4 * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
6 * Originally based upon librte_pmd_pcap code:
8 * Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
9 * Copyright(c) 2014 6WIND S.A.
10 * All rights reserved.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
16 * * Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * * Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in
20 * the documentation and/or other materials provided with the
22 * * Neither the name of Intel Corporation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
27 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
28 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
29 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
30 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
31 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
32 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
33 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
34 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
35 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
36 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 #include <rte_ethdev.h>
41 #include <rte_ethdev_vdev.h>
42 #include <rte_malloc.h>
43 #include <rte_kvargs.h>
46 #include <linux/if_ether.h>
47 #include <linux/if_packet.h>
48 #include <arpa/inet.h>
50 #include <sys/types.h>
51 #include <sys/socket.h>
52 #include <sys/ioctl.h>
57 #define ETH_AF_PACKET_IFACE_ARG "iface"
58 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
59 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
60 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
61 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
63 #define DFLT_BLOCK_SIZE (1 << 12)
64 #define DFLT_FRAME_SIZE (1 << 11)
65 #define DFLT_FRAME_COUNT (1 << 9)
67 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
74 unsigned int framecount;
75 unsigned int framenum;
77 struct rte_mempool *mb_pool;
80 volatile unsigned long rx_pkts;
81 volatile unsigned long err_pkts;
82 volatile unsigned long rx_bytes;
87 unsigned int frame_data_size;
91 unsigned int framecount;
92 unsigned int framenum;
94 volatile unsigned long tx_pkts;
95 volatile unsigned long err_pkts;
96 volatile unsigned long tx_bytes;
99 struct pmd_internals {
104 struct ether_addr eth_addr;
106 struct tpacket_req req;
108 struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
109 struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
112 static const char *valid_arguments[] = {
113 ETH_AF_PACKET_IFACE_ARG,
114 ETH_AF_PACKET_NUM_Q_ARG,
115 ETH_AF_PACKET_BLOCKSIZE_ARG,
116 ETH_AF_PACKET_FRAMESIZE_ARG,
117 ETH_AF_PACKET_FRAMECOUNT_ARG,
121 static struct rte_eth_link pmd_link = {
122 .link_speed = ETH_SPEED_NUM_10G,
123 .link_duplex = ETH_LINK_FULL_DUPLEX,
124 .link_status = ETH_LINK_DOWN,
125 .link_autoneg = ETH_LINK_SPEED_AUTONEG
129 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
132 struct tpacket2_hdr *ppd;
133 struct rte_mbuf *mbuf;
135 struct pkt_rx_queue *pkt_q = queue;
137 unsigned long num_rx_bytes = 0;
138 unsigned int framecount, framenum;
140 if (unlikely(nb_pkts == 0))
144 * Reads the given number of packets from the AF_PACKET socket one by
145 * one and copies the packet data into a newly allocated mbuf.
147 framecount = pkt_q->framecount;
148 framenum = pkt_q->framenum;
149 for (i = 0; i < nb_pkts; i++) {
150 /* point at the next incoming frame */
151 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
152 if ((ppd->tp_status & TP_STATUS_USER) == 0)
155 /* allocate the next mbuf */
156 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
157 if (unlikely(mbuf == NULL))
160 /* packet will fit in the mbuf, go ahead and receive it */
161 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
162 pbuf = (uint8_t *) ppd + ppd->tp_mac;
163 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
165 /* check for vlan info */
166 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
167 mbuf->vlan_tci = ppd->tp_vlan_tci;
168 mbuf->ol_flags |= (PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED);
171 /* release incoming frame and advance ring buffer */
172 ppd->tp_status = TP_STATUS_KERNEL;
173 if (++framenum >= framecount)
175 mbuf->port = pkt_q->in_port;
177 /* account for the receive frame */
180 num_rx_bytes += mbuf->pkt_len;
182 pkt_q->framenum = framenum;
183 pkt_q->rx_pkts += num_rx;
184 pkt_q->rx_bytes += num_rx_bytes;
189 * Callback to handle sending packets through a real NIC.
