1 /*
2  * Copyright (c) 2017, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <rdma/ib_verbs.h>
34 #include <linux/mlx5/fs.h>
35 #include "en.h"
36 #include "ipoib.h"
37 
38 #define IB_DEFAULT_Q_KEY   0xb1b
39 #define MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE 9
40 
41 static int mlx5i_open(struct net_device *netdev);
42 static int mlx5i_close(struct net_device *netdev);
43 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu);
44 
45 static const struct net_device_ops mlx5i_netdev_ops = {
46 	.ndo_open                = mlx5i_open,
47 	.ndo_stop                = mlx5i_close,
48 	.ndo_get_stats64         = mlx5i_get_stats,
49 	.ndo_init                = mlx5i_dev_init,
50 	.ndo_uninit              = mlx5i_dev_cleanup,
51 	.ndo_change_mtu          = mlx5i_change_mtu,
52 	.ndo_do_ioctl            = mlx5i_ioctl,
53 };
54 
55 /* IPoIB mlx5 netdev profile */
56 static void mlx5i_build_nic_params(struct mlx5_core_dev *mdev,
57 				   struct mlx5e_params *params)
58 {
59 	/* Override RQ params as IPoIB supports only LINKED LIST RQ for now */
60 	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ, false);
61 	mlx5e_set_rq_type(mdev, params);
62 	mlx5e_init_rq_type_params(mdev, params);
63 
64 	/* RQ size in ipoib by default is 512 */
65 	params->log_rq_mtu_frames = is_kdump_kernel() ?
66 		MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
67 		MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE;
68 
69 	params->lro_en = false;
70 	params->hard_mtu = MLX5_IB_GRH_BYTES + MLX5_IPOIB_HARD_LEN;
71 	params->tunneled_offload_en = false;
72 }
73 
74 /* Called directly after IPoIB netdevice was created to initialize SW structs */
75 int mlx5i_init(struct mlx5_core_dev *mdev,
76 	       struct net_device *netdev,
77 	       const struct mlx5e_profile *profile,
78 	       void *ppriv)
79 {
80 	struct mlx5e_priv *priv  = mlx5i_epriv(netdev);
81 	int err;
82 
83 	err = mlx5e_netdev_init(netdev, priv, mdev, profile, ppriv);
84 	if (err)
85 		return err;
86 
87 	mlx5e_set_netdev_mtu_boundaries(priv);
88 	netdev->mtu = netdev->max_mtu;
89 
90 	mlx5e_build_nic_params(priv, NULL, &priv->rss_params, &priv->channels.params,
91 			       netdev->mtu);
92 	mlx5i_build_nic_params(mdev, &priv->channels.params);
93 
94 	mlx5e_timestamp_init(priv);
95 
96 	/* netdev init */
97 	netdev->hw_features    |= NETIF_F_SG;
98 	netdev->hw_features    |= NETIF_F_IP_CSUM;
99 	netdev->hw_features    |= NETIF_F_IPV6_CSUM;
100 	netdev->hw_features    |= NETIF_F_GRO;
101 	netdev->hw_features    |= NETIF_F_TSO;
102 	netdev->hw_features    |= NETIF_F_TSO6;
103 	netdev->hw_features    |= NETIF_F_RXCSUM;
104 	netdev->hw_features    |= NETIF_F_RXHASH;
105 
106 	netdev->netdev_ops = &mlx5i_netdev_ops;
107 	netdev->ethtool_ops = &mlx5i_ethtool_ops;
108 
109 	return 0;
110 }
111 
112 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
113 void mlx5i_cleanup(struct mlx5e_priv *priv)
114 {
115 	mlx5e_netdev_cleanup(priv->netdev, priv);
116 }
117 
118 static void mlx5i_grp_sw_update_stats(struct mlx5e_priv *priv)
119 {
120 	struct mlx5e_sw_stats s = { 0 };
121 	int i, j;
122 
123 	for (i = 0; i < priv->max_nch; i++) {
124 		struct mlx5e_channel_stats *channel_stats;
125 		struct mlx5e_rq_stats *rq_stats;
126 
127 		channel_stats = &priv->channel_stats[i];
128 		rq_stats = &channel_stats->rq;
129 
130 		s.rx_packets += rq_stats->packets;
131 		s.rx_bytes   += rq_stats->bytes;
132 
133 		for (j = 0; j < priv->max_opened_tc; j++) {
134 			struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
135 
136 			s.tx_packets           += sq_stats->packets;
137 			s.tx_bytes             += sq_stats->bytes;
138 			s.tx_queue_dropped     += sq_stats->dropped;
139 		}
140 	}
141 
142 	memcpy(&priv->stats.sw, &s, sizeof(s));
143 }
144 
145 void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
