1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2015 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 #define _GNU_SOURCE
35 
36 #include <stdio.h>
37 #include <stdlib.h>
38 #include <stdint.h>
39 #include <inttypes.h>
40 #include <sys/types.h>
41 #include <string.h>
42 #include <sys/queue.h>
43 #include <stdarg.h>
44 #include <errno.h>
45 #include <getopt.h>
46 
47 #include <rte_common.h>
48 #include <rte_vect.h>
49 #include <rte_byteorder.h>
50 #include <rte_log.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_eal.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
64 #include <rte_pci.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_ring.h>
70 #include <rte_mempool.h>
71 #include <rte_mbuf.h>
72 #include <rte_ip.h>
73 #include <rte_tcp.h>
74 #include <rte_udp.h>
75 #include <rte_string_fns.h>
76 
77 #include <cmdline_parse.h>
78 #include <cmdline_parse_etheraddr.h>
79 
80 #include <lthread_api.h>
81 
82 #define APP_LOOKUP_EXACT_MATCH          0
83 #define APP_LOOKUP_LPM                  1
84 #define DO_RFC_1812_CHECKS
85 
86 /* Enable cpu-load stats 0-off, 1-on */
87 #define APP_CPU_LOAD                 1
88 
89 #ifndef APP_LOOKUP_METHOD
90 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
91 #endif
92 
93 /*
94  *  When set to zero, simple forwaring path is eanbled.
95  *  When set to one, optimized forwarding path is enabled.
96  *  Note that LPM optimisation path uses SSE4.1 instructions.
97  */
98 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
99 #define ENABLE_MULTI_BUFFER_OPTIMIZE	0
100 #else
101 #define ENABLE_MULTI_BUFFER_OPTIMIZE	1
102 #endif
103 
104 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
105 #include <rte_hash.h>
106 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
107 #include <rte_lpm.h>
108 #include <rte_lpm6.h>
109 #else
110 #error "APP_LOOKUP_METHOD set to incorrect value"
111 #endif
112 
113 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
114 
115 #define MAX_JUMBO_PKT_LEN  9600
116 
117 #define IPV6_ADDR_LEN 16
118 
119 #define MEMPOOL_CACHE_SIZE 256
120 
121 /*
122  * This expression is used to calculate the number of mbufs needed depending on
123  * user input, taking into account memory for rx and tx hardware rings, cache
124  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
125  * NB_MBUF never goes below a minimum value of 8192
126  */
127 
128 #define NB_MBUF RTE_MAX(\
129 		(nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +       \
130 		nb_ports*nb_lcores*MAX_PKT_BURST +                     \
131 		nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +         \
132 		nb_lcores*MEMPOOL_CACHE_SIZE),                         \
133 		(unsigned)8192)
134 
135 #define MAX_PKT_BURST     32
136 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
137 
138 /*
139  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
140  */
141 #define	MAX_TX_BURST  (MAX_PKT_BURST / 2)
142 #define BURST_SIZE    MAX_TX_BURST
143 
144 #define NB_SOCKETS 8
145 
146 /* Configure how many packets ahead to prefetch, when reading packets */
147 #define PREFETCH_OFFSET	3
148 
149 /* Used to mark destination port as 'invalid'. */
150 #define	BAD_PORT	((uint16_t)-1)
151 
152 #define FWDSTEP	4
153 
154 /*
155  * Configurable number of RX/TX ring descriptors
156  */
157 #define RTE_TEST_RX_DESC_DEFAULT 128
158 #define RTE_TEST_TX_DESC_DEFAULT 128
159 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
160 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
161 
162 /* ethernet addresses of ports */
163 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
164 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
165 
166 static __m128i val_eth[RTE_MAX_ETHPORTS];
167 
168 /* replace first 12B of the ethernet header. */
169 #define	MASK_ETH 0x3f
170 
171 /* mask of enabled ports */
172 static uint32_t enabled_port_mask;
173 static int promiscuous_on; /**< $et in promiscuous mode off by default. */
174 static int numa_on = 1;    /**< NUMA is enabled by default. */
175 
176 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
177 static int ipv6;           /**< ipv6 is false by default. */
178 #endif
179 
180 #if (APP_CPU_LOAD == 1)
181 
182 #define MAX_CPU RTE_MAX_LCORE
183 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
184 
185 #define CPU_PROCESS     0
186 #define CPU_POLL        1
187 #define MAX_CPU_COUNTER 2
188 
189 struct cpu_load {
190 	uint16_t       n_cpu;
191 	uint64_t       counter;
192 	uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
193 } __rte_cache_aligned;
194 
195 static struct cpu_load cpu_load;
196 static int cpu_load_lcore_id = -1;
197 
198 #define SET_CPU_BUSY(thread, counter) \
199 		thread->conf.busy[counter] = 1
200 
201 #define SET_CPU_IDLE(thread, counter) \
202 		thread->conf.busy[counter] = 0
203 
204 #define IS_CPU_BUSY(thread, counter) \
205 		(thread->conf.busy[counter] > 0)
206 
207 #else
208 
209 #define SET_CPU_BUSY(thread, counter)
210 #define SET_CPU_IDLE(thread, counter)
211 #define IS_CPU_BUSY(thread, counter) 0
212 
213 #endif
214 
215 struct mbuf_table {
216 	uint16_t len;
217 	struct rte_mbuf *m_table[MAX_PKT_BURST];
218 };
219 
220 struct lcore_rx_queue {
221 	uint8_t port_id;
222 	uint8_t queue_id;
223 } __rte_cache_aligned;
224 
225 #define MAX_RX_QUEUE_PER_LCORE 16
226 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
227 #define MAX_RX_QUEUE_PER_PORT  128
228 
229 #define MAX_LCORE_PARAMS       1024
230 struct rx_thread_params {
231 	uint8_t port_id;
232 	uint8_t queue_id;
233 	uint8_t lcore_id;
234 	uint8_t thread_id;
235 } __rte_cache_aligned;
236 
237 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
238 static struct rx_thread_params rx_thread_params_array_default[] = {
239 	{0, 0, 2, 0},
240 	{0, 1, 2, 1},
241 	{0, 2, 2, 2},
242 	{1, 0, 2, 3},
243 	{1, 1, 2, 4},
244 	{1, 2, 2, 5},
245 	{2, 0, 2, 6},
246 	{3, 0, 3, 7},
247 	{3, 1, 3, 8},
248 };
249 
250 static struct rx_thread_params *rx_thread_params =
251 		rx_thread_params_array_default;
252 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
253 
254 struct tx_thread_params {
255 	uint8_t lcore_id;
256 	uint8_t thread_id;
257 } __rte_cache_aligned;
258 
259 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
260 static struct tx_thread_params tx_thread_params_array_default[] = {
261 	{4, 0},
262 	{5, 1},
263 	{6, 2},
264 	{7, 3},
265 	{8, 4},
266 	{9, 5},
267 	{10, 6},
268 	{11, 7},
269 	{12, 8},
270 };
271 
272 static struct tx_thread_params *tx_thread_params =
273 		tx_thread_params_array_default;
274 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
275 
276 static struct rte_eth_conf port_conf = {
277 	.rxmode = {
278 		.mq_mode = ETH_MQ_RX_RSS,
279 		.max_rx_pkt_len = ETHER_MAX_LEN,
280 		.split_hdr_size = 0,
281 		.header_split   = 0, /**< Header Split disabled */
282 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
283 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
284 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
285 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
286 	},
287 	.rx_adv_conf = {
288 		.rss_conf = {
289 			.rss_key = NULL,
290 			.rss_hf = ETH_RSS_TCP,
291 		},
292 	},
293 	.txmode = {
294 		.mq_mode = ETH_MQ_TX_NONE,
295 	},
296 };
297 
298 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
299 
300 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
301 
302 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
303 #include <rte_hash_crc.h>
304 #define DEFAULT_HASH_FUNC       rte_hash_crc
305 #else
306 #include <rte_jhash.h>
307 #define DEFAULT_HASH_FUNC       rte_jhash
308 #endif
309 
310 struct ipv4_5tuple {
311 	uint32_t ip_dst;
312 	uint32_t ip_src;
313 	uint16_t port_dst;
314 	uint16_t port_src;
315 	uint8_t  proto;
316 } __attribute__((__packed__));
317 
318 union ipv4_5tuple_host {
319 	struct {
320 		uint8_t  pad0;
321 		uint8_t  proto;
322 		uint16_t pad1;
323 		uint32_t ip_src;
324 		uint32_t ip_dst;
325 		uint16_t port_src;
326 		uint16_t port_dst;
327 	};
328 	__m128i xmm;
329 };
330 
331 #define XMM_NUM_IN_IPV6_5TUPLE 3
332 
333 struct ipv6_5tuple {
334 	uint8_t  ip_dst[IPV6_ADDR_LEN];
335 	uint8_t  ip_src[IPV6_ADDR_LEN];
336 	uint16_t port_dst;
337 	uint16_t port_src;
338 	uint8_t  proto;
339 } __attribute__((__packed__));
340 
341 union ipv6_5tuple_host {
342 	struct {
343 		uint16_t pad0;
344 		uint8_t  proto;
345 		uint8_t  pad1;
346 		uint8_t  ip_src[IPV6_ADDR_LEN];
347 		uint8_t  ip_dst[IPV6_ADDR_LEN];
348 		uint16_t port_src;
349 		uint16_t port_dst;
350 		uint64_t reserve;
351 	};
352 	__m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
353 };
354 
355 struct ipv4_l3fwd_route {
356 	struct ipv4_5tuple key;
357 	uint8_t if_out;
358 };
359 
360 struct ipv6_l3fwd_route {
361 	struct ipv6_5tuple key;
362 	uint8_t if_out;
363 };
364 
365 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
366 	{{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
367 	{{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
368 	{{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
369 	{{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
370 };
371 
372 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
373 	{{
374 	{0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
375 	{0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
376 			0x05},
377 	101, 11, IPPROTO_TCP}, 0},
378 
379 	{{
380 	{0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
381 	{0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
382 			0x05},
383 	102, 12, IPPROTO_TCP}, 1},
384 
385 	{{
386 	{0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
387 	{0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
388 			0x05},
389 	101, 11, IPPROTO_TCP}, 2},
390 
391 	{{
392 	{0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
393 	{0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
394 			0x05},
395 	102, 12, IPPROTO_TCP}, 3},
396 };
397 
398 typedef struct rte_hash lookup_struct_t;
399 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
400 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
401 
402 #ifdef RTE_ARCH_X86_64
403 /* default to 4 million hash entries (approx) */
404 #define L3FWD_HASH_ENTRIES (1024*1024*4)
405 #else
406 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
407 #define L3FWD_HASH_ENTRIES (1024*1024*1)
408 #endif
409 #define HASH_ENTRY_NUMBER_DEFAULT 4
410 
411 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
412 
413 static inline uint32_t
414 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
415 		uint32_t init_val)
416 {
417 	const union ipv4_5tuple_host *k;
418 	uint32_t t;
419 	const uint32_t *p;
420 
421 	k = data;
422 	t = k->proto;
423 	p = (const uint32_t *)&k->port_src;
424 
425 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
426 	init_val = rte_hash_crc_4byte(t, init_val);
427 	init_val = rte_hash_crc_4byte(k->ip_src, init_val);
428 	init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
429 	init_val = rte_hash_crc_4byte(*p, init_val);
430 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
431 	init_val = rte_jhash_1word(t, init_val);
432 	init_val = rte_jhash_1word(k->ip_src, init_val);
433 	init_val = rte_jhash_1word(k->ip_dst, init_val);
434 	init_val = rte_jhash_1word(*p, init_val);
435 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
436 	return init_val;
437 }
438 
439 static inline uint32_t
440 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
441 		uint32_t init_val)
442 {
443 	const union ipv6_5tuple_host *k;
444 	uint32_t t;
445 	const uint32_t *p;
446 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
447 	const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
448 	const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
449 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
450 
451 	k = data;
452 	t = k->proto;
453 	p = (const uint32_t *)&k->port_src;
454 
455 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
456 	ip_src0 = (const uint32_t *) k->ip_src;
457 	ip_src1 = (const uint32_t *)(k->ip_src + 4);
458 	ip_src2 = (const uint32_t *)(k->ip_src + 8);
459 	ip_src3 = (const uint32_t *)(k->ip_src + 12);
460 	ip_dst0 = (const uint32_t *) k->ip_dst;
461 	ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
462 	ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
463 	ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
464 	init_val = rte_hash_crc_4byte(t, init_val);
465 	init_val = rte_hash_crc_4byte(*ip_src0, init_val);
466 	init_val = rte_hash_crc_4byte(*ip_src1, init_val);
467 	init_val = rte_hash_crc_4byte(*ip_src2, init_val);
468 	init_val = rte_hash_crc_4byte(*ip_src3, init_val);
469 	init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
470 	init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
471 	init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
472 	init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
473 	init_val = rte_hash_crc_4byte(*p, init_val);
474 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
475 	init_val = rte_jhash_1word(t, init_val);
476 	init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
477 	init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
478 	init_val = rte_jhash_1word(*p, init_val);
479 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
480 	return init_val;
481 }
482 
483 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
484 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
485 
486 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
487 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
488 
489 #endif
490 
491 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
492 struct ipv4_l3fwd_route {
493 	uint32_t ip;
494 	uint8_t  depth;
495 	uint8_t  if_out;
496 };
497 
498 struct ipv6_l3fwd_route {
499 	uint8_t ip[16];
500 	uint8_t depth;
501 	uint8_t if_out;
502 };
503 
504 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
505 	{IPv4(1, 1, 1, 0), 24, 0},
506 	{IPv4(2, 1, 1, 0), 24, 1},
507 	{IPv4(3, 1, 1, 0), 24, 2},
508 	{IPv4(4, 1, 1, 0), 24, 3},
509 	{IPv4(5, 1, 1, 0), 24, 4},
510 	{IPv4(6, 1, 1, 0), 24, 5},
511 	{IPv4(7, 1, 1, 0), 24, 6},
512 	{IPv4(8, 1, 1, 0), 24, 7},
513 };
514 
515 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
516 	{{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
517 	{{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
518 	{{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
519 	{{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
520 	{{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
521 	{{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
522 	{{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
523 	{{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
524 };
525 
526 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
527 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
528 
529 #define IPV4_L3FWD_LPM_MAX_RULES         1024
530 #define IPV6_L3FWD_LPM_MAX_RULES         1024
531 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
532 
533 typedef struct rte_lpm lookup_struct_t;
534 typedef struct rte_lpm6 lookup6_struct_t;
535 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
536 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
537 #endif
538 
539 struct lcore_conf {
540 	lookup_struct_t *ipv4_lookup_struct;
541 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
542 	lookup6_struct_t *ipv6_lookup_struct;
543 #else
544 	lookup_struct_t *ipv6_lookup_struct;
545 #endif
546 	void *data;
547 } __rte_cache_aligned;
548 
549 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
550 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
551 
552 #define MAX_RX_QUEUE_PER_THREAD 16
553 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
