xref: /dpdk/app/test-flow-perf/main.c (revision ab4bb42406cc7c82ff69b444433f09d873a5466e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2020 Mellanox Technologies, Ltd
3  *
4  * This file contain the application main file
5  * This application provides the user the ability to test the
6  * insertion rate for specific rte_flow rule under stress state ~4M rule/
7  *
8  * Then it will also provide packet per second measurement after installing
9  * all rules, the user may send traffic to test the PPS that match the rules
10  * after all rules are installed, to check performance or functionality after
11  * the stress.
12  *
13  * The flows insertion will go for all ports first, then it will print the
14  * results, after that the application will go into forwarding packets mode
15  * it will start receiving traffic if any and then forwarding it back and
16  * gives packet per second measurement.
17  */
18 
19 #include <stdio.h>
20 #include <stdlib.h>
21 #include <string.h>
22 #include <stdint.h>
23 #include <inttypes.h>
24 #include <stdarg.h>
25 #include <errno.h>
26 #include <getopt.h>
27 #include <stdbool.h>
28 #include <sys/time.h>
29 #include <signal.h>
30 #include <unistd.h>
31 
32 #include <rte_malloc.h>
33 #include <rte_mempool.h>
34 #include <rte_mbuf.h>
35 #include <rte_ethdev.h>
36 #include <rte_flow.h>
37 #include <rte_mtr.h>
38 
39 #include "config.h"
40 #include "flow_gen.h"
41 
42 #define MAX_BATCHES_COUNT          100
43 #define DEFAULT_RULES_COUNT    4000000
44 #define DEFAULT_RULES_BATCH     100000
45 #define DEFAULT_GROUP                0
46 
47 struct rte_flow *flow;
48 static uint8_t flow_group;
49 
50 static uint64_t encap_data;
51 static uint64_t decap_data;
52 
53 static uint64_t flow_items[MAX_ITEMS_NUM];
54 static uint64_t flow_actions[MAX_ACTIONS_NUM];
55 static uint64_t flow_attrs[MAX_ATTRS_NUM];
56 static uint8_t items_idx, actions_idx, attrs_idx;
57 
58 static uint64_t ports_mask;
59 static uint16_t dst_ports[RTE_MAX_ETHPORTS];
60 static volatile bool force_quit;
61 static bool dump_iterations;
62 static bool delete_flag;
63 static bool dump_socket_mem_flag;
64 static bool enable_fwd;
65 static bool unique_data;
66 
67 static uint8_t rx_queues_count;
68 static uint8_t tx_queues_count;
69 static uint8_t rxd_count;
70 static uint8_t txd_count;
71 static uint32_t mbuf_size;
72 static uint32_t mbuf_cache_size;
73 static uint32_t total_mbuf_num;
74 
75 static struct rte_mempool *mbuf_mp;
76 static uint32_t nb_lcores;
77 static uint32_t rules_count;
78 static uint32_t rules_batch;
79 static uint32_t hairpin_queues_num; /* total hairpin q number - default: 0 */
80 static uint32_t nb_lcores;
81 
82 #define MAX_PKT_BURST    32
83 #define LCORE_MODE_PKT    1
84 #define LCORE_MODE_STATS  2
85 #define MAX_STREAMS      64
86 #define METER_CREATE	  1
87 #define METER_DELETE	  2
88 
89 struct stream {
90 	int tx_port;
91 	int tx_queue;
92 	int rx_port;
93 	int rx_queue;
94 };
95 
96 struct lcore_info {
97 	int mode;
98 	int streams_nb;
99 	struct stream streams[MAX_STREAMS];
100 	/* stats */
101 	uint64_t tx_pkts;
102 	uint64_t tx_drops;
103 	uint64_t rx_pkts;
104 	struct rte_mbuf *pkts[MAX_PKT_BURST];
105 } __rte_cache_aligned;
106 
107 static struct lcore_info lcore_infos[RTE_MAX_LCORE];
108 
109 struct used_cpu_time {
110 	double insertion[MAX_PORTS][RTE_MAX_LCORE];
111 	double deletion[MAX_PORTS][RTE_MAX_LCORE];
112 };
113 
114 struct multi_cores_pool {
115 	uint32_t cores_count;
116 	uint32_t rules_count;
117 	struct used_cpu_time meters_record;
118 	struct used_cpu_time flows_record;
119 	int64_t last_alloc[RTE_MAX_LCORE];
120 	int64_t current_alloc[RTE_MAX_LCORE];
121 } __rte_cache_aligned;
122 
123 static struct multi_cores_pool mc_pool = {
124 	.cores_count = 1,
125 };
126 
127 static void
128 usage(char *progname)
129 {
130 	printf("\nusage: %s\n", progname);
131 	printf("\nControl configurations:\n");
132 	printf("  --rules-count=N: to set the number of needed"
133 		" rules to insert, default is %d\n", DEFAULT_RULES_COUNT);
134 	printf("  --rules-batch=N: set number of batched rules,"
135 		" default is %d\n", DEFAULT_RULES_BATCH);
136 	printf("  --dump-iterations: To print rates for each"
137 		" iteration\n");
138 	printf("  --deletion-rate: Enable deletion rate"
139 		" calculations\n");
140 	printf("  --dump-socket-mem: To dump all socket memory\n");
141 	printf("  --enable-fwd: To enable packets forwarding"
142 		" after insertion\n");
143 	printf("  --portmask=N: hexadecimal bitmask of ports used\n");
144 	printf("  --unique-data: flag to set using unique data for all"
145 		" actions that support data, such as header modify and encap actions\n");
146 
147 	printf("To set flow attributes:\n");
148 	printf("  --ingress: set ingress attribute in flows\n");
149 	printf("  --egress: set egress attribute in flows\n");
150 	printf("  --transfer: set transfer attribute in flows\n");
151 	printf("  --group=N: set group for all flows,"
152 		" default is %d\n", DEFAULT_GROUP);
153 	printf("  --cores=N: to set the number of needed "
154 		"cores to insert rte_flow rules, default is 1\n");
155 	printf("  --rxq=N: to set the count of receive queues\n");
156 	printf("  --txq=N: to set the count of send queues\n");
157 	printf("  --rxd=N: to set the count of rxd\n");
158 	printf("  --txd=N: to set the count of txd\n");
159 	printf("  --mbuf-size=N: to set the size of mbuf\n");
160 	printf("  --mbuf-cache-size=N: to set the size of mbuf cache\n");
161 	printf("  --total-mbuf-count=N: to set the count of total mbuf count\n");
162 
163 
164 	printf("To set flow items:\n");
165 	printf("  --ether: add ether layer in flow items\n");
166 	printf("  --vlan: add vlan layer in flow items\n");
167 	printf("  --ipv4: add ipv4 layer in flow items\n");
168 	printf("  --ipv6: add ipv6 layer in flow items\n");
169 	printf("  --tcp: add tcp layer in flow items\n");
170 	printf("  --udp: add udp layer in flow items\n");
171 	printf("  --vxlan: add vxlan layer in flow items\n");
172 	printf("  --vxlan-gpe: add vxlan-gpe layer in flow items\n");
173 	printf("  --gre: add gre layer in flow items\n");
174 	printf("  --geneve: add geneve layer in flow items\n");
175 	printf("  --gtp: add gtp layer in flow items\n");
176 	printf("  --meta: add meta layer in flow items\n");
177 	printf("  --tag: add tag layer in flow items\n");
178 	printf("  --icmpv4: add icmpv4 layer in flow items\n");
179 	printf("  --icmpv6: add icmpv6 layer in flow items\n");
180 
181 	printf("To set flow actions:\n");
182 	printf("  --port-id: add port-id action in flow actions\n");
183 	printf("  --rss: add rss action in flow actions\n");
184 	printf("  --queue: add queue action in flow actions\n");
185 	printf("  --jump: add jump action in flow actions\n");
186 	printf("  --mark: add mark action in flow actions\n");
187 	printf("  --count: add count action in flow actions\n");
188 	printf("  --set-meta: add set meta action in flow actions\n");
189 	printf("  --set-tag: add set tag action in flow actions\n");
190 	printf("  --drop: add drop action in flow actions\n");
191 	printf("  --hairpin-queue=N: add hairpin-queue action in flow actions\n");
192 	printf("  --hairpin-rss=N: add hairpin-rss action in flow actions\n");
193 	printf("  --set-src-mac: add set src mac action to flow actions\n"
194 		"Src mac to be set is random each flow\n");
195 	printf("  --set-dst-mac: add set dst mac action to flow actions\n"
196 		 "Dst mac to be set is random each flow\n");
197 	printf("  --set-src-ipv4: add set src ipv4 action to flow actions\n"
198 		"Src ipv4 to be set is random each flow\n");
199 	printf("  --set-dst-ipv4 add set dst ipv4 action to flow actions\n"
200 		"Dst ipv4 to be set is random each flow\n");
201 	printf("  --set-src-ipv6: add set src ipv6 action to flow actions\n"
202 		"Src ipv6 to be set is random each flow\n");
203 	printf("  --set-dst-ipv6: add set dst ipv6 action to flow actions\n"
204 		"Dst ipv6 to be set is random each flow\n");
205 	printf("  --set-src-tp: add set src tp action to flow actions\n"
206 		"Src tp to be set is random each flow\n");
207 	printf("  --set-dst-tp: add set dst tp action to flow actions\n"
208 		"Dst tp to be set is random each flow\n");
209 	printf("  --inc-tcp-ack: add inc tcp ack action to flow actions\n"
210 		"tcp ack will be increments by 1\n");
211 	printf("  --dec-tcp-ack: add dec tcp ack action to flow actions\n"
212 		"tcp ack will be decrements by 1\n");
213 	printf("  --inc-tcp-seq: add inc tcp seq action to flow actions\n"
