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 <sys/types.h>
23 #include <sys/socket.h>
24 #include <sys/ioctl.h>
30 #define ETH_AF_PACKET_IFACE_ARG "iface"
31 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
32 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
33 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
34 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
35 #define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
37 #define DFLT_FRAME_SIZE (1 << 11)
38 #define DFLT_FRAME_COUNT (1 << 9)
40 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
47 unsigned int framecount;
48 unsigned int framenum;
50 struct rte_mempool *mb_pool;
53 volatile unsigned long rx_pkts;
54 volatile unsigned long err_pkts;
55 volatile unsigned long rx_bytes;
60 unsigned int frame_data_size;
64 unsigned int framecount;
65 unsigned int framenum;
67 volatile unsigned long tx_pkts;
68 volatile unsigned long err_pkts;
69 volatile unsigned long tx_bytes;
72 struct pmd_internals {
77 struct rte_ether_addr eth_addr;
79 struct tpacket_req req;
81 struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
82 struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
85 static const char *valid_arguments[] = {
86 ETH_AF_PACKET_IFACE_ARG,
87 ETH_AF_PACKET_NUM_Q_ARG,
88 ETH_AF_PACKET_BLOCKSIZE_ARG,
89 ETH_AF_PACKET_FRAMESIZE_ARG,
90 ETH_AF_PACKET_FRAMECOUNT_ARG,
91 ETH_AF_PACKET_QDISC_BYPASS_ARG,
95 static struct rte_eth_link pmd_link = {
96 .link_speed = ETH_SPEED_NUM_10G,
97 .link_duplex = ETH_LINK_FULL_DUPLEX,
98 .link_status = ETH_LINK_DOWN,
99 .link_autoneg = ETH_LINK_FIXED,
102 static int af_packet_logtype;
104 #define PMD_LOG(level, fmt, args...) \
105 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
106 "%s(): " fmt "\n", __func__, ##args)
108 #define PMD_LOG_ERRNO(level, fmt, args...) \
109 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
110 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
113 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
116 struct tpacket2_hdr *ppd;
117 struct rte_mbuf *mbuf;
119 struct pkt_rx_queue *pkt_q = queue;
121 unsigned long num_rx_bytes = 0;
122 unsigned int framecount, framenum;
124 if (unlikely(nb_pkts == 0))
128 * Reads the given number of packets from the AF_PACKET socket one by
129 * one and copies the packet data into a newly allocated mbuf.
131 framecount = pkt_q->framecount;
132 framenum = pkt_q->framenum;
133 for (i = 0; i < nb_pkts; i++) {
134 /* point at the next incoming frame */
135 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
136 if ((ppd->tp_status & TP_STATUS_USER) == 0)
139 /* allocate the next mbuf */
140 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
141 if (unlikely(mbuf == NULL))
144 /* packet will fit in the mbuf, go ahead and receive it */
145 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
146 pbuf = (uint8_t *) ppd + ppd->tp_mac;
147 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
149 /* check for vlan info */
150 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
151 mbuf->vlan_tci = ppd->tp_vlan_tci;
152 mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
155 /* release incoming frame and advance ring buffer */
156 ppd->tp_status = TP_STATUS_KERNEL;
157 if (++framenum >= framecount)
159 mbuf->port = pkt_q->in_port;
161 /* account for the receive frame */
164 num_rx_bytes += mbuf->pkt_len;
166 pkt_q->framenum = framenum;
167 pkt_q->rx_pkts += num_rx;
168 pkt_q->rx_bytes += num_rx_bytes;
173 * Callback to handle sending packets through a real NIC.
