net/ice: fix TCP checksum offload
[dpdk.git] / drivers / net / ice / ice_rxtx.c
index 5f4fc42..ddf6a93 100644 (file)
@@ -236,23 +236,12 @@ _ice_rx_queue_release_mbufs(struct ice_rx_queue *rxq)
                        rxq->sw_ring[i].mbuf = NULL;
                }
        }
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
-               if (rxq->rx_nb_avail == 0)
-                       return;
-               for (i = 0; i < rxq->rx_nb_avail; i++) {
-                       struct rte_mbuf *mbuf;
-
-                       mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
-                       rte_pktmbuf_free_seg(mbuf);
-               }
-               rxq->rx_nb_avail = 0;
-#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
-}
+       if (rxq->rx_nb_avail == 0)
+               return;
+       for (i = 0; i < rxq->rx_nb_avail; i++)
+               rte_pktmbuf_free_seg(rxq->rx_stage[rxq->rx_next_avail + i]);
 
-static void
-ice_rx_queue_release_mbufs(struct ice_rx_queue *rxq)
-{
-       rxq->rx_rel_mbufs(rxq);
+       rxq->rx_nb_avail = 0;
 }
 
 /* turn on or off rx queue
@@ -309,16 +298,10 @@ ice_switch_rx_queue(struct ice_hw *hw, uint16_t q_idx, bool on)
 }
 
 static inline int
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
 ice_check_rx_burst_bulk_alloc_preconditions(struct ice_rx_queue *rxq)
-#else
-ice_check_rx_burst_bulk_alloc_preconditions
-       (__rte_unused struct ice_rx_queue *rxq)
-#endif
 {
        int ret = 0;
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
        if (!(rxq->rx_free_thresh >= ICE_RX_MAX_BURST)) {
                PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
                             "rxq->rx_free_thresh=%d, "
@@ -338,9 +321,6 @@ ice_check_rx_burst_bulk_alloc_preconditions
                             rxq->nb_rx_desc, rxq->rx_free_thresh);
                ret = -EINVAL;
        }
-#else
-       ret = -EINVAL;
-#endif
 
        return ret;
 }
@@ -357,17 +337,11 @@ ice_reset_rx_queue(struct ice_rx_queue *rxq)
                return;
        }
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
-       if (ice_check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
-               len = (uint16_t)(rxq->nb_rx_desc + ICE_RX_MAX_BURST);
-       else
-#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
-               len = rxq->nb_rx_desc;
+       len = (uint16_t)(rxq->nb_rx_desc + ICE_RX_MAX_BURST);
 
        for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
                ((volatile char *)rxq->rx_ring)[i] = 0;
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
        memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
        for (i = 0; i < ICE_RX_MAX_BURST; ++i)
                rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
@@ -375,7 +349,6 @@ ice_reset_rx_queue(struct ice_rx_queue *rxq)
        rxq->rx_nb_avail = 0;
        rxq->rx_next_avail = 0;
        rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
-#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
 
        rxq->rx_tail = 0;
        rxq->nb_rx_hold = 0;
@@ -424,12 +397,12 @@ ice_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        /* Init the RX tail register. */
        ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
 
-       err = ice_switch_rx_queue(hw, rxq->reg_idx, TRUE);
+       err = ice_switch_rx_queue(hw, rxq->reg_idx, true);
        if (err) {
                PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
                            rx_queue_id);
 
-               ice_rx_queue_release_mbufs(rxq);
+               rxq->rx_rel_mbufs(rxq);
                ice_reset_rx_queue(rxq);
                return -EINVAL;
        }
@@ -450,13 +423,13 @@ ice_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        if (rx_queue_id < dev->data->nb_rx_queues) {
                rxq = dev->data->rx_queues[rx_queue_id];
 
-               err = ice_switch_rx_queue(hw, rxq->reg_idx, FALSE);
+               err = ice_switch_rx_queue(hw, rxq->reg_idx, false);
                if (err) {
                        PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
                                    rx_queue_id);
                        return -EINVAL;
                }
-               ice_rx_queue_release_mbufs(rxq);
+               rxq->rx_rel_mbufs(rxq);
                ice_reset_rx_queue(rxq);
                dev->data->rx_queue_state[rx_queue_id] =
                        RTE_ETH_QUEUE_STATE_STOPPED;
@@ -508,7 +481,7 @@ ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
        tx_ctx.tso_qnum = txq->reg_idx; /* index for tso state structure */
        tx_ctx.legacy_int = 1; /* Legacy or Advanced Host Interface */
 
