6e56cfb3acdcf782cafd5c5ad16c35f3d9ac843a
[dpdk.git] / examples / l3fwd-vf / main.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44 #include <signal.h>
45
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
48 #include <rte_log.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_memzone.h>
52 #include <rte_eal.h>
53 #include <rte_per_lcore.h>
54 #include <rte_launch.h>
55 #include <rte_atomic.h>
56 #include <rte_spinlock.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
63 #include <rte_pci.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
68 #include <rte_ring.h>
69 #include <rte_mempool.h>
70 #include <rte_mbuf.h>
71 #include <rte_ip.h>
72 #include <rte_tcp.h>
73 #include <rte_udp.h>
74 #include <rte_string_fns.h>
75
76 #define APP_LOOKUP_EXACT_MATCH          0
77 #define APP_LOOKUP_LPM                  1
78 #define DO_RFC_1812_CHECKS
79
80 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
81 #ifndef APP_LOOKUP_METHOD
82 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
83 #endif
84
85 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
86 #include <rte_hash.h>
87 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
88 #include <rte_lpm.h>
89 #else
90 #error "APP_LOOKUP_METHOD set to incorrect value"
91 #endif
92
93 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
94
95 #define MEMPOOL_CACHE_SIZE 256
96
97 /*
98  * This expression is used to calculate the number of mbufs needed depending on user input, taking
99  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
100  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
101  */
102
103 #define NB_MBUF RTE_MAX (                                                                                                                                       \
104                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
105                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
106                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
107                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
108                                 (unsigned)8192)
109
110 /*
111  * RX and TX Prefetch, Host, and Write-back threshold values should be
112  * carefully set for optimal performance. Consult the network
113  * controller's datasheet and supporting DPDK documentation for guidance
114  * on how these parameters should be set.
115  */
116 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
117 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
118 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
119
120 /*
121  * These default values are optimized for use with the Intel(R) 82599 10 GbE
122  * Controller and the DPDK ixgbe PMD. Consider using other values for other
123  * network controllers and/or network drivers.
124  */
125 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
126 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
127 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
128
129 #define MAX_PKT_BURST 32
130 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
131
132 #define NB_SOCKETS 8
133
134 #define SOCKET0 0
135
136 /* Configure how many packets ahead to prefetch, when reading packets */
137 #define PREFETCH_OFFSET 3
138
139 /*
140  * Configurable number of RX/TX ring descriptors
141  */
142 #define RTE_TEST_RX_DESC_DEFAULT 128
143 #define RTE_TEST_TX_DESC_DEFAULT 512
144 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
145 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
146
147 /* ethernet addresses of ports */
148 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
149
150 /* mask of enabled ports */
151 static uint32_t enabled_port_mask = 0;
152 static int numa_on = 1; /**< NUMA is enabled by default. */
153
154 struct mbuf_table {
155         uint16_t len;
156         struct rte_mbuf *m_table[MAX_PKT_BURST];
157 };
158
159 struct lcore_rx_queue {
160         uint8_t port_id;
161         uint8_t queue_id;
162 } __rte_cache_aligned;
163
164 #define MAX_RX_QUEUE_PER_LCORE 16
165 #define MAX_TX_QUEUE_PER_PORT 1
166 #define MAX_RX_QUEUE_PER_PORT 1
167
168 #define MAX_LCORE_PARAMS 1024
169 struct lcore_params {
170         uint8_t port_id;
171         uint8_t queue_id;
172         uint8_t lcore_id;
173 } __rte_cache_aligned;
174
175 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
176 static struct lcore_params lcore_params_array_default[] = {
177         {0, 0, 2},
178         {0, 1, 2},
179         {0, 2, 2},
180         {1, 0, 2},
181         {1, 1, 2},
182         {1, 2, 2},
183         {2, 0, 2},
184         {3, 0, 3},
185         {3, 1, 3},
186 };
187
188 static struct lcore_params * lcore_params = lcore_params_array_default;
189 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
190                                 sizeof(lcore_params_array_default[0]);
191
192 static struct rte_eth_conf port_conf = {
193         .rxmode = {
194                 .mq_mode        = ETH_MQ_RX_RSS,
195                 .max_rx_pkt_len = ETHER_MAX_LEN,
196                 .split_hdr_size = 0,
197                 .header_split   = 0, /**< Header Split disabled */
198                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
