net/iavf: support flexible Rx descriptor in normal path
authorLeyi Rong <leyi.rong@intel.com>
Mon, 20 Apr 2020 06:16:17 +0000 (14:16 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 21 Apr 2020 11:57:09 +0000 (13:57 +0200)
Support flexible Rx descriptor format in normal
path of iAVF PMD.

Signed-off-by: Leyi Rong <leyi.rong@intel.com>
Reviewed-by: Qi Zhang <qi.z.zhang@intel.com>
drivers/net/iavf/iavf.h
drivers/net/iavf/iavf_ethdev.c
drivers/net/iavf/iavf_rxtx.c
drivers/net/iavf/iavf_rxtx.h
drivers/net/iavf/iavf_vchnl.c

index 368ddf6..fdb31b9 100644 (file)
@@ -103,6 +103,7 @@ struct iavf_info {
        struct virtchnl_version_info virtchnl_version;
        struct virtchnl_vf_resource *vf_res; /* VF resource */
        struct virtchnl_vsi_resource *vsi_res; /* LAN VSI */
+       uint64_t supported_rxdid;
 
        volatile enum virtchnl_ops pend_cmd; /* pending command not finished */
        uint32_t cmd_retval; /* return value of the cmd response from PF */
@@ -235,6 +236,7 @@ int iavf_disable_queues(struct iavf_adapter *adapter);
 int iavf_configure_rss_lut(struct iavf_adapter *adapter);
 int iavf_configure_rss_key(struct iavf_adapter *adapter);
 int iavf_configure_queues(struct iavf_adapter *adapter);
+int iavf_get_supported_rxdid(struct iavf_adapter *adapter);
 int iavf_config_irq_map(struct iavf_adapter *adapter);
 void iavf_add_del_all_mac_addr(struct iavf_adapter *adapter, bool add);
 int iavf_dev_link_update(struct rte_eth_dev *dev,
index 9dadee3..4280a92 100644 (file)
@@ -1245,6 +1245,14 @@ iavf_init_vf(struct rte_eth_dev *dev)
                        goto err_rss;
                }
        }
+
+       if (vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {
+               if (iavf_get_supported_rxdid(adapter) != 0) {
+                       PMD_INIT_LOG(ERR, "failed to do get supported rxdid");
+                       goto err_rss;
+               }
+       }
+
        return 0;
 err_rss:
        rte_free(vf->rss_key);
index 9eccb7c..8b2528d 100644 (file)
@@ -346,6 +346,14 @@ iavf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
                return -ENOMEM;
        }
 
+       if (vf->vf_res->vf_cap_flags &
+           VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC &&
+           vf->supported_rxdid & BIT(IAVF_RXDID_COMMS_OVS_1)) {
+               rxq->rxdid = IAVF_RXDID_COMMS_OVS_1;
+       } else {
+               rxq->rxdid = IAVF_RXDID_LEGACY_1;
+       }
+
        rxq->mp = mp;
        rxq->nb_rx_desc = nb_desc;
        rxq->rx_free_thresh = rx_free_thresh;
@@ -720,6 +728,20 @@ iavf_rxd_to_vlan_tci(struct rte_mbuf *mb, volatile union iavf_rx_desc *rxdp)
        }
 }
 
+static inline void
+iavf_flex_rxd_to_vlan_tci(struct rte_mbuf *mb,
+                         volatile union iavf_rx_flex_desc *rxdp)
+{
+       if (rte_le_to_cpu_64(rxdp->wb.status_error0) &
+               (1 << IAVF_RX_FLEX_DESC_STATUS0_L2TAG1P_S)) {
+               mb->ol_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
+               mb->vlan_tci =
+                       rte_le_to_cpu_16(rxdp->wb.l2tag1);
+       } else {
+               mb->vlan_tci = 0;
+       }
+}
+
 /* Translate the rx descriptor status and error fields to pkt flags */
 static inline uint64_t
 iavf_rxd_to_pkt_flags(uint64_t qword)
@@ -754,6 +776,87 @@ iavf_rxd_to_pkt_flags(uint64_t qword)
        return flags;
 }
 
