xref: /dpdk/drivers/baseband/turbo_sw/bbdev_turbo_software.c (revision 429d8d434e9b1f2d5625800c165900eb9ee9e422)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2017 Intel Corporation
3  */
4 
5 #include <string.h>
6 
7 #include <rte_common.h>
8 #include <rte_bus_vdev.h>
9 #include <rte_malloc.h>
10 #include <rte_ring.h>
11 #include <rte_kvargs.h>
12 
13 #include <rte_bbdev.h>
14 #include <rte_bbdev_pmd.h>
15 
16 #include <phy_turbo.h>
17 #include <phy_crc.h>
18 #include <phy_rate_match.h>
19 #include <divide.h>
20 
21 #define DRIVER_NAME turbo_sw
22 
23 /* Turbo SW PMD logging ID */
24 static int bbdev_turbo_sw_logtype;
25 
26 /* Helper macro for logging */
27 #define rte_bbdev_log(level, fmt, ...) \
28 	rte_log(RTE_LOG_ ## level, bbdev_turbo_sw_logtype, fmt "\n", \
29 		##__VA_ARGS__)
30 
31 #define rte_bbdev_log_debug(fmt, ...) \
32 	rte_bbdev_log(DEBUG, RTE_STR(__LINE__) ":%s() " fmt, __func__, \
33 		##__VA_ARGS__)
34 
35 /* Number of columns in sub-block interleaver (36.212, section 5.1.4.1.1) */
36 #define C_SUBBLOCK (32)
37 #define MAX_TB_SIZE (391656)
38 #define MAX_CB_SIZE (6144)
39 #define MAX_KW (18528)
40 
41 /* private data structure */
42 struct bbdev_private {
43 	unsigned int max_nb_queues;  /**< Max number of queues */
44 };
45 
46 /*  Initialisation params structure that can be used by Turbo SW driver */
47 struct turbo_sw_params {
48 	int socket_id;  /*< Turbo SW device socket */
49 	uint16_t queues_num;  /*< Turbo SW device queues number */
50 };
51 
52 /* Accecptable params for Turbo SW devices */
53 #define TURBO_SW_MAX_NB_QUEUES_ARG  "max_nb_queues"
54 #define TURBO_SW_SOCKET_ID_ARG      "socket_id"
55 
56 static const char * const turbo_sw_valid_params[] = {
57 	TURBO_SW_MAX_NB_QUEUES_ARG,
58 	TURBO_SW_SOCKET_ID_ARG
59 };
60 
61 /* queue */
62 struct turbo_sw_queue {
63 	/* Ring for processed (encoded/decoded) operations which are ready to
64 	 * be dequeued.
65 	 */
66 	struct rte_ring *processed_pkts;
67 	/* Stores input for turbo encoder (used when CRC attachment is
68 	 * performed
69 	 */
70 	uint8_t *enc_in;
71 	/* Stores output from turbo encoder */
72 	uint8_t *enc_out;
73 	/* Alpha gamma buf for bblib_turbo_decoder() function */
74 	int8_t *ag;
75 	/* Temp buf for bblib_turbo_decoder() function */
76 	uint16_t *code_block;
77 	/* Input buf for bblib_rate_dematching_lte() function */
78 	uint8_t *deint_input;
79 	/* Output buf for bblib_rate_dematching_lte() function */
80 	uint8_t *deint_output;
81 	/* Output buf for bblib_turbodec_adapter_lte() function */
82 	uint8_t *adapter_output;
83 	/* Operation type of this queue */
84 	enum rte_bbdev_op_type type;
85 } __rte_cache_aligned;
86 
87 /* Calculate index based on Table 5.1.3-3 from TS34.212 */
88 static inline int32_t
89 compute_idx(uint16_t k)
90 {
91 	int32_t result = 0;
92 
93 	if (k < 40 || k > MAX_CB_SIZE)
94 		return -1;
95 
96 	if (k > 2048) {
97 		if ((k - 2048) % 64 != 0)
98 			result = -1;
99 
100 		result = 124 + (k - 2048) / 64;
101 	} else if (k <= 512) {
102 		if ((k - 40) % 8 != 0)
103 			result = -1;
104 
105 		result = (k - 40) / 8 + 1;
106 	} else if (k <= 1024) {
107 		if ((k - 512) % 16 != 0)
108 			result = -1;
109 
110 		result = 60 + (k - 512) / 16;
111 	} else { /* 1024 < k <= 2048 */
112 		if ((k - 1024) % 32 != 0)
113 			result = -1;
114 
115 		result = 92 + (k - 1024) / 32;
116 	}
117 
118 	return result;
119 }
120 
121 /* Read flag value 0/1 from bitmap */
122 static inline bool
123 check_bit(uint32_t bitmap, uint32_t bitmask)
124 {
125 	return bitmap & bitmask;
126 }
127 
128 /* Get device info */
129 static void
130 info_get(struct rte_bbdev *dev, struct rte_bbdev_driver_info *dev_info)
131 {
132 	struct bbdev_private *internals = dev->data->dev_private;
133 
134 	static const struct rte_bbdev_op_cap bbdev_capabilities[] = {
135 		{
136 			.type = RTE_BBDEV_OP_TURBO_DEC,
137 			.cap.turbo_dec = {
138 				.capability_flags =
139 					RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE |
140 					RTE_BBDEV_TURBO_POS_LLR_1_BIT_IN |
141 					RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN |
142 					RTE_BBDEV_TURBO_CRC_TYPE_24B |
143 					RTE_BBDEV_TURBO_EARLY_TERMINATION,
144 				.num_buffers_src = RTE_BBDEV_MAX_CODE_BLOCKS,
145 				.num_buffers_hard_out =
146 						RTE_BBDEV_MAX_CODE_BLOCKS,
147 				.num_buffers_soft_out = 0,
148 			}
149 		},
150 		{
151 			.type   = RTE_BBDEV_OP_TURBO_ENC,
152 			.cap.turbo_enc = {
153 				.capability_flags =
154 						RTE_BBDEV_TURBO_CRC_24B_ATTACH |
155 						RTE_BBDEV_TURBO_CRC_24A_ATTACH |
156 						RTE_BBDEV_TURBO_RATE_MATCH |
157 						RTE_BBDEV_TURBO_RV_INDEX_BYPASS,
158 				.num_buffers_src = RTE_BBDEV_MAX_CODE_BLOCKS,
159 				.num_buffers_dst = RTE_BBDEV_MAX_CODE_BLOCKS,
160 			}
161 		},
162 		RTE_BBDEV_END_OF_CAPABILITIES_LIST()
163 	};
164 
165 	static struct rte_bbdev_queue_conf default_queue_conf = {
166 		.queue_size = RTE_BBDEV_QUEUE_SIZE_LIMIT,
167 	};
168 
169 	static const enum rte_cpu_flag_t cpu_flag = RTE_CPUFLAG_SSE4_2;
170 
171 	default_queue_conf.socket = dev->data->socket_id;
172 
173 	dev_info->driver_name = RTE_STR(DRIVER_NAME);
174 	dev_info->max_num_queues = internals->max_nb_queues;
175 	dev_info->queue_size_lim = RTE_BBDEV_QUEUE_SIZE_LIMIT;