554 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
555 #define MAX_RX_QUEUE_PER_PORT   128
556 
557 #define MAX_RX_THREAD 1024
558 #define MAX_TX_THREAD 1024
559 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
560 
561 /**
562  * Producers and consumers threads configuration
563  */
564 static int lthreads_on = 1; /**< Use lthreads for processing*/
565 
566 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
567 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
568 
569 struct thread_conf {
570 	uint16_t lcore_id;      /**< Initial lcore for rx thread */
571 	uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
572 	uint16_t thread_id;     /**< Thread ID */
573 
574 #if (APP_CPU_LOAD > 0)
575 	int busy[MAX_CPU_COUNTER];
576 #endif
577 };
578 
579 struct thread_rx_conf {
580 	struct thread_conf conf;
581 
582 	uint16_t n_rx_queue;
583 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
584 
585 	uint16_t n_ring;        /**< Number of output rings */
586 	struct rte_ring *ring[RTE_MAX_LCORE];
587 	struct lthread_cond *ready[RTE_MAX_LCORE];
588 
589 #if (APP_CPU_LOAD > 0)
590 	int busy[MAX_CPU_COUNTER];
591 #endif
592 } __rte_cache_aligned;
593 
594 uint16_t n_rx_thread;
595 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
596 
597 struct thread_tx_conf {
598 	struct thread_conf conf;
599 
600 	uint16_t tx_queue_id[RTE_MAX_LCORE];
601 	struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
602 
603 	struct rte_ring *ring;
604 	struct lthread_cond **ready;
605 
606 } __rte_cache_aligned;
607 
608 uint16_t n_tx_thread;
609 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
610 
611 /* Send burst of packets on an output interface */
612 static inline int
613 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint8_t port)
614 {
615 	struct rte_mbuf **m_table;
616 	int ret;
617 	uint16_t queueid;
618 
619 	queueid = qconf->tx_queue_id[port];
620 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
621 
622 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
623 	if (unlikely(ret < n)) {
624 		do {
625 			rte_pktmbuf_free(m_table[ret]);
626 		} while (++ret < n);
627 	}
628 
629 	return 0;
630 }
631 
632 /* Enqueue a single packet, and send burst if queue is filled */
633 static inline int
634 send_single_packet(struct rte_mbuf *m, uint8_t port)
635 {
636 	uint16_t len;
637 	struct thread_tx_conf *qconf;
638 
639 	if (lthreads_on)
640 		qconf = (struct thread_tx_conf *)lthread_get_data();
641 	else
642 		qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
643 
644 	len = qconf->tx_mbufs[port].len;
645 	qconf->tx_mbufs[port].m_table[len] = m;
646 	len++;
647 
648 	/* enough pkts to be sent */
649 	if (unlikely(len == MAX_PKT_BURST)) {
650 		send_burst(qconf, MAX_PKT_BURST, port);
651 		len = 0;
652 	}
653 
654 	qconf->tx_mbufs[port].len = len;
655 	return 0;
656 }
657 
658 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
659 	(ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
660 static inline __attribute__((always_inline)) void
661 send_packetsx4(uint8_t port,
662 	struct rte_mbuf *m[], uint32_t num)
663 {
664 	uint32_t len, j, n;
665 	struct thread_tx_conf *qconf;
666 
667 	if (lthreads_on)
668 		qconf = (struct thread_tx_conf *)lthread_get_data();
669 	else
670 		qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
671 
672 	len = qconf->tx_mbufs[port].len;
673 
674 	/*
675 	 * If TX buffer for that queue is empty, and we have enough packets,
676 	 * then send them straightway.
677 	 */
678 	if (num >= MAX_TX_BURST && len == 0) {
679 		n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
680 		if (unlikely(n < num)) {
681 			do {
682 				rte_pktmbuf_free(m[n]);
683 			} while (++n < num);
684 		}
685 		return;
686 	}
687 
688 	/*
689 	 * Put packets into TX buffer for that queue.
690 	 */
691 
692 	n = len + num;
693 	n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
694 
695 	j = 0;
696 	switch (n % FWDSTEP) {
697 	while (j < n) {
698 	case 0:
699 		qconf->tx_mbufs[port].m_table[len + j] = m[j];
700 		j++;
701 	case 3:
702 		qconf->tx_mbufs[port].m_table[len + j] = m[j];
703 		j++;
704 	case 2:
705 		qconf->tx_mbufs[port].m_table[len + j] = m[j];
706 		j++;
707 	case 1:
708 		qconf->tx_mbufs[port].m_table[len + j] = m[j];
709 		j++;
710 	}
711 	}
712 
713 	len += n;
714 
715 	/* enough pkts to be sent */
716 	if (unlikely(len == MAX_PKT_BURST)) {
717 
718 		send_burst(qconf, MAX_PKT_BURST, port);
719 
720 		/* copy rest of the packets into the TX buffer. */
721 		len = num - n;
722 		j = 0;
723 		switch (len % FWDSTEP) {
724 		while (j < len) {
725 		case 0:
726 			qconf->tx_mbufs[port].m_table[j] = m[n + j];
727 			j++;
728 		case 3:
729 			qconf->tx_mbufs[port].m_table[j] = m[n + j];
730 			j++;
731 		case 2:
732 			qconf->tx_mbufs[port].m_table[j] = m[n + j];
733 			j++;
734 		case 1:
735 			qconf->tx_mbufs[port].m_table[j] = m[n + j];
736 			j++;
737 		}
738 		}
739 	}
740 
741 	qconf->tx_mbufs[port].len = len;
742 }
743 #endif /* APP_LOOKUP_LPM */
744 
745 #ifdef DO_RFC_1812_CHECKS
746 static inline int
747 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
748 {
749 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
750 	/*
751 	 * 1. The packet length reported by the Link Layer must be large
752 	 * enough to hold the minimum length legal IP datagram (20 bytes).
753 	 */
754 	if (link_len < sizeof(struct ipv4_hdr))
755 		return -1;
756 
757 	/* 2. The IP checksum must be correct. */
758 	/* this is checked in H/W */
759 
760 	/*
761 	 * 3. The IP version number must be 4. If the version number is not 4
762 	 * then the packet may be another version of IP, such as IPng or
763 	 * ST-II.
764 	 */
765 	if (((pkt->version_ihl) >> 4) != 4)
766 		return -3;
767 	/*
768 	 * 4. The IP header length field must be large enough to hold the
769 	 * minimum length legal IP datagram (20 bytes = 5 words).
770 	 */
771 	if ((pkt->version_ihl & 0xf) < 5)
772 		return -4;
773 
774 	/*
775 	 * 5. The IP total length field must be large enough to hold the IP
776 	 * datagram header, whose length is specified in the IP header length
777 	 * field.
778 	 */
779 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
780 		return -5;
781 
782 	return 0;
783 }
784 #endif
785 
786 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
787 
788 static __m128i mask0;
789 static __m128i mask1;
790 static __m128i mask2;
791 static inline uint8_t
792 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid,
793 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
794 {
795 	int ret = 0;
796 	union ipv4_5tuple_host key;
797 
798 	ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
799 	__m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
800 	/* Get 5 tuple: dst port, src port, dst IP address, src IP address and
801 	   protocol */
802 	key.xmm = _mm_and_si128(data, mask0);
803 	/* Find destination port */
804 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
805 	return (uint8_t)((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
806 }
807 
808 static inline uint8_t
809 get_ipv6_dst_port(void *ipv6_hdr, uint8_t portid,
810 		lookup_struct_t *ipv6_l3fwd_lookup_struct)
811 {
812 	int ret = 0;
813 	union ipv6_5tuple_host key;
814 
815 	ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
816 	__m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
817 	__m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
818 			sizeof(__m128i)));
819 	__m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
820 			sizeof(__m128i) + sizeof(__m128i)));
821 	/* Get part of 5 tuple: src IP address lower 96 bits and protocol */
822 	key.xmm[0] = _mm_and_si128(data0, mask1);
823 	/* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
824 	   higher 32 bits */
825 	key.xmm[1] = data1;
826 	/* Get part of 5 tuple: dst port and src port and dst IP address higher
827 	   32 bits */
828 	key.xmm[2] = _mm_and_si128(data2, mask2);
829 
830 	/* Find destination port */
831 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
832 	return (uint8_t)((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
833 }
834 #endif
835 
836 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
837 
838 static inline uint8_t
839 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid,
840 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
841 {
842 	uint32_t next_hop;
843 
844 	return (uint8_t)((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
845 		rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
846 		&next_hop) == 0) ? next_hop : portid);
847 }
848 
849 static inline uint8_t
850 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid,
851 		lookup6_struct_t *ipv6_l3fwd_lookup_struct)
852 {
853 	uint8_t next_hop;
854 
855 	return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
856 			((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
857 			next_hop : portid);
858 }
859 #endif
860 
861 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid)
862 		__attribute__((unused));
863 
864 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
865 	(ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
866 
867 #define MASK_ALL_PKTS   0xff
868 #define EXCLUDE_1ST_PKT 0xfe
869 #define EXCLUDE_2ND_PKT 0xfd
870 #define EXCLUDE_3RD_PKT 0xfb
871 #define EXCLUDE_4TH_PKT 0xf7
872 #define EXCLUDE_5TH_PKT 0xef
873 #define EXCLUDE_6TH_PKT 0xdf
874 #define EXCLUDE_7TH_PKT 0xbf
875 #define EXCLUDE_8TH_PKT 0x7f
876 
877 static inline void
878 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid)
879 {
880 	struct ether_hdr *eth_hdr[8];
881 	struct ipv4_hdr *ipv4_hdr[8];
882 	uint8_t dst_port[8];
883 	int32_t ret[8];
884 	union ipv4_5tuple_host key[8];
885 	__m128i data[8];
886 
887 	eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
888 	eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
889 	eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
890 	eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
891 	eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
892 	eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
893 	eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
894 	eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
895 
896 	/* Handle IPv4 headers.*/
897 	ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
898 			sizeof(struct ether_hdr));
899 	ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
900 			sizeof(struct ether_hdr));
901 	ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
902 			sizeof(struct ether_hdr));
903 	ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
904 			sizeof(struct ether_hdr));
905 	ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
906 			sizeof(struct ether_hdr));
907 	ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
908 			sizeof(struct ether_hdr));
909 	ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
910 			sizeof(struct ether_hdr));
911 	ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
912 			sizeof(struct ether_hdr));
913 
914 #ifdef DO_RFC_1812_CHECKS
915 	/* Check to make sure the packet is valid (RFC1812) */
916 	uint8_t valid_mask = MASK_ALL_PKTS;
917 
918 	if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
919 		rte_pktmbuf_free(m[0]);
920 		valid_mask &= EXCLUDE_1ST_PKT;
921 	}
922 	if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
923 		rte_pktmbuf_free(m[1]);
924 		valid_mask &= EXCLUDE_2ND_PKT;
925 	}
926 	if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
927 		rte_pktmbuf_free(m[2]);
928 		valid_mask &= EXCLUDE_3RD_PKT;
929 	}
930 	if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
931 		rte_pktmbuf_free(m[3]);
932 		valid_mask &= EXCLUDE_4TH_PKT;
933 	}
934 	if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
935 		rte_pktmbuf_free(m[4]);
936 		valid_mask &= EXCLUDE_5TH_PKT;
937 	}
938 	if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
939 		rte_pktmbuf_free(m[5]);
940 		valid_mask &= EXCLUDE_6TH_PKT;
941 	}
942 	if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
943 		rte_pktmbuf_free(m[6]);
944 		valid_mask &= EXCLUDE_7TH_PKT;
945 	}
946 	if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
947 		rte_pktmbuf_free(m[7]);
948 		valid_mask &= EXCLUDE_8TH_PKT;
949 	}
950 	if (unlikely(valid_mask != MASK_ALL_PKTS)) {
951 		if (valid_mask == 0)
952 			return;
953 
954 		uint8_t i = 0;
955 
956 		for (i = 0; i < 8; i++)
957 			if ((0x1 << i) & valid_mask)
958 				l3fwd_simple_forward(m[i], portid);
959 	}
960 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
961 
962 	data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
963 			sizeof(struct ether_hdr) +
964 			offsetof(struct ipv4_hdr, time_to_live)));
965 	data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
966 			sizeof(struct ether_hdr) +
967 			offsetof(struct ipv4_hdr, time_to_live)));
968 	data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
969 			sizeof(struct ether_hdr) +
970 			offsetof(struct ipv4_hdr, time_to_live)));
971 	data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
972 			sizeof(struct ether_hdr) +
973 			offsetof(struct ipv4_hdr, time_to_live)));
974 	data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
975 			sizeof(struct ether_hdr) +
976 			offsetof(struct ipv4_hdr, time_to_live)));
977 	data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
978 			sizeof(struct ether_hdr) +
979 			offsetof(struct ipv4_hdr, time_to_live)));
980 	data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
981 			sizeof(struct ether_hdr) +
982 			offsetof(struct ipv4_hdr, time_to_live)));
983 	data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
984 			sizeof(struct ether_hdr) +
985 			offsetof(struct ipv4_hdr, time_to_live)));
986 
987 	key[0].xmm = _mm_and_si128(data[0], mask0);
988 	key[1].xmm = _mm_and_si128(data[1], mask0);
989 	key[2].xmm = _mm_and_si128(data[2], mask0);
990 	key[3].xmm = _mm_and_si128(data[3], mask0);
991 	key[4].xmm = _mm_and_si128(data[4], mask0);
992 	key[5].xmm = _mm_and_si128(data[5], mask0);
993 	key[6].xmm = _mm_and_si128(data[6], mask0);
994 	key[7].xmm = _mm_and_si128(data[7], mask0);
995 
996 	const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
997 			&key[4], &key[5], &key[6], &key[7]};
998 
999 	rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1000 			&key_array[0], 8, ret);
1001 	dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1002 	dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1003 	dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1004 	dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1005 	dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1006 	dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1007 	dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1008 	dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1009 
1010 	if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1011 			(enabled_port_mask & 1 << dst_port[0]) == 0)
1012 		dst_port[0] = portid;
1013 	if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1014 			(enabled_port_mask & 1 << dst_port[1]) == 0)
1015 		dst_port[1] = portid;
1016 	if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1017 			(enabled_port_mask & 1 << dst_port[2]) == 0)
1018 		dst_port[2] = portid;
1019 	if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1020 			(enabled_port_mask & 1 << dst_port[3]) == 0)
1021 		dst_port[3] = portid;
1022 	if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1023 			(enabled_port_mask & 1 << dst_port[4]) == 0)
1024 		dst_port[4] = portid;
1025 	if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1026 			(enabled_port_mask & 1 << dst_port[5]) == 0)
1027 		dst_port[5] = portid;
1028 	if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1029 			(enabled_port_mask & 1 << dst_port[6]) == 0)
