[dpdk-dev] [PATCH v4 7/8] app/crypto-perf: support multiple queue pairs

[Pablo de Lara]

Add support for multiple queue pairs, when there are
more logical cores available than crypto devices enabled.
For instance, if there are 4 cores available and
2 crypto devices, each device will have two queue pairs.

This is useful to have multiple logical cores using
a single crypto device, without needing to initialize
a crypto device per core.

Signed-off-by: Pablo de Lara <pablo.de.lara.guarch at intel.com>
---
 app/test-crypto-perf/Makefile                    |   4 +
 app/test-crypto-perf/cperf_options.h             |   1 +
 app/test-crypto-perf/cperf_options_parsing.c     |   2 +
 app/test-crypto-perf/cperf_test_common.c         |  15 ++--
 app/test-crypto-perf/cperf_test_common.h         |   3 +-
 app/test-crypto-perf/cperf_test_latency.c        |   4 +-
 app/test-crypto-perf/cperf_test_pmd_cyclecount.c |   2 +-
 app/test-crypto-perf/cperf_test_throughput.c     |   4 +-
 app/test-crypto-perf/cperf_test_verify.c         |   4 +-
 app/test-crypto-perf/main.c                      | 100 +++++++++++++++++------
 10 files changed, 95 insertions(+), 44 deletions(-)

diff --git a/app/test-crypto-perf/Makefile b/app/test-crypto-perf/Makefile
index 25ae395..c75d7ed 100644
--- a/app/test-crypto-perf/Makefile
+++ b/app/test-crypto-perf/Makefile
@@ -47,4 +47,8 @@ SRCS-y += cperf_test_verify.c
 SRCS-y += cperf_test_vector_parsing.c
 SRCS-y += cperf_test_common.c
 
+ifeq ($(CONFIG_RTE_LIBRTE_PMD_CRYPTO_SCHEDULER),y)
+LDLIBS += -lrte_pmd_crypto_scheduler
+endif
+
 include $(RTE_SDK)/mk/rte.app.mk
diff --git a/app/test-crypto-perf/cperf_options.h b/app/test-crypto-perf/cperf_options.h
index 6d339f4..da4fb47 100644
--- a/app/test-crypto-perf/cperf_options.h
+++ b/app/test-crypto-perf/cperf_options.h
@@ -74,6 +74,7 @@ struct cperf_options {
 	uint32_t segment_sz;
 	uint32_t test_buffer_size;
 	uint32_t nb_descriptors;
+	uint16_t nb_qps;
 
 	uint32_t sessionless:1;
 	uint32_t out_of_place:1;
diff --git a/app/test-crypto-perf/cperf_options_parsing.c b/app/test-crypto-perf/cperf_options_parsing.c
index 89f86a2..e5fa4a8 100644
--- a/app/test-crypto-perf/cperf_options_parsing.c
+++ b/app/test-crypto-perf/cperf_options_parsing.c
@@ -747,6 +747,7 @@ cperf_options_default(struct cperf_options *opts)
 
 	strncpy(opts->device_type, "crypto_aesni_mb",
 			sizeof(opts->device_type));
+	opts->nb_qps = 1;
 
 	opts->op_type = CPERF_CIPHER_THEN_AUTH;
 
@@ -1032,6 +1033,7 @@ cperf_options_dump(struct cperf_options *opts)
 	printf("#\n");
 	printf("# cryptodev type: %s\n", opts->device_type);
 	printf("#\n");
+	printf("# number of queue pairs per device: %u\n", opts->nb_qps);
 	printf("# crypto operation: %s\n", cperf_op_type_strs[opts->op_type]);
 	printf("# sessionless: %s\n", opts->sessionless ? "yes" : "no");
 	printf("# out of place: %s\n", opts->out_of_place ? "yes" : "no");
diff --git a/app/test-crypto-perf/cperf_test_common.c b/app/test-crypto-perf/cperf_test_common.c
index 25eb970..65f0be7 100644
--- a/app/test-crypto-perf/cperf_test_common.c
+++ b/app/test-crypto-perf/cperf_test_common.c
@@ -108,7 +108,8 @@ cperf_mbuf_create(struct rte_mempool *mempool,
 int
 cperf_alloc_common_memory(const struct cperf_options *options,
 			const struct cperf_test_vector *test_vector,
-			uint8_t dev_id, size_t extra_op_priv_size,
+			uint8_t dev_id, uint16_t qp_id,
+			size_t extra_op_priv_size,
 			struct rte_mempool **pkt_mbuf_pool_in,
 			struct rte_mempool **pkt_mbuf_pool_out,
 			struct rte_mbuf ***mbufs_in,
@@ -118,8 +119,8 @@ cperf_alloc_common_memory(const struct cperf_options *options,
 	unsigned int mbuf_idx = 0;
 	char pool_name[32] = "";
 
