net/sfc: support VXLAN and NVGRE packet types classification
authorAndrew Rybchenko <arybchenko@solarflare.com>
Sun, 24 Dec 2017 10:46:41 +0000 (10:46 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 16 Jan 2018 17:47:49 +0000 (18:47 +0100)
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@solarflare.com>
doc/guides/nics/sfc_efx.rst
drivers/net/sfc/sfc_dp_rx.h
drivers/net/sfc/sfc_ef10_rx.c
drivers/net/sfc/sfc_ethdev.c
drivers/net/sfc/sfc_rx.c

index bde3cc8..994e111 100644 (file)
@@ -124,6 +124,17 @@ with full-feature firmware variant running.
 **sfboot** should be used to configure NIC to run full-feature firmware variant.
 See Solarflare Server Adapter User's Guide for details.
 
+SFN8xxx family adapters provide either inner or outer packet classes.
+If adapter firmware advertises support for tunnels then the PMD
+configures the hardware to report inner classes, and outer classes are
+not reported in received packets.
+However, for VXLAN and GENEVE tunnels the PMD does report UDP as the
+outer layer 4 packet type.
+
+SFN8xxx family adapters report GENEVE packets as VXLAN.
+If UDP ports are configured for only one tunnel type then it is safe to
+treat VXLAN packet type indication as the corresponding UDP tunnel type.
+
 
 Flow API support
 ----------------
index 3f6a604..33e06ac 100644 (file)
@@ -150,7 +150,8 @@ typedef bool (sfc_dp_rx_qrx_ev_t)(struct sfc_dp_rxq *dp_rxq, unsigned int id);
 typedef void (sfc_dp_rx_qpurge_t)(struct sfc_dp_rxq *dp_rxq);
 
 /** Get packet types recognized/classified */
-typedef const uint32_t * (sfc_dp_rx_supported_ptypes_get_t)(void);
+typedef const uint32_t * (sfc_dp_rx_supported_ptypes_get_t)(
+                               uint32_t tunnel_encaps);
 
 /** Get number of pending Rx descriptors */
 typedef unsigned int (sfc_dp_rx_qdesc_npending_t)(struct sfc_dp_rxq *dp_rxq);
@@ -166,6 +167,7 @@ struct sfc_dp_rx {
        unsigned int                            features;
 #define SFC_DP_RX_FEAT_SCATTER                 0x1
 #define SFC_DP_RX_FEAT_MULTI_PROCESS           0x2
+#define SFC_DP_RX_FEAT_TUNNELS                 0x4
        sfc_dp_rx_qcreate_t                     *qcreate;
        sfc_dp_rx_qdestroy_t                    *qdestroy;
        sfc_dp_rx_qstart_t                      *qstart;
index 4c76f74..41c2885 100644 (file)
@@ -251,6 +251,7 @@ static void
 sfc_ef10_rx_ev_to_offloads(struct sfc_ef10_rxq *rxq, const efx_qword_t rx_ev,
                           struct rte_mbuf *m)
 {
+       uint32_t tun_ptype = 0;
        uint32_t l2_ptype = 0;
        uint32_t l3_ptype = 0;
        uint32_t l4_ptype = 0;
@@ -259,15 +260,40 @@ sfc_ef10_rx_ev_to_offloads(struct sfc_ef10_rxq *rxq, const efx_qword_t rx_ev,
        if (unlikely(EFX_TEST_QWORD_BIT(rx_ev, ESF_DZ_RX_PARSE_INCOMPLETE_LBN)))
                goto done;
 
+       switch (EFX_QWORD_FIELD(rx_ev, ESF_EZ_RX_ENCAP_HDR)) {
+       default:
+               /* Unexpected encapsulation tag class */
+               SFC_ASSERT(false);
+               /* FALLTHROUGH */
+       case ESE_EZ_ENCAP_HDR_NONE:
+               break;
+       case ESE_EZ_ENCAP_HDR_VXLAN:
+               /*
+                * It is definitely UDP, but we have no information
+                * about IPv4 vs IPv6 and VLAN tagging.
+                */
+               tun_ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP;
+               break;
+       case ESE_EZ_ENCAP_HDR_GRE:
+               /*
+                * We have no information about IPv4 vs IPv6 and VLAN tagging.
+                */
+               tun_ptype = RTE_PTYPE_TUNNEL_NVGRE;
+               break;
+       }
+
        switch (EFX_QWORD_FIELD(rx_ev, ESF_DZ_RX_ETH_TAG_CLASS)) {
        case ESE_DZ_ETH_TAG_CLASS_NONE:
-               l2_ptype = RTE_PTYPE_L2_ETHER;
+               l2_ptype = (tun_ptype == 0) ? RTE_PTYPE_L2_ETHER :
+                       RTE_PTYPE_INNER_L2_ETHER;
                break;
        case ESE_DZ_ETH_TAG_CLASS_VLAN1:
-               l2_ptype = RTE_PTYPE_L2_ETHER_VLAN;
+               l2_ptype = (tun_ptype == 0) ? RTE_PTYPE_L2_ETHER_VLAN :
+                       RTE_PTYPE_INNER_L2_ETHER_VLAN;
                break;
        case ESE_DZ_ETH_TAG_CLASS_VLAN2:
-               l2_ptype = RTE_PTYPE_L2_ETHER_QINQ;
+               l2_ptype = (tun_ptype == 0) ? RTE_PTYPE_L2_ETHER_QINQ :
+                       RTE_PTYPE_INNER_L2_ETHER_QINQ;
                break;
        default:
                /* Unexpected Eth tag class */
@@ -276,25 +302,31 @@ sfc_ef10_rx_ev_to_offloads(struct sfc_ef10_rxq *rxq, const efx_qword_t rx_ev,
 
