af_packet: add PMD for AF_PACKET-based virtual devices
authorJohn W. Linville <linville@tuxdriver.com>
Mon, 17 Nov 2014 15:57:58 +0000 (10:57 -0500)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Mon, 24 Nov 2014 15:39:49 +0000 (16:39 +0100)
This is a Linux-specific virtual PMD driver backed by an AF_PACKET
socket.  This implementation uses mmap'ed ring buffers to limit copying
and user/kernel transitions.  The PACKET_FANOUT_HASH behavior of
AF_PACKET is used for frame reception.  In the current implementation,
Tx and Rx queues are always paired, and therefore are always equal
in number -- changing this would be a Simple Matter Of Programming.

Interfaces of this type are created with a command line option like
"--vdev=eth_af_packet0,iface=...".  There are a number of options available
as arguments:

 - Interface is chosen by "iface" (required)
 - Number of queue pairs set by "qpairs" (optional, default: 1)
 - AF_PACKET MMAP block size set by "blocksz" (optional, default: 4096)
 - AF_PACKET MMAP frame size set by "framesz" (optional, default: 2048)
 - AF_PACKET MMAP frame count set by "framecnt" (optional, default: 512)

Signed-off-by: John W. Linville <linville@tuxdriver.com>
[Thomas: disable because of incompatibility with some kernels]

config/common_linuxapp
lib/Makefile
lib/librte_eal/linuxapp/eal/Makefile
lib/librte_pmd_af_packet/Makefile [new file with mode: 0644]
lib/librte_pmd_af_packet/rte_eth_af_packet.c [new file with mode: 0644]
lib/librte_pmd_af_packet/rte_eth_af_packet.h [new file with mode: 0644]
mk/rte.app.mk

index 57b61c9..86a0d15 100644 (file)
@@ -246,6 +246,11 @@ CONFIG_RTE_LIBRTE_PMD_PCAP=n
 #
 CONFIG_RTE_LIBRTE_PMD_BOND=y
 
+#
+# Compile software PMD backed by AF_PACKET sockets (Linux only)
+#
+CONFIG_RTE_LIBRTE_PMD_AF_PACKET=n
+
 #
 # Compile Xen PMD
 #
index e3237ff..204ef11 100644 (file)
@@ -47,6 +47,7 @@ DIRS-$(CONFIG_RTE_LIBRTE_I40E_PMD) += librte_pmd_i40e
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_BOND) += librte_pmd_bond
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_RING) += librte_pmd_ring
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_PCAP) += librte_pmd_pcap
+DIRS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += librte_pmd_af_packet
 DIRS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += librte_pmd_virtio
 DIRS-$(CONFIG_RTE_LIBRTE_VMXNET3_PMD) += librte_pmd_vmxnet3
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_XENVIRT) += librte_pmd_xenvirt
index c99433e..06c1dc5 100644 (file)
@@ -44,6 +44,7 @@ CFLAGS += -I$(RTE_SDK)/lib/librte_ether
 CFLAGS += -I$(RTE_SDK)/lib/librte_ivshmem
 CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_ring
 CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_pcap
+CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_af_packet
 CFLAGS += -I$(RTE_SDK)/lib/librte_pmd_xenvirt
 CFLAGS += $(WERROR_FLAGS) -O3
 
diff --git a/lib/librte_pmd_af_packet/Makefile b/lib/librte_pmd_af_packet/Makefile
new file mode 100644 (file)
index 0000000..6955e5c
--- /dev/null
@@ -0,0 +1,60 @@
+#   BSD LICENSE
+#
+#   Copyright(c) 2014 John W. Linville <linville@redhat.com>
+#   Copyright(c) 2010-2014 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.
