xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision d2574c33)
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 <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
250 {
251 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
252 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
253 	struct mlx4_port_gid_table   *port_gid_table;
254 	int free = -1, found = -1;
255 	int ret = 0;
256 	int hw_update = 0;
257 	int i;
258 	struct gid_entry *gids = NULL;
259 
260 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
261 		return -EINVAL;
262 
263 	if (attr->port_num > MLX4_MAX_PORTS)
264 		return -EINVAL;
265 
266 	if (!context)
267 		return -EINVAL;
268 
269 	port_gid_table = &iboe->gids[attr->port_num - 1];
270 	spin_lock_bh(&iboe->lock);
271 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
272 		if (!memcmp(&port_gid_table->gids[i].gid,
273 			    &attr->gid, sizeof(attr->gid)) &&
274 		    port_gid_table->gids[i].gid_type == attr->gid_type)  {
275 			found = i;
276 			break;
277 		}
278 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
279 			free = i; /* HW has space */
280 	}
281 
282 	if (found < 0) {
283 		if (free < 0) {
284 			ret = -ENOSPC;
285 		} else {
286 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
287 			if (!port_gid_table->gids[free].ctx) {
288 				ret = -ENOMEM;
289 			} else {
290 				*context = port_gid_table->gids[free].ctx;
291 				memcpy(&port_gid_table->gids[free].gid,
292 				       &attr->gid, sizeof(attr->gid));
293 				port_gid_table->gids[free].gid_type = attr->gid_type;
294 				port_gid_table->gids[free].ctx->real_index = free;
295 				port_gid_table->gids[free].ctx->refcount = 1;
296 				hw_update = 1;
297 			}
298 		}
299 	} else {
300 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
301 		*context = ctx;
302 		ctx->refcount++;
303 	}
304 	if (!ret && hw_update) {
305 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
306 				     GFP_ATOMIC);
307 		if (!gids) {
308 			ret = -ENOMEM;
309 		} else {
310 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
311 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
312 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
313 			}
314 		}
315 	}
316 	spin_unlock_bh(&iboe->lock);
317 
318 	if (!ret && hw_update) {
319 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
320 		kfree(gids);
321 	}
322 
323 	return ret;
324 }
325 
326 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
327 {
328 	struct gid_cache_context *ctx = *context;
329 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
330 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
331 	struct mlx4_port_gid_table   *port_gid_table;
332 	int ret = 0;
333 	int hw_update = 0;
334 	struct gid_entry *gids = NULL;
335 
336 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
337 		return -EINVAL;
338 
339 	if (attr->port_num > MLX4_MAX_PORTS)
340 		return -EINVAL;
341 
342 	port_gid_table = &iboe->gids[attr->port_num - 1];
343 	spin_lock_bh(&iboe->lock);
344 	if (ctx) {
345 		ctx->refcount--;
346 		if (!ctx->refcount) {
347 			unsigned int real_index = ctx->real_index;
348 
349 			memset(&port_gid_table->gids[real_index].gid, 0,
350 			       sizeof(port_gid_table->gids[real_index].gid));
351 			kfree(port_gid_table->gids[real_index].ctx);
352 			port_gid_table->gids[real_index].ctx = NULL;
353 			hw_update = 1;
354 		}
355 	}
356 	if (!ret && hw_update) {
357 		int i;
358 
359 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
360 				     GFP_ATOMIC);
361 		if (!gids) {
362 			ret = -ENOMEM;
363 		} else {
364 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
365 				memcpy(&gids[i].gid,
366 				       &port_gid_table->gids[i].gid,
367 				       sizeof(union ib_gid));
368 				gids[i].gid_type =
369 				    port_gid_table->gids[i].gid_type;
370 			}
371 		}
372 	}
373 	spin_unlock_bh(&iboe->lock);
374 
375 	if (!ret && hw_update) {
376 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
377 		kfree(gids);
378 	}
379 	return ret;
380 }
381 
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383 				    const struct ib_gid_attr *attr)
384 {
385 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386 	struct gid_cache_context *ctx = NULL;
387 	struct mlx4_port_gid_table   *port_gid_table;
388 	int real_index = -EINVAL;
389 	int i;
390 	unsigned long flags;
391 	u8 port_num = attr->port_num;
392 
393 	if (port_num > MLX4_MAX_PORTS)
394 		return -EINVAL;
395 
396 	if (mlx4_is_bonded(ibdev->dev))
397 		port_num = 1;
398 
399 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
400 		return attr->index;
401 
402 	spin_lock_irqsave(&iboe->lock, flags);
403 	port_gid_table = &iboe->gids[port_num - 1];
404 
405 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
406 		if (!memcmp(&port_gid_table->gids[i].gid,
407 			    &attr->gid, sizeof(attr->gid)) &&
408 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
409 			ctx = port_gid_table->gids[i].ctx;
410 			break;
411 		}
412 	if (ctx)
413 		real_index = ctx->real_index;
414 	spin_unlock_irqrestore(&iboe->lock, flags);
415 	return real_index;
416 }
417 
418 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
419 				    sizeof(((type *)0)->fld) <= (sz))
420 
421 static int mlx4_ib_query_device(struct ib_device *ibdev,
422 				struct ib_device_attr *props,
423 				struct ib_udata *uhw)
424 {
425 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
426 	struct ib_smp *in_mad  = NULL;
427 	struct ib_smp *out_mad = NULL;
428 	int err;
429 	int have_ib_ports;
430 	struct mlx4_uverbs_ex_query_device cmd;
431 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
432 	struct mlx4_clock_params clock_params;
433 
434 	if (uhw->inlen) {
435 		if (uhw->inlen < sizeof(cmd))
436 			return -EINVAL;
437 
438 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
439 		if (err)
440 			return err;
441 
442 		if (cmd.comp_mask)
443 			return -EINVAL;
444 
445 		if (cmd.reserved)
446 			return -EINVAL;
447 	}
448 
449 	resp.response_length = offsetof(typeof(resp), response_length) +
450 		sizeof(resp.response_length);
451 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
452 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
453 	err = -ENOMEM;
454 	if (!in_mad || !out_mad)
455 		goto out;
456 
457 	init_query_mad(in_mad);
458 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
459 
460 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
461 			   1, NULL, NULL, in_mad, out_mad);
462 	if (err)
463 		goto out;
464 
465 	memset(props, 0, sizeof *props);
466 
467 	have_ib_ports = num_ib_ports(dev->dev);
468 
469 	props->fw_ver = dev->dev->caps.fw_ver;
470 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
471 		IB_DEVICE_PORT_ACTIVE_EVENT		|
472 		IB_DEVICE_SYS_IMAGE_GUID		|
473 		IB_DEVICE_RC_RNR_NAK_GEN		|
474 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
475 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
476 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
477 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
478 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
479 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
480 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
481 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
482 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
484 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
485 	if (dev->dev->caps.max_gso_sz &&
486 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
487 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
488 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
489 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
490 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
491 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
492 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
493 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
494 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
496 		props->device_cap_flags |= IB_DEVICE_XRC;
497 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
498 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
499 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
500 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
501 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
502 		else
503 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
504 	}
505 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
506 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
507 
508 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
509 
510 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
511 		0xffffff;
512 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
513 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
514 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
515 
516 	props->max_mr_size	   = ~0ull;
517 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