        switch (EFX_QWORD_FIELD(rx_ev, ESF_DZ_RX_L3_CLASS)) {
        case ESE_DZ_L3_CLASS_IP4_FRAG:
-               l4_ptype = RTE_PTYPE_L4_FRAG;
+               l4_ptype = (tun_ptype == 0) ? RTE_PTYPE_L4_FRAG :
+                       RTE_PTYPE_INNER_L4_FRAG;
                /* FALLTHROUGH */
        case ESE_DZ_L3_CLASS_IP4:
-               l3_ptype = RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
+               l3_ptype = (tun_ptype == 0) ? RTE_PTYPE_L3_IPV4_EXT_UNKNOWN :
+                       RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN;
                ol_flags |= PKT_RX_RSS_HASH |
                        ((EFX_TEST_QWORD_BIT(rx_ev,
                                             ESF_DZ_RX_IPCKSUM_ERR_LBN)) ?
                         PKT_RX_IP_CKSUM_BAD : PKT_RX_IP_CKSUM_GOOD);
                break;
        case ESE_DZ_L3_CLASS_IP6_FRAG:
-               l4_ptype = RTE_PTYPE_L4_FRAG;
+               l4_ptype = (tun_ptype == 0) ? RTE_PTYPE_L4_FRAG :
+                       RTE_PTYPE_INNER_L4_FRAG;
                /* FALLTHROUGH */
        case ESE_DZ_L3_CLASS_IP6:
-               l3_ptype = RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
+               l3_ptype = (tun_ptype == 0) ? RTE_PTYPE_L3_IPV6_EXT_UNKNOWN :
+                       RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN;
                ol_flags |= PKT_RX_RSS_HASH;
                break;
        case ESE_DZ_L3_CLASS_ARP:
                /* Override Layer 2 packet type */
-               l2_ptype = RTE_PTYPE_L2_ETHER_ARP;
+               /* There is no ARP classification for inner packets */
+               if (tun_ptype == 0)
+                       l2_ptype = RTE_PTYPE_L2_ETHER_ARP;
                break;
        default:
                /* Unexpected Layer 3 class */
@@ -303,14 +335,16 @@ sfc_ef10_rx_ev_to_offloads(struct sfc_ef10_rxq *rxq, const efx_qword_t rx_ev,
 
        switch (EFX_QWORD_FIELD(rx_ev, ESF_DZ_RX_L4_CLASS)) {
        case ESE_DZ_L4_CLASS_TCP:
-               l4_ptype = RTE_PTYPE_L4_TCP;
+               l4_ptype = (tun_ptype == 0) ? RTE_PTYPE_L4_TCP :
+                       RTE_PTYPE_INNER_L4_TCP;
                ol_flags |=
                        (EFX_TEST_QWORD_BIT(rx_ev,
                                            ESF_DZ_RX_TCPUDP_CKSUM_ERR_LBN)) ?
                        PKT_RX_L4_CKSUM_BAD : PKT_RX_L4_CKSUM_GOOD;
                break;
        case ESE_DZ_L4_CLASS_UDP:
-               l4_ptype = RTE_PTYPE_L4_UDP;
+               l4_ptype = (tun_ptype == 0) ? RTE_PTYPE_L4_UDP :
+                       RTE_PTYPE_INNER_L4_UDP;
                ol_flags |=
                        (EFX_TEST_QWORD_BIT(rx_ev,
                                            ESF_DZ_RX_TCPUDP_CKSUM_ERR_LBN)) ?
@@ -329,7 +363,7 @@ sfc_ef10_rx_ev_to_offloads(struct sfc_ef10_rxq *rxq, const efx_qword_t rx_ev,
 
 done:
        m->ol_flags = ol_flags;
-       m->packet_type = l2_ptype | l3_ptype | l4_ptype;
+       m->packet_type = tun_ptype | l2_ptype | l3_ptype | l4_ptype;
 }
 
 static uint16_t
@@ -515,7 +549,7 @@ sfc_ef10_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 }
 