214 		"tcp seq will be increments by 1\n");
215 	printf("  --dec-tcp-seq: add dec tcp seq action to flow actions\n"
216 		"tcp seq will be decrements by 1\n");
217 	printf("  --set-ttl: add set ttl action to flow actions\n"
218 		"L3 ttl to be set is random each flow\n");
219 	printf("  --dec-ttl: add dec ttl action to flow actions\n"
220 		"L3 ttl will be decrements by 1\n");
221 	printf("  --set-ipv4-dscp: add set ipv4 dscp action to flow actions\n"
222 		"ipv4 dscp value to be set is random each flow\n");
223 	printf("  --set-ipv6-dscp: add set ipv6 dscp action to flow actions\n"
224 		"ipv6 dscp value to be set is random each flow\n");
225 	printf("  --flag: add flag action to flow actions\n");
226 	printf("  --meter: add meter action to flow actions\n");
227 	printf("  --raw-encap=<data>: add raw encap action to flow actions\n"
228 		"Data is the data needed to be encaped\n"
229 		"Example: raw-encap=ether,ipv4,udp,vxlan\n");
230 	printf("  --raw-decap=<data>: add raw decap action to flow actions\n"
231 		"Data is the data needed to be decaped\n"
232 		"Example: raw-decap=ether,ipv4,udp,vxlan\n");
233 	printf("  --vxlan-encap: add vxlan-encap action to flow actions\n"
234 		"Encapped data is fixed with pattern: ether,ipv4,udp,vxlan\n"
235 		"With fixed values\n");
236 	printf("  --vxlan-decap: add vxlan_decap action to flow actions\n");
237 }
238 
239 static void
240 args_parse(int argc, char **argv)
241 {
242 	uint64_t pm;
243 	char **argvopt;
244 	char *token;
245 	char *end;
246 	int n, opt;
247 	int opt_idx;
248 	size_t i;
249 
250 	static const struct option_dict {
251 		const char *str;
252 		const uint64_t mask;
253 		uint64_t *map;
254 		uint8_t *map_idx;
255 
256 	} flow_options[] = {
257 		{
258 			.str = "ether",
259 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ETH),
260 			.map = &flow_items[0],
261 			.map_idx = &items_idx
262 		},
263 		{
264 			.str = "ipv4",
265 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_IPV4),
266 			.map = &flow_items[0],
267 			.map_idx = &items_idx
268 		},
269 		{
270 			.str = "ipv6",
271 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_IPV6),
272 			.map = &flow_items[0],
273 			.map_idx = &items_idx
274 		},
275 		{
276 			.str = "vlan",
277 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VLAN),
278 			.map = &flow_items[0],
279 			.map_idx = &items_idx
280 		},
281 		{
282 			.str = "tcp",
283 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_TCP),
284 			.map = &flow_items[0],
285 			.map_idx = &items_idx
286 		},
287 		{
288 			.str = "udp",
289 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_UDP),
290 			.map = &flow_items[0],
291 			.map_idx = &items_idx
292 		},
293 		{
294 			.str = "vxlan",
295 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VXLAN),
296 			.map = &flow_items[0],
297 			.map_idx = &items_idx
298 		},
299 		{
300 			.str = "vxlan-gpe",
301 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VXLAN_GPE),
302 			.map = &flow_items[0],
303 			.map_idx = &items_idx
304 		},
305 		{
306 			.str = "gre",
307 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GRE),
308 			.map = &flow_items[0],
309 			.map_idx = &items_idx
310 		},
311 		{
312 			.str = "geneve",
313 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GENEVE),
314 			.map = &flow_items[0],
315 			.map_idx = &items_idx
316 		},
317 		{
318 			.str = "gtp",
319 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GTP),
320 			.map = &flow_items[0],
321 			.map_idx = &items_idx
322 		},
323 		{
324 			.str = "meta",
325 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_META),
326 			.map = &flow_items[0],
327 			.map_idx = &items_idx
328 		},
329 		{
330 			.str = "tag",
331 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_TAG),
332 			.map = &flow_items[0],
333 			.map_idx = &items_idx
334 		},
335 		{
336 			.str = "icmpv4",
337 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ICMP),
338 			.map = &flow_items[0],
339 			.map_idx = &items_idx
340 		},
341 		{
342 			.str = "icmpv6",
343 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ICMP6),
344 			.map = &flow_items[0],
345 			.map_idx = &items_idx
346 		},
347 		{
348 			.str = "ingress",
349 			.mask = INGRESS,
350 			.map = &flow_attrs[0],
351 			.map_idx = &attrs_idx
352 		},
353 		{
354 			.str = "egress",
355 			.mask = EGRESS,
356 			.map = &flow_attrs[0],
357 			.map_idx = &attrs_idx
358 		},
359 		{
360 			.str = "transfer",
361 			.mask = TRANSFER,
362 			.map = &flow_attrs[0],
363 			.map_idx = &attrs_idx
364 		},
365 		{
366 			.str = "port-id",
367 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_PORT_ID),
368 			.map = &flow_actions[0],
369 			.map_idx = &actions_idx
370 		},
371 		{
372 			.str = "rss",
373 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_RSS),
374 			.map = &flow_actions[0],
375 			.map_idx = &actions_idx
376 		},
377 		{
378 			.str = "queue",
379 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_QUEUE),
380 			.map = &flow_actions[0],
381 			.map_idx = &actions_idx
382 		},
383 		{
384 			.str = "jump",
385 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_JUMP),
386 			.map = &flow_actions[0],
387 			.map_idx = &actions_idx
388 		},
389 		{
390 			.str = "mark",
391 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_MARK),
392 			.map = &flow_actions[0],
393 			.map_idx = &actions_idx
394 		},
395 		{
396 			.str = "count",
397 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_COUNT),
398 			.map = &flow_actions[0],
399 			.map_idx = &actions_idx
400 		},
401 		{
402 			.str = "set-meta",
403 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_SET_META),
404 			.map = &flow_actions[0],
405 			.map_idx = &actions_idx
406 		},
407 		{
408 			.str = "set-tag",
409 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_SET_TAG),
410 			.map = &flow_actions[0],
411 			.map_idx = &actions_idx
412 		},
413 		{
414 			.str = "drop",
415 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_DROP),
416 			.map = &flow_actions[0],
417 			.map_idx = &actions_idx
418 		},
419 		{
420 			.str = "set-src-mac",
421 			.mask = FLOW_ACTION_MASK(
422 				RTE_FLOW_ACTION_TYPE_SET_MAC_SRC
423 			),
424 			.map = &flow_actions[0],
425 			.map_idx = &actions_idx
426 		},
427 		{
428 			.str = "set-dst-mac",
429 			.mask = FLOW_ACTION_MASK(
430 				RTE_FLOW_ACTION_TYPE_SET_MAC_DST
431 			),
432 			.map = &flow_actions[0],
433 			.map_idx = &actions_idx
434 		},
435 		{
436 			.str = "set-src-ipv4",
437 			.mask = FLOW_ACTION_MASK(
438 				RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC
439 			),
440 			.map = &flow_actions[0],
441 			.map_idx = &actions_idx
442 		},
443 		{
444 			.str = "set-dst-ipv4",
445 			.mask = FLOW_ACTION_MASK(
446 				RTE_FLOW_ACTION_TYPE_SET_IPV4_DST
447 			),
448 			.map = &flow_actions[0],
449 			.map_idx = &actions_idx
450 		},
451 		{
452 			.str = "set-src-ipv6",
453 			.mask = FLOW_ACTION_MASK(
454 				RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC
455 			),
456 			.map = &flow_actions[0],
457 			.map_idx = &actions_idx
458 		},
459 		{
460 			.str = "set-dst-ipv6",
461 			.mask = FLOW_ACTION_MASK(
462 				RTE_FLOW_ACTION_TYPE_SET_IPV6_DST
463 			),
464 			.map = &flow_actions[0],
465 			.map_idx = &actions_idx
466 		},
467 		{
468 			.str = "set-src-tp",
469 			.mask = FLOW_ACTION_MASK(
470 				RTE_FLOW_ACTION_TYPE_SET_TP_SRC
471 			),
472 			.map = &flow_actions[0],
473 			.map_idx = &actions_idx
474 		},
475 		{
476 			.str = "set-dst-tp",
477 			.mask = FLOW_ACTION_MASK(
478 				RTE_FLOW_ACTION_TYPE_SET_TP_DST
479 			),
480 			.map = &flow_actions[0],
481 			.map_idx = &actions_idx
482 		},
483 		{
484 			.str = "inc-tcp-ack",
485 			.mask = FLOW_ACTION_MASK(
486 				RTE_FLOW_ACTION_TYPE_INC_TCP_ACK
487 			),
488 			.map = &flow_actions[0],
489 			.map_idx = &actions_idx
490 		},
491 		{
492 			.str = "dec-tcp-ack",
493 			.mask = FLOW_ACTION_MASK(
494 				RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK
495 			),
496 			.map = &flow_actions[0],
497 			.map_idx = &actions_idx
498 		},
499 		{
500 			.str = "inc-tcp-seq",
501 			.mask = FLOW_ACTION_MASK(
502 				RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ
503 			),
504 			.map = &flow_actions[0],
505 			.map_idx = &actions_idx
506 		},
507 		{
508 			.str = "dec-tcp-seq",
509 			.mask = FLOW_ACTION_MASK(
510 				RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ
511 			),
512 			.map = &flow_actions[0],
513 			.map_idx = &actions_idx
514 		},
515 		{
