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 "en/params.h"
37 #include "ipoib.h"
38 
39 #define IB_DEFAULT_Q_KEY   0xb1b
40 #define MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE 9
41 
42 static int mlx5i_open(struct net_device *netdev);
43 static int mlx5i_close(struct net_device *netdev);
44 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu);
45 
46 static const struct net_device_ops mlx5i_netdev_ops = {
47 	.ndo_open                = mlx5i_open,
48 	.ndo_stop                = mlx5i_close,
49 	.ndo_get_stats64         = mlx5i_get_stats,
50 	.ndo_init                = mlx5i_dev_init,
51 	.ndo_uninit              = mlx5i_dev_cleanup,
52 	.ndo_change_mtu          = mlx5i_change_mtu,
53 	.ndo_do_ioctl            = mlx5i_ioctl,
54 };
55 
56 /* IPoIB mlx5 netdev profile */
57 static void mlx5i_build_nic_params(struct mlx5_core_dev *mdev,
58 				   struct mlx5e_params *params)
59 {
60 	/* Override RQ params as IPoIB supports only LINKED LIST RQ for now */
61 	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ, false);
62 	mlx5e_set_rq_type(mdev, params);
63 	mlx5e_init_rq_type_params(mdev, params);
64 
65 	/* RQ size in ipoib by default is 512 */
66 	params->log_rq_mtu_frames = is_kdump_kernel() ?
67 		MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
68 		MLX5I_PARAMS_DEFAULT_LOG_RQ_SIZE;
69 
70 	params->lro_en = false;
71 	params->hard_mtu = MLX5_IB_GRH_BYTES + MLX5_IPOIB_HARD_LEN;
72 	params->tunneled_offload_en = false;
73 }
74 
75 /* Called directly after IPoIB netdevice was created to initialize SW structs */
76 int mlx5i_init(struct mlx5_core_dev *mdev, struct net_device *netdev)
77 {
78 	struct mlx5e_priv *priv  = mlx5i_epriv(netdev);
79 
80 	netif_carrier_off(netdev);
81 	mlx5e_set_netdev_mtu_boundaries(priv);
82 	netdev->mtu = netdev->max_mtu;
83 
84 	mlx5e_build_nic_params(priv, NULL, netdev->mtu);
85 	mlx5i_build_nic_params(mdev, &priv->channels.params);
86 
87 	mlx5e_timestamp_init(priv);
88 
89 	/* netdev init */
90 	netdev->hw_features    |= NETIF_F_SG;
91 	netdev->hw_features    |= NETIF_F_IP_CSUM;
92 	netdev->hw_features    |= NETIF_F_IPV6_CSUM;
93 	netdev->hw_features    |= NETIF_F_GRO;
94 	netdev->hw_features    |= NETIF_F_TSO;
95 	netdev->hw_features    |= NETIF_F_TSO6;
96 	netdev->hw_features    |= NETIF_F_RXCSUM;
97 	netdev->hw_features    |= NETIF_F_RXHASH;
98 
99 	netdev->netdev_ops = &mlx5i_netdev_ops;
100 	netdev->ethtool_ops = &mlx5i_ethtool_ops;
101 
102 	return 0;
103 }
104 
105 /* Called directly before IPoIB netdevice is destroyed to cleanup SW structs */
106 void mlx5i_cleanup(struct mlx5e_priv *priv)
107 {
108 	mlx5e_priv_cleanup(priv);
109 }
110 
111 static void mlx5i_grp_sw_update_stats(struct mlx5e_priv *priv)
112 {
113 	struct mlx5e_sw_stats s = { 0 };
114 	int i, j;
115 
116 	for (i = 0; i < priv->max_nch; i++) {
117 		struct mlx5e_channel_stats *channel_stats;
118 		struct mlx5e_rq_stats *rq_stats;
119 
120 		channel_stats = &priv->channel_stats[i];
121 		rq_stats = &channel_stats->rq;
122 
123 		s.rx_packets += rq_stats->packets;
124 		s.rx_bytes   += rq_stats->bytes;
125 
126 		for (j = 0; j < priv->max_opened_tc; j++) {
127 			struct mlx5e_sq_stats *sq_stats = &channel_stats->sq[j];
128 
129 			s.tx_packets           += sq_stats->packets;
130 			s.tx_bytes             += sq_stats->bytes;
131 			s.tx_queue_dropped     += sq_stats->dropped;