146 {
147 	struct mlx5e_priv     *priv   = mlx5i_epriv(dev);
148 	struct mlx5e_sw_stats *sstats = &priv->stats.sw;
149 
150 	mlx5i_grp_sw_update_stats(priv);
151 
152 	stats->rx_packets = sstats->rx_packets;
153 	stats->rx_bytes   = sstats->rx_bytes;
154 	stats->tx_packets = sstats->tx_packets;
155 	stats->tx_bytes   = sstats->tx_bytes;
156 	stats->tx_dropped = sstats->tx_queue_dropped;
157 }
158 
159 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
160 {
161 	struct mlx5_core_dev *mdev = priv->mdev;
162 	struct mlx5i_priv *ipriv = priv->ppriv;
163 	struct mlx5_core_qp *qp = &ipriv->qp;
164 	struct mlx5_qp_context *context;
165 	int ret;
166 
167 	/* QP states */
168 	context = kzalloc(sizeof(*context), GFP_KERNEL);
169 	if (!context)
170 		return -ENOMEM;
171 
172 	context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
173 	context->pri_path.port = 1;
174 	context->pri_path.pkey_index = cpu_to_be16(ipriv->pkey_index);
175 	context->qkey = cpu_to_be32(IB_DEFAULT_Q_KEY);
176 
177 	ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RST2INIT_QP, 0, context, qp);
178 	if (ret) {
179 		mlx5_core_err(mdev, "Failed to modify qp RST2INIT, err: %d\n", ret);
180 		goto err_qp_modify_to_err;
181 	}
182 	memset(context, 0, sizeof(*context));
183 
184 	ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, context, qp);
185 	if (ret) {
186 		mlx5_core_err(mdev, "Failed to modify qp INIT2RTR, err: %d\n", ret);
187 		goto err_qp_modify_to_err;
188 	}
189 
190 	ret = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, 0, context, qp);
191 	if (ret) {
192 		mlx5_core_err(mdev, "Failed to modify qp RTR2RTS, err: %d\n", ret);
193 		goto err_qp_modify_to_err;
194 	}
195 
196 	kfree(context);
197 	return 0;
198 
199 err_qp_modify_to_err:
200 	mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2ERR_QP, 0, &context, qp);
201 	kfree(context);
202 	return ret;
203 }
204 
205 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
206 {
207 	struct mlx5i_priv *ipriv = priv->ppriv;
208 	struct mlx5_core_dev *mdev = priv->mdev;
209 	struct mlx5_qp_context context;
210 	int err;
211 
212 	err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, &context,
213 				  &ipriv->qp);
214 	if (err)
215 		mlx5_core_err(mdev, "Failed to modify qp 2RST, err: %d\n", err);
216 }
217 
218 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
219 
220 int mlx5i_create_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
221 {
222 	u32 *in = NULL;
223 	void *addr_path;
224 	int ret = 0;
225 	int inlen;
226 	void *qpc;
227 
228 	inlen = MLX5_ST_SZ_BYTES(create_qp_in);
229 	in = kvzalloc(inlen, GFP_KERNEL);
230 	if (!in)
231 		return -ENOMEM;
232 
233 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
234 	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
235 	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
236 	MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
237 		 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
238 
239 	addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
240 	MLX5_SET(ads, addr_path, vhca_port_num, 1);
241 	MLX5_SET(ads, addr_path, grh, 1);
242 
243 	ret = mlx5_core_create_qp(mdev, qp, in, inlen);
244 	if (ret) {
245 		mlx5_core_err(mdev, "Failed creating IPoIB QP err : %d\n", ret);
246 		goto out;
247 	}
248 
249 out:
250 	kvfree(in);
251 	return ret;
252 }
253 
254 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, struct mlx5_core_qp *qp)
255 {
256 	mlx5_core_destroy_qp(mdev, qp);
257 }
258 
259 int mlx5i_create_tis(struct mlx5_core_dev *mdev, u32 underlay_qpn, u32 *tisn)
260 {
261 	u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
262 	void *tisc;
263 
264 	tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
265 
266 	MLX5_SET(tisc, tisc, underlay_qpn, underlay_qpn);
267 
268 	return mlx5e_create_tis(mdev, in, tisn);
269 }
