1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2016-2017 Intel Corporation. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 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 15 * distribution. 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. 19 * 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. 31 */ 32 33 #include <rte_malloc.h> 34 #include <rte_cycles.h> 35 #include <rte_crypto.h> 36 #include <rte_cryptodev.h> 37 38 #include "cperf_test_throughput.h" 39 #include "cperf_ops.h" 40 41 struct cperf_throughput_ctx { 42 uint8_t dev_id; 43 uint16_t qp_id; 44 uint8_t lcore_id; 45 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; 50 51 struct rte_mempool *crypto_op_pool; 52 53 struct rte_cryptodev_sym_session *sess; 54 55 cperf_populate_ops_t populate_ops; 56 57 const struct cperf_options *options; 58 const struct cperf_test_vector *test_vector; 59 }; 60 61 static void 62 cperf_throughput_test_free(struct cperf_throughput_ctx *ctx, uint32_t mbuf_nb) 63 { 64 uint32_t i; 65 66 if (ctx) { 67 if (ctx->sess) 68 rte_cryptodev_sym_session_free(ctx->dev_id, ctx->sess); 69 70 if (ctx->mbufs_in) { 71 for (i = 0; i < mbuf_nb; i++) 72 rte_pktmbuf_free(ctx->mbufs_in[i]); 73 74 rte_free(ctx->mbufs_in); 75 } 76 77 if (ctx->mbufs_out) { 78 for (i = 0; i < mbuf_nb; i++) { 79 if (ctx->mbufs_out[i] != NULL) 80 rte_pktmbuf_free(ctx->mbufs_out[i]); 81 } 82 83 rte_free(ctx->mbufs_out); 84 } 85 86 if (ctx->pkt_mbuf_pool_in) 87 rte_mempool_free(ctx->pkt_mbuf_pool_in); 88 89 if (ctx->pkt_mbuf_pool_out) 90 rte_mempool_free(ctx->pkt_mbuf_pool_out); 91 92 if (ctx->crypto_op_pool) 93 rte_mempool_free(ctx->crypto_op_pool); 94 95 rte_free(ctx); 96 } 97 } 98 99 static struct rte_mbuf * 100 cperf_mbuf_create(struct rte_mempool *mempool, 101 uint32_t segments_nb, 102 const struct cperf_options *options, 103 const struct cperf_test_vector *test_vector) 104 { 105 struct rte_mbuf *mbuf; 106 uint32_t segment_sz = options->max_buffer_size / segments_nb; 107 uint32_t last_sz = options->max_buffer_size % segments_nb; 108 uint8_t *mbuf_data; 109 uint8_t *test_data = 110 (options->cipher_op == RTE_CRYPTO_CIPHER_OP_ENCRYPT) ? 111 test_vector->plaintext.data : 112 test_vector->ciphertext.data; 113 114 mbuf = rte_pktmbuf_alloc(mempool); 115 if (mbuf == NULL) 116 goto error; 117 118 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, segment_sz); 119 if (mbuf_data == NULL) 120 goto error; 121 122 memcpy(mbuf_data, test_data, segment_sz); 123 test_data += segment_sz; 124 segments_nb--; 125 126 while (segments_nb) { 127 struct rte_mbuf *m; 128 129 m = rte_pktmbuf_alloc(mempool); 130 if (m == NULL) 131 goto error; 132 133 rte_pktmbuf_chain(mbuf, m); 134 135 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, segment_sz); 136 if (mbuf_data == NULL) 137 goto error; 138 139 memcpy(mbuf_data, test_data, segment_sz); 140 test_data += segment_sz; 141 segments_nb--; 142 } 143 144 if (last_sz) { 145 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, last_sz); 146 if (mbuf_data == NULL) 147 goto error; 148 149 memcpy(mbuf_data, test_data, last_sz); 150 } 151 152 if (options->op_type != CPERF_CIPHER_ONLY) { 153 mbuf_data = (uint8_t *)rte_pktmbuf_append(mbuf, 154 options->auth_digest_sz); 155 if (mbuf_data == NULL) 156 goto error; 157 } 158 159 if (options->op_type == CPERF_AEAD) { 160 uint8_t *aead = (uint8_t *)rte_pktmbuf_prepend(mbuf, 161 RTE_ALIGN_CEIL(options->auth_aad_sz, 