1030 		dst_port[6] = portid;
1031 	if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1032 			(enabled_port_mask & 1 << dst_port[7]) == 0)
1033 		dst_port[7] = portid;
1034 
1035 #ifdef DO_RFC_1812_CHECKS
1036 	/* Update time to live and header checksum */
1037 	--(ipv4_hdr[0]->time_to_live);
1038 	--(ipv4_hdr[1]->time_to_live);
1039 	--(ipv4_hdr[2]->time_to_live);
1040 	--(ipv4_hdr[3]->time_to_live);
1041 	++(ipv4_hdr[0]->hdr_checksum);
1042 	++(ipv4_hdr[1]->hdr_checksum);
1043 	++(ipv4_hdr[2]->hdr_checksum);
1044 	++(ipv4_hdr[3]->hdr_checksum);
1045 	--(ipv4_hdr[4]->time_to_live);
1046 	--(ipv4_hdr[5]->time_to_live);
1047 	--(ipv4_hdr[6]->time_to_live);
1048 	--(ipv4_hdr[7]->time_to_live);
1049 	++(ipv4_hdr[4]->hdr_checksum);
1050 	++(ipv4_hdr[5]->hdr_checksum);
1051 	++(ipv4_hdr[6]->hdr_checksum);
1052 	++(ipv4_hdr[7]->hdr_checksum);
1053 #endif
1054 
1055 	/* dst addr */
1056 	*(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1057 	*(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1058 	*(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1059 	*(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1060 	*(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1061 	*(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1062 	*(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1063 	*(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1064 
1065 	/* src addr */
1066 	ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1067 	ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1068 	ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1069 	ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1070 	ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1071 	ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1072 	ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1073 	ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1074 
1075 	send_single_packet(m[0], (uint8_t)dst_port[0]);
1076 	send_single_packet(m[1], (uint8_t)dst_port[1]);
1077 	send_single_packet(m[2], (uint8_t)dst_port[2]);
1078 	send_single_packet(m[3], (uint8_t)dst_port[3]);
1079 	send_single_packet(m[4], (uint8_t)dst_port[4]);
1080 	send_single_packet(m[5], (uint8_t)dst_port[5]);
1081 	send_single_packet(m[6], (uint8_t)dst_port[6]);
1082 	send_single_packet(m[7], (uint8_t)dst_port[7]);
1083 
1084 }
1085 
1086 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1087 		__m128i mask1, union ipv6_5tuple_host *key)
1088 {
1089 	__m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1090 			__m128i *, sizeof(struct ether_hdr) +
1091 			offsetof(struct ipv6_hdr, payload_len)));
1092 	__m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1093 			__m128i *, sizeof(struct ether_hdr) +
1094 			offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1095 	__m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1096 			__m128i *, sizeof(struct ether_hdr) +
1097 			offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1098 			sizeof(__m128i)));
1099 	key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1100 	key->xmm[1] = tmpdata1;
1101 	key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1102 }
1103 
1104 static inline void
1105 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid)
1106 {
1107 	int32_t ret[8];
1108 	uint8_t dst_port[8];
1109 	struct ether_hdr *eth_hdr[8];
1110 	union ipv6_5tuple_host key[8];
1111 
1112 	__attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1113 
1114 	eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1115 	eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1116 	eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1117 	eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1118 	eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1119 	eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1120 	eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1121 	eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1122 
1123 	/* Handle IPv6 headers.*/
1124 	ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1125 			sizeof(struct ether_hdr));
1126 	ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1127 			sizeof(struct ether_hdr));
1128 	ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1129 			sizeof(struct ether_hdr));
1130 	ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1131 			sizeof(struct ether_hdr));
1132 	ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1133 			sizeof(struct ether_hdr));
1134 	ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1135 			sizeof(struct ether_hdr));
1136 	ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1137 			sizeof(struct ether_hdr));
1138 	ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1139 			sizeof(struct ether_hdr));
1140 
1141 	get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1142 	get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1143 	get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1144 	get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1145 	get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1146 	get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1147 	get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1148 	get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1149 
1150 	const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1151 			&key[4], &key[5], &key[6], &key[7]};
1152 
1153 	rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1154 			&key_array[0], 4, ret);
1155 	dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1156 	dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1157 	dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1158 	dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1159 	dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1160 	dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1161 	dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1162 	dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1163 
1164 	if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1165 			(enabled_port_mask & 1 << dst_port[0]) == 0)
1166 		dst_port[0] = portid;
1167 	if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1168 			(enabled_port_mask & 1 << dst_port[1]) == 0)
1169 		dst_port[1] = portid;
1170 	if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1171 			(enabled_port_mask & 1 << dst_port[2]) == 0)
1172 		dst_port[2] = portid;
1173 	if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1174 			(enabled_port_mask & 1 << dst_port[3]) == 0)
1175 		dst_port[3] = portid;
1176 	if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1177 			(enabled_port_mask & 1 << dst_port[4]) == 0)
1178 		dst_port[4] = portid;
1179 	if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1180 			(enabled_port_mask & 1 << dst_port[5]) == 0)
1181 		dst_port[5] = portid;
1182 	if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1183 			(enabled_port_mask & 1 << dst_port[6]) == 0)
1184 		dst_port[6] = portid;
1185 	if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1186 			(enabled_port_mask & 1 << dst_port[7]) == 0)
1187 		dst_port[7] = portid;
1188 
1189 	/* dst addr */
1190 	*(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1191 	*(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1192 	*(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1193 	*(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1194 	*(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1195 	*(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1196 	*(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1197 	*(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1198 
1199 	/* src addr */
1200 	ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1201 	ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1202 	ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1203 	ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1204 	ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1205 	ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1206 	ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1207 	ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1208 
1209 	send_single_packet(m[0], (uint8_t)dst_port[0]);
1210 	send_single_packet(m[1], (uint8_t)dst_port[1]);
1211 	send_single_packet(m[2], (uint8_t)dst_port[2]);
1212 	send_single_packet(m[3], (uint8_t)dst_port[3]);
1213 	send_single_packet(m[4], (uint8_t)dst_port[4]);
1214 	send_single_packet(m[5], (uint8_t)dst_port[5]);
1215 	send_single_packet(m[6], (uint8_t)dst_port[6]);
1216 	send_single_packet(m[7], (uint8_t)dst_port[7]);
1217 
1218 }
1219 #endif /* APP_LOOKUP_METHOD */
1220 
1221 static inline __attribute__((always_inline)) void
1222 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid)
1223 {
1224 	struct ether_hdr *eth_hdr;
1225 	struct ipv4_hdr *ipv4_hdr;
1226 	uint8_t dst_port;
1227 
1228 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1229 
1230 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1231 		/* Handle IPv4 headers.*/
1232 		ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1233 				sizeof(struct ether_hdr));
1234 
1235 #ifdef DO_RFC_1812_CHECKS
1236 		/* Check to make sure the packet is valid (RFC1812) */
1237 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1238 			rte_pktmbuf_free(m);
1239 			return;
1240 		}
1241 #endif
1242 
1243 		 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1244 			RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1245 		if (dst_port >= RTE_MAX_ETHPORTS ||
1246 				(enabled_port_mask & 1 << dst_port) == 0)
1247 			dst_port = portid;
1248 
1249 #ifdef DO_RFC_1812_CHECKS
1250 		/* Update time to live and header checksum */
1251 		--(ipv4_hdr->time_to_live);
1252 		++(ipv4_hdr->hdr_checksum);
1253 #endif
1254 		/* dst addr */
1255 		*(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1256 
1257 		/* src addr */
1258 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1259 
1260 		send_single_packet(m, dst_port);
1261 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1262 		/* Handle IPv6 headers.*/
1263 		struct ipv6_hdr *ipv6_hdr;
1264 
1265 		ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1266 				sizeof(struct ether_hdr));
1267 
1268 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1269 				RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1270 
1271 		if (dst_port >= RTE_MAX_ETHPORTS ||
1272 				(enabled_port_mask & 1 << dst_port) == 0)
1273 			dst_port = portid;
1274 
1275 		/* dst addr */
1276 		*(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1277 
1278 		/* src addr */
1279 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1280 
1281 		send_single_packet(m, dst_port);
1282 	} else
1283 		/* Free the mbuf that contains non-IPV4/IPV6 packet */
1284 		rte_pktmbuf_free(m);
1285 }
1286 
1287 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1288 	(ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1289 #ifdef DO_RFC_1812_CHECKS
1290 
1291 #define	IPV4_MIN_VER_IHL	0x45
1292 #define	IPV4_MAX_VER_IHL	0x4f
1293 #define	IPV4_MAX_VER_IHL_DIFF	(IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1294 
1295 /* Minimum value of IPV4 total length (20B) in network byte order. */
1296 #define	IPV4_MIN_LEN_BE	(sizeof(struct ipv4_hdr) << 8)
1297 
1298 /*
1299  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1300  * - The IP version number must be 4.
1301  * - The IP header length field must be large enough to hold the
1302  *    minimum length legal IP datagram (20 bytes = 5 words).
1303  * - The IP total length field must be large enough to hold the IP
1304  *   datagram header, whose length is specified in the IP header length
1305  *   field.
1306  * If we encounter invalid IPV4 packet, then set destination port for it
1307  * to BAD_PORT value.
1308  */
1309 static inline __attribute__((always_inline)) void
1310 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1311 {
1312 	uint8_t ihl;
1313 
1314 	if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1315 		ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1316 
1317 		ipv4_hdr->time_to_live--;
1318 		ipv4_hdr->hdr_checksum++;
1319 
1320 		if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1321 				((uint8_t)ipv4_hdr->total_length == 0 &&
1322 				ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1323 			dp[0] = BAD_PORT;
1324 		}
1325 	}
1326 }
1327 
1328 #else
1329 #define	rfc1812_process(mb, dp, ptype)	do { } while (0)
1330 #endif /* DO_RFC_1812_CHECKS */
1331 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1332 
1333 
1334 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1335 	(ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1336 
1337 static inline __attribute__((always_inline)) uint16_t
1338 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint8_t portid)
1339 {
1340 	uint32_t next_hop_ipv4;
1341 	uint8_t next_hop_ipv6;
1342 	struct ipv6_hdr *ipv6_hdr;
1343 	struct ether_hdr *eth_hdr;
1344 
1345 	if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1346 		return (uint16_t) ((rte_lpm_lookup(
1347 				RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1348 				&next_hop_ipv4) == 0) ? next_hop_ipv4 : portid);
1349 
1350 	} else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1351 
1352 		eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1353 		ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1354 
1355 		return (uint16_t) ((rte_lpm6_lookup(
1356 				RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1357 				ipv6_hdr->dst_addr, &next_hop_ipv6) == 0) ? next_hop_ipv6 :
1358 						portid);
1359 
1360 	}
1361 
1362 	return portid;
1363 }
1364 
1365 static inline void
1366 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint8_t portid)
1367 {
1368 	struct ether_hdr *eth_hdr;
1369 	struct ipv4_hdr *ipv4_hdr;
1370 	uint32_t dst_ipv4;
1371 	uint16_t dp;
1372 	__m128i te, ve;
1373 
1374 	eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1375 	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1376 
1377 	dst_ipv4 = ipv4_hdr->dst_addr;
1378 	dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1379 	dp = get_dst_port(pkt, dst_ipv4, portid);
1380 
1381 	te = _mm_load_si128((__m128i *)eth_hdr);
1382 	ve = val_eth[dp];
1383 
1384 	dst_port[0] = dp;
1385 	rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1386 
1387 	te =  _mm_blend_epi16(te, ve, MASK_ETH);
1388 	_mm_store_si128((__m128i *)eth_hdr, te);
1389 }
1390 
1391 /*
1392  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1393  */
1394 static inline void
1395 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1396 		__m128i *dip,
1397 		uint32_t *ipv4_flag)
1398 {
1399 	struct ipv4_hdr *ipv4_hdr;
1400 	struct ether_hdr *eth_hdr;
1401 	uint32_t x0, x1, x2, x3;
1402 
1403 	eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1404 	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1405 	x0 = ipv4_hdr->dst_addr;
1406 	ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1407 
1408 	eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1409 	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1410 	x1 = ipv4_hdr->dst_addr;
1411 	ipv4_flag[0] &= pkt[1]->packet_type;
1412 
1413 	eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1414 	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1415 	x2 = ipv4_hdr->dst_addr;
1416 	ipv4_flag[0] &= pkt[2]->packet_type;
1417 
1418 	eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1419 	ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1420 	x3 = ipv4_hdr->dst_addr;
1421 	ipv4_flag[0] &= pkt[3]->packet_type;
1422 
1423 	dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1424 }
1425 
1426 /*
1427  * Lookup into LPM for destination port.
1428  * If lookup fails, use incoming port (portid) as destination port.