270 
271 static int mlx5i_init_tx(struct mlx5e_priv *priv)
272 {
273 	struct mlx5i_priv *ipriv = priv->ppriv;
274 	int err;
275 
276 	err = mlx5i_create_underlay_qp(priv->mdev, &ipriv->qp);
277 	if (err) {
278 		mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
279 		return err;
280 	}
281 
282 	err = mlx5i_create_tis(priv->mdev, ipriv->qp.qpn, &priv->tisn[0][0]);
283 	if (err) {
284 		mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
285 		goto err_destroy_underlay_qp;
286 	}
287 
288 	return 0;
289 
290 err_destroy_underlay_qp:
291 	mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
292 	return err;
293 }
294 
295 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
296 {
297 	struct mlx5i_priv *ipriv = priv->ppriv;
298 
299 	mlx5e_destroy_tis(priv->mdev, priv->tisn[0][0]);
300 	mlx5i_destroy_underlay_qp(priv->mdev, &ipriv->qp);
301 }
302 
303 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
304 {
305 	struct ttc_params ttc_params = {};
306 	int tt, err;
307 
308 	priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
309 					       MLX5_FLOW_NAMESPACE_KERNEL);
310 
311 	if (!priv->fs.ns)
312 		return -EINVAL;
313 
314 	err = mlx5e_arfs_create_tables(priv);
315 	if (err) {
316 		netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
317 			   err);
318 		priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
319 	}
320 
321 	mlx5e_set_ttc_basic_params(priv, &ttc_params);
322 	mlx5e_set_inner_ttc_ft_params(&ttc_params);
323 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
324 		ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
325 
326 	err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
327 	if (err) {
328 		netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
329 			   err);
330 		goto err_destroy_arfs_tables;
331 	}
332 
333 	mlx5e_set_ttc_ft_params(&ttc_params);
334 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
335 		ttc_params.indir_tirn[tt] = priv->indir_tir[tt].tirn;
336 
337 	err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
338 	if (err) {
339 		netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
340 			   err);
341 		goto err_destroy_inner_ttc_table;
342 	}
343 
344 	return 0;
345 
346 err_destroy_inner_ttc_table:
347 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
348 err_destroy_arfs_tables:
349 	mlx5e_arfs_destroy_tables(priv);
350 
351 	return err;
352 }
353 
354 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
355 {
356 	mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
357 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
358 	mlx5e_arfs_destroy_tables(priv);
359 }
360 
361 static int mlx5i_init_rx(struct mlx5e_priv *priv)
362 {
363 	struct mlx5_core_dev *mdev = priv->mdev;
364 	int err;
365 
366 	mlx5e_create_q_counters(priv);
367 
368 	err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
369 	if (err) {
370 		mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
371 		goto err_destroy_q_counters;
372 	}
373 
374 	err = mlx5e_create_indirect_rqt(priv);
375 	if (err)
376 		goto err_close_drop_rq;
377 
378 	err = mlx5e_create_direct_rqts(priv, priv->direct_tir);
379 	if (err)
380 		goto err_destroy_indirect_rqts;
381 
382 	err = mlx5e_create_indirect_tirs(priv, true);
383 	if (err)
384 		goto err_destroy_direct_rqts;
385 
386 	err = mlx5e_create_direct_tirs(priv, priv->direct_tir);
387 	if (err)
388 		goto err_destroy_indirect_tirs;
389 
390 	err = mlx5i_create_flow_steering(priv);
391 	if (err)
392 		goto err_destroy_direct_tirs;
393 
394 	return 0;
395 
396 err_destroy_direct_tirs:
397 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
398 err_destroy_indirect_tirs:
399 	mlx5e_destroy_indirect_tirs(priv, true);
400 err_destroy_direct_rqts:
401 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
402 err_destroy_indirect_rqts:
