xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c (revision e5a52fd2b8cdb700b3c07b030e050a49ef3156b9)
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 	u32 out[MLX5_ST_SZ_DW(create_qp_out)] = {};
230 	u32 in[MLX5_ST_SZ_DW(create_qp_in)] = {};
231 	struct mlx5i_priv *ipriv = priv->ppriv;
232 	void *addr_path;
233 	int ret = 0;
234 	void *qpc;
235 
236 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
237 	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
238 	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
239 	MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
240 		 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
241 
242 	addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
243 	MLX5_SET(ads, addr_path, vhca_port_num, 1);
244 	MLX5_SET(ads, addr_path, grh, 1);
245 
246 	MLX5_SET(create_qp_in, in, opcode, MLX5_CMD_OP_CREATE_QP);
247 	ret = mlx5_cmd_exec_inout(priv->mdev, create_qp, in, out);
248 	if (ret)
249 		return ret;
250 
251 	ipriv->qpn = MLX5_GET(create_qp_out, out, qpn);
252 
253 	return 0;
254 }
255 
256 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, u32 qpn)
257 {
258 	u32 in[MLX5_ST_SZ_DW(destroy_qp_in)] = {};
259 
260 	MLX5_SET(destroy_qp_in, in, opcode, MLX5_CMD_OP_DESTROY_QP);
261 	MLX5_SET(destroy_qp_in, in, qpn, qpn);
262 	mlx5_cmd_exec_in(mdev, destroy_qp, in);
263 }
264 
265 int mlx5i_update_nic_rx(struct mlx5e_priv *priv)
266 {
267 	return mlx5e_refresh_tirs(priv, true, true);
268 }
269 
270 int mlx5i_create_tis(struct mlx5_core_dev *mdev, u32 underlay_qpn, u32 *tisn)
271 {
272 	u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
273 	void *tisc;
274 
275 	tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
276 
277 	MLX5_SET(tisc, tisc, underlay_qpn, underlay_qpn);
278 
279 	return mlx5e_create_tis(mdev, in, tisn);
280 }
281 
282 static int mlx5i_init_tx(struct mlx5e_priv *priv)
283 {
284 	struct mlx5i_priv *ipriv = priv->ppriv;
285 	int err;
286 
287 	err = mlx5i_create_underlay_qp(priv);
288 	if (err) {
289 		mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
290 		return err;
291 	}
292 
293 	err = mlx5i_create_tis(priv->mdev, ipriv->qpn, &priv->tisn[0][0]);
294 	if (err) {
295 		mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
296 		goto err_destroy_underlay_qp;
297 	}
298 
299 	return 0;
300 
301 err_destroy_underlay_qp:
302 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
303 	return err;
304 }
305 
306 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
307 {
308 	struct mlx5i_priv *ipriv = priv->ppriv;
309 
310 	mlx5e_destroy_tis(priv->mdev, priv->tisn[0][0]);
311 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
312 }
313 
314 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
315 {
316 	struct ttc_params ttc_params = {};
317 	int tt, err;
318 
319 	priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
320 					       MLX5_FLOW_NAMESPACE_KERNEL);
321 
322 	if (!priv->fs.ns)
323 		return -EINVAL;
324 
325 	err = mlx5e_arfs_create_tables(priv);
326 	if (err) {
327 		netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
328 			   err);
329 		priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
330 	}
331 
332 	mlx5e_set_ttc_basic_params(priv, &ttc_params);
333 	mlx5e_set_inner_ttc_ft_params(&ttc_params);
334 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
335 		ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
336 
337 	err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
338 	if (err) {
339 		netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
340 			   err);
341 		goto err_destroy_arfs_tables;
342 	}
343 
344 	mlx5e_set_ttc_ft_params(&ttc_params);
345 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
346 		ttc_params.indir_tirn[tt] = priv->indir_tir[tt].tirn;