-       ice_set_ctx((uint8_t *)&tx_ctx, txq_elem.txqs[0].txq_ctx,
+       ice_set_ctx(hw, (uint8_t *)&tx_ctx, txq_elem.txqs[0].txq_ctx,
                    ice_tlan_ctx_info);
 
        txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
@@ -630,7 +603,7 @@ ice_fdir_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        /* Init the RX tail register. */
        ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
 
-       err = ice_switch_rx_queue(hw, rxq->reg_idx, TRUE);
+       err = ice_switch_rx_queue(hw, rxq->reg_idx, true);
        if (err) {
                PMD_DRV_LOG(ERR, "Failed to switch FDIR RX queue %u on",
                            rx_queue_id);
@@ -680,7 +653,7 @@ ice_fdir_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
        tx_ctx.tso_qnum = txq->reg_idx; /* index for tso state structure */
        tx_ctx.legacy_int = 1; /* Legacy or Advanced Host Interface */
 
-       ice_set_ctx((uint8_t *)&tx_ctx, txq_elem.txqs[0].txq_ctx,
+       ice_set_ctx(hw, (uint8_t *)&tx_ctx, txq_elem.txqs[0].txq_ctx,
                    ice_tlan_ctx_info);
 
        txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
@@ -719,11 +692,6 @@ _ice_tx_queue_release_mbufs(struct ice_tx_queue *txq)
                }
        }
 }
-static void
-ice_tx_queue_release_mbufs(struct ice_tx_queue *txq)
-{
-       txq->tx_rel_mbufs(txq);
-}
 
 static void
 ice_reset_tx_queue(struct ice_tx_queue *txq)
@@ -799,7 +767,7 @@ ice_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
                return -EINVAL;
        }
 
-       ice_tx_queue_release_mbufs(txq);
+       txq->tx_rel_mbufs(txq);
        ice_reset_tx_queue(txq);
        dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 
@@ -816,13 +784,13 @@ ice_fdir_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
        rxq = pf->fdir.rxq;
 
-       err = ice_switch_rx_queue(hw, rxq->reg_idx, FALSE);
+       err = ice_switch_rx_queue(hw, rxq->reg_idx, false);
        if (err) {
                PMD_DRV_LOG(ERR, "Failed to switch FDIR RX queue %u off",
                            rx_queue_id);
                return -EINVAL;
        }
-       ice_rx_queue_release_mbufs(rxq);
+       rxq->rx_rel_mbufs(rxq);
 
        return 0;
 }
@@ -858,7 +826,7 @@ ice_fdir_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
                return -EINVAL;
        }
 
-       ice_tx_queue_release_mbufs(txq);
+       txq->tx_rel_mbufs(txq);
 
        return 0;
 }
@@ -926,13 +894,11 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
        /* Allocate the maximun number of RX ring hardware descriptor. */
        len = ICE_MAX_RING_DESC;
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
        /**
         * Allocating a little more memory because vectorized/bulk_alloc Rx
         * functions doesn't check boundaries each time.
         */
        len += ICE_RX_MAX_BURST;
-#endif
 
        /* Allocate the maximum number of RX ring hardware descriptor. */
        ring_size = sizeof(union ice_rx_flex_desc) * len;
@@ -952,11 +918,8 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->rx_ring_dma = rz->iova;
        rxq->rx_ring = rz->addr;
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       /* always reserve more for bulk alloc */
        len = (uint16_t)(nb_desc + ICE_RX_MAX_BURST);
-#else
-       len = nb_desc;
-#endif
 