518 	props->max_qp		   = dev->dev->quotas.qp;
519 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
520 	props->max_send_sge =
521 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
522 	props->max_recv_sge =
523 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
524 	props->max_sge_rd = MLX4_MAX_SGE_RD;
525 	props->max_cq		   = dev->dev->quotas.cq;
526 	props->max_cqe		   = dev->dev->caps.max_cqes;
527 	props->max_mr		   = dev->dev->quotas.mpt;
528 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
529 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
530 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
531 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
532 	props->max_srq		   = dev->dev->quotas.srq;
533 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
534 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
535 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
536 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
537 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
538 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
539 	props->masked_atomic_cap   = props->atomic_cap;
540 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
541 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
542 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
543 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
544 					   props->max_mcast_grp;
545 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
546 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
547 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
548 	props->max_ah = INT_MAX;
549 
550 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
551 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
552 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
553 			props->rss_caps.max_rwq_indirection_tables =
554 				props->max_qp;
555 			props->rss_caps.max_rwq_indirection_table_size =
556 				dev->dev->caps.max_rss_tbl_sz;
557 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
558 			props->max_wq_type_rq = props->max_qp;
559 		}
560 
561 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
562 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
563 	}
564 
565 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
566 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
567 
568 	if (!mlx4_is_slave(dev->dev))
569 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
570 
571 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
572 		resp.response_length += sizeof(resp.hca_core_clock_offset);
573 		if (!err && !mlx4_is_slave(dev->dev)) {
574 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
575 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
576 		}
577 	}
578 
579 	if (uhw->outlen >= resp.response_length +
580 	    sizeof(resp.max_inl_recv_sz)) {
581 		resp.response_length += sizeof(resp.max_inl_recv_sz);
582 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
583 			sizeof(struct mlx4_wqe_data_seg);
584 	}
585 
586 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
587 		if (props->rss_caps.supported_qpts) {
588 			resp.rss_caps.rx_hash_function =
589 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
590 
591 			resp.rss_caps.rx_hash_fields_mask =
592 				MLX4_IB_RX_HASH_SRC_IPV4 |
593 				MLX4_IB_RX_HASH_DST_IPV4 |
594 				MLX4_IB_RX_HASH_SRC_IPV6 |
595 				MLX4_IB_RX_HASH_DST_IPV6 |
596 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
597 				MLX4_IB_RX_HASH_DST_PORT_TCP |
598 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
599 				MLX4_IB_RX_HASH_DST_PORT_UDP;
600 
601 			if (dev->dev->caps.tunnel_offload_mode ==
602 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
603 				resp.rss_caps.rx_hash_fields_mask |=
604 					MLX4_IB_RX_HASH_INNER;
605 		}
606 		resp.response_length = offsetof(typeof(resp), rss_caps) +
607 				       sizeof(resp.rss_caps);
608 	}
609 
610 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
611 		if (dev->dev->caps.max_gso_sz &&
612 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
613 		    IB_LINK_LAYER_ETHERNET) ||
614 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
615 		    IB_LINK_LAYER_ETHERNET))) {
616 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
617 			resp.tso_caps.supported_qpts |=
618 				1 << IB_QPT_RAW_PACKET;
619 		}
620 		resp.response_length = offsetof(typeof(resp), tso_caps) +
621 				       sizeof(resp.tso_caps);
622 	}
623 
624 	if (uhw->outlen) {
625 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
626 		if (err)
627 			goto out;
628 	}
629 out:
630 	kfree(in_mad);
631 	kfree(out_mad);
632 
633 	return err;
634 }
635 
636 static enum rdma_link_layer
637 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
638 {
639 	struct mlx4_dev *dev = to_mdev(device)->dev;
640 
641 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
642 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
643 }
644 
645 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
646 			      struct ib_port_attr *props, int netw_view)
647 {
648 	struct ib_smp *in_mad  = NULL;
649 	struct ib_smp *out_mad = NULL;
650 	int ext_active_speed;
651 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
652 	int err = -ENOMEM;
653 
654 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
655 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
656 	if (!in_mad || !out_mad)
657 		goto out;
658 
659 	init_query_mad(in_mad);
660 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
661 	in_mad->attr_mod = cpu_to_be32(port);
662 
663 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
664 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
665 
666 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
667 				in_mad, out_mad);
668 	if (err)
669 		goto out;
670 
671 
672 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
673 	props->lmc		= out_mad->data[34] & 0x7;
674 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
675 	props->sm_sl		= out_mad->data[36] & 0xf;
676 	props->state		= out_mad->data[32] & 0xf;
677 	props->phys_state	= out_mad->data[33] >> 4;
678 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
679 	if (netw_view)
680 		props->gid_tbl_len = out_mad->data[50];
681 	else
682 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
683 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
684 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
685 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
686 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
687 	props->active_width	= out_mad->data[31] & 0xf;
688 	props->active_speed	= out_mad->data[35] >> 4;
689 	props->max_mtu		= out_mad->data[41] & 0xf;
690 	props->active_mtu	= out_mad->data[36] >> 4;
691 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
692 	props->max_vl_num	= out_mad->data[37] >> 4;
693 	props->init_type_reply	= out_mad->data[41] >> 4;
694 
695 	/* Check if extended speeds (EDR/FDR/...) are supported */
696 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
697 		ext_active_speed = out_mad->data[62] >> 4;
698 
699 		switch (ext_active_speed) {
700 		case 1:
701 			props->active_speed = IB_SPEED_FDR;
702 			break;
703 		case 2:
704 			props->active_speed = IB_SPEED_EDR;
705 			break;
706 		}
707 	}
708 
709 	/* If reported active speed is QDR, check if is FDR-10 */
710 	if (props->active_speed == IB_SPEED_QDR) {
711 		init_query_mad(in_mad);
712 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
713 		in_mad->attr_mod = cpu_to_be32(port);
714 
715 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
716 				   NULL, NULL, in_mad, out_mad);
717 		if (err)
718 			goto out;
719 
720 		/* Checking LinkSpeedActive for FDR-10 */
721 		if (out_mad->data[15] & 0x1)
722 			props->active_speed = IB_SPEED_FDR10;
723 	}
724 
725 	/* Avoid wrong speed value returned by FW if the IB link is down. */
726 	if (props->state == IB_PORT_DOWN)
727 		 props->active_speed = IB_SPEED_SDR;
728 
729 out:
730 	kfree(in_mad);
731 	kfree(out_mad);
732 	return err;
733 }
734 
735 static u8 state_to_phys_state(enum ib_port_state state)
736 {
737 	return state == IB_PORT_ACTIVE ? 5 : 3;
738 }
739 
740 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
741 			       struct ib_port_attr *props)
742 {
743 
744 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
745 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
746 	struct net_device *ndev;
747 	enum ib_mtu tmp;
748 	struct mlx4_cmd_mailbox *mailbox;
749 	int err = 0;
750 	int is_bonded = mlx4_is_bonded(mdev->dev);
751 
752 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
753 	if (IS_ERR(mailbox))
754 		return PTR_ERR(mailbox);
755 
756 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
757 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
758 			   MLX4_CMD_WRAPPED);
759 	if (err)
760 		goto out;
761 
762 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
