1ffe1cdd1a0c0844b20e0e8f67783bba1011eb64
[dpdk.git] / lib / librte_pmd_af_packet / rte_eth_af_packet.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
5  *
6  *   Originally based upon librte_pmd_pcap code:
7  *
8  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
9  *   Copyright(c) 2014 6WIND S.A.
10  *   All rights reserved.
11  *
12  *   Redistribution and use in source and binary forms, with or without
13  *   modification, are permitted provided that the following conditions
14  *   are met:
15  *
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
21  *       distribution.
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.
25  *
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.
37  */
38
39 #include <rte_mbuf.h>
40 #include <rte_ethdev.h>
41 #include <rte_malloc.h>
42 #include <rte_kvargs.h>
43 #include <rte_dev.h>
44
45 #include <linux/if_ether.h>
46 #include <linux/if_packet.h>
47 #include <arpa/inet.h>
48 #include <net/if.h>
49 #include <sys/types.h>
50 #include <sys/socket.h>
51 #include <sys/ioctl.h>
52 #include <sys/mman.h>
53 #include <unistd.h>
54 #include <poll.h>
55
56 #include "rte_eth_af_packet.h"
57
58 #define ETH_AF_PACKET_IFACE_ARG         "iface"
59 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
60 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
61 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
62 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
63
64 #define DFLT_BLOCK_SIZE         (1 << 12)
65 #define DFLT_FRAME_SIZE         (1 << 11)
66 #define DFLT_FRAME_COUNT        (1 << 9)
67
68 struct pkt_rx_queue {
69         int sockfd;
70
71         struct iovec *rd;
72         uint8_t *map;
73         unsigned int framecount;
74         unsigned int framenum;
75
76         struct rte_mempool *mb_pool;
77
78         volatile unsigned long rx_pkts;
79         volatile unsigned long err_pkts;
80 };
81
82 struct pkt_tx_queue {
83         int sockfd;
84
85         struct iovec *rd;
86         uint8_t *map;
87         unsigned int framecount;
88         unsigned int framenum;
89
90         volatile unsigned long tx_pkts;
91         volatile unsigned long err_pkts;
92 };
93
94 struct pmd_internals {
95         unsigned nb_queues;
96
97         int if_index;
98         struct ether_addr eth_addr;
99
100         struct tpacket_req req;
101
102         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
103         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
104 };
105
106 static const char *valid_arguments[] = {
107         ETH_AF_PACKET_IFACE_ARG,
108         ETH_AF_PACKET_NUM_Q_ARG,
109         ETH_AF_PACKET_BLOCKSIZE_ARG,
110         ETH_AF_PACKET_FRAMESIZE_ARG,
111         ETH_AF_PACKET_FRAMECOUNT_ARG,
112         NULL
113 };
114
115 static const char *drivername = "AF_PACKET PMD";
116
117 static struct rte_eth_link pmd_link = {
118         .link_speed = 10000,
119         .link_duplex = ETH_LINK_FULL_DUPLEX,
120         .link_status = 0
121 };
122
123 static uint16_t
124 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
125 {
126         unsigned i;
127         struct tpacket2_hdr *ppd;
128         struct rte_mbuf *mbuf;
129         uint8_t *pbuf;
130         struct pkt_rx_queue *pkt_q = queue;
131         uint16_t num_rx = 0;
132         unsigned int framecount, framenum;
133
134         if (unlikely(nb_pkts == 0))
135                 return 0;
136
137         /*
138          * Reads the given number of packets from the AF_PACKET socket one by
139          * one and copies the packet data into a newly allocated mbuf.
140          */
141         framecount = pkt_q->framecount;
142         framenum = pkt_q->framenum;
143         for (i = 0; i < nb_pkts; i++) {
144                 /* point at the next incoming frame */
145                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
146                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
147                         break;
148
149                 /* allocate the next mbuf */
150                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
151                 if (unlikely(mbuf == NULL))
152                         break;
153
154                 /* packet will fit in the mbuf, go ahead and receive it */
155                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
156                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
157                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
158
159                 /* release incoming frame and advance ring buffer */
160                 ppd->tp_status = TP_STATUS_KERNEL;
161                 if (++framenum >= framecount)
162                         framenum = 0;
163
164                 /* account for the receive frame */
165                 bufs[i] = mbuf;
166                 num_rx++;
167         }
168         pkt_q->framenum = framenum;
169         pkt_q->rx_pkts += num_rx;
170         return num_rx;
171 }
172
173 /*
174  * Callback to handle sending packets through a real NIC.
