xref: /dpdk/app/test-flow-perf/main.c (revision 452c1916)
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 volatile bool force_quit;
60 static bool dump_iterations;
61 static bool delete_flag;
62 static bool dump_socket_mem_flag;
63 static bool enable_fwd;
64 static bool unique_data;
65 
66 static uint8_t rx_queues_count;
67 static uint8_t tx_queues_count;
68 static uint8_t rxd_count;
69 static uint8_t txd_count;
70 static uint32_t mbuf_size;
71 static uint32_t mbuf_cache_size;
72 static uint32_t total_mbuf_num;
73 
74 static struct rte_mempool *mbuf_mp;
75 static uint32_t nb_lcores;
76 static uint32_t rules_count;
77 static uint32_t rules_batch;
78 static uint32_t hairpin_queues_num; /* total hairpin q number - default: 0 */
79 static uint32_t nb_lcores;
80 
81 #define MAX_PKT_BURST    32
82 #define LCORE_MODE_PKT    1
83 #define LCORE_MODE_STATS  2
84 #define MAX_STREAMS      64
85 #define METER_CREATE	  1
86 #define METER_DELETE	  2
87 
88 struct stream {
89 	int tx_port;
90 	int tx_queue;
91 	int rx_port;
92 	int rx_queue;
93 };
94 
95 struct lcore_info {
96 	int mode;
97 	int streams_nb;
98 	struct stream streams[MAX_STREAMS];
99 	/* stats */
100 	uint64_t tx_pkts;
101 	uint64_t tx_drops;
102 	uint64_t rx_pkts;
103 	struct rte_mbuf *pkts[MAX_PKT_BURST];
104 } __rte_cache_aligned;
105 
106 static struct lcore_info lcore_infos[RTE_MAX_LCORE];
107 
108 struct used_cpu_time {
109 	double insertion[MAX_PORTS][RTE_MAX_LCORE];
110 	double deletion[MAX_PORTS][RTE_MAX_LCORE];
111 };
112 
113 struct multi_cores_pool {
114 	uint32_t cores_count;
115 	uint32_t rules_count;
116 	struct used_cpu_time meters_record;
117 	struct used_cpu_time flows_record;
118 	int64_t last_alloc[RTE_MAX_LCORE];
119 	int64_t current_alloc[RTE_MAX_LCORE];
120 } __rte_cache_aligned;
121 
122 static struct multi_cores_pool mc_pool = {
123 	.cores_count = 1,
124 };
125 
126 static void
127 usage(char *progname)
128 {
129 	printf("\nusage: %s\n", progname);
130 	printf("\nControl configurations:\n");
131 	printf("  --rules-count=N: to set the number of needed"
132 		" rules to insert, default is %d\n", DEFAULT_RULES_COUNT);
133 	printf("  --rules-batch=N: set number of batched rules,"
134 		" default is %d\n", DEFAULT_RULES_BATCH);
135 	printf("  --dump-iterations: To print rates for each"
136 		" iteration\n");
137 	printf("  --deletion-rate: Enable deletion rate"
138 		" calculations\n");
139 	printf("  --dump-socket-mem: To dump all socket memory\n");
140 	printf("  --enable-fwd: To enable packets forwarding"
141 		" after insertion\n");
142 	printf("  --portmask=N: hexadecimal bitmask of ports used\n");
143 	printf("  --unique-data: flag to set using unique data for all"
144 		" actions that support data, such as header modify and encap actions\n");
145 
146 	printf("To set flow attributes:\n");
147 	printf("  --ingress: set ingress attribute in flows\n");
148 	printf("  --egress: set egress attribute in flows\n");
149 	printf("  --transfer: set transfer attribute in flows\n");
150 	printf("  --group=N: set group for all flows,"
151 		" default is %d\n", DEFAULT_GROUP);
152 	printf("  --cores=N: to set the number of needed "
153 		"cores to insert rte_flow rules, default is 1\n");
154 	printf("  --rxq=N: to set the count of receive queues\n");
155 	printf("  --txq=N: to set the count of send queues\n");
156 	printf("  --rxd=N: to set the count of rxd\n");
157 	printf("  --txd=N: to set the count of txd\n");
158 	printf("  --mbuf-size=N: to set the size of mbuf\n");
159 	printf("  --mbuf-cache-size=N: to set the size of mbuf cache\n");
160 	printf("  --total-mbuf-count=N: to set the count of total mbuf count\n");
161 
162 
163 	printf("To set flow items:\n");
164 	printf("  --ether: add ether layer in flow items\n");
165 	printf("  --vlan: add vlan layer in flow items\n");
166 	printf("  --ipv4: add ipv4 layer in flow items\n");
167 	printf("  --ipv6: add ipv6 layer in flow items\n");
168 	printf("  --tcp: add tcp layer in flow items\n");
169 	printf("  --udp: add udp layer in flow items\n");
170 	printf("  --vxlan: add vxlan layer in flow items\n");
171 	printf("  --vxlan-gpe: add vxlan-gpe layer in flow items\n");
172 	printf("  --gre: add gre layer in flow items\n");
173 	printf("  --geneve: add geneve layer in flow items\n");
174 	printf("  --gtp: add gtp layer in flow items\n");
175 	printf("  --meta: add meta layer in flow items\n");
176 	printf("  --tag: add tag layer in flow items\n");
177 	printf("  --icmpv4: add icmpv4 layer in flow items\n");
178 	printf("  --icmpv6: add icmpv6 layer in flow items\n");
179 
180 	printf("To set flow actions:\n");
181 	printf("  --port-id: add port-id action in flow actions\n");
182 	printf("  --rss: add rss action in flow actions\n");
183 	printf("  --queue: add queue action in flow actions\n");
184 	printf("  --jump: add jump action in flow actions\n");
185 	printf("  --mark: add mark action in flow actions\n");
186 	printf("  --count: add count action in flow actions\n");
187 	printf("  --set-meta: add set meta action in flow actions\n");
188 	printf("  --set-tag: add set tag action in flow actions\n");
189 	printf("  --drop: add drop action in flow actions\n");
190 	printf("  --hairpin-queue=N: add hairpin-queue action in flow actions\n");
191 	printf("  --hairpin-rss=N: add hairpin-rss action in flow actions\n");
192 	printf("  --set-src-mac: add set src mac action to flow actions\n"
193 		"Src mac to be set is random each flow\n");
194 	printf("  --set-dst-mac: add set dst mac action to flow actions\n"
195 		 "Dst mac to be set is random each flow\n");
196 	printf("  --set-src-ipv4: add set src ipv4 action to flow actions\n"
197 		"Src ipv4 to be set is random each flow\n");
198 	printf("  --set-dst-ipv4 add set dst ipv4 action to flow actions\n"
199 		"Dst ipv4 to be set is random each flow\n");
200 	printf("  --set-src-ipv6: add set src ipv6 action to flow actions\n"
201 		"Src ipv6 to be set is random each flow\n");
202 	printf("  --set-dst-ipv6: add set dst ipv6 action to flow actions\n"
203 		"Dst ipv6 to be set is random each flow\n");
204 	printf("  --set-src-tp: add set src tp action to flow actions\n"
205 		"Src tp to be set is random each flow\n");
206 	printf("  --set-dst-tp: add set dst tp action to flow actions\n"
207 		"Dst tp to be set is random each flow\n");
208 	printf("  --inc-tcp-ack: add inc tcp ack action to flow actions\n"
209 		"tcp ack will be increments by 1\n");
210 	printf("  --dec-tcp-ack: add dec tcp ack action to flow actions\n"
211 		"tcp ack will be decrements by 1\n");
212 	printf("  --inc-tcp-seq: add inc tcp seq action to flow actions\n"
213 		"tcp seq will be increments by 1\n");
214 	printf("  --dec-tcp-seq: add dec tcp seq action to flow actions\n"
215 		"tcp seq will be decrements by 1\n");
216 	printf("  --set-ttl: add set ttl action to flow actions\n"
217 		"L3 ttl to be set is random each flow\n");
218 	printf("  --dec-ttl: add dec ttl action to flow actions\n"
219 		"L3 ttl will be decrements by 1\n");
220 	printf("  --set-ipv4-dscp: add set ipv4 dscp action to flow actions\n"
221 		"ipv4 dscp value to be set is random each flow\n");
222 	printf("  --set-ipv6-dscp: add set ipv6 dscp action to flow actions\n"
223 		"ipv6 dscp value to be set is random each flow\n");
224 	printf("  --flag: add flag action to flow actions\n");
225 	printf("  --meter: add meter action to flow actions\n");
226 	printf("  --raw-encap=<data>: add raw encap action to flow actions\n"
227 		"Data is the data needed to be encaped\n"
228 		"Example: raw-encap=ether,ipv4,udp,vxlan\n");
229 	printf("  --raw-decap=<data>: add raw decap action to flow actions\n"
230 		"Data is the data needed to be decaped\n"
231 		"Example: raw-decap=ether,ipv4,udp,vxlan\n");
232 	printf("  --vxlan-encap: add vxlan-encap action to flow actions\n"
233 		"Encapped data is fixed with pattern: ether,ipv4,udp,vxlan\n"
234 		"With fixed values\n");
235 	printf("  --vxlan-decap: add vxlan_decap action to flow actions\n");