763 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
764 					   IB_WIDTH_4X : IB_WIDTH_1X;
765 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
766 					   IB_SPEED_FDR : IB_SPEED_QDR;
767 	props->port_cap_flags	= IB_PORT_CM_SUP;
768 	props->ip_gids = true;
769 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
770 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
771 	props->pkey_tbl_len	= 1;
772 	props->max_mtu		= IB_MTU_4096;
773 	props->max_vl_num	= 2;
774 	props->state		= IB_PORT_DOWN;
775 	props->phys_state	= state_to_phys_state(props->state);
776 	props->active_mtu	= IB_MTU_256;
777 	spin_lock_bh(&iboe->lock);
778 	ndev = iboe->netdevs[port - 1];
779 	if (ndev && is_bonded) {
780 		rcu_read_lock(); /* required to get upper dev */
781 		ndev = netdev_master_upper_dev_get_rcu(ndev);
782 		rcu_read_unlock();
783 	}
784 	if (!ndev)
785 		goto out_unlock;
786 
787 	tmp = iboe_get_mtu(ndev->mtu);
788 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
789 
790 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
791 					IB_PORT_ACTIVE : IB_PORT_DOWN;
792 	props->phys_state	= state_to_phys_state(props->state);
793 out_unlock:
794 	spin_unlock_bh(&iboe->lock);
795 out:
796 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
797 	return err;
798 }
799 
800 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
801 			 struct ib_port_attr *props, int netw_view)
802 {
803 	int err;
804 
805 	/* props being zeroed by the caller, avoid zeroing it here */
806 
807 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
808 		ib_link_query_port(ibdev, port, props, netw_view) :
809 				eth_link_query_port(ibdev, port, props);
810 
811 	return err;
812 }
813 
814 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
815 			      struct ib_port_attr *props)
816 {
817 	/* returns host view */
818 	return __mlx4_ib_query_port(ibdev, port, props, 0);
819 }
820 
821 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
822 			union ib_gid *gid, int netw_view)
823 {
824 	struct ib_smp *in_mad  = NULL;
825 	struct ib_smp *out_mad = NULL;
826 	int err = -ENOMEM;
827 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
828 	int clear = 0;
829 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
830 
831 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
832 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
833 	if (!in_mad || !out_mad)
834 		goto out;
835 
836 	init_query_mad(in_mad);
837 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
838 	in_mad->attr_mod = cpu_to_be32(port);
839 
840 	if (mlx4_is_mfunc(dev->dev) && netw_view)
841 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
842 
843 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
844 	if (err)
845 		goto out;
846 
847 	memcpy(gid->raw, out_mad->data + 8, 8);
848 
849 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
850 		if (index) {
851 			/* For any index > 0, return the null guid */
852 			err = 0;
853 			clear = 1;
854 			goto out;
855 		}
856 	}
857 
858 	init_query_mad(in_mad);
859 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
860 	in_mad->attr_mod = cpu_to_be32(index / 8);
861 
862 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
863 			   NULL, NULL, in_mad, out_mad);
864 	if (err)
865 		goto out;
866 
867 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
868 
869 out:
870 	if (clear)
871 		memset(gid->raw + 8, 0, 8);
872 	kfree(in_mad);
873 	kfree(out_mad);
874 	return err;
875 }
876 
877 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
878 			     union ib_gid *gid)
879 {
880 	if (rdma_protocol_ib(ibdev, port))
881 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
882 	return 0;
883 }
884 
885 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
886 {
887 	union sl2vl_tbl_to_u64 sl2vl64;
888 	struct ib_smp *in_mad  = NULL;
889 	struct ib_smp *out_mad = NULL;
890 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
891 	int err = -ENOMEM;
892 	int jj;
893 
894 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
895 		*sl2vl_tbl = 0;
896 		return 0;
897 	}
898 
899 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
900 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
901 	if (!in_mad || !out_mad)
902 		goto out;
903 
904 	init_query_mad(in_mad);
905 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
906 	in_mad->attr_mod = 0;
907 
908 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
909 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
910 
911 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
912 			   in_mad, out_mad);
913 	if (err)
914 		goto out;
915 
916 	for (jj = 0; jj < 8; jj++)
917 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
918 	*sl2vl_tbl = sl2vl64.sl64;
919 
920 out:
921 	kfree(in_mad);
922 	kfree(out_mad);
923 	return err;
924 }
925 
926 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
927 {
928 	u64 sl2vl;
929 	int i;
930 	int err;
931 
932 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
933 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
934 			continue;
935 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
936 		if (err) {
937 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
938 			       i, err);
939 			sl2vl = 0;
940 		}
941 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
942 	}
943 }
944 
945 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
946 			 u16 *pkey, int netw_view)
947 {
948 	struct ib_smp *in_mad  = NULL;
949 	struct ib_smp *out_mad = NULL;
950 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
951 	int err = -ENOMEM;
952 
953 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
954 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
955 	if (!in_mad || !out_mad)
956 		goto out;
957 
958 	init_query_mad(in_mad);
959 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
960 	in_mad->attr_mod = cpu_to_be32(index / 32);
961 
962 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
963 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
964 
965 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
966 			   in_mad, out_mad);
967 	if (err)
968 		goto out;
969 
970 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
971 
972 out:
973 	kfree(in_mad);
974 	kfree(out_mad);
975 	return err;
976 }
977 
978 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
979 {
980 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
981 }
982 
983 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
984 				 struct ib_device_modify *props)
985 {
986 	struct mlx4_cmd_mailbox *mailbox;
987 	unsigned long flags;
988 
989 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
990 		return -EOPNOTSUPP;
991 
992 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
993 		return 0;
994 
995 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
996 		return -EOPNOTSUPP;
997 
998 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
999 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1000 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1001 
1002 	/*
1003 	 * If possible, pass node desc to FW, so it can generate
1004 	 * a 144 trap.  If cmd fails, just ignore.
1005 	 */
1006 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1007 	if (IS_ERR(mailbox))
1008 		return 0;
1009 
1010 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1011 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1012 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1013 
1014 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1015 
1016 	return 0;
1017 }
1018 
1019 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1020 			    u32 cap_mask)
1021 {
1022 	struct mlx4_cmd_mailbox *mailbox;
1023 	int err;
1024 
1025 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1026 	if (IS_ERR(mailbox))
1027 		return PTR_ERR(mailbox);
1028 
1029 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1030 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1031 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1032 	} else {
1033 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1034 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1035 	}
1036 
1037 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1038 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1039 		       MLX4_CMD_WRAPPED);
1040 
1041 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1042 	return err;
1043 }
1044 
1045 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1046 			       struct ib_port_modify *props)
1047 {
1048 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1049 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1050 	struct ib_port_attr attr;
1051 	u32 cap_mask;
1052 	int err;
1053 
1054 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1055 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1056 	 * violations and port capabilities are not meaningful.
