xref: /freebsd/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #define	LINUXKPI_PARAM_PREFIX mlx4_
35 
36 #include <linux/module.h>
37 #include <linux/slab.h>
38 #include <linux/errno.h>
39 #include <linux/etherdevice.h>
40 #include <linux/netdevice.h>
41 #include <linux/if_vlan.h>
42 #include <linux/fs.h>
43 #include <linux/rcupdate.h>
44 #include <linux/notifier.h>
45 #include <linux/delay.h>
46 
47 #include <net/ipv6.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <dev/mlx4/driver.h>
55 #include <dev/mlx4/cmd.h>
56 #include <dev/mlx4/qp.h>
57 #include <linux/sched.h>
58 #include <linux/page.h>
59 #include <linux/printk.h>
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 #include "wc.h"
63 
64 #define DRV_NAME	MLX4_IB_DRV_NAME
65 #ifndef DRV_VERSION
66 #define DRV_VERSION	"3.7.1"
67 #endif
68 #define DRV_RELDATE	"November 2021"
69 
70 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
71 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
72 #define MLX4_IB_CARD_REV_A0   0xA0
73 
74 MODULE_AUTHOR("Roland Dreier");
75 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
76 MODULE_LICENSE("Dual BSD/GPL");
77 
78 int mlx4_ib_sm_guid_assign = 0;
79 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
80 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
81 
82 static const char mlx4_ib_version[] =
83 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
84 	DRV_VERSION " (" DRV_RELDATE ")\n";
85 
86 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
87 
88 static struct workqueue_struct *wq;
89 
90 static void init_query_mad(struct ib_smp *mad)
91 {
92 	mad->base_version  = 1;
93 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
94 	mad->class_version = 1;
95 	mad->method	   = IB_MGMT_METHOD_GET;
96 }
97 
98 static int check_flow_steering_support(struct mlx4_dev *dev)
99 {
100 	int eth_num_ports = 0;
101 	int ib_num_ports = 0;
102 
103 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
104 
105 	if (dmfs) {
106 		int i;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
108 			eth_num_ports++;
109 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
110 			ib_num_ports++;
111 		dmfs &= (!ib_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
113 			(!eth_num_ports ||
114 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
115 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
116 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
117 			dmfs = 0;
118 		}
119 	}
120 	return dmfs;
121 }
122 
123 static int num_ib_ports(struct mlx4_dev *dev)
124 {
125 	int ib_ports = 0;
126 	int i;
127 
128 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
129 		ib_ports++;
130 
131 	return ib_ports;
132 }
133 
134 static if_t mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
135 {
136 	struct mlx4_ib_dev *ibdev = to_mdev(device);
137 	if_t dev;
138 
139 	rcu_read_lock();
140 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
141 
142 #if 0
143 	if (dev) {
144 		if (mlx4_is_bonded(ibdev->dev)) {
145 			if_t upper = NULL;
146 
147 			upper = netdev_master_upper_dev_get_rcu(dev);
148 			if (upper) {
149 				if_t active;
150 
151 				active = bond_option_active_slave_get_rcu(mlx4_netdev_priv(upper));
152 				if (active)
153 					dev = active;
154 			}
155 		}
156 	}
157 #endif
158 	if (dev)
159 		if_ref(dev);
160 
161 	rcu_read_unlock();
162 	return dev;
163 }
164 
165 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
166 				  struct mlx4_ib_dev *ibdev,
167 				  u8 port_num)
168 {
169 	struct mlx4_cmd_mailbox *mailbox;
170 	int err;
171 	struct mlx4_dev *dev = ibdev->dev;
172 	int i;
173 	union ib_gid *gid_tbl;
174 
175 	mailbox = mlx4_alloc_cmd_mailbox(dev);
176 	if (IS_ERR(mailbox))
177 		return -ENOMEM;
178 
179 	gid_tbl = mailbox->buf;
180 
181 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
182 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
183 
184 	err = mlx4_cmd(dev, mailbox->dma,
185 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
186 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
187 		       MLX4_CMD_WRAPPED);
188 	if (mlx4_is_bonded(dev))
189 		err += mlx4_cmd(dev, mailbox->dma,
190 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
191 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
192 				MLX4_CMD_WRAPPED);
193 
194 	mlx4_free_cmd_mailbox(dev, mailbox);
195 	return err;
196 }
197 
198 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
199 				     struct mlx4_ib_dev *ibdev,
200 				     u8 port_num)
201 {
202 	struct mlx4_cmd_mailbox *mailbox;
203 	int err;
204 	struct mlx4_dev *dev = ibdev->dev;
205 	int i;
206 	struct {
207 		union ib_gid	gid;
208 		__be32		rsrvd1[2];
209 		__be16		rsrvd2;
210 		u8		type;
211 		u8		version;
212 		__be32		rsrvd3;
213 	} *gid_tbl;
214 
215 	mailbox = mlx4_alloc_cmd_mailbox(dev);
216 	if (IS_ERR(mailbox))
217 		return -ENOMEM;
218 
219 	gid_tbl = mailbox->buf;
220 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
221 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
222 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
223 			gid_tbl[i].version = 2;
224 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
225 				gid_tbl[i].type = 1;
226 			else
227 				memset(&gid_tbl[i].gid, 0, 12);
228 		}
229 	}
230 
231 	err = mlx4_cmd(dev, mailbox->dma,
232 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
233 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
234 		       MLX4_CMD_WRAPPED);
235 	if (mlx4_is_bonded(dev))
236 		err += mlx4_cmd(dev, mailbox->dma,
237 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
238 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
239 				MLX4_CMD_WRAPPED);
240 
241 	mlx4_free_cmd_mailbox(dev, mailbox);
242 	return err;
243 }
244 
245 static int mlx4_ib_update_gids(struct gid_entry *gids,
246 			       struct mlx4_ib_dev *ibdev,
247 			       u8 port_num)
248 {
249 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
250 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
251 
252 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
253 }
254 
255 static int mlx4_ib_add_gid(struct ib_device *device,
256 			   u8 port_num,
257 			   unsigned int index,
258 			   const union ib_gid *gid,
259 			   const struct ib_gid_attr *attr,
260 			   void **context)
261 {
262 	struct mlx4_ib_dev *ibdev = to_mdev(device);
263 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
264 	struct mlx4_port_gid_table   *port_gid_table;
265 	int free = -1, found = -1;
266 	int ret = 0;
267 	int hw_update = 0;
268 	int i;
269 	struct gid_entry *gids = NULL;
270 
271 	if (!rdma_cap_roce_gid_table(device, port_num))
272 		return -EINVAL;
273 
274 	if (port_num > MLX4_MAX_PORTS)
275 		return -EINVAL;
276 
277 	if (!context)
278 		return -EINVAL;
279 
280 	port_gid_table = &iboe->gids[port_num - 1];
281 	spin_lock_bh(&iboe->lock);
282 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
283 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
284 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
285 			found = i;
286 			break;
287 		}
288 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
289 			free = i; /* HW has space */
290 	}
291 
292 	if (found < 0) {
293 		if (free < 0) {
294 			ret = -ENOSPC;
295 		} else {
296 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
297 			if (!port_gid_table->gids[free].ctx) {
298 				ret = -ENOMEM;
299 			} else {
300 				*context = port_gid_table->gids[free].ctx;
301 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
302 				port_gid_table->gids[free].gid_type = attr->gid_type;
303 				port_gid_table->gids[free].ctx->real_index = free;
304 				port_gid_table->gids[free].ctx->refcount = 1;
305 				hw_update = 1;
306 			}
307 		}
308 	} else {
309 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
310 		*context = ctx;
311 		ctx->refcount++;
312 	}
313 	if (!ret && hw_update) {
314 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
315 		if (!gids) {
316 			ret = -ENOMEM;
317 		} else {
318 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
319 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
320 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
321 			}
322 		}
323 	}
324 	spin_unlock_bh(&iboe->lock);
325 
326 	if (!ret && hw_update) {
327 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
328 		kfree(gids);
329 	}
330 
331 	return ret;
332 }
333 
334 static int mlx4_ib_del_gid(struct ib_device *device,
335 			   u8 port_num,
336 			   unsigned int index,
337 			   void **context)
338 {
339 	struct gid_cache_context *ctx = *context;
340 	struct mlx4_ib_dev *ibdev = to_mdev(device);
341 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
342 	struct mlx4_port_gid_table   *port_gid_table;
343 	int ret = 0;
344 	int hw_update = 0;
345 	struct gid_entry *gids = NULL;
346 
347 	if (!rdma_cap_roce_gid_table(device, port_num))
348 		return -EINVAL;
349 
350 	if (port_num > MLX4_MAX_PORTS)
351 		return -EINVAL;
352 
353 	port_gid_table = &iboe->gids[port_num - 1];
354 	spin_lock_bh(&iboe->lock);
355 	if (ctx) {
356 		ctx->refcount--;
357 		if (!ctx->refcount) {
358 			unsigned int real_index = ctx->real_index;
359 
360 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
361 			kfree(port_gid_table->gids[real_index].ctx);
362 			port_gid_table->gids[real_index].ctx = NULL;
363 			hw_update = 1;
364 		}
365 	}
366 	if (!ret && hw_update) {
367 		int i;
368 
369 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
370 		if (!gids) {
371 			ret = -ENOMEM;
372 		} else {
373 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
374 				memcpy(&gids[i].gid,
375 				       &port_gid_table->gids[i].gid,
376 				       sizeof(union ib_gid));
377 				gids[i].gid_type =
378 				    port_gid_table->gids[i].gid_type;
379 			}
380 		}
381 	}
382 	spin_unlock_bh(&iboe->lock);
383 
384 	if (!ret && hw_update) {
385 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
386 		kfree(gids);
387 	}
388 	return ret;
389 }
390 
391 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
392 				    u8 port_num, int index)
393 {
394 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
395 	struct gid_cache_context *ctx = NULL;
396 	union ib_gid gid;
397 	struct mlx4_port_gid_table   *port_gid_table;
398 	int real_index = -EINVAL;
399 	int i;
400 	int ret;
401 	unsigned long flags;
402 	struct ib_gid_attr attr;
403 
404 	if (port_num > MLX4_MAX_PORTS)
405 		return -EINVAL;
406 
407 	if (mlx4_is_bonded(ibdev->dev))
408 		port_num = 1;
409 
410 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
411 		return index;
412 
413 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
414 	if (ret)
415 		return ret;
416 
417 	if (attr.ndev)
418 		if_rele(attr.ndev);
419 
420 	if (!memcmp(&gid, &zgid, sizeof(gid)))
421 		return -EINVAL;
422 
423 	spin_lock_irqsave(&iboe->lock, flags);
424 	port_gid_table = &iboe->gids[port_num - 1];
425 
426 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
427 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
428 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
429 			ctx = port_gid_table->gids[i].ctx;
430 			break;
431 		}
432 	if (ctx)