+#ifndef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
+/* Translate the rx flex descriptor status to pkt flags */
+static inline void
+iavf_rxd_to_pkt_fields(struct rte_mbuf *mb,
+                      volatile union iavf_rx_flex_desc *rxdp)
+{
+       volatile struct iavf_32b_rx_flex_desc_comms_ovs *desc =
+                       (volatile struct iavf_32b_rx_flex_desc_comms_ovs *)rxdp;
+       uint16_t stat_err;
+
+       stat_err = rte_le_to_cpu_16(desc->status_error0);
+       if (likely(stat_err & (1 << IAVF_RX_FLEX_DESC_STATUS0_RSS_VALID_S))) {
+               mb->ol_flags |= PKT_RX_RSS_HASH;
+               mb->hash.rss = rte_le_to_cpu_32(desc->rss_hash);
+       }
+}
+#endif
+
+#define IAVF_RX_FLEX_ERR0_BITS \
+       ((1 << IAVF_RX_FLEX_DESC_STATUS0_HBO_S) |       \
+        (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_IPE_S) |  \
+        (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_L4E_S) |  \
+        (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_EIPE_S) | \
+        (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_EUDPE_S) |        \
+        (1 << IAVF_RX_FLEX_DESC_STATUS0_RXE_S))
+
+/* Rx L3/L4 checksum */
+static inline uint64_t
+iavf_flex_rxd_error_to_pkt_flags(uint16_t stat_err0)
+{
+       uint64_t flags = 0;
+
+       /* check if HW has decoded the packet and checksum */
+       if (unlikely(!(stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_L3L4P_S))))
+               return 0;
+
+       if (likely(!(stat_err0 & IAVF_RX_FLEX_ERR0_BITS))) {
+               flags |= (PKT_RX_IP_CKSUM_GOOD | PKT_RX_L4_CKSUM_GOOD);
+               return flags;
+       }
+
+       if (unlikely(stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_IPE_S)))
+               flags |= PKT_RX_IP_CKSUM_BAD;
+       else
+               flags |= PKT_RX_IP_CKSUM_GOOD;
+
+       if (unlikely(stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_L4E_S)))
+               flags |= PKT_RX_L4_CKSUM_BAD;
+       else
+               flags |= PKT_RX_L4_CKSUM_GOOD;
+
+       if (unlikely(stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_XSUM_EIPE_S)))
+               flags |= PKT_RX_EIP_CKSUM_BAD;
+
+       return flags;
+}
+
+/* If the number of free RX descriptors is greater than the RX free
+ * threshold of the queue, advance the Receive Descriptor Tail (RDT)
+ * register. Update the RDT with the value of the last processed RX
+ * descriptor minus 1, to guarantee that the RDT register is never
+ * equal to the RDH register, which creates a "full" ring situation
+ * from the hardware point of view.
+ */
+static inline void
+iavf_update_rx_tail(struct iavf_rx_queue *rxq, uint16_t nb_hold, uint16_t rx_id)
+{
+       nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
+
+       if (nb_hold > rxq->rx_free_thresh) {
+               PMD_RX_LOG(DEBUG,
+                          "port_id=%u queue_id=%u rx_tail=%u nb_hold=%u",
+                          rxq->port_id, rxq->queue_id, rx_id, nb_hold);
+               rx_id = (uint16_t)((rx_id == 0) ?
+                       (rxq->nb_rx_desc - 1) : (rx_id - 1));
+               IAVF_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
+               nb_hold = 0;
+       }
+       rxq->nb_rx_hold = nb_hold;
+}
+
 /* implement recv_pkts */
 uint16_t
 iavf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
@@ -854,23 +957,260 @@ iavf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        }
        rxq->rx_tail = rx_id;
 
