build: set toolchain info during meson configure
[dpdk.git] / drivers / crypto / scheduler / scheduler_multicore.c
index 8757552..b2ce44c 100644 (file)
@@ -1,33 +1,5 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2017 Intel Corporation. All rights reserved.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Intel Corporation nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Intel Corporation
  */
 #include <unistd.h>
 
@@ -42,6 +14,8 @@
 
 #define MC_SCHED_BUFFER_SIZE 32
 
+#define CRYPTO_OP_STATUS_BIT_COMPLETE  0x80
+
 /** multi-core scheduler context */
 struct mc_scheduler_ctx {
        uint32_t num_workers;             /**< Number of workers polling */
@@ -136,10 +110,31 @@ static uint16_t
 schedule_dequeue_ordering(void *qp, struct rte_crypto_op **ops,
                uint16_t nb_ops)
 {
-       struct rte_ring *order_ring =
-                       ((struct scheduler_qp_ctx *)qp)->order_ring;
+       struct rte_ring *order_ring = ((struct scheduler_qp_ctx *)qp)->order_ring;
+       struct rte_crypto_op *op;
+       uint32_t nb_objs = rte_ring_count(order_ring);
+       uint32_t nb_ops_to_deq = 0;
+       uint32_t nb_ops_deqd = 0;
+
+       if (nb_objs > nb_ops)
+               nb_objs = nb_ops;
+
+       while (nb_ops_to_deq < nb_objs) {
+               SCHEDULER_GET_RING_OBJ(order_ring, nb_ops_to_deq, op);
+
+               if (!(op->status & CRYPTO_OP_STATUS_BIT_COMPLETE))
+                       break;
+
+               op->status &= ~CRYPTO_OP_STATUS_BIT_COMPLETE;
+               nb_ops_to_deq++;
+       }
 
-       return scheduler_order_drain(order_ring, ops, nb_ops);
+       if (nb_ops_to_deq) {
+               nb_ops_deqd = rte_ring_sc_dequeue_bulk(order_ring,
+                               (void **)ops, nb_ops_to_deq, NULL);
+       }
+
+       return nb_ops_deqd;
 }
 
 static int
@@ -168,11 +163,13 @@ mc_scheduler_worker(struct rte_cryptodev *dev)
        struct scheduler_slave *slave;
        struct rte_crypto_op *enq_ops[MC_SCHED_BUFFER_SIZE];
        struct rte_crypto_op *deq_ops[MC_SCHED_BUFFER_SIZE];
-       struct scheduler_session *sess0, *sess1, *sess2, *sess3;
        uint16_t processed_ops;
-       uint16_t left_op = 0;
-       uint16_t left_op_idx = 0;
+       uint16_t pending_enq_ops = 0;
+       uint16_t pending_enq_ops_idx = 0;
+       uint16_t pending_deq_ops = 0;
+       uint16_t pending_deq_ops_idx = 0;
        uint16_t inflight_ops = 0;
+       const uint8_t reordering_enabled = sched_ctx->reordering_enabled;
 
        for (i = 0; i < (int)sched_ctx->nb_wc; i++) {
                if (sched_ctx->wc_pool[i] == core_id) {
@@ -190,75 +187,51 @@ mc_scheduler_worker(struct rte_cryptodev *dev)
        deq_ring = mc_ctx->sched_deq_ring[worker_idx];
 
