xref: /f-stack/lib/ff_dpdk_if.c (revision 0f5432bb)
1 /*
2  * Copyright (C) 2017 THL A29 Limited, a Tencent company.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions are met:
7  *
8  * 1. Redistributions of source code must retain the above copyright notice, this
9  *   list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright notice,
11  *   this list of conditions and the following disclaimer in the documentation
12  *   and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
16  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
17  * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
18  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
19  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
20  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
21  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
22  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
23  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
24  *
25  */
26 #include <assert.h>
27 #include <unistd.h>
28 #include <sys/mman.h>
29 #include <errno.h>
30 
31 #include <rte_common.h>
32 #include <rte_byteorder.h>
33 #include <rte_log.h>
34 #include <rte_memory.h>
35 #include <rte_memcpy.h>
36 #include <rte_memzone.h>
37 #include <rte_config.h>
38 #include <rte_eal.h>
39 #include <rte_pci.h>
40 #include <rte_mbuf.h>
41 #include <rte_memory.h>
42 #include <rte_lcore.h>
43 #include <rte_launch.h>
44 #include <rte_ethdev.h>
45 #include <rte_debug.h>
46 #include <rte_common.h>
47 #include <rte_ether.h>
48 #include <rte_malloc.h>
49 #include <rte_cycles.h>
50 #include <rte_timer.h>
51 #include <rte_thash.h>
52 #include <rte_ip.h>
53 #include <rte_tcp.h>
54 #include <rte_udp.h>
55 
56 #include "ff_dpdk_if.h"
57 #include "ff_dpdk_pcap.h"
58 #include "ff_dpdk_kni.h"
59 #include "ff_config.h"
60 #include "ff_veth.h"
61 #include "ff_host_interface.h"
62 #include "ff_msg.h"
63 #include "ff_api.h"
64 #include "ff_memory.h"
65 
66 #ifdef FF_KNI
67 #define KNI_MBUF_MAX 2048
68 #define KNI_QUEUE_SIZE 2048
69 
70 int enable_kni;
71 static int kni_accept;
72 #endif
73 
74 static int numa_on;
75 
76 static unsigned idle_sleep;
77 static unsigned pkt_tx_delay;
78 
79 static struct rte_timer freebsd_clock;
80 
81 // Mellanox Linux's driver key
82 static uint8_t default_rsskey_40bytes[40] = {
83     0xd1, 0x81, 0xc6, 0x2c, 0xf7, 0xf4, 0xdb, 0x5b,
84     0x19, 0x83, 0xa2, 0xfc, 0x94, 0x3e, 0x1a, 0xdb,
85     0xd9, 0x38, 0x9e, 0x6b, 0xd1, 0x03, 0x9c, 0x2c,
86     0xa7, 0x44, 0x99, 0xad, 0x59, 0x3d, 0x56, 0xd9,
87     0xf3, 0x25, 0x3c, 0x06, 0x2a, 0xdc, 0x1f, 0xfc
88 };
89 
90 struct lcore_conf lcore_conf;
91 
92 struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
93 
94 static struct rte_ring **dispatch_ring[RTE_MAX_ETHPORTS];
95 static dispatch_func_t packet_dispatcher;
96 
97 static uint16_t rss_reta_size[RTE_MAX_ETHPORTS];
98 
99 static inline int send_single_packet(struct rte_mbuf *m, uint8_t port);
100 
101 struct ff_msg_ring {
102     char ring_name[FF_MSG_NUM][RTE_RING_NAMESIZE];
103     /* ring[0] for lcore recv msg, other send */
104     /* ring[1] for lcore send msg, other read */
105     struct rte_ring *ring[FF_MSG_NUM];
106 } __rte_cache_aligned;
107 
108 static struct ff_msg_ring msg_ring[RTE_MAX_LCORE];
109 static struct rte_mempool *message_pool;
110 static struct ff_dpdk_if_context *veth_ctx[RTE_MAX_ETHPORTS];
111 
112 static struct ff_top_args ff_top_status;
113 static struct ff_traffic_args ff_traffic;
114 extern void ff_hardclock(void);
115 
116 static void
117 ff_hardclock_job(__rte_unused struct rte_timer *timer,
118     __rte_unused void *arg) {
119     ff_hardclock();
120     ff_update_current_ts();
121 }
122 
123 struct ff_dpdk_if_context *
124 ff_dpdk_register_if(void *sc, void *ifp, struct ff_port_cfg *cfg)
125 {
126     struct ff_dpdk_if_context *ctx;
127 
128     ctx = calloc(1, sizeof(struct ff_dpdk_if_context));
129     if (ctx == NULL)
130         return NULL;
131 
132     ctx->sc = sc;
133     ctx->ifp = ifp;
134     ctx->port_id = cfg->port_id;
135     ctx->hw_features = cfg->hw_features;
136 
137     return ctx;
138 }
139 
140 void
141 ff_dpdk_deregister_if(struct ff_dpdk_if_context *ctx)
142 {
143     free(ctx);
144 }
145 
146 static void
147 check_all_ports_link_status(void)
148 {
149     #define CHECK_INTERVAL 100 /* 100ms */
150     #define MAX_CHECK_TIME 90  /* 9s (90 * 100ms) in total */
151 
152     uint16_t portid;
153     uint8_t count, all_ports_up, print_flag = 0;
154     struct rte_eth_link link;
155 
156     printf("\nChecking link status");
157     fflush(stdout);
158 
159     int i, nb_ports;
160     nb_ports = ff_global_cfg.dpdk.nb_ports;
161     for (count = 0; count <= MAX_CHECK_TIME; count++) {
162         all_ports_up = 1;
163         for (i = 0; i < nb_ports; i++) {
164             uint16_t portid = ff_global_cfg.dpdk.portid_list[i];
165             memset(&link, 0, sizeof(link));
166             rte_eth_link_get_nowait(portid, &link);
167 
168             /* print link status if flag set */
169             if (print_flag == 1) {
170                 if (link.link_status) {
171                     printf("Port %d Link Up - speed %u "
172                         "Mbps - %s\n", (int)portid,
173                         (unsigned)link.link_speed,
174                         (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
175                         ("full-duplex") : ("half-duplex\n"));
176                 } else {
177                     printf("Port %d Link Down\n", (int)portid);
178                 }
179                 continue;
180             }
181             /* clear all_ports_up flag if any link down */
182             if (link.link_status == 0) {
183                 all_ports_up = 0;
184                 break;
185             }
186         }
187 
188         /* after finally printing all link status, get out */
189         if (print_flag == 1)
190             break;
191 
192         if (all_ports_up == 0) {
193             printf(".");
194             fflush(stdout);
195             rte_delay_ms(CHECK_INTERVAL);
196         }
197 
198         /* set the print_flag if all ports up or timeout */
199         if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
200             print_flag = 1;
201             printf("done\n");
202         }
203     }
204 }
205 
206 static int
207 init_lcore_conf(void)
208 {
209     uint8_t nb_dev_ports = rte_eth_dev_count_avail();
210     if (nb_dev_ports == 0) {
211         rte_exit(EXIT_FAILURE, "No probed ethernet devices\n");
212     }
213 