176 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
178 struct tpacket2_hdr *ppd;
179 struct rte_mbuf *mbuf;
181 unsigned int framecount, framenum;
183 struct pkt_tx_queue *pkt_q = queue;
185 unsigned long num_tx_bytes = 0;
188 if (unlikely(nb_pkts == 0))
191 memset(&pfd, 0, sizeof(pfd));
192 pfd.fd = pkt_q->sockfd;
193 pfd.events = POLLOUT;
196 framecount = pkt_q->framecount;
197 framenum = pkt_q->framenum;
198 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
199 for (i = 0; i < nb_pkts; i++) {
202 /* drop oversized packets */
203 if (mbuf->pkt_len > pkt_q->frame_data_size) {
204 rte_pktmbuf_free(mbuf);
208 /* insert vlan info if necessary */
209 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
210 if (rte_vlan_insert(&mbuf)) {
211 rte_pktmbuf_free(mbuf);
216 /* point at the next incoming frame */
217 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
218 (poll(&pfd, 1, -1) < 0))
221 /* copy the tx frame data */
222 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
223 sizeof(struct sockaddr_ll);
225 struct rte_mbuf *tmp_mbuf = mbuf;
227 uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
228 memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
230 tmp_mbuf = tmp_mbuf->next;
233 ppd->tp_len = mbuf->pkt_len;
234 ppd->tp_snaplen = mbuf->pkt_len;
236 /* release incoming frame and advance ring buffer */
237 ppd->tp_status = TP_STATUS_SEND_REQUEST;
238 if (++framenum >= framecount)
240 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
243 num_tx_bytes += mbuf->pkt_len;
244 rte_pktmbuf_free(mbuf);
247 /* kick-off transmits */
248 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
249 /* error sending -- no packets transmitted */
254 pkt_q->framenum = framenum;
255 pkt_q->tx_pkts += num_tx;
256 pkt_q->err_pkts += i - num_tx;
257 pkt_q->tx_bytes += num_tx_bytes;
262 eth_dev_start(struct rte_eth_dev *dev)
264 dev->data->dev_link.link_status = ETH_LINK_UP;
269 * This function gets called when the current port gets stopped.
272 eth_dev_stop(struct rte_eth_dev *dev)
276 struct pmd_internals *internals = dev->data->dev_private;
278 for (i = 0; i < internals->nb_queues; i++) {
279 sockfd = internals->rx_queue[i].sockfd;
283 /* Prevent use after free in case tx fd == rx fd */
284 if (sockfd != internals->tx_queue[i].sockfd) {
285 sockfd = internals->tx_queue[i].sockfd;
290 internals->rx_queue[i].sockfd = -1;
291 internals->tx_queue[i].sockfd = -1;
294 dev->data->dev_link.link_status = ETH_LINK_DOWN;
298 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
304 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
306 struct pmd_internals *internals = dev->data->dev_private;
308 dev_info->if_index = internals->if_index;
309 dev_info->max_mac_addrs = 1;
310 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
311 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
312 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
313 dev_info->min_rx_bufsize = 0;
317 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
320 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
321 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
322 const struct pmd_internals *internal = dev->data->dev_private;
324 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
325 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
326 for (i = 0; i < imax; i++) {
327 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
328 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
329 rx_total += igb_stats->q_ipackets[i];
330 rx_bytes_total += igb_stats->q_ibytes[i];
333 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
334 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
335 for (i = 0; i < imax; i++) {
336 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
337 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
338 tx_total += igb_stats->q_opackets[i];
339 tx_err_total += internal->tx_queue[i].err_pkts;
340 tx_bytes_total += igb_stats->q_obytes[i];
343 igb_stats->ipackets = rx_total;
344 igb_stats->ibytes = rx_bytes_total;
345 igb_stats->opackets = tx_total;
346 igb_stats->oerrors = tx_err_total;
347 igb_stats->obytes = tx_bytes_total;
352 eth_stats_reset(struct rte_eth_dev *dev)
355 struct pmd_internals *internal = dev->data->dev_private;
357 for (i = 0; i < internal->nb_queues; i++) {
358 internal->rx_queue[i].rx_pkts = 0;
359 internal->rx_queue[i].rx_bytes = 0;
362 for (i = 0; i < internal->nb_queues; i++) {
363 internal->tx_queue[i].tx_pkts = 0;
364 internal->tx_queue[i].err_pkts = 0;
365 internal->tx_queue[i].tx_bytes = 0;
370 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
375 eth_queue_release(void *q __rte_unused)
