net/bnxt: fix forwarding with higher mbuf size
[dpdk.git] / drivers / net / bnxt / bnxt_rxr.c
index 75d2c76..f60880c 100644 (file)
@@ -17,6 +17,9 @@
 #include "bnxt_rxr.h"
 #include "bnxt_rxq.h"
 #include "hsi_struct_def_dpdk.h"
+#ifdef RTE_LIBRTE_IEEE1588
+#include "bnxt_hwrm.h"
+#endif
 
 /*
  * RX Ring handling
@@ -348,6 +351,30 @@ bnxt_parse_pkt_type(struct rx_pkt_cmpl *rxcmp, struct rx_pkt_cmpl_hi *rxcmp1)
        return pkt_type;
 }
 
+#ifdef RTE_LIBRTE_IEEE1588
+static void
+bnxt_get_rx_ts_thor(struct bnxt *bp, uint32_t rx_ts_cmpl)
+{
+       uint64_t systime_cycles = 0;
+
+       if (!BNXT_CHIP_THOR(bp))
+               return;
+
+       /* On Thor, Rx timestamps are provided directly in the
+        * Rx completion records to the driver. Only 32 bits of
+        * the timestamp is present in the completion. Driver needs
+        * to read the current 48 bit free running timer using the
+        * HWRM_PORT_TS_QUERY command and combine the upper 16 bits
+        * from the HWRM response with the lower 32 bits in the
+        * Rx completion to produce the 48 bit timestamp for the Rx packet
+        */
+       bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
+                               &systime_cycles);
+       bp->ptp_cfg->rx_timestamp = (systime_cycles & 0xFFFF00000000);
+       bp->ptp_cfg->rx_timestamp |= rx_ts_cmpl;
+}
+#endif
+
 static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
                            struct bnxt_rx_queue *rxq, uint32_t *raw_cons)
 {
@@ -362,6 +389,8 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
        int rc = 0;
        uint8_t agg_buf = 0;
        uint16_t cmp_type;
+       uint32_t flags2_f = 0;
+       uint16_t flags_type;
 
        rxcmp = (struct rx_pkt_cmpl *)
            &cpr->cp_desc_ring[cp_cons];
@@ -417,18 +446,22 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
        mbuf->data_len = mbuf->pkt_len;
        mbuf->port = rxq->port_id;
        mbuf->ol_flags = 0;
-       if (rxcmp->flags_type & RX_PKT_CMPL_FLAGS_RSS_VALID) {
+
+       flags_type = rte_le_to_cpu_16(rxcmp->flags_type);
+       if (flags_type & RX_PKT_CMPL_FLAGS_RSS_VALID) {
                mbuf->hash.rss = rxcmp->rss_hash;
                mbuf->ol_flags |= PKT_RX_RSS_HASH;
        } else {
                mbuf->hash.fdir.id = rxcmp1->cfa_code;
                mbuf->ol_flags |= PKT_RX_FDIR | PKT_RX_FDIR_ID;
        }
-
-       if ((rxcmp->flags_type & rte_cpu_to_le_16(RX_PKT_CMPL_FLAGS_MASK)) ==
-            RX_PKT_CMPL_FLAGS_ITYPE_PTP_W_TIMESTAMP)
+#ifdef RTE_LIBRTE_IEEE1588
+       if (unlikely((flags_type & RX_PKT_CMPL_FLAGS_MASK) ==
+                    RX_PKT_CMPL_FLAGS_ITYPE_PTP_W_TIMESTAMP)) {
                mbuf->ol_flags |= PKT_RX_IEEE1588_PTP | PKT_RX_IEEE1588_TMST;
-
+               bnxt_get_rx_ts_thor(rxq->bp, rxcmp1->reorder);
+       }
+#endif
        if (agg_buf)
                bnxt_rx_pages(rxq, mbuf, &tmp_raw_cons, agg_buf);
 
@@ -440,19 +473,41 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
                mbuf->ol_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
        }
 
-       if (likely(RX_CMP_IP_CS_OK(rxcmp1)))
-               mbuf->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
-       else if (likely(RX_CMP_IP_CS_UNKNOWN(rxcmp1)))
-               mbuf->ol_flags |= PKT_RX_IP_CKSUM_UNKNOWN;
-       else
+       flags2_f = flags2_0xf(rxcmp1);
+       /* IP Checksum */
+       if (unlikely(((IS_IP_NONTUNNEL_PKT(flags2_f)) &&
+                     (RX_CMP_IP_CS_ERROR(rxcmp1))) ||
+                    (IS_IP_TUNNEL_PKT(flags2_f) &&
+                     (RX_CMP_IP_OUTER_CS_ERROR(rxcmp1))))) {
                mbuf->ol_flags |= PKT_RX_IP_CKSUM_BAD;
+       } else if (unlikely(RX_CMP_IP_CS_UNKNOWN(rxcmp1))) {
+               mbuf->ol_flags |= PKT_RX_IP_CKSUM_UNKNOWN;
+       } else {
+               mbuf->ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+       }
 