433 		real_index = ctx->real_index;
434 	spin_unlock_irqrestore(&iboe->lock, flags);
435 	return real_index;
436 }
437 
438 static int mlx4_ib_query_device(struct ib_device *ibdev,
439 				struct ib_device_attr *props,
440 				struct ib_udata *uhw)
441 {
442 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
443 	struct ib_smp *in_mad  = NULL;
444 	struct ib_smp *out_mad = NULL;
445 	int err = -ENOMEM;
446 	int have_ib_ports;
447 	struct mlx4_uverbs_ex_query_device cmd;
448 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
449 	struct mlx4_clock_params clock_params;
450 
451 	if (uhw->inlen) {
452 		if (uhw->inlen < sizeof(cmd))
453 			return -EINVAL;
454 
455 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
456 		if (err)
457 			return err;
458 
459 		if (cmd.comp_mask)
460 			return -EINVAL;
461 
462 		if (cmd.reserved)
463 			return -EINVAL;
464 	}
465 
466 	resp.response_length = offsetof(typeof(resp), response_length) +
467 		sizeof(resp.response_length);
468 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
469 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
470 	if (!in_mad || !out_mad)
471 		goto out;
472 
473 	init_query_mad(in_mad);
474 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
475 
476 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
477 			   1, NULL, NULL, in_mad, out_mad);
478 	if (err)
479 		goto out;
480 
481 	memset(props, 0, sizeof *props);
482 
483 	have_ib_ports = num_ib_ports(dev->dev);
484 
485 	props->fw_ver = dev->dev->caps.fw_ver;
486 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
487 		IB_DEVICE_PORT_ACTIVE_EVENT		|
488 		IB_DEVICE_SYS_IMAGE_GUID		|
489 		IB_DEVICE_RC_RNR_NAK_GEN		|
490 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
492 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
493 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
494 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
496 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
497 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
498 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
499 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
500 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
501 	if (dev->dev->caps.max_gso_sz &&
502 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
503 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
504 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
505 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
506 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
507 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
508 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
509 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
510 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
511 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
512 		props->device_cap_flags |= IB_DEVICE_XRC;
513 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
514 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
515 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
516 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
517 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
518 		else
519 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
520 	}
521 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
522 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
523 
524 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
525 
526 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
527 		0xffffff;
528 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
529 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
530 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
531 
532 	props->max_mr_size	   = ~0ull;
533 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
534 	props->max_qp		   = dev->dev->quotas.qp;
535 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
536 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
537 					 dev->dev->caps.max_rq_sg);
538 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
539 	props->max_cq		   = dev->dev->quotas.cq;
540 	props->max_cqe		   = dev->dev->caps.max_cqes;
541 	props->max_mr		   = dev->dev->quotas.mpt;
542 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
543 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
544 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
545 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
546 	props->max_srq		   = dev->dev->quotas.srq;
547 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
548 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
549 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
550 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
551 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
552 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
553 	props->masked_atomic_cap   = props->atomic_cap;
554 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
555 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
556 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
557 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
558 					   props->max_mcast_grp;
559 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
560 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
561 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
562 
563 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
564 		resp.response_length += sizeof(resp.hca_core_clock_offset);
565 		if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
566 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
567 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
568 		}
569 	}
570 
571 	if (uhw->outlen) {
572 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
573 		if (err)
574 			goto out;
575 	}
576 out:
577 	kfree(in_mad);
578 	kfree(out_mad);
579 
580 	return err;
581 }
582 
583 static enum rdma_link_layer
584 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
585 {
586 	struct mlx4_dev *dev = to_mdev(device)->dev;
587 
588 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
589 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
590 }
591 
592 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
593 			      struct ib_port_attr *props, int netw_view)
594 {
595 	struct ib_smp *in_mad  = NULL;
596 	struct ib_smp *out_mad = NULL;
597 	int ext_active_speed;
598 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
599 	int err = -ENOMEM;
600 
601 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
602 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
603 	if (!in_mad || !out_mad)
604 		goto out;
605 
606 	init_query_mad(in_mad);
607 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
608 	in_mad->attr_mod = cpu_to_be32(port);
609 
610 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
611 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
612 
613 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
614 				in_mad, out_mad);
615 	if (err)
616 		goto out;
617 
618 
619 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
620 	props->lmc		= out_mad->data[34] & 0x7;
621 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
622 	props->sm_sl		= out_mad->data[36] & 0xf;
623 	props->state		= out_mad->data[32] & 0xf;
624 	props->phys_state	= out_mad->data[33] >> 4;
625 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
626 	if (netw_view)
627 		props->gid_tbl_len = out_mad->data[50];
628 	else
629 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
630 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
631 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
632 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
633 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
634 	props->active_width	= out_mad->data[31] & 0xf;
635 	props->active_speed	= out_mad->data[35] >> 4;
636 	props->max_mtu		= out_mad->data[41] & 0xf;
637 	props->active_mtu	= out_mad->data[36] >> 4;
638 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
639 	props->max_vl_num	= out_mad->data[37] >> 4;
640 	props->init_type_reply	= out_mad->data[41] >> 4;
641 
642 	/* Check if extended speeds (EDR/FDR/...) are supported */
643 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
644 		ext_active_speed = out_mad->data[62] >> 4;
645 
646 		switch (ext_active_speed) {
647 		case 1:
648 			props->active_speed = IB_SPEED_FDR;
649 			break;
650 		case 2:
651 			props->active_speed = IB_SPEED_EDR;
652 			break;
653 		}
654 	}
655 
656 	/* If reported active speed is QDR, check if is FDR-10 */
657 	if (props->active_speed == IB_SPEED_QDR) {
658 		init_query_mad(in_mad);
659 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
660 		in_mad->attr_mod = cpu_to_be32(port);
661 
662 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
663 				   NULL, NULL, in_mad, out_mad);
664 		if (err)
665 			goto out;
666 
667 		/* Checking LinkSpeedActive for FDR-10 */
668 		if (out_mad->data[15] & 0x1)
669 			props->active_speed = IB_SPEED_FDR10;
670 	}
671 
672 	/* Avoid wrong speed value returned by FW if the IB link is down. */
673 	if (props->state == IB_PORT_DOWN)
674 		 props->active_speed = IB_SPEED_SDR;
675 
676 out:
677 	kfree(in_mad);
678 	kfree(out_mad);
679 	return err;
680 }
681 
682 static u8 state_to_phys_state(enum ib_port_state state)
683 {
684 	return state == IB_PORT_ACTIVE ?
685 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
686 }
687 
688 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
689 			       struct ib_port_attr *props, int netw_view)
690 {
691 
692 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
693 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
694 	if_t ndev;
695 	enum ib_mtu tmp;
696 	struct mlx4_cmd_mailbox *mailbox;
697 	int err = 0;
698 	int is_bonded = mlx4_is_bonded(mdev->dev);
699 
700 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
701 	if (IS_ERR(mailbox))
702 		return PTR_ERR(mailbox);
703 
704 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
705 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
706 			   MLX4_CMD_WRAPPED);
707 	if (err)
708 		goto out;
709 
710 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
711 						IB_WIDTH_4X : IB_WIDTH_1X;
712 	props->active_speed	= IB_SPEED_QDR;
713 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
714 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
715 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
716 	props->pkey_tbl_len	= 1;
717 	props->max_mtu		= IB_MTU_4096;
718 	props->max_vl_num	= 2;
719 	props->state		= IB_PORT_DOWN;
720 	props->phys_state	= state_to_phys_state(props->state);
721 	props->active_mtu	= IB_MTU_256;
722 	spin_lock_bh(&iboe->lock);
723 	ndev = iboe->netdevs[port - 1];
724 	if (ndev && is_bonded) {
725 #if 0
726 		rcu_read_lock(); /* required to get upper dev */
727 		ndev = netdev_master_upper_dev_get_rcu(ndev);
728 		rcu_read_unlock();
729 #endif
730 	}
731 	if (!ndev)
732 		goto out_unlock;
733 
734 	tmp = iboe_get_mtu(if_getmtu(ndev));
735 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
736 
737 	props->state		= ((if_getdrvflags(ndev) & IFF_DRV_RUNNING) != 0 &&
738 				   if_getlinkstate(ndev) == LINK_STATE_UP) ?
