4 * Copyright(c) 2016-2017 Intel Corporation. All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
10 * * Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * * Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
16 * * Neither the name of Intel Corporation nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include <rte_malloc.h>
34 #include <rte_cycles.h>
35 #include <rte_crypto.h>
36 #include <rte_cryptodev.h>
38 #include "cperf_test_throughput.h"
39 #include "cperf_ops.h"
41 struct cperf_throughput_ctx {
46 struct rte_mempool *pkt_mbuf_pool_in;
47 struct rte_mempool *pkt_mbuf_pool_out;
48 struct rte_mbuf **mbufs_in;
49 struct rte_mbuf **mbufs_out;
51 struct rte_mempool *crypto_op_pool;
53 struct rte_cryptodev_sym_session *sess;
55 cperf_populate_ops_t populate_ops;
57 const struct cperf_options *options;
58 const struct cperf_test_vector *test_vector;
62 cperf_throughput_test_free(struct cperf_throughput_ctx *ctx, uint32_t mbuf_nb)
68 rte_cryptodev_sym_session_clear(ctx->dev_id, ctx->sess);
69 rte_cryptodev_sym_session_free(ctx->sess);
73 for (i = 0; i < mbuf_nb; i++)
74 rte_pktmbuf_free(ctx->mbufs_in[i]);
76 rte_free(ctx->mbufs_in);
80 for (i = 0; i < mbuf_nb; i++) {
81 if (ctx->mbufs_out[i] != NULL)
82 rte_pktmbuf_free(ctx->mbufs_out[i]);
85 rte_free(ctx->mbufs_out);
88 if (ctx->pkt_mbuf_pool_in)
89 rte_mempool_free(ctx->pkt_mbuf_pool_in);
91 if (ctx->pkt_mbuf_pool_out)
92 rte_mempool_free(ctx->pkt_mbuf_pool_out);
94 if (ctx->crypto_op_pool)
95 rte_mempool_free(ctx->crypto_op_pool);
101 static struct rte_mbuf *
102 cperf_mbuf_create(struct rte_mempool *mempool,
103 uint32_t segments_nb,
104 const struct cperf_options *options,
105 const struct cperf_test_vector *test_vector)
107 struct rte_mbuf *mbuf;
108 uint32_t segment_sz = options->max_buffer_size / segments_nb;
109 uint32_t last_sz = options->max_buffer_size % segments_nb;
112 (options->cipher_op == RTE_CRYPTO_CIPHER_OP_ENCRYPT) ?
113 test_vector->plaintext.data :
114 test_vector->ciphertext.data;
116 mbuf = rte_pktmbuf_alloc(mempool);
120 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, segment_sz);
121 if (mbuf_data == NULL)
124 memcpy(mbuf_data, test_data, segment_sz);
125 test_data += segment_sz;
128 while (segments_nb) {
131 m = rte_pktmbuf_alloc(mempool);
135 rte_pktmbuf_chain(mbuf, m);
137 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, segment_sz);
138 if (mbuf_data == NULL)
141 memcpy(mbuf_data, test_data, segment_sz);
142 test_data += segment_sz;
147 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, last_sz);
148 if (mbuf_data == NULL)
151 memcpy(mbuf_data, test_data, last_sz);
154 if (options->op_type != CPERF_CIPHER_ONLY) {
155 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf,
157 if (mbuf_data == NULL)
161 if (options->op_type == CPERF_AEAD) {
162 uint8_t *aead = (uint8_t *)rte_pktmbuf_prepend(mbuf,
163 RTE_ALIGN_CEIL(options->aead_aad_sz, 16));
168 memcpy(aead, test_vector->aad.data, test_vector->aad.length);
174 rte_pktmbuf_free(mbuf);
180 cperf_throughput_test_constructor(struct rte_mempool *sess_mp,
181 uint8_t dev_id, uint16_t qp_id,
182 const struct cperf_options *options,
183 const struct cperf_test_vector *test_vector,
184 const struct cperf_op_fns *op_fns)
186 struct cperf_throughput_ctx *ctx = NULL;
187 unsigned int mbuf_idx = 0;
188 char pool_name[32] = "";
190 ctx = rte_malloc(NULL, sizeof(struct cperf_throughput_ctx), 0);
194 ctx->dev_id = dev_id;
197 ctx->populate_ops = op_fns->populate_ops;
198 ctx->options = options;
199 ctx->test_vector = test_vector;
201 /* IV goes at the end of the cryptop operation */
202 uint16_t iv_offset = sizeof(struct rte_crypto_op) +
203 sizeof(struct rte_crypto_sym_op);
