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, struct net_device *netdev)
76 {
77 	struct mlx5e_priv *priv  = mlx5i_epriv(netdev);
78 
79 	netif_carrier_off(netdev);
80 	mlx5e_set_netdev_mtu_boundaries(priv);
81 	netdev->mtu = netdev->max_mtu;
82 
83 	mlx5e_build_nic_params(priv, NULL, netdev->mtu);
84 	mlx5i_build_nic_params(mdev, &priv->channels.params);
85 
86 	mlx5e_timestamp_init(priv);
87 
88 	/* netdev init */
89 	netdev->hw_features    |= NETIF_F_SG;
90 	netdev->hw_features    |= NETIF_F_IP_CSUM;
91 	netdev->hw_features    |= NETIF_F_IPV6_CSUM;
92 	netdev->hw_features    |= NETIF_F_GRO;
93 	netdev->hw_features    |= NETIF_F_TSO;
94 	netdev->hw_features    |= NETIF_F_TSO6;
95 	netdev->hw_features    |= NETIF_F_RXCSUM;
96 	netdev->hw_features    |= NETIF_F_RXHASH;
97 
98 	netdev->netdev_ops = &mlx5i_netdev_ops;
99 	netdev->ethtool_ops = &mlx5i_ethtool_ops;
100 
101 	return 0;
102 }
103 
104 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
105 void mlx5i_cleanup(struct mlx5e_priv *priv)
106 {
107 	mlx5e_priv_cleanup(priv);
108 }
109 
110 static void mlx5i_grp_sw_update_stats(struct mlx5e_priv *priv)
111 {
112 	struct mlx5e_sw_stats s = { 0 };
113 	int i, j;
114 
115 	for (i = 0; i < priv->max_nch; i++) {
116 		struct mlx5e_channel_stats *channel_stats;
117 		struct mlx5e_rq_stats *rq_stats;
118 
119 		channel_stats = &priv->channel_stats[i];
120 		rq_stats = &channel_stats->rq;
121 
122 		s.rx_packets += rq_stats->packets;
123 		s.rx_bytes   += rq_stats->bytes;
124 
125 		for (j = 0; j < priv->max_opened_tc; j++) {
126 			struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
127 
128 			s.tx_packets           += sq_stats->packets;
129 			s.tx_bytes             += sq_stats->bytes;
130 			s.tx_queue_dropped     += sq_stats->dropped;
131 		}
132 	}
133 
134 	memcpy(&priv->stats.sw, &s, sizeof(s));
135 }
136 
137 void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
138 {
139 	struct mlx5e_priv     *priv   = mlx5i_epriv(dev);
140 	struct mlx5e_sw_stats *sstats = &priv->stats.sw;
141 
142 	mlx5i_grp_sw_update_stats(priv);
143 
144 	stats->rx_packets = sstats->rx_packets;
145 	stats->rx_bytes   = sstats->rx_bytes;
146 	stats->tx_packets = sstats->tx_packets;
147 	stats->tx_bytes   = sstats->tx_bytes;
148 	stats->tx_dropped = sstats->tx_queue_dropped;
149 }
150 
151 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
152 {
153 	struct mlx5_core_dev *mdev = priv->mdev;
154 	struct mlx5i_priv *ipriv = priv->ppriv;
155 	int ret;
156 
157 	{
158 		u32 in[MLX5_ST_SZ_DW(rst2init_qp_in)] = {};
159 		u32 *qpc;
160 
161 		qpc = MLX5_ADDR_OF(rst2init_qp_in, in, qpc);
162 
163 		MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
164 		MLX5_SET(qpc, qpc, primary_address_path.pkey_index,
165 			 ipriv->pkey_index);
166 		MLX5_SET(qpc, qpc, primary_address_path.vhca_port_num, 1);
167 		MLX5_SET(qpc, qpc, q_key, IB_DEFAULT_Q_KEY);
168 
169 		MLX5_SET(rst2init_qp_in, in, opcode, MLX5_CMD_OP_RST2INIT_QP);
170 		MLX5_SET(rst2init_qp_in, in, qpn, ipriv->qpn);
171 		ret = mlx5_cmd_exec_in(mdev, rst2init_qp, in);
172 		if (ret)
173 			goto err_qp_modify_to_err;
174 	}
175 	{
176 		u32 in[MLX5_ST_SZ_DW(init2rtr_qp_in)] = {};
177 
178 		MLX5_SET(init2rtr_qp_in, in, opcode, MLX5_CMD_OP_INIT2RTR_QP);
179 		MLX5_SET(init2rtr_qp_in, in, qpn, ipriv->qpn);
180 		ret = mlx5_cmd_exec_in(mdev, init2rtr_qp, in);
181 		if (ret)
182 			goto err_qp_modify_to_err;
183 	}
184 	{
185 		u32 in[MLX5_ST_SZ_DW(rtr2rts_qp_in)] = {};
186 
187 		MLX5_SET(rtr2rts_qp_in, in, opcode, MLX5_CMD_OP_RTR2RTS_QP);
188 		MLX5_SET(rtr2rts_qp_in, in, qpn, ipriv->qpn);
189 		ret = mlx5_cmd_exec_in(mdev, rtr2rts_qp, in);
