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_eth_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->packet_merge.type = MLX5E_PACKET_MERGE_NONE;
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 rtnl_link_stats64 s = {};
114 	int i, j;
115 
116 	for (i = 0; i < priv->stats_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_dropped           += sq_stats->dropped;
132 		}
133 	}
134 
135 	memset(&priv->stats.sw, 0, sizeof(s));
136 
137 	priv->stats.sw.rx_packets = s.rx_packets;
138 	priv->stats.sw.rx_bytes = s.rx_bytes;
139 	priv->stats.sw.tx_packets = s.tx_packets;
140 	priv->stats.sw.tx_bytes = s.tx_bytes;
141 	priv->stats.sw.tx_queue_dropped = s.tx_dropped;
142 }
143 
144 void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
145 {
146 	struct mlx5e_priv     *priv   = mlx5i_epriv(dev);
147 	struct mlx5e_sw_stats *sstats = &priv->stats.sw;
148 
149 	mlx5i_grp_sw_update_stats(priv);
150 
151 	stats->rx_packets = sstats->rx_packets;
152 	stats->rx_bytes   = sstats->rx_bytes;
153 	stats->tx_packets = sstats->tx_packets;
154 	stats->tx_bytes   = sstats->tx_bytes;
155 	stats->tx_dropped = sstats->tx_queue_dropped;
156 }
157 
158 int mlx5i_init_underlay_qp(struct mlx5e_priv *priv)
159 {
160 	struct mlx5_core_dev *mdev = priv->mdev;
161 	struct mlx5i_priv *ipriv = priv->ppriv;
162 	int ret;
163 
164 	{
165 		u32 in[MLX5_ST_SZ_DW(rst2init_qp_in)] = {};
166 		u32 *qpc;
167 
168 		qpc = MLX5_ADDR_OF(rst2init_qp_in, in, qpc);
169 
170 		MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
171 		MLX5_SET(qpc, qpc, primary_address_path.pkey_index,
172 			 ipriv->pkey_index);
173 		MLX5_SET(qpc, qpc, primary_address_path.vhca_port_num, 1);
174 		MLX5_SET(qpc, qpc, q_key, IB_DEFAULT_Q_KEY);
175 
176 		MLX5_SET(rst2init_qp_in, in, opcode, MLX5_CMD_OP_RST2INIT_QP);
177 		MLX5_SET(rst2init_qp_in, in, qpn, ipriv->qpn);
178 		ret = mlx5_cmd_exec_in(mdev, rst2init_qp, in);
179 		if (ret)
180 			goto err_qp_modify_to_err;
181 	}
182 	{
183 		u32 in[MLX5_ST_SZ_DW(init2rtr_qp_in)] = {};
184 
185 		MLX5_SET(init2rtr_qp_in, in, opcode, MLX5_CMD_OP_INIT2RTR_QP);
186 		MLX5_SET(init2rtr_qp_in, in, qpn, ipriv->qpn);
187 		ret = mlx5_cmd_exec_in(mdev, init2rtr_qp, in);
188 		if (ret)
189 			goto err_qp_modify_to_err;
190 	}
191 	{
192 		u32 in[MLX5_ST_SZ_DW(rtr2rts_qp_in)] = {};
193 
194 		MLX5_SET(rtr2rts_qp_in, in, opcode, MLX5_CMD_OP_RTR2RTS_QP);
195 		MLX5_SET(rtr2rts_qp_in, in, qpn, ipriv->qpn);
196 		ret = mlx5_cmd_exec_in(mdev, rtr2rts_qp, in);
197 		if (ret)
198 			goto err_qp_modify_to_err;
199 	}
200 	return 0;
201 
202 err_qp_modify_to_err:
203 	{
204 		u32 in[MLX5_ST_SZ_DW(qp_2err_in)] = {};
205 
206 		MLX5_SET(qp_2err_in, in, opcode, MLX5_CMD_OP_2ERR_QP);
207 		MLX5_SET(qp_2err_in, in, qpn, ipriv->qpn);