199                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
200                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
201                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
202         },
203         .rx_adv_conf = {
204                 .rss_conf = {
205                         .rss_key = NULL,
206                         .rss_hf = ETH_RSS_IP,
207                 },
208         },
209         .txmode = {
210                 .mq_mode = ETH_MQ_TX_NONE,
211         },
212 };
213
214 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
215
216
217 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
218
219 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
220 #include <rte_hash_crc.h>
221 #define DEFAULT_HASH_FUNC       rte_hash_crc
222 #else
223 #include <rte_jhash.h>
224 #define DEFAULT_HASH_FUNC       rte_jhash
225 #endif
226
227 struct ipv4_5tuple {
228         uint32_t ip_dst;
229         uint32_t ip_src;
230         uint16_t port_dst;
231         uint16_t port_src;
232         uint8_t proto;
233 } __attribute__((__packed__));
234
235 struct l3fwd_route {
236         struct ipv4_5tuple key;
237         uint8_t if_out;
238 };
239
240 static struct l3fwd_route l3fwd_route_array[] = {
241         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
242         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
243         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
244         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
245 };
246
247 typedef struct rte_hash lookup_struct_t;
248 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
249
250 #define L3FWD_HASH_ENTRIES      1024
251 struct rte_hash_parameters l3fwd_hash_params = {
252         .name = "l3fwd_hash_0",
253         .entries = L3FWD_HASH_ENTRIES,
254         .bucket_entries = 4,
255         .key_len = sizeof(struct ipv4_5tuple),
256         .hash_func = DEFAULT_HASH_FUNC,
257         .hash_func_init_val = 0,
258         .socket_id = SOCKET0,
259 };
260
261 #define L3FWD_NUM_ROUTES \
262         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
263
264 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
265 #endif
266
267 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
268 struct l3fwd_route {
269         uint32_t ip;
270         uint8_t  depth;
271         uint8_t  if_out;
272 };
273
274 static struct l3fwd_route l3fwd_route_array[] = {
275         {IPv4(1,1,1,0), 24, 0},
276         {IPv4(2,1,1,0), 24, 1},
277         {IPv4(3,1,1,0), 24, 2},
278         {IPv4(4,1,1,0), 24, 3},
279         {IPv4(5,1,1,0), 24, 4},
280         {IPv4(6,1,1,0), 24, 5},
281         {IPv4(7,1,1,0), 24, 6},
282         {IPv4(8,1,1,0), 24, 7},
283 };
284
285 #define L3FWD_NUM_ROUTES \
286         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
287
288 #define L3FWD_LPM_MAX_RULES     1024
289
290 typedef struct rte_lpm lookup_struct_t;
291 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
292 #endif
293
294 struct lcore_conf {
295         uint16_t n_rx_queue;
296         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
297         uint16_t tx_queue_id;
298         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
299         lookup_struct_t * lookup_struct;
300 } __rte_cache_aligned;
301
302 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
303 static rte_spinlock_t spinlock_conf[RTE_MAX_ETHPORTS] = {RTE_SPINLOCK_INITIALIZER};
304 /* Send burst of packets on an output interface */
305 static inline int
306 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
307 {
308         struct rte_mbuf **m_table;
309         int ret;
310         uint16_t queueid;
311
312         queueid = qconf->tx_queue_id;
313         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
314
315         rte_spinlock_lock(&spinlock_conf[port]);
316         ret = rte_eth_tx_burst(port, queueid, m_table, n);
317         rte_spinlock_unlock(&spinlock_conf[port]);
318
319         if (unlikely(ret < n)) {
320                 do {
321                         rte_pktmbuf_free(m_table[ret]);
322                 } while (++ret < n);
323         }
324
325         return 0;
326 }
327
328 /* Enqueue a single packet, and send burst if queue is filled */
329 static inline int
330 send_single_packet(struct rte_mbuf *m, uint8_t port)
331 {
332         uint32_t lcore_id;
333         uint16_t len;
334         struct lcore_conf *qconf;
335
336         lcore_id = rte_lcore_id();
337
338         qconf = &lcore_conf[lcore_id];
339         len = qconf->tx_mbufs[port].len;
340         qconf->tx_mbufs[port].m_table[len] = m;
341         len++;
342
343         /* enough pkts to be sent */
344         if (unlikely(len == MAX_PKT_BURST)) {
345                 send_burst(qconf, MAX_PKT_BURST, port);
346                 len = 0;
347         }
348
349         qconf->tx_mbufs[port].len = len;
350         return 0;
351 }
352
353 #ifdef DO_RFC_1812_CHECKS
354 static inline int
355 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
356 {
357         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
358         /*
359          * 1. The packet length reported by the Link Layer must be large
360          * enough to hold the minimum length legal IP datagram (20 bytes).