-	snprintf(pool_name, sizeof(pool_name), "cperf_pool_in_cdev_%d",
-			dev_id);
+	snprintf(pool_name, sizeof(pool_name), "cperf_pool_in_cdev_%u_qp_%u",
+			dev_id, qp_id);
 
 	uint32_t max_size = options->max_buffer_size + options->digest_sz;
 	uint16_t segments_nb = (max_size % options->segment_sz) ?
@@ -153,8 +154,8 @@ cperf_alloc_common_memory(const struct cperf_options *options,
 			options->pool_sz), 0);
 
 	if (options->out_of_place == 1)	{
-		snprintf(pool_name, sizeof(pool_name), "cperf_pool_out_cdev_%d",
-				dev_id);
+		snprintf(pool_name, sizeof(pool_name), "cperf_pool_out_cdev_%u_qp_%u",
+				dev_id, qp_id);
 
 		*pkt_mbuf_pool_out = rte_pktmbuf_pool_create(
 				pool_name, options->pool_sz, 0, 0,
@@ -173,8 +174,8 @@ cperf_alloc_common_memory(const struct cperf_options *options,
 		}
 	}
 
-	snprintf(pool_name, sizeof(pool_name), "cperf_op_pool_cdev_%d",
-			dev_id);
+	snprintf(pool_name, sizeof(pool_name), "cperf_op_pool_cdev_%u_qp_%u",
+			dev_id, qp_id);
 
 	uint16_t priv_size = RTE_ALIGN_CEIL(test_vector->cipher_iv.length +
 		test_vector->auth_iv.length + test_vector->aead_iv.length +
diff --git a/app/test-crypto-perf/cperf_test_common.h b/app/test-crypto-perf/cperf_test_common.h
index 766d643..ad29431 100644
--- a/app/test-crypto-perf/cperf_test_common.h
+++ b/app/test-crypto-perf/cperf_test_common.h
@@ -43,7 +43,8 @@
 int
 cperf_alloc_common_memory(const struct cperf_options *options,
 			const struct cperf_test_vector *test_vector,
-			uint8_t dev_id, size_t extra_op_priv_size,
+			uint8_t dev_id, uint16_t qp_id,
+			size_t extra_op_priv_size,
 			struct rte_mempool **pkt_mbuf_pool_in,
 			struct rte_mempool **pkt_mbuf_pool_out,
 			struct rte_mbuf ***mbufs_in,
diff --git a/app/test-crypto-perf/cperf_test_latency.c b/app/test-crypto-perf/cperf_test_latency.c
index 2dc9c0c..313665e 100644
--- a/app/test-crypto-perf/cperf_test_latency.c
+++ b/app/test-crypto-perf/cperf_test_latency.c
@@ -124,7 +124,7 @@ cperf_latency_test_constructor(struct rte_mempool *sess_mp,
 	if (ctx->sess == NULL)
 		goto err;
 
-	if (cperf_alloc_common_memory(options, test_vector, dev_id,
+	if (cperf_alloc_common_memory(options, test_vector, dev_id, qp_id,
 			extra_op_priv_size,
 			&ctx->pkt_mbuf_pool_in, &ctx->pkt_mbuf_pool_out,
 			&ctx->mbufs_in, &ctx->mbufs_out,
@@ -417,7 +417,5 @@ cperf_latency_test_destructor(void *arg)
 	if (ctx == NULL)
 		return;
 
