xref: /dpdk/app/test-eventdev/test_perf_common.c (revision efb1a06bb3f8dbcce5e43b49d23d73aaf80b2c8f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Cavium, Inc
3  */
4 
5 #include <math.h>
6 
7 #include "test_perf_common.h"
8 
9 #define NB_CRYPTODEV_DESCRIPTORS 128
10 #define DATA_SIZE		512
11 struct modex_test_data {
12 	enum rte_crypto_asym_xform_type xform_type;
13 	struct {
14 		uint8_t data[DATA_SIZE];
15 		uint16_t len;
16 	} base;
17 	struct {
18 		uint8_t data[DATA_SIZE];
19 		uint16_t len;
20 	} exponent;
21 	struct {
22 		uint8_t data[DATA_SIZE];
23 		uint16_t len;
24 	} modulus;
25 	struct {
26 		uint8_t data[DATA_SIZE];
27 		uint16_t len;
28 	} reminder;
29 	uint16_t result_len;
30 };
31 
32 static struct
33 modex_test_data modex_test_case = {
34 	.xform_type = RTE_CRYPTO_ASYM_XFORM_MODEX,
35 	.base = {
36 		.data = {
37 			0xF8, 0xBA, 0x1A, 0x55, 0xD0, 0x2F, 0x85,
38 			0xAE, 0x96, 0x7B, 0xB6, 0x2F, 0xB6, 0xCD,
39 			0xA8, 0xEB, 0x7E, 0x78, 0xA0, 0x50
40 		},
41 		.len = 20,
42 	},
43 	.exponent = {
44 		.data = {
45 			0x01, 0x00, 0x01
46 		},
47 		.len = 3,
48 	},
49 	.reminder = {
50 		.data = {
51 			0x2C, 0x60, 0x75, 0x45, 0x98, 0x9D, 0xE0, 0x72,
52 			0xA0, 0x9D, 0x3A, 0x9E, 0x03, 0x38, 0x73, 0x3C,
53 			0x31, 0x83, 0x04, 0xFE, 0x75, 0x43, 0xE6, 0x17,
54 			0x5C, 0x01, 0x29, 0x51, 0x69, 0x33, 0x62, 0x2D,
55 			0x78, 0xBE, 0xAE, 0xC4, 0xBC, 0xDE, 0x7E, 0x2C,
56 			0x77, 0x84, 0xF2, 0xC5, 0x14, 0xB5, 0x2F, 0xF7,
57 			0xC5, 0x94, 0xEF, 0x86, 0x75, 0x75, 0xB5, 0x11,
58 			0xE5, 0x0E, 0x0A, 0x29, 0x76, 0xE2, 0xEA, 0x32,
59 			0x0E, 0x43, 0x77, 0x7E, 0x2C, 0x27, 0xAC, 0x3B,
60 			0x86, 0xA5, 0xDB, 0xC9, 0x48, 0x40, 0xE8, 0x99,
61 			0x9A, 0x0A, 0x3D, 0xD6, 0x74, 0xFA, 0x2E, 0x2E,
62 			0x5B, 0xAF, 0x8C, 0x99, 0x44, 0x2A, 0x67, 0x38,
63 			0x27, 0x41, 0x59, 0x9D, 0xB8, 0x51, 0xC9, 0xF7,
64 			0x43, 0x61, 0x31, 0x6E, 0xF1, 0x25, 0x38, 0x7F,
65 			0xAE, 0xC6, 0xD0, 0xBB, 0x29, 0x76, 0x3F, 0x46,
66 			0x2E, 0x1B, 0xE4, 0x67, 0x71, 0xE3, 0x87, 0x5A
67 		},
68 		.len = 128,
69 	},
70 	.modulus = {
71 		.data = {
72 			0xb3, 0xa1, 0xaf, 0xb7, 0x13, 0x08, 0x00, 0x0a,
73 			0x35, 0xdc, 0x2b, 0x20, 0x8d, 0xa1, 0xb5, 0xce,
74 			0x47, 0x8a, 0xc3, 0x80, 0xf4, 0x7d, 0x4a, 0xa2,
75 			0x62, 0xfd, 0x61, 0x7f, 0xb5, 0xa8, 0xde, 0x0a,
76 			0x17, 0x97, 0xa0, 0xbf, 0xdf, 0x56, 0x5a, 0x3d,
77 			0x51, 0x56, 0x4f, 0x70, 0x70, 0x3f, 0x63, 0x6a,
78 			0x44, 0x5b, 0xad, 0x84, 0x0d, 0x3f, 0x27, 0x6e,
79 			0x3b, 0x34, 0x91, 0x60, 0x14, 0xb9, 0xaa, 0x72,
80 			0xfd, 0xa3, 0x64, 0xd2, 0x03, 0xa7, 0x53, 0x87,
81 			0x9e, 0x88, 0x0b, 0xc1, 0x14, 0x93, 0x1a, 0x62,
82 			0xff, 0xb1, 0x5d, 0x74, 0xcd, 0x59, 0x63, 0x18,
83 			0x11, 0x3d, 0x4f, 0xba, 0x75, 0xd4, 0x33, 0x4e,
84 			0x23, 0x6b, 0x7b, 0x57, 0x44, 0xe1, 0xd3, 0x03,
85 			0x13, 0xa6, 0xf0, 0x8b, 0x60, 0xb0, 0x9e, 0xee,
86 			0x75, 0x08, 0x9d, 0x71, 0x63, 0x13, 0xcb, 0xa6,
87 			0x81, 0x92, 0x14, 0x03, 0x22, 0x2d, 0xde, 0x55
88 		},
89 		.len = 128,
90 	},
91 	.result_len = 128,
92 };
93 
94 int
95 perf_test_result(struct evt_test *test, struct evt_options *opt)
96 {
97 	RTE_SET_USED(opt);
98 	int i;
99 	uint64_t total = 0;
100 	struct test_perf *t = evt_test_priv(test);
101 
102 	printf("Packet distribution across worker cores :\n");
103 	for (i = 0; i < t->nb_workers; i++)
104 		total += t->worker[i].processed_pkts;
105 	for (i = 0; i < t->nb_workers; i++)
106 		printf("Worker %d packets: "CLGRN"%"PRIx64" "CLNRM"percentage:"
107 				CLGRN" %3.2f"CLNRM"\n", i,
108 				t->worker[i].processed_pkts,
109 				(((double)t->worker[i].processed_pkts)/total)
110 				* 100);
111 
112 	return t->result;
113 }
114 
115 static inline int
116 perf_producer(void *arg)
117 {
118 	int i;
119 	struct prod_data *p  = arg;
120 	struct test_perf *t = p->t;
121 	struct evt_options *opt = t->opt;
122 	const uint8_t dev_id = p->dev_id;
123 	const uint8_t port = p->port_id;
124 	struct rte_mempool *pool = t->pool;
125 	const uint64_t nb_pkts = t->nb_pkts;
126 	const uint32_t nb_flows = t->nb_flows;
127 	uint32_t flow_counter = 0;
128 	uint64_t count = 0;
129 	struct perf_elt *m[BURST_SIZE + 1] = {NULL};
130 	struct rte_event ev;
131 
132 	if (opt->verbose_level > 1)
133 		printf("%s(): lcore %d dev_id %d port=%d queue %d\n", __func__,
134 				rte_lcore_id(), dev_id, port, p->queue_id);
135 
136 	ev.event = 0;
137 	ev.op = RTE_EVENT_OP_NEW;
138 	ev.queue_id = p->queue_id;
139 	ev.sched_type = t->opt->sched_type_list[0];
140 	ev.priority = RTE_EVENT_DEV_PRIORITY_NORMAL;
141 	ev.event_type =  RTE_EVENT_TYPE_CPU;
142 	ev.sub_event_type = 0; /* stage 0 */
143 
144 	while (count < nb_pkts && t->done == false) {
145 		if (rte_mempool_get_bulk(pool, (void **)m, BURST_SIZE) < 0)
146 			continue;
147 		for (i = 0; i < BURST_SIZE; i++) {
148 			ev.flow_id = flow_counter++ % nb_flows;
149 			ev.event_ptr = m[i];
150 			m[i]->timestamp = rte_get_timer_cycles();
151 			while (rte_event_enqueue_burst(dev_id,
152 						       port, &ev, 1) != 1) {
153 				if (t->done)
154 					break;
155 				rte_pause();
156 				m[i]->timestamp = rte_get_timer_cycles();
157 			}
158 		}
159 		count += BURST_SIZE;
160 	}
161 
162 	return 0;
163 }
164 
165 static inline int
166 perf_producer_burst(void *arg)
167 {
168 	uint32_t i;
169 	uint64_t timestamp;
170 	struct rte_event_dev_info dev_info;
171 	struct prod_data *p  = arg;
172 	struct test_perf *t = p->t;
173 	struct evt_options *opt = t->opt;
174 	const uint8_t dev_id = p->dev_id;
175 	const uint8_t port = p->port_id;
176 	struct rte_mempool *pool = t->pool;
177 	const uint64_t nb_pkts = t->nb_pkts;
178 	const uint32_t nb_flows = t->nb_flows;
179 	uint32_t flow_counter = 0;
180 	uint16_t enq = 0;
181 	uint64_t count = 0;
182 	struct perf_elt *m[MAX_PROD_ENQ_BURST_SIZE + 1];
183 	struct rte_event ev[MAX_PROD_ENQ_BURST_SIZE + 1];
184 	uint32_t burst_size = opt->prod_enq_burst_sz;
185 
186 	memset(m, 0, sizeof(*m) * (MAX_PROD_ENQ_BURST_SIZE + 1));
187 	rte_event_dev_info_get(dev_id, &dev_info);
188 	if (dev_info.max_event_port_enqueue_depth < burst_size)
189 		burst_size = dev_info.max_event_port_enqueue_depth;
190 
191 	if (opt->verbose_level > 1)
192 		printf("%s(): lcore %d dev_id %d port=%d queue %d\n", __func__,
193 				rte_lcore_id(), dev_id, port, p->queue_id);
194 
195 	for (i = 0; i < burst_size; i++) {
196 		ev[i].op = RTE_EVENT_OP_NEW;