-       /* If the number of free RX descriptors is greater than the RX free
-        * threshold of the queue, advance the receive tail register of queue.
-        * Update that register with the value of the last processed RX
-        * descriptor minus 1.
-        */
-       nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
-       if (nb_hold > rxq->rx_free_thresh) {
-               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
-                          "nb_hold=%u nb_rx=%u",
-                          rxq->port_id, rxq->queue_id,
-                          rx_id, nb_hold, nb_rx);
-               rx_id = (uint16_t)((rx_id == 0) ?
-                       (rxq->nb_rx_desc - 1) : (rx_id - 1));
-               IAVF_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
-               nb_hold = 0;
+       iavf_update_rx_tail(rxq, nb_hold, rx_id);
+
+       return nb_rx;
+}
+
+/* implement recv_pkts for flexible Rx descriptor */
+uint16_t
+iavf_recv_pkts_flex_rxd(void *rx_queue,
+                       struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+       volatile union iavf_rx_desc *rx_ring;
+       volatile union iavf_rx_flex_desc *rxdp;
+       struct iavf_rx_queue *rxq;
+       union iavf_rx_flex_desc rxd;
+       struct rte_mbuf *rxe;
+       struct rte_eth_dev *dev;
+       struct rte_mbuf *rxm;
+       struct rte_mbuf *nmb;
+       uint16_t nb_rx;
+       uint16_t rx_stat_err0;
+       uint16_t rx_packet_len;
+       uint16_t rx_id, nb_hold;
+       uint64_t dma_addr;
+       uint64_t pkt_flags;
+       const uint32_t *ptype_tbl;
+
+       nb_rx = 0;
+       nb_hold = 0;
+       rxq = rx_queue;
+       rx_id = rxq->rx_tail;
+       rx_ring = rxq->rx_ring;
+       ptype_tbl = rxq->vsi->adapter->ptype_tbl;
+
+       while (nb_rx < nb_pkts) {
+               rxdp = (volatile union iavf_rx_flex_desc *)&rx_ring[rx_id];
+               rx_stat_err0 = rte_le_to_cpu_16(rxdp->wb.status_error0);
+
+               /* Check the DD bit first */
+               if (!(rx_stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_DD_S)))
+                       break;
+               IAVF_DUMP_RX_DESC(rxq, rxdp, rx_id);
+
+               nmb = rte_mbuf_raw_alloc(rxq->mp);
+               if (unlikely(!nmb)) {
+                       dev = &rte_eth_devices[rxq->port_id];
+                       dev->data->rx_mbuf_alloc_failed++;
+                       PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
+                                  "queue_id=%u", rxq->port_id, rxq->queue_id);
+                       break;
+               }
+
+               rxd = *rxdp;
+               nb_hold++;
+               rxe = rxq->sw_ring[rx_id];
+               rx_id++;
+               if (unlikely(rx_id == rxq->nb_rx_desc))
+                       rx_id = 0;
+
+               /* Prefetch next mbuf */
+               rte_prefetch0(rxq->sw_ring[rx_id]);
+
+               /* When next RX descriptor is on a cache line boundary,
+                * prefetch the next 4 RX descriptors and next 8 pointers
+                * to mbufs.
+                */
+               if ((rx_id & 0x3) == 0) {
+                       rte_prefetch0(&rx_ring[rx_id]);
+                       rte_prefetch0(rxq->sw_ring[rx_id]);
+               }
+               rxm = rxe;
+               rxe = nmb;
+               dma_addr =
+                       rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
+               rxdp->read.hdr_addr = 0;
+               rxdp->read.pkt_addr = dma_addr;
+
+               rx_packet_len = (rte_le_to_cpu_16(rxd.wb.pkt_len) &
+                               IAVF_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
+               rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
+               rxm->nb_segs = 1;
