xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 7bcae826)
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;
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 	err = -ENOMEM;
459 	if (!in_mad || !out_mad)
460 		goto out;
461 
462 	init_query_mad(in_mad);
463 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
464 
465 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
466 			   1, NULL, NULL, in_mad, out_mad);
467 	if (err)
468 		goto out;
469 
470 	memset(props, 0, sizeof *props);
471 
472 	have_ib_ports = num_ib_ports(dev->dev);
473 
474 	props->fw_ver = dev->dev->caps.fw_ver;
475 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
476 		IB_DEVICE_PORT_ACTIVE_EVENT		|
477 		IB_DEVICE_SYS_IMAGE_GUID		|
478 		IB_DEVICE_RC_RNR_NAK_GEN		|
479 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
480 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
481 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
482 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
483 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
484 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
485 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
486 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
487 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
488 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
489 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
490 	if (dev->dev->caps.max_gso_sz &&
491 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
492 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
493 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
494 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
495 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
496 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
497 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
498 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
499 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
500 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
501 		props->device_cap_flags |= IB_DEVICE_XRC;
502 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
503 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
504 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
505 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
506 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
507 		else
508 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
509 	}
510 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
511 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
512 
513 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
514 
515 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
516 		0xffffff;
517 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
518 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
519 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
520 
521 	props->max_mr_size	   = ~0ull;
522 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
523 	props->max_qp		   = dev->dev->quotas.qp;
524 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
525 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
526 					 dev->dev->caps.max_rq_sg);
527 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
528 	props->max_cq		   = dev->dev->quotas.cq;
529 	props->max_cqe		   = dev->dev->caps.max_cqes;
530 	props->max_mr		   = dev->dev->quotas.mpt;
531 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
532 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
533 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
534 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
535 	props->max_srq		   = dev->dev->quotas.srq;
536 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
537 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
538 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
539 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
540 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
541 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
542 	props->masked_atomic_cap   = props->atomic_cap;
543 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
544 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
545 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
546 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
547 					   props->max_mcast_grp;
548 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
549 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
550 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
551 	props->max_ah = INT_MAX;
552 
553 	if (!mlx4_is_slave(dev->dev))
554 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
555 
556 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
557 		resp.response_length += sizeof(resp.hca_core_clock_offset);
558 		if (!err && !mlx4_is_slave(dev->dev)) {
559 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
560 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
561 		}
562 	}
563 
564 	if (uhw->outlen) {
565 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
566 		if (err)
567 			goto out;
568 	}
569 out:
570 	kfree(in_mad);
571 	kfree(out_mad);
572 
573 	return err;
574 }
575 
576 static enum rdma_link_layer
577 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
578 {
579 	struct mlx4_dev *dev = to_mdev(device)->dev;
580 
581 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
582 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
583 }
584 
585 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
586 			      struct ib_port_attr *props, int netw_view)
587 {
588 	struct ib_smp *in_mad  = NULL;
589 	struct ib_smp *out_mad = NULL;
590 	int ext_active_speed;
591 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
592 	int err = -ENOMEM;
593 
594 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
595 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
596 	if (!in_mad || !out_mad)
597 		goto out;
598 
599 	init_query_mad(in_mad);
600 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
601 	in_mad->attr_mod = cpu_to_be32(port);
602 
603 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
604 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
605 
606 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
607 				in_mad, out_mad);
608 	if (err)
609 		goto out;
610 
611 
612 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
613 	props->lmc		= out_mad->data[34] & 0x7;
614 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
615 	props->sm_sl		= out_mad->data[36] & 0xf;
616 	props->state		= out_mad->data[32] & 0xf;
617 	props->phys_state	= out_mad->data[33] >> 4;
618 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
619 	if (netw_view)
620 		props->gid_tbl_len = out_mad->data[50];
621 	else
622 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
623 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
624 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
625 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
626 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
627 	props->active_width	= out_mad->data[31] & 0xf;
628 	props->active_speed	= out_mad->data[35] >> 4;
629 	props->max_mtu		= out_mad->data[41] & 0xf;
630 	props->active_mtu	= out_mad->data[36] >> 4;
631 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
632 	props->max_vl_num	= out_mad->data[37] >> 4;
633 	props->init_type_reply	= out_mad->data[41] >> 4;
634 
635 	/* Check if extended speeds (EDR/FDR/...) are supported */
636 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
637 		ext_active_speed = out_mad->data[62] >> 4;
638 
639 		switch (ext_active_speed) {
640 		case 1:
641 			props->active_speed = IB_SPEED_FDR;
642 			break;
643 		case 2:
644 			props->active_speed = IB_SPEED_EDR;
645 			break;
646 		}
647 	}
648 
649 	/* If reported active speed is QDR, check if is FDR-10 */
650 	if (props->active_speed == IB_SPEED_QDR) {
651 		init_query_mad(in_mad);
652 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
653 		in_mad->attr_mod = cpu_to_be32(port);
654 
655 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
656 				   NULL, NULL, in_mad, out_mad);
657 		if (err)
658 			goto out;
659 
660 		/* Checking LinkSpeedActive for FDR-10 */
661 		if (out_mad->data[15] & 0x1)
662 			props->active_speed = IB_SPEED_FDR10;
663 	}
664 
665 	/* Avoid wrong speed value returned by FW if the IB link is down. */
666 	if (props->state == IB_PORT_DOWN)
667 		 props->active_speed = IB_SPEED_SDR;
668 
669 out:
670 	kfree(in_mad);
671 	kfree(out_mad);
672 	return err;
673 }
674 
675 static u8 state_to_phys_state(enum ib_port_state state)
676 {
677 	return state == IB_PORT_ACTIVE ? 5 : 3;
678 }
679 
680 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
681 			       struct ib_port_attr *props)
682 {
683 
684 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
685 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
686 	struct net_device *ndev;
687 	enum ib_mtu tmp;
688 	struct mlx4_cmd_mailbox *mailbox;
689 	int err = 0;
690 	int is_bonded = mlx4_is_bonded(mdev->dev);
691 
692 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
693 	if (IS_ERR(mailbox))
694 		return PTR_ERR(mailbox);
695 
696 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
697 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
698 			   MLX4_CMD_WRAPPED);
699 	if (err)
700 		goto out;
701 
702 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
703 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
704 					   IB_WIDTH_4X : IB_WIDTH_1X;
705 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
706 					   IB_SPEED_FDR : IB_SPEED_QDR;
707 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
708 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
709 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
710 	props->pkey_tbl_len	= 1;
711 	props->max_mtu		= IB_MTU_4096;
712 	props->max_vl_num	= 2;
713 	props->state		= IB_PORT_DOWN;
714 	props->phys_state	= state_to_phys_state(props->state);
715 	props->active_mtu	= IB_MTU_256;
716 	spin_lock_bh(&iboe->lock);
717 	ndev = iboe->netdevs[port - 1];
718 	if (ndev && is_bonded) {
719 		rcu_read_lock(); /* required to get upper dev */
720 		ndev = netdev_master_upper_dev_get_rcu(ndev);
721 		rcu_read_unlock();
722 	}
723 	if (!ndev)
724 		goto out_unlock;
725 
726 	tmp = iboe_get_mtu(ndev->mtu);
727 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
728 
729 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
730 					IB_PORT_ACTIVE : IB_PORT_DOWN;
731 	props->phys_state	= state_to_phys_state(props->state);
732 out_unlock:
733 	spin_unlock_bh(&iboe->lock);
734 out:
735 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
736 	return err;
737 }
738 
739 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
740 			 struct ib_port_attr *props, int netw_view)
741 {
742 	int err;
743 
744 	/* props being zeroed by the caller, avoid zeroing it here */
745 
746 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
747 		ib_link_query_port(ibdev, port, props, netw_view) :
748 				eth_link_query_port(ibdev, port, props);
749 
750 	return err;
751 }
752 
753 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
754 			      struct ib_port_attr *props)
755 {
756 	/* returns host view */
757 	return __mlx4_ib_query_port(ibdev, port, props, 0);
758 }
759 
760 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
761 			union ib_gid *gid, int netw_view)
762 {
763 	struct ib_smp *in_mad  = NULL;
764 	struct ib_smp *out_mad = NULL;
765 	int err = -ENOMEM;
766 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
767 	int clear = 0;
768 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
769 
770 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
771 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
772 	if (!in_mad || !out_mad)
773 		goto out;
774 
775 	init_query_mad(in_mad);
776 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
777 	in_mad->attr_mod = cpu_to_be32(port);
778 
779 	if (mlx4_is_mfunc(dev->dev) && netw_view)
780 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
781 
782 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
783 	if (err)
784 		goto out;
785 
786 	memcpy(gid->raw, out_mad->data + 8, 8);
787 
788 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
789 		if (index) {
790 			/* For any index > 0, return the null guid */
791 			err = 0;
792 			clear = 1;
793 			goto out;
794 		}
795 	}
796 
797 	init_query_mad(in_mad);
798 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
799 	in_mad->attr_mod = cpu_to_be32(index / 8);
800 
801 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
802 			   NULL, NULL, in_mad, out_mad);
803 	if (err)
804 		goto out;
805 
806 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
807 
808 out:
809 	if (clear)
810 		memset(gid->raw + 8, 0, 8);
811 	kfree(in_mad);
812 	kfree(out_mad);
813 	return err;
814 }
815 
816 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
817 			     union ib_gid *gid)
818 {
819 	int ret;
820 
821 	if (rdma_protocol_ib(ibdev, port))
822 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
823 
824 	if (!rdma_protocol_roce(ibdev, port))
825 		return -ENODEV;
826 
827 	if (!rdma_cap_roce_gid_table(ibdev, port))
828 		return -ENODEV;
829 
830 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
831 	if (ret == -EAGAIN) {
832 		memcpy(gid, &zgid, sizeof(*gid));
833 		return 0;
834 	}
835 
836 	return ret;
837 }
838 
839 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
840 {
841 	union sl2vl_tbl_to_u64 sl2vl64;
842 	struct ib_smp *in_mad  = NULL;
843 	struct ib_smp *out_mad = NULL;
844 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
845 	int err = -ENOMEM;
846 	int jj;
847 
848 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
849 		*sl2vl_tbl = 0;
850 		return 0;
851 	}
852 
853 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
854 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
855 	if (!in_mad || !out_mad)
856 		goto out;
857 
858 	init_query_mad(in_mad);
859 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
860 	in_mad->attr_mod = 0;
861 
862 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
863 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
864 
865 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
866 			   in_mad, out_mad);
867 	if (err)
868 		goto out;
869 
870 	for (jj = 0; jj < 8; jj++)
871 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
872 	*sl2vl_tbl = sl2vl64.sl64;
873 
874 out:
875 	kfree(in_mad);
876 	kfree(out_mad);
877 	return err;
878 }
879 
880 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
881 {
882 	u64 sl2vl;
883 	int i;
884 	int err;
885 
886 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
887 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
888 			continue;
889 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
890 		if (err) {
891 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
892 			       i, err);
893 			sl2vl = 0;
894 		}
895 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
896 	}
897 }
898 
899 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
900 			 u16 *pkey, int netw_view)
901 {
902 	struct ib_smp *in_mad  = NULL;
903 	struct ib_smp *out_mad = NULL;
904 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
905 	int err = -ENOMEM;
906 
907 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
908 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
909 	if (!in_mad || !out_mad)
910 		goto out;
911 
912 	init_query_mad(in_mad);
913 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
914 	in_mad->attr_mod = cpu_to_be32(index / 32);
915 
916 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
917 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
918 
919 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
920 			   in_mad, out_mad);
921 	if (err)
922 		goto out;
923 
924 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
925 
926 out:
927 	kfree(in_mad);
928 	kfree(out_mad);
929 	return err;
930 }
931 
932 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
933 {
934 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
935 }
936 
937 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
938 				 struct ib_device_modify *props)
939 {
940 	struct mlx4_cmd_mailbox *mailbox;
941 	unsigned long flags;
942 
943 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
944 		return -EOPNOTSUPP;
945 
946 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
947 		return 0;
948 
949 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
950 		return -EOPNOTSUPP;
951 
952 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
953 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
954 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
955 
956 	/*
957 	 * If possible, pass node desc to FW, so it can generate
958 	 * a 144 trap.  If cmd fails, just ignore.
