xref: /f-stack/lib/ff_dpdk_if.c (revision ca915d33)
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 static 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[2][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[2];
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;
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         snprintf(msg_ring[i].ring_name[1], RTE_RING_NAMESIZE,
446             "%s%u", FF_MSG_RING_OUT, i);
447 
448         msg_ring[i].ring[0] = create_ring(msg_ring[i].ring_name[0],
449             MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ);
450         if (msg_ring[i].ring[0] == NULL)
451             rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[0]);
452 
453         msg_ring[i].ring[1] = create_ring(msg_ring[i].ring_name[1],
454             MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ);
455         if (msg_ring[i].ring[1] == NULL)
456             rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[0]);
457     }
458 
459     return 0;
460 }
461 
462 #ifdef FF_KNI
463 static int
464 init_kni(void)
465 {
466     int nb_ports = rte_eth_dev_count_avail();
467     kni_accept = 0;
468     if(strcasecmp(ff_global_cfg.kni.method, "accept") == 0)
469         kni_accept = 1;
470 
471     ff_kni_init(nb_ports, ff_global_cfg.kni.tcp_port,
472         ff_global_cfg.kni.udp_port);
473 
474     unsigned socket_id = lcore_conf.socket_id;
475     struct rte_mempool *mbuf_pool = pktmbuf_pool[socket_id];
476 
477     nb_ports = ff_global_cfg.dpdk.nb_ports;
478     int i, ret;
479     for (i = 0; i < nb_ports; i++) {
480         uint16_t port_id = ff_global_cfg.dpdk.portid_list[i];
481         ff_kni_alloc(port_id, socket_id, mbuf_pool, KNI_QUEUE_SIZE);
482     }
483 
484     return 0;
485 }
486 #endif
487 
488 static void
489 set_rss_table(uint16_t port_id, uint16_t reta_size, uint16_t nb_queues)
490 {
491     if (reta_size == 0) {
492         return;
493     }
494 
495     int reta_conf_size = RTE_MAX(1, reta_size / RTE_RETA_GROUP_SIZE);
496     struct rte_eth_rss_reta_entry64 reta_conf[reta_conf_size];
497 
498     /* config HW indirection table */
499     unsigned i, j, hash=0;
500     for (i = 0; i < reta_conf_size; i++) {
501         reta_conf[i].mask = ~0ULL;
502         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++) {
503             reta_conf[i].reta[j] = hash++ % nb_queues;
504         }
505     }
506 
507     if (rte_eth_dev_rss_reta_update(port_id, reta_conf, reta_size)) {
508         rte_exit(EXIT_FAILURE, "port[%d], failed to update rss table\n",
509             port_id);
510     }
511 }
512 
513 static int
514 init_port_start(void)
515 {
516     int nb_ports = ff_global_cfg.dpdk.nb_ports;
517     unsigned socketid = 0;
518     struct rte_mempool *mbuf_pool;
519     uint16_t i;
520 
521     for (i = 0; i < nb_ports; i++) {
522         uint16_t port_id = ff_global_cfg.dpdk.portid_list[i];
523         struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[port_id];
524         uint16_t nb_queues = pconf->nb_lcores;
525 
526         struct rte_eth_dev_info dev_info;
527         struct rte_eth_conf port_conf = {0};
528         struct rte_eth_rxconf rxq_conf;
529         struct rte_eth_txconf txq_conf;
530 
531         rte_eth_dev_info_get(port_id, &dev_info);
532 
533         if (nb_queues > dev_info.max_rx_queues) {
534             rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_rx_queues[%d]\n",
535                 nb_queues,
536                 dev_info.max_rx_queues);
537         }
538 
539         if (nb_queues > dev_info.max_tx_queues) {
540             rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_tx_queues[%d]\n",
541                 nb_queues,
542                 dev_info.max_tx_queues);
543         }
544 
545         struct ether_addr addr;
546         rte_eth_macaddr_get(port_id, &addr);
547         printf("Port %u MAC: %02" PRIx8 " %02" PRIx8 " %02" PRIx8
548                    " %02" PRIx8 " %02" PRIx8 " %02" PRIx8 "\n",
549                 (unsigned)port_id,
550                 addr.addr_bytes[0], addr.addr_bytes[1],
551                 addr.addr_bytes[2], addr.addr_bytes[3],
552                 addr.addr_bytes[4], addr.addr_bytes[5]);
553 
554         rte_memcpy(pconf->mac,
555             addr.addr_bytes, ETHER_ADDR_LEN);
556 
557         /* Set RSS mode */
558         uint64_t default_rss_hf = ETH_RSS_PROTO_MASK;
559         port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS;
560         port_conf.rx_adv_conf.rss_conf.rss_hf = default_rss_hf;
561         port_conf.rx_adv_conf.rss_conf.rss_key = default_rsskey_40bytes;
562         port_conf.rx_adv_conf.rss_conf.rss_key_len = 40;
563         port_conf.rx_adv_conf.rss_conf.rss_hf &= dev_info.flow_type_rss_offloads;
564         if (port_conf.rx_adv_conf.rss_conf.rss_hf !=
565                 ETH_RSS_PROTO_MASK) {
566             printf("Port %u modified RSS hash function based on hardware support,"
567                     "requested:%#"PRIx64" configured:%#"PRIx64"\n",
568                     port_id, default_rss_hf,