+               rxm->next = NULL;
+               rxm->pkt_len = rx_packet_len;
+               rxm->data_len = rx_packet_len;
+               rxm->port = rxq->port_id;
+               rxm->ol_flags = 0;
+               rxm->packet_type = ptype_tbl[IAVF_RX_FLEX_DESC_PTYPE_M &
+                       rte_le_to_cpu_16(rxd.wb.ptype_flex_flags0)];
+               iavf_flex_rxd_to_vlan_tci(rxm, &rxd);
+#ifndef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
+               iavf_rxd_to_pkt_fields(rxm, &rxd);
+#endif
+               pkt_flags = iavf_flex_rxd_error_to_pkt_flags(rx_stat_err0);
+               rxm->ol_flags |= pkt_flags;
+
+               rx_pkts[nb_rx++] = rxm;
        }
-       rxq->nb_rx_hold = nb_hold;
+       rxq->rx_tail = rx_id;
+
+       iavf_update_rx_tail(rxq, nb_hold, rx_id);
+
+       return nb_rx;
+}
+
+/* implement recv_scattered_pkts for flexible Rx descriptor */
+uint16_t
+iavf_recv_scattered_pkts_flex_rxd(void *rx_queue, struct rte_mbuf **rx_pkts,
+                                 uint16_t nb_pkts)
+{
+       struct iavf_rx_queue *rxq = rx_queue;
+       union iavf_rx_flex_desc rxd;
+       struct rte_mbuf *rxe;
+       struct rte_mbuf *first_seg = rxq->pkt_first_seg;
+       struct rte_mbuf *last_seg = rxq->pkt_last_seg;
+       struct rte_mbuf *nmb, *rxm;
+       uint16_t rx_id = rxq->rx_tail;
+       uint16_t nb_rx = 0, nb_hold = 0, rx_packet_len;
+       struct rte_eth_dev *dev;
+       uint16_t rx_stat_err0;
+       uint64_t dma_addr;
+       uint64_t pkt_flags;
+
+       volatile union iavf_rx_desc *rx_ring = rxq->rx_ring;
+       volatile union iavf_rx_flex_desc *rxdp;
+       const uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl;
+
+       while (nb_rx < nb_pkts) {
+               rxdp = (volatile union iavf_rx_flex_desc *)&rx_ring[rx_id];
+               rx_stat_err0 = rte_le_to_cpu_16(rxdp->wb.status_error0);
+
+               /* Check the DD bit */
+               if (!(rx_stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_DD_S)))
+                       break;
+               IAVF_DUMP_RX_DESC(rxq, rxdp, rx_id);
+
+               nmb = rte_mbuf_raw_alloc(rxq->mp);
+               if (unlikely(!nmb)) {
+                       PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
+                                  "queue_id=%u", rxq->port_id, rxq->queue_id);
+                       dev = &rte_eth_devices[rxq->port_id];
+                       dev->data->rx_mbuf_alloc_failed++;
+                       break;
+               }
+
+               rxd = *rxdp;
+               nb_hold++;
+               rxe = rxq->sw_ring[rx_id];
+               rx_id++;
+               if (rx_id == rxq->nb_rx_desc)
+                       rx_id = 0;
+
+               /* Prefetch next mbuf */
+               rte_prefetch0(rxq->sw_ring[rx_id]);
+
+               /* When next RX descriptor is on a cache line boundary,
+                * prefetch the next 4 RX descriptors and next 8 pointers
+                * to mbufs.
+                */
+               if ((rx_id & 0x3) == 0) {
+                       rte_prefetch0(&rx_ring[rx_id]);
+                       rte_prefetch0(rxq->sw_ring[rx_id]);
+               }
+
+               rxm = rxe;
+               rxe = nmb;
+               dma_addr =
+                       rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
+
+               /* Set data buffer address and data length of the mbuf */
+               rxdp->read.hdr_addr = 0;
+               rxdp->read.pkt_addr = dma_addr;
+               rx_packet_len = rte_le_to_cpu_16(rxd.wb.pkt_len) &
+                               IAVF_RX_FLX_DESC_PKT_LEN_M;
+               rxm->data_len = rx_packet_len;
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
+
