xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <net/ipv6.h>
43 #include <net/addrconf.h>
44 #include <net/devlink.h>
45 
46 #include <rdma/ib_smi.h>
47 #include <rdma/ib_user_verbs.h>
48 #include <rdma/ib_addr.h>
49 #include <rdma/ib_cache.h>
50 
51 #include <net/bonding.h>
52 
53 #include <linux/mlx4/driver.h>
54 #include <linux/mlx4/cmd.h>
55 #include <linux/mlx4/qp.h>
56 
57 #include "mlx4_ib.h"
58 #include <rdma/mlx4-abi.h>
59 
60 #define DRV_NAME	MLX4_IB_DRV_NAME
61 #define DRV_VERSION	"2.2-1"
62 #define DRV_RELDATE	"Feb 2014"
63 
64 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
65 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
66 #define MLX4_IB_CARD_REV_A0   0xA0
67 
68 MODULE_AUTHOR("Roland Dreier");
69 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
70 MODULE_LICENSE("Dual BSD/GPL");
71 MODULE_VERSION(DRV_VERSION);
72 
73 int mlx4_ib_sm_guid_assign = 0;
74 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
75 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
76 
77 static const char mlx4_ib_version[] =
78 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
79 	DRV_VERSION " (" DRV_RELDATE ")\n";
80 
81 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
82 
83 static struct workqueue_struct *wq;
84 
85 static void init_query_mad(struct ib_smp *mad)
86 {
87 	mad->base_version  = 1;
88 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
89 	mad->class_version = 1;
90 	mad->method	   = IB_MGMT_METHOD_GET;
91 }
92 
93 static int check_flow_steering_support(struct mlx4_dev *dev)
94 {
95 	int eth_num_ports = 0;
96 	int ib_num_ports = 0;
97 
98 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
99 
100 	if (dmfs) {
101 		int i;
102 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
103 			eth_num_ports++;
104 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
105 			ib_num_ports++;
106 		dmfs &= (!ib_num_ports ||
107 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
108 			(!eth_num_ports ||
109 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
110 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
111 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
112 			dmfs = 0;
113 		}
114 	}
115 	return dmfs;
116 }
117 
118 static int num_ib_ports(struct mlx4_dev *dev)
119 {
120 	int ib_ports = 0;
121 	int i;
122 
123 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
124 		ib_ports++;
125 
126 	return ib_ports;
127 }
128 
129 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
130 {
131 	struct mlx4_ib_dev *ibdev = to_mdev(device);
132 	struct net_device *dev;
133 
134 	rcu_read_lock();
135 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
136 
137 	if (dev) {
138 		if (mlx4_is_bonded(ibdev->dev)) {
139 			struct net_device *upper = NULL;
140 
141 			upper = netdev_master_upper_dev_get_rcu(dev);
142 			if (upper) {
143 				struct net_device *active;
144 
145 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
146 				if (active)
147 					dev = active;
148 			}
149 		}
150 	}
151 	if (dev)
152 		dev_hold(dev);
153 
154 	rcu_read_unlock();
155 	return dev;
156 }
157 
158 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
159 				  struct mlx4_ib_dev *ibdev,
160 				  u8 port_num)
161 {
162 	struct mlx4_cmd_mailbox *mailbox;
163 	int err;
164 	struct mlx4_dev *dev = ibdev->dev;
165 	int i;
166 	union ib_gid *gid_tbl;
167 
168 	mailbox = mlx4_alloc_cmd_mailbox(dev);
169 	if (IS_ERR(mailbox))
170 		return -ENOMEM;
171 
172 	gid_tbl = mailbox->buf;
173 
174 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
175 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
176 
177 	err = mlx4_cmd(dev, mailbox->dma,
178 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
179 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
180 		       MLX4_CMD_WRAPPED);
181 	if (mlx4_is_bonded(dev))
182 		err += mlx4_cmd(dev, mailbox->dma,
183 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
184 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
185 				MLX4_CMD_WRAPPED);
186 
187 	mlx4_free_cmd_mailbox(dev, mailbox);
188 	return err;
189 }
190 
191 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
192 				     struct mlx4_ib_dev *ibdev,
193 				     u8 port_num)
194 {
195 	struct mlx4_cmd_mailbox *mailbox;
196 	int err;
197 	struct mlx4_dev *dev = ibdev->dev;
198 	int i;
199 	struct {
200 		union ib_gid	gid;
201 		__be32		rsrvd1[2];
202 		__be16		rsrvd2;
203 		u8		type;
204 		u8		version;
205 		__be32		rsrvd3;
206 	} *gid_tbl;
207 
208 	mailbox = mlx4_alloc_cmd_mailbox(dev);
209 	if (IS_ERR(mailbox))
210 		return -ENOMEM;
211 
212 	gid_tbl = mailbox->buf;
213 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
214 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
215 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
216 			gid_tbl[i].version = 2;
217 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
218 				gid_tbl[i].type = 1;
219 			else
220 				memset(&gid_tbl[i].gid, 0, 12);
221 		}
222 	}
223 
224 	err = mlx4_cmd(dev, mailbox->dma,
225 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
226 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
227 		       MLX4_CMD_WRAPPED);
228 	if (mlx4_is_bonded(dev))
229 		err += mlx4_cmd(dev, mailbox->dma,
230 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
231 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
232 				MLX4_CMD_WRAPPED);
233 
234 	mlx4_free_cmd_mailbox(dev, mailbox);
235 	return err;
236 }
237 
238 static int mlx4_ib_update_gids(struct gid_entry *gids,
239 			       struct mlx4_ib_dev *ibdev,
240 			       u8 port_num)
241 {
242 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
243 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
244 
245 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
246 }
247 
248 static int mlx4_ib_add_gid(struct ib_device *device,
249 			   u8 port_num,
250 			   unsigned int index,
251 			   const union ib_gid *gid,
252 			   const struct ib_gid_attr *attr,
253 			   void **context)
254 {
255 	struct mlx4_ib_dev *ibdev = to_mdev(device);
256 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
257 	struct mlx4_port_gid_table   *port_gid_table;
258 	int free = -1, found = -1;
259 	int ret = 0;
260 	int hw_update = 0;
261 	int i;
262 	struct gid_entry *gids = NULL;
263 
264 	if (!rdma_cap_roce_gid_table(device, port_num))
265 		return -EINVAL;
266 
267 	if (port_num > MLX4_MAX_PORTS)
268 		return -EINVAL;
269 
270 	if (!context)
271 		return -EINVAL;
272 
273 	port_gid_table = &iboe->gids[port_num - 1];
274 	spin_lock_bh(&iboe->lock);
275 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
276 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
277 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
278 			found = i;
279 			break;
280 		}
281 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
282 			free = i; /* HW has space */
283 	}
284 
285 	if (found < 0) {
286 		if (free < 0) {
287 			ret = -ENOSPC;
288 		} else {
289 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
290 			if (!port_gid_table->gids[free].ctx) {
291 				ret = -ENOMEM;
292 			} else {
293 				*context = port_gid_table->gids[free].ctx;
294 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
295 				port_gid_table->gids[free].gid_type = attr->gid_type;
296 				port_gid_table->gids[free].ctx->real_index = free;
297 				port_gid_table->gids[free].ctx->refcount = 1;
298 				hw_update = 1;
299 			}
300 		}
301 	} else {
302 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
303 		*context = ctx;
304 		ctx->refcount++;
305 	}
306 	if (!ret && hw_update) {
307 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
308 		if (!gids) {
309 			ret = -ENOMEM;
310 		} else {
311 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
312 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
313 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
314 			}
315 		}
316 	}
317 	spin_unlock_bh(&iboe->lock);
318 
319 	if (!ret && hw_update) {
320 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
321 		kfree(gids);
322 	}
323 
324 	return ret;
325 }
326 
327 static int mlx4_ib_del_gid(struct ib_device *device,
328 			   u8 port_num,
329 			   unsigned int index,
330 			   void **context)
331 {
332 	struct gid_cache_context *ctx = *context;
333 	struct mlx4_ib_dev *ibdev = to_mdev(device);
334 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
335 	struct mlx4_port_gid_table   *port_gid_table;
336 	int ret = 0;
337 	int hw_update = 0;
338 	struct gid_entry *gids = NULL;
339 
340 	if (!rdma_cap_roce_gid_table(device, port_num))
341 		return -EINVAL;
342 
343 	if (port_num > MLX4_MAX_PORTS)
344 		return -EINVAL;
345 
346 	port_gid_table = &iboe->gids[port_num - 1];
347 	spin_lock_bh(&iboe->lock);
348 	if (ctx) {
349 		ctx->refcount--;
350 		if (!ctx->refcount) {
351 			unsigned int real_index = ctx->real_index;
352 
353 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
354 			kfree(port_gid_table->gids[real_index].ctx);
355 			port_gid_table->gids[real_index].ctx = NULL;
356 			hw_update = 1;
357 		}
358 	}
359 	if (!ret && hw_update) {
360 		int i;
361 
362 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
363 		if (!gids) {
364 			ret = -ENOMEM;
365 		} else {
366 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
367 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
368 		}
369 	}
370 	spin_unlock_bh(&iboe->lock);
371 
372 	if (!ret && hw_update) {
373 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
374 		kfree(gids);
375 	}
376 	return ret;
377 }
378 
379 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
380 				    u8 port_num, int index)
381 {
382 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
383 	struct gid_cache_context *ctx = NULL;
384 	union ib_gid gid;
385 	struct mlx4_port_gid_table   *port_gid_table;
386 	int real_index = -EINVAL;
387 	int i;
388 	int ret;
389 	unsigned long flags;
390 	struct ib_gid_attr attr;
391 
392 	if (port_num > MLX4_MAX_PORTS)
393 		return -EINVAL;
394 
395 	if (mlx4_is_bonded(ibdev->dev))
396 		port_num = 1;
397 
398 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
399 		return index;
400 
401 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
402 	if (ret)
403 		return ret;
404 
405 	if (attr.ndev)
406 		dev_put(attr.ndev);
407 
408 	if (!memcmp(&gid, &zgid, sizeof(gid)))
409 		return -EINVAL;
410 
411 	spin_lock_irqsave(&iboe->lock, flags);
412 	port_gid_table = &iboe->gids[port_num - 1];
413 
414 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
415 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
416 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
417 			ctx = port_gid_table->gids[i].ctx;
418 			break;
419 		}
420 	if (ctx)
421 		real_index = ctx->real_index;
422 	spin_unlock_irqrestore(&iboe->lock, flags);
423 	return real_index;
424 }
425 
426 static int mlx4_ib_query_device(struct ib_device *ibdev,
427 				struct ib_device_attr *props,
428 				struct ib_udata *uhw)
429 {
430 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
431 	struct ib_smp *in_mad  = NULL;
432 	struct ib_smp *out_mad = NULL;
433 	int err = -ENOMEM;
434 	int have_ib_ports;
435 	struct mlx4_uverbs_ex_query_device cmd;
436 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
437 	struct mlx4_clock_params clock_params;
438 
439 	if (uhw->inlen) {
440 		if (uhw->inlen < sizeof(cmd))
441 			return -EINVAL;
442 
443 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
444 		if (err)
445 			return err;
446 
447 		if (cmd.comp_mask)
448 			return -EINVAL;
449 
450 		if (cmd.reserved)
451 			return -EINVAL;
452 	}
453 
454 	resp.response_length = offsetof(typeof(resp), response_length) +
455 		sizeof(resp.response_length);
456 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
457 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
458 	if (!in_mad || !out_mad)
459 		goto out;
460 
461 	init_query_mad(in_mad);
462 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
463 
464 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
465 			   1, NULL, NULL, in_mad, out_mad);
466 	if (err)
467 		goto out;
468 
469 	memset(props, 0, sizeof *props);
470 
471 	have_ib_ports = num_ib_ports(dev->dev);
472 
473 	props->fw_ver = dev->dev->caps.fw_ver;
474 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
475 		IB_DEVICE_PORT_ACTIVE_EVENT		|
476 		IB_DEVICE_SYS_IMAGE_GUID		|
477 		IB_DEVICE_RC_RNR_NAK_GEN		|
478 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
479 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
480 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
481 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
482 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
484 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
485 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
486 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
488 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
489 	if (dev->dev->caps.max_gso_sz &&
490 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
491 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
492 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
493 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
494 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
495 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
496 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
497 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
498 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
499 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
500 		props->device_cap_flags |= IB_DEVICE_XRC;
501 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
502 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
503 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
504 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
505 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
506 		else
507 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
508 	}
509 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
510 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
511 