361          */
362         if (link_len < sizeof(struct ipv4_hdr))
363                 return -1;
364
365         /* 2. The IP checksum must be correct. */
366         /* this is checked in H/W */
367
368         /*
369          * 3. The IP version number must be 4. If the version number is not 4
370          * then the packet may be another version of IP, such as IPng or
371          * ST-II.
372          */
373         if (((pkt->version_ihl) >> 4) != 4)
374                 return -3;
375         /*
376          * 4. The IP header length field must be large enough to hold the
377          * minimum length legal IP datagram (20 bytes = 5 words).
378          */
379         if ((pkt->version_ihl & 0xf) < 5)
380                 return -4;
381
382         /*
383          * 5. The IP total length field must be large enough to hold the IP
384          * datagram header, whose length is specified in the IP header length
385          * field.
386          */
387         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
388                 return -5;
389
390         return 0;
391 }
392 #endif
393
394 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
395 static void
396 print_key(struct ipv4_5tuple key)
397 {
398         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
399                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
400 }
401
402 static inline uint8_t
403 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
404 {
405         struct ipv4_5tuple key;
406         struct tcp_hdr *tcp;
407         struct udp_hdr *udp;
408         int ret = 0;
409
410         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
411         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
412         key.proto = ipv4_hdr->next_proto_id;
413
414         switch (ipv4_hdr->next_proto_id) {
415         case IPPROTO_TCP:
416                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
417                                         sizeof(struct ipv4_hdr));
418                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
419                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
420                 break;
421
422         case IPPROTO_UDP:
423                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
424                                         sizeof(struct ipv4_hdr));
425                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
426                 key.port_src = rte_be_to_cpu_16(udp->src_port);
427                 break;
428
429         default:
430                 key.port_dst = 0;
431                 key.port_src = 0;
432         }
433
434         /* Find destination port */
435         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
436         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
437 }
438 #endif
439
440 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
441 static inline uint8_t
442 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
443 {
444         uint8_t next_hop;
445
446         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
447                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
448                         next_hop : portid);
449 }
450 #endif
451
452 static inline void
453 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
454 {
455         struct ether_hdr *eth_hdr;
456         struct ipv4_hdr *ipv4_hdr;
457         void *tmp;
458         uint8_t dst_port;
459
460         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
461
462         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
463                                 sizeof(struct ether_hdr));
464
465 #ifdef DO_RFC_1812_CHECKS
466         /* Check to make sure the packet is valid (RFC1812) */
467         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
468                 rte_pktmbuf_free(m);
469                 return;
470         }
471 #endif
472
473         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
474         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
475                 dst_port = portid;
476
477         /* 02:00:00:00:00:xx */
478         tmp = &eth_hdr->d_addr.addr_bytes[0];
479         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
480
481 #ifdef DO_RFC_1812_CHECKS
482         /* Update time to live and header checksum */
483         --(ipv4_hdr->time_to_live);
484         ++(ipv4_hdr->hdr_checksum);
485 #endif
486
487         /* src addr */
488         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
489
490         send_single_packet(m, dst_port);
491
492 }
493
494 /* main processing loop */
495 static int
496 main_loop(__attribute__((unused)) void *dummy)
497 {
498         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
499         unsigned lcore_id;
500         uint64_t prev_tsc, diff_tsc, cur_tsc;
501         int i, j, nb_rx;
502         uint8_t portid, queueid;
503         struct lcore_conf *qconf;
504         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
505
506         prev_tsc = 0;
507
508         lcore_id = rte_lcore_id();
509         qconf = &lcore_conf[lcore_id];
510
511         if (qconf->n_rx_queue == 0) {
512                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
513                 return 0;
514         }
515
516         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
517
518         for (i = 0; i < qconf->n_rx_queue; i++) {
519
520                 portid = qconf->rx_queue_list[i].port_id;
521                 queueid = qconf->rx_queue_list[i].queue_id;
