xref: /freebsd/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c (revision 4652422eb477731f284b1345afeefef7f269da50)
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	"3.6.0"
67 #endif
68 #define DRV_RELDATE	"December 2020"
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 struct ifnet *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
135 {
136 	struct mlx4_ib_dev *ibdev = to_mdev(device);
137 	struct ifnet *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 			struct ifnet *upper = NULL;
146 
147 			upper = netdev_master_upper_dev_get_rcu(dev);
148 			if (upper) {
149 				struct ifnet *active;
150 
151 				active = bond_option_active_slave_get_rcu(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 	struct ifnet *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(ndev->if_mtu);
735 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
736 
737 	props->state		= ((ndev->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
738 				   ndev->if_link_state == 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 struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1043 						  struct ib_udata *udata)
1044 {
1045 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1046 	struct mlx4_ib_ucontext *context;
1047 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1048 	struct mlx4_ib_alloc_ucontext_resp resp;
1049 	int err;
1050 
1051 	if (!dev->ib_active)
1052 		return ERR_PTR(-EAGAIN);
1053 
1054 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1055 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1056 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1057 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1058 	} else {
1059 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1060 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1061 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1062 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1063 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1064 	}
1065 
1066 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1067 	if (!context)
1068 		return ERR_PTR(-ENOMEM);
1069 
1070 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1071 	if (err) {
1072 		kfree(context);
1073 		return ERR_PTR(err);
1074 	}
1075 
1076 	INIT_LIST_HEAD(&context->db_page_list);
1077 	mutex_init(&context->db_page_mutex);
1078 
1079 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1080 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1081 	else
1082 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1083 
1084 	if (err) {
1085 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1086 		kfree(context);
1087 		return ERR_PTR(-EFAULT);
1088 	}
1089 
1090 	return &context->ibucontext;
1091 }
1092 
1093 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1094 {
1095 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1096 
1097 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1098 	kfree(context);
1099 
1100 	return 0;
1101 }
1102 
1103 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1104 {
1105 	/* vma_open is called when a new VMA is created on top of our VMA.
1106 	 * This is done through either mremap flow or split_vma (usually due
1107 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1108 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1109 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1110 	 * calling us again and trying to do incorrect actions. We assume that
1111 	 * the original vma size is exactly a single page that there will be no
1112 	 * "splitting" operations on.
1113 	 */
1114 	area->vm_ops = NULL;
1115 }
1116 
1117 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1118 {
1119 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1120 
1121 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1122 	 * file itself is closed, therefore no sync is needed with the regular
1123 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1124 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1125 	 * The close operation is usually called under mm->mmap_sem except when
1126 	 * process is exiting.  The exiting case is handled explicitly as part
1127 	 * of mlx4_ib_disassociate_ucontext.
1128 	 */
1129 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1130 				area->vm_private_data;
1131 
1132 	/* set the vma context pointer to null in the mlx4_ib driver's private
1133 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1134 	 */
1135 	mlx4_ib_vma_priv_data->vma = NULL;
1136 }
1137 
1138 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1139 	.open = mlx4_ib_vma_open,
1140 	.close = mlx4_ib_vma_close
1141 };
1142 
1143 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1144 				 struct mlx4_ib_vma_private_data *vma_private_data)
1145 {
1146 	vma_private_data->vma = vma;
1147 	vma->vm_private_data = vma_private_data;
1148 	vma->vm_ops =  &mlx4_ib_vm_ops;
1149 }
1150 
1151 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1152 {
1153 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1154 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1155 
1156 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1157 		return -EINVAL;
1158 
1159 	if (vma->vm_pgoff == 0) {
1160 		/* We prevent double mmaping on same context */
1161 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1162 			return -EINVAL;
1163 
1164 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1165 
1166 		if (io_remap_pfn_range(vma, vma->vm_start,
1167 				       to_mucontext(context)->uar.pfn,
1168 				       PAGE_SIZE, vma->vm_page_prot))
1169 			return -EAGAIN;
1170 
1171 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1172 
1173 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1174 		/* We prevent double mmaping on same context */
1175 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1176 			return -EINVAL;
1177 
1178 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1179 
1180 		if (io_remap_pfn_range(vma, vma->vm_start,
1181 				       to_mucontext(context)->uar.pfn +
1182 				       dev->dev->caps.num_uars,
1183 				       PAGE_SIZE, vma->vm_page_prot))
1184 			return -EAGAIN;
1185 
1186 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1187 
1188 	} else if (vma->vm_pgoff == 3) {
1189 		struct mlx4_clock_params params;
1190 		int ret;
1191 
1192 		/* We prevent double mmaping on same context */
1193 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1194 			return -EINVAL;
1195 
1196 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1197 
1198 		if (ret)
1199 			return ret;
1200 
1201 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1202 		if (io_remap_pfn_range(vma, vma->vm_start,
1203 				       (pci_resource_start(dev->dev->persist->pdev,
1204 							   params.bar) +
1205 					params.offset)
1206 				       >> PAGE_SHIFT,
1207 				       PAGE_SIZE, vma->vm_page_prot))
1208 			return -EAGAIN;
1209 
1210 		mlx4_ib_set_vma_data(vma,
1211 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1212 	} else {
1213 		return -EINVAL;
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1220 				      struct ib_ucontext *context,
1221 				      struct ib_udata *udata)
1222 {
1223 	struct mlx4_ib_pd *pd;
1224 	int err;
1225 
1226 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1227 	if (!pd)
1228 		return ERR_PTR(-ENOMEM);
1229 
1230 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1231 	if (err) {
1232 		kfree(pd);
1233 		return ERR_PTR(err);
1234 	}
1235 
1236 	if (context)
1237 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1238 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1239 			kfree(pd);
1240 			return ERR_PTR(-EFAULT);
1241 		}
1242 
1243 	return &pd->ibpd;
1244 }
1245 
1246 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1247 {
1248 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1249 	kfree(pd);
1250 
1251 	return 0;
1252 }
1253 
1254 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1255 					  struct ib_ucontext *context,
1256 					  struct ib_udata *udata)
1257 {
1258 	struct mlx4_ib_xrcd *xrcd;
1259 	struct ib_cq_init_attr cq_attr = {};
1260 	int err;
1261 
1262 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1263 		return ERR_PTR(-ENOSYS);
1264 
1265 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1266 	if (!xrcd)
1267 		return ERR_PTR(-ENOMEM);
1268 
1269 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1270 	if (err)
1271 		goto err1;
1272 
1273 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1274 	if (IS_ERR(xrcd->pd)) {
1275 		err = PTR_ERR(xrcd->pd);
1276 		goto err2;
1277 	}
1278 
1279 	cq_attr.cqe = 1;
1280 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1281 	if (IS_ERR(xrcd->cq)) {
1282 		err = PTR_ERR(xrcd->cq);
1283 		goto err3;
1284 	}
1285 
1286 	return &xrcd->ibxrcd;
1287 
1288 err3:
1289 	ib_dealloc_pd(xrcd->pd);
1290 err2:
1291 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1292 err1:
1293 	kfree(xrcd);
1294 	return ERR_PTR(err);
1295 }
1296 
1297 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1298 {
1299 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1300 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1301 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1302 	kfree(xrcd);
1303 
1304 	return 0;
1305 }
1306 
1307 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1308 {
1309 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1310 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1311 	struct mlx4_ib_gid_entry *ge;
1312 
1313 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1314 	if (!ge)
1315 		return -ENOMEM;