569                     port_conf.rx_adv_conf.rss_conf.rss_hf);
570         }
571 
572         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) {
573             port_conf.txmode.offloads |=
574                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
575         }
576 
577         /* Set Rx VLAN stripping */
578         if (ff_global_cfg.dpdk.vlan_strip) {
579             if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_STRIP) {
580                 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
581             }
582         }
583 
584         /* Enable HW CRC stripping */
585         port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_KEEP_CRC;
586 
587         /* FIXME: Enable TCP LRO ?*/
588         #if 0
589         if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO) {
590             printf("LRO is supported\n");
591             port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_TCP_LRO;
592             pconf->hw_features.rx_lro = 1;
593         }
594         #endif
595 
596         /* Set Rx checksum checking */
597         if ((dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM) &&
598             (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_UDP_CKSUM) &&
599             (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_CKSUM)) {
600             printf("RX checksum offload supported\n");
601             port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM;
602             pconf->hw_features.rx_csum = 1;
603         }
604 
605         if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)) {
606             printf("TX ip checksum offload supported\n");
607             port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
608             pconf->hw_features.tx_csum_ip = 1;
609         }
610 
611         if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_CKSUM) &&
612             (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_CKSUM)) {
613             printf("TX TCP&UDP checksum offload supported\n");
614             port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM | DEV_TX_OFFLOAD_TCP_CKSUM;
615             pconf->hw_features.tx_csum_l4 = 1;
616         }
617 
618         if (ff_global_cfg.dpdk.tso) {
619             if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) {
620                 printf("TSO is supported\n");
621                 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_TSO;
622                 pconf->hw_features.tx_tso = 1;
623             }
624         } else {
625             printf("TSO is disabled\n");
626         }
627 
628         if (dev_info.reta_size) {
629             /* reta size must be power of 2 */
630             assert((dev_info.reta_size & (dev_info.reta_size - 1)) == 0);
631 
632             rss_reta_size[port_id] = dev_info.reta_size;
633             printf("port[%d]: rss table size: %d\n", port_id,
634                 dev_info.reta_size);
635         }
636 
637         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
638             continue;
639         }
640 
641         int ret = rte_eth_dev_configure(port_id, nb_queues, nb_queues, &port_conf);
642         if (ret != 0) {
643             return ret;
644         }
645 
646         static uint16_t nb_rxd = RX_QUEUE_SIZE;
647         static uint16_t nb_txd = TX_QUEUE_SIZE;
648         ret = rte_eth_dev_adjust_nb_rx_tx_desc(port_id, &nb_rxd, &nb_txd);
649         if (ret < 0)
650             printf("Could not adjust number of descriptors "
651                     "for port%u (%d)\n", (unsigned)port_id, ret);
652 
653         uint16_t q;
654         for (q = 0; q < nb_queues; q++) {
655             if (numa_on) {
656                 uint16_t lcore_id = lcore_conf.port_cfgs[port_id].lcore_list[q];
657                 socketid = rte_lcore_to_socket_id(lcore_id);
658             }
659             mbuf_pool = pktmbuf_pool[socketid];
660 
661             txq_conf = dev_info.default_txconf;
662             txq_conf.offloads = port_conf.txmode.offloads;
663             ret = rte_eth_tx_queue_setup(port_id, q, nb_txd,
664                 socketid, &txq_conf);
665             if (ret < 0) {
666                 return ret;
667             }
668 
669             rxq_conf = dev_info.default_rxconf;
670             rxq_conf.offloads = port_conf.rxmode.offloads;
671             ret = rte_eth_rx_queue_setup(port_id, q, nb_rxd,
672                 socketid, &rxq_conf, mbuf_pool);
673             if (ret < 0) {
674                 return ret;
675             }
676         }
677 
678         ret = rte_eth_dev_start(port_id);
679         if (ret < 0) {
680             return ret;
681         }
682 
683         if (nb_queues > 1) {
684             /* set HW rss hash function to Toeplitz. */
685             if (!rte_eth_dev_filter_supported(port_id, RTE_ETH_FILTER_HASH)) {
686                 struct rte_eth_hash_filter_info info = {0};
687                 info.info_type = RTE_ETH_HASH_FILTER_GLOBAL_CONFIG;
688                 info.info.global_conf.hash_func = RTE_ETH_HASH_FUNCTION_TOEPLITZ;