190 		if (ret)
191 			goto err_qp_modify_to_err;
192 	}
193 	return 0;
194 
195 err_qp_modify_to_err:
196 	{
197 		u32 in[MLX5_ST_SZ_DW(qp_2err_in)] = {};
198 
199 		MLX5_SET(qp_2err_in, in, opcode, MLX5_CMD_OP_2ERR_QP);
200 		MLX5_SET(qp_2err_in, in, qpn, ipriv->qpn);
201 		mlx5_cmd_exec_in(mdev, qp_2err, in);
202 	}
203 	return ret;
204 }
205 
206 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
207 {
208 	struct mlx5i_priv *ipriv = priv->ppriv;
209 	struct mlx5_core_dev *mdev = priv->mdev;
210 	u32 in[MLX5_ST_SZ_DW(qp_2rst_in)] = {};
211 
212 	MLX5_SET(qp_2rst_in, in, opcode, MLX5_CMD_OP_2RST_QP);
213 	MLX5_SET(qp_2rst_in, in, qpn, ipriv->qpn);
214 	mlx5_cmd_exec_in(mdev, qp_2rst, in);
215 }
216 
217 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
218 
219 int mlx5i_create_underlay_qp(struct mlx5e_priv *priv)
220 {
221 	unsigned char *dev_addr = priv->netdev->dev_addr;
222 	u32 out[MLX5_ST_SZ_DW(create_qp_out)] = {};
223 	u32 in[MLX5_ST_SZ_DW(create_qp_in)] = {};
224 	struct mlx5i_priv *ipriv = priv->ppriv;
225 	void *addr_path;
226 	int qpn = 0;
227 	int ret = 0;
228 	void *qpc;
229 
230 	if (MLX5_CAP_GEN(priv->mdev, mkey_by_name)) {
231 		qpn = (dev_addr[1] << 16) + (dev_addr[2] << 8) + dev_addr[3];
232 		MLX5_SET(create_qp_in, in, input_qpn, qpn);
233 	}
234 
235 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
236 	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
237 	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
238 	MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
239 		 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
240 
241 	addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
242 	MLX5_SET(ads, addr_path, vhca_port_num, 1);
243 	MLX5_SET(ads, addr_path, grh, 1);
244 
245 	MLX5_SET(create_qp_in, in, opcode, MLX5_CMD_OP_CREATE_QP);
246 	ret = mlx5_cmd_exec_inout(priv->mdev, create_qp, in, out);
247 	if (ret)
248 		return ret;
249 
250 	ipriv->qpn = MLX5_GET(create_qp_out, out, qpn);
251 
252 	return 0;
253 }
254 
255 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, u32 qpn)
256 {
257 	u32 in[MLX5_ST_SZ_DW(destroy_qp_in)] = {};
258 
259 	MLX5_SET(destroy_qp_in, in, opcode, MLX5_CMD_OP_DESTROY_QP);
260 	MLX5_SET(destroy_qp_in, in, qpn, qpn);
261 	mlx5_cmd_exec_in(mdev, destroy_qp, in);
262 }
263 
264 int mlx5i_update_nic_rx(struct mlx5e_priv *priv)
265 {
266 	return mlx5e_refresh_tirs(priv, true, true);
267 }
268 
269 int mlx5i_create_tis(struct mlx5_core_dev *mdev, u32 underlay_qpn, u32 *tisn)
270 {
271 	u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
272 	void *tisc;
273 
274 	tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
275 
276 	MLX5_SET(tisc, tisc, underlay_qpn, underlay_qpn);
277 
278 	return mlx5e_create_tis(mdev, in, tisn);
279 }
280 
281 static int mlx5i_init_tx(struct mlx5e_priv *priv)
282 {
283 	struct mlx5i_priv *ipriv = priv->ppriv;
284 	int err;
285 
286 	err = mlx5i_create_underlay_qp(priv);
287 	if (err) {
288 		mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
289 		return err;
290 	}
291 
292 	err = mlx5i_create_tis(priv->mdev, ipriv->qpn, &priv->tisn[0][0]);
293 	if (err) {
294 		mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
295 		goto err_destroy_underlay_qp;
296 	}
297 
298 	return 0;
299 
300 err_destroy_underlay_qp:
301 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
302 	return err;
303 }
304 
305 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
306 {
307 	struct mlx5i_priv *ipriv = priv->ppriv;
308 
309 	mlx5e_destroy_tis(priv->mdev, priv->tisn[0][0]);
310 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
311 }
312 