516 			.str = "set-ttl",
517 			.mask = FLOW_ACTION_MASK(
518 				RTE_FLOW_ACTION_TYPE_SET_TTL
519 			),
520 			.map = &flow_actions[0],
521 			.map_idx = &actions_idx
522 		},
523 		{
524 			.str = "dec-ttl",
525 			.mask = FLOW_ACTION_MASK(
526 				RTE_FLOW_ACTION_TYPE_DEC_TTL
527 			),
528 			.map = &flow_actions[0],
529 			.map_idx = &actions_idx
530 		},
531 		{
532 			.str = "set-ipv4-dscp",
533 			.mask = FLOW_ACTION_MASK(
534 				RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP
535 			),
536 			.map = &flow_actions[0],
537 			.map_idx = &actions_idx
538 		},
539 		{
540 			.str = "set-ipv6-dscp",
541 			.mask = FLOW_ACTION_MASK(
542 				RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP
543 			),
544 			.map = &flow_actions[0],
545 			.map_idx = &actions_idx
546 		},
547 		{
548 			.str = "flag",
549 			.mask = FLOW_ACTION_MASK(
550 				RTE_FLOW_ACTION_TYPE_FLAG
551 			),
552 			.map = &flow_actions[0],
553 			.map_idx = &actions_idx
554 		},
555 		{
556 			.str = "meter",
557 			.mask = FLOW_ACTION_MASK(
558 				RTE_FLOW_ACTION_TYPE_METER
559 			),
560 			.map = &flow_actions[0],
561 			.map_idx = &actions_idx
562 		},
563 		{
564 			.str = "vxlan-encap",
565 			.mask = FLOW_ACTION_MASK(
566 				RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP
567 			),
568 			.map = &flow_actions[0],
569 			.map_idx = &actions_idx
570 		},
571 		{
572 			.str = "vxlan-decap",
573 			.mask = FLOW_ACTION_MASK(
574 				RTE_FLOW_ACTION_TYPE_VXLAN_DECAP
575 			),
576 			.map = &flow_actions[0],
577 			.map_idx = &actions_idx
578 		},
579 	};
580 
581 	static const struct option lgopts[] = {
582 		/* Control */
583 		{ "help",                       0, 0, 0 },
584 		{ "rules-count",                1, 0, 0 },
585 		{ "rules-batch",                1, 0, 0 },
586 		{ "dump-iterations",            0, 0, 0 },
587 		{ "deletion-rate",              0, 0, 0 },
588 		{ "dump-socket-mem",            0, 0, 0 },
589 		{ "enable-fwd",                 0, 0, 0 },
590 		{ "unique-data",                0, 0, 0 },
591 		{ "portmask",                   1, 0, 0 },
592 		{ "cores",                      1, 0, 0 },
593 		{ "meter-profile-alg",          1, 0, 0 },
594 		{ "rxq",                        1, 0, 0 },
595 		{ "txq",                        1, 0, 0 },
596 		{ "rxd",                        1, 0, 0 },
597 		{ "txd",                        1, 0, 0 },
598 		{ "mbuf-size",                  1, 0, 0 },
599 		{ "mbuf-cache-size",            1, 0, 0 },
600 		{ "total-mbuf-count",           1, 0, 0 },
601 		/* Attributes */
602 		{ "ingress",                    0, 0, 0 },
603 		{ "egress",                     0, 0, 0 },
604 		{ "transfer",                   0, 0, 0 },
605 		{ "group",                      1, 0, 0 },
606 		/* Items */
607 		{ "ether",                      0, 0, 0 },
608 		{ "vlan",                       0, 0, 0 },
609 		{ "ipv4",                       0, 0, 0 },
610 		{ "ipv6",                       0, 0, 0 },
611 		{ "tcp",                        0, 0, 0 },
612 		{ "udp",                        0, 0, 0 },
613 		{ "vxlan",                      0, 0, 0 },
614 		{ "vxlan-gpe",                  0, 0, 0 },
615 		{ "gre",                        0, 0, 0 },
616 		{ "geneve",                     0, 0, 0 },
617 		{ "gtp",                        0, 0, 0 },
618 		{ "meta",                       0, 0, 0 },
619 		{ "tag",                        0, 0, 0 },
620 		{ "icmpv4",                     0, 0, 0 },
621 		{ "icmpv6",                     0, 0, 0 },
622 		/* Actions */
623 		{ "port-id",                    2, 0, 0 },
624 		{ "rss",                        0, 0, 0 },
625 		{ "queue",                      0, 0, 0 },
626 		{ "jump",                       0, 0, 0 },
627 		{ "mark",                       0, 0, 0 },
628 		{ "count",                      0, 0, 0 },
629 		{ "set-meta",                   0, 0, 0 },
630 		{ "set-tag",                    0, 0, 0 },
631 		{ "drop",                       0, 0, 0 },
632 		{ "hairpin-queue",              1, 0, 0 },
633 		{ "hairpin-rss",                1, 0, 0 },
634 		{ "set-src-mac",                0, 0, 0 },
635 		{ "set-dst-mac",                0, 0, 0 },
636 		{ "set-src-ipv4",               0, 0, 0 },
637 		{ "set-dst-ipv4",               0, 0, 0 },
638 		{ "set-src-ipv6",               0, 0, 0 },
639 		{ "set-dst-ipv6",               0, 0, 0 },
640 		{ "set-src-tp",                 0, 0, 0 },
641 		{ "set-dst-tp",                 0, 0, 0 },
642 		{ "inc-tcp-ack",                0, 0, 0 },
643 		{ "dec-tcp-ack",                0, 0, 0 },
644 		{ "inc-tcp-seq",                0, 0, 0 },
645 		{ "dec-tcp-seq",                0, 0, 0 },
646 		{ "set-ttl",                    0, 0, 0 },
647 		{ "dec-ttl",                    0, 0, 0 },
648 		{ "set-ipv4-dscp",              0, 0, 0 },
649 		{ "set-ipv6-dscp",              0, 0, 0 },
650 		{ "flag",                       0, 0, 0 },
651 		{ "meter",                      0, 0, 0 },
652 		{ "raw-encap",                  1, 0, 0 },
653 		{ "raw-decap",                  1, 0, 0 },
654 		{ "vxlan-encap",                0, 0, 0 },
655 		{ "vxlan-decap",                0, 0, 0 },
656 	};
657 
658 	RTE_ETH_FOREACH_DEV(i)
659 		ports_mask |= 1 << i;
660 
661 	for (i = 0; i < RTE_MAX_ETHPORTS; i++)
662 		dst_ports[i] = PORT_ID_DST;
663 
664 	hairpin_queues_num = 0;
665 	argvopt = argv;
666 
667 	printf(":: Flow -> ");
668 	while ((opt = getopt_long(argc, argvopt, "",
669 				lgopts, &opt_idx)) != EOF) {
670 		switch (opt) {
671 		case 0:
672 			if (strcmp(lgopts[opt_idx].name, "help") == 0) {
673 				usage(argv[0]);
674 				exit(EXIT_SUCCESS);
675 			}
676 
677 			if (strcmp(lgopts[opt_idx].name, "group") == 0) {
678 				n = atoi(optarg);
679 				if (n >= 0)
680 					flow_group = n;
681 				else
682 					rte_exit(EXIT_FAILURE,
683 						"flow group should be >= 0\n");
684 				printf("group %d / ", flow_group);
685 			}
686 
687 			for (i = 0; i < RTE_DIM(flow_options); i++)
688 				if (strcmp(lgopts[opt_idx].name,
689 						flow_options[i].str) == 0) {
690 					flow_options[i].map[
691 					(*flow_options[i].map_idx)++] =
692 						flow_options[i].mask;
693 					printf("%s / ", flow_options[i].str);
694 				}
695 
696 			if (strcmp(lgopts[opt_idx].name,
697 					"hairpin-rss") == 0) {
698 				n = atoi(optarg);
699 				if (n > 0)
700 					hairpin_queues_num = n;
701 				else
702 					rte_exit(EXIT_FAILURE,
703 						"Hairpin queues should be > 0\n");
704 
705 				flow_actions[actions_idx++] =
706 					HAIRPIN_RSS_ACTION;
707 				printf("hairpin-rss / ");
708 			}
709 			if (strcmp(lgopts[opt_idx].name,
710 					"hairpin-queue") == 0) {
711 				n = atoi(optarg);
712 				if (n > 0)
713 					hairpin_queues_num = n;
714 				else
715 					rte_exit(EXIT_FAILURE,
716 						"Hairpin queues should be > 0\n");
717 
718 				flow_actions[actions_idx++] =
719 					HAIRPIN_QUEUE_ACTION;
720 				printf("hairpin-queue / ");
721 			}
722 
723 			if (strcmp(lgopts[opt_idx].name, "raw-encap") == 0) {
724 				printf("raw-encap ");
725 				flow_actions[actions_idx++] =
726 					FLOW_ITEM_MASK(
727 						RTE_FLOW_ACTION_TYPE_RAW_ENCAP
728 					);
729 
730 				token = strtok(optarg, ",");
731 				while (token != NULL) {
732 					for (i = 0; i < RTE_DIM(flow_options); i++) {
733 						if (strcmp(flow_options[i].str, token) == 0) {
734 							printf("%s,", token);
735 							encap_data |= flow_options[i].mask;
736 							break;
737 						}
738 						/* Reached last item with no match */
739 						if (i == (RTE_DIM(flow_options) - 1))
740 							rte_exit(EXIT_FAILURE,
741 								"Invalid encap item: %s\n", token);
742 					}
743 					token = strtok(NULL, ",");
744 				}
745 				printf(" / ");
746 			}
747 			if (strcmp(lgopts[opt_idx].name, "raw-decap") == 0) {
748 				printf("raw-decap ");
749 				flow_actions[actions_idx++] =
750 					FLOW_ITEM_MASK(
751 						RTE_FLOW_ACTION_TYPE_RAW_DECAP
752 					);
753 
754 				token = strtok(optarg, ",");
755 				while (token != NULL) {
756 					for (i = 0; i < RTE_DIM(flow_options); i++) {
757 						if (strcmp(flow_options[i].str, token) == 0) {
758 							printf("%s,", token);
759 							decap_data |= flow_options[i].mask;
760 							break;
761 						}
762 						/* Reached last item with no match */
763 						if (i == (RTE_DIM(flow_options) - 1))
764 							rte_exit(EXIT_FAILURE,
765 								"Invalid decap item %s\n", token);
766 					}
767 					token = strtok(NULL, ",");
768 				}
769 				printf(" / ");
770 			}
771 			/* Control */
772 			if (strcmp(lgopts[opt_idx].name,
773 					"rules-batch") == 0) {
