xref: /dpdk/app/test/test_distributor.c (revision 68a03efeed657e6e05f281479b33b51102797e15)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include "test.h"
6 
7 #include <unistd.h>
8 #include <string.h>
9 #include <rte_cycles.h>
10 #include <rte_errno.h>
11 #include <rte_mempool.h>
12 #include <rte_mbuf.h>
13 #include <rte_mbuf_dyn.h>
14 #include <rte_distributor.h>
15 #include <rte_string_fns.h>
16 
17 #define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */
18 #define BURST 32
19 #define BIG_BATCH 1024
20 
21 typedef uint32_t seq_dynfield_t;
22 static int seq_dynfield_offset = -1;
23 
24 static inline seq_dynfield_t *
25 seq_field(struct rte_mbuf *mbuf)
26 {
27 	return RTE_MBUF_DYNFIELD(mbuf, seq_dynfield_offset, seq_dynfield_t *);
28 }
29 
30 struct worker_params {
31 	char name[64];
32 	struct rte_distributor *dist;
33 };
34 
35 struct worker_params worker_params;
36 
37 /* statics - all zero-initialized by default */
38 static volatile int quit;      /**< general quit variable for all threads */
39 static volatile int zero_quit; /**< var for when we just want thr0 to quit*/
40 static volatile int zero_sleep; /**< thr0 has quit basic loop and is sleeping*/
41 static volatile unsigned worker_idx;
42 static volatile unsigned zero_idx;
43 
44 struct worker_stats {
45 	volatile unsigned handled_packets;
46 } __rte_cache_aligned;
47 struct worker_stats worker_stats[RTE_MAX_LCORE];
48 
49 /* returns the total count of the number of packets handled by the worker
50  * functions given below.
51  */
52 static inline unsigned
53 total_packet_count(void)
54 {
55 	unsigned i, count = 0;
56 	for (i = 0; i < worker_idx; i++)
57 		count += __atomic_load_n(&worker_stats[i].handled_packets,
58 				__ATOMIC_RELAXED);
59 	return count;
60 }
61 
62 /* resets the packet counts for a new test */
63 static inline void
64 clear_packet_count(void)
65 {
66 	unsigned int i;
67 	for (i = 0; i < RTE_MAX_LCORE; i++)
68 		__atomic_store_n(&worker_stats[i].handled_packets, 0,
69 			__ATOMIC_RELAXED);
70 }
71 
72 /* this is the basic worker function for sanity test
73  * it does nothing but return packets and count them.
74  */
75 static int
76 handle_work(void *arg)
77 {
78 	struct rte_mbuf *buf[8] __rte_cache_aligned;
79 	struct worker_params *wp = arg;
80 	struct rte_distributor *db = wp->dist;
81 	unsigned int num;
82 	unsigned int id = __atomic_fetch_add(&worker_idx, 1, __ATOMIC_RELAXED);
83 
84 	num = rte_distributor_get_pkt(db, id, buf, NULL, 0);
85 	while (!quit) {
86 		__atomic_fetch_add(&worker_stats[id].handled_packets, num,
87 				__ATOMIC_RELAXED);
88 		num = rte_distributor_get_pkt(db, id,
89 				buf, buf, num);
90 	}
91 	__atomic_fetch_add(&worker_stats[id].handled_packets, num,
92 			__ATOMIC_RELAXED);
93 	rte_distributor_return_pkt(db, id, buf, num);
94 	return 0;
95 }
96 
97 /* do basic sanity testing of the distributor. This test tests the following:
98  * - send 32 packets through distributor with the same tag and ensure they
99  *   all go to the one worker
100  * - send 32 packets through the distributor with two different tags and
101  *   verify that they go equally to two different workers.
102  * - send 32 packets with different tags through the distributors and
103  *   just verify we get all packets back.
104  * - send 1024 packets through the distributor, gathering the returned packets
105  *   as we go. Then verify that we correctly got all 1024 pointers back again,
106  *   not necessarily in the same order (as different flows).