132 		}
133 	}
134 
135 	memcpy(&priv->stats.sw, &s, sizeof(s));
136 }
137 
138 void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
139 {
140 	struct mlx5e_priv     *priv   = mlx5i_epriv(dev);
141 	struct mlx5e_sw_stats *sstats = &priv->stats.sw;
142 
143 	mlx5i_grp_sw_update_stats(priv);
144 
145 	stats->rx_packets = sstats->rx_packets;
146 	stats->rx_bytes   = sstats->rx_bytes;
147 	stats->tx_packets = sstats->tx_packets;
148 	stats->tx_bytes   = sstats->tx_bytes;
149 	stats->tx_dropped = sstats->tx_queue_dropped;
150 }
151 
152 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
153 {
154 	struct mlx5_core_dev *mdev = priv->mdev;
155 	struct mlx5i_priv *ipriv = priv->ppriv;
156 	int ret;
157 
158 	{
159 		u32 in[MLX5_ST_SZ_DW(rst2init_qp_in)] = {};
160 		u32 *qpc;
161 
162 		qpc = MLX5_ADDR_OF(rst2init_qp_in, in, qpc);
163 
164 		MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
165 		MLX5_SET(qpc, qpc, primary_address_path.pkey_index,
166 			 ipriv->pkey_index);
167 		MLX5_SET(qpc, qpc, primary_address_path.vhca_port_num, 1);
168 		MLX5_SET(qpc, qpc, q_key, IB_DEFAULT_Q_KEY);
169 
170 		MLX5_SET(rst2init_qp_in, in, opcode, MLX5_CMD_OP_RST2INIT_QP);
171 		MLX5_SET(rst2init_qp_in, in, qpn, ipriv->qpn);
172 		ret = mlx5_cmd_exec_in(mdev, rst2init_qp, in);
173 		if (ret)
174 			goto err_qp_modify_to_err;
175 	}
176 	{
177 		u32 in[MLX5_ST_SZ_DW(init2rtr_qp_in)] = {};
178 
179 		MLX5_SET(init2rtr_qp_in, in, opcode, MLX5_CMD_OP_INIT2RTR_QP);
180 		MLX5_SET(init2rtr_qp_in, in, qpn, ipriv->qpn);
181 		ret = mlx5_cmd_exec_in(mdev, init2rtr_qp, in);
182 		if (ret)
183 			goto err_qp_modify_to_err;
184 	}
185 	{
186 		u32 in[MLX5_ST_SZ_DW(rtr2rts_qp_in)] = {};
187 
188 		MLX5_SET(rtr2rts_qp_in, in, opcode, MLX5_CMD_OP_RTR2RTS_QP);
189 		MLX5_SET(rtr2rts_qp_in, in, qpn, ipriv->qpn);
190 		ret = mlx5_cmd_exec_in(mdev, rtr2rts_qp, in);
191 		if (ret)
192 			goto err_qp_modify_to_err;
193 	}
194 	return 0;
195 
196 err_qp_modify_to_err:
197 	{
198 		u32 in[MLX5_ST_SZ_DW(qp_2err_in)] = {};
199 
200 		MLX5_SET(qp_2err_in, in, opcode, MLX5_CMD_OP_2ERR_QP);
201 		MLX5_SET(qp_2err_in, in, qpn, ipriv->qpn);
202 		mlx5_cmd_exec_in(mdev, qp_2err, in);
203 	}
204 	return ret;
205 }
206 
207 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
208 {
209 	struct mlx5i_priv *ipriv = priv->ppriv;
210 	struct mlx5_core_dev *mdev = priv->mdev;
211 	u32 in[MLX5_ST_SZ_DW(qp_2rst_in)] = {};
212 
213 	MLX5_SET(qp_2rst_in, in, opcode, MLX5_CMD_OP_2RST_QP);
214 	MLX5_SET(qp_2rst_in, in, qpn, ipriv->qpn);
215 	mlx5_cmd_exec_in(mdev, qp_2rst, in);
216 }
217 
218 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
219 
220 int mlx5i_create_underlay_qp(struct mlx5e_priv *priv)
221 {
222 	unsigned char *dev_addr = priv->netdev->dev_addr;
223 	u32 out[MLX5_ST_SZ_DW(create_qp_out)] = {};
224 	u32 in[MLX5_ST_SZ_DW(create_qp_in)] = {};
225 	struct mlx5i_priv *ipriv = priv->ppriv;
226 	void *addr_path;
227 	int qpn = 0;
228 	int ret = 0;
229 	void *qpc;
230 
231 	if (MLX5_CAP_GEN(priv->mdev, mkey_by_name)) {
232 		qpn = (dev_addr[1] << 16) + (dev_addr[2] << 8) + dev_addr[3];
233 		MLX5_SET(create_qp_in, in, input_qpn, qpn);
234 	}
235 
236 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
237 	MLX5_SET(qpc, qpc, ts_format, mlx5_get_qp_default_ts(priv->mdev));
238 	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