1316 
1317 	ge->gid = *gid;
1318 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1319 		ge->port = mqp->port;
1320 		ge->added = 1;
1321 	}
1322 
1323 	mutex_lock(&mqp->mutex);
1324 	list_add_tail(&ge->list, &mqp->gid_list);
1325 	mutex_unlock(&mqp->mutex);
1326 
1327 	return 0;
1328 }
1329 
1330 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1331 					  struct mlx4_ib_counters *ctr_table)
1332 {
1333 	struct counter_index *counter, *tmp_count;
1334 
1335 	mutex_lock(&ctr_table->mutex);
1336 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1337 				 list) {
1338 		if (counter->allocated)
1339 			mlx4_counter_free(ibdev->dev, counter->index);
1340 		list_del(&counter->list);
1341 		kfree(counter);
1342 	}
1343 	mutex_unlock(&ctr_table->mutex);
1344 }
1345 
1346 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1347 		   union ib_gid *gid)
1348 {
1349 	struct ifnet *ndev;
1350 	int ret = 0;
1351 
1352 	if (!mqp->port)
1353 		return 0;
1354 
1355 	spin_lock_bh(&mdev->iboe.lock);
1356 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1357 	if (ndev)
1358 		if_ref(ndev);
1359 	spin_unlock_bh(&mdev->iboe.lock);
1360 
1361 	if (ndev) {
1362 		ret = 1;
1363 		if_rele(ndev);
1364 	}
1365 
1366 	return ret;
1367 }
1368 
1369 struct mlx4_ib_steering {
1370 	struct list_head list;
1371 	struct mlx4_flow_reg_id reg_id;
1372 	union ib_gid gid;
1373 };
1374 
1375 #define LAST_ETH_FIELD vlan_tag
1376 #define LAST_IB_FIELD sl
1377 #define LAST_IPV4_FIELD dst_ip
1378 #define LAST_TCP_UDP_FIELD src_port
1379 
1380 /* Field is the last supported field */
1381 #define FIELDS_NOT_SUPPORTED(filter, field)\
1382 	memchr_inv((void *)&filter.field  +\
1383 		   sizeof(filter.field), 0,\
1384 		   sizeof(filter) -\
1385 		   offsetof(typeof(filter), field) -\
1386 		   sizeof(filter.field))
1387 
1388 static int parse_flow_attr(struct mlx4_dev *dev,
1389 			   u32 qp_num,
1390 			   union ib_flow_spec *ib_spec,
1391 			   struct _rule_hw *mlx4_spec)
1392 {
1393 	enum mlx4_net_trans_rule_id type;
1394 
1395 	switch (ib_spec->type) {
1396 	case IB_FLOW_SPEC_ETH:
1397 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1398 			return -ENOTSUPP;
1399 
1400 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1401 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1402 		       ETH_ALEN);
1403 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1404 		       ETH_ALEN);
1405 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1406 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1407 		break;
1408 	case IB_FLOW_SPEC_IB:
1409 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1410 			return -ENOTSUPP;
1411 
1412 		type = MLX4_NET_TRANS_RULE_ID_IB;
1413 		mlx4_spec->ib.l3_qpn =
1414 			cpu_to_be32(qp_num);
1415 		mlx4_spec->ib.qpn_mask =
1416 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1417 		break;
1418 
1419 
1420 	case IB_FLOW_SPEC_IPV4:
1421 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1422 			return -ENOTSUPP;
1423 
1424 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1425 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1426 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1427 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1428 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1429 		break;
1430 
1431 	case IB_FLOW_SPEC_TCP:
1432 	case IB_FLOW_SPEC_UDP:
1433 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1434 			return -ENOTSUPP;
1435 
1436 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1437 					MLX4_NET_TRANS_RULE_ID_TCP :
1438 					MLX4_NET_TRANS_RULE_ID_UDP;
1439 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1440 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1441 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1442 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1443 		break;
1444 
1445 	default:
1446 		return -EINVAL;
1447 	}
1448 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1449 	    mlx4_hw_rule_sz(dev, type) < 0)
1450 		return -EINVAL;
1451 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1452 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1453 	return mlx4_hw_rule_sz(dev, type);
1454 }
1455 
1456 struct default_rules {
1457 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1458 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1459 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1460 	__u8  link_layer;
1461 };
1462 static const struct default_rules default_table[] = {
1463 	{
1464 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1465 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1466 		.rules_create_list = {IB_FLOW_SPEC_IB},
1467 		.link_layer = IB_LINK_LAYER_INFINIBAND
1468 	}
1469 };
1470 
1471 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1472 					 struct ib_flow_attr *flow_attr)
1473 {
1474 	int i, j, k;
1475 	void *ib_flow;
1476 	const struct default_rules *pdefault_rules = default_table;
1477 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1478 
1479 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1480 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1481 		memset(&field_types, 0, sizeof(field_types));
1482 
1483 		if (link_layer != pdefault_rules->link_layer)
1484 			continue;
1485 
1486 		ib_flow = flow_attr + 1;
1487 		/* we assume the specs are sorted */
1488 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1489 		     j < flow_attr->num_of_specs; k++) {
1490 			union ib_flow_spec *current_flow =
1491 				(union ib_flow_spec *)ib_flow;
1492 
1493 			/* same layer but different type */
1494 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1495 			     (pdefault_rules->mandatory_fields[k] &
1496 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1497 			    (current_flow->type !=
1498 			     pdefault_rules->mandatory_fields[k]))
1499 				goto out;
1500 
1501 			/* same layer, try match next one */
1502 			if (current_flow->type ==
1503 			    pdefault_rules->mandatory_fields[k]) {
1504 				j++;
1505 				ib_flow +=
1506 					((union ib_flow_spec *)ib_flow)->size;
1507 			}
1508 		}
1509 
1510 		ib_flow = flow_attr + 1;
1511 		for (j = 0; j < flow_attr->num_of_specs;
1512 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1513 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1514 				/* same layer and same type */
1515 				if (((union ib_flow_spec *)ib_flow)->type ==
1516 				    pdefault_rules->mandatory_not_fields[k])
1517 					goto out;
1518 
1519 		return i;
1520 	}
1521 out:
1522 	return -1;
1523 }
1524 
1525 static int __mlx4_ib_create_default_rules(
1526 		struct mlx4_ib_dev *mdev,
1527 		struct ib_qp *qp,
1528 		const struct default_rules *pdefault_rules,
1529 		struct _rule_hw *mlx4_spec) {
1530 	int size = 0;
1531 	int i;
1532 
1533 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1534 		int ret;
1535 		union ib_flow_spec ib_spec;
1536 		switch (pdefault_rules->rules_create_list[i]) {
1537 		case 0:
1538 			/* no rule */
1539 			continue;
1540 		case IB_FLOW_SPEC_IB:
1541 			ib_spec.type = IB_FLOW_SPEC_IB;
1542 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1543 
1544 			break;
1545 		default:
1546 			/* invalid rule */
1547 			return -EINVAL;
1548 		}
1549 		/* We must put empty rule, qpn is being ignored */
1550 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1551 				      mlx4_spec);
1552 		if (ret < 0) {
1553 			pr_info("invalid parsing\n");
1554 			return -EINVAL;
1555 		}
1556 
1557 		mlx4_spec = (void *)mlx4_spec + ret;
1558 		size += ret;
1559 	}
1560 	return size;
1561 }
1562 
1563 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1564 			  int domain,
1565 			  enum mlx4_net_trans_promisc_mode flow_type,
1566 			  u64 *reg_id)
1567 {
1568 	int ret, i;
1569 	int size = 0;
1570 	void *ib_flow;
1571 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1572 	struct mlx4_cmd_mailbox *mailbox;
1573 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1574 	int default_flow;
1575 
1576 	static const u16 __mlx4_domain[] = {
1577 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1578 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1579 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1580 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1581 	};
1582 
1583 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1584 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1585 		return -EINVAL;
1586 	}
1587 
1588 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1589 		pr_err("Invalid domain value %d\n", domain);
1590 		return -EINVAL;
1591 	}
1592 
1593 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1594 		return -EINVAL;
1595 
1596 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1597 	if (IS_ERR(mailbox))
1598 		return PTR_ERR(mailbox);
1599 	ctrl = mailbox->buf;
1600 
1601 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1602 				 flow_attr->priority);
1603 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1604 	ctrl->port = flow_attr->port;
1605 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1606 
1607 	ib_flow = flow_attr + 1;
1608 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1609 	/* Add default flows */
1610 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1611 	if (default_flow >= 0) {
1612 		ret = __mlx4_ib_create_default_rules(
1613 				mdev, qp, default_table + default_flow,
1614 				mailbox->buf + size);
1615 		if (ret < 0) {
1616 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1617 			return -EINVAL;
1618 		}
1619 		size += ret;
1620 	}