774 				n = atoi(optarg);
775 				if (n >= DEFAULT_RULES_BATCH)
776 					rules_batch = n;
777 				else {
778 					rte_exit(EXIT_FAILURE,
779 						"rules_batch should be >= %d\n",
780 						DEFAULT_RULES_BATCH);
781 				}
782 			}
783 			if (strcmp(lgopts[opt_idx].name,
784 					"rules-count") == 0) {
785 				n = atoi(optarg);
786 				if (n >= (int) rules_batch)
787 					rules_count = n;
788 				else {
789 					rte_exit(EXIT_FAILURE,
790 						"rules_count should be >= %d\n",
791 						rules_batch);
792 				}
793 			}
794 			if (strcmp(lgopts[opt_idx].name,
795 					"dump-iterations") == 0)
796 				dump_iterations = true;
797 			if (strcmp(lgopts[opt_idx].name,
798 					"unique-data") == 0)
799 				unique_data = true;
800 			if (strcmp(lgopts[opt_idx].name,
801 					"deletion-rate") == 0)
802 				delete_flag = true;
803 			if (strcmp(lgopts[opt_idx].name,
804 					"dump-socket-mem") == 0)
805 				dump_socket_mem_flag = true;
806 			if (strcmp(lgopts[opt_idx].name,
807 					"enable-fwd") == 0)
808 				enable_fwd = true;
809 			if (strcmp(lgopts[opt_idx].name,
810 					"portmask") == 0) {
811 				/* parse hexadecimal string */
812 				end = NULL;
813 				pm = strtoull(optarg, &end, 16);
814 				if ((optarg[0] == '\0') || (end == NULL) || (*end != '\0'))
815 					rte_exit(EXIT_FAILURE, "Invalid fwd port mask\n");
816 				ports_mask = pm;
817 			}
818 			if (strcmp(lgopts[opt_idx].name,
819 					"port-id") == 0) {
820 				uint16_t port_idx = 0;
821 				char *token;
822 
823 				token = strtok(optarg, ",");
824 				while (token != NULL) {
825 					dst_ports[port_idx++] = atoi(token);
826 					token = strtok(NULL, ",");
827 				}
828 			}
829 			if (strcmp(lgopts[opt_idx].name, "rxq") == 0) {
830 				n = atoi(optarg);
831 				rx_queues_count = (uint8_t) n;
832 			}
833 			if (strcmp(lgopts[opt_idx].name, "txq") == 0) {
834 				n = atoi(optarg);
835 				tx_queues_count = (uint8_t) n;
836 			}
837 			if (strcmp(lgopts[opt_idx].name, "rxd") == 0) {
838 				n = atoi(optarg);
839 				rxd_count = (uint8_t) n;
840 			}
841 			if (strcmp(lgopts[opt_idx].name, "txd") == 0) {
842 				n = atoi(optarg);
843 				txd_count = (uint8_t) n;
844 			}
845 			if (strcmp(lgopts[opt_idx].name, "mbuf-size") == 0) {
846 				n = atoi(optarg);
847 				mbuf_size = (uint32_t) n;
848 			}
849 			if (strcmp(lgopts[opt_idx].name, "mbuf-cache-size") == 0) {
850 				n = atoi(optarg);
851 				mbuf_cache_size = (uint32_t) n;
852 			}
853 			if (strcmp(lgopts[opt_idx].name, "total-mbuf-count") == 0) {
854 				n = atoi(optarg);
855 				total_mbuf_num = (uint32_t) n;
856 			}
857 			if (strcmp(lgopts[opt_idx].name, "cores") == 0) {
858 				n = atoi(optarg);
859 				if ((int) rte_lcore_count() <= n) {
860 					rte_exit(EXIT_FAILURE,
861 						"Error: you need %d cores to run on multi-cores\n"
862 						"Existing cores are: %d\n", n, rte_lcore_count());
863 				}
864 				if (n <= RTE_MAX_LCORE && n > 0)
865 					mc_pool.cores_count = n;
866 				else {
867 					rte_exit(EXIT_FAILURE,
868 						"Error: cores count must be > 0 and < %d\n",
869 						RTE_MAX_LCORE);
870 				}
871 			}
872 			break;
873 		default:
874 			usage(argv[0]);
875 			rte_exit(EXIT_FAILURE, "Invalid option: %s\n",
876 					argv[optind]);
877 			break;
878 		}
879 	}
880 	printf("end_flow\n");
881 }
882 
883 /* Dump the socket memory statistics on console */
884 static size_t
885 dump_socket_mem(FILE *f)
886 {
887 	struct rte_malloc_socket_stats socket_stats;
888 	unsigned int i = 0;
889 	size_t total = 0;
890 	size_t alloc = 0;
891 	size_t free = 0;
892 	unsigned int n_alloc = 0;
893 	unsigned int n_free = 0;
894 	bool active_nodes = false;
895 
896 
897 	for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
898 		if (rte_malloc_get_socket_stats(i, &socket_stats) ||
899 		    !socket_stats.heap_totalsz_bytes)
900 			continue;
901 		active_nodes = true;
902 		total += socket_stats.heap_totalsz_bytes;
903 		alloc += socket_stats.heap_allocsz_bytes;
904 		free += socket_stats.heap_freesz_bytes;
905 		n_alloc += socket_stats.alloc_count;
906 		n_free += socket_stats.free_count;
907 		if (dump_socket_mem_flag) {
908 			fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
909 			fprintf(f,
910 				"\nSocket %u:\nsize(M) total: %.6lf\nalloc:"
911 				" %.6lf(%.3lf%%)\nfree: %.6lf"
912 				"\nmax: %.6lf"
913 				"\ncount alloc: %u\nfree: %u\n",
914 				i,
915 				socket_stats.heap_totalsz_bytes / 1.0e6,
916 				socket_stats.heap_allocsz_bytes / 1.0e6,
917 				(double)socket_stats.heap_allocsz_bytes * 100 /
918 				(double)socket_stats.heap_totalsz_bytes,
919 				socket_stats.heap_freesz_bytes / 1.0e6,
920 				socket_stats.greatest_free_size / 1.0e6,
921 				socket_stats.alloc_count,
922 				socket_stats.free_count);
923 				fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
924 		}
925 	}
926 	if (dump_socket_mem_flag && active_nodes) {
927 		fprintf(f,
928 			"\nTotal: size(M)\ntotal: %.6lf"
929 			"\nalloc: %.6lf(%.3lf%%)\nfree: %.6lf"
930 			"\ncount alloc: %u\nfree: %u\n",
931 			total / 1.0e6, alloc / 1.0e6,
932 			(double)alloc * 100 / (double)total, free / 1.0e6,
933 			n_alloc, n_free);
934 		fprintf(f, "::::::::::::::::::::::::::::::::::::::::\n");
935 	}
936 	return alloc;
937 }
938 
939 static void
940 print_flow_error(struct rte_flow_error error)
941 {
942 	printf("Flow can't be created %d message: %s\n",
943 		error.type,
944 		error.message ? error.message : "(no stated reason)");
945 }
946 
947 static inline void
948 print_rules_batches(double *cpu_time_per_batch)
949 {
950 	uint8_t idx;
951 	double delta;
952 	double rate;
953 
954 	for (idx = 0; idx < MAX_BATCHES_COUNT; idx++) {
955 		if (!cpu_time_per_batch[idx])
956 			break;
957 		delta = (double)(rules_batch / cpu_time_per_batch[idx]);
958 		rate = delta / 1000; /* Save rate in K unit. */
959 		printf(":: Rules batch #%d: %d rules "
960 			"in %f sec[ Rate = %f K Rule/Sec ]\n",
961 			idx, rules_batch,
962 			cpu_time_per_batch[idx], rate);
963 	}
964 }
965 
966 
967 static inline int
968 has_meter(void)
969 {
970 	int i;
971 
972 	for (i = 0; i < MAX_ACTIONS_NUM; i++) {
973 		if (flow_actions[i] == 0)
974 			break;
975 		if (flow_actions[i]
976 				& FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_METER))
977 			return 1;
978 	}
979 	return 0;
980 }
981 
982 static void
983 create_meter_rule(int port_id, uint32_t counter)
984 {
985 	int ret;
986 	struct rte_mtr_params params;
987 	uint32_t default_prof_id = 100;
988 	struct rte_mtr_error error;
989 
990 	memset(&params, 0, sizeof(struct rte_mtr_params));
991 	params.meter_enable = 1;
992 	params.stats_mask = 0xffff;
993 	params.use_prev_mtr_color = 0;
994 	params.dscp_table = NULL;
995 
996 	/*create meter*/
997 	params.meter_profile_id = default_prof_id;
998 	ret = rte_mtr_create(port_id, counter, &params, 1, &error);
999 	if (ret != 0) {
1000 		printf("Port %u create meter idx(%d) error(%d) message: %s\n",
1001 			port_id, counter, error.type,
1002 			error.message ? error.message : "(no stated reason)");
1003 		rte_exit(EXIT_FAILURE, "Error in creating meter\n");
1004 	}
1005 }
1006 
1007 static void
1008 destroy_meter_rule(int port_id, uint32_t counter)
1009 {
1010 	struct rte_mtr_error error;
1011 
1012 	if (rte_mtr_destroy(port_id, counter, &error)) {
1013 		printf("Port %u destroy meter(%d) error(%d) message: %s\n",
1014 			port_id, counter, error.type,
1015 			error.message ? error.message : "(no stated reason)");
1016 		rte_exit(EXIT_FAILURE, "Error in deleting meter rule\n");
1017 	}
1018 }
1019 
1020 static void
1021 meters_handler(int port_id, uint8_t core_id, uint8_t ops)
1022 {
1023 	uint64_t start_batch;
1024 	double cpu_time_used, insertion_rate;
1025 	int rules_count_per_core, rules_batch_idx;
1026 	uint32_t counter, start_counter = 0, end_counter;
1027 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1028 
1029 	rules_count_per_core = rules_count / mc_pool.cores_count;
1030 
1031 	if (core_id)
1032 		start_counter = core_id * rules_count_per_core;
1033 	end_counter = (core_id + 1) * rules_count_per_core;
1034 
1035 	cpu_time_used = 0;
1036 	start_batch = rte_get_timer_cycles();
1037 	for (counter = start_counter; counter < end_counter; counter++) {
1038 		if (ops == METER_CREATE)
1039 			create_meter_rule(port_id, counter);
1040 		else
1041 			destroy_meter_rule(port_id, counter);
1042 		/*
1043 		 * Save the insertion rate for rules batch.
1044 		 * Check if the insertion reached the rules
1045 		 * patch counter, then save the insertion rate
1046 		 * for this batch.