512 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
513 
514 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
515 		0xffffff;
516 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
517 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
518 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
519 
520 	props->max_mr_size	   = ~0ull;
521 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
522 	props->max_qp		   = dev->dev->quotas.qp;
523 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
524 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
525 					 dev->dev->caps.max_rq_sg);
526 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
527 	props->max_cq		   = dev->dev->quotas.cq;
528 	props->max_cqe		   = dev->dev->caps.max_cqes;
529 	props->max_mr		   = dev->dev->quotas.mpt;
530 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
531 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
532 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
533 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
534 	props->max_srq		   = dev->dev->quotas.srq;
535 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
536 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
537 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
538 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
539 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
540 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
541 	props->masked_atomic_cap   = props->atomic_cap;
542 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
543 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
544 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
545 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
546 					   props->max_mcast_grp;
547 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
548 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
549 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
550 
551 	if (!mlx4_is_slave(dev->dev))
552 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
553 
554 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
555 		resp.response_length += sizeof(resp.hca_core_clock_offset);
556 		if (!err && !mlx4_is_slave(dev->dev)) {
557 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
558 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
559 		}
560 	}
561 
562 	if (uhw->outlen) {
563 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
564 		if (err)
565 			goto out;
566 	}
567 out:
568 	kfree(in_mad);
569 	kfree(out_mad);
570 
571 	return err;
572 }
573 
574 static enum rdma_link_layer
575 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
576 {
577 	struct mlx4_dev *dev = to_mdev(device)->dev;
578 
579 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
580 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
581 }
582 
583 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
584 			      struct ib_port_attr *props, int netw_view)
585 {
586 	struct ib_smp *in_mad  = NULL;
587 	struct ib_smp *out_mad = NULL;
588 	int ext_active_speed;
589 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
590 	int err = -ENOMEM;
591 
592 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
593 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
594 	if (!in_mad || !out_mad)
595 		goto out;
596 
597 	init_query_mad(in_mad);
598 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
599 	in_mad->attr_mod = cpu_to_be32(port);
600 
601 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
602 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
603 
604 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
605 				in_mad, out_mad);
606 	if (err)
607 		goto out;
608 
609 
610 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
611 	props->lmc		= out_mad->data[34] & 0x7;
612 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
613 	props->sm_sl		= out_mad->data[36] & 0xf;
614 	props->state		= out_mad->data[32] & 0xf;
615 	props->phys_state	= out_mad->data[33] >> 4;
616 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
617 	if (netw_view)
618 		props->gid_tbl_len = out_mad->data[50];
619 	else
620 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
621 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
622 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
623 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
624 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
625 	props->active_width	= out_mad->data[31] & 0xf;
626 	props->active_speed	= out_mad->data[35] >> 4;
627 	props->max_mtu		= out_mad->data[41] & 0xf;
628 	props->active_mtu	= out_mad->data[36] >> 4;
629 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
630 	props->max_vl_num	= out_mad->data[37] >> 4;
631 	props->init_type_reply	= out_mad->data[41] >> 4;
632 
633 	/* Check if extended speeds (EDR/FDR/...) are supported */
634 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
635 		ext_active_speed = out_mad->data[62] >> 4;
636 
637 		switch (ext_active_speed) {
638 		case 1:
639 			props->active_speed = IB_SPEED_FDR;
640 			break;
641 		case 2:
642 			props->active_speed = IB_SPEED_EDR;
643 			break;
644 		}
645 	}
646 
647 	/* If reported active speed is QDR, check if is FDR-10 */
648 	if (props->active_speed == IB_SPEED_QDR) {
649 		init_query_mad(in_mad);
650 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
651 		in_mad->attr_mod = cpu_to_be32(port);
652 
653 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
654 				   NULL, NULL, in_mad, out_mad);
655 		if (err)
656 			goto out;
657 
658 		/* Checking LinkSpeedActive for FDR-10 */
659 		if (out_mad->data[15] & 0x1)
660 			props->active_speed = IB_SPEED_FDR10;
661 	}
662 
663 	/* Avoid wrong speed value returned by FW if the IB link is down. */
664 	if (props->state == IB_PORT_DOWN)
665 		 props->active_speed = IB_SPEED_SDR;
666 
667 out:
668 	kfree(in_mad);
669 	kfree(out_mad);
670 	return err;
671 }
672 
673 static u8 state_to_phys_state(enum ib_port_state state)
674 {
675 	return state == IB_PORT_ACTIVE ? 5 : 3;
676 }
677 
678 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
679 			       struct ib_port_attr *props, int netw_view)
680 {
681 
682 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
683 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
684 	struct net_device *ndev;
685 	enum ib_mtu tmp;
686 	struct mlx4_cmd_mailbox *mailbox;
687 	int err = 0;
688 	int is_bonded = mlx4_is_bonded(mdev->dev);
689 
690 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
691 	if (IS_ERR(mailbox))
692 		return PTR_ERR(mailbox);
693 
694 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
695 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
696 			   MLX4_CMD_WRAPPED);
697 	if (err)
698 		goto out;
699 
700 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
701 						IB_WIDTH_4X : IB_WIDTH_1X;
702 	props->active_speed	= IB_SPEED_QDR;
703 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
704 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
705 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
706 	props->pkey_tbl_len	= 1;
707 	props->max_mtu		= IB_MTU_4096;
708 	props->max_vl_num	= 2;
709 	props->state		= IB_PORT_DOWN;
710 	props->phys_state	= state_to_phys_state(props->state);
711 	props->active_mtu	= IB_MTU_256;
712 	spin_lock_bh(&iboe->lock);
713 	ndev = iboe->netdevs[port - 1];
714 	if (ndev && is_bonded) {
715 		rcu_read_lock(); /* required to get upper dev */
716 		ndev = netdev_master_upper_dev_get_rcu(ndev);
717 		rcu_read_unlock();
718 	}
719 	if (!ndev)
720 		goto out_unlock;
721 
722 	tmp = iboe_get_mtu(ndev->mtu);
723 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
724 
725 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
726 					IB_PORT_ACTIVE : IB_PORT_DOWN;
727 	props->phys_state	= state_to_phys_state(props->state);
728 out_unlock:
729 	spin_unlock_bh(&iboe->lock);
730 out:
731 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
732 	return err;
733 }
734 
735 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
736 			 struct ib_port_attr *props, int netw_view)
737 {
738 	int err;
739 
740 	memset(props, 0, sizeof *props);
741 
742 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
743 		ib_link_query_port(ibdev, port, props, netw_view) :
744 				eth_link_query_port(ibdev, port, props, netw_view);
745 
746 	return err;
747 }
748 
749 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
750 			      struct ib_port_attr *props)
751 {
752 	/* returns host view */
753 	return __mlx4_ib_query_port(ibdev, port, props, 0);
754 }
755 
756 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
757 			union ib_gid *gid, int netw_view)
758 {
759 	struct ib_smp *in_mad  = NULL;
760 	struct ib_smp *out_mad = NULL;
761 	int err = -ENOMEM;
762 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
763 	int clear = 0;
764 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
765 
766 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
767 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
768 	if (!in_mad || !out_mad)
769 		goto out;
770 
771 	init_query_mad(in_mad);
772 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
773 	in_mad->attr_mod = cpu_to_be32(port);
774 
775 	if (mlx4_is_mfunc(dev->dev) && netw_view)
776 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
777 
778 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
779 	if (err)
780 		goto out;
781 
782 	memcpy(gid->raw, out_mad->data + 8, 8);
783 
784 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
785 		if (index) {
786 			/* For any index > 0, return the null guid */
787 			err = 0;
788 			clear = 1;
789 			goto out;
790 		}
791 	}
792 
793 	init_query_mad(in_mad);
794 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
795 	in_mad->attr_mod = cpu_to_be32(index / 8);
796 
797 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
798 			   NULL, NULL, in_mad, out_mad);
799 	if (err)
800 		goto out;
801 
802 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
803 
804 out:
805 	if (clear)
806 		memset(gid->raw + 8, 0, 8);
807 	kfree(in_mad);
808 	kfree(out_mad);
809 	return err;
810 }
811 
812 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
813 			     union ib_gid *gid)
814 {
815 	int ret;
816 
817 	if (rdma_protocol_ib(ibdev, port))
818 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
819 
820 	if (!rdma_protocol_roce(ibdev, port))
821 		return -ENODEV;
822 
823 	if (!rdma_cap_roce_gid_table(ibdev, port))
824 		return -ENODEV;
825 
826 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
827 	if (ret == -EAGAIN) {
828 		memcpy(gid, &zgid, sizeof(*gid));
829 		return 0;
830 	}
831 
832 	return ret;
833 }
834 
835 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
836 {
837 	union sl2vl_tbl_to_u64 sl2vl64;
838 	struct ib_smp *in_mad  = NULL;
839 	struct ib_smp *out_mad = NULL;
840 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
841 	int err = -ENOMEM;
842 	int jj;
843 
844 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
845 		*sl2vl_tbl = 0;
846 		return 0;
847 	}
848 
849 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
850 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
851 	if (!in_mad || !out_mad)
852 		goto out;
853 
854 	init_query_mad(in_mad);
855 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
856 	in_mad->attr_mod = 0;
857 
858 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
859 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
860 
861 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
862 			   in_mad, out_mad);
863 	if (err)
864 		goto out;
865 
866 	for (jj = 0; jj < 8; jj++)
867 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
868 	*sl2vl_tbl = sl2vl64.sl64;
869 
870 out:
871 	kfree(in_mad);
872 	kfree(out_mad);
873 	return err;
874 }
875 
876 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
877 {
878 	u64 sl2vl;
879 	int i;
880 	int err;
881 
882 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
883 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
884 			continue;
885 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
886 		if (err) {
887 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
888 			       i, err);
889 			sl2vl = 0;
890 		}
891 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
892 	}
893 }
894 
895 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
896 			 u16 *pkey, int netw_view)
897 {
898 	struct ib_smp *in_mad  = NULL;
899 	struct ib_smp *out_mad = NULL;
900 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
901 	int err = -ENOMEM;
902 
903 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
904 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
905 	if (!in_mad || !out_mad)
906 		goto out;
907 
908 	init_query_mad(in_mad);
909 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
910 	in_mad->attr_mod = cpu_to_be32(index / 32);
911 
912 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
913 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
914 
915 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
916 			   in_mad, out_mad);
917 	if (err)
918 		goto out;
919 
920 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
921 
922 out:
923 	kfree(in_mad);
924 	kfree(out_mad);
925 	return err;
926 }
927 
928 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
929 {
930 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
931 }
932 
933 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
934 				 struct ib_device_modify *props)
935 {
936 	struct mlx4_cmd_mailbox *mailbox;
937 	unsigned long flags;
938 
939 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
940 		return -EOPNOTSUPP;
941 
942 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
943 		return 0;
944 
945 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
946 		return -EOPNOTSUPP;
947 
948 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
949 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
950 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
951 
952 	/*
953 	 * If possible, pass node desc to FW, so it can generate
954 	 * a 144 trap.  If cmd fails, just ignore.