1621 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1622 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1623 				      mailbox->buf + size);
1624 		if (ret < 0) {
1625 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1626 			return -EINVAL;
1627 		}
1628 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1629 		size += ret;
1630 	}
1631 
1632 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1633 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1634 			   MLX4_CMD_WRAPPED);
1635 	if (ret == -ENOMEM)
1636 		pr_err("mcg table is full. Fail to register network rule.\n");
1637 	else if (ret == -ENXIO)
1638 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1639 	else if (ret)
1640 		pr_err("Invalid argument. Fail to register network rule.\n");
1641 
1642 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1643 	return ret;
1644 }
1645 
1646 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1647 {
1648 	int err;
1649 	err = mlx4_cmd(dev, reg_id, 0, 0,
1650 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1651 		       MLX4_CMD_WRAPPED);
1652 	if (err)
1653 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1654 		       (long long)reg_id);
1655 	return err;
1656 }
1657 
1658 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1659 				    u64 *reg_id)
1660 {
1661 	void *ib_flow;
1662 	union ib_flow_spec *ib_spec;
1663 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1664 	int err = 0;
1665 
1666 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1667 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1668 		return 0; /* do nothing */
1669 
1670 	ib_flow = flow_attr + 1;
1671 	ib_spec = (union ib_flow_spec *)ib_flow;
1672 
1673 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1674 		return 0; /* do nothing */
1675 
1676 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1677 				    flow_attr->port, qp->qp_num,
1678 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1679 				    reg_id);
1680 	return err;
1681 }
1682 
1683 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1684 				      struct ib_flow_attr *flow_attr,
1685 				      enum mlx4_net_trans_promisc_mode *type)
1686 {
1687 	int err = 0;
1688 
1689 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1690 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1691 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1692 		return -EOPNOTSUPP;
1693 	}
1694 
1695 	if (flow_attr->num_of_specs == 0) {
1696 		type[0] = MLX4_FS_MC_SNIFFER;
1697 		type[1] = MLX4_FS_UC_SNIFFER;
1698 	} else {
1699 		union ib_flow_spec *ib_spec;
1700 
1701 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1702 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1703 			return -EINVAL;
1704 
1705 		/* if all is zero than MC and UC */
1706 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1707 			type[0] = MLX4_FS_MC_SNIFFER;
1708 			type[1] = MLX4_FS_UC_SNIFFER;
1709 		} else {
1710 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1711 					    ib_spec->eth.mask.dst_mac[1],
1712 					    ib_spec->eth.mask.dst_mac[2],
1713 					    ib_spec->eth.mask.dst_mac[3],
1714 					    ib_spec->eth.mask.dst_mac[4],
1715 					    ib_spec->eth.mask.dst_mac[5]};
1716 
1717 			/* Above xor was only on MC bit, non empty mask is valid
1718 			 * only if this bit is set and rest are zero.
1719 			 */
1720 			if (!is_zero_ether_addr(&mac[0]))
1721 				return -EINVAL;
1722 
1723 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1724 				type[0] = MLX4_FS_MC_SNIFFER;
1725 			else
1726 				type[0] = MLX4_FS_UC_SNIFFER;
1727 		}
1728 	}
1729 
1730 	return err;
1731 }
1732 
1733 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1734 				    struct ib_flow_attr *flow_attr,
1735 				    int domain)
1736 {
1737 	int err = 0, i = 0, j = 0;
1738 	struct mlx4_ib_flow *mflow;
1739 	enum mlx4_net_trans_promisc_mode type[2];
1740 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1741 	int is_bonded = mlx4_is_bonded(dev);
1742 
1743 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1744 		return ERR_PTR(-EINVAL);
1745 
1746 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1747 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1748 		return ERR_PTR(-EOPNOTSUPP);
1749 
1750 	memset(type, 0, sizeof(type));
1751 
1752 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1753 	if (!mflow) {
1754 		err = -ENOMEM;
1755 		goto err_free;
1756 	}
1757 
1758 	switch (flow_attr->type) {
1759 	case IB_FLOW_ATTR_NORMAL:
1760 		/* If dont trap flag (continue match) is set, under specific
1761 		 * condition traffic be replicated to given qp,
1762 		 * without stealing it
1763 		 */
1764 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1765 			err = mlx4_ib_add_dont_trap_rule(dev,
1766 							 flow_attr,
1767 							 type);
1768 			if (err)
1769 				goto err_free;
1770 		} else {
1771 			type[0] = MLX4_FS_REGULAR;
1772 		}
1773 		break;
1774 
1775 	case IB_FLOW_ATTR_ALL_DEFAULT:
1776 		type[0] = MLX4_FS_ALL_DEFAULT;
1777 		break;
1778 
1779 	case IB_FLOW_ATTR_MC_DEFAULT:
1780 		type[0] = MLX4_FS_MC_DEFAULT;
1781 		break;
1782 
1783 	case IB_FLOW_ATTR_SNIFFER:
1784 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1785 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1786 		break;
1787 
1788 	default:
1789 		err = -EINVAL;
1790 		goto err_free;
1791 	}
1792 
1793 	while (i < ARRAY_SIZE(type) && type[i]) {
1794 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1795 					    &mflow->reg_id[i].id);
1796 		if (err)
1797 			goto err_create_flow;
1798 		if (is_bonded) {
1799 			/* Application always sees one port so the mirror rule
1800 			 * must be on port #2
1801 			 */
1802 			flow_attr->port = 2;
1803 			err = __mlx4_ib_create_flow(qp, flow_attr,
1804 						    domain, type[j],
1805 						    &mflow->reg_id[j].mirror);
1806 			flow_attr->port = 1;
1807 			if (err)
1808 				goto err_create_flow;
1809 			j++;
1810 		}
1811 
1812 		i++;
1813 	}
1814 
1815 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1816 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1817 					       &mflow->reg_id[i].id);
1818 		if (err)
1819 			goto err_create_flow;
1820 
1821 		if (is_bonded) {
1822 			flow_attr->port = 2;
1823 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1824 						       &mflow->reg_id[j].mirror);
1825 			flow_attr->port = 1;
1826 			if (err)
1827 				goto err_create_flow;
1828 			j++;
1829 		}
1830 		/* function to create mirror rule */
1831 		i++;
1832 	}
1833 
1834 	return &mflow->ibflow;
1835 
1836 err_create_flow:
1837 	while (i) {
1838 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1839 					     mflow->reg_id[i].id);
1840 		i--;
1841 	}
1842 
1843 	while (j) {
1844 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1845 					     mflow->reg_id[j].mirror);
1846 		j--;
1847 	}
1848 err_free:
1849 	kfree(mflow);
1850 	return ERR_PTR(err);
1851 }
1852 
1853 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1854 {
1855 	int err, ret = 0;
1856 	int i = 0;
1857 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1858 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1859 
1860 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1861 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1862 		if (err)
1863 			ret = err;
1864 		if (mflow->reg_id[i].mirror) {
1865 			err = __mlx4_ib_destroy_flow(mdev->dev,
1866 						     mflow->reg_id[i].mirror);
1867 			if (err)
1868 				ret = err;
1869 		}
1870 		i++;
1871 	}
1872 
1873 	kfree(mflow);
1874 	return ret;
1875 }
1876 
1877 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1878 {
1879 	int err;
1880 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1881 	struct mlx4_dev	*dev = mdev->dev;
1882 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1883 	struct mlx4_ib_steering *ib_steering = NULL;
1884 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1885 	struct mlx4_flow_reg_id	reg_id;
1886 
1887 	if (mdev->dev->caps.steering_mode ==
1888 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1889 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1890 		if (!ib_steering)
1891 			return -ENOMEM;
1892 	}
1893 
1894 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1895 				    !!(mqp->flags &
1896 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1897 				    prot, &reg_id.id);
1898 	if (err) {
1899 		pr_err("multicast attach op failed, err %d\n", err);
1900 		goto err_malloc;
1901 	}
1902 
1903 	reg_id.mirror = 0;
1904 	if (mlx4_is_bonded(dev)) {
1905 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1906 					    (mqp->port == 1) ? 2 : 1,
1907 					    !!(mqp->flags &
1908 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1909 					    prot, &reg_id.mirror);
1910 		if (err)
1911 			goto err_add;
1912 	}
1913 
1914 	err = add_gid_entry(ibqp, gid);
1915 	if (err)
1916 		goto err_add;
1917 
1918 	if (ib_steering) {
1919 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1920 		ib_steering->reg_id = reg_id;
1921 		mutex_lock(&mqp->mutex);
1922 		list_add(&ib_steering->list, &mqp->steering_rules);
1923 		mutex_unlock(&mqp->mutex);
1924 	}
1925 	return 0;
1926 
1927 err_add:
1928 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1929 			      prot, reg_id.id);
1930 	if (reg_id.mirror)
1931 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1932 				      prot, reg_id.mirror);
1933 err_malloc:
1934 	kfree(ib_steering);
1935 
1936 	return err;
1937 }
1938 
1939 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1940 {
1941 	struct mlx4_ib_gid_entry *ge;
1942 	struct mlx4_ib_gid_entry *tmp;
1943 	struct mlx4_ib_gid_entry *ret = NULL;