689 
690                 if (rte_eth_dev_filter_ctrl(port_id, RTE_ETH_FILTER_HASH,
691                     RTE_ETH_FILTER_SET, &info) < 0) {
692                     rte_exit(EXIT_FAILURE, "port[%d] set hash func failed\n",
693                         port_id);
694                 }
695             }
696 
697             set_rss_table(port_id, dev_info.reta_size, nb_queues);
698         }
699 
700         /* Enable RX in promiscuous mode for the Ethernet device. */
701         if (ff_global_cfg.dpdk.promiscuous) {
702             rte_eth_promiscuous_enable(port_id);
703             ret = rte_eth_promiscuous_get(port_id);
704             if (ret == 1) {
705                 printf("set port %u to promiscuous mode ok\n", port_id);
706             } else {
707                 printf("set port %u to promiscuous mode error\n", port_id);
708             }
709         }
710 
711         /* Enable pcap dump */
712         if (pconf->pcap) {
713             ff_enable_pcap(pconf->pcap);
714         }
715     }
716 
717     if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
718         check_all_ports_link_status();
719     }
720 
721     return 0;
722 }
723 
724 static int
725 init_clock(void)
726 {
727     rte_timer_subsystem_init();
728     uint64_t hz = rte_get_timer_hz();
729     uint64_t intrs = MS_PER_S/ff_global_cfg.freebsd.hz;
730     uint64_t tsc = (hz + MS_PER_S - 1) / MS_PER_S*intrs;
731 
732     rte_timer_init(&freebsd_clock);
733     rte_timer_reset(&freebsd_clock, tsc, PERIODICAL,
734         rte_lcore_id(), &ff_hardclock_job, NULL);
735 
736     ff_update_current_ts();
737 
738     return 0;
739 }
740 
741 int
742 ff_dpdk_init(int argc, char **argv)
743 {
744     if (ff_global_cfg.dpdk.nb_procs < 1 ||
745         ff_global_cfg.dpdk.nb_procs > RTE_MAX_LCORE ||
746         ff_global_cfg.dpdk.proc_id >= ff_global_cfg.dpdk.nb_procs ||
747         ff_global_cfg.dpdk.proc_id < 0) {
748         printf("param num_procs[%d] or proc_id[%d] error!\n",
749             ff_global_cfg.dpdk.nb_procs,
750             ff_global_cfg.dpdk.proc_id);
751         exit(1);
752     }
753 
754     int ret = rte_eal_init(argc, argv);
755     if (ret < 0) {
756         rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
757     }
758 
759     numa_on = ff_global_cfg.dpdk.numa_on;
760 
761     idle_sleep = ff_global_cfg.dpdk.idle_sleep;
762     pkt_tx_delay = ff_global_cfg.dpdk.pkt_tx_delay > BURST_TX_DRAIN_US ? \
763         BURST_TX_DRAIN_US : ff_global_cfg.dpdk.pkt_tx_delay;
764 
765     init_lcore_conf();
766 
767     init_mem_pool();
768 
769     init_dispatch_ring();
770 
771     init_msg_ring();
772 
773 #ifdef FF_KNI
774     enable_kni = ff_global_cfg.kni.enable;
775     if (enable_kni) {
776         init_kni();
777     }
778 #endif
779 
780 #ifdef FF_USE_PAGE_ARRAY
781     ff_mmap_init();
782 #endif
783 
784     ret = init_port_start();
785     if (ret < 0) {
786         rte_exit(EXIT_FAILURE, "init_port_start failed\n");
787     }
788 
789     init_clock();
790 
791     return 0;
792 }
793 
794 static void
795 ff_veth_input(const struct ff_dpdk_if_context *ctx, struct rte_mbuf *pkt)
796 {
797     uint8_t rx_csum = ctx->hw_features.rx_csum;
798     if (rx_csum) {
799         if (pkt->ol_flags & (PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD)) {
800             rte_pktmbuf_free(pkt);
801             return;
802         }
803     }
804 
805     /*
806      * FIXME: should we save pkt->vlan_tci
807      * if (pkt->ol_flags & PKT_RX_VLAN_PKT)
808      */
809 
810     void *data = rte_pktmbuf_mtod(pkt, void*);
811     uint16_t len = rte_pktmbuf_data_len(pkt);
812 
813     void *hdr = ff_mbuf_gethdr(pkt, pkt->pkt_len, data, len, rx_csum);
814     if (hdr == NULL) {
815         rte_pktmbuf_free(pkt);
816         return;
817     }
818 
819     struct rte_mbuf *pn = pkt->next;
820     void *prev = hdr;
821     while(pn != NULL) {
822         data = rte_pktmbuf_mtod(pn, void*);
823         len = rte_pktmbuf_data_len(pn);
824 
825         void *mb = ff_mbuf_get(prev, data, len);
826         if (mb == NULL) {
827             ff_mbuf_free(hdr);
828             rte_pktmbuf_free(pkt);
829             return;
830         }
831         pn = pn->next;
832         prev = mb;
833     }
834 
835     ff_veth_process_packet(ctx->ifp, hdr);
836 }
837 
838 static enum FilterReturn
839 protocol_filter(const void *data, uint16_t len)
840 {
841     if(len < ETHER_HDR_LEN)
842         return FILTER_UNKNOWN;
843 
844     const struct ether_hdr *hdr;
845     hdr = (const struct ether_hdr *)data;
846     uint16_t eth_frame_type = rte_be_to_cpu_16(hdr->ether_type);
847 
848     if(eth_frame_type == ETHER_TYPE_ARP)
849         return FILTER_ARP;
850 
851 #ifndef FF_KNI
852     return FILTER_UNKNOWN;
853 #else
854     if (!enable_kni) {
855         return FILTER_UNKNOWN;
856     }
857 
858     if(eth_frame_type != ETHER_TYPE_IPv4
859 #ifdef INET6
860             && eth_frame_type != ETHER_TYPE_IPv6
861 #endif
862             )
863         return FILTER_UNKNOWN;
864 
865     return ff_kni_proto_filter(data + ETHER_HDR_LEN,