1429  */
1430 static inline void
1431 processx4_step2(__m128i dip,
1432 		uint32_t ipv4_flag,
1433 		uint8_t portid,
1434 		struct rte_mbuf *pkt[FWDSTEP],
1435 		uint16_t dprt[FWDSTEP])
1436 {
1437 	rte_xmm_t dst;
1438 	const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1439 			4, 5, 6, 7, 0, 1, 2, 3);
1440 
1441 	/* Byte swap 4 IPV4 addresses. */
1442 	dip = _mm_shuffle_epi8(dip, bswap_mask);
1443 
1444 	/* if all 4 packets are IPV4. */
1445 	if (likely(ipv4_flag)) {
1446 		rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1447 				dst.u32, portid);
1448 
1449 		/* get rid of unused upper 16 bit for each dport. */
1450 		dst.x = _mm_packs_epi32(dst.x, dst.x);
1451 		*(uint64_t *)dprt = dst.u64[0];
1452 	} else {
1453 		dst.x = dip;
1454 		dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1455 		dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1456 		dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1457 		dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1458 	}
1459 }
1460 
1461 /*
1462  * Update source and destination MAC addresses in the ethernet header.
1463  * Perform RFC1812 checks and updates for IPV4 packets.
1464  */
1465 static inline void
1466 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1467 {
1468 	__m128i te[FWDSTEP];
1469 	__m128i ve[FWDSTEP];
1470 	__m128i *p[FWDSTEP];
1471 
1472 	p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1473 	p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1474 	p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1475 	p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1476 
1477 	ve[0] = val_eth[dst_port[0]];
1478 	te[0] = _mm_load_si128(p[0]);
1479 
1480 	ve[1] = val_eth[dst_port[1]];
1481 	te[1] = _mm_load_si128(p[1]);
1482 
1483 	ve[2] = val_eth[dst_port[2]];
1484 	te[2] = _mm_load_si128(p[2]);
1485 
1486 	ve[3] = val_eth[dst_port[3]];
1487 	te[3] = _mm_load_si128(p[3]);
1488 
1489 	/* Update first 12 bytes, keep rest bytes intact. */
1490 	te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1491 	te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1492 	te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1493 	te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1494 
1495 	_mm_store_si128(p[0], te[0]);
1496 	_mm_store_si128(p[1], te[1]);
1497 	_mm_store_si128(p[2], te[2]);
1498 	_mm_store_si128(p[3], te[3]);
1499 
1500 	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1501 			&dst_port[0], pkt[0]->packet_type);
1502 	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1503 			&dst_port[1], pkt[1]->packet_type);
1504 	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1505 			&dst_port[2], pkt[2]->packet_type);
1506 	rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1507 			&dst_port[3], pkt[3]->packet_type);
1508 }
1509 
1510 /*
1511  * We group consecutive packets with the same destionation port into one burst.
1512  * To avoid extra latency this is done together with some other packet
1513  * processing, but after we made a final decision about packet's destination.
1514  * To do this we maintain:
1515  * pnum - array of number of consecutive packets with the same dest port for
1516  * each packet in the input burst.
1517  * lp - pointer to the last updated element in the pnum.
1518  * dlp - dest port value lp corresponds to.
1519  */
1520 
1521 #define	GRPSZ	(1 << FWDSTEP)
1522 #define	GRPMSK	(GRPSZ - 1)
1523 
1524 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)	do { \
1525 	if (likely((dlp) == (dcp)[(idx)])) {         \
1526 		(lp)[0]++;                           \
1527 	} else {                                     \
1528 		(dlp) = (dcp)[idx];                  \
1529 		(lp) = (pn) + (idx);                 \
1530 		(lp)[0] = 1;                         \
1531 	}                                            \
1532 } while (0)
1533 
1534 /*
1535  * Group consecutive packets with the same destination port in bursts of 4.
1536  * Suppose we have array of destionation ports:
1537  * dst_port[] = {a, b, c, d,, e, ... }
1538  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1539  * We doing 4 comparisions at once and the result is 4 bit mask.
1540  * This mask is used as an index into prebuild array of pnum values.
1541  */
1542 static inline uint16_t *
1543 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1544 {
1545 	static const struct {
1546 		uint64_t pnum; /* prebuild 4 values for pnum[]. */
1547 		int32_t  idx;  /* index for new last updated elemnet. */
1548 		uint16_t lpv;  /* add value to the last updated element. */
1549 	} gptbl[GRPSZ] = {
1550 	{
1551 		/* 0: a != b, b != c, c != d, d != e */
1552 		.pnum = UINT64_C(0x0001000100010001),
1553 		.idx = 4,
1554 		.lpv = 0,
1555 	},
1556 	{
1557 		/* 1: a == b, b != c, c != d, d != e */
1558 		.pnum = UINT64_C(0x0001000100010002),
1559 		.idx = 4,
1560 		.lpv = 1,
1561 	},
1562 	{
1563 		/* 2: a != b, b == c, c != d, d != e */
1564 		.pnum = UINT64_C(0x0001000100020001),
1565 		.idx = 4,
1566 		.lpv = 0,
1567 	},
1568 	{
1569 		/* 3: a == b, b == c, c != d, d != e */
1570 		.pnum = UINT64_C(0x0001000100020003),
1571 		.idx = 4,
1572 		.lpv = 2,
1573 	},
1574 	{
1575 		/* 4: a != b, b != c, c == d, d != e */
1576 		.pnum = UINT64_C(0x0001000200010001),
1577 		.idx = 4,
1578 		.lpv = 0,
1579 	},
1580 	{
1581 		/* 5: a == b, b != c, c == d, d != e */
1582 		.pnum = UINT64_C(0x0001000200010002),
1583 		.idx = 4,
1584 		.lpv = 1,
1585 	},
1586 	{
1587 		/* 6: a != b, b == c, c == d, d != e */
1588 		.pnum = UINT64_C(0x0001000200030001),
1589 		.idx = 4,
1590 		.lpv = 0,
1591 	},
1592 	{
1593 		/* 7: a == b, b == c, c == d, d != e */
1594 		.pnum = UINT64_C(0x0001000200030004),
1595 		.idx = 4,
1596 		.lpv = 3,
1597 	},
1598 	{
1599 		/* 8: a != b, b != c, c != d, d == e */
1600 		.pnum = UINT64_C(0x0002000100010001),
1601 		.idx = 3,
1602 		.lpv = 0,
1603 	},
1604 	{
1605 		/* 9: a == b, b != c, c != d, d == e */
1606 		.pnum = UINT64_C(0x0002000100010002),
1607 		.idx = 3,
1608 		.lpv = 1,
1609 	},
1610 	{
1611 		/* 0xa: a != b, b == c, c != d, d == e */
1612 		.pnum = UINT64_C(0x0002000100020001),
1613 		.idx = 3,
1614 		.lpv = 0,
1615 	},
1616 	{
1617 		/* 0xb: a == b, b == c, c != d, d == e */
1618 		.pnum = UINT64_C(0x0002000100020003),
1619 		.idx = 3,
1620 		.lpv = 2,
1621 	},
1622 	{
1623 		/* 0xc: a != b, b != c, c == d, d == e */
1624 		.pnum = UINT64_C(0x0002000300010001),
1625 		.idx = 2,
1626 		.lpv = 0,
1627 	},
1628 	{
1629 		/* 0xd: a == b, b != c, c == d, d == e */
1630 		.pnum = UINT64_C(0x0002000300010002),
1631 		.idx = 2,
1632 		.lpv = 1,
1633 	},
1634 	{
1635 		/* 0xe: a != b, b == c, c == d, d == e */
1636 		.pnum = UINT64_C(0x0002000300040001),
1637 		.idx = 1,
1638 		.lpv = 0,
1639 	},
1640 	{
1641 		/* 0xf: a == b, b == c, c == d, d == e */
1642 		.pnum = UINT64_C(0x0002000300040005),
1643 		.idx = 0,
1644 		.lpv = 4,
1645 	},
1646 	};
1647 
1648 	union {
1649 		uint16_t u16[FWDSTEP + 1];
1650 		uint64_t u64;
1651 	} *pnum = (void *)pn;
1652 
1653 	int32_t v;
1654 
1655 	dp1 = _mm_cmpeq_epi16(dp1, dp2);
1656 	dp1 = _mm_unpacklo_epi16(dp1, dp1);
1657 	v = _mm_movemask_ps((__m128)dp1);
1658 
1659 	/* update last port counter. */
1660 	lp[0] += gptbl[v].lpv;
1661 
1662 	/* if dest port value has changed. */
1663 	if (v != GRPMSK) {
1664 		pnum->u64 = gptbl[v].pnum;
1665 		pnum->u16[FWDSTEP] = 1;
1666 		lp = pnum->u16 + gptbl[v].idx;
1667 	}
1668 
1669 	return lp;
1670 }
1671 
1672 #endif /* APP_LOOKUP_METHOD */
1673 
1674 static void
1675 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1676 		uint8_t portid) {
1677 
1678 	int j;
1679 
1680 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1681 	(ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1682 	int32_t k;
1683 	uint16_t dlp;
1684 	uint16_t *lp;
1685 	uint16_t dst_port[MAX_PKT_BURST];
1686 	__m128i dip[MAX_PKT_BURST / FWDSTEP];
1687 	uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1688 	uint16_t pnum[MAX_PKT_BURST + 1];
1689 #endif
1690 
1691 
1692 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1693 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1694 	{
1695 		/*
1696 		 * Send nb_rx - nb_rx%8 packets
1697 		 * in groups of 8.
1698 		 */
1699 		int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1700 
1701 		for (j = 0; j < n; j += 8) {
1702 			uint32_t pkt_type =
1703 				pkts_burst[j]->packet_type &
1704 				pkts_burst[j+1]->packet_type &
1705 				pkts_burst[j+2]->packet_type &
1706 				pkts_burst[j+3]->packet_type &
1707 				pkts_burst[j+4]->packet_type &
1708 				pkts_burst[j+5]->packet_type &
1709 				pkts_burst[j+6]->packet_type &
1710 				pkts_burst[j+7]->packet_type;
1711 			if (pkt_type & RTE_PTYPE_L3_IPV4) {
1712 				simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1713 			} else if (pkt_type &
1714 				RTE_PTYPE_L3_IPV6) {
1715 				simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1716 			} else {
1717 				l3fwd_simple_forward(pkts_burst[j], portid);
1718 				l3fwd_simple_forward(pkts_burst[j+1], portid);
1719 				l3fwd_simple_forward(pkts_burst[j+2], portid);
1720 				l3fwd_simple_forward(pkts_burst[j+3], portid);
1721 				l3fwd_simple_forward(pkts_burst[j+4], portid);
1722 				l3fwd_simple_forward(pkts_burst[j+5], portid);
1723 				l3fwd_simple_forward(pkts_burst[j+6], portid);
1724 				l3fwd_simple_forward(pkts_burst[j+7], portid);
1725 			}
1726 		}
1727 		for (; j < nb_rx ; j++)
1728 			l3fwd_simple_forward(pkts_burst[j], portid);
1729 	}
1730 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1731 
1732 	k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1733 	for (j = 0; j != k; j += FWDSTEP)
1734 		processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1735 				&ipv4_flag[j / FWDSTEP]);
1736 
1737 	k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1738 	for (j = 0; j != k; j += FWDSTEP)
1739 		processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1740 				portid, &pkts_burst[j], &dst_port[j]);
1741 
1742 	/*
1743 	 * Finish packet processing and group consecutive
1744 	 * packets with the same destination port.
1745 	 */
1746 	k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1747 	if (k != 0) {
1748 		__m128i dp1, dp2;
1749 
1750 		lp = pnum;
1751 		lp[0] = 1;
1752 
1753 		processx4_step3(pkts_burst, dst_port);
1754 
1755 		/* dp1: <d[0], d[1], d[2], d[3], ... > */
1756 		dp1 = _mm_loadu_si128((__m128i *)dst_port);
1757 
1758 		for (j = FWDSTEP; j != k; j += FWDSTEP) {
1759 			processx4_step3(&pkts_burst[j], &dst_port[j]);
1760 
1761 			/*
1762 			 * dp2:
1763 			 * <d[j-3], d[j-2], d[j-1], d[j], ... >
1764 			 */
1765 			dp2 = _mm_loadu_si128(
1766 					(__m128i *)&dst_port[j - FWDSTEP + 1]);
1767 			lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1768 
1769 			/*
1770 			 * dp1:
1771 			 * <d[j], d[j+1], d[j+2], d[j+3], ... >
1772 			 */
1773 			dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1774 					sizeof(dst_port[0]));
1775 		}
1776 
1777 		/*
1778 		 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1779 		 */
1780 		dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1781 		lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1782 
1783 		/*
1784 		 * remove values added by the last repeated
1785 		 * dst port.
1786 		 */
1787 		lp[0]--;
1788 		dlp = dst_port[j - 1];
1789 	} else {
1790 		/* set dlp and lp to the never used values. */
1791 		dlp = BAD_PORT - 1;
1792 		lp = pnum + MAX_PKT_BURST;
1793 	}
1794 
1795 	/* Process up to last 3 packets one by one. */
1796 	switch (nb_rx % FWDSTEP) {
1797 	case 3:
1798 		process_packet(pkts_burst[j], dst_port + j, portid);
1799 		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1800 		j++;
1801 	case 2:
1802 		process_packet(pkts_burst[j], dst_port + j, portid);
1803 		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1804 		j++;
1805 	case 1:
1806 		process_packet(pkts_burst[j], dst_port + j, portid);
1807 		GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1808 		j++;
1809 	}
1810 
1811 	/*
1812 	 * Send packets out, through destination port.
1813 	 * Consecuteve pacekts with the same destination port
1814 	 * are already grouped together.
1815 	 * If destination port for the packet equals BAD_PORT,
1816 	 * then free the packet without sending it out.
