xref: /dpdk/drivers/crypto/scheduler/scheduler_multicore.c (revision e977e4199a8d6bab72cf94e154adcad1fb964e5e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Intel Corporation
3  */
4 #include <unistd.h>
5 
6 #include <rte_cryptodev.h>
7 #include <rte_malloc.h>
8 
9 #include "rte_cryptodev_scheduler_operations.h"
10 #include "scheduler_pmd_private.h"
11 
12 #define MC_SCHED_ENQ_RING_NAME_PREFIX	"MCS_ENQR_"
13 #define MC_SCHED_DEQ_RING_NAME_PREFIX	"MCS_DEQR_"
14 
15 #define MC_SCHED_BUFFER_SIZE 32
16 
17 #define CRYPTO_OP_STATUS_BIT_COMPLETE	0x80
18 
19 /** multi-core scheduler context */
20 struct mc_scheduler_ctx {
21 	uint32_t num_workers;             /**< Number of workers polling */
22 	uint32_t stop_signal;
23 
24 	struct rte_ring *sched_enq_ring[RTE_CRYPTODEV_SCHEDULER_MAX_NB_WORKER_CORES];
25 	struct rte_ring *sched_deq_ring[RTE_CRYPTODEV_SCHEDULER_MAX_NB_WORKER_CORES];
26 };
27 
28 struct mc_scheduler_qp_ctx {
29 	struct scheduler_slave slaves[RTE_CRYPTODEV_SCHEDULER_MAX_NB_SLAVES];
30 	uint32_t nb_slaves;
31 
32 	uint32_t last_enq_worker_idx;
33 	uint32_t last_deq_worker_idx;
34 
35 	struct mc_scheduler_ctx *mc_private_ctx;
36 };
37 
38 static uint16_t
39 schedule_enqueue(void *qp, struct rte_crypto_op **ops, uint16_t nb_ops)
40 {
41 	struct mc_scheduler_qp_ctx *mc_qp_ctx =
42 			((struct scheduler_qp_ctx *)qp)->private_qp_ctx;
43 	struct mc_scheduler_ctx *mc_ctx = mc_qp_ctx->mc_private_ctx;
44 	uint32_t worker_idx = mc_qp_ctx->last_enq_worker_idx;
45 	uint16_t i, processed_ops = 0;
46 
47 	if (unlikely(nb_ops == 0))
48 		return 0;
49 
50 	for (i = 0; i <  mc_ctx->num_workers && nb_ops != 0; i++) {
51 		struct rte_ring *enq_ring = mc_ctx->sched_enq_ring[worker_idx];
52 		uint16_t nb_queue_ops = rte_ring_enqueue_burst(enq_ring,
53 			(void *)(&ops[processed_ops]), nb_ops, NULL);
54 
55 		nb_ops -= nb_queue_ops;
56 		processed_ops += nb_queue_ops;
57 
58 		if (++worker_idx == mc_ctx->num_workers)
59 			worker_idx = 0;
60 	}
61 	mc_qp_ctx->last_enq_worker_idx = worker_idx;
62 
63 	return processed_ops;
64 }
65 
66 static uint16_t
67 schedule_enqueue_ordering(void *qp, struct rte_crypto_op **ops,
68 		uint16_t nb_ops)
69 {
70 	struct rte_ring *order_ring =
71 			((struct scheduler_qp_ctx *)qp)->order_ring;
72 	uint16_t nb_ops_to_enq = get_max_enqueue_order_count(order_ring,
73 			nb_ops);
74 	uint16_t nb_ops_enqd = schedule_enqueue(qp, ops,
75 			nb_ops_to_enq);
76 
77 	scheduler_order_insert(order_ring, ops, nb_ops_enqd);
78 
79 	return nb_ops_enqd;
80 }
81 
82 
83 static uint16_t
84 schedule_dequeue(void *qp, struct rte_crypto_op **ops, uint16_t nb_ops)
85 {
86 	struct mc_scheduler_qp_ctx *mc_qp_ctx =
87 			((struct scheduler_qp_ctx *)qp)->private_qp_ctx;
88 	struct mc_scheduler_ctx *mc_ctx = mc_qp_ctx->mc_private_ctx;
89 	uint32_t worker_idx = mc_qp_ctx->last_deq_worker_idx;
90 	uint16_t i, processed_ops = 0;
91 
92 	for (i = 0; i < mc_ctx->num_workers && nb_ops != 0; i++) {
93 		struct rte_ring *deq_ring = mc_ctx->sched_deq_ring[worker_idx];