236 }
237 
238 static void
239 args_parse(int argc, char **argv)
240 {
241 	uint64_t pm;
242 	char **argvopt;
243 	char *token;
244 	char *end;
245 	int n, opt;
246 	int opt_idx;
247 	size_t i;
248 
249 	static const struct option_dict {
250 		const char *str;
251 		const uint64_t mask;
252 		uint64_t *map;
253 		uint8_t *map_idx;
254 
255 	} flow_options[] = {
256 		{
257 			.str = "ether",
258 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ETH),
259 			.map = &flow_items[0],
260 			.map_idx = &items_idx
261 		},
262 		{
263 			.str = "ipv4",
264 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_IPV4),
265 			.map = &flow_items[0],
266 			.map_idx = &items_idx
267 		},
268 		{
269 			.str = "ipv6",
270 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_IPV6),
271 			.map = &flow_items[0],
272 			.map_idx = &items_idx
273 		},
274 		{
275 			.str = "vlan",
276 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VLAN),
277 			.map = &flow_items[0],
278 			.map_idx = &items_idx
279 		},
280 		{
281 			.str = "tcp",
282 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_TCP),
283 			.map = &flow_items[0],
284 			.map_idx = &items_idx
285 		},
286 		{
287 			.str = "udp",
288 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_UDP),
289 			.map = &flow_items[0],
290 			.map_idx = &items_idx
291 		},
292 		{
293 			.str = "vxlan",
294 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VXLAN),
295 			.map = &flow_items[0],
296 			.map_idx = &items_idx
297 		},
298 		{
299 			.str = "vxlan-gpe",
300 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_VXLAN_GPE),
301 			.map = &flow_items[0],
302 			.map_idx = &items_idx
303 		},
304 		{
305 			.str = "gre",
306 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GRE),
307 			.map = &flow_items[0],
308 			.map_idx = &items_idx
309 		},
310 		{
311 			.str = "geneve",
312 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GENEVE),
313 			.map = &flow_items[0],
314 			.map_idx = &items_idx
315 		},
316 		{
317 			.str = "gtp",
318 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_GTP),
319 			.map = &flow_items[0],
320 			.map_idx = &items_idx
321 		},
322 		{
323 			.str = "meta",
324 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_META),
325 			.map = &flow_items[0],
326 			.map_idx = &items_idx
327 		},
328 		{
329 			.str = "tag",
330 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_TAG),
331 			.map = &flow_items[0],
332 			.map_idx = &items_idx
333 		},
334 		{
335 			.str = "icmpv4",
336 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ICMP),
337 			.map = &flow_items[0],
338 			.map_idx = &items_idx
339 		},
340 		{
341 			.str = "icmpv6",
342 			.mask = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ICMP6),
343 			.map = &flow_items[0],
344 			.map_idx = &items_idx
345 		},
346 		{
347 			.str = "ingress",
348 			.mask = INGRESS,
349 			.map = &flow_attrs[0],
350 			.map_idx = &attrs_idx
351 		},
352 		{
353 			.str = "egress",
354 			.mask = EGRESS,
355 			.map = &flow_attrs[0],
356 			.map_idx = &attrs_idx
357 		},
358 		{
359 			.str = "transfer",
360 			.mask = TRANSFER,
361 			.map = &flow_attrs[0],
362 			.map_idx = &attrs_idx
363 		},
364 		{
365 			.str = "port-id",
366 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_PORT_ID),
367 			.map = &flow_actions[0],
368 			.map_idx = &actions_idx
369 		},
370 		{
371 			.str = "rss",
372 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_RSS),
373 			.map = &flow_actions[0],
374 			.map_idx = &actions_idx
375 		},
376 		{
377 			.str = "queue",
378 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_QUEUE),
379 			.map = &flow_actions[0],
380 			.map_idx = &actions_idx
381 		},
382 		{
383 			.str = "jump",
384 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_JUMP),
385 			.map = &flow_actions[0],
386 			.map_idx = &actions_idx
387 		},
388 		{
389 			.str = "mark",
390 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_MARK),
391 			.map = &flow_actions[0],
392 			.map_idx = &actions_idx
393 		},
394 		{
395 			.str = "count",
396 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_COUNT),
397 			.map = &flow_actions[0],
398 			.map_idx = &actions_idx
399 		},
400 		{
401 			.str = "set-meta",
402 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_SET_META),
403 			.map = &flow_actions[0],
404 			.map_idx = &actions_idx
405 		},
406 		{
407 			.str = "set-tag",
408 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_SET_TAG),
409 			.map = &flow_actions[0],
410 			.map_idx = &actions_idx
411 		},
412 		{
413 			.str = "drop",
414 			.mask = FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_DROP),
415 			.map = &flow_actions[0],
416 			.map_idx = &actions_idx
417 		},
418 		{
419 			.str = "set-src-mac",
420 			.mask = FLOW_ACTION_MASK(
421 				RTE_FLOW_ACTION_TYPE_SET_MAC_SRC
422 			),
423 			.map = &flow_actions[0],
424 			.map_idx = &actions_idx
425 		},
426 		{
427 			.str = "set-dst-mac",
428 			.mask = FLOW_ACTION_MASK(
429 				RTE_FLOW_ACTION_TYPE_SET_MAC_DST
430 			),
431 			.map = &flow_actions[0],
432 			.map_idx = &actions_idx
433 		},
434 		{
435 			.str = "set-src-ipv4",
436 			.mask = FLOW_ACTION_MASK(
437 				RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC
438 			),
439 			.map = &flow_actions[0],
440 			.map_idx = &actions_idx
441 		},
442 		{
443 			.str = "set-dst-ipv4",
444 			.mask = FLOW_ACTION_MASK(
445 				RTE_FLOW_ACTION_TYPE_SET_IPV4_DST
446 			),
447 			.map = &flow_actions[0],
448 			.map_idx = &actions_idx
449 		},
450 		{
451 			.str = "set-src-ipv6",
452 			.mask = FLOW_ACTION_MASK(
453 				RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC
454 			),
455 			.map = &flow_actions[0],
456 			.map_idx = &actions_idx
457 		},
458 		{
459 			.str = "set-dst-ipv6",
460 			.mask = FLOW_ACTION_MASK(
461 				RTE_FLOW_ACTION_TYPE_SET_IPV6_DST
462 			),
463 			.map = &flow_actions[0],
464 			.map_idx = &actions_idx
465 		},
466 		{
467 			.str = "set-src-tp",
468 			.mask = FLOW_ACTION_MASK(
469 				RTE_FLOW_ACTION_TYPE_SET_TP_SRC
470 			),
471 			.map = &flow_actions[0],
472 			.map_idx = &actions_idx
473 		},
474 		{
475 			.str = "set-dst-tp",
476 			.mask = FLOW_ACTION_MASK(
477 				RTE_FLOW_ACTION_TYPE_SET_TP_DST
478 			),
479 			.map = &flow_actions[0],
480 			.map_idx = &actions_idx
481 		},
482 		{
483 			.str = "inc-tcp-ack",
484 			.mask = FLOW_ACTION_MASK(
485 				RTE_FLOW_ACTION_TYPE_INC_TCP_ACK
486 			),
487 			.map = &flow_actions[0],
488 			.map_idx = &actions_idx
489 		},
490 		{
491 			.str = "dec-tcp-ack",
492 			.mask = FLOW_ACTION_MASK(
493 				RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK
494 			),
495 			.map = &flow_actions[0],
496 			.map_idx = &actions_idx
497 		},
498 		{
499 			.str = "inc-tcp-seq",
500 			.mask = FLOW_ACTION_MASK(
501 				RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ
502 			),
503 			.map = &flow_actions[0],
504 			.map_idx = &actions_idx
505 		},
506 		{
507 			.str = "dec-tcp-seq",
508 			.mask = FLOW_ACTION_MASK(
509 				RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ
510 			),
511 			.map = &flow_actions[0],
512 			.map_idx = &actions_idx
513 		},
514 		{
515 			.str = "set-ttl",
516 			.mask = FLOW_ACTION_MASK(
517 				RTE_FLOW_ACTION_TYPE_SET_TTL
518 			),
519 			.map = &flow_actions[0],
520 			.map_idx = &actions_idx
521 		},
522 		{
523 			.str = "dec-ttl",
524 			.mask = FLOW_ACTION_MASK(
525 				RTE_FLOW_ACTION_TYPE_DEC_TTL
526 			),
527 			.map = &flow_actions[0],
528 			.map_idx = &actions_idx
529 		},
530 		{
531 			.str = "set-ipv4-dscp",
532 			.mask = FLOW_ACTION_MASK(
533 				RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP
534 			),
535 			.map = &flow_actions[0],
536 			.map_idx = &actions_idx
537 		},
538 		{
539 			.str = "set-ipv6-dscp",
540 			.mask = FLOW_ACTION_MASK(
541 				RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP
542 			),