347 
348 	err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
349 	if (err) {
350 		netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
351 			   err);
352 		goto err_destroy_inner_ttc_table;
353 	}
354 
355 	return 0;
356 
357 err_destroy_inner_ttc_table:
358 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
359 err_destroy_arfs_tables:
360 	mlx5e_arfs_destroy_tables(priv);
361 
362 	return err;
363 }
364 
365 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
366 {
367 	mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
368 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
369 	mlx5e_arfs_destroy_tables(priv);
370 }
371 
372 static int mlx5i_init_rx(struct mlx5e_priv *priv)
373 {
374 	struct mlx5_core_dev *mdev = priv->mdev;
375 	int err;
376 
377 	mlx5e_create_q_counters(priv);
378 
379 	err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
380 	if (err) {
381 		mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
382 		goto err_destroy_q_counters;
383 	}
384 
385 	err = mlx5e_create_indirect_rqt(priv);
386 	if (err)
387 		goto err_close_drop_rq;
388 
389 	err = mlx5e_create_direct_rqts(priv, priv->direct_tir);
390 	if (err)
391 		goto err_destroy_indirect_rqts;
392 
393 	err = mlx5e_create_indirect_tirs(priv, true);
394 	if (err)
395 		goto err_destroy_direct_rqts;
396 
397 	err = mlx5e_create_direct_tirs(priv, priv->direct_tir);
398 	if (err)
399 		goto err_destroy_indirect_tirs;
400 
401 	err = mlx5i_create_flow_steering(priv);
402 	if (err)
403 		goto err_destroy_direct_tirs;
404 
405 	return 0;
406 
407 err_destroy_direct_tirs:
408 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
409 err_destroy_indirect_tirs:
410 	mlx5e_destroy_indirect_tirs(priv);
411 err_destroy_direct_rqts:
412 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
413 err_destroy_indirect_rqts:
414 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
415 err_close_drop_rq:
416 	mlx5e_close_drop_rq(&priv->drop_rq);
417 err_destroy_q_counters:
418 	mlx5e_destroy_q_counters(priv);
419 	return err;
420 }
421 
422 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
423 {
424 	mlx5i_destroy_flow_steering(priv);
425 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
426 	mlx5e_destroy_indirect_tirs(priv);
427 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir);
428 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
429 	mlx5e_close_drop_rq(&priv->drop_rq);
430 	mlx5e_destroy_q_counters(priv);
431 }
432 
433 /* The stats groups order is opposite to the update_stats() order calls */
434 static mlx5e_stats_grp_t mlx5i_stats_grps[] = {
435 	&MLX5E_STATS_GRP(sw),
436 	&MLX5E_STATS_GRP(qcnt),
437 	&MLX5E_STATS_GRP(vnic_env),
438 	&MLX5E_STATS_GRP(vport),
439 	&MLX5E_STATS_GRP(802_3),
440 	&MLX5E_STATS_GRP(2863),
441 	&MLX5E_STATS_GRP(2819),
442 	&MLX5E_STATS_GRP(phy),
443 	&MLX5E_STATS_GRP(pcie),
444 	&MLX5E_STATS_GRP(per_prio),
445 	&MLX5E_STATS_GRP(pme),
446 	&MLX5E_STATS_GRP(channels),
447 	&MLX5E_STATS_GRP(per_port_buff_congest),
448 };
449 
450 static unsigned int mlx5i_stats_grps_num(struct mlx5e_priv *priv)
451 {
452 	return ARRAY_SIZE(mlx5i_stats_grps);
453 }
454 
455 static const struct mlx5e_profile mlx5i_nic_profile = {
456 	.init		   = mlx5i_init,
457 	.cleanup	   = mlx5i_cleanup,
458 	.init_tx	   = mlx5i_init_tx,
459 	.cleanup_tx	   = mlx5i_cleanup_tx,
460 	.init_rx	   = mlx5i_init_rx,
461 	.cleanup_rx	   = mlx5i_cleanup_rx,
462 	.enable		   = NULL, /* mlx5i_enable */
463 	.disable	   = NULL, /* mlx5i_disable */
464 	.update_rx	   = mlx5i_update_nic_rx,
465 	.update_stats	   = NULL, /* mlx5i_update_stats */
466 	.update_carrier    = NULL, /* no HW update in IB link */
467 	.rx_handlers.handle_rx_cqe       = mlx5i_handle_rx_cqe,
468 	.rx_handlers.handle_rx_cqe_mpwqe = NULL, /* Not supported */