176 	dev_info->hardware_accelerated = false;
177 	dev_info->max_queue_priority = 0;
178 	dev_info->default_queue_conf = default_queue_conf;
179 	dev_info->capabilities = bbdev_capabilities;
180 	dev_info->cpu_flag_reqs = &cpu_flag;
181 	dev_info->min_alignment = 64;
182 
183 	rte_bbdev_log_debug("got device info from %u\n", dev->data->dev_id);
184 }
185 
186 /* Release queue */
187 static int
188 q_release(struct rte_bbdev *dev, uint16_t q_id)
189 {
190 	struct turbo_sw_queue *q = dev->data->queues[q_id].queue_private;
191 
192 	if (q != NULL) {
193 		rte_ring_free(q->processed_pkts);
194 		rte_free(q->enc_out);
195 		rte_free(q->enc_in);
196 		rte_free(q->ag);
197 		rte_free(q->code_block);
198 		rte_free(q->deint_input);
199 		rte_free(q->deint_output);
200 		rte_free(q->adapter_output);
201 		rte_free(q);
202 		dev->data->queues[q_id].queue_private = NULL;
203 	}
204 
205 	rte_bbdev_log_debug("released device queue %u:%u",
206 			dev->data->dev_id, q_id);
207 	return 0;
208 }
209 
210 /* Setup a queue */
211 static int
212 q_setup(struct rte_bbdev *dev, uint16_t q_id,
213 		const struct rte_bbdev_queue_conf *queue_conf)
214 {
215 	int ret;
216 	struct turbo_sw_queue *q;
217 	char name[RTE_RING_NAMESIZE];
218 
219 	/* Allocate the queue data structure. */
220 	q = rte_zmalloc_socket(RTE_STR(DRIVER_NAME), sizeof(*q),
221 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
222 	if (q == NULL) {
223 		rte_bbdev_log(ERR, "Failed to allocate queue memory");
224 		return -ENOMEM;
225 	}
226 
227 	/* Allocate memory for encoder output. */
228 	ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_enc_out%u:%u",
229 			dev->data->dev_id, q_id);
230 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
231 		rte_bbdev_log(ERR,
232 				"Creating queue name for device %u queue %u failed",
233 				dev->data->dev_id, q_id);
234 		return -ENAMETOOLONG;
235 	}
236 	q->enc_out = rte_zmalloc_socket(name,
237 			((MAX_TB_SIZE >> 3) + 3) * sizeof(*q->enc_out) * 3,
238 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
239 	if (q->enc_out == NULL) {
240 		rte_bbdev_log(ERR,
241 			"Failed to allocate queue memory for %s", name);
242 		goto free_q;
243 	}
244 
245 	/* Allocate memory for rate matching output. */
246 	ret = snprintf(name, RTE_RING_NAMESIZE,
247 			RTE_STR(DRIVER_NAME)"_enc_in%u:%u", dev->data->dev_id,
248 			q_id);
249 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
250 		rte_bbdev_log(ERR,
251 				"Creating queue name for device %u queue %u failed",
252 				dev->data->dev_id, q_id);
253 		return -ENAMETOOLONG;
254 	}
255 	q->enc_in = rte_zmalloc_socket(name,
256 			(MAX_CB_SIZE >> 3) * sizeof(*q->enc_in),
257 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
258 	if (q->enc_in == NULL) {
259 		rte_bbdev_log(ERR,
260 			"Failed to allocate queue memory for %s", name);
261 		goto free_q;
262 	}
263 
264 	/* Allocate memory for Aplha Gamma temp buffer. */
265 	ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_ag%u:%u",
266 			dev->data->dev_id, q_id);
267 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
268 		rte_bbdev_log(ERR,
269 				"Creating queue name for device %u queue %u failed",
270 				dev->data->dev_id, q_id);
271 		return -ENAMETOOLONG;
272 	}
273 	q->ag = rte_zmalloc_socket(name,
274 			MAX_CB_SIZE * 10 * sizeof(*q->ag),
275 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
276 	if (q->ag == NULL) {
277 		rte_bbdev_log(ERR,
278 			"Failed to allocate queue memory for %s", name);
279 		goto free_q;
280 	}
281 
282 	/* Allocate memory for code block temp buffer. */
283 	ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_cb%u:%u",
284 			dev->data->dev_id, q_id);
285 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
286 		rte_bbdev_log(ERR,
287 				"Creating queue name for device %u queue %u failed",
288 				dev->data->dev_id, q_id);
289 		return -ENAMETOOLONG;
290 	}
291 	q->code_block = rte_zmalloc_socket(name,
292 			(6144 >> 3) * sizeof(*q->code_block),
293 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
294 	if (q->code_block == NULL) {
295 		rte_bbdev_log(ERR,
296 			"Failed to allocate queue memory for %s", name);
297 		goto free_q;
298 	}
299 
300 	/* Allocate memory for Deinterleaver input. */
301 	ret = snprintf(name, RTE_RING_NAMESIZE,
302 			RTE_STR(DRIVER_NAME)"_deint_input%u:%u",
303 			dev->data->dev_id, q_id);
304 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
305 		rte_bbdev_log(ERR,
306 				"Creating queue name for device %u queue %u failed",
307 				dev->data->dev_id, q_id);
308 		return -ENAMETOOLONG;
309 	}
310 	q->deint_input = rte_zmalloc_socket(name,
311 			MAX_KW * sizeof(*q->deint_input),
312 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
313 	if (q->deint_input == NULL) {
314 		rte_bbdev_log(ERR,
315 			"Failed to allocate queue memory for %s", name);
316 		goto free_q;
317 	}
318 
319 	/* Allocate memory for Deinterleaver output. */
320 	ret = snprintf(name, RTE_RING_NAMESIZE,
321 			RTE_STR(DRIVER_NAME)"_deint_output%u:%u",
322 			dev->data->dev_id, q_id);
323 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
324 		rte_bbdev_log(ERR,
325 				"Creating queue name for device %u queue %u failed",
326 				dev->data->dev_id, q_id);
327 		return -ENAMETOOLONG;