        /* Allocate the software ring. */
        rxq->sw_ring = rte_zmalloc_socket(NULL,
@@ -970,24 +933,21 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
        }
 
        ice_reset_rx_queue(rxq);
-       rxq->q_set = TRUE;
+       rxq->q_set = true;
        dev->data->rx_queues[queue_idx] = rxq;
        rxq->rx_rel_mbufs = _ice_rx_queue_release_mbufs;
 
        use_def_burst_func = ice_check_rx_burst_bulk_alloc_preconditions(rxq);
 
        if (!use_def_burst_func) {
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
                PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
                             "satisfied. Rx Burst Bulk Alloc function will be "
                             "used on port=%d, queue=%d.",
                             rxq->port_id, rxq->queue_id);
-#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
        } else {
                PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
-                            "not satisfied, Scattered Rx is requested, "
-                            "or RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC is "
-                            "not enabled on port=%d, queue=%d.",
+                            "not satisfied, Scattered Rx is requested. "
+                            "on port=%d, queue=%d.",
                             rxq->port_id, rxq->queue_id);
                ad->rx_bulk_alloc_allowed = false;
        }
@@ -1005,7 +965,7 @@ ice_rx_queue_release(void *rxq)
                return;
        }
 
-       ice_rx_queue_release_mbufs(q);
+       q->rx_rel_mbufs(q);
        rte_free(q->sw_ring);
        rte_free(q);
 }
@@ -1183,7 +1143,7 @@ ice_tx_queue_setup(struct rte_eth_dev *dev,
        }
 
        ice_reset_tx_queue(txq);
-       txq->q_set = TRUE;
+       txq->q_set = true;
        dev->data->tx_queues[queue_idx] = txq;
        txq->tx_rel_mbufs = _ice_tx_queue_release_mbufs;
        ice_set_tx_function_flag(dev, txq);
@@ -1201,7 +1161,7 @@ ice_tx_queue_release(void *txq)
                return;
        }
 
-       ice_tx_queue_release_mbufs(q);
+       q->tx_rel_mbufs(q);
        rte_free(q->sw_ring);
        rte_free(q);
 }
@@ -1399,7 +1359,6 @@ ice_rxd_to_pkt_fields(struct rte_mbuf *mb,
 #endif
 }
 
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
 #define ICE_LOOK_AHEAD 8
 #if (ICE_LOOK_AHEAD != 8)
 #error "PMD ICE: ICE_LOOK_AHEAD must be 8\n"
@@ -1620,15 +1579,6 @@ ice_recv_pkts_bulk_alloc(void *rx_queue,
 
        return nb_rx;
 }
-#else
-static uint16_t
-ice_recv_pkts_bulk_alloc(void __rte_unused *rx_queue,
-                        struct rte_mbuf __rte_unused **rx_pkts,
-                        uint16_t __rte_unused nb_pkts)
-{
-       return 0;
-}
-#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
 
 static uint16_t
 ice_recv_scattered_pkts(void *rx_queue,
@@ -1872,9 +1822,7 @@ ice_dev_supported_ptypes_get(struct rte_eth_dev *dev)
                ptypes = ptypes_os;
 
        if (dev->rx_pkt_burst == ice_recv_pkts ||
-#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
            dev->rx_pkt_burst == ice_recv_pkts_bulk_alloc ||
-#endif
            dev->rx_pkt_burst == ice_recv_scattered_pkts)
                return ptypes;
 
@@ -1945,24 +1893,6 @@ ice_tx_descriptor_status(void *tx_queue, uint16_t offset)
        return RTE_ETH_TX_DESC_FULL;
 }
 
-void
-ice_clear_queues(struct rte_eth_dev *dev)
-{
-       uint16_t i;
-
-       PMD_INIT_FUNC_TRACE();
-
-       for (i = 0; i < dev->data->nb_tx_queues; i++) {
-               ice_tx_queue_release_mbufs(dev->data->tx_queues[i]);
-               ice_reset_tx_queue(dev->data->tx_queues[i]);
-       }
-
-       for (i = 0; i < dev->data->nb_rx_queues; i++) {
-               ice_rx_queue_release_mbufs(dev->data->rx_queues[i]);
-               ice_reset_rx_queue(dev->data->rx_queues[i]);
-       }
-}
-
 void
 ice_free_queues(struct rte_eth_dev *dev)
 {
@@ -1975,6 +1905,7 @@ ice_free_queues(struct rte_eth_dev *dev)
                        continue;
                ice_rx_queue_release(dev->data->rx_queues[i]);
                dev->data->rx_queues[i] = NULL;
+               rte_eth_dma_zone_free(dev, "rx_ring", i);
        }
        dev->data->nb_rx_queues = 0;
 