197 		ev[i].queue_id = p->queue_id;
198 		ev[i].sched_type = t->opt->sched_type_list[0];
199 		ev[i].priority = RTE_EVENT_DEV_PRIORITY_NORMAL;
200 		ev[i].event_type =  RTE_EVENT_TYPE_CPU;
201 		ev[i].sub_event_type = 0; /* stage 0 */
202 	}
203 
204 	while (count < nb_pkts && t->done == false) {
205 		if (rte_mempool_get_bulk(pool, (void **)m, burst_size) < 0)
206 			continue;
207 		timestamp = rte_get_timer_cycles();
208 		for (i = 0; i < burst_size; i++) {
209 			ev[i].flow_id = flow_counter++ % nb_flows;
210 			ev[i].event_ptr = m[i];
211 			m[i]->timestamp = timestamp;
212 		}
213 		enq = rte_event_enqueue_burst(dev_id, port, ev, burst_size);
214 		while (enq < burst_size) {
215 			enq += rte_event_enqueue_burst(dev_id, port,
216 							ev + enq,
217 							burst_size - enq);
218 			if (t->done)
219 				break;
220 			rte_pause();
221 			timestamp = rte_get_timer_cycles();
222 			for (i = enq; i < burst_size; i++)
223 				m[i]->timestamp = timestamp;
224 		}
225 		count += burst_size;
226 	}
227 	return 0;
228 }
229 
230 static inline int
231 perf_event_timer_producer(void *arg)
232 {
233 	int i;
234 	struct prod_data *p  = arg;
235 	struct test_perf *t = p->t;
236 	struct evt_options *opt = t->opt;
237 	uint32_t flow_counter = 0;
238 	uint64_t count = 0;
239 	uint64_t arm_latency = 0;
240 	const uint8_t nb_timer_adptrs = opt->nb_timer_adptrs;
241 	const uint32_t nb_flows = t->nb_flows;
242 	const uint64_t nb_timers = opt->nb_timers;
243 	struct rte_mempool *pool = t->pool;
244 	struct perf_elt *m[BURST_SIZE + 1] = {NULL};
245 	struct rte_event_timer_adapter **adptr = t->timer_adptr;
246 	struct rte_event_timer tim;
247 	uint64_t timeout_ticks = opt->expiry_nsec / opt->timer_tick_nsec;
248 
249 	memset(&tim, 0, sizeof(struct rte_event_timer));
250 	timeout_ticks =
251 		opt->optm_timer_tick_nsec
252 			? ceil((double)(timeout_ticks * opt->timer_tick_nsec) /
253 			       opt->optm_timer_tick_nsec)
254 			: timeout_ticks;
255 	timeout_ticks += timeout_ticks ? 0 : 1;
256 	tim.ev.event_type = RTE_EVENT_TYPE_TIMER;
257 	tim.ev.op = RTE_EVENT_OP_NEW;
258 	tim.ev.sched_type = t->opt->sched_type_list[0];
259 	tim.ev.queue_id = p->queue_id;
260 	tim.ev.priority = RTE_EVENT_DEV_PRIORITY_NORMAL;
261 	tim.state = RTE_EVENT_TIMER_NOT_ARMED;
262 	tim.timeout_ticks = timeout_ticks;
263 
264 	if (opt->verbose_level > 1)
265 		printf("%s(): lcore %d\n", __func__, rte_lcore_id());
266 
267 	while (count < nb_timers && t->done == false) {
268 		if (rte_mempool_get_bulk(pool, (void **)m, BURST_SIZE) < 0)
269 			continue;
270 		for (i = 0; i < BURST_SIZE; i++) {
271 			rte_prefetch0(m[i + 1]);
272 			m[i]->tim = tim;
273 			m[i]->tim.ev.flow_id = flow_counter++ % nb_flows;
274 			m[i]->tim.ev.event_ptr = m[i];
275 			m[i]->timestamp = rte_get_timer_cycles();
276 			while (rte_event_timer_arm_burst(
277 			       adptr[flow_counter % nb_timer_adptrs],
278 			       (struct rte_event_timer **)&m[i], 1) != 1) {
279 				if (t->done)
280 					break;
281 				m[i]->timestamp = rte_get_timer_cycles();
282 			}
283 			arm_latency += rte_get_timer_cycles() - m[i]->timestamp;
284 		}
285 		count += BURST_SIZE;
286 	}
287 	fflush(stdout);
288 	rte_delay_ms(1000);
289 	printf("%s(): lcore %d Average event timer arm latency = %.3f us\n",
290 			__func__, rte_lcore_id(),
291 			count ? (float)(arm_latency / count) /
292 			(rte_get_timer_hz() / 1000000) : 0);
293 	return 0;
294 }
295 
296 static inline int
297 perf_event_timer_producer_burst(void *arg)
298 {
299 	int i;
300 	struct prod_data *p  = arg;
301 	struct test_perf *t = p->t;
302 	struct evt_options *opt = t->opt;
303 	uint32_t flow_counter = 0;
304 	uint64_t count = 0;
305 	uint64_t arm_latency = 0;
306 	const uint8_t nb_timer_adptrs = opt->nb_timer_adptrs;
307 	const uint32_t nb_flows = t->nb_flows;
308 	const uint64_t nb_timers = opt->nb_timers;
309 	struct rte_mempool *pool = t->pool;
310 	struct perf_elt *m[BURST_SIZE + 1] = {NULL};
311 	struct rte_event_timer_adapter **adptr = t->timer_adptr;
312 	struct rte_event_timer tim;
313 	uint64_t timeout_ticks = opt->expiry_nsec / opt->timer_tick_nsec;
314 
315 	memset(&tim, 0, sizeof(struct rte_event_timer));
316 	timeout_ticks =
317 		opt->optm_timer_tick_nsec
318 			? ceil((double)(timeout_ticks * opt->timer_tick_nsec) /
319 			       opt->optm_timer_tick_nsec)
320 			: timeout_ticks;
321 	timeout_ticks += timeout_ticks ? 0 : 1;
322 	tim.ev.event_type = RTE_EVENT_TYPE_TIMER;
323 	tim.ev.op = RTE_EVENT_OP_NEW;
324 	tim.ev.sched_type = t->opt->sched_type_list[0];
325 	tim.ev.queue_id = p->queue_id;
326 	tim.ev.priority = RTE_EVENT_DEV_PRIORITY_NORMAL;
327 	tim.state = RTE_EVENT_TIMER_NOT_ARMED;
328 	tim.timeout_ticks = timeout_ticks;
329 
330 	if (opt->verbose_level > 1)
331 		printf("%s(): lcore %d\n", __func__, rte_lcore_id());
332 
333 	while (count < nb_timers && t->done == false) {
334 		if (rte_mempool_get_bulk(pool, (void **)m, BURST_SIZE) < 0)
335 			continue;
336 		for (i = 0; i < BURST_SIZE; i++) {
337 			rte_prefetch0(m[i + 1]);
338 			m[i]->tim = tim;
339 			m[i]->tim.ev.flow_id = flow_counter++ % nb_flows;
340 			m[i]->tim.ev.event_ptr = m[i];
341 			m[i]->timestamp = rte_get_timer_cycles();
342 		}
343 		rte_event_timer_arm_tmo_tick_burst(
344 				adptr[flow_counter % nb_timer_adptrs],
345 				(struct rte_event_timer **)m,
346 				tim.timeout_ticks,
347 				BURST_SIZE);
348 		arm_latency += rte_get_timer_cycles() - m[i - 1]->timestamp;
349 		count += BURST_SIZE;
350 	}
351 	fflush(stdout);
352 	rte_delay_ms(1000);
353 	printf("%s(): lcore %d Average event timer arm latency = %.3f us\n",
354 			__func__, rte_lcore_id(),
355 			count ? (float)(arm_latency / count) /
356 			(rte_get_timer_hz() / 1000000) : 0);
357 	return 0;
358 }
359 
360 static inline void
361 crypto_adapter_enq_op_new(struct prod_data *p)
362 {
363 	struct test_perf *t = p->t;
364 	const uint32_t nb_flows = t->nb_flows;
365 	const uint64_t nb_pkts = t->nb_pkts;
366 	struct rte_mempool *pool = t->pool;
367 	struct evt_options *opt = t->opt;
368 	uint16_t qp_id = p->ca.cdev_qp_id;
369 	uint8_t cdev_id = p->ca.cdev_id;
370 	uint32_t flow_counter = 0;
371 	struct rte_crypto_op *op;
372 	struct rte_mbuf *m;
373 	uint64_t count = 0;
374 	uint16_t len;
375 
376 	if (opt->verbose_level > 1)
377 		printf("%s(): lcore %d queue %d cdev_id %u cdev_qp_id %u\n",
