your
side?

Cheers,
Maxime