1047 		 */
1048 		if (!((counter + 1) % rules_batch)) {
1049 			rules_batch_idx = ((counter + 1) / rules_batch) - 1;
1050 			cpu_time_per_batch[rules_batch_idx] =
1051 				((double)(rte_get_timer_cycles() - start_batch))
1052 				/ rte_get_timer_hz();
1053 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1054 			start_batch = rte_get_timer_cycles();
1055 		}
1056 	}
1057 
1058 	/* Print insertion rates for all batches */
1059 	if (dump_iterations)
1060 		print_rules_batches(cpu_time_per_batch);
1061 
1062 	insertion_rate =
1063 		((double) (rules_count_per_core / cpu_time_used) / 1000);
1064 
1065 	/* Insertion rate for all rules in one core */
1066 	printf(":: Port %d :: Core %d Meter %s :: start @[%d] - end @[%d],"
1067 		" use:%.02fs, rate:%.02fk Rule/Sec\n",
1068 		port_id, core_id, ops == METER_CREATE ? "create" : "delete",
1069 		start_counter, end_counter - 1,
1070 		cpu_time_used, insertion_rate);
1071 
1072 	if (ops == METER_CREATE)
1073 		mc_pool.meters_record.insertion[port_id][core_id]
1074 			= cpu_time_used;
1075 	else
1076 		mc_pool.meters_record.deletion[port_id][core_id]
1077 			= cpu_time_used;
1078 }
1079 
1080 static void
1081 destroy_meter_profile(void)
1082 {
1083 	struct rte_mtr_error error;
1084 	uint16_t nr_ports;
1085 	int port_id;
1086 
1087 	nr_ports = rte_eth_dev_count_avail();
1088 	for (port_id = 0; port_id < nr_ports; port_id++) {
1089 		/* If port outside portmask */
1090 		if (!((ports_mask >> port_id) & 0x1))
1091 			continue;
1092 
1093 		if (rte_mtr_meter_profile_delete
1094 			(port_id, DEFAULT_METER_PROF_ID, &error)) {
1095 			printf("Port %u del profile error(%d) message: %s\n",
1096 				port_id, error.type,
1097 				error.message ? error.message : "(no stated reason)");
1098 			rte_exit(EXIT_FAILURE, "Error: Destroy meter profile Failed!\n");
1099 		}
1100 	}
1101 }
1102 
1103 static void
1104 create_meter_profile(void)
1105 {
1106 	uint16_t nr_ports;
1107 	int ret, port_id;
1108 	struct rte_mtr_meter_profile mp;
1109 	struct rte_mtr_error error;
1110 
1111 	/*
1112 	 *currently , only create one meter file for one port
1113 	 *1 meter profile -> N meter rules -> N rte flows
1114 	 */
1115 	memset(&mp, 0, sizeof(struct rte_mtr_meter_profile));
1116 	nr_ports = rte_eth_dev_count_avail();
1117 	for (port_id = 0; port_id < nr_ports; port_id++) {
1118 		/* If port outside portmask */
1119 		if (!((ports_mask >> port_id) & 0x1))
1120 			continue;
1121 
1122 		mp.alg = RTE_MTR_SRTCM_RFC2697;
1123 		mp.srtcm_rfc2697.cir = METER_CIR;
1124 		mp.srtcm_rfc2697.cbs = METER_CIR / 8;
1125 		mp.srtcm_rfc2697.ebs = 0;
1126 
1127 		ret = rte_mtr_meter_profile_add
1128 			(port_id, DEFAULT_METER_PROF_ID, &mp, &error);
1129 		if (ret != 0) {
1130 			printf("Port %u create Profile error(%d) message: %s\n",
1131 				port_id, error.type,
1132 				error.message ? error.message : "(no stated reason)");
1133 			rte_exit(EXIT_FAILURE, "Error: Creation meter profile Failed!\n");
1134 		}
1135 	}
1136 }
1137 
1138 static inline void
1139 destroy_flows(int port_id, uint8_t core_id, struct rte_flow **flows_list)
1140 {
1141 	struct rte_flow_error error;
1142 	clock_t start_batch, end_batch;
1143 	double cpu_time_used = 0;
1144 	double deletion_rate;
1145 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1146 	double delta;
1147 	uint32_t i;
1148 	int rules_batch_idx;
1149 	int rules_count_per_core;
1150 
1151 	rules_count_per_core = rules_count / mc_pool.cores_count;
1152 	/* If group > 0 , should add 1 flow which created in group 0 */
1153 	if (flow_group > 0 && core_id == 0)
1154 		rules_count_per_core++;
1155 
1156 	start_batch = rte_get_timer_cycles();
1157 	for (i = 0; i < (uint32_t) rules_count_per_core; i++) {
1158 		if (flows_list[i] == 0)
1159 			break;
1160 
1161 		memset(&error, 0x33, sizeof(error));
1162 		if (rte_flow_destroy(port_id, flows_list[i], &error)) {
1163 			print_flow_error(error);
1164 			rte_exit(EXIT_FAILURE, "Error in deleting flow\n");
1165 		}
1166 
1167 		/*
1168 		 * Save the deletion rate for rules batch.
1169 		 * Check if the deletion reached the rules
1170 		 * patch counter, then save the deletion rate
1171 		 * for this batch.
1172 		 */
1173 		if (!((i + 1) % rules_batch)) {
1174 			end_batch = rte_get_timer_cycles();
1175 			delta = (double) (end_batch - start_batch);
1176 			rules_batch_idx = ((i + 1) / rules_batch) - 1;
1177 			cpu_time_per_batch[rules_batch_idx] = delta / rte_get_timer_hz();
1178 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1179 			start_batch = rte_get_timer_cycles();
1180 		}
1181 	}
1182 
1183 	/* Print deletion rates for all batches */
1184 	if (dump_iterations)
1185 		print_rules_batches(cpu_time_per_batch);
1186 
1187 	/* Deletion rate for all rules */
1188 	deletion_rate = ((double) (rules_count_per_core / cpu_time_used) / 1000);
1189 	printf(":: Port %d :: Core %d :: Rules deletion rate -> %f K Rule/Sec\n",
1190 		port_id, core_id, deletion_rate);
1191 	printf(":: Port %d :: Core %d :: The time for deleting %d rules is %f seconds\n",
1192 		port_id, core_id, rules_count_per_core, cpu_time_used);
1193 
1194 	mc_pool.flows_record.deletion[port_id][core_id] = cpu_time_used;
1195 }
1196 
1197 static struct rte_flow **
1198 insert_flows(int port_id, uint8_t core_id, uint16_t dst_port_id)
1199 {
1200 	struct rte_flow **flows_list;
1201 	struct rte_flow_error error;
1202 	clock_t start_batch, end_batch;
1203 	double first_flow_latency;
1204 	double cpu_time_used;
1205 	double insertion_rate;
1206 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1207 	double delta;
1208 	uint32_t flow_index;
1209 	uint32_t counter, start_counter = 0, end_counter;
1210 	uint64_t global_items[MAX_ITEMS_NUM] = { 0 };
1211 	uint64_t global_actions[MAX_ACTIONS_NUM] = { 0 };
1212 	int rules_batch_idx;
1213 	int rules_count_per_core;
1214 
1215 	rules_count_per_core = rules_count / mc_pool.cores_count;
1216 
1217 	/* Set boundaries of rules for each core. */
1218 	if (core_id)
1219 		start_counter = core_id * rules_count_per_core;
1220 	end_counter = (core_id + 1) * rules_count_per_core;
1221 
1222 	global_items[0] = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ETH);
1223 	global_actions[0] = FLOW_ITEM_MASK(RTE_FLOW_ACTION_TYPE_JUMP);
1224 
1225 	flows_list = rte_zmalloc("flows_list",
1226 		(sizeof(struct rte_flow *) * rules_count_per_core) + 1, 0);
1227 	if (flows_list == NULL)
1228 		rte_exit(EXIT_FAILURE, "No Memory available!\n");
1229 
1230 	cpu_time_used = 0;
1231 	flow_index = 0;
1232 	if (flow_group > 0 && core_id == 0) {
1233 		/*
1234 		 * Create global rule to jump into flow_group,
1235 		 * this way the app will avoid the default rules.
1236 		 *
1237 		 * This rule will be created only once.
1238 		 *
1239 		 * Global rule:
1240 		 * group 0 eth / end actions jump group <flow_group>
1241 		 */
1242 		flow = generate_flow(port_id, 0, flow_attrs,
1243 			global_items, global_actions,
1244 			flow_group, 0, 0, 0, 0, dst_port_id, core_id,
1245 			rx_queues_count, unique_data, &error);
1246 
1247 		if (flow == NULL) {
1248 			print_flow_error(error);
1249 			rte_exit(EXIT_FAILURE, "Error in creating flow\n");
1250 		}
1251 		flows_list[flow_index++] = flow;
1252 	}
1253 
1254 	start_batch = rte_get_timer_cycles();
1255 	for (counter = start_counter; counter < end_counter; counter++) {
1256 		flow = generate_flow(port_id, flow_group,
1257 			flow_attrs, flow_items, flow_actions,
1258 			JUMP_ACTION_TABLE, counter,
1259 			hairpin_queues_num, encap_data,
1260 			decap_data, dst_port_id,
1261 			core_id, rx_queues_count,
1262 			unique_data, &error);
1263 
1264 		if (!counter) {
1265 			first_flow_latency = (double) (rte_get_timer_cycles() - start_batch);
1266 			first_flow_latency /= rte_get_timer_hz();
1267 			/* In millisecond */
1268 			first_flow_latency *= 1000;
1269 			printf(":: First Flow Latency :: Port %d :: First flow "
1270 				"installed in %f milliseconds\n",
1271 				port_id, first_flow_latency);
1272 		}
1273 
1274 		if (force_quit)
1275 			counter = end_counter;
1276 
1277 		if (!flow) {
1278 			print_flow_error(error);
1279 			rte_exit(EXIT_FAILURE, "Error in creating flow\n");
1280 		}
1281 
1282 		flows_list[flow_index++] = flow;
1283 
1284 		/*
1285 		 * Save the insertion rate for rules batch.
1286 		 * Check if the insertion reached the rules
1287 		 * patch counter, then save the insertion rate
1288 		 * for this batch.