522                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
523                         portid, queueid);
524         }
525
526         while (1) {
527
528                 cur_tsc = rte_rdtsc();
529
530                 /*
531                  * TX burst queue drain
532                  */
533                 diff_tsc = cur_tsc - prev_tsc;
534                 if (unlikely(diff_tsc > drain_tsc)) {
535
536                         /*
537                          * This could be optimized (use queueid instead of
538                          * portid), but it is not called so often
539                          */
540                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
541                                 if (qconf->tx_mbufs[portid].len == 0)
542                                         continue;
543                                 send_burst(&lcore_conf[lcore_id],
544                                         qconf->tx_mbufs[portid].len,
545                                         portid);
546                                 qconf->tx_mbufs[portid].len = 0;
547                         }
548
549                         prev_tsc = cur_tsc;
550                 }
551
552                 /*
553                  * Read packet from RX queues
554                  */
555                 for (i = 0; i < qconf->n_rx_queue; ++i) {
556
557                         portid = qconf->rx_queue_list[i].port_id;
558                         queueid = qconf->rx_queue_list[i].queue_id;
559                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
560
561                         /* Prefetch first packets */
562                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
563                                 rte_prefetch0(rte_pktmbuf_mtod(
564                                                 pkts_burst[j], void *));
565                         }
566
567                         /* Prefetch and forward already prefetched packets */
568                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
569                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
570                                                 j + PREFETCH_OFFSET], void *));
571                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
572                         }
573
574                         /* Forward remaining prefetched packets */
575                         for (; j < nb_rx; j++) {
576                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
577                         }
578                 }
579         }
580 }
581
582 static int
583 check_lcore_params(void)
584 {
585         uint8_t queue, lcore;
586         uint16_t i;
587         int socketid;
588
589         for (i = 0; i < nb_lcore_params; ++i) {
590                 queue = lcore_params[i].queue_id;
591                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
592                         printf("invalid queue number: %hhu\n", queue);
593                         return -1;
594                 }
595                 lcore = lcore_params[i].lcore_id;
596                 if (!rte_lcore_is_enabled(lcore)) {
597                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
598                         return -1;
599                 }
600                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
601                         (numa_on == 0)) {
602                         printf("warning: lcore %hhu is on socket %d with numa off \n",
603                                 lcore, socketid);
604                 }
605         }
606         return 0;
607 }
608
609 static int
610 check_port_config(const unsigned nb_ports)
611 {
612         unsigned portid;
613         uint16_t i;
614
615         for (i = 0; i < nb_lcore_params; ++i) {
616                 portid = lcore_params[i].port_id;
617                 if ((enabled_port_mask & (1 << portid)) == 0) {
618                         printf("port %u is not enabled in port mask\n", portid);
619                         return -1;
620                 }
621                 if (portid >= nb_ports) {
622                         printf("port %u is not present on the board\n", portid);
623                         return -1;
624                 }
625         }
626         return 0;
627 }
628
629 static uint8_t
630 get_port_n_rx_queues(const uint8_t port)
631 {
632         int queue = -1;
633         uint16_t i;
634
635         for (i = 0; i < nb_lcore_params; ++i) {
636                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
637                         queue = lcore_params[i].queue_id;
638         }
639         return (uint8_t)(++queue);
640 }
641
642 static int
643 init_lcore_rx_queues(void)
644 {
645         uint16_t i, nb_rx_queue;
646         uint8_t lcore;
647
648         for (i = 0; i < nb_lcore_params; ++i) {
649                 lcore = lcore_params[i].lcore_id;
650                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
651                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
652                         printf("error: too many queues (%u) for lcore: %u\n",
653                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
654                         return -1;
655                 } else {
656                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
657                                 lcore_params[i].port_id;
658                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
659                                 lcore_params[i].queue_id;