214     if (ff_global_cfg.dpdk.max_portid >= nb_dev_ports) {
215         rte_exit(EXIT_FAILURE, "this machine doesn't have port %d.\n",
216                  ff_global_cfg.dpdk.max_portid);
217     }
218 
219     lcore_conf.port_cfgs = ff_global_cfg.dpdk.port_cfgs;
220     lcore_conf.proc_id = ff_global_cfg.dpdk.proc_id;
221 
222     uint16_t proc_id;
223     for (proc_id = 0; proc_id < ff_global_cfg.dpdk.nb_procs; proc_id++) {
224         uint16_t lcore_id = ff_global_cfg.dpdk.proc_lcore[proc_id];
225         if (!lcore_config[lcore_id].detected) {
226             rte_exit(EXIT_FAILURE, "lcore %u unavailable\n", lcore_id);
227         }
228     }
229 
230     uint16_t socket_id = 0;
231     if (numa_on) {
232         socket_id = rte_lcore_to_socket_id(rte_lcore_id());
233     }
234 
235     lcore_conf.socket_id = socket_id;
236 
237     uint16_t lcore_id = ff_global_cfg.dpdk.proc_lcore[lcore_conf.proc_id];
238     int j;
239     for (j = 0; j < ff_global_cfg.dpdk.nb_ports; ++j) {
240         uint16_t port_id = ff_global_cfg.dpdk.portid_list[j];
241         struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[port_id];
242 
243         int queueid = -1;
244         int i;
245         for (i = 0; i < pconf->nb_lcores; i++) {
246             if (pconf->lcore_list[i] == lcore_id) {
247                 queueid = i;
248             }
249         }
250         if (queueid < 0) {
251             continue;
252         }
253         printf("lcore: %u, port: %u, queue: %u\n", lcore_id, port_id, queueid);
254         uint16_t nb_rx_queue = lcore_conf.nb_rx_queue;
255         lcore_conf.rx_queue_list[nb_rx_queue].port_id = port_id;
256         lcore_conf.rx_queue_list[nb_rx_queue].queue_id = queueid;
257         lcore_conf.nb_rx_queue++;
258 
259         lcore_conf.tx_queue_id[port_id] = queueid;
260         lcore_conf.tx_port_id[lcore_conf.nb_tx_port] = port_id;
261         lcore_conf.nb_tx_port++;
262 
263         lcore_conf.pcap[port_id] = pconf->pcap;
264         lcore_conf.nb_queue_list[port_id] = pconf->nb_lcores;
265     }
266 
267     if (lcore_conf.nb_rx_queue == 0) {
268         rte_exit(EXIT_FAILURE, "lcore %u has nothing to do\n", lcore_id);
269     }
270 
271     return 0;
272 }
273 
274 static int
275 init_mem_pool(void)
276 {
277     uint8_t nb_ports = ff_global_cfg.dpdk.nb_ports;
278     uint32_t nb_lcores = ff_global_cfg.dpdk.nb_procs;
279     uint32_t nb_tx_queue = nb_lcores;
280     uint32_t nb_rx_queue = lcore_conf.nb_rx_queue * nb_lcores;
281 
282     unsigned nb_mbuf = RTE_ALIGN_CEIL (
283         (nb_rx_queue*RX_QUEUE_SIZE          +
284         nb_ports*nb_lcores*MAX_PKT_BURST    +
285         nb_ports*nb_tx_queue*TX_QUEUE_SIZE  +
286         nb_lcores*MEMPOOL_CACHE_SIZE +
287 #ifdef FF_KNI
288         nb_ports*KNI_MBUF_MAX +
289         nb_ports*KNI_QUEUE_SIZE +
290 #endif
291         nb_lcores*nb_ports*DISPATCH_RING_SIZE),
292         (unsigned)8192);
293 
294     unsigned socketid = 0;
295     uint16_t i, lcore_id;
296     char s[64];
297 
298     for (i = 0; i < ff_global_cfg.dpdk.nb_procs; i++) {
299         lcore_id = ff_global_cfg.dpdk.proc_lcore[i];
300         if (numa_on) {
301             socketid = rte_lcore_to_socket_id(lcore_id);
302         }
303 
304         if (socketid >= NB_SOCKETS) {
305             rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
306                 socketid, i, NB_SOCKETS);
307         }
308 
309         if (pktmbuf_pool[socketid] != NULL) {
310             continue;
311         }
312 
313         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
314             snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
315             pktmbuf_pool[socketid] =
316                 rte_pktmbuf_pool_create(s, nb_mbuf,
317                     MEMPOOL_CACHE_SIZE, 0,
318                     RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
319         } else {
320             snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
321             pktmbuf_pool[socketid] = rte_mempool_lookup(s);
322         }
323 
324         if (pktmbuf_pool[socketid] == NULL) {
325             rte_exit(EXIT_FAILURE, "Cannot create mbuf pool on socket %d\n", socketid);
326         } else {
327             printf("create mbuf pool on socket %d\n", socketid);
328         }
329 
330 #ifdef FF_USE_PAGE_ARRAY
331         nb_mbuf = RTE_ALIGN_CEIL (
332             nb_ports*nb_lcores*MAX_PKT_BURST    +
333             nb_ports*nb_tx_queue*TX_QUEUE_SIZE  +
334             nb_lcores*MEMPOOL_CACHE_SIZE,
335             (unsigned)4096);
336         ff_init_ref_pool(nb_mbuf, socketid);
337 #endif
338     }
339 
340     return 0;
341 }
342 
343 static struct rte_ring *
344 create_ring(const char *name, unsigned count, int socket_id, unsigned flags)
345 {
346     struct rte_ring *ring;
347 
348     if (name == NULL) {
349         rte_exit(EXIT_FAILURE, "create ring failed, no name!\n");
350     }
351 
352     if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
353         ring = rte_ring_create(name, count, socket_id, flags);
354     } else {
355         ring = rte_ring_lookup(name);
356     }
357 
358     if (ring == NULL) {
359         rte_exit(EXIT_FAILURE, "create ring:%s failed!\n", name);
360     }
361 
362     return ring;
363 }
364 
365 static int
366 init_dispatch_ring(void)
367 {
368     int j;
369     char name_buf[RTE_RING_NAMESIZE];
370     int queueid;
371 
372     unsigned socketid = lcore_conf.socket_id;
373 
374     /* Create ring according to ports actually being used. */
375     int nb_ports = ff_global_cfg.dpdk.nb_ports;
376     for (j = 0; j < nb_ports; j++) {
377         uint16_t portid = ff_global_cfg.dpdk.portid_list[j];
378         struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[portid];
379         int nb_queues = pconf->nb_lcores;
380         if (dispatch_ring[portid] == NULL) {
381             snprintf(name_buf, RTE_RING_NAMESIZE, "ring_ptr_p%d", portid);
382 
383             dispatch_ring[portid] = rte_zmalloc(name_buf,
384                 sizeof(struct rte_ring *) * nb_queues,
385                 RTE_CACHE_LINE_SIZE);
386             if (dispatch_ring[portid] == NULL) {
387                 rte_exit(EXIT_FAILURE, "rte_zmalloc(%s (struct rte_ring*)) "
388                     "failed\n", name_buf);
389             }
390         }
391 
392         for(queueid = 0; queueid < nb_queues; ++queueid) {
393             snprintf(name_buf, RTE_RING_NAMESIZE, "dispatch_ring_p%d_q%d",
394                 portid, queueid);
395             dispatch_ring[portid][queueid] = create_ring(name_buf,