469 	.max_tc		   = MLX5I_MAX_NUM_TC,
470 	.rq_groups	   = MLX5E_NUM_RQ_GROUPS(REGULAR),
471 	.stats_grps        = mlx5i_stats_grps,
472 	.stats_grps_num    = mlx5i_stats_grps_num,
473 };
474 
475 /* mlx5i netdev NDos */
476 
477 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
478 {
479 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
480 	struct mlx5e_channels new_channels = {};
481 	struct mlx5e_params *params;
482 	int err = 0;
483 
484 	mutex_lock(&priv->state_lock);
485 
486 	params = &priv->channels.params;
487 
488 	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
489 		params->sw_mtu = new_mtu;
490 		netdev->mtu = params->sw_mtu;
491 		goto out;
492 	}
493 
494 	new_channels.params = *params;
495 	new_channels.params.sw_mtu = new_mtu;
496 
497 	err = mlx5e_safe_switch_channels(priv, &new_channels, NULL, NULL);
498 	if (err)
499 		goto out;
500 
501 	netdev->mtu = new_channels.params.sw_mtu;
502 
503 out:
504 	mutex_unlock(&priv->state_lock);
505 	return err;
506 }
507 
508 int mlx5i_dev_init(struct net_device *dev)
509 {
510 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
511 	struct mlx5i_priv    *ipriv  = priv->ppriv;
512 
513 	/* Set dev address using underlay QP */
514 	dev->dev_addr[1] = (ipriv->qpn >> 16) & 0xff;
515 	dev->dev_addr[2] = (ipriv->qpn >>  8) & 0xff;
516 	dev->dev_addr[3] = (ipriv->qpn) & 0xff;
517 
518 	/* Add QPN to net-device mapping to HT */
519 	mlx5i_pkey_add_qpn(dev, ipriv->qpn);
520 
521 	return 0;
522 }
523 
524 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
525 {
526 	struct mlx5e_priv *priv = mlx5i_epriv(dev);
527 
528 	switch (cmd) {
529 	case SIOCSHWTSTAMP:
530 		return mlx5e_hwstamp_set(priv, ifr);
531 	case SIOCGHWTSTAMP:
532 		return mlx5e_hwstamp_get(priv, ifr);
533 	default:
534 		return -EOPNOTSUPP;
535 	}
536 }
537 
538 void mlx5i_dev_cleanup(struct net_device *dev)
539 {
540 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
541 	struct mlx5i_priv    *ipriv = priv->ppriv;
542 
543 	mlx5i_uninit_underlay_qp(priv);
544 
545 	/* Delete QPN to net-device mapping from HT */
546 	mlx5i_pkey_del_qpn(dev, ipriv->qpn);
547 }
548 
549 static int mlx5i_open(struct net_device *netdev)
550 {
551 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
552 	struct mlx5i_priv *ipriv = epriv->ppriv;
553 	struct mlx5_core_dev *mdev = epriv->mdev;
554 	int err;
555 
556 	mutex_lock(&epriv->state_lock);
557 
558 	set_bit(MLX5E_STATE_OPENED, &epriv->state);
559 
560 	err = mlx5i_init_underlay_qp(epriv);
561 	if (err) {
562 		mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
563 		goto err_clear_state_opened_flag;
564 	}
565 
566 	err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qpn);
567 	if (err) {
568 		mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
569 		goto err_reset_qp;
570 	}
571 
572 	err = mlx5e_open_channels(epriv, &epriv->channels);
573 	if (err)
574 		goto err_remove_fs_underlay_qp;
575 
576 	epriv->profile->update_rx(epriv);
577 	mlx5e_activate_priv_channels(epriv);
578 
579 	mutex_unlock(&epriv->state_lock);
580 	return 0;
581 
582 err_remove_fs_underlay_qp:
583 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
584 err_reset_qp:
585 	mlx5i_uninit_underlay_qp(epriv);
586 err_clear_state_opened_flag:
587 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
588 	mutex_unlock(&epriv->state_lock);
589 	return err;
590 }
591 
592 static int mlx5i_close(struct net_device *netdev)
593 {
594 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
595 	struct mlx5i_priv *ipriv = epriv->ppriv;
596 	struct mlx5_core_dev *mdev = epriv->mdev;
597 
598 	/* May already be CLOSED in case a previous configuration operation
599 	 * (e.g RX/TX queue size change) that involves close&open failed.