208 		mlx5_cmd_exec_in(mdev, qp_2err, in);
209 	}
210 	return ret;
211 }
212 
213 void mlx5i_uninit_underlay_qp(struct mlx5e_priv *priv)
214 {
215 	struct mlx5i_priv *ipriv = priv->ppriv;
216 	struct mlx5_core_dev *mdev = priv->mdev;
217 	u32 in[MLX5_ST_SZ_DW(qp_2rst_in)] = {};
218 
219 	MLX5_SET(qp_2rst_in, in, opcode, MLX5_CMD_OP_2RST_QP);
220 	MLX5_SET(qp_2rst_in, in, qpn, ipriv->qpn);
221 	mlx5_cmd_exec_in(mdev, qp_2rst, in);
222 }
223 
224 #define MLX5_QP_ENHANCED_ULP_STATELESS_MODE 2
225 
226 int mlx5i_create_underlay_qp(struct mlx5e_priv *priv)
227 {
228 	const unsigned char *dev_addr = priv->netdev->dev_addr;
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 qpn = 0;
234 	int ret = 0;
235 	void *qpc;
236 
237 	if (MLX5_CAP_GEN(priv->mdev, mkey_by_name)) {
238 		qpn = (dev_addr[1] << 16) + (dev_addr[2] << 8) + dev_addr[3];
239 		MLX5_SET(create_qp_in, in, input_qpn, qpn);
240 	}
241 
242 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
243 	MLX5_SET(qpc, qpc, ts_format, mlx5_get_qp_default_ts(priv->mdev));
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 	int err;
324 
325 	priv->fs.ns = mlx5_get_flow_namespace(priv->mdev,
326 					       MLX5_FLOW_NAMESPACE_KERNEL);
327 
328 	if (!priv->fs.ns)
329 		return -EINVAL;
330 
331 	err = mlx5e_arfs_create_tables(priv);
332 	if (err) {
333 		netdev_err(priv->netdev, "Failed to create arfs tables, err=%d\n",
334 			   err);
335 		priv->netdev->hw_features &= ~NETIF_F_NTUPLE;
336 	}
337 
338 	err = mlx5e_create_ttc_table(priv);
339 	if (err) {
340 		netdev_err(priv->netdev, "Failed to create ttc table, err=%d\n",
341 			   err);
342 		goto err_destroy_arfs_tables;
343 	}
344 
345 	mlx5e_ethtool_init_steering(priv);
346 
347 	return 0;
348 
349 err_destroy_arfs_tables:
350 	mlx5e_arfs_destroy_tables(priv);
351 
352 	return err;
353 }
354 
355 static void mlx5i_destroy_flow_steering(struct mlx5e_priv *priv)
356 {
357 	mlx5e_destroy_ttc_table(priv);
358 	mlx5e_arfs_destroy_tables(priv);
359 	mlx5e_ethtool_cleanup_steering(priv);
360 }
361 
362 static int mlx5i_init_rx(struct mlx5e_priv *priv)
363 {
364 	struct mlx5_core_dev *mdev = priv->mdev;
365 	int err;
366 
367 	priv->rx_res = mlx5e_rx_res_alloc();
368 	if (!priv->rx_res)
369 		return -ENOMEM;
370 
371 	mlx5e_create_q_counters(priv);
372 
373 	err = mlx5e_open_drop_rq(priv, &priv->drop_rq);
374 	if (err) {
375 		mlx5_core_err(mdev, "open drop rq failed, %d\n", err);
376 		goto err_destroy_q_counters;
377 	}
378 
379 	err = mlx5e_rx_res_init(priv->rx_res, priv->mdev, 0,
380 				priv->max_nch, priv->drop_rq.rqn,
381 				&priv->channels.params.packet_merge,
382 				priv->channels.params.num_channels);
383 	if (err)
384 		goto err_close_drop_rq;
385 
386 	err = mlx5i_create_flow_steering(priv);