107  */
108 static int
109 sanity_test(struct worker_params *wp, struct rte_mempool *p)
110 {
111 	struct rte_distributor *db = wp->dist;
112 	struct rte_mbuf *bufs[BURST];
113 	struct rte_mbuf *returns[BURST*2];
114 	unsigned int i, count;
115 	unsigned int retries;
116 	unsigned int processed;
117 
118 	printf("=== Basic distributor sanity tests ===\n");
119 	clear_packet_count();
120 	if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
121 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
122 		return -1;
123 	}
124 
125 	/* now set all hash values in all buffers to zero, so all pkts go to the
126 	 * one worker thread */
127 	for (i = 0; i < BURST; i++)
128 		bufs[i]->hash.usr = 0;
129 
130 	processed = 0;
131 	while (processed < BURST)
132 		processed += rte_distributor_process(db, &bufs[processed],
133 			BURST - processed);
134 
135 	count = 0;
136 	do {
137 
138 		rte_distributor_flush(db);
139 		count += rte_distributor_returned_pkts(db,
140 				returns, BURST*2);
141 	} while (count < BURST);
142 
143 	if (total_packet_count() != BURST) {
144 		printf("Line %d: Error, not all packets flushed. "
145 				"Expected %u, got %u\n",
146 				__LINE__, BURST, total_packet_count());
147 		rte_mempool_put_bulk(p, (void *)bufs, BURST);
148 		return -1;
149 	}
150 
151 	for (i = 0; i < rte_lcore_count() - 1; i++)
152 		printf("Worker %u handled %u packets\n", i,
153 			__atomic_load_n(&worker_stats[i].handled_packets,
154 					__ATOMIC_RELAXED));
155 	printf("Sanity test with all zero hashes done.\n");
156 
157 	/* pick two flows and check they go correctly */
158 	if (rte_lcore_count() >= 3) {
159 		clear_packet_count();
160 		for (i = 0; i < BURST; i++)
161 			bufs[i]->hash.usr = (i & 1) << 8;
162 
163 		rte_distributor_process(db, bufs, BURST);
164 		count = 0;
165 		do {
166 			rte_distributor_flush(db);
167 			count += rte_distributor_returned_pkts(db,
168 					returns, BURST*2);
169 		} while (count < BURST);
170 		if (total_packet_count() != BURST) {
171 			printf("Line %d: Error, not all packets flushed. "
172 					"Expected %u, got %u\n",
173 					__LINE__, BURST, total_packet_count());
174 			rte_mempool_put_bulk(p, (void *)bufs, BURST);
175 			return -1;
176 		}
177 
178 		for (i = 0; i < rte_lcore_count() - 1; i++)
179 			printf("Worker %u handled %u packets\n", i,
180 				__atomic_load_n(
181 					&worker_stats[i].handled_packets,
182 					__ATOMIC_RELAXED));
183 		printf("Sanity test with two hash values done\n");
184 	}
185 
186 	/* give a different hash value to each packet,
187 	 * so load gets distributed */
188 	clear_packet_count();
189 	for (i = 0; i < BURST; i++)
190 		bufs[i]->hash.usr = i+1;
191 
192 	rte_distributor_process(db, bufs, BURST);
193 	count = 0;
194 	do {
195 		rte_distributor_flush(db);
196 		count += rte_distributor_returned_pkts(db,
197 				returns, BURST*2);
198 	} while (count < BURST);
199 	if (total_packet_count() != BURST) {
200 		printf("Line %d: Error, not all packets flushed. "
201 				"Expected %u, got %u\n",
202 				__LINE__, BURST, total_packet_count());
203 		rte_mempool_put_bulk(p, (void *)bufs, BURST);
204 		return -1;
205 	}
206 
207 	for (i = 0; i < rte_lcore_count() - 1; i++)
208 		printf("Worker %u handled %u packets\n", i,
209 			__atomic_load_n(&worker_stats[i].handled_packets,
210 					__ATOMIC_RELAXED));
211 	printf("Sanity test with non-zero hashes done\n");
212 
213 	rte_mempool_put_bulk(p, (void *)bufs, BURST);
214 
215 	/* sanity test with BIG_BATCH packets to ensure they all arrived back
216 	 * from the returned packets function */