739 					IB_PORT_ACTIVE : IB_PORT_DOWN;
740 	props->phys_state	= state_to_phys_state(props->state);
741 out_unlock:
742 	spin_unlock_bh(&iboe->lock);
743 out:
744 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
745 	return err;
746 }
747 
748 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
749 			 struct ib_port_attr *props, int netw_view)
750 {
751 	int err;
752 
753 	memset(props, 0, sizeof *props);
754 
755 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
756 		ib_link_query_port(ibdev, port, props, netw_view) :
757 				eth_link_query_port(ibdev, port, props, netw_view);
758 
759 	return err;
760 }
761 
762 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
763 			      struct ib_port_attr *props)
764 {
765 	/* returns host view */
766 	return __mlx4_ib_query_port(ibdev, port, props, 0);
767 }
768 
769 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
770 			union ib_gid *gid, int netw_view)
771 {
772 	struct ib_smp *in_mad  = NULL;
773 	struct ib_smp *out_mad = NULL;
774 	int err = -ENOMEM;
775 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
776 	int clear = 0;
777 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
778 
779 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
780 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
781 	if (!in_mad || !out_mad)
782 		goto out;
783 
784 	init_query_mad(in_mad);
785 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
786 	in_mad->attr_mod = cpu_to_be32(port);
787 
788 	if (mlx4_is_mfunc(dev->dev) && netw_view)
789 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
790 
791 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
792 	if (err)
793 		goto out;
794 
795 	memcpy(gid->raw, out_mad->data + 8, 8);
796 
797 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
798 		if (index) {
799 			/* For any index > 0, return the null guid */
800 			err = 0;
801 			clear = 1;
802 			goto out;
803 		}
804 	}
805 
806 	init_query_mad(in_mad);
807 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
808 	in_mad->attr_mod = cpu_to_be32(index / 8);
809 
810 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
811 			   NULL, NULL, in_mad, out_mad);
812 	if (err)
813 		goto out;
814 
815 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
816 
817 out:
818 	if (clear)
819 		memset(gid->raw + 8, 0, 8);
820 	kfree(in_mad);
821 	kfree(out_mad);
822 	return err;
823 }
824 
825 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
826 			     union ib_gid *gid)
827 {
828 	int ret;
829 
830 	if (rdma_protocol_ib(ibdev, port))
831 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
832 
833 	if (!rdma_protocol_roce(ibdev, port))
834 		return -ENODEV;
835 
836 	if (!rdma_cap_roce_gid_table(ibdev, port))
837 		return -ENODEV;
838 
839 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
840 	if (ret == -EAGAIN) {
841 		memcpy(gid, &zgid, sizeof(*gid));
842 		return 0;
843 	}
844 
845 	return ret;
846 }
847 
848 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
849 {
850 	union sl2vl_tbl_to_u64 sl2vl64;
851 	struct ib_smp *in_mad  = NULL;
852 	struct ib_smp *out_mad = NULL;
853 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
854 	int err = -ENOMEM;
855 	int jj;
856 
857 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
858 		*sl2vl_tbl = 0;
859 		return 0;
860 	}
861 
862 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
863 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
864 	if (!in_mad || !out_mad)
865 		goto out;
866 
867 	init_query_mad(in_mad);
868 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
869 	in_mad->attr_mod = 0;
870 
871 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
872 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
873 
874 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
875 			   in_mad, out_mad);
876 	if (err)
877 		goto out;
878 
879 	for (jj = 0; jj < 8; jj++)
880 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
881 	*sl2vl_tbl = sl2vl64.sl64;
882 
883 out:
884 	kfree(in_mad);
885 	kfree(out_mad);
886 	return err;
887 }
888 
889 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
890 {
891 	u64 sl2vl;
892 	int i;
893 	int err;
894 
895 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
896 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
897 			continue;
898 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
899 		if (err) {
900 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
901 			       i, err);
902 			sl2vl = 0;
903 		}
904 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
905 	}
906 }
907 
908 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
909 			 u16 *pkey, int netw_view)
910 {
911 	struct ib_smp *in_mad  = NULL;
912 	struct ib_smp *out_mad = NULL;
913 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
914 	int err = -ENOMEM;
915 
916 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
917 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
918 	if (!in_mad || !out_mad)
919 		goto out;
920 
921 	init_query_mad(in_mad);
922 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
923 	in_mad->attr_mod = cpu_to_be32(index / 32);
924 
925 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
926 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
927 
928 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
929 			   in_mad, out_mad);
930 	if (err)
931 		goto out;
932 
933 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
934 
935 out:
936 	kfree(in_mad);
937 	kfree(out_mad);
938 	return err;
939 }
940 
941 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
942 {
943 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
944 }
945 
946 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
947 				 struct ib_device_modify *props)
948 {
949 	struct mlx4_cmd_mailbox *mailbox;
950 	unsigned long flags;
951 
952 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
953 		return -EOPNOTSUPP;
954 
955 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
956 		return 0;
957 
958 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
959 		return -EOPNOTSUPP;
960 
961 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
962 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
963 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
964 
965 	/*
966 	 * If possible, pass node desc to FW, so it can generate
967 	 * a 144 trap.  If cmd fails, just ignore.
968 	 */
969 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
970 	if (IS_ERR(mailbox))
971 		return 0;
972 
973 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
974 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
975 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
976 
977 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
978 
979 	return 0;
980 }
981 
982 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
983 			    u32 cap_mask)
984 {
985 	struct mlx4_cmd_mailbox *mailbox;
986 	int err;
987 
988 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
989 	if (IS_ERR(mailbox))
990 		return PTR_ERR(mailbox);
991 
992 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
993 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
994 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
995 	} else {
996 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
997 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
998 	}
999 
1000 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1001 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1002 		       MLX4_CMD_WRAPPED);
1003 
1004 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1005 	return err;
1006 }
1007 
1008 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1009 			       struct ib_port_modify *props)
1010 {
1011 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1012 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1013 	struct ib_port_attr attr;
1014 	u32 cap_mask;
1015 	int err;
1016 
1017 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1018 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1019 	 * violations and port capabilities are not meaningful.
1020 	 */
1021 	if (is_eth)
1022 		return 0;
1023 
1024 	mutex_lock(&mdev->cap_mask_mutex);
1025 
1026 	err = mlx4_ib_query_port(ibdev, port, &attr);
1027 	if (err)
1028 		goto out;
1029 
1030 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1031 		~props->clr_port_cap_mask;
1032 
1033 	err = mlx4_ib_SET_PORT(mdev, port,
1034 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1035 			       cap_mask);
1036 
1037 out:
1038 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1039 	return err;
1040 }
1041 
1042 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1043 				  struct ib_udata *udata)
1044 {
1045 	struct ib_device *ibdev = uctx->device;
1046 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1047 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1048 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1049 	struct mlx4_ib_alloc_ucontext_resp resp;
1050 	int err;
1051 
1052 	if (!dev->ib_active)
1053 		return -EAGAIN;
1054 
1055 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1056 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1057 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1058 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1059 	} else {
1060 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1061 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1062 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1063 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1064 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1065 	}
1066 
1067 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1068 	if (err)
1069 		return err;
1070 
1071 	INIT_LIST_HEAD(&context->db_page_list);
1072 	mutex_init(&context->db_page_mutex);
1073 
1074 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1075 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1076 	else
1077 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1078 
1079 	if (err) {
1080 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1081 		return -EFAULT;
1082 	}
1083 
1084 	return err;
1085 }
1086 
1087 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1088 {
1089 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1090 
1091 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1092 }
1093 
1094 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1095 {
1096 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1097 
1098 	switch (vma->vm_pgoff) {
1099 	case 0:
1100 		return rdma_user_mmap_io(context, vma,
1101 					 to_mucontext(context)->uar.pfn,
1102 					 PAGE_SIZE,
1103 					 pgprot_noncached(vma->vm_page_prot),
1104 					 NULL);
1105 
1106 	case 1:
1107 		if (dev->dev->caps.bf_reg_size == 0)
1108 			return -EINVAL;
1109 		return rdma_user_mmap_io(
1110 			context, vma,
1111 			to_mucontext(context)->uar.pfn +
1112 				dev->dev->caps.num_uars,
1113 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1114 			NULL);
1115 
1116 	case 3: {
1117 		struct mlx4_clock_params params;
1118 		int ret;
1119 
1120 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1121 		if (ret)
1122 			return ret;
1123 
1124 		return rdma_user_mmap_io(
1125 			context, vma,
1126 			(pci_resource_start(dev->dev->persist->pdev,
1127 					    params.bar) +
1128 			 params.offset) >>
1129 				PAGE_SHIFT,
1130 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1131 			NULL);
1132 	}
1133 
1134 	default:
1135 		return -EINVAL;
1136 	}
1137 }
1138 
1139 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1140 {
1141 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1142 	struct ib_device *ibdev = ibpd->device;
1143 	int err;
1144 
1145 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1146 	if (err)
1147 		return err;
1148 
1149 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1150 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1151 		return -EFAULT;
1152 	}
1153 	return 0;
1154 }
1155 
1156 static void mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1157 {
1158 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1159 }
1160 
1161 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1162 					  struct ib_udata *udata)
1163 {
1164 	struct mlx4_ib_xrcd *xrcd;
1165 	struct ib_cq_init_attr cq_attr = {};
1166 	int err;
1167 
1168 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1169 		return ERR_PTR(-ENOSYS);
1170 
1171 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1172 	if (!xrcd)
1173 		return ERR_PTR(-ENOMEM);
1174 
1175 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1176 	if (err)
1177 		goto err1;
1178 
1179 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1180 	if (IS_ERR(xrcd->pd)) {
1181 		err = PTR_ERR(xrcd->pd);
1182 		goto err2;
1183 	}
1184 
1185 	cq_attr.cqe = 1;
1186 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1187 	if (IS_ERR(xrcd->cq)) {
1188 		err = PTR_ERR(xrcd->cq);
1189 		goto err3;
1190 	}
1191 
1192 	return &xrcd->ibxrcd;
1193 
1194 err3:
1195 	ib_dealloc_pd(xrcd->pd);
1196 err2:
1197 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1198 err1:
1199 	kfree(xrcd);
1200 	return ERR_PTR(err);
1201 }
1202 
1203 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1204 {
