drivers/net: share vdev data to secondary process
[dpdk.git] / drivers / net / af_packet / rte_eth_af_packet.c
index a03966a..b394d3c 100644 (file)
@@ -1,47 +1,17 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
- *
- *   Originally based upon librte_pmd_pcap code:
- *
- *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
- *   Copyright(c) 2014 6WIND S.A.
- *   All rights reserved.
- *
- *   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 Intel Corporation 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) 2014 John W. Linville <linville@tuxdriver.com>
+ * Originally based upon librte_pmd_pcap code:
+ * Copyright(c) 2010-2015 Intel Corporation.
+ * Copyright(c) 2014 6WIND S.A.
+ * All rights reserved.
  */
 
 #include <rte_mbuf.h>
-#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
 #include <rte_ethdev_vdev.h>
 #include <rte_malloc.h>
 #include <rte_kvargs.h>
-#include <rte_vdev.h>
+#include <rte_bus_vdev.h>
 
 #include <linux/if_ether.h>
 #include <linux/if_packet.h>
@@ -59,6 +29,7 @@
 #define ETH_AF_PACKET_BLOCKSIZE_ARG    "blocksz"
 #define ETH_AF_PACKET_FRAMESIZE_ARG    "framesz"
 #define ETH_AF_PACKET_FRAMECOUNT_ARG   "framecnt"
+#define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
 
 #define DFLT_BLOCK_SIZE                (1 << 12)
 #define DFLT_FRAME_SIZE                (1 << 11)
@@ -75,7 +46,7 @@ struct pkt_rx_queue {
        unsigned int framenum;
 
        struct rte_mempool *mb_pool;
-       uint8_t in_port;
+       uint16_t in_port;
 
        volatile unsigned long rx_pkts;
        volatile unsigned long err_pkts;
@@ -115,6 +86,7 @@ static const char *valid_arguments[] = {
        ETH_AF_PACKET_BLOCKSIZE_ARG,
        ETH_AF_PACKET_FRAMESIZE_ARG,
        ETH_AF_PACKET_FRAMECOUNT_ARG,
+       ETH_AF_PACKET_QDISC_BYPASS_ARG,
        NULL
 };
 
@@ -122,7 +94,7 @@ static struct rte_eth_link pmd_link = {
        .link_speed = ETH_SPEED_NUM_10G,
        .link_duplex = ETH_LINK_FULL_DUPLEX,
        .link_status = ETH_LINK_DOWN,
-       .link_autoneg = ETH_LINK_SPEED_AUTONEG
+       .link_autoneg = ETH_LINK_AUTONEG
 };
 
 static uint16_t
@@ -165,7 +137,7 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                /* check for vlan info */
                if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
                        mbuf->vlan_tci = ppd->tp_vlan_tci;
-                       mbuf->ol_flags |= (PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED);
+                       mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
                }
 
                /* release incoming frame and advance ring buffer */
@@ -216,7 +188,7 @@ eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                mbuf = *bufs++;
 
                /* drop oversized packets */
-               if (rte_pktmbuf_data_len(mbuf) > pkt_q->frame_data_size) {
+               if (mbuf->pkt_len > pkt_q->frame_data_size) {
                        rte_pktmbuf_free(mbuf);
                        continue;
                }
@@ -237,8 +209,17 @@ eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                /* copy the tx frame data */
                pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
                        sizeof(struct sockaddr_ll);
-               memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
-               ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
+
+               struct rte_mbuf *tmp_mbuf = mbuf;
+               while (tmp_mbuf) {
+                       uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
+                       memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
+                       pbuf += data_len;
+                       tmp_mbuf = tmp_mbuf->next;
+               }
+
+               ppd->tp_len = mbuf->pkt_len;
+               ppd->tp_snaplen = mbuf->pkt_len;
 
                /* release incoming frame and advance ring buffer */
                ppd->tp_status = TP_STATUS_SEND_REQUEST;
@@ -252,8 +233,11 @@ eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        }
 
        /* kick-off transmits */
-       if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)
-               num_tx = 0; /* error sending -- no packets transmitted */
+       if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
+               /* error sending -- no packets transmitted */
+               num_tx = 0;
+               num_tx_bytes = 0;
+       }
 
        pkt_q->framenum = framenum;
        pkt_q->tx_pkts += num_tx;
@@ -317,7 +301,7 @@ eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        dev_info->min_rx_bufsize = 0;
 }
 
-static void
+static int
 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 {
        unsigned i, imax;
@@ -350,6 +334,7 @@ eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
        igb_stats->opackets = tx_total;
        igb_stats->oerrors = tx_err_total;
        igb_stats->obytes = tx_bytes_total;
+       return 0;
 }
 
 static void
@@ -410,7 +395,7 @@ eth_rx_queue_setup(struct rte_eth_dev *dev,
        if (data_size > buf_size) {
                RTE_LOG(ERR, PMD,
                        "%s: %d bytes will not fit in mbuf (%d bytes)\n",
-                       dev->data->name, data_size, buf_size);
+                       dev->device->name, data_size, buf_size);
                return -ENOMEM;
        }
 
@@ -547,6 +532,7 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
                        unsigned int blockcnt,
                        unsigned int framesize,
                        unsigned int framecnt,
+                      unsigned int qdisc_bypass,
                        struct pmd_internals **internals,
                        struct rte_eth_dev **eth_dev,
                        struct rte_kvargs *kvlist)
@@ -568,9 +554,6 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
 #if defined(PACKET_FANOUT)
        int fanout_arg;
 #endif
-#if defined(PACKET_QDISC_BYPASS)
-       int bypass;
-#endif
 
        for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
                pair = &kvlist->pairs[k_idx];
@@ -581,25 +564,17 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
                RTE_LOG(ERR, PMD,
                        "%s: no interface specified for AF_PACKET ethdev\n",
                        name);
-               goto error_early;
+               return -1;
        }
 
