xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision 4c3d7961d9002bb715a8ee76bcf464d633316d4c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2015 6WIND S.A.
3  * Copyright 2015 Mellanox Technologies, Ltd
4  */
5 
6 #include <stddef.h>
7 #include <unistd.h>
8 #include <string.h>
9 #include <stdint.h>
10 #include <stdlib.h>
11 #include <errno.h>
12 
13 #include <ethdev_driver.h>
14 #include <bus_pci_driver.h>
15 #include <rte_mbuf.h>
16 #include <rte_common.h>
17 #include <rte_interrupts.h>
18 #include <rte_malloc.h>
19 #include <rte_string_fns.h>
20 #include <rte_rwlock.h>
21 #include <rte_cycles.h>
22 
23 #include <mlx5_malloc.h>
24 
25 #include "mlx5_rxtx.h"
26 #include "mlx5_rx.h"
27 #include "mlx5_tx.h"
28 #include "mlx5_autoconf.h"
29 #include "mlx5_devx.h"
30 #include "rte_pmd_mlx5.h"
31 
32 /**
33  * Get the interface index from device name.
34  *
35  * @param[in] dev
36  *   Pointer to Ethernet device.
37  *
38  * @return
39  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
40  */
41 unsigned int
42 mlx5_ifindex(const struct rte_eth_dev *dev)
43 {
44 	struct mlx5_priv *priv = dev->data->dev_private;
45 	unsigned int ifindex;
46 
47 	MLX5_ASSERT(priv);
48 	MLX5_ASSERT(priv->if_index);
49 	if (priv->master && priv->sh->bond.ifindex > 0)
50 		ifindex = priv->sh->bond.ifindex;
51 	else
52 		ifindex = priv->if_index;
53 	if (!ifindex)
54 		rte_errno = ENXIO;
55 	return ifindex;
56 }
57 
58 /**
59  * DPDK callback for Ethernet device configuration.
60  *
61  * @param dev
62  *   Pointer to Ethernet device structure.
63  *
64  * @return
65  *   0 on success, a negative errno value otherwise and rte_errno is set.
66  */
67 int
68 mlx5_dev_configure(struct rte_eth_dev *dev)
69 {
70 	struct mlx5_priv *priv = dev->data->dev_private;
71 	unsigned int rxqs_n = dev->data->nb_rx_queues;
72 	unsigned int txqs_n = dev->data->nb_tx_queues;
73 	const uint8_t use_app_rss_key =
74 		!!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
75 	int ret = 0;
76 
77 	if (use_app_rss_key &&
78 	    (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
79 	     MLX5_RSS_HASH_KEY_LEN)) {
80 		DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
81 			dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
82 		rte_errno = EINVAL;
83 		return -rte_errno;
84 	}
85 	priv->rss_conf.rss_key = mlx5_realloc(priv->rss_conf.rss_key,
86 					      MLX5_MEM_RTE,
87 					      MLX5_RSS_HASH_KEY_LEN, 0,
88 					      SOCKET_ID_ANY);
89 	if (!priv->rss_conf.rss_key) {
90 		DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
91 			dev->data->port_id, rxqs_n);
92 		rte_errno = ENOMEM;
93 		return -rte_errno;
94 	}
95 
96 	if ((dev->data->dev_conf.txmode.offloads &
97 			RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP) &&
98 			rte_mbuf_dyn_tx_timestamp_register(NULL, NULL) != 0) {
99 		DRV_LOG(ERR, "port %u cannot register Tx timestamp field/flag",
100 			dev->data->port_id);
101 		return -rte_errno;
102 	}
103 	memcpy(priv->rss_conf.rss_key,
104 	       use_app_rss_key ?