192 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
194 struct tpacket2_hdr *ppd;
195 struct rte_mbuf *mbuf;
197 unsigned int framecount, framenum;
199 struct pkt_tx_queue *pkt_q = queue;
201 unsigned long num_tx_bytes = 0;
204 if (unlikely(nb_pkts == 0))
207 memset(&pfd, 0, sizeof(pfd));
208 pfd.fd = pkt_q->sockfd;
209 pfd.events = POLLOUT;
212 framecount = pkt_q->framecount;
213 framenum = pkt_q->framenum;
214 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
215 for (i = 0; i < nb_pkts; i++) {
218 /* drop oversized packets */
219 if (rte_pktmbuf_data_len(mbuf) > pkt_q->frame_data_size) {
220 rte_pktmbuf_free(mbuf);
224 /* insert vlan info if necessary */
225 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
226 if (rte_vlan_insert(&mbuf)) {
227 rte_pktmbuf_free(mbuf);
232 /* point at the next incoming frame */
233 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
234 (poll(&pfd, 1, -1) < 0))
237 /* copy the tx frame data */
238 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
239 sizeof(struct sockaddr_ll);
240 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
241 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
243 /* release incoming frame and advance ring buffer */
244 ppd->tp_status = TP_STATUS_SEND_REQUEST;
245 if (++framenum >= framecount)
247 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
250 num_tx_bytes += mbuf->pkt_len;
251 rte_pktmbuf_free(mbuf);
254 /* kick-off transmits */
255 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
256 /* error sending -- no packets transmitted */
261 pkt_q->framenum = framenum;
262 pkt_q->tx_pkts += num_tx;
263 pkt_q->err_pkts += i - num_tx;
264 pkt_q->tx_bytes += num_tx_bytes;
269 eth_dev_start(struct rte_eth_dev *dev)
271 dev->data->dev_link.link_status = ETH_LINK_UP;
276 * This function gets called when the current port gets stopped.
279 eth_dev_stop(struct rte_eth_dev *dev)
283 struct pmd_internals *internals = dev->data->dev_private;
285 for (i = 0; i < internals->nb_queues; i++) {
286 sockfd = internals->rx_queue[i].sockfd;
290 /* Prevent use after free in case tx fd == rx fd */
291 if (sockfd != internals->tx_queue[i].sockfd) {
292 sockfd = internals->tx_queue[i].sockfd;
297 internals->rx_queue[i].sockfd = -1;
298 internals->tx_queue[i].sockfd = -1;
301 dev->data->dev_link.link_status = ETH_LINK_DOWN;
305 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
311 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
313 struct pmd_internals *internals = dev->data->dev_private;
315 dev_info->if_index = internals->if_index;
316 dev_info->max_mac_addrs = 1;
317 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
318 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
319 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
320 dev_info->min_rx_bufsize = 0;
324 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
327 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
328 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
329 const struct pmd_internals *internal = dev->data->dev_private;
331 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
332 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
333 for (i = 0; i < imax; i++) {
334 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
335 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
336 rx_total += igb_stats->q_ipackets[i];
337 rx_bytes_total += igb_stats->q_ibytes[i];
340 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
341 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
342 for (i = 0; i < imax; i++) {
343 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
344 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
345 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
346 tx_total += igb_stats->q_opackets[i];
347 tx_err_total += igb_stats->q_errors[i];
348 tx_bytes_total += igb_stats->q_obytes[i];
351 igb_stats->ipackets = rx_total;
352 igb_stats->ibytes = rx_bytes_total;
353 igb_stats->opackets = tx_total;
354 igb_stats->oerrors = tx_err_total;
355 igb_stats->obytes = tx_bytes_total;
359 eth_stats_reset(struct rte_eth_dev *dev)
362 struct pmd_internals *internal = dev->data->dev_private;
364 for (i = 0; i < internal->nb_queues; i++) {
365 internal->rx_queue[i].rx_pkts = 0;
366 internal->rx_queue[i].rx_bytes = 0;
369 for (i = 0; i < internal->nb_queues; i++) {
370 internal->tx_queue[i].tx_pkts = 0;
371 internal->tx_queue[i].err_pkts = 0;
372 internal->tx_queue[i].tx_bytes = 0;
377 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