1205 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1206 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1207 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1208 	kfree(xrcd);
1209 
1210 	return 0;
1211 }
1212 
1213 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1214 {
1215 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1216 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1217 	struct mlx4_ib_gid_entry *ge;
1218 
1219 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1220 	if (!ge)
1221 		return -ENOMEM;
1222 
1223 	ge->gid = *gid;
1224 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1225 		ge->port = mqp->port;
1226 		ge->added = 1;
1227 	}
1228 
1229 	mutex_lock(&mqp->mutex);
1230 	list_add_tail(&ge->list, &mqp->gid_list);
1231 	mutex_unlock(&mqp->mutex);
1232 
1233 	return 0;
1234 }
1235 
1236 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1237 					  struct mlx4_ib_counters *ctr_table)
1238 {
1239 	struct counter_index *counter, *tmp_count;
1240 
1241 	mutex_lock(&ctr_table->mutex);
1242 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1243 				 list) {
1244 		if (counter->allocated)
1245 			mlx4_counter_free(ibdev->dev, counter->index);
1246 		list_del(&counter->list);
1247 		kfree(counter);
1248 	}
1249 	mutex_unlock(&ctr_table->mutex);
1250 }
1251 
1252 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1253 		   union ib_gid *gid)
1254 {
1255 	if_t ndev;
1256 	int ret = 0;
1257 
1258 	if (!mqp->port)
1259 		return 0;
1260 
1261 	spin_lock_bh(&mdev->iboe.lock);
1262 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1263 	if (ndev)
1264 		if_ref(ndev);
1265 	spin_unlock_bh(&mdev->iboe.lock);
1266 
1267 	if (ndev) {
1268 		ret = 1;
1269 		if_rele(ndev);
1270 	}
1271 
1272 	return ret;
1273 }
1274 
1275 struct mlx4_ib_steering {
1276 	struct list_head list;
1277 	struct mlx4_flow_reg_id reg_id;
1278 	union ib_gid gid;
1279 };
1280 
1281 #define LAST_ETH_FIELD vlan_tag
1282 #define LAST_IB_FIELD sl
1283 #define LAST_IPV4_FIELD dst_ip
1284 #define LAST_TCP_UDP_FIELD src_port
1285 
1286 /* Field is the last supported field */
1287 #define FIELDS_NOT_SUPPORTED(filter, field)\
1288 	memchr_inv((void *)&filter.field  +\
1289 		   sizeof(filter.field), 0,\
1290 		   sizeof(filter) -\
1291 		   offsetof(typeof(filter), field) -\
1292 		   sizeof(filter.field))
1293 
1294 static int parse_flow_attr(struct mlx4_dev *dev,
1295 			   u32 qp_num,
1296 			   union ib_flow_spec *ib_spec,
1297 			   struct _rule_hw *mlx4_spec)
1298 {
1299 	enum mlx4_net_trans_rule_id type;
1300 
1301 	switch (ib_spec->type) {
1302 	case IB_FLOW_SPEC_ETH:
1303 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1304 			return -ENOTSUPP;
1305 
1306 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1307 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1308 		       ETH_ALEN);
1309 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1310 		       ETH_ALEN);
1311 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1312 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1313 		break;
1314 	case IB_FLOW_SPEC_IB:
1315 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1316 			return -ENOTSUPP;
1317 
1318 		type = MLX4_NET_TRANS_RULE_ID_IB;
1319 		mlx4_spec->ib.l3_qpn =
1320 			cpu_to_be32(qp_num);
1321 		mlx4_spec->ib.qpn_mask =
1322 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1323 		break;
1324 
1325 
1326 	case IB_FLOW_SPEC_IPV4:
1327 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1328 			return -ENOTSUPP;
1329 
1330 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1331 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1332 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1333 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1334 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1335 		break;
1336 
1337 	case IB_FLOW_SPEC_TCP:
1338 	case IB_FLOW_SPEC_UDP:
1339 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1340 			return -ENOTSUPP;
1341 
1342 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1343 					MLX4_NET_TRANS_RULE_ID_TCP :
1344 					MLX4_NET_TRANS_RULE_ID_UDP;
1345 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1346 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1347 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1348 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1349 		break;
1350 
1351 	default:
1352 		return -EINVAL;
1353 	}
1354 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1355 	    mlx4_hw_rule_sz(dev, type) < 0)
1356 		return -EINVAL;
1357 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1358 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1359 	return mlx4_hw_rule_sz(dev, type);
1360 }
1361 
1362 struct default_rules {
1363 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1364 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1365 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1366 	__u8  link_layer;
1367 };
1368 static const struct default_rules default_table[] = {
1369 	{
1370 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1371 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1372 		.rules_create_list = {IB_FLOW_SPEC_IB},
1373 		.link_layer = IB_LINK_LAYER_INFINIBAND
1374 	}
1375 };
1376 
1377 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1378 					 struct ib_flow_attr *flow_attr)
1379 {
1380 	int i, j, k;
1381 	void *ib_flow;
1382 	const struct default_rules *pdefault_rules = default_table;
1383 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1384 
1385 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1386 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1387 		memset(&field_types, 0, sizeof(field_types));
1388 
1389 		if (link_layer != pdefault_rules->link_layer)
1390 			continue;
1391 
1392 		ib_flow = flow_attr + 1;
1393 		/* we assume the specs are sorted */
1394 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1395 		     j < flow_attr->num_of_specs; k++) {
1396 			union ib_flow_spec *current_flow =
1397 				(union ib_flow_spec *)ib_flow;
1398 
1399 			/* same layer but different type */
1400 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1401 			     (pdefault_rules->mandatory_fields[k] &
1402 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1403 			    (current_flow->type !=
1404 			     pdefault_rules->mandatory_fields[k]))
1405 				goto out;
1406 
1407 			/* same layer, try match next one */
1408 			if (current_flow->type ==
1409 			    pdefault_rules->mandatory_fields[k]) {
1410 				j++;
1411 				ib_flow +=
1412 					((union ib_flow_spec *)ib_flow)->size;
1413 			}
1414 		}
1415 
1416 		ib_flow = flow_attr + 1;
1417 		for (j = 0; j < flow_attr->num_of_specs;
1418 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1419 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1420 				/* same layer and same type */
1421 				if (((union ib_flow_spec *)ib_flow)->type ==
1422 				    pdefault_rules->mandatory_not_fields[k])
1423 					goto out;
1424 
1425 		return i;
1426 	}
1427 out:
1428 	return -1;
1429 }
1430 
1431 static int __mlx4_ib_create_default_rules(
1432 		struct mlx4_ib_dev *mdev,
1433 		struct ib_qp *qp,
1434 		const struct default_rules *pdefault_rules,
1435 		struct _rule_hw *mlx4_spec) {
1436 	int size = 0;
1437 	int i;
1438 
1439 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1440 		int ret;
1441 		union ib_flow_spec ib_spec;
1442 		switch (pdefault_rules->rules_create_list[i]) {
1443 		case 0:
1444 			/* no rule */
1445 			continue;
1446 		case IB_FLOW_SPEC_IB:
1447 			ib_spec.type = IB_FLOW_SPEC_IB;
1448 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1449 
1450 			break;
1451 		default:
1452 			/* invalid rule */
1453 			return -EINVAL;
1454 		}
1455 		/* We must put empty rule, qpn is being ignored */
1456 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1457 				      mlx4_spec);
1458 		if (ret < 0) {
1459 			pr_info("invalid parsing\n");
1460 			return -EINVAL;
1461 		}
1462 
1463 		mlx4_spec = (void *)mlx4_spec + ret;
1464 		size += ret;
1465 	}
1466 	return size;
1467 }
1468 
1469 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1470 			  int domain,
1471 			  enum mlx4_net_trans_promisc_mode flow_type,
1472 			  u64 *reg_id)
1473 {
1474 	int ret, i;
1475 	int size = 0;
1476 	void *ib_flow;
1477 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1478 	struct mlx4_cmd_mailbox *mailbox;
1479 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1480 	int default_flow;
1481 
1482 	static const u16 __mlx4_domain[] = {
1483 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1484 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1485 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1486 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1487 	};
1488 
1489 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1490 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1491 		return -EINVAL;
1492 	}
1493 
1494 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1495 		pr_err("Invalid domain value %d\n", domain);
1496 		return -EINVAL;
1497 	}
1498 
1499 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1500 		return -EINVAL;
1501 
1502 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1503 	if (IS_ERR(mailbox))
1504 		return PTR_ERR(mailbox);
1505 	ctrl = mailbox->buf;
1506 
1507 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1508 				 flow_attr->priority);
1509 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1510 	ctrl->port = flow_attr->port;
1511 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1512 
1513 	ib_flow = flow_attr + 1;
1514 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1515 	/* Add default flows */
1516 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1517 	if (default_flow >= 0) {
1518 		ret = __mlx4_ib_create_default_rules(
1519 				mdev, qp, default_table + default_flow,
1520 				mailbox->buf + size);
1521 		if (ret < 0) {
1522 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1523 			return -EINVAL;
1524 		}
1525 		size += ret;
1526 	}
1527 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1528 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1529 				      mailbox->buf + size);
1530 		if (ret < 0) {
1531 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1532 			return -EINVAL;
1533 		}
1534 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1535 		size += ret;
1536 	}
1537 
1538 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1539 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1540 			   MLX4_CMD_WRAPPED);
1541 	if (ret == -ENOMEM)
1542 		pr_err("mcg table is full. Fail to register network rule.\n");
1543 	else if (ret == -ENXIO)
1544 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1545 	else if (ret)
1546 		pr_err("Invalid argument. Fail to register network rule.\n");
1547 
1548 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1549 	return ret;
1550 }
1551 
1552 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1553 {
1554 	int err;
1555 	err = mlx4_cmd(dev, reg_id, 0, 0,
1556 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1557 		       MLX4_CMD_WRAPPED);
1558 	if (err)
1559 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1560 		       (long long)reg_id);
1561 	return err;
1562 }
1563 
1564 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1565 				    u64 *reg_id)
1566 {
1567 	void *ib_flow;
1568 	union ib_flow_spec *ib_spec;
1569 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1570 	int err = 0;
1571 
1572 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1573 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1574 		return 0; /* do nothing */
1575 
1576 	ib_flow = flow_attr + 1;
1577 	ib_spec = (union ib_flow_spec *)ib_flow;
1578 
1579 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1580 		return 0; /* do nothing */
1581 
1582 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1583 				    flow_attr->port, qp->qp_num,
1584 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1585 				    reg_id);
1586 	return err;
1587 }
1588 
1589 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1590 				      struct ib_flow_attr *flow_attr,
1591 				      enum mlx4_net_trans_promisc_mode *type)
1592 {
1593 	int err = 0;
1594 
1595 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1596 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1597 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1598 		return -EOPNOTSUPP;
1599 	}
1600 
1601 	if (flow_attr->num_of_specs == 0) {
1602 		type[0] = MLX4_FS_MC_SNIFFER;
1603 		type[1] = MLX4_FS_UC_SNIFFER;
1604 	} else {
1605 		union ib_flow_spec *ib_spec;
1606 
1607 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1608 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1609 			return -EINVAL;
1610 
1611 		/* if all is zero than MC and UC */
1612 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1613 			type[0] = MLX4_FS_MC_SNIFFER;
1614 			type[1] = MLX4_FS_UC_SNIFFER;
1615 		} else {
1616 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1617 					    ib_spec->eth.mask.dst_mac[1],
1618 					    ib_spec->eth.mask.dst_mac[2],
1619 					    ib_spec->eth.mask.dst_mac[3],
1620 					    ib_spec->eth.mask.dst_mac[4],
1621 					    ib_spec->eth.mask.dst_mac[5]};
1622 
1623 			/* Above xor was only on MC bit, non empty mask is valid
1624 			 * only if this bit is set and rest are zero.