-       if (likely(RX_CMP_L4_CS_OK(rxcmp1)))
-               mbuf->ol_flags |= PKT_RX_L4_CKSUM_GOOD;
-       else if (likely(RX_CMP_L4_CS_UNKNOWN(rxcmp1)))
+       /* L4 Checksum */
+       if (likely(IS_L4_NONTUNNEL_PKT(flags2_f))) {
+               if (unlikely(RX_CMP_L4_INNER_CS_ERR2(rxcmp1)))
+                       mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
+               else
+                       mbuf->ol_flags |= PKT_RX_L4_CKSUM_GOOD;
+       } else if (IS_L4_TUNNEL_PKT(flags2_f)) {
+               if (unlikely(RX_CMP_L4_INNER_CS_ERR2(rxcmp1)))
+                       mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
+               else
+                       mbuf->ol_flags |= PKT_RX_L4_CKSUM_GOOD;
+               if (unlikely(RX_CMP_L4_OUTER_CS_ERR2(rxcmp1))) {
+                       mbuf->ol_flags |= PKT_RX_OUTER_L4_CKSUM_BAD;
+               } else if (unlikely(IS_L4_TUNNEL_PKT_ONLY_INNER_L4_CS
+                                   (flags2_f))) {
+                       mbuf->ol_flags |= PKT_RX_OUTER_L4_CKSUM_UNKNOWN;
+               } else {
+                       mbuf->ol_flags |= PKT_RX_OUTER_L4_CKSUM_GOOD;
+               }
+       } else if (unlikely(RX_CMP_L4_CS_UNKNOWN(rxcmp1))) {
                mbuf->ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
-       else
-               mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
+       }
 
        mbuf->packet_type = bnxt_parse_pkt_type(rxcmp, rxcmp1);
 
@@ -516,6 +571,9 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        int rc = 0;
        bool evt = false;
 
+       if (unlikely(is_bnxt_in_error(rxq->bp)))
+               return 0;
+
        /* If Rx Q was stopped return. RxQ0 cannot be stopped. */
        if (unlikely(((rxq->rx_deferred_start ||
                       !rte_spinlock_trylock(&rxq->lock)) &&
@@ -541,7 +599,7 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                                nb_rx_pkts++;
                        if (rc == -EBUSY)       /* partial completion */
                                break;
-               } else {
+               } else if (!BNXT_NUM_ASYNC_CPR(rxq->bp)) {
                        evt =
                        bnxt_event_hwrm_resp_handler(rxq->bp,
                                                     (struct cmpl_base *)rxcmp);
@@ -602,6 +660,20 @@ done:
        return nb_rx_pkts;
 }
 
+/*
+ * Dummy DPDK callback for RX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ */
+uint16_t
+bnxt_dummy_recv_pkts(void *rx_queue __rte_unused,
+                    struct rte_mbuf **rx_pkts __rte_unused,
+                    uint16_t nb_pkts __rte_unused)
+{
+       return 0;
+}
+
 void bnxt_free_rx_rings(struct bnxt *bp)
 {
        int i;
@@ -637,12 +709,11 @@ void bnxt_free_rx_rings(struct bnxt *bp)
 int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
 {
        struct bnxt_cp_ring_info *cpr;
+       struct bnxt_cp_ring_info *nqr;
        struct bnxt_rx_ring_info *rxr;
        struct bnxt_ring *ring;
 
-       rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-               RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
-       rxq->rx_buf_size = rxq->rx_buf_use_size + sizeof(struct rte_mbuf);
+       rxq->rx_buf_size = BNXT_MAX_PKT_LEN + sizeof(struct rte_mbuf);
 
        rxr = rte_zmalloc_socket("bnxt_rx_ring",
                                 sizeof(struct bnxt_rx_ring_info),
@@ -685,6 +756,32 @@ int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
        ring->vmem_size = 0;
        ring->vmem = NULL;
 
+       if (BNXT_HAS_NQ(rxq->bp)) {
+               nqr = rte_zmalloc_socket("bnxt_rx_ring_cq",
+                                        sizeof(struct bnxt_cp_ring_info),
+                                        RTE_CACHE_LINE_SIZE, socket_id);
+               if (nqr == NULL)
+                       return -ENOMEM;
+
+               rxq->nq_ring = nqr;
+
+               ring = rte_zmalloc_socket("bnxt_rx_ring_struct",
+                                         sizeof(struct bnxt_ring),
+                                         RTE_CACHE_LINE_SIZE, socket_id);
+               if (ring == NULL)
+                       return -ENOMEM;
+
+               nqr->cp_ring_struct = ring;
+               ring->ring_size =
+                       rte_align32pow2(rxr->rx_ring_struct->ring_size *
+                                       (2 + AGG_RING_SIZE_FACTOR));
+               ring->ring_mask = ring->ring_size - 1;
+               ring->bd = (void *)nqr->cp_desc_ring;
+               ring->bd_dma = nqr->cp_desc_mapping;
+               ring->vmem_size = 0;
+               ring->vmem = NULL;
+       }
+
        /* Allocate Aggregator rings */
        ring = rte_zmalloc_socket("bnxt_rx_ring_struct",
                                   sizeof(struct bnxt_ring),
@@ -727,8 +824,7 @@ int bnxt_init_one_rx_ring(struct bnxt_rx_queue *rxq)
        uint16_t size;
 
        size = rte_pktmbuf_data_room_size(rxq->mb_pool) - RTE_PKTMBUF_HEADROOM;
-       if (rxq->rx_buf_use_size <= size)
-               size = rxq->rx_buf_use_size;
+       size = RTE_MIN(BNXT_MAX_PKT_LEN, size);
 
        type = RX_PROD_PKT_BD_TYPE_RX_PROD_PKT | RX_PROD_PKT_BD_FLAGS_EOP_PAD;