382 eth_queue_release(void *q __rte_unused)
387 eth_link_update(struct rte_eth_dev *dev __rte_unused,
388 int wait_to_complete __rte_unused)
394 eth_rx_queue_setup(struct rte_eth_dev *dev,
395 uint16_t rx_queue_id,
396 uint16_t nb_rx_desc __rte_unused,
397 unsigned int socket_id __rte_unused,
398 const struct rte_eth_rxconf *rx_conf __rte_unused,
399 struct rte_mempool *mb_pool)
401 struct pmd_internals *internals = dev->data->dev_private;
402 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
403 unsigned int buf_size, data_size;
405 pkt_q->mb_pool = mb_pool;
407 /* Now get the space available for data in the mbuf */
408 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
409 RTE_PKTMBUF_HEADROOM;
410 data_size = internals->req.tp_frame_size;
411 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
413 if (data_size > buf_size) {
415 "%s: %d bytes will not fit in mbuf (%d bytes)\n",
416 dev->device->name, data_size, buf_size);
420 dev->data->rx_queues[rx_queue_id] = pkt_q;
421 pkt_q->in_port = dev->data->port_id;
427 eth_tx_queue_setup(struct rte_eth_dev *dev,
428 uint16_t tx_queue_id,
429 uint16_t nb_tx_desc __rte_unused,
430 unsigned int socket_id __rte_unused,
431 const struct rte_eth_txconf *tx_conf __rte_unused)
434 struct pmd_internals *internals = dev->data->dev_private;
436 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
441 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
443 struct pmd_internals *internals = dev->data->dev_private;
444 struct ifreq ifr = { .ifr_mtu = mtu };
447 unsigned int data_size = internals->req.tp_frame_size -
449 sizeof(struct sockaddr_ll);
454 s = socket(PF_INET, SOCK_DGRAM, 0);
458 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", internals->if_name);
459 ret = ioctl(s, SIOCSIFMTU, &ifr);
469 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
474 s = socket(PF_INET, SOCK_DGRAM, 0);
478 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", if_name);
479 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
481 ifr.ifr_flags &= mask;
482 ifr.ifr_flags |= flags;
483 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
490 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
492 struct pmd_internals *internals = dev->data->dev_private;
494 eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
498 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
500 struct pmd_internals *internals = dev->data->dev_private;
502 eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
505 static const struct eth_dev_ops ops = {
506 .dev_start = eth_dev_start,
507 .dev_stop = eth_dev_stop,
508 .dev_close = eth_dev_close,
509 .dev_configure = eth_dev_configure,
510 .dev_infos_get = eth_dev_info,
511 .mtu_set = eth_dev_mtu_set,
512 .promiscuous_enable = eth_dev_promiscuous_enable,
513 .promiscuous_disable = eth_dev_promiscuous_disable,
514 .rx_queue_setup = eth_rx_queue_setup,
515 .tx_queue_setup = eth_tx_queue_setup,
516 .rx_queue_release = eth_queue_release,
517 .tx_queue_release = eth_queue_release,
518 .link_update = eth_link_update,
519 .stats_get = eth_stats_get,
520 .stats_reset = eth_stats_reset,
524 * Opens an AF_PACKET socket
527 open_packet_iface(const char *key __rte_unused,
528 const char *value __rte_unused,
531 int *sockfd = extra_args;
533 /* Open an AF_PACKET socket... */
534 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
536 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
543 static struct rte_vdev_driver pmd_af_packet_drv;
546 rte_pmd_init_internals(struct rte_vdev_device *dev,
548 const unsigned nb_queues,
549 unsigned int blocksize,
550 unsigned int blockcnt,
551 unsigned int framesize,
552 unsigned int framecnt,
553 struct pmd_internals **internals,
554 struct rte_eth_dev **eth_dev,
555 struct rte_kvargs *kvlist)
557 const char *name = rte_vdev_device_name(dev);
558 const unsigned int numa_node = dev->device.numa_node;
559 struct rte_eth_dev_data *data = NULL;
560 struct rte_kvargs_pair *pair = NULL;
564 struct sockaddr_ll sockaddr;
565 struct tpacket_req *req;
566 struct pkt_rx_queue *rx_queue;
567 struct pkt_tx_queue *tx_queue;
568 int rc, tpver, discard;
570 unsigned int i, q, rdsize;
571 #if defined(PACKET_FANOUT)
574 #if defined(PACKET_QDISC_BYPASS)
578 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
579 pair = &kvlist->pairs[k_idx];
580 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
585 "%s: no interface specified for AF_PACKET ethdev\n",
591 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