600 	 */
601 	mutex_lock(&epriv->state_lock);
602 
603 	if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
604 		goto unlock;
605 
606 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
607 
608 	netif_carrier_off(epriv->netdev);
609 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
610 	mlx5e_deactivate_priv_channels(epriv);
611 	mlx5e_close_channels(&epriv->channels);
612 	mlx5i_uninit_underlay_qp(epriv);
613 unlock:
614 	mutex_unlock(&epriv->state_lock);
615 	return 0;
616 }
617 
618 /* IPoIB RDMA netdev callbacks */
619 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
620 			      union ib_gid *gid, u16 lid, int set_qkey,
621 			      u32 qkey)
622 {
623 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
624 	struct mlx5_core_dev *mdev  = epriv->mdev;
625 	struct mlx5i_priv    *ipriv = epriv->ppriv;
626 	int err;
627 
628 	mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
629 		      gid->raw);
630 	err = mlx5_core_attach_mcg(mdev, gid, ipriv->qpn);
631 	if (err)
632 		mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
633 			       ipriv->qpn, gid->raw);
634 
635 	if (set_qkey) {
636 		mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
637 			      netdev->name, qkey);
638 		ipriv->qkey = qkey;
639 	}
640 
641 	return err;
642 }
643 
644 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
645 			      union ib_gid *gid, u16 lid)
646 {
647 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
648 	struct mlx5_core_dev *mdev  = epriv->mdev;
649 	struct mlx5i_priv    *ipriv = epriv->ppriv;
650 	int err;
651 
652 	mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
653 		      gid->raw);
654 
655 	err = mlx5_core_detach_mcg(mdev, gid, ipriv->qpn);
656 	if (err)
657 		mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
658 			      ipriv->qpn, gid->raw);
659 
660 	return err;
661 }
662 
663 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
664 		      struct ib_ah *address, u32 dqpn)
665 {
666 	struct mlx5e_priv *epriv = mlx5i_epriv(dev);
667 	struct mlx5e_txqsq *sq   = epriv->txq2sq[skb_get_queue_mapping(skb)];
668 	struct mlx5_ib_ah *mah   = to_mah(address);
669 	struct mlx5i_priv *ipriv = epriv->ppriv;
670 
671 	mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey, netdev_xmit_more());
672 
673 	return NETDEV_TX_OK;
674 }
675 
676 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
677 {
678 	struct mlx5i_priv *ipriv = netdev_priv(netdev);
679 
680 	ipriv->pkey_index = (u16)id;
681 }
682 
683 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
684 {
685 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
686 		return -EOPNOTSUPP;
687 
688 	if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
689 		mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
690 		return -EOPNOTSUPP;
691 	}
692 
693 	return 0;
694 }
695 
696 static void mlx5_rdma_netdev_free(struct net_device *netdev)
697 {