+               /* If this is the first buffer of the received packet, set the
+                * pointer to the first mbuf of the packet and initialize its
+                * context. Otherwise, update the total length and the number
+                * of segments of the current scattered packet, and update the
+                * pointer to the last mbuf of the current packet.
+                */
+               if (!first_seg) {
+                       first_seg = rxm;
+                       first_seg->nb_segs = 1;
+                       first_seg->pkt_len = rx_packet_len;
+               } else {
+                       first_seg->pkt_len =
+                               (uint16_t)(first_seg->pkt_len +
+                                               rx_packet_len);
+                       first_seg->nb_segs++;
+                       last_seg->next = rxm;
+               }
+
+               /* If this is not the last buffer of the received packet,
+                * update the pointer to the last mbuf of the current scattered
+                * packet and continue to parse the RX ring.
+                */
+               if (!(rx_stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_EOF_S))) {
+                       last_seg = rxm;
+                       continue;
+               }
+
+               /* This is the last buffer of the received packet. If the CRC
+                * is not stripped by the hardware:
+                *  - Subtract the CRC length from the total packet length.
+                *  - If the last buffer only contains the whole CRC or a part
+                *  of it, free the mbuf associated to the last buffer. If part
+                *  of the CRC is also contained in the previous mbuf, subtract
+                *  the length of that CRC part from the data length of the
+                *  previous mbuf.
+                */
+               rxm->next = NULL;
+               if (unlikely(rxq->crc_len > 0)) {
+                       first_seg->pkt_len -= RTE_ETHER_CRC_LEN;
+                       if (rx_packet_len <= RTE_ETHER_CRC_LEN) {
+                               rte_pktmbuf_free_seg(rxm);
+                               first_seg->nb_segs--;
+                               last_seg->data_len =
+                                       (uint16_t)(last_seg->data_len -
+                                       (RTE_ETHER_CRC_LEN - rx_packet_len));
+                               last_seg->next = NULL;
+                       } else {
+                               rxm->data_len = (uint16_t)(rx_packet_len -
+                                                       RTE_ETHER_CRC_LEN);
+                       }
+               }
+
+               first_seg->port = rxq->port_id;
+               first_seg->ol_flags = 0;
+               first_seg->packet_type = ptype_tbl[IAVF_RX_FLEX_DESC_PTYPE_M &
+                       rte_le_to_cpu_16(rxd.wb.ptype_flex_flags0)];
+               iavf_flex_rxd_to_vlan_tci(first_seg, &rxd);
+#ifndef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
+               iavf_rxd_to_pkt_fields(first_seg, &rxd);
+#endif
+               pkt_flags = iavf_flex_rxd_error_to_pkt_flags(rx_stat_err0);
+
+               first_seg->ol_flags |= pkt_flags;
+
+               /* Prefetch data of first segment, if configured to do so. */
+               rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
+                                         first_seg->data_off));
+               rx_pkts[nb_rx++] = first_seg;
+               first_seg = NULL;
+       }
+
+       /* Record index of the next RX descriptor to probe. */
+       rxq->rx_tail = rx_id;
+       rxq->pkt_first_seg = first_seg;
+       rxq->pkt_last_seg = last_seg;
+
+       iavf_update_rx_tail(rxq, nb_hold, rx_id);
 