328 	}
329 	q->deint_output = rte_zmalloc_socket(NULL,
330 			MAX_KW * sizeof(*q->deint_output),
331 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
332 	if (q->deint_output == NULL) {
333 		rte_bbdev_log(ERR,
334 			"Failed to allocate queue memory for %s", name);
335 		goto free_q;
336 	}
337 
338 	/* Allocate memory for Adapter output. */
339 	ret = snprintf(name, RTE_RING_NAMESIZE,
340 			RTE_STR(DRIVER_NAME)"_adapter_output%u:%u",
341 			dev->data->dev_id, q_id);
342 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
343 		rte_bbdev_log(ERR,
344 				"Creating queue name for device %u queue %u failed",
345 				dev->data->dev_id, q_id);
346 		return -ENAMETOOLONG;
347 	}
348 	q->adapter_output = rte_zmalloc_socket(NULL,
349 			MAX_CB_SIZE * 6 * sizeof(*q->adapter_output),
350 			RTE_CACHE_LINE_SIZE, queue_conf->socket);
351 	if (q->adapter_output == NULL) {
352 		rte_bbdev_log(ERR,
353 			"Failed to allocate queue memory for %s", name);
354 		goto free_q;
355 	}
356 
357 	/* Create ring for packets awaiting to be dequeued. */
358 	ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"%u:%u",
359 			dev->data->dev_id, q_id);
360 	if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
361 		rte_bbdev_log(ERR,
362 				"Creating queue name for device %u queue %u failed",
363 				dev->data->dev_id, q_id);
364 		return -ENAMETOOLONG;
365 	}
366 	q->processed_pkts = rte_ring_create(name, queue_conf->queue_size,
367 			queue_conf->socket, RING_F_SP_ENQ | RING_F_SC_DEQ);
368 	if (q->processed_pkts == NULL) {
369 		rte_bbdev_log(ERR, "Failed to create ring for %s", name);
370 		goto free_q;
371 	}
372 
373 	q->type = queue_conf->op_type;
374 
375 	dev->data->queues[q_id].queue_private = q;
376 	rte_bbdev_log_debug("setup device queue %s", name);
377 	return 0;
378 
379 free_q:
380 	rte_ring_free(q->processed_pkts);
381 	rte_free(q->enc_out);
382 	rte_free(q->enc_in);
383 	rte_free(q->ag);
384 	rte_free(q->code_block);
385 	rte_free(q->deint_input);
386 	rte_free(q->deint_output);
387 	rte_free(q->adapter_output);
388 	rte_free(q);
389 	return -EFAULT;
390 }
391 
392 static const struct rte_bbdev_ops pmd_ops = {
393 	.info_get = info_get,
394 	.queue_setup = q_setup,
395 	.queue_release = q_release
396 };
397 
398 /* Checks if the encoder input buffer is correct.
399  * Returns 0 if it's valid, -1 otherwise.
400  */
401 static inline int
402 is_enc_input_valid(const uint16_t k, const int32_t k_idx,
403 		const uint16_t in_length)
404 {
405 	if (k_idx < 0) {
406 		rte_bbdev_log(ERR, "K Index is invalid");
407 		return -1;
408 	}
409 
410 	if (in_length - (k >> 3) < 0) {
411 		rte_bbdev_log(ERR,
412 				"Mismatch between input length (%u bytes) and K (%u bits)",
413 				in_length, k);
414 		return -1;
415 	}
416 
417 	if (k > MAX_CB_SIZE) {
418 		rte_bbdev_log(ERR, "CB size (%u) is too big, max: %d",
419 				k, MAX_CB_SIZE);
420 		return -1;
421 	}
422 
423 	return 0;
424 }
425 
426 /* Checks if the decoder input buffer is correct.
427  * Returns 0 if it's valid, -1 otherwise.
428  */
429 static inline int
430 is_dec_input_valid(int32_t k_idx, int16_t kw, int16_t in_length)
431 {
432 	if (k_idx < 0) {
433 		rte_bbdev_log(ERR, "K index is invalid");
434 		return -1;
435 	}
436 
437 	if (in_length - kw < 0) {
438 		rte_bbdev_log(ERR,
439 				"Mismatch between input length (%u) and kw (%u)",
440 				in_length, kw);
441 		return -1;
442 	}
443 
444 	if (kw > MAX_KW) {
445 		rte_bbdev_log(ERR, "Input length (%u) is too big, max: %d",
446 				kw, MAX_KW);
447 		return -1;
448 	}
449 
450 	return 0;
451 }
452 
453 static inline void
454 process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
455 		uint8_t r, uint8_t c, uint16_t k, uint16_t ncb,
456 		uint32_t e, struct rte_mbuf *m_in, struct rte_mbuf *m_out,
457 		uint16_t in_offset, uint16_t out_offset, uint16_t total_left)
458 {
459 	int ret;
460 	int16_t k_idx;
461 	uint16_t m;
462 	uint8_t *in, *out0, *out1, *out2, *tmp_out, *rm_out;
463 	uint64_t first_3_bytes = 0;
464 	struct rte_bbdev_op_turbo_enc *enc = &op->turbo_enc;
465 	struct bblib_crc_request crc_req;
466 	struct bblib_crc_response crc_resp;
467 	struct bblib_turbo_encoder_request turbo_req;
468 	struct bblib_turbo_encoder_response turbo_resp;
469 	struct bblib_rate_match_dl_request rm_req;
470 	struct bblib_rate_match_dl_response rm_resp;
471 
472 	k_idx = compute_idx(k);
473 	in = rte_pktmbuf_mtod_offset(m_in, uint8_t *, in_offset);
474 
475 	/* CRC24A (for TB) */
476 	if ((enc->op_flags & RTE_BBDEV_TURBO_CRC_24A_ATTACH) &&
477 		(enc->code_block_mode == 1)) {
478 		ret = is_enc_input_valid(k - 24, k_idx, total_left);
479 		if (ret != 0) {
480 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
481 			return;
482 		}
483 		crc_req.data = in;
484 		crc_req.len = (k - 24) >> 3;
485 		/* Check if there is a room for CRC bits. If not use
486 		 * the temporary buffer.
487 		 */
488 		if (rte_pktmbuf_append(m_in, 3) == NULL) {
489 			rte_memcpy(q->enc_in, in, (k - 24) >> 3);
490 			in = q->enc_in;
491 		} else {
492 			/* Store 3 first bytes of next CB as they will be
493 			 * overwritten by CRC bytes. If it is the last CB then
494 			 * there is no point to store 3 next bytes and this
495 			 * if..else branch will be omitted.