698 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
699 	struct mlx5i_priv *ipriv = priv->ppriv;
700 	const struct mlx5e_profile *profile = priv->profile;
701 
702 	mlx5e_detach_netdev(priv);
703 	profile->cleanup(priv);
704 
705 	if (!ipriv->sub_interface) {
706 		mlx5i_pkey_qpn_ht_cleanup(netdev);
707 		mlx5e_destroy_mdev_resources(priv->mdev);
708 	}
709 }
710 
711 static bool mlx5_is_sub_interface(struct mlx5_core_dev *mdev)
712 {
713 	return mdev->mlx5e_res.pdn != 0;
714 }
715 
716 static const struct mlx5e_profile *mlx5_get_profile(struct mlx5_core_dev *mdev)
717 {
718 	if (mlx5_is_sub_interface(mdev))
719 		return mlx5i_pkey_get_profile();
720 	return &mlx5i_nic_profile;
721 }
722 
723 static int mlx5_rdma_setup_rn(struct ib_device *ibdev, u8 port_num,
724 			      struct net_device *netdev, void *param)
725 {
726 	struct mlx5_core_dev *mdev = (struct mlx5_core_dev *)param;
727 	const struct mlx5e_profile *prof = mlx5_get_profile(mdev);
728 	struct mlx5i_priv *ipriv;
729 	struct mlx5e_priv *epriv;
730 	struct rdma_netdev *rn;
731 	int err;
732 
733 	ipriv = netdev_priv(netdev);
734 	epriv = mlx5i_epriv(netdev);
735 
736 	ipriv->sub_interface = mlx5_is_sub_interface(mdev);
737 	if (!ipriv->sub_interface) {
738 		err = mlx5i_pkey_qpn_ht_init(netdev);
739 		if (err) {
740 			mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
741 			return err;
742 		}
743 
744 		/* This should only be called once per mdev */
745 		err = mlx5e_create_mdev_resources(mdev);
746 		if (err)
747 			goto destroy_ht;
748 	}
749 
750 	prof->init(mdev, netdev, prof, ipriv);
751 
752 	err = mlx5e_attach_netdev(epriv);
753 	if (err)
754 		goto detach;
755 	netif_carrier_off(netdev);
756 
757 	/* set rdma_netdev func pointers */
758 	rn = &ipriv->rn;
759 	rn->hca  = ibdev;
760 	rn->send = mlx5i_xmit;
761 	rn->attach_mcast = mlx5i_attach_mcast;
762 	rn->detach_mcast = mlx5i_detach_mcast;
763 	rn->set_id = mlx5i_set_pkey_index;
764 
765 	netdev->priv_destructor = mlx5_rdma_netdev_free;
766 	netdev->needs_free_netdev = 1;
767 
768 	return 0;
769 
770 detach:
771 	prof->cleanup(epriv);
772 	if (ipriv->sub_interface)
773 		return err;
774 	mlx5e_destroy_mdev_resources(mdev);
775 destroy_ht:
776 	mlx5i_pkey_qpn_ht_cleanup(netdev);
777 	return err;
778 }
779 
780 int mlx5_rdma_rn_get_params(struct mlx5_core_dev *mdev,
781 			    struct ib_device *device,
782 			    struct rdma_netdev_alloc_params *params)
783 {
784 	int nch;
785 	int rc;
786 
787 	rc = mlx5i_check_required_hca_cap(mdev);
788 	if (rc)
789 		return rc;
790 
791 	nch = mlx5e_get_max_num_channels(mdev);
792 
793 	*params = (struct rdma_netdev_alloc_params){
794 		.sizeof_priv = sizeof(struct mlx5i_priv) +
795 			       sizeof(struct mlx5e_priv),
796 		.txqs = nch * MLX5E_MAX_NUM_TC,
797 		.rxqs = nch,
798 		.param = mdev,
799 		.initialize_rdma_netdev = mlx5_rdma_setup_rn,
800 	};
801 
802 	return 0;
803 }
804 EXPORT_SYMBOL(mlx5_rdma_rn_get_params);
805