@@ -1983,6 +1914,7 @@ ice_free_queues(struct rte_eth_dev *dev)
                        continue;
                ice_tx_queue_release(dev->data->tx_queues[i]);
                dev->data->tx_queues[i] = NULL;
+               rte_eth_dma_zone_free(dev, "tx_ring", i);
        }
        dev->data->nb_tx_queues = 0;
 }
@@ -2040,7 +1972,7 @@ ice_fdir_setup_tx_resources(struct ice_pf *pf)
         * don't need to allocate software ring and reset for the fdir
         * program queue just set the queue has been configured.
         */
-       txq->q_set = TRUE;
+       txq->q_set = true;
        pf->fdir.txq = txq;
 
        txq->tx_rel_mbufs = _ice_tx_queue_release_mbufs;
@@ -2101,7 +2033,7 @@ ice_fdir_setup_rx_resources(struct ice_pf *pf)
         * Don't need to allocate software ring and reset for the fdir
         * rx queue, just set the queue has been configured.
         */
-       rxq->q_set = TRUE;
+       rxq->q_set = true;
        pf->fdir.rxq = rxq;
 
        rxq->rx_rel_mbufs = _ice_rx_queue_release_mbufs;
@@ -2302,7 +2234,7 @@ ice_txd_enable_checksum(uint64_t ol_flags,
        switch (ol_flags & PKT_TX_L4_MASK) {
        case PKT_TX_TCP_CKSUM:
                *td_cmd |= ICE_TX_DESC_CMD_L4T_EOFT_TCP;
-               *td_offset |= (sizeof(struct rte_tcp_hdr) >> 2) <<
+               *td_offset |= (tx_offload.l4_len >> 2) <<
                              ICE_TX_DESC_LEN_L4_LEN_S;
                break;
        case PKT_TX_SCTP_CKSUM:
@@ -2421,6 +2353,24 @@ ice_set_tso_ctx(struct rte_mbuf *mbuf, union ice_tx_offload tx_offload)
        return ctx_desc;
 }
 
+/* HW requires that TX buffer size ranges from 1B up to (16K-1)B. */
+#define ICE_MAX_DATA_PER_TXD \
+       (ICE_TXD_QW1_TX_BUF_SZ_M >> ICE_TXD_QW1_TX_BUF_SZ_S)
+/* Calculate the number of TX descriptors needed for each pkt */
+static inline uint16_t
+ice_calc_pkt_desc(struct rte_mbuf *tx_pkt)
+{
+       struct rte_mbuf *txd = tx_pkt;
+       uint16_t count = 0;
+
+       while (txd != NULL) {
+               count += DIV_ROUND_UP(txd->data_len, ICE_MAX_DATA_PER_TXD);
+               txd = txd->next;
+       }
+
+       return count;
+}
+
 uint16_t
 ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
@@ -2440,6 +2390,7 @@ ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
        uint32_t td_offset = 0;
        uint32_t td_tag = 0;
        uint16_t tx_last;
+       uint16_t slen;
        uint64_t buf_dma_addr;
        uint64_t ol_flags;
        union ice_tx_offload tx_offload = {0};
@@ -2452,7 +2403,7 @@ ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 
        /* Check if the descriptor ring needs to be cleaned. */
        if (txq->nb_tx_free < txq->tx_free_thresh)
-               ice_xmit_cleanup(txq);
+               (void)ice_xmit_cleanup(txq);
 
        for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
                tx_pkt = *tx_pkts++;
@@ -2471,8 +2422,15 @@ ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                /* The number of descriptors that must be allocated for
                 * a packet equals to the number of the segments of that
                 * packet plus the number of context descriptor if needed.
+                * Recalculate the needed tx descs when TSO enabled in case
+                * the mbuf data size exceeds max data size that hw allows
+                * per tx desc.
                 */
-               nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
+               if (ol_flags & PKT_TX_TCP_SEG)
+                       nb_used = (uint16_t)(ice_calc_pkt_desc(tx_pkt) +
+                                            nb_ctx);
+               else
+                       nb_used = (uint16_t)(tx_pkt->nb_segs + nb_ctx);
                tx_last = (uint16_t)(tx_id + nb_used - 1);
 