1817 	 */
1818 	for (j = 0; j < nb_rx; j += k) {
1819 
1820 		int32_t m;
1821 		uint16_t pn;
1822 
1823 		pn = dst_port[j];
1824 		k = pnum[j];
1825 
1826 		if (likely(pn != BAD_PORT))
1827 			send_packetsx4(pn, pkts_burst + j, k);
1828 		else
1829 			for (m = j; m != j + k; m++)
1830 				rte_pktmbuf_free(pkts_burst[m]);
1831 
1832 	}
1833 
1834 #endif /* APP_LOOKUP_METHOD */
1835 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1836 
1837 	/* Prefetch first packets */
1838 	for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1839 		rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1840 
1841 	/* Prefetch and forward already prefetched packets */
1842 	for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1843 		rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1844 				j + PREFETCH_OFFSET], void *));
1845 		l3fwd_simple_forward(pkts_burst[j], portid);
1846 	}
1847 
1848 	/* Forward remaining prefetched packets */
1849 	for (; j < nb_rx; j++)
1850 		l3fwd_simple_forward(pkts_burst[j], portid);
1851 
1852 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1853 
1854 }
1855 
1856 #if (APP_CPU_LOAD > 0)
1857 
1858 /*
1859  * CPU-load stats collector
1860  */
1861 static int
1862 cpu_load_collector(__rte_unused void *arg) {
1863 	unsigned i, j, k;
1864 	uint64_t hits;
1865 	uint64_t prev_tsc, diff_tsc, cur_tsc;
1866 	uint64_t total[MAX_CPU] = { 0 };
1867 	unsigned min_cpu = MAX_CPU;
1868 	unsigned max_cpu = 0;
1869 	unsigned cpu_id;
1870 	int busy_total = 0;
1871 	int busy_flag = 0;
1872 
1873 	unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1874 	struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1875 
1876 	struct thread_conf *thread_conf;
1877 
1878 	const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1879 		US_PER_S * CPU_LOAD_TIMEOUT_US;
1880 
1881 	prev_tsc = 0;
1882 	/*
1883 	 * Wait for all threads
1884 	 */
1885 
1886 	printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1887 			n_tx_thread);
1888 
1889 	while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1890 		rte_pause();
1891 
1892 	while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1893 		rte_pause();
1894 
1895 	for (i = 0; i < n_rx_thread; i++) {
1896 
1897 		thread_conf = &rx_thread[i].conf;
1898 		cpu_id = thread_conf->cpu_id;
1899 		thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1900 
1901 		if (cpu_id > max_cpu)
1902 			max_cpu = cpu_id;
1903 		if (cpu_id < min_cpu)
1904 			min_cpu = cpu_id;
1905 	}
1906 	for (i = 0; i < n_tx_thread; i++) {
1907 
1908 		thread_conf = &tx_thread[i].conf;
1909 		cpu_id = thread_conf->cpu_id;
1910 		thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1911 
1912 		if (thread_conf->cpu_id > max_cpu)
1913 			max_cpu = thread_conf->cpu_id;
1914 		if (thread_conf->cpu_id < min_cpu)
1915 			min_cpu = thread_conf->cpu_id;
1916 	}
1917 
1918 	while (1) {
1919 
1920 		cpu_load.counter++;
1921 		for (i = min_cpu; i <= max_cpu; i++) {
1922 			for (j = 0; j < MAX_CPU_COUNTER; j++) {
1923 				for (k = 0; k < n_thread_per_cpu[i]; k++)
1924 					if (thread_per_cpu[i][k]->busy[j]) {
1925 						busy_flag = 1;
1926 						break;
1927 					}
1928 				if (busy_flag) {
1929 					cpu_load.hits[j][i]++;
1930 					busy_total = 1;
1931 					busy_flag = 0;
1932 				}
1933 			}
1934 
1935 			if (busy_total) {
1936 				total[i]++;
1937 				busy_total = 0;
1938 			}
1939 		}
1940 
1941 		cur_tsc = rte_rdtsc();
1942 
1943 		diff_tsc = cur_tsc - prev_tsc;
1944 		if (unlikely(diff_tsc > interval_tsc)) {
1945 
1946 			printf("\033c");
1947 
1948 			printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1949 					n_rx_thread, n_tx_thread);
1950 
1951 			printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1952 
1953 			for (i = min_cpu; i <= max_cpu; i++) {
1954 				hits = 0;
1955 				printf("CPU %d:", i);
1956 				for (j = 0; j < MAX_CPU_COUNTER; j++) {
1957 					printf("%7" PRIu64 "",
1958 							cpu_load.hits[j][i] * 100 / cpu_load.counter);
1959 					hits += cpu_load.hits[j][i];
1960 					cpu_load.hits[j][i] = 0;
1961 				}
1962 				printf("%7" PRIu64 "\n",
1963 						100 - total[i] * 100 / cpu_load.counter);
1964 				total[i] = 0;
1965 			}
1966 			cpu_load.counter = 0;
1967 
1968 			prev_tsc = cur_tsc;
1969 		}
1970 
1971 	}
1972 }
1973 #endif /* APP_CPU_LOAD */
1974 
1975 /*
1976  * Null processing lthread loop
1977  *
1978  * This loop is used to start empty scheduler on lcore.
1979  */
1980 static void
1981 lthread_null(__rte_unused void *args)
1982 {
1983 	int lcore_id = rte_lcore_id();
1984 
1985 	RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
1986 	lthread_exit(NULL);
1987 }
1988 
1989 /* main processing loop */
1990 static void
1991 lthread_tx_per_ring(void *dummy)
1992 {
1993 	int nb_rx;
1994 	uint8_t portid;
1995 	struct rte_ring *ring;
1996 	struct thread_tx_conf *tx_conf;
1997 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1998 	struct lthread_cond *ready;
1999 
2000 	tx_conf = (struct thread_tx_conf *)dummy;
2001 	ring = tx_conf->ring;
2002 	ready = *tx_conf->ready;
2003 
2004 	lthread_set_data((void *)tx_conf);
2005 
2006 	/*
2007 	 * Move this lthread to lcore
2008 	 */
2009 	lthread_set_affinity(tx_conf->conf.lcore_id);
2010 
2011 	RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2012 
2013 	nb_rx = 0;
2014 	rte_atomic16_inc(&tx_counter);
2015 	while (1) {
2016 
2017 		/*
2018 		 * Read packet from ring
2019 		 */
2020 		SET_CPU_BUSY(tx_conf, CPU_POLL);
2021 		nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2022 				MAX_PKT_BURST);
2023 		SET_CPU_IDLE(tx_conf, CPU_POLL);
2024 
2025 		if (nb_rx > 0) {
2026 			SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2027 			portid = pkts_burst[0]->port;
2028 			process_burst(pkts_burst, nb_rx, portid);
2029 			SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2030 			lthread_yield();
2031 		} else
2032 			lthread_cond_wait(ready, 0);
2033 
2034 	}
2035 }
2036 
2037 /*
2038  * Main tx-lthreads spawner lthread.
2039  *
2040  * This lthread is used to spawn one new lthread per ring from producers.
2041  *
2042  */
2043 static void
2044 lthread_tx(void *args)
2045 {
2046 	struct lthread *lt;
2047 
2048 	unsigned lcore_id;
2049 	uint8_t portid;
2050 	struct thread_tx_conf *tx_conf;
2051 
2052 	tx_conf = (struct thread_tx_conf *)args;
2053 	lthread_set_data((void *)tx_conf);
2054 
2055 	/*
2056 	 * Move this lthread to the selected lcore
2057 	 */
2058 	lthread_set_affinity(tx_conf->conf.lcore_id);
2059 
2060 	/*
2061 	 * Spawn tx readers (one per input ring)
2062 	 */
2063 	lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2064 			(void *)tx_conf);
2065 
2066 	lcore_id = rte_lcore_id();
2067 
2068 	RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2069 
2070 	tx_conf->conf.cpu_id = sched_getcpu();
2071 	while (1) {
2072 
2073 		lthread_sleep(BURST_TX_DRAIN_US * 1000);
2074 
2075 		/*
2076 		 * TX burst queue drain
2077 		 */
2078 		for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2079 			if (tx_conf->tx_mbufs[portid].len == 0)
2080 				continue;
2081 			SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2082 			send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2083 			SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2084 			tx_conf->tx_mbufs[portid].len = 0;
2085 		}
2086 
2087 	}
2088 }
2089 
2090 static void
2091 lthread_rx(void *dummy)
2092 {
2093 	int ret;
2094 	uint16_t nb_rx;
2095 	int i;
2096 	uint8_t portid, queueid;
2097 	int worker_id;
2098 	int len[RTE_MAX_LCORE] = { 0 };
2099 	int old_len, new_len;
2100 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2101 	struct thread_rx_conf *rx_conf;
2102 
2103 	rx_conf = (struct thread_rx_conf *)dummy;
2104 	lthread_set_data((void *)rx_conf);
2105 
2106 	/*
2107 	 * Move this lthread to lcore
2108 	 */
2109 	lthread_set_affinity(rx_conf->conf.lcore_id);
2110 
2111 	if (rx_conf->n_rx_queue == 0) {
2112 		RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2113 		return;
2114 	}
2115 
2116 	RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2117 
2118 	for (i = 0; i < rx_conf->n_rx_queue; i++) {
2119 
2120 		portid = rx_conf->rx_queue_list[i].port_id;
2121 		queueid = rx_conf->rx_queue_list[i].queue_id;
2122 		RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
2123 				rte_lcore_id(), portid, queueid);
2124 	}
2125 
2126 	/*
2127 	 * Init all condition variables (one per rx thread)
2128 	 */
2129 	for (i = 0; i < rx_conf->n_rx_queue; i++)
2130 		lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2131 
2132 	worker_id = 0;
2133 
2134 	rx_conf->conf.cpu_id = sched_getcpu();
2135 	rte_atomic16_inc(&rx_counter);
2136 	while (1) {
2137 
2138 		/*
2139 		 * Read packet from RX queues
2140 		 */
2141 		for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2142 			portid = rx_conf->rx_queue_list[i].port_id;
2143 			queueid = rx_conf->rx_queue_list[i].queue_id;
2144 
2145 			SET_CPU_BUSY(rx_conf, CPU_POLL);
2146 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2147 				MAX_PKT_BURST);
2148 			SET_CPU_IDLE(rx_conf, CPU_POLL);
2149 
2150 			if (nb_rx != 0) {
2151 				worker_id = (worker_id + 1) % rx_conf->n_ring;
2152 				old_len = len[worker_id];
2153 
2154 				SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2155 				ret = rte_ring_sp_enqueue_burst(
2156 						rx_conf->ring[worker_id],
2157 						(void **) pkts_burst,
2158 						nb_rx);
2159 
2160 				new_len = old_len + ret;
2161 
2162 				if (new_len >= BURST_SIZE) {
2163 					lthread_cond_signal(rx_conf->ready[worker_id]);
2164 					new_len = 0;
2165 				}
2166 
2167 				len[worker_id] = new_len;
2168 
2169 				if (unlikely(ret < nb_rx)) {
2170 					uint32_t k;
2171 
2172 					for (k = ret; k < nb_rx; k++) {
2173 						struct rte_mbuf *m = pkts_burst[k];
2174 
2175 						rte_pktmbuf_free(m);
2176 					}
2177 				}
2178 				SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2179 			}
2180 
2181 			lthread_yield();
2182 		}
2183 	}
2184 }
2185 
2186 /*
2187  * Start scheduler with initial lthread on lcore
2188  *
2189  * This lthread loop spawns all rx and tx lthreads on master lcore
2190  */
2191 
2192 static void
2193 lthread_spawner(__rte_unused void *arg) {
2194 	struct lthread *lt[MAX_THREAD];
2195 	int i;
2196 	int n_thread = 0;
2197 
2198 	printf("Entering lthread_spawner\n");
2199 
2200 	/*
2201 	 * Create producers (rx threads) on default lcore
2202 	 */
2203 	for (i = 0; i < n_rx_thread; i++) {
2204 		rx_thread[i].conf.thread_id = i;
2205 		lthread_create(&lt[n_thread], -1, lthread_rx,
2206 				(void *)&rx_thread[i]);
2207 		n_thread++;
2208 	}
2209 
2210 	/*
2211 	 * Wait for all producers. Until some producers can be started on the same
2212 	 * scheduler as this lthread, yielding is required to let them to run and
2213 	 * prevent deadlock here.
2214 	 */
2215 	while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2216 		lthread_sleep(100000);
2217 
2218 	/*
2219 	 * Create consumers (tx threads) on default lcore_id
2220 	 */
2221 	for (i = 0; i < n_tx_thread; i++) {
2222 		tx_thread[i].conf.thread_id = i;
2223 		lthread_create(&lt[n_thread], -1, lthread_tx,
2224 				(void *)&tx_thread[i]);
2225 		n_thread++;
2226 	}
2227 
2228 	/*
2229 	 * Wait for all threads finished
2230 	 */
2231 	for (i = 0; i < n_thread; i++)
2232 		lthread_join(lt[i], NULL);
2233 
2234 }
2235 
2236 /*
2237  * Start master scheduler with initial lthread spawning rx and tx lthreads
2238  * (main_lthread_master).
2239  */
2240 static int
2241 lthread_master_spawner(__rte_unused void *arg) {
2242 	struct lthread *lt;
2243 	int lcore_id = rte_lcore_id();
2244 
2245 	RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2246 	lthread_create(&lt, -1, lthread_spawner, NULL);
2247 	lthread_run();
2248 
2249 	return 0;
2250 }
2251 
2252 /*
2253  * Start scheduler on lcore.