-	rte_cryptodev_stop(ctx->dev_id);
-
 	cperf_latency_test_free(ctx);
 }
diff --git a/app/test-crypto-perf/cperf_test_pmd_cyclecount.c b/app/test-crypto-perf/cperf_test_pmd_cyclecount.c
index 81f403c..13a270f 100644
--- a/app/test-crypto-perf/cperf_test_pmd_cyclecount.c
+++ b/app/test-crypto-perf/cperf_test_pmd_cyclecount.c
@@ -143,7 +143,7 @@ cperf_pmd_cyclecount_test_constructor(struct rte_mempool *sess_mp,
 	if (ctx->sess == NULL)
 		goto err;
 
-	if (cperf_alloc_common_memory(options, test_vector, dev_id, 0,
+	if (cperf_alloc_common_memory(options, test_vector, dev_id, qp_id, 0,
 			&ctx->pkt_mbuf_pool_in, &ctx->pkt_mbuf_pool_out,
 			&ctx->mbufs_in, &ctx->mbufs_out,
 			&ctx->crypto_op_pool) < 0)
diff --git a/app/test-crypto-perf/cperf_test_throughput.c b/app/test-crypto-perf/cperf_test_throughput.c
index 8d54642..bc7d889 100644
--- a/app/test-crypto-perf/cperf_test_throughput.c
+++ b/app/test-crypto-perf/cperf_test_throughput.c
@@ -107,7 +107,7 @@ cperf_throughput_test_constructor(struct rte_mempool *sess_mp,
 	if (ctx->sess == NULL)
 		goto err;
 
-	if (cperf_alloc_common_memory(options, test_vector, dev_id, 0,
+	if (cperf_alloc_common_memory(options, test_vector, dev_id, qp_id, 0,
 			&ctx->pkt_mbuf_pool_in, &ctx->pkt_mbuf_pool_out,
 			&ctx->mbufs_in, &ctx->mbufs_out,
 			&ctx->crypto_op_pool) < 0)
@@ -368,7 +368,5 @@ cperf_throughput_test_destructor(void *arg)
 	if (ctx == NULL)
 		return;
 
-	rte_cryptodev_stop(ctx->dev_id);
-
 	cperf_throughput_test_free(ctx);
 }
diff --git a/app/test-crypto-perf/cperf_test_verify.c b/app/test-crypto-perf/cperf_test_verify.c
index 9a806fa..b122022 100644
--- a/app/test-crypto-perf/cperf_test_verify.c
+++ b/app/test-crypto-perf/cperf_test_verify.c
@@ -111,7 +111,7 @@ cperf_verify_test_constructor(struct rte_mempool *sess_mp,
 	if (ctx->sess == NULL)
 		goto err;
 
-	if (cperf_alloc_common_memory(options, test_vector, dev_id, 0,
+	if (cperf_alloc_common_memory(options, test_vector, dev_id, qp_id, 0,
 			&ctx->pkt_mbuf_pool_in, &ctx->pkt_mbuf_pool_out,
 			&ctx->mbufs_in, &ctx->mbufs_out,
 			&ctx->crypto_op_pool) < 0)
@@ -464,7 +464,5 @@ cperf_verify_test_destructor(void *arg)
 	if (ctx == NULL)
 		return;
 
-	rte_cryptodev_stop(ctx->dev_id);
-
 	cperf_verify_test_free(ctx);
 }
diff --git a/app/test-crypto-perf/main.c b/app/test-crypto-perf/main.c
index ffa7180..aaa5830 100644
--- a/app/test-crypto-perf/main.c
+++ b/app/test-crypto-perf/main.c
@@ -35,6 +35,9 @@
 
 #include <rte_eal.h>
 #include <rte_cryptodev.h>
+#ifdef RTE_LIBRTE_PMD_CRYPTO_SCHEDULER
+#include <rte_cryptodev_scheduler.h>
+#endif
 
