> From: Burakov, Anatoly <anatoly.bura...@intel.com>
> Sent: Wednesday 3 March 2021 11:53
> To: Walsh, Conor <conor.wa...@intel.com>; jer...@marvell.com;
> step...@networkplumber.org; Iremonger, Bernard
> <bernard.iremon...@intel.com>; Ananyev, Konstantin
> <konstantin.anan...@intel.com>; Medvedkin, Vladimir
> <vladimir.medved...@intel.com>
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v3 4/5] examples/l3fwd: implement FIB
> lookup method
>
> On 19-Feb-21 3:09 PM, Conor Walsh wrote:
> > This patch implements the Forwarding Information Base (FIB) library
> > in l3fwd using the function calls and infrastructure introduced in
> > the previous patch.
> >
> > Signed-off-by: Conor Walsh <conor.wa...@intel.com>
> > ---
>
> <snip>
>
> > + || defined RTE_ARCH_PPC_64
> > +#define FIB_SEND_MULTI
> > +#endif
> > +
> > +static struct rte_fib *ipv4_l3fwd_fib_lookup_struct[NB_SOCKETS];
> > +static struct rte_fib6 *ipv6_l3fwd_fib_lookup_struct[NB_SOCKETS];
> > +
> > +/* Parse packet type and ip address. */
> > +static inline void
> > +fib_parse_packet(struct rte_mbuf *mbuf,
> > + uint32_t *ipv4, uint32_t *ipv4_cnt,
> > + uint8_t ipv6[RTE_FIB6_IPV6_ADDR_SIZE],
> > + uint32_t *ipv6_cnt, uint8_t *ip_type)
>
> Nitpicking, but here and in a bunch of other places, the indentation is
> quite odd :)
Hi Anatoly,
I have corrected the indentation issues in the v4 of the patchset.
>
> <snip>
>
> > +
> > +/* Bulk parse, fib lookup and send. */
> > +static inline void
> > +fib_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
> > + uint16_t portid, struct lcore_conf *qconf)
> > +{
> > + uint32_t ipv4_arr[nb_rx];
> > + uint8_t ipv6_arr[nb_rx][RTE_FIB6_IPV6_ADDR_SIZE];
> > + uint16_t hops[nb_rx];
> > + uint64_t hopsv4[nb_rx], hopsv6[nb_rx];
> > + uint8_t type_arr[nb_rx];
> > + uint32_t ipv4_cnt = 0, ipv6_cnt = 0;
> > + uint32_t ipv4_reassem = 0, ipv6_reassem = 0;
>
> I don't quite follow the naming here - this looks like it's
> "reassembling" something but i don't see any IP reassembly going on?
> Artifacts of copy-paste?
For FIB lookups the packets need to be split into ipv4 and ipv6.
These lookups create separate arrays of 'hops' for ipv4 and ipv6 which need to
be assembled into a single 'hops' array.
I use ipv4_reassem And ipv6_reassem as counters for this purpose.
To try and avoid confusion I have renamed them as follows ipv4_reassem ->
ipv4_arr_assem and ipv6_reassem -> ipv6_arr_assem in v4.
>
> > + int32_t i;
> > +
> > + /* Prefetch first packets. */
> > + for (i = 0; i < FIB_PREFETCH_OFFSET && i < nb_rx; i++)
> > + rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i], void *));
> > +
> > + /* Parse packet info and prefetch. */
> > + for (i = 0; i < (nb_rx - FIB_PREFETCH_OFFSET); i++) {
> > + /* Prefetch packet. */
> > + rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
> > + i + FIB_PREFETCH_OFFSET],
> > + void *));
> > + fib_parse_packet(pkts_burst[i],
> > + &ipv4_arr[ipv4_cnt], &ipv4_cnt,
> > + ipv6_arr[ipv6_cnt], &ipv6_cnt,
> > + &type_arr[i]);
> > + }
> > +
> > + /* Parse remaining packet info. */
> > + for (; i < nb_rx; i++)
> > + fib_parse_packet(pkts_burst[i],
> > + &ipv4_arr[ipv4_cnt], &ipv4_cnt,
> > + ipv6_arr[ipv6_cnt], &ipv6_cnt,
> > + &type_arr[i]);
> > +
> > + /* Lookup IPv4 hops if IPv4 packets are present. */
> > + if (likely(ipv4_cnt > 0))
> > + rte_fib_lookup_bulk(qconf->ipv4_lookup_struct,
> > + ipv4_arr, hopsv4, ipv4_cnt);
> > +
> > + /* Lookup IPv6 hops if IPv6 packets are present. */
> > + if (ipv6_cnt > 0)
> > + rte_fib6_lookup_bulk(qconf->ipv6_lookup_struct,
> > + ipv6_arr, hopsv6, ipv6_cnt);
> > +
> > + /* Add IPv4 and IPv6 hops to one array depending on type. */
> > + for (i = 0; i < nb_rx; i++) {
> > + if (type_arr[i]) {
> > + if (hopsv4[ipv4_reassem] != FIB_DEFAULT_HOP)
> > + hops[i] = (uint16_t)hopsv4[ipv4_reassem];
> > + else
> > + hops[i] = portid;
> > + ipv4_reassem++;
>
> Nitpicking, but could be made slightly more concise and readable:
>
> const uint16_t nh = (uint16_t)hopsv4[ipv4_reassem++];
> hops[i] = nh != FIB_DEFAULT_HOP ? nh : portid;
>
> Same for IPv6.
I have changed this in the v4 of the patchset which will be released soon.
Thanks,
Conor.
>
> --
> Thanks,
> Anatoly
