1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
3 * Originally based upon librte_pmd_pcap code:
4 * Copyright(c) 2010-2015 Intel Corporation.
5 * Copyright(c) 2014 6WIND S.A.
9 #include <rte_string_fns.h>
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_vdev.h>
13 #include <rte_malloc.h>
14 #include <rte_kvargs.h>
15 #include <rte_bus_vdev.h>
18 #include <linux/if_ether.h>
19 #include <linux/if_packet.h>
20 #include <arpa/inet.h>
22 #include <net/if_arp.h>
23 #include <sys/types.h>
24 #include <sys/socket.h>
25 #include <sys/ioctl.h>
31 #define ETH_AF_PACKET_IFACE_ARG "iface"
32 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
33 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
34 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
35 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
36 #define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
38 #define DFLT_FRAME_SIZE (1 << 11)
39 #define DFLT_FRAME_COUNT (1 << 9)
46 unsigned int framecount;
47 unsigned int framenum;
49 struct rte_mempool *mb_pool;
52 volatile unsigned long rx_pkts;
53 volatile unsigned long rx_bytes;
58 unsigned int frame_data_size;
62 unsigned int framecount;
63 unsigned int framenum;
65 volatile unsigned long tx_pkts;
66 volatile unsigned long err_pkts;
67 volatile unsigned long tx_bytes;
70 struct pmd_internals {
75 struct rte_ether_addr eth_addr;
77 struct tpacket_req req;
79 struct pkt_rx_queue *rx_queue;
80 struct pkt_tx_queue *tx_queue;
83 static const char *valid_arguments[] = {
84 ETH_AF_PACKET_IFACE_ARG,
85 ETH_AF_PACKET_NUM_Q_ARG,
86 ETH_AF_PACKET_BLOCKSIZE_ARG,
87 ETH_AF_PACKET_FRAMESIZE_ARG,
88 ETH_AF_PACKET_FRAMECOUNT_ARG,
89 ETH_AF_PACKET_QDISC_BYPASS_ARG,
93 static struct rte_eth_link pmd_link = {
94 .link_speed = ETH_SPEED_NUM_10G,
95 .link_duplex = ETH_LINK_FULL_DUPLEX,
96 .link_status = ETH_LINK_DOWN,
97 .link_autoneg = ETH_LINK_FIXED,
100 RTE_LOG_REGISTER(af_packet_logtype, pmd.net.packet, NOTICE);
102 #define PMD_LOG(level, fmt, args...) \
103 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
104 "%s(): " fmt "\n", __func__, ##args)
106 #define PMD_LOG_ERRNO(level, fmt, args...) \
107 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
108 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
111 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
114 struct tpacket2_hdr *ppd;
115 struct rte_mbuf *mbuf;
117 struct pkt_rx_queue *pkt_q = queue;
119 unsigned long num_rx_bytes = 0;
120 unsigned int framecount, framenum;
122 if (unlikely(nb_pkts == 0))
126 * Reads the given number of packets from the AF_PACKET socket one by
127 * one and copies the packet data into a newly allocated mbuf.
129 framecount = pkt_q->framecount;
130 framenum = pkt_q->framenum;
131 for (i = 0; i < nb_pkts; i++) {
132 /* point at the next incoming frame */
133 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
134 if ((ppd->tp_status & TP_STATUS_USER) == 0)
137 /* allocate the next mbuf */
138 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
139 if (unlikely(mbuf == NULL))
142 /* packet will fit in the mbuf, go ahead and receive it */
143 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
144 pbuf = (uint8_t *) ppd + ppd->tp_mac;
145 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
147 /* check for vlan info */
148 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
149 mbuf->vlan_tci = ppd->tp_vlan_tci;
150 mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
153 /* release incoming frame and advance ring buffer */
154 ppd->tp_status = TP_STATUS_KERNEL;
155 if (++framenum >= framecount)
157 mbuf->port = pkt_q->in_port;
159 /* account for the receive frame */
162 num_rx_bytes += mbuf->pkt_len;
164 pkt_q->framenum = framenum;
165 pkt_q->rx_pkts += num_rx;
166 pkt_q->rx_bytes += num_rx_bytes;
171 * Callback to handle sending packets through a real NIC.