1625 			 */
1626 			if (!is_zero_ether_addr(&mac[0]))
1627 				return -EINVAL;
1628 
1629 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1630 				type[0] = MLX4_FS_MC_SNIFFER;
1631 			else
1632 				type[0] = MLX4_FS_UC_SNIFFER;
1633 		}
1634 	}
1635 
1636 	return err;
1637 }
1638 
1639 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1640 				    struct ib_flow_attr *flow_attr,
1641 				    int domain, struct ib_udata *udata)
1642 {
1643 	int err = 0, i = 0, j = 0;
1644 	struct mlx4_ib_flow *mflow;
1645 	enum mlx4_net_trans_promisc_mode type[2];
1646 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1647 	int is_bonded = mlx4_is_bonded(dev);
1648 
1649 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1650 		return ERR_PTR(-EINVAL);
1651 
1652 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1653 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1654 		return ERR_PTR(-EOPNOTSUPP);
1655 
1656 	if (udata &&
1657 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1658 		return ERR_PTR(-EOPNOTSUPP);
1659 
1660 	memset(type, 0, sizeof(type));
1661 
1662 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1663 	if (!mflow) {
1664 		err = -ENOMEM;
1665 		goto err_free;
1666 	}
1667 
1668 	switch (flow_attr->type) {
1669 	case IB_FLOW_ATTR_NORMAL:
1670 		/* If dont trap flag (continue match) is set, under specific
1671 		 * condition traffic be replicated to given qp,
1672 		 * without stealing it
1673 		 */
1674 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1675 			err = mlx4_ib_add_dont_trap_rule(dev,
1676 							 flow_attr,
1677 							 type);
1678 			if (err)
1679 				goto err_free;
1680 		} else {
1681 			type[0] = MLX4_FS_REGULAR;
1682 		}
1683 		break;
1684 
1685 	case IB_FLOW_ATTR_ALL_DEFAULT:
1686 		type[0] = MLX4_FS_ALL_DEFAULT;
1687 		break;
1688 
1689 	case IB_FLOW_ATTR_MC_DEFAULT:
1690 		type[0] = MLX4_FS_MC_DEFAULT;
1691 		break;
1692 
1693 	case IB_FLOW_ATTR_SNIFFER:
1694 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1695 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1696 		break;
1697 
1698 	default:
1699 		err = -EINVAL;
1700 		goto err_free;
1701 	}
1702 
1703 	while (i < ARRAY_SIZE(type) && type[i]) {
1704 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1705 					    &mflow->reg_id[i].id);
1706 		if (err)
1707 			goto err_create_flow;
1708 		if (is_bonded) {
1709 			/* Application always sees one port so the mirror rule
1710 			 * must be on port #2
1711 			 */
1712 			flow_attr->port = 2;
1713 			err = __mlx4_ib_create_flow(qp, flow_attr,
1714 						    domain, type[j],
1715 						    &mflow->reg_id[j].mirror);
1716 			flow_attr->port = 1;
1717 			if (err)
1718 				goto err_create_flow;
1719 			j++;
1720 		}
1721 
1722 		i++;
1723 	}
1724 
1725 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1726 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1727 					       &mflow->reg_id[i].id);
1728 		if (err)
1729 			goto err_create_flow;
1730 
1731 		if (is_bonded) {
1732 			flow_attr->port = 2;
1733 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1734 						       &mflow->reg_id[j].mirror);
1735 			flow_attr->port = 1;
1736 			if (err)
1737 				goto err_create_flow;
1738 			j++;
1739 		}
1740 		/* function to create mirror rule */
1741 		i++;
1742 	}
1743 
1744 	return &mflow->ibflow;
1745 
1746 err_create_flow:
1747 	while (i) {
1748 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1749 					     mflow->reg_id[i].id);
1750 		i--;
1751 	}
1752 
1753 	while (j) {
1754 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1755 					     mflow->reg_id[j].mirror);
1756 		j--;
1757 	}
1758 err_free:
1759 	kfree(mflow);
1760 	return ERR_PTR(err);
1761 }
1762 
1763 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1764 {
1765 	int err, ret = 0;
1766 	int i = 0;
1767 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1768 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1769 
1770 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1771 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1772 		if (err)
1773 			ret = err;
1774 		if (mflow->reg_id[i].mirror) {
1775 			err = __mlx4_ib_destroy_flow(mdev->dev,
1776 						     mflow->reg_id[i].mirror);
1777 			if (err)
1778 				ret = err;
1779 		}
1780 		i++;
1781 	}
1782 
1783 	kfree(mflow);
1784 	return ret;
1785 }
1786 
1787 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1788 {
1789 	int err;
1790 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1791 	struct mlx4_dev	*dev = mdev->dev;
1792 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1793 	struct mlx4_ib_steering *ib_steering = NULL;
1794 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1795 	struct mlx4_flow_reg_id	reg_id;
1796 
1797 	if (mdev->dev->caps.steering_mode ==
1798 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1799 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1800 		if (!ib_steering)
1801 			return -ENOMEM;
1802 	}
1803 
1804 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1805 				    !!(mqp->flags &
1806 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1807 				    prot, &reg_id.id);
1808 	if (err) {
1809 		pr_err("multicast attach op failed, err %d\n", err);
1810 		goto err_malloc;
1811 	}
1812 
1813 	reg_id.mirror = 0;
1814 	if (mlx4_is_bonded(dev)) {
1815 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1816 					    (mqp->port == 1) ? 2 : 1,
1817 					    !!(mqp->flags &
1818 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1819 					    prot, &reg_id.mirror);
1820 		if (err)
1821 			goto err_add;
1822 	}
1823 
1824 	err = add_gid_entry(ibqp, gid);
1825 	if (err)
1826 		goto err_add;
1827 
1828 	if (ib_steering) {
1829 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1830 		ib_steering->reg_id = reg_id;
1831 		mutex_lock(&mqp->mutex);
1832 		list_add(&ib_steering->list, &mqp->steering_rules);
1833 		mutex_unlock(&mqp->mutex);
1834 	}
1835 	return 0;
1836 
1837 err_add:
1838 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1839 			      prot, reg_id.id);
1840 	if (reg_id.mirror)
1841 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1842 				      prot, reg_id.mirror);
1843 err_malloc:
1844 	kfree(ib_steering);
1845 
1846 	return err;
1847 }
1848 
1849 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1850 {
1851 	struct mlx4_ib_gid_entry *ge;
1852 	struct mlx4_ib_gid_entry *tmp;
1853 	struct mlx4_ib_gid_entry *ret = NULL;
1854 
1855 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1856 		if (!memcmp(raw, ge->gid.raw, 16)) {
1857 			ret = ge;
1858 			break;
1859 		}
1860 	}
1861 
1862 	return ret;
1863 }
1864 
1865 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1866 {
1867 	int err;
1868 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1869 	struct mlx4_dev *dev = mdev->dev;
1870 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1871 	if_t ndev;
1872 	struct mlx4_ib_gid_entry *ge;
1873 	struct mlx4_flow_reg_id reg_id = {0, 0};
1874 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1875 
1876 	if (mdev->dev->caps.steering_mode ==
1877 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1878 		struct mlx4_ib_steering *ib_steering;
1879 
1880 		mutex_lock(&mqp->mutex);
1881 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1882 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1883 				list_del(&ib_steering->list);
1884 				break;
1885 			}
1886 		}
1887 		mutex_unlock(&mqp->mutex);
1888 		if (&ib_steering->list == &mqp->steering_rules) {
1889 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1890 			return -EINVAL;
1891 		}
1892 		reg_id = ib_steering->reg_id;
1893 		kfree(ib_steering);
1894 	}
1895 
1896 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1897 				    prot, reg_id.id);
1898 	if (err)
1899 		return err;
1900 
1901 	if (mlx4_is_bonded(dev)) {
1902 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1903 					    prot, reg_id.mirror);
1904 		if (err)
1905 			return err;
1906 	}
1907 
1908 	mutex_lock(&mqp->mutex);
1909 	ge = find_gid_entry(mqp, gid->raw);
1910 	if (ge) {
1911 		spin_lock_bh(&mdev->iboe.lock);
1912 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1913 		if (ndev)
1914 			if_ref(ndev);
1915 		spin_unlock_bh(&mdev->iboe.lock);
1916 		if (ndev)
1917 			if_rele(ndev);
1918 		list_del(&ge->list);
1919 		kfree(ge);
1920 	} else
1921 		pr_warn("could not find mgid entry\n");
1922 
1923 	mutex_unlock(&mqp->mutex);
1924 
1925 	return 0;
1926 }
1927 
1928 static int init_node_data(struct mlx4_ib_dev *dev)
1929 {
1930 	struct ib_smp *in_mad  = NULL;
1931 	struct ib_smp *out_mad = NULL;
1932 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1933 	int err = -ENOMEM;
1934 
1935 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1936 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1937 	if (!in_mad || !out_mad)
1938 		goto out;
1939 
1940 	init_query_mad(in_mad);
1941 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1942 	if (mlx4_is_master(dev->dev))
1943 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1944 
1945 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1946 	if (err)
1947 		goto out;
1948 
1949 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1950 
1951 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1952 
1953 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1954 	if (err)
1955 		goto out;
1956 
1957 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1958 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1959 
1960 out:
1961 	kfree(in_mad);
1962 	kfree(out_mad);
1963 	return err;
1964 }
1965 
1966 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1967 			char *buf)
1968 {
1969 	struct mlx4_ib_dev *dev =
1970 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1971 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1972 }
1973 
1974 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1975 			char *buf)
1976 {
1977 	struct mlx4_ib_dev *dev =
1978 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1979 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1980 }
1981 
1982 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1983 			  char *buf)
1984 {
1985 	struct mlx4_ib_dev *dev =
1986 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1987 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1988 		       dev->dev->board_id);
1989 }
1990 
1991 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1992 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1993 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1994 
1995 static struct device_attribute *mlx4_class_attributes[] = {
1996 	&dev_attr_hw_rev,
1997 	&dev_attr_hca_type,
1998 	&dev_attr_board_id
1999 };
2000 
2001 struct diag_counter {
2002 	const char *name;
2003 	u32 offset;
2004 };
2005 
2006 #define DIAG_COUNTER(_name, _offset)			\
2007 	{ .name = #_name, .offset = _offset }
2008 
2009 static const struct diag_counter diag_basic[] = {
2010 	DIAG_COUNTER(rq_num_lle, 0x00),
2011 	DIAG_COUNTER(sq_num_lle, 0x04),
2012 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2013 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2014 	DIAG_COUNTER(rq_num_lpe, 0x18),
2015 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2016 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2017 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2018 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2019 	DIAG_COUNTER(sq_num_bre, 0x34),
2020 	DIAG_COUNTER(sq_num_rire, 0x44),
2021 	DIAG_COUNTER(rq_num_rire, 0x48),
2022 	DIAG_COUNTER(sq_num_rae, 0x4C),
2023 	DIAG_COUNTER(rq_num_rae, 0x50),
2024 	DIAG_COUNTER(sq_num_roe, 0x54),
2025 	DIAG_COUNTER(sq_num_tree, 0x5C),
2026 	DIAG_COUNTER(sq_num_rree, 0x64),
2027 	DIAG_COUNTER(rq_num_rnr, 0x68),
2028 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2029 	DIAG_COUNTER(rq_num_oos, 0x100),
2030 	DIAG_COUNTER(sq_num_oos, 0x104),
2031 };
2032 
2033 static const struct diag_counter diag_ext[] = {
2034 	DIAG_COUNTER(rq_num_dup, 0x130),
2035 	DIAG_COUNTER(sq_num_to, 0x134),
2036 };
2037 
2038 static const struct diag_counter diag_device_only[] = {
2039 	DIAG_COUNTER(num_cqovf, 0x1A0),
2040 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2041 };
2042 
2043 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2044 						    u8 port_num)
2045 {
2046 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2047 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2048 
2049 	if (!diag[!!port_num].name)
2050 		return NULL;
2051 
2052 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2053 					  diag[!!port_num].num_counters,
2054 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2055 }
2056 
2057 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2058 				struct rdma_hw_stats *stats,
2059 				u8 port, int index)
2060 {