313 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
314 {
315 	struct ttc_params ttc_params = {};
316 	int tt, err;
317 
318 	priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
319 					       MLX5_FLOW_NAMESPACE_KERNEL);
320 
321 	if (!priv->fs.ns)
322 		return -EINVAL;
323 
324 	err = mlx5e_arfs_create_tables(priv);
325 	if (err) {
326 		netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
327 			   err);
328 		priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
329 	}
330 
331 	mlx5e_set_ttc_basic_params(priv, &ttc_params);
332 	mlx5e_set_inner_ttc_ft_params(&ttc_params);
333 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
334 		ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
335 
336 	err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
337 	if (err) {
338 		netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
339 			   err);
340 		goto err_destroy_arfs_tables;
341 	}
342 
343 	mlx5e_set_ttc_ft_params(&ttc_params);
344 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
345 		ttc_params.indir_tirn[tt] = priv->indir_tir[tt].tirn;
346 
347 	err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
348 	if (err) {
349 		netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
350 			   err);
351 		goto err_destroy_inner_ttc_table;
352 	}
353 
354 	return 0;
355 
356 err_destroy_inner_ttc_table:
357 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
358 err_destroy_arfs_tables:
359 	mlx5e_arfs_destroy_tables(priv);
360 
361 	return err;
362 }
363 
364 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
365 {
366 	mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
367 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
368 	mlx5e_arfs_destroy_tables(priv);
369 }
370 
371 static int mlx5i_init_rx(struct mlx5e_priv *priv)
372 {
373 	struct mlx5_core_dev *mdev = priv->mdev;
374 	int err;
375 
376 	mlx5e_create_q_counters(priv);
377 
378 	err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
379 	if (err) {
380 		mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
381 		goto err_destroy_q_counters;
382 	}
383 
384 	err = mlx5e_create_indirect_rqt(priv);
385 	if (err)
386 		goto err_close_drop_rq;
387 
388 	err = mlx5e_create_direct_rqts(priv, priv->direct_tir);
389 	if (err)
390 		goto err_destroy_indirect_rqts;
391 
392 	err = mlx5e_create_indirect_tirs(priv, true);
393 	if (err)
394 		goto err_destroy_direct_rqts;
395 
396 	err = mlx5e_create_direct_tirs(priv, priv->direct_tir);
397 	if (err)
398 		goto err_destroy_indirect_tirs;
399 
400 	err = mlx5i_create_flow_steering(priv);
401 	if (err)
402 		goto err_destroy_direct_tirs;
403 
404 	return 0;
405 
406 err_destroy_direct_tirs:
407 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
408 err_destroy_indirect_tirs:
409 	mlx5e_destroy_indirect_tirs(priv);
410 err_destroy_direct_rqts:
411 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
412 err_destroy_indirect_rqts:
413 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
414 err_close_drop_rq:
415 	mlx5e_close_drop_rq(&priv->drop_rq);
416 err_destroy_q_counters:
417 	mlx5e_destroy_q_counters(priv);
418 	return err;
419 }
420 
421 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
422 {
423 	mlx5i_destroy_flow_steering(priv);
424 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
425 	mlx5e_destroy_indirect_tirs(priv);
426 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
427 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
428 	mlx5e_close_drop_rq(&priv->drop_rq);
429 	mlx5e_destroy_q_counters(priv);
430 }
431 
432 /* The stats groups order is opposite to the update_stats() order calls */