174 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
176 struct tpacket2_hdr *ppd;
177 struct rte_mbuf *mbuf;
179 unsigned int framecount, framenum;
181 struct pkt_tx_queue *pkt_q = queue;
183 unsigned long num_tx_bytes = 0;
186 if (unlikely(nb_pkts == 0))
189 memset(&pfd, 0, sizeof(pfd));
190 pfd.fd = pkt_q->sockfd;
191 pfd.events = POLLOUT;
194 framecount = pkt_q->framecount;
195 framenum = pkt_q->framenum;
196 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
197 for (i = 0; i < nb_pkts; i++) {
200 /* drop oversized packets */
201 if (mbuf->pkt_len > pkt_q->frame_data_size) {
202 rte_pktmbuf_free(mbuf);
206 /* insert vlan info if necessary */
207 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
208 if (rte_vlan_insert(&mbuf)) {
209 rte_pktmbuf_free(mbuf);
214 /* point at the next incoming frame */
215 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
216 (poll(&pfd, 1, -1) < 0))
219 /* copy the tx frame data */
220 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
221 sizeof(struct sockaddr_ll);
223 struct rte_mbuf *tmp_mbuf = mbuf;
225 uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
226 memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
228 tmp_mbuf = tmp_mbuf->next;
231 ppd->tp_len = mbuf->pkt_len;
232 ppd->tp_snaplen = mbuf->pkt_len;
234 /* release incoming frame and advance ring buffer */
235 ppd->tp_status = TP_STATUS_SEND_REQUEST;
236 if (++framenum >= framecount)
238 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
241 num_tx_bytes += mbuf->pkt_len;
242 rte_pktmbuf_free(mbuf);
245 /* kick-off transmits */
246 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
247 errno != ENOBUFS && errno != EAGAIN) {
249 * In case of a ENOBUFS/EAGAIN error all of the enqueued
250 * packets will be considered successful even though only some
258 pkt_q->framenum = framenum;
259 pkt_q->tx_pkts += num_tx;
260 pkt_q->err_pkts += i - num_tx;
261 pkt_q->tx_bytes += num_tx_bytes;
266 eth_dev_start(struct rte_eth_dev *dev)
268 dev->data->dev_link.link_status = ETH_LINK_UP;
273 * This function gets called when the current port gets stopped.
276 eth_dev_stop(struct rte_eth_dev *dev)
280 struct pmd_internals *internals = dev->data->dev_private;
282 for (i = 0; i < internals->nb_queues; i++) {
283 sockfd = internals->rx_queue[i].sockfd;
287 /* Prevent use after free in case tx fd == rx fd */
288 if (sockfd != internals->tx_queue[i].sockfd) {
289 sockfd = internals->tx_queue[i].sockfd;
294 internals->rx_queue[i].sockfd = -1;
295 internals->tx_queue[i].sockfd = -1;
298 dev->data->dev_link.link_status = ETH_LINK_DOWN;
302 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
308 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
310 struct pmd_internals *internals = dev->data->dev_private;
312 dev_info->if_index = internals->if_index;
313 dev_info->max_mac_addrs = 1;
314 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
315 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
316 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
317 dev_info->min_rx_bufsize = 0;
318 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
319 DEV_TX_OFFLOAD_VLAN_INSERT;
325 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
328 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
329 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
330 const struct pmd_internals *internal = dev->data->dev_private;
332 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
333 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
334 for (i = 0; i < imax; i++) {
335 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
336 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
337 rx_total += igb_stats->q_ipackets[i];
338 rx_bytes_total += igb_stats->q_ibytes[i];
341 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
342 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
343 for (i = 0; i < imax; i++) {
344 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_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 += internal->tx_queue[i].err_pkts;
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;
360 eth_stats_reset(struct rte_eth_dev *dev)
363 struct pmd_internals *internal = dev->data->dev_private;
365 for (i = 0; i < internal->nb_queues; i++) {
366 internal->rx_queue[i].rx_pkts = 0;
367 internal->rx_queue[i].rx_bytes = 0;
370 for (i = 0; i < internal->nb_queues; i++) {
371 internal->tx_queue[i].tx_pkts = 0;
372 internal->tx_queue[i].err_pkts = 0;
373 internal->tx_queue[i].tx_bytes = 0;
380 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
385 eth_queue_release(void *q __rte_unused)
390 eth_link_update(struct rte_eth_dev *dev __rte_unused,
391 int wait_to_complete __rte_unused)
397 eth_rx_queue_setup(struct rte_eth_dev *dev,
398 uint16_t rx_queue_id,
399 uint16_t nb_rx_desc __rte_unused,
400 unsigned int socket_id __rte_unused,
401 const struct rte_eth_rxconf *rx_conf __rte_unused,