403 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
404 err_close_drop_rq:
405 	mlx5e_close_drop_rq(&priv->drop_rq);
406 err_destroy_q_counters:
407 	mlx5e_destroy_q_counters(priv);
408 	return err;
409 }
410 
411 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
412 {
413 	mlx5i_destroy_flow_steering(priv);
414 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
415 	mlx5e_destroy_indirect_tirs(priv, true);
416 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
417 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
418 	mlx5e_close_drop_rq(&priv->drop_rq);
419 	mlx5e_destroy_q_counters(priv);
420 }
421 
422 /* The stats groups order is opposite to the update_stats() order calls */
423 static mlx5e_stats_grp_t mlx5i_stats_grps[] = {
424 	&MLX5E_STATS_GRP(sw),
425 	&MLX5E_STATS_GRP(qcnt),
426 	&MLX5E_STATS_GRP(vnic_env),
427 	&MLX5E_STATS_GRP(vport),
428 	&MLX5E_STATS_GRP(802_3),
429 	&MLX5E_STATS_GRP(2863),
430 	&MLX5E_STATS_GRP(2819),
431 	&MLX5E_STATS_GRP(phy),
432 	&MLX5E_STATS_GRP(pcie),
433 	&MLX5E_STATS_GRP(per_prio),
434 	&MLX5E_STATS_GRP(pme),
435 	&MLX5E_STATS_GRP(channels),
436 	&MLX5E_STATS_GRP(per_port_buff_congest),
437 };
438 
439 static unsigned int mlx5i_stats_grps_num(struct mlx5e_priv *priv)
440 {
441 	return ARRAY_SIZE(mlx5i_stats_grps);
442 }
443 
444 static const struct mlx5e_profile mlx5i_nic_profile = {
445 	.init		   = mlx5i_init,
446 	.cleanup	   = mlx5i_cleanup,
447 	.init_tx	   = mlx5i_init_tx,
448 	.cleanup_tx	   = mlx5i_cleanup_tx,
449 	.init_rx	   = mlx5i_init_rx,
450 	.cleanup_rx	   = mlx5i_cleanup_rx,
451 	.enable		   = NULL, /* mlx5i_enable */
452 	.disable	   = NULL, /* mlx5i_disable */
453 	.update_rx	   = mlx5e_update_nic_rx,
454 	.update_stats	   = NULL, /* mlx5i_update_stats */
455 	.update_carrier    = NULL, /* no HW update in IB link */
456 	.rx_handlers.handle_rx_cqe       = mlx5i_handle_rx_cqe,
457 	.rx_handlers.handle_rx_cqe_mpwqe = NULL, /* Not supported */
458 	.max_tc		   = MLX5I_MAX_NUM_TC,
459 	.rq_groups	   = MLX5E_NUM_RQ_GROUPS(REGULAR),
460 	.stats_grps        = mlx5i_stats_grps,
461 	.stats_grps_num    = mlx5i_stats_grps_num,
462 };
463 
464 /* mlx5i netdev NDos */
465 
466 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
467 {
468 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
469 	struct mlx5e_channels new_channels = {};
470 	struct mlx5e_params *params;
471 	int err = 0;
472 
473 	mutex_lock(&priv->state_lock);
474 
475 	params = &priv->channels.params;
476 
477 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
478 		params->sw_mtu = new_mtu;
479 		netdev->mtu = params->sw_mtu;
480 		goto out;
481 	}
482 
483 	new_channels.params = *params;
484 	new_channels.params.sw_mtu = new_mtu;
485 
486 	err = mlx5e_safe_switch_channels(priv, &new_channels, NULL);
487 	if (err)
488 		goto out;
489 
490 	netdev->mtu = new_channels.params.sw_mtu;
491 
492 out:
493 	mutex_unlock(&priv->state_lock);
494 	return err;
495 }
496 
497 int mlx5i_dev_init(struct net_device *dev)
498 {
499 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
500 	struct mlx5i_priv    *ipriv  = priv->ppriv;
501 
502 	/* Set dev address using underlay QP */
503 	dev->dev_addr[1] = (ipriv->qp.qpn >> 16) & 0xff;
504 	dev->dev_addr[2] = (ipriv->qp.qpn >>  8) & 0xff;
505 	dev->dev_addr[3] = (ipriv->qp.qpn) & 0xff;
506 
507 	/* Add QPN to net-device mapping to HT */
508 	mlx5i_pkey_add_qpn(dev ,ipriv->qp.qpn);
509 
510 	return 0;
511 }
512 
513 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
514 {
515 	struct mlx5e_priv *priv = mlx5i_epriv(dev);
516 
517 	switch (cmd) {
518 	case SIOCSHWTSTAMP:
519 		return mlx5e_hwstamp_set(priv, ifr);
520 	case SIOCGHWTSTAMP:
521 		return mlx5e_hwstamp_get(priv, ifr);
522 	default:
523 		return -EOPNOTSUPP;
524 	}
525 }
526 
527 void mlx5i_dev_cleanup(struct net_device *dev)
528 {