959 	 */
960 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
961 	if (IS_ERR(mailbox))
962 		return 0;
963 
964 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
965 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
966 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
967 
968 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
969 
970 	return 0;
971 }
972 
973 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
974 			    u32 cap_mask)
975 {
976 	struct mlx4_cmd_mailbox *mailbox;
977 	int err;
978 
979 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
980 	if (IS_ERR(mailbox))
981 		return PTR_ERR(mailbox);
982 
983 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
984 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
985 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
986 	} else {
987 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
988 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
989 	}
990 
991 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
992 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
993 		       MLX4_CMD_WRAPPED);
994 
995 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
996 	return err;
997 }
998 
999 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1000 			       struct ib_port_modify *props)
1001 {
1002 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1003 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1004 	struct ib_port_attr attr;
1005 	u32 cap_mask;
1006 	int err;
1007 
1008 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1009 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1010 	 * violations and port capabilities are not meaningful.
1011 	 */
1012 	if (is_eth)
1013 		return 0;
1014 
1015 	mutex_lock(&mdev->cap_mask_mutex);
1016 
1017 	err = ib_query_port(ibdev, port, &attr);
1018 	if (err)
1019 		goto out;
1020 
1021 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1022 		~props->clr_port_cap_mask;
1023 
1024 	err = mlx4_ib_SET_PORT(mdev, port,
1025 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1026 			       cap_mask);
1027 
1028 out:
1029 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1030 	return err;
1031 }
1032 
1033 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1034 						  struct ib_udata *udata)
1035 {
1036 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1037 	struct mlx4_ib_ucontext *context;
1038 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1039 	struct mlx4_ib_alloc_ucontext_resp resp;
1040 	int err;
1041 
1042 	if (!dev->ib_active)
1043 		return ERR_PTR(-EAGAIN);
1044 
1045 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1046 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1047 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1048 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1049 	} else {
1050 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1051 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1052 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1053 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1054 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1055 	}
1056 
1057 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1058 	if (!context)
1059 		return ERR_PTR(-ENOMEM);
1060 
1061 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1062 	if (err) {
1063 		kfree(context);
1064 		return ERR_PTR(err);
1065 	}
1066 
1067 	INIT_LIST_HEAD(&context->db_page_list);
1068 	mutex_init(&context->db_page_mutex);
1069 
1070 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1071 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1072 	else
1073 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1074 
1075 	if (err) {
1076 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1077 		kfree(context);
1078 		return ERR_PTR(-EFAULT);
1079 	}
1080 
1081 	return &context->ibucontext;
1082 }
1083 
1084 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1085 {
1086 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1087 
1088 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1089 	kfree(context);
1090 
1091 	return 0;
1092 }
1093 
1094 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1095 {
1096 	/* vma_open is called when a new VMA is created on top of our VMA.
1097 	 * This is done through either mremap flow or split_vma (usually due
1098 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1099 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1100 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1101 	 * calling us again and trying to do incorrect actions. We assume that
1102 	 * the original vma size is exactly a single page that there will be no
1103 	 * "splitting" operations on.
1104 	 */
1105 	area->vm_ops = NULL;
1106 }
1107 
1108 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1109 {
1110 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1111 
1112 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1113 	 * file itself is closed, therefore no sync is needed with the regular
1114 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1115 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1116 	 * The close operation is usually called under mm->mmap_sem except when
1117 	 * process is exiting.  The exiting case is handled explicitly as part
1118 	 * of mlx4_ib_disassociate_ucontext.
1119 	 */
1120 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1121 				area->vm_private_data;
1122 
1123 	/* set the vma context pointer to null in the mlx4_ib driver's private
1124 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1125 	 */
1126 	mlx4_ib_vma_priv_data->vma = NULL;
1127 }
1128 
1129 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1130 	.open = mlx4_ib_vma_open,
1131 	.close = mlx4_ib_vma_close
1132 };
1133 
1134 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1135 {
1136 	int i;
1137 	int ret = 0;
1138 	struct vm_area_struct *vma;
1139 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1140 	struct task_struct *owning_process  = NULL;
1141 	struct mm_struct   *owning_mm       = NULL;
1142 
1143 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1144 	if (!owning_process)
1145 		return;
1146 
1147 	owning_mm = get_task_mm(owning_process);
1148 	if (!owning_mm) {
1149 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1150 		while (1) {
1151 			/* make sure that task is dead before returning, it may
1152 			 * prevent a rare case of module down in parallel to a
1153 			 * call to mlx4_ib_vma_close.
1154 			 */
1155 			put_task_struct(owning_process);
1156 			msleep(1);
1157 			owning_process = get_pid_task(ibcontext->tgid,
1158 						      PIDTYPE_PID);
1159 			if (!owning_process ||
1160 			    owning_process->state == TASK_DEAD) {
1161 				pr_info("disassociate ucontext done, task was terminated\n");
1162 				/* in case task was dead need to release the task struct */
1163 				if (owning_process)
1164 					put_task_struct(owning_process);
1165 				return;
1166 			}
1167 		}
1168 	}
1169 
1170 	/* need to protect from a race on closing the vma as part of
1171 	 * mlx4_ib_vma_close().