1057 	 */
1058 	if (is_eth)
1059 		return 0;
1060 
1061 	mutex_lock(&mdev->cap_mask_mutex);
1062 
1063 	err = ib_query_port(ibdev, port, &attr);
1064 	if (err)
1065 		goto out;
1066 
1067 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1068 		~props->clr_port_cap_mask;
1069 
1070 	err = mlx4_ib_SET_PORT(mdev, port,
1071 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1072 			       cap_mask);
1073 
1074 out:
1075 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1076 	return err;
1077 }
1078 
1079 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1080 				  struct ib_udata *udata)
1081 {
1082 	struct ib_device *ibdev = uctx->device;
1083 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1084 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1085 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1086 	struct mlx4_ib_alloc_ucontext_resp resp;
1087 	int err;
1088 
1089 	if (!dev->ib_active)
1090 		return -EAGAIN;
1091 
1092 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1093 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1094 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1095 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1096 	} else {
1097 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1098 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1099 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1100 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1101 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1102 	}
1103 
1104 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1105 	if (err)
1106 		return err;
1107 
1108 	INIT_LIST_HEAD(&context->db_page_list);
1109 	mutex_init(&context->db_page_mutex);
1110 
1111 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1112 	mutex_init(&context->wqn_ranges_mutex);
1113 
1114 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1115 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1116 	else
1117 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1118 
1119 	if (err) {
1120 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1121 		return -EFAULT;
1122 	}
1123 
1124 	return err;
1125 }
1126 
1127 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1128 {
1129 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1130 
1131 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1132 }
1133 
1134 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1135 {
1136 }
1137 
1138 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1139 {
1140 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1141 
1142 	switch (vma->vm_pgoff) {
1143 	case 0:
1144 		return rdma_user_mmap_io(context, vma,
1145 					 to_mucontext(context)->uar.pfn,
1146 					 PAGE_SIZE,
1147 					 pgprot_noncached(vma->vm_page_prot));
1148 
1149 	case 1:
1150 		if (dev->dev->caps.bf_reg_size == 0)
1151 			return -EINVAL;
1152 		return rdma_user_mmap_io(
1153 			context, vma,
1154 			to_mucontext(context)->uar.pfn +
1155 				dev->dev->caps.num_uars,
1156 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot));
1157 
1158 	case 3: {
1159 		struct mlx4_clock_params params;
1160 		int ret;
1161 
1162 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1163 		if (ret)
1164 			return ret;
1165 
1166 		return rdma_user_mmap_io(
1167 			context, vma,
1168 			(pci_resource_start(dev->dev->persist->pdev,
1169 					    params.bar) +
1170 			 params.offset) >>
1171 				PAGE_SHIFT,
1172 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot));
1173 	}
1174 
1175 	default:
1176 		return -EINVAL;
1177 	}
1178 }
1179 
1180 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_ucontext *context,
1181 			    struct ib_udata *udata)
1182 {
1183 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1184 	struct ib_device *ibdev = ibpd->device;
1185 	int err;
1186 
1187 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1188 	if (err)
1189 		return err;
1190 
1191 	if (context && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1192 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1193 		return -EFAULT;
1194 	}
1195 	return 0;
1196 }
1197 
1198 static void mlx4_ib_dealloc_pd(struct ib_pd *pd)
1199 {
1200 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1201 }
1202 
1203 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1204 					  struct ib_ucontext *context,
1205 					  struct ib_udata *udata)
1206 {
1207 	struct mlx4_ib_xrcd *xrcd;
1208 	struct ib_cq_init_attr cq_attr = {};
1209 	int err;
1210 
1211 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1212 		return ERR_PTR(-ENOSYS);
1213 
1214 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1215 	if (!xrcd)
1216 		return ERR_PTR(-ENOMEM);
1217 
1218 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1219 	if (err)
1220 		goto err1;
1221 
1222 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1223 	if (IS_ERR(xrcd->pd)) {
1224 		err = PTR_ERR(xrcd->pd);
1225 		goto err2;
1226 	}
1227 
1228 	cq_attr.cqe = 1;
1229 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1230 	if (IS_ERR(xrcd->cq)) {
1231 		err = PTR_ERR(xrcd->cq);
1232 		goto err3;
1233 	}
1234 
1235 	return &xrcd->ibxrcd;
1236 
1237 err3:
1238 	ib_dealloc_pd(xrcd->pd);
1239 err2:
1240 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1241 err1:
1242 	kfree(xrcd);
1243 	return ERR_PTR(err);
1244 }
1245 
1246 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1247 {
1248 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1249 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1250 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1251 	kfree(xrcd);
1252 
1253 	return 0;
1254 }
1255 
1256 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1257 {
1258 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1259 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1260 	struct mlx4_ib_gid_entry *ge;
1261 
1262 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1263 	if (!ge)
1264 		return -ENOMEM;
1265 
1266 	ge->gid = *gid;
1267 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1268 		ge->port = mqp->port;
1269 		ge->added = 1;
1270 	}
1271 
1272 	mutex_lock(&mqp->mutex);
1273 	list_add_tail(&ge->list, &mqp->gid_list);
1274 	mutex_unlock(&mqp->mutex);
1275 
1276 	return 0;
1277 }
1278 
1279 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1280 					  struct mlx4_ib_counters *ctr_table)
1281 {
1282 	struct counter_index *counter, *tmp_count;
1283 
1284 	mutex_lock(&ctr_table->mutex);
1285 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1286 				 list) {
1287 		if (counter->allocated)
1288 			mlx4_counter_free(ibdev->dev, counter->index);
1289 		list_del(&counter->list);
1290 		kfree(counter);
1291 	}
1292 	mutex_unlock(&ctr_table->mutex);
1293 }
1294 
1295 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1296 		   union ib_gid *gid)
1297 {
1298 	struct net_device *ndev;
1299 	int ret = 0;
1300 
1301 	if (!mqp->port)
1302 		return 0;
1303 
1304 	spin_lock_bh(&mdev->iboe.lock);
1305 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1306 	if (ndev)
1307 		dev_hold(ndev);
1308 	spin_unlock_bh(&mdev->iboe.lock);
1309 
1310 	if (ndev) {
1311 		ret = 1;
1312 		dev_put(ndev);
1313 	}
1314 
1315 	return ret;
1316 }
1317 
1318 struct mlx4_ib_steering {
1319 	struct list_head list;
1320 	struct mlx4_flow_reg_id reg_id;
1321 	union ib_gid gid;
1322 };
1323 
1324 #define LAST_ETH_FIELD vlan_tag
1325 #define LAST_IB_FIELD sl
1326 #define LAST_IPV4_FIELD dst_ip
1327 #define LAST_TCP_UDP_FIELD src_port
1328 
1329 /* Field is the last supported field */
1330 #define FIELDS_NOT_SUPPORTED(filter, field)\
1331 	memchr_inv((void *)&filter.field  +\
1332 		   sizeof(filter.field), 0,\
1333 		   sizeof(filter) -\
1334 		   offsetof(typeof(filter), field) -\
1335 		   sizeof(filter.field))
1336 
1337 static int parse_flow_attr(struct mlx4_dev *dev,
1338 			   u32 qp_num,
1339 			   union ib_flow_spec *ib_spec,
1340 			   struct _rule_hw *mlx4_spec)
1341 {
1342 	enum mlx4_net_trans_rule_id type;
1343 
1344 	switch (ib_spec->type) {
1345 	case IB_FLOW_SPEC_ETH:
1346 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1347 			return -ENOTSUPP;
1348 
1349 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1350 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1351 		       ETH_ALEN);
1352 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1353 		       ETH_ALEN);
1354 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1355 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1356 		break;
1357 	case IB_FLOW_SPEC_IB:
1358 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1359 			return -ENOTSUPP;
1360 
1361 		type = MLX4_NET_TRANS_RULE_ID_IB;
1362 		mlx4_spec->ib.l3_qpn =
1363 			cpu_to_be32(qp_num);
1364 		mlx4_spec->ib.qpn_mask =
1365 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1366 		break;
1367 
1368 
1369 	case IB_FLOW_SPEC_IPV4:
1370 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1371 			return -ENOTSUPP;
1372 
1373 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1374 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1375 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1376 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1377 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1378 		break;
1379 
1380 	case IB_FLOW_SPEC_TCP:
1381 	case IB_FLOW_SPEC_UDP:
1382 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1383 			return -ENOTSUPP;
1384 
1385 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1386 					MLX4_NET_TRANS_RULE_ID_TCP :
1387 					MLX4_NET_TRANS_RULE_ID_UDP;