543 			.map = &flow_actions[0],
544 			.map_idx = &actions_idx
545 		},
546 		{
547 			.str = "flag",
548 			.mask = FLOW_ACTION_MASK(
549 				RTE_FLOW_ACTION_TYPE_FLAG
550 			),
551 			.map = &flow_actions[0],
552 			.map_idx = &actions_idx
553 		},
554 		{
555 			.str = "meter",
556 			.mask = FLOW_ACTION_MASK(
557 				RTE_FLOW_ACTION_TYPE_METER
558 			),
559 			.map = &flow_actions[0],
560 			.map_idx = &actions_idx
561 		},
562 		{
563 			.str = "vxlan-encap",
564 			.mask = FLOW_ACTION_MASK(
565 				RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP
566 			),
567 			.map = &flow_actions[0],
568 			.map_idx = &actions_idx
569 		},
570 		{
571 			.str = "vxlan-decap",
572 			.mask = FLOW_ACTION_MASK(
573 				RTE_FLOW_ACTION_TYPE_VXLAN_DECAP
574 			),
575 			.map = &flow_actions[0],
576 			.map_idx = &actions_idx
577 		},
578 	};
579 
580 	static const struct option lgopts[] = {
581 		/* Control */
582 		{ "help",                       0, 0, 0 },
583 		{ "rules-count",                1, 0, 0 },
584 		{ "rules-batch",                1, 0, 0 },
585 		{ "dump-iterations",            0, 0, 0 },
586 		{ "deletion-rate",              0, 0, 0 },
587 		{ "dump-socket-mem",            0, 0, 0 },
588 		{ "enable-fwd",                 0, 0, 0 },
589 		{ "unique-data",                0, 0, 0 },
590 		{ "portmask",                   1, 0, 0 },
591 		{ "cores",                      1, 0, 0 },
592 		{ "meter-profile-alg",          1, 0, 0 },
593 		{ "rxq",                        1, 0, 0 },
594 		{ "txq",                        1, 0, 0 },
595 		{ "rxd",                        1, 0, 0 },
596 		{ "txd",                        1, 0, 0 },
597 		{ "mbuf-size",                  1, 0, 0 },
598 		{ "mbuf-cache-size",            1, 0, 0 },
599 		{ "total-mbuf-count",           1, 0, 0 },
600 		/* Attributes */
601 		{ "ingress",                    0, 0, 0 },
602 		{ "egress",                     0, 0, 0 },
603 		{ "transfer",                   0, 0, 0 },
604 		{ "group",                      1, 0, 0 },
605 		/* Items */
606 		{ "ether",                      0, 0, 0 },
607 		{ "vlan",                       0, 0, 0 },
608 		{ "ipv4",                       0, 0, 0 },
609 		{ "ipv6",                       0, 0, 0 },
610 		{ "tcp",                        0, 0, 0 },
611 		{ "udp",                        0, 0, 0 },
612 		{ "vxlan",                      0, 0, 0 },
613 		{ "vxlan-gpe",                  0, 0, 0 },
614 		{ "gre",                        0, 0, 0 },
615 		{ "geneve",                     0, 0, 0 },
616 		{ "gtp",                        0, 0, 0 },
617 		{ "meta",                       0, 0, 0 },
618 		{ "tag",                        0, 0, 0 },
619 		{ "icmpv4",                     0, 0, 0 },
620 		{ "icmpv6",                     0, 0, 0 },
621 		/* Actions */
622 		{ "port-id",                    0, 0, 0 },
623 		{ "rss",                        0, 0, 0 },
624 		{ "queue",                      0, 0, 0 },
625 		{ "jump",                       0, 0, 0 },
626 		{ "mark",                       0, 0, 0 },
627 		{ "count",                      0, 0, 0 },
628 		{ "set-meta",                   0, 0, 0 },
629 		{ "set-tag",                    0, 0, 0 },
630 		{ "drop",                       0, 0, 0 },
631 		{ "hairpin-queue",              1, 0, 0 },
632 		{ "hairpin-rss",                1, 0, 0 },
633 		{ "set-src-mac",                0, 0, 0 },
634 		{ "set-dst-mac",                0, 0, 0 },
635 		{ "set-src-ipv4",               0, 0, 0 },
636 		{ "set-dst-ipv4",               0, 0, 0 },
637 		{ "set-src-ipv6",               0, 0, 0 },
638 		{ "set-dst-ipv6",               0, 0, 0 },
639 		{ "set-src-tp",                 0, 0, 0 },
640 		{ "set-dst-tp",                 0, 0, 0 },
641 		{ "inc-tcp-ack",                0, 0, 0 },
642 		{ "dec-tcp-ack",                0, 0, 0 },
643 		{ "inc-tcp-seq",                0, 0, 0 },
644 		{ "dec-tcp-seq",                0, 0, 0 },
645 		{ "set-ttl",                    0, 0, 0 },
646 		{ "dec-ttl",                    0, 0, 0 },
647 		{ "set-ipv4-dscp",              0, 0, 0 },
648 		{ "set-ipv6-dscp",              0, 0, 0 },
649 		{ "flag",                       0, 0, 0 },
650 		{ "meter",                      0, 0, 0 },
651 		{ "raw-encap",                  1, 0, 0 },
652 		{ "raw-decap",                  1, 0, 0 },
653 		{ "vxlan-encap",                0, 0, 0 },
654 		{ "vxlan-decap",                0, 0, 0 },
655 	};
656 
657 	RTE_ETH_FOREACH_DEV(i)
658 		ports_mask |= 1 << i;
659 
660 	hairpin_queues_num = 0;
661 	argvopt = argv;
662 
663 	printf(":: Flow -> ");
664 	while ((opt = getopt_long(argc, argvopt, "",
665 				lgopts, &opt_idx)) != EOF) {
666 		switch (opt) {
667 		case 0:
668 			if (strcmp(lgopts[opt_idx].name, "help") == 0) {
669 				usage(argv[0]);
670 				exit(EXIT_SUCCESS);
671 			}
672 
673 			if (strcmp(lgopts[opt_idx].name, "group") == 0) {
674 				n = atoi(optarg);
675 				if (n >= 0)
676 					flow_group = n;
677 				else
678 					rte_exit(EXIT_FAILURE,
679 						"flow group should be >= 0\n");
680 				printf("group %d / ", flow_group);
681 			}
682 
683 			for (i = 0; i < RTE_DIM(flow_options); i++)
684 				if (strcmp(lgopts[opt_idx].name,
685 						flow_options[i].str) == 0) {
686 					flow_options[i].map[
687 					(*flow_options[i].map_idx)++] =
688 						flow_options[i].mask;
689 					printf("%s / ", flow_options[i].str);
690 				}
691 
692 			if (strcmp(lgopts[opt_idx].name,
693 					"hairpin-rss") == 0) {
694 				n = atoi(optarg);
695 				if (n > 0)
696 					hairpin_queues_num = n;
697 				else
698 					rte_exit(EXIT_FAILURE,
699 						"Hairpin queues should be > 0\n");
700 
701 				flow_actions[actions_idx++] =
702 					HAIRPIN_RSS_ACTION;
703 				printf("hairpin-rss / ");
704 			}
705 			if (strcmp(lgopts[opt_idx].name,
706 					"hairpin-queue") == 0) {
707 				n = atoi(optarg);
708 				if (n > 0)
709 					hairpin_queues_num = n;
710 				else
711 					rte_exit(EXIT_FAILURE,
712 						"Hairpin queues should be > 0\n");
713 
714 				flow_actions[actions_idx++] =
715 					HAIRPIN_QUEUE_ACTION;
716 				printf("hairpin-queue / ");
717 			}
718 
719 			if (strcmp(lgopts[opt_idx].name, "raw-encap") == 0) {
720 				printf("raw-encap ");
721 				flow_actions[actions_idx++] =
722 					FLOW_ITEM_MASK(
723 						RTE_FLOW_ACTION_TYPE_RAW_ENCAP
724 					);
725 
726 				token = strtok(optarg, ",");
727 				while (token != NULL) {
728 					for (i = 0; i < RTE_DIM(flow_options); i++) {
729 						if (strcmp(flow_options[i].str, token) == 0) {
730 							printf("%s,", token);
731 							encap_data |= flow_options[i].mask;
732 							break;
733 						}
734 						/* Reached last item with no match */
735 						if (i == (RTE_DIM(flow_options) - 1))
736 							rte_exit(EXIT_FAILURE,
737 								"Invalid encap item: %s\n", token);
738 					}
739 					token = strtok(NULL, ",");
740 				}
741 				printf(" / ");
742 			}
743 			if (strcmp(lgopts[opt_idx].name, "raw-decap") == 0) {
744 				printf("raw-decap ");
745 				flow_actions[actions_idx++] =
746 					FLOW_ITEM_MASK(
747 						RTE_FLOW_ACTION_TYPE_RAW_DECAP
748 					);
749 
750 				token = strtok(optarg, ",");
751 				while (token != NULL) {
752 					for (i = 0; i < RTE_DIM(flow_options); i++) {
753 						if (strcmp(flow_options[i].str, token) == 0) {
754 							printf("%s,", token);
755 							decap_data |= flow_options[i].mask;
756 							break;
757 						}
758 						/* Reached last item with no match */
759 						if (i == (RTE_DIM(flow_options) - 1))
760 							rte_exit(EXIT_FAILURE,
761 								"Invalid decap item %s\n", token);
762 					}
763 					token = strtok(NULL, ",");
764 				}
765 				printf(" / ");
766 			}
767 			/* Control */
768 			if (strcmp(lgopts[opt_idx].name,
769 					"rules-batch") == 0) {
770 				n = atoi(optarg);
771 				if (n >= DEFAULT_RULES_BATCH)
772 					rules_batch = n;
773 				else {
774 					rte_exit(EXIT_FAILURE,
775 						"rules_batch should be >= %d\n",
776 						DEFAULT_RULES_BATCH);
777 				}
778 			}
