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>
17 #include <linux/if_ether.h>
18 #include <linux/if_packet.h>
19 #include <arpa/inet.h>
21 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/ioctl.h>
28 #define ETH_AF_PACKET_IFACE_ARG "iface"
29 #define ETH_AF_PACKET_NUM_Q_ARG "qpairs"
30 #define ETH_AF_PACKET_BLOCKSIZE_ARG "blocksz"
31 #define ETH_AF_PACKET_FRAMESIZE_ARG "framesz"
32 #define ETH_AF_PACKET_FRAMECOUNT_ARG "framecnt"
33 #define ETH_AF_PACKET_QDISC_BYPASS_ARG "qdisc_bypass"
35 #define DFLT_BLOCK_SIZE (1 << 12)
36 #define DFLT_FRAME_SIZE (1 << 11)
37 #define DFLT_FRAME_COUNT (1 << 9)
39 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
46 unsigned int framecount;
47 unsigned int framenum;
49 struct rte_mempool *mb_pool;
52 volatile unsigned long rx_pkts;
53 volatile unsigned long err_pkts;
54 volatile unsigned long rx_bytes;
59 unsigned int frame_data_size;
63 unsigned int framecount;
64 unsigned int framenum;
66 volatile unsigned long tx_pkts;
67 volatile unsigned long err_pkts;
68 volatile unsigned long tx_bytes;
71 struct pmd_internals {
76 struct rte_ether_addr eth_addr;
78 struct tpacket_req req;
80 struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
81 struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
84 static const char *valid_arguments[] = {
85 ETH_AF_PACKET_IFACE_ARG,
86 ETH_AF_PACKET_NUM_Q_ARG,
87 ETH_AF_PACKET_BLOCKSIZE_ARG,
88 ETH_AF_PACKET_FRAMESIZE_ARG,
89 ETH_AF_PACKET_FRAMECOUNT_ARG,
90 ETH_AF_PACKET_QDISC_BYPASS_ARG,
94 static struct rte_eth_link pmd_link = {
95 .link_speed = ETH_SPEED_NUM_10G,
96 .link_duplex = ETH_LINK_FULL_DUPLEX,
97 .link_status = ETH_LINK_DOWN,
98 .link_autoneg = ETH_LINK_FIXED,
101 static int af_packet_logtype;
103 #define PMD_LOG(level, fmt, args...) \
104 rte_log(RTE_LOG_ ## level, af_packet_logtype, \
105 "%s(): " fmt "\n", __func__, ##args)
108 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
111 struct tpacket2_hdr *ppd;
112 struct rte_mbuf *mbuf;
114 struct pkt_rx_queue *pkt_q = queue;
116 unsigned long num_rx_bytes = 0;
117 unsigned int framecount, framenum;
119 if (unlikely(nb_pkts == 0))
123 * Reads the given number of packets from the AF_PACKET socket one by
124 * one and copies the packet data into a newly allocated mbuf.
126 framecount = pkt_q->framecount;
127 framenum = pkt_q->framenum;
128 for (i = 0; i < nb_pkts; i++) {
129 /* point at the next incoming frame */
130 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
131 if ((ppd->tp_status & TP_STATUS_USER) == 0)
134 /* allocate the next mbuf */
135 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
136 if (unlikely(mbuf == NULL))
139 /* packet will fit in the mbuf, go ahead and receive it */
140 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
141 pbuf = (uint8_t *) ppd + ppd->tp_mac;
142 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
144 /* check for vlan info */
145 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
146 mbuf->vlan_tci = ppd->tp_vlan_tci;
147 mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
150 /* release incoming frame and advance ring buffer */
151 ppd->tp_status = TP_STATUS_KERNEL;
152 if (++framenum >= framecount)
154 mbuf->port = pkt_q->in_port;
156 /* account for the receive frame */
159 num_rx_bytes += mbuf->pkt_len;
161 pkt_q->framenum = framenum;
162 pkt_q->rx_pkts += num_rx;
163 pkt_q->rx_bytes += num_rx_bytes;
168 * Callback to handle sending packets through a real NIC.