396                 DISPATCH_RING_SIZE, socketid, RING_F_SC_DEQ);
397 
398             if (dispatch_ring[portid][queueid] == NULL)
399                 rte_panic("create ring:%s failed!\n", name_buf);
400 
401             printf("create ring:%s success, %u ring entries are now free!\n",
402                 name_buf, rte_ring_free_count(dispatch_ring[portid][queueid]));
403         }
404     }
405 
406     return 0;
407 }
408 
409 static void
410 ff_msg_init(struct rte_mempool *mp,
411     __attribute__((unused)) void *opaque_arg,
412     void *obj, __attribute__((unused)) unsigned i)
413 {
414     struct ff_msg *msg = (struct ff_msg *)obj;
415     msg->msg_type = FF_UNKNOWN;
416     msg->buf_addr = (char *)msg + sizeof(struct ff_msg);
417     msg->buf_len = mp->elt_size - sizeof(struct ff_msg);
418 }
419 
420 static int
421 init_msg_ring(void)
422 {
423     uint16_t i, j;
424     uint16_t nb_procs = ff_global_cfg.dpdk.nb_procs;
425     unsigned socketid = lcore_conf.socket_id;
426 
427     /* Create message buffer pool */
428     if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
429         message_pool = rte_mempool_create(FF_MSG_POOL,
430            MSG_RING_SIZE * 2 * nb_procs,
431            MAX_MSG_BUF_SIZE, MSG_RING_SIZE / 2, 0,
432            NULL, NULL, ff_msg_init, NULL,
433            socketid, 0);
434     } else {
435         message_pool = rte_mempool_lookup(FF_MSG_POOL);
436     }
437 
438     if (message_pool == NULL) {
439         rte_panic("Create msg mempool failed\n");
440     }
441 
442     for(i = 0; i < nb_procs; ++i) {
443         snprintf(msg_ring[i].ring_name[0], RTE_RING_NAMESIZE,
444             "%s%u", FF_MSG_RING_IN, i);
445         msg_ring[i].ring[0] = create_ring(msg_ring[i].ring_name[0],
446             MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ);
447         if (msg_ring[i].ring[0] == NULL)
448             rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[0]);
449 
450         for (j = FF_SYSCTL; j < FF_MSG_NUM; j++) {
451             snprintf(msg_ring[i].ring_name[j], RTE_RING_NAMESIZE,
452                 "%s%u_%u", FF_MSG_RING_OUT, i, j);
453             msg_ring[i].ring[j] = create_ring(msg_ring[i].ring_name[j],
454                 MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ);
455             if (msg_ring[i].ring[j] == NULL)
456                 rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[j]);
457         }
458     }
459 
460     return 0;
461 }
462 
463 #ifdef FF_KNI
464 static int
465 init_kni(void)
466 {
467     int nb_ports = rte_eth_dev_count_avail();
468     kni_accept = 0;
469     if(strcasecmp(ff_global_cfg.kni.method, "accept") == 0)
470         kni_accept = 1;
471 
472     ff_kni_init(nb_ports, ff_global_cfg.kni.tcp_port,
473         ff_global_cfg.kni.udp_port);
474 
475     unsigned socket_id = lcore_conf.socket_id;
476     struct rte_mempool *mbuf_pool = pktmbuf_pool[socket_id];
477 
478     nb_ports = ff_global_cfg.dpdk.nb_ports;
479     int i, ret;
480     for (i = 0; i < nb_ports; i++) {
481         uint16_t port_id = ff_global_cfg.dpdk.portid_list[i];
482         ff_kni_alloc(port_id, socket_id, mbuf_pool, KNI_QUEUE_SIZE);
483     }
484 
485     return 0;
486 }
487 #endif
488 
489 static void
490 set_rss_table(uint16_t port_id, uint16_t reta_size, uint16_t nb_queues)
491 {
492     if (reta_size == 0) {
493         return;
494     }
495 
496     int reta_conf_size = RTE_MAX(1, reta_size / RTE_RETA_GROUP_SIZE);
497     struct rte_eth_rss_reta_entry64 reta_conf[reta_conf_size];
498 
499     /* config HW indirection table */
500     unsigned i, j, hash=0;
501     for (i = 0; i < reta_conf_size; i++) {
502         reta_conf[i].mask = ~0ULL;
503         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++) {
504             reta_conf[i].reta[j] = hash++ % nb_queues;
505         }
506     }
507 
508     if (rte_eth_dev_rss_reta_update(port_id, reta_conf, reta_size)) {
509         rte_exit(EXIT_FAILURE, "port[%d], failed to update rss table\n",
510             port_id);
511     }
512 }
513 
514 static int
515 init_port_start(void)
516 {
517     int nb_ports = ff_global_cfg.dpdk.nb_ports;
518     unsigned socketid = 0;
519     struct rte_mempool *mbuf_pool;
520     uint16_t i;
521 
522     for (i = 0; i < nb_ports; i++) {
523         uint16_t port_id = ff_global_cfg.dpdk.portid_list[i];
524         struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[port_id];
525         uint16_t nb_queues = pconf->nb_lcores;
526 
527         struct rte_eth_dev_info dev_info;
528         struct rte_eth_conf port_conf = {0};
529         struct rte_eth_rxconf rxq_conf;
530         struct rte_eth_txconf txq_conf;
531 
532         rte_eth_dev_info_get(port_id, &dev_info);
533 
534         if (nb_queues > dev_info.max_rx_queues) {
535             rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_rx_queues[%d]\n",
536                 nb_queues,
537                 dev_info.max_rx_queues);
538         }
539 
540         if (nb_queues > dev_info.max_tx_queues) {
541             rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_tx_queues[%d]\n",
542                 nb_queues,
543                 dev_info.max_tx_queues);
544         }
545 
546         struct ether_addr addr;
547         rte_eth_macaddr_get(port_id, &addr);
548         printf("Port %u MAC: %02" PRIx8 " %02" PRIx8 " %02" PRIx8
549                    " %02" PRIx8 " %02" PRIx8 " %02" PRIx8 "\n",
550                 (unsigned)port_id,
551                 addr.addr_bytes[0], addr.addr_bytes[1],
552                 addr.addr_bytes[2], addr.addr_bytes[3],
553                 addr.addr_bytes[4], addr.addr_bytes[5]);
554 
555         rte_memcpy(pconf->mac,
556             addr.addr_bytes, ETHER_ADDR_LEN);
557 
558         /* Set RSS mode */
559         uint64_t default_rss_hf = ETH_RSS_PROTO_MASK;
560         port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
561         port_conf.rx_adv_conf.rss_conf.rss_hf = default_rss_hf;
562         port_conf.rx_adv_conf.rss_conf.rss_key = default_rsskey_40bytes;
563         port_conf.rx_adv_conf.rss_conf.rss_key_len = 40;
564         port_conf.rx_adv_conf.rss_conf.rss_hf &= dev_info.flow_type_rss_offloads;
565         if (port_conf.rx_adv_conf.rss_conf.rss_hf !=
566                 ETH_RSS_PROTO_MASK) {
567             printf("Port %u modified RSS hash function based on hardware support,"
568                     "requested:%#"PRIx64" configured:%#"PRIx64"\n",
569                     port_id, default_rss_hf,
570                     port_conf.rx_adv_conf.rss_conf.rss_hf);