217 	clear_packet_count();
218 	struct rte_mbuf *many_bufs[BIG_BATCH], *return_bufs[BIG_BATCH];
219 	unsigned num_returned = 0;
220 	unsigned int num_being_processed = 0;
221 	unsigned int return_buffer_capacity = 127;/* RTE_DISTRIB_RETURNS_MASK */
222 
223 	/* flush out any remaining packets */
224 	rte_distributor_flush(db);
225 	rte_distributor_clear_returns(db);
226 
227 	if (rte_mempool_get_bulk(p, (void *)many_bufs, BIG_BATCH) != 0) {
228 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
229 		return -1;
230 	}
231 	for (i = 0; i < BIG_BATCH; i++)
232 		many_bufs[i]->hash.usr = i << 2;
233 
234 	printf("=== testing big burst (%s) ===\n", wp->name);
235 	for (i = 0; i < BIG_BATCH/BURST; i++) {
236 		rte_distributor_process(db,
237 				&many_bufs[i*BURST], BURST);
238 		num_being_processed += BURST;
239 		do {
240 			count = rte_distributor_returned_pkts(db,
241 					&return_bufs[num_returned],
242 					BIG_BATCH - num_returned);
243 			num_being_processed -= count;
244 			num_returned += count;
245 			rte_distributor_flush(db);
246 		} while (num_being_processed + BURST > return_buffer_capacity);
247 	}
248 	retries = 0;
249 	do {
250 		rte_distributor_flush(db);
251 		count = rte_distributor_returned_pkts(db,
252 				&return_bufs[num_returned],
253 				BIG_BATCH - num_returned);
254 		num_returned += count;
255 		retries++;
256 	} while ((num_returned < BIG_BATCH) && (retries < 100));
257 
258 	if (num_returned != BIG_BATCH) {
259 		printf("line %d: Missing packets, expected %d\n",
260 				__LINE__, num_returned);
261 		rte_mempool_put_bulk(p, (void *)many_bufs, BIG_BATCH);
262 		return -1;
263 	}
264 
265 	/* big check -  make sure all packets made it back!! */
266 	for (i = 0; i < BIG_BATCH; i++) {
267 		unsigned j;
268 		struct rte_mbuf *src = many_bufs[i];
269 		for (j = 0; j < BIG_BATCH; j++) {
270 			if (return_bufs[j] == src)
271 				break;
272 		}
273 
274 		if (j == BIG_BATCH) {
275 			printf("Error: could not find source packet #%u\n", i);
276 			rte_mempool_put_bulk(p, (void *)many_bufs, BIG_BATCH);
277 			return -1;
278 		}
279 	}
280 	printf("Sanity test of returned packets done\n");
281 
282 	rte_mempool_put_bulk(p, (void *)many_bufs, BIG_BATCH);
283 
284 	printf("\n");
285 	return 0;
286 }
287 
288 
289 /* to test that the distributor does not lose packets, we use this worker
290  * function which frees mbufs when it gets them. The distributor thread does
291  * the mbuf allocation. If distributor drops packets we'll eventually run out
292  * of mbufs.
293  */
294 static int
295 handle_work_with_free_mbufs(void *arg)
296 {
297 	struct rte_mbuf *buf[8] __rte_cache_aligned;
298 	struct worker_params *wp = arg;
299 	struct rte_distributor *d = wp->dist;
300 	unsigned int i;
301 	unsigned int num;
302 	unsigned int id = __atomic_fetch_add(&worker_idx, 1, __ATOMIC_RELAXED);
303 
304 	num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
305 	while (!quit) {
306 		__atomic_fetch_add(&worker_stats[id].handled_packets, num,
307 				__ATOMIC_RELAXED);
308 		for (i = 0; i < num; i++)
309 			rte_pktmbuf_free(buf[i]);
310 		num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
311 	}
312 	__atomic_fetch_add(&worker_stats[id].handled_packets, num,
313 			__ATOMIC_RELAXED);
314 	rte_distributor_return_pkt(d, id, buf, num);
315 	return 0;
316 }
317 
318 /* Perform a sanity test of the distributor with a large number of packets,
319  * where we allocate a new set of mbufs for each burst. The workers then
320  * free the mbufs. This ensures that we don't have any packet leaks in the
321  * library.
