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_malloc.h>
42 #include <rte_kvargs.h>
45 #include <linux/if_ether.h>
46 #include <linux/if_packet.h>
47 #include <arpa/inet.h>
49 #include <sys/types.h>
50 #include <sys/socket.h>
51 #include <sys/ioctl.h>
56 #define ETH_AF_PACKET_IFACE_ARG "iface"
57 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
58 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
59 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
60 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
62 #define DFLT_BLOCK_SIZE (1 << 12)
63 #define DFLT_FRAME_SIZE (1 << 11)
64 #define DFLT_FRAME_COUNT (1 << 9)
66 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
73 unsigned int framecount;
74 unsigned int framenum;
76 struct rte_mempool *mb_pool;
79 volatile unsigned long rx_pkts;
80 volatile unsigned long err_pkts;
81 volatile unsigned long rx_bytes;
86 unsigned int frame_data_size;
90 unsigned int framecount;
91 unsigned int framenum;
93 volatile unsigned long tx_pkts;
94 volatile unsigned long err_pkts;
95 volatile unsigned long tx_bytes;
98 struct pmd_internals {
103 struct ether_addr eth_addr;
105 struct tpacket_req req;
107 struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
108 struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
111 static const char *valid_arguments[] = {
112 ETH_AF_PACKET_IFACE_ARG,
113 ETH_AF_PACKET_NUM_Q_ARG,
114 ETH_AF_PACKET_BLOCKSIZE_ARG,
115 ETH_AF_PACKET_FRAMESIZE_ARG,
116 ETH_AF_PACKET_FRAMECOUNT_ARG,
120 static struct rte_eth_link pmd_link = {
121 .link_speed = ETH_SPEED_NUM_10G,
122 .link_duplex = ETH_LINK_FULL_DUPLEX,
123 .link_status = ETH_LINK_DOWN,
124 .link_autoneg = ETH_LINK_SPEED_AUTONEG
128 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
131 struct tpacket2_hdr *ppd;
132 struct rte_mbuf *mbuf;
134 struct pkt_rx_queue *pkt_q = queue;
136 unsigned long num_rx_bytes = 0;
137 unsigned int framecount, framenum;
139 if (unlikely(nb_pkts == 0))
143 * Reads the given number of packets from the AF_PACKET socket one by
144 * one and copies the packet data into a newly allocated mbuf.
146 framecount = pkt_q->framecount;
147 framenum = pkt_q->framenum;
148 for (i = 0; i < nb_pkts; i++) {
149 /* point at the next incoming frame */
150 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
151 if ((ppd->tp_status & TP_STATUS_USER) == 0)
154 /* allocate the next mbuf */
155 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
156 if (unlikely(mbuf == NULL))
159 /* packet will fit in the mbuf, go ahead and receive it */
160 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
161 pbuf = (uint8_t *) ppd + ppd->tp_mac;
162 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
164 /* release incoming frame and advance ring buffer */
165 ppd->tp_status = TP_STATUS_KERNEL;
166 if (++framenum >= framecount)
168 mbuf->port = pkt_q->in_port;
170 /* account for the receive frame */
173 num_rx_bytes += mbuf->pkt_len;
175 pkt_q->framenum = framenum;
176 pkt_q->rx_pkts += num_rx;
177 pkt_q->rx_bytes += num_rx_bytes;
182 * Callback to handle sending packets through a real NIC.