496 			 */
497 			first_3_bytes = *((uint64_t *)&in[(k - 32) >> 3]);
498 		}
499 
500 		crc_resp.data = in;
501 		bblib_lte_crc24a_gen(&crc_req, &crc_resp);
502 	} else if (enc->op_flags & RTE_BBDEV_TURBO_CRC_24B_ATTACH) {
503 		/* CRC24B */
504 		ret = is_enc_input_valid(k - 24, k_idx, total_left);
505 		if (ret != 0) {
506 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
507 			return;
508 		}
509 		crc_req.data = in;
510 		crc_req.len = (k - 24) >> 3;
511 		/* Check if there is a room for CRC bits. If this is the last
512 		 * CB in TB. If not use temporary buffer.
513 		 */
514 		if ((c - r == 1) && (rte_pktmbuf_append(m_in, 3) == NULL)) {
515 			rte_memcpy(q->enc_in, in, (k - 24) >> 3);
516 			in = q->enc_in;
517 		} else if (c - r > 1) {
518 			/* Store 3 first bytes of next CB as they will be
519 			 * overwritten by CRC bytes. If it is the last CB then
520 			 * there is no point to store 3 next bytes and this
521 			 * if..else branch will be omitted.
522 			 */
523 			first_3_bytes = *((uint64_t *)&in[(k - 32) >> 3]);
524 		}
525 
526 		crc_resp.data = in;
527 		bblib_lte_crc24b_gen(&crc_req, &crc_resp);
528 	} else {
529 		ret = is_enc_input_valid(k, k_idx, total_left);
530 		if (ret != 0) {
531 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
532 			return;
533 		}
534 	}
535 
536 	/* Turbo encoder */
537 
538 	/* Each bit layer output from turbo encoder is (k+4) bits long, i.e.
539 	 * input length + 4 tail bits. That's (k/8) + 1 bytes after rounding up.
540 	 * So dst_data's length should be 3*(k/8) + 3 bytes.
541 	 * In Rate-matching bypass case outputs pointers passed to encoder
542 	 * (out0, out1 and out2) can directly point to addresses of output from
543 	 * turbo_enc entity.
544 	 */
545 	if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH) {
546 		out0 = q->enc_out;
547 		out1 = RTE_PTR_ADD(out0, (k >> 3) + 1);
548 		out2 = RTE_PTR_ADD(out1, (k >> 3) + 1);
549 	} else {
550 		out0 = (uint8_t *)rte_pktmbuf_append(m_out, (k >> 3) * 3 + 2);
551 		if (out0 == NULL) {
552 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
553 			rte_bbdev_log(ERR,
554 					"Too little space in output mbuf");
555 			return;
556 		}
557 		enc->output.length += (k >> 3) * 3 + 2;
558 		/* rte_bbdev_op_data.offset can be different than the
559 		 * offset of the appended bytes
560 		 */
561 		out0 = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
562 		out1 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
563 				out_offset + (k >> 3) + 1);
564 		out2 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
565 				out_offset + 2 * ((k >> 3) + 1));
566 	}
567 
568 	turbo_req.case_id = k_idx;
569 	turbo_req.input_win = in;
570 	turbo_req.length = k >> 3;
571 	turbo_resp.output_win_0 = out0;
572 	turbo_resp.output_win_1 = out1;
573 	turbo_resp.output_win_2 = out2;
574 	if (bblib_turbo_encoder(&turbo_req, &turbo_resp) != 0) {
575 		op->status |= 1 << RTE_BBDEV_DRV_ERROR;
576 		rte_bbdev_log(ERR, "Turbo Encoder failed");
577 		return;
578 	}
579 
580 	/* Restore 3 first bytes of next CB if they were overwritten by CRC*/
581 	if (first_3_bytes != 0)
582 		*((uint64_t *)&in[(k - 32) >> 3]) = first_3_bytes;
583 
584 	/* Rate-matching */
585 	if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH) {
586 		/* get output data starting address */
587 		rm_out = (uint8_t *)rte_pktmbuf_append(m_out, (e >> 3));
588 		if (rm_out == NULL) {
589 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
590 			rte_bbdev_log(ERR,
591 					"Too little space in output mbuf");
592 			return;
593 		}
594 		/* rte_bbdev_op_data.offset can be different than the offset
595 		 * of the appended bytes
596 		 */
597 		rm_out = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
598 
599 		/* index of current code block */
600 		rm_req.r = r;
601 		/* total number of code block */
602 		rm_req.C = c;
603 		/* For DL - 1, UL - 0 */
604 		rm_req.direction = 1;
605 		/* According to 3ggp 36.212 Spec 5.1.4.1.2 section Nsoft, KMIMO
606 		 * and MDL_HARQ are used for Ncb calculation. As Ncb is already
607 		 * known we can adjust those parameters
608 		 */
609 		rm_req.Nsoft = ncb * rm_req.C;
610 		rm_req.KMIMO = 1;
611 		rm_req.MDL_HARQ = 1;
612 		/* According to 3ggp 36.212 Spec 5.1.4.1.2 section Nl, Qm and G
613 		 * are used for E calculation. As E is already known we can
614 		 * adjust those parameters
615 		 */
616 		rm_req.NL = e;
617 		rm_req.Qm = 1;
618 		rm_req.G = rm_req.NL * rm_req.Qm * rm_req.C;
619 
620 		rm_req.rvidx = enc->rv_index;
621 		rm_req.Kidx = k_idx - 1;
622 		rm_req.nLen = k + 4;
623 		rm_req.tin0 = out0;
624 		rm_req.tin1 = out1;
625 		rm_req.tin2 = out2;
626 		rm_resp.output = rm_out;
627 		rm_resp.OutputLen = (e >> 3);
628 		if (enc->op_flags & RTE_BBDEV_TURBO_RV_INDEX_BYPASS)
629 			rm_req.bypass_rvidx = 1;
630 		else
631 			rm_req.bypass_rvidx = 0;
632 
633 		if (bblib_rate_match_dl(&rm_req, &rm_resp) != 0) {
634 			op->status |= 1 << RTE_BBDEV_DRV_ERROR;
635 			rte_bbdev_log(ERR, "Rate matching failed");
636 			return;
637 		}
638 		enc->output.length += rm_resp.OutputLen;
639 	} else {
640 		/* Rate matching is bypassed */
641 
642 		/* Completing last byte of out0 (where 4 tail bits are stored)
643 		 * by moving first 4 bits from out1
644 		 */
645 		tmp_out = (uint8_t *) --out1;
646 		*tmp_out = *tmp_out | ((*(tmp_out + 1) & 0xF0) >> 4);
647 		tmp_out++;
648 		/* Shifting out1 data by 4 bits to the left */
649 		for (m = 0; m < k >> 3; ++m) {
650 			uint8_t *first = tmp_out;