387 	if (err)
388 		goto err_destroy_rx_res;
389 
390 	return 0;
391 
392 err_destroy_rx_res:
393 	mlx5e_rx_res_destroy(priv->rx_res);
394 err_close_drop_rq:
395 	mlx5e_close_drop_rq(&priv->drop_rq);
396 err_destroy_q_counters:
397 	mlx5e_destroy_q_counters(priv);
398 	mlx5e_rx_res_free(priv->rx_res);
399 	priv->rx_res = NULL;
400 	return err;
401 }
402 
403 static void mlx5i_cleanup_rx(struct mlx5e_priv *priv)
404 {
405 	mlx5i_destroy_flow_steering(priv);
406 	mlx5e_rx_res_destroy(priv->rx_res);
407 	mlx5e_close_drop_rq(&priv->drop_rq);
408 	mlx5e_destroy_q_counters(priv);
409 	mlx5e_rx_res_free(priv->rx_res);
410 	priv->rx_res = NULL;
411 }
412 
413 /* The stats groups order is opposite to the update_stats() order calls */
414 static mlx5e_stats_grp_t mlx5i_stats_grps[] = {
415 	&MLX5E_STATS_GRP(sw),
416 	&MLX5E_STATS_GRP(qcnt),
417 	&MLX5E_STATS_GRP(vnic_env),
418 	&MLX5E_STATS_GRP(vport),
419 	&MLX5E_STATS_GRP(802_3),
420 	&MLX5E_STATS_GRP(2863),
421 	&MLX5E_STATS_GRP(2819),
422 	&MLX5E_STATS_GRP(phy),
423 	&MLX5E_STATS_GRP(pcie),
424 	&MLX5E_STATS_GRP(per_prio),
425 	&MLX5E_STATS_GRP(pme),
426 	&MLX5E_STATS_GRP(channels),
427 	&MLX5E_STATS_GRP(per_port_buff_congest),
428 };
429 
430 static unsigned int mlx5i_stats_grps_num(struct mlx5e_priv *priv)
431 {
432 	return ARRAY_SIZE(mlx5i_stats_grps);
433 }
434 
435 static const struct mlx5e_profile mlx5i_nic_profile = {
436 	.init		   = mlx5i_init,
437 	.cleanup	   = mlx5i_cleanup,
438 	.init_tx	   = mlx5i_init_tx,
439 	.cleanup_tx	   = mlx5i_cleanup_tx,
440 	.init_rx	   = mlx5i_init_rx,
441 	.cleanup_rx	   = mlx5i_cleanup_rx,
442 	.enable		   = NULL, /* mlx5i_enable */
443 	.disable	   = NULL, /* mlx5i_disable */
444 	.update_rx	   = mlx5i_update_nic_rx,
445 	.update_stats	   = NULL, /* mlx5i_update_stats */
446 	.update_carrier    = NULL, /* no HW update in IB link */
447 	.rx_handlers       = &mlx5i_rx_handlers,
448 	.max_tc		   = MLX5I_MAX_NUM_TC,
449 	.rq_groups	   = MLX5E_NUM_RQ_GROUPS(REGULAR),
450 	.stats_grps        = mlx5i_stats_grps,
451 	.stats_grps_num    = mlx5i_stats_grps_num,
452 };
453 
454 /* mlx5i netdev NDos */
455 
456 static int mlx5i_change_mtu(struct net_device *netdev, int new_mtu)
457 {
458 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
459 	struct mlx5e_params new_params;
460 	int err = 0;
461 
462 	mutex_lock(&priv->state_lock);
463 
464 	new_params = priv->channels.params;
465 	new_params.sw_mtu = new_mtu;
466 
467 	err = mlx5e_safe_switch_params(priv, &new_params, NULL, NULL, true);
468 	if (err)
469 		goto out;
470 
471 	netdev->mtu = new_params.sw_mtu;
472 
473 out:
474 	mutex_unlock(&priv->state_lock);
475 	return err;
476 }
477 
478 int mlx5i_dev_init(struct net_device *dev)
479 {