16)); 162 163 if (aead == NULL) 164 goto error; 165 166 memcpy(aead, test_vector->aad.data, test_vector->aad.length); 167 } 168 169 return mbuf; 170 error: 171 if (mbuf != NULL) 172 rte_pktmbuf_free(mbuf); 173 174 return NULL; 175 } 176 177 void * 178 cperf_throughput_test_constructor(uint8_t dev_id, uint16_t qp_id, 179 const struct cperf_options *options, 180 const struct cperf_test_vector *test_vector, 181 const struct cperf_op_fns *op_fns) 182 { 183 struct cperf_throughput_ctx *ctx = NULL; 184 unsigned int mbuf_idx = 0; 185 char pool_name[32] = ""; 186 187 ctx = rte_malloc(NULL, sizeof(struct cperf_throughput_ctx), 0); 188 if (ctx == NULL) 189 goto err; 190 191 ctx->dev_id = dev_id; 192 ctx->qp_id = qp_id; 193 194 ctx->populate_ops = op_fns->populate_ops; 195 ctx->options = options; 196 ctx->test_vector = test_vector; 197 198 ctx->sess = op_fns->sess_create(dev_id, options, test_vector); 199 if (ctx->sess == NULL) 200 goto err; 201 202 snprintf(pool_name, sizeof(pool_name), "cperf_pool_in_cdev_%d", 203 dev_id); 204 205 ctx->pkt_mbuf_pool_in = rte_pktmbuf_pool_create(pool_name, 206 options->pool_sz * options->segments_nb, 0, 0, 207 RTE_PKTMBUF_HEADROOM + 208 RTE_CACHE_LINE_ROUNDUP( 209 (options->max_buffer_size / options->segments_nb) + 210 (options->max_buffer_size % options->segments_nb) + 211 options->auth_digest_sz), 212 rte_socket_id()); 213 214 if (ctx->pkt_mbuf_pool_in == NULL) 215 goto err; 216 217 /* Generate mbufs_in with plaintext populated for test */ 218 ctx->mbufs_in = rte_malloc(NULL, 219 (sizeof(struct rte_mbuf *) * ctx->options->pool_sz), 0); 220 221 for (mbuf_idx = 0; mbuf_idx < options->pool_sz; mbuf_idx++) { 222 ctx->mbufs_in[mbuf_idx] = cperf_mbuf_create( 223 ctx->pkt_mbuf_pool_in, options->segments_nb, 224 options, test_vector); 225 if (ctx->mbufs_in[mbuf_idx] == NULL) 226 goto err; 227 } 228 229 if (options->out_of_place == 1) { 230 231 snprintf(pool_name, sizeof(pool_name), "cperf_pool_out_cdev_%d", 232 dev_id); 233 234 ctx->pkt_mbuf_pool_out = rte_pktmbuf_pool_create( 235 pool_name, options->pool_sz, 0, 0, 236 RTE_PKTMBUF_HEADROOM + 237 RTE_CACHE_LINE_ROUNDUP( 238 options->max_buffer_size + 239 options->auth_digest_sz), 240 rte_socket_id()); 241 242 if (ctx->pkt_mbuf_pool_out == NULL) 243 goto err; 244 } 245 246 ctx->mbufs_out = rte_malloc(NULL, 247 (sizeof(struct rte_mbuf *) * 248 ctx->options->pool_sz), 0); 249 250 for (mbuf_idx = 0; mbuf_idx < options->pool_sz; mbuf_idx++) { 251 if (options->out_of_place == 1) { 252 ctx->mbufs_out[mbuf_idx] = cperf_mbuf_create( 253 ctx->pkt_mbuf_pool_out, 1, 254 options, test_vector); 255 if (ctx->mbufs_out[mbuf_idx] == NULL) 256 goto err; 257 } else { 258 ctx->mbufs_out[mbuf_idx] = NULL; 259 } 260 } 261 262 snprintf(pool_name, sizeof(pool_name), "cperf_op_pool_cdev_%d", 263 dev_id); 264 265 uint16_t priv_size = test_vector->iv.length; 266 267 ctx->crypto_op_pool = rte_crypto_op_pool_create(pool_name, 268 RTE_CRYPTO_OP_TYPE_SYMMETRIC, options->pool_sz, 269 512, priv_size, rte_socket_id()); 270 if (ctx->crypto_op_pool == NULL) 271 goto err; 272 273 return ctx; 274 err: 275 cperf_throughput_test_free(ctx, mbuf_idx); 276 277 return NULL; 278 } 279 280 int 281 cperf_throughput_test_runner(void *test_ctx) 282 { 283 struct cperf_throughput_ctx *ctx = test_ctx; 284 uint16_t test_burst_size; 285 uint8_t burst_size_idx = 0; 286 287 static int only_once; 288 289 struct rte_crypto_op *ops[ctx->options->max_burst_size]; 290 struct rte_crypto_op *ops_processed[ctx->options->max_burst_size]; 291 uint64_t i; 292 293 uint32_t lcore = rte_lcore_id(); 294 295 #ifdef CPERF_LINEARIZATION_ENABLE 296 struct rte_cryptodev_info dev_info; 297 int linearize = 0; 298 299 /* Check if source mbufs require coalescing */ 300 if (ctx->options->segments_nb > 1) { 301 rte_cryptodev_info_get(ctx->dev_id, &dev_info); 302 if ((dev_info.feature_flags & 303 RTE_CRYPTODEV_FF_MBUF_SCATTER_GATHER) == 0) 304 linearize = 1; 305 } 306 #endif /* CPERF_LINEARIZATION_ENABLE */ 307 308 ctx->lcore_id = lcore; 309 310 /* Warm up the host CPU before starting the test */ 311 for (i = 0; i < ctx->options->total_ops; i++) 312 rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, NULL, 0); 313 314 /* Get first size from range or list */ 315 if (ctx->options->inc_burst_size != 0) 316 test_burst_size = ctx->options->min_burst_size; 317 else 318 test_burst_size = ctx->options->burst_size_list[0]; 319 320 uint16_t iv_offset = sizeof(struct rte_crypto_op) + 321 sizeof(struct rte_crypto_sym_op); 322 323 while (test_burst_size <= ctx->options->max_burst_size) { 324 uint64_t ops_enqd = 0, ops_enqd_total = 0, ops_enqd_failed = 0; 325 uint64_t ops_deqd = 0, ops_deqd_total = 0, ops_deqd_failed = 0; 326 327 uint64_t m_idx = 0, tsc_start, tsc_end, tsc_duration; 328 329 uint16_t ops_unused = 0; 330 331 tsc_start = rte_rdtsc_precise(); 332 333 while (ops_enqd_total < ctx->options->total_ops) { 334 335 uint16_t burst_size = ((ops_enqd_total + test_burst_size) 336 <= ctx->options->total_ops) ? 337 test_burst_size : 338 ctx->options->total_ops - 339 ops_enqd_total; 340 341 uint16_t ops_needed = burst_size - ops_unused; 342 343 /* Allocate crypto ops from pool */ 344 if (ops_needed != rte_crypto_op_bulk_alloc( 345 ctx->crypto_op_pool, 346 RTE_CRYPTO_OP_TYPE_SYMMETRIC, 347 ops, ops_needed)) 348 return -1; 349 350 /* Setup crypto op, attach mbuf etc */ 351 (ctx->populate_ops)(ops, &ctx->mbufs_in[m_idx], 352 &ctx->mbufs_out[m_idx], 353 ops_needed, ctx->sess, ctx->options, 354 ctx->test_vector, iv_offset); 355 356 /** 357 * When ops_needed is smaller than ops_enqd, the 358 * unused ops need to be moved to the front for 359 * next round use. 360 */ 361 if (unlikely(ops_enqd > ops_needed)) { 362 size_t nb_b_to_mov = ops_unused * sizeof( 363 struct rte_crypto_op *); 364 365 memmove(&ops[ops_needed], &ops[ops_enqd], 366 nb_b_to_mov); 367 } 368 369 #ifdef CPERF_LINEARIZATION_ENABLE 370 if (linearize) { 371 /* PMD doesn't support scatter-gather and source buffer 372 * is segmented. 373 * We need to linearize it before enqueuing. 374 */ 375 for (i = 0; i < burst_size; i++) 376 rte_pktmbuf_linearize(ops[i]->sym->m_src); 377 } 378 #endif /* CPERF_LINEARIZATION_ENABLE */ 379 380 /* Enqueue burst of ops on crypto device */ 381 ops_enqd = rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, 382 ops, burst_size); 383 if (ops_enqd < burst_size) 384 ops_enqd_failed++; 385 386 /** 387 * Calculate number of ops not enqueued (mainly for hw 388 * accelerators whose ingress queue can fill up). 389 */ 390 ops_unused = burst_size - ops_enqd; 391 ops_enqd_total += ops_enqd; 392 393 394 /* Dequeue processed burst of ops from crypto device */ 395 ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id, 396 ops_processed, test_burst_size); 397 398 if (likely(ops_deqd)) { 399 /* free crypto ops so they can be reused. We don't free 400 * the mbufs here as we don't want to reuse them as 401 * the crypto operation will change the data and cause 402 * failures. 