866         len - ETHER_HDR_LEN, eth_frame_type);
867 #endif
868 }
869 
870 static inline void
871 pktmbuf_deep_attach(struct rte_mbuf *mi, const struct rte_mbuf *m)
872 {
873     struct rte_mbuf *md;
874     void *src, *dst;
875 
876     dst = rte_pktmbuf_mtod(mi, void *);
877     src = rte_pktmbuf_mtod(m, void *);
878 
879     mi->data_len = m->data_len;
880     rte_memcpy(dst, src, m->data_len);
881 
882     mi->port = m->port;
883     mi->vlan_tci = m->vlan_tci;
884     mi->vlan_tci_outer = m->vlan_tci_outer;
885     mi->tx_offload = m->tx_offload;
886     mi->hash = m->hash;
887     mi->ol_flags = m->ol_flags;
888     mi->packet_type = m->packet_type;
889 }
890 
891 /* copied from rte_pktmbuf_clone */
892 static inline struct rte_mbuf *
893 pktmbuf_deep_clone(const struct rte_mbuf *md,
894     struct rte_mempool *mp)
895 {
896     struct rte_mbuf *mc, *mi, **prev;
897     uint32_t pktlen;
898     uint8_t nseg;
899 
900     if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
901         return NULL;
902 
903     mi = mc;
904     prev = &mi->next;
905     pktlen = md->pkt_len;
906     nseg = 0;
907 
908     do {
909         nseg++;
910         pktmbuf_deep_attach(mi, md);
911         *prev = mi;
912         prev = &mi->next;
913     } while ((md = md->next) != NULL &&
914         (mi = rte_pktmbuf_alloc(mp)) != NULL);
915 
916     *prev = NULL;
917     mc->nb_segs = nseg;
918     mc->pkt_len = pktlen;
919 
920     /* Allocation of new indirect segment failed */
921     if (unlikely (mi == NULL)) {
922         rte_pktmbuf_free(mc);
923         return NULL;
924     }
925 
926     __rte_mbuf_sanity_check(mc, 1);
927     return mc;
928 }
929 
930 static inline void
931 process_packets(uint16_t port_id, uint16_t queue_id, struct rte_mbuf **bufs,
932     uint16_t count, const struct ff_dpdk_if_context *ctx, int pkts_from_ring)
933 {
934     struct lcore_conf *qconf = &lcore_conf;
935     uint16_t nb_queues = qconf->nb_queue_list[port_id];
936 
937     uint16_t i;
938     for (i = 0; i < count; i++) {
939         struct rte_mbuf *rtem = bufs[i];
940 
941         if (unlikely(qconf->pcap[port_id] != NULL)) {
942             if (!pkts_from_ring) {
943                 ff_dump_packets(qconf->pcap[port_id], rtem);
944             }
945         }
946 
947         void *data = rte_pktmbuf_mtod(rtem, void*);
948         uint16_t len = rte_pktmbuf_data_len(rtem);
949 
950         if (!pkts_from_ring) {
951             ff_traffic.rx_packets++;
952             ff_traffic.rx_bytes += len;
953         }
954 
955         if (!pkts_from_ring && packet_dispatcher) {
956             int ret = (*packet_dispatcher)(data, &len, queue_id, nb_queues);
957             if (ret == FF_DISPATCH_RESPONSE) {
958                 rte_pktmbuf_pkt_len(rtem) = rte_pktmbuf_data_len(rtem) = len;
959                 send_single_packet(rtem, port_id);
960                 continue;
961             }
962 
963             if (ret == FF_DISPATCH_ERROR || ret >= nb_queues) {
964                 rte_pktmbuf_free(rtem);
965                 continue;
966             }
967 
968             if (ret != queue_id) {
969                 ret = rte_ring_enqueue(dispatch_ring[port_id][ret], rtem);
970                 if (ret < 0)
971                     rte_pktmbuf_free(rtem);
972 
973                 continue;
974             }
975         }
976 
977         enum FilterReturn filter = protocol_filter(data, len);
978         if (filter == FILTER_ARP) {
979             struct rte_mempool *mbuf_pool;
980             struct rte_mbuf *mbuf_clone;
981             if (!pkts_from_ring) {
982                 uint16_t j;
983                 for(j = 0; j < nb_queues; ++j) {
984                     if(j == queue_id)
985                         continue;
986 
987                     unsigned socket_id = 0;
988                     if (numa_on) {
989                         uint16_t lcore_id = qconf->port_cfgs[port_id].lcore_list[j];
990                         socket_id = rte_lcore_to_socket_id(lcore_id);
991                     }
992                     mbuf_pool = pktmbuf_pool[socket_id];
993                     mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool);
994                     if(mbuf_clone) {
995                         int ret = rte_ring_enqueue(dispatch_ring[port_id][j],
996                             mbuf_clone);
997                         if (ret < 0)
998                             rte_pktmbuf_free(mbuf_clone);
999                     }
1000                 }
1001             }
1002 
1003 #ifdef FF_KNI
1004             if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) {
1005                 mbuf_pool = pktmbuf_pool[qconf->socket_id];
1006                 mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool);
1007                 if(mbuf_clone) {
1008                     ff_kni_enqueue(port_id, mbuf_clone);
1009                 }
1010             }
1011 #endif
1012             ff_veth_input(ctx, rtem);
1013 #ifdef FF_KNI
1014         } else if (enable_kni &&