779 			if (strcmp(lgopts[opt_idx].name,
780 					"rules-count") == 0) {
781 				n = atoi(optarg);
782 				if (n >= (int) rules_batch)
783 					rules_count = n;
784 				else {
785 					rte_exit(EXIT_FAILURE,
786 						"rules_count should be >= %d\n",
787 						rules_batch);
788 				}
789 			}
790 			if (strcmp(lgopts[opt_idx].name,
791 					"dump-iterations") == 0)
792 				dump_iterations = true;
793 			if (strcmp(lgopts[opt_idx].name,
794 					"unique-data") == 0)
795 				unique_data = true;
796 			if (strcmp(lgopts[opt_idx].name,
797 					"deletion-rate") == 0)
798 				delete_flag = true;
799 			if (strcmp(lgopts[opt_idx].name,
800 					"dump-socket-mem") == 0)
801 				dump_socket_mem_flag = true;
802 			if (strcmp(lgopts[opt_idx].name,
803 					"enable-fwd") == 0)
804 				enable_fwd = true;
805 			if (strcmp(lgopts[opt_idx].name,
806 					"portmask") == 0) {
807 				/* parse hexadecimal string */
808 				end = NULL;
809 				pm = strtoull(optarg, &end, 16);
810 				if ((optarg[0] == '\0') || (end == NULL) || (*end != '\0'))
811 					rte_exit(EXIT_FAILURE, "Invalid fwd port mask\n");
812 				ports_mask = pm;
813 			}
814 			if (strcmp(lgopts[opt_idx].name, "rxq") == 0) {
815 				n = atoi(optarg);
816 				rx_queues_count = (uint8_t) n;
817 			}
818 			if (strcmp(lgopts[opt_idx].name, "txq") == 0) {
819 				n = atoi(optarg);
820 				tx_queues_count = (uint8_t) n;
821 			}
822 			if (strcmp(lgopts[opt_idx].name, "rxd") == 0) {
823 				n = atoi(optarg);
824 				rxd_count = (uint8_t) n;
825 			}
826 			if (strcmp(lgopts[opt_idx].name, "txd") == 0) {
827 				n = atoi(optarg);
828 				txd_count = (uint8_t) n;
829 			}
830 			if (strcmp(lgopts[opt_idx].name, "mbuf-size") == 0) {
831 				n = atoi(optarg);
832 				mbuf_size = (uint32_t) n;
833 			}
834 			if (strcmp(lgopts[opt_idx].name, "mbuf-cache-size") == 0) {
835 				n = atoi(optarg);
836 				mbuf_cache_size = (uint32_t) n;
837 			}
838 			if (strcmp(lgopts[opt_idx].name, "total-mbuf-count") == 0) {
839 				n = atoi(optarg);
840 				total_mbuf_num = (uint32_t) n;
841 			}
842 			if (strcmp(lgopts[opt_idx].name, "cores") == 0) {
843 				n = atoi(optarg);
844 				if ((int) rte_lcore_count() <= n) {
845 					rte_exit(EXIT_FAILURE,
846 						"Error: you need %d cores to run on multi-cores\n"
847 						"Existing cores are: %d\n", n, rte_lcore_count());
848 				}
849 				if (n <= RTE_MAX_LCORE && n > 0)
850 					mc_pool.cores_count = n;
851 				else {
852 					rte_exit(EXIT_FAILURE,
853 						"Error: cores count must be > 0 and < %d\n",
854 						RTE_MAX_LCORE);
855 				}
856 			}
857 			break;
858 		default:
859 			usage(argv[0]);
860 			rte_exit(EXIT_FAILURE, "Invalid option: %s\n",
861 					argv[optind]);
862 			break;
863 		}
864 	}
865 	printf("end_flow\n");
866 }
867 
868 /* Dump the socket memory statistics on console */
869 static size_t
870 dump_socket_mem(FILE *f)
871 {
872 	struct rte_malloc_socket_stats socket_stats;
873 	unsigned int i = 0;
874 	size_t total = 0;
875 	size_t alloc = 0;
876 	size_t free = 0;
877 	unsigned int n_alloc = 0;
878 	unsigned int n_free = 0;
879 	bool active_nodes = false;
880 
881 
882 	for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
883 		if (rte_malloc_get_socket_stats(i, &socket_stats) ||
884 		    !socket_stats.heap_totalsz_bytes)
885 			continue;
886 		active_nodes = true;
887 		total += socket_stats.heap_totalsz_bytes;
888 		alloc += socket_stats.heap_allocsz_bytes;
889 		free += socket_stats.heap_freesz_bytes;
890 		n_alloc += socket_stats.alloc_count;
891 		n_free += socket_stats.free_count;
892 		if (dump_socket_mem_flag) {
893 			fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
894 			fprintf(f,
895 				"\nSocket %u:\nsize(M) total: %.6lf\nalloc:"
896 				" %.6lf(%.3lf%%)\nfree: %.6lf"
897 				"\nmax: %.6lf"
898 				"\ncount alloc: %u\nfree: %u\n",
899 				i,
900 				socket_stats.heap_totalsz_bytes / 1.0e6,
901 				socket_stats.heap_allocsz_bytes / 1.0e6,
902 				(double)socket_stats.heap_allocsz_bytes * 100 /
903 				(double)socket_stats.heap_totalsz_bytes,
904 				socket_stats.heap_freesz_bytes / 1.0e6,
905 				socket_stats.greatest_free_size / 1.0e6,
906 				socket_stats.alloc_count,
907 				socket_stats.free_count);
908 				fprintf(f, "::::::::::::::::::::::::::::::::::::::::");
909 		}
910 	}
911 	if (dump_socket_mem_flag && active_nodes) {
912 		fprintf(f,
913 			"\nTotal: size(M)\ntotal: %.6lf"
914 			"\nalloc: %.6lf(%.3lf%%)\nfree: %.6lf"
915 			"\ncount alloc: %u\nfree: %u\n",
916 			total / 1.0e6, alloc / 1.0e6,
917 			(double)alloc * 100 / (double)total, free / 1.0e6,
918 			n_alloc, n_free);
919 		fprintf(f, "::::::::::::::::::::::::::::::::::::::::\n");
920 	}
921 	return alloc;
922 }
923 
924 static void
925 print_flow_error(struct rte_flow_error error)
926 {
927 	printf("Flow can't be created %d message: %s\n",
928 		error.type,
929 		error.message ? error.message : "(no stated reason)");
930 }
931 
932 static inline void
933 print_rules_batches(double *cpu_time_per_batch)
934 {
935 	uint8_t idx;
936 	double delta;
937 	double rate;
938 
939 	for (idx = 0; idx < MAX_BATCHES_COUNT; idx++) {
940 		if (!cpu_time_per_batch[idx])
941 			break;
942 		delta = (double)(rules_batch / cpu_time_per_batch[idx]);
943 		rate = delta / 1000; /* Save rate in K unit. */
944 		printf(":: Rules batch #%d: %d rules "
945 			"in %f sec[ Rate = %f K Rule/Sec ]\n",
946 			idx, rules_batch,
947 			cpu_time_per_batch[idx], rate);
948 	}
949 }
950 
951 
952 static inline int
953 has_meter(void)
954 {
955 	int i;
956 
957 	for (i = 0; i < MAX_ACTIONS_NUM; i++) {
958 		if (flow_actions[i] == 0)
959 			break;
960 		if (flow_actions[i]
961 				& FLOW_ACTION_MASK(RTE_FLOW_ACTION_TYPE_METER))
962 			return 1;
963 	}
964 	return 0;
965 }
966 
967 static void
968 create_meter_rule(int port_id, uint32_t counter)
969 {
970 	int ret;
971 	struct rte_mtr_params params;
972 	uint32_t default_prof_id = 100;
973 	struct rte_mtr_error error;
974 
975 	memset(&params, 0, sizeof(struct rte_mtr_params));
976 	params.meter_enable = 1;
977 	params.stats_mask = 0xffff;
978 	params.use_prev_mtr_color = 0;
979 	params.dscp_table = NULL;
980 
981 	/*create meter*/
982 	params.meter_profile_id = default_prof_id;
983 	ret = rte_mtr_create(port_id, counter, &params, 1, &error);
984 	if (ret != 0) {
985 		printf("Port %u create meter idx(%d) error(%d) message: %s\n",
986 			port_id, counter, error.type,
987 			error.message ? error.message : "(no stated reason)");
988 		rte_exit(EXIT_FAILURE, "Error in creating meter\n");
989 	}
990 }
991 
992 static void
993 destroy_meter_rule(int port_id, uint32_t counter)
994 {
995 	struct rte_mtr_error error;
996 
997 	if (rte_mtr_destroy(port_id, counter, &error)) {
998 		printf("Port %u destroy meter(%d) error(%d) message: %s\n",
999 			port_id, counter, error.type,
1000 			error.message ? error.message : "(no stated reason)");
1001 		rte_exit(EXIT_FAILURE, "Error in deleting meter rule\n");
1002 	}
1003 }
1004 
1005 static void
1006 meters_handler(int port_id, uint8_t core_id, uint8_t ops)
1007 {
1008 	uint64_t start_batch;
1009 	double cpu_time_used, insertion_rate;
1010 	int rules_count_per_core, rules_batch_idx;
1011 	uint32_t counter, start_counter = 0, end_counter;
1012 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1013 
1014 	rules_count_per_core = rules_count / mc_pool.cores_count;
1015 
1016 	if (core_id)
1017 		start_counter = core_id * rules_count_per_core;
1018 	end_counter = (core_id + 1) * rules_count_per_core;
1019 
1020 	cpu_time_used = 0;
1021 	start_batch = rte_get_timer_cycles();
1022 	for (counter = start_counter; counter < end_counter; counter++) {
1023 		if (ops == METER_CREATE)
1024 			create_meter_rule(port_id, counter);
1025 		else
1026 			destroy_meter_rule(port_id, counter);
1027 		/*
1028 		 * Save the insertion rate for rules batch.
1029 		 * Check if the insertion reached the rules
1030 		 * patch counter, then save the insertion rate
1031 		 * for this batch.