529 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
530 	struct mlx5i_priv    *ipriv = priv->ppriv;
531 
532 	mlx5i_uninit_underlay_qp(priv);
533 
534 	/* Delete QPN to net-device mapping from HT */
535 	mlx5i_pkey_del_qpn(dev, ipriv->qp.qpn);
536 }
537 
538 static int mlx5i_open(struct net_device *netdev)
539 {
540 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
541 	struct mlx5i_priv *ipriv = epriv->ppriv;
542 	struct mlx5_core_dev *mdev = epriv->mdev;
543 	int err;
544 
545 	mutex_lock(&epriv->state_lock);
546 
547 	set_bit(MLX5E_STATE_OPENED, &epriv->state);
548 
549 	err = mlx5i_init_underlay_qp(epriv);
550 	if (err) {
551 		mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
552 		goto err_clear_state_opened_flag;
553 	}
554 
555 	err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qp.qpn);
556 	if (err) {
557 		mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
558 		goto err_reset_qp;
559 	}
560 
561 	err = mlx5e_open_channels(epriv, &epriv->channels);
562 	if (err)
563 		goto err_remove_fs_underlay_qp;
564 
565 	epriv->profile->update_rx(epriv);
566 	mlx5e_activate_priv_channels(epriv);
567 
568 	mutex_unlock(&epriv->state_lock);
569 	return 0;
570 
571 err_remove_fs_underlay_qp:
572 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
573 err_reset_qp:
574 	mlx5i_uninit_underlay_qp(epriv);
575 err_clear_state_opened_flag:
576 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
577 	mutex_unlock(&epriv->state_lock);
578 	return err;
579 }
580 
581 static int mlx5i_close(struct net_device *netdev)
582 {
583 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
584 	struct mlx5i_priv *ipriv = epriv->ppriv;
585 	struct mlx5_core_dev *mdev = epriv->mdev;
586 
587 	/* May already be CLOSED in case a previous configuration operation
588 	 * (e.g RX/TX queue size change) that involves close&open failed.
589 	 */
590 	mutex_lock(&epriv->state_lock);
591 
592 	if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
593 		goto unlock;
594 
595 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
596 
597 	netif_carrier_off(epriv->netdev);
598 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qp.qpn);
599 	mlx5e_deactivate_priv_channels(epriv);
600 	mlx5e_close_channels(&epriv->channels);
601 	mlx5i_uninit_underlay_qp(epriv);
602 unlock:
603 	mutex_unlock(&epriv->state_lock);
604 	return 0;
605 }
606 
607 /* IPoIB RDMA netdev callbacks */
608 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
609 			      union ib_gid *gid, u16 lid, int set_qkey,
610 			      u32 qkey)
611 {
612 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
613 	struct mlx5_core_dev *mdev  = epriv->mdev;
614 	struct mlx5i_priv    *ipriv = epriv->ppriv;
615 	int err;
616 
617 	mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
618 	err = mlx5_core_attach_mcg(mdev, gid, ipriv->qp.qpn);
619 	if (err)
620 		mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
621 			       ipriv->qp.qpn, gid->raw);
622 
623 	if (set_qkey) {
624 		mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
625 			      netdev->name, qkey);
626 		ipriv->qkey = qkey;
627 	}
628 
629 	return err;
630 }
631 
632 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
633 			      union ib_gid *gid, u16 lid)
634 {
635 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
636 	struct mlx5_core_dev *mdev  = epriv->mdev;
637 	struct mlx5i_priv    *ipriv = epriv->ppriv;
638 	int err;
639 
640 	mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qp.qpn, gid->raw);
641 
642 	err = mlx5_core_detach_mcg(mdev, gid, ipriv->qp.qpn);
643 	if (err)
644 		mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
645 			      ipriv->qp.qpn, gid->raw);
646 
647 	return err;
648 }
649 
650 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
651 		      struct ib_ah *address, u32 dqpn)
652 {