433 static mlx5e_stats_grp_t mlx5i_stats_grps[] = {
434 	&MLX5E_STATS_GRP(sw),
435 	&MLX5E_STATS_GRP(qcnt),
436 	&MLX5E_STATS_GRP(vnic_env),
437 	&MLX5E_STATS_GRP(vport),
438 	&MLX5E_STATS_GRP(802_3),
439 	&MLX5E_STATS_GRP(2863),
440 	&MLX5E_STATS_GRP(2819),
441 	&MLX5E_STATS_GRP(phy),
442 	&MLX5E_STATS_GRP(pcie),
443 	&MLX5E_STATS_GRP(per_prio),
444 	&MLX5E_STATS_GRP(pme),
445 	&MLX5E_STATS_GRP(channels),
446 	&MLX5E_STATS_GRP(per_port_buff_congest),
447 };
448 
449 static unsigned int mlx5i_stats_grps_num(struct mlx5e_priv *priv)
450 {
451 	return ARRAY_SIZE(mlx5i_stats_grps);
452 }
453 
454 static const struct mlx5e_profile mlx5i_nic_profile = {
455 	.init		   = mlx5i_init,
456 	.cleanup	   = mlx5i_cleanup,
457 	.init_tx	   = mlx5i_init_tx,
458 	.cleanup_tx	   = mlx5i_cleanup_tx,
459 	.init_rx	   = mlx5i_init_rx,
460 	.cleanup_rx	   = mlx5i_cleanup_rx,
461 	.enable		   = NULL, /* mlx5i_enable */
462 	.disable	   = NULL, /* mlx5i_disable */
463 	.update_rx	   = mlx5i_update_nic_rx,
464 	.update_stats	   = NULL, /* mlx5i_update_stats */
465 	.update_carrier    = NULL, /* no HW update in IB link */
466 	.rx_handlers       = &mlx5i_rx_handlers,
467 	.max_tc		   = MLX5I_MAX_NUM_TC,
468 	.rq_groups	   = MLX5E_NUM_RQ_GROUPS(REGULAR),
469 	.stats_grps        = mlx5i_stats_grps,
470 	.stats_grps_num    = mlx5i_stats_grps_num,
471 };
472 
473 /* mlx5i netdev NDos */
474 
475 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
476 {
477 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
478 	struct mlx5e_channels new_channels = {};
479 	struct mlx5e_params *params;
480 	int err = 0;
481 
482 	mutex_lock(&priv->state_lock);
483 
484 	params = &priv->channels.params;
485 
486 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
487 		params->sw_mtu = new_mtu;
488 		netdev->mtu = params->sw_mtu;
489 		goto out;
490 	}
491 
492 	new_channels.params = *params;
493 	new_channels.params.sw_mtu = new_mtu;
494 
495 	err = mlx5e_safe_switch_channels(priv, &new_channels, NULL, NULL);
496 	if (err)
497 		goto out;
498 
499 	netdev->mtu = new_channels.params.sw_mtu;
500 
501 out:
502 	mutex_unlock(&priv->state_lock);
503 	return err;
504 }
505 
506 int mlx5i_dev_init(struct net_device *dev)
507 {
508 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
509 	struct mlx5i_priv    *ipriv  = priv->ppriv;
510 
511 	/* Set dev address using underlay QP */
512 	dev->dev_addr[1] = (ipriv->qpn >> 16) & 0xff;
513 	dev->dev_addr[2] = (ipriv->qpn >>  8) & 0xff;
514 	dev->dev_addr[3] = (ipriv->qpn) & 0xff;
515 
516 	/* Add QPN to net-device mapping to HT */
517 	mlx5i_pkey_add_qpn(dev, ipriv->qpn);
518 
519 	return 0;
520 }
521 
522 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
523 {
524 	struct mlx5e_priv *priv = mlx5i_epriv(dev);
525 
526 	switch (cmd) {
527 	case SIOCSHWTSTAMP:
528 		return mlx5e_hwstamp_set(priv, ifr);
529 	case SIOCGHWTSTAMP:
530 		return mlx5e_hwstamp_get(priv, ifr);
531 	default:
532 		return -EOPNOTSUPP;
533 	}
534 }
535 
536 void mlx5i_dev_cleanup(struct net_device *dev)
537 {
538 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
539 	struct mlx5i_priv    *ipriv = priv->ppriv;
540 
541 	mlx5i_uninit_underlay_qp(priv);
542 
543 	/* Delete QPN to net-device mapping from HT */
544 	mlx5i_pkey_del_qpn(dev, ipriv->qpn);
545 }
546 
547 static int mlx5i_open(struct net_device *netdev)
548 {
549 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
550 	struct mlx5i_priv *ipriv = epriv->ppriv;
551 	struct mlx5_core_dev *mdev = epriv->mdev;
552 	int err;
553 
554 	mutex_lock(&epriv->state_lock);
555 
556 	set_bit(MLX5E_STATE_OPENED, &epriv->state);
557 
558 	err = mlx5i_init_underlay_qp(epriv);