105 	       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
106 	       rss_hash_default_key,
107 	       MLX5_RSS_HASH_KEY_LEN);
108 	priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
109 	priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
110 	priv->rxq_privs = mlx5_realloc(priv->rxq_privs,
111 				       MLX5_MEM_RTE | MLX5_MEM_ZERO,
112 				       sizeof(void *) * rxqs_n, 0,
113 				       SOCKET_ID_ANY);
114 	if (rxqs_n && priv->rxq_privs == NULL) {
115 		DRV_LOG(ERR, "port %u cannot allocate rxq private data",
116 			dev->data->port_id);
117 		rte_errno = ENOMEM;
118 		return -rte_errno;
119 	}
120 	priv->txqs = (void *)dev->data->tx_queues;
121 	if (txqs_n != priv->txqs_n) {
122 		DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
123 			dev->data->port_id, priv->txqs_n, txqs_n);
124 		priv->txqs_n = txqs_n;
125 	}
126 	if (priv->ext_txqs && txqs_n >= MLX5_EXTERNAL_TX_QUEUE_ID_MIN) {
127 		DRV_LOG(ERR, "port %u cannot handle this many Tx queues (%u), "
128 			"the maximal number of internal Tx queues is %u",
129 			dev->data->port_id, txqs_n,
130 			MLX5_EXTERNAL_TX_QUEUE_ID_MIN - 1);
131 		rte_errno = EINVAL;
132 		return -rte_errno;
133 	}
134 	if (rxqs_n > priv->sh->dev_cap.ind_table_max_size) {
135 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
136 			dev->data->port_id, rxqs_n);
137 		rte_errno = EINVAL;
138 		return -rte_errno;
139 	}
140 	if (priv->ext_rxqs && rxqs_n >= RTE_PMD_MLX5_EXTERNAL_RX_QUEUE_ID_MIN) {
141 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u), "
142 			"the maximal number of internal Rx queues is %u",
143 			dev->data->port_id, rxqs_n,
144 			RTE_PMD_MLX5_EXTERNAL_RX_QUEUE_ID_MIN - 1);
145 		rte_errno = EINVAL;
146 		return -rte_errno;
147 	}
148 	if (rxqs_n != priv->rxqs_n) {
149 		DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
150 			dev->data->port_id, priv->rxqs_n, rxqs_n);
151 		priv->rxqs_n = rxqs_n;
152 	}
153 	priv->skip_default_rss_reta = 0;
154 	ret = mlx5_proc_priv_init(dev);
155 	if (ret)
156 		return ret;
157 	ret = mlx5_dev_set_mtu(dev, dev->data->mtu);
158 	if (ret) {
159 		DRV_LOG(ERR, "port %u failed to set MTU to %u", dev->data->port_id,
160 			dev->data->mtu);
161 		return ret;
162 	}
163 	return 0;
164 }
165 
166 /**
167  * Configure default RSS reta.
168  *
169  * @param dev
170  *   Pointer to Ethernet device structure.
171  *
172  * @return
173  *   0 on success, a negative errno value otherwise and rte_errno is set.
174  */
175 int
176 mlx5_dev_configure_rss_reta(struct rte_eth_dev *dev)
177 {
178 	struct mlx5_priv *priv = dev->data->dev_private;
179 	unsigned int rxqs_n = dev->data->nb_rx_queues;
180 	unsigned int i;
181 	unsigned int j;
182 	unsigned int reta_idx_n;
183 	int ret = 0;
184 	unsigned int *rss_queue_arr = NULL;
185 	unsigned int rss_queue_n = 0;
186 
187 	if (priv->skip_default_rss_reta)
188 		return ret;
189 	rss_queue_arr = mlx5_malloc(0, rxqs_n * sizeof(unsigned int), 0,
190 				    SOCKET_ID_ANY);
191 	if (!rss_queue_arr) {
192 		DRV_LOG(ERR, "port %u cannot allocate RSS queue list (%u)",
193 			dev->data->port_id, rxqs_n);
194 		rte_errno = ENOMEM;
195 		return -rte_errno;
196 	}
197 	for (i = 0, j = 0; i < rxqs_n; i++) {
198 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, i);
199 
200 		if (rxq_ctrl && !rxq_ctrl->is_hairpin)
201 			rss_queue_arr[j++] = i;
202 	}
203 	rss_queue_n = j;
204 	if (rss_queue_n > priv->sh->dev_cap.ind_table_max_size) {
205 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
206 			dev->data->port_id, rss_queue_n);
207 		rte_errno = EINVAL;
208 		mlx5_free(rss_queue_arr);
209 		return -rte_errno;
210 	}
211 	DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
212 		dev->data->port_id, priv->rxqs_n, rxqs_n);
213 	priv->rxqs_n = rxqs_n;
214 	/*
215 	 * If the requested number of RX queues is not a power of two,
216 	 * use the maximum indirection table size for better balancing.