1388 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1389 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1390 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1391 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1392 		break;
1393 
1394 	default:
1395 		return -EINVAL;
1396 	}
1397 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1398 	    mlx4_hw_rule_sz(dev, type) < 0)
1399 		return -EINVAL;
1400 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1401 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1402 	return mlx4_hw_rule_sz(dev, type);
1403 }
1404 
1405 struct default_rules {
1406 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1407 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1408 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1409 	__u8  link_layer;
1410 };
1411 static const struct default_rules default_table[] = {
1412 	{
1413 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1414 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1415 		.rules_create_list = {IB_FLOW_SPEC_IB},
1416 		.link_layer = IB_LINK_LAYER_INFINIBAND
1417 	}
1418 };
1419 
1420 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1421 					 struct ib_flow_attr *flow_attr)
1422 {
1423 	int i, j, k;
1424 	void *ib_flow;
1425 	const struct default_rules *pdefault_rules = default_table;
1426 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1427 
1428 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1429 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1430 		memset(&field_types, 0, sizeof(field_types));
1431 
1432 		if (link_layer != pdefault_rules->link_layer)
1433 			continue;
1434 
1435 		ib_flow = flow_attr + 1;
1436 		/* we assume the specs are sorted */
1437 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1438 		     j < flow_attr->num_of_specs; k++) {
1439 			union ib_flow_spec *current_flow =
1440 				(union ib_flow_spec *)ib_flow;
1441 
1442 			/* same layer but different type */
1443 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1444 			     (pdefault_rules->mandatory_fields[k] &
1445 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1446 			    (current_flow->type !=
1447 			     pdefault_rules->mandatory_fields[k]))
1448 				goto out;
1449 
1450 			/* same layer, try match next one */
1451 			if (current_flow->type ==
1452 			    pdefault_rules->mandatory_fields[k]) {
1453 				j++;
1454 				ib_flow +=
1455 					((union ib_flow_spec *)ib_flow)->size;
1456 			}
1457 		}
1458 
1459 		ib_flow = flow_attr + 1;
1460 		for (j = 0; j < flow_attr->num_of_specs;
1461 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1462 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1463 				/* same layer and same type */
1464 				if (((union ib_flow_spec *)ib_flow)->type ==
1465 				    pdefault_rules->mandatory_not_fields[k])
1466 					goto out;
1467 
1468 		return i;
1469 	}
1470 out:
1471 	return -1;
1472 }
1473 
1474 static int __mlx4_ib_create_default_rules(
1475 		struct mlx4_ib_dev *mdev,
1476 		struct ib_qp *qp,
1477 		const struct default_rules *pdefault_rules,
1478 		struct _rule_hw *mlx4_spec) {
1479 	int size = 0;
1480 	int i;
1481 
1482 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1483 		int ret;
1484 		union ib_flow_spec ib_spec;
1485 		switch (pdefault_rules->rules_create_list[i]) {
1486 		case 0:
1487 			/* no rule */
1488 			continue;
1489 		case IB_FLOW_SPEC_IB:
1490 			ib_spec.type = IB_FLOW_SPEC_IB;
1491 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1492 
1493 			break;
1494 		default:
1495 			/* invalid rule */
1496 			return -EINVAL;
1497 		}
1498 		/* We must put empty rule, qpn is being ignored */
1499 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1500 				      mlx4_spec);
1501 		if (ret < 0) {
1502 			pr_info("invalid parsing\n");
1503 			return -EINVAL;
1504 		}
1505 
1506 		mlx4_spec = (void *)mlx4_spec + ret;
1507 		size += ret;
1508 	}
1509 	return size;
1510 }
1511 
1512 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1513 			  int domain,
1514 			  enum mlx4_net_trans_promisc_mode flow_type,
1515 			  u64 *reg_id)
1516 {
1517 	int ret, i;
1518 	int size = 0;
1519 	void *ib_flow;
1520 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1521 	struct mlx4_cmd_mailbox *mailbox;
1522 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1523 	int default_flow;
1524 
1525 	static const u16 __mlx4_domain[] = {
1526 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1527 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1528 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1529 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1530 	};
1531 
1532 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1533 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1534 		return -EINVAL;
1535 	}
1536 
1537 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1538 		pr_err("Invalid domain value %d\n", domain);
1539 		return -EINVAL;
1540 	}
1541 
1542 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1543 		return -EINVAL;
1544 
1545 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1546 	if (IS_ERR(mailbox))
1547 		return PTR_ERR(mailbox);
1548 	ctrl = mailbox->buf;
1549 
1550 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1551 				 flow_attr->priority);
1552 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1553 	ctrl->port = flow_attr->port;
1554 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1555 
1556 	ib_flow = flow_attr + 1;
1557 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1558 	/* Add default flows */
1559 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1560 	if (default_flow >= 0) {
1561 		ret = __mlx4_ib_create_default_rules(
1562 				mdev, qp, default_table + default_flow,
1563 				mailbox->buf + size);
1564 		if (ret < 0) {
1565 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1566 			return -EINVAL;
1567 		}
1568 		size += ret;
1569 	}
1570 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1571 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1572 				      mailbox->buf + size);
1573 		if (ret < 0) {
1574 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1575 			return -EINVAL;
1576 		}
1577 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1578 		size += ret;
1579 	}
1580 
1581 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1582 	    flow_attr->num_of_specs == 1) {
1583 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1584 		enum ib_flow_spec_type header_spec =
1585 			((union ib_flow_spec *)(flow_attr + 1))->type;
1586 
1587 		if (header_spec == IB_FLOW_SPEC_ETH)
1588 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1589 	}
1590 
1591 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1592 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1593 			   MLX4_CMD_NATIVE);
1594 	if (ret == -ENOMEM)
1595 		pr_err("mcg table is full. Fail to register network rule.\n");
1596 	else if (ret == -ENXIO)
1597 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1598 	else if (ret)
1599 		pr_err("Invalid argument. Fail to register network rule.\n");
1600 
1601 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1602 	return ret;
1603 }
1604 
1605 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1606 {
1607 	int err;
1608 	err = mlx4_cmd(dev, reg_id, 0, 0,
1609 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1610 		       MLX4_CMD_NATIVE);
1611 	if (err)
1612 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1613 		       reg_id);
1614 	return err;
1615 }
1616 
1617 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1618 				    u64 *reg_id)
1619 {
1620 	void *ib_flow;
1621 	union ib_flow_spec *ib_spec;
1622 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1623 	int err = 0;
1624 
1625 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1626 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1627 		return 0; /* do nothing */
1628 
1629 	ib_flow = flow_attr + 1;
1630 	ib_spec = (union ib_flow_spec *)ib_flow;
1631 
1632 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1633 		return 0; /* do nothing */
1634 
1635 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1636 				    flow_attr->port, qp->qp_num,
1637 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1638 				    reg_id);
1639 	return err;
1640 }
1641 
1642 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1643 				      struct ib_flow_attr *flow_attr,
1644 				      enum mlx4_net_trans_promisc_mode *type)
1645 {
1646 	int err = 0;
1647 
1648 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1649 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1650 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1651 		return -EOPNOTSUPP;
1652 	}
1653 
1654 	if (flow_attr->num_of_specs == 0) {
1655 		type[0] = MLX4_FS_MC_SNIFFER;
1656 		type[1] = MLX4_FS_UC_SNIFFER;
1657 	} else {
1658 		union ib_flow_spec *ib_spec;
1659 
1660 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1661 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1662 			return -EINVAL;
1663 
1664 		/* if all is zero than MC and UC */
1665 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1666 			type[0] = MLX4_FS_MC_SNIFFER;
1667 			type[1] = MLX4_FS_UC_SNIFFER;
1668 		} else {
1669 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1670 					    ib_spec->eth.mask.dst_mac[1],
1671 					    ib_spec->eth.mask.dst_mac[2],
1672 					    ib_spec->eth.mask.dst_mac[3],
1673 					    ib_spec->eth.mask.dst_mac[4],
1674 					    ib_spec->eth.mask.dst_mac[5]};
1675 
1676 			/* Above xor was only on MC bit, non empty mask is valid
1677 			 * only if this bit is set and rest are zero.