559 	if (err) {
560 		mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
561 		goto err_clear_state_opened_flag;
562 	}
563 
564 	err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qpn);
565 	if (err) {
566 		mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
567 		goto err_reset_qp;
568 	}
569 
570 	err = mlx5e_open_channels(epriv, &epriv->channels);
571 	if (err)
572 		goto err_remove_fs_underlay_qp;
573 
574 	epriv->profile->update_rx(epriv);
575 	mlx5e_activate_priv_channels(epriv);
576 
577 	mutex_unlock(&epriv->state_lock);
578 	return 0;
579 
580 err_remove_fs_underlay_qp:
581 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
582 err_reset_qp:
583 	mlx5i_uninit_underlay_qp(epriv);
584 err_clear_state_opened_flag:
585 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
586 	mutex_unlock(&epriv->state_lock);
587 	return err;
588 }
589 
590 static int mlx5i_close(struct net_device *netdev)
591 {
592 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
593 	struct mlx5i_priv *ipriv = epriv->ppriv;
594 	struct mlx5_core_dev *mdev = epriv->mdev;
595 
596 	/* May already be CLOSED in case a previous configuration operation
597 	 * (e.g RX/TX queue size change) that involves close&open failed.
598 	 */
599 	mutex_lock(&epriv->state_lock);
600 
601 	if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
602 		goto unlock;
603 
604 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
605 
606 	netif_carrier_off(epriv->netdev);
607 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
608 	mlx5e_deactivate_priv_channels(epriv);
609 	mlx5e_close_channels(&epriv->channels);
610 	mlx5i_uninit_underlay_qp(epriv);
611 unlock:
612 	mutex_unlock(&epriv->state_lock);
613 	return 0;
614 }
615 
616 /* IPoIB RDMA netdev callbacks */
617 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
618 			      union ib_gid *gid, u16 lid, int set_qkey,
619 			      u32 qkey)
620 {
621 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
622 	struct mlx5_core_dev *mdev  = epriv->mdev;
623 	struct mlx5i_priv    *ipriv = epriv->ppriv;
624 	int err;
625 
626 	mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
627 		      gid->raw);
628 	err = mlx5_core_attach_mcg(mdev, gid, ipriv->qpn);
629 	if (err)
630 		mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
631 			       ipriv->qpn, gid->raw);
632 
633 	if (set_qkey) {
634 		mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
635 			      netdev->name, qkey);
636 		ipriv->qkey = qkey;
637 	}
638 
639 	return err;
640 }
641 
642 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
643 			      union ib_gid *gid, u16 lid)
644 {
645 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
646 	struct mlx5_core_dev *mdev  = epriv->mdev;
647 	struct mlx5i_priv    *ipriv = epriv->ppriv;
648 	int err;
649 
650 	mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
651 		      gid->raw);
652 
653 	err = mlx5_core_detach_mcg(mdev, gid, ipriv->qpn);
654 	if (err)
655 		mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
656 			      ipriv->qpn, gid->raw);
657 
658 	return err;
659 }
660 
661 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
662 		      struct ib_ah *address, u32 dqpn)
663 {
664 	struct mlx5e_priv *epriv = mlx5i_epriv(dev);
665 	struct mlx5e_txqsq *sq   = epriv->txq2sq[skb_get_queue_mapping(skb)];
666 	struct mlx5_ib_ah *mah   = to_mah(address);
667 	struct mlx5i_priv *ipriv = epriv->ppriv;
668 
669 	mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey, netdev_xmit_more());
670 
671 	return NETDEV_TX_OK;
672 }
673 
674 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
675 {