205 ctx->sess = op_fns->sess_create(sess_mp, dev_id, options, test_vector,
207 if (ctx->sess == NULL)
210 snprintf(pool_name, sizeof(pool_name), "cperf_pool_in_cdev_%d",
213 ctx->pkt_mbuf_pool_in = rte_pktmbuf_pool_create(pool_name,
214 options->pool_sz * options->segments_nb, 0, 0,
215 RTE_PKTMBUF_HEADROOM +
216 RTE_CACHE_LINE_ROUNDUP(
217 (options->max_buffer_size / options->segments_nb) +
218 (options->max_buffer_size % options->segments_nb) +
222 if (ctx->pkt_mbuf_pool_in == NULL)
225 /* Generate mbufs_in with plaintext populated for test */
226 ctx->mbufs_in = rte_malloc(NULL,
227 (sizeof(struct rte_mbuf *) * ctx->options->pool_sz), 0);
229 for (mbuf_idx = 0; mbuf_idx < options->pool_sz; mbuf_idx++) {
230 ctx->mbufs_in[mbuf_idx] = cperf_mbuf_create(
231 ctx->pkt_mbuf_pool_in, options->segments_nb,
232 options, test_vector);
233 if (ctx->mbufs_in[mbuf_idx] == NULL)
237 if (options->out_of_place == 1) {
239 snprintf(pool_name, sizeof(pool_name), "cperf_pool_out_cdev_%d",
242 ctx->pkt_mbuf_pool_out = rte_pktmbuf_pool_create(
243 pool_name, options->pool_sz, 0, 0,
244 RTE_PKTMBUF_HEADROOM +
245 RTE_CACHE_LINE_ROUNDUP(
246 options->max_buffer_size +
250 if (ctx->pkt_mbuf_pool_out == NULL)
254 ctx->mbufs_out = rte_malloc(NULL,
255 (sizeof(struct rte_mbuf *) *
256 ctx->options->pool_sz), 0);
258 for (mbuf_idx = 0; mbuf_idx < options->pool_sz; mbuf_idx++) {
259 if (options->out_of_place == 1) {
260 ctx->mbufs_out[mbuf_idx] = cperf_mbuf_create(
261 ctx->pkt_mbuf_pool_out, 1,
262 options, test_vector);
263 if (ctx->mbufs_out[mbuf_idx] == NULL)
266 ctx->mbufs_out[mbuf_idx] = NULL;
270 snprintf(pool_name, sizeof(pool_name), "cperf_op_pool_cdev_%d",
273 uint16_t priv_size = test_vector->cipher_iv.length +
274 test_vector->auth_iv.length;
276 ctx->crypto_op_pool = rte_crypto_op_pool_create(pool_name,
277 RTE_CRYPTO_OP_TYPE_SYMMETRIC, options->pool_sz,
278 512, priv_size, rte_socket_id());
279 if (ctx->crypto_op_pool == NULL)
284 cperf_throughput_test_free(ctx, mbuf_idx);
290 cperf_throughput_test_runner(void *test_ctx)
292 struct cperf_throughput_ctx *ctx = test_ctx;
293 uint16_t test_burst_size;
294 uint8_t burst_size_idx = 0;
296 static int only_once;
298 struct rte_crypto_op *ops[ctx->options->max_burst_size];
299 struct rte_crypto_op *ops_processed[ctx->options->max_burst_size];
302 uint32_t lcore = rte_lcore_id();
304 #ifdef CPERF_LINEARIZATION_ENABLE
305 struct rte_cryptodev_info dev_info;
308 /* Check if source mbufs require coalescing */
309 if (ctx->options->segments_nb > 1) {
310 rte_cryptodev_info_get(ctx->dev_id, &dev_info);
311 if ((dev_info.feature_flags &
312 RTE_CRYPTODEV_FF_MBUF_SCATTER_GATHER) == 0)
315 #endif /* CPERF_LINEARIZATION_ENABLE */
317 ctx->lcore_id = lcore;
319 /* Warm up the host CPU before starting the test */
320 for (i = 0; i < ctx->options->total_ops; i++)
321 rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, NULL, 0);
323 /* Get first size from range or list */
324 if (ctx->options->inc_burst_size != 0)
325 test_burst_size = ctx->options->min_burst_size;
327 test_burst_size = ctx->options->burst_size_list[0];
329 uint16_t iv_offset = sizeof(struct rte_crypto_op) +
330 sizeof(struct rte_crypto_sym_op);
332 while (test_burst_size <= ctx->options->max_burst_size) {
333 uint64_t ops_enqd = 0, ops_enqd_total = 0, ops_enqd_failed = 0;
334 uint64_t ops_deqd = 0, ops_deqd_total = 0, ops_deqd_failed = 0;
336 uint64_t m_idx = 0, tsc_start, tsc_end, tsc_duration;
338 uint16_t ops_unused = 0;
340 tsc_start = rte_rdtsc_precise();
342 while (ops_enqd_total < ctx->options->total_ops) {
344 uint16_t burst_size = ((ops_enqd_total + test_burst_size)
345 <= ctx->options->total_ops) ?