2254  */
2255 static int
2256 sched_spawner(__rte_unused void *arg) {
2257 	struct lthread *lt;
2258 	int lcore_id = rte_lcore_id();
2259 
2260 #if (APP_CPU_LOAD)
2261 	if (lcore_id == cpu_load_lcore_id) {
2262 		cpu_load_collector(arg);
2263 		return 0;
2264 	}
2265 #endif /* APP_CPU_LOAD */
2266 
2267 	RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2268 	lthread_create(&lt, -1, lthread_null, NULL);
2269 	lthread_run();
2270 
2271 	return 0;
2272 }
2273 
2274 /* main processing loop */
2275 static int
2276 pthread_tx(void *dummy)
2277 {
2278 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2279 	uint64_t prev_tsc, diff_tsc, cur_tsc;
2280 	int nb_rx;
2281 	uint8_t portid;
2282 	struct thread_tx_conf *tx_conf;
2283 
2284 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2285 		US_PER_S * BURST_TX_DRAIN_US;
2286 
2287 	prev_tsc = 0;
2288 
2289 	tx_conf = (struct thread_tx_conf *)dummy;
2290 
2291 	RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2292 
2293 	tx_conf->conf.cpu_id = sched_getcpu();
2294 	rte_atomic16_inc(&tx_counter);
2295 	while (1) {
2296 
2297 		cur_tsc = rte_rdtsc();
2298 
2299 		/*
2300 		 * TX burst queue drain
2301 		 */
2302 		diff_tsc = cur_tsc - prev_tsc;
2303 		if (unlikely(diff_tsc > drain_tsc)) {
2304 
2305 			/*
2306 			 * This could be optimized (use queueid instead of
2307 			 * portid), but it is not called so often
2308 			 */
2309 			SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2310 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2311 				if (tx_conf->tx_mbufs[portid].len == 0)
2312 					continue;
2313 				send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2314 				tx_conf->tx_mbufs[portid].len = 0;
2315 			}
2316 			SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2317 
2318 			prev_tsc = cur_tsc;
2319 		}
2320 
2321 		/*
2322 		 * Read packet from ring
2323 		 */
2324 		SET_CPU_BUSY(tx_conf, CPU_POLL);
2325 		nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2326 				(void **)pkts_burst, MAX_PKT_BURST);
2327 		SET_CPU_IDLE(tx_conf, CPU_POLL);
2328 
2329 		if (unlikely(nb_rx == 0)) {
2330 			sched_yield();
2331 			continue;
2332 		}
2333 
2334 		SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2335 		portid = pkts_burst[0]->port;
2336 		process_burst(pkts_burst, nb_rx, portid);
2337 		SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2338 
2339 	}
2340 }
2341 
2342 static int
2343 pthread_rx(void *dummy)
2344 {
2345 	int i;
2346 	int worker_id;
2347 	uint32_t n;
2348 	uint32_t nb_rx;
2349 	unsigned lcore_id;
2350 	uint8_t portid, queueid;
2351 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2352 
2353 	struct thread_rx_conf *rx_conf;
2354 
2355 	lcore_id = rte_lcore_id();
2356 	rx_conf = (struct thread_rx_conf *)dummy;
2357 
2358 	if (rx_conf->n_rx_queue == 0) {
2359 		RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2360 		return 0;
2361 	}
2362 
2363 	RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2364 
2365 	for (i = 0; i < rx_conf->n_rx_queue; i++) {
2366 
2367 		portid = rx_conf->rx_queue_list[i].port_id;
2368 		queueid = rx_conf->rx_queue_list[i].queue_id;
2369 		RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
2370 				lcore_id, portid, queueid);
2371 	}
2372 
2373 	worker_id = 0;
2374 	rx_conf->conf.cpu_id = sched_getcpu();
2375 	rte_atomic16_inc(&rx_counter);
2376 	while (1) {
2377 
2378 		/*
2379 		 * Read packet from RX queues
2380 		 */
2381 		for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2382 			portid = rx_conf->rx_queue_list[i].port_id;
2383 			queueid = rx_conf->rx_queue_list[i].queue_id;
2384 
2385 			SET_CPU_BUSY(rx_conf, CPU_POLL);
2386 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2387 				MAX_PKT_BURST);
2388 			SET_CPU_IDLE(rx_conf, CPU_POLL);
2389 
2390 			if (nb_rx == 0) {
2391 				sched_yield();
2392 				continue;
2393 			}
2394 
2395 			SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2396 			worker_id = (worker_id + 1) % rx_conf->n_ring;
2397 			n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2398 					(void **)pkts_burst, nb_rx);
2399 
2400 			if (unlikely(n != nb_rx)) {
2401 				uint32_t k;
2402 
2403 				for (k = n; k < nb_rx; k++) {
2404 					struct rte_mbuf *m = pkts_burst[k];
2405 
2406 					rte_pktmbuf_free(m);
2407 				}
2408 			}
2409 
2410 			SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2411 
2412 		}
2413 	}
2414 }
2415 
2416 /*
2417  * P-Thread spawner.
2418  */
2419 static int
2420 pthread_run(__rte_unused void *arg) {
2421 	int lcore_id = rte_lcore_id();
2422 	int i;
2423 
2424 	for (i = 0; i < n_rx_thread; i++)
2425 		if (rx_thread[i].conf.lcore_id == lcore_id) {
2426 			printf("Start rx thread on %d...\n", lcore_id);
2427 			RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2428 			RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2429 			pthread_rx((void *)&rx_thread[i]);
2430 			return 0;
2431 		}
2432 
2433 	for (i = 0; i < n_tx_thread; i++)
2434 		if (tx_thread[i].conf.lcore_id == lcore_id) {
2435 			printf("Start tx thread on %d...\n", lcore_id);
2436 			RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2437 			RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2438 			pthread_tx((void *)&tx_thread[i]);
2439 			return 0;
2440 		}
2441 
2442 #if (APP_CPU_LOAD)
2443 	if (lcore_id == cpu_load_lcore_id)
2444 		cpu_load_collector(arg);
2445 #endif /* APP_CPU_LOAD */
2446 
2447 	return 0;
2448 }
2449 
2450 static int
2451 check_lcore_params(void)
2452 {
2453 	uint8_t queue, lcore;
2454 	uint16_t i;
2455 	int socketid;
2456 
2457 	for (i = 0; i < nb_rx_thread_params; ++i) {
2458 		queue = rx_thread_params[i].queue_id;
2459 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
2460 			printf("invalid queue number: %hhu\n", queue);
2461 			return -1;
2462 		}
2463 		lcore = rx_thread_params[i].lcore_id;
2464 		if (!rte_lcore_is_enabled(lcore)) {
2465 			printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2466 			return -1;
2467 		}
2468 		socketid = rte_lcore_to_socket_id(lcore);
2469 		if ((socketid != 0) && (numa_on == 0))
2470 			printf("warning: lcore %hhu is on socket %d with numa off\n",
2471 				lcore, socketid);
2472 	}
2473 	return 0;
2474 }
2475 
2476 static int
2477 check_port_config(const unsigned nb_ports)
2478 {
2479 	unsigned portid;
2480 	uint16_t i;
2481 
2482 	for (i = 0; i < nb_rx_thread_params; ++i) {
2483 		portid = rx_thread_params[i].port_id;
2484 		if ((enabled_port_mask & (1 << portid)) == 0) {
2485 			printf("port %u is not enabled in port mask\n", portid);
2486 			return -1;
2487 		}
2488 		if (portid >= nb_ports) {
2489 			printf("port %u is not present on the board\n", portid);
2490 			return -1;
2491 		}
2492 	}
2493 	return 0;
2494 }
2495 
2496 static uint8_t
2497 get_port_n_rx_queues(const uint8_t port)
2498 {
2499 	int queue = -1;
2500 	uint16_t i;
2501 
2502 	for (i = 0; i < nb_rx_thread_params; ++i)
2503 		if (rx_thread_params[i].port_id == port &&
2504 				rx_thread_params[i].queue_id > queue)
2505 			queue = rx_thread_params[i].queue_id;
2506 
2507 	return (uint8_t)(++queue);
2508 }
2509 
2510 static int
2511 init_rx_rings(void)
2512 {
2513 	unsigned socket_io;
2514 	struct thread_rx_conf *rx_conf;
2515 	struct thread_tx_conf *tx_conf;
2516 	unsigned rx_thread_id, tx_thread_id;
2517 	char name[256];
2518 	struct rte_ring *ring = NULL;
2519 
2520 	for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2521 
2522 		tx_conf = &tx_thread[tx_thread_id];
2523 
2524 		printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2525 				tx_conf->conf.thread_id);
2526 
2527 		rx_thread_id = tx_conf->conf.thread_id;
2528 		if (rx_thread_id > n_tx_thread) {
2529 			printf("connection from tx-thread %u to rx-thread %u fails "
2530 					"(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2531 			return -1;
2532 		}
2533 
2534 		rx_conf = &rx_thread[rx_thread_id];
2535 		socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2536 
2537 		snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2538 				socket_io, rx_thread_id, tx_thread_id);
2539 
2540 		ring = rte_ring_create(name, 1024 * 4, socket_io,
2541 				RING_F_SP_ENQ | RING_F_SC_DEQ);
2542 
2543 		if (ring == NULL) {
2544 			rte_panic("Cannot create ring to connect rx-thread %u "
2545 					"with tx-thread %u\n", rx_thread_id, tx_thread_id);
2546 		}
2547 
2548 		rx_conf->ring[rx_conf->n_ring] = ring;
2549 
2550 		tx_conf->ring = ring;
2551 		tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2552 
2553 		rx_conf->n_ring++;
2554 	}
2555 	return 0;
2556 }
2557 
2558 static int
2559 init_rx_queues(void)
2560 {
2561 	uint16_t i, nb_rx_queue;
2562 	uint8_t thread;
2563 
2564 	n_rx_thread = 0;
2565 
2566 	for (i = 0; i < nb_rx_thread_params; ++i) {
2567 		thread = rx_thread_params[i].thread_id;
2568 		nb_rx_queue = rx_thread[thread].n_rx_queue;
2569 
2570 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2571 			printf("error: too many queues (%u) for thread: %u\n",
2572 				(unsigned)nb_rx_queue + 1, (unsigned)thread);
2573 			return -1;
2574 		}
2575 
2576 		rx_thread[thread].conf.thread_id = thread;
2577 		rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2578 		rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2579 			rx_thread_params[i].port_id;
2580 		rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2581 			rx_thread_params[i].queue_id;
2582 		rx_thread[thread].n_rx_queue++;
2583 
2584 		if (thread >= n_rx_thread)
2585 			n_rx_thread = thread + 1;
2586 
2587 	}
2588 	return 0;
2589 }
2590 
2591 static int
2592 init_tx_threads(void)
2593 {
2594 	int i;
2595 
2596 	n_tx_thread = 0;
2597 	for (i = 0; i < nb_tx_thread_params; ++i) {
2598 		tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2599 		tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2600 		n_tx_thread++;
2601 	}
2602 	return 0;
2603 }
2604 
2605 /* display usage */
2606 static void
2607 print_usage(const char *prgname)
2608 {
2609 	printf("%s [EAL options] -- -p PORTMASK -P"
2610 		"  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2611 		"  [--tx (lcore,thread)[,(lcore,thread]]"
2612 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2613 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2614 		"  -P : enable promiscuous mode\n"
2615 		"  --rx (port,queue,lcore,thread): rx queues configuration\n"
2616 		"  --tx (lcore,thread): tx threads configuration\n"
2617 		"  --stat-lcore LCORE: use lcore for stat collector\n"
2618 		"  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2619 		"  --no-numa: optional, disable numa awareness\n"
2620 		"  --ipv6: optional, specify it if running ipv6 packets\n"
2621 		"  --enable-jumbo: enable jumbo frame"
2622 		" which max packet len is PKTLEN in decimal (64-9600)\n"
2623 		"  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2624 		"  --no-lthreads: turn off lthread model\n",
2625 		prgname);
2626 }
2627 
2628 static int parse_max_pkt_len(const char *pktlen)
2629 {
2630 	char *end = NULL;
2631 	unsigned long len;
2632 
2633 	/* parse decimal string */
2634 	len = strtoul(pktlen, &end, 10);
2635 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2636 		return -1;
2637 
2638 	if (len == 0)
2639 		return -1;
2640 
2641 	return len;
2642 }
2643 
2644 static int
2645 parse_portmask(const char *portmask)
2646 {
2647 	char *end = NULL;
2648 	unsigned long pm;
2649 
2650 	/* parse hexadecimal string */
2651 	pm = strtoul(portmask, &end, 16);
2652 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2653 		return -1;
2654 
2655 	if (pm == 0)
2656 		return -1;
2657 
2658 	return pm;
2659 }
2660 
2661 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2662 static int
2663 parse_hash_entry_number(const char *hash_entry_num)
2664 {
2665 	char *end = NULL;
2666 	unsigned long hash_en;
2667 
2668 	/* parse hexadecimal string */
2669 	hash_en = strtoul(hash_entry_num, &end, 16);
2670 	if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2671 		return -1;
2672 
2673 	if (hash_en == 0)
2674 		return -1;
2675 
2676 	return hash_en;
2677 }
2678 #endif
2679 
2680 static int
2681 parse_rx_config(const char *q_arg)
2682 {
2683 	char s[256];
2684 	const char *p, *p0 = q_arg;
2685 	char *end;
2686 	enum fieldnames {
2687 		FLD_PORT = 0,
2688 		FLD_QUEUE,
2689 		FLD_LCORE,
2690 		FLD_THREAD,
2691 		_NUM_FLD
2692 	};
2693 	unsigned long int_fld[_NUM_FLD];
2694 	char *str_fld[_NUM_FLD];
2695 	int i;
2696 	unsigned size;
2697 
2698 	nb_rx_thread_params = 0;
2699 
2700 	while ((p = strchr(p0, '(')) != NULL) {
2701 		++p;
2702 		p0 = strchr(p, ')');
2703 		if (p0 == NULL)
2704 			return -1;
2705 
2706 		size = p0 - p;
2707 		if (size >= sizeof(s))
2708 			return -1;
2709 
2710 		snprintf(s, sizeof(s), "%.*s", size, p);
2711 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2712 			return -1;
2713 		for (i = 0; i < _NUM_FLD; i++) {
2714 			errno = 0;
2715 			int_fld[i] = strtoul(str_fld[i], &end, 0);
2716 			if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2717 				return -1;
2718 		}
2719 		if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2720 			printf("exceeded max number of rx params: %hu\n",
2721 					nb_rx_thread_params);
2722 			return -1;
2723 		}
2724 		rx_thread_params_array[nb_rx_thread_params].port_id =
2725 				(uint8_t)int_fld[FLD_PORT];
2726 		rx_thread_params_array[nb_rx_thread_params].queue_id =
2727 				(uint8_t)int_fld[FLD_QUEUE];
2728 		rx_thread_params_array[nb_rx_thread_params].lcore_id =
2729 				(uint8_t)int_fld[FLD_LCORE];
2730 		rx_thread_params_array[nb_rx_thread_params].thread_id =
2731 				(uint8_t)int_fld[FLD_THREAD];
2732 		++nb_rx_thread_params;
2733 	}
2734 	rx_thread_params = rx_thread_params_array;
2735 	return 0;
2736 }
2737 
2738 static int
2739 parse_tx_config(const char *q_arg)
2740 {
2741 	char s[256];
2742 	const char *p, *p0 = q_arg;
2743 	char *end;
2744 	enum fieldnames {
2745 		FLD_LCORE = 0,
2746 		FLD_THREAD,
2747 		_NUM_FLD
2748 	};
2749 	unsigned long int_fld[_NUM_FLD];
2750 	char *str_fld[_NUM_FLD];
2751 	int i;
2752 	unsigned size;
2753 
2754 	nb_tx_thread_params = 0;
2755 
2756 	while ((p = strchr(p0, '(')) != NULL) {
2757 		++p;
2758 		p0 = strchr(p, ')');
2759 		if (p0 == NULL)
2760 			return -1;
2761 
2762 		size = p0 - p;
2763 		if (size >= sizeof(s))