1289 		 */
1290 		if (!((counter + 1) % rules_batch)) {
1291 			end_batch = rte_get_timer_cycles();
1292 			delta = (double) (end_batch - start_batch);
1293 			rules_batch_idx = ((counter + 1) / rules_batch) - 1;
1294 			cpu_time_per_batch[rules_batch_idx] = delta / rte_get_timer_hz();
1295 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1296 			start_batch = rte_get_timer_cycles();
1297 		}
1298 	}
1299 
1300 	/* Print insertion rates for all batches */
1301 	if (dump_iterations)
1302 		print_rules_batches(cpu_time_per_batch);
1303 
1304 	printf(":: Port %d :: Core %d boundaries :: start @[%d] - end @[%d]\n",
1305 		port_id, core_id, start_counter, end_counter - 1);
1306 
1307 	/* Insertion rate for all rules in one core */
1308 	insertion_rate = ((double) (rules_count_per_core / cpu_time_used) / 1000);
1309 	printf(":: Port %d :: Core %d :: Rules insertion rate -> %f K Rule/Sec\n",
1310 		port_id, core_id, insertion_rate);
1311 	printf(":: Port %d :: Core %d :: The time for creating %d in rules %f seconds\n",
1312 		port_id, core_id, rules_count_per_core, cpu_time_used);
1313 
1314 	mc_pool.flows_record.insertion[port_id][core_id] = cpu_time_used;
1315 	return flows_list;
1316 }
1317 
1318 static void
1319 flows_handler(uint8_t core_id)
1320 {
1321 	struct rte_flow **flows_list;
1322 	uint16_t port_idx = 0;
1323 	uint16_t nr_ports;
1324 	int port_id;
1325 
1326 	nr_ports = rte_eth_dev_count_avail();
1327 
1328 	if (rules_batch > rules_count)
1329 		rules_batch = rules_count;
1330 
1331 	printf(":: Rules Count per port: %d\n\n", rules_count);
1332 
1333 	for (port_id = 0; port_id < nr_ports; port_id++) {
1334 		/* If port outside portmask */
1335 		if (!((ports_mask >> port_id) & 0x1))
1336 			continue;
1337 
1338 		/* Insertion part. */
1339 		mc_pool.last_alloc[core_id] = (int64_t)dump_socket_mem(stdout);
1340 		if (has_meter())
1341 			meters_handler(port_id, core_id, METER_CREATE);
1342 		flows_list = insert_flows(port_id, core_id,
1343 						dst_ports[port_idx++]);
1344 		if (flows_list == NULL)
1345 			rte_exit(EXIT_FAILURE, "Error: Insertion Failed!\n");
1346 		mc_pool.current_alloc[core_id] = (int64_t)dump_socket_mem(stdout);
1347 
1348 		/* Deletion part. */
1349 		if (delete_flag) {
1350 			destroy_flows(port_id, core_id, flows_list);
1351 			if (has_meter())
1352 				meters_handler(port_id, core_id, METER_DELETE);
1353 		}
1354 	}
1355 }
1356 
1357 static void
1358 dump_used_cpu_time(const char *item,
1359 		uint16_t port, struct used_cpu_time *used_time)
1360 {
1361 	uint32_t i;
1362 	/* Latency: total count of rte rules divided
1363 	 * over max time used by thread between all
1364 	 * threads time.
1365 	 *
1366 	 * Throughput: total count of rte rules divided
1367 	 * over the average of the time cosumed by all
1368 	 * threads time.
1369 	 */
1370 	double insertion_latency_time;
1371 	double insertion_throughput_time;
1372 	double deletion_latency_time;
1373 	double deletion_throughput_time;
1374 	double insertion_latency, insertion_throughput;
1375 	double deletion_latency, deletion_throughput;
1376 
1377 	/* Save first insertion/deletion rates from first thread.
1378 	 * Start comparing with all threads, if any thread used
1379 	 * time more than current saved, replace it.
1380 	 *
1381 	 * Thus in the end we will have the max time used for
1382 	 * insertion/deletion by one thread.
1383 	 *
1384 	 * As for memory consumption, save the min of all threads
1385 	 * of last alloc, and save the max for all threads for
1386 	 * current alloc.
1387 	 */
1388 
1389 	insertion_latency_time = used_time->insertion[port][0];
1390 	deletion_latency_time = used_time->deletion[port][0];
1391 	insertion_throughput_time = used_time->insertion[port][0];
1392 	deletion_throughput_time = used_time->deletion[port][0];
1393 
1394 	i = mc_pool.cores_count;
1395 	while (i-- > 1) {
1396 		insertion_throughput_time += used_time->insertion[port][i];
1397 		deletion_throughput_time += used_time->deletion[port][i];
1398 		if (insertion_latency_time < used_time->insertion[port][i])
1399 			insertion_latency_time = used_time->insertion[port][i];
1400 		if (deletion_latency_time < used_time->deletion[port][i])
1401 			deletion_latency_time = used_time->deletion[port][i];
1402 	}
1403 
1404 	insertion_latency = ((double) (mc_pool.rules_count
1405 				/ insertion_latency_time) / 1000);
1406 	deletion_latency = ((double) (mc_pool.rules_count
1407 				/ deletion_latency_time) / 1000);
1408 
1409 	insertion_throughput_time /= mc_pool.cores_count;
1410 	deletion_throughput_time /= mc_pool.cores_count;
1411 	insertion_throughput = ((double) (mc_pool.rules_count
1412 				/ insertion_throughput_time) / 1000);
1413 	deletion_throughput = ((double) (mc_pool.rules_count
1414 				/ deletion_throughput_time) / 1000);
1415 
1416 	/* Latency stats */
1417 	printf("\n%s\n:: [Latency | Insertion] All Cores :: Port %d :: ",
1418 		item, port);
1419 	printf("Total flows insertion rate -> %f K Rules/Sec\n",
1420 		insertion_latency);
1421 	printf(":: [Latency | Insertion] All Cores :: Port %d :: ", port);
1422 	printf("The time for creating %d rules is %f seconds\n",
1423 		mc_pool.rules_count, insertion_latency_time);
1424 
1425 	/* Throughput stats */
1426 	printf(":: [Throughput | Insertion] All Cores :: Port %d :: ", port);
1427 	printf("Total flows insertion rate -> %f K Rules/Sec\n",
1428 		insertion_throughput);
1429 	printf(":: [Throughput | Insertion] All Cores :: Port %d :: ", port);
1430 	printf("The average time for creating %d rules is %f seconds\n",
1431 		mc_pool.rules_count, insertion_throughput_time);
1432 
1433 	if (delete_flag) {
1434 	/* Latency stats */
1435 		printf(":: [Latency | Deletion] All Cores :: Port %d :: Total "
1436 			"deletion rate -> %f K Rules/Sec\n",
1437 			port, deletion_latency);
1438 		printf(":: [Latency | Deletion] All Cores :: Port %d :: ",
1439 			port);
1440 		printf("The time for deleting %d rules is %f seconds\n",
1441 			mc_pool.rules_count, deletion_latency_time);
1442 
1443 		/* Throughput stats */
1444 		printf(":: [Throughput | Deletion] All Cores :: Port %d :: Total "
1445 			"deletion rate -> %f K Rules/Sec\n",
1446 			port, deletion_throughput);
1447 		printf(":: [Throughput | Deletion] All Cores :: Port %d :: ",
1448 			port);
1449 		printf("The average time for deleting %d rules is %f seconds\n",
1450 			mc_pool.rules_count, deletion_throughput_time);
1451 	}
1452 }
1453 
1454 static void
1455 dump_used_mem(uint16_t port)
1456 {
1457 	uint32_t i;
1458 	int64_t last_alloc, current_alloc;
1459 	int flow_size_in_bytes;
1460 
1461 	last_alloc = mc_pool.last_alloc[0];
1462 	current_alloc = mc_pool.current_alloc[0];
1463 
1464 	i = mc_pool.cores_count;
1465 	while (i-- > 1) {
1466 		if (last_alloc > mc_pool.last_alloc[i])
1467 			last_alloc = mc_pool.last_alloc[i];
1468 		if (current_alloc < mc_pool.current_alloc[i])
1469 			current_alloc = mc_pool.current_alloc[i];
1470 	}
1471 
1472 	flow_size_in_bytes = (current_alloc - last_alloc) / mc_pool.rules_count;
1473 	printf("\n:: Port %d :: rte_flow size in DPDK layer: %d Bytes\n",
1474 		port, flow_size_in_bytes);
1475 }
1476 
1477 static int
1478 run_rte_flow_handler_cores(void *data __rte_unused)
1479 {
1480 	uint16_t port;
1481 	int lcore_counter = 0;
1482 	int lcore_id = rte_lcore_id();
1483 	int i;
1484 
1485 	RTE_LCORE_FOREACH(i) {
1486 		/*  If core not needed return. */
1487 		if (lcore_id == i) {
1488 			printf(":: lcore %d mapped with index %d\n", lcore_id, lcore_counter);
1489 			if (lcore_counter >= (int) mc_pool.cores_count)
1490 				return 0;
1491 			break;
1492 		}
1493 		lcore_counter++;
1494 	}
1495 	lcore_id = lcore_counter;
1496 
1497 	if (lcore_id >= (int) mc_pool.cores_count)
1498 		return 0;
1499 
1500 	mc_pool.rules_count = rules_count;
1501 
1502 	flows_handler(lcore_id);
1503 
1504 	/* Only main core to print total results. */
1505 	if (lcore_id != 0)
1506 		return 0;
1507 
1508 	/* Make sure all cores finished insertion/deletion process. */
1509 	rte_eal_mp_wait_lcore();
1510 
1511 	RTE_ETH_FOREACH_DEV(port) {
1512 		/* If port outside portmask */
1513 		if (!((ports_mask >> port) & 0x1))
1514 			continue;
1515 		if (has_meter())
1516 			dump_used_cpu_time("Meters:",
1517 				port, &mc_pool.meters_record);
1518 		dump_used_cpu_time("Flows:",
1519 			port, &mc_pool.flows_record);
1520 		dump_used_mem(port);
1521 	}
1522 
1523 	return 0;
1524 }
1525 
1526 static void
1527 signal_handler(int signum)
1528 {
1529 	if (signum == SIGINT || signum == SIGTERM) {
1530 		printf("\n\nSignal %d received, preparing to exit...\n",
1531 					signum);
1532 		printf("Error: Stats are wrong due to sudden signal!\n\n");
1533 		force_quit = true;
1534 	}
1535 }
1536 
1537 static inline uint16_t
1538 do_rx(struct lcore_info *li, uint16_t rx_port, uint16_t rx_queue)
1539 {
1540 	uint16_t cnt = 0;
1541 	cnt = rte_eth_rx_burst(rx_port, rx_queue, li->pkts, MAX_PKT_BURST);
1542 	li->rx_pkts += cnt;
1543 	return cnt;
1544 }
1545 
1546 static inline void
1547 do_tx(struct lcore_info *li, uint16_t cnt, uint16_t tx_port,
1548 			uint16_t tx_queue)
1549 {
1550 	uint16_t nr_tx = 0;
1551 	uint16_t i;
1552 
1553 	nr_tx = rte_eth_tx_burst(tx_port, tx_queue, li->pkts, cnt);
1554 	li->tx_pkts  += nr_tx;
1555 	li->tx_drops += cnt - nr_tx;
1556 
1557 	for (i = nr_tx; i < cnt; i++)
1558 		rte_pktmbuf_free(li->pkts[i]);
1559 }
1560 
1561 /*
1562  * Method to convert numbers into pretty numbers that easy
1563  * to read. The design here is to add comma after each three
1564  * digits and set all of this inside buffer.