402 struct rte_mempool *mb_pool)
404 struct pmd_internals *internals = dev->data->dev_private;
405 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
406 unsigned int buf_size, data_size;
408 pkt_q->mb_pool = mb_pool;
410 /* Now get the space available for data in the mbuf */
411 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
412 RTE_PKTMBUF_HEADROOM;
413 data_size = internals->req.tp_frame_size;
414 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
416 if (data_size > buf_size) {
418 "%s: %d bytes will not fit in mbuf (%d bytes)",
419 dev->device->name, data_size, buf_size);
423 dev->data->rx_queues[rx_queue_id] = pkt_q;
424 pkt_q->in_port = dev->data->port_id;
430 eth_tx_queue_setup(struct rte_eth_dev *dev,
431 uint16_t tx_queue_id,
432 uint16_t nb_tx_desc __rte_unused,
433 unsigned int socket_id __rte_unused,
434 const struct rte_eth_txconf *tx_conf __rte_unused)
437 struct pmd_internals *internals = dev->data->dev_private;
439 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
444 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
446 struct pmd_internals *internals = dev->data->dev_private;
447 struct ifreq ifr = { .ifr_mtu = mtu };
450 unsigned int data_size = internals->req.tp_frame_size -
456 s = socket(PF_INET, SOCK_DGRAM, 0);
460 strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
461 ret = ioctl(s, SIOCSIFMTU, &ifr);
471 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
473 struct pmd_internals *internals = dev->data->dev_private;
474 struct ifreq ifr = { };
475 int sockfd = internals->rx_queue[0].sockfd;
479 PMD_LOG(ERR, "receive socket not found");
483 strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
484 ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
485 memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
486 ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
489 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
497 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
503 s = socket(PF_INET, SOCK_DGRAM, 0);
507 strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
508 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
512 ifr.ifr_flags &= mask;
513 ifr.ifr_flags |= flags;
514 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
524 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
526 struct pmd_internals *internals = dev->data->dev_private;
528 return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
532 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
534 struct pmd_internals *internals = dev->data->dev_private;
536 return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
539 static const struct eth_dev_ops ops = {
540 .dev_start = eth_dev_start,
541 .dev_stop = eth_dev_stop,
542 .dev_close = eth_dev_close,
543 .dev_configure = eth_dev_configure,
544 .dev_infos_get = eth_dev_info,
545 .mac_addr_set = eth_dev_macaddr_set,
546 .mtu_set = eth_dev_mtu_set,
547 .promiscuous_enable = eth_dev_promiscuous_enable,
548 .promiscuous_disable = eth_dev_promiscuous_disable,
549 .rx_queue_setup = eth_rx_queue_setup,
550 .tx_queue_setup = eth_tx_queue_setup,
551 .rx_queue_release = eth_queue_release,
552 .tx_queue_release = eth_queue_release,
553 .link_update = eth_link_update,
554 .stats_get = eth_stats_get,
555 .stats_reset = eth_stats_reset,
559 * Opens an AF_PACKET socket
562 open_packet_iface(const char *key __rte_unused,
563 const char *value __rte_unused,
566 int *sockfd = extra_args;
568 /* Open an AF_PACKET socket... */
569 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
571 PMD_LOG(ERR, "Could not open AF_PACKET socket");
579 rte_pmd_init_internals(struct rte_vdev_device *dev,
581 const unsigned nb_queues,
582 unsigned int blocksize,
583 unsigned int blockcnt,
584 unsigned int framesize,
585 unsigned int framecnt,
586 unsigned int qdisc_bypass,
587 struct pmd_internals **internals,
588 struct rte_eth_dev **eth_dev,
589 struct rte_kvargs *kvlist)
591 const char *name = rte_vdev_device_name(dev);
592 const unsigned int numa_node = dev->device.numa_node;
593 struct rte_eth_dev_data *data = NULL;
594 struct rte_kvargs_pair *pair = NULL;
598 struct sockaddr_ll sockaddr;
599 struct tpacket_req *req;
600 struct pkt_rx_queue *rx_queue;
601 struct pkt_tx_queue *tx_queue;
602 int rc, tpver, discard;
604 unsigned int i, q, rdsize;
605 #if defined(PACKET_FANOUT)
609 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
610 pair = &kvlist->pairs[k_idx];
611 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
616 "%s: no interface specified for AF_PACKET ethdev",
622 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
625 *internals = rte_zmalloc_socket(name, sizeof(**internals),
627 if (*internals == NULL)
631 (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
633 sizeof(struct pkt_rx_queue),
635 (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
637 sizeof(struct pkt_tx_queue),
