net/avp: add packet receive functions
authorAllain Legacy <allain.legacy@windriver.com>
Tue, 28 Mar 2017 11:54:03 +0000 (07:54 -0400)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 4 Apr 2017 17:02:18 +0000 (19:02 +0200)
Adds function required for receiving packets from the host application via
AVP device queues.  Both the simple and scattered functions are supported.

Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Signed-off-by: Matt Peters <matt.peters@windriver.com>
Acked-by: Vincent Jardin <vincent.jardin@6wind.com>
doc/guides/nics/features/avp.ini
drivers/net/avp/avp_ethdev.c

index 45a2185..e748ea8 100644 (file)
@@ -5,6 +5,9 @@
 ;
 [Features]
 Link status          = Y
+Jumbo frame          = Y
+Scattered Rx         = Y
+Unicast MAC filter   = Y
 VLAN offload         = Y
 Linux UIO            = Y
 x86-64               = Y
index 8945a45..b3daa15 100644 (file)
@@ -79,11 +79,19 @@ static int avp_dev_tx_queue_setup(struct rte_eth_dev *dev,
                                  unsigned int socket_id,
                                  const struct rte_eth_txconf *tx_conf);
 
+static uint16_t avp_recv_scattered_pkts(void *rx_queue,
+                                       struct rte_mbuf **rx_pkts,
+                                       uint16_t nb_pkts);
+
+static uint16_t avp_recv_pkts(void *rx_queue,
+                             struct rte_mbuf **rx_pkts,
+                             uint16_t nb_pkts);
 static void avp_dev_rx_queue_release(void *rxq);
 static void avp_dev_tx_queue_release(void *txq);
 #define AVP_DEV_TO_PCI(eth_dev) RTE_DEV_TO_PCI((eth_dev)->device)
 
 
+#define AVP_MAX_RX_BURST 64
 #define AVP_MAX_MAC_ADDRS 1
 #define AVP_MIN_RX_BUFSIZE ETHER_MIN_LEN
 
@@ -301,6 +309,15 @@ avp_dev_ctrl_set_config(struct rte_eth_dev *eth_dev,
        return ret == 0 ? request.result : ret;
 }
 
+/* translate from host mbuf virtual address to guest virtual address */
+static inline void *
+avp_dev_translate_buffer(struct avp_dev *avp, void *host_mbuf_address)
+{
+       return RTE_PTR_ADD(RTE_PTR_SUB(host_mbuf_address,
+                                      (uintptr_t)avp->host_mbuf_addr),
+                          (uintptr_t)avp->mbuf_addr);
+}
+
 /* translate from host physical address to guest virtual address */
 static void *
 avp_dev_translate_address(struct rte_eth_dev *eth_dev,
@@ -627,6 +644,7 @@ eth_avp_dev_init(struct rte_eth_dev *eth_dev)
 
        pci_dev = AVP_DEV_TO_PCI(eth_dev);
        eth_dev->dev_ops = &avp_eth_dev_ops;
+       eth_dev->rx_pkt_burst = &avp_recv_pkts;
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
                /*
@@ -635,6 +653,10 @@ eth_avp_dev_init(struct rte_eth_dev *eth_dev)
                 * be mapped to the same virtual address so all pointers should
                 * be valid.
                 */
+               if (eth_dev->data->scattered_rx) {
+                       PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
+                       eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
+               }
                return 0;
        }
 
@@ -701,6 +723,38 @@ static struct eth_driver rte_avp_pmd = {
        .dev_private_size = sizeof(struct avp_adapter),
 };
 
+static int
+avp_dev_enable_scattered(struct rte_eth_dev *eth_dev,
+                        struct avp_dev *avp)
+{
+       unsigned int max_rx_pkt_len;
+
+       max_rx_pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
+
+       if ((max_rx_pkt_len > avp->guest_mbuf_size) ||
+           (max_rx_pkt_len > avp->host_mbuf_size)) {
+               /*
+                * If the guest MTU is greater than either the host or guest
+                * buffers then chained mbufs have to be enabled in the TX
+                * direction.  It is assumed that the application will not need
+                * to send packets larger than their max_rx_pkt_len (MRU).
+                */
+               return 1;
+       }
+
+       if ((avp->max_rx_pkt_len > avp->guest_mbuf_size) ||
+           (avp->max_rx_pkt_len > avp->host_mbuf_size)) {
+               /*
+                * If the host MRU is greater than its own mbuf size or the
+                * guest mbuf size then chained mbufs have to be enabled in the
+                * RX direction.
+                */
+               return 1;
+       }
+
+       return 0;
+}
+
 static int
 avp_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
                       uint16_t rx_queue_id,
@@ -727,6 +781,14 @@ avp_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
        avp->guest_mbuf_size = (uint16_t)(mbp_priv->mbuf_data_room_size);
        avp->guest_mbuf_size -= RTE_PKTMBUF_HEADROOM;
 