1565  *
1566  * For example if n = 1799321, the output will be
1567  * 1,799,321 after this method which is easier to read.
1568  */
1569 static char *
1570 pretty_number(uint64_t n, char *buf)
1571 {
1572 	char p[6][4];
1573 	int i = 0;
1574 	int off = 0;
1575 
1576 	while (n > 1000) {
1577 		sprintf(p[i], "%03d", (int)(n % 1000));
1578 		n /= 1000;
1579 		i += 1;
1580 	}
1581 
1582 	sprintf(p[i++], "%d", (int)n);
1583 
1584 	while (i--)
1585 		off += sprintf(buf + off, "%s,", p[i]);
1586 	buf[strlen(buf) - 1] = '\0';
1587 
1588 	return buf;
1589 }
1590 
1591 static void
1592 packet_per_second_stats(void)
1593 {
1594 	struct lcore_info *old;
1595 	struct lcore_info *li, *oli;
1596 	int nr_lines = 0;
1597 	int i;
1598 
1599 	old = rte_zmalloc("old",
1600 		sizeof(struct lcore_info) * RTE_MAX_LCORE, 0);
1601 	if (old == NULL)
1602 		rte_exit(EXIT_FAILURE, "No Memory available!\n");
1603 
1604 	memcpy(old, lcore_infos,
1605 		sizeof(struct lcore_info) * RTE_MAX_LCORE);
1606 
1607 	while (!force_quit) {
1608 		uint64_t total_tx_pkts = 0;
1609 		uint64_t total_rx_pkts = 0;
1610 		uint64_t total_tx_drops = 0;
1611 		uint64_t tx_delta, rx_delta, drops_delta;
1612 		char buf[3][32];
1613 		int nr_valid_core = 0;
1614 
1615 		sleep(1);
1616 
1617 		if (nr_lines) {
1618 			char go_up_nr_lines[16];
1619 
1620 			sprintf(go_up_nr_lines, "%c[%dA\r", 27, nr_lines);
1621 			printf("%s\r", go_up_nr_lines);
1622 		}
1623 
1624 		printf("\n%6s %16s %16s %16s\n", "core", "tx", "tx drops", "rx");
1625 		printf("%6s %16s %16s %16s\n", "------", "----------------",
1626 			"----------------", "----------------");
1627 		nr_lines = 3;
1628 		for (i = 0; i < RTE_MAX_LCORE; i++) {
1629 			li  = &lcore_infos[i];
1630 			oli = &old[i];
1631 			if (li->mode != LCORE_MODE_PKT)
1632 				continue;
1633 
1634 			tx_delta    = li->tx_pkts  - oli->tx_pkts;
1635 			rx_delta    = li->rx_pkts  - oli->rx_pkts;
1636 			drops_delta = li->tx_drops - oli->tx_drops;
1637 			printf("%6d %16s %16s %16s\n", i,
1638 				pretty_number(tx_delta,    buf[0]),
1639 				pretty_number(drops_delta, buf[1]),
1640 				pretty_number(rx_delta,    buf[2]));
1641 
1642 			total_tx_pkts  += tx_delta;
1643 			total_rx_pkts  += rx_delta;
1644 			total_tx_drops += drops_delta;
1645 
1646 			nr_valid_core++;
1647 			nr_lines += 1;
1648 		}
1649 
1650 		if (nr_valid_core > 1) {
1651 			printf("%6s %16s %16s %16s\n", "total",
1652 				pretty_number(total_tx_pkts,  buf[0]),
1653 				pretty_number(total_tx_drops, buf[1]),
1654 				pretty_number(total_rx_pkts,  buf[2]));
1655 			nr_lines += 1;
1656 		}
1657 
1658 		memcpy(old, lcore_infos,
1659 			sizeof(struct lcore_info) * RTE_MAX_LCORE);
1660 	}
1661 }
1662 
1663 static int
1664 start_forwarding(void *data __rte_unused)
1665 {
1666 	int lcore = rte_lcore_id();
1667 	int stream_id;
1668 	uint16_t cnt;
1669 	struct lcore_info *li = &lcore_infos[lcore];
1670 
1671 	if (!li->mode)
1672 		return 0;
1673 
1674 	if (li->mode == LCORE_MODE_STATS) {
1675 		printf(":: started stats on lcore %u\n", lcore);
1676 		packet_per_second_stats();
1677 		return 0;
1678 	}
1679 
1680 	while (!force_quit)
1681 		for (stream_id = 0; stream_id < MAX_STREAMS; stream_id++) {
1682 			if (li->streams[stream_id].rx_port == -1)
1683 				continue;
1684 
1685 			cnt = do_rx(li,
1686 					li->streams[stream_id].rx_port,
1687 					li->streams[stream_id].rx_queue);
1688 			if (cnt)
1689 				do_tx(li, cnt,
1690 					li->streams[stream_id].tx_port,
1691 					li->streams[stream_id].tx_queue);
1692 		}
1693 	return 0;
1694 }
1695 
1696 static void
1697 init_lcore_info(void)
1698 {
1699 	int i, j;
1700 	unsigned int lcore;
1701 	uint16_t nr_port;
1702 	uint16_t queue;
1703 	int port;
1704 	int stream_id = 0;
1705 	int streams_per_core;
1706 	int unassigned_streams;
1707 	int nb_fwd_streams;
1708 	nr_port = rte_eth_dev_count_avail();
1709 
1710 	/* First logical core is reserved for stats printing */
1711 	lcore = rte_get_next_lcore(-1, 0, 0);
1712 	lcore_infos[lcore].mode = LCORE_MODE_STATS;
1713 
1714 	/*
1715 	 * Initialize all cores
1716 	 * All cores at first must have -1 value in all streams
1717 	 * This means that this stream is not used, or not set
1718 	 * yet.
1719 	 */
1720 	for (i = 0; i < RTE_MAX_LCORE; i++)
1721 		for (j = 0; j < MAX_STREAMS; j++) {
1722 			lcore_infos[i].streams[j].tx_port = -1;
1723 			lcore_infos[i].streams[j].rx_port = -1;
1724 			lcore_infos[i].streams[j].tx_queue = -1;
1725 			lcore_infos[i].streams[j].rx_queue = -1;
1726 			lcore_infos[i].streams_nb = 0;
1727 		}
1728 
1729 	/*
1730 	 * Calculate the total streams count.
1731 	 * Also distribute those streams count between the available
1732 	 * logical cores except first core, since it's reserved for
1733 	 * stats prints.
1734 	 */
1735 	nb_fwd_streams = nr_port * rx_queues_count;
1736 	if ((int)(nb_lcores - 1) >= nb_fwd_streams)
1737 		for (i = 0; i < (int)(nb_lcores - 1); i++) {
1738 			lcore = rte_get_next_lcore(lcore, 0, 0);
1739 			lcore_infos[lcore].streams_nb = 1;
1740 		}
1741 	else {
1742 		streams_per_core = nb_fwd_streams / (nb_lcores - 1);
1743 		unassigned_streams = nb_fwd_streams % (nb_lcores - 1);
1744 		for (i = 0; i < (int)(nb_lcores - 1); i++) {
1745 			lcore = rte_get_next_lcore(lcore, 0, 0);
1746 			lcore_infos[lcore].streams_nb = streams_per_core;
1747 			if (unassigned_streams) {
1748 				lcore_infos[lcore].streams_nb++;
1749 				unassigned_streams--;
1750 			}
1751 		}
1752 	}
1753 
1754 	/*
1755 	 * Set the streams for the cores according to each logical
1756 	 * core stream count.
1757 	 * The streams is built on the design of what received should
1758 	 * forward as well, this means that if you received packets on
1759 	 * port 0 queue 0 then the same queue should forward the
1760 	 * packets, using the same logical core.