185 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
187 struct tpacket2_hdr *ppd;
188 struct rte_mbuf *mbuf;
190 unsigned int framecount, framenum;
192 struct pkt_tx_queue *pkt_q = queue;
194 unsigned long num_tx_bytes = 0;
197 if (unlikely(nb_pkts == 0))
200 memset(&pfd, 0, sizeof(pfd));
201 pfd.fd = pkt_q->sockfd;
202 pfd.events = POLLOUT;
205 framecount = pkt_q->framecount;
206 framenum = pkt_q->framenum;
207 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
208 for (i = 0; i < nb_pkts; i++) {
211 /* drop oversized packets */
212 if (rte_pktmbuf_data_len(mbuf) > pkt_q->frame_data_size) {
213 rte_pktmbuf_free(mbuf);
217 /* point at the next incoming frame */
218 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
219 (poll(&pfd, 1, -1) < 0))
222 /* copy the tx frame data */
223 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
224 sizeof(struct sockaddr_ll);
225 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
226 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
228 /* release incoming frame and advance ring buffer */
229 ppd->tp_status = TP_STATUS_SEND_REQUEST;
230 if (++framenum >= framecount)
232 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
235 num_tx_bytes += mbuf->pkt_len;
236 rte_pktmbuf_free(mbuf);
239 /* kick-off transmits */
240 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)
241 num_tx = 0; /* error sending -- no packets transmitted */
243 pkt_q->framenum = framenum;
244 pkt_q->tx_pkts += num_tx;
245 pkt_q->err_pkts += i - num_tx;
246 pkt_q->tx_bytes += num_tx_bytes;
251 eth_dev_start(struct rte_eth_dev *dev)
253 dev->data->dev_link.link_status = ETH_LINK_UP;
258 * This function gets called when the current port gets stopped.
261 eth_dev_stop(struct rte_eth_dev *dev)
265 struct pmd_internals *internals = dev->data->dev_private;
267 for (i = 0; i < internals->nb_queues; i++) {
268 sockfd = internals->rx_queue[i].sockfd;
272 /* Prevent use after free in case tx fd == rx fd */
273 if (sockfd != internals->tx_queue[i].sockfd) {
274 sockfd = internals->tx_queue[i].sockfd;
279 internals->rx_queue[i].sockfd = -1;
280 internals->tx_queue[i].sockfd = -1;
283 dev->data->dev_link.link_status = ETH_LINK_DOWN;
287 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
293 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
295 struct pmd_internals *internals = dev->data->dev_private;
297 dev_info->if_index = internals->if_index;
298 dev_info->max_mac_addrs = 1;
299 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
300 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
301 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
302 dev_info->min_rx_bufsize = 0;
306 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
309 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
310 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
311 const struct pmd_internals *internal = dev->data->dev_private;
313 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
314 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
315 for (i = 0; i < imax; i++) {
316 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
317 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
318 rx_total += igb_stats->q_ipackets[i];
319 rx_bytes_total += igb_stats->q_ibytes[i];
322 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
323 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
324 for (i = 0; i < imax; i++) {
325 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
326 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
327 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
328 tx_total += igb_stats->q_opackets[i];
329 tx_err_total += igb_stats->q_errors[i];
330 tx_bytes_total += igb_stats->q_obytes[i];
333 igb_stats->ipackets = rx_total;
334 igb_stats->ibytes = rx_bytes_total;
335 igb_stats->opackets = tx_total;
336 igb_stats->oerrors = tx_err_total;
337 igb_stats->obytes = tx_bytes_total;
341 eth_stats_reset(struct rte_eth_dev *dev)
344 struct pmd_internals *internal = dev->data->dev_private;
346 for (i = 0; i < internal->nb_queues; i++) {
347 internal->rx_queue[i].rx_pkts = 0;
348 internal->rx_queue[i].rx_bytes = 0;
351 for (i = 0; i < internal->nb_queues; i++) {
352 internal->tx_queue[i].tx_pkts = 0;
353 internal->tx_queue[i].err_pkts = 0;
354 internal->tx_queue[i].tx_bytes = 0;
359 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
364 eth_queue_release(void *q __rte_unused)