+       if (avp_dev_enable_scattered(eth_dev, avp)) {
+               if (!eth_dev->data->scattered_rx) {
+                       PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
+                       eth_dev->data->scattered_rx = 1;
+                       eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
+               }
+       }
+
        PMD_DRV_LOG(DEBUG, "AVP max_rx_pkt_len=(%u,%u) mbuf_size=(%u,%u)\n",
                    avp->max_rx_pkt_len,
                    eth_dev->data->dev_conf.rxmode.max_rx_pkt_len,
@@ -797,6 +859,395 @@ avp_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
        return 0;
 }
 
+static inline int
+_avp_cmp_ether_addr(struct ether_addr *a, struct ether_addr *b)
+{
+       uint16_t *_a = (uint16_t *)&a->addr_bytes[0];
+       uint16_t *_b = (uint16_t *)&b->addr_bytes[0];
+       return (_a[0] ^ _b[0]) | (_a[1] ^ _b[1]) | (_a[2] ^ _b[2]);
+}
+
+static inline int
+_avp_mac_filter(struct avp_dev *avp, struct rte_mbuf *m)
+{
+       struct ether_hdr *eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
+
+       if (likely(_avp_cmp_ether_addr(&avp->ethaddr, &eth->d_addr) == 0)) {
+               /* allow all packets destined to our address */
+               return 0;
+       }
+
+       if (likely(is_broadcast_ether_addr(&eth->d_addr))) {
+               /* allow all broadcast packets */
+               return 0;
+       }
+
+       if (likely(is_multicast_ether_addr(&eth->d_addr))) {
+               /* allow all multicast packets */
+               return 0;
+       }
+
+       if (avp->flags & AVP_F_PROMISC) {
+               /* allow all packets when in promiscuous mode */
+               return 0;
+       }
+
+       return -1;
+}
+
+#ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
+static inline void
+__avp_dev_buffer_sanity_check(struct avp_dev *avp, struct rte_avp_desc *buf)
+{
+       struct rte_avp_desc *first_buf;
+       struct rte_avp_desc *pkt_buf;
+       unsigned int pkt_len;
+       unsigned int nb_segs;
+       void *pkt_data;
+       unsigned int i;
+
+       first_buf = avp_dev_translate_buffer(avp, buf);
+
+       i = 0;
+       pkt_len = 0;
+       nb_segs = first_buf->nb_segs;
+       do {
+               /* Adjust pointers for guest addressing */
+               pkt_buf = avp_dev_translate_buffer(avp, buf);
+               if (pkt_buf == NULL)
+                       rte_panic("bad buffer: segment %u has an invalid address %p\n",
+                                 i, buf);
+               pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
+               if (pkt_data == NULL)
+                       rte_panic("bad buffer: segment %u has a NULL data pointer\n",
+                                 i);
+               if (pkt_buf->data_len == 0)
+                       rte_panic("bad buffer: segment %u has 0 data length\n",
+                                 i);
+               pkt_len += pkt_buf->data_len;
+               nb_segs--;
+               i++;
+
+       } while (nb_segs && (buf = pkt_buf->next) != NULL);
+
+       if (nb_segs != 0)
+               rte_panic("bad buffer: expected %u segments found %u\n",
+                         first_buf->nb_segs, (first_buf->nb_segs - nb_segs));
+       if (pkt_len != first_buf->pkt_len)
+               rte_panic("bad buffer: expected length %u found %u\n",
+                         first_buf->pkt_len, pkt_len);
+}
+
+#define avp_dev_buffer_sanity_check(a, b) \
+       __avp_dev_buffer_sanity_check((a), (b))
+
+#else /* RTE_LIBRTE_AVP_DEBUG_BUFFERS */
+
+#define avp_dev_buffer_sanity_check(a, b) do {} while (0)
+
+#endif
+
+/*
+ * Copy a host buffer chain to a set of mbufs. This function assumes that
+ * there exactly the required number of mbufs to copy all source bytes.
+ */
+static inline struct rte_mbuf *
+avp_dev_copy_from_buffers(struct avp_dev *avp,
+                         struct rte_avp_desc *buf,
+                         struct rte_mbuf **mbufs,
+                         unsigned int count)
+{
+       struct rte_mbuf *m_previous = NULL;
+       struct rte_avp_desc *pkt_buf;
+       unsigned int total_length = 0;
+       unsigned int copy_length;
+       unsigned int src_offset;
+       struct rte_mbuf *m;
+       uint16_t ol_flags;
+       uint16_t vlan_tci;
+       void *pkt_data;
+       unsigned int i;
+
+       avp_dev_buffer_sanity_check(avp, buf);
+
+       /* setup the first source buffer */
+       pkt_buf = avp_dev_translate_buffer(avp, buf);
+       pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
+       total_length = pkt_buf->pkt_len;
+       src_offset = 0;
+
+       if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
+               ol_flags = PKT_RX_VLAN_PKT;
+               vlan_tci = pkt_buf->vlan_tci;
+       } else {
+               ol_flags = 0;
+               vlan_tci = 0;
+       }
+
+       for (i = 0; (i < count) && (buf != NULL); i++) {