217 	 * The result is always rounded to the next power of two.
218 	 */
219 	reta_idx_n = (1 << log2above((rss_queue_n & (rss_queue_n - 1)) ?
220 				     priv->sh->dev_cap.ind_table_max_size :
221 				     rss_queue_n));
222 	ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
223 	if (ret) {
224 		mlx5_free(rss_queue_arr);
225 		return ret;
226 	}
227 	/*
228 	 * When the number of RX queues is not a power of two,
229 	 * the remaining table entries are padded with reused WQs
230 	 * and hashes are not spread uniformly.
231 	 */
232 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
233 		(*priv->reta_idx)[i] = rss_queue_arr[j];
234 		if (++j == rss_queue_n)
235 			j = 0;
236 	}
237 	mlx5_free(rss_queue_arr);
238 	return ret;
239 }
240 
241 /**
242  * Sets default tuning parameters.
243  *
244  * @param dev
245  *   Pointer to Ethernet device.
246  * @param[out] info
247  *   Info structure output buffer.
248  */
249 static void
250 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
251 {
252 	struct mlx5_priv *priv = dev->data->dev_private;
253 
254 	/* Minimum CPU utilization. */
255 	info->default_rxportconf.ring_size = 256;
256 	info->default_txportconf.ring_size = 256;
257 	info->default_rxportconf.burst_size = MLX5_RX_DEFAULT_BURST;
258 	info->default_txportconf.burst_size = MLX5_TX_DEFAULT_BURST;
259 	if (priv->link_speed_capa >> rte_bsf32(RTE_ETH_LINK_SPEED_100G)) {
260 		/* if supports at least 100G */
261 		info->default_rxportconf.nb_queues = 16;
262 		info->default_txportconf.nb_queues = 16;
263 		if (dev->data->nb_rx_queues > 2 ||
264 		    dev->data->nb_tx_queues > 2) {
265 			/* Max Throughput. */
266 			info->default_rxportconf.ring_size = 2048;
267 			info->default_txportconf.ring_size = 2048;
268 		}
269 	} else {
270 		info->default_rxportconf.nb_queues = 8;
271 		info->default_txportconf.nb_queues = 8;
272 		if (dev->data->nb_rx_queues > 2 ||
273 		    dev->data->nb_tx_queues > 2) {
274 			/* Max Throughput. */
275 			info->default_rxportconf.ring_size = 4096;
276 			info->default_txportconf.ring_size = 4096;
277 		}
278 	}
279 }
280 
281 /**
282  * Sets tx mbuf limiting parameters.
283  *
284  * @param dev
285  *   Pointer to Ethernet device.
286  * @param[out] info
287  *   Info structure output buffer.
288  */
289 static void
290 mlx5_set_txlimit_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
291 {
292 	struct mlx5_priv *priv = dev->data->dev_private;
293 	struct mlx5_port_config *config = &priv->config;
294 	unsigned int inlen;
295 	uint16_t nb_max;
296 
297 	inlen = (config->txq_inline_max == MLX5_ARG_UNSET) ?