175  */
176 static uint16_t
177 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
178 {
179         struct tpacket2_hdr *ppd;
180         struct rte_mbuf *mbuf;
181         uint8_t *pbuf;
182         unsigned int framecount, framenum;
183         struct pollfd pfd;
184         struct pkt_tx_queue *pkt_q = queue;
185         uint16_t num_tx = 0;
186         int i;
187
188         if (unlikely(nb_pkts == 0))
189                 return 0;
190
191         memset(&pfd, 0, sizeof(pfd));
192         pfd.fd = pkt_q->sockfd;
193         pfd.events = POLLOUT;
194         pfd.revents = 0;
195
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++) {
200                 /* point at the next incoming frame */
201                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
202                     (poll(&pfd, 1, -1) < 0))
203                                 continue;
204
205                 /* copy the tx frame data */
206                 mbuf = bufs[num_tx];
207                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
208                         sizeof(struct sockaddr_ll);
209                 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
210                 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
211
212                 /* release incoming frame and advance ring buffer */
213                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
214                 if (++framenum >= framecount)
215                         framenum = 0;
216                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
217
218                 num_tx++;
219                 rte_pktmbuf_free(mbuf);
220         }
221
222         /* kick-off transmits */
223         sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0);
224
225         pkt_q->framenum = framenum;
226         pkt_q->tx_pkts += num_tx;
227         pkt_q->err_pkts += nb_pkts - num_tx;
228         return num_tx;
229 }
230
231 static int
232 eth_dev_start(struct rte_eth_dev *dev)
233 {
234         dev->data->dev_link.link_status = 1;
235         return 0;
236 }
237
238 /*
239  * This function gets called when the current port gets stopped.
240  */
241 static void
242 eth_dev_stop(struct rte_eth_dev *dev)
243 {
244         unsigned i;
245         int sockfd;
246         struct pmd_internals *internals = dev->data->dev_private;
247
248         for (i = 0; i < internals->nb_queues; i++) {
249                 sockfd = internals->rx_queue[i].sockfd;
250                 if (sockfd != -1)
251                         close(sockfd);
252                 sockfd = internals->tx_queue[i].sockfd;
253                 if (sockfd != -1)
254                         close(sockfd);
255         }
256
257         dev->data->dev_link.link_status = 0;
258 }
259
260 static int
261 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
262 {
263         return 0;
264 }
265
266 static void
267 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
268 {
269         struct pmd_internals *internals = dev->data->dev_private;
270
271         dev_info->driver_name = drivername;
272         dev_info->if_index = internals->if_index;
273         dev_info->max_mac_addrs = 1;
274         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
275         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
276         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
277         dev_info->min_rx_bufsize = 0;
278         dev_info->pci_dev = NULL;
279 }
280
281 static void
282 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
283 {
284         unsigned i, imax;
285         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
286         const struct pmd_internals *internal = dev->data->dev_private;
287
288         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
289                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
290         for (i = 0; i < imax; i++) {
291                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
292                 rx_total += igb_stats->q_ipackets[i];
293         }
294
295         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
296                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
297         for (i = 0; i < imax; i++) {
298                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
299                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
300                 tx_total += igb_stats->q_opackets[i];
301                 tx_err_total += igb_stats->q_errors[i];
302         }
303
304         igb_stats->ipackets = rx_total;
305         igb_stats->opackets = tx_total;
306         igb_stats->oerrors = tx_err_total;
307 }
308
309 static void
310 eth_stats_reset(struct rte_eth_dev *dev)
311 {
312         unsigned i;
313         struct pmd_internals *internal = dev->data->dev_private;
314
315         for (i = 0; i < internal->nb_queues; i++)
316                 internal->rx_queue[i].rx_pkts = 0;
317
318         for (i = 0; i < internal->nb_queues; i++) {
319                 internal->tx_queue[i].tx_pkts = 0;
320                 internal->tx_queue[i].err_pkts = 0;
321         }
322 }
323