1172 	 */
1173 	down_read(&owning_mm->mmap_sem);
1174 	for (i = 0; i < HW_BAR_COUNT; i++) {
1175 		vma = context->hw_bar_info[i].vma;
1176 		if (!vma)
1177 			continue;
1178 
1179 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1180 				   context->hw_bar_info[i].vma->vm_start,
1181 				   PAGE_SIZE);
1182 		if (ret) {
1183 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1184 			BUG_ON(1);
1185 		}
1186 
1187 		/* context going to be destroyed, should not access ops any more */
1188 		context->hw_bar_info[i].vma->vm_ops = NULL;
1189 	}
1190 
1191 	up_read(&owning_mm->mmap_sem);
1192 	mmput(owning_mm);
1193 	put_task_struct(owning_process);
1194 }
1195 
1196 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1197 				 struct mlx4_ib_vma_private_data *vma_private_data)
1198 {
1199 	vma_private_data->vma = vma;
1200 	vma->vm_private_data = vma_private_data;
1201 	vma->vm_ops =  &mlx4_ib_vm_ops;
1202 }
1203 
1204 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1205 {
1206 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1207 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1208 
1209 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1210 		return -EINVAL;
1211 
1212 	if (vma->vm_pgoff == 0) {
1213 		/* We prevent double mmaping on same context */
1214 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1215 			return -EINVAL;
1216 
1217 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1218 
1219 		if (io_remap_pfn_range(vma, vma->vm_start,
1220 				       to_mucontext(context)->uar.pfn,
1221 				       PAGE_SIZE, vma->vm_page_prot))
1222 			return -EAGAIN;
1223 
1224 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1225 
1226 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1227 		/* We prevent double mmaping on same context */
1228 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1229 			return -EINVAL;
1230 
1231 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1232 
1233 		if (io_remap_pfn_range(vma, vma->vm_start,
1234 				       to_mucontext(context)->uar.pfn +
1235 				       dev->dev->caps.num_uars,
1236 				       PAGE_SIZE, vma->vm_page_prot))
1237 			return -EAGAIN;
1238 
1239 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1240 
1241 	} else if (vma->vm_pgoff == 3) {
1242 		struct mlx4_clock_params params;
1243 		int ret;
1244 
1245 		/* We prevent double mmaping on same context */
1246 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1247 			return -EINVAL;
1248 
1249 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1250 
1251 		if (ret)
1252 			return ret;
1253 
1254 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1255 		if (io_remap_pfn_range(vma, vma->vm_start,
1256 				       (pci_resource_start(dev->dev->persist->pdev,
1257 							   params.bar) +
1258 					params.offset)
1259 				       >> PAGE_SHIFT,
1260 				       PAGE_SIZE, vma->vm_page_prot))
1261 			return -EAGAIN;
1262 
1263 		mlx4_ib_set_vma_data(vma,
1264 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1265 	} else {
1266 		return -EINVAL;
1267 	}
1268 
1269 	return 0;
1270 }
1271 
1272 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1273 				      struct ib_ucontext *context,
1274 				      struct ib_udata *udata)
1275 {
1276 	struct mlx4_ib_pd *pd;
1277 	int err;
1278 
1279 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1280 	if (!pd)
1281 		return ERR_PTR(-ENOMEM);
1282 
1283 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1284 	if (err) {
1285 		kfree(pd);
1286 		return ERR_PTR(err);
1287 	}
1288 
1289 	if (context)
1290 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1291 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1292 			kfree(pd);
1293 			return ERR_PTR(-EFAULT);
1294 		}
1295 
1296 	return &pd->ibpd;
1297 }
1298 
1299 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1300 {
1301 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1302 	kfree(pd);
1303 
1304 	return 0;
1305 }
1306 
1307 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1308 					  struct ib_ucontext *context,
1309 					  struct ib_udata *udata)
1310 {
1311 	struct mlx4_ib_xrcd *xrcd;
1312 	struct ib_cq_init_attr cq_attr = {};
1313 	int err;
1314 
1315 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1316 		return ERR_PTR(-ENOSYS);
1317 
1318 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1319 	if (!xrcd)
1320 		return ERR_PTR(-ENOMEM);
1321 
1322 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1323 	if (err)
1324 		goto err1;
1325 
1326 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1327 	if (IS_ERR(xrcd->pd)) {
1328 		err = PTR_ERR(xrcd->pd);
1329 		goto err2;
1330 	}
1331 
1332 	cq_attr.cqe = 1;
1333 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1334 	if (IS_ERR(xrcd->cq)) {
1335 		err = PTR_ERR(xrcd->cq);
1336 		goto err3;
1337 	}
1338 
1339 	return &xrcd->ibxrcd;
1340 
1341 err3:
1342 	ib_dealloc_pd(xrcd->pd);
1343 err2:
1344 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1345 err1:
1346 	kfree(xrcd);
1347 	return ERR_PTR(err);
1348 }
1349 
1350 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1351 {
1352 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1353 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1354 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1355 	kfree(xrcd);
1356 
1357 	return 0;
1358 }
1359 
1360 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1361 {
1362 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1363 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1364 	struct mlx4_ib_gid_entry *ge;
1365 
1366 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1367 	if (!ge)
1368 		return -ENOMEM;
1369 
1370 	ge->gid = *gid;
1371 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1372 		ge->port = mqp->port;
1373 		ge->added = 1;
1374 	}
1375 
1376 	mutex_lock(&mqp->mutex);
1377 	list_add_tail(&ge->list, &mqp->gid_list);
1378 	mutex_unlock(&mqp->mutex);
1379 
1380 	return 0;
1381 }
1382 
1383 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1384 					  struct mlx4_ib_counters *ctr_table)
1385 {
1386 	struct counter_index *counter, *tmp_count;
1387 
1388 	mutex_lock(&ctr_table->mutex);
1389 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1390 				 list) {
1391 		if (counter->allocated)
1392 			mlx4_counter_free(ibdev->dev, counter->index);
1393 		list_del(&counter->list);
1394 		kfree(counter);
1395 	}
1396 	mutex_unlock(&ctr_table->mutex);
1397 }
1398 
1399 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1400 		   union ib_gid *gid)
1401 {
1402 	struct net_device *ndev;
1403 	int ret = 0;
1404 
1405 	if (!mqp->port)
1406 		return 0;
1407 
1408 	spin_lock_bh(&mdev->iboe.lock);
1409 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1410 	if (ndev)
1411 		dev_hold(ndev);
1412 	spin_unlock_bh(&mdev->iboe.lock);
1413 
1414 	if (ndev) {
1415 		ret = 1;
1416 		dev_put(ndev);
1417 	}
1418 
1419 	return ret;
1420 }
1421 
1422 struct mlx4_ib_steering {
1423 	struct list_head list;
1424 	struct mlx4_flow_reg_id reg_id;
1425 	union ib_gid gid;
1426 };
1427 
1428 #define LAST_ETH_FIELD vlan_tag
1429 #define LAST_IB_FIELD sl
1430 #define LAST_IPV4_FIELD dst_ip
1431 #define LAST_TCP_UDP_FIELD src_port
1432 
1433 /* Field is the last supported field */
1434 #define FIELDS_NOT_SUPPORTED(filter, field)\
1435 	memchr_inv((void *)&filter.field  +\
1436 		   sizeof(filter.field), 0,\
1437 		   sizeof(filter) -\
1438 		   offsetof(typeof(filter), field) -\
1439 		   sizeof(filter.field))
1440 
1441 static int parse_flow_attr(struct mlx4_dev *dev,
1442 			   u32 qp_num,
1443 			   union ib_flow_spec *ib_spec,
1444 			   struct _rule_hw *mlx4_spec)
1445 {
1446 	enum mlx4_net_trans_rule_id type;
1447 
1448 	switch (ib_spec->type) {
1449 	case IB_FLOW_SPEC_ETH:
1450 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1451 			return -ENOTSUPP;
1452 
1453 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1454 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1455 		       ETH_ALEN);
1456 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1457 		       ETH_ALEN);
1458 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1459 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1460 		break;
1461 	case IB_FLOW_SPEC_IB:
1462 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1463 			return -ENOTSUPP;
1464 
1465 		type = MLX4_NET_TRANS_RULE_ID_IB;
1466 		mlx4_spec->ib.l3_qpn =
1467 			cpu_to_be32(qp_num);
1468 		mlx4_spec->ib.qpn_mask =
1469 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1470 		break;
1471 
1472 
1473 	case IB_FLOW_SPEC_IPV4:
1474 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1475 			return -ENOTSUPP;
1476 
1477 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1478 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1479 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1480 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1481 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1482 		break;
1483 
1484 	case IB_FLOW_SPEC_TCP:
1485 	case IB_FLOW_SPEC_UDP:
1486 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1487 			return -ENOTSUPP;
1488 
1489 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1490 					MLX4_NET_TRANS_RULE_ID_TCP :
1491 					MLX4_NET_TRANS_RULE_ID_UDP;
1492 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1493 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1494 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1495 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1496 		break;
1497 
1498 	default:
1499 		return -EINVAL;
1500 	}
1501 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1502 	    mlx4_hw_rule_sz(dev, type) < 0)
1503 		return -EINVAL;
1504 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1505 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1506 	return mlx4_hw_rule_sz(dev, type);
1507 }
1508 
1509 struct default_rules {
1510 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1511 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1512 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1513 	__u8  link_layer;
1514 };
1515 static const struct default_rules default_table[] = {
1516 	{
1517 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1518 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1519 		.rules_create_list = {IB_FLOW_SPEC_IB},
1520 		.link_layer = IB_LINK_LAYER_INFINIBAND
1521 	}
1522 };
1523 
1524 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1525 					 struct ib_flow_attr *flow_attr)
1526 {
1527 	int i, j, k;
1528 	void *ib_flow;
1529 	const struct default_rules *pdefault_rules = default_table;
1530 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1531 
1532 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1533 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1534 		memset(&field_types, 0, sizeof(field_types));
1535 
1536 		if (link_layer != pdefault_rules->link_layer)
1537 			continue;
1538 
1539 		ib_flow = flow_attr + 1;
1540 		/* we assume the specs are sorted */
1541 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1542 		     j < flow_attr->num_of_specs; k++) {
1543 			union ib_flow_spec *current_flow =
1544 				(union ib_flow_spec *)ib_flow;
1545 
1546 			/* same layer but different type */
1547 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1548 			     (pdefault_rules->mandatory_fields[k] &
1549 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1550 			    (current_flow->type !=
1551 			     pdefault_rules->mandatory_fields[k]))
1552 				goto out;
1553 
1554 			/* same layer, try match next one */
1555 			if (current_flow->type ==
1556 			    pdefault_rules->mandatory_fields[k]) {