380 eth_link_update(struct rte_eth_dev *dev __rte_unused,
381 int wait_to_complete __rte_unused)
387 eth_rx_queue_setup(struct rte_eth_dev *dev,
388 uint16_t rx_queue_id,
389 uint16_t nb_rx_desc __rte_unused,
390 unsigned int socket_id __rte_unused,
391 const struct rte_eth_rxconf *rx_conf __rte_unused,
392 struct rte_mempool *mb_pool)
394 struct pmd_internals *internals = dev->data->dev_private;
395 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
396 unsigned int buf_size, data_size;
398 pkt_q->mb_pool = mb_pool;
400 /* Now get the space available for data in the mbuf */
401 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
402 RTE_PKTMBUF_HEADROOM;
403 data_size = internals->req.tp_frame_size;
404 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
406 if (data_size > buf_size) {
408 "%s: %d bytes will not fit in mbuf (%d bytes)",
409 dev->device->name, data_size, buf_size);
413 dev->data->rx_queues[rx_queue_id] = pkt_q;
414 pkt_q->in_port = dev->data->port_id;
420 eth_tx_queue_setup(struct rte_eth_dev *dev,
421 uint16_t tx_queue_id,
422 uint16_t nb_tx_desc __rte_unused,
423 unsigned int socket_id __rte_unused,
424 const struct rte_eth_txconf *tx_conf __rte_unused)
427 struct pmd_internals *internals = dev->data->dev_private;
429 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
434 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
436 struct pmd_internals *internals = dev->data->dev_private;
437 struct ifreq ifr = { .ifr_mtu = mtu };
440 unsigned int data_size = internals->req.tp_frame_size -
446 s = socket(PF_INET, SOCK_DGRAM, 0);
450 strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
451 ret = ioctl(s, SIOCSIFMTU, &ifr);
461 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
466 s = socket(PF_INET, SOCK_DGRAM, 0);
470 strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
471 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
473 ifr.ifr_flags &= mask;
474 ifr.ifr_flags |= flags;
475 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
482 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
484 struct pmd_internals *internals = dev->data->dev_private;
486 eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
490 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
492 struct pmd_internals *internals = dev->data->dev_private;
494 eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
497 static const struct eth_dev_ops ops = {
498 .dev_start = eth_dev_start,
499 .dev_stop = eth_dev_stop,
500 .dev_close = eth_dev_close,
501 .dev_configure = eth_dev_configure,
502 .dev_infos_get = eth_dev_info,
503 .mtu_set = eth_dev_mtu_set,
504 .promiscuous_enable = eth_dev_promiscuous_enable,
505 .promiscuous_disable = eth_dev_promiscuous_disable,
506 .rx_queue_setup = eth_rx_queue_setup,
507 .tx_queue_setup = eth_tx_queue_setup,
508 .rx_queue_release = eth_queue_release,
509 .tx_queue_release = eth_queue_release,
510 .link_update = eth_link_update,
511 .stats_get = eth_stats_get,
512 .stats_reset = eth_stats_reset,
516 * Opens an AF_PACKET socket
519 open_packet_iface(const char *key __rte_unused,
520 const char *value __rte_unused,
523 int *sockfd = extra_args;
525 /* Open an AF_PACKET socket... */
526 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
528 PMD_LOG(ERR, "Could not open AF_PACKET socket");
536 rte_pmd_init_internals(struct rte_vdev_device *dev,
538 const unsigned nb_queues,
539 unsigned int blocksize,
540 unsigned int blockcnt,
541 unsigned int framesize,
542 unsigned int framecnt,
543 unsigned int qdisc_bypass,
544 struct pmd_internals **internals,
545 struct rte_eth_dev **eth_dev,
546 struct rte_kvargs *kvlist)
548 const char *name = rte_vdev_device_name(dev);
549 const unsigned int numa_node = dev->device.numa_node;
550 struct rte_eth_dev_data *data = NULL;
551 struct rte_kvargs_pair *pair = NULL;
555 struct sockaddr_ll sockaddr;
556 struct tpacket_req *req;
557 struct pkt_rx_queue *rx_queue;
558 struct pkt_tx_queue *tx_queue;
559 int rc, tpver, discard;
561 unsigned int i, q, rdsize;
562 #if defined(PACKET_FANOUT)
566 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
567 pair = &kvlist->pairs[k_idx];
568 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
573 "%s: no interface specified for AF_PACKET ethdev",
579 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
582 *internals = rte_zmalloc_socket(name, sizeof(**internals),
584 if (*internals == NULL)
587 for (q = 0; q < nb_queues; q++) {
588 (*internals)->rx_queue[q].map = MAP_FAILED;
589 (*internals)->tx_queue[q].map = MAP_FAILED;
592 req = &((*internals)->req);
594 req->tp_block_size = blocksize;