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+#
+# library name
+#
+LIB = librte_pmd_af_packet.a
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+
+#
+# all source are stored in SRCS-y
+#
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += rte_eth_af_packet.c
+
+#
+# Export include files
+#
+SYMLINK-y-include += rte_eth_af_packet.h
+
+# this lib depends upon:
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += lib/librte_mbuf
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += lib/librte_ether
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += lib/librte_malloc
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += lib/librte_kvargs
+
+include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/lib/librte_pmd_af_packet/rte_eth_af_packet.c b/lib/librte_pmd_af_packet/rte_eth_af_packet.c
new file mode 100644 (file)
index 0000000..3b5bd58
--- /dev/null
@@ -0,0 +1,826 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
+ *
+ *   Originally based upon librte_pmd_pcap code:
+ *
+ *   Copyright(c) 2010-2014 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.
+ */
+
+#include <rte_mbuf.h>
+#include <rte_ethdev.h>
+#include <rte_malloc.h>
+#include <rte_kvargs.h>
+#include <rte_dev.h>
+
+#include <linux/if_ether.h>
+#include <linux/if_packet.h>
+#include <arpa/inet.h>
+#include <net/if.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <poll.h>
+
+#include "rte_eth_af_packet.h"
+
+#define ETH_AF_PACKET_IFACE_ARG                "iface"
+#define ETH_AF_PACKET_NUM_Q_ARG                "qpairs"
+#define ETH_AF_PACKET_BLOCKSIZE_ARG    "blocksz"
+#define ETH_AF_PACKET_FRAMESIZE_ARG    "framesz"
+#define ETH_AF_PACKET_FRAMECOUNT_ARG   "framecnt"
+
+#define DFLT_BLOCK_SIZE                (1 << 12)
+#define DFLT_FRAME_SIZE                (1 << 11)
+#define DFLT_FRAME_COUNT       (1 << 9)
+
+struct pkt_rx_queue {
+       int sockfd;
+
+       struct iovec *rd;
+       uint8_t *map;
+       unsigned int framecount;
+       unsigned int framenum;
+
+       struct rte_mempool *mb_pool;
+
+       volatile unsigned long rx_pkts;
+       volatile unsigned long err_pkts;
+};
+
+struct pkt_tx_queue {
+       int sockfd;
+
+       struct iovec *rd;
+       uint8_t *map;
+       unsigned int framecount;
+       unsigned int framenum;
+
+       volatile unsigned long tx_pkts;
+       volatile unsigned long err_pkts;
+};
+
+struct pmd_internals {
+       unsigned nb_queues;
+
+       int if_index;
+       struct ether_addr eth_addr;
+
+       struct tpacket_req req;
+
+       struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
+       struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
+};
+
+static const char *valid_arguments[] = {
+       ETH_AF_PACKET_IFACE_ARG,
+       ETH_AF_PACKET_NUM_Q_ARG,
+       ETH_AF_PACKET_BLOCKSIZE_ARG,
+       ETH_AF_PACKET_FRAMESIZE_ARG,
+       ETH_AF_PACKET_FRAMECOUNT_ARG,
+       NULL
+};
+
+static const char *drivername = "AF_PACKET PMD";
+
+static struct rte_eth_link pmd_link = {
+       .link_speed = 10000,
+       .link_duplex = ETH_LINK_FULL_DUPLEX,
+       .link_status = 0
+};
+
+static uint16_t
+eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+       unsigned i;
+       struct tpacket2_hdr *ppd;
+       struct rte_mbuf *mbuf;
+       uint8_t *pbuf;
+       struct pkt_rx_queue *pkt_q = queue;
+       uint16_t num_rx = 0;
+       unsigned int framecount, framenum;
+
+       if (unlikely(nb_pkts == 0))
+               return 0;
+
+       /*
+        * Reads the given number of packets from the AF_PACKET socket one by
+        * one and copies the packet data into a newly allocated mbuf.