+               /* fill each destination buffer */
+               m = mbufs[i];
+
+               if (m_previous != NULL)
+                       m_previous->next = m;
+
+               m_previous = m;
+
+               do {
+                       /*
+                        * Copy as many source buffers as will fit in the
+                        * destination buffer.
+                        */
+                       copy_length = RTE_MIN((avp->guest_mbuf_size -
+                                              rte_pktmbuf_data_len(m)),
+                                             (pkt_buf->data_len -
+                                              src_offset));
+                       rte_memcpy(RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
+                                              rte_pktmbuf_data_len(m)),
+                                  RTE_PTR_ADD(pkt_data, src_offset),
+                                  copy_length);
+                       rte_pktmbuf_data_len(m) += copy_length;
+                       src_offset += copy_length;
+
+                       if (likely(src_offset == pkt_buf->data_len)) {
+                               /* need a new source buffer */
+                               buf = pkt_buf->next;
+                               if (buf != NULL) {
+                                       pkt_buf = avp_dev_translate_buffer(
+                                               avp, buf);
+                                       pkt_data = avp_dev_translate_buffer(
+                                               avp, pkt_buf->data);
+                                       src_offset = 0;
+                               }
+                       }
+
+                       if (unlikely(rte_pktmbuf_data_len(m) ==
+                                    avp->guest_mbuf_size)) {
+                               /* need a new destination mbuf */
+                               break;
+                       }
+
+               } while (buf != NULL);
+       }
+
+       m = mbufs[0];
+       m->ol_flags = ol_flags;
+       m->nb_segs = count;
+       rte_pktmbuf_pkt_len(m) = total_length;
+       m->vlan_tci = vlan_tci;
+
+       __rte_mbuf_sanity_check(m, 1);
+
+       return m;
+}
+
+static uint16_t
+avp_recv_scattered_pkts(void *rx_queue,
+                       struct rte_mbuf **rx_pkts,
+                       uint16_t nb_pkts)
+{
+       struct avp_queue *rxq = (struct avp_queue *)rx_queue;
+       struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
+       struct rte_mbuf *mbufs[RTE_AVP_MAX_MBUF_SEGMENTS];
+       struct avp_dev *avp = rxq->avp;
+       struct rte_avp_desc *pkt_buf;
+       struct rte_avp_fifo *free_q;
+       struct rte_avp_fifo *rx_q;
+       struct rte_avp_desc *buf;
+       unsigned int count, avail, n;
+       unsigned int guest_mbuf_size;
+       struct rte_mbuf *m;
+       unsigned int required;
+       unsigned int buf_len;
+       unsigned int port_id;
+       unsigned int i;
+
+       guest_mbuf_size = avp->guest_mbuf_size;
+       port_id = avp->port_id;
+       rx_q = avp->rx_q[rxq->queue_id];
+       free_q = avp->free_q[rxq->queue_id];
+
+       /* setup next queue to service */
+       rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
+               (rxq->queue_id + 1) : rxq->queue_base;
+
+       /* determine how many slots are available in the free queue */
+       count = avp_fifo_free_count(free_q);
+
+       /* determine how many packets are available in the rx queue */
+       avail = avp_fifo_count(rx_q);
+
+       /* determine how many packets can be received */
+       count = RTE_MIN(count, avail);
+       count = RTE_MIN(count, nb_pkts);
+       count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
+
+       if (unlikely(count == 0)) {
+               /* no free buffers, or no buffers on the rx queue */
+               return 0;
+       }
+
+       /* retrieve pending packets */
+       n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
+       PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
+                  count, rx_q);
+
+       count = 0;
+       for (i = 0; i < n; i++) {
+               /* prefetch next entry while processing current one */
+               if (i + 1 < n) {
+                       pkt_buf = avp_dev_translate_buffer(avp,
+                                                          avp_bufs[i + 1]);
+                       rte_prefetch0(pkt_buf);
+               }
+               buf = avp_bufs[i];
+
+               /* Peek into the first buffer to determine the total length */
+               pkt_buf = avp_dev_translate_buffer(avp, buf);
+               buf_len = pkt_buf->pkt_len;
+
+               /* Allocate enough mbufs to receive the entire packet */
+               required = (buf_len + guest_mbuf_size - 1) / guest_mbuf_size;
+               if (rte_pktmbuf_alloc_bulk(avp->pool, mbufs, required)) {