 #include "cperf.h"
 #include "cperf_options.h"
@@ -90,7 +93,7 @@ cperf_initialize_cryptodev(struct cperf_options *opts, uint8_t *enabled_cdevs,
 			struct rte_mempool *session_pool_socket[])
 {
 	uint8_t enabled_cdev_count = 0, nb_lcores, cdev_id;
-	unsigned int i;
+	unsigned int i, j;
 	int ret;
 
 	enabled_cdev_count = rte_cryptodev_devices_get(opts->device_type,
@@ -119,21 +122,53 @@ cperf_initialize_cryptodev(struct cperf_options *opts, uint8_t *enabled_cdevs,
 			max_sess_size = sess_size;
 	}
 
+	/*
+	 * Calculate number of needed queue pairs, based on the amount
+	 * of available number of logical cores and crypto devices.
+	 * For instance, if there are 4 cores and 2 crypto devices,
+	 * 2 queue pairs will be set up per device.
+	 */
+	opts->nb_qps = (nb_lcores % enabled_cdev_count) ?
+				(nb_lcores / enabled_cdev_count) + 1 :
+				nb_lcores / enabled_cdev_count;
+
 	for (i = 0; i < enabled_cdev_count &&
 			i < RTE_CRYPTO_MAX_DEVS; i++) {
 		cdev_id = enabled_cdevs[i];
+#ifdef RTE_LIBRTE_PMD_CRYPTO_SCHEDULER
+		/*
+		 * If multi-core scheduler is used, limit the number
+		 * of queue pairs to 1, as there is no way to know
+		 * how many cores are being used by the PMD, and
+		 * how many will be available for the application.
+		 */
+		if (!strcmp((const char *)opts->device_type, "crypto_scheduler") &&
+				rte_cryptodev_scheduler_mode_get(cdev_id) ==
+				CDEV_SCHED_MODE_MULTICORE)
+			opts->nb_qps = 1;
+#endif
+
+		struct rte_cryptodev_info cdev_info;
 		uint8_t socket_id = rte_cryptodev_socket_id(cdev_id);
 
+		rte_cryptodev_info_get(cdev_id, &cdev_info);
+		if (opts->nb_qps > cdev_info.max_nb_queue_pairs) {
+			printf("Number of needed queue pairs is higher "
+				"than the maximum number of queue pairs "
+				"per device.\n");
+			printf("Lower the number of cores or increase "
+				"the number of crypto devices\n");
+			return -EINVAL;
+		}
 		struct rte_cryptodev_config conf = {
-				.nb_queue_pairs = 1,
-				.socket_id = socket_id
+			.nb_queue_pairs = opts->nb_qps,
+			.socket_id = socket_id
 		};
 
 		struct rte_cryptodev_qp_conf qp_conf = {
-			    .nb_descriptors = opts->nb_descriptors
+			.nb_descriptors = opts->nb_descriptors
 		};
 
-
 		if (session_pool_socket[socket_id] == NULL) {
 			char mp_name[RTE_MEMPOOL_NAMESIZE];
 			struct rte_mempool *sess_mp;
@@ -165,14 +200,16 @@ cperf_initialize_cryptodev(struct cperf_options *opts, uint8_t *enabled_cdevs,
 			return -EINVAL;
 		}
 
-		ret = rte_cryptodev_queue_pair_setup(cdev_id, 0,
+		for (j = 0; j < opts->nb_qps; j++) {
+			ret = rte_cryptodev_queue_pair_setup(cdev_id, j,
 				&qp_conf, socket_id,
 				session_pool_socket[socket_id]);
 			if (ret < 0) {
 				printf("Failed to setup queue pair %u on "
-					"cryptodev %u",	0, cdev_id);
+					"cryptodev %u",	j, cdev_id);
 				return -EINVAL;
 			}
+		}
 
 		ret = rte_cryptodev_start(cdev_id);
 		if (ret < 0) {
@@ -417,11 +454,12 @@ main(int argc, char **argv)
 		goto err;
 	}
 