239 	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
240 	MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
241 		 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
242 
243 	addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
244 	MLX5_SET(ads, addr_path, vhca_port_num, 1);
245 	MLX5_SET(ads, addr_path, grh, 1);
246 
247 	MLX5_SET(create_qp_in, in, opcode, MLX5_CMD_OP_CREATE_QP);
248 	ret = mlx5_cmd_exec_inout(priv->mdev, create_qp, in, out);
249 	if (ret)
250 		return ret;
251 
252 	ipriv->qpn = MLX5_GET(create_qp_out, out, qpn);
253 
254 	return 0;
255 }
256 
257 void mlx5i_destroy_underlay_qp(struct mlx5_core_dev *mdev, u32 qpn)
258 {
259 	u32 in[MLX5_ST_SZ_DW(destroy_qp_in)] = {};
260 
261 	MLX5_SET(destroy_qp_in, in, opcode, MLX5_CMD_OP_DESTROY_QP);
262 	MLX5_SET(destroy_qp_in, in, qpn, qpn);
263 	mlx5_cmd_exec_in(mdev, destroy_qp, in);
264 }
265 
266 int mlx5i_update_nic_rx(struct mlx5e_priv *priv)
267 {
268 	return mlx5e_refresh_tirs(priv, true, true);
269 }
270 
271 int mlx5i_create_tis(struct mlx5_core_dev *mdev, u32 underlay_qpn, u32 *tisn)
272 {
273 	u32 in[MLX5_ST_SZ_DW(create_tis_in)] = {};
274 	void *tisc;
275 
276 	tisc = MLX5_ADDR_OF(create_tis_in, in, ctx);
277 
278 	MLX5_SET(tisc, tisc, underlay_qpn, underlay_qpn);
279 
280 	return mlx5e_create_tis(mdev, in, tisn);
281 }
282 
283 static int mlx5i_init_tx(struct mlx5e_priv *priv)
284 {
285 	struct mlx5i_priv *ipriv = priv->ppriv;
286 	int err;
287 
288 	err = mlx5i_create_underlay_qp(priv);
289 	if (err) {
290 		mlx5_core_warn(priv->mdev, "create underlay QP failed, %d\n", err);
291 		return err;
292 	}
293 
294 	err = mlx5i_create_tis(priv->mdev, ipriv->qpn, &priv->tisn[0][0]);
295 	if (err) {
296 		mlx5_core_warn(priv->mdev, "create tis failed, %d\n", err);
297 		goto err_destroy_underlay_qp;
298 	}
299 
300 	return 0;
301 
302 err_destroy_underlay_qp:
303 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
304 	return err;
305 }
306 
307 static void mlx5i_cleanup_tx(struct mlx5e_priv *priv)
308 {
309 	struct mlx5i_priv *ipriv = priv->ppriv;
310 
311 	mlx5e_destroy_tis(priv->mdev, priv->tisn[0][0]);
312 	mlx5i_destroy_underlay_qp(priv->mdev, ipriv->qpn);
313 }
314 
315 static int mlx5i_create_flow_steering(struct mlx5e_priv *priv)
316 {
317 	struct ttc_params ttc_params = {};
318 	int tt, err;
319 
320 	priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
321 					       MLX5_FLOW_NAMESPACE_KERNEL);
322 
323 	if (!priv->fs.ns)
324 		return -EINVAL;
325 
326 	err = mlx5e_arfs_create_tables(priv);
327 	if (err) {
328 		netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
329 			   err);
330 		priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
331 	}
332 
333 	mlx5e_set_ttc_basic_params(priv, &ttc_params);
334 	mlx5e_set_inner_ttc_ft_params(&ttc_params);
335 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
336 		ttc_params.indir_tirn[tt] = priv->inner_indir_tir[tt].tirn;
337 
338 	err = mlx5e_create_inner_ttc_table(priv, &ttc_params, &priv->fs.inner_ttc);
339 	if (err) {
340 		netdev_err(priv->netdev, "Failed to create inner ttc table, err=%d\n",
341 			   err);
342 		goto err_destroy_arfs_tables;
343 	}
344 
345 	mlx5e_set_ttc_ft_params(&ttc_params);
346 	for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++)
347 		ttc_params.indir_tirn[tt] = priv->indir_tir[tt].tirn;