480 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
481 	struct mlx5i_priv    *ipriv  = priv->ppriv;
482 	u8 addr_mod[3];
483 
484 	/* Set dev address using underlay QP */
485 	addr_mod[0] = (ipriv->qpn >> 16) & 0xff;
486 	addr_mod[1] = (ipriv->qpn >>  8) & 0xff;
487 	addr_mod[2] = (ipriv->qpn) & 0xff;
488 	dev_addr_mod(dev, 1, addr_mod, sizeof(addr_mod));
489 
490 	/* Add QPN to net-device mapping to HT */
491 	mlx5i_pkey_add_qpn(dev, ipriv->qpn);
492 
493 	return 0;
494 }
495 
496 int mlx5i_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
497 {
498 	struct mlx5e_priv *priv = mlx5i_epriv(dev);
499 
500 	switch (cmd) {
501 	case SIOCSHWTSTAMP:
502 		return mlx5e_hwstamp_set(priv, ifr);
503 	case SIOCGHWTSTAMP:
504 		return mlx5e_hwstamp_get(priv, ifr);
505 	default:
506 		return -EOPNOTSUPP;
507 	}
508 }
509 
510 void mlx5i_dev_cleanup(struct net_device *dev)
511 {
512 	struct mlx5e_priv    *priv   = mlx5i_epriv(dev);
513 	struct mlx5i_priv    *ipriv = priv->ppriv;
514 
515 	mlx5i_uninit_underlay_qp(priv);
516 
517 	/* Delete QPN to net-device mapping from HT */
518 	mlx5i_pkey_del_qpn(dev, ipriv->qpn);
519 }
520 
521 static int mlx5i_open(struct net_device *netdev)
522 {
523 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
524 	struct mlx5i_priv *ipriv = epriv->ppriv;
525 	struct mlx5_core_dev *mdev = epriv->mdev;
526 	int err;
527 
528 	mutex_lock(&epriv->state_lock);
529 
530 	set_bit(MLX5E_STATE_OPENED, &epriv->state);
531 
532 	err = mlx5i_init_underlay_qp(epriv);
533 	if (err) {
534 		mlx5_core_warn(mdev, "prepare underlay qp state failed, %d\n", err);
535 		goto err_clear_state_opened_flag;
536 	}
537 
538 	err = mlx5_fs_add_rx_underlay_qpn(mdev, ipriv->qpn);
539 	if (err) {
540 		mlx5_core_warn(mdev, "attach underlay qp to ft failed, %d\n", err);
541 		goto err_reset_qp;
542 	}
543 
544 	err = mlx5e_open_channels(epriv, &epriv->channels);
545 	if (err)
546 		goto err_remove_fs_underlay_qp;
547 
548 	epriv->profile->update_rx(epriv);
549 	mlx5e_activate_priv_channels(epriv);
550 
551 	mutex_unlock(&epriv->state_lock);
552 	return 0;
553 
554 err_remove_fs_underlay_qp:
555 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
556 err_reset_qp:
557 	mlx5i_uninit_underlay_qp(epriv);
558 err_clear_state_opened_flag:
559 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
560 	mutex_unlock(&epriv->state_lock);
561 	return err;
562 }
563 
564 static int mlx5i_close(struct net_device *netdev)
565 {
566 	struct mlx5e_priv *epriv = mlx5i_epriv(netdev);
567 	struct mlx5i_priv *ipriv = epriv->ppriv;
568 	struct mlx5_core_dev *mdev = epriv->mdev;
569 
570 	/* May already be CLOSED in case a previous configuration operation
571 	 * (e.g RX/TX queue size change) that involves close&open failed.