298 		MLX5_SEND_DEF_INLINE_LEN :
299 		(unsigned int)config->txq_inline_max;
300 	MLX5_ASSERT(config->txq_inline_min >= 0);
301 	inlen = RTE_MAX(inlen, (unsigned int)config->txq_inline_min);
302 	inlen = RTE_MIN(inlen, MLX5_WQE_SIZE_MAX +
303 			       MLX5_ESEG_MIN_INLINE_SIZE -
304 			       MLX5_WQE_CSEG_SIZE -
305 			       MLX5_WQE_ESEG_SIZE -
306 			       MLX5_WQE_DSEG_SIZE * 2);
307 	nb_max = (MLX5_WQE_SIZE_MAX +
308 		  MLX5_ESEG_MIN_INLINE_SIZE -
309 		  MLX5_WQE_CSEG_SIZE -
310 		  MLX5_WQE_ESEG_SIZE -
311 		  MLX5_WQE_DSEG_SIZE -
312 		  inlen) / MLX5_WSEG_SIZE;
313 	info->tx_desc_lim.nb_seg_max = nb_max;
314 	info->tx_desc_lim.nb_mtu_seg_max = nb_max;
315 }
316 
317 /**
318  * DPDK callback to get information about the device.
319  *
320  * @param dev
321  *   Pointer to Ethernet device structure.
322  * @param[out] info
323  *   Info structure output buffer.
324  */
325 int
326 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
327 {
328 	struct mlx5_priv *priv = dev->data->dev_private;
329 	unsigned int max;
330 
331 	/* FIXME: we should ask the device for these values. */
332 	info->min_rx_bufsize = 32;
333 	info->max_rx_pktlen = 65536;
334 	info->max_lro_pkt_size = MLX5_MAX_LRO_SIZE;
335 	/*
336 	 * Since we need one CQ per QP, the limit is the minimum number
337 	 * between the two values.
338 	 */
339 	max = RTE_MIN(priv->sh->dev_cap.max_cq, priv->sh->dev_cap.max_qp);
340 	/* max_rx_queues is uint16_t. */
341 	max = RTE_MIN(max, (unsigned int)UINT16_MAX);
342 	info->max_rx_queues = max;
343 	info->max_tx_queues = max;
344 	info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
345 	info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
346 	info->rx_seg_capa.max_nseg = MLX5_MAX_RXQ_NSEG;
347 	info->rx_seg_capa.multi_pools = !priv->config.mprq.enabled;
348 	info->rx_seg_capa.offset_allowed = !priv->config.mprq.enabled;
349 	info->rx_seg_capa.offset_align_log2 = 0;
350 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
351 				 info->rx_queue_offload_capa);
352 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
353 	info->dev_capa = RTE_ETH_DEV_CAPA_FLOW_SHARED_OBJECT_KEEP;
354 	info->if_index = mlx5_ifindex(dev);
355 	info->reta_size = priv->reta_idx_n ?
356 		priv->reta_idx_n : priv->sh->dev_cap.ind_table_max_size;
357 	info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
358 	info->speed_capa = priv->link_speed_capa;
359 	info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
360 	mlx5_set_default_params(dev, info);
361 	mlx5_set_txlimit_params(dev, info);
362 	info->rx_desc_lim.nb_max =
363 		1 << priv->sh->cdev->config.hca_attr.log_max_wq_sz;
364 	info->tx_desc_lim.nb_max =
365 		1 << priv->sh->cdev->config.hca_attr.log_max_wq_sz;
366 	if (priv->sh->cdev->config.hca_attr.mem_rq_rmp &&
367 	    priv->obj_ops.rxq_obj_new == devx_obj_ops.rxq_obj_new)
368 		info->dev_capa |= RTE_ETH_DEV_CAPA_RXQ_SHARE;
369 	info->switch_info.name = dev->data->name;
370 	info->switch_info.domain_id = priv->domain_id;
371 	info->switch_info.port_id = priv->representor_id;
372 	info->switch_info.rx_domain = 0; /* No sub Rx domains. */
373 	if (priv->representor) {
374 		uint16_t port_id;
375 
376 		MLX5_ETH_FOREACH_DEV(port_id, dev->device) {
377 			struct mlx5_priv *opriv =
378 				rte_eth_devices[port_id].data->dev_private;
379 
380 			if (!opriv ||
381 			    opriv->representor ||
382 			    opriv->sh != priv->sh ||
383 			    opriv->domain_id != priv->domain_id)
384 				continue;
385 			/*
386 			 * Override switch name with that of the master
387 			 * device.