639 if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
643 for (q = 0; q < nb_queues; q++) {
644 (*internals)->rx_queue[q].map = MAP_FAILED;
645 (*internals)->tx_queue[q].map = MAP_FAILED;
646 (*internals)->rx_queue[q].sockfd = -1;
647 (*internals)->tx_queue[q].sockfd = -1;
650 req = &((*internals)->req);
652 req->tp_block_size = blocksize;
653 req->tp_block_nr = blockcnt;
654 req->tp_frame_size = framesize;
655 req->tp_frame_nr = framecnt;
657 ifnamelen = strlen(pair->value);
658 if (ifnamelen < sizeof(ifr.ifr_name)) {
659 memcpy(ifr.ifr_name, pair->value, ifnamelen);
660 ifr.ifr_name[ifnamelen] = '\0';
663 "%s: I/F name too long (%s)",
667 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
668 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
671 (*internals)->if_name = strdup(pair->value);
672 if ((*internals)->if_name == NULL)
674 (*internals)->if_index = ifr.ifr_ifindex;
676 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
677 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
680 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
682 memset(&sockaddr, 0, sizeof(sockaddr));
683 sockaddr.sll_family = AF_PACKET;
684 sockaddr.sll_protocol = htons(ETH_P_ALL);
685 sockaddr.sll_ifindex = (*internals)->if_index;
687 #if defined(PACKET_FANOUT)
688 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
689 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
690 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
691 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
695 for (q = 0; q < nb_queues; q++) {
696 /* Open an AF_PACKET socket for this queue... */
697 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
700 "%s: could not open AF_PACKET socket",
706 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
707 &tpver, sizeof(tpver));
710 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
716 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
717 &discard, sizeof(discard));
720 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
725 #if defined(PACKET_QDISC_BYPASS)
726 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
727 &qdisc_bypass, sizeof(qdisc_bypass));
730 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
735 RTE_SET_USED(qdisc_bypass);
738 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
741 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
746 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
749 "%s: could not set PACKET_TX_RING on AF_PACKET "
750 "socket for %s", name, pair->value);
754 rx_queue = &((*internals)->rx_queue[q]);
755 rx_queue->framecount = req->tp_frame_nr;
757 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
758 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
760 if (rx_queue->map == MAP_FAILED) {
762 "%s: call to mmap failed on AF_PACKET socket for %s",
767 /* rdsize is same for both Tx and Rx */
768 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
770 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
771 if (rx_queue->rd == NULL)
773 for (i = 0; i < req->tp_frame_nr; ++i) {
774 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
775 rx_queue->rd[i].iov_len = req->tp_frame_size;
777 rx_queue->sockfd = qsockfd;
779 tx_queue = &((*internals)->tx_queue[q]);
780 tx_queue->framecount = req->tp_frame_nr;
781 tx_queue->frame_data_size = req->tp_frame_size;
782 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
783 sizeof(struct sockaddr_ll);
785 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
787 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
788 if (tx_queue->rd == NULL)
790 for (i = 0; i < req->tp_frame_nr; ++i) {
791 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
792 tx_queue->rd[i].iov_len = req->tp_frame_size;
794 tx_queue->sockfd = qsockfd;
796 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
799 "%s: could not bind AF_PACKET socket to %s",
804 #if defined(PACKET_FANOUT)
805 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
806 &fanout_arg, sizeof(fanout_arg));
809 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
816 /* reserve an ethdev entry */
817 *eth_dev = rte_eth_vdev_allocate(dev, 0);
818 if (*eth_dev == NULL)
822 * now put it all together
823 * - store queue data in internals,
824 * - store numa_node in eth_dev
825 * - point eth_dev_data to internals
826 * - and point eth_dev structure to new eth_dev_data structure
829 (*internals)->nb_queues = nb_queues;
831 data = (*eth_dev)->data;
832 data->dev_private = *internals;
833 data->nb_rx_queues = (uint16_t)nb_queues;
834 data->nb_tx_queues = (uint16_t)nb_queues;
835 data->dev_link = pmd_link;
836 data->mac_addrs = &(*internals)->eth_addr;
838 (*eth_dev)->dev_ops = &ops;
845 for (q = 0; q < nb_queues; q++) {
846 if ((*internals)->rx_queue[q].map != MAP_FAILED)
847 munmap((*internals)->rx_queue[q].map,
848 2 * req->tp_block_size * req->tp_block_nr);
850 rte_free((*internals)->rx_queue[q].rd);