322  */
323 static int
324 sanity_test_with_mbuf_alloc(struct worker_params *wp, struct rte_mempool *p)
325 {
326 	struct rte_distributor *d = wp->dist;
327 	unsigned i;
328 	struct rte_mbuf *bufs[BURST];
329 	unsigned int processed;
330 
331 	printf("=== Sanity test with mbuf alloc/free (%s) ===\n", wp->name);
332 
333 	clear_packet_count();
334 	for (i = 0; i < ((1<<ITER_POWER)); i += BURST) {
335 		unsigned j;
336 		while (rte_mempool_get_bulk(p, (void *)bufs, BURST) < 0)
337 			rte_distributor_process(d, NULL, 0);
338 		for (j = 0; j < BURST; j++) {
339 			bufs[j]->hash.usr = (i+j) << 1;
340 		}
341 
342 		processed = 0;
343 		while (processed < BURST)
344 			processed += rte_distributor_process(d,
345 				&bufs[processed], BURST - processed);
346 	}
347 
348 	rte_distributor_flush(d);
349 
350 	rte_delay_us(10000);
351 
352 	if (total_packet_count() < (1<<ITER_POWER)) {
353 		printf("Line %u: Packet count is incorrect, %u, expected %u\n",
354 				__LINE__, total_packet_count(),
355 				(1<<ITER_POWER));
356 		return -1;
357 	}
358 
359 	printf("Sanity test with mbuf alloc/free passed\n\n");
360 	return 0;
361 }
362 
363 static int
364 handle_work_for_shutdown_test(void *arg)
365 {
366 	struct rte_mbuf *buf[8] __rte_cache_aligned;
367 	struct worker_params *wp = arg;
368 	struct rte_distributor *d = wp->dist;
369 	unsigned int num;
370 	unsigned int zero_id = 0;
371 	unsigned int zero_unset;
372 	const unsigned int id = __atomic_fetch_add(&worker_idx, 1,
373 			__ATOMIC_RELAXED);
374 
375 	num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
376 
377 	if (num > 0) {
378 		zero_unset = RTE_MAX_LCORE;
379 		__atomic_compare_exchange_n(&zero_idx, &zero_unset, id,
380 			false, __ATOMIC_ACQ_REL, __ATOMIC_ACQUIRE);
381 	}
382 	zero_id = __atomic_load_n(&zero_idx, __ATOMIC_ACQUIRE);
383 
384 	/* wait for quit single globally, or for worker zero, wait
385 	 * for zero_quit */
386 	while (!quit && !(id == zero_id && zero_quit)) {
387 		__atomic_fetch_add(&worker_stats[id].handled_packets, num,
388 				__ATOMIC_RELAXED);
389 		num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
390 
391 		if (num > 0) {
392 			zero_unset = RTE_MAX_LCORE;
393 			__atomic_compare_exchange_n(&zero_idx, &zero_unset, id,
394 				false, __ATOMIC_ACQ_REL, __ATOMIC_ACQUIRE);
395 		}
396 		zero_id = __atomic_load_n(&zero_idx, __ATOMIC_ACQUIRE);
397 	}
398 
399 	__atomic_fetch_add(&worker_stats[id].handled_packets, num,
400 			__ATOMIC_RELAXED);
401 	if (id == zero_id) {
402 		rte_distributor_return_pkt(d, id, NULL, 0);
403 
404 		/* for worker zero, allow it to restart to pick up last packet
405 		 * when all workers are shutting down.
406 		 */
407 		__atomic_store_n(&zero_sleep, 1, __ATOMIC_RELEASE);
408 		while (zero_quit)
409 			usleep(100);
410 		__atomic_store_n(&zero_sleep, 0, __ATOMIC_RELEASE);
411 
412 		num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
413 
414 		while (!quit) {
415 			__atomic_fetch_add(&worker_stats[id].handled_packets,
416 					num, __ATOMIC_RELAXED);
417 			num = rte_distributor_get_pkt(d, id, buf, NULL, 0);
418 		}
419 	}
420 	rte_distributor_return_pkt(d, id, buf, num);
421 	return 0;
422 }
423 
424 
425 /* Perform a sanity test of the distributor with a large number of packets,
426  * where we allocate a new set of mbufs for each burst. The workers then
427  * free the mbufs. This ensures that we don't have any packet leaks in the
428  * library.
429  */
430 static int
431 sanity_test_with_worker_shutdown(struct worker_params *wp,
432 		struct rte_mempool *p)
433 {
434 	struct rte_distributor *d = wp->dist;
435 	struct rte_mbuf *bufs[BURST];
436 	struct rte_mbuf *bufs2[BURST];
437 	unsigned int i;
438 	unsigned int failed = 0;
439 	unsigned int processed = 0;
440 
441 	printf("=== Sanity test of worker shutdown ===\n");
442 
443 	clear_packet_count();
444 
445 	if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
446 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
447 		return -1;
448 	}
449 
450 	/*
451 	 * Now set all hash values in all buffers to same value so all
452 	 * pkts go to the one worker thread
453 	 */
454 	for (i = 0; i < BURST; i++)
455 		bufs[i]->hash.usr = 1;
456 
457 	processed = 0;
458 	while (processed < BURST)
459 		processed += rte_distributor_process(d, &bufs[processed],
460 			BURST - processed);
461 	rte_distributor_flush(d);
462 
463 	/* at this point, we will have processed some packets and have a full
464 	 * backlog for the other ones at worker 0.