571         }
572 
573         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) {
574             port_conf.txmode.offloads |=
575                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
576         }
577 
578         /* Set Rx VLAN stripping */
579         if (ff_global_cfg.dpdk.vlan_strip) {
580             if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_STRIP) {
581                 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
582             }
583         }
584 
585         /* Enable HW CRC stripping */
586         port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_KEEP_CRC;
587 
588         /* FIXME: Enable TCP LRO ?*/
589         #if 0
590         if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO) {
591             printf("LRO is supported\n");
592             port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_TCP_LRO;
593             pconf->hw_features.rx_lro = 1;
594         }
595         #endif
596 
597         /* Set Rx checksum checking */
598         if ((dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM) &&
599             (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_UDP_CKSUM) &&
600             (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_CKSUM)) {
601             printf("RX checksum offload supported\n");
602             port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM;
603             pconf->hw_features.rx_csum = 1;
604         }
605 
606         if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)) {
607             printf("TX ip checksum offload supported\n");
608             port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
609             pconf->hw_features.tx_csum_ip = 1;
610         }
611 
612         if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_CKSUM) &&
613             (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_CKSUM)) {
614             printf("TX TCP&UDP checksum offload supported\n");
615             port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM | DEV_TX_OFFLOAD_TCP_CKSUM;
616             pconf->hw_features.tx_csum_l4 = 1;
617         }
618 
619         if (ff_global_cfg.dpdk.tso) {
620             if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) {
621                 printf("TSO is supported\n");
622                 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_TSO;
623                 pconf->hw_features.tx_tso = 1;
624             }
625         } else {
626             printf("TSO is disabled\n");
627         }
628 
629         if (dev_info.reta_size) {
630             /* reta size must be power of 2 */
631             assert((dev_info.reta_size & (dev_info.reta_size - 1)) == 0);
632 
633             rss_reta_size[port_id] = dev_info.reta_size;
634             printf("port[%d]: rss table size: %d\n", port_id,
635                 dev_info.reta_size);
636         }
637 
638         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
639             continue;
640         }
641 
642         int ret = rte_eth_dev_configure(port_id, nb_queues, nb_queues, &port_conf);
643         if (ret != 0) {
644             return ret;
645         }
646 
647         static uint16_t nb_rxd = RX_QUEUE_SIZE;
648         static uint16_t nb_txd = TX_QUEUE_SIZE;
649         ret = rte_eth_dev_adjust_nb_rx_tx_desc(port_id, &nb_rxd, &nb_txd);
650         if (ret < 0)
651             printf("Could not adjust number of descriptors "
652                     "for port%u (%d)\n", (unsigned)port_id, ret);
653 
654         uint16_t q;
655         for (q = 0; q < nb_queues; q++) {
656             if (numa_on) {
657                 uint16_t lcore_id = lcore_conf.port_cfgs[port_id].lcore_list[q];
658                 socketid = rte_lcore_to_socket_id(lcore_id);
659             }
660             mbuf_pool = pktmbuf_pool[socketid];
661 
662             txq_conf = dev_info.default_txconf;
663             txq_conf.offloads = port_conf.txmode.offloads;
664             ret = rte_eth_tx_queue_setup(port_id, q, nb_txd,
665                 socketid, &txq_conf);
666             if (ret < 0) {
667                 return ret;
668             }
669 
670             rxq_conf = dev_info.default_rxconf;
671             rxq_conf.offloads = port_conf.rxmode.offloads;
672             ret = rte_eth_rx_queue_setup(port_id, q, nb_rxd,
673                 socketid, &rxq_conf, mbuf_pool);
674             if (ret < 0) {
675                 return ret;
676             }
677         }
678 
679         ret = rte_eth_dev_start(port_id);
680         if (ret < 0) {
681             return ret;
682         }
683 
684         if (nb_queues > 1) {
685             /* set HW rss hash function to Toeplitz. */
686             if (!rte_eth_dev_filter_supported(port_id, RTE_ETH_FILTER_HASH)) {
687                 struct rte_eth_hash_filter_info info = {0};
688                 info.info_type = RTE_ETH_HASH_FILTER_GLOBAL_CONFIG;
689                 info.info.global_conf.hash_func = RTE_ETH_HASH_FUNCTION_TOEPLITZ;
690 
691                 if (rte_eth_dev_filter_ctrl(port_id, RTE_ETH_FILTER_HASH,
692                     RTE_ETH_FILTER_SET, &info) < 0) {
693                     rte_exit(EXIT_FAILURE, "port[%d] set hash func failed\n",
694                         port_id);
695                 }
696             }
697 
698             set_rss_table(port_id, dev_info.reta_size, nb_queues);
699         }
700 
701         /* Enable RX in promiscuous mode for the Ethernet device. */
702         if (ff_global_cfg.dpdk.promiscuous) {
703             rte_eth_promiscuous_enable(port_id);
704             ret = rte_eth_promiscuous_get(port_id);
705             if (ret == 1) {
706                 printf("set port %u to promiscuous mode ok\n", port_id);
707             } else {
708                 printf("set port %u to promiscuous mode error\n", port_id);
709             }
710         }
711 
712         /* Enable pcap dump */
713         if (pconf->pcap) {
714             ff_enable_pcap(pconf->pcap);
715         }
716     }
717 
718     if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
719         check_all_ports_link_status();
720     }
721 
722     return 0;
723 }
724 
725 static int
726 init_clock(void)
727 {
728     rte_timer_subsystem_init();
729     uint64_t hz = rte_get_timer_hz();
730     uint64_t intrs = MS_PER_S/ff_global_cfg.freebsd.hz;
731     uint64_t tsc = (hz + MS_PER_S - 1) / MS_PER_S*intrs;
732 
733     rte_timer_init(&freebsd_clock);
734     rte_timer_reset(&freebsd_clock, tsc, PERIODICAL,