955 	 */
956 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
957 	if (IS_ERR(mailbox))
958 		return 0;
959 
960 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
961 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
962 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
963 
964 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
965 
966 	return 0;
967 }
968 
969 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
970 			    u32 cap_mask)
971 {
972 	struct mlx4_cmd_mailbox *mailbox;
973 	int err;
974 
975 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
976 	if (IS_ERR(mailbox))
977 		return PTR_ERR(mailbox);
978 
979 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
980 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
981 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
982 	} else {
983 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
984 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
985 	}
986 
987 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
988 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
989 		       MLX4_CMD_WRAPPED);
990 
991 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
992 	return err;
993 }
994 
995 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
996 			       struct ib_port_modify *props)
997 {
998 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
999 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1000 	struct ib_port_attr attr;
1001 	u32 cap_mask;
1002 	int err;
1003 
1004 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1005 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1006 	 * violations and port capabilities are not meaningful.
1007 	 */
1008 	if (is_eth)
1009 		return 0;
1010 
1011 	mutex_lock(&mdev->cap_mask_mutex);
1012 
1013 	err = mlx4_ib_query_port(ibdev, port, &attr);
1014 	if (err)
1015 		goto out;
1016 
1017 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1018 		~props->clr_port_cap_mask;
1019 
1020 	err = mlx4_ib_SET_PORT(mdev, port,
1021 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1022 			       cap_mask);
1023 
1024 out:
1025 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1026 	return err;
1027 }
1028 
1029 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1030 						  struct ib_udata *udata)
1031 {
1032 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1033 	struct mlx4_ib_ucontext *context;
1034 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1035 	struct mlx4_ib_alloc_ucontext_resp resp;
1036 	int err;
1037 
1038 	if (!dev->ib_active)
1039 		return ERR_PTR(-EAGAIN);
1040 
1041 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1042 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1043 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1044 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1045 	} else {
1046 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1047 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1048 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1049 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1050 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1051 	}
1052 
1053 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1054 	if (!context)
1055 		return ERR_PTR(-ENOMEM);
1056 
1057 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1058 	if (err) {
1059 		kfree(context);
1060 		return ERR_PTR(err);
1061 	}
1062 
1063 	INIT_LIST_HEAD(&context->db_page_list);
1064 	mutex_init(&context->db_page_mutex);
1065 
1066 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1067 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1068 	else
1069 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1070 
1071 	if (err) {
1072 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1073 		kfree(context);
1074 		return ERR_PTR(-EFAULT);
1075 	}
1076 
1077 	return &context->ibucontext;
1078 }
1079 
1080 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1081 {
1082 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1083 
1084 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1085 	kfree(context);
1086 
1087 	return 0;
1088 }
1089 
1090 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1091 {
1092 	/* vma_open is called when a new VMA is created on top of our VMA.
1093 	 * This is done through either mremap flow or split_vma (usually due
1094 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1095 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1096 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1097 	 * calling us again and trying to do incorrect actions. We assume that
1098 	 * the original vma size is exactly a single page that there will be no
1099 	 * "splitting" operations on.
1100 	 */
1101 	area->vm_ops = NULL;
1102 }
1103 
1104 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1105 {
1106 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1107 
1108 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1109 	 * file itself is closed, therefore no sync is needed with the regular
1110 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1111 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1112 	 * The close operation is usually called under mm->mmap_sem except when
1113 	 * process is exiting.  The exiting case is handled explicitly as part
1114 	 * of mlx4_ib_disassociate_ucontext.
1115 	 */
1116 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1117 				area->vm_private_data;
1118 
1119 	/* set the vma context pointer to null in the mlx4_ib driver's private
1120 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1121 	 */
1122 	mlx4_ib_vma_priv_data->vma = NULL;
1123 }
1124 
1125 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1126 	.open = mlx4_ib_vma_open,
1127 	.close = mlx4_ib_vma_close
1128 };
1129 
1130 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1131 {
1132 	int i;
1133 	int ret = 0;
1134 	struct vm_area_struct *vma;
1135 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1136 	struct task_struct *owning_process  = NULL;
1137 	struct mm_struct   *owning_mm       = NULL;
1138 
1139 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1140 	if (!owning_process)
1141 		return;
1142 
1143 	owning_mm = get_task_mm(owning_process);
1144 	if (!owning_mm) {
1145 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1146 		while (1) {
1147 			/* make sure that task is dead before returning, it may
1148 			 * prevent a rare case of module down in parallel to a
1149 			 * call to mlx4_ib_vma_close.
1150 			 */
1151 			put_task_struct(owning_process);
1152 			msleep(1);
1153 			owning_process = get_pid_task(ibcontext->tgid,
1154 						      PIDTYPE_PID);
1155 			if (!owning_process ||
1156 			    owning_process->state == TASK_DEAD) {
1157 				pr_info("disassociate ucontext done, task was terminated\n");
1158 				/* in case task was dead need to release the task struct */
1159 				if (owning_process)
1160 					put_task_struct(owning_process);
1161 				return;
1162 			}
1163 		}
1164 	}
1165 
1166 	/* need to protect from a race on closing the vma as part of
1167 	 * mlx4_ib_vma_close().
1168 	 */
1169 	down_read(&owning_mm->mmap_sem);
1170 	for (i = 0; i < HW_BAR_COUNT; i++) {
1171 		vma = context->hw_bar_info[i].vma;
1172 		if (!vma)
1173 			continue;
1174 
1175 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1176 				   context->hw_bar_info[i].vma->vm_start,
1177 				   PAGE_SIZE);
1178 		if (ret) {
1179 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1180 			BUG_ON(1);
1181 		}
1182 
1183 		/* context going to be destroyed, should not access ops any more */
1184 		context->hw_bar_info[i].vma->vm_ops = NULL;
1185 	}
1186 
1187 	up_read(&owning_mm->mmap_sem);
1188 	mmput(owning_mm);
1189 	put_task_struct(owning_process);
1190 }
1191 
1192 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1193 				 struct mlx4_ib_vma_private_data *vma_private_data)
1194 {
1195 	vma_private_data->vma = vma;
1196 	vma->vm_private_data = vma_private_data;
1197 	vma->vm_ops =  &mlx4_ib_vm_ops;
1198 }
1199 
1200 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1201 {
1202 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1203 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1204 
1205 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1206 		return -EINVAL;
1207 
1208 	if (vma->vm_pgoff == 0) {
1209 		/* We prevent double mmaping on same context */
1210 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1211 			return -EINVAL;
1212 
1213 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1214 
1215 		if (io_remap_pfn_range(vma, vma->vm_start,
1216 				       to_mucontext(context)->uar.pfn,
1217 				       PAGE_SIZE, vma->vm_page_prot))
1218 			return -EAGAIN;
1219 
1220 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1221 
1222 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1223 		/* We prevent double mmaping on same context */
1224 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1225 			return -EINVAL;
1226 
1227 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1228 
1229 		if (io_remap_pfn_range(vma, vma->vm_start,
1230 				       to_mucontext(context)->uar.pfn +
1231 				       dev->dev->caps.num_uars,
1232 				       PAGE_SIZE, vma->vm_page_prot))
1233 			return -EAGAIN;
1234 
1235 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1236 
1237 	} else if (vma->vm_pgoff == 3) {
1238 		struct mlx4_clock_params params;
1239 		int ret;
1240 
1241 		/* We prevent double mmaping on same context */
1242 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1243 			return -EINVAL;
1244 
1245 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1246 
1247 		if (ret)
1248 			return ret;
1249 
1250 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1251 		if (io_remap_pfn_range(vma, vma->vm_start,
1252 				       (pci_resource_start(dev->dev->persist->pdev,
1253 							   params.bar) +
1254 					params.offset)
1255 				       >> PAGE_SHIFT,
1256 				       PAGE_SIZE, vma->vm_page_prot))
1257 			return -EAGAIN;
1258 
1259 		mlx4_ib_set_vma_data(vma,
1260 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1261 	} else {
1262 		return -EINVAL;
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1269 				      struct ib_ucontext *context,
1270 				      struct ib_udata *udata)
1271 {
1272 	struct mlx4_ib_pd *pd;
1273 	int err;
1274 
1275 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1276 	if (!pd)
1277 		return ERR_PTR(-ENOMEM);
1278 
1279 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1280 	if (err) {
1281 		kfree(pd);
1282 		return ERR_PTR(err);
1283 	}
1284 
1285 	if (context)
1286 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1287 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1288 			kfree(pd);
1289 			return ERR_PTR(-EFAULT);
1290 		}
1291 
1292 	return &pd->ibpd;
1293 }
1294 
1295 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1296 {
1297 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1298 	kfree(pd);
1299 
1300 	return 0;
1301 }
1302 
1303 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1304 					  struct ib_ucontext *context,
1305 					  struct ib_udata *udata)
1306 {
1307 	struct mlx4_ib_xrcd *xrcd;
1308 	struct ib_cq_init_attr cq_attr = {};
1309 	int err;
1310 
1311 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1312 		return ERR_PTR(-ENOSYS);
1313 
1314 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1315 	if (!xrcd)
1316 		return ERR_PTR(-ENOMEM);
1317 
1318 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1319 	if (err)
1320 		goto err1;
1321 
1322 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1323 	if (IS_ERR(xrcd->pd)) {
1324 		err = PTR_ERR(xrcd->pd);
1325 		goto err2;
1326 	}
1327 
1328 	cq_attr.cqe = 1;
1329 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1330 	if (IS_ERR(xrcd->cq)) {
1331 		err = PTR_ERR(xrcd->cq);
1332 		goto err3;
1333 	}
1334 
1335 	return &xrcd->ibxrcd;
1336 
1337 err3:
1338 	ib_dealloc_pd(xrcd->pd);
1339 err2:
1340 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1341 err1:
1342 	kfree(xrcd);
1343 	return ERR_PTR(err);
1344 }
1345 
1346 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1347 {
1348 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1349 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1350 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1351 	kfree(xrcd);
1352 
1353 	return 0;
1354 }
1355 
1356 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1357 {
1358 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1359 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1360 	struct mlx4_ib_gid_entry *ge;
1361 
1362 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1363 	if (!ge)
1364 		return -ENOMEM;
1365 
1366 	ge->gid = *gid;