2061 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2062 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2063 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2064 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2065 	int ret;
2066 	int i;
2067 
2068 	ret = mlx4_query_diag_counters(dev->dev,
2069 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2070 				       diag[!!port].offset, hw_value,
2071 				       diag[!!port].num_counters, port);
2072 
2073 	if (ret)
2074 		return ret;
2075 
2076 	for (i = 0; i < diag[!!port].num_counters; i++)
2077 		stats->value[i] = hw_value[i];
2078 
2079 	return diag[!!port].num_counters;
2080 }
2081 
2082 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2083 					 const char ***name,
2084 					 u32 **offset,
2085 					 u32 *num,
2086 					 bool port)
2087 {
2088 	u32 num_counters;
2089 
2090 	num_counters = ARRAY_SIZE(diag_basic);
2091 
2092 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2093 		num_counters += ARRAY_SIZE(diag_ext);
2094 
2095 	if (!port)
2096 		num_counters += ARRAY_SIZE(diag_device_only);
2097 
2098 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2099 	if (!*name)
2100 		return -ENOMEM;
2101 
2102 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2103 	if (!*offset)
2104 		goto err_name;
2105 
2106 	*num = num_counters;
2107 
2108 	return 0;
2109 
2110 err_name:
2111 	kfree(*name);
2112 	return -ENOMEM;
2113 }
2114 
2115 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2116 				       const char **name,
2117 				       u32 *offset,
2118 				       bool port)
2119 {
2120 	int i;
2121 	int j;
2122 
2123 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2124 		name[i] = diag_basic[i].name;
2125 		offset[i] = diag_basic[i].offset;
2126 	}
2127 
2128 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2129 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2130 			name[j] = diag_ext[i].name;
2131 			offset[j] = diag_ext[i].offset;
2132 		}
2133 	}
2134 
2135 	if (!port) {
2136 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2137 			name[j] = diag_device_only[i].name;
2138 			offset[j] = diag_device_only[i].offset;
2139 		}
2140 	}
2141 }
2142 
2143 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2144 {
2145 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2146 	int i;
2147 	int ret;
2148 	bool per_port = !!(ibdev->dev->caps.flags2 &
2149 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2150 
2151 	if (mlx4_is_slave(ibdev->dev))
2152 		return 0;
2153 
2154 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2155 		/* i == 1 means we are building port counters */
2156 		if (i && !per_port)
2157 			continue;
2158 
2159 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2160 						    &diag[i].offset,
2161 						    &diag[i].num_counters, i);
2162 		if (ret)
2163 			goto err_alloc;
2164 
2165 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2166 					   diag[i].offset, i);
2167 	}
2168 
2169 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2170 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2171 
2172 	return 0;
2173 
2174 err_alloc:
2175 	if (i) {
2176 		kfree(diag[i - 1].name);
2177 		kfree(diag[i - 1].offset);
2178 	}
2179 
2180 	return ret;
2181 }
2182 
2183 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2184 {
2185 	int i;
2186 
2187 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2188 		kfree(ibdev->diag_counters[i].offset);
2189 		kfree(ibdev->diag_counters[i].name);
2190 	}
2191 }
2192 
2193 #define MLX4_IB_INVALID_MAC	((u64)-1)
2194 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2195 			       if_t dev,
2196 			       int port)
2197 {
2198 	u64 new_smac = 0;
2199 	u64 release_mac = MLX4_IB_INVALID_MAC;
2200 	struct mlx4_ib_qp *qp;
2201 
2202 	new_smac = mlx4_mac_to_u64(if_getlladdr(dev));
2203 
2204 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2205 
2206 	/* no need for update QP1 and mac registration in non-SRIOV */
2207 	if (!mlx4_is_mfunc(ibdev->dev))
2208 		return;
2209 
2210 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2211 	qp = ibdev->qp1_proxy[port - 1];
2212 	if (qp) {
2213 		int new_smac_index;
2214 		u64 old_smac;
2215 		struct mlx4_update_qp_params update_params;
2216 
2217 		mutex_lock(&qp->mutex);
2218 		old_smac = qp->pri.smac;
2219 		if (new_smac == old_smac)
2220 			goto unlock;
2221 
2222 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2223 
2224 		if (new_smac_index < 0)
2225 			goto unlock;
2226 
2227 		update_params.smac_index = new_smac_index;
2228 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2229 				   &update_params)) {
2230 			release_mac = new_smac;
2231 			goto unlock;
2232 		}
2233 		/* if old port was zero, no mac was yet registered for this QP */
2234 		if (qp->pri.smac_port)
2235 			release_mac = old_smac;
2236 		qp->pri.smac = new_smac;
2237 		qp->pri.smac_port = port;
2238 		qp->pri.smac_index = new_smac_index;
2239 	}
2240 
2241 unlock:
2242 	if (release_mac != MLX4_IB_INVALID_MAC)
2243 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2244 	if (qp)
2245 		mutex_unlock(&qp->mutex);
2246 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2247 }
2248 
2249 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2250 				 if_t dev,
2251 				 unsigned long event)
2252 
2253 {
2254 	struct mlx4_ib_iboe *iboe;
2255 	int update_qps_port = -1;
2256 	int port;
2257 
2258 	iboe = &ibdev->iboe;
2259 
2260 	spin_lock_bh(&iboe->lock);
2261 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2262 
2263 		iboe->netdevs[port - 1] =
2264 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2265 
2266 		if (dev == iboe->netdevs[port - 1] &&
2267 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2268 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2269 			update_qps_port = port;
2270 
2271 	}
2272 	spin_unlock_bh(&iboe->lock);
2273 
2274 	if (update_qps_port > 0)
2275 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2276 }
2277 
2278 static int mlx4_ib_netdev_event(struct notifier_block *this,
2279 				unsigned long event, void *ptr)
2280 {
2281 	if_t dev = netdev_notifier_info_to_ifp(ptr);
2282 	struct mlx4_ib_dev *ibdev;
2283 
2284 	if (if_getvnet(dev) != &init_net)
2285 		return NOTIFY_DONE;
2286 
2287 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2288 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2289 
2290 	return NOTIFY_DONE;
2291 }
2292 
2293 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2294 {
2295 	int port;
2296 	int slave;
2297 	int i;
2298 
2299 	if (mlx4_is_master(ibdev->dev)) {
2300 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2301 		     ++slave) {
2302 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2303 				for (i = 0;
2304 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2305 				     ++i) {
2306 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2307 					/* master has the identity virt2phys pkey mapping */
2308 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2309 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2310 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2311 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2312 				}
2313 			}
2314 		}
2315 		/* initialize pkey cache */
2316 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2317 			for (i = 0;
2318 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2319 			     ++i)
2320 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2321 					(i) ? 0 : 0xFFFF;
2322 		}
2323 	}
2324 }
2325 
2326 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2327 {
2328 	int i, j, eq = 0, total_eqs = 0;
2329 
2330 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2331 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2332 	if (!ibdev->eq_table)
2333 		return;
2334 
2335 	for (i = 1; i <= dev->caps.num_ports; i++) {
2336 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2337 		     j++, total_eqs++) {
2338 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2339 				continue;
2340 			ibdev->eq_table[eq] = total_eqs;
2341 			if (!mlx4_assign_eq(dev, i,
2342 					    &ibdev->eq_table[eq]))
2343 				eq++;
2344 			else
2345 				ibdev->eq_table[eq] = -1;
2346 		}
2347 	}
2348 
2349 	for (i = eq; i < dev->caps.num_comp_vectors;
2350 	     ibdev->eq_table[i++] = -1)
2351 		;
2352 
2353 	/* Advertise the new number of EQs to clients */
2354 	ibdev->ib_dev.num_comp_vectors = eq;
2355 }
2356 
2357 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2358 {
2359 	int i;
2360 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2361 
2362 	/* no eqs were allocated */
2363 	if (!ibdev->eq_table)
2364 		return;
2365 
2366 	/* Reset the advertised EQ number */
2367 	ibdev->ib_dev.num_comp_vectors = 0;
2368 
2369 	for (i = 0; i < total_eqs; i++)
2370 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2371 
2372 	kfree(ibdev->eq_table);
2373 	ibdev->eq_table = NULL;
2374 }
2375 
2376 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2377 			       struct ib_port_immutable *immutable)
2378 {
2379 	struct ib_port_attr attr;
2380 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2381 	int err;
2382 
2383 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2384 	if (err)
2385 		return err;
2386 
2387 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2388 	immutable->gid_tbl_len = attr.gid_tbl_len;
2389 
2390 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2391 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2392 	} else {
2393 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2394 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2395 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2396 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2397 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2398 	}
2399 
2400 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2401 
2402 	return 0;
2403 }
2404 
2405 static void get_fw_ver_str(struct ib_device *device, char *str,
2406 			   size_t str_len)
2407 {
2408 	struct mlx4_ib_dev *dev =
2409 		container_of(device, struct mlx4_ib_dev, ib_dev);
2410 	snprintf(str, str_len, "%d.%d.%d",
2411 		 (int) (dev->dev->caps.fw_ver >> 32),
2412 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2413 		 (int) dev->dev->caps.fw_ver & 0xffff);
2414 }
2415 
2416 static void *mlx4_ib_add(struct mlx4_dev *dev)
2417 {
2418 	struct mlx4_ib_dev *ibdev;
2419 	int num_ports;
2420 	int i, j;
2421 	int err;
2422 	struct mlx4_ib_iboe *iboe;
2423 	int ib_num_ports = 0;
2424 	int num_req_counters;
2425 	int allocated;
2426 	u32 counter_index;
2427 	struct counter_index *new_counter_index = NULL;
2428 
2429 	pr_info_once("%s", mlx4_ib_version);
2430 
2431 	num_ports = 0;
2432 	mlx4_foreach_ib_transport_port(i, dev)
2433 		num_ports++;
2434 
2435 	/* No point in registering a device with no ports... */
2436 	if (num_ports == 0)
2437 		return NULL;
2438 
2439 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2440 	if (!ibdev) {
2441 		dev_err(&dev->persist->pdev->dev,
2442 			"Device struct alloc failed\n");
2443 		return NULL;
2444 	}
2445 
2446 	iboe = &ibdev->iboe;
2447 
2448 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2449 		goto err_dealloc;
2450 
2451 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2452 		goto err_pd;
2453 
2454 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2455 				 PAGE_SIZE);
2456 	if (!ibdev->uar_map)
2457 		goto err_uar;
2458 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2459 
2460 	ibdev->dev = dev;
2461 	ibdev->bond_next_port	= 0;
2462 
2463 	INIT_IB_DEVICE_OPS(&ibdev->ib_dev.ops, mlx4, MLX4);
2464 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2465 	ibdev->ib_dev.owner		= THIS_MODULE;
2466 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2467 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2468 	ibdev->num_ports		= num_ports;
2469 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2470 						1 : ibdev->num_ports;
2471 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2472 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2473 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2474 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2475 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2476 
2477 	if (dev->caps.userspace_caps)
2478 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2479 	else
2480 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2481 
2482 	ibdev->ib_dev.uverbs_cmd_mask	=