465 	 */
466 
467 	/* get more buffers to queue up, again setting them to the same flow */
468 	if (rte_mempool_get_bulk(p, (void *)bufs2, BURST) != 0) {
469 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
470 		rte_mempool_put_bulk(p, (void *)bufs, BURST);
471 		return -1;
472 	}
473 	for (i = 0; i < BURST; i++)
474 		bufs2[i]->hash.usr = 1;
475 
476 	/* get worker zero to quit */
477 	zero_quit = 1;
478 	rte_distributor_process(d, bufs2, BURST);
479 
480 	/* flush the distributor */
481 	rte_distributor_flush(d);
482 	while (!__atomic_load_n(&zero_sleep, __ATOMIC_ACQUIRE))
483 		rte_distributor_flush(d);
484 
485 	zero_quit = 0;
486 	while (__atomic_load_n(&zero_sleep, __ATOMIC_ACQUIRE))
487 		rte_delay_us(100);
488 
489 	for (i = 0; i < rte_lcore_count() - 1; i++)
490 		printf("Worker %u handled %u packets\n", i,
491 			__atomic_load_n(&worker_stats[i].handled_packets,
492 					__ATOMIC_RELAXED));
493 
494 	if (total_packet_count() != BURST * 2) {
495 		printf("Line %d: Error, not all packets flushed. "
496 				"Expected %u, got %u\n",
497 				__LINE__, BURST * 2, total_packet_count());
498 		failed = 1;
499 	}
500 
501 	rte_mempool_put_bulk(p, (void *)bufs, BURST);
502 	rte_mempool_put_bulk(p, (void *)bufs2, BURST);
503 
504 	if (failed)
505 		return -1;
506 
507 	printf("Sanity test with worker shutdown passed\n\n");
508 	return 0;
509 }
510 
511 /* Test that the flush function is able to move packets between workers when
512  * one worker shuts down..
513  */
514 static int
515 test_flush_with_worker_shutdown(struct worker_params *wp,
516 		struct rte_mempool *p)
517 {
518 	struct rte_distributor *d = wp->dist;
519 	struct rte_mbuf *bufs[BURST];
520 	unsigned int i;
521 	unsigned int failed = 0;
522 	unsigned int processed;
523 
524 	printf("=== Test flush fn with worker shutdown (%s) ===\n", wp->name);
525 
526 	clear_packet_count();
527 	if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
528 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
529 		return -1;
530 	}
531 
532 	/* now set all hash values in all buffers to zero, so all pkts go to the
533 	 * one worker thread */
534 	for (i = 0; i < BURST; i++)
535 		bufs[i]->hash.usr = 0;
536 
537 	processed = 0;
538 	while (processed < BURST)
539 		processed += rte_distributor_process(d, &bufs[processed],
540 			BURST - processed);
541 	/* at this point, we will have processed some packets and have a full
542 	 * backlog for the other ones at worker 0.
543 	 */
544 
545 	/* get worker zero to quit */
546 	zero_quit = 1;
547 
548 	/* flush the distributor */
549 	rte_distributor_flush(d);
550 
551 	while (!__atomic_load_n(&zero_sleep, __ATOMIC_ACQUIRE))
552 		rte_distributor_flush(d);
553 
554 	zero_quit = 0;
555 
556 	while (__atomic_load_n(&zero_sleep, __ATOMIC_ACQUIRE))
557 		rte_delay_us(100);
558 
559 	for (i = 0; i < rte_lcore_count() - 1; i++)
560 		printf("Worker %u handled %u packets\n", i,
561 			__atomic_load_n(&worker_stats[i].handled_packets,
562 					__ATOMIC_RELAXED));
563 
564 	if (total_packet_count() != BURST) {
565 		printf("Line %d: Error, not all packets flushed. "
566 				"Expected %u, got %u\n",
567 				__LINE__, BURST, total_packet_count());
568 		failed = 1;
569 	}
570 
571 	rte_mempool_put_bulk(p, (void *)bufs, BURST);
572 
573 	if (failed)
574 		return -1;
575 
576 	printf("Flush test with worker shutdown passed\n\n");
577 	return 0;
578 }
579 
580 static int
581 handle_and_mark_work(void *arg)
582 {
583 	struct rte_mbuf *buf[8] __rte_cache_aligned;
584 	struct worker_params *wp = arg;
585 	struct rte_distributor *db = wp->dist;
586 	unsigned int num, i;
587 	unsigned int id = __atomic_fetch_add(&worker_idx, 1, __ATOMIC_RELAXED);
588 	num = rte_distributor_get_pkt(db, id, buf, NULL, 0);
589 	while (!quit) {
590 		__atomic_fetch_add(&worker_stats[id].handled_packets, num,
591 				__ATOMIC_RELAXED);
592 		for (i = 0; i < num; i++)
593 			*seq_field(buf[i]) += id + 1;
594 		num = rte_distributor_get_pkt(db, id,
595 				buf, buf, num);
596 	}
597 	__atomic_fetch_add(&worker_stats[id].handled_packets, num,
598 			__ATOMIC_RELAXED);
599 	rte_distributor_return_pkt(db, id, buf, num);
600 	return 0;
601 }
602 
603 /* sanity_mark_test sends packets to workers which mark them.