1032 		 */
1033 		if (!((counter + 1) % rules_batch)) {
1034 			rules_batch_idx = ((counter + 1) / rules_batch) - 1;
1035 			cpu_time_per_batch[rules_batch_idx] =
1036 				((double)(rte_get_timer_cycles() - start_batch))
1037 				/ rte_get_timer_hz();
1038 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1039 			start_batch = rte_get_timer_cycles();
1040 		}
1041 	}
1042 
1043 	/* Print insertion rates for all batches */
1044 	if (dump_iterations)
1045 		print_rules_batches(cpu_time_per_batch);
1046 
1047 	insertion_rate =
1048 		((double) (rules_count_per_core / cpu_time_used) / 1000);
1049 
1050 	/* Insertion rate for all rules in one core */
1051 	printf(":: Port %d :: Core %d Meter %s :: start @[%d] - end @[%d],"
1052 		" use:%.02fs, rate:%.02fk Rule/Sec\n",
1053 		port_id, core_id, ops == METER_CREATE ? "create" : "delete",
1054 		start_counter, end_counter - 1,
1055 		cpu_time_used, insertion_rate);
1056 
1057 	if (ops == METER_CREATE)
1058 		mc_pool.meters_record.insertion[port_id][core_id]
1059 			= cpu_time_used;
1060 	else
1061 		mc_pool.meters_record.deletion[port_id][core_id]
1062 			= cpu_time_used;
1063 }
1064 
1065 static void
1066 destroy_meter_profile(void)
1067 {
1068 	struct rte_mtr_error error;
1069 	uint16_t nr_ports;
1070 	int port_id;
1071 
1072 	nr_ports = rte_eth_dev_count_avail();
1073 	for (port_id = 0; port_id < nr_ports; port_id++) {
1074 		/* If port outside portmask */
1075 		if (!((ports_mask >> port_id) & 0x1))
1076 			continue;
1077 
1078 		if (rte_mtr_meter_profile_delete
1079 			(port_id, DEFAULT_METER_PROF_ID, &error)) {
1080 			printf("Port %u del profile error(%d) message: %s\n",
1081 				port_id, error.type,
1082 				error.message ? error.message : "(no stated reason)");
1083 			rte_exit(EXIT_FAILURE, "Error: Destroy meter profile Failed!\n");
1084 		}
1085 	}
1086 }
1087 
1088 static void
1089 create_meter_profile(void)
1090 {
1091 	uint16_t nr_ports;
1092 	int ret, port_id;
1093 	struct rte_mtr_meter_profile mp;
1094 	struct rte_mtr_error error;
1095 
1096 	/*
1097 	 *currently , only create one meter file for one port
1098 	 *1 meter profile -> N meter rules -> N rte flows
1099 	 */
1100 	memset(&mp, 0, sizeof(struct rte_mtr_meter_profile));
1101 	nr_ports = rte_eth_dev_count_avail();
1102 	for (port_id = 0; port_id < nr_ports; port_id++) {
1103 		/* If port outside portmask */
1104 		if (!((ports_mask >> port_id) & 0x1))
1105 			continue;
1106 
1107 		mp.alg = RTE_MTR_SRTCM_RFC2697;
1108 		mp.srtcm_rfc2697.cir = METER_CIR;
1109 		mp.srtcm_rfc2697.cbs = METER_CIR / 8;
1110 		mp.srtcm_rfc2697.ebs = 0;
1111 
1112 		ret = rte_mtr_meter_profile_add
1113 			(port_id, DEFAULT_METER_PROF_ID, &mp, &error);
1114 		if (ret != 0) {
1115 			printf("Port %u create Profile error(%d) message: %s\n",
1116 				port_id, error.type,
1117 				error.message ? error.message : "(no stated reason)");
1118 			rte_exit(EXIT_FAILURE, "Error: Creation meter profile Failed!\n");
1119 		}
1120 	}
1121 }
1122 
1123 static inline void
1124 destroy_flows(int port_id, uint8_t core_id, struct rte_flow **flows_list)
1125 {
1126 	struct rte_flow_error error;
1127 	clock_t start_batch, end_batch;
1128 	double cpu_time_used = 0;
1129 	double deletion_rate;
1130 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1131 	double delta;
1132 	uint32_t i;
1133 	int rules_batch_idx;
1134 	int rules_count_per_core;
1135 
1136 	rules_count_per_core = rules_count / mc_pool.cores_count;
1137 	/* If group > 0 , should add 1 flow which created in group 0 */
1138 	if (flow_group > 0 && core_id == 0)
1139 		rules_count_per_core++;
1140 
1141 	start_batch = rte_get_timer_cycles();
1142 	for (i = 0; i < (uint32_t) rules_count_per_core; i++) {
1143 		if (flows_list[i] == 0)
1144 			break;
1145 
1146 		memset(&error, 0x33, sizeof(error));
1147 		if (rte_flow_destroy(port_id, flows_list[i], &error)) {
1148 			print_flow_error(error);
1149 			rte_exit(EXIT_FAILURE, "Error in deleting flow\n");
1150 		}
1151 
1152 		/*
1153 		 * Save the deletion rate for rules batch.
1154 		 * Check if the deletion reached the rules
1155 		 * patch counter, then save the deletion rate
1156 		 * for this batch.
1157 		 */
1158 		if (!((i + 1) % rules_batch)) {
1159 			end_batch = rte_get_timer_cycles();
1160 			delta = (double) (end_batch - start_batch);
1161 			rules_batch_idx = ((i + 1) / rules_batch) - 1;
1162 			cpu_time_per_batch[rules_batch_idx] = delta / rte_get_timer_hz();
1163 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1164 			start_batch = rte_get_timer_cycles();
1165 		}
1166 	}
1167 
1168 	/* Print deletion rates for all batches */
1169 	if (dump_iterations)
1170 		print_rules_batches(cpu_time_per_batch);
1171 
1172 	/* Deletion rate for all rules */
1173 	deletion_rate = ((double) (rules_count_per_core / cpu_time_used) / 1000);
1174 	printf(":: Port %d :: Core %d :: Rules deletion rate -> %f K Rule/Sec\n",
1175 		port_id, core_id, deletion_rate);
1176 	printf(":: Port %d :: Core %d :: The time for deleting %d rules is %f seconds\n",
1177 		port_id, core_id, rules_count_per_core, cpu_time_used);
1178 
1179 	mc_pool.flows_record.deletion[port_id][core_id] = cpu_time_used;
1180 }
1181 
1182 static struct rte_flow **
1183 insert_flows(int port_id, uint8_t core_id)
1184 {
1185 	struct rte_flow **flows_list;
1186 	struct rte_flow_error error;
1187 	clock_t start_batch, end_batch;
1188 	double first_flow_latency;
1189 	double cpu_time_used;
1190 	double insertion_rate;
1191 	double cpu_time_per_batch[MAX_BATCHES_COUNT] = { 0 };
1192 	double delta;
1193 	uint32_t flow_index;
1194 	uint32_t counter, start_counter = 0, end_counter;
1195 	uint64_t global_items[MAX_ITEMS_NUM] = { 0 };
1196 	uint64_t global_actions[MAX_ACTIONS_NUM] = { 0 };
1197 	int rules_batch_idx;
1198 	int rules_count_per_core;
1199 
1200 	rules_count_per_core = rules_count / mc_pool.cores_count;
1201 
1202 	/* Set boundaries of rules for each core. */
1203 	if (core_id)
1204 		start_counter = core_id * rules_count_per_core;
1205 	end_counter = (core_id + 1) * rules_count_per_core;
1206 
1207 	global_items[0] = FLOW_ITEM_MASK(RTE_FLOW_ITEM_TYPE_ETH);
1208 	global_actions[0] = FLOW_ITEM_MASK(RTE_FLOW_ACTION_TYPE_JUMP);
1209 
1210 	flows_list = rte_zmalloc("flows_list",
1211 		(sizeof(struct rte_flow *) * rules_count_per_core) + 1, 0);
1212 	if (flows_list == NULL)
1213 		rte_exit(EXIT_FAILURE, "No Memory available!\n");
1214 
1215 	cpu_time_used = 0;
1216 	flow_index = 0;
1217 	if (flow_group > 0 && core_id == 0) {
1218 		/*
1219 		 * Create global rule to jump into flow_group,
1220 		 * this way the app will avoid the default rules.
1221 		 *
1222 		 * This rule will be created only once.
1223 		 *
1224 		 * Global rule:
1225 		 * group 0 eth / end actions jump group <flow_group>
1226 		 */
1227 		flow = generate_flow(port_id, 0, flow_attrs,
1228 			global_items, global_actions,
1229 			flow_group, 0, 0, 0, 0, core_id, rx_queues_count,
1230 			unique_data, &error);
1231 
1232 		if (flow == NULL) {
1233 			print_flow_error(error);
1234 			rte_exit(EXIT_FAILURE, "Error in creating flow\n");
1235 		}
1236 		flows_list[flow_index++] = flow;
1237 	}
1238 
1239 	start_batch = rte_get_timer_cycles();
1240 	for (counter = start_counter; counter < end_counter; counter++) {
1241 		flow = generate_flow(port_id, flow_group,
1242 			flow_attrs, flow_items, flow_actions,
1243 			JUMP_ACTION_TABLE, counter,
1244 			hairpin_queues_num,
1245 			encap_data, decap_data,
1246 			core_id, rx_queues_count,
1247 			unique_data, &error);
1248 
1249 		if (!counter) {
1250 			first_flow_latency = (double) (rte_get_timer_cycles() - start_batch);
1251 			first_flow_latency /= rte_get_timer_hz();
1252 			/* In millisecond */
1253 			first_flow_latency *= 1000;
1254 			printf(":: First Flow Latency :: Port %d :: First flow "
1255 				"installed in %f milliseconds\n",
1256 				port_id, first_flow_latency);
1257 		}
1258 
1259 		if (force_quit)
1260 			counter = end_counter;
1261 
1262 		if (!flow) {
1263 			print_flow_error(error);
1264 			rte_exit(EXIT_FAILURE, "Error in creating flow\n");
1265 		}
1266 
1267 		flows_list[flow_index++] = flow;
1268 
1269 		/*
1270 		 * Save the insertion rate for rules batch.
1271 		 * Check if the insertion reached the rules
1272 		 * patch counter, then save the insertion rate
1273 		 * for this batch.
