mbuf: add namespace to offload flags
[dpdk.git] / drivers / net / thunderx / nicvf_rxtx.c
index 6cae834..defa551 100644 (file)
@@ -1,33 +1,5 @@
-/*
- *   BSD LICENSE
- *
- *   Copyright (C) Cavium networks Ltd. 2016.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Cavium networks nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2016 Cavium, Inc
  */
 
 #include <unistd.h>
@@ -41,7 +13,7 @@
 #include <rte_common.h>
 #include <rte_cycles.h>
 #include <rte_errno.h>
-#include <rte_ethdev.h>
+#include <ethdev_driver.h>
 #include <rte_ether.h>
 #include <rte_log.h>
 #include <rte_mbuf.h>
@@ -53,7 +25,7 @@
 #include "nicvf_rxtx.h"
 #include "nicvf_logs.h"
 
-static inline void __hot
+static inline void __rte_hot
 fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
 {
        /* Local variable sqe to avoid read from sq desc memory*/
@@ -70,10 +42,10 @@ fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
        ol_flags = pkt->ol_flags & NICVF_TX_OFFLOAD_MASK;
        if (unlikely(ol_flags)) {
                /* L4 cksum */
-               uint64_t l4_flags = ol_flags & PKT_TX_L4_MASK;
-               if (l4_flags == PKT_TX_TCP_CKSUM)
+               uint64_t l4_flags = ol_flags & RTE_MBUF_F_TX_L4_MASK;
+               if (l4_flags == RTE_MBUF_F_TX_TCP_CKSUM)
                        sqe.hdr.csum_l4 = SEND_L4_CSUM_TCP;
-               else if (l4_flags == PKT_TX_UDP_CKSUM)
+               else if (l4_flags == RTE_MBUF_F_TX_UDP_CKSUM)
                        sqe.hdr.csum_l4 = SEND_L4_CSUM_UDP;
                else
                        sqe.hdr.csum_l4 = SEND_L4_CSUM_DISABLE;
@@ -82,14 +54,22 @@ fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
                sqe.hdr.l4_offset = pkt->l3_len + pkt->l2_len;
 
                /* L3 cksum */
-               if (ol_flags & PKT_TX_IP_CKSUM)
+               if (ol_flags & RTE_MBUF_F_TX_IP_CKSUM)
                        sqe.hdr.csum_l3 = 1;
        }
 
        entry->buff[0] = sqe.buff[0];
 }
 
-void __hot
+static inline void __rte_hot
+fill_sq_desc_header_zero_w1(union sq_entry_t *entry,
+                               struct rte_mbuf *pkt)
+{
+       fill_sq_desc_header(entry, pkt);
+       entry->buff[1] = 0ULL;
+}
+
+void __rte_hot
 nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq)
 {
        int j = 0;
@@ -112,7 +92,7 @@ nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq)
        NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
 }
 
-void __hot
+void __rte_hot
 nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq)
 {
        uint32_t n = 0;
@@ -135,7 +115,7 @@ nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq)
        NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
 }
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_free_tx_desc(struct nicvf_txq *sq)
 {
        return ((sq->head - sq->tail - 1) & sq->qlen_mask);
@@ -144,7 +124,7 @@ nicvf_free_tx_desc(struct nicvf_txq *sq)
 /* Send Header + Packet */
 #define TX_DESC_PER_PKT 2
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_free_xmitted_buffers(struct nicvf_txq *sq, struct rte_mbuf **tx_pkts,
                            uint16_t nb_pkts)
 {
@@ -162,7 +142,7 @@ nicvf_free_xmitted_buffers(struct nicvf_txq *sq, struct rte_mbuf **tx_pkts,
        return free_desc;
 }
 
-uint16_t __hot
+uint16_t __rte_hot
 nicvf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
        int i;
@@ -190,16 +170,18 @@ nicvf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                free_desc -= TX_DESC_PER_PKT;
        }
 
-       sq->tail = tail;
-       sq->xmit_bufs += i;
-       rte_wmb();
+       if (likely(i)) {
+               sq->tail = tail;
+               sq->xmit_bufs += i;
+               rte_wmb();
 
-       /* Inform HW to xmit the packets */
-       nicvf_addr_write(sq->sq_door, i * TX_DESC_PER_PKT);
+               /* Inform HW to xmit the packets */
+               nicvf_addr_write(sq->sq_door, i * TX_DESC_PER_PKT);
+       }
        return i;
 }
 
-uint16_t __hot
+uint16_t __rte_hot
 nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
                         uint16_t nb_pkts)
 {
@@ -230,7 +212,7 @@ nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
                used_bufs += nb_segs;
 
                txbuffs[tail] = NULL;
-               fill_sq_desc_header(desc_ptr + tail, pkt);
+               fill_sq_desc_header_zero_w1(desc_ptr + tail, pkt);
                tail = (tail + 1) & qlen_mask;
 
                txbuffs[tail] = pkt;
@@ -246,13 +228,15 @@ nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
                }
        }
 