735         rte_lcore_id(), &ff_hardclock_job, NULL);
736 
737     ff_update_current_ts();
738 
739     return 0;
740 }
741 
742 int
743 ff_dpdk_init(int argc, char **argv)
744 {
745     if (ff_global_cfg.dpdk.nb_procs < 1 ||
746         ff_global_cfg.dpdk.nb_procs > RTE_MAX_LCORE ||
747         ff_global_cfg.dpdk.proc_id >= ff_global_cfg.dpdk.nb_procs ||
748         ff_global_cfg.dpdk.proc_id < 0) {
749         printf("param num_procs[%d] or proc_id[%d] error!\n",
750             ff_global_cfg.dpdk.nb_procs,
751             ff_global_cfg.dpdk.proc_id);
752         exit(1);
753     }
754 
755     int ret = rte_eal_init(argc, argv);
756     if (ret < 0) {
757         rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
758     }
759 
760     numa_on = ff_global_cfg.dpdk.numa_on;
761 
762     idle_sleep = ff_global_cfg.dpdk.idle_sleep;
763     pkt_tx_delay = ff_global_cfg.dpdk.pkt_tx_delay > BURST_TX_DRAIN_US ? \
764         BURST_TX_DRAIN_US : ff_global_cfg.dpdk.pkt_tx_delay;
765 
766     init_lcore_conf();
767 
768     init_mem_pool();
769 
770     init_dispatch_ring();
771 
772     init_msg_ring();
773 
774 #ifdef FF_KNI
775     enable_kni = ff_global_cfg.kni.enable;
776     if (enable_kni) {
777         init_kni();
778     }
779 #endif
780 
781 #ifdef FF_USE_PAGE_ARRAY
782     ff_mmap_init();
783 #endif
784 
785     ret = init_port_start();
786     if (ret < 0) {
787         rte_exit(EXIT_FAILURE, "init_port_start failed\n");
788     }
789 
790     init_clock();
791 
792     return 0;
793 }
794 
795 static void
796 ff_veth_input(const struct ff_dpdk_if_context *ctx, struct rte_mbuf *pkt)
797 {
798     uint8_t rx_csum = ctx->hw_features.rx_csum;
799     if (rx_csum) {
800         if (pkt->ol_flags & (PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD)) {
801             rte_pktmbuf_free(pkt);
802             return;
803         }
804     }
805 
806     /*
807      * FIXME: should we save pkt->vlan_tci
808      * if (pkt->ol_flags & PKT_RX_VLAN_PKT)
809      */
810 
811     void *data = rte_pktmbuf_mtod(pkt, void*);
812     uint16_t len = rte_pktmbuf_data_len(pkt);
813 
814     void *hdr = ff_mbuf_gethdr(pkt, pkt->pkt_len, data, len, rx_csum);
815     if (hdr == NULL) {
816         rte_pktmbuf_free(pkt);
817         return;
818     }
819 
820     struct rte_mbuf *pn = pkt->next;
821     void *prev = hdr;
822     while(pn != NULL) {
823         data = rte_pktmbuf_mtod(pn, void*);
824         len = rte_pktmbuf_data_len(pn);
825 
826         void *mb = ff_mbuf_get(prev, data, len);
827         if (mb == NULL) {
828             ff_mbuf_free(hdr);
829             rte_pktmbuf_free(pkt);
830             return;
831         }
832         pn = pn->next;
833         prev = mb;
834     }
835 
836     ff_veth_process_packet(ctx->ifp, hdr);
837 }
838 
839 static enum FilterReturn
840 protocol_filter(const void *data, uint16_t len)
841 {
842     if(len < ETHER_HDR_LEN)
843         return FILTER_UNKNOWN;
844 
845     const struct ether_hdr *hdr;
846     hdr = (const struct ether_hdr *)data;
847     uint16_t eth_frame_type = rte_be_to_cpu_16(hdr->ether_type);
848 
849     if(eth_frame_type == ETHER_TYPE_ARP)
850         return FILTER_ARP;
851 
852 #ifdef INET6
853     if (eth_frame_type == ETHER_TYPE_IPv6) {
854         return ff_kni_proto_filter(data + ETHER_HDR_LEN,
855             len - ETHER_HDR_LEN, eth_frame_type);
856     }
857 #endif
858 
859 #ifndef FF_KNI
860     return FILTER_UNKNOWN;
861 #else
862     if (!enable_kni) {
863         return FILTER_UNKNOWN;
864     }
865 
866     if(eth_frame_type != ETHER_TYPE_IPv4)
867         return FILTER_UNKNOWN;
868 
869     return ff_kni_proto_filter(data + ETHER_HDR_LEN,
870         len - ETHER_HDR_LEN, eth_frame_type);
871 #endif
872 }
873 
874 static inline void
875 pktmbuf_deep_attach(struct rte_mbuf *mi, const struct rte_mbuf *m)
876 {
877     struct rte_mbuf *md;
878     void *src, *dst;
879 
880     dst = rte_pktmbuf_mtod(mi, void *);
881     src = rte_pktmbuf_mtod(m, void *);
882 
883     mi->data_len = m->data_len;
884     rte_memcpy(dst, src, m->data_len);
885 
886     mi->port = m->port;
887     mi->vlan_tci = m->vlan_tci;
888     mi->vlan_tci_outer = m->vlan_tci_outer;
889     mi->tx_offload = m->tx_offload;
890     mi->hash = m->hash;
891     mi->ol_flags = m->ol_flags;
892     mi->packet_type = m->packet_type;
893 }
894 
895 /* copied from rte_pktmbuf_clone */
896 static inline struct rte_mbuf *
897 pktmbuf_deep_clone(const struct rte_mbuf *md,
898     struct rte_mempool *mp)
899 {
900     struct rte_mbuf *mc, *mi, **prev;
901     uint32_t pktlen;
902     uint8_t nseg;
903 
904     if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
905         return NULL;
906 
907     mi = mc;
908     prev = &mi->next;
909     pktlen = md->pkt_len;
910     nseg = 0;
911 
912     do {
913         nseg++;
914         pktmbuf_deep_attach(mi, md);
915         *prev = mi;
916         prev = &mi->next;
917     } while ((md = md->next) != NULL &&
918         (mi = rte_pktmbuf_alloc(mp)) != NULL);
919 
920     *prev = NULL;
921     mc->nb_segs = nseg;
922     mc->pkt_len = pktlen;
923 
924     /* Allocation of new indirect segment failed */
925     if (unlikely (mi == NULL)) {
926         rte_pktmbuf_free(mc);
927         return NULL;
928     }
929 
930     __rte_mbuf_sanity_check(mc, 1);
931     return mc;
932 }
933 
934 static inline void
935 process_packets(uint16_t port_id, uint16_t queue_id, struct rte_mbuf **bufs,
936     uint16_t count, const struct ff_dpdk_if_context *ctx, int pkts_from_ring)
937 {
938     struct lcore_conf *qconf = &lcore_conf;
939     uint16_t nb_queues = qconf->nb_queue_list[port_id];
940 
941     uint16_t i;
942     for (i = 0; i < count; i++) {
943         struct rte_mbuf *rtem = bufs[i];
944 
945         if (unlikely(qconf->pcap[port_id] != NULL)) {
946             if (!pkts_from_ring) {
947                 ff_dump_packets(qconf->pcap[port_id], rtem);
948             }
949         }
950 
951         void *data = rte_pktmbuf_mtod(rtem, void*);
952         uint16_t len = rte_pktmbuf_data_len(rtem);
953 
954         if (!pkts_from_ring) {
955             ff_traffic.rx_packets++;
956             ff_traffic.rx_bytes += len;
957         }
958 
959         if (!pkts_from_ring && packet_dispatcher) {
960             int ret = (*packet_dispatcher)(data, &len, queue_id, nb_queues);
961             if (ret == FF_DISPATCH_RESPONSE) {
962                 rte_pktmbuf_pkt_len(rtem) = rte_pktmbuf_data_len(rtem) = len;
963                 send_single_packet(rtem, port_id);
964                 continue;