660                         lcore_conf[lcore].n_rx_queue++;
661                 }
662         }
663         return 0;
664 }
665
666 /* display usage */
667 static void
668 print_usage(const char *prgname)
669 {
670         printf ("%s [EAL options] -- -p PORTMASK"
671                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
672                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
673                 "  --config (port,queue,lcore): rx queues configuration\n"
674                 "  --no-numa: optional, disable numa awareness\n",
675                 prgname);
676 }
677
678 /* Custom handling of signals to handle process terminal */
679 static void
680 signal_handler(int signum)
681 {
682         uint8_t portid;
683         uint8_t nb_ports = rte_eth_dev_count();
684
685         /* When we receive a SIGINT signal */
686         if (signum == SIGINT) {
687                 for (portid = 0; portid < nb_ports; portid++) {
688                         /* skip ports that are not enabled */
689                         if ((enabled_port_mask & (1 << portid)) == 0)
690                                 continue;
691                         rte_eth_dev_close(portid);
692                 }
693         }
694         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
695 }
696 static int
697 parse_portmask(const char *portmask)
698 {
699         char *end = NULL;
700         unsigned long pm;
701
702         /* parse hexadecimal string */
703         pm = strtoul(portmask, &end, 16);
704         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
705                 return -1;
706
707         if (pm == 0)
708                 return -1;
709
710         return pm;
711 }
712
713 static int
714 parse_config(const char *q_arg)
715 {
716         char s[256];
717         const char *p, *p0 = q_arg;
718         char *end;
719         enum fieldnames {
720                 FLD_PORT = 0,
721                 FLD_QUEUE,
722                 FLD_LCORE,
723                 _NUM_FLD
724         };
725         unsigned long int_fld[_NUM_FLD];
726         char *str_fld[_NUM_FLD];
727         int i;
728         unsigned size;
729
730         nb_lcore_params = 0;
731
732         while ((p = strchr(p0,'(')) != NULL) {
733                 ++p;
734                 if((p0 = strchr(p,')')) == NULL)
735                         return -1;
736
737                 size = p0 - p;
738                 if(size >= sizeof(s))
739                         return -1;
740
741                 snprintf(s, sizeof(s), "%.*s", size, p);
742                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
743                         return -1;
744                 for (i = 0; i < _NUM_FLD; i++){
745                         errno = 0;
746                         int_fld[i] = strtoul(str_fld[i], &end, 0);
747                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
748                                 return -1;
749                 }
750                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
751                         printf("exceeded max number of lcore params: %hu\n",
752                                 nb_lcore_params);
753                         return -1;
754                 }
755                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
756                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
757                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
758                 ++nb_lcore_params;
759         }
760         lcore_params = lcore_params_array;
761         return 0;
762 }
763
764 /* Parse the argument given in the command line of the application */
765 static int
766 parse_args(int argc, char **argv)
767 {
768         int opt, ret;
769         char **argvopt;
770         int option_index;
771         char *prgname = argv[0];
772         static struct option lgopts[] = {
773                 {"config", 1, 0, 0},
774                 {"no-numa", 0, 0, 0},
775                 {NULL, 0, 0, 0}
776         };
777
778         argvopt = argv;
779
780         while ((opt = getopt_long(argc, argvopt, "p:",
781                                 lgopts, &option_index)) != EOF) {
782
783                 switch (opt) {
784                 /* portmask */
785                 case 'p':
786                         enabled_port_mask = parse_portmask(optarg);
787                         if (enabled_port_mask == 0) {
788                                 printf("invalid portmask\n");
789                                 print_usage(prgname);
790                                 return -1;
791                         }
792                         break;
793
794                 /* long options */
795                 case 0:
796                         if (!strcmp(lgopts[option_index].name, "config")) {
797                                 ret = parse_config(optarg);
798                                 if (ret) {
799                                         printf("invalid config\n");
800                                         print_usage(prgname);
801                                         return -1;
802                                 }
803                         }
804
805                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
806                                 printf("numa is disabled \n");
807                                 numa_on = 0;
808                         }
809                         break;
810
811                 default:
812                         print_usage(prgname);
813                         return -1;
814                 }