851 rte_free((*internals)->tx_queue[q].rd);
852 if (((*internals)->rx_queue[q].sockfd >= 0) &&
853 ((*internals)->rx_queue[q].sockfd != qsockfd))
854 close((*internals)->rx_queue[q].sockfd);
857 rte_free((*internals)->rx_queue);
858 rte_free((*internals)->tx_queue);
859 free((*internals)->if_name);
860 rte_free(*internals);
865 rte_eth_from_packet(struct rte_vdev_device *dev,
867 struct rte_kvargs *kvlist)
869 const char *name = rte_vdev_device_name(dev);
870 struct pmd_internals *internals = NULL;
871 struct rte_eth_dev *eth_dev = NULL;
872 struct rte_kvargs_pair *pair = NULL;
874 unsigned int blockcount;
875 unsigned int blocksize;
876 unsigned int framesize = DFLT_FRAME_SIZE;
877 unsigned int framecount = DFLT_FRAME_COUNT;
878 unsigned int qpairs = 1;
879 unsigned int qdisc_bypass = 1;
881 /* do some parameter checking */
885 blocksize = getpagesize();
888 * Walk arguments for configurable settings
890 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
891 pair = &kvlist->pairs[k_idx];
892 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
893 qpairs = atoi(pair->value);
896 "%s: invalid qpairs value",
902 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
903 blocksize = atoi(pair->value);
906 "%s: invalid blocksize value",
912 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
913 framesize = atoi(pair->value);
916 "%s: invalid framesize value",
922 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
923 framecount = atoi(pair->value);
926 "%s: invalid framecount value",
932 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
933 qdisc_bypass = atoi(pair->value);
934 if (qdisc_bypass > 1) {
936 "%s: invalid bypass value",
944 if (framesize > blocksize) {
946 "%s: AF_PACKET MMAP frame size exceeds block size!",
951 blockcount = framecount / (blocksize / framesize);
954 "%s: invalid AF_PACKET MMAP parameters", name);
958 PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
959 PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
960 PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
961 PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
962 PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
964 if (rte_pmd_init_internals(dev, *sockfd, qpairs,
965 blocksize, blockcount,
966 framesize, framecount,
968 &internals, ð_dev,
972 eth_dev->rx_pkt_burst = eth_af_packet_rx;
973 eth_dev->tx_pkt_burst = eth_af_packet_tx;
975 rte_eth_dev_probing_finish(eth_dev);
980 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
983 struct rte_kvargs *kvlist;
985 struct rte_eth_dev *eth_dev;
986 const char *name = rte_vdev_device_name(dev);
988 PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
990 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
991 eth_dev = rte_eth_dev_attach_secondary(name);
993 PMD_LOG(ERR, "Failed to probe %s", name);
996 /* TODO: request info from primary to set up Rx and Tx */
997 eth_dev->dev_ops = &ops;
998 eth_dev->device = &dev->device;
999 rte_eth_dev_probing_finish(eth_dev);
1003 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1004 if (kvlist == NULL) {
1010 * If iface argument is passed we open the NICs and use them for
1013 if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1015 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1016 &open_packet_iface, &sockfd);
1021 if (dev->device.numa_node == SOCKET_ID_ANY)
1022 dev->device.numa_node = rte_socket_id();
1024 ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1025 close(sockfd); /* no longer needed */
1028 rte_kvargs_free(kvlist);
1033 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1035 struct rte_eth_dev *eth_dev = NULL;
1036 struct pmd_internals *internals;
1037 struct tpacket_req *req;
1040 PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
1046 /* find the ethdev entry */
1047 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1048 if (eth_dev == NULL)
1051 /* mac_addrs must not be freed alone because part of dev_private */
1052 eth_dev->data->mac_addrs = NULL;
1054 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1055 return rte_eth_dev_release_port(eth_dev);
1057 internals = eth_dev->data->dev_private;
1058 req = &internals->req;
1059 for (q = 0; q < internals->nb_queues; q++) {
1060 munmap(internals->rx_queue[q].map,
1061 2 * req->tp_block_size * req->tp_block_nr);
1062 rte_free(internals->rx_queue[q].rd);
1063 rte_free(internals->tx_queue[q].rd);
1065 free(internals->if_name);
1066 rte_free(internals->rx_queue);
1067 rte_free(internals->tx_queue);
1069 rte_eth_dev_release_port(eth_dev);
1074 static struct rte_vdev_driver pmd_af_packet_drv = {
1075 .probe = rte_pmd_af_packet_probe,
1076 .remove = rte_pmd_af_packet_remove,
1079 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1080 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1081 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1087 "qdisc_bypass=<0|1>");