651 			uint8_t second = *(tmp_out + 1);
652 			*first = (*first << 4) | ((second & 0xF0) >> 4);
653 			tmp_out++;
654 		}
655 		/* Shifting out2 data by 8 bits to the left */
656 		for (m = 0; m < (k >> 3) + 1; ++m) {
657 			*tmp_out = *(tmp_out + 1);
658 			tmp_out++;
659 		}
660 		*tmp_out = 0;
661 	}
662 }
663 
664 static inline void
665 enqueue_enc_one_op(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op)
666 {
667 	uint8_t c, r, crc24_bits = 0;
668 	uint16_t k, ncb;
669 	uint32_t e;
670 	struct rte_bbdev_op_turbo_enc *enc = &op->turbo_enc;
671 	uint16_t in_offset = enc->input.offset;
672 	uint16_t out_offset = enc->output.offset;
673 	struct rte_mbuf *m_in = enc->input.data;
674 	struct rte_mbuf *m_out = enc->output.data;
675 	uint16_t total_left = enc->input.length;
676 
677 	/* Clear op status */
678 	op->status = 0;
679 
680 	if (total_left > MAX_TB_SIZE >> 3) {
681 		rte_bbdev_log(ERR, "TB size (%u) is too big, max: %d",
682 				total_left, MAX_TB_SIZE);
683 		op->status = 1 << RTE_BBDEV_DATA_ERROR;
684 		return;
685 	}
686 
687 	if (m_in == NULL || m_out == NULL) {
688 		rte_bbdev_log(ERR, "Invalid mbuf pointer");
689 		op->status = 1 << RTE_BBDEV_DATA_ERROR;
690 		return;
691 	}
692 
693 	if ((enc->op_flags & RTE_BBDEV_TURBO_CRC_24B_ATTACH) ||
694 		(enc->op_flags & RTE_BBDEV_TURBO_CRC_24A_ATTACH))
695 		crc24_bits = 24;
696 
697 	if (enc->code_block_mode == 0) { /* For Transport Block mode */
698 		c = enc->tb_params.c;
699 		r = enc->tb_params.r;
700 	} else {/* For Code Block mode */
701 		c = 1;
702 		r = 0;
703 	}
704 
705 	while (total_left > 0 && r < c) {
706 		if (enc->code_block_mode == 0) {
707 			k = (r < enc->tb_params.c_neg) ?
708 				enc->tb_params.k_neg : enc->tb_params.k_pos;
709 			ncb = (r < enc->tb_params.c_neg) ?
710 				enc->tb_params.ncb_neg : enc->tb_params.ncb_pos;
711 			e = (r < enc->tb_params.cab) ?
712 				enc->tb_params.ea : enc->tb_params.eb;
713 		} else {
714 			k = enc->cb_params.k;
715 			ncb = enc->cb_params.ncb;
716 			e = enc->cb_params.e;
717 		}
718 
719 		process_enc_cb(q, op, r, c, k, ncb, e, m_in,
720 				m_out, in_offset, out_offset, total_left);
721 		/* Update total_left */
722 		total_left -= (k - crc24_bits) >> 3;
723 		/* Update offsets for next CBs (if exist) */
724 		in_offset += (k - crc24_bits) >> 3;
725 		if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH)
726 			out_offset += e >> 3;
727 		else
728 			out_offset += (k >> 3) * 3 + 2;
729 		r++;
730 	}
731 
732 	/* check if all input data was processed */
733 	if (total_left != 0) {
734 		op->status |= 1 << RTE_BBDEV_DATA_ERROR;
735 		rte_bbdev_log(ERR,
736 				"Mismatch between mbuf length and included CBs sizes");
737 	}
738 }
739 
740 static inline uint16_t
741 enqueue_enc_all_ops(struct turbo_sw_queue *q, struct rte_bbdev_enc_op **ops,
742 		uint16_t nb_ops)
743 {
744 	uint16_t i;
745 
746 	for (i = 0; i < nb_ops; ++i)
747 		enqueue_enc_one_op(q, ops[i]);
748 
749 	return rte_ring_enqueue_burst(q->processed_pkts, (void **)ops, nb_ops,
750 			NULL);
751 }
752 
753 /* Remove the padding bytes from a cyclic buffer.
754  * The input buffer is a data stream wk as described in 3GPP TS 36.212 section
755  * 5.1.4.1.2 starting from w0 and with length Ncb bytes.
756  * The output buffer is a data stream wk with pruned padding bytes. It's length
757  * is 3*D bytes and the order of non-padding bytes is preserved.
758  */
759 static inline void
760 remove_nulls_from_circular_buf(const uint8_t *in, uint8_t *out, uint16_t k,
761 		uint16_t ncb)
762 {
763 	uint32_t in_idx, out_idx, c_idx;
764 	const uint32_t d = k + 4;
765 	const uint32_t kw = (ncb / 3);
766 	const uint32_t nd = kw - d;
767 	const uint32_t r_subblock = kw / C_SUBBLOCK;
768 	/* Inter-column permutation pattern */
769 	const uint32_t P[C_SUBBLOCK] = {0, 16, 8, 24, 4, 20, 12, 28, 2, 18, 10,
770 			26, 6, 22, 14, 30, 1, 17, 9, 25, 5, 21, 13, 29, 3, 19,
771 			11, 27, 7, 23, 15, 31};
772 	in_idx = 0;
773 	out_idx = 0;
774 
775 	/* The padding bytes are at the first Nd positions in the first row. */
776 	for (c_idx = 0; in_idx < kw; in_idx += r_subblock, ++c_idx) {
777 		if (P[c_idx] < nd) {
778 			rte_memcpy(&out[out_idx], &in[in_idx + 1],
779 					r_subblock - 1);
780 			out_idx += r_subblock - 1;
781 		} else {
782 			rte_memcpy(&out[out_idx], &in[in_idx], r_subblock);
783 			out_idx += r_subblock;
784 		}
785 	}
786 
787 	/* First and second parity bits sub-blocks are interlaced. */
788 	for (c_idx = 0; in_idx < ncb - 2 * r_subblock;
789 			in_idx += 2 * r_subblock, ++c_idx) {
790 		uint32_t second_block_c_idx = P[c_idx];
791 		uint32_t third_block_c_idx = P[c_idx] + 1;
792 
793 		if (second_block_c_idx < nd && third_block_c_idx < nd) {
794 			rte_memcpy(&out[out_idx], &in[in_idx + 2],
795 					2 * r_subblock - 2);
796 			out_idx += 2 * r_subblock - 2;
797 		} else if (second_block_c_idx >= nd &&
798 				third_block_c_idx >= nd) {
799 			rte_memcpy(&out[out_idx], &in[in_idx], 2 * r_subblock);
800 			out_idx += 2 * r_subblock;
801 		} else if (second_block_c_idx < nd) {
802 			out[out_idx++] = in[in_idx];
803 			rte_memcpy(&out[out_idx], &in[in_idx + 2],
804 					2 * r_subblock - 2);
805 			out_idx += 2 * r_subblock - 2;
806 		} else {
807 			rte_memcpy(&out[out_idx], &in[in_idx + 1],
808 					2 * r_subblock - 1);
809 			out_idx += 2 * r_subblock - 1;
810 		}
811 	}
812 
813 	/* Last interlaced row is different - its last byte is the only padding
814 	 * byte. We can have from 2 up to 26 padding bytes (Nd) per sub-block.