348 
349 	err = mlx5e_create_ttc_table(priv, &ttc_params, &priv->fs.ttc);
350 	if (err) {
351 		netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
352 			   err);
353 		goto err_destroy_inner_ttc_table;
354 	}
355 
356 	return 0;
357 
358 err_destroy_inner_ttc_table:
359 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
360 err_destroy_arfs_tables:
361 	mlx5e_arfs_destroy_tables(priv);
362 
363 	return err;
364 }
365 
366 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
367 {
368 	mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
369 	mlx5e_destroy_inner_ttc_table(priv, &priv->fs.inner_ttc);
370 	mlx5e_arfs_destroy_tables(priv);
371 }
372 
373 static int mlx5i_init_rx(struct mlx5e_priv *priv)
374 {
375 	struct mlx5_core_dev *mdev = priv->mdev;
376 	u16 max_nch = priv->max_nch;
377 	int err;
378 
379 	mlx5e_create_q_counters(priv);
380 
381 	err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
382 	if (err) {
383 		mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
384 		goto err_destroy_q_counters;
385 	}
386 
387 	err = mlx5e_create_indirect_rqt(priv);
388 	if (err)
389 		goto err_close_drop_rq;
390 
391 	err = mlx5e_create_direct_rqts(priv, priv->direct_tir, max_nch);
392 	if (err)
393 		goto err_destroy_indirect_rqts;
394 
395 	err = mlx5e_create_indirect_tirs(priv, true);
396 	if (err)
397 		goto err_destroy_direct_rqts;
398 
399 	err = mlx5e_create_direct_tirs(priv, priv->direct_tir, max_nch);
400 	if (err)
401 		goto err_destroy_indirect_tirs;
402 
403 	err = mlx5i_create_flow_steering(priv);
404 	if (err)
405 		goto err_destroy_direct_tirs;
406 
407 	return 0;
408 
409 err_destroy_direct_tirs:
410 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir, max_nch);
411 err_destroy_indirect_tirs:
412 	mlx5e_destroy_indirect_tirs(priv);
413 err_destroy_direct_rqts:
414 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir, max_nch);
415 err_destroy_indirect_rqts:
416 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
417 err_close_drop_rq:
418 	mlx5e_close_drop_rq(&priv->drop_rq);
419 err_destroy_q_counters:
420 	mlx5e_destroy_q_counters(priv);
421 	return err;
422 }
423 
424 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
425 {
426 	u16 max_nch = priv->max_nch;
427 
428 	mlx5i_destroy_flow_steering(priv);
429 	mlx5e_destroy_direct_tirs(priv, priv->direct_tir, max_nch);
430 	mlx5e_destroy_indirect_tirs(priv);
431 	mlx5e_destroy_direct_rqts(priv, priv->direct_tir, max_nch);
432 	mlx5e_destroy_rqt(priv, &priv->indir_rqt);
433 	mlx5e_close_drop_rq(&priv->drop_rq);
434 	mlx5e_destroy_q_counters(priv);
435 }
436 
437 /* The stats groups order is opposite to the update_stats() order calls */
438 static mlx5e_stats_grp_t mlx5i_stats_grps[] = {
439 	&MLX5E_STATS_GRP(sw),
440 	&MLX5E_STATS_GRP(qcnt),
441 	&MLX5E_STATS_GRP(vnic_env),
442 	&MLX5E_STATS_GRP(vport),
443 	&MLX5E_STATS_GRP(802_3),
444 	&MLX5E_STATS_GRP(2863),
445 	&MLX5E_STATS_GRP(2819),
446 	&MLX5E_STATS_GRP(phy),
447 	&MLX5E_STATS_GRP(pcie),
448 	&MLX5E_STATS_GRP(per_prio),
449 	&MLX5E_STATS_GRP(pme),
450 	&MLX5E_STATS_GRP(channels),
451 	&MLX5E_STATS_GRP(per_port_buff_congest),
452 };
453 
454 static unsigned int mlx5i_stats_grps_num(struct mlx5e_priv *priv)
455 {
456 	return ARRAY_SIZE(mlx5i_stats_grps);
457 }
458 
459 static const struct mlx5e_profile mlx5i_nic_profile = {
460 	.init		   = mlx5i_init,