+        */
+       framecount = pkt_q->framecount;
+       framenum = pkt_q->framenum;
+       for (i = 0; i < nb_pkts; i++) {
+               /* point at the next incoming frame */
+               ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
+               if ((ppd->tp_status & TP_STATUS_USER) == 0)
+                       break;
+
+               /* allocate the next mbuf */
+               mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
+               if (unlikely(mbuf == NULL))
+                       break;
+
+               /* packet will fit in the mbuf, go ahead and receive it */
+               rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
+               pbuf = (uint8_t *) ppd + ppd->tp_mac;
+               memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
+
+               /* release incoming frame and advance ring buffer */
+               ppd->tp_status = TP_STATUS_KERNEL;
+               if (++framenum >= framecount)
+                       framenum = 0;
+
+               /* account for the receive frame */
+               bufs[i] = mbuf;
+               num_rx++;
+       }
+       pkt_q->framenum = framenum;
+       pkt_q->rx_pkts += num_rx;
+       return num_rx;
+}
+
+/*
+ * Callback to handle sending packets through a real NIC.
+ */
+static uint16_t
+eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+       struct tpacket2_hdr *ppd;
+       struct rte_mbuf *mbuf;
+       uint8_t *pbuf;
+       unsigned int framecount, framenum;
+       struct pollfd pfd;
+       struct pkt_tx_queue *pkt_q = queue;
+       uint16_t num_tx = 0;
+       int i;
+
+       if (unlikely(nb_pkts == 0))
+               return 0;
+
+       memset(&pfd, 0, sizeof(pfd));
+       pfd.fd = pkt_q->sockfd;
+       pfd.events = POLLOUT;
+       pfd.revents = 0;
+
+       framecount = pkt_q->framecount;
+       framenum = pkt_q->framenum;
+       ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
+       for (i = 0; i < nb_pkts; i++) {
+               /* point at the next incoming frame */
+               if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
+                   (poll(&pfd, 1, -1) < 0))
+                               continue;
+
+               /* copy the tx frame data */
+               mbuf = bufs[num_tx];
+               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);
+
+               /* release incoming frame and advance ring buffer */
+               ppd->tp_status = TP_STATUS_SEND_REQUEST;
+               if (++framenum >= framecount)
+                       framenum = 0;
+               ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
+
+               num_tx++;
+               rte_pktmbuf_free(mbuf);
+       }
+
+       /* kick-off transmits */
+       sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0);
+
+       pkt_q->framenum = framenum;
+       pkt_q->tx_pkts += num_tx;
+       pkt_q->err_pkts += nb_pkts - num_tx;
+       return num_tx;
+}
+
+static int
+eth_dev_start(struct rte_eth_dev *dev)
+{
+       dev->data->dev_link.link_status = 1;
+       return 0;
+}
+
+/*
+ * This function gets called when the current port gets stopped.
+ */
+static void
+eth_dev_stop(struct rte_eth_dev *dev)
+{
+       unsigned i;
+       int sockfd;
+       struct pmd_internals *internals = dev->data->dev_private;
+
+       for (i = 0; i < internals->nb_queues; i++) {
+               sockfd = internals->rx_queue[i].sockfd;
+               if (sockfd != -1)
+                       close(sockfd);
+               sockfd = internals->tx_queue[i].sockfd;
+               if (sockfd != -1)
+                       close(sockfd);
+       }
+
+       dev->data->dev_link.link_status = 0;
+}
+
+static int
+eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
+{
+       return 0;
+}
+
+static void
+eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
+{
+       struct pmd_internals *internals = dev->data->dev_private;
+
+       dev_info->driver_name = drivername;
+       dev_info->if_index = internals->if_index;
+       dev_info->max_mac_addrs = 1;
+       dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
+       dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
+       dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
+       dev_info->min_rx_bufsize = 0;
+       dev_info->pci_dev = NULL;
+}
+
+static void
+eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
+{
+       unsigned i, imax;
+       unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
+       const struct pmd_internals *internal = dev->data->dev_private;
+
+       memset(igb_stats, 0, sizeof(*igb_stats));
+
+       imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
+               internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