1015             ((filter == FILTER_KNI && kni_accept) ||
1016             (filter == FILTER_UNKNOWN && !kni_accept)) ) {
1017             ff_kni_enqueue(port_id, rtem);
1018 #endif
1019         } else {
1020             ff_veth_input(ctx, rtem);
1021         }
1022     }
1023 }
1024 
1025 static inline int
1026 process_dispatch_ring(uint16_t port_id, uint16_t queue_id,
1027     struct rte_mbuf **pkts_burst, const struct ff_dpdk_if_context *ctx)
1028 {
1029     /* read packet from ring buf and to process */
1030     uint16_t nb_rb;
1031     nb_rb = rte_ring_dequeue_burst(dispatch_ring[port_id][queue_id],
1032         (void **)pkts_burst, MAX_PKT_BURST, NULL);
1033 
1034     if(nb_rb > 0) {
1035         process_packets(port_id, queue_id, pkts_burst, nb_rb, ctx, 1);
1036     }
1037 
1038     return 0;
1039 }
1040 
1041 static inline void
1042 handle_sysctl_msg(struct ff_msg *msg)
1043 {
1044     int ret = ff_sysctl(msg->sysctl.name, msg->sysctl.namelen,
1045         msg->sysctl.old, msg->sysctl.oldlenp, msg->sysctl.new,
1046         msg->sysctl.newlen);
1047 
1048     if (ret < 0) {
1049         msg->result = errno;
1050     } else {
1051         msg->result = 0;
1052     }
1053 }
1054 
1055 static inline void
1056 handle_ioctl_msg(struct ff_msg *msg)
1057 {
1058     int fd, ret;
1059 #ifdef INET6
1060     if (msg->msg_type == FF_IOCTL6) {
1061         fd = ff_socket(AF_INET6, SOCK_DGRAM, 0);
1062     } else
1063 #endif
1064         fd = ff_socket(AF_INET, SOCK_DGRAM, 0);
1065 
1066     if (fd < 0) {
1067         ret = -1;
1068         goto done;
1069     }
1070 
1071     ret = ff_ioctl_freebsd(fd, msg->ioctl.cmd, msg->ioctl.data);
1072 
1073     ff_close(fd);
1074 
1075 done:
1076     if (ret < 0) {
1077         msg->result = errno;
1078     } else {
1079         msg->result = 0;
1080     }
1081 }
1082 
1083 static inline void
1084 handle_route_msg(struct ff_msg *msg)
1085 {
1086     int ret = ff_rtioctl(msg->route.fib, msg->route.data,
1087         &msg->route.len, msg->route.maxlen);
1088     if (ret < 0) {
1089         msg->result = errno;
1090     } else {
1091         msg->result = 0;
1092     }
1093 }
1094 
1095 static inline void
1096 handle_top_msg(struct ff_msg *msg)
1097 {
1098     msg->top = ff_top_status;
1099     msg->result = 0;
1100 }
1101 
1102 #ifdef FF_NETGRAPH
1103 static inline void
1104 handle_ngctl_msg(struct ff_msg *msg)
1105 {
1106     int ret = ff_ngctl(msg->ngctl.cmd, msg->ngctl.data);
1107     if (ret < 0) {
1108         msg->result = errno;
1109     } else {
1110         msg->result = 0;
1111         msg->ngctl.ret = ret;
1112     }
1113 }
1114 #endif
1115 
1116 #ifdef FF_IPFW
1117 static inline void
1118 handle_ipfw_msg(struct ff_msg *msg)
1119 {
1120     int fd, ret;
1121     fd = ff_socket(AF_INET, SOCK_RAW, IPPROTO_RAW);
1122     if (fd < 0) {
1123         ret = -1;
1124         goto done;
1125     }
1126 
1127     switch (msg->ipfw.cmd) {
1128         case FF_IPFW_GET:
1129             ret = ff_getsockopt_freebsd(fd, msg->ipfw.level,
1130                 msg->ipfw.optname, msg->ipfw.optval,
1131                 msg->ipfw.optlen);
1132             break;
1133         case FF_IPFW_SET:
1134             ret = ff_setsockopt_freebsd(fd, msg->ipfw.level,
1135                 msg->ipfw.optname, msg->ipfw.optval,
1136                 *(msg->ipfw.optlen));
1137             break;
1138         default:
1139             ret = -1;
1140             errno = ENOTSUP;
1141             break;
1142     }
1143 
1144     ff_close(fd);
1145 
1146 done:
1147     if (ret < 0) {
1148         msg->result = errno;
1149     } else {
1150         msg->result = 0;
1151     }
1152 }
1153 #endif
1154 
1155 static inline void
1156 handle_traffic_msg(struct ff_msg *msg)
1157 {
1158     msg->traffic = ff_traffic;
1159     msg->result = 0;
1160 }
1161 
1162 static inline void
1163 handle_default_msg(struct ff_msg *msg)
1164 {
1165     msg->result = ENOTSUP;
1166 }
1167 
1168 static inline void
1169 handle_msg(struct ff_msg *msg, uint16_t proc_id)
1170 {
1171     switch (msg->msg_type) {
1172         case FF_SYSCTL:
1173             handle_sysctl_msg(msg);
1174             break;
1175         case FF_IOCTL:
1176 #ifdef INET6
1177         case FF_IOCTL6:
1178 #endif
1179             handle_ioctl_msg(msg);
1180             break;
1181         case FF_ROUTE:
1182             handle_route_msg(msg);
1183             break;
1184         case FF_TOP:
1185             handle_top_msg(msg);
1186             break;
1187 #ifdef FF_NETGRAPH
1188         case FF_NGCTL:
1189             handle_ngctl_msg(msg);
1190             break;
1191 #endif
1192 #ifdef FF_IPFW
1193         case FF_IPFW_CTL:
1194             handle_ipfw_msg(msg);
1195             break;
1196 #endif
1197         case FF_TRAFFIC:
1198             handle_traffic_msg(msg);
1199             break;
1200         default:
1201             handle_default_msg(msg);
1202             break;
1203     }
1204     rte_ring_enqueue(msg_ring[proc_id].ring[1], msg);
1205 }
1206 
1207 static inline int
1208 process_msg_ring(uint16_t proc_id)
1209 {
1210     void *msg;