1944 
1945 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1946 		if (!memcmp(raw, ge->gid.raw, 16)) {
1947 			ret = ge;
1948 			break;
1949 		}
1950 	}
1951 
1952 	return ret;
1953 }
1954 
1955 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1956 {
1957 	int err;
1958 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1959 	struct mlx4_dev *dev = mdev->dev;
1960 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1961 	struct ifnet *ndev;
1962 	struct mlx4_ib_gid_entry *ge;
1963 	struct mlx4_flow_reg_id reg_id = {0, 0};
1964 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1965 
1966 	if (mdev->dev->caps.steering_mode ==
1967 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1968 		struct mlx4_ib_steering *ib_steering;
1969 
1970 		mutex_lock(&mqp->mutex);
1971 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1972 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1973 				list_del(&ib_steering->list);
1974 				break;
1975 			}
1976 		}
1977 		mutex_unlock(&mqp->mutex);
1978 		if (&ib_steering->list == &mqp->steering_rules) {
1979 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1980 			return -EINVAL;
1981 		}
1982 		reg_id = ib_steering->reg_id;
1983 		kfree(ib_steering);
1984 	}
1985 
1986 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1987 				    prot, reg_id.id);
1988 	if (err)
1989 		return err;
1990 
1991 	if (mlx4_is_bonded(dev)) {
1992 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1993 					    prot, reg_id.mirror);
1994 		if (err)
1995 			return err;
1996 	}
1997 
1998 	mutex_lock(&mqp->mutex);
1999 	ge = find_gid_entry(mqp, gid->raw);
2000 	if (ge) {
2001 		spin_lock_bh(&mdev->iboe.lock);
2002 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2003 		if (ndev)
2004 			if_ref(ndev);
2005 		spin_unlock_bh(&mdev->iboe.lock);
2006 		if (ndev)
2007 			if_rele(ndev);
2008 		list_del(&ge->list);
2009 		kfree(ge);
2010 	} else
2011 		pr_warn("could not find mgid entry\n");
2012 
2013 	mutex_unlock(&mqp->mutex);
2014 
2015 	return 0;
2016 }
2017 
2018 static int init_node_data(struct mlx4_ib_dev *dev)
2019 {
2020 	struct ib_smp *in_mad  = NULL;
2021 	struct ib_smp *out_mad = NULL;
2022 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2023 	int err = -ENOMEM;
2024 
2025 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2026 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2027 	if (!in_mad || !out_mad)
2028 		goto out;
2029 
2030 	init_query_mad(in_mad);
2031 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2032 	if (mlx4_is_master(dev->dev))
2033 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2034 
2035 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2036 	if (err)
2037 		goto out;
2038 
2039 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2040 
2041 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2042 
2043 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2044 	if (err)
2045 		goto out;
2046 
2047 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2048 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2049 
2050 out:
2051 	kfree(in_mad);
2052 	kfree(out_mad);
2053 	return err;
2054 }
2055 
2056 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2057 			char *buf)
2058 {
2059 	struct mlx4_ib_dev *dev =
2060 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2061 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2062 }
2063 
2064 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2065 			char *buf)
2066 {
2067 	struct mlx4_ib_dev *dev =
2068 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2069 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2070 }
2071 
2072 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2073 			  char *buf)
2074 {
2075 	struct mlx4_ib_dev *dev =
2076 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2077 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2078 		       dev->dev->board_id);
2079 }
2080 
2081 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2082 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2083 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2084 
2085 static struct device_attribute *mlx4_class_attributes[] = {
2086 	&dev_attr_hw_rev,
2087 	&dev_attr_hca_type,
2088 	&dev_attr_board_id
2089 };
2090 
2091 struct diag_counter {
2092 	const char *name;
2093 	u32 offset;
2094 };
2095 
2096 #define DIAG_COUNTER(_name, _offset)			\
2097 	{ .name = #_name, .offset = _offset }
2098 
2099 static const struct diag_counter diag_basic[] = {
2100 	DIAG_COUNTER(rq_num_lle, 0x00),
2101 	DIAG_COUNTER(sq_num_lle, 0x04),
2102 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2103 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2104 	DIAG_COUNTER(rq_num_lpe, 0x18),
2105 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2106 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2107 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2108 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2109 	DIAG_COUNTER(sq_num_bre, 0x34),
2110 	DIAG_COUNTER(sq_num_rire, 0x44),
2111 	DIAG_COUNTER(rq_num_rire, 0x48),
2112 	DIAG_COUNTER(sq_num_rae, 0x4C),
2113 	DIAG_COUNTER(rq_num_rae, 0x50),
2114 	DIAG_COUNTER(sq_num_roe, 0x54),
2115 	DIAG_COUNTER(sq_num_tree, 0x5C),
2116 	DIAG_COUNTER(sq_num_rree, 0x64),
2117 	DIAG_COUNTER(rq_num_rnr, 0x68),
2118 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2119 	DIAG_COUNTER(rq_num_oos, 0x100),
2120 	DIAG_COUNTER(sq_num_oos, 0x104),
2121 };
2122 
2123 static const struct diag_counter diag_ext[] = {
2124 	DIAG_COUNTER(rq_num_dup, 0x130),
2125 	DIAG_COUNTER(sq_num_to, 0x134),
2126 };
2127 
2128 static const struct diag_counter diag_device_only[] = {
2129 	DIAG_COUNTER(num_cqovf, 0x1A0),
2130 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2131 };
2132 
2133 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2134 						    u8 port_num)
2135 {
2136 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2137 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2138 
2139 	if (!diag[!!port_num].name)
2140 		return NULL;
2141 
2142 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2143 					  diag[!!port_num].num_counters,
2144 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2145 }
2146 
2147 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2148 				struct rdma_hw_stats *stats,
2149 				u8 port, int index)
2150 {
2151 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2152 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2153 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2154 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2155 	int ret;
2156 	int i;
2157 
2158 	ret = mlx4_query_diag_counters(dev->dev,
2159 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2160 				       diag[!!port].offset, hw_value,
2161 				       diag[!!port].num_counters, port);
2162 
2163 	if (ret)
2164 		return ret;
2165 
2166 	for (i = 0; i < diag[!!port].num_counters; i++)
2167 		stats->value[i] = hw_value[i];
2168 
2169 	return diag[!!port].num_counters;
2170 }
2171 
2172 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2173 					 const char ***name,
2174 					 u32 **offset,
2175 					 u32 *num,
2176 					 bool port)
2177 {
2178 	u32 num_counters;
2179 
2180 	num_counters = ARRAY_SIZE(diag_basic);
2181 
2182 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2183 		num_counters += ARRAY_SIZE(diag_ext);
2184 
2185 	if (!port)
2186 		num_counters += ARRAY_SIZE(diag_device_only);
2187 
2188 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2189 	if (!*name)
2190 		return -ENOMEM;
2191 
2192 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2193 	if (!*offset)
2194 		goto err_name;
2195 
2196 	*num = num_counters;
2197 
2198 	return 0;
2199 
2200 err_name:
2201 	kfree(*name);
2202 	return -ENOMEM;
2203 }
2204 
2205 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2206 				       const char **name,
2207 				       u32 *offset,
2208 				       bool port)
2209 {
2210 	int i;
2211 	int j;
2212 
2213 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2214 		name[i] = diag_basic[i].name;
2215 		offset[i] = diag_basic[i].offset;
2216 	}
2217 
2218 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2219 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2220 			name[j] = diag_ext[i].name;
2221 			offset[j] = diag_ext[i].offset;
2222 		}
2223 	}
2224 
2225 	if (!port) {
2226 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2227 			name[j] = diag_device_only[i].name;
2228 			offset[j] = diag_device_only[i].offset;
2229 		}
2230 	}
2231 }
2232 
2233 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2234 {
2235 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2236 	int i;
2237 	int ret;
2238 	bool per_port = !!(ibdev->dev->caps.flags2 &
2239 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2240 
2241 	if (mlx4_is_slave(ibdev->dev))
2242 		return 0;
2243 
2244 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2245 		/* i == 1 means we are building port counters */
2246 		if (i && !per_port)
2247 			continue;
2248 
2249 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2250 						    &diag[i].offset,
2251 						    &diag[i].num_counters, i);
2252 		if (ret)
2253 			goto err_alloc;
2254 
2255 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2256 					   diag[i].offset, i);
2257 	}
2258 
2259 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2260 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2261 
2262 	return 0;
2263 
2264 err_alloc:
2265 	if (i) {
2266 		kfree(diag[i - 1].name);
2267 		kfree(diag[i - 1].offset);
2268 	}