1557 				j++;
1558 				ib_flow +=
1559 					((union ib_flow_spec *)ib_flow)->size;
1560 			}
1561 		}
1562 
1563 		ib_flow = flow_attr + 1;
1564 		for (j = 0; j < flow_attr->num_of_specs;
1565 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1566 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1567 				/* same layer and same type */
1568 				if (((union ib_flow_spec *)ib_flow)->type ==
1569 				    pdefault_rules->mandatory_not_fields[k])
1570 					goto out;
1571 
1572 		return i;
1573 	}
1574 out:
1575 	return -1;
1576 }
1577 
1578 static int __mlx4_ib_create_default_rules(
1579 		struct mlx4_ib_dev *mdev,
1580 		struct ib_qp *qp,
1581 		const struct default_rules *pdefault_rules,
1582 		struct _rule_hw *mlx4_spec) {
1583 	int size = 0;
1584 	int i;
1585 
1586 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1587 		int ret;
1588 		union ib_flow_spec ib_spec;
1589 		switch (pdefault_rules->rules_create_list[i]) {
1590 		case 0:
1591 			/* no rule */
1592 			continue;
1593 		case IB_FLOW_SPEC_IB:
1594 			ib_spec.type = IB_FLOW_SPEC_IB;
1595 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1596 
1597 			break;
1598 		default:
1599 			/* invalid rule */
1600 			return -EINVAL;
1601 		}
1602 		/* We must put empty rule, qpn is being ignored */
1603 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1604 				      mlx4_spec);
1605 		if (ret < 0) {
1606 			pr_info("invalid parsing\n");
1607 			return -EINVAL;
1608 		}
1609 
1610 		mlx4_spec = (void *)mlx4_spec + ret;
1611 		size += ret;
1612 	}
1613 	return size;
1614 }
1615 
1616 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1617 			  int domain,
1618 			  enum mlx4_net_trans_promisc_mode flow_type,
1619 			  u64 *reg_id)
1620 {
1621 	int ret, i;
1622 	int size = 0;
1623 	void *ib_flow;
1624 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1625 	struct mlx4_cmd_mailbox *mailbox;
1626 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1627 	int default_flow;
1628 
1629 	static const u16 __mlx4_domain[] = {
1630 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1631 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1632 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1633 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1634 	};
1635 
1636 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1637 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1638 		return -EINVAL;
1639 	}
1640 
1641 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1642 		pr_err("Invalid domain value %d\n", domain);
1643 		return -EINVAL;
1644 	}
1645 
1646 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1647 		return -EINVAL;
1648 
1649 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1650 	if (IS_ERR(mailbox))
1651 		return PTR_ERR(mailbox);
1652 	ctrl = mailbox->buf;
1653 
1654 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1655 				 flow_attr->priority);
1656 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1657 	ctrl->port = flow_attr->port;
1658 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1659 
1660 	ib_flow = flow_attr + 1;
1661 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1662 	/* Add default flows */
1663 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1664 	if (default_flow >= 0) {
1665 		ret = __mlx4_ib_create_default_rules(
1666 				mdev, qp, default_table + default_flow,
1667 				mailbox->buf + size);
1668 		if (ret < 0) {
1669 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1670 			return -EINVAL;
1671 		}
1672 		size += ret;
1673 	}
1674 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1675 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1676 				      mailbox->buf + size);
1677 		if (ret < 0) {
1678 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1679 			return -EINVAL;
1680 		}
1681 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1682 		size += ret;
1683 	}
1684 
1685 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1686 	    flow_attr->num_of_specs == 1) {
1687 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1688 		enum ib_flow_spec_type header_spec =
1689 			((union ib_flow_spec *)(flow_attr + 1))->type;
1690 
1691 		if (header_spec == IB_FLOW_SPEC_ETH)
1692 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1693 	}
1694 
1695 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1696 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1697 			   MLX4_CMD_NATIVE);
1698 	if (ret == -ENOMEM)
1699 		pr_err("mcg table is full. Fail to register network rule.\n");
1700 	else if (ret == -ENXIO)
1701 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1702 	else if (ret)
1703 		pr_err("Invalid argument. Fail to register network rule.\n");
1704 
1705 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1706 	return ret;
1707 }
1708 
1709 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1710 {
1711 	int err;
1712 	err = mlx4_cmd(dev, reg_id, 0, 0,
1713 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1714 		       MLX4_CMD_NATIVE);
1715 	if (err)
1716 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1717 		       reg_id);
1718 	return err;
1719 }
1720 
1721 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1722 				    u64 *reg_id)
1723 {
1724 	void *ib_flow;
1725 	union ib_flow_spec *ib_spec;
1726 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1727 	int err = 0;
1728 
1729 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1730 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1731 		return 0; /* do nothing */
1732 
1733 	ib_flow = flow_attr + 1;
1734 	ib_spec = (union ib_flow_spec *)ib_flow;
1735 
1736 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1737 		return 0; /* do nothing */
1738 
1739 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1740 				    flow_attr->port, qp->qp_num,
1741 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1742 				    reg_id);
1743 	return err;
1744 }
1745 
1746 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1747 				      struct ib_flow_attr *flow_attr,
1748 				      enum mlx4_net_trans_promisc_mode *type)
1749 {
1750 	int err = 0;
1751 
1752 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1753 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1754 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1755 		return -EOPNOTSUPP;
1756 	}
1757 
1758 	if (flow_attr->num_of_specs == 0) {
1759 		type[0] = MLX4_FS_MC_SNIFFER;
1760 		type[1] = MLX4_FS_UC_SNIFFER;
1761 	} else {
1762 		union ib_flow_spec *ib_spec;
1763 
1764 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1765 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1766 			return -EINVAL;
1767 
1768 		/* if all is zero than MC and UC */
1769 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1770 			type[0] = MLX4_FS_MC_SNIFFER;
1771 			type[1] = MLX4_FS_UC_SNIFFER;
1772 		} else {
1773 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1774 					    ib_spec->eth.mask.dst_mac[1],
1775 					    ib_spec->eth.mask.dst_mac[2],
1776 					    ib_spec->eth.mask.dst_mac[3],
1777 					    ib_spec->eth.mask.dst_mac[4],
1778 					    ib_spec->eth.mask.dst_mac[5]};
1779 
1780 			/* Above xor was only on MC bit, non empty mask is valid
1781 			 * only if this bit is set and rest are zero.
1782 			 */
1783 			if (!is_zero_ether_addr(&mac[0]))
1784 				return -EINVAL;
1785 
1786 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1787 				type[0] = MLX4_FS_MC_SNIFFER;
1788 			else
1789 				type[0] = MLX4_FS_UC_SNIFFER;
1790 		}
1791 	}
1792 
1793 	return err;
1794 }
1795 
1796 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1797 				    struct ib_flow_attr *flow_attr,
1798 				    int domain)
1799 {
1800 	int err = 0, i = 0, j = 0;
1801 	struct mlx4_ib_flow *mflow;
1802 	enum mlx4_net_trans_promisc_mode type[2];
1803 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1804 	int is_bonded = mlx4_is_bonded(dev);
1805 
1806 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1807 		return ERR_PTR(-EINVAL);
1808 
1809 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1810 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1811 		return ERR_PTR(-EOPNOTSUPP);
1812 
1813 	memset(type, 0, sizeof(type));
1814 
1815 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1816 	if (!mflow) {
1817 		err = -ENOMEM;
1818 		goto err_free;
1819 	}
1820 
1821 	switch (flow_attr->type) {
1822 	case IB_FLOW_ATTR_NORMAL:
1823 		/* If dont trap flag (continue match) is set, under specific
1824 		 * condition traffic be replicated to given qp,
1825 		 * without stealing it
1826 		 */
1827 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1828 			err = mlx4_ib_add_dont_trap_rule(dev,
1829 							 flow_attr,
1830 							 type);
1831 			if (err)
1832 				goto err_free;
1833 		} else {
1834 			type[0] = MLX4_FS_REGULAR;
1835 		}
1836 		break;
1837 
1838 	case IB_FLOW_ATTR_ALL_DEFAULT:
1839 		type[0] = MLX4_FS_ALL_DEFAULT;
1840 		break;
1841 
1842 	case IB_FLOW_ATTR_MC_DEFAULT:
1843 		type[0] = MLX4_FS_MC_DEFAULT;
1844 		break;
1845 
1846 	case IB_FLOW_ATTR_SNIFFER:
1847 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1848 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1849 		break;
1850 
1851 	default:
1852 		err = -EINVAL;
1853 		goto err_free;
1854 	}
1855 
1856 	while (i < ARRAY_SIZE(type) && type[i]) {
1857 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1858 					    &mflow->reg_id[i].id);
1859 		if (err)
1860 			goto err_create_flow;
1861 		if (is_bonded) {
1862 			/* Application always sees one port so the mirror rule
1863 			 * must be on port #2
1864 			 */
1865 			flow_attr->port = 2;
1866 			err = __mlx4_ib_create_flow(qp, flow_attr,
1867 						    domain, type[j],
1868 						    &mflow->reg_id[j].mirror);
1869 			flow_attr->port = 1;
1870 			if (err)
1871 				goto err_create_flow;
1872 			j++;
1873 		}
1874 
1875 		i++;
1876 	}
1877 
1878 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1879 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1880 					       &mflow->reg_id[i].id);
1881 		if (err)
1882 			goto err_create_flow;
1883 
1884 		if (is_bonded) {
1885 			flow_attr->port = 2;
1886 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1887 						       &mflow->reg_id[j].mirror);
1888 			flow_attr->port = 1;
1889 			if (err)
1890 				goto err_create_flow;
1891 			j++;
1892 		}
1893 		/* function to create mirror rule */
1894 		i++;
1895 	}
1896 
1897 	return &mflow->ibflow;
1898 
1899 err_create_flow:
1900 	while (i) {
1901 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1902 					     mflow->reg_id[i].id);
1903 		i--;
1904 	}
1905 
1906 	while (j) {
1907 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1908 					     mflow->reg_id[j].mirror);
1909 		j--;
1910 	}
1911 err_free:
1912 	kfree(mflow);
1913 	return ERR_PTR(err);
1914 }
1915 
1916 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1917 {
1918 	int err, ret = 0;
1919 	int i = 0;
1920 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1921 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1922 
1923 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1924 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1925 		if (err)
1926 			ret = err;
1927 		if (mflow->reg_id[i].mirror) {
1928 			err = __mlx4_ib_destroy_flow(mdev->dev,
1929 						     mflow->reg_id[i].mirror);
1930 			if (err)
1931 				ret = err;
1932 		}
1933 		i++;
1934 	}
1935 
1936 	kfree(mflow);
1937 	return ret;
1938 }
1939 
1940 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1941 {
1942 	int err;
1943 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1944 	struct mlx4_dev	*dev = mdev->dev;