378 		       __func__, rte_lcore_id(), p->queue_id, p->ca.cdev_id,
379 		       p->ca.cdev_qp_id);
380 
381 	len = opt->mbuf_sz ? opt->mbuf_sz : RTE_ETHER_MIN_LEN;
382 
383 	while (count < nb_pkts && t->done == false) {
384 		if (opt->crypto_op_type == RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
385 			struct rte_crypto_sym_op *sym_op;
386 
387 			op = rte_crypto_op_alloc(t->ca_op_pool,
388 					 RTE_CRYPTO_OP_TYPE_SYMMETRIC);
389 			m = rte_pktmbuf_alloc(pool);
390 			if (m == NULL)
391 				continue;
392 
393 			rte_pktmbuf_append(m, len);
394 			sym_op = op->sym;
395 			sym_op->m_src = m;
396 			sym_op->cipher.data.offset = 0;
397 			sym_op->cipher.data.length = len;
398 			rte_crypto_op_attach_sym_session(
399 				op, p->ca.crypto_sess[flow_counter++ % nb_flows]);
400 		} else {
401 			struct rte_crypto_asym_op *asym_op;
402 			uint8_t *result = rte_zmalloc(NULL,
403 					modex_test_case.result_len, 0);
404 
405 			op = rte_crypto_op_alloc(t->ca_op_pool,
406 					 RTE_CRYPTO_OP_TYPE_ASYMMETRIC);
407 			asym_op = op->asym;
408 			asym_op->modex.base.data = modex_test_case.base.data;
409 			asym_op->modex.base.length = modex_test_case.base.len;
410 			asym_op->modex.result.data = result;
411 			asym_op->modex.result.length = modex_test_case.result_len;
412 			rte_crypto_op_attach_asym_session(
413 				op, p->ca.crypto_sess[flow_counter++ % nb_flows]);
414 		}
415 		while (rte_cryptodev_enqueue_burst(cdev_id, qp_id, &op, 1) != 1 &&
416 				t->done == false)
417 			rte_pause();
418 
419 		count++;
420 	}
421 }
422 
423 static inline void
424 crypto_adapter_enq_op_fwd(struct prod_data *p)
425 {
426 	const uint8_t dev_id = p->dev_id;
427 	const uint8_t port = p->port_id;
428 	struct test_perf *t = p->t;
429 	const uint32_t nb_flows = t->nb_flows;
430 	const uint64_t nb_pkts = t->nb_pkts;
431 	struct rte_mempool *pool = t->pool;
432 	struct evt_options *opt = t->opt;
433 	uint32_t flow_counter = 0;
434 	struct rte_crypto_op *op;
435 	struct rte_event ev;
436 	struct rte_mbuf *m;
437 	uint64_t count = 0;
438 	uint16_t len;
439 
440 	if (opt->verbose_level > 1)
441 		printf("%s(): lcore %d port %d queue %d cdev_id %u cdev_qp_id %u\n",
442 		       __func__, rte_lcore_id(), port, p->queue_id,
443 		       p->ca.cdev_id, p->ca.cdev_qp_id);
444 
445 	ev.event = 0;
446 	ev.op = RTE_EVENT_OP_NEW;
447 	ev.queue_id = p->queue_id;
448 	ev.sched_type = RTE_SCHED_TYPE_ATOMIC;
449 	ev.event_type = RTE_EVENT_TYPE_CPU;
450 	len = opt->mbuf_sz ? opt->mbuf_sz : RTE_ETHER_MIN_LEN;
451 
452 	while (count < nb_pkts && t->done == false) {
453 		if (opt->crypto_op_type == RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
454 			struct rte_crypto_sym_op *sym_op;
455 
456 			op = rte_crypto_op_alloc(t->ca_op_pool,
457 					 RTE_CRYPTO_OP_TYPE_SYMMETRIC);
458 			m = rte_pktmbuf_alloc(pool);
459 			if (m == NULL)
460 				continue;
461 
462 			rte_pktmbuf_append(m, len);
463 			sym_op = op->sym;
464 			sym_op->m_src = m;
465 			sym_op->cipher.data.offset = 0;
466 			sym_op->cipher.data.length = len;
467 			rte_crypto_op_attach_sym_session(
468 				op, p->ca.crypto_sess[flow_counter++ % nb_flows]);
469 		} else {
470 			struct rte_crypto_asym_op *asym_op;
471 			uint8_t *result = rte_zmalloc(NULL,
472 					modex_test_case.result_len, 0);
473 
474 			op = rte_crypto_op_alloc(t->ca_op_pool,
475 					 RTE_CRYPTO_OP_TYPE_ASYMMETRIC);
476 			asym_op = op->asym;
477 			asym_op->modex.base.data = modex_test_case.base.data;
478 			asym_op->modex.base.length = modex_test_case.base.len;
479 			asym_op->modex.result.data = result;
480 			asym_op->modex.result.length = modex_test_case.result_len;
481 			rte_crypto_op_attach_asym_session(
482 				op, p->ca.crypto_sess[flow_counter++ % nb_flows]);
483 		}
484 		ev.event_ptr = op;
485 
486 		while (rte_event_crypto_adapter_enqueue(dev_id, port, &ev, 1) != 1 &&
487 		       t->done == false)
488 			rte_pause();
489 
490 		count++;
491 	}
492 }
493 
494 static inline int
495 perf_event_crypto_producer(void *arg)
496 {
497 	struct prod_data *p = arg;
498 	struct evt_options *opt = p->t->opt;
499 
500 	if (opt->crypto_adptr_mode == RTE_EVENT_CRYPTO_ADAPTER_OP_NEW)
501 		crypto_adapter_enq_op_new(p);
502 	else
503 		crypto_adapter_enq_op_fwd(p);
504 
505 	return 0;
506 }
507 
508 static int
509 perf_producer_wrapper(void *arg)
510 {
511 	struct prod_data *p  = arg;
512 	struct test_perf *t = p->t;
513 	bool burst = evt_has_burst_mode(p->dev_id);
514 
515 	/* In case of synthetic producer, launch perf_producer or
516 	 * perf_producer_burst depending on producer enqueue burst size
517 	 */
518 	if (t->opt->prod_type == EVT_PROD_TYPE_SYNT &&
519 			t->opt->prod_enq_burst_sz == 1)
520 		return perf_producer(arg);
521 	else if (t->opt->prod_type == EVT_PROD_TYPE_SYNT &&
522 			t->opt->prod_enq_burst_sz > 1) {
523 		if (!burst)
524 			evt_err("This event device does not support burst mode");
525 		else
526 			return perf_producer_burst(arg);
527 	}
528 	else if (t->opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR &&
529 			!t->opt->timdev_use_burst)
530 		return perf_event_timer_producer(arg);
531 	else if (t->opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR &&
532 			t->opt->timdev_use_burst)
533 		return perf_event_timer_producer_burst(arg);
534 	else if (t->opt->prod_type == EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR)
535 		return perf_event_crypto_producer(arg);
536 	return 0;
537 }
538 
539 static inline uint64_t
540 processed_pkts(struct test_perf *t)
541 {
542 	uint8_t i;
543 	uint64_t total = 0;
544 
545 	for (i = 0; i < t->nb_workers; i++)
546 		total += t->worker[i].processed_pkts;
547 
548 	return total;
549 }
550 
551 static inline uint64_t
552 total_latency(struct test_perf *t)
553 {
554 	uint8_t i;
555 	uint64_t total = 0;
556 
557 	for (i = 0; i < t->nb_workers; i++)
558 		total += t->worker[i].latency;
559 
560 	return total;
561 }
562 
563 
564 int
565 perf_launch_lcores(struct evt_test *test, struct evt_options *opt,
566 		int (*worker)(void *))
567 {
568 	int ret, lcore_id;
569 	struct test_perf *t = evt_test_priv(test);
570 
571 	int port_idx = 0;
572 	/* launch workers */
573 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
574 		if (!(opt->wlcores[lcore_id]))
575 			continue;
576 
577 		ret = rte_eal_remote_launch(worker,