2764 			return -1;
2765 
2766 		snprintf(s, sizeof(s), "%.*s", size, p);
2767 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2768 			return -1;
2769 		for (i = 0; i < _NUM_FLD; i++) {
2770 			errno = 0;
2771 			int_fld[i] = strtoul(str_fld[i], &end, 0);
2772 			if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2773 				return -1;
2774 		}
2775 		if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2776 			printf("exceeded max number of tx params: %hu\n",
2777 				nb_tx_thread_params);
2778 			return -1;
2779 		}
2780 		tx_thread_params_array[nb_tx_thread_params].lcore_id =
2781 				(uint8_t)int_fld[FLD_LCORE];
2782 		tx_thread_params_array[nb_tx_thread_params].thread_id =
2783 				(uint8_t)int_fld[FLD_THREAD];
2784 		++nb_tx_thread_params;
2785 	}
2786 	tx_thread_params = tx_thread_params_array;
2787 
2788 	return 0;
2789 }
2790 
2791 #if (APP_CPU_LOAD > 0)
2792 static int
2793 parse_stat_lcore(const char *stat_lcore)
2794 {
2795 	char *end = NULL;
2796 	unsigned long lcore_id;
2797 
2798 	lcore_id = strtoul(stat_lcore, &end, 10);
2799 	if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2800 		return -1;
2801 
2802 	return lcore_id;
2803 }
2804 #endif
2805 
2806 static void
2807 parse_eth_dest(const char *optarg)
2808 {
2809 	uint8_t portid;
2810 	char *port_end;
2811 	uint8_t c, *dest, peer_addr[6];
2812 
2813 	errno = 0;
2814 	portid = strtoul(optarg, &port_end, 10);
2815 	if (errno != 0 || port_end == optarg || *port_end++ != ',')
2816 		rte_exit(EXIT_FAILURE,
2817 		"Invalid eth-dest: %s", optarg);
2818 	if (portid >= RTE_MAX_ETHPORTS)
2819 		rte_exit(EXIT_FAILURE,
2820 		"eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2821 		portid, RTE_MAX_ETHPORTS);
2822 
2823 	if (cmdline_parse_etheraddr(NULL, port_end,
2824 		&peer_addr, sizeof(peer_addr)) < 0)
2825 		rte_exit(EXIT_FAILURE,
2826 		"Invalid ethernet address: %s\n",
2827 		port_end);
2828 	dest = (uint8_t *)&dest_eth_addr[portid];
2829 	for (c = 0; c < 6; c++)
2830 		dest[c] = peer_addr[c];
2831 	*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2832 }
2833 
2834 #define CMD_LINE_OPT_RX_CONFIG "rx"
2835 #define CMD_LINE_OPT_TX_CONFIG "tx"
2836 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2837 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2838 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2839 #define CMD_LINE_OPT_IPV6 "ipv6"
2840 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2841 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2842 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2843 
2844 /* Parse the argument given in the command line of the application */
2845 static int
2846 parse_args(int argc, char **argv)
2847 {
2848 	int opt, ret;
2849 	char **argvopt;
2850 	int option_index;
2851 	char *prgname = argv[0];
2852 	static struct option lgopts[] = {
2853 		{CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2854 		{CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2855 		{CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2856 		{CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2857 		{CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2858 		{CMD_LINE_OPT_IPV6, 0, 0, 0},
2859 		{CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2860 		{CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2861 		{CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2862 		{NULL, 0, 0, 0}
2863 	};
2864 
2865 	argvopt = argv;
2866 
2867 	while ((opt = getopt_long(argc, argvopt, "p:P",
2868 				lgopts, &option_index)) != EOF) {
2869 
2870 		switch (opt) {
2871 		/* portmask */
2872 		case 'p':
2873 			enabled_port_mask = parse_portmask(optarg);
2874 			if (enabled_port_mask == 0) {
2875 				printf("invalid portmask\n");
2876 				print_usage(prgname);
2877 				return -1;
2878 			}
2879 			break;
2880 		case 'P':
2881 			printf("Promiscuous mode selected\n");
2882 			promiscuous_on = 1;
2883 			break;
2884 
2885 		/* long options */
2886 		case 0:
2887 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2888 				sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2889 				ret = parse_rx_config(optarg);
2890 				if (ret) {
2891 					printf("invalid rx-config\n");
2892 					print_usage(prgname);
2893 					return -1;
2894 				}
2895 			}
2896 
2897 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2898 				sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2899 				ret = parse_tx_config(optarg);
2900 				if (ret) {
2901 					printf("invalid tx-config\n");
2902 					print_usage(prgname);
2903 					return -1;
2904 				}
2905 			}
2906 
2907 #if (APP_CPU_LOAD > 0)
2908 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2909 					sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2910 				cpu_load_lcore_id = parse_stat_lcore(optarg);
2911 			}
2912 #endif
2913 
2914 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2915 				sizeof(CMD_LINE_OPT_ETH_DEST)))
2916 					parse_eth_dest(optarg);
2917 
2918 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2919 				sizeof(CMD_LINE_OPT_NO_NUMA))) {
2920 				printf("numa is disabled\n");
2921 				numa_on = 0;
2922 			}
2923 
2924 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2925 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2926 				sizeof(CMD_LINE_OPT_IPV6))) {
2927 				printf("ipv6 is specified\n");
2928 				ipv6 = 1;
2929 			}
2930 #endif
2931 
2932 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2933 					sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2934 				printf("l-threads model is disabled\n");
2935 				lthreads_on = 0;
2936 			}
2937 
2938 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2939 				sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2940 				struct option lenopts = {"max-pkt-len", required_argument, 0,
2941 						0};
2942 
2943 				printf("jumbo frame is enabled - disabling simple TX path\n");
2944 				port_conf.rxmode.jumbo_frame = 1;
2945 
2946 				/* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2947 				if (0 == getopt_long(argc, argvopt, "", &lenopts,
2948 						&option_index)) {
2949 
2950 					ret = parse_max_pkt_len(optarg);
2951 					if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2952 						printf("invalid packet length\n");
2953 						print_usage(prgname);
2954 						return -1;
2955 					}
2956 					port_conf.rxmode.max_rx_pkt_len = ret;
2957 				}
2958 				printf("set jumbo frame max packet length to %u\n",
2959 						(unsigned int)port_conf.rxmode.max_rx_pkt_len);
2960 			}
2961 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2962 			if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2963 				sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2964 				ret = parse_hash_entry_number(optarg);
2965 				if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2966 					hash_entry_number = ret;
2967 				} else {
2968 					printf("invalid hash entry number\n");
2969 					print_usage(prgname);
2970 					return -1;
2971 				}
2972 			}
2973 #endif
2974 			break;
2975 
2976 		default:
2977 			print_usage(prgname);
2978 			return -1;
2979 		}
2980 	}
2981 
2982 	if (optind >= 0)
2983 		argv[optind-1] = prgname;
2984 
2985 	ret = optind-1;
2986 	optind = 0; /* reset getopt lib */
2987 	return ret;
2988 }
2989 
2990 static void
2991 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
2992 {
2993 	char buf[ETHER_ADDR_FMT_SIZE];
2994 
2995 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
2996 	printf("%s%s", name, buf);
2997 }
2998 
2999 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3000 
3001 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3002 		union ipv4_5tuple_host *key2)
3003 {
3004 	key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3005 	key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3006 	key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3007 	key2->port_src = rte_cpu_to_be_16(key1->port_src);
3008 	key2->proto = key1->proto;
3009 	key2->pad0 = 0;
3010 	key2->pad1 = 0;
3011 }
3012 
3013 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3014 		union ipv6_5tuple_host *key2)
3015 {
3016 	uint32_t i;
3017 
3018 	for (i = 0; i < 16; i++) {
3019 		key2->ip_dst[i] = key1->ip_dst[i];
3020 		key2->ip_src[i] = key1->ip_src[i];
3021 	}
3022 	key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3023 	key2->port_src = rte_cpu_to_be_16(key1->port_src);
3024 	key2->proto = key1->proto;
3025 	key2->pad0 = 0;
3026 	key2->pad1 = 0;
3027 	key2->reserve = 0;
3028 }
3029 
3030 #define BYTE_VALUE_MAX 256
3031 #define ALL_32_BITS 0xffffffff
3032 #define BIT_8_TO_15 0x0000ff00
3033 static inline void
3034 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3035 {
3036 	uint32_t i;
3037 	int32_t ret;
3038 	uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3039 
3040 	mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3041 	for (i = 0; i < array_len; i++) {
3042 		struct ipv4_l3fwd_route  entry;
3043 		union ipv4_5tuple_host newkey;
3044 
3045 		entry = ipv4_l3fwd_route_array[i];
3046 		convert_ipv4_5tuple(&entry.key, &newkey);
3047 		ret = rte_hash_add_key(h, (void *)&newkey);
3048 		if (ret < 0) {
3049 			rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3050 				" to the l3fwd hash.\n", i);
3051 		}
3052 		ipv4_l3fwd_out_if[ret] = entry.if_out;
3053 	}
3054 	printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3055 }
3056 
3057 #define BIT_16_TO_23 0x00ff0000
3058 static inline void
3059 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3060 {
3061 	uint32_t i;
3062 	int32_t ret;
3063 	uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3064 
3065 	mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3066 	mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3067 	for (i = 0; i < array_len; i++) {
3068 		struct ipv6_l3fwd_route entry;
3069 		union ipv6_5tuple_host newkey;
3070 
3071 		entry = ipv6_l3fwd_route_array[i];
3072 		convert_ipv6_5tuple(&entry.key, &newkey);
3073 		ret = rte_hash_add_key(h, (void *)&newkey);
3074 		if (ret < 0) {
3075 			rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3076 				" to the l3fwd hash.\n", i);
3077 		}
3078 		ipv6_l3fwd_out_if[ret] = entry.if_out;
3079 	}
3080 	printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3081 }
3082 
3083 #define NUMBER_PORT_USED 4
3084 static inline void
3085 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3086 		unsigned int nr_flow)
3087 {
3088 	unsigned i;
3089 
3090 	mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3091 
3092 	for (i = 0; i < nr_flow; i++) {
3093 		struct ipv4_l3fwd_route entry;
3094 		union ipv4_5tuple_host newkey;
3095 		uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3096 		uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3097 				BYTE_VALUE_MAX);
3098 		uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3099 				BYTE_VALUE_MAX));
3100 		/* Create the ipv4 exact match flow */
3101 		memset(&entry, 0, sizeof(entry));
3102 		switch (i & (NUMBER_PORT_USED - 1)) {
3103 		case 0:
3104 			entry = ipv4_l3fwd_route_array[0];
3105 			entry.key.ip_dst = IPv4(101, c, b, a);
3106 			break;
3107 		case 1:
3108 			entry = ipv4_l3fwd_route_array[1];
3109 			entry.key.ip_dst = IPv4(201, c, b, a);
3110 			break;
3111 		case 2:
3112 			entry = ipv4_l3fwd_route_array[2];
3113 			entry.key.ip_dst = IPv4(111, c, b, a);
3114 			break;
3115 		case 3:
3116 			entry = ipv4_l3fwd_route_array[3];
3117 			entry.key.ip_dst = IPv4(211, c, b, a);
3118 			break;
3119 		};
3120 		convert_ipv4_5tuple(&entry.key, &newkey);
3121 		int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3122 
3123 		if (ret < 0)
3124 			rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3125 
3126 		ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3127 
3128 	}
3129 	printf("Hash: Adding 0x%x keys\n", nr_flow);
3130 }
3131 
3132 static inline void
3133 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3134 		unsigned int nr_flow)
3135 {
3136 	unsigned i;
3137 
3138 	mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3139 	mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3140 	for (i = 0; i < nr_flow; i++) {
3141 		struct ipv6_l3fwd_route entry;
3142 		union ipv6_5tuple_host newkey;
3143 
3144 		uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3145 		uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3146 				BYTE_VALUE_MAX);
3147 		uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3148 				BYTE_VALUE_MAX));
3149 
3150 		/* Create the ipv6 exact match flow */
3151 		memset(&entry, 0, sizeof(entry));
3152 		switch (i & (NUMBER_PORT_USED - 1)) {
3153 		case 0:
3154 			entry = ipv6_l3fwd_route_array[0];
3155 			break;
3156 		case 1:
3157 			entry = ipv6_l3fwd_route_array[1];
3158 			break;
3159 		case 2:
3160 			entry = ipv6_l3fwd_route_array[2];
3161 			break;
3162 		case 3:
3163 			entry = ipv6_l3fwd_route_array[3];
3164 			break;
3165 		};
3166 		entry.key.ip_dst[13] = c;
3167 		entry.key.ip_dst[14] = b;
3168 		entry.key.ip_dst[15] = a;
3169 		convert_ipv6_5tuple(&entry.key, &newkey);
3170 		int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3171 
3172 		if (ret < 0)
3173 			rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3174 
3175 		ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3176 
3177 	}
3178 	printf("Hash: Adding 0x%x keys\n", nr_flow);
3179 }
3180 
3181 static void
3182 setup_hash(int socketid)
3183 {
3184 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3185 		.name = NULL,
3186 		.entries = L3FWD_HASH_ENTRIES,
3187 		.key_len = sizeof(union ipv4_5tuple_host),
3188 		.hash_func = ipv4_hash_crc,
3189 		.hash_func_init_val = 0,
3190 	};
3191 
3192 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3193 		.name = NULL,
3194 		.entries = L3FWD_HASH_ENTRIES,
3195 		.key_len = sizeof(union ipv6_5tuple_host),
3196 		.hash_func = ipv6_hash_crc,
3197 		.hash_func_init_val = 0,
3198 	};
3199 
3200 	char s[64];
3201 
3202 	/* create ipv4 hash */
3203 	snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3204 	ipv4_l3fwd_hash_params.name = s;
3205 	ipv4_l3fwd_hash_params.socket_id = socketid;
3206 	ipv4_l3fwd_lookup_struct[socketid] =
3207 			rte_hash_create(&ipv4_l3fwd_hash_params);
3208 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3209 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3210 				"socket %d\n", socketid);