572 	 */
573 	mutex_lock(&epriv->state_lock);
574 
575 	if (!test_bit(MLX5E_STATE_OPENED, &epriv->state))
576 		goto unlock;
577 
578 	clear_bit(MLX5E_STATE_OPENED, &epriv->state);
579 
580 	netif_carrier_off(epriv->netdev);
581 	mlx5_fs_remove_rx_underlay_qpn(mdev, ipriv->qpn);
582 	mlx5e_deactivate_priv_channels(epriv);
583 	mlx5e_close_channels(&epriv->channels);
584 	mlx5i_uninit_underlay_qp(epriv);
585 unlock:
586 	mutex_unlock(&epriv->state_lock);
587 	return 0;
588 }
589 
590 /* IPoIB RDMA netdev callbacks */
591 static int mlx5i_attach_mcast(struct net_device *netdev, struct ib_device *hca,
592 			      union ib_gid *gid, u16 lid, int set_qkey,
593 			      u32 qkey)
594 {
595 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
596 	struct mlx5_core_dev *mdev  = epriv->mdev;
597 	struct mlx5i_priv    *ipriv = epriv->ppriv;
598 	int err;
599 
600 	mlx5_core_dbg(mdev, "attaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
601 		      gid->raw);
602 	err = mlx5_core_attach_mcg(mdev, gid, ipriv->qpn);
603 	if (err)
604 		mlx5_core_warn(mdev, "failed attaching QPN 0x%x, MGID %pI6\n",
605 			       ipriv->qpn, gid->raw);
606 
607 	if (set_qkey) {
608 		mlx5_core_dbg(mdev, "%s setting qkey 0x%x\n",
609 			      netdev->name, qkey);
610 		ipriv->qkey = qkey;
611 	}
612 
613 	return err;
614 }
615 
616 static int mlx5i_detach_mcast(struct net_device *netdev, struct ib_device *hca,
617 			      union ib_gid *gid, u16 lid)
618 {
619 	struct mlx5e_priv    *epriv = mlx5i_epriv(netdev);
620 	struct mlx5_core_dev *mdev  = epriv->mdev;
621 	struct mlx5i_priv    *ipriv = epriv->ppriv;
622 	int err;
623 
624 	mlx5_core_dbg(mdev, "detaching QPN 0x%x, MGID %pI6\n", ipriv->qpn,
625 		      gid->raw);
626 
627 	err = mlx5_core_detach_mcg(mdev, gid, ipriv->qpn);
628 	if (err)
629 		mlx5_core_dbg(mdev, "failed detaching QPN 0x%x, MGID %pI6\n",
630 			      ipriv->qpn, gid->raw);
631 
632 	return err;
633 }
634 
635 static int mlx5i_xmit(struct net_device *dev, struct sk_buff *skb,
636 		      struct ib_ah *address, u32 dqpn)
637 {
638 	struct mlx5e_priv *epriv = mlx5i_epriv(dev);
639 	struct mlx5e_txqsq *sq   = epriv->txq2sq[skb_get_queue_mapping(skb)];
640 	struct mlx5_ib_ah *mah   = to_mah(address);
641 	struct mlx5i_priv *ipriv = epriv->ppriv;
642 
643 	mlx5i_sq_xmit(sq, skb, &mah->av, dqpn, ipriv->qkey, netdev_xmit_more());
644 
645 	return NETDEV_TX_OK;
646 }
647 
648 static void mlx5i_set_pkey_index(struct net_device *netdev, int id)
649 {
650 	struct mlx5i_priv *ipriv = netdev_priv(netdev);
651 
652 	ipriv->pkey_index = (u16)id;
653 }
654 
655 static int mlx5i_check_required_hca_cap(struct mlx5_core_dev *mdev)
656 {
657 	if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_IB)
658 		return -EOPNOTSUPP;
659 
660 	if (!MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads)) {
661 		mlx5_core_warn(mdev, "IPoIB enhanced offloads are not supported\n");
662 		return -EOPNOTSUPP;
663 	}
664 
665 	return 0;
666 }
667 
668 static void mlx5_rdma_netdev_free(struct net_device *netdev)
669 {
670 	struct mlx5e_priv *priv = mlx5i_epriv(netdev);