604  * Every packet has also encoded sequence number.
605  * The returned packets are sorted and verified if they were handled
606  * by proper workers.
607  */
608 static int
609 sanity_mark_test(struct worker_params *wp, struct rte_mempool *p)
610 {
611 	const unsigned int buf_count = 24;
612 	const unsigned int burst = 8;
613 	const unsigned int shift = 12;
614 	const unsigned int seq_shift = 10;
615 
616 	struct rte_distributor *db = wp->dist;
617 	struct rte_mbuf *bufs[buf_count];
618 	struct rte_mbuf *returns[buf_count];
619 	unsigned int i, count, id;
620 	unsigned int sorted[buf_count], seq;
621 	unsigned int failed = 0;
622 	unsigned int processed;
623 
624 	printf("=== Marked packets test ===\n");
625 	clear_packet_count();
626 	if (rte_mempool_get_bulk(p, (void *)bufs, buf_count) != 0) {
627 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
628 		return -1;
629 	}
630 
631 	/* bufs' hashes will be like these below, but shifted left.
632 	 * The shifting is for avoiding collisions with backlogs
633 	 * and in-flight tags left by previous tests.
634 	 * [1, 1, 1, 1, 1, 1, 1, 1
635 	 *  1, 1, 1, 1, 2, 2, 2, 2
636 	 *  2, 2, 2, 2, 1, 1, 1, 1]
637 	 */
638 	for (i = 0; i < burst; i++) {
639 		bufs[0 * burst + i]->hash.usr = 1 << shift;
640 		bufs[1 * burst + i]->hash.usr = ((i < burst / 2) ? 1 : 2)
641 			<< shift;
642 		bufs[2 * burst + i]->hash.usr = ((i < burst / 2) ? 2 : 1)
643 			<< shift;
644 	}
645 	/* Assign a sequence number to each packet. The sequence is shifted,
646 	 * so that lower bits will hold mark from worker.
647 	 */
648 	for (i = 0; i < buf_count; i++)
649 		*seq_field(bufs[i]) = i << seq_shift;
650 
651 	count = 0;
652 	for (i = 0; i < buf_count/burst; i++) {
653 		processed = 0;
654 		while (processed < burst)
655 			processed += rte_distributor_process(db,
656 				&bufs[i * burst + processed],
657 				burst - processed);
658 		count += rte_distributor_returned_pkts(db, &returns[count],
659 			buf_count - count);
660 	}
661 
662 	do {
663 		rte_distributor_flush(db);
664 		count += rte_distributor_returned_pkts(db, &returns[count],
665 			buf_count - count);
666 	} while (count < buf_count);
667 
668 	for (i = 0; i < rte_lcore_count() - 1; i++)
669 		printf("Worker %u handled %u packets\n", i,
670 			__atomic_load_n(&worker_stats[i].handled_packets,
671 					__ATOMIC_RELAXED));
672 
673 	/* Sort returned packets by sent order (sequence numbers). */
674 	for (i = 0; i < buf_count; i++) {
675 		seq = *seq_field(returns[i]) >> seq_shift;
676 		id = *seq_field(returns[i]) - (seq << seq_shift);
677 		sorted[seq] = id;
678 	}
679 
680 	/* Verify that packets [0-11] and [20-23] were processed
681 	 * by the same worker
682 	 */
683 	for (i = 1; i < 12; i++) {
684 		if (sorted[i] != sorted[0]) {
685 			printf("Packet number %u processed by worker %u,"
686 				" but should be processes by worker %u\n",
687 				i, sorted[i], sorted[0]);
688 			failed = 1;
689 		}
690 	}
691 	for (i = 20; i < 24; i++) {
692 		if (sorted[i] != sorted[0]) {
693 			printf("Packet number %u processed by worker %u,"
694 				" but should be processes by worker %u\n",
695 				i, sorted[i], sorted[0]);
696 			failed = 1;
697 		}
698 	}
699 	/* And verify that packets [12-19] were processed
700 	 * by the another worker
701 	 */
702 	for (i = 13; i < 20; i++) {
703 		if (sorted[i] != sorted[12]) {