1367 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1368 		ge->port = mqp->port;
1369 		ge->added = 1;
1370 	}
1371 
1372 	mutex_lock(&mqp->mutex);
1373 	list_add_tail(&ge->list, &mqp->gid_list);
1374 	mutex_unlock(&mqp->mutex);
1375 
1376 	return 0;
1377 }
1378 
1379 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1380 					  struct mlx4_ib_counters *ctr_table)
1381 {
1382 	struct counter_index *counter, *tmp_count;
1383 
1384 	mutex_lock(&ctr_table->mutex);
1385 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1386 				 list) {
1387 		if (counter->allocated)
1388 			mlx4_counter_free(ibdev->dev, counter->index);
1389 		list_del(&counter->list);
1390 		kfree(counter);
1391 	}
1392 	mutex_unlock(&ctr_table->mutex);
1393 }
1394 
1395 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1396 		   union ib_gid *gid)
1397 {
1398 	struct net_device *ndev;
1399 	int ret = 0;
1400 
1401 	if (!mqp->port)
1402 		return 0;
1403 
1404 	spin_lock_bh(&mdev->iboe.lock);
1405 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1406 	if (ndev)
1407 		dev_hold(ndev);
1408 	spin_unlock_bh(&mdev->iboe.lock);
1409 
1410 	if (ndev) {
1411 		ret = 1;
1412 		dev_put(ndev);
1413 	}
1414 
1415 	return ret;
1416 }
1417 
1418 struct mlx4_ib_steering {
1419 	struct list_head list;
1420 	struct mlx4_flow_reg_id reg_id;
1421 	union ib_gid gid;
1422 };
1423 
1424 #define LAST_ETH_FIELD vlan_tag
1425 #define LAST_IB_FIELD sl
1426 #define LAST_IPV4_FIELD dst_ip
1427 #define LAST_TCP_UDP_FIELD src_port
1428 
1429 /* Field is the last supported field */
1430 #define FIELDS_NOT_SUPPORTED(filter, field)\
1431 	memchr_inv((void *)&filter.field  +\
1432 		   sizeof(filter.field), 0,\
1433 		   sizeof(filter) -\
1434 		   offsetof(typeof(filter), field) -\
1435 		   sizeof(filter.field))
1436 
1437 static int parse_flow_attr(struct mlx4_dev *dev,
1438 			   u32 qp_num,
1439 			   union ib_flow_spec *ib_spec,
1440 			   struct _rule_hw *mlx4_spec)
1441 {
1442 	enum mlx4_net_trans_rule_id type;
1443 
1444 	switch (ib_spec->type) {
1445 	case IB_FLOW_SPEC_ETH:
1446 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1447 			return -ENOTSUPP;
1448 
1449 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1450 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1451 		       ETH_ALEN);
1452 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1453 		       ETH_ALEN);
1454 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1455 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1456 		break;
1457 	case IB_FLOW_SPEC_IB:
1458 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1459 			return -ENOTSUPP;
1460 
1461 		type = MLX4_NET_TRANS_RULE_ID_IB;
1462 		mlx4_spec->ib.l3_qpn =
1463 			cpu_to_be32(qp_num);
1464 		mlx4_spec->ib.qpn_mask =
1465 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1466 		break;
1467 
1468 
1469 	case IB_FLOW_SPEC_IPV4:
1470 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1471 			return -ENOTSUPP;
1472 
1473 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1474 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1475 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1476 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1477 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1478 		break;
1479 
1480 	case IB_FLOW_SPEC_TCP:
1481 	case IB_FLOW_SPEC_UDP:
1482 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1483 			return -ENOTSUPP;
1484 
1485 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1486 					MLX4_NET_TRANS_RULE_ID_TCP :
1487 					MLX4_NET_TRANS_RULE_ID_UDP;
1488 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1489 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1490 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1491 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1492 		break;
1493 
1494 	default:
1495 		return -EINVAL;
1496 	}
1497 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1498 	    mlx4_hw_rule_sz(dev, type) < 0)
1499 		return -EINVAL;
1500 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1501 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1502 	return mlx4_hw_rule_sz(dev, type);
1503 }
1504 
1505 struct default_rules {
1506 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1507 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1508 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1509 	__u8  link_layer;
1510 };
1511 static const struct default_rules default_table[] = {
1512 	{
1513 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1514 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1515 		.rules_create_list = {IB_FLOW_SPEC_IB},
1516 		.link_layer = IB_LINK_LAYER_INFINIBAND
1517 	}
1518 };
1519 
1520 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1521 					 struct ib_flow_attr *flow_attr)
1522 {
1523 	int i, j, k;
1524 	void *ib_flow;
1525 	const struct default_rules *pdefault_rules = default_table;
1526 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1527 
1528 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1529 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1530 		memset(&field_types, 0, sizeof(field_types));
1531 
1532 		if (link_layer != pdefault_rules->link_layer)
1533 			continue;
1534 
1535 		ib_flow = flow_attr + 1;
1536 		/* we assume the specs are sorted */
1537 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1538 		     j < flow_attr->num_of_specs; k++) {
1539 			union ib_flow_spec *current_flow =
1540 				(union ib_flow_spec *)ib_flow;
1541 
1542 			/* same layer but different type */
1543 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1544 			     (pdefault_rules->mandatory_fields[k] &
1545 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1546 			    (current_flow->type !=
1547 			     pdefault_rules->mandatory_fields[k]))
1548 				goto out;
1549 
1550 			/* same layer, try match next one */
1551 			if (current_flow->type ==
1552 			    pdefault_rules->mandatory_fields[k]) {
1553 				j++;
1554 				ib_flow +=
1555 					((union ib_flow_spec *)ib_flow)->size;
1556 			}
1557 		}
1558 
1559 		ib_flow = flow_attr + 1;
1560 		for (j = 0; j < flow_attr->num_of_specs;
1561 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1562 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1563 				/* same layer and same type */
1564 				if (((union ib_flow_spec *)ib_flow)->type ==
1565 				    pdefault_rules->mandatory_not_fields[k])
1566 					goto out;
1567 
1568 		return i;
1569 	}
1570 out:
1571 	return -1;
1572 }
1573 
1574 static int __mlx4_ib_create_default_rules(
1575 		struct mlx4_ib_dev *mdev,
1576 		struct ib_qp *qp,
1577 		const struct default_rules *pdefault_rules,
1578 		struct _rule_hw *mlx4_spec) {
1579 	int size = 0;
1580 	int i;
1581 
1582 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1583 		int ret;
1584 		union ib_flow_spec ib_spec;
1585 		switch (pdefault_rules->rules_create_list[i]) {
1586 		case 0:
1587 			/* no rule */
1588 			continue;
1589 		case IB_FLOW_SPEC_IB:
1590 			ib_spec.type = IB_FLOW_SPEC_IB;
1591 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1592 
1593 			break;
1594 		default:
1595 			/* invalid rule */
1596 			return -EINVAL;
1597 		}
1598 		/* We must put empty rule, qpn is being ignored */
1599 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1600 				      mlx4_spec);
1601 		if (ret < 0) {
1602 			pr_info("invalid parsing\n");
1603 			return -EINVAL;
1604 		}
1605 
1606 		mlx4_spec = (void *)mlx4_spec + ret;
1607 		size += ret;
1608 	}
1609 	return size;
1610 }
1611 
1612 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1613 			  int domain,
1614 			  enum mlx4_net_trans_promisc_mode flow_type,
1615 			  u64 *reg_id)
1616 {
1617 	int ret, i;
1618 	int size = 0;
1619 	void *ib_flow;
1620 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1621 	struct mlx4_cmd_mailbox *mailbox;
1622 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1623 	int default_flow;
1624 
1625 	static const u16 __mlx4_domain[] = {
1626 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1627 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1628 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1629 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1630 	};
1631 
1632 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1633 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1634 		return -EINVAL;
1635 	}
1636 
1637 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1638 		pr_err("Invalid domain value %d\n", domain);
1639 		return -EINVAL;
1640 	}
1641 
1642 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1643 		return -EINVAL;
1644 
1645 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1646 	if (IS_ERR(mailbox))
1647 		return PTR_ERR(mailbox);
1648 	ctrl = mailbox->buf;
1649 
1650 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1651 				 flow_attr->priority);
1652 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1653 	ctrl->port = flow_attr->port;
1654 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1655 
1656 	ib_flow = flow_attr + 1;
1657 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1658 	/* Add default flows */
1659 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1660 	if (default_flow >= 0) {
1661 		ret = __mlx4_ib_create_default_rules(
1662 				mdev, qp, default_table + default_flow,
1663 				mailbox->buf + size);
1664 		if (ret < 0) {
1665 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1666 			return -EINVAL;
1667 		}
1668 		size += ret;
1669 	}
1670 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1671 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1672 				      mailbox->buf + size);
1673 		if (ret < 0) {
1674 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1675 			return -EINVAL;
1676 		}
1677 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1678 		size += ret;
1679 	}
1680 
1681 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1682 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1683 			   MLX4_CMD_WRAPPED);
1684 	if (ret == -ENOMEM)
1685 		pr_err("mcg table is full. Fail to register network rule.\n");
1686 	else if (ret == -ENXIO)
1687 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1688 	else if (ret)
1689 		pr_err("Invalid argument. Fail to register network rule.\n");
1690 
1691 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1692 	return ret;
1693 }
1694 
1695 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1696 {
1697 	int err;
1698 	err = mlx4_cmd(dev, reg_id, 0, 0,
1699 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1700 		       MLX4_CMD_WRAPPED);
1701 	if (err)
1702 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1703 		       reg_id);
1704 	return err;
1705 }
1706 
1707 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1708 				    u64 *reg_id)
1709 {
1710 	void *ib_flow;
1711 	union ib_flow_spec *ib_spec;
1712 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1713 	int err = 0;
1714 
1715 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1716 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1717 		return 0; /* do nothing */
1718 
1719 	ib_flow = flow_attr + 1;
1720 	ib_spec = (union ib_flow_spec *)ib_flow;
1721 
1722 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1723 		return 0; /* do nothing */
1724 
1725 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1726 				    flow_attr->port, qp->qp_num,
1727 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1728 				    reg_id);
1729 	return err;
1730 }
1731 
1732 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1733 				      struct ib_flow_attr *flow_attr,
1734 				      enum mlx4_net_trans_promisc_mode *type)
1735 {
1736 	int err = 0;
1737 
1738 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1739 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1740 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1741 		return -EOPNOTSUPP;
1742 	}
1743 
1744 	if (flow_attr->num_of_specs == 0) {
1745 		type[0] = MLX4_FS_MC_SNIFFER;
1746 		type[1] = MLX4_FS_UC_SNIFFER;
1747 	} else {
1748 		union ib_flow_spec *ib_spec;
1749 
1750 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1751 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1752 			return -EINVAL;
1753 
1754 		/* if all is zero than MC and UC */
1755 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1756 			type[0] = MLX4_FS_MC_SNIFFER;
1757 			type[1] = MLX4_FS_UC_SNIFFER;
1758 		} else {
1759 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1760 					    ib_spec->eth.mask.dst_mac[1],
1761 					    ib_spec->eth.mask.dst_mac[2],
1762 					    ib_spec->eth.mask.dst_mac[3],
1763 					    ib_spec->eth.mask.dst_mac[4],
1764 					    ib_spec->eth.mask.dst_mac[5]};
1765 
1766 			/* Above xor was only on MC bit, non empty mask is valid
1767 			 * only if this bit is set and rest are zero.