388 			 */
389 			info->switch_info.name = opriv->dev_data->name;
390 			break;
391 		}
392 	}
393 	return 0;
394 }
395 
396 /**
397  * Calculate representor ID from port switch info.
398  *
399  * Uint16 representor ID bits definition:
400  *   pf: 2
401  *   type: 2
402  *   vf/sf: 12
403  *
404  * @param info
405  *   Port switch info.
406  * @param hpf_type
407  *   Use this type if port is HPF.
408  *
409  * @return
410  *   Encoded representor ID.
411  */
412 uint16_t
413 mlx5_representor_id_encode(const struct mlx5_switch_info *info,
414 			   enum rte_eth_representor_type hpf_type)
415 {
416 	enum rte_eth_representor_type type;
417 	uint16_t repr = info->port_name;
418 	int32_t pf = info->pf_num;
419 
420 	switch (info->name_type) {
421 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
422 		if (!info->representor)
423 			return UINT16_MAX;
424 		type = RTE_ETH_REPRESENTOR_PF;
425 		pf = info->mpesw_owner;
426 		break;
427 	case MLX5_PHYS_PORT_NAME_TYPE_PFSF:
428 		type = RTE_ETH_REPRESENTOR_SF;
429 		break;
430 	case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
431 		type = hpf_type;
432 		repr = UINT16_MAX;
433 		break;
434 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
435 	default:
436 		type = RTE_ETH_REPRESENTOR_VF;
437 		break;
438 	}
439 	return MLX5_REPRESENTOR_ID(pf, type, repr);
440 }
441 
442 /**
443  * DPDK callback to get information about representor.
444  *
445  * Representor ID bits definition:
446  *   vf/sf: 12
447  *   type: 2
448  *   pf: 2
449  *
450  * @param dev
451  *   Pointer to Ethernet device structure.
452  * @param[out] info
453  *   Nullable info structure output buffer.
454  *
455  * @return
456  *   negative on error, or the number of representor ranges.
457  */
458 int
459 mlx5_representor_info_get(struct rte_eth_dev *dev,
460 			  struct rte_eth_representor_info *info)
461 {
462 	struct mlx5_priv *priv = dev->data->dev_private;
463 	int n_type = 5; /* Representor types: PF, VF, HPF@VF, SF and HPF@SF. */
464 	int n_pf = 2; /* Number of PFs. */
465 	int i = 0, pf;
466 	int n_entries;
467 
468 	if (info == NULL)
469 		goto out;
470 
471 	n_entries = n_type * n_pf;
472 	if ((uint32_t)n_entries > info->nb_ranges_alloc)
473 		n_entries = info->nb_ranges_alloc;
474 
475 	info->controller = 0;
476 	info->pf = 0;
477 	if (mlx5_is_port_on_mpesw_device(priv)) {
478 		info->pf = priv->mpesw_port;
479 		/* PF range, both ports will show the same information. */
480 		info->ranges[i].type = RTE_ETH_REPRESENTOR_PF;
481 		info->ranges[i].controller = 0;
482 		info->ranges[i].pf = priv->mpesw_owner + 1;
483 		info->ranges[i].vf = 0;
484 		/*
485 		 * The representor indexes should be the values set of "priv->mpesw_port".