578 				 &t->worker[port_idx], lcore_id);
579 		if (ret) {
580 			evt_err("failed to launch worker %d", lcore_id);
581 			return ret;
582 		}
583 		port_idx++;
584 	}
585 
586 	/* launch producers */
587 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
588 		if (!(opt->plcores[lcore_id]))
589 			continue;
590 
591 		ret = rte_eal_remote_launch(perf_producer_wrapper,
592 				&t->prod[port_idx], lcore_id);
593 		if (ret) {
594 			evt_err("failed to launch perf_producer %d", lcore_id);
595 			return ret;
596 		}
597 		port_idx++;
598 	}
599 
600 	const uint64_t total_pkts = t->outstand_pkts;
601 
602 	uint64_t dead_lock_cycles = rte_get_timer_cycles();
603 	int64_t dead_lock_remaining  =  total_pkts;
604 	const uint64_t dead_lock_sample = rte_get_timer_hz() * 5;
605 
606 	uint64_t perf_cycles = rte_get_timer_cycles();
607 	int64_t perf_remaining  = total_pkts;
608 	const uint64_t perf_sample = rte_get_timer_hz();
609 
610 	static float total_mpps;
611 	static uint64_t samples;
612 
613 	const uint64_t freq_mhz = rte_get_timer_hz() / 1000000;
614 	int64_t remaining = t->outstand_pkts - processed_pkts(t);
615 
616 	while (t->done == false) {
617 		const uint64_t new_cycles = rte_get_timer_cycles();
618 
619 		if ((new_cycles - perf_cycles) > perf_sample) {
620 			const uint64_t latency = total_latency(t);
621 			const uint64_t pkts = processed_pkts(t);
622 
623 			remaining = t->outstand_pkts - pkts;
624 			float mpps = (float)(perf_remaining-remaining)/1000000;
625 
626 			perf_remaining = remaining;
627 			perf_cycles = new_cycles;
628 			total_mpps += mpps;
629 			++samples;
630 			if (opt->fwd_latency && pkts > 0) {
631 				printf(CLGRN"\r%.3f mpps avg %.3f mpps [avg fwd latency %.3f us] "CLNRM,
632 					mpps, total_mpps/samples,
633 					(float)(latency/pkts)/freq_mhz);
634 			} else {
635 				printf(CLGRN"\r%.3f mpps avg %.3f mpps"CLNRM,
636 					mpps, total_mpps/samples);
637 			}
638 			fflush(stdout);
639 
640 			if (remaining <= 0) {
641 				t->result = EVT_TEST_SUCCESS;
642 				if (opt->prod_type == EVT_PROD_TYPE_SYNT ||
643 				    opt->prod_type ==
644 					    EVT_PROD_TYPE_EVENT_TIMER_ADPTR ||
645 				    opt->prod_type ==
646 					    EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR) {
647 					t->done = true;
648 					break;
649 				}
650 			}
651 		}
652 
653 		if (new_cycles - dead_lock_cycles > dead_lock_sample &&
654 		    (opt->prod_type == EVT_PROD_TYPE_SYNT ||
655 		     opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR ||
656 		     opt->prod_type == EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR)) {
657 			remaining = t->outstand_pkts - processed_pkts(t);
658 			if (dead_lock_remaining == remaining) {
659 				rte_event_dev_dump(opt->dev_id, stdout);
660 				evt_err("No schedules for seconds, deadlock");
661 				t->done = true;
662 				break;
663 			}
664 			dead_lock_remaining = remaining;
665 			dead_lock_cycles = new_cycles;
666 		}
667 	}
668 	printf("\n");
669 	return 0;
670 }
671 
672 static int
673 perf_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
674 		struct rte_event_port_conf prod_conf)
675 {
676 	int ret = 0;
677 	uint16_t prod;
678 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
679 
680 	memset(&queue_conf, 0,
681 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
682 	queue_conf.ev.sched_type = opt->sched_type_list[0];
683 	RTE_ETH_FOREACH_DEV(prod) {
684 		uint32_t cap;
685 
686 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
687 				prod, &cap);
688 		if (ret) {
689 			evt_err("failed to get event rx adapter[%d]"
690 					" capabilities",
691 					opt->dev_id);
692 			return ret;
693 		}
694 		queue_conf.ev.queue_id = prod * stride;
695 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
696 				&prod_conf);
697 		if (ret) {
698 			evt_err("failed to create rx adapter[%d]", prod);
699 			return ret;
700 		}
701 		ret = rte_event_eth_rx_adapter_queue_add(prod, prod, -1,
702 				&queue_conf);
703 		if (ret) {
704 			evt_err("failed to add rx queues to adapter[%d]", prod);
705 			return ret;
706 		}
707 
708 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
709 			uint32_t service_id;
710 
711 			rte_event_eth_rx_adapter_service_id_get(prod,
712 					&service_id);
713 			ret = evt_service_setup(service_id);
714 			if (ret) {
715 				evt_err("Failed to setup service core"
716 						" for Rx adapter\n");
717 				return ret;
718 			}
719 		}
720 	}
721 
722 	return ret;
723 }
724 
725 static int
726 perf_event_timer_adapter_setup(struct test_perf *t)
727 {
728 	int i;
729 	int ret;
730 	struct rte_event_timer_adapter_info adapter_info;
731 	struct rte_event_timer_adapter *wl;
732 	uint8_t nb_producers = evt_nr_active_lcores(t->opt->plcores);
733 	uint8_t flags = RTE_EVENT_TIMER_ADAPTER_F_ADJUST_RES;
734 
735 	if (nb_producers == 1)
736 		flags |= RTE_EVENT_TIMER_ADAPTER_F_SP_PUT;
737 
738 	for (i = 0; i < t->opt->nb_timer_adptrs; i++) {
739 		struct rte_event_timer_adapter_conf config = {
740 			.event_dev_id = t->opt->dev_id,
741 			.timer_adapter_id = i,
742 			.timer_tick_ns = t->opt->timer_tick_nsec,
743 			.max_tmo_ns = t->opt->max_tmo_nsec,
744 			.nb_timers = t->opt->pool_sz,
745 			.flags = flags,
746 		};
747 
748 		wl = rte_event_timer_adapter_create(&config);
749 		if (wl == NULL) {
750 			evt_err("failed to create event timer ring %d", i);
751 			return rte_errno;
752 		}
753 
754 		memset(&adapter_info, 0,
755 				sizeof(struct rte_event_timer_adapter_info));
756 		rte_event_timer_adapter_get_info(wl, &adapter_info);
757 		t->opt->optm_timer_tick_nsec = adapter_info.min_resolution_ns;
758 
759 		if (!(adapter_info.caps &
760 				RTE_EVENT_TIMER_ADAPTER_CAP_INTERNAL_PORT)) {
761 			uint32_t service_id = -1U;
762 
763 			rte_event_timer_adapter_service_id_get(wl,
764 					&service_id);
765 			ret = evt_service_setup(service_id);
766 			if (ret) {
767 				evt_err("Failed to setup service core"
768 						" for timer adapter\n");
769 				return ret;
770 			}
771 			rte_service_runstate_set(service_id, 1);
772 		}
773 		t->timer_adptr[i] = wl;
774 	}
775 	return 0;
776 }
777 
778 static int
779 perf_event_crypto_adapter_setup(struct test_perf *t, struct prod_data *p)
780 {
781 	struct evt_options *opt = t->opt;
782 	uint32_t cap;
783 	int ret;
784 
785 	ret = rte_event_crypto_adapter_caps_get(p->dev_id, p->ca.cdev_id, &cap);