94 		uint16_t nb_deq_ops = rte_ring_dequeue_burst(deq_ring,
95 			(void *)(&ops[processed_ops]), nb_ops, NULL);
96 
97 		nb_ops -= nb_deq_ops;
98 		processed_ops += nb_deq_ops;
99 		if (++worker_idx == mc_ctx->num_workers)
100 			worker_idx = 0;
101 	}
102 
103 	mc_qp_ctx->last_deq_worker_idx = worker_idx;
104 
105 	return processed_ops;
106 
107 }
108 
109 static uint16_t
110 schedule_dequeue_ordering(void *qp, struct rte_crypto_op **ops,
111 		uint16_t nb_ops)
112 {
113 	struct rte_ring *order_ring = ((struct scheduler_qp_ctx *)qp)->order_ring;
114 	struct rte_crypto_op *op;
115 	uint32_t nb_objs = rte_ring_count(order_ring);
116 	uint32_t nb_ops_to_deq = 0;
117 	uint32_t nb_ops_deqd = 0;
118 
119 	if (nb_objs > nb_ops)
120 		nb_objs = nb_ops;
121 
122 	while (nb_ops_to_deq < nb_objs) {
123 		SCHEDULER_GET_RING_OBJ(order_ring, nb_ops_to_deq, op);
124 
125 		if (!(op->status & CRYPTO_OP_STATUS_BIT_COMPLETE))
126 			break;
127 
128 		op->status &= ~CRYPTO_OP_STATUS_BIT_COMPLETE;
129 		nb_ops_to_deq++;
130 	}
131 
132 	if (nb_ops_to_deq) {
133 		nb_ops_deqd = rte_ring_sc_dequeue_bulk(order_ring,
134 				(void **)ops, nb_ops_to_deq, NULL);
135 	}
136 
137 	return nb_ops_deqd;
138 }
139 
140 static int
141 slave_attach(__rte_unused struct rte_cryptodev *dev,
142 		__rte_unused uint8_t slave_id)
143 {
144 	return 0;
145 }
146 
147 static int
148 slave_detach(__rte_unused struct rte_cryptodev *dev,
149 		__rte_unused uint8_t slave_id)
150 {
151 	return 0;
152 }
153 
154 static int
155 mc_scheduler_worker(struct rte_cryptodev *dev)
156 {
157 	struct scheduler_ctx *sched_ctx = dev->data->dev_private;
158 	struct mc_scheduler_ctx *mc_ctx = sched_ctx->private_ctx;
159 	struct rte_ring *enq_ring;
160 	struct rte_ring *deq_ring;
161 	uint32_t core_id = rte_lcore_id();
162 	int i, worker_idx = -1;
163 	struct scheduler_slave *slave;
164 	struct rte_crypto_op *enq_ops[MC_SCHED_BUFFER_SIZE];
165 	struct rte_crypto_op *deq_ops[MC_SCHED_BUFFER_SIZE];
166 	uint16_t processed_ops;
167 	uint16_t pending_enq_ops = 0;
168 	uint16_t pending_enq_ops_idx = 0;
169 	uint16_t pending_deq_ops = 0;
170 	uint16_t pending_deq_ops_idx = 0;
171 	uint16_t inflight_ops = 0;
172 	const uint8_t reordering_enabled = sched_ctx->reordering_enabled;
173 
174 	for (i = 0; i < (int)sched_ctx->nb_wc; i++) {
175 		if (sched_ctx->wc_pool[i] == core_id) {
176 			worker_idx = i;
177 			break;
178 		}
179 	}
180 	if (worker_idx == -1) {
181 		CS_LOG_ERR("worker on core %u:cannot find worker index!\n", core_id);
182 		return -1;
183 	}
184 
185 	slave = &sched_ctx->slaves[worker_idx];
186 	enq_ring = mc_ctx->sched_enq_ring[worker_idx];
187 	deq_ring = mc_ctx->sched_deq_ring[worker_idx];
188 
189 	while (!mc_ctx->stop_signal) {
190 		if (pending_enq_ops) {
191 			processed_ops =
192 				rte_cryptodev_enqueue_burst(slave->dev_id,
193 					slave->qp_id, &enq_ops[pending_enq_ops_idx],
194 					pending_enq_ops);
195 			pending_enq_ops -= processed_ops;
196 			pending_enq_ops_idx += processed_ops;
197 			inflight_ops += processed_ops;
198 		} else {