        return nb_rx;
 }
@@ -1027,30 +1367,90 @@ iavf_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        rxq->pkt_first_seg = first_seg;
        rxq->pkt_last_seg = last_seg;
 
-       /* If the number of free RX descriptors is greater than the RX free
-        * threshold of the queue, advance the Receive Descriptor Tail (RDT)
-        * register. Update the RDT with the value of the last processed RX
-        * descriptor minus 1, to guarantee that the RDT register is never
-        * equal to the RDH register, which creates a "full" ring situtation
-        * from the hardware point of view.
+       iavf_update_rx_tail(rxq, nb_hold, rx_id);
+
+       return nb_rx;
+}
+
+#define IAVF_LOOK_AHEAD 8
+static inline int
+iavf_rx_scan_hw_ring_flex_rxd(struct iavf_rx_queue *rxq)
+{
+       volatile union iavf_rx_flex_desc *rxdp;
+       struct rte_mbuf **rxep;
+       struct rte_mbuf *mb;
+       uint16_t stat_err0;
+       uint16_t pkt_len;
+       int32_t s[IAVF_LOOK_AHEAD], nb_dd;
+       int32_t i, j, nb_rx = 0;
+       uint64_t pkt_flags;
+       const uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl;
+
+       rxdp = (volatile union iavf_rx_flex_desc *)&rxq->rx_ring[rxq->rx_tail];
+       rxep = &rxq->sw_ring[rxq->rx_tail];
+
+       stat_err0 = rte_le_to_cpu_16(rxdp->wb.status_error0);
+
+       /* Make sure there is at least 1 packet to receive */
+       if (!(stat_err0 & (1 << IAVF_RX_FLEX_DESC_STATUS0_DD_S)))
+               return 0;
+
+       /* Scan LOOK_AHEAD descriptors at a time to determine which
+        * descriptors reference packets that are ready to be received.
         */
-       nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
-       if (nb_hold > rxq->rx_free_thresh) {
-               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
-                          "nb_hold=%u nb_rx=%u",
-                          rxq->port_id, rxq->queue_id,
-                          rx_id, nb_hold, nb_rx);
-               rx_id = (uint16_t)(rx_id == 0 ?
-                       (rxq->nb_rx_desc - 1) : (rx_id - 1));
-               IAVF_PCI_REG_WRITE(rxq->qrx_tail, rx_id);
-               nb_hold = 0;
+       for (i = 0; i < IAVF_RX_MAX_BURST; i += IAVF_LOOK_AHEAD,
+            rxdp += IAVF_LOOK_AHEAD, rxep += IAVF_LOOK_AHEAD) {
+               /* Read desc statuses backwards to avoid race condition */
+               for (j = IAVF_LOOK_AHEAD - 1; j >= 0; j--)
+                       s[j] = rte_le_to_cpu_16(rxdp[j].wb.status_error0);
+
+               rte_smp_rmb();
+
+               /* Compute how many status bits were set */
+               for (j = 0, nb_dd = 0; j < IAVF_LOOK_AHEAD; j++)
+                       nb_dd += s[j] & (1 << IAVF_RX_FLEX_DESC_STATUS0_DD_S);
+
+               nb_rx += nb_dd;
+
+               /* Translate descriptor info to mbuf parameters */
+               for (j = 0; j < nb_dd; j++) {
+                       IAVF_DUMP_RX_DESC(rxq, &rxdp[j],
+                                         rxq->rx_tail +
+                                         i * IAVF_LOOK_AHEAD + j);
+
+                       mb = rxep[j];
+                       pkt_len = (rte_le_to_cpu_16(rxdp[j].wb.pkt_len) &
+                               IAVF_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+                       mb->data_len = pkt_len;
+                       mb->pkt_len = pkt_len;
+                       mb->ol_flags = 0;
+
+                       mb->packet_type = ptype_tbl[IAVF_RX_FLEX_DESC_PTYPE_M &
+                               rte_le_to_cpu_16(rxdp[j].wb.ptype_flex_flags0)];
+                       iavf_flex_rxd_to_vlan_tci(mb, &rxdp[j]);
+#ifndef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
+                       iavf_rxd_to_pkt_fields(mb, &rxdp[j]);
+#endif
+                       stat_err0 = rte_le_to_cpu_16(rxdp[j].wb.status_error0);
+                       pkt_flags = iavf_flex_rxd_error_to_pkt_flags(stat_err0);
+
+                       mb->ol_flags |= pkt_flags;
+               }
+
+               for (j = 0; j < IAVF_LOOK_AHEAD; j++)
+                       rxq->rx_stage[i + j] = rxep[j];
+
+               if (nb_dd != IAVF_LOOK_AHEAD)
+                       break;
        }
-       rxq->nb_rx_hold = nb_hold;
+
+       /* Clear software ring entries */
+       for (i = 0; i < nb_rx; i++)
+               rxq->sw_ring[rxq->rx_tail + i] = NULL;
 