324 static void
325 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
326 {
327 }
328
329 static void
330 eth_queue_release(void *q __rte_unused)
331 {
332 }
333
334 static int
335 eth_link_update(struct rte_eth_dev *dev __rte_unused,
336                 int wait_to_complete __rte_unused)
337 {
338         return 0;
339 }
340
341 static int
342 eth_rx_queue_setup(struct rte_eth_dev *dev,
343                    uint16_t rx_queue_id,
344                    uint16_t nb_rx_desc __rte_unused,
345                    unsigned int socket_id __rte_unused,
346                    const struct rte_eth_rxconf *rx_conf __rte_unused,
347                    struct rte_mempool *mb_pool)
348 {
349         struct pmd_internals *internals = dev->data->dev_private;
350         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
351         struct rte_pktmbuf_pool_private *mbp_priv;
352         uint16_t buf_size;
353
354         pkt_q->mb_pool = mb_pool;
355
356         /* Now get the space available for data in the mbuf */
357         mbp_priv = rte_mempool_get_priv(pkt_q->mb_pool);
358         buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
359                                RTE_PKTMBUF_HEADROOM);
360
361         if (ETH_FRAME_LEN > buf_size) {
362                 RTE_LOG(ERR, PMD,
363                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
364                         dev->data->name, ETH_FRAME_LEN, buf_size);
365                 return -ENOMEM;
366         }
367
368         dev->data->rx_queues[rx_queue_id] = pkt_q;
369
370         return 0;
371 }
372
373 static int
374 eth_tx_queue_setup(struct rte_eth_dev *dev,
375                    uint16_t tx_queue_id,
376                    uint16_t nb_tx_desc __rte_unused,
377                    unsigned int socket_id __rte_unused,
378                    const struct rte_eth_txconf *tx_conf __rte_unused)
379 {
380
381         struct pmd_internals *internals = dev->data->dev_private;
382
383         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
384         return 0;
385 }
386
387 static struct eth_dev_ops ops = {
388         .dev_start = eth_dev_start,
389         .dev_stop = eth_dev_stop,
390         .dev_close = eth_dev_close,
391         .dev_configure = eth_dev_configure,
392         .dev_infos_get = eth_dev_info,
393         .rx_queue_setup = eth_rx_queue_setup,
394         .tx_queue_setup = eth_tx_queue_setup,
395         .rx_queue_release = eth_queue_release,
396         .tx_queue_release = eth_queue_release,
397         .link_update = eth_link_update,
398         .stats_get = eth_stats_get,
399         .stats_reset = eth_stats_reset,
400 };
401
402 /*
403  * Opens an AF_PACKET socket
404  */
405 static int
406 open_packet_iface(const char *key __rte_unused,
407                   const char *value __rte_unused,
408                   void *extra_args)
409 {
410         int *sockfd = extra_args;
411
412         /* Open an AF_PACKET socket... */
413         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
414         if (*sockfd == -1) {
415                 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
416                 return -1;
417         }
418
419         return 0;
420 }
421
422 static int
423 rte_pmd_init_internals(const char *name,
424                        const int sockfd,
425                        const unsigned nb_queues,
426                        unsigned int blocksize,
427                        unsigned int blockcnt,
428                        unsigned int framesize,
429                        unsigned int framecnt,
430                        const unsigned numa_node,
431                        struct pmd_internals **internals,
432                        struct rte_eth_dev **eth_dev,
433                        struct rte_kvargs *kvlist)
434 {
435         struct rte_eth_dev_data *data = NULL;
436         struct rte_pci_device *pci_dev = NULL;
437         struct rte_kvargs_pair *pair = NULL;
438         struct ifreq ifr;
439         size_t ifnamelen;
440         unsigned k_idx;
441         struct sockaddr_ll sockaddr;
442         struct tpacket_req *req;
443         struct pkt_rx_queue *rx_queue;
444         struct pkt_tx_queue *tx_queue;
445         int rc, qsockfd, tpver, discard;
446         unsigned int i, q, rdsize;
447         int fanout_arg __rte_unused, bypass __rte_unused;
448
449         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
450                 pair = &kvlist->pairs[k_idx];
451                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
452                         break;
453         }
454         if (pair == NULL) {
455                 RTE_LOG(ERR, PMD,
456                         "%s: no interface specified for AF_PACKET ethdev\n",
457                         name);
458                 goto error;
459         }
460
461         RTE_LOG(INFO, PMD,
462                 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
463                 name, numa_node);
464
465         /*
466          * now do all data allocation - for eth_dev structure, dummy pci driver
467          * and internal (private) data
468          */