199 			processed_ops = rte_ring_dequeue_burst(enq_ring, (void *)enq_ops,
200 							MC_SCHED_BUFFER_SIZE, NULL);
201 			if (processed_ops) {
202 				pending_enq_ops_idx = rte_cryptodev_enqueue_burst(
203 							slave->dev_id, slave->qp_id,
204 							enq_ops, processed_ops);
205 				pending_enq_ops = processed_ops - pending_enq_ops_idx;
206 				inflight_ops += pending_enq_ops_idx;
207 			}
208 		}
209 
210 		if (pending_deq_ops) {
211 			processed_ops = rte_ring_enqueue_burst(
212 					deq_ring, (void *)&deq_ops[pending_deq_ops_idx],
213 							pending_deq_ops, NULL);
214 			pending_deq_ops -= processed_ops;
215 			pending_deq_ops_idx += processed_ops;
216 		} else if (inflight_ops) {
217 			processed_ops = rte_cryptodev_dequeue_burst(slave->dev_id,
218 					slave->qp_id, deq_ops, MC_SCHED_BUFFER_SIZE);
219 			if (processed_ops) {
220 				inflight_ops -= processed_ops;
221 				if (reordering_enabled) {
222 					uint16_t j;
223 
224 					for (j = 0; j < processed_ops; j++) {
225 						deq_ops[j]->status |=
226 							CRYPTO_OP_STATUS_BIT_COMPLETE;
227 					}
228 				} else {
229 					pending_deq_ops_idx = rte_ring_enqueue_burst(
230 						deq_ring, (void *)deq_ops, processed_ops,
231 						NULL);
232 					pending_deq_ops = processed_ops -
233 								pending_deq_ops_idx;
234 				}
235 			}
236 		}
237 
238 		rte_pause();
239 	}
240 
241 	return 0;
242 }
243 
244 static int
245 scheduler_start(struct rte_cryptodev *dev)
246 {
247 	struct scheduler_ctx *sched_ctx = dev->data->dev_private;
248 	struct mc_scheduler_ctx *mc_ctx = sched_ctx->private_ctx;
249 	uint16_t i;
250 
251 	mc_ctx->stop_signal = 0;
252 
253 	for (i = 0; i < sched_ctx->nb_wc; i++)
254 		rte_eal_remote_launch(
255 			(lcore_function_t *)mc_scheduler_worker, dev,
256 					sched_ctx->wc_pool[i]);
257 
258 	if (sched_ctx->reordering_enabled) {
259 		dev->enqueue_burst = &schedule_enqueue_ordering;
260 		dev->dequeue_burst = &schedule_dequeue_ordering;
261 	} else {
262 		dev->enqueue_burst = &schedule_enqueue;
263 		dev->dequeue_burst = &schedule_dequeue;
264 	}
265 
266 	for (i = 0; i < dev->data->nb_queue_pairs; i++) {
267 		struct scheduler_qp_ctx *qp_ctx = dev->data->queue_pairs[i];
268 		struct mc_scheduler_qp_ctx *mc_qp_ctx =
269 				qp_ctx->private_qp_ctx;
270 		uint32_t j;
271 
272 		memset(mc_qp_ctx->slaves, 0,
273 				RTE_CRYPTODEV_SCHEDULER_MAX_NB_SLAVES *
274 				sizeof(struct scheduler_slave));
275 		for (j = 0; j < sched_ctx->nb_slaves; j++) {
276 			mc_qp_ctx->slaves[j].dev_id =
277 					sched_ctx->slaves[j].dev_id;
278 			mc_qp_ctx->slaves[j].qp_id = i;
279 		}
280 
281 		mc_qp_ctx->nb_slaves = sched_ctx->nb_slaves;
282 
283 		mc_qp_ctx->last_enq_worker_idx = 0;
284 		mc_qp_ctx->last_deq_worker_idx = 0;
285 	}
286 
287 	return 0;
288 }
289 
290 static int
291 scheduler_stop(struct rte_cryptodev *dev)
292 {
293 	struct scheduler_ctx *sched_ctx = dev->data->dev_private;
294 	struct mc_scheduler_ctx *mc_ctx = sched_ctx->private_ctx;
295 	uint16_t i;
296 
297 	mc_ctx->stop_signal = 1;
298 