171 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
173 struct tpacket2_hdr *ppd;
174 struct rte_mbuf *mbuf;
176 unsigned int framecount, framenum;
178 struct pkt_tx_queue *pkt_q = queue;
180 unsigned long num_tx_bytes = 0;
183 if (unlikely(nb_pkts == 0))
186 memset(&pfd, 0, sizeof(pfd));
187 pfd.fd = pkt_q->sockfd;
188 pfd.events = POLLOUT;
191 framecount = pkt_q->framecount;
192 framenum = pkt_q->framenum;
193 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
194 for (i = 0; i < nb_pkts; i++) {
197 /* drop oversized packets */
198 if (mbuf->pkt_len > pkt_q->frame_data_size) {
199 rte_pktmbuf_free(mbuf);
203 /* insert vlan info if necessary */
204 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
205 if (rte_vlan_insert(&mbuf)) {
206 rte_pktmbuf_free(mbuf);
211 /* point at the next incoming frame */
212 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
213 (poll(&pfd, 1, -1) < 0))
216 /* copy the tx frame data */
217 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
218 sizeof(struct sockaddr_ll);
220 struct rte_mbuf *tmp_mbuf = mbuf;
222 uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
223 memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
225 tmp_mbuf = tmp_mbuf->next;
228 ppd->tp_len = mbuf->pkt_len;
229 ppd->tp_snaplen = mbuf->pkt_len;
231 /* release incoming frame and advance ring buffer */
232 ppd->tp_status = TP_STATUS_SEND_REQUEST;
233 if (++framenum >= framecount)
235 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
238 num_tx_bytes += mbuf->pkt_len;
239 rte_pktmbuf_free(mbuf);
242 /* kick-off transmits */
243 if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
244 /* error sending -- no packets transmitted */
249 pkt_q->framenum = framenum;
250 pkt_q->tx_pkts += num_tx;
251 pkt_q->err_pkts += i - num_tx;
252 pkt_q->tx_bytes += num_tx_bytes;
257 eth_dev_start(struct rte_eth_dev *dev)
259 dev->data->dev_link.link_status = ETH_LINK_UP;
264 * This function gets called when the current port gets stopped.
267 eth_dev_stop(struct rte_eth_dev *dev)
271 struct pmd_internals *internals = dev->data->dev_private;
273 for (i = 0; i < internals->nb_queues; i++) {
274 sockfd = internals->rx_queue[i].sockfd;
278 /* Prevent use after free in case tx fd == rx fd */
279 if (sockfd != internals->tx_queue[i].sockfd) {
280 sockfd = internals->tx_queue[i].sockfd;
285 internals->rx_queue[i].sockfd = -1;
286 internals->tx_queue[i].sockfd = -1;
289 dev->data->dev_link.link_status = ETH_LINK_DOWN;
293 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
299 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
301 struct pmd_internals *internals = dev->data->dev_private;
303 dev_info->if_index = internals->if_index;
304 dev_info->max_mac_addrs = 1;
305 dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
306 dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
307 dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
308 dev_info->min_rx_bufsize = 0;
312 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
315 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
316 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
317 const struct pmd_internals *internal = dev->data->dev_private;
319 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
320 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
321 for (i = 0; i < imax; i++) {
322 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
323 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
324 rx_total += igb_stats->q_ipackets[i];
325 rx_bytes_total += igb_stats->q_ibytes[i];
328 imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
329 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
330 for (i = 0; i < imax; i++) {
331 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
332 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
333 tx_total += igb_stats->q_opackets[i];
334 tx_err_total += internal->tx_queue[i].err_pkts;
335 tx_bytes_total += igb_stats->q_obytes[i];
338 igb_stats->ipackets = rx_total;
339 igb_stats->ibytes = rx_bytes_total;
340 igb_stats->opackets = tx_total;
341 igb_stats->oerrors = tx_err_total;
342 igb_stats->obytes = tx_bytes_total;
347 eth_stats_reset(struct rte_eth_dev *dev)
350 struct pmd_internals *internal = dev->data->dev_private;
352 for (i = 0; i < internal->nb_queues; i++) {
353 internal->rx_queue[i].rx_pkts = 0;
354 internal->rx_queue[i].rx_bytes = 0;
357 for (i = 0; i < internal->nb_queues; i++) {
358 internal->tx_queue[i].tx_pkts = 0;
359 internal->tx_queue[i].err_pkts = 0;