3211 
3212 	/* create ipv6 hash */
3213 	snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3214 	ipv6_l3fwd_hash_params.name = s;
3215 	ipv6_l3fwd_hash_params.socket_id = socketid;
3216 	ipv6_l3fwd_lookup_struct[socketid] =
3217 			rte_hash_create(&ipv6_l3fwd_hash_params);
3218 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3219 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3220 				"socket %d\n", socketid);
3221 
3222 	if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3223 		/* For testing hash matching with a large number of flows we
3224 		 * generate millions of IP 5-tuples with an incremented dst
3225 		 * address to initialize the hash table. */
3226 		if (ipv6 == 0) {
3227 			/* populate the ipv4 hash */
3228 			populate_ipv4_many_flow_into_table(
3229 				ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3230 		} else {
3231 			/* populate the ipv6 hash */
3232 			populate_ipv6_many_flow_into_table(
3233 				ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3234 		}
3235 	} else {
3236 		/* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3237 		 * the hash table */
3238 		if (ipv6 == 0) {
3239 			/* populate the ipv4 hash */
3240 			populate_ipv4_few_flow_into_table(
3241 					ipv4_l3fwd_lookup_struct[socketid]);
3242 		} else {
3243 			/* populate the ipv6 hash */
3244 			populate_ipv6_few_flow_into_table(
3245 					ipv6_l3fwd_lookup_struct[socketid]);
3246 		}
3247 	}
3248 }
3249 #endif
3250 
3251 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3252 static void
3253 setup_lpm(int socketid)
3254 {
3255 	struct rte_lpm6_config config;
3256 	struct rte_lpm_config lpm_ipv4_config;
3257 	unsigned i;
3258 	int ret;
3259 	char s[64];
3260 
3261 	/* create the LPM table */
3262 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3263 	lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3264 	lpm_ipv4_config.number_tbl8s = 256;
3265 	lpm_ipv4_config.flags = 0;
3266 	ipv4_l3fwd_lookup_struct[socketid] =
3267 			rte_lpm_create(s, socketid, &lpm_ipv4_config);
3268 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3269 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3270 				" on socket %d\n", socketid);
3271 
3272 	/* populate the LPM table */
3273 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3274 
3275 		/* skip unused ports */
3276 		if ((1 << ipv4_l3fwd_route_array[i].if_out &
3277 				enabled_port_mask) == 0)
3278 			continue;
3279 
3280 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3281 			ipv4_l3fwd_route_array[i].ip,
3282 			ipv4_l3fwd_route_array[i].depth,
3283 			ipv4_l3fwd_route_array[i].if_out);
3284 
3285 		if (ret < 0) {
3286 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3287 				"l3fwd LPM table on socket %d\n",
3288 				i, socketid);
3289 		}
3290 
3291 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
3292 			(unsigned)ipv4_l3fwd_route_array[i].ip,
3293 			ipv4_l3fwd_route_array[i].depth,
3294 			ipv4_l3fwd_route_array[i].if_out);
3295 	}
3296 
3297 	/* create the LPM6 table */
3298 	snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3299 
3300 	config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3301 	config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3302 	config.flags = 0;
3303 	ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3304 				&config);
3305 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3306 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3307 				" on socket %d\n", socketid);
3308 
3309 	/* populate the LPM table */
3310 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3311 
3312 		/* skip unused ports */
3313 		if ((1 << ipv6_l3fwd_route_array[i].if_out &
3314 				enabled_port_mask) == 0)
3315 			continue;
3316 
3317 		ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3318 			ipv6_l3fwd_route_array[i].ip,
3319 			ipv6_l3fwd_route_array[i].depth,
3320 			ipv6_l3fwd_route_array[i].if_out);
3321 
3322 		if (ret < 0) {
3323 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3324 				"l3fwd LPM table on socket %d\n",
3325 				i, socketid);
3326 		}
3327 
3328 		printf("LPM: Adding route %s / %d (%d)\n",
3329 			"IPV6",
3330 			ipv6_l3fwd_route_array[i].depth,
3331 			ipv6_l3fwd_route_array[i].if_out);
3332 	}
3333 }
3334 #endif
3335 
3336 static int
3337 init_mem(unsigned nb_mbuf)
3338 {
3339 	struct lcore_conf *qconf;
3340 	int socketid;
3341 	unsigned lcore_id;
3342 	char s[64];
3343 
3344 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3345 		if (rte_lcore_is_enabled(lcore_id) == 0)
3346 			continue;
3347 
3348 		if (numa_on)
3349 			socketid = rte_lcore_to_socket_id(lcore_id);
3350 		else
3351 			socketid = 0;
3352 
3353 		if (socketid >= NB_SOCKETS) {
3354 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3355 				socketid, lcore_id, NB_SOCKETS);
3356 		}
3357 		if (pktmbuf_pool[socketid] == NULL) {
3358 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3359 			pktmbuf_pool[socketid] =
3360 				rte_pktmbuf_pool_create(s, nb_mbuf,
3361 					MEMPOOL_CACHE_SIZE, 0,
3362 					RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3363 			if (pktmbuf_pool[socketid] == NULL)
3364 				rte_exit(EXIT_FAILURE,
3365 						"Cannot init mbuf pool on socket %d\n", socketid);
3366 			else
3367 				printf("Allocated mbuf pool on socket %d\n", socketid);
3368 
3369 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3370 			setup_lpm(socketid);
3371 #else
3372 			setup_hash(socketid);
3373 #endif
3374 		}
3375 		qconf = &lcore_conf[lcore_id];
3376 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3377 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3378 	}
3379 	return 0;
3380 }
3381 
3382 /* Check the link status of all ports in up to 9s, and print them finally */
3383 static void
3384 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
3385 {
3386 #define CHECK_INTERVAL 100 /* 100ms */
3387 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3388 	uint8_t portid, count, all_ports_up, print_flag = 0;
3389 	struct rte_eth_link link;
3390 
3391 	printf("\nChecking link status");
3392 	fflush(stdout);
3393 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
3394 		all_ports_up = 1;
3395 		for (portid = 0; portid < port_num; portid++) {
3396 			if ((port_mask & (1 << portid)) == 0)
3397 				continue;
3398 			memset(&link, 0, sizeof(link));
3399 			rte_eth_link_get_nowait(portid, &link);
3400 			/* print link status if flag set */
3401 			if (print_flag == 1) {
3402 				if (link.link_status)
3403 					printf("Port %d Link Up - speed %u "
3404 						"Mbps - %s\n", (uint8_t)portid,
3405 						(unsigned)link.link_speed,
3406 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3407 					("full-duplex") : ("half-duplex\n"));
3408 				else
3409 					printf("Port %d Link Down\n",
3410 						(uint8_t)portid);
3411 				continue;
3412 			}
3413 			/* clear all_ports_up flag if any link down */
3414 			if (link.link_status == ETH_LINK_DOWN) {
3415 				all_ports_up = 0;
3416 				break;
3417 			}
3418 		}
3419 		/* after finally printing all link status, get out */
3420 		if (print_flag == 1)
3421 			break;
3422 
3423 		if (all_ports_up == 0) {
3424 			printf(".");
3425 			fflush(stdout);
3426 			rte_delay_ms(CHECK_INTERVAL);
3427 		}
3428 
3429 		/* set the print_flag if all ports up or timeout */
3430 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3431 			print_flag = 1;
3432 			printf("done\n");
3433 		}
3434 	}
3435 }
3436 
3437 int
3438 main(int argc, char **argv)
3439 {
3440 	struct rte_eth_dev_info dev_info;
3441 	struct rte_eth_txconf *txconf;
3442 	int ret;
3443 	int i;
3444 	unsigned nb_ports;
3445 	uint16_t queueid;
3446 	unsigned lcore_id;
3447 	uint32_t n_tx_queue, nb_lcores;
3448 	uint8_t portid, nb_rx_queue, queue, socketid;
3449 
3450 	/* init EAL */
3451 	ret = rte_eal_init(argc, argv);
3452 	if (ret < 0)
3453 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3454 	argc -= ret;
3455 	argv += ret;
3456 
3457 	/* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3458 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3459 		dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3460 				((uint64_t)portid << 40);
3461 		*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3462 	}
3463 
3464 	/* parse application arguments (after the EAL ones) */
3465 	ret = parse_args(argc, argv);
3466 	if (ret < 0)
3467 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3468 
3469 	if (check_lcore_params() < 0)
3470 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3471 
3472 	printf("Initializing rx-queues...\n");
3473 	ret = init_rx_queues();
3474 	if (ret < 0)
3475 		rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3476 
3477 	printf("Initializing tx-threads...\n");
3478 	ret = init_tx_threads();
3479 	if (ret < 0)
3480 		rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3481 
3482 	printf("Initializing rings...\n");
3483 	ret = init_rx_rings();
3484 	if (ret < 0)
3485 		rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3486 
3487 	nb_ports = rte_eth_dev_count();
3488 
3489 	if (check_port_config(nb_ports) < 0)
3490 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3491 
3492 	nb_lcores = rte_lcore_count();
3493 
3494 	/* initialize all ports */
3495 	for (portid = 0; portid < nb_ports; portid++) {
3496 		/* skip ports that are not enabled */
3497 		if ((enabled_port_mask & (1 << portid)) == 0) {
3498 			printf("\nSkipping disabled port %d\n", portid);
3499 			continue;
3500 		}
3501 
3502 		/* init port */
3503 		printf("Initializing port %d ... ", portid);
3504 		fflush(stdout);
3505 
3506 		nb_rx_queue = get_port_n_rx_queues(portid);
3507 		n_tx_queue = nb_lcores;
3508 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3509 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3510 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3511 			nb_rx_queue, (unsigned)n_tx_queue);
3512 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
3513 					(uint16_t)n_tx_queue, &port_conf);
3514 		if (ret < 0)
3515 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3516 				ret, portid);
3517 
3518 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3519 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
3520 		printf(", ");
3521 		print_ethaddr("Destination:",
3522 			(const struct ether_addr *)&dest_eth_addr[portid]);
3523 		printf(", ");
3524 
3525 		/*
3526 		 * prepare src MACs for each port.
3527 		 */
3528 		ether_addr_copy(&ports_eth_addr[portid],
3529 			(struct ether_addr *)(val_eth + portid) + 1);
3530 
3531 		/* init memory */
3532 		ret = init_mem(NB_MBUF);
3533 		if (ret < 0)
3534 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
3535 
3536 		/* init one TX queue per couple (lcore,port) */
3537 		queueid = 0;
3538 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3539 			if (rte_lcore_is_enabled(lcore_id) == 0)
3540 				continue;
3541 
3542 			if (numa_on)
3543 				socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3544 			else
3545 				socketid = 0;
3546 
3547 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3548 			fflush(stdout);
3549 
3550 			rte_eth_dev_info_get(portid, &dev_info);
3551 			txconf = &dev_info.default_txconf;
3552 			if (port_conf.rxmode.jumbo_frame)
3553 				txconf->txq_flags = 0;
3554 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3555 						     socketid, txconf);
3556 			if (ret < 0)
3557 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3558 					"port=%d\n", ret, portid);
3559 
3560 			tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3561 			queueid++;
3562 		}
3563 		printf("\n");
3564 	}
3565 
3566 	for (i = 0; i < n_rx_thread; i++) {
3567 		lcore_id = rx_thread[i].conf.lcore_id;
3568 
3569 		if (rte_lcore_is_enabled(lcore_id) == 0) {
3570 			rte_exit(EXIT_FAILURE,
3571 					"Cannot start Rx thread on lcore %u: lcore disabled\n",
3572 					lcore_id
3573 				);
3574 		}
3575 
3576 		printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3577 				i, lcore_id);
3578 		fflush(stdout);
3579 
3580 		/* init RX queues */
3581 		for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3582 			portid = rx_thread[i].rx_queue_list[queue].port_id;
3583 			queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3584 
3585 			if (numa_on)
3586 				socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3587 			else
3588 				socketid = 0;
3589 
3590 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3591 			fflush(stdout);
3592 
3593 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3594 					socketid,
3595 					NULL,
3596 					pktmbuf_pool[socketid]);
3597 			if (ret < 0)
3598 				rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3599 						"port=%d\n", ret, portid);
3600 		}
3601 	}
3602 
3603 	printf("\n");
3604 
3605 	/* start ports */
3606 	for (portid = 0; portid < nb_ports; portid++) {
3607 		if ((enabled_port_mask & (1 << portid)) == 0)
3608 			continue;
3609 
3610 		/* Start device */
3611 		ret = rte_eth_dev_start(portid);
3612 		if (ret < 0)
3613 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3614 				ret, portid);
3615 
3616 		/*
3617 		 * If enabled, put device in promiscuous mode.
3618 		 * This allows IO forwarding mode to forward packets
3619 		 * to itself through 2 cross-connected  ports of the
3620 		 * target machine.
3621 		 */
3622 		if (promiscuous_on)
3623 			rte_eth_promiscuous_enable(portid);
3624 	}
3625 
3626 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
3627 
3628 	if (lthreads_on) {
3629 		printf("Starting L-Threading Model\n");
3630 
3631 #if (APP_CPU_LOAD > 0)
3632 		if (cpu_load_lcore_id > 0)
3633 			/* Use one lcore for cpu load collector */
3634 			nb_lcores--;
3635 #endif
3636 
3637 		lthread_num_schedulers_set(nb_lcores);
3638 		rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3639 		lthread_master_spawner(NULL);
3640 
3641 	} else {
3642 		printf("Starting P-Threading Model\n");
3643 		/* launch per-lcore init on every lcore */
3644 		rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3645 		RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3646 			if (rte_eal_wait_lcore(lcore_id) < 0)
3647 				return -1;
3648 		}
3649 	}
3650 
3651 	return 0;
3652 }
3653