1274 		 */
1275 		if (!((counter + 1) % rules_batch)) {
1276 			end_batch = rte_get_timer_cycles();
1277 			delta = (double) (end_batch - start_batch);
1278 			rules_batch_idx = ((counter + 1) / rules_batch) - 1;
1279 			cpu_time_per_batch[rules_batch_idx] = delta / rte_get_timer_hz();
1280 			cpu_time_used += cpu_time_per_batch[rules_batch_idx];
1281 			start_batch = rte_get_timer_cycles();
1282 		}
1283 	}
1284 
1285 	/* Print insertion rates for all batches */
1286 	if (dump_iterations)
1287 		print_rules_batches(cpu_time_per_batch);
1288 
1289 	printf(":: Port %d :: Core %d boundaries :: start @[%d] - end @[%d]\n",
1290 		port_id, core_id, start_counter, end_counter - 1);
1291 
1292 	/* Insertion rate for all rules in one core */
1293 	insertion_rate = ((double) (rules_count_per_core / cpu_time_used) / 1000);
1294 	printf(":: Port %d :: Core %d :: Rules insertion rate -> %f K Rule/Sec\n",
1295 		port_id, core_id, insertion_rate);
1296 	printf(":: Port %d :: Core %d :: The time for creating %d in rules %f seconds\n",
1297 		port_id, core_id, rules_count_per_core, cpu_time_used);
1298 
1299 	mc_pool.flows_record.insertion[port_id][core_id] = cpu_time_used;
1300 	return flows_list;
1301 }
1302 
1303 static void
1304 flows_handler(uint8_t core_id)
1305 {
1306 	struct rte_flow **flows_list;
1307 	uint16_t nr_ports;
1308 	int port_id;
1309 
1310 	nr_ports = rte_eth_dev_count_avail();
1311 
1312 	if (rules_batch > rules_count)
1313 		rules_batch = rules_count;
1314 
1315 	printf(":: Rules Count per port: %d\n\n", rules_count);
1316 
1317 	for (port_id = 0; port_id < nr_ports; port_id++) {
1318 		/* If port outside portmask */
1319 		if (!((ports_mask >> port_id) & 0x1))
1320 			continue;
1321 
1322 		/* Insertion part. */
1323 		mc_pool.last_alloc[core_id] = (int64_t)dump_socket_mem(stdout);
1324 		if (has_meter())
1325 			meters_handler(port_id, core_id, METER_CREATE);
1326 		flows_list = insert_flows(port_id, core_id);
1327 		if (flows_list == NULL)
1328 			rte_exit(EXIT_FAILURE, "Error: Insertion Failed!\n");
1329 		mc_pool.current_alloc[core_id] = (int64_t)dump_socket_mem(stdout);
1330 
1331 		/* Deletion part. */
1332 		if (delete_flag) {
1333 			destroy_flows(port_id, core_id, flows_list);
1334 			if (has_meter())
1335 				meters_handler(port_id, core_id, METER_DELETE);
1336 		}
1337 	}
1338 }
1339 
1340 static void
1341 dump_used_cpu_time(const char *item,
1342 		uint16_t port, struct used_cpu_time *used_time)
1343 {
1344 	uint32_t i;
1345 	/* Latency: total count of rte rules divided
1346 	 * over max time used by thread between all
1347 	 * threads time.
1348 	 *
1349 	 * Throughput: total count of rte rules divided
1350 	 * over the average of the time cosumed by all
1351 	 * threads time.
1352 	 */
1353 	double insertion_latency_time;
1354 	double insertion_throughput_time;
1355 	double deletion_latency_time;
1356 	double deletion_throughput_time;
1357 	double insertion_latency, insertion_throughput;
1358 	double deletion_latency, deletion_throughput;
1359 
1360 	/* Save first insertion/deletion rates from first thread.
1361 	 * Start comparing with all threads, if any thread used
1362 	 * time more than current saved, replace it.
1363 	 *
1364 	 * Thus in the end we will have the max time used for
1365 	 * insertion/deletion by one thread.
1366 	 *
1367 	 * As for memory consumption, save the min of all threads
1368 	 * of last alloc, and save the max for all threads for
1369 	 * current alloc.
1370 	 */
1371 
1372 	insertion_latency_time = used_time->insertion[port][0];
1373 	deletion_latency_time = used_time->deletion[port][0];
1374 	insertion_throughput_time = used_time->insertion[port][0];
1375 	deletion_throughput_time = used_time->deletion[port][0];
1376 
1377 	i = mc_pool.cores_count;
1378 	while (i-- > 1) {
1379 		insertion_throughput_time += used_time->insertion[port][i];
1380 		deletion_throughput_time += used_time->deletion[port][i];
1381 		if (insertion_latency_time < used_time->insertion[port][i])
1382 			insertion_latency_time = used_time->insertion[port][i];
1383 		if (deletion_latency_time < used_time->deletion[port][i])
1384 			deletion_latency_time = used_time->deletion[port][i];
1385 	}
1386 
1387 	insertion_latency = ((double) (mc_pool.rules_count
1388 				/ insertion_latency_time) / 1000);
1389 	deletion_latency = ((double) (mc_pool.rules_count
1390 				/ deletion_latency_time) / 1000);
1391 
1392 	insertion_throughput_time /= mc_pool.cores_count;
1393 	deletion_throughput_time /= mc_pool.cores_count;
1394 	insertion_throughput = ((double) (mc_pool.rules_count
1395 				/ insertion_throughput_time) / 1000);
1396 	deletion_throughput = ((double) (mc_pool.rules_count
1397 				/ deletion_throughput_time) / 1000);
1398 
1399 	/* Latency stats */
1400 	printf("\n%s\n:: [Latency | Insertion] All Cores :: Port %d :: ",
1401 		item, port);
1402 	printf("Total flows insertion rate -> %f K Rules/Sec\n",
1403 		insertion_latency);
1404 	printf(":: [Latency | Insertion] All Cores :: Port %d :: ", port);
1405 	printf("The time for creating %d rules is %f seconds\n",
1406 		mc_pool.rules_count, insertion_latency_time);
1407 
1408 	/* Throughput stats */
1409 	printf(":: [Throughput | Insertion] All Cores :: Port %d :: ", port);
1410 	printf("Total flows insertion rate -> %f K Rules/Sec\n",
1411 		insertion_throughput);
1412 	printf(":: [Throughput | Insertion] All Cores :: Port %d :: ", port);
1413 	printf("The average time for creating %d rules is %f seconds\n",
1414 		mc_pool.rules_count, insertion_throughput_time);
1415 
1416 	if (delete_flag) {
1417 	/* Latency stats */
1418 		printf(":: [Latency | Deletion] All Cores :: Port %d :: Total "
1419 			"deletion rate -> %f K Rules/Sec\n",
1420 			port, deletion_latency);
1421 		printf(":: [Latency | Deletion] All Cores :: Port %d :: ",
1422 			port);
1423 		printf("The time for deleting %d rules is %f seconds\n",
1424 			mc_pool.rules_count, deletion_latency_time);
1425 
1426 		/* Throughput stats */
1427 		printf(":: [Throughput | Deletion] All Cores :: Port %d :: Total "
1428 			"deletion rate -> %f K Rules/Sec\n",
1429 			port, deletion_throughput);
1430 		printf(":: [Throughput | Deletion] All Cores :: Port %d :: ",
1431 			port);
1432 		printf("The average time for deleting %d rules is %f seconds\n",
1433 			mc_pool.rules_count, deletion_throughput_time);
1434 	}
1435 }
1436 
1437 static void
1438 dump_used_mem(uint16_t port)
1439 {
1440 	uint32_t i;
1441 	int64_t last_alloc, current_alloc;
1442 	int flow_size_in_bytes;
1443 
1444 	last_alloc = mc_pool.last_alloc[0];
1445 	current_alloc = mc_pool.current_alloc[0];
1446 
1447 	i = mc_pool.cores_count;
1448 	while (i-- > 1) {
1449 		if (last_alloc > mc_pool.last_alloc[i])
1450 			last_alloc = mc_pool.last_alloc[i];
1451 		if (current_alloc < mc_pool.current_alloc[i])
1452 			current_alloc = mc_pool.current_alloc[i];
1453 	}
1454 
1455 	flow_size_in_bytes = (current_alloc - last_alloc) / mc_pool.rules_count;
1456 	printf("\n:: Port %d :: rte_flow size in DPDK layer: %d Bytes\n",
1457 		port, flow_size_in_bytes);
1458 }
1459 
1460 static int
1461 run_rte_flow_handler_cores(void *data __rte_unused)
1462 {
1463 	uint16_t port;
1464 	int lcore_counter = 0;
1465 	int lcore_id = rte_lcore_id();
1466 	int i;
1467 
1468 	RTE_LCORE_FOREACH(i) {
1469 		/*  If core not needed return. */
1470 		if (lcore_id == i) {
1471 			printf(":: lcore %d mapped with index %d\n", lcore_id, lcore_counter);
1472 			if (lcore_counter >= (int) mc_pool.cores_count)
1473 				return 0;
1474 			break;
1475 		}
1476 		lcore_counter++;
1477 	}
1478 	lcore_id = lcore_counter;
1479 
1480 	if (lcore_id >= (int) mc_pool.cores_count)
1481 		return 0;
1482 
1483 	mc_pool.rules_count = rules_count;
1484 
1485 	flows_handler(lcore_id);
1486 
1487 	/* Only main core to print total results. */
1488 	if (lcore_id != 0)
1489 		return 0;
1490 
1491 	/* Make sure all cores finished insertion/deletion process. */
1492 	rte_eal_mp_wait_lcore();
1493 
1494 	RTE_ETH_FOREACH_DEV(port) {
1495 		/* If port outside portmask */
1496 		if (!((ports_mask >> port) & 0x1))
1497 			continue;
1498 		if (has_meter())
1499 			dump_used_cpu_time("Meters:",
1500 				port, &mc_pool.meters_record);
1501 		dump_used_cpu_time("Flows:",
1502 			port, &mc_pool.flows_record);
1503 		dump_used_mem(port);
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 static void
1510 signal_handler(int signum)
1511 {
1512 	if (signum == SIGINT || signum == SIGTERM) {
1513 		printf("\n\nSignal %d received, preparing to exit...\n",
1514 					signum);
1515 		printf("Error: Stats are wrong due to sudden signal!\n\n");
1516 		force_quit = true;
1517 	}
1518 }
1519 
1520 static inline uint16_t
1521 do_rx(struct lcore_info *li, uint16_t rx_port, uint16_t rx_queue)
1522 {
1523 	uint16_t cnt = 0;
1524 	cnt = rte_eth_rx_burst(rx_port, rx_queue, li->pkts, MAX_PKT_BURST);
1525 	li->rx_pkts += cnt;
1526 	return cnt;
1527 }
1528 
1529 static inline void
1530 do_tx(struct lcore_info *li, uint16_t cnt, uint16_t tx_port,
1531 			uint16_t tx_queue)
1532 {
1533 	uint16_t nr_tx = 0;
1534 	uint16_t i;
1535 
1536 	nr_tx = rte_eth_tx_burst(tx_port, tx_queue, li->pkts, cnt);
1537 	li->tx_pkts  += nr_tx;
1538 	li->tx_drops += cnt - nr_tx;
1539 
1540 	for (i = nr_tx; i < cnt; i++)
1541 		rte_pktmbuf_free(li->pkts[i]);
1542 }
1543 
1544 /*
1545  * Method to convert numbers into pretty numbers that easy
1546  * to read. The design here is to add comma after each three
1547  * digits and set all of this inside buffer.