815         }
816
817         if (optind >= 0)
818                 argv[optind-1] = prgname;
819
820         ret = optind-1;
821         optind = 0; /* reset getopt lib */
822         return ret;
823 }
824
825 static void
826 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
827 {
828         char buf[ETHER_ADDR_FMT_SIZE];
829         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
830         printf("%s%s", name, buf);
831 }
832
833 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
834 static void
835 setup_hash(int socketid)
836 {
837         unsigned i;
838         int ret;
839         char s[64];
840
841         /* create  hashes */
842         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
843         l3fwd_hash_params.name = s;
844         l3fwd_hash_params.socket_id = socketid;
845         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
846         if (l3fwd_lookup_struct[socketid] == NULL)
847                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
848                                 "socket %d\n", socketid);
849
850         /* populate the hash */
851         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
852                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
853                                 (void *) &l3fwd_route_array[i].key);
854                 if (ret < 0) {
855                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
856                                 "l3fwd hash on socket %d\n", i, socketid);
857                 }
858                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
859                 printf("Hash: Adding key\n");
860                 print_key(l3fwd_route_array[i].key);
861         }
862 }
863 #endif
864
865 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
866 static void
867 setup_lpm(int socketid)
868 {
869         unsigned i;
870         int ret;
871         char s[64];
872
873         /* create the LPM table */
874         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
875         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
876                                 L3FWD_LPM_MAX_RULES, 0);
877         if (l3fwd_lookup_struct[socketid] == NULL)
878                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
879                                 " on socket %d\n", socketid);
880
881         /* populate the LPM table */
882         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
883                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
884                         l3fwd_route_array[i].ip,
885                         l3fwd_route_array[i].depth,
886                         l3fwd_route_array[i].if_out);
887
888                 if (ret < 0) {
889                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
890                                 "l3fwd LPM table on socket %d\n",
891                                 i, socketid);
892                 }
893
894                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
895                         (unsigned)l3fwd_route_array[i].ip,
896                         l3fwd_route_array[i].depth,
897                         l3fwd_route_array[i].if_out);
898         }
899 }
900 #endif
901
902 static int
903 init_mem(unsigned nb_mbuf)
904 {
905         struct lcore_conf *qconf;
906         int socketid;
907         unsigned lcore_id;
908         char s[64];
909
910         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
911                 if (rte_lcore_is_enabled(lcore_id) == 0)
912                         continue;
913
914                 if (numa_on)
915                         socketid = rte_lcore_to_socket_id(lcore_id);
916                 else
917                         socketid = 0;
918
919                 if (socketid >= NB_SOCKETS) {
920                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
921                                 socketid, lcore_id, NB_SOCKETS);
922                 }
923                 if (pktmbuf_pool[socketid] == NULL) {
924                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
925                         pktmbuf_pool[socketid] = rte_pktmbuf_pool_create(s,
926                                 nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
927                                 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
928                         if (pktmbuf_pool[socketid] == NULL)
929                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
930                         else
931                                 printf("Allocated mbuf pool on socket %d\n", socketid);
932
933 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
934                         setup_lpm(socketid);
935 #else
936                         setup_hash(socketid);
937 #endif
938                 }
939                 qconf = &lcore_conf[lcore_id];
940                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
941         }
942         return 0;
943 }
944
945 int
946 main(int argc, char **argv)
947 {
948         struct lcore_conf *qconf;
949         struct rte_eth_dev_info dev_info;
950         struct rte_eth_txconf *txconf;
951         int ret;
952         unsigned nb_ports;
953         uint16_t queueid;
954         unsigned lcore_id;
955         uint32_t nb_lcores;
956         uint16_t n_tx_queue;
957         uint8_t portid, nb_rx_queue, queue, socketid;
958
959         signal(SIGINT, signal_handler);
960         /* init EAL */
961         ret = rte_eal_init(argc, argv);