469         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
470         if (data == NULL)
471                 goto error;
472
473         pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
474         if (pci_dev == NULL)
475                 goto error;
476
477         *internals = rte_zmalloc_socket(name, sizeof(**internals),
478                                         0, numa_node);
479         if (*internals == NULL)
480                 goto error;
481
482         for (q = 0; q < nb_queues; q++) {
483                 (*internals)->rx_queue[q].map = MAP_FAILED;
484                 (*internals)->tx_queue[q].map = MAP_FAILED;
485         }
486
487         req = &((*internals)->req);
488
489         req->tp_block_size = blocksize;
490         req->tp_block_nr = blockcnt;
491         req->tp_frame_size = framesize;
492         req->tp_frame_nr = framecnt;
493
494         ifnamelen = strlen(pair->value);
495         if (ifnamelen < sizeof(ifr.ifr_name)) {
496                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
497                 ifr.ifr_name[ifnamelen] = '\0';
498         } else {
499                 RTE_LOG(ERR, PMD,
500                         "%s: I/F name too long (%s)\n",
501                         name, pair->value);
502                 goto error;
503         }
504         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
505                 RTE_LOG(ERR, PMD,
506                         "%s: ioctl failed (SIOCGIFINDEX)\n",
507                         name);
508                 goto error;
509         }
510         (*internals)->if_index = ifr.ifr_ifindex;
511
512         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
513                 RTE_LOG(ERR, PMD,
514                         "%s: ioctl failed (SIOCGIFHWADDR)\n",
515                         name);
516                 goto error;
517         }
518         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
519
520         memset(&sockaddr, 0, sizeof(sockaddr));
521         sockaddr.sll_family = AF_PACKET;
522         sockaddr.sll_protocol = htons(ETH_P_ALL);
523         sockaddr.sll_ifindex = (*internals)->if_index;
524
525 #if defined(PACKET_FANOUT)
526         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
527         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
528 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
529         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
530 #endif
531 #endif
532
533         for (q = 0; q < nb_queues; q++) {
534                 /* Open an AF_PACKET socket for this queue... */
535                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
536                 if (qsockfd == -1) {
537                         RTE_LOG(ERR, PMD,
538                                 "%s: could not open AF_PACKET socket\n",
539                                 name);
540                         return -1;
541                 }
542
543                 tpver = TPACKET_V2;
544                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
545                                 &tpver, sizeof(tpver));
546                 if (rc == -1) {
547                         RTE_LOG(ERR, PMD,
548                                 "%s: could not set PACKET_VERSION on AF_PACKET "
549                                 "socket for %s\n", name, pair->value);
550                         goto error;
551                 }
552
553                 discard = 1;
554                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
555                                 &discard, sizeof(discard));
556                 if (rc == -1) {
557                         RTE_LOG(ERR, PMD,
558                                 "%s: could not set PACKET_LOSS on "
559                                 "AF_PACKET socket for %s\n", name, pair->value);
560                         goto error;
561                 }
562
563 #if defined(PACKET_QDISC_BYPASS)
564                 bypass = 1;
565                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
566                                 &bypass, sizeof(bypass));
567                 if (rc == -1) {
568                         RTE_LOG(ERR, PMD,
569                                 "%s: could not set PACKET_QDISC_BYPASS "
570                                 "on AF_PACKET socket for %s\n", name,
571                                 pair->value);
572                         goto error;
573                 }
574 #endif
575
576                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
577                 if (rc == -1) {
578                         RTE_LOG(ERR, PMD,
579                                 "%s: could not set PACKET_RX_RING on AF_PACKET "
580                                 "socket for %s\n", name, pair->value);
581                         goto error;
582                 }
583
584                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
585                 if (rc == -1) {
586                         RTE_LOG(ERR, PMD,
587                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
588                                 "socket for %s\n", name, pair->value);