965             }
966 
967             if (ret == FF_DISPATCH_ERROR || ret >= nb_queues) {
968                 rte_pktmbuf_free(rtem);
969                 continue;
970             }
971 
972             if (ret != queue_id) {
973                 ret = rte_ring_enqueue(dispatch_ring[port_id][ret], rtem);
974                 if (ret < 0)
975                     rte_pktmbuf_free(rtem);
976 
977                 continue;
978             }
979         }
980 
981         enum FilterReturn filter = protocol_filter(data, len);
982 #ifdef INET6
983         if (filter == FILTER_ARP || filter == FILTER_NDP) {
984 #else
985         if (filter == FILTER_ARP) {
986 #endif
987             struct rte_mempool *mbuf_pool;
988             struct rte_mbuf *mbuf_clone;
989             if (!pkts_from_ring) {
990                 uint16_t j;
991                 for(j = 0; j < nb_queues; ++j) {
992                     if(j == queue_id)
993                         continue;
994 
995                     unsigned socket_id = 0;
996                     if (numa_on) {
997                         uint16_t lcore_id = qconf->port_cfgs[port_id].lcore_list[j];
998                         socket_id = rte_lcore_to_socket_id(lcore_id);
999                     }
1000                     mbuf_pool = pktmbuf_pool[socket_id];
1001                     mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool);
1002                     if(mbuf_clone) {
1003                         int ret = rte_ring_enqueue(dispatch_ring[port_id][j],
1004                             mbuf_clone);
1005                         if (ret < 0)
1006                             rte_pktmbuf_free(mbuf_clone);
1007                     }
1008                 }
1009             }
1010 
1011 #ifdef FF_KNI
1012             if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) {
1013                 mbuf_pool = pktmbuf_pool[qconf->socket_id];
1014                 mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool);
1015                 if(mbuf_clone) {
1016                     ff_kni_enqueue(port_id, mbuf_clone);
1017                 }
1018             }
1019 #endif
1020             ff_veth_input(ctx, rtem);
1021 #ifdef FF_KNI
1022         } else if (enable_kni &&
1023             ((filter == FILTER_KNI && kni_accept) ||
1024             (filter == FILTER_UNKNOWN && !kni_accept)) ) {
1025             ff_kni_enqueue(port_id, rtem);
1026 #endif
1027         } else {
1028             ff_veth_input(ctx, rtem);
1029         }
1030     }
1031 }
1032 
1033 static inline int
1034 process_dispatch_ring(uint16_t port_id, uint16_t queue_id,
1035     struct rte_mbuf **pkts_burst, const struct ff_dpdk_if_context *ctx)
1036 {
1037     /* read packet from ring buf and to process */
1038     uint16_t nb_rb;
1039     nb_rb = rte_ring_dequeue_burst(dispatch_ring[port_id][queue_id],
1040         (void **)pkts_burst, MAX_PKT_BURST, NULL);
1041 
1042     if(nb_rb > 0) {
1043         process_packets(port_id, queue_id, pkts_burst, nb_rb, ctx, 1);
1044     }
1045 
1046     return 0;
1047 }
1048 
1049 static inline void
1050 handle_sysctl_msg(struct ff_msg *msg)
1051 {
1052     int ret = ff_sysctl(msg->sysctl.name, msg->sysctl.namelen,
1053         msg->sysctl.old, msg->sysctl.oldlenp, msg->sysctl.new,
1054         msg->sysctl.newlen);
1055 
1056     if (ret < 0) {
1057         msg->result = errno;
1058     } else {
1059         msg->result = 0;
1060     }
1061 }
1062 
1063 static inline void
1064 handle_ioctl_msg(struct ff_msg *msg)
1065 {
1066     int fd, ret;
1067 #ifdef INET6
1068     if (msg->msg_type == FF_IOCTL6) {
1069         fd = ff_socket(AF_INET6, SOCK_DGRAM, 0);
1070     } else
1071 #endif
1072         fd = ff_socket(AF_INET, SOCK_DGRAM, 0);
1073 
1074     if (fd < 0) {
1075         ret = -1;
1076         goto done;
1077     }
1078 
1079     ret = ff_ioctl_freebsd(fd, msg->ioctl.cmd, msg->ioctl.data);
1080 
1081     ff_close(fd);
1082 
1083 done:
1084     if (ret < 0) {
1085         msg->result = errno;
1086     } else {
1087         msg->result = 0;
1088     }
1089 }
1090 
1091 static inline void
1092 handle_route_msg(struct ff_msg *msg)
1093 {
1094     int ret = ff_rtioctl(msg->route.fib, msg->route.data,
1095         &msg->route.len, msg->route.maxlen);
1096     if (ret < 0) {
1097         msg->result = errno;
1098     } else {
1099         msg->result = 0;
1100     }
1101 }
1102 
1103 static inline void
1104 handle_top_msg(struct ff_msg *msg)
1105 {
1106     msg->top = ff_top_status;
1107     msg->result = 0;
1108 }
1109 
1110 #ifdef FF_NETGRAPH
1111 static inline void
1112 handle_ngctl_msg(struct ff_msg *msg)
1113 {
1114     int ret = ff_ngctl(msg->ngctl.cmd, msg->ngctl.data);
1115     if (ret < 0) {
1116         msg->result = errno;
1117     } else {
1118         msg->result = 0;
1119         msg->ngctl.ret = ret;
1120     }
1121 }
1122 #endif
1123 
1124 #ifdef FF_IPFW
1125 static inline void
1126 handle_ipfw_msg(struct ff_msg *msg)
1127 {
1128     int fd, ret;
1129     fd = ff_socket(AF_INET, SOCK_RAW, IPPROTO_RAW);
1130     if (fd < 0) {
1131         ret = -1;
1132         goto done;
1133     }
1134 
1135     switch (msg->ipfw.cmd) {
1136         case FF_IPFW_GET:
1137             ret = ff_getsockopt_freebsd(fd, msg->ipfw.level,
1138                 msg->ipfw.optname, msg->ipfw.optval,
1139                 msg->ipfw.optlen);
1140             break;
1141         case FF_IPFW_SET:
1142             ret = ff_setsockopt_freebsd(fd, msg->ipfw.level,
1143                 msg->ipfw.optname, msg->ipfw.optval,
1144                 *(msg->ipfw.optlen));
1145             break;
1146         default:
1147             ret = -1;
1148             errno = ENOTSUP;
1149             break;
1150     }
1151 
1152     ff_close(fd);
1153 
1154 done:
1155     if (ret < 0) {
1156         msg->result = errno;
1157     } else {
1158         msg->result = 0;
1159     }
1160 }
1161 #endif
1162 
1163 static inline void
1164 handle_traffic_msg(struct ff_msg *msg)
1165 {
1166     msg->traffic = ff_traffic;
1167     msg->result = 0;
1168 }
1169 
1170 static inline void
1171 handle_default_msg(struct ff_msg *msg)
1172 {
1173     msg->result = ENOTSUP;
1174 }
1175 
1176 static inline void
1177 handle_msg(struct ff_msg *msg, uint16_t proc_id)
1178 {
1179     switch (msg->msg_type) {
1180         case FF_SYSCTL:
1181             handle_sysctl_msg(msg);
1182             break;
1183         case FF_IOCTL:
1184 #ifdef INET6
1185         case FF_IOCTL6:
1186 #endif
1187             handle_ioctl_msg(msg);
1188             break;
1189         case FF_ROUTE:
1190             handle_route_msg(msg);
1191             break;
1192         case FF_TOP:
1193             handle_top_msg(msg);
1194             break;
1195 #ifdef FF_NETGRAPH
1196         case FF_NGCTL:
1197             handle_ngctl_msg(msg);
1198             break;
1199 #endif
1200 #ifdef FF_IPFW
1201         case FF_IPFW_CTL:
1202             handle_ipfw_msg(msg);
1203             break;
1204 #endif
1205         case FF_TRAFFIC:
1206             handle_traffic_msg(msg);
1207             break;
1208         default:
1209             handle_default_msg(msg);
1210             break;
1211     }
1212     rte_ring_enqueue(msg_ring[proc_id].ring[msg->msg_type], msg);
1213 }
1214 
1215 static inline int
1216 process_msg_ring(uint16_t proc_id)
1217 {
1218     void *msg;
1219     int ret = rte_ring_dequeue(msg_ring[proc_id].ring[0], &msg);
1220 
1221     if (unlikely(ret == 0)) {
1222         handle_msg((struct ff_msg *)msg, proc_id);
1223     }
1224 
1225     return 0;
1226 }
1227 
1228 /* Send burst of packets on an output interface */
1229 static inline int
1230 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
1231 {
1232     struct rte_mbuf **m_table;
1233     int ret;
1234     uint16_t queueid;
1235 
1236     queueid = qconf->tx_queue_id[port];
1237     m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
1238 
1239     if (unlikely(qconf->pcap[port] != NULL)) {
1240         uint16_t i;
1241         for (i = 0; i < n; i++) {
1242             ff_dump_packets(qconf->pcap[port], m_table[i]);
1243         }
1244     }
1245 
1246     ret = rte_eth_tx_burst(port, queueid, m_table, n);
1247     ff_traffic.tx_packets += ret;
1248     uint16_t i;
1249     for (i = 0; i < ret; i++) {
1250         ff_traffic.tx_bytes += rte_pktmbuf_pkt_len(m_table[i]);
1251 #ifdef FF_USE_PAGE_ARRAY
1252         if (qconf->tx_mbufs[port].bsd_m_table[i])
1253             ff_enq_tx_bsdmbuf(port, qconf->tx_mbufs[port].bsd_m_table[i], m_table[i]->nb_segs);
1254 #endif
1255     }
1256     if (unlikely(ret < n)) {
1257         do {
1258             rte_pktmbuf_free(m_table[ret]);
1259 #ifdef FF_USE_PAGE_ARRAY
1260             if ( qconf->tx_mbufs[port].bsd_m_table[ret] )
1261                 ff_mbuf_free(qconf->tx_mbufs[port].bsd_m_table[ret]);
1262 #endif
1263         } while (++ret < n);
1264     }
1265     return 0;
1266 }
1267 
1268 /* Enqueue a single packet, and send burst if queue is filled */
1269 static inline int
1270 send_single_packet(struct rte_mbuf *m, uint8_t port)
1271 {
1272     uint16_t len;
1273     struct lcore_conf *qconf;
1274 
1275     qconf = &lcore_conf;
1276     len = qconf->tx_mbufs[port].len;
1277     qconf->tx_mbufs[port].m_table[len] = m;
1278     len++;
1279 
1280     /* enough pkts to be sent */
1281     if (unlikely(len == MAX_PKT_BURST)) {
1282         send_burst(qconf, MAX_PKT_BURST, port);
1283         len = 0;
1284     }
1285 
1286     qconf->tx_mbufs[port].len = len;
1287     return 0;
1288 }
1289 
1290 int
1291 ff_dpdk_if_send(struct ff_dpdk_if_context *ctx, void *m,
1292     int total)
1293 {
1294 #ifdef FF_USE_PAGE_ARRAY
1295     struct lcore_conf *qconf = &lcore_conf;
1296     int    len = 0;
1297 
1298     len = ff_if_send_onepkt(ctx, m,total);
1299     if (unlikely(len == MAX_PKT_BURST)) {
1300         send_burst(qconf, MAX_PKT_BURST, ctx->port_id);
1301         len = 0;
1302     }
1303     qconf->tx_mbufs[ctx->port_id].len = len;
1304     return 0;
1305 #endif
1306     struct rte_mempool *mbuf_pool = pktmbuf_pool[lcore_conf.socket_id];
1307     struct rte_mbuf *head = rte_pktmbuf_alloc(mbuf_pool);
1308     if (head == NULL) {
1309         ff_mbuf_free(m);
1310         return -1;
1311     }
1312 
1313     head->pkt_len = total;
1314     head->nb_segs = 0;
1315 
1316     int off = 0;
1317     struct rte_mbuf *cur = head, *prev = NULL;
1318     while(total > 0) {
1319         if (cur == NULL) {
1320             cur = rte_pktmbuf_alloc(mbuf_pool);
1321             if (cur == NULL) {
1322                 rte_pktmbuf_free(head);
1323                 ff_mbuf_free(m);
1324                 return -1;
1325             }
1326         }
1327 
1328         if (prev != NULL) {
1329             prev->next = cur;
1330         }
1331         head->nb_segs++;
1332 
1333         prev = cur;
1334         void *data = rte_pktmbuf_mtod(cur, void*);
1335         int len = total > RTE_MBUF_DEFAULT_DATAROOM ? RTE_MBUF_DEFAULT_DATAROOM : total;
1336         int ret = ff_mbuf_copydata(m, data, off, len);
1337         if (ret < 0) {
1338             rte_pktmbuf_free(head);
1339             ff_mbuf_free(m);
1340             return -1;
1341         }
1342 
1343 
1344         cur->data_len = len;
1345         off += len;
1346         total -= len;
1347         cur = NULL;
1348     }
1349 
1350     struct ff_tx_offload offload = {0};
1351     ff_mbuf_tx_offload(m, &offload);
1352 
1353     void *data = rte_pktmbuf_mtod(head, void*);
1354 
1355     if (offload.ip_csum) {
1356         /* ipv6 not supported yet */
1357         struct ipv4_hdr *iph;
1358         int iph_len;
1359         iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN);
1360         iph_len = (iph->version_ihl & 0x0f) << 2;
1361 
1362         head->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
1363         head->l2_len = ETHER_HDR_LEN;
1364         head->l3_len = iph_len;
1365     }
1366 
1367     if (ctx->hw_features.tx_csum_l4) {
1368         struct ipv4_hdr *iph;
1369         int iph_len;
1370         iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN);
1371         iph_len = (iph->version_ihl & 0x0f) << 2;
1372 
1373         if (offload.tcp_csum) {
1374             head->ol_flags |= PKT_TX_TCP_CKSUM;
1375             head->l2_len = ETHER_HDR_LEN;
1376             head->l3_len = iph_len;
1377         }
1378 
1379         /*
1380          *  TCP segmentation offload.
1381          *
1382          *  - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag
1383          *    implies PKT_TX_TCP_CKSUM)
1384          *  - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
1385          *  - if it's IPv4, set the PKT_TX_IP_CKSUM flag and
1386          *    write the IP checksum to 0 in the packet
1387          *  - fill the mbuf offload information: l2_len,
1388          *    l3_len, l4_len, tso_segsz
1389          *  - calculate the pseudo header checksum without taking ip_len
1390          *    in account, and set it in the TCP header. Refer to
1391          *    rte_ipv4_phdr_cksum() and rte_ipv6_phdr_cksum() that can be
1392          *    used as helpers.