595 req->tp_block_nr = blockcnt;
596 req->tp_frame_size = framesize;
597 req->tp_frame_nr = framecnt;
599 ifnamelen = strlen(pair->value);
600 if (ifnamelen < sizeof(ifr.ifr_name)) {
601 memcpy(ifr.ifr_name, pair->value, ifnamelen);
602 ifr.ifr_name[ifnamelen] = '\0';
605 "%s: I/F name too long (%s)",
609 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
610 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
613 (*internals)->if_name = strdup(pair->value);
614 if ((*internals)->if_name == NULL)
616 (*internals)->if_index = ifr.ifr_ifindex;
618 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
619 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
622 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
624 memset(&sockaddr, 0, sizeof(sockaddr));
625 sockaddr.sll_family = AF_PACKET;
626 sockaddr.sll_protocol = htons(ETH_P_ALL);
627 sockaddr.sll_ifindex = (*internals)->if_index;
629 #if defined(PACKET_FANOUT)
630 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
631 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
632 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
633 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
637 for (q = 0; q < nb_queues; q++) {
638 /* Open an AF_PACKET socket for this queue... */
639 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
642 "%s: could not open AF_PACKET socket",
648 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
649 &tpver, sizeof(tpver));
652 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
658 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
659 &discard, sizeof(discard));
662 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
667 #if defined(PACKET_QDISC_BYPASS)
668 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
669 &qdisc_bypass, sizeof(qdisc_bypass));
672 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
677 RTE_SET_USED(qdisc_bypass);
680 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
683 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
688 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
691 "%s: could not set PACKET_TX_RING on AF_PACKET "
692 "socket for %s", name, pair->value);
696 rx_queue = &((*internals)->rx_queue[q]);
697 rx_queue->framecount = req->tp_frame_nr;
699 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
700 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
702 if (rx_queue->map == MAP_FAILED) {
704 "%s: call to mmap failed on AF_PACKET socket for %s",
709 /* rdsize is same for both Tx and Rx */
710 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
712 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
713 if (rx_queue->rd == NULL)
715 for (i = 0; i < req->tp_frame_nr; ++i) {
716 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
717 rx_queue->rd[i].iov_len = req->tp_frame_size;
719 rx_queue->sockfd = qsockfd;
721 tx_queue = &((*internals)->tx_queue[q]);
722 tx_queue->framecount = req->tp_frame_nr;
723 tx_queue->frame_data_size = req->tp_frame_size;
724 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
725 sizeof(struct sockaddr_ll);
727 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
729 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
730 if (tx_queue->rd == NULL)
732 for (i = 0; i < req->tp_frame_nr; ++i) {
733 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
734 tx_queue->rd[i].iov_len = req->tp_frame_size;
736 tx_queue->sockfd = qsockfd;
738 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
741 "%s: could not bind AF_PACKET socket to %s",
746 #if defined(PACKET_FANOUT)
747 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
748 &fanout_arg, sizeof(fanout_arg));
751 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
758 /* reserve an ethdev entry */
759 *eth_dev = rte_eth_vdev_allocate(dev, 0);
760 if (*eth_dev == NULL)
764 * now put it all together
765 * - store queue data in internals,
766 * - store numa_node in eth_dev
767 * - point eth_dev_data to internals
768 * - and point eth_dev structure to new eth_dev_data structure
771 (*internals)->nb_queues = nb_queues;
773 data = (*eth_dev)->data;
774 data->dev_private = *internals;
775 data->nb_rx_queues = (uint16_t)nb_queues;
776 data->nb_tx_queues = (uint16_t)nb_queues;
777 data->dev_link = pmd_link;
778 data->mac_addrs = &(*internals)->eth_addr;
780 (*eth_dev)->dev_ops = &ops;
787 for (q = 0; q < nb_queues; q++) {
788 munmap((*internals)->rx_queue[q].map,
789 2 * req->tp_block_size * req->tp_block_nr);
791 rte_free((*internals)->rx_queue[q].rd);
792 rte_free((*internals)->tx_queue[q].rd);
793 if (((*internals)->rx_queue[q].sockfd != 0) &&
794 ((*internals)->rx_queue[q].sockfd != qsockfd))