704 			printf("Packet number %u processed by worker %u,"
705 				" but should be processes by worker %u\n",
706 				i, sorted[i], sorted[12]);
707 			failed = 1;
708 		}
709 	}
710 
711 	rte_mempool_put_bulk(p, (void *)bufs, buf_count);
712 
713 	if (failed)
714 		return -1;
715 
716 	printf("Marked packets test passed\n");
717 	return 0;
718 }
719 
720 static
721 int test_error_distributor_create_name(void)
722 {
723 	struct rte_distributor *d = NULL;
724 	struct rte_distributor *db = NULL;
725 	char *name = NULL;
726 
727 	d = rte_distributor_create(name, rte_socket_id(),
728 			rte_lcore_count() - 1,
729 			RTE_DIST_ALG_SINGLE);
730 	if (d != NULL || rte_errno != EINVAL) {
731 		printf("ERROR: No error on create() with NULL name param\n");
732 		return -1;
733 	}
734 
735 	db = rte_distributor_create(name, rte_socket_id(),
736 			rte_lcore_count() - 1,
737 			RTE_DIST_ALG_BURST);
738 	if (db != NULL || rte_errno != EINVAL) {
739 		printf("ERROR: No error on create() with NULL param\n");
740 		return -1;
741 	}
742 
743 	return 0;
744 }
745 
746 
747 static
748 int test_error_distributor_create_numworkers(void)
749 {
750 	struct rte_distributor *ds = NULL;
751 	struct rte_distributor *db = NULL;
752 
753 	ds = rte_distributor_create("test_numworkers", rte_socket_id(),
754 			RTE_MAX_LCORE + 10,
755 			RTE_DIST_ALG_SINGLE);
756 	if (ds != NULL || rte_errno != EINVAL) {
757 		printf("ERROR: No error on create() with num_workers > MAX\n");
758 		return -1;
759 	}
760 
761 	db = rte_distributor_create("test_numworkers", rte_socket_id(),
762 			RTE_MAX_LCORE + 10,
763 			RTE_DIST_ALG_BURST);
764 	if (db != NULL || rte_errno != EINVAL) {
765 		printf("ERROR: No error on create() num_workers > MAX\n");
766 		return -1;
767 	}
768 
769 	return 0;
770 }
771 
772 
773 /* Useful function which ensures that all worker functions terminate */
774 static void
775 quit_workers(struct worker_params *wp, struct rte_mempool *p)
776 {
777 	struct rte_distributor *d = wp->dist;
778 	const unsigned num_workers = rte_lcore_count() - 1;
779 	unsigned i;
780 	struct rte_mbuf *bufs[RTE_MAX_LCORE];
781 	struct rte_mbuf *returns[RTE_MAX_LCORE];
782 	if (rte_mempool_get_bulk(p, (void *)bufs, num_workers) != 0) {
783 		printf("line %d: Error getting mbufs from pool\n", __LINE__);
784 		return;
785 	}
786 
787 	zero_quit = 0;
788 	quit = 1;
789 	for (i = 0; i < num_workers; i++) {
790 		bufs[i]->hash.usr = i << 1;
791 		rte_distributor_process(d, &bufs[i], 1);
792 	}
793 
794 	rte_distributor_process(d, NULL, 0);
795 	rte_distributor_flush(d);
796 	rte_eal_mp_wait_lcore();
797 
798 	while (rte_distributor_returned_pkts(d, returns, RTE_MAX_LCORE))
799 		;
800 
801 	rte_distributor_clear_returns(d);
802 	rte_mempool_put_bulk(p, (void *)bufs, num_workers);
803 
804 	quit = 0;
805 	worker_idx = 0;
806 	zero_idx = RTE_MAX_LCORE;
807 	zero_quit = 0;
808 	zero_sleep = 0;
809 }
810 
811 static int
812 test_distributor(void)
813 {
814 	static struct rte_distributor *ds;
815 	static struct rte_distributor *db;
816 	static struct rte_distributor *dist[2];
817 	static struct rte_mempool *p;
818 	int i;
819 
820 	static const struct rte_mbuf_dynfield seq_dynfield_desc = {
821 		.name = "test_distributor_dynfield_seq",
822 		.size = sizeof(seq_dynfield_t),
823 		.align = __alignof__(seq_dynfield_t),
824 	};
825 	seq_dynfield_offset =