1678 			 */
1679 			if (!is_zero_ether_addr(&mac[0]))
1680 				return -EINVAL;
1681 
1682 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1683 				type[0] = MLX4_FS_MC_SNIFFER;
1684 			else
1685 				type[0] = MLX4_FS_UC_SNIFFER;
1686 		}
1687 	}
1688 
1689 	return err;
1690 }
1691 
1692 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1693 				    struct ib_flow_attr *flow_attr,
1694 				    int domain, struct ib_udata *udata)
1695 {
1696 	int err = 0, i = 0, j = 0;
1697 	struct mlx4_ib_flow *mflow;
1698 	enum mlx4_net_trans_promisc_mode type[2];
1699 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1700 	int is_bonded = mlx4_is_bonded(dev);
1701 
1702 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1703 		return ERR_PTR(-EINVAL);
1704 
1705 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1706 		return ERR_PTR(-EOPNOTSUPP);
1707 
1708 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1709 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1710 		return ERR_PTR(-EOPNOTSUPP);
1711 
1712 	if (udata &&
1713 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1714 		return ERR_PTR(-EOPNOTSUPP);
1715 
1716 	memset(type, 0, sizeof(type));
1717 
1718 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1719 	if (!mflow) {
1720 		err = -ENOMEM;
1721 		goto err_free;
1722 	}
1723 
1724 	switch (flow_attr->type) {
1725 	case IB_FLOW_ATTR_NORMAL:
1726 		/* If dont trap flag (continue match) is set, under specific
1727 		 * condition traffic be replicated to given qp,
1728 		 * without stealing it
1729 		 */
1730 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1731 			err = mlx4_ib_add_dont_trap_rule(dev,
1732 							 flow_attr,
1733 							 type);
1734 			if (err)
1735 				goto err_free;
1736 		} else {
1737 			type[0] = MLX4_FS_REGULAR;
1738 		}
1739 		break;
1740 
1741 	case IB_FLOW_ATTR_ALL_DEFAULT:
1742 		type[0] = MLX4_FS_ALL_DEFAULT;
1743 		break;
1744 
1745 	case IB_FLOW_ATTR_MC_DEFAULT:
1746 		type[0] = MLX4_FS_MC_DEFAULT;
1747 		break;
1748 
1749 	case IB_FLOW_ATTR_SNIFFER:
1750 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1751 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1752 		break;
1753 
1754 	default:
1755 		err = -EINVAL;
1756 		goto err_free;
1757 	}
1758 
1759 	while (i < ARRAY_SIZE(type) && type[i]) {
1760 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1761 					    &mflow->reg_id[i].id);
1762 		if (err)
1763 			goto err_create_flow;
1764 		if (is_bonded) {
1765 			/* Application always sees one port so the mirror rule
1766 			 * must be on port #2
1767 			 */
1768 			flow_attr->port = 2;
1769 			err = __mlx4_ib_create_flow(qp, flow_attr,
1770 						    domain, type[j],
1771 						    &mflow->reg_id[j].mirror);
1772 			flow_attr->port = 1;
1773 			if (err)
1774 				goto err_create_flow;
1775 			j++;
1776 		}
1777 
1778 		i++;
1779 	}
1780 
1781 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1782 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1783 					       &mflow->reg_id[i].id);
1784 		if (err)
1785 			goto err_create_flow;
1786 
1787 		if (is_bonded) {
1788 			flow_attr->port = 2;
1789 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1790 						       &mflow->reg_id[j].mirror);
1791 			flow_attr->port = 1;
1792 			if (err)
1793 				goto err_create_flow;
1794 			j++;
1795 		}
1796 		/* function to create mirror rule */
1797 		i++;
1798 	}
1799 
1800 	return &mflow->ibflow;
1801 
1802 err_create_flow:
1803 	while (i) {
1804 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1805 					     mflow->reg_id[i].id);
1806 		i--;
1807 	}
1808 
1809 	while (j) {
1810 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1811 					     mflow->reg_id[j].mirror);
1812 		j--;
1813 	}
1814 err_free:
1815 	kfree(mflow);
1816 	return ERR_PTR(err);
1817 }
1818 
1819 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1820 {
1821 	int err, ret = 0;
1822 	int i = 0;
1823 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1824 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1825 
1826 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1827 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1828 		if (err)
1829 			ret = err;
1830 		if (mflow->reg_id[i].mirror) {
1831 			err = __mlx4_ib_destroy_flow(mdev->dev,
1832 						     mflow->reg_id[i].mirror);
1833 			if (err)
1834 				ret = err;
1835 		}
1836 		i++;
1837 	}
1838 
1839 	kfree(mflow);
1840 	return ret;
1841 }
1842 
1843 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1844 {
1845 	int err;
1846 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1847 	struct mlx4_dev	*dev = mdev->dev;
1848 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1849 	struct mlx4_ib_steering *ib_steering = NULL;
1850 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1851 	struct mlx4_flow_reg_id	reg_id;
1852 
1853 	if (mdev->dev->caps.steering_mode ==
1854 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1855 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1856 		if (!ib_steering)
1857 			return -ENOMEM;
1858 	}
1859 
1860 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1861 				    !!(mqp->flags &
1862 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1863 				    prot, &reg_id.id);
1864 	if (err) {
1865 		pr_err("multicast attach op failed, err %d\n", err);
1866 		goto err_malloc;
1867 	}
1868 
1869 	reg_id.mirror = 0;
1870 	if (mlx4_is_bonded(dev)) {
1871 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1872 					    (mqp->port == 1) ? 2 : 1,
1873 					    !!(mqp->flags &
1874 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1875 					    prot, &reg_id.mirror);
1876 		if (err)
1877 			goto err_add;
1878 	}
1879 
1880 	err = add_gid_entry(ibqp, gid);
1881 	if (err)
1882 		goto err_add;
1883 
1884 	if (ib_steering) {
1885 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1886 		ib_steering->reg_id = reg_id;
1887 		mutex_lock(&mqp->mutex);
1888 		list_add(&ib_steering->list, &mqp->steering_rules);
1889 		mutex_unlock(&mqp->mutex);
1890 	}
1891 	return 0;
1892 
1893 err_add:
1894 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1895 			      prot, reg_id.id);
1896 	if (reg_id.mirror)
1897 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1898 				      prot, reg_id.mirror);
1899 err_malloc:
1900 	kfree(ib_steering);
1901 
1902 	return err;
1903 }
1904 
1905 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1906 {
1907 	struct mlx4_ib_gid_entry *ge;
1908 	struct mlx4_ib_gid_entry *tmp;
1909 	struct mlx4_ib_gid_entry *ret = NULL;
1910 
1911 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1912 		if (!memcmp(raw, ge->gid.raw, 16)) {
1913 			ret = ge;
1914 			break;
1915 		}
1916 	}
1917 
1918 	return ret;
1919 }
1920 
1921 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1922 {
1923 	int err;
1924 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1925 	struct mlx4_dev *dev = mdev->dev;
1926 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1927 	struct net_device *ndev;
1928 	struct mlx4_ib_gid_entry *ge;
1929 	struct mlx4_flow_reg_id reg_id = {0, 0};
1930 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1931 
1932 	if (mdev->dev->caps.steering_mode ==
1933 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1934 		struct mlx4_ib_steering *ib_steering;
1935 
1936 		mutex_lock(&mqp->mutex);
1937 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1938 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1939 				list_del(&ib_steering->list);
1940 				break;
1941 			}
1942 		}
1943 		mutex_unlock(&mqp->mutex);
1944 		if (&ib_steering->list == &mqp->steering_rules) {
1945 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1946 			return -EINVAL;
1947 		}
1948 		reg_id = ib_steering->reg_id;
1949 		kfree(ib_steering);
1950 	}
1951 
1952 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1953 				    prot, reg_id.id);
1954 	if (err)
1955 		return err;
1956 
1957 	if (mlx4_is_bonded(dev)) {
1958 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1959 					    prot, reg_id.mirror);
1960 		if (err)
1961 			return err;
1962 	}
1963 
1964 	mutex_lock(&mqp->mutex);
1965 	ge = find_gid_entry(mqp, gid->raw);
1966 	if (ge) {
1967 		spin_lock_bh(&mdev->iboe.lock);