653 	struct mlx5e_priv *epriv = mlx5i_epriv(dev);
654 	struct mlx5e_txqsq *sq   = epriv->txq2sq[skb_get_queue_mapping(skb)];
655 	struct mlx5_ib_ah *mah   = to_mah(address);
656 	struct mlx5i_priv *ipriv = epriv->ppriv;
657 
658 	return mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey, netdev_xmit_more());
659 }
660 
661 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
662 {
663 	struct mlx5i_priv *ipriv = netdev_priv(netdev);
664 
665 	ipriv->pkey_index = (u16)id;
666 }
667 
668 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
669 {
670 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
671 		return -EOPNOTSUPP;
672 
673 	if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
674 		mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
675 		return -EOPNOTSUPP;
676 	}
677 
678 	return 0;
679 }
680 
681 static void mlx5_rdma_netdev_free(struct net_device *netdev)
682 {
683 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
684 	struct mlx5i_priv *ipriv = priv->ppriv;
685 	const struct mlx5e_profile *profile = priv->profile;
686 
687 	mlx5e_detach_netdev(priv);
688 	profile->cleanup(priv);
689 
690 	if (!ipriv->sub_interface) {
691 		mlx5i_pkey_qpn_ht_cleanup(netdev);
692 		mlx5e_destroy_mdev_resources(priv->mdev);
693 	}
694 }
695 
696 static bool mlx5_is_sub_interface(struct mlx5_core_dev *mdev)
697 {
698 	return mdev->mlx5e_res.pdn != 0;
699 }
700 
701 static const struct mlx5e_profile *mlx5_get_profile(struct mlx5_core_dev *mdev)
702 {
703 	if (mlx5_is_sub_interface(mdev))
704 		return mlx5i_pkey_get_profile();
705 	return &mlx5i_nic_profile;
706 }
707 
708 static int mlx5_rdma_setup_rn(struct ib_device *ibdev, u8 port_num,
709 			      struct net_device *netdev, void *param)
710 {
711 	struct mlx5_core_dev *mdev = (struct mlx5_core_dev *)param;
712 	const struct mlx5e_profile *prof = mlx5_get_profile(mdev);
713 	struct mlx5i_priv *ipriv;
714 	struct mlx5e_priv *epriv;
715 	struct rdma_netdev *rn;
716 	int err;
717 
718 	ipriv = netdev_priv(netdev);
719 	epriv = mlx5i_epriv(netdev);
720 
721 	ipriv->sub_interface = mlx5_is_sub_interface(mdev);
722 	if (!ipriv->sub_interface) {
723 		err = mlx5i_pkey_qpn_ht_init(netdev);
724 		if (err) {
725 			mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
726 			return err;
727 		}
728 
729 		/* This should only be called once per mdev */
730 		err = mlx5e_create_mdev_resources(mdev);
731 		if (err)
732 			goto destroy_ht;
733 	}
734 
735 	prof->init(mdev, netdev, prof, ipriv);
736 
737 	err = mlx5e_attach_netdev(epriv);
738 	if (err)
739 		goto detach;
740 	netif_carrier_off(netdev);
741 
742 	/* set rdma_netdev func pointers */
743 	rn = &ipriv->rn;
744 	rn->hca  = ibdev;
745 	rn->send = mlx5i_xmit;
746 	rn->attach_mcast = mlx5i_attach_mcast;
747 	rn->detach_mcast = mlx5i_detach_mcast;
748 	rn->set_id = mlx5i_set_pkey_index;
749 
750 	netdev->priv_destructor = mlx5_rdma_netdev_free;
751 	netdev->needs_free_netdev = 1;
752 
753 	return 0;
754 
755 detach:
756 	prof->cleanup(epriv);
757 	if (ipriv->sub_interface)
758 		return err;
759 	mlx5e_destroy_mdev_resources(mdev);
760 destroy_ht:
761 	mlx5i_pkey_qpn_ht_cleanup(netdev);
762 	return err;
763 }
764 
765 int mlx5_rdma_rn_get_params(struct mlx5_core_dev *mdev,
766 			    struct ib_device *device,
767 			    struct rdma_netdev_alloc_params *params)
768 {
769 	int nch;
770 	int rc;
771 
772 	rc = mlx5i_check_required_hca_cap(mdev);
773 	if (rc)
774 		return rc;
775 
776 	nch = mlx5e_get_max_num_channels(mdev);
777 
778 	*params = (struct rdma_netdev_alloc_params){
779 		.sizeof_priv = sizeof(struct mlx5i_priv) +
780 			       sizeof(struct mlx5e_priv),
781 		.txqs = nch * MLX5E_MAX_NUM_TC,
782 		.rxqs = nch,
783 		.param = mdev,
784 		.initialize_rdma_netdev = mlx5_rdma_setup_rn,
785 	};
786 
787 	return 0;
788 }
789 EXPORT_SYMBOL(mlx5_rdma_rn_get_params);
790