2269 
2270 	return ret;
2271 }
2272 
2273 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2274 {
2275 	int i;
2276 
2277 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2278 		kfree(ibdev->diag_counters[i].offset);
2279 		kfree(ibdev->diag_counters[i].name);
2280 	}
2281 }
2282 
2283 #define MLX4_IB_INVALID_MAC	((u64)-1)
2284 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2285 			       struct ifnet *dev,
2286 			       int port)
2287 {
2288 	u64 new_smac = 0;
2289 	u64 release_mac = MLX4_IB_INVALID_MAC;
2290 	struct mlx4_ib_qp *qp;
2291 
2292 	new_smac = mlx4_mac_to_u64(IF_LLADDR(dev));
2293 
2294 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2295 
2296 	/* no need for update QP1 and mac registration in non-SRIOV */
2297 	if (!mlx4_is_mfunc(ibdev->dev))
2298 		return;
2299 
2300 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2301 	qp = ibdev->qp1_proxy[port - 1];
2302 	if (qp) {
2303 		int new_smac_index;
2304 		u64 old_smac;
2305 		struct mlx4_update_qp_params update_params;
2306 
2307 		mutex_lock(&qp->mutex);
2308 		old_smac = qp->pri.smac;
2309 		if (new_smac == old_smac)
2310 			goto unlock;
2311 
2312 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2313 
2314 		if (new_smac_index < 0)
2315 			goto unlock;
2316 
2317 		update_params.smac_index = new_smac_index;
2318 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2319 				   &update_params)) {
2320 			release_mac = new_smac;
2321 			goto unlock;
2322 		}
2323 		/* if old port was zero, no mac was yet registered for this QP */
2324 		if (qp->pri.smac_port)
2325 			release_mac = old_smac;
2326 		qp->pri.smac = new_smac;
2327 		qp->pri.smac_port = port;
2328 		qp->pri.smac_index = new_smac_index;
2329 	}
2330 
2331 unlock:
2332 	if (release_mac != MLX4_IB_INVALID_MAC)
2333 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2334 	if (qp)
2335 		mutex_unlock(&qp->mutex);
2336 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2337 }
2338 
2339 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2340 				 struct ifnet *dev,
2341 				 unsigned long event)
2342 
2343 {
2344 	struct mlx4_ib_iboe *iboe;
2345 	int update_qps_port = -1;
2346 	int port;
2347 
2348 	iboe = &ibdev->iboe;
2349 
2350 	spin_lock_bh(&iboe->lock);
2351 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2352 
2353 		iboe->netdevs[port - 1] =
2354 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2355 
2356 		if (dev == iboe->netdevs[port - 1] &&
2357 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2358 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2359 			update_qps_port = port;
2360 
2361 	}
2362 	spin_unlock_bh(&iboe->lock);
2363 
2364 	if (update_qps_port > 0)
2365 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2366 }
2367 
2368 static int mlx4_ib_netdev_event(struct notifier_block *this,
2369 				unsigned long event, void *ptr)
2370 {
2371 	struct ifnet *dev = netdev_notifier_info_to_ifp(ptr);
2372 	struct mlx4_ib_dev *ibdev;
2373 
2374 	if (dev->if_vnet != &init_net)
2375 		return NOTIFY_DONE;
2376 
2377 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2378 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2379 
2380 	return NOTIFY_DONE;
2381 }
2382 
2383 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2384 {
2385 	int port;
2386 	int slave;
2387 	int i;
2388 
2389 	if (mlx4_is_master(ibdev->dev)) {
2390 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2391 		     ++slave) {
2392 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2393 				for (i = 0;
2394 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2395 				     ++i) {
2396 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2397 					/* master has the identity virt2phys pkey mapping */
2398 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2399 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2400 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2401 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2402 				}
2403 			}
2404 		}
2405 		/* initialize pkey cache */
2406 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2407 			for (i = 0;
2408 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2409 			     ++i)
2410 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2411 					(i) ? 0 : 0xFFFF;
2412 		}
2413 	}
2414 }
2415 
2416 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2417 {
2418 	int i, j, eq = 0, total_eqs = 0;
2419 
2420 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2421 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2422 	if (!ibdev->eq_table)
2423 		return;
2424 
2425 	for (i = 1; i <= dev->caps.num_ports; i++) {
2426 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2427 		     j++, total_eqs++) {
2428 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2429 				continue;
2430 			ibdev->eq_table[eq] = total_eqs;
2431 			if (!mlx4_assign_eq(dev, i,
2432 					    &ibdev->eq_table[eq]))
2433 				eq++;
2434 			else
2435 				ibdev->eq_table[eq] = -1;
2436 		}
2437 	}
2438 
2439 	for (i = eq; i < dev->caps.num_comp_vectors;
2440 	     ibdev->eq_table[i++] = -1)
2441 		;
2442 
2443 	/* Advertise the new number of EQs to clients */
2444 	ibdev->ib_dev.num_comp_vectors = eq;
2445 }
2446 
2447 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2448 {
2449 	int i;
2450 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2451 
2452 	/* no eqs were allocated */
2453 	if (!ibdev->eq_table)
2454 		return;
2455 
2456 	/* Reset the advertised EQ number */
2457 	ibdev->ib_dev.num_comp_vectors = 0;
2458 
2459 	for (i = 0; i < total_eqs; i++)
2460 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2461 
2462 	kfree(ibdev->eq_table);
2463 	ibdev->eq_table = NULL;
2464 }
2465 
2466 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2467 			       struct ib_port_immutable *immutable)
2468 {
2469 	struct ib_port_attr attr;
2470 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2471 	int err;
2472 
2473 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2474 	if (err)
2475 		return err;
2476 
2477 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2478 	immutable->gid_tbl_len = attr.gid_tbl_len;
2479 
2480 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2481 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2482 	} else {
2483 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2484 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2485 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2486 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2487 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2488 	}
2489 
2490 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2491 
2492 	return 0;
2493 }
2494 
2495 static void get_fw_ver_str(struct ib_device *device, char *str,
2496 			   size_t str_len)
2497 {
2498 	struct mlx4_ib_dev *dev =
2499 		container_of(device, struct mlx4_ib_dev, ib_dev);
2500 	snprintf(str, str_len, "%d.%d.%d",
2501 		 (int) (dev->dev->caps.fw_ver >> 32),
2502 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2503 		 (int) dev->dev->caps.fw_ver & 0xffff);
2504 }
2505 
2506 static void *mlx4_ib_add(struct mlx4_dev *dev)
2507 {
2508 	struct mlx4_ib_dev *ibdev;
2509 	int num_ports;
2510 	int i, j;
2511 	int err;
2512 	struct mlx4_ib_iboe *iboe;
2513 	int ib_num_ports = 0;
2514 	int num_req_counters;
2515 	int allocated;
2516 	u32 counter_index;
2517 	struct counter_index *new_counter_index = NULL;
2518 
2519 	pr_info_once("%s", mlx4_ib_version);
2520 
2521 	num_ports = 0;
2522 	mlx4_foreach_ib_transport_port(i, dev)
2523 		num_ports++;
2524 
2525 	/* No point in registering a device with no ports... */
2526 	if (num_ports == 0)
2527 		return NULL;
2528 
2529 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2530 	if (!ibdev) {
2531 		dev_err(&dev->persist->pdev->dev,
2532 			"Device struct alloc failed\n");
2533 		return NULL;
2534 	}
2535 
2536 	iboe = &ibdev->iboe;
2537 
2538 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2539 		goto err_dealloc;
2540 
2541 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2542 		goto err_pd;
2543 
2544 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2545 				 PAGE_SIZE);
2546 	if (!ibdev->uar_map)
2547 		goto err_uar;
2548 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2549 
2550 	ibdev->dev = dev;
2551 	ibdev->bond_next_port	= 0;
2552 
2553 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2554 	ibdev->ib_dev.owner		= THIS_MODULE;
2555 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2556 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2557 	ibdev->num_ports		= num_ports;
2558 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2559 						1 : ibdev->num_ports;
2560 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2561 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2562 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2563 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2564 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2565 
2566 	if (dev->caps.userspace_caps)
2567 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2568 	else
2569 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2570 
2571 	ibdev->ib_dev.uverbs_cmd_mask	=