795 close((*internals)->rx_queue[q].sockfd);
797 free((*internals)->if_name);
798 rte_free(*internals);
803 rte_eth_from_packet(struct rte_vdev_device *dev,
805 struct rte_kvargs *kvlist)
807 const char *name = rte_vdev_device_name(dev);
808 struct pmd_internals *internals = NULL;
809 struct rte_eth_dev *eth_dev = NULL;
810 struct rte_kvargs_pair *pair = NULL;
812 unsigned int blockcount;
813 unsigned int blocksize;
814 unsigned int framesize = DFLT_FRAME_SIZE;
815 unsigned int framecount = DFLT_FRAME_COUNT;
816 unsigned int qpairs = 1;
817 unsigned int qdisc_bypass = 1;
819 /* do some parameter checking */
823 blocksize = getpagesize();
826 * Walk arguments for configurable settings
828 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
829 pair = &kvlist->pairs[k_idx];
830 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
831 qpairs = atoi(pair->value);
833 qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
835 "%s: invalid qpairs value",
841 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
842 blocksize = atoi(pair->value);
845 "%s: invalid blocksize value",
851 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
852 framesize = atoi(pair->value);
855 "%s: invalid framesize value",
861 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
862 framecount = atoi(pair->value);
865 "%s: invalid framecount value",
871 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
872 qdisc_bypass = atoi(pair->value);
873 if (qdisc_bypass > 1) {
875 "%s: invalid bypass value",
883 if (framesize > blocksize) {
885 "%s: AF_PACKET MMAP frame size exceeds block size!",
890 blockcount = framecount / (blocksize / framesize);
893 "%s: invalid AF_PACKET MMAP parameters", name);
897 PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
898 PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
899 PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
900 PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
901 PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
903 if (rte_pmd_init_internals(dev, *sockfd, qpairs,
904 blocksize, blockcount,
905 framesize, framecount,
907 &internals, ð_dev,
911 eth_dev->rx_pkt_burst = eth_af_packet_rx;
912 eth_dev->tx_pkt_burst = eth_af_packet_tx;
914 rte_eth_dev_probing_finish(eth_dev);
919 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
922 struct rte_kvargs *kvlist;
924 struct rte_eth_dev *eth_dev;
925 const char *name = rte_vdev_device_name(dev);
927 PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
929 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
930 eth_dev = rte_eth_dev_attach_secondary(name);
932 PMD_LOG(ERR, "Failed to probe %s", name);
935 /* TODO: request info from primary to set up Rx and Tx */
936 eth_dev->dev_ops = &ops;
937 eth_dev->device = &dev->device;
938 rte_eth_dev_probing_finish(eth_dev);
942 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
943 if (kvlist == NULL) {
949 * If iface argument is passed we open the NICs and use them for
952 if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
954 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
955 &open_packet_iface, &sockfd);
960 if (dev->device.numa_node == SOCKET_ID_ANY)
961 dev->device.numa_node = rte_socket_id();
963 ret = rte_eth_from_packet(dev, &sockfd, kvlist);
964 close(sockfd); /* no longer needed */
967 rte_kvargs_free(kvlist);
972 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
974 struct rte_eth_dev *eth_dev = NULL;
975 struct pmd_internals *internals;
978 PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
984 /* find the ethdev entry */
985 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
989 /* mac_addrs must not be freed alone because part of dev_private */
990 eth_dev->data->mac_addrs = NULL;
992 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
993 return rte_eth_dev_release_port(eth_dev);
995 internals = eth_dev->data->dev_private;
996 for (q = 0; q < internals->nb_queues; q++) {
997 rte_free(internals->rx_queue[q].rd);
998 rte_free(internals->tx_queue[q].rd);
1000 free(internals->if_name);
1002 rte_eth_dev_release_port(eth_dev);
1007 static struct rte_vdev_driver pmd_af_packet_drv = {
1008 .probe = rte_pmd_af_packet_probe,
1009 .remove = rte_pmd_af_packet_remove,
1012 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1013 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1014 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1020 "qdisc_bypass=<0|1>");
1022 RTE_INIT(af_packet_init_log)
1024 af_packet_logtype = rte_log_register("pmd.net.packet");
1025 if (af_packet_logtype >= 0)
1026 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);