+       for (i = 0; i < imax; i++) {
+               igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
+               rx_total += igb_stats->q_ipackets[i];
+       }
+
+       imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
+               internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
+       for (i = 0; i < imax; i++) {
+               igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
+               igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
+               tx_total += igb_stats->q_opackets[i];
+               tx_err_total += igb_stats->q_errors[i];
+       }
+
+       igb_stats->ipackets = rx_total;
+       igb_stats->opackets = tx_total;
+       igb_stats->oerrors = tx_err_total;
+}
+
+static void
+eth_stats_reset(struct rte_eth_dev *dev)
+{
+       unsigned i;
+       struct pmd_internals *internal = dev->data->dev_private;
+
+       for (i = 0; i < internal->nb_queues; i++)
+               internal->rx_queue[i].rx_pkts = 0;
+
+       for (i = 0; i < internal->nb_queues; i++) {
+               internal->tx_queue[i].tx_pkts = 0;
+               internal->tx_queue[i].err_pkts = 0;
+       }
+}
+
+static void
+eth_dev_close(struct rte_eth_dev *dev __rte_unused)
+{
+}
+
+static void
+eth_queue_release(void *q __rte_unused)
+{
+}
+
+static int
+eth_link_update(struct rte_eth_dev *dev __rte_unused,
+                int wait_to_complete __rte_unused)
+{
+       return 0;
+}
+
+static int
+eth_rx_queue_setup(struct rte_eth_dev *dev,
+                   uint16_t rx_queue_id,
+                   uint16_t nb_rx_desc __rte_unused,
+                   unsigned int socket_id __rte_unused,
+                   const struct rte_eth_rxconf *rx_conf __rte_unused,
+                   struct rte_mempool *mb_pool)
+{
+       struct pmd_internals *internals = dev->data->dev_private;
+       struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
+       struct rte_pktmbuf_pool_private *mbp_priv;
+       uint16_t buf_size;
+
+       pkt_q->mb_pool = mb_pool;
+
+       /* Now get the space available for data in the mbuf */
+       mbp_priv = rte_mempool_get_priv(pkt_q->mb_pool);
+       buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
+                              RTE_PKTMBUF_HEADROOM);
+
+       if (ETH_FRAME_LEN > buf_size) {
+               RTE_LOG(ERR, PMD,
+                       "%s: %d bytes will not fit in mbuf (%d bytes)\n",
+                       dev->data->name, ETH_FRAME_LEN, buf_size);
+               return -ENOMEM;
+       }
+
+       dev->data->rx_queues[rx_queue_id] = pkt_q;
+
+       return 0;
+}
+
+static int
+eth_tx_queue_setup(struct rte_eth_dev *dev,
+                   uint16_t tx_queue_id,
+                   uint16_t nb_tx_desc __rte_unused,
+                   unsigned int socket_id __rte_unused,
+                   const struct rte_eth_txconf *tx_conf __rte_unused)
+{
+
+       struct pmd_internals *internals = dev->data->dev_private;
+
+       dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
+       return 0;
+}
+
+static struct eth_dev_ops ops = {
+       .dev_start = eth_dev_start,
+       .dev_stop = eth_dev_stop,
+       .dev_close = eth_dev_close,
+       .dev_configure = eth_dev_configure,
+       .dev_infos_get = eth_dev_info,
+       .rx_queue_setup = eth_rx_queue_setup,
+       .tx_queue_setup = eth_tx_queue_setup,
+       .rx_queue_release = eth_queue_release,
+       .tx_queue_release = eth_queue_release,
+       .link_update = eth_link_update,
+       .stats_get = eth_stats_get,
+       .stats_reset = eth_stats_reset,
+};
+
+/*
+ * Opens an AF_PACKET socket
+ */
+static int
+open_packet_iface(const char *key __rte_unused,
+                  const char *value __rte_unused,
+                  void *extra_args)
+{
+       int *sockfd = extra_args;
+
+       /* Open an AF_PACKET socket... */
+       *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+       if (*sockfd == -1) {
+               RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
+               return -1;
+       }
+
+       return 0;
+}
+
+static int
+rte_pmd_init_internals(const char *name,
+                       const int sockfd,
+                       const unsigned nb_queues,
+                       unsigned int blocksize,
+                       unsigned int blockcnt,
+                       unsigned int framesize,
+                       unsigned int framecnt,
+                       const unsigned numa_node,
+                       struct pmd_internals **internals,
+                       struct rte_eth_dev **eth_dev,
+                       struct rte_kvargs *kvlist)
+{
+       struct rte_eth_dev_data *data = NULL;
+       struct rte_pci_device *pci_dev = NULL;