786 	if (ret) {
787 		evt_err("Failed to get crypto adapter capabilities");
788 		return ret;
789 	}
790 
791 	if (((opt->crypto_adptr_mode == RTE_EVENT_CRYPTO_ADAPTER_OP_NEW) &&
792 	     !(cap & RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_NEW)) ||
793 	    ((opt->crypto_adptr_mode == RTE_EVENT_CRYPTO_ADAPTER_OP_FORWARD) &&
794 	     !(cap & RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_FWD))) {
795 		evt_err("crypto adapter %s mode unsupported\n",
796 			opt->crypto_adptr_mode ? "OP_FORWARD" : "OP_NEW");
797 		return -ENOTSUP;
798 	} else if (!(cap & RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA)) {
799 		evt_err("Storing crypto session not supported");
800 		return -ENOTSUP;
801 	}
802 
803 	if (cap & RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_QP_EV_BIND) {
804 		struct rte_event response_info;
805 
806 		response_info.event = 0;
807 		response_info.sched_type = RTE_SCHED_TYPE_ATOMIC;
808 		response_info.queue_id = p->queue_id;
809 		ret = rte_event_crypto_adapter_queue_pair_add(
810 			TEST_PERF_CA_ID, p->ca.cdev_id, p->ca.cdev_qp_id,
811 			&response_info);
812 	} else {
813 		ret = rte_event_crypto_adapter_queue_pair_add(
814 			TEST_PERF_CA_ID, p->ca.cdev_id, p->ca.cdev_qp_id, NULL);
815 	}
816 
817 	return ret;
818 }
819 
820 static struct rte_cryptodev_sym_session *
821 cryptodev_sym_sess_create(struct prod_data *p, struct test_perf *t)
822 {
823 	struct rte_crypto_sym_xform cipher_xform;
824 	struct rte_cryptodev_sym_session *sess;
825 
826 	cipher_xform.type = RTE_CRYPTO_SYM_XFORM_CIPHER;
827 	cipher_xform.cipher.algo = RTE_CRYPTO_CIPHER_NULL;
828 	cipher_xform.cipher.op = RTE_CRYPTO_CIPHER_OP_ENCRYPT;
829 	cipher_xform.next = NULL;
830 
831 	sess = rte_cryptodev_sym_session_create(t->ca_sess_pool);
832 	if (sess == NULL) {
833 		evt_err("Failed to create sym session");
834 		return NULL;
835 	}
836 
837 	if (rte_cryptodev_sym_session_init(p->ca.cdev_id, sess, &cipher_xform,
838 					   t->ca_sess_priv_pool)) {
839 		evt_err("Failed to init session");
840 		return NULL;
841 	}
842 
843 	return sess;
844 }
845 
846 static void *
847 cryptodev_asym_sess_create(struct prod_data *p, struct test_perf *t)
848 {
849 	const struct rte_cryptodev_asymmetric_xform_capability *capability;
850 	struct rte_cryptodev_asym_capability_idx cap_idx;
851 	struct rte_crypto_asym_xform xform;
852 	void *sess;
853 
854 	xform.next = NULL;
855 	xform.xform_type = RTE_CRYPTO_ASYM_XFORM_MODEX;
856 	cap_idx.type = xform.xform_type;
857 	capability = rte_cryptodev_asym_capability_get(p->ca.cdev_id, &cap_idx);
858 	if (capability == NULL) {
859 		evt_err("Device doesn't support MODEX. Test Skipped\n");
860 		return NULL;
861 	}
862 
863 	xform.modex.modulus.data = modex_test_case.modulus.data;
864 	xform.modex.modulus.length = modex_test_case.modulus.len;
865 	xform.modex.exponent.data = modex_test_case.exponent.data;
866 	xform.modex.exponent.length = modex_test_case.exponent.len;
867 
868 	if (rte_cryptodev_asym_session_create(p->ca.cdev_id, &xform,
869 			t->ca_asym_sess_pool, &sess)) {
870 		evt_err("Failed to create asym session");
871 		return NULL;
872 	}
873 
874 	return sess;
875 }
876 
877 int
878 perf_event_dev_port_setup(struct evt_test *test, struct evt_options *opt,
879 				uint8_t stride, uint8_t nb_queues,
880 				const struct rte_event_port_conf *port_conf)
881 {
882 	struct test_perf *t = evt_test_priv(test);
883 	uint16_t port, prod;
884 	int ret = -1;
885 
886 	/* setup one port per worker, linking to all queues */
887 	for (port = 0; port < evt_nr_active_lcores(opt->wlcores);
888 				port++) {
889 		struct worker_data *w = &t->worker[port];
890 
891 		w->dev_id = opt->dev_id;
892 		w->port_id = port;
893 		w->t = t;
894 		w->processed_pkts = 0;
895 		w->latency = 0;
896 
897 		struct rte_event_port_conf conf = *port_conf;
898 		conf.event_port_cfg |= RTE_EVENT_PORT_CFG_HINT_WORKER;
899 
900 		ret = rte_event_port_setup(opt->dev_id, port, &conf);
901 		if (ret) {
902 			evt_err("failed to setup port %d", port);
903 			return ret;
904 		}
905 
906 		ret = rte_event_port_link(opt->dev_id, port, NULL, NULL, 0);
907 		if (ret != nb_queues) {
908 			evt_err("failed to link all queues to port %d", port);
909 			return -EINVAL;
910 		}
911 	}
912 
913 	/* port for producers, no links */
914 	if (opt->prod_type == EVT_PROD_TYPE_ETH_RX_ADPTR) {
915 		for ( ; port < perf_nb_event_ports(opt); port++) {
916 			struct prod_data *p = &t->prod[port];
917 			p->t = t;
918 		}
919 
920 		struct rte_event_port_conf conf = *port_conf;
921 		conf.event_port_cfg |= RTE_EVENT_PORT_CFG_HINT_PRODUCER;
922 
923 		ret = perf_event_rx_adapter_setup(opt, stride, conf);
924 		if (ret)
925 			return ret;
926 	} else if (opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR) {
927 		prod = 0;
928 		for ( ; port < perf_nb_event_ports(opt); port++) {
929 			struct prod_data *p = &t->prod[port];
930 			p->queue_id = prod * stride;
931 			p->t = t;
932 			prod++;
933 		}
934 
935 		ret = perf_event_timer_adapter_setup(t);
936 		if (ret)
937 			return ret;
938 	} else if (opt->prod_type == EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR) {
939 		struct rte_event_port_conf conf = *port_conf;
940 		uint8_t cdev_id = 0;
941 		uint16_t qp_id = 0;
942 
943 		ret = rte_event_crypto_adapter_create(TEST_PERF_CA_ID,
944 						      opt->dev_id, &conf, 0);
945 		if (ret) {
946 			evt_err("Failed to create crypto adapter");
947 			return ret;
948 		}
949 
950 		prod = 0;
951 		for (; port < perf_nb_event_ports(opt); port++) {
952 			union rte_event_crypto_metadata m_data;
953 			struct prod_data *p = &t->prod[port];
954 			uint32_t flow_id;
955 
956 			if (qp_id == rte_cryptodev_queue_pair_count(cdev_id)) {
957 				cdev_id++;
958 				qp_id = 0;
959 			}
960 
961 			p->dev_id = opt->dev_id;
962 			p->port_id = port;
963 			p->queue_id = prod * stride;
964 			p->ca.cdev_id = cdev_id;
965 			p->ca.cdev_qp_id = qp_id;
966 			p->ca.crypto_sess = rte_zmalloc_socket(
967 				NULL, sizeof(void *) * t->nb_flows,
968 				RTE_CACHE_LINE_SIZE, opt->socket_id);
969 			p->t = t;
970 
971 			m_data.request_info.cdev_id = p->ca.cdev_id;
972 			m_data.request_info.queue_pair_id = p->ca.cdev_qp_id;
973 			m_data.response_info.sched_type = RTE_SCHED_TYPE_ATOMIC;