360 internal->tx_queue[i].tx_bytes = 0;
365 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
370 eth_queue_release(void *q __rte_unused)
375 eth_link_update(struct rte_eth_dev *dev __rte_unused,
376 int wait_to_complete __rte_unused)
382 eth_rx_queue_setup(struct rte_eth_dev *dev,
383 uint16_t rx_queue_id,
384 uint16_t nb_rx_desc __rte_unused,
385 unsigned int socket_id __rte_unused,
386 const struct rte_eth_rxconf *rx_conf __rte_unused,
387 struct rte_mempool *mb_pool)
389 struct pmd_internals *internals = dev->data->dev_private;
390 struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
391 unsigned int buf_size, data_size;
393 pkt_q->mb_pool = mb_pool;
395 /* Now get the space available for data in the mbuf */
396 buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
397 RTE_PKTMBUF_HEADROOM;
398 data_size = internals->req.tp_frame_size;
399 data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
401 if (data_size > buf_size) {
403 "%s: %d bytes will not fit in mbuf (%d bytes)",
404 dev->device->name, data_size, buf_size);
408 dev->data->rx_queues[rx_queue_id] = pkt_q;
409 pkt_q->in_port = dev->data->port_id;
415 eth_tx_queue_setup(struct rte_eth_dev *dev,
416 uint16_t tx_queue_id,
417 uint16_t nb_tx_desc __rte_unused,
418 unsigned int socket_id __rte_unused,
419 const struct rte_eth_txconf *tx_conf __rte_unused)
422 struct pmd_internals *internals = dev->data->dev_private;
424 dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
429 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
431 struct pmd_internals *internals = dev->data->dev_private;
432 struct ifreq ifr = { .ifr_mtu = mtu };
435 unsigned int data_size = internals->req.tp_frame_size -
441 s = socket(PF_INET, SOCK_DGRAM, 0);
445 strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
446 ret = ioctl(s, SIOCSIFMTU, &ifr);
456 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
461 s = socket(PF_INET, SOCK_DGRAM, 0);
465 strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
466 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
468 ifr.ifr_flags &= mask;
469 ifr.ifr_flags |= flags;
470 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
477 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
479 struct pmd_internals *internals = dev->data->dev_private;
481 eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
485 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
487 struct pmd_internals *internals = dev->data->dev_private;
489 eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
492 static const struct eth_dev_ops ops = {
493 .dev_start = eth_dev_start,
494 .dev_stop = eth_dev_stop,
495 .dev_close = eth_dev_close,
496 .dev_configure = eth_dev_configure,
497 .dev_infos_get = eth_dev_info,
498 .mtu_set = eth_dev_mtu_set,
499 .promiscuous_enable = eth_dev_promiscuous_enable,
500 .promiscuous_disable = eth_dev_promiscuous_disable,
501 .rx_queue_setup = eth_rx_queue_setup,
502 .tx_queue_setup = eth_tx_queue_setup,
503 .rx_queue_release = eth_queue_release,
504 .tx_queue_release = eth_queue_release,
505 .link_update = eth_link_update,
506 .stats_get = eth_stats_get,
507 .stats_reset = eth_stats_reset,
511 * Opens an AF_PACKET socket
514 open_packet_iface(const char *key __rte_unused,
515 const char *value __rte_unused,
518 int *sockfd = extra_args;
520 /* Open an AF_PACKET socket... */
521 *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
523 PMD_LOG(ERR, "Could not open AF_PACKET socket");
530 static struct rte_vdev_driver pmd_af_packet_drv;
533 rte_pmd_init_internals(struct rte_vdev_device *dev,
535 const unsigned nb_queues,
536 unsigned int blocksize,
537 unsigned int blockcnt,
538 unsigned int framesize,
539 unsigned int framecnt,
540 unsigned int qdisc_bypass,
541 struct pmd_internals **internals,
542 struct rte_eth_dev **eth_dev,
543 struct rte_kvargs *kvlist)
545 const char *name = rte_vdev_device_name(dev);
546 const unsigned int numa_node = dev->device.numa_node;
547 struct rte_eth_dev_data *data = NULL;
548 struct rte_kvargs_pair *pair = NULL;
552 struct sockaddr_ll sockaddr;
553 struct tpacket_req *req;
554 struct pkt_rx_queue *rx_queue;
555 struct pkt_tx_queue *tx_queue;
556 int rc, tpver, discard;
558 unsigned int i, q, rdsize;
559 #if defined(PACKET_FANOUT)
563 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
564 pair = &kvlist->pairs[k_idx];
565 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
570 "%s: no interface specified for AF_PACKET ethdev",