2572 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2573 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2574 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2575 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2576 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2577 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2578 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2579 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2580 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2581 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2582 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2583 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2584 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2585 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2586 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2587 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2588 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2589 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2590 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2591 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2592 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2593 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2594 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2595 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2596 
2597 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2598 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2599 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2600 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2601 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2602 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2603 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2604 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2605 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2606 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2607 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2608 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2609 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2610 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2611 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2612 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2613 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2614 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2615 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2616 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2617 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2618 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2619 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2620 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2621 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2622 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2623 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2624 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2625 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2626 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2627 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2628 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2629 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2630 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2631 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2632 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2633 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2634 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2635 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2636 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2637 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2638 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2639 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2640 
2641 	if (!mlx4_is_slave(ibdev->dev)) {
2642 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2643 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2644 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2645 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2646 	}
2647 
2648 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2649 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2650 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2651 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2652 
2653 		ibdev->ib_dev.uverbs_cmd_mask |=
2654 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2655 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2656 	}
2657 
2658 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2659 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2660 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2661 		ibdev->ib_dev.uverbs_cmd_mask |=
2662 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2663 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2664 	}
2665 
2666 	if (check_flow_steering_support(dev)) {
2667 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2668 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2669 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2670 
2671 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2672 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2673 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2674 	}
2675 
2676 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2677 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2678 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2679 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2680 
2681 	mlx4_ib_alloc_eqs(dev, ibdev);
2682 
2683 	spin_lock_init(&iboe->lock);
2684 
2685 	if (init_node_data(ibdev))
2686 		goto err_map;
2687 	mlx4_init_sl2vl_tbl(ibdev);
2688 
2689 	for (i = 0; i < ibdev->num_ports; ++i) {
2690 		mutex_init(&ibdev->counters_table[i].mutex);
2691 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2692 	}
2693 
2694 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2695 	for (i = 0; i < num_req_counters; ++i) {
2696 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2697 		allocated = 0;
2698 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2699 						IB_LINK_LAYER_ETHERNET) {
2700 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2701 			/* if failed to allocate a new counter, use default */
2702 			if (err)
2703 				counter_index =
2704 					mlx4_get_default_counter_index(dev,
2705 								       i + 1);
2706 			else
2707 				allocated = 1;
2708 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2709 			counter_index = mlx4_get_default_counter_index(dev,
2710 								       i + 1);
2711 		}
2712 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2713 					    GFP_KERNEL);
2714 		if (!new_counter_index) {
2715 			if (allocated)
2716 				mlx4_counter_free(ibdev->dev, counter_index);
2717 			goto err_counter;
2718 		}
2719 		new_counter_index->index = counter_index;
2720 		new_counter_index->allocated = allocated;
2721 		list_add_tail(&new_counter_index->list,
2722 			      &ibdev->counters_table[i].counters_list);
2723 		ibdev->counters_table[i].default_counter = counter_index;
2724 		pr_info("counter index %d for port %d allocated %d\n",
2725 			counter_index, i + 1, allocated);
2726 	}
2727 	if (mlx4_is_bonded(dev))
2728 		for (i = 1; i < ibdev->num_ports ; ++i) {
2729 			new_counter_index =
2730 					kmalloc(sizeof(struct counter_index),
2731 						GFP_KERNEL);
2732 			if (!new_counter_index)
2733 				goto err_counter;
2734 			new_counter_index->index = counter_index;
2735 			new_counter_index->allocated = 0;
2736 			list_add_tail(&new_counter_index->list,
2737 				      &ibdev->counters_table[i].counters_list);
2738 			ibdev->counters_table[i].default_counter =
2739 								counter_index;
2740 		}
2741 
2742 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2743 		ib_num_ports++;
2744 
2745 	spin_lock_init(&ibdev->sm_lock);
2746 	mutex_init(&ibdev->cap_mask_mutex);
2747 	INIT_LIST_HEAD(&ibdev->qp_list);
2748 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2749 
2750 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2751 	    ib_num_ports) {
2752 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2753 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2754 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2755 					    &ibdev->steer_qpn_base, 0);
2756 		if (err)
2757 			goto err_counter;
2758 
2759 		ibdev->ib_uc_qpns_bitmap =
2760 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2761 				sizeof(long),
2762 				GFP_KERNEL);
2763 		if (!ibdev->ib_uc_qpns_bitmap) {
2764 			dev_err(&dev->persist->pdev->dev,
2765 				"bit map alloc failed\n");
2766 			goto err_steer_qp_release;
2767 		}
2768 
2769 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2770 
2771 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2772 				dev, ibdev->steer_qpn_base,
2773 				ibdev->steer_qpn_base +
2774 				ibdev->steer_qpn_count - 1);
2775 		if (err)
2776 			goto err_steer_free_bitmap;
2777 	}
2778 
2779 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2780 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2781 
2782 	if (mlx4_ib_alloc_diag_counters(ibdev))
2783 		goto err_steer_free_bitmap;
2784 
2785 	if (ib_register_device(&ibdev->ib_dev, NULL))
2786 		goto err_diag_counters;
2787 
2788 	if (mlx4_ib_mad_init(ibdev))
2789 		goto err_reg;
2790 
2791 	if (mlx4_ib_init_sriov(ibdev))
2792 		goto err_mad;
2793 
2794 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2795 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2796 		if (!iboe->nb.notifier_call) {
2797 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2798 			err = register_netdevice_notifier(&iboe->nb);
2799 			if (err) {
2800 				iboe->nb.notifier_call = NULL;
2801 				goto err_notif;
2802 			}
2803 		}
2804 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2805 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2806 			if (err) {