+       struct rte_kvargs_pair *pair = NULL;
+       struct ifreq ifr;
+       size_t ifnamelen;
+       unsigned k_idx;
+       struct sockaddr_ll sockaddr;
+       struct tpacket_req *req;
+       struct pkt_rx_queue *rx_queue;
+       struct pkt_tx_queue *tx_queue;
+       int rc, tpver, discard, bypass;
+       unsigned int i, q, rdsize;
+       int qsockfd, fanout_arg;
+
+       for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
+               pair = &kvlist->pairs[k_idx];
+               if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
+                       break;
+       }
+       if (pair == NULL) {
+               RTE_LOG(ERR, PMD,
+                       "%s: no interface specified for AF_PACKET ethdev\n",
+                       name);
+               goto error;
+       }
+
+       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;
+
+       pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
+       if (pci_dev == NULL)
+               goto error;
+
+       *internals = rte_zmalloc_socket(name, sizeof(**internals),
+                                       0, numa_node);
+       if (*internals == NULL)
+               goto error;
+
+       req = &((*internals)->req);
+
+       req->tp_block_size = blocksize;
+       req->tp_block_nr = blockcnt;
+       req->tp_frame_size = framesize;
+       req->tp_frame_nr = framecnt;
+
+       ifnamelen = strlen(pair->value);
+       if (ifnamelen < sizeof(ifr.ifr_name)) {
+               memcpy(ifr.ifr_name, pair->value, ifnamelen);
+               ifr.ifr_name[ifnamelen] = '\0';
+       } else {
+               RTE_LOG(ERR, PMD,
+                       "%s: I/F name too long (%s)\n",
+                       name, pair->value);
+               goto error;
+       }
+       if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
+               RTE_LOG(ERR, PMD,
+                       "%s: ioctl failed (SIOCGIFINDEX)\n",
+                       name);
+               goto error;
+       }
+       (*internals)->if_index = ifr.ifr_ifindex;
+
+       if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
+               RTE_LOG(ERR, PMD,
+                       "%s: ioctl failed (SIOCGIFHWADDR)\n",
+                       name);
+               goto error;
+       }
+       memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
+
+       memset(&sockaddr, 0, sizeof(sockaddr));
+       sockaddr.sll_family = AF_PACKET;
+       sockaddr.sll_protocol = htons(ETH_P_ALL);
+       sockaddr.sll_ifindex = (*internals)->if_index;
+
+       fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
+       fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG |
+                      PACKET_FANOUT_FLAG_ROLLOVER) << 16;
+
+       for (q = 0; q < nb_queues; q++) {
+               /* Open an AF_PACKET socket for this queue... */
+               qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+               if (qsockfd == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not open AF_PACKET socket\n",
+                               name);
+                       return -1;
+               }
+
+               tpver = TPACKET_V2;
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
+                               &tpver, sizeof(tpver));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_VERSION on AF_PACKET "
+                               "socket for %s\n", name, pair->value);
+                       goto error;
+               }
+
+               discard = 1;
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
+                               &discard, sizeof(discard));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_LOSS on "
+                               "AF_PACKET socket for %s\n", name, pair->value);
+                       goto error;
+               }
+
+               bypass = 1;
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
+                               &bypass, sizeof(bypass));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_QDISC_BYPASS "
+                               "on AF_PACKET socket for %s\n", name,
+                               pair->value);
+                       goto error;
+               }
+
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_RX_RING on AF_PACKET "
+                               "socket for %s\n", name, pair->value);
+                       goto error;
+               }
+
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_TX_RING on AF_PACKET "
+                               "socket for %s\n", name, pair->value);
+                       goto error;
+               }
+
+               rx_queue = &((*internals)->rx_queue[q]);
+               rx_queue->framecount = req->tp_frame_nr;
+
+               rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