        while (!mc_ctx->stop_signal) {
-               if (left_op) {
+               if (pending_enq_ops) {
                        processed_ops =
                                rte_cryptodev_enqueue_burst(slave->dev_id,
-                                               slave->qp_id,
-                                               &enq_ops[left_op_idx], left_op);
-
-                       left_op -= processed_ops;
-                       left_op_idx += processed_ops;
+                                       slave->qp_id, &enq_ops[pending_enq_ops_idx],
+                                       pending_enq_ops);
+                       pending_enq_ops -= processed_ops;
+                       pending_enq_ops_idx += processed_ops;
+                       inflight_ops += processed_ops;
                } else {
-                       uint16_t nb_deq_ops = rte_ring_dequeue_burst(enq_ring,
-                               (void *)enq_ops, MC_SCHED_BUFFER_SIZE, NULL);
-                       if (nb_deq_ops) {
-                               uint16_t i;
-
-                               for (i = 0; i < nb_deq_ops && i < 4; i++)
-                                       rte_prefetch0(enq_ops[i]->sym->session);
-
-                               for (i = 0; (i < (nb_deq_ops - 8))
-                                                       && (nb_deq_ops > 8); i += 4) {
-                                       sess0 = (struct scheduler_session *)
-                                               enq_ops[i]->sym->session->_private;
-                                       sess1 = (struct scheduler_session *)
-                                               enq_ops[i+1]->sym->session->_private;
-                                       sess2 = (struct scheduler_session *)
-                                               enq_ops[i+2]->sym->session->_private;
-                                       sess3 = (struct scheduler_session *)
-                                               enq_ops[i+3]->sym->session->_private;
-
-                                       enq_ops[i]->sym->session =
-                                                       sess0->sessions[worker_idx];
-                                       enq_ops[i + 1]->sym->session =
-                                                       sess1->sessions[worker_idx];
-                                       enq_ops[i + 2]->sym->session =
-                                                       sess2->sessions[worker_idx];
-                                       enq_ops[i + 3]->sym->session =
-                                                       sess3->sessions[worker_idx];
-
-                                       rte_prefetch0(enq_ops[i + 4]->sym->session);
-                                       rte_prefetch0(enq_ops[i + 5]->sym->session);
-                                       rte_prefetch0(enq_ops[i + 6]->sym->session);
-                                       rte_prefetch0(enq_ops[i + 7]->sym->session);
-                               }
-
-                               for (; i < nb_deq_ops; i++) {
-                                       sess0 = (struct scheduler_session *)
-                                               enq_ops[i]->sym->session->_private;
-                                       enq_ops[i]->sym->session =
-                                                       sess0->sessions[worker_idx];
-                               }
-
-                               processed_ops = rte_cryptodev_enqueue_burst(slave->dev_id,
-                                               slave->qp_id, enq_ops, nb_deq_ops);
-
-                               if (unlikely(processed_ops < nb_deq_ops)) {
-                                       left_op = nb_deq_ops - processed_ops;
-                                       left_op_idx = processed_ops;
-                               }
-
-                               inflight_ops += processed_ops;
+                       processed_ops = rte_ring_dequeue_burst(enq_ring, (void *)enq_ops,
+                                                       MC_SCHED_BUFFER_SIZE, NULL);
+                       if (processed_ops) {
+                               pending_enq_ops_idx = rte_cryptodev_enqueue_burst(
+                                                       slave->dev_id, slave->qp_id,
+                                                       enq_ops, processed_ops);
+                               pending_enq_ops = processed_ops - pending_enq_ops_idx;
+                               inflight_ops += pending_enq_ops_idx;
                        }
                }
 
-               if (inflight_ops > 0) {
+               if (pending_deq_ops) {
+                       processed_ops = rte_ring_enqueue_burst(
+                                       deq_ring, (void *)&deq_ops[pending_deq_ops_idx],
+                                                       pending_deq_ops, NULL);
+                       pending_deq_ops -= processed_ops;
+                       pending_deq_ops_idx += processed_ops;
+               } else if (inflight_ops) {
                        processed_ops = rte_cryptodev_dequeue_burst(slave->dev_id,
                                        slave->qp_id, deq_ops, MC_SCHED_BUFFER_SIZE);
                        if (processed_ops) {
-                               uint16_t nb_enq_ops = rte_ring_enqueue_burst(deq_ring,
-                                       (void *)deq_ops, processed_ops, NULL);
-                               inflight_ops -= nb_enq_ops;
+                               inflight_ops -= processed_ops;
+                               if (reordering_enabled) {
+                                       uint16_t j;
+
+                                       for (j = 0; j < processed_ops; j++) {
+                                               deq_ops[j]->status |=
+                                                       CRYPTO_OP_STATUS_BIT_COMPLETE;
+                                       }
+                               } else {
+                                       pending_deq_ops_idx = rte_ring_enqueue_burst(
+                                               deq_ring, (void *)deq_ops, processed_ops,
+                                               NULL);
+                                       pending_deq_ops = processed_ops -
+                                                               pending_deq_ops_idx;
+                               }
                        }
                }
 
@@ -319,10 +292,11 @@ scheduler_stop(struct rte_cryptodev *dev)
 {
        struct scheduler_ctx *sched_ctx = dev->data->dev_private;
        struct mc_scheduler_ctx *mc_ctx = sched_ctx->private_ctx;
+       uint16_t i;
 
        mc_ctx->stop_signal = 1;
 
-       for (uint16_t i = 0; i < sched_ctx->nb_wc; i++)
+       for (i = 0; i < sched_ctx->nb_wc; i++)
                rte_eal_wait_lcore(sched_ctx->wc_pool[i]);
 
        return 0;
@@ -355,6 +329,7 @@ scheduler_create_private_ctx(struct rte_cryptodev *dev)
 {
        struct scheduler_ctx *sched_ctx = dev->data->dev_private;
        struct mc_scheduler_ctx *mc_ctx;
+       uint16_t i;
 
        if (sched_ctx->private_ctx)
                rte_free(sched_ctx->private_ctx);
@@ -367,7 +342,7 @@ scheduler_create_private_ctx(struct rte_cryptodev *dev)
        }
 
        mc_ctx->num_workers = sched_ctx->nb_wc;
-       for (uint16_t i = 0; i < sched_ctx->nb_wc; i++) {
+       for (i = 0; i < sched_ctx->nb_wc; i++) {
                char r_name[16];
 
                snprintf(r_name, sizeof(r_name), MC_SCHED_ENQ_RING_NAME_PREFIX "%u", i);