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