+                                   PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
+                                   qsockfd, 0);
+               if (rx_queue->map == MAP_FAILED) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: call to mmap failed on AF_PACKET socket for %s\n",
+                               name, pair->value);
+                       goto error;
+               }
+
+               /* rdsize is same for both Tx and Rx */
+               rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
+
+               rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
+               for (i = 0; i < req->tp_frame_nr; ++i) {
+                       rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
+                       rx_queue->rd[i].iov_len = req->tp_frame_size;
+               }
+               rx_queue->sockfd = qsockfd;
+
+               tx_queue = &((*internals)->tx_queue[q]);
+               tx_queue->framecount = req->tp_frame_nr;
+
+               tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
+
+               tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
+               for (i = 0; i < req->tp_frame_nr; ++i) {
+                       tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
+                       tx_queue->rd[i].iov_len = req->tp_frame_size;
+               }
+               tx_queue->sockfd = qsockfd;
+
+               rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not bind AF_PACKET socket to %s\n",
+                               name, pair->value);
+                       goto error;
+               }
+
+               rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
+                               &fanout_arg, sizeof(fanout_arg));
+               if (rc == -1) {
+                       RTE_LOG(ERR, PMD,
+                               "%s: could not set PACKET_FANOUT on AF_PACKET socket "
+                               "for %s\n", name, pair->value);
+                       goto error;
+               }
+       }
+
+       /* reserve an ethdev entry */
+       *eth_dev = rte_eth_dev_allocate(name);
+       if (*eth_dev == NULL)
+               goto error;
+
+       /*
+        * now put it all together
+        * - store queue data in internals,
+        * - store numa_node info in pci_driver
+        * - point eth_dev_data to internals and pci_driver
+        * - and point eth_dev structure to new eth_dev_data structure
+        */
+
+       (*internals)->nb_queues = nb_queues;
+
+       data->dev_private = *internals;
+       data->port_id = (*eth_dev)->data->port_id;
+       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;
+
+       pci_dev->numa_node = numa_node;
+
+       (*eth_dev)->data = data;
+       (*eth_dev)->dev_ops = &ops;
+       (*eth_dev)->pci_dev = pci_dev;
+
+       return 0;
+
+error:
+       if (data)
+               rte_free(data);
+       if (pci_dev)
+               rte_free(pci_dev);
+       for (q = 0; q < nb_queues; q++) {
+               if ((*internals)->rx_queue[q].rd)
+                       rte_free((*internals)->rx_queue[q].rd);
+               if ((*internals)->tx_queue[q].rd)
+                       rte_free((*internals)->tx_queue[q].rd);
+       }
+       if (*internals)
+               rte_free(*internals);
+       return -1;
+}
+
+static int
+rte_eth_from_packet(const char *name,
+                    int const *sockfd,
+                    const unsigned numa_node,
+                    struct rte_kvargs *kvlist)
+{
+       struct pmd_internals *internals = NULL;
+       struct rte_eth_dev *eth_dev = NULL;
+       struct rte_kvargs_pair *pair = NULL;
+       unsigned k_idx;
+       unsigned int blockcount;
+       unsigned int blocksize = DFLT_BLOCK_SIZE;
+       unsigned int framesize = DFLT_FRAME_SIZE;
+       unsigned int framecount = DFLT_FRAME_COUNT;
+       unsigned int qpairs = 1;
+
+       /* do some parameter checking */
+       if (*sockfd < 0)
+               return -1;
+
+       /*
+        * Walk arguments for configurable settings
+        */
+       for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
+               pair = &kvlist->pairs[k_idx];
+               if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
+                       qpairs = atoi(pair->value);
+                       if (qpairs < 1 ||
+                           qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
+                               RTE_LOG(ERR, PMD,
+                                       "%s: invalid qpairs value\n",
+                                       name);
+                               return -1;
+                       }
+                       continue;
+               }