486 		 * In the real case now, only 1 PF/UPLINK representor is supported.
487 		 * The port index will always be the value of "owner + 1".
488 		 */
489 		info->ranges[i].id_base =
490 			MLX5_REPRESENTOR_ID(priv->mpesw_owner, info->ranges[i].type,
491 					    info->ranges[i].pf);
492 		info->ranges[i].id_end =
493 			MLX5_REPRESENTOR_ID(priv->mpesw_owner, info->ranges[i].type,
494 					    info->ranges[i].pf);
495 		snprintf(info->ranges[i].name, sizeof(info->ranges[i].name),
496 			 "pf%d", info->ranges[i].pf);
497 		i++;
498 	} else if (priv->pf_bond >= 0)
499 		info->pf = priv->pf_bond;
500 	for (pf = 0; pf < n_pf; ++pf) {
501 		/* VF range. */
502 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
503 		info->ranges[i].controller = 0;
504 		info->ranges[i].pf = pf;
505 		info->ranges[i].vf = 0;
506 		info->ranges[i].id_base =
507 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
508 		info->ranges[i].id_end =
509 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
510 		snprintf(info->ranges[i].name,
511 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
512 		i++;
513 		if (i == n_entries)
514 			break;
515 		/* HPF range of VF type. */
516 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
517 		info->ranges[i].controller = 0;
518 		info->ranges[i].pf = pf;
519 		info->ranges[i].vf = UINT16_MAX;
520 		info->ranges[i].id_base =
521 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
522 		info->ranges[i].id_end =
523 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
524 		snprintf(info->ranges[i].name,
525 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
526 		i++;
527 		if (i == n_entries)
528 			break;
529 		/* SF range. */
530 		info->ranges[i].type = RTE_ETH_REPRESENTOR_SF;
531 		info->ranges[i].controller = 0;
532 		info->ranges[i].pf = pf;
533 		info->ranges[i].vf = 0;
534 		info->ranges[i].id_base =
535 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
536 		info->ranges[i].id_end =
537 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
538 		snprintf(info->ranges[i].name,
539 			 sizeof(info->ranges[i].name), "pf%dsf", pf);
540 		i++;
541 		if (i == n_entries)
542 			break;
543 		/* HPF range of SF type. */
544 		info->ranges[i].type = RTE_ETH_REPRESENTOR_SF;
545 		info->ranges[i].controller = 0;
546 		info->ranges[i].pf = pf;
547 		info->ranges[i].vf = UINT16_MAX;
548 		info->ranges[i].id_base =
549 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
550 		info->ranges[i].id_end =
551 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
552 		snprintf(info->ranges[i].name,
553 			 sizeof(info->ranges[i].name), "pf%dsf", pf);
554 		i++;
555 		if (i == n_entries)
556 			break;
557 	}
558 	info->nb_ranges = i;
559 out:
560 	return n_type * n_pf;
561 }
562 
563 /**
564  * Get firmware version of a device.
565  *
566  * @param dev
567  *   Ethernet device port.
568  * @param fw_ver
569  *   String output allocated by caller.
570  * @param fw_size
571  *   Size of the output string, including terminating null byte.
572  *
573  * @return
574  *   0 on success, or the size of the non truncated string if too big.
575  */
576 int
577 mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
578 {
579 	struct mlx5_priv *priv = dev->data->dev_private;
580 	struct mlx5_dev_cap *attr = &priv->sh->dev_cap;
581 	size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
582 
583 	if (fw_size < size)
584 		return size;
585 	if (fw_ver != NULL)
586 		strlcpy(fw_ver, attr->fw_ver, fw_size);
587 	return 0;
588 }
589 
590 /**
591  * Get supported packet types.
592  *
593  * @param dev
594  *   Pointer to Ethernet device structure.
595  *
596  * @return
597  *   A pointer to the supported Packet types array.