1945 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1946 	struct mlx4_ib_steering *ib_steering = NULL;
1947 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1948 	struct mlx4_flow_reg_id	reg_id;
1949 
1950 	if (mdev->dev->caps.steering_mode ==
1951 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1952 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1953 		if (!ib_steering)
1954 			return -ENOMEM;
1955 	}
1956 
1957 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1958 				    !!(mqp->flags &
1959 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1960 				    prot, &reg_id.id);
1961 	if (err) {
1962 		pr_err("multicast attach op failed, err %d\n", err);
1963 		goto err_malloc;
1964 	}
1965 
1966 	reg_id.mirror = 0;
1967 	if (mlx4_is_bonded(dev)) {
1968 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1969 					    (mqp->port == 1) ? 2 : 1,
1970 					    !!(mqp->flags &
1971 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1972 					    prot, &reg_id.mirror);
1973 		if (err)
1974 			goto err_add;
1975 	}
1976 
1977 	err = add_gid_entry(ibqp, gid);
1978 	if (err)
1979 		goto err_add;
1980 
1981 	if (ib_steering) {
1982 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1983 		ib_steering->reg_id = reg_id;
1984 		mutex_lock(&mqp->mutex);
1985 		list_add(&ib_steering->list, &mqp->steering_rules);
1986 		mutex_unlock(&mqp->mutex);
1987 	}
1988 	return 0;
1989 
1990 err_add:
1991 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1992 			      prot, reg_id.id);
1993 	if (reg_id.mirror)
1994 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1995 				      prot, reg_id.mirror);
1996 err_malloc:
1997 	kfree(ib_steering);
1998 
1999 	return err;
2000 }
2001 
2002 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2003 {
2004 	struct mlx4_ib_gid_entry *ge;
2005 	struct mlx4_ib_gid_entry *tmp;
2006 	struct mlx4_ib_gid_entry *ret = NULL;
2007 
2008 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2009 		if (!memcmp(raw, ge->gid.raw, 16)) {
2010 			ret = ge;
2011 			break;
2012 		}
2013 	}
2014 
2015 	return ret;
2016 }
2017 
2018 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2019 {
2020 	int err;
2021 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2022 	struct mlx4_dev *dev = mdev->dev;
2023 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2024 	struct net_device *ndev;
2025 	struct mlx4_ib_gid_entry *ge;
2026 	struct mlx4_flow_reg_id reg_id = {0, 0};
2027 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2028 
2029 	if (mdev->dev->caps.steering_mode ==
2030 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2031 		struct mlx4_ib_steering *ib_steering;
2032 
2033 		mutex_lock(&mqp->mutex);
2034 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2035 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2036 				list_del(&ib_steering->list);
2037 				break;
2038 			}
2039 		}
2040 		mutex_unlock(&mqp->mutex);
2041 		if (&ib_steering->list == &mqp->steering_rules) {
2042 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2043 			return -EINVAL;
2044 		}
2045 		reg_id = ib_steering->reg_id;
2046 		kfree(ib_steering);
2047 	}
2048 
2049 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2050 				    prot, reg_id.id);
2051 	if (err)
2052 		return err;
2053 
2054 	if (mlx4_is_bonded(dev)) {
2055 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2056 					    prot, reg_id.mirror);
2057 		if (err)
2058 			return err;
2059 	}
2060 
2061 	mutex_lock(&mqp->mutex);
2062 	ge = find_gid_entry(mqp, gid->raw);
2063 	if (ge) {
2064 		spin_lock_bh(&mdev->iboe.lock);
2065 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2066 		if (ndev)
2067 			dev_hold(ndev);
2068 		spin_unlock_bh(&mdev->iboe.lock);
2069 		if (ndev)
2070 			dev_put(ndev);
2071 		list_del(&ge->list);
2072 		kfree(ge);
2073 	} else
2074 		pr_warn("could not find mgid entry\n");
2075 
2076 	mutex_unlock(&mqp->mutex);
2077 
2078 	return 0;
2079 }
2080 
2081 static int init_node_data(struct mlx4_ib_dev *dev)
2082 {
2083 	struct ib_smp *in_mad  = NULL;
2084 	struct ib_smp *out_mad = NULL;
2085 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2086 	int err = -ENOMEM;
2087 
2088 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2089 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2090 	if (!in_mad || !out_mad)
2091 		goto out;
2092 
2093 	init_query_mad(in_mad);
2094 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2095 	if (mlx4_is_master(dev->dev))
2096 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2097 
2098 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2099 	if (err)
2100 		goto out;
2101 
2102 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2103 
2104 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2105 
2106 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2107 	if (err)
2108 		goto out;
2109 
2110 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2111 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2112 
2113 out:
2114 	kfree(in_mad);
2115 	kfree(out_mad);
2116 	return err;
2117 }
2118 
2119 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2120 			char *buf)
2121 {
2122 	struct mlx4_ib_dev *dev =
2123 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2124 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2125 }
2126 
2127 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2128 			char *buf)
2129 {
2130 	struct mlx4_ib_dev *dev =
2131 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2132 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2133 }
2134 
2135 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2136 			  char *buf)
2137 {
2138 	struct mlx4_ib_dev *dev =
2139 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2140 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2141 		       dev->dev->board_id);
2142 }
2143 
2144 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2145 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2146 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2147 
2148 static struct device_attribute *mlx4_class_attributes[] = {
2149 	&dev_attr_hw_rev,
2150 	&dev_attr_hca_type,
2151 	&dev_attr_board_id
2152 };
2153 
2154 struct diag_counter {
2155 	const char *name;
2156 	u32 offset;
2157 };
2158 
2159 #define DIAG_COUNTER(_name, _offset)			\
2160 	{ .name = #_name, .offset = _offset }
2161 
2162 static const struct diag_counter diag_basic[] = {
2163 	DIAG_COUNTER(rq_num_lle, 0x00),
2164 	DIAG_COUNTER(sq_num_lle, 0x04),
2165 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2166 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2167 	DIAG_COUNTER(rq_num_lpe, 0x18),
2168 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2169 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2170 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2171 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2172 	DIAG_COUNTER(sq_num_bre, 0x34),
2173 	DIAG_COUNTER(sq_num_rire, 0x44),
2174 	DIAG_COUNTER(rq_num_rire, 0x48),
2175 	DIAG_COUNTER(sq_num_rae, 0x4C),
2176 	DIAG_COUNTER(rq_num_rae, 0x50),
2177 	DIAG_COUNTER(sq_num_roe, 0x54),
2178 	DIAG_COUNTER(sq_num_tree, 0x5C),
2179 	DIAG_COUNTER(sq_num_rree, 0x64),
2180 	DIAG_COUNTER(rq_num_rnr, 0x68),
2181 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2182 	DIAG_COUNTER(rq_num_oos, 0x100),
2183 	DIAG_COUNTER(sq_num_oos, 0x104),
2184 };
2185 
2186 static const struct diag_counter diag_ext[] = {
2187 	DIAG_COUNTER(rq_num_dup, 0x130),
2188 	DIAG_COUNTER(sq_num_to, 0x134),
2189 };
2190 
2191 static const struct diag_counter diag_device_only[] = {
2192 	DIAG_COUNTER(num_cqovf, 0x1A0),
2193 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2194 };
2195 
2196 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2197 						    u8 port_num)
2198 {
2199 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2200 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2201 
2202 	if (!diag[!!port_num].name)
2203 		return NULL;
2204 
2205 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2206 					  diag[!!port_num].num_counters,
2207 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2208 }
2209 
2210 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2211 				struct rdma_hw_stats *stats,
2212 				u8 port, int index)
2213 {
2214 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2215 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2216 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2217 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2218 	int ret;
2219 	int i;
2220 
2221 	ret = mlx4_query_diag_counters(dev->dev,
2222 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2223 				       diag[!!port].offset, hw_value,
2224 				       diag[!!port].num_counters, port);
2225 
2226 	if (ret)
2227 		return ret;
2228 
2229 	for (i = 0; i < diag[!!port].num_counters; i++)
2230 		stats->value[i] = hw_value[i];
2231 
2232 	return diag[!!port].num_counters;
2233 }
2234 
2235 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2236 					 const char ***name,
2237 					 u32 **offset,
2238 					 u32 *num,
2239 					 bool port)
2240 {
2241 	u32 num_counters;
2242 
2243 	num_counters = ARRAY_SIZE(diag_basic);
2244 
2245 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2246 		num_counters += ARRAY_SIZE(diag_ext);
2247 
2248 	if (!port)
2249 		num_counters += ARRAY_SIZE(diag_device_only);
2250 
2251 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2252 	if (!*name)
2253 		return -ENOMEM;
2254 
2255 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2256 	if (!*offset)
2257 		goto err_name;
2258 
2259 	*num = num_counters;
2260 
2261 	return 0;
2262 
2263 err_name:
2264 	kfree(*name);
2265 	return -ENOMEM;
2266 }
2267 
2268 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2269 				       const char **name,
2270 				       u32 *offset,
2271 				       bool port)
2272 {
2273 	int i;
2274 	int j;
2275 
2276 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2277 		name[i] = diag_basic[i].name;
2278 		offset[i] = diag_basic[i].offset;
2279 	}
2280 
2281 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2282 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2283 			name[j] = diag_ext[i].name;
2284 			offset[j] = diag_ext[i].offset;
2285 		}
2286 	}
2287 
2288 	if (!port) {
2289 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2290 			name[j] = diag_device_only[i].name;
2291 			offset[j] = diag_device_only[i].offset;
2292 		}
2293 	}
2294 }
2295 
2296 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2297 {
2298 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2299 	int i;
2300 	int ret;
2301 	bool per_port = !!(ibdev->dev->caps.flags2 &
2302 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2303 
2304 	if (mlx4_is_slave(ibdev->dev))
2305 		return 0;
2306 
2307 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2308 		/* i == 1 means we are building port counters */
2309 		if (i && !per_port)
2310 			continue;
2311 
2312 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2313 						    &diag[i].offset,
2314 						    &diag[i].num_counters, i);
2315 		if (ret)
2316 			goto err_alloc;
2317 
2318 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2319 					   diag[i].offset, i);
2320 	}
2321 
2322 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2323 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2324 
2325 	return 0;
2326 
2327 err_alloc:
2328 	if (i) {
2329 		kfree(diag[i - 1].name);
2330 		kfree(diag[i - 1].offset);
2331 	}
2332 
2333 	return ret;
2334 }
2335 
2336 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2337 {
2338 	int i;
2339 