461 	.cleanup	   = mlx5i_cleanup,
462 	.init_tx	   = mlx5i_init_tx,
463 	.cleanup_tx	   = mlx5i_cleanup_tx,
464 	.init_rx	   = mlx5i_init_rx,
465 	.cleanup_rx	   = mlx5i_cleanup_rx,
466 	.enable		   = NULL, /* mlx5i_enable */
467 	.disable	   = NULL, /* mlx5i_disable */
468 	.update_rx	   = mlx5i_update_nic_rx,
469 	.update_stats	   = NULL, /* mlx5i_update_stats */
470 	.update_carrier    = NULL, /* no HW update in IB link */
471 	.rx_handlers       = &mlx5i_rx_handlers,
472 	.max_tc		   = MLX5I_MAX_NUM_TC,
473 	.rq_groups	   = MLX5E_NUM_RQ_GROUPS(REGULAR),
474 	.stats_grps        = mlx5i_stats_grps,
475 	.stats_grps_num    = mlx5i_stats_grps_num,
476 	.rx_ptp_support    = false,
477 };
478 
479 /* mlx5i netdev NDos */
480 
481 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
482 {
483 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
484 	struct mlx5e_params new_params;
485 	int err = 0;
486 
487 	mutex_lock(&priv->state_lock);
488 
489 	new_params = priv->channels.params;
490 	new_params.sw_mtu = new_mtu;
491 
492 	err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, true);
493 	if (err)
494 		goto out;
495 
496 	netdev->mtu = new_params.sw_mtu;
497 
498 out:
499 	mutex_unlock(&priv->state_lock);
500 	return err;
501 }
502 
503 int mlx5i_dev_init(struct net_device *dev)
504 {
505 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
506 	struct mlx5i_priv    *ipriv  = priv->ppriv;
507 
508 	/* Set dev address using underlay QP */
509 	dev->dev_addr[1] = (ipriv->qpn >> 16) & 0xff;
510 	dev->dev_addr[2] = (ipriv->qpn >>  8) & 0xff;
511 	dev->dev_addr[3] = (ipriv->qpn) & 0xff;
512 
513 	/* Add QPN to net-device mapping to HT */
514 	mlx5i_pkey_add_qpn(dev, ipriv->qpn);
515 
516 	return 0;
517 }
518 
519 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
520 {
521 	struct mlx5e_priv *priv = mlx5i_epriv(dev);
522 
523 	switch (cmd) {
524 	case SIOCSHWTSTAMP:
525 		return mlx5e_hwstamp_set(priv, ifr);
526 	case SIOCGHWTSTAMP:
527 		return mlx5e_hwstamp_get(priv, ifr);
528 	default:
529 		return -EOPNOTSUPP;
530 	}
531 }
532 
533 void mlx5i_dev_cleanup(struct net_device *dev)
534 {
535 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
536 	struct mlx5i_priv    *ipriv = priv->ppriv;
537 
538 	mlx5i_uninit_underlay_qp(priv);
539 
540 	/* Delete QPN to net-device mapping from HT */
541 	mlx5i_pkey_del_qpn(dev, ipriv->qpn);
542 }
543 
544 static int mlx5i_open(struct net_device *netdev)
545 {
546 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
547 	struct mlx5i_priv *ipriv = epriv->ppriv;
548 	struct mlx5_core_dev *mdev = epriv->mdev;
549 	int err;
550 
551 	mutex_lock(&epriv->state_lock);
552 
553 	set_bit(MLX5E_STATE_OPENED, &epriv->state);
554 
555 	err = mlx5i_init_underlay_qp(epriv);
556 	if (err) {
557 		mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
558 		goto err_clear_state_opened_flag;
559 	}
560 
561 	err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qpn);
562 	if (err) {
563 		mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
564 		goto err_reset_qp;
565 	}
566 
567 	err = mlx5e_open_channels(epriv, &epriv->channels);
568 	if (err)
569 		goto err_remove_fs_underlay_qp;
570 
571 	epriv->profile->update_rx(epriv);
572 	mlx5e_activate_priv_channels(epriv);
573 
574 	mutex_unlock(&epriv->state_lock);
575 	return 0;
576 
577 err_remove_fs_underlay_qp:
578 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
579 err_reset_qp:
580 	mlx5i_uninit_underlay_qp(epriv);
581 err_clear_state_opened_flag:
582 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
583 	mutex_unlock(&epriv->state_lock);
584 	return err;
585 }
586 
587 static int mlx5i_close(struct net_device *netdev)
588 {
589 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
590 	struct mlx5i_priv *ipriv = epriv->ppriv;
591 	struct mlx5_core_dev *mdev = epriv->mdev;
592 
593 	/* May already be CLOSED in case a previous configuration operation
594 	 * (e.g RX/TX queue size change) that involves close&open failed.
595 	 */
596 	mutex_lock(&epriv->state_lock);
597 
598 	if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
599 		goto unlock;
600 
601 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
602 
603 	netif_carrier_off(epriv->netdev);
604 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
605 	mlx5e_deactivate_priv_channels(epriv);
606 	mlx5e_close_channels(&epriv->channels);
607 	mlx5i_uninit_underlay_qp(epriv);
608 unlock:
609 	mutex_unlock(&epriv->state_lock);
610 	return 0;
611 }
612 
613 /* IPoIB RDMA netdev callbacks */
614 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
615 			      union ib_gid *gid, u16 lid, int set_qkey,
616 			      u32 qkey)
617 {
618 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
619 	struct mlx5_core_dev *mdev  = epriv->mdev;
620 	struct mlx5i_priv    *ipriv = epriv->ppriv;
621 	int err;
622 
623 	mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
624 		      gid->raw);
625 	err = mlx5_core_attach_mcg(mdev, gid, ipriv->qpn);
626 	if (err)
627 		mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
628 			       ipriv->qpn, gid->raw);
629 
630 	if (set_qkey) {
631 		mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
632 			      netdev->name, qkey);
633 		ipriv->qkey = qkey;
634 	}
635 
636 	return err;
637 }
638 
639 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
640 			      union ib_gid *gid, u16 lid)
641 {
642 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
643 	struct mlx5_core_dev *mdev  = epriv->mdev;
644 	struct mlx5i_priv    *ipriv = epriv->ppriv;
645 	int err;
646 
647 	mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
648 		      gid->raw);
649 
650 	err = mlx5_core_detach_mcg(mdev, gid, ipriv->qpn);
651 	if (err)
652 		mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
653 			      ipriv->qpn, gid->raw);
654 
655 	return err;
656 }
657 
658 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
659 		      struct ib_ah *address, u32 dqpn)
660 {
661 	struct mlx5e_priv *epriv = mlx5i_epriv(dev);
662 	struct mlx5e_txqsq *sq   = epriv->txq2sq[skb_get_queue_mapping(skb)];
663 	struct mlx5_ib_ah *mah   = to_mah(address);
664 	struct mlx5i_priv *ipriv = epriv->ppriv;
665 
666 	mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey, netdev_xmit_more());
667 
668 	return NETDEV_TX_OK;
669 }
670 
671 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
672 {
673 	struct mlx5i_priv *ipriv = netdev_priv(netdev);
674 
675 	ipriv->pkey_index = (u16)id;
676 }
677 
678 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
679 {
680 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
681 		return -EOPNOTSUPP;
682 
683 	if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