369 eth_link_update(struct rte_eth_dev *dev __rte_unused,
370 int wait_to_complete __rte_unused)
376 eth_rx_queue_setup(struct rte_eth_dev *dev,
377 uint16_t rx_queue_id,
378 uint16_t nb_rx_desc __rte_unused,
379 unsigned int socket_id __rte_unused,
380 const struct rte_eth_rxconf *rx_conf __rte_unused,
381 struct rte_mempool *mb_pool)
383 struct pmd_internals *internals = dev->data->dev_private;
384 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
385 unsigned int buf_size, data_size;
387 pkt_q->mb_pool = mb_pool;
389 /* Now get the space available for data in the mbuf */
390 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
391 RTE_PKTMBUF_HEADROOM;
392 data_size = internals->req.tp_frame_size;
393 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
395 if (data_size > buf_size) {
397 "%s: %d bytes will not fit in mbuf (%d bytes)\n",
398 dev->data->name, data_size, buf_size);
402 dev->data->rx_queues[rx_queue_id] = pkt_q;
403 pkt_q->in_port = dev->data->port_id;
409 eth_tx_queue_setup(struct rte_eth_dev *dev,
410 uint16_t tx_queue_id,
411 uint16_t nb_tx_desc __rte_unused,
412 unsigned int socket_id __rte_unused,
413 const struct rte_eth_txconf *tx_conf __rte_unused)
416 struct pmd_internals *internals = dev->data->dev_private;
418 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
423 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
425 struct pmd_internals *internals = dev->data->dev_private;
426 struct ifreq ifr = { .ifr_mtu = mtu };
429 unsigned int data_size = internals->req.tp_frame_size -
431 sizeof(struct sockaddr_ll);
436 s = socket(PF_INET, SOCK_DGRAM, 0);
440 strncpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
441 ret = ioctl(s, SIOCSIFMTU, &ifr);
451 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
456 s = socket(PF_INET, SOCK_DGRAM, 0);
460 strncpy(ifr.ifr_name, if_name, IFNAMSIZ);
461 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
463 ifr.ifr_flags &= mask;
464 ifr.ifr_flags |= flags;
465 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
472 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
474 struct pmd_internals *internals = dev->data->dev_private;
476 eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
480 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
482 struct pmd_internals *internals = dev->data->dev_private;
484 eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
487 static const struct eth_dev_ops ops = {
488 .dev_start = eth_dev_start,
489 .dev_stop = eth_dev_stop,
490 .dev_close = eth_dev_close,
491 .dev_configure = eth_dev_configure,
492 .dev_infos_get = eth_dev_info,
493 .mtu_set = eth_dev_mtu_set,
494 .promiscuous_enable = eth_dev_promiscuous_enable,
495 .promiscuous_disable = eth_dev_promiscuous_disable,
496 .rx_queue_setup = eth_rx_queue_setup,
497 .tx_queue_setup = eth_tx_queue_setup,
498 .rx_queue_release = eth_queue_release,
499 .tx_queue_release = eth_queue_release,
500 .link_update = eth_link_update,
501 .stats_get = eth_stats_get,
502 .stats_reset = eth_stats_reset,
506 * Opens an AF_PACKET socket
509 open_packet_iface(const char *key __rte_unused,
510 const char *value __rte_unused,
513 int *sockfd = extra_args;
515 /* Open an AF_PACKET socket... */
516 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
518 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
525 static struct rte_vdev_driver pmd_af_packet_drv;
528 rte_pmd_init_internals(const char *name,
530 const unsigned nb_queues,
531 unsigned int blocksize,
532 unsigned int blockcnt,
533 unsigned int framesize,
534 unsigned int framecnt,
535 const unsigned numa_node,
536 struct pmd_internals **internals,
537 struct rte_eth_dev **eth_dev,
538 struct rte_kvargs *kvlist)
540 struct rte_eth_dev_data *data = NULL;
541 struct rte_kvargs_pair *pair = NULL;
545 struct sockaddr_ll sockaddr;
546 struct tpacket_req *req;
547 struct pkt_rx_queue *rx_queue;
548 struct pkt_tx_queue *tx_queue;
549 int rc, tpver, discard;
551 unsigned int i, q, rdsize;
552 int fanout_arg __rte_unused, bypass __rte_unused;
554 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
555 pair = &kvlist->pairs[k_idx];
556 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
561 "%s: no interface specified for AF_PACKET ethdev\n",
567 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
571 * now do all data allocation - for eth_dev structure, dummy pci driver
572 * and internal (private) data
574 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