598  */
599 const uint32_t *
600 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev, size_t *no_of_elements)
601 {
602 	static const uint32_t ptypes[] = {
603 		/* refers to rxq_cq_to_pkt_type() */
604 		RTE_PTYPE_L2_ETHER,
605 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
606 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
607 		RTE_PTYPE_L4_NONFRAG,
608 		RTE_PTYPE_L4_FRAG,
609 		RTE_PTYPE_L4_TCP,
610 		RTE_PTYPE_L4_UDP,
611 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
612 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
613 		RTE_PTYPE_INNER_L4_NONFRAG,
614 		RTE_PTYPE_INNER_L4_FRAG,
615 		RTE_PTYPE_INNER_L4_TCP,
616 		RTE_PTYPE_INNER_L4_UDP,
617 	};
618 
619 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
620 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
621 	    dev->rx_pkt_burst == mlx5_rx_burst_vec ||
622 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq_vec) {
623 		*no_of_elements = RTE_DIM(ptypes);
624 		return ptypes;
625 	}
626 	return NULL;
627 }
628 
629 /**
630  * DPDK callback to change the MTU.
631  *
632  * @param dev
633  *   Pointer to Ethernet device structure.
634  * @param in_mtu
635  *   New MTU.
636  *
637  * @return
638  *   0 on success, a negative errno value otherwise and rte_errno is set.
639  */
640 int
641 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
642 {
643 	struct mlx5_priv *priv = dev->data->dev_private;
644 	uint16_t kern_mtu = 0;
645 	int ret;
646 
647 	ret = mlx5_get_mtu(dev, &kern_mtu);
648 	if (ret)
649 		return ret;
650 	/* Set kernel interface MTU first. */
651 	ret = mlx5_set_mtu(dev, mtu);
652 	if (ret)
653 		return ret;
654 	ret = mlx5_get_mtu(dev, &kern_mtu);
655 	if (ret)
656 		return ret;
657 	if (kern_mtu == mtu) {
658 		priv->mtu = mtu;
659 		DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
660 			dev->data->port_id, mtu);
661 		return 0;
662 	}
663 	rte_errno = EAGAIN;
664 	return -rte_errno;
665 }
666 
667 /**
668  * Configure the RX function to use.
669  *
670  * @param dev
671  *   Pointer to private data structure.
672  *
673  * @return
674  *   Pointer to selected Rx burst function.
675  */
676 eth_rx_burst_t
677 mlx5_select_rx_function(struct rte_eth_dev *dev)
678 {
679 	eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
680 
681 	MLX5_ASSERT(dev != NULL);
682 	if (mlx5_check_vec_rx_support(dev) > 0) {
683 		if (mlx5_mprq_enabled(dev)) {
684 			rx_pkt_burst = mlx5_rx_burst_mprq_vec;
685 			DRV_LOG(DEBUG, "port %u selected vectorized"
686 				" MPRQ Rx function", dev->data->port_id);
687 		} else {
688 			rx_pkt_burst = mlx5_rx_burst_vec;
689 			DRV_LOG(DEBUG, "port %u selected vectorized"
690 				" SPRQ Rx function", dev->data->port_id);
691 		}
692 	} else if (mlx5_mprq_enabled(dev)) {
693 		rx_pkt_burst = mlx5_rx_burst_mprq;
694 		DRV_LOG(DEBUG, "port %u selected MPRQ Rx function",
695 			dev->data->port_id);
696 	} else {
697 		DRV_LOG(DEBUG, "port %u selected SPRQ Rx function",
698 			dev->data->port_id);
699 	}
700 	return rx_pkt_burst;
701 }
702 
703 /**
704  * Get the E-Switch parameters by port id.
705  *
706  * @param[in] port
707  *   Device port id.
708  * @param[in] valid
709  *   Device port id is valid, skip check. This flag is useful
710  *   when trials are performed from probing and device is not
711  *   flagged as valid yet (in attaching process).