595 * now do all data allocation - for eth_dev structure, dummy pci driver
596 * and internal (private) data
598 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
602 *internals = rte_zmalloc_socket(name, sizeof(**internals),
604 if (*internals == NULL)
607 for (q = 0; q < nb_queues; q++) {
608 (*internals)->rx_queue[q].map = MAP_FAILED;
609 (*internals)->tx_queue[q].map = MAP_FAILED;
612 req = &((*internals)->req);
614 req->tp_block_size = blocksize;
615 req->tp_block_nr = blockcnt;
616 req->tp_frame_size = framesize;
617 req->tp_frame_nr = framecnt;
619 ifnamelen = strlen(pair->value);
620 if (ifnamelen < sizeof(ifr.ifr_name)) {
621 memcpy(ifr.ifr_name, pair->value, ifnamelen);
622 ifr.ifr_name[ifnamelen] = '\0';
625 "%s: I/F name too long (%s)\n",
629 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
631 "%s: ioctl failed (SIOCGIFINDEX)\n",
635 (*internals)->if_name = strdup(pair->value);
636 if ((*internals)->if_name == NULL)
638 (*internals)->if_index = ifr.ifr_ifindex;
640 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
642 "%s: ioctl failed (SIOCGIFHWADDR)\n",
646 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
648 memset(&sockaddr, 0, sizeof(sockaddr));
649 sockaddr.sll_family = AF_PACKET;
650 sockaddr.sll_protocol = htons(ETH_P_ALL);
651 sockaddr.sll_ifindex = (*internals)->if_index;
653 #if defined(PACKET_FANOUT)
654 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
655 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
656 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
657 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
661 for (q = 0; q < nb_queues; q++) {
662 /* Open an AF_PACKET socket for this queue... */
663 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
666 "%s: could not open AF_PACKET socket\n",
672 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
673 &tpver, sizeof(tpver));
676 "%s: could not set PACKET_VERSION on AF_PACKET "
677 "socket for %s\n", name, pair->value);
682 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
683 &discard, sizeof(discard));
686 "%s: could not set PACKET_LOSS on "
687 "AF_PACKET socket for %s\n", name, pair->value);
691 #if defined(PACKET_QDISC_BYPASS)
693 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
694 &bypass, sizeof(bypass));
697 "%s: could not set PACKET_QDISC_BYPASS "
698 "on AF_PACKET socket for %s\n", name,
704 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
707 "%s: could not set PACKET_RX_RING on AF_PACKET "
708 "socket for %s\n", name, pair->value);
712 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
715 "%s: could not set PACKET_TX_RING on AF_PACKET "
716 "socket for %s\n", name, pair->value);
720 rx_queue = &((*internals)->rx_queue[q]);
721 rx_queue->framecount = req->tp_frame_nr;
723 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
724 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
726 if (rx_queue->map == MAP_FAILED) {
728 "%s: call to mmap failed on AF_PACKET socket for %s\n",
733 /* rdsize is same for both Tx and Rx */
734 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
736 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
737 if (rx_queue->rd == NULL)
739 for (i = 0; i < req->tp_frame_nr; ++i) {
740 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
741 rx_queue->rd[i].iov_len = req->tp_frame_size;
743 rx_queue->sockfd = qsockfd;
745 tx_queue = &((*internals)->tx_queue[q]);
746 tx_queue->framecount = req->tp_frame_nr;
747 tx_queue->frame_data_size = req->tp_frame_size;
748 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
749 sizeof(struct sockaddr_ll);
751 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
753 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
754 if (tx_queue->rd == NULL)
756 for (i = 0; i < req->tp_frame_nr; ++i) {
757 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
758 tx_queue->rd[i].iov_len = req->tp_frame_size;
760 tx_queue->sockfd = qsockfd;
762 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
765 "%s: could not bind AF_PACKET socket to %s\n",
770 #if defined(PACKET_FANOUT)
771 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
772 &fanout_arg, sizeof(fanout_arg));
775 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
776 "for %s\n", name, pair->value);
782 /* reserve an ethdev entry */
783 *eth_dev = rte_eth_vdev_allocate(dev, 0);
784 if (*eth_dev == NULL)
788 * now put it all together
789 * - store queue data in internals,
790 * - store numa_node in eth_dev