974 			m_data.response_info.queue_id = p->queue_id;
975 
976 			for (flow_id = 0; flow_id < t->nb_flows; flow_id++) {
977 				m_data.response_info.flow_id = flow_id;
978 				if (opt->crypto_op_type ==
979 						RTE_CRYPTO_OP_TYPE_SYMMETRIC) {
980 					struct rte_cryptodev_sym_session *sess;
981 
982 					sess = cryptodev_sym_sess_create(p, t);
983 					if (sess == NULL)
984 						return -ENOMEM;
985 
986 					rte_cryptodev_session_event_mdata_set(
987 						cdev_id,
988 						sess,
989 						RTE_CRYPTO_OP_TYPE_SYMMETRIC,
990 						RTE_CRYPTO_OP_WITH_SESSION,
991 						&m_data, sizeof(m_data));
992 					p->ca.crypto_sess[flow_id] = sess;
993 				} else {
994 					void *sess;
995 
996 					sess = cryptodev_asym_sess_create(p, t);
997 					if (sess == NULL)
998 						return -ENOMEM;
999 					rte_cryptodev_session_event_mdata_set(
1000 						cdev_id,
1001 						sess,
1002 						RTE_CRYPTO_OP_TYPE_ASYMMETRIC,
1003 						RTE_CRYPTO_OP_WITH_SESSION,
1004 						&m_data, sizeof(m_data));
1005 					p->ca.crypto_sess[flow_id] = sess;
1006 				}
1007 			}
1008 
1009 			conf.event_port_cfg |=
1010 				RTE_EVENT_PORT_CFG_HINT_PRODUCER |
1011 				RTE_EVENT_PORT_CFG_HINT_CONSUMER;
1012 
1013 			ret = rte_event_port_setup(opt->dev_id, port, &conf);
1014 			if (ret) {
1015 				evt_err("failed to setup port %d", port);
1016 				return ret;
1017 			}
1018 
1019 			ret = perf_event_crypto_adapter_setup(t, p);
1020 			if (ret)
1021 				return ret;
1022 
1023 			qp_id++;
1024 			prod++;
1025 		}
1026 	} else {
1027 		prod = 0;
1028 		for ( ; port < perf_nb_event_ports(opt); port++) {
1029 			struct prod_data *p = &t->prod[port];
1030 
1031 			p->dev_id = opt->dev_id;
1032 			p->port_id = port;
1033 			p->queue_id = prod * stride;
1034 			p->t = t;
1035 
1036 			struct rte_event_port_conf conf = *port_conf;
1037 			conf.event_port_cfg |=
1038 				RTE_EVENT_PORT_CFG_HINT_PRODUCER |
1039 				RTE_EVENT_PORT_CFG_HINT_CONSUMER;
1040 
1041 			ret = rte_event_port_setup(opt->dev_id, port, &conf);
1042 			if (ret) {
1043 				evt_err("failed to setup port %d", port);
1044 				return ret;
1045 			}
1046 			prod++;
1047 		}
1048 	}
1049 
1050 	return ret;
1051 }
1052 
1053 int
1054 perf_opt_check(struct evt_options *opt, uint64_t nb_queues)
1055 {
1056 	unsigned int lcores;
1057 
1058 	/* N producer + N worker + main when producer cores are used
1059 	 * Else N worker + main when Rx adapter is used
1060 	 */
1061 	lcores = opt->prod_type == EVT_PROD_TYPE_SYNT ? 3 : 2;
1062 
1063 	if (rte_lcore_count() < lcores) {
1064 		evt_err("test need minimum %d lcores", lcores);
1065 		return -1;
1066 	}
1067 
1068 	/* Validate worker lcores */
1069 	if (evt_lcores_has_overlap(opt->wlcores, rte_get_main_lcore())) {
1070 		evt_err("worker lcores overlaps with main lcore");
1071 		return -1;
1072 	}
1073 	if (evt_lcores_has_overlap_multi(opt->wlcores, opt->plcores)) {
1074 		evt_err("worker lcores overlaps producer lcores");
1075 		return -1;
1076 	}
1077 	if (evt_has_disabled_lcore(opt->wlcores)) {
1078 		evt_err("one or more workers lcores are not enabled");
1079 		return -1;
1080 	}
1081 	if (!evt_has_active_lcore(opt->wlcores)) {
1082 		evt_err("minimum one worker is required");
1083 		return -1;
1084 	}
1085 
1086 	if (opt->prod_type == EVT_PROD_TYPE_SYNT ||
1087 	    opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR ||
1088 	    opt->prod_type == EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR) {
1089 		/* Validate producer lcores */
1090 		if (evt_lcores_has_overlap(opt->plcores,
1091 					rte_get_main_lcore())) {
1092 			evt_err("producer lcores overlaps with main lcore");
1093 			return -1;
1094 		}
1095 		if (evt_has_disabled_lcore(opt->plcores)) {
1096 			evt_err("one or more producer lcores are not enabled");
1097 			return -1;
1098 		}
1099 		if (!evt_has_active_lcore(opt->plcores)) {
1100 			evt_err("minimum one producer is required");
1101 			return -1;
1102 		}
1103 	}
1104 
1105 	if (evt_has_invalid_stage(opt))
1106 		return -1;
1107 
1108 	if (evt_has_invalid_sched_type(opt))
1109 		return -1;
1110 
1111 	if (nb_queues > EVT_MAX_QUEUES) {
1112 		evt_err("number of queues exceeds %d", EVT_MAX_QUEUES);
1113 		return -1;
1114 	}
1115 	if (perf_nb_event_ports(opt) > EVT_MAX_PORTS) {
1116 		evt_err("number of ports exceeds %d", EVT_MAX_PORTS);
1117 		return -1;
1118 	}
1119 
1120 	/* Fixups */
1121 	if ((opt->nb_stages == 1 &&
1122 			opt->prod_type != EVT_PROD_TYPE_EVENT_TIMER_ADPTR) &&
1123 			opt->fwd_latency) {
1124 		evt_info("fwd_latency is valid when nb_stages > 1, disabling");
1125 		opt->fwd_latency = 0;
1126 	}
1127 
1128 	if (opt->fwd_latency && !opt->q_priority) {
1129 		evt_info("enabled queue priority for latency measurement");
1130 		opt->q_priority = 1;
1131 	}
1132 	if (opt->nb_pkts == 0)
1133 		opt->nb_pkts = INT64_MAX/evt_nr_active_lcores(opt->plcores);
1134 
1135 	return 0;
1136 }
1137 
1138 void
1139 perf_opt_dump(struct evt_options *opt, uint8_t nb_queues)
1140 {
1141 	evt_dump("nb_prod_lcores", "%d", evt_nr_active_lcores(opt->plcores));
1142 	evt_dump_producer_lcores(opt);
1143 	evt_dump("nb_worker_lcores", "%d", evt_nr_active_lcores(opt->wlcores));
1144 	evt_dump_worker_lcores(opt);
1145 	evt_dump_nb_stages(opt);
1146 	evt_dump("nb_evdev_ports", "%d", perf_nb_event_ports(opt));
1147 	evt_dump("nb_evdev_queues", "%d", nb_queues);
1148 	evt_dump_queue_priority(opt);
1149 	evt_dump_sched_type_list(opt);
1150 	evt_dump_producer_type(opt);
1151 	evt_dump("prod_enq_burst_sz", "%d", opt->prod_enq_burst_sz);
1152 }
1153 
1154 static void
1155 perf_event_port_flush(uint8_t dev_id __rte_unused, struct rte_event ev,
1156 		      void *args)
1157 {
1158 	rte_mempool_put(args, ev.event_ptr);
1159 }
1160 
1161 void
1162 perf_worker_cleanup(struct rte_mempool *const pool, uint8_t dev_id,
1163 		    uint8_t port_id, struct rte_event events[], uint16_t nb_enq,
1164 		    uint16_t nb_deq)
1165 {
1166 	int i;
1167 
1168 	if (nb_deq) {
1169 		for (i = nb_enq; i < nb_deq; i++)
1170 			rte_mempool_put(pool, events[i].event_ptr);
1171 
1172 		for (i = 0; i < nb_deq; i++)
1173 			events[i].op = RTE_EVENT_OP_RELEASE;
1174 		rte_event_enqueue_burst(dev_id, port_id, events, nb_deq);
1175 	}
1176 	rte_event_port_quiesce(dev_id, port_id, perf_event_port_flush, pool);
1177 }
1178 
1179 void