589                         goto error;
590                 }
591
592                 rx_queue = &((*internals)->rx_queue[q]);
593                 rx_queue->framecount = req->tp_frame_nr;
594
595                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
596                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
597                                     qsockfd, 0);
598                 if (rx_queue->map == MAP_FAILED) {
599                         RTE_LOG(ERR, PMD,
600                                 "%s: call to mmap failed on AF_PACKET socket for %s\n",
601                                 name, pair->value);
602                         goto error;
603                 }
604
605                 /* rdsize is same for both Tx and Rx */
606                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
607
608                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
609                 if (rx_queue->rd == NULL)
610                         goto error;
611                 for (i = 0; i < req->tp_frame_nr; ++i) {
612                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
613                         rx_queue->rd[i].iov_len = req->tp_frame_size;
614                 }
615                 rx_queue->sockfd = qsockfd;
616
617                 tx_queue = &((*internals)->tx_queue[q]);
618                 tx_queue->framecount = req->tp_frame_nr;
619
620                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
621
622                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
623                 if (tx_queue->rd == NULL)
624                         goto error;
625                 for (i = 0; i < req->tp_frame_nr; ++i) {
626                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
627                         tx_queue->rd[i].iov_len = req->tp_frame_size;
628                 }
629                 tx_queue->sockfd = qsockfd;
630
631                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
632                 if (rc == -1) {
633                         RTE_LOG(ERR, PMD,
634                                 "%s: could not bind AF_PACKET socket to %s\n",
635                                 name, pair->value);
636                         goto error;
637                 }
638
639 #if defined(PACKET_FANOUT)
640                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
641                                 &fanout_arg, sizeof(fanout_arg));
642                 if (rc == -1) {
643                         RTE_LOG(ERR, PMD,
644                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
645                                 "for %s\n", name, pair->value);
646                         goto error;
647                 }
648 #endif
649         }
650
651         /* reserve an ethdev entry */
652         *eth_dev = rte_eth_dev_allocate(name);
653         if (*eth_dev == NULL)
654                 goto error;
655
656         /*
657          * now put it all together
658          * - store queue data in internals,
659          * - store numa_node info in pci_driver
660          * - point eth_dev_data to internals and pci_driver
661          * - and point eth_dev structure to new eth_dev_data structure
662          */
663
664         (*internals)->nb_queues = nb_queues;
665
666         data->dev_private = *internals;
667         data->port_id = (*eth_dev)->data->port_id;
668         data->nb_rx_queues = (uint16_t)nb_queues;
669         data->nb_tx_queues = (uint16_t)nb_queues;
670         data->dev_link = pmd_link;
671         data->mac_addrs = &(*internals)->eth_addr;
672
673         pci_dev->numa_node = numa_node;
674
675         (*eth_dev)->data = data;
676         (*eth_dev)->dev_ops = &ops;
677         (*eth_dev)->pci_dev = pci_dev;
678
679         return 0;
680
681 error:
682         if (data)
683                 rte_free(data);
684         if (pci_dev)
685                 rte_free(pci_dev);
686         if (*internals) {
687                 for (q = 0; q < nb_queues; q++) {
688                         munmap((*internals)->rx_queue[q].map,
689                                2 * req->tp_block_size * req->tp_block_nr);
690                         if ((*internals)->rx_queue[q].rd)
691                                 rte_free((*internals)->rx_queue[q].rd);
692                         if ((*internals)->tx_queue[q].rd)
693                                 rte_free((*internals)->tx_queue[q].rd);
694                 }
695                 rte_free(*internals);
696         }
697         return -1;
698 }
699
700 static int
701 rte_eth_from_packet(const char *name,
702                     int const *sockfd,
703                     const unsigned numa_node,
704                     struct rte_kvargs *kvlist)
705 {
706         struct pmd_internals *internals = NULL;
707         struct rte_eth_dev *eth_dev = NULL;
708         struct rte_kvargs_pair *pair = NULL;
709         unsigned k_idx;
710         unsigned int blockcount;
711         unsigned int blocksize = DFLT_BLOCK_SIZE;
712         unsigned int framesize = DFLT_FRAME_SIZE;