+               if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
+                       blocksize = atoi(pair->value);
+                       if (!blocksize) {
+                               RTE_LOG(ERR, PMD,
+                                       "%s: invalid blocksize value\n",
+                                       name);
+                               return -1;
+                       }
+                       continue;
+               }
+               if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
+                       framesize = atoi(pair->value);
+                       if (!framesize) {
+                               RTE_LOG(ERR, PMD,
+                                       "%s: invalid framesize value\n",
+                                       name);
+                               return -1;
+                       }
+                       continue;
+               }
+               if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
+                       framecount = atoi(pair->value);
+                       if (!framecount) {
+                               RTE_LOG(ERR, PMD,
+                                       "%s: invalid framecount value\n",
+                                       name);
+                               return -1;
+                       }
+                       continue;
+               }
+       }
+
+       if (framesize > blocksize) {
+               RTE_LOG(ERR, PMD,
+                       "%s: AF_PACKET MMAP frame size exceeds block size!\n",
+                       name);
+               return -1;
+       }
+
+       blockcount = framecount / (blocksize / framesize);
+       if (!blockcount) {
+               RTE_LOG(ERR, PMD,
+                       "%s: invalid AF_PACKET MMAP parameters\n", name);
+               return -1;
+       }
+
+       RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
+       RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
+       RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
+       RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
+       RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
+
+       if (rte_pmd_init_internals(name, *sockfd, qpairs,
+                                  blocksize, blockcount,
+                                  framesize, framecount,
+                                  numa_node, &internals, &eth_dev,
+                                  kvlist) < 0)
+               return -1;
+
+       eth_dev->rx_pkt_burst = eth_af_packet_rx;
+       eth_dev->tx_pkt_burst = eth_af_packet_tx;
+
+       return 0;
+}
+
+int
+rte_pmd_af_packet_devinit(const char *name, const char *params)
+{
+       unsigned numa_node;
+       int ret;
+       struct rte_kvargs *kvlist;
+       int sockfd = -1;
+
+       RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
+
+       numa_node = rte_socket_id();
+
+       kvlist = rte_kvargs_parse(params, valid_arguments);
+       if (kvlist == NULL)
+               return -1;
+
+       /*
+        * If iface argument is passed we open the NICs and use them for
+        * reading / writing
+        */
+       if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
+
+               ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
+                                        &open_packet_iface, &sockfd);
+               if (ret < 0)
+                       return -1;
+       }
+
+       ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
+       close(sockfd); /* no longer needed */
+
+       if (ret < 0)
+               return -1;
+
+       return 0;
+}
+
+static struct rte_driver pmd_af_packet_drv = {
+       .name = "eth_af_packet",
+       .type = PMD_VDEV,
+       .init = rte_pmd_af_packet_devinit,
+};
+
+PMD_REGISTER_DRIVER(pmd_af_packet_drv);
diff --git a/lib/librte_pmd_af_packet/rte_eth_af_packet.h b/lib/librte_pmd_af_packet/rte_eth_af_packet.h
new file mode 100644 (file)
index 0000000..5d1bc7e
--- /dev/null
@@ -0,0 +1,53 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   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.
+ */
+
+#ifndef _RTE_ETH_AF_PACKET_H_
+#define _RTE_ETH_AF_PACKET_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_PMD_AF_PACKET_MAX_RINGS 16
+
+/**
+ * For use by the EAL only. Called as part of EAL init to set up any dummy NICs
+ * configured on command line.
+ */
+int rte_pmd_af_packet_devinit(const char *name, const char *params);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
index 285b65c..59468b0 100644 (file)
@@ -215,6 +215,10 @@ ifeq ($(CONFIG_RTE_LIBRTE_PMD_PCAP),y)
 LDLIBS += -lrte_pmd_pcap -lpcap
 endif
 
+ifeq ($(CONFIG_RTE_LIBRTE_PMD_AF_PACKET),y)
+LDLIBS += -lrte_pmd_af_packet
+endif
+
 endif # plugins
 
 LDLIBS += $(EXECENV_LDLIBS)