1768 			 */
1769 			if (!is_zero_ether_addr(&mac[0]))
1770 				return -EINVAL;
1771 
1772 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1773 				type[0] = MLX4_FS_MC_SNIFFER;
1774 			else
1775 				type[0] = MLX4_FS_UC_SNIFFER;
1776 		}
1777 	}
1778 
1779 	return err;
1780 }
1781 
1782 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1783 				    struct ib_flow_attr *flow_attr,
1784 				    int domain)
1785 {
1786 	int err = 0, i = 0, j = 0;
1787 	struct mlx4_ib_flow *mflow;
1788 	enum mlx4_net_trans_promisc_mode type[2];
1789 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1790 	int is_bonded = mlx4_is_bonded(dev);
1791 
1792 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1793 		return ERR_PTR(-EINVAL);
1794 
1795 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1796 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1797 		return ERR_PTR(-EOPNOTSUPP);
1798 
1799 	memset(type, 0, sizeof(type));
1800 
1801 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1802 	if (!mflow) {
1803 		err = -ENOMEM;
1804 		goto err_free;
1805 	}
1806 
1807 	switch (flow_attr->type) {
1808 	case IB_FLOW_ATTR_NORMAL:
1809 		/* If dont trap flag (continue match) is set, under specific
1810 		 * condition traffic be replicated to given qp,
1811 		 * without stealing it
1812 		 */
1813 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1814 			err = mlx4_ib_add_dont_trap_rule(dev,
1815 							 flow_attr,
1816 							 type);
1817 			if (err)
1818 				goto err_free;
1819 		} else {
1820 			type[0] = MLX4_FS_REGULAR;
1821 		}
1822 		break;
1823 
1824 	case IB_FLOW_ATTR_ALL_DEFAULT:
1825 		type[0] = MLX4_FS_ALL_DEFAULT;
1826 		break;
1827 
1828 	case IB_FLOW_ATTR_MC_DEFAULT:
1829 		type[0] = MLX4_FS_MC_DEFAULT;
1830 		break;
1831 
1832 	case IB_FLOW_ATTR_SNIFFER:
1833 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1834 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1835 		break;
1836 
1837 	default:
1838 		err = -EINVAL;
1839 		goto err_free;
1840 	}
1841 
1842 	while (i < ARRAY_SIZE(type) && type[i]) {
1843 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1844 					    &mflow->reg_id[i].id);
1845 		if (err)
1846 			goto err_create_flow;
1847 		if (is_bonded) {
1848 			/* Application always sees one port so the mirror rule
1849 			 * must be on port #2
1850 			 */
1851 			flow_attr->port = 2;
1852 			err = __mlx4_ib_create_flow(qp, flow_attr,
1853 						    domain, type[j],
1854 						    &mflow->reg_id[j].mirror);
1855 			flow_attr->port = 1;
1856 			if (err)
1857 				goto err_create_flow;
1858 			j++;
1859 		}
1860 
1861 		i++;
1862 	}
1863 
1864 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1865 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1866 					       &mflow->reg_id[i].id);
1867 		if (err)
1868 			goto err_create_flow;
1869 
1870 		if (is_bonded) {
1871 			flow_attr->port = 2;
1872 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1873 						       &mflow->reg_id[j].mirror);
1874 			flow_attr->port = 1;
1875 			if (err)
1876 				goto err_create_flow;
1877 			j++;
1878 		}
1879 		/* function to create mirror rule */
1880 		i++;
1881 	}
1882 
1883 	return &mflow->ibflow;
1884 
1885 err_create_flow:
1886 	while (i) {
1887 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1888 					     mflow->reg_id[i].id);
1889 		i--;
1890 	}
1891 
1892 	while (j) {
1893 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1894 					     mflow->reg_id[j].mirror);
1895 		j--;
1896 	}
1897 err_free:
1898 	kfree(mflow);
1899 	return ERR_PTR(err);
1900 }
1901 
1902 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1903 {
1904 	int err, ret = 0;
1905 	int i = 0;
1906 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1907 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1908 
1909 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1910 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1911 		if (err)
1912 			ret = err;
1913 		if (mflow->reg_id[i].mirror) {
1914 			err = __mlx4_ib_destroy_flow(mdev->dev,
1915 						     mflow->reg_id[i].mirror);
1916 			if (err)
1917 				ret = err;
1918 		}
1919 		i++;
1920 	}
1921 
1922 	kfree(mflow);
1923 	return ret;
1924 }
1925 
1926 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1927 {
1928 	int err;
1929 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1930 	struct mlx4_dev	*dev = mdev->dev;
1931 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1932 	struct mlx4_ib_steering *ib_steering = NULL;
1933 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1934 	struct mlx4_flow_reg_id	reg_id;
1935 
1936 	if (mdev->dev->caps.steering_mode ==
1937 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1938 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1939 		if (!ib_steering)
1940 			return -ENOMEM;
1941 	}
1942 
1943 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1944 				    !!(mqp->flags &
1945 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1946 				    prot, &reg_id.id);
1947 	if (err) {
1948 		pr_err("multicast attach op failed, err %d\n", err);
1949 		goto err_malloc;
1950 	}
1951 
1952 	reg_id.mirror = 0;
1953 	if (mlx4_is_bonded(dev)) {
1954 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1955 					    (mqp->port == 1) ? 2 : 1,
1956 					    !!(mqp->flags &
1957 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1958 					    prot, &reg_id.mirror);
1959 		if (err)
1960 			goto err_add;
1961 	}
1962 
1963 	err = add_gid_entry(ibqp, gid);
1964 	if (err)
1965 		goto err_add;
1966 
1967 	if (ib_steering) {
1968 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1969 		ib_steering->reg_id = reg_id;
1970 		mutex_lock(&mqp->mutex);
1971 		list_add(&ib_steering->list, &mqp->steering_rules);
1972 		mutex_unlock(&mqp->mutex);
1973 	}
1974 	return 0;
1975 
1976 err_add:
1977 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1978 			      prot, reg_id.id);
1979 	if (reg_id.mirror)
1980 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1981 				      prot, reg_id.mirror);
1982 err_malloc:
1983 	kfree(ib_steering);
1984 
1985 	return err;
1986 }
1987 
1988 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1989 {
1990 	struct mlx4_ib_gid_entry *ge;
1991 	struct mlx4_ib_gid_entry *tmp;
1992 	struct mlx4_ib_gid_entry *ret = NULL;
1993 
1994 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1995 		if (!memcmp(raw, ge->gid.raw, 16)) {
1996 			ret = ge;
1997 			break;
1998 		}
1999 	}
2000 
2001 	return ret;
2002 }
2003 
2004 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2005 {
2006 	int err;
2007 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2008 	struct mlx4_dev *dev = mdev->dev;
2009 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2010 	struct net_device *ndev;
2011 	struct mlx4_ib_gid_entry *ge;
2012 	struct mlx4_flow_reg_id reg_id = {0, 0};
2013 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2014 
2015 	if (mdev->dev->caps.steering_mode ==
2016 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2017 		struct mlx4_ib_steering *ib_steering;
2018 
2019 		mutex_lock(&mqp->mutex);
2020 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2021 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2022 				list_del(&ib_steering->list);
2023 				break;
2024 			}
2025 		}
2026 		mutex_unlock(&mqp->mutex);
2027 		if (&ib_steering->list == &mqp->steering_rules) {
2028 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2029 			return -EINVAL;
2030 		}
2031 		reg_id = ib_steering->reg_id;
2032 		kfree(ib_steering);
2033 	}
2034 
2035 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2036 				    prot, reg_id.id);
2037 	if (err)
2038 		return err;
2039 
2040 	if (mlx4_is_bonded(dev)) {
2041 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2042 					    prot, reg_id.mirror);
2043 		if (err)
2044 			return err;
2045 	}
2046 
2047 	mutex_lock(&mqp->mutex);
2048 	ge = find_gid_entry(mqp, gid->raw);
2049 	if (ge) {
2050 		spin_lock_bh(&mdev->iboe.lock);
2051 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2052 		if (ndev)
2053 			dev_hold(ndev);
2054 		spin_unlock_bh(&mdev->iboe.lock);
2055 		if (ndev)
2056 			dev_put(ndev);
2057 		list_del(&ge->list);
2058 		kfree(ge);
2059 	} else
2060 		pr_warn("could not find mgid entry\n");
2061 
2062 	mutex_unlock(&mqp->mutex);
2063 
2064 	return 0;
2065 }
2066 
2067 static int init_node_data(struct mlx4_ib_dev *dev)
2068 {
2069 	struct ib_smp *in_mad  = NULL;
2070 	struct ib_smp *out_mad = NULL;
2071 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2072 	int err = -ENOMEM;
2073 
2074 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2075 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2076 	if (!in_mad || !out_mad)
2077 		goto out;
2078 
2079 	init_query_mad(in_mad);
2080 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2081 	if (mlx4_is_master(dev->dev))
2082 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2083 
2084 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2085 	if (err)
2086 		goto out;
2087 
2088 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2089 
2090 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2091 
2092 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2093 	if (err)
2094 		goto out;
2095 
2096 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2097 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2098 
2099 out:
2100 	kfree(in_mad);
2101 	kfree(out_mad);
2102 	return err;
2103 }
2104 
2105 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2106 			char *buf)
2107 {
2108 	struct mlx4_ib_dev *dev =
2109 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2110 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2111 }
2112 
2113 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2114 			char *buf)
2115 {
2116 	struct mlx4_ib_dev *dev =
2117 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2118 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2119 }
2120 
2121 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2122 			  char *buf)
2123 {
2124 	struct mlx4_ib_dev *dev =
2125 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2126 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2127 		       dev->dev->board_id);
2128 }
2129 
2130 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2131 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2132 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2133 
2134 static struct device_attribute *mlx4_class_attributes[] = {
2135 	&dev_attr_hw_rev,
2136 	&dev_attr_hca_type,
2137 	&dev_attr_board_id
2138 };
2139 
2140 struct diag_counter {
2141 	const char *name;
2142 	u32 offset;
2143 };
2144 
2145 #define DIAG_COUNTER(_name, _offset)			\
2146 	{ .name = #_name, .offset = _offset }
2147 
2148 static const struct diag_counter diag_basic[] = {
2149 	DIAG_COUNTER(rq_num_lle, 0x00),
2150 	DIAG_COUNTER(sq_num_lle, 0x04),
2151 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2152 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2153 	DIAG_COUNTER(rq_num_lpe, 0x18),
2154 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2155 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2156 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2157 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2158 	DIAG_COUNTER(sq_num_bre, 0x34),
2159 	DIAG_COUNTER(sq_num_rire, 0x44),
2160 	DIAG_COUNTER(rq_num_rire, 0x48),
2161 	DIAG_COUNTER(sq_num_rae, 0x4C),
2162 	DIAG_COUNTER(rq_num_rae, 0x50),
2163 	DIAG_COUNTER(sq_num_roe, 0x54),
2164 	DIAG_COUNTER(sq_num_tree, 0x5C),
2165 	DIAG_COUNTER(sq_num_rree, 0x64),
2166 	DIAG_COUNTER(rq_num_rnr, 0x68),
2167 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2168 	DIAG_COUNTER(rq_num_oos, 0x100),
2169 	DIAG_COUNTER(sq_num_oos, 0x104),
2170 };
2171 
2172 static const struct diag_counter diag_ext[] = {
2173 	DIAG_COUNTER(rq_num_dup, 0x130),
2174 	DIAG_COUNTER(sq_num_to, 0x134),
2175 };
2176 
2177 static const struct diag_counter diag_device_only[] = {
2178 	DIAG_COUNTER(num_cqovf, 0x1A0),
2179 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2180 };
2181 
2182 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2183 						    u8 port_num)
2184 {
2185 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2186 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2187 
2188 	if (!diag[!!port_num].name)
2189 		return NULL;
2190 
2191 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2192 					  diag[!!port_num].num_counters,
2193 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2194 }
2195 
2196 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2197 				struct rdma_hw_stats *stats,
2198 				u8 port, int index)
2199 {