1180 perf_eventdev_destroy(struct evt_test *test, struct evt_options *opt)
1181 {
1182 	int i;
1183 	struct test_perf *t = evt_test_priv(test);
1184 
1185 	if (opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR) {
1186 		for (i = 0; i < opt->nb_timer_adptrs; i++)
1187 			rte_event_timer_adapter_stop(t->timer_adptr[i]);
1188 	}
1189 	rte_event_dev_stop(opt->dev_id);
1190 	rte_event_dev_close(opt->dev_id);
1191 }
1192 
1193 static inline void
1194 perf_elt_init(struct rte_mempool *mp, void *arg __rte_unused,
1195 	    void *obj, unsigned i __rte_unused)
1196 {
1197 	memset(obj, 0, mp->elt_size);
1198 }
1199 
1200 #define NB_RX_DESC			128
1201 #define NB_TX_DESC			512
1202 int
1203 perf_ethdev_setup(struct evt_test *test, struct evt_options *opt)
1204 {
1205 	uint16_t i;
1206 	int ret;
1207 	struct test_perf *t = evt_test_priv(test);
1208 	struct rte_eth_conf port_conf = {
1209 		.rxmode = {
1210 			.mq_mode = RTE_ETH_MQ_RX_RSS,
1211 			.split_hdr_size = 0,
1212 		},
1213 		.rx_adv_conf = {
1214 			.rss_conf = {
1215 				.rss_key = NULL,
1216 				.rss_hf = RTE_ETH_RSS_IP,
1217 			},
1218 		},
1219 	};
1220 
1221 	if (opt->prod_type != EVT_PROD_TYPE_ETH_RX_ADPTR)
1222 		return 0;
1223 
1224 	if (!rte_eth_dev_count_avail()) {
1225 		evt_err("No ethernet ports found.");
1226 		return -ENODEV;
1227 	}
1228 
1229 	RTE_ETH_FOREACH_DEV(i) {
1230 		struct rte_eth_dev_info dev_info;
1231 		struct rte_eth_conf local_port_conf = port_conf;
1232 
1233 		ret = rte_eth_dev_info_get(i, &dev_info);
1234 		if (ret != 0) {
1235 			evt_err("Error during getting device (port %u) info: %s\n",
1236 					i, strerror(-ret));
1237 			return ret;
1238 		}
1239 
1240 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1241 			dev_info.flow_type_rss_offloads;
1242 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1243 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
1244 			evt_info("Port %u modified RSS hash function based on hardware support,"
1245 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
1246 				i,
1247 				port_conf.rx_adv_conf.rss_conf.rss_hf,
1248 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1249 		}
1250 
1251 		if (rte_eth_dev_configure(i, 1, 1, &local_port_conf) < 0) {
1252 			evt_err("Failed to configure eth port [%d]", i);
1253 			return -EINVAL;
1254 		}
1255 
1256 		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
1257 				rte_socket_id(), NULL, t->pool) < 0) {
1258 			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
1259 					i, 0);
1260 			return -EINVAL;
1261 		}
1262 
1263 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
1264 					rte_socket_id(), NULL) < 0) {
1265 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
1266 					i, 0);
1267 			return -EINVAL;
1268 		}
1269 
1270 		ret = rte_eth_promiscuous_enable(i);
1271 		if (ret != 0) {
1272 			evt_err("Failed to enable promiscuous mode for eth port [%d]: %s",
1273 				i, rte_strerror(-ret));
1274 			return ret;
1275 		}
1276 	}
1277 
1278 	return 0;
1279 }
1280 
1281 void
1282 perf_ethdev_rx_stop(struct evt_test *test, struct evt_options *opt)
1283 {
1284 	uint16_t i;
1285 	RTE_SET_USED(test);
1286 
1287 	if (opt->prod_type == EVT_PROD_TYPE_ETH_RX_ADPTR) {
1288 		RTE_ETH_FOREACH_DEV(i) {
1289 			rte_event_eth_rx_adapter_stop(i);
1290 			rte_event_eth_rx_adapter_queue_del(i, i, -1);
1291 			rte_eth_dev_rx_queue_stop(i, 0);
1292 		}
1293 	}
1294 }
1295 
1296 void
1297 perf_ethdev_destroy(struct evt_test *test, struct evt_options *opt)
1298 {
1299 	uint16_t i;
1300 	RTE_SET_USED(test);
1301 
1302 	if (opt->prod_type == EVT_PROD_TYPE_ETH_RX_ADPTR) {
1303 		RTE_ETH_FOREACH_DEV(i) {
1304 			rte_event_eth_tx_adapter_stop(i);
1305 			rte_event_eth_tx_adapter_queue_del(i, i, -1);
1306 			rte_eth_dev_tx_queue_stop(i, 0);
1307 			rte_eth_dev_stop(i);
1308 		}
1309 	}
1310 }
1311 
1312 int
1313 perf_cryptodev_setup(struct evt_test *test, struct evt_options *opt)
1314 {
1315 	uint8_t cdev_count, cdev_id, nb_plcores, nb_qps;
1316 	struct test_perf *t = evt_test_priv(test);
1317 	unsigned int max_session_size;
1318 	uint32_t nb_sessions;
1319 	int ret;
1320 
1321 	if (opt->prod_type != EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR)
1322 		return 0;
1323 
1324 	cdev_count = rte_cryptodev_count();
1325 	if (cdev_count == 0) {
1326 		evt_err("No crypto devices available\n");
1327 		return -ENODEV;
1328 	}
1329 
1330 	t->ca_op_pool = rte_crypto_op_pool_create(
1331 		"crypto_op_pool", opt->crypto_op_type, opt->pool_sz,
1332 		128, sizeof(union rte_event_crypto_metadata),
1333 		rte_socket_id());
1334 	if (t->ca_op_pool == NULL) {
1335 		evt_err("Failed to create crypto op pool");
1336 		return -ENOMEM;
1337 	}
1338 
1339 	nb_sessions = evt_nr_active_lcores(opt->plcores) * t->nb_flows;
1340 	t->ca_asym_sess_pool = rte_cryptodev_asym_session_pool_create(
1341 		"ca_asym_sess_pool", nb_sessions, 0,
1342 		sizeof(union rte_event_crypto_metadata), SOCKET_ID_ANY);
1343 	if (t->ca_asym_sess_pool == NULL) {
1344 		evt_err("Failed to create sym session pool");
1345 		ret = -ENOMEM;
1346 		goto err;
1347 	}
1348 
1349 	t->ca_sess_pool = rte_cryptodev_sym_session_pool_create(
1350 		"ca_sess_pool", nb_sessions, 0, 0,
1351 		sizeof(union rte_event_crypto_metadata), SOCKET_ID_ANY);
1352 	if (t->ca_sess_pool == NULL) {
1353 		evt_err("Failed to create sym session pool");
1354 		ret = -ENOMEM;
1355 		goto err;
1356 	}
1357 
1358 	max_session_size = 0;
1359 	for (cdev_id = 0; cdev_id < cdev_count; cdev_id++) {
1360 		unsigned int session_size;
1361 
1362 		session_size =
1363 			rte_cryptodev_sym_get_private_session_size(cdev_id);
1364 		if (session_size > max_session_size)
1365 			max_session_size = session_size;
1366 	}
1367 
1368 	max_session_size += sizeof(union rte_event_crypto_metadata);
1369 	t->ca_sess_priv_pool = rte_mempool_create(
1370 		"ca_sess_priv_pool", nb_sessions, max_session_size, 0, 0, NULL,
1371 		NULL, NULL, NULL, SOCKET_ID_ANY, 0);
1372 	if (t->ca_sess_priv_pool == NULL) {
1373 		evt_err("failed to create sym session private pool");
1374 		ret = -ENOMEM;
1375 		goto err;
1376 	}
1377 
1378 	/*
1379 	 * Calculate number of needed queue pairs, based on the amount of
1380 	 * available number of logical cores and crypto devices. For instance,
1381 	 * if there are 4 cores and 2 crypto devices, 2 queue pairs will be set
1382 	 * up per device.