578 *internals = rte_zmalloc_socket(name, sizeof(**internals),
580 if (*internals == NULL)
583 for (q = 0; q < nb_queues; q++) {
584 (*internals)->rx_queue[q].map = MAP_FAILED;
585 (*internals)->tx_queue[q].map = MAP_FAILED;
588 req = &((*internals)->req);
590 req->tp_block_size = blocksize;
591 req->tp_block_nr = blockcnt;
592 req->tp_frame_size = framesize;
593 req->tp_frame_nr = framecnt;
595 ifnamelen = strlen(pair->value);
596 if (ifnamelen < sizeof(ifr.ifr_name)) {
597 memcpy(ifr.ifr_name, pair->value, ifnamelen);
598 ifr.ifr_name[ifnamelen] = '\0';
601 "%s: I/F name too long (%s)\n",
605 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
607 "%s: ioctl failed (SIOCGIFINDEX)\n",
611 (*internals)->if_name = strdup(pair->value);
612 (*internals)->if_index = ifr.ifr_ifindex;
614 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
616 "%s: ioctl failed (SIOCGIFHWADDR)\n",
620 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
622 memset(&sockaddr, 0, sizeof(sockaddr));
623 sockaddr.sll_family = AF_PACKET;
624 sockaddr.sll_protocol = htons(ETH_P_ALL);
625 sockaddr.sll_ifindex = (*internals)->if_index;
627 #if defined(PACKET_FANOUT)
628 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
629 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
630 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
631 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
635 for (q = 0; q < nb_queues; q++) {
636 /* Open an AF_PACKET socket for this queue... */
637 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
640 "%s: could not open AF_PACKET socket\n",
646 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
647 &tpver, sizeof(tpver));
650 "%s: could not set PACKET_VERSION on AF_PACKET "
651 "socket for %s\n", name, pair->value);
656 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
657 &discard, sizeof(discard));
660 "%s: could not set PACKET_LOSS on "
661 "AF_PACKET socket for %s\n", name, pair->value);
665 #if defined(PACKET_QDISC_BYPASS)
667 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
668 &bypass, sizeof(bypass));
671 "%s: could not set PACKET_QDISC_BYPASS "
672 "on AF_PACKET socket for %s\n", name,
678 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
681 "%s: could not set PACKET_RX_RING on AF_PACKET "
682 "socket for %s\n", name, pair->value);
686 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
689 "%s: could not set PACKET_TX_RING on AF_PACKET "
690 "socket for %s\n", name, pair->value);
694 rx_queue = &((*internals)->rx_queue[q]);
695 rx_queue->framecount = req->tp_frame_nr;
697 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
698 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
700 if (rx_queue->map == MAP_FAILED) {
702 "%s: call to mmap failed on AF_PACKET socket for %s\n",
707 /* rdsize is same for both Tx and Rx */
708 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
710 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
711 if (rx_queue->rd == NULL)
713 for (i = 0; i < req->tp_frame_nr; ++i) {
714 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
715 rx_queue->rd[i].iov_len = req->tp_frame_size;
717 rx_queue->sockfd = qsockfd;
719 tx_queue = &((*internals)->tx_queue[q]);
720 tx_queue->framecount = req->tp_frame_nr;
721 tx_queue->frame_data_size = req->tp_frame_size;
722 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
723 sizeof(struct sockaddr_ll);
725 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
727 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
728 if (tx_queue->rd == NULL)
730 for (i = 0; i < req->tp_frame_nr; ++i) {
731 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
732 tx_queue->rd[i].iov_len = req->tp_frame_size;
734 tx_queue->sockfd = qsockfd;
736 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
739 "%s: could not bind AF_PACKET socket to %s\n",
744 #if defined(PACKET_FANOUT)
745 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
746 &fanout_arg, sizeof(fanout_arg));
749 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
750 "for %s\n", name, pair->value);
756 /* reserve an ethdev entry */
757 *eth_dev = rte_eth_dev_allocate(name);
758 if (*eth_dev == NULL)
762 * now put it all together
763 * - store queue data in internals,
764 * - store numa_node in eth_dev
765 * - point eth_dev_data to internals
766 * - and point eth_dev structure to new eth_dev_data structure
769 (*internals)->nb_queues = nb_queues;
771 data->dev_private = *internals;