2340 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2341 		kfree(ibdev->diag_counters[i].offset);
2342 		kfree(ibdev->diag_counters[i].name);
2343 	}
2344 }
2345 
2346 #define MLX4_IB_INVALID_MAC	((u64)-1)
2347 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2348 			       struct net_device *dev,
2349 			       int port)
2350 {
2351 	u64 new_smac = 0;
2352 	u64 release_mac = MLX4_IB_INVALID_MAC;
2353 	struct mlx4_ib_qp *qp;
2354 
2355 	read_lock(&dev_base_lock);
2356 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2357 	read_unlock(&dev_base_lock);
2358 
2359 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2360 
2361 	/* no need for update QP1 and mac registration in non-SRIOV */
2362 	if (!mlx4_is_mfunc(ibdev->dev))
2363 		return;
2364 
2365 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2366 	qp = ibdev->qp1_proxy[port - 1];
2367 	if (qp) {
2368 		int new_smac_index;
2369 		u64 old_smac;
2370 		struct mlx4_update_qp_params update_params;
2371 
2372 		mutex_lock(&qp->mutex);
2373 		old_smac = qp->pri.smac;
2374 		if (new_smac == old_smac)
2375 			goto unlock;
2376 
2377 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2378 
2379 		if (new_smac_index < 0)
2380 			goto unlock;
2381 
2382 		update_params.smac_index = new_smac_index;
2383 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2384 				   &update_params)) {
2385 			release_mac = new_smac;
2386 			goto unlock;
2387 		}
2388 		/* if old port was zero, no mac was yet registered for this QP */
2389 		if (qp->pri.smac_port)
2390 			release_mac = old_smac;
2391 		qp->pri.smac = new_smac;
2392 		qp->pri.smac_port = port;
2393 		qp->pri.smac_index = new_smac_index;
2394 	}
2395 
2396 unlock:
2397 	if (release_mac != MLX4_IB_INVALID_MAC)
2398 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2399 	if (qp)
2400 		mutex_unlock(&qp->mutex);
2401 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2402 }
2403 
2404 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2405 				 struct net_device *dev,
2406 				 unsigned long event)
2407 
2408 {
2409 	struct mlx4_ib_iboe *iboe;
2410 	int update_qps_port = -1;
2411 	int port;
2412 
2413 	ASSERT_RTNL();
2414 
2415 	iboe = &ibdev->iboe;
2416 
2417 	spin_lock_bh(&iboe->lock);
2418 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2419 
2420 		iboe->netdevs[port - 1] =
2421 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2422 
2423 		if (dev == iboe->netdevs[port - 1] &&
2424 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2425 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2426 			update_qps_port = port;
2427 
2428 	}
2429 	spin_unlock_bh(&iboe->lock);
2430 
2431 	if (update_qps_port > 0)
2432 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2433 }
2434 
2435 static int mlx4_ib_netdev_event(struct notifier_block *this,
2436 				unsigned long event, void *ptr)
2437 {
2438 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2439 	struct mlx4_ib_dev *ibdev;
2440 
2441 	if (!net_eq(dev_net(dev), &init_net))
2442 		return NOTIFY_DONE;
2443 
2444 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2445 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2446 
2447 	return NOTIFY_DONE;
2448 }
2449 
2450 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2451 {
2452 	int port;
2453 	int slave;
2454 	int i;
2455 
2456 	if (mlx4_is_master(ibdev->dev)) {
2457 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2458 		     ++slave) {
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.virt2phys_pkey[slave][port - 1][i] =
2464 					/* master has the identity virt2phys pkey mapping */
2465 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2466 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2467 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2468 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2469 				}
2470 			}
2471 		}
2472 		/* initialize pkey cache */
2473 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2474 			for (i = 0;
2475 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2476 			     ++i)
2477 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2478 					(i) ? 0 : 0xFFFF;
2479 		}
2480 	}
2481 }
2482 
2483 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2484 {
2485 	int i, j, eq = 0, total_eqs = 0;
2486 
2487 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2488 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2489 	if (!ibdev->eq_table)
2490 		return;
2491 
2492 	for (i = 1; i <= dev->caps.num_ports; i++) {
2493 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2494 		     j++, total_eqs++) {
2495 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2496 				continue;
2497 			ibdev->eq_table[eq] = total_eqs;
2498 			if (!mlx4_assign_eq(dev, i,
2499 					    &ibdev->eq_table[eq]))
2500 				eq++;
2501 			else
2502 				ibdev->eq_table[eq] = -1;
2503 		}
2504 	}
2505 
2506 	for (i = eq; i < dev->caps.num_comp_vectors;
2507 	     ibdev->eq_table[i++] = -1)
2508 		;
2509 
2510 	/* Advertise the new number of EQs to clients */
2511 	ibdev->ib_dev.num_comp_vectors = eq;
2512 }
2513 
2514 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2515 {
2516 	int i;
2517 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2518 
2519 	/* no eqs were allocated */
2520 	if (!ibdev->eq_table)
2521 		return;
2522 
2523 	/* Reset the advertised EQ number */
2524 	ibdev->ib_dev.num_comp_vectors = 0;
2525 
2526 	for (i = 0; i < total_eqs; i++)
2527 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2528 
2529 	kfree(ibdev->eq_table);
2530 	ibdev->eq_table = NULL;
2531 }
2532 
2533 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2534 			       struct ib_port_immutable *immutable)
2535 {
2536 	struct ib_port_attr attr;
2537 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2538 	int err;
2539 
2540 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2541 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2542 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2543 	} else {
2544 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2545 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2546 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2547 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2548 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2549 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2550 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2551 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2552 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2553 	}
2554 
2555 	err = ib_query_port(ibdev, port_num, &attr);
2556 	if (err)
2557 		return err;
2558 
2559 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2560 	immutable->gid_tbl_len = attr.gid_tbl_len;
2561 
2562 	return 0;
2563 }
2564 
2565 static void get_fw_ver_str(struct ib_device *device, char *str,
2566 			   size_t str_len)
2567 {
2568 	struct mlx4_ib_dev *dev =
2569 		container_of(device, struct mlx4_ib_dev, ib_dev);
2570 	snprintf(str, str_len, "%d.%d.%d",
2571 		 (int) (dev->dev->caps.fw_ver >> 32),
2572 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2573 		 (int) dev->dev->caps.fw_ver & 0xffff);
2574 }
2575 
2576 static void *mlx4_ib_add(struct mlx4_dev *dev)
2577 {
2578 	struct mlx4_ib_dev *ibdev;
2579 	int num_ports = 0;
2580 	int i, j;
2581 	int err;
2582 	struct mlx4_ib_iboe *iboe;
2583 	int ib_num_ports = 0;
2584 	int num_req_counters;
2585 	int allocated;
2586 	u32 counter_index;
2587 	struct counter_index *new_counter_index = NULL;
2588 
2589 	pr_info_once("%s", mlx4_ib_version);
2590 
2591 	num_ports = 0;
2592 	mlx4_foreach_ib_transport_port(i, dev)
2593 		num_ports++;
2594 
2595 	/* No point in registering a device with no ports... */
2596 	if (num_ports == 0)
2597 		return NULL;
2598 
2599 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2600 	if (!ibdev) {
2601 		dev_err(&dev->persist->pdev->dev,
2602 			"Device struct alloc failed\n");
2603 		return NULL;
2604 	}
2605 
2606 	iboe = &ibdev->iboe;
2607 
2608 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2609 		goto err_dealloc;
2610 
2611 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2612 		goto err_pd;
2613 
2614 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2615 				 PAGE_SIZE);
2616 	if (!ibdev->uar_map)
2617 		goto err_uar;
2618 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2619 
2620 	ibdev->dev = dev;
2621 	ibdev->bond_next_port	= 0;
2622 
2623 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2624 	ibdev->ib_dev.owner		= THIS_MODULE;
2625 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2626 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2627 	ibdev->num_ports		= num_ports;
2628 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2629 						1 : ibdev->num_ports;
2630 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2631 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2632 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2633 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2634 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2635 
2636 	if (dev->caps.userspace_caps)
2637 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2638 	else
2639 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2640 
2641 	ibdev->ib_dev.uverbs_cmd_mask	=
2642 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2643 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2644 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2645 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2646 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2647 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2648 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2649 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2650 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2651 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2652 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2653 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2654 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2655 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2656 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2657 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2658 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2659 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2660 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2661 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2662 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2663 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2664 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2665 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2666 