        return nb_rx;
 }
 
-#define IAVF_LOOK_AHEAD 8
 static inline int
 iavf_rx_scan_hw_ring(struct iavf_rx_queue *rxq)
 {
@@ -1219,7 +1619,10 @@ rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        if (rxq->rx_nb_avail)
                return iavf_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
 
-       nb_rx = (uint16_t)iavf_rx_scan_hw_ring(rxq);
+       if (rxq->rxdid == IAVF_RXDID_COMMS_OVS_1)
+               nb_rx = (uint16_t)iavf_rx_scan_hw_ring_flex_rxd(rxq);
+       else
+               nb_rx = (uint16_t)iavf_rx_scan_hw_ring(rxq);
        rxq->rx_next_avail = 0;
        rxq->rx_nb_avail = nb_rx;
        rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
@@ -1663,6 +2066,7 @@ iavf_set_rx_function(struct rte_eth_dev *dev)
 {
        struct iavf_adapter *adapter =
                IAVF_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
+       struct iavf_info *vf = IAVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
 #ifdef RTE_ARCH_X86
        struct iavf_rx_queue *rxq;
        int i;
@@ -1702,7 +2106,10 @@ iavf_set_rx_function(struct rte_eth_dev *dev)
        if (dev->data->scattered_rx) {
                PMD_DRV_LOG(DEBUG, "Using a Scattered Rx callback (port=%d).",
                            dev->data->port_id);
-               dev->rx_pkt_burst = iavf_recv_scattered_pkts;
+               if (vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC)
+                       dev->rx_pkt_burst = iavf_recv_scattered_pkts_flex_rxd;
+               else
+                       dev->rx_pkt_burst = iavf_recv_scattered_pkts;
        } else if (adapter->rx_bulk_alloc_allowed) {
                PMD_DRV_LOG(DEBUG, "Using bulk Rx callback (port=%d).",
                            dev->data->port_id);
@@ -1710,7 +2117,10 @@ iavf_set_rx_function(struct rte_eth_dev *dev)
        } else {
                PMD_DRV_LOG(DEBUG, "Using Basic Rx callback (port=%d).",
                            dev->data->port_id);
-               dev->rx_pkt_burst = iavf_recv_pkts;
+               if (vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC)
+                       dev->rx_pkt_burst = iavf_recv_pkts_flex_rxd;
+               else
+                       dev->rx_pkt_burst = iavf_recv_pkts;
        }
 }
 
@@ -1797,6 +2207,7 @@ iavf_dev_rxq_count(struct rte_eth_dev *dev, uint16_t queue_id)
 
        rxq = dev->data->rx_queues[queue_id];
        rxdp = &rxq->rx_ring[rxq->rx_tail];
+
        while ((desc < rxq->nb_rx_desc) &&
               ((rte_le_to_cpu_64(rxdp->wb.qword1.status_error_len) &
                 IAVF_RXD_QW1_STATUS_MASK) >> IAVF_RXD_QW1_STATUS_SHIFT) &
index d04233b..8246e79 100644 (file)
 /* HW desc structure, both 16-byte and 32-byte types are supported */
 #ifdef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
 #define iavf_rx_desc iavf_16byte_rx_desc
+#define iavf_rx_flex_desc iavf_16b_rx_flex_desc
 #else
 #define iavf_rx_desc iavf_32byte_rx_desc
+#define iavf_rx_flex_desc iavf_32b_rx_flex_desc
 #endif
 
 struct iavf_rxq_ops {
@@ -87,6 +89,7 @@ struct iavf_rx_queue {
        struct rte_mbuf *pkt_first_seg; /* first segment of current packet */
        struct rte_mbuf *pkt_last_seg;  /* last segment of current packet */
        struct rte_mbuf fake_mbuf;      /* dummy mbuf */
+       uint8_t rxdid;
 
        /* used for VPMD */
        uint16_t rxrearm_nb;       /* number of remaining to be re-armed */
@@ -379,9 +382,15 @@ void iavf_dev_tx_queue_release(void *txq);
 void iavf_stop_queues(struct rte_eth_dev *dev);
 uint16_t iavf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                       uint16_t nb_pkts);
+uint16_t iavf_recv_pkts_flex_rxd(void *rx_queue,
+                                struct rte_mbuf **rx_pkts,
+                                uint16_t nb_pkts);
 uint16_t iavf_recv_scattered_pkts(void *rx_queue,
                                 struct rte_mbuf **rx_pkts,
                                 uint16_t nb_pkts);
+uint16_t iavf_recv_scattered_pkts_flex_rxd(void *rx_queue,
+                                          struct rte_mbuf **rx_pkts,
+                                          uint16_t nb_pkts);
 uint16_t iavf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                       uint16_t nb_pkts);
 uint16_t iavf_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
index b7fb05d..b913f06 100644 (file)
@@ -88,6 +88,7 @@ iavf_execute_vf_cmd(struct iavf_adapter *adapter, struct iavf_cmd_info *args)
                break;
        case VIRTCHNL_OP_VERSION:
        case VIRTCHNL_OP_GET_VF_RESOURCES:
+       case VIRTCHNL_OP_GET_SUPPORTED_RXDIDS:
                /* for init virtchnl ops, need to poll the response */
                do {
                        ret = iavf_read_msg_from_pf(adapter, args->out_size,
@@ -338,7 +339,8 @@ iavf_get_vf_resource(struct iavf_adapter *adapter)
         * add advanced/optional offload capabilities
         */
 