1211     int ret = rte_ring_dequeue(msg_ring[proc_id].ring[0], &msg);
1212 
1213     if (unlikely(ret == 0)) {
1214         handle_msg((struct ff_msg *)msg, proc_id);
1215     }
1216 
1217     return 0;
1218 }
1219 
1220 /* Send burst of packets on an output interface */
1221 static inline int
1222 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
1223 {
1224     struct rte_mbuf **m_table;
1225     int ret;
1226     uint16_t queueid;
1227 
1228     queueid = qconf->tx_queue_id[port];
1229     m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
1230 
1231     if (unlikely(qconf->pcap[port] != NULL)) {
1232         uint16_t i;
1233         for (i = 0; i < n; i++) {
1234             ff_dump_packets(qconf->pcap[port], m_table[i]);
1235         }
1236     }
1237 
1238     ret = rte_eth_tx_burst(port, queueid, m_table, n);
1239     ff_traffic.tx_packets += ret;
1240     uint16_t i;
1241     for (i = 0; i < ret; i++) {
1242         ff_traffic.tx_bytes += rte_pktmbuf_pkt_len(m_table[i]);
1243 #ifdef FF_USE_PAGE_ARRAY
1244         if (qconf->tx_mbufs[port].bsd_m_table[i])
1245             ff_enq_tx_bsdmbuf(port, qconf->tx_mbufs[port].bsd_m_table[i], m_table[i]->nb_segs);
1246 #endif
1247     }
1248     if (unlikely(ret < n)) {
1249         do {
1250             rte_pktmbuf_free(m_table[ret]);
1251 #ifdef FF_USE_PAGE_ARRAY
1252             if ( qconf->tx_mbufs[port].bsd_m_table[ret] )
1253                 ff_mbuf_free(qconf->tx_mbufs[port].bsd_m_table[ret]);
1254 #endif
1255         } while (++ret < n);
1256     }
1257     return 0;
1258 }
1259 
1260 /* Enqueue a single packet, and send burst if queue is filled */
1261 static inline int
1262 send_single_packet(struct rte_mbuf *m, uint8_t port)
1263 {
1264     uint16_t len;
1265     struct lcore_conf *qconf;
1266 
1267     qconf = &lcore_conf;
1268     len = qconf->tx_mbufs[port].len;
1269     qconf->tx_mbufs[port].m_table[len] = m;
1270     len++;
1271 
1272     /* enough pkts to be sent */
1273     if (unlikely(len == MAX_PKT_BURST)) {
1274         send_burst(qconf, MAX_PKT_BURST, port);
1275         len = 0;
1276     }
1277 
1278     qconf->tx_mbufs[port].len = len;
1279     return 0;
1280 }
1281 
1282 int
1283 ff_dpdk_if_send(struct ff_dpdk_if_context *ctx, void *m,
1284     int total)
1285 {
1286 #ifdef FF_USE_PAGE_ARRAY
1287     struct lcore_conf *qconf = &lcore_conf;
1288     int    len = 0;
1289 
1290     len = ff_if_send_onepkt(ctx, m,total);
1291     if (unlikely(len == MAX_PKT_BURST)) {
1292         send_burst(qconf, MAX_PKT_BURST, ctx->port_id);
1293         len = 0;
1294     }
1295     qconf->tx_mbufs[ctx->port_id].len = len;
1296     return 0;
1297 #endif
1298     struct rte_mempool *mbuf_pool = pktmbuf_pool[lcore_conf.socket_id];
1299     struct rte_mbuf *head = rte_pktmbuf_alloc(mbuf_pool);
1300     if (head == NULL) {
1301         ff_mbuf_free(m);
1302         return -1;
1303     }
1304 
1305     head->pkt_len = total;
1306     head->nb_segs = 0;
1307 
1308     int off = 0;
1309     struct rte_mbuf *cur = head, *prev = NULL;
1310     while(total > 0) {
1311         if (cur == NULL) {
1312             cur = rte_pktmbuf_alloc(mbuf_pool);
1313             if (cur == NULL) {
1314                 rte_pktmbuf_free(head);
1315                 ff_mbuf_free(m);
1316                 return -1;
1317             }
1318         }
1319 
1320         if (prev != NULL) {
1321             prev->next = cur;
1322         }
1323         head->nb_segs++;
1324 
1325         prev = cur;
1326         void *data = rte_pktmbuf_mtod(cur, void*);
1327         int len = total > RTE_MBUF_DEFAULT_DATAROOM ? RTE_MBUF_DEFAULT_DATAROOM : total;
1328         int ret = ff_mbuf_copydata(m, data, off, len);
1329         if (ret < 0) {
1330             rte_pktmbuf_free(head);
1331             ff_mbuf_free(m);
1332             return -1;
1333         }
1334 
1335 
1336         cur->data_len = len;
1337         off += len;
1338         total -= len;
1339         cur = NULL;
1340     }
1341 
1342     struct ff_tx_offload offload = {0};
1343     ff_mbuf_tx_offload(m, &offload);
1344 
1345     void *data = rte_pktmbuf_mtod(head, void*);
1346 
1347     if (offload.ip_csum) {
1348         /* ipv6 not supported yet */
1349         struct ipv4_hdr *iph;
1350         int iph_len;
1351         iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN);
1352         iph_len = (iph->version_ihl & 0x0f) << 2;
1353 
1354         head->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
1355         head->l2_len = ETHER_HDR_LEN;
1356         head->l3_len = iph_len;
1357     }
1358 
1359     if (ctx->hw_features.tx_csum_l4) {
1360         struct ipv4_hdr *iph;
1361         int iph_len;
1362         iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN);
1363         iph_len = (iph->version_ihl & 0x0f) << 2;
1364 
1365         if (offload.tcp_csum) {
1366             head->ol_flags |= PKT_TX_TCP_CKSUM;
1367             head->l2_len = ETHER_HDR_LEN;
1368             head->l3_len = iph_len;
1369         }
1370 
1371         /*
1372          *  TCP segmentation offload.