826 		rte_mbuf_dynfield_register(&seq_dynfield_desc);
827 	if (seq_dynfield_offset < 0) {
828 		printf("Error registering mbuf field\n");
829 		return TEST_FAILED;
830 	}
831 
832 	if (rte_lcore_count() < 2) {
833 		printf("Not enough cores for distributor_autotest, expecting at least 2\n");
834 		return TEST_SKIPPED;
835 	}
836 
837 	if (db == NULL) {
838 		db = rte_distributor_create("Test_dist_burst", rte_socket_id(),
839 				rte_lcore_count() - 1,
840 				RTE_DIST_ALG_BURST);
841 		if (db == NULL) {
842 			printf("Error creating burst distributor\n");
843 			return -1;
844 		}
845 	} else {
846 		rte_distributor_flush(db);
847 		rte_distributor_clear_returns(db);
848 	}
849 
850 	if (ds == NULL) {
851 		ds = rte_distributor_create("Test_dist_single",
852 				rte_socket_id(),
853 				rte_lcore_count() - 1,
854 			RTE_DIST_ALG_SINGLE);
855 		if (ds == NULL) {
856 			printf("Error creating single distributor\n");
857 			return -1;
858 		}
859 	} else {
860 		rte_distributor_flush(ds);
861 		rte_distributor_clear_returns(ds);
862 	}
863 
864 	const unsigned nb_bufs = (511 * rte_lcore_count()) < BIG_BATCH ?
865 			(BIG_BATCH * 2) - 1 : (511 * rte_lcore_count());
866 	if (p == NULL) {
867 		p = rte_pktmbuf_pool_create("DT_MBUF_POOL", nb_bufs, BURST,
868 			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
869 		if (p == NULL) {
870 			printf("Error creating mempool\n");
871 			return -1;
872 		}
873 	}
874 
875 	dist[0] = ds;
876 	dist[1] = db;
877 
878 	for (i = 0; i < 2; i++) {
879 
880 		worker_params.dist = dist[i];
881 		if (i)
882 			strlcpy(worker_params.name, "burst",
883 					sizeof(worker_params.name));
884 		else
885 			strlcpy(worker_params.name, "single",
886 					sizeof(worker_params.name));
887 
888 		rte_eal_mp_remote_launch(handle_work,
889 				&worker_params, SKIP_MAIN);
890 		if (sanity_test(&worker_params, p) < 0)
891 			goto err;
892 		quit_workers(&worker_params, p);
893 
894 		rte_eal_mp_remote_launch(handle_work_with_free_mbufs,
895 				&worker_params, SKIP_MAIN);
896 		if (sanity_test_with_mbuf_alloc(&worker_params, p) < 0)
897 			goto err;
898 		quit_workers(&worker_params, p);
899 
900 		if (rte_lcore_count() > 2) {
901 			rte_eal_mp_remote_launch(handle_work_for_shutdown_test,
902 					&worker_params,
903 					SKIP_MAIN);
904 			if (sanity_test_with_worker_shutdown(&worker_params,
905 					p) < 0)
906 				goto err;
907 			quit_workers(&worker_params, p);
908 
909 			rte_eal_mp_remote_launch(handle_work_for_shutdown_test,
910 					&worker_params,
911 					SKIP_MAIN);
912 			if (test_flush_with_worker_shutdown(&worker_params,
913 					p) < 0)
914 				goto err;
915 			quit_workers(&worker_params, p);
916 
917 			rte_eal_mp_remote_launch(handle_and_mark_work,
918 					&worker_params, SKIP_MAIN);
919 			if (sanity_mark_test(&worker_params, p) < 0)
920 				goto err;
921 			quit_workers(&worker_params, p);
922 
923 		} else {
924 			printf("Too few cores to run worker shutdown test\n");
925 		}
926 
927 	}
928 
929 	if (test_error_distributor_create_numworkers() == -1 ||
930 			test_error_distributor_create_name() == -1) {
931 		printf("rte_distributor_create parameter check tests failed");
932 		return -1;
933 	}
934 
935 	return 0;
936 
937 err:
938 	quit_workers(&worker_params, p);
939 	return -1;
940 }
941 
942 REGISTER_TEST_COMMAND(distributor_autotest, test_distributor);
943