962         if (ret < 0)
963                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
964         argc -= ret;
965         argv += ret;
966
967         /* parse application arguments (after the EAL ones) */
968         ret = parse_args(argc, argv);
969         if (ret < 0)
970                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
971
972         if (check_lcore_params() < 0)
973                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
974
975         ret = init_lcore_rx_queues();
976         if (ret < 0)
977                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
978
979         nb_ports = rte_eth_dev_count();
980         if (nb_ports > RTE_MAX_ETHPORTS)
981                 nb_ports = RTE_MAX_ETHPORTS;
982
983         if (check_port_config(nb_ports) < 0)
984                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
985
986         nb_lcores = rte_lcore_count();
987
988         /* initialize all ports */
989         for (portid = 0; portid < nb_ports; portid++) {
990                 /* skip ports that are not enabled */
991                 if ((enabled_port_mask & (1 << portid)) == 0) {
992                         printf("\nSkipping disabled port %d\n", portid);
993                         continue;
994                 }
995
996                 /* init port */
997                 printf("Initializing port %d ... ", portid );
998                 fflush(stdout);
999
1000                 /* must always equal(=1) */
1001                 nb_rx_queue = get_port_n_rx_queues(portid);
1002                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1003
1004                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1005                         nb_rx_queue, (unsigned)1 );
1006                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1007                 if (ret < 0)
1008                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1009                                 ret, portid);
1010
1011                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1012                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1013                 printf(", ");
1014
1015                 ret = init_mem(NB_MBUF);
1016                 if (ret < 0)
1017                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1018
1019                 /* init one TX queue */
1020                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1021
1022                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1023                 fflush(stdout);
1024
1025                 rte_eth_dev_info_get(portid, &dev_info);
1026                 txconf = &dev_info.default_txconf;
1027                 if (port_conf.rxmode.jumbo_frame)
1028                         txconf->txq_flags = 0;
1029                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1030                                                  socketid, txconf);
1031                 if (ret < 0)
1032                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1033                                 "port=%d\n", ret, portid);
1034
1035                 printf("\n");
1036         }
1037
1038         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1039                 if (rte_lcore_is_enabled(lcore_id) == 0)
1040                         continue;
1041                 qconf = &lcore_conf[lcore_id];
1042                 qconf->tx_queue_id = 0;
1043
1044                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1045                 fflush(stdout);
1046                 /* init RX queues */
1047                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1048                         portid = qconf->rx_queue_list[queue].port_id;
1049                         queueid = qconf->rx_queue_list[queue].queue_id;
1050
1051                         if (numa_on)
1052                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1053                         else
1054                                 socketid = 0;
1055
1056                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1057                         fflush(stdout);
1058
1059                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1060                                                 socketid, NULL,
1061                                                 pktmbuf_pool[socketid]);
1062                         if (ret < 0)
1063                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1064                                                 "port=%d\n", ret, portid);
1065                 }
1066         }
1067         printf("\n");
1068
1069         /* start ports */
1070         for (portid = 0; portid < nb_ports; portid++) {
1071                 if ((enabled_port_mask & (1 << portid)) == 0) {
1072                         continue;
1073                 }
1074                 /* Start device */
1075                 ret = rte_eth_dev_start(portid);
1076                 if (ret < 0)
1077                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1078                                 ret, portid);
1079
1080                 printf("done: Port %d\n", portid);
1081
1082         }
1083
1084         /* launch per-lcore init on every lcore */
1085         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1086         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1087                 if (rte_eal_wait_lcore(lcore_id) < 0)
1088                         return -1;
1089         }
1090
1091         return 0;
1092 }