676 	struct mlx5i_priv *ipriv = netdev_priv(netdev);
677 
678 	ipriv->pkey_index = (u16)id;
679 }
680 
681 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
682 {
683 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
684 		return -EOPNOTSUPP;
685 
686 	if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
687 		mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
688 		return -EOPNOTSUPP;
689 	}
690 
691 	return 0;
692 }
693 
694 static void mlx5_rdma_netdev_free(struct net_device *netdev)
695 {
696 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
697 	struct mlx5i_priv *ipriv = priv->ppriv;
698 	const struct mlx5e_profile *profile = priv->profile;
699 
700 	mlx5e_detach_netdev(priv);
701 	profile->cleanup(priv);
702 
703 	if (!ipriv->sub_interface) {
704 		mlx5i_pkey_qpn_ht_cleanup(netdev);
705 		mlx5e_destroy_mdev_resources(priv->mdev);
706 	}
707 }
708 
709 static bool mlx5_is_sub_interface(struct mlx5_core_dev *mdev)
710 {
711 	return mdev->mlx5e_res.pdn != 0;
712 }
713 
714 static const struct mlx5e_profile *mlx5_get_profile(struct mlx5_core_dev *mdev)
715 {
716 	if (mlx5_is_sub_interface(mdev))
717 		return mlx5i_pkey_get_profile();
718 	return &mlx5i_nic_profile;
719 }
720 
721 static int mlx5_rdma_setup_rn(struct ib_device *ibdev, u8 port_num,
722 			      struct net_device *netdev, void *param)
723 {
724 	struct mlx5_core_dev *mdev = (struct mlx5_core_dev *)param;
725 	const struct mlx5e_profile *prof = mlx5_get_profile(mdev);
726 	struct mlx5i_priv *ipriv;
727 	struct mlx5e_priv *epriv;
728 	struct rdma_netdev *rn;
729 	int err;
730 
731 	ipriv = netdev_priv(netdev);
732 	epriv = mlx5i_epriv(netdev);
733 
734 	ipriv->sub_interface = mlx5_is_sub_interface(mdev);
735 	if (!ipriv->sub_interface) {
736 		err = mlx5i_pkey_qpn_ht_init(netdev);
737 		if (err) {
738 			mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
739 			return err;
740 		}
741 
742 		/* This should only be called once per mdev */
743 		err = mlx5e_create_mdev_resources(mdev);
744 		if (err)
745 			goto destroy_ht;
746 	}
747 
748 	err = mlx5e_priv_init(epriv, netdev, mdev);
749 	if (err)
750 		goto destroy_mdev_resources;
751 
752 	epriv->profile = prof;
753 	epriv->ppriv = ipriv;
754 
755 	prof->init(mdev, netdev);
756 
757 	err = mlx5e_attach_netdev(epriv);
758 	if (err)
759 		goto detach;
760 	netif_carrier_off(netdev);
761 
762 	/* set rdma_netdev func pointers */
763 	rn = &ipriv->rn;
764 	rn->hca  = ibdev;
765 	rn->send = mlx5i_xmit;
766 	rn->attach_mcast = mlx5i_attach_mcast;
767 	rn->detach_mcast = mlx5i_detach_mcast;
768 	rn->set_id = mlx5i_set_pkey_index;
769 
770 	netdev->priv_destructor = mlx5_rdma_netdev_free;
771 	netdev->needs_free_netdev = 1;
772 
773 	return 0;
774 
775 detach:
776 	prof->cleanup(epriv);
777 	if (ipriv->sub_interface)
778 		return err;
779 destroy_mdev_resources:
780 	mlx5e_destroy_mdev_resources(mdev);
781 destroy_ht:
782 	mlx5i_pkey_qpn_ht_cleanup(netdev);
783 	return err;
784 }
785 
786 int mlx5_rdma_rn_get_params(struct mlx5_core_dev *mdev,
787 			    struct ib_device *device,
788 			    struct rdma_netdev_alloc_params *params)
789 {
790 	int nch;
791 	int rc;
792 
793 	rc = mlx5i_check_required_hca_cap(mdev);
794 	if (rc)
795 		return rc;
796 
797 	nch = mlx5e_get_max_num_channels(mdev);
798 
799 	*params = (struct rdma_netdev_alloc_params){
800 		.sizeof_priv = sizeof(struct mlx5i_priv) +
801 			       sizeof(struct mlx5e_priv),
802 		.txqs = nch * MLX5E_MAX_NUM_TC,
803 		.rxqs = nch,
804 		.param = mdev,
805 		.initialize_rdma_netdev = mlx5_rdma_setup_rn,
806 	};
807 
808 	return 0;
809 }
810 EXPORT_SYMBOL(mlx5_rdma_rn_get_params);
811