671 	struct mlx5_core_dev *mdev = priv->mdev;
672 	struct mlx5i_priv *ipriv = priv->ppriv;
673 	const struct mlx5e_profile *profile = priv->profile;
674 
675 	mlx5e_detach_netdev(priv);
676 	profile->cleanup(priv);
677 
678 	if (!ipriv->sub_interface) {
679 		mlx5i_pkey_qpn_ht_cleanup(netdev);
680 		mlx5e_destroy_mdev_resources(mdev);
681 	}
682 }
683 
684 static bool mlx5_is_sub_interface(struct mlx5_core_dev *mdev)
685 {
686 	return mdev->mlx5e_res.hw_objs.pdn != 0;
687 }
688 
689 static const struct mlx5e_profile *mlx5_get_profile(struct mlx5_core_dev *mdev)
690 {
691 	if (mlx5_is_sub_interface(mdev))
692 		return mlx5i_pkey_get_profile();
693 	return &mlx5i_nic_profile;
694 }
695 
696 static int mlx5_rdma_setup_rn(struct ib_device *ibdev, u32 port_num,
697 			      struct net_device *netdev, void *param)
698 {
699 	struct mlx5_core_dev *mdev = (struct mlx5_core_dev *)param;
700 	const struct mlx5e_profile *prof = mlx5_get_profile(mdev);
701 	struct mlx5i_priv *ipriv;
702 	struct mlx5e_priv *epriv;
703 	struct rdma_netdev *rn;
704 	int err;
705 
706 	ipriv = netdev_priv(netdev);
707 	epriv = mlx5i_epriv(netdev);
708 
709 	ipriv->sub_interface = mlx5_is_sub_interface(mdev);
710 	if (!ipriv->sub_interface) {
711 		err = mlx5i_pkey_qpn_ht_init(netdev);
712 		if (err) {
713 			mlx5_core_warn(mdev, "allocate qpn_to_netdev ht failed\n");
714 			return err;
715 		}
716 
717 		/* This should only be called once per mdev */
718 		err = mlx5e_create_mdev_resources(mdev);
719 		if (err)
720 			goto destroy_ht;
721 	}
722 
723 	err = mlx5e_priv_init(epriv, prof, netdev, mdev);
724 	if (err)
725 		goto destroy_mdev_resources;
726 
727 	epriv->profile = prof;
728 	epriv->ppriv = ipriv;
729 
730 	prof->init(mdev, netdev);
731 
732 	err = mlx5e_attach_netdev(epriv);
733 	if (err)
734 		goto detach;
735 	netif_carrier_off(netdev);
736 
737 	/* set rdma_netdev func pointers */
738 	rn = &ipriv->rn;
739 	rn->hca  = ibdev;
740 	rn->send = mlx5i_xmit;
741 	rn->attach_mcast = mlx5i_attach_mcast;
742 	rn->detach_mcast = mlx5i_detach_mcast;
743 	rn->set_id = mlx5i_set_pkey_index;
744 
745 	netdev->priv_destructor = mlx5_rdma_netdev_free;
746 	netdev->needs_free_netdev = 1;
747 
748 	return 0;
749 
750 detach:
751 	prof->cleanup(epriv);
752 	if (ipriv->sub_interface)
753 		return err;
754 destroy_mdev_resources:
755 	mlx5e_destroy_mdev_resources(mdev);
756 destroy_ht:
757 	mlx5i_pkey_qpn_ht_cleanup(netdev);
758 	return err;
759 }
760 
761 int mlx5_rdma_rn_get_params(struct mlx5_core_dev *mdev,
762 			    struct ib_device *device,
763 			    struct rdma_netdev_alloc_params *params)
764 {
765 	int nch;
766 	int rc;
767 
768 	rc = mlx5i_check_required_hca_cap(mdev);
769 	if (rc)
770 		return rc;
771 
772 	nch = mlx5e_get_max_num_channels(mdev);
773 
774 	*params = (struct rdma_netdev_alloc_params){
775 		.sizeof_priv = sizeof(struct mlx5i_priv) +
776 			       sizeof(struct mlx5e_priv),
777 		.txqs = nch * MLX5E_MAX_NUM_TC,
778 		.rxqs = nch,
779 		.param = mdev,
780 		.initialize_rdma_netdev = mlx5_rdma_setup_rn,
781 	};
782 
783 	return 0;
784 }
785 EXPORT_SYMBOL(mlx5_rdma_rn_get_params);
786