2667 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2668 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2669 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2670 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2671 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2672 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2673 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2674 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2675 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2676 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2677 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2678 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2679 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2680 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2681 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2682 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2683 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2684 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2685 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2686 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2687 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2688 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2689 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2690 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2691 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2692 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2693 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2694 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2695 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2696 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2697 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2698 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2699 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2700 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2701 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2702 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2703 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2704 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2705 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2706 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2707 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2708 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2709 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2710 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2711 
2712 	if (!mlx4_is_slave(ibdev->dev)) {
2713 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2714 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2715 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2716 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2717 	}
2718 
2719 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2720 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2721 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2722 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2723 
2724 		ibdev->ib_dev.uverbs_cmd_mask |=
2725 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2726 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2727 	}
2728 
2729 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2730 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2731 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2732 		ibdev->ib_dev.uverbs_cmd_mask |=
2733 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2734 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2735 	}
2736 
2737 	if (check_flow_steering_support(dev)) {
2738 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2739 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2740 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2741 
2742 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2743 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2744 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2745 	}
2746 
2747 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2748 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2749 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2750 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2751 
2752 	mlx4_ib_alloc_eqs(dev, ibdev);
2753 
2754 	spin_lock_init(&iboe->lock);
2755 
2756 	if (init_node_data(ibdev))
2757 		goto err_map;
2758 	mlx4_init_sl2vl_tbl(ibdev);
2759 
2760 	for (i = 0; i < ibdev->num_ports; ++i) {
2761 		mutex_init(&ibdev->counters_table[i].mutex);
2762 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2763 	}
2764 
2765 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2766 	for (i = 0; i < num_req_counters; ++i) {
2767 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2768 		allocated = 0;
2769 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2770 						IB_LINK_LAYER_ETHERNET) {
2771 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2772 			/* if failed to allocate a new counter, use default */
2773 			if (err)
2774 				counter_index =
2775 					mlx4_get_default_counter_index(dev,
2776 								       i + 1);
2777 			else
2778 				allocated = 1;
2779 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2780 			counter_index = mlx4_get_default_counter_index(dev,
2781 								       i + 1);
2782 		}
2783 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2784 					    GFP_KERNEL);
2785 		if (!new_counter_index) {
2786 			if (allocated)
2787 				mlx4_counter_free(ibdev->dev, counter_index);
2788 			goto err_counter;
2789 		}
2790 		new_counter_index->index = counter_index;
2791 		new_counter_index->allocated = allocated;
2792 		list_add_tail(&new_counter_index->list,
2793 			      &ibdev->counters_table[i].counters_list);
2794 		ibdev->counters_table[i].default_counter = counter_index;
2795 		pr_info("counter index %d for port %d allocated %d\n",
2796 			counter_index, i + 1, allocated);
2797 	}
2798 	if (mlx4_is_bonded(dev))
2799 		for (i = 1; i < ibdev->num_ports ; ++i) {
2800 			new_counter_index =
2801 					kmalloc(sizeof(struct counter_index),
2802 						GFP_KERNEL);
2803 			if (!new_counter_index)
2804 				goto err_counter;
2805 			new_counter_index->index = counter_index;
2806 			new_counter_index->allocated = 0;
2807 			list_add_tail(&new_counter_index->list,
2808 				      &ibdev->counters_table[i].counters_list);
2809 			ibdev->counters_table[i].default_counter =
2810 								counter_index;
2811 		}
2812 
2813 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2814 		ib_num_ports++;
2815 
2816 	spin_lock_init(&ibdev->sm_lock);
2817 	mutex_init(&ibdev->cap_mask_mutex);
2818 	INIT_LIST_HEAD(&ibdev->qp_list);
2819 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2820 
2821 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2822 	    ib_num_ports) {
2823 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2824 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2825 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2826 					    &ibdev->steer_qpn_base, 0);
2827 		if (err)
2828 			goto err_counter;
2829 
2830 		ibdev->ib_uc_qpns_bitmap =
2831 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2832 				sizeof(long),
2833 				GFP_KERNEL);
2834 		if (!ibdev->ib_uc_qpns_bitmap)
2835 			goto err_steer_qp_release;
2836 
2837 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2838 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2839 				    ibdev->steer_qpn_count);
2840 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2841 					dev, ibdev->steer_qpn_base,
2842 					ibdev->steer_qpn_base +
2843 					ibdev->steer_qpn_count - 1);
2844 			if (err)
2845 				goto err_steer_free_bitmap;
2846 		} else {
2847 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2848 				    ibdev->steer_qpn_count);
2849 		}
2850 	}
2851 
2852 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2853 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2854 
2855 	if (mlx4_ib_alloc_diag_counters(ibdev))
2856 		goto err_steer_free_bitmap;
2857 
2858 	if (ib_register_device(&ibdev->ib_dev, NULL))
2859 		goto err_diag_counters;
2860 
2861 	if (mlx4_ib_mad_init(ibdev))
2862 		goto err_reg;
2863 
2864 	if (mlx4_ib_init_sriov(ibdev))
2865 		goto err_mad;
2866 
2867 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2868 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2869 		if (!iboe->nb.notifier_call) {
2870 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2871 			err = register_netdevice_notifier(&iboe->nb);
2872 			if (err) {
2873 				iboe->nb.notifier_call = NULL;
2874 				goto err_notif;
2875 			}
2876 		}
2877 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2878 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2879 			if (err) {
2880 				goto err_notif;
2881 			}
2882 		}
2883 	}
2884 
2885 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2886 		if (device_create_file(&ibdev->ib_dev.dev,
2887 				       mlx4_class_attributes[j]))
2888 			goto err_notif;
2889 	}
2890 
2891 	ibdev->ib_active = true;
2892 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2893 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2894 					 &ibdev->ib_dev);
2895 
2896 	if (mlx4_is_mfunc(ibdev->dev))
2897 		init_pkeys(ibdev);
2898 
2899 	/* create paravirt contexts for any VFs which are active */
2900 	if (mlx4_is_master(ibdev->dev)) {
2901 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2902 			if (j == mlx4_master_func_num(ibdev->dev))
2903 				continue;
2904 			if (mlx4_is_slave_active(ibdev->dev, j))
2905 				do_slave_init(ibdev, j, 1);
2906 		}
2907 	}
2908 	return ibdev;
2909 
2910 err_notif:
2911 	if (ibdev->iboe.nb.notifier_call) {
2912 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2913 			pr_warn("failure unregistering notifier\n");
2914 		ibdev->iboe.nb.notifier_call = NULL;
2915 	}
2916 	flush_workqueue(wq);
2917 
2918 	mlx4_ib_close_sriov(ibdev);
2919 
2920 err_mad:
2921 	mlx4_ib_mad_cleanup(ibdev);
2922 
2923 err_reg:
2924 	ib_unregister_device(&ibdev->ib_dev);
2925 
2926 err_diag_counters:
2927 	mlx4_ib_diag_cleanup(ibdev);
2928 
2929 err_steer_free_bitmap:
2930 	kfree(ibdev->ib_uc_qpns_bitmap);
2931 
2932 err_steer_qp_release:
2933 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2934 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2935 				      ibdev->steer_qpn_count);
2936 err_counter:
2937 	for (i = 0; i < ibdev->num_ports; ++i)
2938 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2939 
2940 err_map:
2941 	iounmap(ibdev->uar_map);
2942 
2943 err_uar:
2944 	mlx4_uar_free(dev, &ibdev->priv_uar);
2945 
2946 err_pd:
2947 	mlx4_pd_free(dev, ibdev->priv_pdn);
2948 
2949 err_dealloc:
2950 	ib_dealloc_device(&ibdev->ib_dev);
2951 
2952 	return NULL;
2953 }
2954 
2955 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2956 {
2957 	int offset;
2958 
2959 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2960 
2961 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2962 					 dev->steer_qpn_count,
2963 					 get_count_order(count));
2964 	if (offset < 0)
2965 		return offset;
2966 
2967 	*qpn = dev->steer_qpn_base + offset;
2968 	return 0;
2969 }
2970 
2971 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2972 {
2973 	if (!qpn ||
2974 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2975 		return;
2976 
2977 	BUG_ON(qpn < dev->steer_qpn_base);
2978 
2979 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2980 			      qpn - dev->steer_qpn_base,
2981 			      get_count_order(count));
2982 }
2983 
2984 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2985 			 int is_attach)
2986 {
2987 	int err;
2988 	size_t flow_size;
2989 	struct ib_flow_attr *flow = NULL;
2990 	struct ib_flow_spec_ib *ib_spec;
2991 
2992 	if (is_attach) {
2993 		flow_size = sizeof(struct ib_flow_attr) +
2994 			    sizeof(struct ib_flow_spec_ib);
2995 		flow = kzalloc(flow_size, GFP_KERNEL);
2996 		if (!flow)
2997 			return -ENOMEM;
2998 		flow->port = mqp->port;
2999 		flow->num_of_specs = 1;
3000 		flow->size = flow_size;
3001 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3002 		ib_spec->type = IB_FLOW_SPEC_IB;
3003 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3004 		/* Add an empty rule for IB L2 */
3005 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3006 
3007 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3008 					    IB_FLOW_DOMAIN_NIC,
3009 					    MLX4_FS_REGULAR,
3010 					    &mqp->reg_id);
3011 	} else {
3012 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3013 	}
3014 	kfree(flow);
3015 	return err;
3016 }
3017 