-       sq->tail = tail;
-       sq->xmit_bufs += used_bufs;
-       rte_wmb();
+       if (likely(used_desc)) {
+               sq->tail = tail;
+               sq->xmit_bufs += used_bufs;
+               rte_wmb();
 
-       /* Inform HW to xmit the packets */
-       nicvf_addr_write(sq->sq_door, used_desc);
-       return nb_pkts;
+               /* Inform HW to xmit the packets */
+               nicvf_addr_write(sq->sq_door, used_desc);
+       }
+       return i;
 }
 
 static const uint32_t ptype_table[16][16] __rte_cache_aligned = {
@@ -349,13 +333,27 @@ static const uint32_t ptype_table[16][16] __rte_cache_aligned = {
        [L3_OTHER][L4_NVGRE] = RTE_PTYPE_TUNNEL_NVGRE,
 };
 
-static inline uint32_t __hot
+static inline uint32_t __rte_hot
 nicvf_rx_classify_pkt(cqe_rx_word0_t cqe_rx_w0)
 {
        return ptype_table[cqe_rx_w0.l3_type][cqe_rx_w0.l4_type];
 }
 
-static inline int __hot
+static inline uint64_t __rte_hot
+nicvf_set_olflags(const cqe_rx_word0_t cqe_rx_w0)
+{
+       static const uint64_t flag_table[3] __rte_cache_aligned = {
+               RTE_MBUF_F_RX_IP_CKSUM_GOOD | RTE_MBUF_F_RX_L4_CKSUM_GOOD,
+               RTE_MBUF_F_RX_IP_CKSUM_BAD | RTE_MBUF_F_RX_L4_CKSUM_UNKNOWN,
+               RTE_MBUF_F_RX_IP_CKSUM_GOOD | RTE_MBUF_F_RX_L4_CKSUM_BAD,
+       };
+
+       const uint8_t idx = (cqe_rx_w0.err_opcode == CQE_RX_ERR_L4_CHK) << 1 |
+               (cqe_rx_w0.err_opcode == CQE_RX_ERR_IP_CHK);
+       return flag_table[idx];
+}
+
+static inline int __rte_hot
 nicvf_fill_rbdr(struct nicvf_rxq *rxq, int to_fill)
 {
        int i;
@@ -387,15 +385,14 @@ nicvf_fill_rbdr(struct nicvf_rxq *rxq, int to_fill)
                ltail++;
        }
 
-       while (__atomic_load_n(&rbdr->tail, __ATOMIC_RELAXED) != next_tail)
-               rte_pause();
+       rte_wait_until_equal_32(&rbdr->tail, next_tail, __ATOMIC_RELAXED);
 
        __atomic_store_n(&rbdr->tail, ltail, __ATOMIC_RELEASE);
        nicvf_addr_write(door, to_fill);
        return to_fill;
 }
 
-static inline int32_t __hot
+static inline int32_t __rte_hot
 nicvf_rx_pkts_to_process(struct nicvf_rxq *rxq, uint16_t nb_pkts,
                         int32_t available_space)
 {
@@ -406,18 +403,20 @@ nicvf_rx_pkts_to_process(struct nicvf_rxq *rxq, uint16_t nb_pkts,
        return RTE_MIN(nb_pkts, available_space);
 }
 
-static inline void __hot
+static inline void __rte_hot
 nicvf_rx_offload(cqe_rx_word0_t cqe_rx_w0, cqe_rx_word2_t cqe_rx_w2,
                 struct rte_mbuf *pkt)
 {
        if (likely(cqe_rx_w0.rss_alg)) {
                pkt->hash.rss = cqe_rx_w2.rss_tag;
-               pkt->ol_flags |= PKT_RX_RSS_HASH;
+               pkt->ol_flags |= RTE_MBUF_F_RX_RSS_HASH;
+
        }
 }
 
-uint16_t __hot
-nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+static __rte_always_inline uint16_t
+nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
+               const uint32_t flag)
 {
        uint32_t i, to_process;
        struct cqe_rx_t *cqe_rx;
@@ -448,7 +447,19 @@ nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                rb0_ptr = *((uint64_t *)cqe_rx + rbptr_offset);
                pkt = (struct rte_mbuf *)nicvf_mbuff_phy2virt
                                (rb0_ptr - cqe_rx_w1.align_pad, mbuf_phys_off);
-               pkt->ol_flags = 0;
+
+               if (flag & NICVF_RX_OFFLOAD_NONE)
+                       pkt->ol_flags = 0;
+               if (flag & NICVF_RX_OFFLOAD_CKSUM)
+                       pkt->ol_flags = nicvf_set_olflags(cqe_rx_w0);
+               if (flag & NICVF_RX_OFFLOAD_VLAN_STRIP) {
+                       if (unlikely(cqe_rx_w0.vlan_stripped)) {
+                               pkt->ol_flags |= RTE_MBUF_F_RX_VLAN
+                                                       | RTE_MBUF_F_RX_VLAN_STRIPPED;
+                               pkt->vlan_tci =
+                                       rte_cpu_to_be_16(cqe_rx_w2.vlan_tci);
+                       }
+               }
                pkt->data_len = cqe_rx_w3.rb0_sz;
                pkt->pkt_len = cqe_rx_w3.rb0_sz;
                pkt->packet_type = nicvf_rx_classify_pkt(cqe_rx_w0);
@@ -473,11 +484,43 @@ nicvf_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        return to_process;
 }
 