1968 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1969 		if (ndev)
1970 			dev_hold(ndev);
1971 		spin_unlock_bh(&mdev->iboe.lock);
1972 		if (ndev)
1973 			dev_put(ndev);
1974 		list_del(&ge->list);
1975 		kfree(ge);
1976 	} else
1977 		pr_warn("could not find mgid entry\n");
1978 
1979 	mutex_unlock(&mqp->mutex);
1980 
1981 	return 0;
1982 }
1983 
1984 static int init_node_data(struct mlx4_ib_dev *dev)
1985 {
1986 	struct ib_smp *in_mad  = NULL;
1987 	struct ib_smp *out_mad = NULL;
1988 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1989 	int err = -ENOMEM;
1990 
1991 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1992 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1993 	if (!in_mad || !out_mad)
1994 		goto out;
1995 
1996 	init_query_mad(in_mad);
1997 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1998 	if (mlx4_is_master(dev->dev))
1999 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2000 
2001 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2002 	if (err)
2003 		goto out;
2004 
2005 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2006 
2007 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2008 
2009 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2010 	if (err)
2011 		goto out;
2012 
2013 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2014 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2015 
2016 out:
2017 	kfree(in_mad);
2018 	kfree(out_mad);
2019 	return err;
2020 }
2021 
2022 static ssize_t hca_type_show(struct device *device,
2023 			     struct device_attribute *attr, char *buf)
2024 {
2025 	struct mlx4_ib_dev *dev =
2026 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2027 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2028 }
2029 static DEVICE_ATTR_RO(hca_type);
2030 
2031 static ssize_t hw_rev_show(struct device *device,
2032 			   struct device_attribute *attr, char *buf)
2033 {
2034 	struct mlx4_ib_dev *dev =
2035 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2036 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2037 }
2038 static DEVICE_ATTR_RO(hw_rev);
2039 
2040 static ssize_t board_id_show(struct device *device,
2041 			     struct device_attribute *attr, char *buf)
2042 {
2043 	struct mlx4_ib_dev *dev =
2044 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2045 
2046 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2047 		       dev->dev->board_id);
2048 }
2049 static DEVICE_ATTR_RO(board_id);
2050 
2051 static struct attribute *mlx4_class_attributes[] = {
2052 	&dev_attr_hw_rev.attr,
2053 	&dev_attr_hca_type.attr,
2054 	&dev_attr_board_id.attr,
2055 	NULL
2056 };
2057 
2058 static const struct attribute_group mlx4_attr_group = {
2059 	.attrs = mlx4_class_attributes,
2060 };
2061 
2062 struct diag_counter {
2063 	const char *name;
2064 	u32 offset;
2065 };
2066 
2067 #define DIAG_COUNTER(_name, _offset)			\
2068 	{ .name = #_name, .offset = _offset }
2069 
2070 static const struct diag_counter diag_basic[] = {
2071 	DIAG_COUNTER(rq_num_lle, 0x00),
2072 	DIAG_COUNTER(sq_num_lle, 0x04),
2073 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2074 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2075 	DIAG_COUNTER(rq_num_lpe, 0x18),
2076 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2077 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2078 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2079 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2080 	DIAG_COUNTER(sq_num_bre, 0x34),
2081 	DIAG_COUNTER(sq_num_rire, 0x44),
2082 	DIAG_COUNTER(rq_num_rire, 0x48),
2083 	DIAG_COUNTER(sq_num_rae, 0x4C),
2084 	DIAG_COUNTER(rq_num_rae, 0x50),
2085 	DIAG_COUNTER(sq_num_roe, 0x54),
2086 	DIAG_COUNTER(sq_num_tree, 0x5C),
2087 	DIAG_COUNTER(sq_num_rree, 0x64),
2088 	DIAG_COUNTER(rq_num_rnr, 0x68),
2089 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2090 	DIAG_COUNTER(rq_num_oos, 0x100),
2091 	DIAG_COUNTER(sq_num_oos, 0x104),
2092 };
2093 
2094 static const struct diag_counter diag_ext[] = {
2095 	DIAG_COUNTER(rq_num_dup, 0x130),
2096 	DIAG_COUNTER(sq_num_to, 0x134),
2097 };
2098 
2099 static const struct diag_counter diag_device_only[] = {
2100 	DIAG_COUNTER(num_cqovf, 0x1A0),
2101 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2102 };
2103 
2104 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2105 						    u8 port_num)
2106 {
2107 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2108 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2109 
2110 	if (!diag[!!port_num].name)
2111 		return NULL;
2112 
2113 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2114 					  diag[!!port_num].num_counters,
2115 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2116 }
2117 
2118 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2119 				struct rdma_hw_stats *stats,
2120 				u8 port, int index)
2121 {
2122 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2123 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2124 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2125 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2126 	int ret;
2127 	int i;
2128 
2129 	ret = mlx4_query_diag_counters(dev->dev,
2130 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2131 				       diag[!!port].offset, hw_value,
2132 				       diag[!!port].num_counters, port);
2133 
2134 	if (ret)
2135 		return ret;
2136 
2137 	for (i = 0; i < diag[!!port].num_counters; i++)
2138 		stats->value[i] = hw_value[i];
2139 
2140 	return diag[!!port].num_counters;
2141 }
2142 
2143 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2144 					 const char ***name,
2145 					 u32 **offset,
2146 					 u32 *num,
2147 					 bool port)
2148 {
2149 	u32 num_counters;
2150 
2151 	num_counters = ARRAY_SIZE(diag_basic);
2152 
2153 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2154 		num_counters += ARRAY_SIZE(diag_ext);
2155 
2156 	if (!port)
2157 		num_counters += ARRAY_SIZE(diag_device_only);
2158 
2159 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2160 	if (!*name)
2161 		return -ENOMEM;
2162 
2163 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2164 	if (!*offset)
2165 		goto err_name;
2166 
2167 	*num = num_counters;
2168 
2169 	return 0;
2170 
2171 err_name:
2172 	kfree(*name);
2173 	return -ENOMEM;
2174 }
2175 
2176 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2177 				       const char **name,
2178 				       u32 *offset,
2179 				       bool port)
2180 {
2181 	int i;
2182 	int j;
2183 
2184 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2185 		name[i] = diag_basic[i].name;
2186 		offset[i] = diag_basic[i].offset;
2187 	}
2188 
2189 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2190 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2191 			name[j] = diag_ext[i].name;
2192 			offset[j] = diag_ext[i].offset;
2193 		}
2194 	}
2195 
2196 	if (!port) {
2197 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2198 			name[j] = diag_device_only[i].name;
2199 			offset[j] = diag_device_only[i].offset;
2200 		}
2201 	}
2202 }
2203 
2204 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2205 	.alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2206 	.get_hw_stats = mlx4_ib_get_hw_stats,
2207 };
2208 
2209 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2210 {
2211 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2212 	int i;
2213 	int ret;
2214 	bool per_port = !!(ibdev->dev->caps.flags2 &
2215 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2216 
2217 	if (mlx4_is_slave(ibdev->dev))
2218 		return 0;
2219 
2220 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2221 		/* i == 1 means we are building port counters */
2222 		if (i && !per_port)
2223 			continue;
2224 
2225 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2226 						    &diag[i].offset,
2227 						    &diag[i].num_counters, i);
2228 		if (ret)
2229 			goto err_alloc;
2230 
2231 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2232 					   diag[i].offset, i);
2233 	}
2234 
2235 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2236 
2237 	return 0;
2238 
2239 err_alloc:
2240 	if (i) {
2241 		kfree(diag[i - 1].name);
2242 		kfree(diag[i - 1].offset);
2243 	}
2244 
2245 	return ret;