2200 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2201 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2202 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2203 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2204 	int ret;
2205 	int i;
2206 
2207 	ret = mlx4_query_diag_counters(dev->dev,
2208 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2209 				       diag[!!port].offset, hw_value,
2210 				       diag[!!port].num_counters, port);
2211 
2212 	if (ret)
2213 		return ret;
2214 
2215 	for (i = 0; i < diag[!!port].num_counters; i++)
2216 		stats->value[i] = hw_value[i];
2217 
2218 	return diag[!!port].num_counters;
2219 }
2220 
2221 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2222 					 const char ***name,
2223 					 u32 **offset,
2224 					 u32 *num,
2225 					 bool port)
2226 {
2227 	u32 num_counters;
2228 
2229 	num_counters = ARRAY_SIZE(diag_basic);
2230 
2231 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2232 		num_counters += ARRAY_SIZE(diag_ext);
2233 
2234 	if (!port)
2235 		num_counters += ARRAY_SIZE(diag_device_only);
2236 
2237 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2238 	if (!*name)
2239 		return -ENOMEM;
2240 
2241 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2242 	if (!*offset)
2243 		goto err_name;
2244 
2245 	*num = num_counters;
2246 
2247 	return 0;
2248 
2249 err_name:
2250 	kfree(*name);
2251 	return -ENOMEM;
2252 }
2253 
2254 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2255 				       const char **name,
2256 				       u32 *offset,
2257 				       bool port)
2258 {
2259 	int i;
2260 	int j;
2261 
2262 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2263 		name[i] = diag_basic[i].name;
2264 		offset[i] = diag_basic[i].offset;
2265 	}
2266 
2267 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2268 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2269 			name[j] = diag_ext[i].name;
2270 			offset[j] = diag_ext[i].offset;
2271 		}
2272 	}
2273 
2274 	if (!port) {
2275 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2276 			name[j] = diag_device_only[i].name;
2277 			offset[j] = diag_device_only[i].offset;
2278 		}
2279 	}
2280 }
2281 
2282 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2283 {
2284 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2285 	int i;
2286 	int ret;
2287 	bool per_port = !!(ibdev->dev->caps.flags2 &
2288 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2289 
2290 	if (mlx4_is_slave(ibdev->dev))
2291 		return 0;
2292 
2293 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2294 		/* i == 1 means we are building port counters */
2295 		if (i && !per_port)
2296 			continue;
2297 
2298 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2299 						    &diag[i].offset,
2300 						    &diag[i].num_counters, i);
2301 		if (ret)
2302 			goto err_alloc;
2303 
2304 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2305 					   diag[i].offset, i);
2306 	}
2307 
2308 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2309 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2310 
2311 	return 0;
2312 
2313 err_alloc:
2314 	if (i) {
2315 		kfree(diag[i - 1].name);
2316 		kfree(diag[i - 1].offset);
2317 	}
2318 
2319 	return ret;
2320 }
2321 
2322 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2323 {
2324 	int i;
2325 
2326 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2327 		kfree(ibdev->diag_counters[i].offset);
2328 		kfree(ibdev->diag_counters[i].name);
2329 	}
2330 }
2331 
2332 #define MLX4_IB_INVALID_MAC	((u64)-1)
2333 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2334 			       struct net_device *dev,
2335 			       int port)
2336 {
2337 	u64 new_smac = 0;
2338 	u64 release_mac = MLX4_IB_INVALID_MAC;
2339 	struct mlx4_ib_qp *qp;
2340 
2341 	read_lock(&dev_base_lock);
2342 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2343 	read_unlock(&dev_base_lock);
2344 
2345 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2346 
2347 	/* no need for update QP1 and mac registration in non-SRIOV */
2348 	if (!mlx4_is_mfunc(ibdev->dev))
2349 		return;
2350 
2351 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2352 	qp = ibdev->qp1_proxy[port - 1];
2353 	if (qp) {
2354 		int new_smac_index;
2355 		u64 old_smac;
2356 		struct mlx4_update_qp_params update_params;
2357 
2358 		mutex_lock(&qp->mutex);
2359 		old_smac = qp->pri.smac;
2360 		if (new_smac == old_smac)
2361 			goto unlock;
2362 
2363 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2364 
2365 		if (new_smac_index < 0)
2366 			goto unlock;
2367 
2368 		update_params.smac_index = new_smac_index;
2369 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2370 				   &update_params)) {
2371 			release_mac = new_smac;
2372 			goto unlock;
2373 		}
2374 		/* if old port was zero, no mac was yet registered for this QP */
2375 		if (qp->pri.smac_port)
2376 			release_mac = old_smac;
2377 		qp->pri.smac = new_smac;
2378 		qp->pri.smac_port = port;
2379 		qp->pri.smac_index = new_smac_index;
2380 	}
2381 
2382 unlock:
2383 	if (release_mac != MLX4_IB_INVALID_MAC)
2384 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2385 	if (qp)
2386 		mutex_unlock(&qp->mutex);
2387 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2388 }
2389 
2390 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2391 				 struct net_device *dev,
2392 				 unsigned long event)
2393 
2394 {
2395 	struct mlx4_ib_iboe *iboe;
2396 	int update_qps_port = -1;
2397 	int port;
2398 
2399 	ASSERT_RTNL();
2400 
2401 	iboe = &ibdev->iboe;
2402 
2403 	spin_lock_bh(&iboe->lock);
2404 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2405 
2406 		iboe->netdevs[port - 1] =
2407 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2408 
2409 		if (dev == iboe->netdevs[port - 1] &&
2410 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2411 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2412 			update_qps_port = port;
2413 
2414 	}
2415 	spin_unlock_bh(&iboe->lock);
2416 
2417 	if (update_qps_port > 0)
2418 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2419 }
2420 
2421 static int mlx4_ib_netdev_event(struct notifier_block *this,
2422 				unsigned long event, void *ptr)
2423 {
2424 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2425 	struct mlx4_ib_dev *ibdev;
2426 
2427 	if (!net_eq(dev_net(dev), &init_net))
2428 		return NOTIFY_DONE;
2429 
2430 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2431 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2432 
2433 	return NOTIFY_DONE;
2434 }
2435 
2436 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2437 {
2438 	int port;
2439 	int slave;
2440 	int i;
2441 
2442 	if (mlx4_is_master(ibdev->dev)) {
2443 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2444 		     ++slave) {
2445 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2446 				for (i = 0;
2447 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2448 				     ++i) {
2449 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2450 					/* master has the identity virt2phys pkey mapping */
2451 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2452 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2453 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2454 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2455 				}
2456 			}
2457 		}
2458 		/* initialize pkey cache */
2459 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2460 			for (i = 0;
2461 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2462 			     ++i)
2463 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2464 					(i) ? 0 : 0xFFFF;
2465 		}
2466 	}
2467 }
2468 
2469 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2470 {
2471 	int i, j, eq = 0, total_eqs = 0;
2472 
2473 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2474 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2475 	if (!ibdev->eq_table)
2476 		return;
2477 
2478 	for (i = 1; i <= dev->caps.num_ports; i++) {
2479 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2480 		     j++, total_eqs++) {
2481 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2482 				continue;
2483 			ibdev->eq_table[eq] = total_eqs;
2484 			if (!mlx4_assign_eq(dev, i,
2485 					    &ibdev->eq_table[eq]))
2486 				eq++;
2487 			else
2488 				ibdev->eq_table[eq] = -1;
2489 		}
2490 	}
2491 
2492 	for (i = eq; i < dev->caps.num_comp_vectors;
2493 	     ibdev->eq_table[i++] = -1)
2494 		;
2495 
2496 	/* Advertise the new number of EQs to clients */
2497 	ibdev->ib_dev.num_comp_vectors = eq;
2498 }
2499 
2500 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2501 {
2502 	int i;
2503 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2504 
2505 	/* no eqs were allocated */
2506 	if (!ibdev->eq_table)
2507 		return;
2508 
2509 	/* Reset the advertised EQ number */
2510 	ibdev->ib_dev.num_comp_vectors = 0;
2511 
2512 	for (i = 0; i < total_eqs; i++)
2513 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2514 
2515 	kfree(ibdev->eq_table);
2516 	ibdev->eq_table = NULL;
2517 }
2518 
2519 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2520 			       struct ib_port_immutable *immutable)
2521 {
2522 	struct ib_port_attr attr;
2523 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2524 	int err;
2525 
2526 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2527 	if (err)
2528 		return err;
2529 
2530 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2531 	immutable->gid_tbl_len = attr.gid_tbl_len;
2532 
2533 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2534 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2535 	} else {
2536 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2537 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2538 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2539 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2540 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2541 	}
2542 
2543 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2544 
2545 	return 0;
2546 }
2547 
2548 static void get_fw_ver_str(struct ib_device *device, char *str,
2549 			   size_t str_len)
2550 {
2551 	struct mlx4_ib_dev *dev =
2552 		container_of(device, struct mlx4_ib_dev, ib_dev);
2553 	snprintf(str, str_len, "%d.%d.%d",
2554 		 (int) (dev->dev->caps.fw_ver >> 32),
2555 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2556 		 (int) dev->dev->caps.fw_ver & 0xffff);
2557 }
2558 
2559 static void *mlx4_ib_add(struct mlx4_dev *dev)
2560 {
2561 	struct mlx4_ib_dev *ibdev;
2562 	int num_ports = 0;
2563 	int i, j;
2564 	int err;
2565 	struct mlx4_ib_iboe *iboe;
2566 	int ib_num_ports = 0;
2567 	int num_req_counters;
2568 	int allocated;
2569 	u32 counter_index;
2570 	struct counter_index *new_counter_index = NULL;
2571 
2572 	pr_info_once("%s", mlx4_ib_version);
2573 
2574 	num_ports = 0;
2575 	mlx4_foreach_ib_transport_port(i, dev)
2576 		num_ports++;
2577 
2578 	/* No point in registering a device with no ports... */
2579 	if (num_ports == 0)
2580 		return NULL;
2581 
2582 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2583 	if (!ibdev) {
2584 		dev_err(&dev->persist->pdev->dev,
2585 			"Device struct alloc failed\n");
2586 		return NULL;
2587 	}
2588 
2589 	iboe = &ibdev->iboe;
2590 
2591 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2592 		goto err_dealloc;
2593 
2594 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2595 		goto err_pd;
2596 
2597 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2598 				 PAGE_SIZE);
2599 	if (!ibdev->uar_map)
2600 		goto err_uar;
2601 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2602 
2603 	ibdev->dev = dev;
2604 	ibdev->bond_next_port	= 0;
2605 
2606 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2607 	ibdev->ib_dev.owner		= THIS_MODULE;
2608 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2609 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2610 	ibdev->num_ports		= num_ports;
2611 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2612 						1 : ibdev->num_ports;
2613 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2614 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2615 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2616 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2617 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2618 
2619 	if (dev->caps.userspace_caps)
2620 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2621 	else
2622 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2623 
2624 	ibdev->ib_dev.uverbs_cmd_mask	=