-static inline uint16_t __hot
+uint16_t __rte_hot
+nicvf_recv_pkts_no_offload(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_NONE);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_cksum(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_CKSUM);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_NONE | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_cksum_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_CKSUM | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+static __rte_always_inline uint16_t __rte_hot
 nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
                        uint64_t mbuf_phys_off,
                        struct rte_mbuf **rx_pkt, uint8_t rbptr_offset,
-                       uint64_t mbuf_init)
+                       uint64_t mbuf_init, const uint32_t flag)
 {
        struct rte_mbuf *pkt, *seg, *prev;
        cqe_rx_word0_t cqe_rx_w0;
@@ -495,12 +538,22 @@ nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
        pkt = (struct rte_mbuf *)nicvf_mbuff_phy2virt
                        (rb_ptr[0] - cqe_rx_w1.align_pad, mbuf_phys_off);
 
-       pkt->ol_flags = 0;
        pkt->pkt_len = cqe_rx_w1.pkt_len;
        pkt->data_len = rb_sz[nicvf_frag_num(0)];
        nicvf_mbuff_init_mseg_update(
                                pkt, mbuf_init, cqe_rx_w1.align_pad, nb_segs);
        pkt->packet_type = nicvf_rx_classify_pkt(cqe_rx_w0);
+       if (flag & NICVF_RX_OFFLOAD_NONE)
+               pkt->ol_flags = 0;
+       if (flag & NICVF_RX_OFFLOAD_CKSUM)
+               pkt->ol_flags = nicvf_set_olflags(cqe_rx_w0);
+       if (flag & NICVF_RX_OFFLOAD_VLAN_STRIP) {
+               if (unlikely(cqe_rx_w0.vlan_stripped)) {
+                       pkt->ol_flags |= RTE_MBUF_F_RX_VLAN
+                               | RTE_MBUF_F_RX_VLAN_STRIPPED;
+                       pkt->vlan_tci = rte_cpu_to_be_16(cqe_rx_w2.vlan_tci);
+               }
+       }
        nicvf_rx_offload(cqe_rx_w0, cqe_rx_w2, pkt);
 
        *rx_pkt = pkt;
@@ -519,9 +572,9 @@ nicvf_process_cq_mseg_entry(struct cqe_rx_t *cqe_rx,
        return nb_segs;
 }
 
-uint16_t __hot
+static __rte_always_inline uint16_t __rte_hot
 nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
-                        uint16_t nb_pkts)
+                        uint16_t nb_pkts, const uint32_t flag)
 {
        union cq_entry_t *cq_entry;
        struct cqe_rx_t *cqe_rx;
@@ -543,7 +596,7 @@ nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
                cq_entry = &desc[cqe_head];
                cqe_rx = (struct cqe_rx_t *)cq_entry;
                nb_segs = nicvf_process_cq_mseg_entry(cqe_rx, mbuf_phys_off,
-                       rx_pkts + i, rbptr_offset, mbuf_init);
+                       rx_pkts + i, rbptr_offset, mbuf_init, flag);
                buffers_consumed += nb_segs;
                cqe_head = (cqe_head + 1) & cqe_mask;
                nicvf_prefetch_store_keep(rx_pkts[i]);
@@ -563,12 +616,44 @@ nicvf_recv_pkts_multiseg(void *rx_queue, struct rte_mbuf **rx_pkts,
        return to_process;
 }
 
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_no_offload(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_NONE);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_cksum(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_CKSUM);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_vlan_strip(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_NONE | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
+uint16_t __rte_hot
+nicvf_recv_pkts_multiseg_cksum_vlan_strip(void *rx_queue,
+               struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+       return nicvf_recv_pkts_multiseg(rx_queue, rx_pkts, nb_pkts,
+                       NICVF_RX_OFFLOAD_CKSUM | NICVF_RX_OFFLOAD_VLAN_STRIP);
+}
+
 uint32_t
-nicvf_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t queue_idx)
+nicvf_dev_rx_queue_count(void *rx_queue)
 {
        struct nicvf_rxq *rxq;
 
-       rxq = dev->data->rx_queues[queue_idx];
+       rxq = rx_queue;
        return nicvf_addr_read(rxq->cq_status) & NICVF_CQ_CQE_COUNT_MASK;
 }