684 		mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
685 		return -EOPNOTSUPP;
686 	}
687 
688 	return 0;
689 }
690 
691 static void mlx5_rdma_netdev_free(struct net_device *netdev)
692 {
693 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
694 	struct mlx5_core_dev *mdev = priv->mdev;
695 	struct mlx5i_priv *ipriv = priv->ppriv;
696 	const struct mlx5e_profile *profile = priv->profile;
697 
698 	mlx5e_detach_netdev(priv);
699 	profile->cleanup(priv);
700 
701 	if (!ipriv->sub_interface) {
702 		mlx5i_pkey_qpn_ht_cleanup(netdev);
703 		mlx5e_destroy_mdev_resources(mdev);
704 	}
705 }
706 
707 static bool mlx5_is_sub_interface(struct mlx5_core_dev *mdev)
708 {
709 	return mdev->mlx5e_res.hw_objs.pdn != 0;
710 }
711 
712 static const struct mlx5e_profile *mlx5_get_profile(struct mlx5_core_dev *mdev)
713 {
714 	if (mlx5_is_sub_interface(mdev))
715 		return mlx5i_pkey_get_profile();
716 	return &mlx5i_nic_profile;
717 }
718 
719 static int mlx5_rdma_setup_rn(struct ib_device *ibdev, u32 port_num,
720 			      struct net_device *netdev, void *param)
721 {
722 	struct mlx5_core_dev *mdev = (struct mlx5_core_dev *)param;
723 	const struct mlx5e_profile *prof = mlx5_get_profile(mdev);
724 	struct mlx5i_priv *ipriv;
725 	struct mlx5e_priv *epriv;
726 	struct rdma_netdev *rn;
727 	int err;
728 
729 	ipriv = netdev_priv(netdev);
730 	epriv = mlx5i_epriv(netdev);
731 
732 	ipriv->sub_interface = mlx5_is_sub_interface(mdev);
733 	if (!ipriv->sub_interface) {
734 		err = mlx5i_pkey_qpn_ht_init(netdev);
735 		if (err) {
736 			mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
737 			return err;
738 		}
739 
740 		/* This should only be called once per mdev */
741 		err = mlx5e_create_mdev_resources(mdev);
742 		if (err)
743 			goto destroy_ht;
744 	}
745 
746 	err = mlx5e_priv_init(epriv, netdev, mdev);
747 	if (err)
748 		goto destroy_mdev_resources;
749 
750 	epriv->profile = prof;
751 	epriv->ppriv = ipriv;
752 
753 	prof->init(mdev, netdev);
754 
755 	err = mlx5e_attach_netdev(epriv);
756 	if (err)
757 		goto detach;
758 	netif_carrier_off(netdev);
759 
760 	/* set rdma_netdev func pointers */
761 	rn = &ipriv->rn;
762 	rn->hca  = ibdev;
763 	rn->send = mlx5i_xmit;
764 	rn->attach_mcast = mlx5i_attach_mcast;
765 	rn->detach_mcast = mlx5i_detach_mcast;
766 	rn->set_id = mlx5i_set_pkey_index;
767 
768 	netdev->priv_destructor = mlx5_rdma_netdev_free;
769 	netdev->needs_free_netdev = 1;
770 
771 	return 0;
772 
773 detach:
774 	prof->cleanup(epriv);
775 	if (ipriv->sub_interface)
776 		return err;
777 destroy_mdev_resources:
778 	mlx5e_destroy_mdev_resources(mdev);
779 destroy_ht:
780 	mlx5i_pkey_qpn_ht_cleanup(netdev);
781 	return err;
782 }
783 
784 int mlx5_rdma_rn_get_params(struct mlx5_core_dev *mdev,
785 			    struct ib_device *device,
786 			    struct rdma_netdev_alloc_params *params)
787 {
788 	int nch;
789 	int rc;
790 
791 	rc = mlx5i_check_required_hca_cap(mdev);
792 	if (rc)
793 		return rc;
794 
795 	nch = mlx5e_get_max_num_channels(mdev);
796 
797 	*params = (struct rdma_netdev_alloc_params){
798 		.sizeof_priv = sizeof(struct mlx5i_priv) +
799 			       sizeof(struct mlx5e_priv),
800 		.txqs = nch * MLX5E_MAX_NUM_TC,
801 		.rxqs = nch,
802 		.param = mdev,
803 		.initialize_rdma_netdev = mlx5_rdma_setup_rn,
804 	};
805 
806 	return 0;
807 }
808 EXPORT_SYMBOL(mlx5_rdma_rn_get_params);
809