1383 	 */
1384 	nb_plcores = evt_nr_active_lcores(opt->plcores);
1385 	nb_qps = (nb_plcores % cdev_count) ? (nb_plcores / cdev_count) + 1 :
1386 					     nb_plcores / cdev_count;
1387 	for (cdev_id = 0; cdev_id < cdev_count; cdev_id++) {
1388 		struct rte_cryptodev_qp_conf qp_conf;
1389 		struct rte_cryptodev_config conf;
1390 		struct rte_cryptodev_info info;
1391 		int qp_id;
1392 
1393 		rte_cryptodev_info_get(cdev_id, &info);
1394 		if (nb_qps > info.max_nb_queue_pairs) {
1395 			evt_err("Not enough queue pairs per cryptodev (%u)",
1396 				nb_qps);
1397 			ret = -EINVAL;
1398 			goto err;
1399 		}
1400 
1401 		conf.nb_queue_pairs = nb_qps;
1402 		conf.socket_id = SOCKET_ID_ANY;
1403 		conf.ff_disable = RTE_CRYPTODEV_FF_SECURITY;
1404 
1405 		ret = rte_cryptodev_configure(cdev_id, &conf);
1406 		if (ret) {
1407 			evt_err("Failed to configure cryptodev (%u)", cdev_id);
1408 			goto err;
1409 		}
1410 
1411 		qp_conf.nb_descriptors = NB_CRYPTODEV_DESCRIPTORS;
1412 		qp_conf.mp_session = t->ca_sess_pool;
1413 		qp_conf.mp_session_private = t->ca_sess_priv_pool;
1414 
1415 		for (qp_id = 0; qp_id < conf.nb_queue_pairs; qp_id++) {
1416 			ret = rte_cryptodev_queue_pair_setup(
1417 				cdev_id, qp_id, &qp_conf,
1418 				rte_cryptodev_socket_id(cdev_id));
1419 			if (ret) {
1420 				evt_err("Failed to setup queue pairs on cryptodev %u\n",
1421 					cdev_id);
1422 				goto err;
1423 			}
1424 		}
1425 	}
1426 
1427 	return 0;
1428 err:
1429 	for (cdev_id = 0; cdev_id < cdev_count; cdev_id++)
1430 		rte_cryptodev_close(cdev_id);
1431 
1432 	rte_mempool_free(t->ca_op_pool);
1433 	rte_mempool_free(t->ca_sess_pool);
1434 	rte_mempool_free(t->ca_sess_priv_pool);
1435 	rte_mempool_free(t->ca_asym_sess_pool);
1436 
1437 	return ret;
1438 }
1439 
1440 void
1441 perf_cryptodev_destroy(struct evt_test *test, struct evt_options *opt)
1442 {
1443 	uint8_t cdev_id, cdev_count = rte_cryptodev_count();
1444 	struct test_perf *t = evt_test_priv(test);
1445 	uint16_t port;
1446 
1447 	if (opt->prod_type != EVT_PROD_TYPE_EVENT_CRYPTO_ADPTR)
1448 		return;
1449 
1450 	for (port = t->nb_workers; port < perf_nb_event_ports(opt); port++) {
1451 		struct rte_cryptodev_sym_session *sess;
1452 		struct prod_data *p = &t->prod[port];
1453 		uint32_t flow_id;
1454 		uint8_t cdev_id;
1455 
1456 		for (flow_id = 0; flow_id < t->nb_flows; flow_id++) {
1457 			sess = p->ca.crypto_sess[flow_id];
1458 			cdev_id = p->ca.cdev_id;
1459 			rte_cryptodev_sym_session_clear(cdev_id, sess);
1460 			rte_cryptodev_sym_session_free(sess);
1461 		}
1462 
1463 		rte_event_crypto_adapter_queue_pair_del(
1464 			TEST_PERF_CA_ID, p->ca.cdev_id, p->ca.cdev_qp_id);
1465 	}
1466 
1467 	rte_event_crypto_adapter_free(TEST_PERF_CA_ID);
1468 
1469 	for (cdev_id = 0; cdev_id < cdev_count; cdev_id++) {
1470 		rte_cryptodev_stop(cdev_id);
1471 		rte_cryptodev_close(cdev_id);
1472 	}
1473 
1474 	rte_mempool_free(t->ca_op_pool);
1475 	rte_mempool_free(t->ca_sess_pool);
1476 	rte_mempool_free(t->ca_sess_priv_pool);
1477 	rte_mempool_free(t->ca_asym_sess_pool);
1478 }
1479 
1480 int
1481 perf_mempool_setup(struct evt_test *test, struct evt_options *opt)
1482 {
1483 	struct test_perf *t = evt_test_priv(test);
1484 
1485 	if (opt->prod_type == EVT_PROD_TYPE_SYNT ||
1486 			opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR) {
1487 		t->pool = rte_mempool_create(test->name, /* mempool name */
1488 				opt->pool_sz, /* number of elements*/
1489 				sizeof(struct perf_elt), /* element size*/
1490 				512, /* cache size*/
1491 				0, NULL, NULL,
1492 				perf_elt_init, /* obj constructor */
1493 				NULL, opt->socket_id, 0); /* flags */
1494 	} else {
1495 		t->pool = rte_pktmbuf_pool_create(test->name, /* mempool name */
1496 				opt->pool_sz, /* number of elements*/
1497 				512, /* cache size*/
1498 				0,
1499 				RTE_MBUF_DEFAULT_BUF_SIZE,
1500 				opt->socket_id); /* flags */
1501 
1502 	}
1503 
1504 	if (t->pool == NULL) {
1505 		evt_err("failed to create mempool");
1506 		return -ENOMEM;
1507 	}
1508 
1509 	return 0;
1510 }
1511 
1512 void
1513 perf_mempool_destroy(struct evt_test *test, struct evt_options *opt)
1514 {
1515 	RTE_SET_USED(opt);
1516 	struct test_perf *t = evt_test_priv(test);
1517 
1518 	rte_mempool_free(t->pool);
1519 }
1520 
1521 int
1522 perf_test_setup(struct evt_test *test, struct evt_options *opt)
1523 {
1524 	void *test_perf;
1525 
1526 	test_perf = rte_zmalloc_socket(test->name, sizeof(struct test_perf),
1527 				RTE_CACHE_LINE_SIZE, opt->socket_id);
1528 	if (test_perf  == NULL) {
1529 		evt_err("failed to allocate test_perf memory");
1530 		goto nomem;
1531 	}
1532 	test->test_priv = test_perf;
1533 
1534 	struct test_perf *t = evt_test_priv(test);
1535 
1536 	if (opt->prod_type == EVT_PROD_TYPE_EVENT_TIMER_ADPTR) {
1537 		t->outstand_pkts = opt->nb_timers *
1538 			evt_nr_active_lcores(opt->plcores);
1539 		t->nb_pkts = opt->nb_timers;
1540 	} else {
1541 		t->outstand_pkts = opt->nb_pkts *
1542 			evt_nr_active_lcores(opt->plcores);
1543 		t->nb_pkts = opt->nb_pkts;
1544 	}
1545 
1546 	t->nb_workers = evt_nr_active_lcores(opt->wlcores);
1547 	t->done = false;
1548 	t->nb_flows = opt->nb_flows;
1549 	t->result = EVT_TEST_FAILED;
1550 	t->opt = opt;
1551 	memcpy(t->sched_type_list, opt->sched_type_list,
1552 			sizeof(opt->sched_type_list));
1553 	return 0;
1554 nomem:
1555 	return -ENOMEM;
1556 }
1557 
1558 void
1559 perf_test_destroy(struct evt_test *test, struct evt_options *opt)
1560 {
1561 	RTE_SET_USED(opt);
1562 
1563 	rte_free(test->test_priv);
1564 }
1565