2483 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2484 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2485 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2486 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2487 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2488 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2489 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2490 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2491 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2492 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2493 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2494 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2495 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2496 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2497 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2498 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2499 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2500 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2501 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2502 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2503 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2504 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2505 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2506 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2507 
2508 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2509 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2510 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2511 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2512 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2513 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2514 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2515 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2516 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2517 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2518 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2519 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2520 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2521 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2522 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2523 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2524 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2525 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2526 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2527 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2528 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2529 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2530 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2531 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2532 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2533 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2534 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2535 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2536 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2537 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2538 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2539 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2540 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2541 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2542 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2543 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2544 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2545 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2546 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2547 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2548 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2549 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2550 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2551 
2552 	if (!mlx4_is_slave(ibdev->dev)) {
2553 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2554 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2555 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2556 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2557 	}
2558 
2559 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2560 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2561 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2562 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2563 
2564 		ibdev->ib_dev.uverbs_cmd_mask |=
2565 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2566 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2567 	}
2568 
2569 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2570 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2571 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2572 		ibdev->ib_dev.uverbs_cmd_mask |=
2573 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2574 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2575 	}
2576 
2577 	if (check_flow_steering_support(dev)) {
2578 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2579 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2580 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2581 
2582 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2583 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2584 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2585 	}
2586 
2587 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2588 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2589 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2590 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2591 
2592 	mlx4_ib_alloc_eqs(dev, ibdev);
2593 
2594 	spin_lock_init(&iboe->lock);
2595 
2596 	if (init_node_data(ibdev))
2597 		goto err_map;
2598 	mlx4_init_sl2vl_tbl(ibdev);
2599 
2600 	for (i = 0; i < ibdev->num_ports; ++i) {
2601 		mutex_init(&ibdev->counters_table[i].mutex);
2602 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2603 	}
2604 
2605 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2606 	for (i = 0; i < num_req_counters; ++i) {
2607 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2608 		allocated = 0;
2609 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2610 						IB_LINK_LAYER_ETHERNET) {
2611 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2612 			/* if failed to allocate a new counter, use default */
2613 			if (err)
2614 				counter_index =
2615 					mlx4_get_default_counter_index(dev,
2616 								       i + 1);
2617 			else
2618 				allocated = 1;
2619 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2620 			counter_index = mlx4_get_default_counter_index(dev,
2621 								       i + 1);
2622 		}
2623 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2624 					    GFP_KERNEL);
2625 		if (!new_counter_index) {
2626 			if (allocated)
2627 				mlx4_counter_free(ibdev->dev, counter_index);
2628 			goto err_counter;
2629 		}
2630 		new_counter_index->index = counter_index;
2631 		new_counter_index->allocated = allocated;
2632 		list_add_tail(&new_counter_index->list,
2633 			      &ibdev->counters_table[i].counters_list);
2634 		ibdev->counters_table[i].default_counter = counter_index;
2635 		pr_info("counter index %d for port %d allocated %d\n",
2636 			counter_index, i + 1, allocated);
2637 	}
2638 	if (mlx4_is_bonded(dev))
2639 		for (i = 1; i < ibdev->num_ports ; ++i) {
2640 			new_counter_index =
2641 					kmalloc(sizeof(struct counter_index),
2642 						GFP_KERNEL);
2643 			if (!new_counter_index)
2644 				goto err_counter;
2645 			new_counter_index->index = counter_index;
2646 			new_counter_index->allocated = 0;
2647 			list_add_tail(&new_counter_index->list,
2648 				      &ibdev->counters_table[i].counters_list);
2649 			ibdev->counters_table[i].default_counter =
2650 								counter_index;
2651 		}
2652 
2653 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2654 		ib_num_ports++;
2655 
2656 	spin_lock_init(&ibdev->sm_lock);
2657 	mutex_init(&ibdev->cap_mask_mutex);
2658 	INIT_LIST_HEAD(&ibdev->qp_list);
2659 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2660 
2661 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2662 	    ib_num_ports) {
2663 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2664 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2665 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2666 					    &ibdev->steer_qpn_base, 0);
2667 		if (err)
2668 			goto err_counter;
2669 
2670 		ibdev->ib_uc_qpns_bitmap =
2671 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2672 				sizeof(long),
2673 				GFP_KERNEL);
2674 		if (!ibdev->ib_uc_qpns_bitmap) {
2675 			dev_err(&dev->persist->pdev->dev,
2676 				"bit map alloc failed\n");
2677 			goto err_steer_qp_release;
2678 		}
2679 
2680 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2681 
2682 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2683 				dev, ibdev->steer_qpn_base,
2684 				ibdev->steer_qpn_base +
2685 				ibdev->steer_qpn_count - 1);
2686 		if (err)
2687 			goto err_steer_free_bitmap;
2688 	}
2689 
2690 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2691 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2692 
2693 	if (mlx4_ib_alloc_diag_counters(ibdev))
2694 		goto err_steer_free_bitmap;
2695 
2696 	if (ib_register_device(&ibdev->ib_dev, NULL))
2697 		goto err_diag_counters;
2698 
2699 	if (mlx4_ib_mad_init(ibdev))
2700 		goto err_reg;
2701 
2702 	if (mlx4_ib_init_sriov(ibdev))
2703 		goto err_mad;
2704 
2705 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2706 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2707 		if (!iboe->nb.notifier_call) {
2708 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2709 			err = register_netdevice_notifier(&iboe->nb);
2710 			if (err) {
2711 				iboe->nb.notifier_call = NULL;
2712 				goto err_notif;
2713 			}
2714 		}
2715 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2716 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2717 			if (err) {
2718 				goto err_notif;
2719 			}
2720 		}
2721 	}
2722 
2723 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2724 		if (device_create_file(&ibdev->ib_dev.dev,
2725 				       mlx4_class_attributes[j]))
2726 			goto err_notif;
2727 	}
2728 
2729 	ibdev->ib_active = true;
2730 
2731 	if (mlx4_is_mfunc(ibdev->dev))
2732 		init_pkeys(ibdev);
2733 
2734 	/* create paravirt contexts for any VFs which are active */
2735 	if (mlx4_is_master(ibdev->dev)) {
2736 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2737 			if (j == mlx4_master_func_num(ibdev->dev))
2738 				continue;
2739 			if (mlx4_is_slave_active(ibdev->dev, j))
2740 				do_slave_init(ibdev, j, 1);
2741 		}
2742 	}
2743 	return ibdev;
2744 
2745 err_notif:
2746 	if (ibdev->iboe.nb.notifier_call) {
2747 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2748 			pr_warn("failure unregistering notifier\n");
2749 		ibdev->iboe.nb.notifier_call = NULL;
2750 	}
2751 	flush_workqueue(wq);
2752 
2753 	mlx4_ib_close_sriov(ibdev);
2754 
2755 err_mad:
2756 	mlx4_ib_mad_cleanup(ibdev);
2757 
2758 err_reg:
2759 	ib_unregister_device(&ibdev->ib_dev);
2760 
2761 err_diag_counters:
2762 	mlx4_ib_diag_cleanup(ibdev);
2763 
2764 err_steer_free_bitmap:
2765 	kfree(ibdev->ib_uc_qpns_bitmap);
2766 
2767 err_steer_qp_release:
2768 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2769 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2770 				      ibdev->steer_qpn_count);
2771 err_counter:
2772 	for (i = 0; i < ibdev->num_ports; ++i)
2773 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2774 
2775 err_map:
2776 	iounmap(ibdev->uar_map);
2777 
2778 err_uar:
2779 	mlx4_uar_free(dev, &ibdev->priv_uar);
2780 
2781 err_pd:
2782 	mlx4_pd_free(dev, ibdev->priv_pdn);
2783 
2784 err_dealloc:
2785 	ib_dealloc_device(&ibdev->ib_dev);
2786 
2787 	return NULL;
2788 }
2789 
2790 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2791 {
2792 	int offset;
2793 
2794 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2795 
2796 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2797 					 dev->steer_qpn_count,
2798 					 get_count_order(count));
2799 	if (offset < 0)
2800 		return offset;
2801 
2802 	*qpn = dev->steer_qpn_base + offset;
2803 	return 0;
2804 }
2805 
2806 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2807 {
2808 	if (!qpn ||
2809 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2810 		return;
2811 
2812 	BUG_ON(qpn < dev->steer_qpn_base);
2813 
2814 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2815 			      qpn - dev->steer_qpn_base,
2816 			      get_count_order(count));
2817 }
2818 
2819 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2820 			 int is_attach)
2821 {
2822 	int err;
2823 	size_t flow_size;
2824 	struct ib_flow_attr *flow = NULL;
2825 	struct ib_flow_spec_ib *ib_spec;
2826 
2827 	if (is_attach) {
2828 		flow_size = sizeof(struct ib_flow_attr) +
2829 			    sizeof(struct ib_flow_spec_ib);
2830 		flow = kzalloc(flow_size, GFP_KERNEL);
2831 		if (!flow)
2832 			return -ENOMEM;
2833 		flow->port = mqp->port;
2834 		flow->num_of_specs = 1;
2835 		flow->size = flow_size;
2836 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2837 		ib_spec->type = IB_FLOW_SPEC_IB;