+	nb_cryptodevs = cperf_initialize_cryptodev(&opts, enabled_cdevs,
+			session_pool_socket);
+
 	if (!opts.silent)
 		cperf_options_dump(&opts);
 
-	nb_cryptodevs = cperf_initialize_cryptodev(&opts, enabled_cdevs,
-			session_pool_socket);
 	if (nb_cryptodevs < 1) {
 		RTE_LOG(ERR, USER1, "Failed to initialise requested crypto "
 				"device type\n");
@@ -471,23 +509,29 @@ main(int argc, char **argv)
 	if (!opts.silent)
 		show_test_vector(t_vec);
 
+	uint16_t total_nb_qps = nb_cryptodevs * opts.nb_qps;
+
 	i = 0;
+	uint8_t qp_id = 0, cdev_index = 0;
 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 
-		if (i == nb_cryptodevs)
+		if (i == total_nb_qps)
 			break;
 
-		cdev_id = enabled_cdevs[i];
+		cdev_id = enabled_cdevs[cdev_index];
 
 		uint8_t socket_id = rte_cryptodev_socket_id(cdev_id);
 
-		ctx[cdev_id] = cperf_testmap[opts.test].constructor(
-				session_pool_socket[socket_id], cdev_id, 0,
+		ctx[i] = cperf_testmap[opts.test].constructor(
+				session_pool_socket[socket_id], cdev_id, qp_id,
 				&opts, t_vec, &op_fns);
-		if (ctx[cdev_id] == NULL) {
+		if (ctx[i] == NULL) {
 			RTE_LOG(ERR, USER1, "Test run constructor failed\n");
 			goto err;
 		}
+		qp_id = (qp_id + 1) % opts.nb_qps;
+		if (qp_id == 0)
+			cdev_index++;
 		i++;
 	}
 
@@ -501,19 +545,17 @@ main(int argc, char **argv)
 		i = 0;
 		RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 
-			if (i == nb_cryptodevs)
+			if (i == total_nb_qps)
 				break;
 
-			cdev_id = enabled_cdevs[i];
-
 			rte_eal_remote_launch(cperf_testmap[opts.test].runner,
-				ctx[cdev_id], lcore_id);
+				ctx[i], lcore_id);
 			i++;
 		}
 		i = 0;
 		RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 
-			if (i == nb_cryptodevs)
+			if (i == total_nb_qps)
 				break;
 			rte_eal_wait_lcore(lcore_id);
 			i++;
@@ -532,15 +574,17 @@ main(int argc, char **argv)
 	i = 0;
 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 
-		if (i == nb_cryptodevs)
+		if (i == total_nb_qps)
 			break;
 
-		cdev_id = enabled_cdevs[i];
-
-		cperf_testmap[opts.test].destructor(ctx[cdev_id]);
+		cperf_testmap[opts.test].destructor(ctx[i]);
 		i++;
 	}
 
+	for (i = 0; i < nb_cryptodevs &&
+			i < RTE_CRYPTO_MAX_DEVS; i++)
+		rte_cryptodev_stop(enabled_cdevs[i]);
+
 	free_test_vector(t_vec, &opts);
 
 	printf("\n");
@@ -549,16 +593,20 @@ main(int argc, char **argv)
 err:
 	i = 0;
 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
-		if (i == nb_cryptodevs)
+		if (i == total_nb_qps)
 			break;
 
 		cdev_id = enabled_cdevs[i];
 
-		if (ctx[cdev_id] && cperf_testmap[opts.test].destructor)
-			cperf_testmap[opts.test].destructor(ctx[cdev_id]);
+		if (ctx[i] && cperf_testmap[opts.test].destructor)
+			cperf_testmap[opts.test].destructor(ctx[i]);
 		i++;
 	}
 
+	for (i = 0; i < nb_cryptodevs &&
+			i < RTE_CRYPTO_MAX_DEVS; i++)
+		rte_cryptodev_stop(enabled_cdevs[i]);
+
 	free_test_vector(t_vec, &opts);
 
 	printf("\n");
-- 
2.9.4