                /* Circular ring */
@@ -2562,15 +2520,37 @@ ice_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                        txe->mbuf = m_seg;
 
                        /* Setup TX Descriptor */
+                       slen = m_seg->data_len;
                        buf_dma_addr = rte_mbuf_data_iova(m_seg);
+
+                       while ((ol_flags & PKT_TX_TCP_SEG) &&
+                               unlikely(slen > ICE_MAX_DATA_PER_TXD)) {
+                               txd->buf_addr = rte_cpu_to_le_64(buf_dma_addr);
+                               txd->cmd_type_offset_bsz =
+                               rte_cpu_to_le_64(ICE_TX_DESC_DTYPE_DATA |
+                               ((uint64_t)td_cmd << ICE_TXD_QW1_CMD_S) |
+                               ((uint64_t)td_offset << ICE_TXD_QW1_OFFSET_S) |
+                               ((uint64_t)ICE_MAX_DATA_PER_TXD <<
+                                ICE_TXD_QW1_TX_BUF_SZ_S) |
+                               ((uint64_t)td_tag << ICE_TXD_QW1_L2TAG1_S));
+
+                               buf_dma_addr += ICE_MAX_DATA_PER_TXD;
+                               slen -= ICE_MAX_DATA_PER_TXD;
+
+                               txe->last_id = tx_last;
+                               tx_id = txe->next_id;
+                               txe = txn;
+                               txd = &tx_ring[tx_id];
+                               txn = &sw_ring[txe->next_id];
+                       }
+
                        txd->buf_addr = rte_cpu_to_le_64(buf_dma_addr);
                        txd->cmd_type_offset_bsz =
                                rte_cpu_to_le_64(ICE_TX_DESC_DTYPE_DATA |
-                               ((uint64_t)td_cmd  << ICE_TXD_QW1_CMD_S) |
+                               ((uint64_t)td_cmd << ICE_TXD_QW1_CMD_S) |
                                ((uint64_t)td_offset << ICE_TXD_QW1_OFFSET_S) |
-                               ((uint64_t)m_seg->data_len  <<
-                                ICE_TXD_QW1_TX_BUF_SZ_S) |
-                               ((uint64_t)td_tag  << ICE_TXD_QW1_L2TAG1_S));
+                               ((uint64_t)slen << ICE_TXD_QW1_TX_BUF_SZ_S) |
+                               ((uint64_t)td_tag << ICE_TXD_QW1_L2TAG1_S));
 
                        txe->last_id = tx_last;
                        tx_id = txe->next_id;
@@ -2607,7 +2587,7 @@ end_of_tx:
        return nb_tx;
 }
 
-static inline int __attribute__((always_inline))
+static __rte_always_inline int
 ice_tx_free_bufs(struct ice_tx_queue *txq)
 {
        struct ice_tx_entry *txep;
@@ -2896,7 +2876,7 @@ ice_xmit_pkts_simple(void *tx_queue,
        return nb_tx;
 }
 
-void __attribute__((cold))
+void __rte_cold
 ice_set_rx_function(struct rte_eth_dev *dev)
 {
        PMD_INIT_FUNC_TRACE();
@@ -3006,7 +2986,7 @@ ice_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
        return ret;
 }
 
-void __attribute__((cold))
+void __rte_cold
 ice_set_tx_function_flag(struct rte_eth_dev *dev, struct ice_tx_queue *txq)
 {
        struct ice_adapter *ad =
@@ -3075,7 +3055,7 @@ ice_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
        return i;
 }
 
-void __attribute__((cold))
+void __rte_cold
 ice_set_tx_function(struct rte_eth_dev *dev)
 {
        struct ice_adapter *ad =
@@ -3704,7 +3684,7 @@ ice_get_default_pkt_type(uint16_t ptype)
        return type_table[ptype];
 }
 
-void __attribute__((cold))
+void __rte_cold
 ice_set_default_ptype_table(struct rte_eth_dev *dev)
 {
        struct ice_adapter *ad =