815 	 * After interlacing the 1st and 2nd parity sub-blocks we can have 0, 1
816 	 * or 2 padding bytes each time we make a step of 2 * R_SUBBLOCK bytes
817 	 * (moving to another column). 2nd parity sub-block uses the same
818 	 * inter-column permutation pattern as the systematic and 1st parity
819 	 * sub-blocks but it adds '1' to the resulting index and calculates the
820 	 * modulus of the result and Kw. Last column is mapped to itself (id 31)
821 	 * so the first byte taken from the 2nd parity sub-block will be the
822 	 * 32nd (31+1) byte, then 64th etc. (step is C_SUBBLOCK == 32) and the
823 	 * last byte will be the first byte from the sub-block:
824 	 * (32 + 32 * (R_SUBBLOCK-1)) % Kw == Kw % Kw == 0. Nd can't  be smaller
825 	 * than 2 so we know that bytes with ids 0 and 1 must be the padding
826 	 * bytes. The bytes from the 1st parity sub-block are the bytes from the
827 	 * 31st column - Nd can't be greater than 26 so we are sure that there
828 	 * are no padding bytes in 31st column.
829 	 */
830 	rte_memcpy(&out[out_idx], &in[in_idx], 2 * r_subblock - 1);
831 }
832 
833 static inline void
834 move_padding_bytes(const uint8_t *in, uint8_t *out, uint16_t k,
835 		uint16_t ncb)
836 {
837 	uint16_t d = k + 4;
838 	uint16_t kpi = ncb / 3;
839 	uint16_t nd = kpi - d;
840 
841 	rte_memcpy(&out[nd], in, d);
842 	rte_memcpy(&out[nd + kpi + 64], &in[kpi], d);
843 	rte_memcpy(&out[nd + 2 * (kpi + 64)], &in[2 * kpi], d);
844 }
845 
846 static inline void
847 process_dec_cb(struct turbo_sw_queue *q, struct rte_bbdev_dec_op *op,
848 		uint8_t c, uint16_t k, uint16_t kw, struct rte_mbuf *m_in,
849 		struct rte_mbuf *m_out, uint16_t in_offset, uint16_t out_offset,
850 		bool check_crc_24b, uint16_t total_left)
851 {
852 	int ret;
853 	int32_t k_idx;
854 	int32_t iter_cnt;
855 	uint8_t *in, *out, *adapter_input;
856 	int32_t ncb, ncb_without_null;
857 	struct bblib_turbo_adapter_ul_response adapter_resp;
858 	struct bblib_turbo_adapter_ul_request adapter_req;
859 	struct bblib_turbo_decoder_request turbo_req;
860 	struct bblib_turbo_decoder_response turbo_resp;
861 	struct rte_bbdev_op_turbo_dec *dec = &op->turbo_dec;
862 
863 	k_idx = compute_idx(k);
864 
865 	ret = is_dec_input_valid(k_idx, kw, total_left);
866 	if (ret != 0) {
867 		op->status |= 1 << RTE_BBDEV_DATA_ERROR;
868 		return;
869 	}
870 
871 	in = rte_pktmbuf_mtod_offset(m_in, uint8_t *, in_offset);
872 	ncb = kw;
873 	ncb_without_null = (k + 4) * 3;
874 
875 	if (check_bit(dec->op_flags, RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE)) {
876 		struct bblib_deinterleave_ul_request deint_req;
877 		struct bblib_deinterleave_ul_response deint_resp;
878 
879 		/* SW decoder accepts only a circular buffer without NULL bytes
880 		 * so the input needs to be converted.
881 		 */
882 		remove_nulls_from_circular_buf(in, q->deint_input, k, ncb);
883 
884 		deint_req.pharqbuffer = q->deint_input;
885 		deint_req.ncb = ncb_without_null;
886 		deint_resp.pinteleavebuffer = q->deint_output;
887 		bblib_deinterleave_ul(&deint_req, &deint_resp);
888 	} else
889 		move_padding_bytes(in, q->deint_output, k, ncb);
890 
891 	adapter_input = q->deint_output;
892 
893 	if (dec->op_flags & RTE_BBDEV_TURBO_POS_LLR_1_BIT_IN)
894 		adapter_req.isinverted = 1;
895 	else if (dec->op_flags & RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN)
896 		adapter_req.isinverted = 0;
897 	else {
898 		op->status |= 1 << RTE_BBDEV_DRV_ERROR;
899 		rte_bbdev_log(ERR, "LLR format wasn't specified");
900 		return;
901 	}
902 
903 	adapter_req.ncb = ncb_without_null;
904 	adapter_req.pinteleavebuffer = adapter_input;
905 	adapter_resp.pharqout = q->adapter_output;
906 	bblib_turbo_adapter_ul(&adapter_req, &adapter_resp);
907 
908 	out = (uint8_t *)rte_pktmbuf_append(m_out, (k >> 3));
909 	if (out == NULL) {
910 		op->status |= 1 << RTE_BBDEV_DATA_ERROR;
911 		rte_bbdev_log(ERR, "Too little space in output mbuf");
912 		return;
913 	}
914 	/* rte_bbdev_op_data.offset can be different than the offset of the
915 	 * appended bytes
916 	 */
917 	out = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
918 	if (check_crc_24b)
919 		turbo_req.c = c + 1;
920 	else
921 		turbo_req.c = c;
922 	turbo_req.input = (int8_t *)q->adapter_output;
923 	turbo_req.k = k;
924 	turbo_req.k_idx = k_idx;
925 	turbo_req.max_iter_num = dec->iter_max;
926 	turbo_resp.ag_buf = q->ag;
927 	turbo_resp.cb_buf = q->code_block;
928 	turbo_resp.output = out;
929 	iter_cnt = bblib_turbo_decoder(&turbo_req, &turbo_resp);
930 	dec->hard_output.length += (k >> 3);
931 
932 	if (iter_cnt > 0) {
933 		/* Temporary solution for returned iter_count from SDK */
934 		iter_cnt = (iter_cnt - 1) / 2;
935 		dec->iter_count = RTE_MAX(iter_cnt, dec->iter_count);
936 	} else {
937 		op->status |= 1 << RTE_BBDEV_DATA_ERROR;
938 		rte_bbdev_log(ERR, "Turbo Decoder failed");
939 		return;
940 	}
941 }
942 
943 static inline void
944 enqueue_dec_one_op(struct turbo_sw_queue *q, struct rte_bbdev_dec_op *op)
945 {
946 	uint8_t c, r = 0;
947 	uint16_t kw, k = 0;
948 	struct rte_bbdev_op_turbo_dec *dec = &op->turbo_dec;
949 	struct rte_mbuf *m_in = dec->input.data;
950 	struct rte_mbuf *m_out = dec->hard_output.data;
951 	uint16_t in_offset = dec->input.offset;
952 	uint16_t total_left = dec->input.length;
953 	uint16_t out_offset = dec->hard_output.offset;
954 
955 	/* Clear op status */
956 	op->status = 0;
957 
958 	if (m_in == NULL || m_out == NULL) {
959 		rte_bbdev_log(ERR, "Invalid mbuf pointer");
960 		op->status = 1 << RTE_BBDEV_DATA_ERROR;
961 		return;
962 	}
963 
964 	if (dec->code_block_mode == 0) { /* For Transport Block mode */
965 		c = dec->tb_params.c;
966 	} else { /* For Code Block mode */
967 		k = dec->cb_params.k;
968 		c = 1;
969 	}
970 
971 	while (total_left > 0) {
972 		if (dec->code_block_mode == 0)
973 			k = (r < dec->tb_params.c_neg) ?
