Worrying about the performance of functions like promiscuous enabling is
misplaced since this routine would hardly be used in a performance path.

The real issue is the syntax.  Has Petri defined what he wants for these
yet?  This seems like a simply attribute on the API odp_pktio_t object.
Either it is or is not in promiscuous mode.  As such, to be consistent with
similar attributes one would expect the getter to be:

odp_bool_t odp_pktio_promisc(odp_pktio_t io);

and the setter to be:

void odp_pktio_set_promisc(odp_pktio_t io, odp_bool_t truefalse);

Bill



On Thu, Nov 27, 2014 at 3:12 PM, Ola Liljedahl <[email protected]>
wrote:

> On 27 November 2014 at 18:51, Maxim Uvarov <[email protected]>
> wrote:
> > Define API and implement promisc functions for linux-generic.
> >
> > Signed-off-by: Maxim Uvarov <[email protected]>
> > ---
> >  platform/linux-generic/include/api/odp_packet_io.h | 24 +++++++
> >  platform/linux-generic/odp_packet_io.c             | 74
> ++++++++++++++++++++++
> >  2 files changed, 98 insertions(+)
> >
> > diff --git a/platform/linux-generic/include/api/odp_packet_io.h
> b/platform/linux-generic/include/api/odp_packet_io.h
> > index 667395c..c7d1e40 100644
> > --- a/platform/linux-generic/include/api/odp_packet_io.h
> > +++ b/platform/linux-generic/include/api/odp_packet_io.h
> > @@ -149,6 +149,30 @@ int odp_pktio_set_mtu(odp_pktio_t id, int mtu);
> >  int odp_pktio_mtu(odp_pktio_t id);
> >
> >  /**
> > + * Enable promiscuous mode on a packet IO interface.
> > + *
> > + * @param[in] id       ODP packet IO handle.
> > + * @param[in] enable    1 enabled, 0 disabled.
> > + *
> > + * @retval  0 on success.
> > + * @retval -1 on a bad pktio id
> > + * @retval -1 any other error
> > + */
> > +int odp_pktio_promisc_set(odp_pktio_t id, odp_bool_t enable);
> > +
> > +/**
> > + * Determine if promiscuous mode is enabled for a packet IO interface.
> > + *
> > + * @param[in] id ODP packet IO handle.
> > + *
> > + * @retval  1 if promiscuous mode is enabled.
> > + * @retval  0 if promiscuous mode is disabled.
> > + * @retval -1 on a bad pktio id
> > + * @retval -1 any other error
> > +*/
> > +int odp_pktio_promisc_enabled(odp_pktio_t id);
> > +
> > +/**
> >   * @}
> >   */
> >
> > diff --git a/platform/linux-generic/odp_packet_io.c
> b/platform/linux-generic/odp_packet_io.c
> > index c523350..9140620 100644
> > --- a/platform/linux-generic/odp_packet_io.c
> > +++ b/platform/linux-generic/odp_packet_io.c
> > @@ -542,3 +542,77 @@ int odp_pktio_mtu(odp_pktio_t id)
> >
> >         return ifr.ifr_mtu;
> >  }
> > +
> > +int odp_pktio_promisc_set(odp_pktio_t id, odp_bool_t enable)
> > +{
> > +       pktio_entry_t *entry;
> > +       int sockfd;
> > +       struct ifreq ifr;
> > +       int ret;
> > +
> > +       entry = get_entry(id);
> > +       if (entry == NULL) {
> > +               ODP_DBG("pktio entry %d does not exist\n", id);
> > +               return -1;
> > +       }
> > +
> > +       if (entry->s.pkt_sock_mmap.sockfd)
> > +               sockfd = entry->s.pkt_sock_mmap.sockfd;
> > +       else
> > +               sockfd = entry->s.pkt_sock.sockfd;
> > +
> > +       strncpy(ifr.ifr_name, entry->s.name, IFNAMSIZ - 1);
> > +       ifr.ifr_name[IFNAMSIZ] = 0;
> > +
> > +       ret = ioctl(sockfd, SIOCGIFFLAGS, &ifr);
> > +       if (ret < 0) {
> > +               ODP_DBG("ioctl SIOCGIFFLAGS error\n");
> > +               return -1;
> > +       }
> > +
> > +       if (enable)
> > +               ifr.ifr_flags |= IFF_PROMISC;
> > +       else
> > +               ifr.ifr_flags &= ~(IFF_PROMISC);
> As some people are concerned with branching caused by different values
> of the enable parameter, we can express this in a branch-less way:
> #define UPDATE(old, mask, ena) ((old) & ~(mask)) | (((ena) ? ~0U : 0) &
> (mask))
> ifr.ifr_flags = UPDATE(ifr.ifr_flags, IFF_PROMISC, enable);
>
> GCC for ARMv7 generates pretty compact code for this, no branches
> (uses on ITE (if-then-else) instruction though).
>
> Branch-less code is preferred if the branches are difficult to predict
> (e.g. more or less random). But don't overdo it since modern branch
> predictors are quite good.
>
>
> > +
> > +       ret = ioctl(sockfd, SIOCSIFFLAGS, &ifr);
> > +       if (ret < 0) {
> > +               ODP_DBG("ioctl SIOCSIFFLAGS error\n");
> > +               return -1;
> > +       }
> > +
> > +       return 0;
> > +}
> > +
> > +int odp_pktio_promisc_enabled(odp_pktio_t id)
> > +{
> > +       pktio_entry_t *entry;
> > +       int sockfd;
> > +       struct ifreq ifr;
> > +       int ret;
> > +
> > +       entry = get_entry(id);
> > +       if (entry == NULL) {
> > +               ODP_DBG("pktio entry %d does not exist\n", id);
> > +               return -1;
> > +       }
> > +
> > +       if (entry->s.pkt_sock_mmap.sockfd)
> > +               sockfd = entry->s.pkt_sock_mmap.sockfd;
> > +       else
> > +               sockfd = entry->s.pkt_sock.sockfd;
> > +
> > +       strncpy(ifr.ifr_name, entry->s.name, IFNAMSIZ - 1);
> > +       ifr.ifr_name[IFNAMSIZ] = 0;
> > +
> > +       ret = ioctl(sockfd, SIOCGIFFLAGS, &ifr);
> > +       if (ret < 0) {
> > +               ODP_DBG("ioctl SIOCGIFFLAGS error\n");
> > +               return -1;
> > +       }
> > +
> > +       if (ifr.ifr_flags & IFF_PROMISC)
> > +               return 1;
> > +       else
> > +               return 0;
> > +}
> > --
> > 1.8.5.1.163.gd7aced9
> >
> >
> > _______________________________________________
> > lng-odp mailing list
> > [email protected]
> > http://lists.linaro.org/mailman/listinfo/lng-odp
>
> _______________________________________________
> lng-odp mailing list
> [email protected]
> http://lists.linaro.org/mailman/listinfo/lng-odp
>
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to