3018 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3019 {
3020 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3021 	int p;
3022 	int i;
3023 
3024 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3025 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3026 	ibdev->ib_active = false;
3027 	flush_workqueue(wq);
3028 
3029 	mlx4_ib_close_sriov(ibdev);
3030 	mlx4_ib_mad_cleanup(ibdev);
3031 	ib_unregister_device(&ibdev->ib_dev);
3032 	mlx4_ib_diag_cleanup(ibdev);
3033 	if (ibdev->iboe.nb.notifier_call) {
3034 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3035 			pr_warn("failure unregistering notifier\n");
3036 		ibdev->iboe.nb.notifier_call = NULL;
3037 	}
3038 
3039 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
3040 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3041 				      ibdev->steer_qpn_count);
3042 		kfree(ibdev->ib_uc_qpns_bitmap);
3043 	}
3044 
3045 	iounmap(ibdev->uar_map);
3046 	for (p = 0; p < ibdev->num_ports; ++p)
3047 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3048 
3049 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3050 		mlx4_CLOSE_PORT(dev, p);
3051 
3052 	mlx4_ib_free_eqs(dev, ibdev);
3053 
3054 	mlx4_uar_free(dev, &ibdev->priv_uar);
3055 	mlx4_pd_free(dev, ibdev->priv_pdn);
3056 	ib_dealloc_device(&ibdev->ib_dev);
3057 }
3058 
3059 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3060 {
3061 	struct mlx4_ib_demux_work **dm = NULL;
3062 	struct mlx4_dev *dev = ibdev->dev;
3063 	int i;
3064 	unsigned long flags;
3065 	struct mlx4_active_ports actv_ports;
3066 	unsigned int ports;
3067 	unsigned int first_port;
3068 
3069 	if (!mlx4_is_master(dev))
3070 		return;
3071 
3072 	actv_ports = mlx4_get_active_ports(dev, slave);
3073 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3074 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3075 
3076 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3077 	if (!dm)
3078 		return;
3079 
3080 	for (i = 0; i < ports; i++) {
3081 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3082 		if (!dm[i]) {
3083 			while (--i >= 0)
3084 				kfree(dm[i]);
3085 			goto out;
3086 		}
3087 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3088 		dm[i]->port = first_port + i + 1;
3089 		dm[i]->slave = slave;
3090 		dm[i]->do_init = do_init;
3091 		dm[i]->dev = ibdev;
3092 	}
3093 	/* initialize or tear down tunnel QPs for the slave */
3094 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3095 	if (!ibdev->sriov.is_going_down) {
3096 		for (i = 0; i < ports; i++)
3097 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3098 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3099 	} else {
3100 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3101 		for (i = 0; i < ports; i++)
3102 			kfree(dm[i]);
3103 	}
3104 out:
3105 	kfree(dm);
3106 	return;
3107 }
3108 
3109 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3110 {
3111 	struct mlx4_ib_qp *mqp;
3112 	unsigned long flags_qp;
3113 	unsigned long flags_cq;
3114 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3115 	struct list_head    cq_notify_list;
3116 	struct mlx4_cq *mcq;
3117 	unsigned long flags;
3118 
3119 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3120 	INIT_LIST_HEAD(&cq_notify_list);
3121 
3122 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3123 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3124 
3125 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3126 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3127 		if (mqp->sq.tail != mqp->sq.head) {
3128 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3129 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3130 			if (send_mcq->mcq.comp &&
3131 			    mqp->ibqp.send_cq->comp_handler) {
3132 				if (!send_mcq->mcq.reset_notify_added) {
3133 					send_mcq->mcq.reset_notify_added = 1;
3134 					list_add_tail(&send_mcq->mcq.reset_notify,
3135 						      &cq_notify_list);
3136 				}
3137 			}
3138 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3139 		}
3140 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3141 		/* Now, handle the QP's receive queue */
3142 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3143 		/* no handling is needed for SRQ */
3144 		if (!mqp->ibqp.srq) {
3145 			if (mqp->rq.tail != mqp->rq.head) {
3146 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3147 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3148 				if (recv_mcq->mcq.comp &&
3149 				    mqp->ibqp.recv_cq->comp_handler) {
3150 					if (!recv_mcq->mcq.reset_notify_added) {
3151 						recv_mcq->mcq.reset_notify_added = 1;
3152 						list_add_tail(&recv_mcq->mcq.reset_notify,
3153 							      &cq_notify_list);
3154 					}
3155 				}
3156 				spin_unlock_irqrestore(&recv_mcq->lock,
3157 						       flags_cq);
3158 			}
3159 		}
3160 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3161 	}
3162 
3163 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3164 		mcq->comp(mcq);
3165 	}
3166 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3167 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3168 }
3169 
3170 static void handle_bonded_port_state_event(struct work_struct *work)
3171 {
3172 	struct ib_event_work *ew =
3173 		container_of(work, struct ib_event_work, work);
3174 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3175 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3176 	int i;
3177 	struct ib_event ibev;
3178 
3179 	kfree(ew);
3180 	spin_lock_bh(&ibdev->iboe.lock);
3181 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3182 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3183 		enum ib_port_state curr_port_state;
3184 
3185 		if (!curr_netdev)
3186 			continue;
3187 
3188 		curr_port_state =
3189 			(netif_running(curr_netdev) &&
3190 			 netif_carrier_ok(curr_netdev)) ?
3191 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3192 
3193 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3194 			curr_port_state : IB_PORT_ACTIVE;
3195 	}
3196 	spin_unlock_bh(&ibdev->iboe.lock);
3197 
3198 	ibev.device = &ibdev->ib_dev;
3199 	ibev.element.port_num = 1;
3200 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3201 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3202 
3203 	ib_dispatch_event(&ibev);
3204 }
3205 
3206 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3207 {
3208 	u64 sl2vl;
3209 	int err;
3210 
3211 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3212 	if (err) {
3213 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3214 		       port, err);
3215 		sl2vl = 0;
3216 	}
3217 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3218 }
3219 
3220 static void ib_sl2vl_update_work(struct work_struct *work)
3221 {
3222 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3223 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3224 	int port = ew->port;
3225 
3226 	mlx4_ib_sl2vl_update(mdev, port);
3227 
3228 	kfree(ew);
3229 }
3230 
3231 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3232 				     int port)
3233 {
3234 	struct ib_event_work *ew;
3235 
3236 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3237 	if (ew) {
3238 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3239 		ew->port = port;
3240 		ew->ib_dev = ibdev;
3241 		queue_work(wq, &ew->work);
3242 	}
3243 }
3244 
3245 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3246 			  enum mlx4_dev_event event, unsigned long param)
3247 {
3248 	struct ib_event ibev;
3249 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3250 	struct mlx4_eqe *eqe = NULL;
3251 	struct ib_event_work *ew;
3252 	int p = 0;
3253 
3254 	if (mlx4_is_bonded(dev) &&
3255 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3256 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3257 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3258 		if (!ew)
3259 			return;
3260 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3261 		ew->ib_dev = ibdev;
3262 		queue_work(wq, &ew->work);
3263 		return;
3264 	}
3265 
3266 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3267 		eqe = (struct mlx4_eqe *)param;
3268 	else
3269 		p = (int) param;
3270 
3271 	switch (event) {
3272 	case MLX4_DEV_EVENT_PORT_UP:
3273 		if (p > ibdev->num_ports)
3274 			return;
3275 		if (!mlx4_is_slave(dev) &&
3276 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3277 			IB_LINK_LAYER_INFINIBAND) {
3278 			if (mlx4_is_master(dev))
3279 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3280 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3281 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3282 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3283 		}
3284 		ibev.event = IB_EVENT_PORT_ACTIVE;
3285 		break;
3286 
3287 	case MLX4_DEV_EVENT_PORT_DOWN:
3288 		if (p > ibdev->num_ports)
3289 			return;
3290 		ibev.event = IB_EVENT_PORT_ERR;
3291 		break;
3292 
3293 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3294 		ibdev->ib_active = false;
3295 		ibev.event = IB_EVENT_DEVICE_FATAL;
3296 		mlx4_ib_handle_catas_error(ibdev);
3297 		break;
3298 
3299 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3300 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3301 		if (!ew)
3302 			break;
3303 
3304 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3305 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3306 		ew->ib_dev = ibdev;
3307 		/* need to queue only for port owner, which uses GEN_EQE */
3308 		if (mlx4_is_master(dev))
3309 			queue_work(wq, &ew->work);
3310 		else
3311 			handle_port_mgmt_change_event(&ew->work);
3312 		return;
3313 
3314 	case MLX4_DEV_EVENT_SLAVE_INIT:
3315 		/* here, p is the slave id */
3316 		do_slave_init(ibdev, p, 1);
3317 		if (mlx4_is_master(dev)) {
3318 			int i;
3319 
3320 			for (i = 1; i <= ibdev->num_ports; i++) {
3321 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3322 					== IB_LINK_LAYER_INFINIBAND)
3323 					mlx4_ib_slave_alias_guid_event(ibdev,
3324 								       p, i,
3325 								       1);
3326 			}
3327 		}
3328 		return;
3329 
3330 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3331 		if (mlx4_is_master(dev)) {
3332 			int i;
3333 
3334 			for (i = 1; i <= ibdev->num_ports; i++) {
3335 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3336 					== IB_LINK_LAYER_INFINIBAND)
3337 					mlx4_ib_slave_alias_guid_event(ibdev,
3338 								       p, i,
3339 								       0);
3340 			}
3341 		}
3342 		/* here, p is the slave id */
3343 		do_slave_init(ibdev, p, 0);
3344 		return;
3345 
3346 	default:
3347 		return;
3348 	}
3349 
3350 	ibev.device	      = ibdev_ptr;
3351 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3352 
3353 	ib_dispatch_event(&ibev);
3354 }
3355 
3356 static struct mlx4_interface mlx4_ib_interface = {
3357 	.add		= mlx4_ib_add,
3358 	.remove		= mlx4_ib_remove,
3359 	.event		= mlx4_ib_event,
3360 	.protocol	= MLX4_PROT_IB_IPV6,
3361 	.flags		= MLX4_INTFF_BONDING
3362 };
3363 
3364 static int __init mlx4_ib_init(void)
3365 {
3366 	int err;
3367 
3368 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3369 	if (!wq)
3370 		return -ENOMEM;
3371 
3372 	err = mlx4_ib_mcg_init();
3373 	if (err)
3374 		goto clean_wq;
3375 
3376 	err = mlx4_register_interface(&mlx4_ib_interface);
3377 	if (err)
3378 		goto clean_mcg;
3379 
3380 	return 0;
3381 
3382 clean_mcg:
3383 	mlx4_ib_mcg_destroy();
3384 
3385 clean_wq:
3386 	destroy_workqueue(wq);
3387 	return err;
3388 }
3389 
3390 static void __exit mlx4_ib_cleanup(void)
3391 {
3392 	mlx4_unregister_interface(&mlx4_ib_interface);
3393 	mlx4_ib_mcg_destroy();
3394 	destroy_workqueue(wq);
3395 }
3396 
3397 module_init(mlx4_ib_init);
3398 module_exit(mlx4_ib_cleanup);
3399