791 * - point eth_dev_data to internals
792 * - and point eth_dev structure to new eth_dev_data structure
795 (*internals)->nb_queues = nb_queues;
797 rte_memcpy(data, (*eth_dev)->data, sizeof(*data));
798 data->dev_private = *internals;
799 data->nb_rx_queues = (uint16_t)nb_queues;
800 data->nb_tx_queues = (uint16_t)nb_queues;
801 data->dev_link = pmd_link;
802 data->mac_addrs = &(*internals)->eth_addr;
804 (*eth_dev)->data = data;
805 (*eth_dev)->dev_ops = &ops;
806 (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
813 for (q = 0; q < nb_queues; q++) {
814 munmap((*internals)->rx_queue[q].map,
815 2 * req->tp_block_size * req->tp_block_nr);
817 rte_free((*internals)->rx_queue[q].rd);
818 rte_free((*internals)->tx_queue[q].rd);
819 if (((*internals)->rx_queue[q].sockfd != 0) &&
820 ((*internals)->rx_queue[q].sockfd != qsockfd))
821 close((*internals)->rx_queue[q].sockfd);
823 free((*internals)->if_name);
824 rte_free(*internals);
831 rte_eth_from_packet(struct rte_vdev_device *dev,
833 struct rte_kvargs *kvlist)
835 const char *name = rte_vdev_device_name(dev);
836 struct pmd_internals *internals = NULL;
837 struct rte_eth_dev *eth_dev = NULL;
838 struct rte_kvargs_pair *pair = NULL;
840 unsigned int blockcount;
841 unsigned int blocksize = DFLT_BLOCK_SIZE;
842 unsigned int framesize = DFLT_FRAME_SIZE;
843 unsigned int framecount = DFLT_FRAME_COUNT;
844 unsigned int qpairs = 1;
846 /* do some parameter checking */
851 * Walk arguments for configurable settings
853 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
854 pair = &kvlist->pairs[k_idx];
855 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
856 qpairs = atoi(pair->value);
858 qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
860 "%s: invalid qpairs value\n",
866 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
867 blocksize = atoi(pair->value);
870 "%s: invalid blocksize value\n",
876 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
877 framesize = atoi(pair->value);
880 "%s: invalid framesize value\n",
886 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
887 framecount = atoi(pair->value);
890 "%s: invalid framecount value\n",
898 if (framesize > blocksize) {
900 "%s: AF_PACKET MMAP frame size exceeds block size!\n",
905 blockcount = framecount / (blocksize / framesize);
908 "%s: invalid AF_PACKET MMAP parameters\n", name);
912 RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
913 RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
914 RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
915 RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
916 RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
918 if (rte_pmd_init_internals(dev, *sockfd, qpairs,
919 blocksize, blockcount,
920 framesize, framecount,
921 &internals, ð_dev,
925 eth_dev->rx_pkt_burst = eth_af_packet_rx;
926 eth_dev->tx_pkt_burst = eth_af_packet_tx;
932 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
935 struct rte_kvargs *kvlist;
938 RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n",
939 rte_vdev_device_name(dev));
941 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
942 if (kvlist == NULL) {
948 * If iface argument is passed we open the NICs and use them for
951 if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
953 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
954 &open_packet_iface, &sockfd);
959 if (dev->device.numa_node == SOCKET_ID_ANY)
960 dev->device.numa_node = rte_socket_id();
962 ret = rte_eth_from_packet(dev, &sockfd, kvlist);
963 close(sockfd); /* no longer needed */
966 rte_kvargs_free(kvlist);
971 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
973 struct rte_eth_dev *eth_dev = NULL;
974 struct pmd_internals *internals;
977 RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
983 /* find the ethdev entry */
984 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
988 internals = eth_dev->data->dev_private;
989 for (q = 0; q < internals->nb_queues; q++) {
990 rte_free(internals->rx_queue[q].rd);
991 rte_free(internals->tx_queue[q].rd);
993 free(internals->if_name);
995 rte_free(eth_dev->data->dev_private);
996 rte_free(eth_dev->data);
998 rte_eth_dev_release_port(eth_dev);
1003 static struct rte_vdev_driver pmd_af_packet_drv = {
1004 .probe = rte_pmd_af_packet_probe,
1005 .remove = rte_pmd_af_packet_remove,
1008 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1009 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1010 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,