-       caps = IAVF_BASIC_OFFLOAD_CAPS | VIRTCHNL_VF_CAP_ADV_LINK_SPEED;
+       caps = IAVF_BASIC_OFFLOAD_CAPS | VIRTCHNL_VF_CAP_ADV_LINK_SPEED |
+               VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC;
 
        args.in_args = (uint8_t *)&caps;
        args.in_args_size = sizeof(caps);
@@ -375,6 +377,32 @@ iavf_get_vf_resource(struct iavf_adapter *adapter)
        return 0;
 }
 
+int
+iavf_get_supported_rxdid(struct iavf_adapter *adapter)
+{
+       struct iavf_info *vf = IAVF_DEV_PRIVATE_TO_VF(adapter);
+       struct iavf_cmd_info args;
+       int ret;
+
+       args.ops = VIRTCHNL_OP_GET_SUPPORTED_RXDIDS;
+       args.in_args = NULL;
+       args.in_args_size = 0;
+       args.out_buffer = vf->aq_resp;
+       args.out_size = IAVF_AQ_BUF_SZ;
+
+       ret = iavf_execute_vf_cmd(adapter, &args);
+       if (ret) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to execute command of OP_GET_SUPPORTED_RXDIDS");
+               return ret;
+       }
+
+       vf->supported_rxdid =
+               ((struct virtchnl_supported_rxdids *)args.out_buffer)->supported_rxdids;
+
+       return 0;
+}
+
 int
 iavf_enable_queues(struct iavf_adapter *adapter)
 {
@@ -567,6 +595,31 @@ iavf_configure_queues(struct iavf_adapter *adapter)
                        vc_qp->rxq.ring_len = rxq[i]->nb_rx_desc;
                        vc_qp->rxq.dma_ring_addr = rxq[i]->rx_ring_phys_addr;
                        vc_qp->rxq.databuffer_size = rxq[i]->rx_buf_len;
+
+#ifndef RTE_LIBRTE_IAVF_16BYTE_RX_DESC
+                       if (vf->vf_res->vf_cap_flags &
+                           VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC &&
+                           vf->supported_rxdid & BIT(IAVF_RXDID_COMMS_OVS_1)) {
+                               vc_qp->rxq.rxdid = IAVF_RXDID_COMMS_OVS_1;
+                               PMD_DRV_LOG(NOTICE, "request RXDID == %d in "
+                                           "Queue[%d]", vc_qp->rxq.rxdid, i);
+                       } else {
+                               vc_qp->rxq.rxdid = IAVF_RXDID_LEGACY_1;
+                               PMD_DRV_LOG(NOTICE, "request RXDID == %d in "
+                                           "Queue[%d]", vc_qp->rxq.rxdid, i);
+                       }
+#else
+                       if (vf->vf_res->vf_cap_flags &
+                           VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC &&
+                           vf->supported_rxdid & BIT(IAVF_RXDID_LEGACY_0)) {
+                               vc_qp->rxq.rxdid = IAVF_RXDID_LEGACY_0;
+                               PMD_DRV_LOG(NOTICE, "request RXDID == %d in "
+                                           "Queue[%d]", vc_qp->rxq.rxdid, i);
+                       } else {
+                               PMD_DRV_LOG(ERR, "RXDID == 0 is not supported");
+                               return -1;
+                       }
+#endif
                }
        }