1373          *
1374          *  - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag
1375          *    implies PKT_TX_TCP_CKSUM)
1376          *  - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
1377          *  - if it's IPv4, set the PKT_TX_IP_CKSUM flag and
1378          *    write the IP checksum to 0 in the packet
1379          *  - fill the mbuf offload information: l2_len,
1380          *    l3_len, l4_len, tso_segsz
1381          *  - calculate the pseudo header checksum without taking ip_len
1382          *    in account, and set it in the TCP header. Refer to
1383          *    rte_ipv4_phdr_cksum() and rte_ipv6_phdr_cksum() that can be
1384          *    used as helpers.
1385          */
1386         if (offload.tso_seg_size) {
1387             struct tcp_hdr *tcph;
1388             int tcph_len;
1389             tcph = (struct tcp_hdr *)((char *)iph + iph_len);
1390             tcph_len = (tcph->data_off & 0xf0) >> 2;
1391             tcph->cksum = rte_ipv4_phdr_cksum(iph, PKT_TX_TCP_SEG);
1392 
1393             head->ol_flags |= PKT_TX_TCP_SEG;
1394             head->l4_len = tcph_len;
1395             head->tso_segsz = offload.tso_seg_size;
1396         }
1397 
1398         if (offload.udp_csum) {
1399             head->ol_flags |= PKT_TX_UDP_CKSUM;
1400             head->l2_len = ETHER_HDR_LEN;
1401             head->l3_len = iph_len;
1402         }
1403     }
1404 
1405     ff_mbuf_free(m);
1406 
1407     return send_single_packet(head, ctx->port_id);
1408 }
1409 
1410 static int
1411 main_loop(void *arg)
1412 {
1413     struct loop_routine *lr = (struct loop_routine *)arg;
1414 
1415     struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1416     uint64_t prev_tsc, diff_tsc, cur_tsc, usch_tsc, div_tsc, usr_tsc, sys_tsc, end_tsc, idle_sleep_tsc;
1417     int i, j, nb_rx, idle;
1418     uint16_t port_id, queue_id;
1419     struct lcore_conf *qconf;
1420     uint64_t drain_tsc = 0;
1421     struct ff_dpdk_if_context *ctx;
1422 
1423     if (pkt_tx_delay) {
1424         drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * pkt_tx_delay;
1425     }
1426 
1427     prev_tsc = 0;
1428     usch_tsc = 0;
1429 
1430     qconf = &lcore_conf;
1431 
1432     while (1) {
1433         cur_tsc = rte_rdtsc();
1434         if (unlikely(freebsd_clock.expire < cur_tsc)) {
1435             rte_timer_manage();
1436         }
1437 
1438         idle = 1;
1439         sys_tsc = 0;
1440         usr_tsc = 0;
1441 
1442         /*
1443          * TX burst queue drain
1444          */
1445         diff_tsc = cur_tsc - prev_tsc;
1446         if (unlikely(diff_tsc >= drain_tsc)) {
1447             for (i = 0; i < qconf->nb_tx_port; i++) {
1448                 port_id = qconf->tx_port_id[i];
1449                 if (qconf->tx_mbufs[port_id].len == 0)
1450                     continue;
1451 
1452                 idle = 0;
1453 
1454                 send_burst(qconf,
1455                     qconf->tx_mbufs[port_id].len,
1456                     port_id);
1457                 qconf->tx_mbufs[port_id].len = 0;
1458             }
1459 
1460             prev_tsc = cur_tsc;
1461         }
1462 
1463         /*
1464          * Read packet from RX queues
1465          */
1466         for (i = 0; i < qconf->nb_rx_queue; ++i) {
1467             port_id = qconf->rx_queue_list[i].port_id;
1468             queue_id = qconf->rx_queue_list[i].queue_id;
1469             ctx = veth_ctx[port_id];
1470 
1471 #ifdef FF_KNI
1472             if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) {
1473                 ff_kni_process(port_id, queue_id, pkts_burst, MAX_PKT_BURST);
1474             }
1475 #endif
1476 
1477             process_dispatch_ring(port_id, queue_id, pkts_burst, ctx);
1478 
1479             nb_rx = rte_eth_rx_burst(port_id, queue_id, pkts_burst,
1480                 MAX_PKT_BURST);
1481             if (nb_rx == 0)
1482                 continue;
1483 
1484             idle = 0;
1485 
1486             /* Prefetch first packets */
1487             for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1488                 rte_prefetch0(rte_pktmbuf_mtod(
1489                         pkts_burst[j], void *));
1490             }
1491 
1492             /* Prefetch and handle already prefetched packets */
1493             for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1494                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1495                         j + PREFETCH_OFFSET], void *));
1496                 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0);