712  * @param[out] es_domain_id
713  *   E-Switch domain id.
714  * @param[out] es_port_id
715  *   The port id of the port in the E-Switch.
716  *
717  * @return
718  *   pointer to device private data structure containing data needed
719  *   on success, NULL otherwise and rte_errno is set.
720  */
721 struct mlx5_priv *
722 mlx5_port_to_eswitch_info(uint16_t port, bool valid)
723 {
724 	struct rte_eth_dev *dev;
725 	struct mlx5_priv *priv;
726 
727 	if (port >= RTE_MAX_ETHPORTS) {
728 		rte_errno = EINVAL;
729 		return NULL;
730 	}
731 	if (!valid && !rte_eth_dev_is_valid_port(port)) {
732 		rte_errno = ENODEV;
733 		return NULL;
734 	}
735 	dev = &rte_eth_devices[port];
736 	priv = dev->data->dev_private;
737 	if (!priv->sh->esw_mode) {
738 		rte_errno = EINVAL;
739 		return NULL;
740 	}
741 	return priv;
742 }
743 
744 /**
745  * Get the E-Switch parameters by device instance.
746  *
747  * @param[in] port
748  *   Device port id.
749  * @param[out] es_domain_id
750  *   E-Switch domain id.
751  * @param[out] es_port_id
752  *   The port id of the port in the E-Switch.
753  *
754  * @return
755  *   pointer to device private data structure containing data needed
756  *   on success, NULL otherwise and rte_errno is set.
757  */
758 struct mlx5_priv *
759 mlx5_dev_to_eswitch_info(struct rte_eth_dev *dev)
760 {
761 	struct mlx5_priv *priv;
762 
763 	priv = dev->data->dev_private;
764 	if (!priv->sh->esw_mode) {
765 		rte_errno = EINVAL;
766 		return NULL;
767 	}
768 	return priv;
769 }
770 
771 /**
772  * DPDK callback to retrieve hairpin capabilities.
773  *
774  * @param dev
775  *   Pointer to Ethernet device structure.
776  * @param[out] cap
777  *   Storage for hairpin capability data.
778  *
779  * @return
780  *   0 on success, a negative errno value otherwise and rte_errno is set.
781  */
782 int
783 mlx5_hairpin_cap_get(struct rte_eth_dev *dev, struct rte_eth_hairpin_cap *cap)
784 {
785 	struct mlx5_priv *priv = dev->data->dev_private;
786 	struct mlx5_hca_attr *hca_attr;
787 
788 	if (!mlx5_devx_obj_ops_en(priv->sh)) {
789 		rte_errno = ENOTSUP;
790 		return -rte_errno;
791 	}
792 	cap->max_nb_queues = UINT16_MAX;
793 	cap->max_rx_2_tx = 1;
794 	cap->max_tx_2_rx = 1;
795 	cap->max_nb_desc = 8192;
796 	hca_attr = &priv->sh->cdev->config.hca_attr;
797 	cap->rx_cap.locked_device_memory = hca_attr->hairpin_data_buffer_locked;
798 	cap->rx_cap.rte_memory = 0;
799 	cap->tx_cap.locked_device_memory = 0;
800 	cap->tx_cap.rte_memory = hca_attr->hairpin_sq_wq_in_host_mem;
801 	return 0;
802 }
803 
804 /**
805  * Indicate to ethdev layer, what configuration must be restored.
806  *
807  * @param[in] dev
808  *   Pointer to Ethernet device structure.
809  * @param[in] op
810  *   Type of operation which might require.
811  * @param[out] flags
812  *   Restore flags will be stored here.
813  */
814 uint64_t
815 mlx5_get_restore_flags(__rte_unused struct rte_eth_dev *dev,
816 		       __rte_unused enum rte_eth_dev_operation op)
817 {
818 	/* mlx5 PMD does not require any configuration restore. */
819 	return 0;
820 }
821