347 ctx->options->total_ops -
350 uint16_t ops_needed = burst_size - ops_unused;
352 /* Allocate crypto ops from pool */
353 if (ops_needed != rte_crypto_op_bulk_alloc(
355 RTE_CRYPTO_OP_TYPE_SYMMETRIC,
358 "Failed to allocate more crypto operations "
359 "from the the crypto operation pool.\n"
360 "Consider increasing the pool size "
365 /* Setup crypto op, attach mbuf etc */
366 (ctx->populate_ops)(ops, &ctx->mbufs_in[m_idx],
367 &ctx->mbufs_out[m_idx],
368 ops_needed, ctx->sess, ctx->options,
369 ctx->test_vector, iv_offset);
372 * When ops_needed is smaller than ops_enqd, the
373 * unused ops need to be moved to the front for
376 if (unlikely(ops_enqd > ops_needed)) {
377 size_t nb_b_to_mov = ops_unused * sizeof(
378 struct rte_crypto_op *);
380 memmove(&ops[ops_needed], &ops[ops_enqd],
384 #ifdef CPERF_LINEARIZATION_ENABLE
386 /* PMD doesn't support scatter-gather and source buffer
388 * We need to linearize it before enqueuing.
390 for (i = 0; i < burst_size; i++)
391 rte_pktmbuf_linearize(ops[i]->sym->m_src);
393 #endif /* CPERF_LINEARIZATION_ENABLE */
395 /* Enqueue burst of ops on crypto device */
396 ops_enqd = rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id,
398 if (ops_enqd < burst_size)
402 * Calculate number of ops not enqueued (mainly for hw
403 * accelerators whose ingress queue can fill up).
405 ops_unused = burst_size - ops_enqd;
406 ops_enqd_total += ops_enqd;
409 /* Dequeue processed burst of ops from crypto device */
410 ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id,
411 ops_processed, test_burst_size);
413 if (likely(ops_deqd)) {
414 /* free crypto ops so they can be reused. We don't free
415 * the mbufs here as we don't want to reuse them as
416 * the crypto operation will change the data and cause
419 rte_mempool_put_bulk(ctx->crypto_op_pool,
420 (void **)ops_processed, ops_deqd);
422 ops_deqd_total += ops_deqd;
425 * Count dequeue polls which didn't return any
426 * processed operations. This statistic is mainly
427 * relevant to hw accelerators.
433 m_idx = m_idx + test_burst_size > ctx->options->pool_sz ?
437 /* Dequeue any operations still in the crypto device */
439 while (ops_deqd_total < ctx->options->total_ops) {
440 /* Sending 0 length burst to flush sw crypto device */
441 rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, NULL, 0);
444 ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id,
445 ops_processed, test_burst_size);
449 rte_mempool_put_bulk(ctx->crypto_op_pool,
450 (void **)ops_processed, ops_deqd);
452 ops_deqd_total += ops_deqd;
456 tsc_end = rte_rdtsc_precise();
457 tsc_duration = (tsc_end - tsc_start);
459 /* Calculate average operations processed per second */
460 double ops_per_second = ((double)ctx->options->total_ops /
461 tsc_duration) * rte_get_tsc_hz();
463 /* Calculate average throughput (Gbps) in bits per second */
464 double throughput_gbps = ((ops_per_second *
465 ctx->options->test_buffer_size * 8) / 1000000000);
467 /* Calculate average cycles per packet */
468 double cycles_per_packet = ((double)tsc_duration /
469 ctx->options->total_ops);
471 if (!ctx->options->csv) {
473 printf("%12s%12s%12s%12s%12s%12s%12s%12s%12s%12s\n\n",
474 "lcore id", "Buf Size", "Burst Size",
475 "Enqueued", "Dequeued", "Failed Enq",
476 "Failed Deq", "MOps", "Gbps",
480 printf("%12u%12u%12u%12"PRIu64"%12"PRIu64"%12"PRIu64
481 "%12"PRIu64"%12.4f%12.4f%12.2f\n",
483 ctx->options->test_buffer_size,
489 ops_per_second/1000000,
494 printf("#lcore id,Buffer Size(B),"
495 "Burst Size,Enqueued,Dequeued,Failed Enq,"
496 "Failed Deq,Ops(Millions),Throughput(Gbps),"
500 printf("%u;%u;%u;%"PRIu64";%"PRIu64";%"PRIu64";%"PRIu64";"
503 ctx->options->test_buffer_size,
509 ops_per_second/1000000,
514 /* Get next size from range or list */
515 if (ctx->options->inc_burst_size != 0)
516 test_burst_size += ctx->options->inc_burst_size;
518 if (++burst_size_idx == ctx->options->burst_size_count)
520 test_burst_size = ctx->options->burst_size_list[burst_size_idx];
530 cperf_throughput_test_destructor(void *arg)
532 struct cperf_throughput_ctx *ctx = arg;
537 rte_cryptodev_stop(ctx->dev_id);
539 cperf_throughput_test_free(ctx, ctx->options->pool_sz);