1761 	 */
1762 	lcore = rte_get_next_lcore(-1, 0, 0);
1763 	for (port = 0; port < nr_port; port++) {
1764 		/* Create FWD stream */
1765 		for (queue = 0; queue < rx_queues_count; queue++) {
1766 			if (!lcore_infos[lcore].streams_nb ||
1767 				!(stream_id % lcore_infos[lcore].streams_nb)) {
1768 				lcore = rte_get_next_lcore(lcore, 0, 0);
1769 				lcore_infos[lcore].mode = LCORE_MODE_PKT;
1770 				stream_id = 0;
1771 			}
1772 			lcore_infos[lcore].streams[stream_id].rx_queue = queue;
1773 			lcore_infos[lcore].streams[stream_id].tx_queue = queue;
1774 			lcore_infos[lcore].streams[stream_id].rx_port = port;
1775 			lcore_infos[lcore].streams[stream_id].tx_port = port;
1776 			stream_id++;
1777 		}
1778 	}
1779 
1780 	/* Print all streams */
1781 	printf(":: Stream -> core id[N]: (rx_port, rx_queue)->(tx_port, tx_queue)\n");
1782 	for (i = 0; i < RTE_MAX_LCORE; i++)
1783 		for (j = 0; j < MAX_STREAMS; j++) {
1784 			/* No streams for this core */
1785 			if (lcore_infos[i].streams[j].tx_port == -1)
1786 				break;
1787 			printf("Stream -> core id[%d]: (%d,%d)->(%d,%d)\n",
1788 				i,
1789 				lcore_infos[i].streams[j].rx_port,
1790 				lcore_infos[i].streams[j].rx_queue,
1791 				lcore_infos[i].streams[j].tx_port,
1792 				lcore_infos[i].streams[j].tx_queue);
1793 		}
1794 }
1795 
1796 static void
1797 init_port(void)
1798 {
1799 	int ret;
1800 	uint16_t std_queue;
1801 	uint16_t hairpin_queue;
1802 	uint16_t port_id;
1803 	uint16_t nr_ports;
1804 	uint16_t nr_queues;
1805 	struct rte_eth_hairpin_conf hairpin_conf = {
1806 		.peer_count = 1,
1807 	};
1808 	struct rte_eth_conf port_conf = {
1809 		.rx_adv_conf = {
1810 			.rss_conf.rss_hf =
1811 				GET_RSS_HF(),
1812 		}
1813 	};
1814 	struct rte_eth_txconf txq_conf;
1815 	struct rte_eth_rxconf rxq_conf;
1816 	struct rte_eth_dev_info dev_info;
1817 
1818 	nr_queues = rx_queues_count;
1819 	if (hairpin_queues_num != 0)
1820 		nr_queues = rx_queues_count + hairpin_queues_num;
1821 
1822 	nr_ports = rte_eth_dev_count_avail();
1823 	if (nr_ports == 0)
1824 		rte_exit(EXIT_FAILURE, "Error: no port detected\n");
1825 
1826 	mbuf_mp = rte_pktmbuf_pool_create("mbuf_pool",
1827 					total_mbuf_num, mbuf_cache_size,
1828 					0, mbuf_size,
1829 					rte_socket_id());
1830 	if (mbuf_mp == NULL)
1831 		rte_exit(EXIT_FAILURE, "Error: can't init mbuf pool\n");
1832 
1833 	for (port_id = 0; port_id < nr_ports; port_id++) {
1834 		uint64_t rx_metadata = 0;
1835 
1836 		rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
1837 		rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
1838 
1839 		ret = rte_eth_rx_metadata_negotiate(port_id, &rx_metadata);
1840 		if (ret == 0) {
1841 			if (!(rx_metadata & RTE_ETH_RX_METADATA_USER_FLAG)) {
1842 				printf(":: flow action FLAG will not affect Rx mbufs on port=%u\n",
1843 				       port_id);
1844 			}
1845 
1846 			if (!(rx_metadata & RTE_ETH_RX_METADATA_USER_MARK)) {
1847 				printf(":: flow action MARK will not affect Rx mbufs on port=%u\n",
1848 				       port_id);
1849 			}
1850 		} else if (ret != -ENOTSUP) {
1851 			rte_exit(EXIT_FAILURE, "Error when negotiating Rx meta features on port=%u: %s\n",
1852 				 port_id, rte_strerror(-ret));
1853 		}
1854 
1855 		ret = rte_eth_dev_info_get(port_id, &dev_info);
1856 		if (ret != 0)
1857 			rte_exit(EXIT_FAILURE,
1858 				"Error during getting device"
1859 				" (port %u) info: %s\n",
1860 				port_id, strerror(-ret));
1861 
1862 		port_conf.txmode.offloads &= dev_info.tx_offload_capa;
1863 		port_conf.rxmode.offloads &= dev_info.rx_offload_capa;
1864 
1865 		printf(":: initializing port: %d\n", port_id);
1866 
1867 		ret = rte_eth_dev_configure(port_id, nr_queues,
1868 				nr_queues, &port_conf);
1869 		if (ret < 0)
1870 			rte_exit(EXIT_FAILURE,
1871 				":: cannot configure device: err=%d, port=%u\n",
1872 				ret, port_id);
1873 
1874 		rxq_conf = dev_info.default_rxconf;
1875 		for (std_queue = 0; std_queue < rx_queues_count; std_queue++) {
1876 			ret = rte_eth_rx_queue_setup(port_id, std_queue, rxd_count,
1877 					rte_eth_dev_socket_id(port_id),
1878 					&rxq_conf,
1879 					mbuf_mp);
1880 			if (ret < 0)
1881 				rte_exit(EXIT_FAILURE,
1882 					":: Rx queue setup failed: err=%d, port=%u\n",
1883 					ret, port_id);
1884 		}
1885 
1886 		txq_conf = dev_info.default_txconf;
1887 		for (std_queue = 0; std_queue < tx_queues_count; std_queue++) {
1888 			ret = rte_eth_tx_queue_setup(port_id, std_queue, txd_count,
1889 					rte_eth_dev_socket_id(port_id),
1890 					&txq_conf);
1891 			if (ret < 0)
1892 				rte_exit(EXIT_FAILURE,
1893 					":: Tx queue setup failed: err=%d, port=%u\n",
1894 					ret, port_id);
1895 		}
1896 
1897 		/* Catch all packets from traffic generator. */
1898 		ret = rte_eth_promiscuous_enable(port_id);
1899 		if (ret != 0)
1900 			rte_exit(EXIT_FAILURE,
1901 				":: promiscuous mode enable failed: err=%s, port=%u\n",
1902 				rte_strerror(-ret), port_id);
1903 
1904 		if (hairpin_queues_num != 0) {
1905 			/*
1906 			 * Configure peer which represents hairpin Tx.
1907 			 * Hairpin queue numbers start after standard queues
1908 			 * (rx_queues_count and tx_queues_count).
1909 			 */
1910 			for (hairpin_queue = rx_queues_count, std_queue = 0;
1911 					hairpin_queue < nr_queues;
1912 					hairpin_queue++, std_queue++) {
1913 				hairpin_conf.peers[0].port = port_id;
1914 				hairpin_conf.peers[0].queue =
1915 					std_queue + tx_queues_count;
1916 				ret = rte_eth_rx_hairpin_queue_setup(
1917 						port_id, hairpin_queue,
1918 						rxd_count, &hairpin_conf);
1919 				if (ret != 0)
1920 					rte_exit(EXIT_FAILURE,
1921 						":: Hairpin rx queue setup failed: err=%d, port=%u\n",
1922 						ret, port_id);
1923 			}
1924 
1925 			for (hairpin_queue = tx_queues_count, std_queue = 0;
1926 					hairpin_queue < nr_queues;
1927 					hairpin_queue++, std_queue++) {
1928 				hairpin_conf.peers[0].port = port_id;
1929 				hairpin_conf.peers[0].queue =
1930 					std_queue + rx_queues_count;
1931 				ret = rte_eth_tx_hairpin_queue_setup(
1932 						port_id, hairpin_queue,
1933 						txd_count, &hairpin_conf);
1934 				if (ret != 0)
1935 					rte_exit(EXIT_FAILURE,
1936 						":: Hairpin tx queue setup failed: err=%d, port=%u\n",
1937 						ret, port_id);
1938 			}
1939 		}
1940 
1941 		ret = rte_eth_dev_start(port_id);
1942 		if (ret < 0)
1943 			rte_exit(EXIT_FAILURE,
1944 				"rte_eth_dev_start:err=%d, port=%u\n",
1945 				ret, port_id);
1946 
1947 		printf(":: initializing port: %d done\n", port_id);
1948 	}
1949 }
1950 
1951 int
1952 main(int argc, char **argv)
1953 {
1954 	int ret;
1955 	uint16_t port;
1956 	struct rte_flow_error error;
1957 
1958 	ret = rte_eal_init(argc, argv);
1959 	if (ret < 0)
1960 		rte_exit(EXIT_FAILURE, "EAL init failed\n");
1961 
1962 	force_quit = false;
1963 	dump_iterations = false;
1964 	rules_count = DEFAULT_RULES_COUNT;
1965 	rules_batch = DEFAULT_RULES_BATCH;
1966 	delete_flag = false;
1967 	dump_socket_mem_flag = false;
1968 	flow_group = DEFAULT_GROUP;
1969 	unique_data = false;
1970 
1971 	rx_queues_count = (uint8_t) RXQ_NUM;
1972 	tx_queues_count = (uint8_t) TXQ_NUM;
1973 	rxd_count = (uint8_t) NR_RXD;
1974 	txd_count = (uint8_t) NR_TXD;
1975 	mbuf_size = (uint32_t) MBUF_SIZE;
1976 	mbuf_cache_size = (uint32_t) MBUF_CACHE_SIZE;
1977 	total_mbuf_num = (uint32_t) TOTAL_MBUF_NUM;
1978 
1979 	signal(SIGINT, signal_handler);
1980 	signal(SIGTERM, signal_handler);
1981 
1982 	argc -= ret;
1983 	argv += ret;
1984 	if (argc > 1)
1985 		args_parse(argc, argv);
1986 
1987 	init_port();
1988 
1989 	nb_lcores = rte_lcore_count();
1990 	if (nb_lcores <= 1)
1991 		rte_exit(EXIT_FAILURE, "This app needs at least two cores\n");
1992 
1993 	printf(":: Flows Count per port: %d\n\n", rules_count);
1994 
1995 	if (has_meter())
1996 		create_meter_profile();
1997 	rte_eal_mp_remote_launch(run_rte_flow_handler_cores, NULL, CALL_MAIN);
1998 
1999 	if (enable_fwd) {
2000 		init_lcore_info();
2001 		rte_eal_mp_remote_launch(start_forwarding, NULL, CALL_MAIN);
2002 	}
2003 	if (has_meter() && delete_flag)
2004 		destroy_meter_profile();
2005 
2006 	RTE_ETH_FOREACH_DEV(port) {
2007 		rte_flow_flush(port, &error);
2008 		if (rte_eth_dev_stop(port) != 0)
2009 			printf("Failed to stop device on port %u\n", port);
2010 		rte_eth_dev_close(port);
2011 	}
2012 	printf("\nBye ...\n");
2013 	return 0;
2014 }
2015