974 				dec->tb_params.k_neg : dec->tb_params.k_pos;
975 
976 		/* Calculates circular buffer size (Kw).
977 		 * According to 3gpp 36.212 section 5.1.4.2
978 		 *   Kw = 3 * Kpi,
979 		 * where:
980 		 *   Kpi = nCol * nRow
981 		 * where nCol is 32 and nRow can be calculated from:
982 		 *   D =< nCol * nRow
983 		 * where D is the size of each output from turbo encoder block
984 		 * (k + 4).
985 		 */
986 		kw = RTE_ALIGN_CEIL(k + 4, C_SUBBLOCK) * 3;
987 
988 		process_dec_cb(q, op, c, k, kw, m_in, m_out, in_offset,
989 				out_offset, check_bit(dec->op_flags,
990 				RTE_BBDEV_TURBO_CRC_TYPE_24B), total_left);
991 		/* As a result of decoding we get Code Block with included
992 		 * decoded CRC24 at the end of Code Block. Type of CRC24 is
993 		 * specified by flag.
994 		 */
995 
996 		/* Update total_left */
997 		total_left -= kw;
998 		/* Update offsets for next CBs (if exist) */
999 		in_offset += kw;
1000 		out_offset += (k >> 3);
1001 		r++;
1002 	}
1003 	if (total_left != 0) {
1004 		op->status |= 1 << RTE_BBDEV_DATA_ERROR;
1005 		rte_bbdev_log(ERR,
1006 				"Mismatch between mbuf length and included Circular buffer sizes");
1007 	}
1008 }
1009 
1010 static inline uint16_t
1011 enqueue_dec_all_ops(struct turbo_sw_queue *q, struct rte_bbdev_dec_op **ops,
1012 		uint16_t nb_ops)
1013 {
1014 	uint16_t i;
1015 
1016 	for (i = 0; i < nb_ops; ++i)
1017 		enqueue_dec_one_op(q, ops[i]);
1018 
1019 	return rte_ring_enqueue_burst(q->processed_pkts, (void **)ops, nb_ops,
1020 			NULL);
1021 }
1022 
1023 /* Enqueue burst */
1024 static uint16_t
1025 enqueue_enc_ops(struct rte_bbdev_queue_data *q_data,
1026 		struct rte_bbdev_enc_op **ops, uint16_t nb_ops)
1027 {
1028 	void *queue = q_data->queue_private;
1029 	struct turbo_sw_queue *q = queue;
1030 	uint16_t nb_enqueued = 0;
1031 
1032 	nb_enqueued = enqueue_enc_all_ops(q, ops, nb_ops);
1033 
1034 	q_data->queue_stats.enqueue_err_count += nb_ops - nb_enqueued;
1035 	q_data->queue_stats.enqueued_count += nb_enqueued;
1036 
1037 	return nb_enqueued;
1038 }
1039 
1040 /* Enqueue burst */
1041 static uint16_t
1042 enqueue_dec_ops(struct rte_bbdev_queue_data *q_data,
1043 		 struct rte_bbdev_dec_op **ops, uint16_t nb_ops)
1044 {
1045 	void *queue = q_data->queue_private;
1046 	struct turbo_sw_queue *q = queue;
1047 	uint16_t nb_enqueued = 0;
1048 
1049 	nb_enqueued = enqueue_dec_all_ops(q, ops, nb_ops);
1050 
1051 	q_data->queue_stats.enqueue_err_count += nb_ops - nb_enqueued;
1052 	q_data->queue_stats.enqueued_count += nb_enqueued;
1053 
1054 	return nb_enqueued;
1055 }
1056 
1057 /* Dequeue decode burst */
1058 static uint16_t
1059 dequeue_dec_ops(struct rte_bbdev_queue_data *q_data,
1060 		struct rte_bbdev_dec_op **ops, uint16_t nb_ops)
1061 {
1062 	struct turbo_sw_queue *q = q_data->queue_private;
1063 	uint16_t nb_dequeued = rte_ring_dequeue_burst(q->processed_pkts,
1064 			(void **)ops, nb_ops, NULL);
1065 	q_data->queue_stats.dequeued_count += nb_dequeued;
1066 
1067 	return nb_dequeued;
1068 }
1069 
1070 /* Dequeue encode burst */
1071 static uint16_t
1072 dequeue_enc_ops(struct rte_bbdev_queue_data *q_data,
1073 		struct rte_bbdev_enc_op **ops, uint16_t nb_ops)
1074 {
1075 	struct turbo_sw_queue *q = q_data->queue_private;
1076 	uint16_t nb_dequeued = rte_ring_dequeue_burst(q->processed_pkts,
1077 			(void **)ops, nb_ops, NULL);
1078 	q_data->queue_stats.dequeued_count += nb_dequeued;
1079 
1080 	return nb_dequeued;
1081 }
1082 
1083 /* Parse 16bit integer from string argument */
1084 static inline int
1085 parse_u16_arg(const char *key, const char *value, void *extra_args)
1086 {
1087 	uint16_t *u16 = extra_args;
1088 	unsigned int long result;
1089 
1090 	if ((value == NULL) || (extra_args == NULL))
1091 		return -EINVAL;
1092 	errno = 0;
1093 	result = strtoul(value, NULL, 0);
1094 	if ((result >= (1 << 16)) || (errno != 0)) {
1095 		rte_bbdev_log(ERR, "Invalid value %lu for %s", result, key);
1096 		return -ERANGE;
1097 	}
1098 	*u16 = (uint16_t)result;
1099 	return 0;
1100 }
1101 
1102 /* Parse parameters used to create device */