2807 				goto err_notif;
2808 			}
2809 		}
2810 	}
2811 
2812 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2813 		if (device_create_file(&ibdev->ib_dev.dev,
2814 				       mlx4_class_attributes[j]))
2815 			goto err_notif;
2816 	}
2817 
2818 	ibdev->ib_active = true;
2819 
2820 	if (mlx4_is_mfunc(ibdev->dev))
2821 		init_pkeys(ibdev);
2822 
2823 	/* create paravirt contexts for any VFs which are active */
2824 	if (mlx4_is_master(ibdev->dev)) {
2825 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2826 			if (j == mlx4_master_func_num(ibdev->dev))
2827 				continue;
2828 			if (mlx4_is_slave_active(ibdev->dev, j))
2829 				do_slave_init(ibdev, j, 1);
2830 		}
2831 	}
2832 	return ibdev;
2833 
2834 err_notif:
2835 	if (ibdev->iboe.nb.notifier_call) {
2836 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2837 			pr_warn("failure unregistering notifier\n");
2838 		ibdev->iboe.nb.notifier_call = NULL;
2839 	}
2840 	flush_workqueue(wq);
2841 
2842 	mlx4_ib_close_sriov(ibdev);
2843 
2844 err_mad:
2845 	mlx4_ib_mad_cleanup(ibdev);
2846 
2847 err_reg:
2848 	ib_unregister_device(&ibdev->ib_dev);
2849 
2850 err_diag_counters:
2851 	mlx4_ib_diag_cleanup(ibdev);
2852 
2853 err_steer_free_bitmap:
2854 	kfree(ibdev->ib_uc_qpns_bitmap);
2855 
2856 err_steer_qp_release:
2857 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2858 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2859 				      ibdev->steer_qpn_count);
2860 err_counter:
2861 	for (i = 0; i < ibdev->num_ports; ++i)
2862 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2863 
2864 err_map:
2865 	iounmap(ibdev->uar_map);
2866 
2867 err_uar:
2868 	mlx4_uar_free(dev, &ibdev->priv_uar);
2869 
2870 err_pd:
2871 	mlx4_pd_free(dev, ibdev->priv_pdn);
2872 
2873 err_dealloc:
2874 	ib_dealloc_device(&ibdev->ib_dev);
2875 
2876 	return NULL;
2877 }
2878 
2879 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2880 {
2881 	int offset;
2882 
2883 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2884 
2885 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2886 					 dev->steer_qpn_count,
2887 					 get_count_order(count));
2888 	if (offset < 0)
2889 		return offset;
2890 
2891 	*qpn = dev->steer_qpn_base + offset;
2892 	return 0;
2893 }
2894 
2895 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2896 {
2897 	if (!qpn ||
2898 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2899 		return;
2900 
2901 	BUG_ON(qpn < dev->steer_qpn_base);
2902 
2903 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2904 			      qpn - dev->steer_qpn_base,
2905 			      get_count_order(count));
2906 }
2907 
2908 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2909 			 int is_attach)
2910 {
2911 	int err;
2912 	size_t flow_size;
2913 	struct ib_flow_attr *flow = NULL;
2914 	struct ib_flow_spec_ib *ib_spec;
2915 
2916 	if (is_attach) {
2917 		flow_size = sizeof(struct ib_flow_attr) +
2918 			    sizeof(struct ib_flow_spec_ib);
2919 		flow = kzalloc(flow_size, GFP_KERNEL);
2920 		if (!flow)
2921 			return -ENOMEM;
2922 		flow->port = mqp->port;
2923 		flow->num_of_specs = 1;
2924 		flow->size = flow_size;
2925 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2926 		ib_spec->type = IB_FLOW_SPEC_IB;
2927 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2928 		/* Add an empty rule for IB L2 */
2929 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2930 
2931 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2932 					    IB_FLOW_DOMAIN_NIC,
2933 					    MLX4_FS_REGULAR,
2934 					    &mqp->reg_id);
2935 	} else {
2936 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2937 	}
2938 	kfree(flow);
2939 	return err;
2940 }
2941 
2942 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2943 {
2944 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2945 	int p;
2946 
2947 	ibdev->ib_active = false;
2948 	flush_workqueue(wq);
2949 
2950 	mlx4_ib_close_sriov(ibdev);
2951 	mlx4_ib_mad_cleanup(ibdev);
2952 	ib_unregister_device(&ibdev->ib_dev);
2953 	mlx4_ib_diag_cleanup(ibdev);
2954 	if (ibdev->iboe.nb.notifier_call) {
2955 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2956 			pr_warn("failure unregistering notifier\n");
2957 		ibdev->iboe.nb.notifier_call = NULL;
2958 	}
2959 
2960 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2961 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2962 				      ibdev->steer_qpn_count);
2963 		kfree(ibdev->ib_uc_qpns_bitmap);
2964 	}
2965 
2966 	iounmap(ibdev->uar_map);
2967 	for (p = 0; p < ibdev->num_ports; ++p)
2968 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2969 
2970 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2971 		mlx4_CLOSE_PORT(dev, p);
2972 
2973 	mlx4_ib_free_eqs(dev, ibdev);
2974 
2975 	mlx4_uar_free(dev, &ibdev->priv_uar);
2976 	mlx4_pd_free(dev, ibdev->priv_pdn);
2977 	ib_dealloc_device(&ibdev->ib_dev);
2978 }
2979 
2980 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2981 {
2982 	struct mlx4_ib_demux_work **dm = NULL;
2983 	struct mlx4_dev *dev = ibdev->dev;
2984 	int i;
2985 	unsigned long flags;
2986 	struct mlx4_active_ports actv_ports;
2987 	unsigned int ports;
2988 	unsigned int first_port;
2989 
2990 	if (!mlx4_is_master(dev))
2991 		return;
2992 
2993 	actv_ports = mlx4_get_active_ports(dev, slave);
2994 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2995 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2996 
2997 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2998 	if (!dm) {
2999 		pr_err("failed to allocate memory for tunneling qp update\n");
3000 		return;
3001 	}
3002 
3003 	for (i = 0; i < ports; i++) {
3004 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3005 		if (!dm[i]) {
3006 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
3007 			while (--i >= 0)
3008 				kfree(dm[i]);
3009 			goto out;
3010 		}
3011 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3012 		dm[i]->port = first_port + i + 1;
3013 		dm[i]->slave = slave;
3014 		dm[i]->do_init = do_init;
3015 		dm[i]->dev = ibdev;
3016 	}
3017 	/* initialize or tear down tunnel QPs for the slave */
3018 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3019 	if (!ibdev->sriov.is_going_down) {
3020 		for (i = 0; i < ports; i++)
3021 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3022 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3023 	} else {
3024 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3025 		for (i = 0; i < ports; i++)
3026 			kfree(dm[i]);
3027 	}
3028 out:
3029 	kfree(dm);
3030 	return;
3031 }
3032 
3033 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3034 {
3035 	struct mlx4_ib_qp *mqp;
3036 	unsigned long flags_qp;
3037 	unsigned long flags_cq;
3038 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3039 	struct list_head    cq_notify_list;
3040 	struct mlx4_cq *mcq;
3041 	unsigned long flags;
3042 
3043 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3044 	INIT_LIST_HEAD(&cq_notify_list);
3045 
3046 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3047 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3048 
3049 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3050 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3051 		if (mqp->sq.tail != mqp->sq.head) {
3052 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3053 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3054 			if (send_mcq->mcq.comp &&
3055 			    mqp->ibqp.send_cq->comp_handler) {
3056 				if (!send_mcq->mcq.reset_notify_added) {
3057 					send_mcq->mcq.reset_notify_added = 1;
3058 					list_add_tail(&send_mcq->mcq.reset_notify,
3059 						      &cq_notify_list);
3060 				}
3061 			}
3062 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3063 		}
3064 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3065 		/* Now, handle the QP's receive queue */
3066 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3067 		/* no handling is needed for SRQ */
3068 		if (!mqp->ibqp.srq) {
3069 			if (mqp->rq.tail != mqp->rq.head) {
3070 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3071 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3072 				if (recv_mcq->mcq.comp &&
3073 				    mqp->ibqp.recv_cq->comp_handler) {
3074 					if (!recv_mcq->mcq.reset_notify_added) {
3075 						recv_mcq->mcq.reset_notify_added = 1;
3076 						list_add_tail(&recv_mcq->mcq.reset_notify,
3077 							      &cq_notify_list);
3078 					}
3079 				}
3080 				spin_unlock_irqrestore(&recv_mcq->lock,
3081 						       flags_cq);
3082 			}
3083 		}
3084 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3085 	}
3086 
3087 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3088 		mcq->comp(mcq);
3089 	}
3090 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3091 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3092 }
3093 
3094 static void handle_bonded_port_state_event(struct work_struct *work)
3095 {
3096 	struct ib_event_work *ew =
3097 		container_of(work, struct ib_event_work, work);
3098 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3099 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3100 	int i;
3101 	struct ib_event ibev;
3102 
3103 	kfree(ew);
3104 	spin_lock_bh(&ibdev->iboe.lock);
3105 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3106 		struct ifnet *curr_netdev = ibdev->iboe.netdevs[i];
3107 		enum ib_port_state curr_port_state;
3108 
3109 		if (!curr_netdev)
3110 			continue;
3111 
3112 		curr_port_state =
3113 			((curr_netdev->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
3114 			 curr_netdev->if_link_state == LINK_STATE_UP) ?