2625 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2626 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2627 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2628 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2629 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2630 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2631 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2632 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2633 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2634 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2635 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2636 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2637 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2638 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2639 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2640 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2641 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2642 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2643 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2644 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2645 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2646 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2647 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2648 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2649 
2650 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2651 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2652 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2653 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2654 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2655 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2656 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2657 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2658 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2659 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2660 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2661 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2662 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2663 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2664 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2665 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2666 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2667 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2668 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2669 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2670 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2671 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2672 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2673 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2674 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2675 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2676 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2677 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2678 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2679 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2680 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2681 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2682 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2683 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2684 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2685 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2686 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2687 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2688 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2689 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2690 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2691 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2692 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2693 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2694 
2695 	if (!mlx4_is_slave(ibdev->dev)) {
2696 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2697 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2698 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2699 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2700 	}
2701 
2702 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2703 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2704 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2705 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2706 
2707 		ibdev->ib_dev.uverbs_cmd_mask |=
2708 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2709 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2710 	}
2711 
2712 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2713 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2714 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2715 		ibdev->ib_dev.uverbs_cmd_mask |=
2716 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2717 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2718 	}
2719 
2720 	if (check_flow_steering_support(dev)) {
2721 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2722 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2723 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2724 
2725 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2726 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2727 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2728 	}
2729 
2730 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2731 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2732 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2733 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2734 
2735 	mlx4_ib_alloc_eqs(dev, ibdev);
2736 
2737 	spin_lock_init(&iboe->lock);
2738 
2739 	if (init_node_data(ibdev))
2740 		goto err_map;
2741 	mlx4_init_sl2vl_tbl(ibdev);
2742 
2743 	for (i = 0; i < ibdev->num_ports; ++i) {
2744 		mutex_init(&ibdev->counters_table[i].mutex);
2745 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2746 	}
2747 
2748 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2749 	for (i = 0; i < num_req_counters; ++i) {
2750 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2751 		allocated = 0;
2752 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2753 						IB_LINK_LAYER_ETHERNET) {
2754 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2755 			/* if failed to allocate a new counter, use default */
2756 			if (err)
2757 				counter_index =
2758 					mlx4_get_default_counter_index(dev,
2759 								       i + 1);
2760 			else
2761 				allocated = 1;
2762 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2763 			counter_index = mlx4_get_default_counter_index(dev,
2764 								       i + 1);
2765 		}
2766 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2767 					    GFP_KERNEL);
2768 		if (!new_counter_index) {
2769 			if (allocated)
2770 				mlx4_counter_free(ibdev->dev, counter_index);
2771 			goto err_counter;
2772 		}
2773 		new_counter_index->index = counter_index;
2774 		new_counter_index->allocated = allocated;
2775 		list_add_tail(&new_counter_index->list,
2776 			      &ibdev->counters_table[i].counters_list);
2777 		ibdev->counters_table[i].default_counter = counter_index;
2778 		pr_info("counter index %d for port %d allocated %d\n",
2779 			counter_index, i + 1, allocated);
2780 	}
2781 	if (mlx4_is_bonded(dev))
2782 		for (i = 1; i < ibdev->num_ports ; ++i) {
2783 			new_counter_index =
2784 					kmalloc(sizeof(struct counter_index),
2785 						GFP_KERNEL);
2786 			if (!new_counter_index)
2787 				goto err_counter;
2788 			new_counter_index->index = counter_index;
2789 			new_counter_index->allocated = 0;
2790 			list_add_tail(&new_counter_index->list,
2791 				      &ibdev->counters_table[i].counters_list);
2792 			ibdev->counters_table[i].default_counter =
2793 								counter_index;
2794 		}
2795 
2796 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2797 		ib_num_ports++;
2798 
2799 	spin_lock_init(&ibdev->sm_lock);
2800 	mutex_init(&ibdev->cap_mask_mutex);
2801 	INIT_LIST_HEAD(&ibdev->qp_list);
2802 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2803 
2804 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2805 	    ib_num_ports) {
2806 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2807 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2808 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2809 					    &ibdev->steer_qpn_base, 0);
2810 		if (err)
2811 			goto err_counter;
2812 
2813 		ibdev->ib_uc_qpns_bitmap =
2814 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2815 				sizeof(long),
2816 				GFP_KERNEL);
2817 		if (!ibdev->ib_uc_qpns_bitmap) {
2818 			dev_err(&dev->persist->pdev->dev,
2819 				"bit map alloc failed\n");
2820 			goto err_steer_qp_release;
2821 		}
2822 
2823 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2824 
2825 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2826 				dev, ibdev->steer_qpn_base,
2827 				ibdev->steer_qpn_base +
2828 				ibdev->steer_qpn_count - 1);
2829 		if (err)
2830 			goto err_steer_free_bitmap;
2831 	}
2832 
2833 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2834 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2835 
2836 	if (mlx4_ib_alloc_diag_counters(ibdev))
2837 		goto err_steer_free_bitmap;
2838 
2839 	if (ib_register_device(&ibdev->ib_dev, NULL))
2840 		goto err_diag_counters;
2841 
2842 	if (mlx4_ib_mad_init(ibdev))
2843 		goto err_reg;
2844 
2845 	if (mlx4_ib_init_sriov(ibdev))
2846 		goto err_mad;
2847 
2848 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2849 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2850 		if (!iboe->nb.notifier_call) {
2851 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2852 			err = register_netdevice_notifier(&iboe->nb);
2853 			if (err) {
2854 				iboe->nb.notifier_call = NULL;
2855 				goto err_notif;
2856 			}
2857 		}
2858 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2859 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2860 			if (err) {
2861 				goto err_notif;
2862 			}
2863 		}
2864 	}
2865 
2866 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2867 		if (device_create_file(&ibdev->ib_dev.dev,
2868 				       mlx4_class_attributes[j]))
2869 			goto err_notif;
2870 	}
2871 
2872 	ibdev->ib_active = true;
2873 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2874 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2875 					 &ibdev->ib_dev);
2876 
2877 	if (mlx4_is_mfunc(ibdev->dev))
2878 		init_pkeys(ibdev);
2879 
2880 	/* create paravirt contexts for any VFs which are active */
2881 	if (mlx4_is_master(ibdev->dev)) {
2882 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2883 			if (j == mlx4_master_func_num(ibdev->dev))
2884 				continue;
2885 			if (mlx4_is_slave_active(ibdev->dev, j))
2886 				do_slave_init(ibdev, j, 1);
2887 		}
2888 	}
2889 	return ibdev;
2890 
2891 err_notif:
2892 	if (ibdev->iboe.nb.notifier_call) {
2893 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2894 			pr_warn("failure unregistering notifier\n");
2895 		ibdev->iboe.nb.notifier_call = NULL;
2896 	}
2897 	flush_workqueue(wq);
2898 
2899 	mlx4_ib_close_sriov(ibdev);
2900 
2901 err_mad:
2902 	mlx4_ib_mad_cleanup(ibdev);
2903 
2904 err_reg:
2905 	ib_unregister_device(&ibdev->ib_dev);
2906 
2907 err_diag_counters:
2908 	mlx4_ib_diag_cleanup(ibdev);
2909 
2910 err_steer_free_bitmap:
2911 	kfree(ibdev->ib_uc_qpns_bitmap);
2912 
2913 err_steer_qp_release:
2914 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2915 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2916 				      ibdev->steer_qpn_count);
2917 err_counter:
2918 	for (i = 0; i < ibdev->num_ports; ++i)
2919 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2920 
2921 err_map:
2922 	iounmap(ibdev->uar_map);
2923 
2924 err_uar:
2925 	mlx4_uar_free(dev, &ibdev->priv_uar);
2926 
2927 err_pd:
2928 	mlx4_pd_free(dev, ibdev->priv_pdn);
2929 
2930 err_dealloc:
2931 	ib_dealloc_device(&ibdev->ib_dev);
2932 
2933 	return NULL;
2934 }
2935 
2936 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2937 {
2938 	int offset;
2939 
2940 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2941 
2942 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2943 					 dev->steer_qpn_count,
2944 					 get_count_order(count));
2945 	if (offset < 0)
2946 		return offset;
2947 
2948 	*qpn = dev->steer_qpn_base + offset;
2949 	return 0;
2950 }
2951 
2952 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2953 {
2954 	if (!qpn ||
2955 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2956 		return;
2957 
2958 	BUG_ON(qpn < dev->steer_qpn_base);
2959 
2960 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2961 			      qpn - dev->steer_qpn_base,
2962 			      get_count_order(count));
2963 }
2964 
2965 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2966 			 int is_attach)
2967 {
2968 	int err;
2969 	size_t flow_size;
2970 	struct ib_flow_attr *flow = NULL;
2971 	struct ib_flow_spec_ib *ib_spec;
2972 
2973 	if (is_attach) {