1497             }
1498 
1499             /* Handle remaining prefetched packets */
1500             for (; j < nb_rx; j++) {
1501                 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0);
1502             }
1503         }
1504 
1505         process_msg_ring(qconf->proc_id);
1506 
1507         div_tsc = rte_rdtsc();
1508 
1509         if (likely(lr->loop != NULL && (!idle || cur_tsc - usch_tsc >= drain_tsc))) {
1510             usch_tsc = cur_tsc;
1511             lr->loop(lr->arg);
1512         }
1513 
1514         idle_sleep_tsc = rte_rdtsc();
1515         if (likely(idle && idle_sleep)) {
1516             usleep(idle_sleep);
1517             end_tsc = rte_rdtsc();
1518         } else {
1519             end_tsc = idle_sleep_tsc;
1520         }
1521 
1522         if (usch_tsc == cur_tsc) {
1523             usr_tsc = idle_sleep_tsc - div_tsc;
1524         }
1525 
1526         if (!idle) {
1527             sys_tsc = div_tsc - cur_tsc;
1528             ff_top_status.sys_tsc += sys_tsc;
1529         }
1530 
1531         ff_top_status.usr_tsc += usr_tsc;
1532         ff_top_status.work_tsc += end_tsc - cur_tsc;
1533         ff_top_status.idle_tsc += end_tsc - cur_tsc - usr_tsc - sys_tsc;
1534 
1535         ff_top_status.loops++;
1536     }
1537 
1538     return 0;
1539 }
1540 
1541 int
1542 ff_dpdk_if_up(void) {
1543     int i;
1544     struct lcore_conf *qconf = &lcore_conf;
1545     for (i = 0; i < qconf->nb_tx_port; i++) {
1546         uint16_t port_id = qconf->tx_port_id[i];
1547 
1548         struct ff_port_cfg *pconf = &qconf->port_cfgs[port_id];
1549         veth_ctx[port_id] = ff_veth_attach(pconf);
1550         if (veth_ctx[port_id] == NULL) {
1551             rte_exit(EXIT_FAILURE, "ff_veth_attach failed");
1552         }
1553     }
1554 
1555     return 0;
1556 }
1557 
1558 void
1559 ff_dpdk_run(loop_func_t loop, void *arg) {
1560     struct loop_routine *lr = rte_malloc(NULL,
1561         sizeof(struct loop_routine), 0);
1562     lr->loop = loop;
1563     lr->arg = arg;
1564     rte_eal_mp_remote_launch(main_loop, lr, CALL_MASTER);
1565     rte_eal_mp_wait_lcore();
1566     rte_free(lr);
1567 }
1568 
1569 void
1570 ff_dpdk_pktmbuf_free(void *m)
1571 {
1572     rte_pktmbuf_free((struct rte_mbuf *)m);
1573 }
1574 
1575 static uint32_t
1576 toeplitz_hash(unsigned keylen, const uint8_t *key,
1577     unsigned datalen, const uint8_t *data)
1578 {
1579     uint32_t hash = 0, v;
1580     u_int i, b;
1581 
1582     /* XXXRW: Perhaps an assertion about key length vs. data length? */
1583 
1584     v = (key[0]<<24) + (key[1]<<16) + (key[2] <<8) + key[3];
1585     for (i = 0; i < datalen; i++) {
1586         for (b = 0; b < 8; b++) {
1587             if (data[i] & (1<<(7-b)))
1588                 hash ^= v;
1589             v <<= 1;
1590             if ((i + 4) < keylen &&
1591                 (key[i+4] & (1<<(7-b))))
1592                 v |= 1;
1593         }
1594     }
1595     return (hash);
1596 }
1597 
1598 int
1599 ff_rss_check(void *softc, uint32_t saddr, uint32_t daddr,
1600     uint16_t sport, uint16_t dport)
1601 {
1602     struct lcore_conf *qconf = &lcore_conf;
1603     struct ff_dpdk_if_context *ctx = ff_veth_softc_to_hostc(softc);
1604     uint16_t nb_queues = qconf->nb_queue_list[ctx->port_id];
1605 
1606     if (nb_queues <= 1) {
1607         return 1;
1608     }
1609 
1610     uint16_t reta_size = rss_reta_size[ctx->port_id];
1611     uint16_t queueid = qconf->tx_queue_id[ctx->port_id];
1612 
1613     uint8_t data[sizeof(saddr) + sizeof(daddr) + sizeof(sport) +
1614         sizeof(dport)];
1615 
1616     unsigned datalen = 0;
1617 
1618     bcopy(&saddr, &data[datalen], sizeof(saddr));
1619     datalen += sizeof(saddr);
1620 
1621     bcopy(&daddr, &data[datalen], sizeof(daddr));
1622     datalen += sizeof(daddr);
1623 
1624     bcopy(&sport, &data[datalen], sizeof(sport));
1625     datalen += sizeof(sport);
1626 
1627     bcopy(&dport, &data[datalen], sizeof(dport));
1628     datalen += sizeof(dport);
1629 
1630     uint32_t hash = toeplitz_hash(sizeof(default_rsskey_40bytes),
1631         default_rsskey_40bytes, datalen, data);
1632 
1633     return ((hash & (reta_size - 1)) % nb_queues) == queueid;
1634 }
1635 
1636 void
1637 ff_regist_packet_dispatcher(dispatch_func_t func)
1638 {
1639     packet_dispatcher = func;
1640 }
1641 
1642 uint64_t
1643 ff_get_tsc_ns()
1644 {
1645     uint64_t cur_tsc = rte_rdtsc();
1646     uint64_t hz = rte_get_tsc_hz();
1647     return ((double)cur_tsc/(double)hz) * NS_PER_S;
1648 }
1649 
1650