1103 static int
1104 parse_turbo_sw_params(struct turbo_sw_params *params, const char *input_args)
1105 {
1106 	struct rte_kvargs *kvlist = NULL;
1107 	int ret = 0;
1108 
1109 	if (params == NULL)
1110 		return -EINVAL;
1111 	if (input_args) {
1112 		kvlist = rte_kvargs_parse(input_args, turbo_sw_valid_params);
1113 		if (kvlist == NULL)
1114 			return -EFAULT;
1115 
1116 		ret = rte_kvargs_process(kvlist, turbo_sw_valid_params[0],
1117 					&parse_u16_arg, &params->queues_num);
1118 		if (ret < 0)
1119 			goto exit;
1120 
1121 		ret = rte_kvargs_process(kvlist, turbo_sw_valid_params[1],
1122 					&parse_u16_arg, &params->socket_id);
1123 		if (ret < 0)
1124 			goto exit;
1125 
1126 		if (params->socket_id >= RTE_MAX_NUMA_NODES) {
1127 			rte_bbdev_log(ERR, "Invalid socket, must be < %u",
1128 					RTE_MAX_NUMA_NODES);
1129 			goto exit;
1130 		}
1131 	}
1132 
1133 exit:
1134 	if (kvlist)
1135 		rte_kvargs_free(kvlist);
1136 	return ret;
1137 }
1138 
1139 /* Create device */
1140 static int
1141 turbo_sw_bbdev_create(struct rte_vdev_device *vdev,
1142 		struct turbo_sw_params *init_params)
1143 {
1144 	struct rte_bbdev *bbdev;
1145 	const char *name = rte_vdev_device_name(vdev);
1146 
1147 	bbdev = rte_bbdev_allocate(name);
1148 	if (bbdev == NULL)
1149 		return -ENODEV;
1150 
1151 	bbdev->data->dev_private = rte_zmalloc_socket(name,
1152 			sizeof(struct bbdev_private), RTE_CACHE_LINE_SIZE,
1153 			init_params->socket_id);
1154 	if (bbdev->data->dev_private == NULL) {
1155 		rte_bbdev_release(bbdev);
1156 		return -ENOMEM;
1157 	}
1158 
1159 	bbdev->dev_ops = &pmd_ops;
1160 	bbdev->device = &vdev->device;
1161 	bbdev->data->socket_id = init_params->socket_id;
1162 	bbdev->intr_handle = NULL;
1163 
1164 	/* register rx/tx burst functions for data path */
1165 	bbdev->dequeue_enc_ops = dequeue_enc_ops;
1166 	bbdev->dequeue_dec_ops = dequeue_dec_ops;
1167 	bbdev->enqueue_enc_ops = enqueue_enc_ops;
1168 	bbdev->enqueue_dec_ops = enqueue_dec_ops;
1169 	((struct bbdev_private *) bbdev->data->dev_private)->max_nb_queues =
1170 			init_params->queues_num;
1171 
1172 	return 0;
1173 }
1174 
1175 /* Initialise device */
1176 static int
1177 turbo_sw_bbdev_probe(struct rte_vdev_device *vdev)
1178 {
1179 	struct turbo_sw_params init_params = {
1180 		rte_socket_id(),
1181 		RTE_BBDEV_DEFAULT_MAX_NB_QUEUES
1182 	};
1183 	const char *name;
1184 	const char *input_args;
1185 
1186 	if (vdev == NULL)
1187 		return -EINVAL;
1188 
1189 	name = rte_vdev_device_name(vdev);
1190 	if (name == NULL)
1191 		return -EINVAL;
1192 	input_args = rte_vdev_device_args(vdev);
1193 	parse_turbo_sw_params(&init_params, input_args);
1194 
1195 	rte_bbdev_log_debug(
1196 			"Initialising %s on NUMA node %d with max queues: %d\n",
1197 			name, init_params.socket_id, init_params.queues_num);
1198 
1199 	return turbo_sw_bbdev_create(vdev, &init_params);
1200 }
1201 
1202 /* Uninitialise device */
1203 static int
1204 turbo_sw_bbdev_remove(struct rte_vdev_device *vdev)
1205 {
1206 	struct rte_bbdev *bbdev;
1207 	const char *name;
1208 
1209 	if (vdev == NULL)
1210 		return -EINVAL;
1211 
1212 	name = rte_vdev_device_name(vdev);
1213 	if (name == NULL)
1214 		return -EINVAL;
1215 
1216 	bbdev = rte_bbdev_get_named_dev(name);
1217 	if (bbdev == NULL)
1218 		return -EINVAL;
1219 
1220 	rte_free(bbdev->data->dev_private);
1221 
1222 	return rte_bbdev_release(bbdev);
1223 }
1224 
1225 static struct rte_vdev_driver bbdev_turbo_sw_pmd_drv = {
1226 	.probe = turbo_sw_bbdev_probe,
1227 	.remove = turbo_sw_bbdev_remove
1228 };
1229 
1230 RTE_PMD_REGISTER_VDEV(DRIVER_NAME, bbdev_turbo_sw_pmd_drv);
1231 RTE_PMD_REGISTER_PARAM_STRING(DRIVER_NAME,
1232 	TURBO_SW_MAX_NB_QUEUES_ARG"=<int> "
1233 	TURBO_SW_SOCKET_ID_ARG"=<int>");
1234 
1235 RTE_INIT(null_bbdev_init_log);
1236 static void
1237 null_bbdev_init_log(void)
1238 {
1239 	bbdev_turbo_sw_logtype = rte_log_register("pmd.bb.turbo_sw");
1240 	if (bbdev_turbo_sw_logtype >= 0)
1241 		rte_log_set_level(bbdev_turbo_sw_logtype, RTE_LOG_NOTICE);
1242 }
1243