        RTE_LOG(INFO, PMD,
                "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
                name, numa_node);
 
-       /*
-        * now do all data allocation - for eth_dev structure, dummy pci driver
-        * and internal (private) data
-        */
-       data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
-       if (data == NULL)
-               goto error_early;
-
        *internals = rte_zmalloc_socket(name, sizeof(**internals),
                                        0, numa_node);
        if (*internals == NULL)
-               goto error_early;
+               return -1;
 
        for (q = 0; q < nb_queues; q++) {
                (*internals)->rx_queue[q].map = MAP_FAILED;
@@ -621,22 +596,24 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
                RTE_LOG(ERR, PMD,
                        "%s: I/F name too long (%s)\n",
                        name, pair->value);
-               goto error_early;
+               return -1;
        }
        if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
                RTE_LOG(ERR, PMD,
                        "%s: ioctl failed (SIOCGIFINDEX)\n",
                        name);
-               goto error_early;
+               return -1;
        }
        (*internals)->if_name = strdup(pair->value);
+       if ((*internals)->if_name == NULL)
+               return -1;
        (*internals)->if_index = ifr.ifr_ifindex;
 
        if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
                RTE_LOG(ERR, PMD,
                        "%s: ioctl failed (SIOCGIFHWADDR)\n",
                        name);
-               goto error_early;
+               return -1;
        }
        memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 
@@ -684,9 +661,8 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
                }
 
 #if defined(PACKET_QDISC_BYPASS)
-               bypass = 1;
                rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
-                               &bypass, sizeof(bypass));
+                               &qdisc_bypass, sizeof(qdisc_bypass));
                if (rc == -1) {
                        RTE_LOG(ERR, PMD,
                                "%s: could not set PACKET_QDISC_BYPASS "
@@ -694,6 +670,8 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
                                pair->value);
                        goto error;
                }
+#else
+               RTE_SET_USED(qdisc_bypass);
 #endif
 
                rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
@@ -789,16 +767,14 @@ rte_pmd_init_internals(struct rte_vdev_device *dev,
 
        (*internals)->nb_queues = nb_queues;
 
-       rte_memcpy(data, (*eth_dev)->data, sizeof(*data));
+       data = (*eth_dev)->data;
        data->dev_private = *internals;
        data->nb_rx_queues = (uint16_t)nb_queues;
        data->nb_tx_queues = (uint16_t)nb_queues;
        data->dev_link = pmd_link;
        data->mac_addrs = &(*internals)->eth_addr;
 
-       (*eth_dev)->data = data;
        (*eth_dev)->dev_ops = &ops;
-       (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 
        return 0;
 
@@ -817,8 +793,6 @@ error:
        }
        free((*internals)->if_name);
        rte_free(*internals);
-error_early:
-       rte_free(data);
        return -1;
 }
 
@@ -837,6 +811,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
        unsigned int framesize = DFLT_FRAME_SIZE;
        unsigned int framecount = DFLT_FRAME_COUNT;
        unsigned int qpairs = 1;
+       unsigned int qdisc_bypass = 1;
 
        /* do some parameter checking */
        if (*sockfd < 0)
@@ -888,6 +863,16 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
                        }
                        continue;
                }
+               if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
+                       qdisc_bypass = atoi(pair->value);
+                       if (qdisc_bypass > 1) {
+                               RTE_LOG(ERR, PMD,
+                                       "%s: invalid bypass value\n",
+                                       name);
+                               return -1;
+                       }
+                       continue;
+               }
        }
 
        if (framesize > blocksize) {
@@ -913,6 +898,7 @@ rte_eth_from_packet(struct rte_vdev_device *dev,
        if (rte_pmd_init_internals(dev, *sockfd, qpairs,
                                   blocksize, blockcount,
                                   framesize, framecount,
+                                  qdisc_bypass,
                                   &internals, &eth_dev,
                                   kvlist) < 0)
                return -1;
@@ -929,9 +915,22 @@ rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
        int ret = 0;
        struct rte_kvargs *kvlist;
        int sockfd = -1;
+       struct rte_eth_dev *eth_dev;
+       const char *name = rte_vdev_device_name(dev);
 
-       RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n",
-               rte_vdev_device_name(dev));
+       RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
+
+       if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
+           strlen(rte_vdev_device_args(dev)) == 0) {
+               eth_dev = rte_eth_dev_attach_secondary(name);
+               if (!eth_dev) {
+                       RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
+                       return -1;
+               }
+               /* TODO: request info from primary to set up Rx and Tx */
+               eth_dev->dev_ops = &ops;
+               return 0;
+       }
 
        kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
        if (kvlist == NULL) {
@@ -988,7 +987,6 @@ rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
        free(internals->if_name);
 
        rte_free(eth_dev->data->dev_private);
-       rte_free(eth_dev->data);
 
        rte_eth_dev_release_port(eth_dev);
 
@@ -1007,4 +1005,5 @@ RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
        "qpairs=<int> "
        "blocksz=<int> "
        "framesz=<int> "
-       "framecnt=<int>");
+       "framecnt=<int> "
+       "qdisc_bypass=<0|1>");