+                       rxq->dev_data->rx_mbuf_alloc_failed++;
+                       continue;
+               }
+
+               /* Copy the data from the buffers to our mbufs */
+               m = avp_dev_copy_from_buffers(avp, buf, mbufs, required);
+
+               /* finalize mbuf */
+               m->port = port_id;
+
+               if (_avp_mac_filter(avp, m) != 0) {
+                       /* silently discard packets not destined to our MAC */
+                       rte_pktmbuf_free(m);
+                       continue;
+               }
+
+               /* return new mbuf to caller */
+               rx_pkts[count++] = m;
+               rxq->bytes += buf_len;
+       }
+
+       rxq->packets += count;
+
+       /* return the buffers to the free queue */
+       avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
+
+       return count;
+}
+
+
+static uint16_t
+avp_recv_pkts(void *rx_queue,
+             struct rte_mbuf **rx_pkts,
+             uint16_t nb_pkts)
+{
+       struct avp_queue *rxq = (struct avp_queue *)rx_queue;
+       struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
+       struct avp_dev *avp = rxq->avp;
+       struct rte_avp_desc *pkt_buf;
+       struct rte_avp_fifo *free_q;
+       struct rte_avp_fifo *rx_q;
+       unsigned int count, avail, n;
+       unsigned int pkt_len;
+       struct rte_mbuf *m;
+       char *pkt_data;
+       unsigned int i;
+
+       rx_q = avp->rx_q[rxq->queue_id];
+       free_q = avp->free_q[rxq->queue_id];
+
+       /* setup next queue to service */
+       rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
+               (rxq->queue_id + 1) : rxq->queue_base;
+
+       /* determine how many slots are available in the free queue */
+       count = avp_fifo_free_count(free_q);
+
+       /* determine how many packets are available in the rx queue */
+       avail = avp_fifo_count(rx_q);
+
+       /* determine how many packets can be received */
+       count = RTE_MIN(count, avail);
+       count = RTE_MIN(count, nb_pkts);
+       count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
+
+       if (unlikely(count == 0)) {
+               /* no free buffers, or no buffers on the rx queue */
+               return 0;
+       }
+
+       /* retrieve pending packets */
+       n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
+       PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
+                  count, rx_q);
+
+       count = 0;
+       for (i = 0; i < n; i++) {
+               /* prefetch next entry while processing current one */
+               if (i < n - 1) {
+                       pkt_buf = avp_dev_translate_buffer(avp,
+                                                          avp_bufs[i + 1]);
+                       rte_prefetch0(pkt_buf);
+               }
+
+               /* Adjust host pointers for guest addressing */
+               pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
+               pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
+               pkt_len = pkt_buf->pkt_len;
+
+               if (unlikely((pkt_len > avp->guest_mbuf_size) ||
+                            (pkt_buf->nb_segs > 1))) {
+                       /*
+                        * application should be using the scattered receive
+                        * function
+                        */
+                       rxq->errors++;
+                       continue;
+               }
+
+               /* process each packet to be transmitted */
+               m = rte_pktmbuf_alloc(avp->pool);
+               if (unlikely(m == NULL)) {
+                       rxq->dev_data->rx_mbuf_alloc_failed++;
+                       continue;
+               }
+
+               /* copy data out of the host buffer to our buffer */
+               m->data_off = RTE_PKTMBUF_HEADROOM;
+               rte_memcpy(rte_pktmbuf_mtod(m, void *), pkt_data, pkt_len);
+
+               /* initialize the local mbuf */
+               rte_pktmbuf_data_len(m) = pkt_len;
+               rte_pktmbuf_pkt_len(m) = pkt_len;
+               m->port = avp->port_id;
+
+               if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
+                       m->ol_flags = PKT_RX_VLAN_PKT;
+                       m->vlan_tci = pkt_buf->vlan_tci;
+               }
+
+               if (_avp_mac_filter(avp, m) != 0) {
+                       /* silently discard packets not destined to our MAC */
+                       rte_pktmbuf_free(m);
+                       continue;
+               }
+
+               /* return new mbuf to caller */
+               rx_pkts[count++] = m;
+               rxq->bytes += pkt_len;
+       }
+
+       rxq->packets += count;
+
+       /* return the buffers to the free queue */
+       avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
+
+       return count;
+}
+
 static void
 avp_dev_rx_queue_release(void *rx_queue)
 {