3115 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3116 
3117 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3118 			curr_port_state : IB_PORT_ACTIVE;
3119 	}
3120 	spin_unlock_bh(&ibdev->iboe.lock);
3121 
3122 	ibev.device = &ibdev->ib_dev;
3123 	ibev.element.port_num = 1;
3124 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3125 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3126 
3127 	ib_dispatch_event(&ibev);
3128 }
3129 
3130 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3131 {
3132 	u64 sl2vl;
3133 	int err;
3134 
3135 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3136 	if (err) {
3137 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3138 		       port, err);
3139 		sl2vl = 0;
3140 	}
3141 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3142 }
3143 
3144 static void ib_sl2vl_update_work(struct work_struct *work)
3145 {
3146 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3147 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3148 	int port = ew->port;
3149 
3150 	mlx4_ib_sl2vl_update(mdev, port);
3151 
3152 	kfree(ew);
3153 }
3154 
3155 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3156 				     int port)
3157 {
3158 	struct ib_event_work *ew;
3159 
3160 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3161 	if (ew) {
3162 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3163 		ew->port = port;
3164 		ew->ib_dev = ibdev;
3165 		queue_work(wq, &ew->work);
3166 	} else {
3167 		pr_err("failed to allocate memory for sl2vl update work\n");
3168 	}
3169 }
3170 
3171 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3172 			  enum mlx4_dev_event event, unsigned long param)
3173 {
3174 	struct ib_event ibev;
3175 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3176 	struct mlx4_eqe *eqe = NULL;
3177 	struct ib_event_work *ew;
3178 	int p = 0;
3179 
3180 	if (mlx4_is_bonded(dev) &&
3181 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3182 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3183 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3184 		if (!ew)
3185 			return;
3186 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3187 		ew->ib_dev = ibdev;
3188 		queue_work(wq, &ew->work);
3189 		return;
3190 	}
3191 
3192 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3193 		eqe = (struct mlx4_eqe *)param;
3194 	else
3195 		p = (int) param;
3196 
3197 	switch (event) {
3198 	case MLX4_DEV_EVENT_PORT_UP:
3199 		if (p > ibdev->num_ports)
3200 			return;
3201 		if (!mlx4_is_slave(dev) &&
3202 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3203 			IB_LINK_LAYER_INFINIBAND) {
3204 			if (mlx4_is_master(dev))
3205 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3206 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3207 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3208 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3209 		}
3210 		ibev.event = IB_EVENT_PORT_ACTIVE;
3211 		break;
3212 
3213 	case MLX4_DEV_EVENT_PORT_DOWN:
3214 		if (p > ibdev->num_ports)
3215 			return;
3216 		ibev.event = IB_EVENT_PORT_ERR;
3217 		break;
3218 
3219 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3220 		ibdev->ib_active = false;
3221 		ibev.event = IB_EVENT_DEVICE_FATAL;
3222 		mlx4_ib_handle_catas_error(ibdev);
3223 		break;
3224 
3225 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3226 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3227 		if (!ew) {
3228 			pr_err("failed to allocate memory for events work\n");
3229 			break;
3230 		}
3231 
3232 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3233 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3234 		ew->ib_dev = ibdev;
3235 		/* need to queue only for port owner, which uses GEN_EQE */
3236 		if (mlx4_is_master(dev))
3237 			queue_work(wq, &ew->work);
3238 		else
3239 			handle_port_mgmt_change_event(&ew->work);
3240 		return;
3241 
3242 	case MLX4_DEV_EVENT_SLAVE_INIT:
3243 		/* here, p is the slave id */
3244 		do_slave_init(ibdev, p, 1);
3245 		if (mlx4_is_master(dev)) {
3246 			int i;
3247 
3248 			for (i = 1; i <= ibdev->num_ports; i++) {
3249 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3250 					== IB_LINK_LAYER_INFINIBAND)
3251 					mlx4_ib_slave_alias_guid_event(ibdev,
3252 								       p, i,
3253 								       1);
3254 			}
3255 		}
3256 		return;
3257 
3258 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3259 		if (mlx4_is_master(dev)) {
3260 			int i;
3261 
3262 			for (i = 1; i <= ibdev->num_ports; i++) {
3263 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3264 					== IB_LINK_LAYER_INFINIBAND)
3265 					mlx4_ib_slave_alias_guid_event(ibdev,
3266 								       p, i,
3267 								       0);
3268 			}
3269 		}
3270 		/* here, p is the slave id */
3271 		do_slave_init(ibdev, p, 0);
3272 		return;
3273 
3274 	default:
3275 		return;
3276 	}
3277 
3278 	ibev.device	      = ibdev_ptr;
3279 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3280 
3281 	ib_dispatch_event(&ibev);
3282 }
3283 
3284 static struct mlx4_interface mlx4_ib_interface = {
3285 	.add		= mlx4_ib_add,
3286 	.remove		= mlx4_ib_remove,
3287 	.event		= mlx4_ib_event,
3288 	.protocol	= MLX4_PROT_IB_IPV6,
3289 	.flags		= MLX4_INTFF_BONDING
3290 };
3291 
3292 static int __init mlx4_ib_init(void)
3293 {
3294 	int err;
3295 
3296 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3297 	if (!wq)
3298 		return -ENOMEM;
3299 
3300 	err = mlx4_ib_mcg_init();
3301 	if (err)
3302 		goto clean_wq;
3303 
3304 	err = mlx4_register_interface(&mlx4_ib_interface);
3305 	if (err)
3306 		goto clean_mcg;
3307 
3308 	return 0;
3309 
3310 clean_mcg:
3311 	mlx4_ib_mcg_destroy();
3312 
3313 clean_wq:
3314 	destroy_workqueue(wq);
3315 	return err;
3316 }
3317 
3318 static void __exit mlx4_ib_cleanup(void)
3319 {
3320 	mlx4_unregister_interface(&mlx4_ib_interface);
3321 	mlx4_ib_mcg_destroy();
3322 	destroy_workqueue(wq);
3323 }
3324 
3325 module_init_order(mlx4_ib_init, SI_ORDER_SEVENTH);
3326 module_exit_order(mlx4_ib_cleanup, SI_ORDER_SEVENTH);
3327 
3328 static int
3329 mlx4ib_evhand(module_t mod, int event, void *arg)
3330 {
3331 	return (0);
3332 }
3333 
3334 static moduledata_t mlx4ib_mod = {
3335 	.name = "mlx4ib",
3336 	.evhand = mlx4ib_evhand,
3337 };
3338 
3339 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_LAST, SI_ORDER_ANY);
3340 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
3341 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);
3342 MODULE_DEPEND(mlx4ib, linuxkpi, 1, 1, 1);
3343