2974 		flow_size = sizeof(struct ib_flow_attr) +
2975 			    sizeof(struct ib_flow_spec_ib);
2976 		flow = kzalloc(flow_size, GFP_KERNEL);
2977 		if (!flow)
2978 			return -ENOMEM;
2979 		flow->port = mqp->port;
2980 		flow->num_of_specs = 1;
2981 		flow->size = flow_size;
2982 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2983 		ib_spec->type = IB_FLOW_SPEC_IB;
2984 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2985 		/* Add an empty rule for IB L2 */
2986 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2987 
2988 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2989 					    IB_FLOW_DOMAIN_NIC,
2990 					    MLX4_FS_REGULAR,
2991 					    &mqp->reg_id);
2992 	} else {
2993 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2994 	}
2995 	kfree(flow);
2996 	return err;
2997 }
2998 
2999 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3000 {
3001 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3002 	int p;
3003 	int i;
3004 
3005 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3006 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3007 	ibdev->ib_active = false;
3008 	flush_workqueue(wq);
3009 
3010 	mlx4_ib_close_sriov(ibdev);
3011 	mlx4_ib_mad_cleanup(ibdev);
3012 	ib_unregister_device(&ibdev->ib_dev);
3013 	mlx4_ib_diag_cleanup(ibdev);
3014 	if (ibdev->iboe.nb.notifier_call) {
3015 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3016 			pr_warn("failure unregistering notifier\n");
3017 		ibdev->iboe.nb.notifier_call = NULL;
3018 	}
3019 
3020 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
3021 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3022 				      ibdev->steer_qpn_count);
3023 		kfree(ibdev->ib_uc_qpns_bitmap);
3024 	}
3025 
3026 	iounmap(ibdev->uar_map);
3027 	for (p = 0; p < ibdev->num_ports; ++p)
3028 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3029 
3030 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3031 		mlx4_CLOSE_PORT(dev, p);
3032 
3033 	mlx4_ib_free_eqs(dev, ibdev);
3034 
3035 	mlx4_uar_free(dev, &ibdev->priv_uar);
3036 	mlx4_pd_free(dev, ibdev->priv_pdn);
3037 	ib_dealloc_device(&ibdev->ib_dev);
3038 }
3039 
3040 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3041 {
3042 	struct mlx4_ib_demux_work **dm = NULL;
3043 	struct mlx4_dev *dev = ibdev->dev;
3044 	int i;
3045 	unsigned long flags;
3046 	struct mlx4_active_ports actv_ports;
3047 	unsigned int ports;
3048 	unsigned int first_port;
3049 
3050 	if (!mlx4_is_master(dev))
3051 		return;
3052 
3053 	actv_ports = mlx4_get_active_ports(dev, slave);
3054 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3055 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3056 
3057 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3058 	if (!dm) {
3059 		pr_err("failed to allocate memory for tunneling qp update\n");
3060 		return;
3061 	}
3062 
3063 	for (i = 0; i < ports; i++) {
3064 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3065 		if (!dm[i]) {
3066 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
3067 			while (--i >= 0)
3068 				kfree(dm[i]);
3069 			goto out;
3070 		}
3071 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3072 		dm[i]->port = first_port + i + 1;
3073 		dm[i]->slave = slave;
3074 		dm[i]->do_init = do_init;
3075 		dm[i]->dev = ibdev;
3076 	}
3077 	/* initialize or tear down tunnel QPs for the slave */
3078 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3079 	if (!ibdev->sriov.is_going_down) {
3080 		for (i = 0; i < ports; i++)
3081 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3082 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3083 	} else {
3084 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3085 		for (i = 0; i < ports; i++)
3086 			kfree(dm[i]);
3087 	}
3088 out:
3089 	kfree(dm);
3090 	return;
3091 }
3092 
3093 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3094 {
3095 	struct mlx4_ib_qp *mqp;
3096 	unsigned long flags_qp;
3097 	unsigned long flags_cq;
3098 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3099 	struct list_head    cq_notify_list;
3100 	struct mlx4_cq *mcq;
3101 	unsigned long flags;
3102 
3103 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3104 	INIT_LIST_HEAD(&cq_notify_list);
3105 
3106 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3107 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3108 
3109 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3110 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3111 		if (mqp->sq.tail != mqp->sq.head) {
3112 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3113 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3114 			if (send_mcq->mcq.comp &&
3115 			    mqp->ibqp.send_cq->comp_handler) {
3116 				if (!send_mcq->mcq.reset_notify_added) {
3117 					send_mcq->mcq.reset_notify_added = 1;
3118 					list_add_tail(&send_mcq->mcq.reset_notify,
3119 						      &cq_notify_list);
3120 				}
3121 			}
3122 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3123 		}
3124 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3125 		/* Now, handle the QP's receive queue */
3126 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3127 		/* no handling is needed for SRQ */
3128 		if (!mqp->ibqp.srq) {
3129 			if (mqp->rq.tail != mqp->rq.head) {
3130 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3131 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3132 				if (recv_mcq->mcq.comp &&
3133 				    mqp->ibqp.recv_cq->comp_handler) {
3134 					if (!recv_mcq->mcq.reset_notify_added) {
3135 						recv_mcq->mcq.reset_notify_added = 1;
3136 						list_add_tail(&recv_mcq->mcq.reset_notify,
3137 							      &cq_notify_list);
3138 					}
3139 				}
3140 				spin_unlock_irqrestore(&recv_mcq->lock,
3141 						       flags_cq);
3142 			}
3143 		}
3144 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3145 	}
3146 
3147 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3148 		mcq->comp(mcq);
3149 	}
3150 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3151 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3152 }
3153 
3154 static void handle_bonded_port_state_event(struct work_struct *work)
3155 {
3156 	struct ib_event_work *ew =
3157 		container_of(work, struct ib_event_work, work);
3158 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3159 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3160 	int i;
3161 	struct ib_event ibev;
3162 
3163 	kfree(ew);
3164 	spin_lock_bh(&ibdev->iboe.lock);
3165 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3166 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3167 		enum ib_port_state curr_port_state;
3168 
3169 		if (!curr_netdev)
3170 			continue;
3171 
3172 		curr_port_state =
3173 			(netif_running(curr_netdev) &&
3174 			 netif_carrier_ok(curr_netdev)) ?
3175 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3176 
3177 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3178 			curr_port_state : IB_PORT_ACTIVE;
3179 	}
3180 	spin_unlock_bh(&ibdev->iboe.lock);
3181 
3182 	ibev.device = &ibdev->ib_dev;
3183 	ibev.element.port_num = 1;
3184 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3185 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3186 
3187 	ib_dispatch_event(&ibev);
3188 }
3189 
3190 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3191 {
3192 	u64 sl2vl;
3193 	int err;
3194 
3195 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3196 	if (err) {
3197 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3198 		       port, err);
3199 		sl2vl = 0;
3200 	}
3201 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3202 }
3203 
3204 static void ib_sl2vl_update_work(struct work_struct *work)
3205 {
3206 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3207 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3208 	int port = ew->port;
3209 
3210 	mlx4_ib_sl2vl_update(mdev, port);
3211 
3212 	kfree(ew);
3213 }
3214 
3215 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3216 				     int port)
3217 {
3218 	struct ib_event_work *ew;
3219 
3220 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3221 	if (ew) {
3222 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3223 		ew->port = port;
3224 		ew->ib_dev = ibdev;
3225 		queue_work(wq, &ew->work);
3226 	} else {
3227 		pr_err("failed to allocate memory for sl2vl update work\n");
3228 	}
3229 }
3230 
3231 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3232 			  enum mlx4_dev_event event, unsigned long param)
3233 {
3234 	struct ib_event ibev;
3235 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3236 	struct mlx4_eqe *eqe = NULL;
3237 	struct ib_event_work *ew;
3238 	int p = 0;
3239 
3240 	if (mlx4_is_bonded(dev) &&
3241 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3242 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3243 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3244 		if (!ew)
3245 			return;
3246 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3247 		ew->ib_dev = ibdev;
3248 		queue_work(wq, &ew->work);
3249 		return;
3250 	}
3251 
3252 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3253 		eqe = (struct mlx4_eqe *)param;
3254 	else
3255 		p = (int) param;
3256 
3257 	switch (event) {
3258 	case MLX4_DEV_EVENT_PORT_UP:
3259 		if (p > ibdev->num_ports)
3260 			return;
3261 		if (!mlx4_is_slave(dev) &&
3262 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3263 			IB_LINK_LAYER_INFINIBAND) {
3264 			if (mlx4_is_master(dev))
3265 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3266 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3267 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3268 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3269 		}
3270 		ibev.event = IB_EVENT_PORT_ACTIVE;
3271 		break;
3272 
3273 	case MLX4_DEV_EVENT_PORT_DOWN:
3274 		if (p > ibdev->num_ports)
3275 			return;
3276 		ibev.event = IB_EVENT_PORT_ERR;
3277 		break;
3278 
3279 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3280 		ibdev->ib_active = false;
3281 		ibev.event = IB_EVENT_DEVICE_FATAL;
3282 		mlx4_ib_handle_catas_error(ibdev);
3283 		break;
3284 
3285 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3286 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3287 		if (!ew) {
3288 			pr_err("failed to allocate memory for events work\n");
3289 			break;
3290 		}
3291 
3292 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3293 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3294 		ew->ib_dev = ibdev;
3295 		/* need to queue only for port owner, which uses GEN_EQE */
3296 		if (mlx4_is_master(dev))
3297 			queue_work(wq, &ew->work);
3298 		else
3299 			handle_port_mgmt_change_event(&ew->work);
3300 		return;
3301 
3302 	case MLX4_DEV_EVENT_SLAVE_INIT:
3303 		/* here, p is the slave id */
3304 		do_slave_init(ibdev, p, 1);
3305 		if (mlx4_is_master(dev)) {
3306 			int i;
3307 
3308 			for (i = 1; i <= ibdev->num_ports; i++) {
3309 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3310 					== IB_LINK_LAYER_INFINIBAND)
3311 					mlx4_ib_slave_alias_guid_event(ibdev,
3312 								       p, i,
3313 								       1);
3314 			}
3315 		}
3316 		return;
3317 
3318 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3319 		if (mlx4_is_master(dev)) {
3320 			int i;
3321 
3322 			for (i = 1; i <= ibdev->num_ports; i++) {
3323 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3324 					== IB_LINK_LAYER_INFINIBAND)
3325 					mlx4_ib_slave_alias_guid_event(ibdev,
3326 								       p, i,
3327 								       0);
3328 			}
3329 		}
3330 		/* here, p is the slave id */
3331 		do_slave_init(ibdev, p, 0);
3332 		return;
3333 
3334 	default:
3335 		return;
3336 	}
3337 
3338 	ibev.device	      = ibdev_ptr;
3339 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3340 
3341 	ib_dispatch_event(&ibev);
3342 }
3343 
3344 static struct mlx4_interface mlx4_ib_interface = {
3345 	.add		= mlx4_ib_add,
3346 	.remove		= mlx4_ib_remove,
3347 	.event		= mlx4_ib_event,
3348 	.protocol	= MLX4_PROT_IB_IPV6,
3349 	.flags		= MLX4_INTFF_BONDING
3350 };
3351 
3352 static int __init mlx4_ib_init(void)
3353 {
3354 	int err;
3355 
3356 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3357 	if (!wq)
3358 		return -ENOMEM;
3359 
3360 	err = mlx4_ib_mcg_init();
3361 	if (err)
3362 		goto clean_wq;
3363 
3364 	err = mlx4_register_interface(&mlx4_ib_interface);
3365 	if (err)
3366 		goto clean_mcg;
3367 
3368 	return 0;
3369 
3370 clean_mcg:
3371 	mlx4_ib_mcg_destroy();
3372 
3373 clean_wq:
3374 	destroy_workqueue(wq);
3375 	return err;
3376 }
3377 
3378 static void __exit mlx4_ib_cleanup(void)
3379 {
3380 	mlx4_unregister_interface(&mlx4_ib_interface);
3381 	mlx4_ib_mcg_destroy();
3382 	destroy_workqueue(wq);
3383 }
3384 
3385 module_init(mlx4_ib_init);
3386 module_exit(mlx4_ib_cleanup);
3387