1393          */
1394         if (offload.tso_seg_size) {
1395             struct tcp_hdr *tcph;
1396             int tcph_len;
1397             tcph = (struct tcp_hdr *)((char *)iph + iph_len);
1398             tcph_len = (tcph->data_off & 0xf0) >> 2;
1399             tcph->cksum = rte_ipv4_phdr_cksum(iph, PKT_TX_TCP_SEG);
1400 
1401             head->ol_flags |= PKT_TX_TCP_SEG;
1402             head->l4_len = tcph_len;
1403             head->tso_segsz = offload.tso_seg_size;
1404         }
1405 
1406         if (offload.udp_csum) {
1407             head->ol_flags |= PKT_TX_UDP_CKSUM;
1408             head->l2_len = ETHER_HDR_LEN;
1409             head->l3_len = iph_len;
1410         }
1411     }
1412 
1413     ff_mbuf_free(m);
1414 
1415     return send_single_packet(head, ctx->port_id);
1416 }
1417 
1418 static int
1419 main_loop(void *arg)
1420 {
1421     struct loop_routine *lr = (struct loop_routine *)arg;
1422 
1423     struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1424     uint64_t prev_tsc, diff_tsc, cur_tsc, usch_tsc, div_tsc, usr_tsc, sys_tsc, end_tsc, idle_sleep_tsc;
1425     int i, j, nb_rx, idle;
1426     uint16_t port_id, queue_id;
1427     struct lcore_conf *qconf;
1428     uint64_t drain_tsc = 0;
1429     struct ff_dpdk_if_context *ctx;
1430 
1431     if (pkt_tx_delay) {
1432         drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * pkt_tx_delay;
1433     }
1434 
1435     prev_tsc = 0;
1436     usch_tsc = 0;
1437 
1438     qconf = &lcore_conf;
1439 
1440     while (1) {
1441         cur_tsc = rte_rdtsc();
1442         if (unlikely(freebsd_clock.expire < cur_tsc)) {
1443             rte_timer_manage();
1444         }
1445 
1446         idle = 1;
1447         sys_tsc = 0;
1448         usr_tsc = 0;
1449 
1450         /*
1451          * TX burst queue drain
1452          */
1453         diff_tsc = cur_tsc - prev_tsc;
1454         if (unlikely(diff_tsc >= drain_tsc)) {
1455             for (i = 0; i < qconf->nb_tx_port; i++) {
1456                 port_id = qconf->tx_port_id[i];
1457                 if (qconf->tx_mbufs[port_id].len == 0)
1458                     continue;
1459 
1460                 idle = 0;
1461 
1462                 send_burst(qconf,
1463                     qconf->tx_mbufs[port_id].len,
1464                     port_id);
1465                 qconf->tx_mbufs[port_id].len = 0;
1466             }
1467 
1468             prev_tsc = cur_tsc;
1469         }
1470 
1471         /*
1472          * Read packet from RX queues
1473          */
1474         for (i = 0; i < qconf->nb_rx_queue; ++i) {
1475             port_id = qconf->rx_queue_list[i].port_id;
1476             queue_id = qconf->rx_queue_list[i].queue_id;
1477             ctx = veth_ctx[port_id];
1478 
1479 #ifdef FF_KNI
1480             if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) {
1481                 ff_kni_process(port_id, queue_id, pkts_burst, MAX_PKT_BURST);
1482             }
1483 #endif
1484 
1485             process_dispatch_ring(port_id, queue_id, pkts_burst, ctx);
1486 
1487             nb_rx = rte_eth_rx_burst(port_id, queue_id, pkts_burst,
1488                 MAX_PKT_BURST);
1489             if (nb_rx == 0)
1490                 continue;
1491 
1492             idle = 0;
1493 
1494             /* Prefetch first packets */
1495             for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1496                 rte_prefetch0(rte_pktmbuf_mtod(
1497                         pkts_burst[j], void *));
1498             }
1499 
1500             /* Prefetch and handle already prefetched packets */
1501             for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1502                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1503                         j + PREFETCH_OFFSET], void *));
1504                 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0);
1505             }
1506 
1507             /* Handle remaining prefetched packets */
1508             for (; j < nb_rx; j++) {
1509                 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0);
1510             }
1511         }
1512 
1513         process_msg_ring(qconf->proc_id);
1514 
1515         div_tsc = rte_rdtsc();
1516 
1517         if (likely(lr->loop != NULL && (!idle || cur_tsc - usch_tsc >= drain_tsc))) {
1518             usch_tsc = cur_tsc;
1519             lr->loop(lr->arg);
1520         }
1521 
1522         idle_sleep_tsc = rte_rdtsc();
1523         if (likely(idle && idle_sleep)) {
1524             usleep(idle_sleep);
1525             end_tsc = rte_rdtsc();
1526         } else {
1527             end_tsc = idle_sleep_tsc;
1528         }
1529 
1530         if (usch_tsc == cur_tsc) {
1531             usr_tsc = idle_sleep_tsc - div_tsc;
1532         }
1533 
1534         if (!idle) {
1535             sys_tsc = div_tsc - cur_tsc;
1536             ff_top_status.sys_tsc += sys_tsc;
1537         }
1538 
1539         ff_top_status.usr_tsc += usr_tsc;
1540         ff_top_status.work_tsc += end_tsc - cur_tsc;
1541         ff_top_status.idle_tsc += end_tsc - cur_tsc - usr_tsc - sys_tsc;
1542 
1543         ff_top_status.loops++;
1544     }
1545 
1546     return 0;
1547 }
1548 
1549 int
1550 ff_dpdk_if_up(void) {
1551     int i;
1552     struct lcore_conf *qconf = &lcore_conf;
1553     for (i = 0; i < qconf->nb_tx_port; i++) {
1554         uint16_t port_id = qconf->tx_port_id[i];
1555 
1556         struct ff_port_cfg *pconf = &qconf->port_cfgs[port_id];
1557         veth_ctx[port_id] = ff_veth_attach(pconf);
1558         if (veth_ctx[port_id] == NULL) {
1559             rte_exit(EXIT_FAILURE, "ff_veth_attach failed");
1560         }
1561     }
1562 
1563     return 0;
1564 }
1565 
1566 void
1567 ff_dpdk_run(loop_func_t loop, void *arg) {
1568     struct loop_routine *lr = rte_malloc(NULL,
1569         sizeof(struct loop_routine), 0);
1570     lr->loop = loop;
1571     lr->arg = arg;
1572     rte_eal_mp_remote_launch(main_loop, lr, CALL_MASTER);
1573     rte_eal_mp_wait_lcore();
1574     rte_free(lr);
1575 }
1576 
1577 void
1578 ff_dpdk_pktmbuf_free(void *m)
1579 {
1580     rte_pktmbuf_free((struct rte_mbuf *)m);
1581 }
1582 
1583 static uint32_t
1584 toeplitz_hash(unsigned keylen, const uint8_t *key,
1585     unsigned datalen, const uint8_t *data)
1586 {
1587     uint32_t hash = 0, v;
1588     u_int i, b;
1589 
1590     /* XXXRW: Perhaps an assertion about key length vs. data length? */
1591 
1592     v = (key[0]<<24) + (key[1]<<16) + (key[2] <<8) + key[3];
1593     for (i = 0; i < datalen; i++) {
1594         for (b = 0; b < 8; b++) {
1595             if (data[i] & (1<<(7-b)))
1596                 hash ^= v;
1597             v <<= 1;
1598             if ((i + 4) < keylen &&
1599                 (key[i+4] & (1<<(7-b))))
1600                 v |= 1;
1601         }
1602     }
1603     return (hash);
1604 }
1605 
1606 int
1607 ff_rss_check(void *softc, uint32_t saddr, uint32_t daddr,
1608     uint16_t sport, uint16_t dport)
1609 {
1610     struct lcore_conf *qconf = &lcore_conf;
1611     struct ff_dpdk_if_context *ctx = ff_veth_softc_to_hostc(softc);
1612     uint16_t nb_queues = qconf->nb_queue_list[ctx->port_id];
1613 
1614     if (nb_queues <= 1) {
1615         return 1;
1616     }
1617 
1618     uint16_t reta_size = rss_reta_size[ctx->port_id];
1619     uint16_t queueid = qconf->tx_queue_id[ctx->port_id];
1620 
1621     uint8_t data[sizeof(saddr) + sizeof(daddr) + sizeof(sport) +
1622         sizeof(dport)];
1623 
1624     unsigned datalen = 0;
1625 
1626     bcopy(&saddr, &data[datalen], sizeof(saddr));
1627     datalen += sizeof(saddr);
1628 
1629     bcopy(&daddr, &data[datalen], sizeof(daddr));
1630     datalen += sizeof(daddr);
1631 
1632     bcopy(&sport, &data[datalen], sizeof(sport));
1633     datalen += sizeof(sport);
1634 
1635     bcopy(&dport, &data[datalen], sizeof(dport));
1636     datalen += sizeof(dport);
1637 
1638     uint32_t hash = toeplitz_hash(sizeof(default_rsskey_40bytes),
1639         default_rsskey_40bytes, datalen, data);
1640 
1641     return ((hash & (reta_size - 1)) % nb_queues) == queueid;
1642 }
1643 
1644 void
1645 ff_regist_packet_dispatcher(dispatch_func_t func)
1646 {
1647     packet_dispatcher = func;
1648 }
1649 
1650 uint64_t
1651 ff_get_tsc_ns()
1652 {
1653     uint64_t cur_tsc = rte_rdtsc();
1654     uint64_t hz = rte_get_tsc_hz();
1655     return ((double)cur_tsc/(double)hz) * NS_PER_S;
1656 }
1657 
1658