Re: [Qemu-devel] [virtio-dev] Re: [PATCH v3 1/2] virtio-net: enable configurable tx queue size

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Jun 27, 2017 at 09:06:07AM +0800, Wei Wang wrote:
> On 06/27/2017 05:21 AM, Michael S. Tsirkin wrote:
> > On Mon, Jun 26, 2017 at 06:34:25PM +0800, Wei Wang wrote:
> > > On 06/26/2017 04:05 PM, Jason Wang wrote:
> > > > 
> > > > On 2017年06月26日 12:55, Wei Wang wrote:
> > > > > On 06/26/2017 11:18 AM, Jason Wang wrote:
> > > > > > On 2017年06月23日 10:32, Wei Wang wrote:
> > > > > > > This patch enables the virtio-net tx queue size to be configurable
> > > > > > > between 256 (the default queue size) and 1024 by the user when the
> > > > > > > vhost-user backend is used.
> > > > > > > 
> > > > > > > Currently, the maximum tx queue size for other backends is 512 due
> > > > > > > to the following limitations:
> > > > > > > - QEMU backend: the QEMU backend implementation in some cases may
> > > > > > > send 1024+1 iovs to writev.
> > > > > > > - Vhost_net backend: there are possibilities that the guest sends
> > > > > > > a vring_desc of memory which corsses a MemoryRegion thereby
> > > > > > > generating more than 1024 iovs after translattion from guest-physical
> > > > > > > address in the backend.
> > > > > > > 
> > > > > > > Signed-off-by: Wei Wang <wei.w.wang@intel.com>
> > > > > > > ---
> > > > > > >    hw/net/virtio-net.c            | 45
> > > > > > > +++++++++++++++++++++++++++++++++---------
> > > > > > >    include/hw/virtio/virtio-net.h |  1 +
> > > > > > >    2 files changed, 37 insertions(+), 9 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> > > > > > > index 91eddaf..d13ca60 100644
> > > > > > > --- a/hw/net/virtio-net.c
> > > > > > > +++ b/hw/net/virtio-net.c
> > > > > > > @@ -34,8 +34,11 @@
> > > > > > >      /* previously fixed value */
> > > > > > >    #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256
> > > > > > > +
> > > > > > > +    /*
> > > > > > > +     * Currently, backends other than vhost-user don't
> > > > > > > support 1024 queue
> > > > > > > +     * size.
> > > > > > > +     */
> > > > > > > +    if (n->net_conf.tx_queue_size == VIRTQUEUE_MAX_SIZE &&
> > > > > > > +        nc->peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) {
> > > > > > > +        n->net_conf.tx_queue_size = VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE;
> > > > > > Do we really want assume all vhost-user backend support 1024
> > > > > > queue size?
> > > > > > 
> > > > >   Do you know any vhost-user backend implementation that doesn't support
> > > > > 1024 tx queue size?
> > > > > 
> > > > > 
> > > > I don't but the issue is vhost-user uditis an open protocol, there could
> > > > be even closed source implementation on this. So we can't audit all
> > > > kinds of backends.
> > > > 
> > > This is the discussion left before. Here is my thought:
> > > 
> > > Until now, nobody is officially using 1024 tx queue size, including the
> > > unknown
> > > vhost-user backend implementation. So, applying this patch won't affect any
> > > already deployed products.
> > > 
> > > If someday, people whose products are based on the unknown vhost-user
> > > implementation change to try with 1024 tx queue size and apply this patch,
> > > and find 1024 doesn't work for them, they can simply set the value to 512
> > > (which
> > > is better than the 256 size they were using) or apply the 2nd patch entitled
> > > "VIRTIO_F_MAX_CHAIN_SIZE" to use 1024 queue size.
> > > 
> > > I didn't see any big issue here. Would you see any issues?
> > > 
> > > @Michael, what is your opinion?
> > Since the default does not change, I think it's fine.
> > You need to both have a non-default value and be using
> > a backend that does not support the 1024 size.
> > Seems unlikely to me.
> > 
> 
> Were you referring to the use of 512 queue size? I think that's already
> supported
> for all the backends.
> 
> The above code only makes a non-vhost-user backend, which doesn't support
> 1024,
> use the default value when the user specified 1024 queue size. It won't make
> any
> change if 512 was given.
> 
> Now, we treat all vhost-user backends as the ones that support 1024. If a
> user got an
> unknown issue due to their special vhost-user backend implementation(maybe
> bugs
> generated by their own) with 1024 queue size, they'll need to re-start with
> 512, and
> the code will work with 512.
> 
> Best,
> Wei

I am just saying since default queue size is unchanged, your patches
are safe.

-- 
MST