713         unsigned int framecount = DFLT_FRAME_COUNT;
714         unsigned int qpairs = 1;
715
716         /* do some parameter checking */
717         if (*sockfd < 0)
718                 return -1;
719
720         /*
721          * Walk arguments for configurable settings
722          */
723         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
724                 pair = &kvlist->pairs[k_idx];
725                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
726                         qpairs = atoi(pair->value);
727                         if (qpairs < 1 ||
728                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
729                                 RTE_LOG(ERR, PMD,
730                                         "%s: invalid qpairs value\n",
731                                         name);
732                                 return -1;
733                         }
734                         continue;
735                 }
736                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
737                         blocksize = atoi(pair->value);
738                         if (!blocksize) {
739                                 RTE_LOG(ERR, PMD,
740                                         "%s: invalid blocksize value\n",
741                                         name);
742                                 return -1;
743                         }
744                         continue;
745                 }
746                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
747                         framesize = atoi(pair->value);
748                         if (!framesize) {
749                                 RTE_LOG(ERR, PMD,
750                                         "%s: invalid framesize value\n",
751                                         name);
752                                 return -1;
753                         }
754                         continue;
755                 }
756                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
757                         framecount = atoi(pair->value);
758                         if (!framecount) {
759                                 RTE_LOG(ERR, PMD,
760                                         "%s: invalid framecount value\n",
761                                         name);
762                                 return -1;
763                         }
764                         continue;
765                 }
766         }
767
768         if (framesize > blocksize) {
769                 RTE_LOG(ERR, PMD,
770                         "%s: AF_PACKET MMAP frame size exceeds block size!\n",
771                         name);
772                 return -1;
773         }
774
775         blockcount = framecount / (blocksize / framesize);
776         if (!blockcount) {
777                 RTE_LOG(ERR, PMD,
778                         "%s: invalid AF_PACKET MMAP parameters\n", name);
779                 return -1;
780         }
781
782         RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
783         RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
784         RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
785         RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
786         RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
787
788         if (rte_pmd_init_internals(name, *sockfd, qpairs,
789                                    blocksize, blockcount,
790                                    framesize, framecount,
791                                    numa_node, &internals, &eth_dev,
792                                    kvlist) < 0)
793                 return -1;
794
795         eth_dev->rx_pkt_burst = eth_af_packet_rx;
796         eth_dev->tx_pkt_burst = eth_af_packet_tx;
797
798         return 0;
799 }
800
801 int
802 rte_pmd_af_packet_devinit(const char *name, const char *params)
803 {
804         unsigned numa_node;
805         int ret;
806         struct rte_kvargs *kvlist;
807         int sockfd = -1;
808
809         RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
810
811         numa_node = rte_socket_id();
812
813         kvlist = rte_kvargs_parse(params, valid_arguments);
814         if (kvlist == NULL)
815                 return -1;
816
817         /*
818          * If iface argument is passed we open the NICs and use them for
819          * reading / writing
820          */
821         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
822
823                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
824                                          &open_packet_iface, &sockfd);
825                 if (ret < 0)
826                         return -1;
827         }
828
829         ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
830         close(sockfd); /* no longer needed */
831
832         if (ret < 0)
833                 return -1;
834
835         return 0;
836 }
837
838 static struct rte_driver pmd_af_packet_drv = {
839         .name = "eth_af_packet",
840         .type = PMD_VDEV,
841         .init = rte_pmd_af_packet_devinit,
842 };
843
844 PMD_REGISTER_DRIVER(pmd_af_packet_drv);