Re: [dpdk-dev] [PATCH 4/5] virtio: fix ring size negotiation

[Stephen Hemminger <stephen@networkplumber.org>]

On Thu, 16 Apr 2015 06:26:02 +0000
"Ouyang, Changchun" <changchun.ouyang@intel.com> wrote:

> 
> 
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Thursday, April 16, 2015 1:48 PM
> To: Ouyang, Changchun
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH 4/5] virtio: fix ring size negotiation
> 
> No warning is needed, it just works.
> 
> I know it works, but the upper user don’t know the descriptor number is reduced.
> I concern it is not so user-friendly here.
> 
> 
> On Wed, Apr 15, 2015 at 8:39 PM, Ouyang, Changchun <changchun.ouyang@intel.com<mailto:changchun.ouyang@intel.com>> wrote:
> 
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org<mailto:dev-bounces@dpdk.org>] On Behalf Of Stephen
> > Hemminger
> > Sent: Wednesday, April 15, 2015 11:20 PM
> > To: dev@dpdk.org<mailto:dev@dpdk.org>
> > Subject: [dpdk-dev] [PATCH 4/5] virtio: fix ring size negotiation
> >
> > This fixes another of the issues with running virtio on non-KVM
> > envirionments. For example, Google Compute Engine reports a ring size of
> > 16K.
> >
> > If guest virtio requests more slots than available then the queue should just
> 
> I suspect 'more' here should be 'less'?
> 
> > be initialized to the smaller value.
> >
> > Conversely, if the number of descriptors requested exceeds the virtio host
> > queue size, then just silently use the smaller host size.
> >
> > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org<mailto:stephen@networkplumber.org>>
> > ---
> >  lib/librte_pmd_virtio/virtio_ethdev.c | 18 +++++++++++++-----
> >  1 file changed, 13 insertions(+), 5 deletions(-)
> >
> > diff --git a/lib/librte_pmd_virtio/virtio_ethdev.c
> > b/lib/librte_pmd_virtio/virtio_ethdev.c
> > index 3cb9c6a..db0232e 100644
> > --- a/lib/librte_pmd_virtio/virtio_ethdev.c
> > +++ b/lib/librte_pmd_virtio/virtio_ethdev.c
> > @@ -267,13 +267,21 @@ int virtio_dev_queue_setup(struct rte_eth_dev
> > *dev,
> >       if (vq_size == 0) {
> >               PMD_INIT_LOG(ERR, "%s: virtqueue does not exist",
> > __func__);
> >               return -EINVAL;
> > -     } else if (!rte_is_power_of_2(vq_size)) {
> > +     }
> > +
> > +     if (!rte_is_power_of_2(vq_size)) {
> >               PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2",
> > __func__);
> >               return -EINVAL;
> > -     } else if (nb_desc != vq_size) {
> > -             PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to
> > vq size (%d), fall to vq size",
> > -                     nb_desc, vq_size);
> > -             nb_desc = vq_size;
> > +     }
> > +
> > +     if (nb_desc < vq_size) {
> > +             if (!rte_is_power_of_2(nb_desc)) {
> > +                     PMD_INIT_LOG(ERR,
> > +                                  "nb_desc(%u) size is not powerof 2",
> > +                                  nb_desc);
> > +                     return -EINVAL;
> > +             }
> > +             vq_size = nb_desc;
> Don't we need a warning when nb_desc > vq_size?


No warning is needed. This will actually be a common case
for many applications.  

IMHO application should not have to worry about
what type of network device it is running on and therefore would likely
pass a reasonably large number of receive descriptors (say 512) and since
the default KVM/QEMU ring size is 256, the receive queue would be limited
by host not application.

The whole idea of application passing number of receive descriptors to
DPDK is bogus because each device driver has different timing, and may
use different number of receive descriptors per packet.