772 data->port_id = (*eth_dev)->data->port_id;
773 data->nb_rx_queues = (uint16_t)nb_queues;
774 data->nb_tx_queues = (uint16_t)nb_queues;
775 data->dev_link = pmd_link;
776 data->mac_addrs = &(*internals)->eth_addr;
778 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
780 (*eth_dev)->data = data;
781 (*eth_dev)->dev_ops = &ops;
782 (*eth_dev)->driver = NULL;
783 (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
784 (*eth_dev)->data->drv_name = pmd_af_packet_drv.driver.name;
785 (*eth_dev)->data->kdrv = RTE_KDRV_NONE;
786 (*eth_dev)->data->numa_node = numa_node;
793 for (q = 0; q < nb_queues; q++) {
794 munmap((*internals)->rx_queue[q].map,
795 2 * req->tp_block_size * req->tp_block_nr);
797 rte_free((*internals)->rx_queue[q].rd);
798 rte_free((*internals)->tx_queue[q].rd);
799 if (((*internals)->rx_queue[q].sockfd != 0) &&
800 ((*internals)->rx_queue[q].sockfd != qsockfd))
801 close((*internals)->rx_queue[q].sockfd);
803 free((*internals)->if_name);
804 rte_free(*internals);
811 rte_eth_from_packet(const char *name,
813 const unsigned numa_node,
814 struct rte_kvargs *kvlist)
816 struct pmd_internals *internals = NULL;
817 struct rte_eth_dev *eth_dev = NULL;
818 struct rte_kvargs_pair *pair = NULL;
820 unsigned int blockcount;
821 unsigned int blocksize = DFLT_BLOCK_SIZE;
822 unsigned int framesize = DFLT_FRAME_SIZE;
823 unsigned int framecount = DFLT_FRAME_COUNT;
824 unsigned int qpairs = 1;
826 /* do some parameter checking */
831 * Walk arguments for configurable settings
833 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
834 pair = &kvlist->pairs[k_idx];
835 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
836 qpairs = atoi(pair->value);
838 qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
840 "%s: invalid qpairs value\n",
846 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
847 blocksize = atoi(pair->value);
850 "%s: invalid blocksize value\n",
856 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
857 framesize = atoi(pair->value);
860 "%s: invalid framesize value\n",
866 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
867 framecount = atoi(pair->value);
870 "%s: invalid framecount value\n",
878 if (framesize > blocksize) {
880 "%s: AF_PACKET MMAP frame size exceeds block size!\n",
885 blockcount = framecount / (blocksize / framesize);
888 "%s: invalid AF_PACKET MMAP parameters\n", name);
892 RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
893 RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
894 RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
895 RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
896 RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
898 if (rte_pmd_init_internals(name, *sockfd, qpairs,
899 blocksize, blockcount,
900 framesize, framecount,
901 numa_node, &internals, ð_dev,
905 eth_dev->rx_pkt_burst = eth_af_packet_rx;
906 eth_dev->tx_pkt_burst = eth_af_packet_tx;
912 rte_pmd_af_packet_probe(const char *name, const char *params)
916 struct rte_kvargs *kvlist;
919 RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
921 numa_node = rte_socket_id();
923 kvlist = rte_kvargs_parse(params, valid_arguments);
924 if (kvlist == NULL) {
930 * If iface argument is passed we open the NICs and use them for
933 if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
935 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
936 &open_packet_iface, &sockfd);
941 ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
942 close(sockfd); /* no longer needed */
945 rte_kvargs_free(kvlist);
950 rte_pmd_af_packet_remove(const char *name)
952 struct rte_eth_dev *eth_dev = NULL;
953 struct pmd_internals *internals;
956 RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
962 /* find the ethdev entry */
963 eth_dev = rte_eth_dev_allocated(name);
967 internals = eth_dev->data->dev_private;
968 for (q = 0; q < internals->nb_queues; q++) {
969 rte_free(internals->rx_queue[q].rd);
970 rte_free(internals->tx_queue[q].rd);
972 free(internals->if_name);
974 rte_free(eth_dev->data->dev_private);
975 rte_free(eth_dev->data);
977 rte_eth_dev_release_port(eth_dev);
982 static struct rte_vdev_driver pmd_af_packet_drv = {
983 .probe = rte_pmd_af_packet_probe,
984 .remove = rte_pmd_af_packet_remove,
987 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
988 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
989 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,