1548  *
1549  * For example if n = 1799321, the output will be
1550  * 1,799,321 after this method which is easier to read.
1551  */
1552 static char *
1553 pretty_number(uint64_t n, char *buf)
1554 {
1555 	char p[6][4];
1556 	int i = 0;
1557 	int off = 0;
1558 
1559 	while (n > 1000) {
1560 		sprintf(p[i], "%03d", (int)(n % 1000));
1561 		n /= 1000;
1562 		i += 1;
1563 	}
1564 
1565 	sprintf(p[i++], "%d", (int)n);
1566 
1567 	while (i--)
1568 		off += sprintf(buf + off, "%s,", p[i]);
1569 	buf[strlen(buf) - 1] = '\0';
1570 
1571 	return buf;
1572 }
1573 
1574 static void
1575 packet_per_second_stats(void)
1576 {
1577 	struct lcore_info *old;
1578 	struct lcore_info *li, *oli;
1579 	int nr_lines = 0;
1580 	int i;
1581 
1582 	old = rte_zmalloc("old",
1583 		sizeof(struct lcore_info) * RTE_MAX_LCORE, 0);
1584 	if (old == NULL)
1585 		rte_exit(EXIT_FAILURE, "No Memory available!\n");
1586 
1587 	memcpy(old, lcore_infos,
1588 		sizeof(struct lcore_info) * RTE_MAX_LCORE);
1589 
1590 	while (!force_quit) {
1591 		uint64_t total_tx_pkts = 0;
1592 		uint64_t total_rx_pkts = 0;
1593 		uint64_t total_tx_drops = 0;
1594 		uint64_t tx_delta, rx_delta, drops_delta;
1595 		char buf[3][32];
1596 		int nr_valid_core = 0;
1597 
1598 		sleep(1);
1599 
1600 		if (nr_lines) {
1601 			char go_up_nr_lines[16];
1602 
1603 			sprintf(go_up_nr_lines, "%c[%dA\r", 27, nr_lines);
1604 			printf("%s\r", go_up_nr_lines);
1605 		}
1606 
1607 		printf("\n%6s %16s %16s %16s\n", "core", "tx", "tx drops", "rx");
1608 		printf("%6s %16s %16s %16s\n", "------", "----------------",
1609 			"----------------", "----------------");
1610 		nr_lines = 3;
1611 		for (i = 0; i < RTE_MAX_LCORE; i++) {
1612 			li  = &lcore_infos[i];
1613 			oli = &old[i];
1614 			if (li->mode != LCORE_MODE_PKT)
1615 				continue;
1616 
1617 			tx_delta    = li->tx_pkts  - oli->tx_pkts;
1618 			rx_delta    = li->rx_pkts  - oli->rx_pkts;
1619 			drops_delta = li->tx_drops - oli->tx_drops;
1620 			printf("%6d %16s %16s %16s\n", i,
1621 				pretty_number(tx_delta,    buf[0]),
1622 				pretty_number(drops_delta, buf[1]),
1623 				pretty_number(rx_delta,    buf[2]));
1624 
1625 			total_tx_pkts  += tx_delta;
1626 			total_rx_pkts  += rx_delta;
1627 			total_tx_drops += drops_delta;
1628 
1629 			nr_valid_core++;
1630 			nr_lines += 1;
1631 		}
1632 
1633 		if (nr_valid_core > 1) {
1634 			printf("%6s %16s %16s %16s\n", "total",
1635 				pretty_number(total_tx_pkts,  buf[0]),
1636 				pretty_number(total_tx_drops, buf[1]),
1637 				pretty_number(total_rx_pkts,  buf[2]));
1638 			nr_lines += 1;
1639 		}
1640 
1641 		memcpy(old, lcore_infos,
1642 			sizeof(struct lcore_info) * RTE_MAX_LCORE);
1643 	}
1644 }
1645 
1646 static int
1647 start_forwarding(void *data __rte_unused)
1648 {
1649 	int lcore = rte_lcore_id();
1650 	int stream_id;
1651 	uint16_t cnt;
1652 	struct lcore_info *li = &lcore_infos[lcore];
1653 
1654 	if (!li->mode)
1655 		return 0;
1656 
1657 	if (li->mode == LCORE_MODE_STATS) {
1658 		printf(":: started stats on lcore %u\n", lcore);
1659 		packet_per_second_stats();
1660 		return 0;
1661 	}
1662 
1663 	while (!force_quit)
1664 		for (stream_id = 0; stream_id < MAX_STREAMS; stream_id++) {
1665 			if (li->streams[stream_id].rx_port == -1)
1666 				continue;
1667 
1668 			cnt = do_rx(li,
1669 					li->streams[stream_id].rx_port,
1670 					li->streams[stream_id].rx_queue);
1671 			if (cnt)
1672 				do_tx(li, cnt,
1673 					li->streams[stream_id].tx_port,
1674 					li->streams[stream_id].tx_queue);
1675 		}
1676 	return 0;
1677 }
1678 
1679 static void
1680 init_lcore_info(void)
1681 {
1682 	int i, j;
1683 	unsigned int lcore;
1684 	uint16_t nr_port;
1685 	uint16_t queue;
1686 	int port;
1687 	int stream_id = 0;
1688 	int streams_per_core;
1689 	int unassigned_streams;
1690 	int nb_fwd_streams;
1691 	nr_port = rte_eth_dev_count_avail();
1692 
1693 	/* First logical core is reserved for stats printing */
1694 	lcore = rte_get_next_lcore(-1, 0, 0);
1695 	lcore_infos[lcore].mode = LCORE_MODE_STATS;
1696 
1697 	/*
1698 	 * Initialize all cores
1699 	 * All cores at first must have -1 value in all streams
1700 	 * This means that this stream is not used, or not set
1701 	 * yet.
1702 	 */
1703 	for (i = 0; i < RTE_MAX_LCORE; i++)
1704 		for (j = 0; j < MAX_STREAMS; j++) {
1705 			lcore_infos[i].streams[j].tx_port = -1;
1706 			lcore_infos[i].streams[j].rx_port = -1;
1707 			lcore_infos[i].streams[j].tx_queue = -1;
1708 			lcore_infos[i].streams[j].rx_queue = -1;
1709 			lcore_infos[i].streams_nb = 0;
1710 		}
1711 
1712 	/*
1713 	 * Calculate the total streams count.
1714 	 * Also distribute those streams count between the available
1715 	 * logical cores except first core, since it's reserved for
1716 	 * stats prints.
1717 	 */
1718 	nb_fwd_streams = nr_port * rx_queues_count;
1719 	if ((int)(nb_lcores - 1) >= nb_fwd_streams)
1720 		for (i = 0; i < (int)(nb_lcores - 1); i++) {
1721 			lcore = rte_get_next_lcore(lcore, 0, 0);
1722 			lcore_infos[lcore].streams_nb = 1;
1723 		}
1724 	else {
1725 		streams_per_core = nb_fwd_streams / (nb_lcores - 1);
1726 		unassigned_streams = nb_fwd_streams % (nb_lcores - 1);
1727 		for (i = 0; i < (int)(nb_lcores - 1); i++) {
1728 			lcore = rte_get_next_lcore(lcore, 0, 0);
1729 			lcore_infos[lcore].streams_nb = streams_per_core;
1730 			if (unassigned_streams) {
1731 				lcore_infos[lcore].streams_nb++;
1732 				unassigned_streams--;
1733 			}
1734 		}
1735 	}
1736 
1737 	/*
1738 	 * Set the streams for the cores according to each logical
1739 	 * core stream count.
1740 	 * The streams is built on the design of what received should
1741 	 * forward as well, this means that if you received packets on
1742 	 * port 0 queue 0 then the same queue should forward the
1743 	 * packets, using the same logical core.