299 	for (i = 0; i < sched_ctx->nb_wc; i++)
300 		rte_eal_wait_lcore(sched_ctx->wc_pool[i]);
301 
302 	return 0;
303 }
304 
305 static int
306 scheduler_config_qp(struct rte_cryptodev *dev, uint16_t qp_id)
307 {
308 	struct scheduler_qp_ctx *qp_ctx = dev->data->queue_pairs[qp_id];
309 	struct mc_scheduler_qp_ctx *mc_qp_ctx;
310 	struct scheduler_ctx *sched_ctx = dev->data->dev_private;
311 	struct mc_scheduler_ctx *mc_ctx = sched_ctx->private_ctx;
312 
313 	mc_qp_ctx = rte_zmalloc_socket(NULL, sizeof(*mc_qp_ctx), 0,
314 			rte_socket_id());
315 	if (!mc_qp_ctx) {
316 		CS_LOG_ERR("failed allocate memory for private queue pair");
317 		return -ENOMEM;
318 	}
319 
320 	mc_qp_ctx->mc_private_ctx = mc_ctx;
321 	qp_ctx->private_qp_ctx = (void *)mc_qp_ctx;
322 
323 
324 	return 0;
325 }
326 
327 static int
328 scheduler_create_private_ctx(struct rte_cryptodev *dev)
329 {
330 	struct scheduler_ctx *sched_ctx = dev->data->dev_private;
331 	struct mc_scheduler_ctx *mc_ctx = NULL;
332 	uint16_t i;
333 
334 	if (sched_ctx->private_ctx) {
335 		rte_free(sched_ctx->private_ctx);
336 		sched_ctx->private_ctx = NULL;
337 	}
338 
339 	mc_ctx = rte_zmalloc_socket(NULL, sizeof(struct mc_scheduler_ctx), 0,
340 			rte_socket_id());
341 	if (!mc_ctx) {
342 		CS_LOG_ERR("failed allocate memory");
343 		return -ENOMEM;
344 	}
345 
346 	mc_ctx->num_workers = sched_ctx->nb_wc;
347 	for (i = 0; i < sched_ctx->nb_wc; i++) {
348 		char r_name[16];
349 
350 		snprintf(r_name, sizeof(r_name), MC_SCHED_ENQ_RING_NAME_PREFIX "%u", i);
351 		mc_ctx->sched_enq_ring[i] = rte_ring_create(r_name, PER_SLAVE_BUFF_SIZE,
352 					rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
353 		if (!mc_ctx->sched_enq_ring[i]) {
354 			CS_LOG_ERR("Cannot create ring for worker %u", i);
355 			goto exit;
356 		}
357 		snprintf(r_name, sizeof(r_name), MC_SCHED_DEQ_RING_NAME_PREFIX "%u", i);
358 		mc_ctx->sched_deq_ring[i] = rte_ring_create(r_name, PER_SLAVE_BUFF_SIZE,
359 					rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
360 		if (!mc_ctx->sched_deq_ring[i]) {
361 			CS_LOG_ERR("Cannot create ring for worker %u", i);
362 			goto exit;
363 		}
364 	}
365 
366 	sched_ctx->private_ctx = (void *)mc_ctx;
367 
368 	return 0;
369 
370 exit:
371 	for (i = 0; i < sched_ctx->nb_wc; i++) {
372 		rte_ring_free(mc_ctx->sched_enq_ring[i]);
373 		rte_ring_free(mc_ctx->sched_deq_ring[i]);
374 	}
375 	rte_free(mc_ctx);
376 
377 	return -1;
378 }
379 
380 struct rte_cryptodev_scheduler_ops scheduler_mc_ops = {
381 	slave_attach,
382 	slave_detach,
383 	scheduler_start,
384 	scheduler_stop,
385 	scheduler_config_qp,
386 	scheduler_create_private_ctx,
387 	NULL,	/* option_set */
388 	NULL	/* option_get */
389 };
390 
391 struct rte_cryptodev_scheduler mc_scheduler = {
392 		.name = "multicore-scheduler",
393 		.description = "scheduler which will run burst across multiple cpu cores",
394 		.mode = CDEV_SCHED_MODE_MULTICORE,
395 		.ops = &scheduler_mc_ops
396 };
397 
398 struct rte_cryptodev_scheduler *multicore_scheduler = &mc_scheduler;
399