2838 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2839 		/* Add an empty rule for IB L2 */
2840 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2841 
2842 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2843 					    IB_FLOW_DOMAIN_NIC,
2844 					    MLX4_FS_REGULAR,
2845 					    &mqp->reg_id);
2846 	} else {
2847 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2848 	}
2849 	kfree(flow);
2850 	return err;
2851 }
2852 
2853 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2854 {
2855 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2856 	int p;
2857 
2858 	ibdev->ib_active = false;
2859 	flush_workqueue(wq);
2860 
2861 	mlx4_ib_close_sriov(ibdev);
2862 	mlx4_ib_mad_cleanup(ibdev);
2863 	ib_unregister_device(&ibdev->ib_dev);
2864 	mlx4_ib_diag_cleanup(ibdev);
2865 	if (ibdev->iboe.nb.notifier_call) {
2866 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2867 			pr_warn("failure unregistering notifier\n");
2868 		ibdev->iboe.nb.notifier_call = NULL;
2869 	}
2870 
2871 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2872 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2873 				      ibdev->steer_qpn_count);
2874 		kfree(ibdev->ib_uc_qpns_bitmap);
2875 	}
2876 
2877 	iounmap(ibdev->uar_map);
2878 	for (p = 0; p < ibdev->num_ports; ++p)
2879 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2880 
2881 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2882 		mlx4_CLOSE_PORT(dev, p);
2883 
2884 	mlx4_ib_free_eqs(dev, ibdev);
2885 
2886 	mlx4_uar_free(dev, &ibdev->priv_uar);
2887 	mlx4_pd_free(dev, ibdev->priv_pdn);
2888 	ib_dealloc_device(&ibdev->ib_dev);
2889 }
2890 
2891 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2892 {
2893 	struct mlx4_ib_demux_work **dm = NULL;
2894 	struct mlx4_dev *dev = ibdev->dev;
2895 	int i;
2896 	unsigned long flags;
2897 	struct mlx4_active_ports actv_ports;
2898 	unsigned int ports;
2899 	unsigned int first_port;
2900 
2901 	if (!mlx4_is_master(dev))
2902 		return;
2903 
2904 	actv_ports = mlx4_get_active_ports(dev, slave);
2905 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2906 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2907 
2908 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2909 	if (!dm) {
2910 		pr_err("failed to allocate memory for tunneling qp update\n");
2911 		return;
2912 	}
2913 
2914 	for (i = 0; i < ports; i++) {
2915 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2916 		if (!dm[i]) {
2917 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2918 			while (--i >= 0)
2919 				kfree(dm[i]);
2920 			goto out;
2921 		}
2922 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2923 		dm[i]->port = first_port + i + 1;
2924 		dm[i]->slave = slave;
2925 		dm[i]->do_init = do_init;
2926 		dm[i]->dev = ibdev;
2927 	}
2928 	/* initialize or tear down tunnel QPs for the slave */
2929 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2930 	if (!ibdev->sriov.is_going_down) {
2931 		for (i = 0; i < ports; i++)
2932 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2933 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2934 	} else {
2935 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2936 		for (i = 0; i < ports; i++)
2937 			kfree(dm[i]);
2938 	}
2939 out:
2940 	kfree(dm);
2941 	return;
2942 }
2943 
2944 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2945 {
2946 	struct mlx4_ib_qp *mqp;
2947 	unsigned long flags_qp;
2948 	unsigned long flags_cq;
2949 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2950 	struct list_head    cq_notify_list;
2951 	struct mlx4_cq *mcq;
2952 	unsigned long flags;
2953 
2954 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2955 	INIT_LIST_HEAD(&cq_notify_list);
2956 
2957 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2958 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2959 
2960 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2961 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2962 		if (mqp->sq.tail != mqp->sq.head) {
2963 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2964 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2965 			if (send_mcq->mcq.comp &&
2966 			    mqp->ibqp.send_cq->comp_handler) {
2967 				if (!send_mcq->mcq.reset_notify_added) {
2968 					send_mcq->mcq.reset_notify_added = 1;
2969 					list_add_tail(&send_mcq->mcq.reset_notify,
2970 						      &cq_notify_list);
2971 				}
2972 			}
2973 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2974 		}
2975 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2976 		/* Now, handle the QP's receive queue */
2977 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2978 		/* no handling is needed for SRQ */
2979 		if (!mqp->ibqp.srq) {
2980 			if (mqp->rq.tail != mqp->rq.head) {
2981 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2982 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2983 				if (recv_mcq->mcq.comp &&
2984 				    mqp->ibqp.recv_cq->comp_handler) {
2985 					if (!recv_mcq->mcq.reset_notify_added) {
2986 						recv_mcq->mcq.reset_notify_added = 1;
2987 						list_add_tail(&recv_mcq->mcq.reset_notify,
2988 							      &cq_notify_list);
2989 					}
2990 				}
2991 				spin_unlock_irqrestore(&recv_mcq->lock,
2992 						       flags_cq);
2993 			}
2994 		}
2995 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2996 	}
2997 
2998 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2999 		mcq->comp(mcq);
3000 	}
3001 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3002 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3003 }
3004 
3005 static void handle_bonded_port_state_event(struct work_struct *work)
3006 {
3007 	struct ib_event_work *ew =
3008 		container_of(work, struct ib_event_work, work);
3009 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3010 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3011 	int i;
3012 	struct ib_event ibev;
3013 
3014 	kfree(ew);
3015 	spin_lock_bh(&ibdev->iboe.lock);
3016 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3017 		if_t curr_netdev = ibdev->iboe.netdevs[i];
3018 		enum ib_port_state curr_port_state;
3019 
3020 		if (!curr_netdev)
3021 			continue;
3022 
3023 		curr_port_state =
3024 			((if_getdrvflags(curr_netdev) & IFF_DRV_RUNNING) != 0 &&
3025 			 if_getlinkstate(curr_netdev) == LINK_STATE_UP) ?
3026 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3027 
3028 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3029 			curr_port_state : IB_PORT_ACTIVE;
3030 	}
3031 	spin_unlock_bh(&ibdev->iboe.lock);
3032 
3033 	ibev.device = &ibdev->ib_dev;
3034 	ibev.element.port_num = 1;
3035 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3036 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3037 
3038 	ib_dispatch_event(&ibev);
3039 }
3040 
3041 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3042 {
3043 	u64 sl2vl;
3044 	int err;
3045 
3046 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3047 	if (err) {
3048 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3049 		       port, err);
3050 		sl2vl = 0;
3051 	}
3052 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3053 }
3054 
3055 static void ib_sl2vl_update_work(struct work_struct *work)
3056 {
3057 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3058 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3059 	int port = ew->port;
3060 
3061 	mlx4_ib_sl2vl_update(mdev, port);
3062 
3063 	kfree(ew);
3064 }
3065 
3066 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3067 				     int port)
3068 {
3069 	struct ib_event_work *ew;
3070 
3071 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3072 	if (ew) {
3073 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3074 		ew->port = port;
3075 		ew->ib_dev = ibdev;
3076 		queue_work(wq, &ew->work);
3077 	} else {
3078 		pr_err("failed to allocate memory for sl2vl update work\n");
3079 	}
3080 }
3081 
3082 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3083 			  enum mlx4_dev_event event, unsigned long param)
3084 {
3085 	struct ib_event ibev;
3086 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3087 	struct mlx4_eqe *eqe = NULL;
3088 	struct ib_event_work *ew;
3089 	int p = 0;
3090 
3091 	if (mlx4_is_bonded(dev) &&
3092 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3093 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3094 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3095 		if (!ew)
3096 			return;
3097 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3098 		ew->ib_dev = ibdev;
3099 		queue_work(wq, &ew->work);
3100 		return;
3101 	}
3102 
3103 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3104 		eqe = (struct mlx4_eqe *)param;
3105 	else
3106 		p = (int) param;
3107 
3108 	switch (event) {
3109 	case MLX4_DEV_EVENT_PORT_UP:
3110 		if (p > ibdev->num_ports)
3111 			return;
3112 		if (!mlx4_is_slave(dev) &&
3113 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3114 			IB_LINK_LAYER_INFINIBAND) {
3115 			if (mlx4_is_master(dev))
3116 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3117 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3118 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3119 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3120 		}
3121 		ibev.event = IB_EVENT_PORT_ACTIVE;
3122 		break;
3123 
3124 	case MLX4_DEV_EVENT_PORT_DOWN:
3125 		if (p > ibdev->num_ports)
3126 			return;
3127 		ibev.event = IB_EVENT_PORT_ERR;
3128 		break;
3129 
3130 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3131 		ibdev->ib_active = false;
3132 		ibev.event = IB_EVENT_DEVICE_FATAL;
3133 		mlx4_ib_handle_catas_error(ibdev);
3134 		break;
3135 
3136 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3137 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3138 		if (!ew) {
3139 			pr_err("failed to allocate memory for events work\n");
3140 			break;
3141 		}
3142 
3143 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3144 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3145 		ew->ib_dev = ibdev;
3146 		/* need to queue only for port owner, which uses GEN_EQE */
3147 		if (mlx4_is_master(dev))
3148 			queue_work(wq, &ew->work);
3149 		else
3150 			handle_port_mgmt_change_event(&ew->work);
3151 		return;
3152 
3153 	case MLX4_DEV_EVENT_SLAVE_INIT:
3154 		/* here, p is the slave id */
3155 		do_slave_init(ibdev, p, 1);
3156 		if (mlx4_is_master(dev)) {
3157 			int i;
3158 
3159 			for (i = 1; i <= ibdev->num_ports; i++) {
3160 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3161 					== IB_LINK_LAYER_INFINIBAND)
3162 					mlx4_ib_slave_alias_guid_event(ibdev,
3163 								       p, i,
3164 								       1);
3165 			}
3166 		}
3167 		return;
3168 
3169 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3170 		if (mlx4_is_master(dev)) {
3171 			int i;
3172 
3173 			for (i = 1; i <= ibdev->num_ports; i++) {
3174 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3175 					== IB_LINK_LAYER_INFINIBAND)
3176 					mlx4_ib_slave_alias_guid_event(ibdev,
3177 								       p, i,
3178 								       0);
3179 			}
3180 		}
3181 		/* here, p is the slave id */
3182 		do_slave_init(ibdev, p, 0);
3183 		return;
3184 
3185 	default:
3186 		return;
3187 	}
3188 
3189 	ibev.device	      = ibdev_ptr;
3190 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3191 
3192 	ib_dispatch_event(&ibev);
3193 }
3194 
3195 static struct mlx4_interface mlx4_ib_interface = {
3196 	.add		= mlx4_ib_add,
3197 	.remove		= mlx4_ib_remove,
3198 	.event		= mlx4_ib_event,
3199 	.protocol	= MLX4_PROT_IB_IPV6,
3200 	.flags		= MLX4_INTFF_BONDING
3201 };
3202 
3203 static int __init mlx4_ib_init(void)
3204 {
3205 	int err;
3206 
3207 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3208 	if (!wq)
3209 		return -ENOMEM;
3210 
3211 	err = mlx4_ib_mcg_init();
3212 	if (err)
3213 		goto clean_wq;
3214 
3215 	err = mlx4_register_interface(&mlx4_ib_interface);
3216 	if (err)
3217 		goto clean_mcg;
3218 
3219 	return 0;
3220 
3221 clean_mcg:
3222 	mlx4_ib_mcg_destroy();
3223 
3224 clean_wq:
3225 	destroy_workqueue(wq);
3226 	return err;
3227 }
3228 
3229 static void __exit mlx4_ib_cleanup(void)
3230 {
3231 	mlx4_unregister_interface(&mlx4_ib_interface);
3232 	mlx4_ib_mcg_destroy();
3233 	destroy_workqueue(wq);
3234 }
3235 
3236 module_init_order(mlx4_ib_init, SI_ORDER_SEVENTH);
3237 module_exit_order(mlx4_ib_cleanup, SI_ORDER_SEVENTH);
3238 
3239 static int
3240 mlx4ib_evhand(module_t mod, int event, void *arg)
3241 {
3242 	return (0);
3243 }
3244 
3245 static moduledata_t mlx4ib_mod = {
3246 	.name = "mlx4ib",
3247 	.evhand = mlx4ib_evhand,
3248 };
3249 
3250 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_LAST, SI_ORDER_ANY);
3251 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
3252 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);
3253 MODULE_DEPEND(mlx4ib, linuxkpi, 1, 1, 1);
3254