576 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
579 *internals = rte_zmalloc_socket(name, sizeof(**internals),
581 if (*internals == NULL)
584 for (q = 0; q < nb_queues; q++) {
585 (*internals)->rx_queue[q].map = MAP_FAILED;
586 (*internals)->tx_queue[q].map = MAP_FAILED;
589 req = &((*internals)->req);
591 req->tp_block_size = blocksize;
592 req->tp_block_nr = blockcnt;
593 req->tp_frame_size = framesize;
594 req->tp_frame_nr = framecnt;
596 ifnamelen = strlen(pair->value);
597 if (ifnamelen < sizeof(ifr.ifr_name)) {
598 memcpy(ifr.ifr_name, pair->value, ifnamelen);
599 ifr.ifr_name[ifnamelen] = '\0';
602 "%s: I/F name too long (%s)",
606 if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
608 "%s: ioctl failed (SIOCGIFINDEX)",
612 (*internals)->if_name = strdup(pair->value);
613 if ((*internals)->if_name == NULL)
615 (*internals)->if_index = ifr.ifr_ifindex;
617 if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
619 "%s: ioctl failed (SIOCGIFHWADDR)",
623 memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
625 memset(&sockaddr, 0, sizeof(sockaddr));
626 sockaddr.sll_family = AF_PACKET;
627 sockaddr.sll_protocol = htons(ETH_P_ALL);
628 sockaddr.sll_ifindex = (*internals)->if_index;
630 #if defined(PACKET_FANOUT)
631 fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
632 fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
633 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
634 fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
638 for (q = 0; q < nb_queues; q++) {
639 /* Open an AF_PACKET socket for this queue... */
640 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
643 "%s: could not open AF_PACKET socket",
649 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
650 &tpver, sizeof(tpver));
653 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
659 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
660 &discard, sizeof(discard));
663 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
668 #if defined(PACKET_QDISC_BYPASS)
669 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
670 &qdisc_bypass, sizeof(qdisc_bypass));
673 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
678 RTE_SET_USED(qdisc_bypass);
681 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
684 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
689 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
692 "%s: could not set PACKET_TX_RING on AF_PACKET "
693 "socket for %s", name, pair->value);
697 rx_queue = &((*internals)->rx_queue[q]);
698 rx_queue->framecount = req->tp_frame_nr;
700 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
701 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
703 if (rx_queue->map == MAP_FAILED) {
705 "%s: call to mmap failed on AF_PACKET socket for %s",
710 /* rdsize is same for both Tx and Rx */
711 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
713 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
714 if (rx_queue->rd == NULL)
716 for (i = 0; i < req->tp_frame_nr; ++i) {
717 rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
718 rx_queue->rd[i].iov_len = req->tp_frame_size;
720 rx_queue->sockfd = qsockfd;
722 tx_queue = &((*internals)->tx_queue[q]);
723 tx_queue->framecount = req->tp_frame_nr;
724 tx_queue->frame_data_size = req->tp_frame_size;
725 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
726 sizeof(struct sockaddr_ll);
728 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
730 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
731 if (tx_queue->rd == NULL)
733 for (i = 0; i < req->tp_frame_nr; ++i) {
734 tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
735 tx_queue->rd[i].iov_len = req->tp_frame_size;
737 tx_queue->sockfd = qsockfd;
739 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
742 "%s: could not bind AF_PACKET socket to %s",
747 #if defined(PACKET_FANOUT)
748 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
749 &fanout_arg, sizeof(fanout_arg));
752 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
753 "for %s", name, pair->value);
759 /* reserve an ethdev entry */
760 *eth_dev = rte_eth_vdev_allocate(dev, 0);
761 if (*eth_dev == NULL)
765 * now put it all together
766 * - store queue data in internals,
767 * - store numa_node in eth_dev
768 * - point eth_dev_data to internals
769 * - and point eth_dev structure to new eth_dev_data structure
772 (*internals)->nb_queues = nb_queues;
774 data = (*eth_dev)->data;
775 data->dev_private = *internals;
776 data->nb_rx_queues = (uint16_t)nb_queues;