 static const uint32_t *
-sfc_ef10_supported_ptypes_get(void)
+sfc_ef10_supported_ptypes_get(uint32_t tunnel_encaps)
 {
        static const uint32_t ef10_native_ptypes[] = {
                RTE_PTYPE_L2_ETHER,
@@ -529,8 +563,47 @@ sfc_ef10_supported_ptypes_get(void)
                RTE_PTYPE_L4_UDP,
                RTE_PTYPE_UNKNOWN
        };
+       static const uint32_t ef10_overlay_ptypes[] = {
+               RTE_PTYPE_L2_ETHER,
+               RTE_PTYPE_L2_ETHER_ARP,
+               RTE_PTYPE_L2_ETHER_VLAN,
+               RTE_PTYPE_L2_ETHER_QINQ,
+               RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
+               RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
+               RTE_PTYPE_L4_FRAG,
+               RTE_PTYPE_L4_TCP,
+               RTE_PTYPE_L4_UDP,
+               RTE_PTYPE_TUNNEL_VXLAN,
+               RTE_PTYPE_TUNNEL_NVGRE,
+               RTE_PTYPE_INNER_L2_ETHER,
+               RTE_PTYPE_INNER_L2_ETHER_VLAN,
+               RTE_PTYPE_INNER_L2_ETHER_QINQ,
+               RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
+               RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
+               RTE_PTYPE_INNER_L4_FRAG,
+               RTE_PTYPE_INNER_L4_TCP,
+               RTE_PTYPE_INNER_L4_UDP,
+               RTE_PTYPE_UNKNOWN
+       };
 
-       return ef10_native_ptypes;
+       /*
+        * The function returns static set of supported packet types,
+        * so we can't build it dynamically based on supported tunnel
+        * encapsulations and should limit to known sets.
+        */
+       switch (tunnel_encaps) {
+       case (1u << EFX_TUNNEL_PROTOCOL_VXLAN |
+             1u << EFX_TUNNEL_PROTOCOL_GENEVE |
+             1u << EFX_TUNNEL_PROTOCOL_NVGRE):
+               return ef10_overlay_ptypes;
+       default:
+               RTE_LOG(ERR, PMD,
+                       "Unexpected set of supported tunnel encapsulations: %#x\n",
+                       tunnel_encaps);
+               /* FALLTHROUGH */
+       case 0:
+               return ef10_native_ptypes;
+       }
 }
 
 static sfc_dp_rx_qdesc_npending_t sfc_ef10_rx_qdesc_npending;
@@ -707,7 +780,8 @@ struct sfc_dp_rx sfc_ef10_rx = {
                .type           = SFC_DP_RX,
                .hw_fw_caps     = SFC_DP_HW_FW_CAP_EF10,
        },
-       .features               = SFC_DP_RX_FEAT_MULTI_PROCESS,
+       .features               = SFC_DP_RX_FEAT_MULTI_PROCESS |
+                                 SFC_DP_RX_FEAT_TUNNELS,
        .qcreate                = sfc_ef10_rx_qcreate,
        .qdestroy               = sfc_ef10_rx_qdestroy,
        .qstart                 = sfc_ef10_rx_qstart,
index 837fd55..0fea997 100644 (file)
@@ -131,6 +131,10 @@ sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
                DEV_RX_OFFLOAD_UDP_CKSUM |
                DEV_RX_OFFLOAD_TCP_CKSUM;
 
+       if ((encp->enc_tunnel_encapsulations_supported != 0) &&
+           (sa->dp_rx->features & SFC_DP_RX_FEAT_TUNNELS))
+               dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
+
        dev_info->tx_offload_capa =
                DEV_TX_OFFLOAD_IPV4_CKSUM |
                DEV_TX_OFFLOAD_UDP_CKSUM |
@@ -183,8 +187,10 @@ static const uint32_t *
 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 {
        struct sfc_adapter *sa = dev->data->dev_private;
+       const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
+       uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 
-       return sa->dp_rx->supported_ptypes_get();
+       return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 }
 
 static int
index 22bf372..70a72b3 100644 (file)
@@ -193,7 +193,7 @@ sfc_efx_rx_desc_flags_to_packet_type(const unsigned int desc_flags)
 }
 
 static const uint32_t *
-sfc_efx_supported_ptypes_get(void)
+sfc_efx_supported_ptypes_get(__rte_unused uint32_t tunnel_encaps)
 {
        static const uint32_t ptypes[] = {
                RTE_PTYPE_L2_ETHER,
@@ -947,6 +947,10 @@ sfc_rx_qinit(struct sfc_adapter *sa, unsigned int sw_index,
                sa->eth_dev->data->dev_conf.rxmode.enable_scatter ?
                EFX_RXQ_FLAG_SCATTER : EFX_RXQ_FLAG_NONE;
 
+       if ((encp->enc_tunnel_encapsulations_supported != 0) &&
+           (sa->dp_rx->features & SFC_DP_RX_FEAT_TUNNELS))
+               rxq_info->type_flags |= EFX_RXQ_FLAG_INNER_CLASSES;
+
        rc = sfc_ev_qinit(sa, SFC_EVQ_TYPE_RX, sw_index,
                          rxq_info->entries, socket_id, &evq);
        if (rc != 0)