403 */ 404 rte_mempool_put_bulk(ctx->crypto_op_pool, 405 (void **)ops_processed, ops_deqd); 406 407 ops_deqd_total += ops_deqd; 408 } else { 409 /** 410 * Count dequeue polls which didn't return any 411 * processed operations. This statistic is mainly 412 * relevant to hw accelerators. 413 */ 414 ops_deqd_failed++; 415 } 416 417 m_idx += ops_needed; 418 m_idx = m_idx + test_burst_size > ctx->options->pool_sz ? 419 0 : m_idx; 420 } 421 422 /* Dequeue any operations still in the crypto device */ 423 424 while (ops_deqd_total < ctx->options->total_ops) { 425 /* Sending 0 length burst to flush sw crypto device */ 426 rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, NULL, 0); 427 428 /* dequeue burst */ 429 ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id, 430 ops_processed, test_burst_size); 431 if (ops_deqd == 0) 432 ops_deqd_failed++; 433 else { 434 rte_mempool_put_bulk(ctx->crypto_op_pool, 435 (void **)ops_processed, ops_deqd); 436 437 ops_deqd_total += ops_deqd; 438 } 439 } 440 441 tsc_end = rte_rdtsc_precise(); 442 tsc_duration = (tsc_end - tsc_start); 443 444 /* Calculate average operations processed per second */ 445 double ops_per_second = ((double)ctx->options->total_ops / 446 tsc_duration) * rte_get_tsc_hz(); 447 448 /* Calculate average throughput (Gbps) in bits per second */ 449 double throughput_gbps = ((ops_per_second * 450 ctx->options->test_buffer_size * 8) / 1000000000); 451 452 /* Calculate average cycles per packet */ 453 double cycles_per_packet = ((double)tsc_duration / 454 ctx->options->total_ops); 455 456 if (!ctx->options->csv) { 457 if (!only_once) 458 printf("%12s%12s%12s%12s%12s%12s%12s%12s%12s%12s\n\n", 459 "lcore id", "Buf Size", "Burst Size", 460 "Enqueued", "Dequeued", "Failed Enq", 461 "Failed Deq", "MOps", "Gbps", 462 "Cycles/Buf"); 463 only_once = 1; 464 465 printf("%12u%12u%12u%12"PRIu64"%12"PRIu64"%12"PRIu64 466 "%12"PRIu64"%12.4f%12.4f%12.2f\n", 467 ctx->lcore_id, 468 ctx->options->test_buffer_size, 469 test_burst_size, 470 ops_enqd_total, 471 ops_deqd_total, 472 ops_enqd_failed, 473 ops_deqd_failed, 474 ops_per_second/1000000, 475 throughput_gbps, 476 cycles_per_packet); 477 } else { 478 if (!only_once) 479 printf("# lcore id, Buffer Size(B)," 480 "Burst Size,Enqueued,Dequeued,Failed Enq," 481 "Failed Deq,Ops(Millions),Throughput(Gbps)," 482 "Cycles/Buf\n\n"); 483 only_once = 1; 484 485 printf("%10u;%10u;%u;%"PRIu64";%"PRIu64";%"PRIu64";%"PRIu64";" 486 "%.f3;%.f3;%.f3\n", 487 ctx->lcore_id, 488 ctx->options->test_buffer_size, 489 test_burst_size, 490 ops_enqd_total, 491 ops_deqd_total, 492 ops_enqd_failed, 493 ops_deqd_failed, 494 ops_per_second/1000000, 495 throughput_gbps, 496 cycles_per_packet); 497 } 498 499 /* Get next size from range or list */ 500 if (ctx->options->inc_burst_size != 0) 501 test_burst_size += ctx->options->inc_burst_size; 502 else { 503 if (++burst_size_idx == ctx->options->burst_size_count) 504 break; 505 test_burst_size = ctx->options->burst_size_list[burst_size_idx]; 506 } 507 508 } 509 510 return 0; 511 } 512 513 514 void 515 cperf_throughput_test_destructor(void *arg) 516 { 517 struct cperf_throughput_ctx *ctx = arg; 518 519 if (ctx == NULL) 520 return; 521 522 cperf_throughput_test_free(ctx, ctx->options->pool_sz); 523 } 524