1744 	 */
1745 	lcore = rte_get_next_lcore(-1, 0, 0);
1746 	for (port = 0; port < nr_port; port++) {
1747 		/* Create FWD stream */
1748 		for (queue = 0; queue < rx_queues_count; queue++) {
1749 			if (!lcore_infos[lcore].streams_nb ||
1750 				!(stream_id % lcore_infos[lcore].streams_nb)) {
1751 				lcore = rte_get_next_lcore(lcore, 0, 0);
1752 				lcore_infos[lcore].mode = LCORE_MODE_PKT;
1753 				stream_id = 0;
1754 			}
1755 			lcore_infos[lcore].streams[stream_id].rx_queue = queue;
1756 			lcore_infos[lcore].streams[stream_id].tx_queue = queue;
1757 			lcore_infos[lcore].streams[stream_id].rx_port = port;
1758 			lcore_infos[lcore].streams[stream_id].tx_port = port;
1759 			stream_id++;
1760 		}
1761 	}
1762 
1763 	/* Print all streams */
1764 	printf(":: Stream -> core id[N]: (rx_port, rx_queue)->(tx_port, tx_queue)\n");
1765 	for (i = 0; i < RTE_MAX_LCORE; i++)
1766 		for (j = 0; j < MAX_STREAMS; j++) {
1767 			/* No streams for this core */
1768 			if (lcore_infos[i].streams[j].tx_port == -1)
1769 				break;
1770 			printf("Stream -> core id[%d]: (%d,%d)->(%d,%d)\n",
1771 				i,
1772 				lcore_infos[i].streams[j].rx_port,
1773 				lcore_infos[i].streams[j].rx_queue,
1774 				lcore_infos[i].streams[j].tx_port,
1775 				lcore_infos[i].streams[j].tx_queue);
1776 		}
1777 }
1778 
1779 static void
1780 init_port(void)
1781 {
1782 	int ret;
1783 	uint16_t std_queue;
1784 	uint16_t hairpin_queue;
1785 	uint16_t port_id;
1786 	uint16_t nr_ports;
1787 	uint16_t nr_queues;
1788 	struct rte_eth_hairpin_conf hairpin_conf = {
1789 		.peer_count = 1,
1790 	};
1791 	struct rte_eth_conf port_conf = {
1792 		.rx_adv_conf = {
1793 			.rss_conf.rss_hf =
1794 				GET_RSS_HF(),
1795 		}
1796 	};
1797 	struct rte_eth_txconf txq_conf;
1798 	struct rte_eth_rxconf rxq_conf;
1799 	struct rte_eth_dev_info dev_info;
1800 
1801 	nr_queues = rx_queues_count;
1802 	if (hairpin_queues_num != 0)
1803 		nr_queues = rx_queues_count + hairpin_queues_num;
1804 
1805 	nr_ports = rte_eth_dev_count_avail();
1806 	if (nr_ports == 0)
1807 		rte_exit(EXIT_FAILURE, "Error: no port detected\n");
1808 
1809 	mbuf_mp = rte_pktmbuf_pool_create("mbuf_pool",
1810 					total_mbuf_num, mbuf_cache_size,
1811 					0, mbuf_size,
1812 					rte_socket_id());
1813 	if (mbuf_mp == NULL)
1814 		rte_exit(EXIT_FAILURE, "Error: can't init mbuf pool\n");
1815 
1816 	for (port_id = 0; port_id < nr_ports; port_id++) {
1817 		uint64_t rx_metadata = 0;
1818 
1819 		rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
1820 		rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
1821 
1822 		ret = rte_eth_rx_metadata_negotiate(port_id, &rx_metadata);
1823 		if (ret == 0) {
1824 			if (!(rx_metadata & RTE_ETH_RX_METADATA_USER_FLAG)) {
1825 				printf(":: flow action FLAG will not affect Rx mbufs on port=%u\n",
1826 				       port_id);
1827 			}
1828 
1829 			if (!(rx_metadata & RTE_ETH_RX_METADATA_USER_MARK)) {
1830 				printf(":: flow action MARK will not affect Rx mbufs on port=%u\n",
1831 				       port_id);
1832 			}
1833 		} else if (ret != -ENOTSUP) {
1834 			rte_exit(EXIT_FAILURE, "Error when negotiating Rx meta features on port=%u: %s\n",
1835 				 port_id, rte_strerror(-ret));
1836 		}
1837 
1838 		ret = rte_eth_dev_info_get(port_id, &dev_info);
1839 		if (ret != 0)
1840 			rte_exit(EXIT_FAILURE,
1841 				"Error during getting device"
1842 				" (port %u) info: %s\n",
1843 				port_id, strerror(-ret));
1844 
1845 		port_conf.txmode.offloads &= dev_info.tx_offload_capa;
1846 		port_conf.rxmode.offloads &= dev_info.rx_offload_capa;
1847 
1848 		printf(":: initializing port: %d\n", port_id);
1849 
1850 		ret = rte_eth_dev_configure(port_id, nr_queues,
1851 				nr_queues, &port_conf);
1852 		if (ret < 0)
1853 			rte_exit(EXIT_FAILURE,
1854 				":: cannot configure device: err=%d, port=%u\n",
1855 				ret, port_id);
1856 
1857 		rxq_conf = dev_info.default_rxconf;
1858 		for (std_queue = 0; std_queue < rx_queues_count; std_queue++) {
1859 			ret = rte_eth_rx_queue_setup(port_id, std_queue, rxd_count,
1860 					rte_eth_dev_socket_id(port_id),
1861 					&rxq_conf,
1862 					mbuf_mp);
1863 			if (ret < 0)
1864 				rte_exit(EXIT_FAILURE,
1865 					":: Rx queue setup failed: err=%d, port=%u\n",
1866 					ret, port_id);
1867 		}
1868 
1869 		txq_conf = dev_info.default_txconf;
1870 		for (std_queue = 0; std_queue < tx_queues_count; std_queue++) {
1871 			ret = rte_eth_tx_queue_setup(port_id, std_queue, txd_count,
1872 					rte_eth_dev_socket_id(port_id),
1873 					&txq_conf);
1874 			if (ret < 0)
1875 				rte_exit(EXIT_FAILURE,
1876 					":: Tx queue setup failed: err=%d, port=%u\n",
1877 					ret, port_id);
1878 		}
1879 
1880 		/* Catch all packets from traffic generator. */
1881 		ret = rte_eth_promiscuous_enable(port_id);
1882 		if (ret != 0)
1883 			rte_exit(EXIT_FAILURE,
1884 				":: promiscuous mode enable failed: err=%s, port=%u\n",
1885 				rte_strerror(-ret), port_id);
1886 
1887 		if (hairpin_queues_num != 0) {
1888 			/*
1889 			 * Configure peer which represents hairpin Tx.
1890 			 * Hairpin queue numbers start after standard queues
1891 			 * (rx_queues_count and tx_queues_count).
1892 			 */
1893 			for (hairpin_queue = rx_queues_count, std_queue = 0;
1894 					hairpin_queue < nr_queues;
1895 					hairpin_queue++, std_queue++) {
1896 				hairpin_conf.peers[0].port = port_id;
1897 				hairpin_conf.peers[0].queue =
1898 					std_queue + tx_queues_count;
1899 				ret = rte_eth_rx_hairpin_queue_setup(
1900 						port_id, hairpin_queue,
1901 						rxd_count, &hairpin_conf);
1902 				if (ret != 0)
1903 					rte_exit(EXIT_FAILURE,
1904 						":: Hairpin rx queue setup failed: err=%d, port=%u\n",
1905 						ret, port_id);
1906 			}
1907 
1908 			for (hairpin_queue = tx_queues_count, std_queue = 0;
1909 					hairpin_queue < nr_queues;
1910 					hairpin_queue++, std_queue++) {
1911 				hairpin_conf.peers[0].port = port_id;
1912 				hairpin_conf.peers[0].queue =
1913 					std_queue + rx_queues_count;
1914 				ret = rte_eth_tx_hairpin_queue_setup(
1915 						port_id, hairpin_queue,
1916 						txd_count, &hairpin_conf);
1917 				if (ret != 0)
1918 					rte_exit(EXIT_FAILURE,
1919 						":: Hairpin tx queue setup failed: err=%d, port=%u\n",
1920 						ret, port_id);
1921 			}
1922 		}
1923 
1924 		ret = rte_eth_dev_start(port_id);
1925 		if (ret < 0)
1926 			rte_exit(EXIT_FAILURE,
1927 				"rte_eth_dev_start:err=%d, port=%u\n",
1928 				ret, port_id);
1929 
1930 		printf(":: initializing port: %d done\n", port_id);
1931 	}
1932 }
1933 
1934 int
1935 main(int argc, char **argv)
1936 {
1937 	int ret;
1938 	uint16_t port;
1939 	struct rte_flow_error error;
1940 
1941 	ret = rte_eal_init(argc, argv);
1942 	if (ret < 0)
1943 		rte_exit(EXIT_FAILURE, "EAL init failed\n");
1944 
1945 	force_quit = false;
1946 	dump_iterations = false;
1947 	rules_count = DEFAULT_RULES_COUNT;
1948 	rules_batch = DEFAULT_RULES_BATCH;
1949 	delete_flag = false;
1950 	dump_socket_mem_flag = false;
1951 	flow_group = DEFAULT_GROUP;
1952 	unique_data = false;
1953 
1954 	rx_queues_count = (uint8_t) RXQ_NUM;
1955 	tx_queues_count = (uint8_t) TXQ_NUM;
1956 	rxd_count = (uint8_t) NR_RXD;
1957 	txd_count = (uint8_t) NR_TXD;
1958 	mbuf_size = (uint32_t) MBUF_SIZE;
1959 	mbuf_cache_size = (uint32_t) MBUF_CACHE_SIZE;
1960 	total_mbuf_num = (uint32_t) TOTAL_MBUF_NUM;
1961 
1962 	signal(SIGINT, signal_handler);
1963 	signal(SIGTERM, signal_handler);
1964 
1965 	argc -= ret;
1966 	argv += ret;
1967 	if (argc > 1)
1968 		args_parse(argc, argv);
1969 
1970 	init_port();
1971 
1972 	nb_lcores = rte_lcore_count();
1973 	if (nb_lcores <= 1)
1974 		rte_exit(EXIT_FAILURE, "This app needs at least two cores\n");
1975 
1976 	printf(":: Flows Count per port: %d\n\n", rules_count);
1977 
1978 	if (has_meter())
1979 		create_meter_profile();
1980 	rte_eal_mp_remote_launch(run_rte_flow_handler_cores, NULL, CALL_MAIN);
1981 
1982 	if (enable_fwd) {
1983 		init_lcore_info();
1984 		rte_eal_mp_remote_launch(start_forwarding, NULL, CALL_MAIN);
1985 	}
1986 	if (has_meter() && delete_flag)
1987 		destroy_meter_profile();
1988 
1989 	RTE_ETH_FOREACH_DEV(port) {
1990 		rte_flow_flush(port, &error);
1991 		if (rte_eth_dev_stop(port) != 0)
1992 			printf("Failed to stop device on port %u\n", port);
1993 		rte_eth_dev_close(port);
1994 	}
1995 	printf("\nBye ...\n");
1996 	return 0;
1997 }
1998