777 data->nb_tx_queues = (uint16_t)nb_queues;
778 data->dev_link = pmd_link;
779 data->mac_addrs = &(*internals)->eth_addr;
781 (*eth_dev)->dev_ops = &ops;
788 for (q = 0; q < nb_queues; q++) {
789 munmap((*internals)->rx_queue[q].map,
790 2 * req->tp_block_size * req->tp_block_nr);
792 rte_free((*internals)->rx_queue[q].rd);
793 rte_free((*internals)->tx_queue[q].rd);
794 if (((*internals)->rx_queue[q].sockfd != 0) &&
795 ((*internals)->rx_queue[q].sockfd != qsockfd))
796 close((*internals)->rx_queue[q].sockfd);
798 free((*internals)->if_name);
799 rte_free(*internals);
804 rte_eth_from_packet(struct rte_vdev_device *dev,
806 struct rte_kvargs *kvlist)
808 const char *name = rte_vdev_device_name(dev);
809 struct pmd_internals *internals = NULL;
810 struct rte_eth_dev *eth_dev = NULL;
811 struct rte_kvargs_pair *pair = NULL;
813 unsigned int blockcount;
814 unsigned int blocksize = DFLT_BLOCK_SIZE;
815 unsigned int framesize = DFLT_FRAME_SIZE;
816 unsigned int framecount = DFLT_FRAME_COUNT;
817 unsigned int qpairs = 1;
818 unsigned int qdisc_bypass = 1;
820 /* do some parameter checking */
825 * Walk arguments for configurable settings
827 for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
828 pair = &kvlist->pairs[k_idx];
829 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
830 qpairs = atoi(pair->value);
832 qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
834 "%s: invalid qpairs value",
840 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
841 blocksize = atoi(pair->value);
844 "%s: invalid blocksize value",
850 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
851 framesize = atoi(pair->value);
854 "%s: invalid framesize value",
860 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
861 framecount = atoi(pair->value);
864 "%s: invalid framecount value",
870 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
871 qdisc_bypass = atoi(pair->value);
872 if (qdisc_bypass > 1) {
874 "%s: invalid bypass value",
882 if (framesize > blocksize) {
884 "%s: AF_PACKET MMAP frame size exceeds block size!",
889 blockcount = framecount / (blocksize / framesize);
892 "%s: invalid AF_PACKET MMAP parameters", name);
896 PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
897 PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
898 PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
899 PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
900 PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
902 if (rte_pmd_init_internals(dev, *sockfd, qpairs,
903 blocksize, blockcount,
904 framesize, framecount,
906 &internals, ð_dev,
910 eth_dev->rx_pkt_burst = eth_af_packet_rx;
911 eth_dev->tx_pkt_burst = eth_af_packet_tx;
913 rte_eth_dev_probing_finish(eth_dev);
918 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
921 struct rte_kvargs *kvlist;
923 struct rte_eth_dev *eth_dev;
924 const char *name = rte_vdev_device_name(dev);
926 PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
928 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
929 eth_dev = rte_eth_dev_attach_secondary(name);
931 PMD_LOG(ERR, "Failed to probe %s", name);
934 /* TODO: request info from primary to set up Rx and Tx */
935 eth_dev->dev_ops = &ops;
936 eth_dev->device = &dev->device;
937 rte_eth_dev_probing_finish(eth_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 PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
983 /* find the ethdev entry */
984 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
988 /* mac_addrs must not be freed alone because part of dev_private */
989 eth_dev->data->mac_addrs = NULL;
991 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
992 return rte_eth_dev_release_port(eth_dev);
994 internals = eth_dev->data->dev_private;
995 for (q = 0; q < internals->nb_queues; q++) {
996 rte_free(internals->rx_queue[q].rd);
997 rte_free(internals->tx_queue[q].rd);
999 free(internals->if_name);
1001 rte_eth_dev_release_port(eth_dev);
1006 static struct rte_vdev_driver pmd_af_packet_drv = {
1007 .probe = rte_pmd_af_packet_probe,
1008 .remove = rte_pmd_af_packet_remove,
1011 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1012 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1013 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1019 "qdisc_bypass=<0|1>");
1021 RTE_INIT(af_packet_init_log)
1023 af_packet_logtype = rte_log_register("pmd.net.packet");
1024 if (af_packet_logtype >= 0)
1025 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);