2246 }
2247 
2248 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2249 {
2250 	int i;
2251 
2252 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2253 		kfree(ibdev->diag_counters[i].offset);
2254 		kfree(ibdev->diag_counters[i].name);
2255 	}
2256 }
2257 
2258 #define MLX4_IB_INVALID_MAC	((u64)-1)
2259 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2260 			       struct net_device *dev,
2261 			       int port)
2262 {
2263 	u64 new_smac = 0;
2264 	u64 release_mac = MLX4_IB_INVALID_MAC;
2265 	struct mlx4_ib_qp *qp;
2266 
2267 	read_lock(&dev_base_lock);
2268 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2269 	read_unlock(&dev_base_lock);
2270 
2271 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2272 
2273 	/* no need for update QP1 and mac registration in non-SRIOV */
2274 	if (!mlx4_is_mfunc(ibdev->dev))
2275 		return;
2276 
2277 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2278 	qp = ibdev->qp1_proxy[port - 1];
2279 	if (qp) {
2280 		int new_smac_index;
2281 		u64 old_smac;
2282 		struct mlx4_update_qp_params update_params;
2283 
2284 		mutex_lock(&qp->mutex);
2285 		old_smac = qp->pri.smac;
2286 		if (new_smac == old_smac)
2287 			goto unlock;
2288 
2289 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2290 
2291 		if (new_smac_index < 0)
2292 			goto unlock;
2293 
2294 		update_params.smac_index = new_smac_index;
2295 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2296 				   &update_params)) {
2297 			release_mac = new_smac;
2298 			goto unlock;
2299 		}
2300 		/* if old port was zero, no mac was yet registered for this QP */
2301 		if (qp->pri.smac_port)
2302 			release_mac = old_smac;
2303 		qp->pri.smac = new_smac;
2304 		qp->pri.smac_port = port;
2305 		qp->pri.smac_index = new_smac_index;
2306 	}
2307 
2308 unlock:
2309 	if (release_mac != MLX4_IB_INVALID_MAC)
2310 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2311 	if (qp)
2312 		mutex_unlock(&qp->mutex);
2313 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2314 }
2315 
2316 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2317 				 struct net_device *dev,
2318 				 unsigned long event)
2319 
2320 {
2321 	struct mlx4_ib_iboe *iboe;
2322 	int update_qps_port = -1;
2323 	int port;
2324 
2325 	ASSERT_RTNL();
2326 
2327 	iboe = &ibdev->iboe;
2328 
2329 	spin_lock_bh(&iboe->lock);
2330 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2331 
2332 		iboe->netdevs[port - 1] =
2333 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2334 
2335 		if (dev == iboe->netdevs[port - 1] &&
2336 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2337 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2338 			update_qps_port = port;
2339 
2340 		if (dev == iboe->netdevs[port - 1] &&
2341 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2342 			enum ib_port_state port_state;
2343 			struct ib_event ibev = { };
2344 
2345 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2346 						     &port_state))
2347 				continue;
2348 
2349 			if (event == NETDEV_UP &&
2350 			    (port_state != IB_PORT_ACTIVE ||
2351 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2352 				continue;
2353 			if (event == NETDEV_DOWN &&
2354 			    (port_state != IB_PORT_DOWN ||
2355 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2356 				continue;
2357 			iboe->last_port_state[port - 1] = port_state;
2358 
2359 			ibev.device = &ibdev->ib_dev;
2360 			ibev.element.port_num = port;
2361 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2362 							  IB_EVENT_PORT_ERR;
2363 			ib_dispatch_event(&ibev);
2364 		}
2365 
2366 	}
2367 	spin_unlock_bh(&iboe->lock);
2368 
2369 	if (update_qps_port > 0)
2370 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2371 }
2372 
2373 static int mlx4_ib_netdev_event(struct notifier_block *this,
2374 				unsigned long event, void *ptr)
2375 {
2376 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2377 	struct mlx4_ib_dev *ibdev;
2378 
2379 	if (!net_eq(dev_net(dev), &init_net))
2380 		return NOTIFY_DONE;
2381 
2382 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2383 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2384 
2385 	return NOTIFY_DONE;
2386 }
2387 
2388 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2389 {
2390 	int port;
2391 	int slave;
2392 	int i;
2393 
2394 	if (mlx4_is_master(ibdev->dev)) {
2395 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2396 		     ++slave) {
2397 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2398 				for (i = 0;
2399 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2400 				     ++i) {
2401 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2402 					/* master has the identity virt2phys pkey mapping */
2403 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2404 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2405 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2406 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2407 				}
2408 			}
2409 		}
2410 		/* initialize pkey cache */
2411 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2412 			for (i = 0;
2413 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2414 			     ++i)
2415 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2416 					(i) ? 0 : 0xFFFF;
2417 		}
2418 	}
2419 }
2420 
2421 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2422 {
2423 	int i, j, eq = 0, total_eqs = 0;
2424 
2425 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2426 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2427 	if (!ibdev->eq_table)
2428 		return;
2429 
2430 	for (i = 1; i <= dev->caps.num_ports; i++) {
2431 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2432 		     j++, total_eqs++) {
2433 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2434 				continue;
2435 			ibdev->eq_table[eq] = total_eqs;
2436 			if (!mlx4_assign_eq(dev, i,
2437 					    &ibdev->eq_table[eq]))
2438 				eq++;
2439 			else
2440 				ibdev->eq_table[eq] = -1;
2441 		}
2442 	}
2443 
2444 	for (i = eq; i < dev->caps.num_comp_vectors;
2445 	     ibdev->eq_table[i++] = -1)
2446 		;
2447 
2448 	/* Advertise the new number of EQs to clients */
2449 	ibdev->ib_dev.num_comp_vectors = eq;
2450 }
2451 
2452 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2453 {
2454 	int i;
2455 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2456 
2457 	/* no eqs were allocated */
2458 	if (!ibdev->eq_table)
2459 		return;
2460 
2461 	/* Reset the advertised EQ number */
2462 	ibdev->ib_dev.num_comp_vectors = 0;
2463 
2464 	for (i = 0; i < total_eqs; i++)
2465 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2466 
2467 	kfree(ibdev->eq_table);
2468 	ibdev->eq_table = NULL;
2469 }
2470 
2471 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2472 			       struct ib_port_immutable *immutable)
2473 {
2474 	struct ib_port_attr attr;
2475 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2476 	int err;
2477 
2478 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2479 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2480 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2481 	} else {
2482 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2483 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2484 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2485 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2486 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2487 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2488 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2489 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2490 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2491 	}
2492 
2493 	err = ib_query_port(ibdev, port_num, &attr);
2494 	if (err)
2495 		return err;
2496 
2497 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2498 	immutable->gid_tbl_len = attr.gid_tbl_len;
2499 
2500 	return 0;
2501 }
2502 
2503 static void get_fw_ver_str(struct ib_device *device, char *str)
2504 {
2505 	struct mlx4_ib_dev *dev =
2506 		container_of(device, struct mlx4_ib_dev, ib_dev);
2507 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2508 		 (int) (dev->dev->caps.fw_ver >> 32),
2509 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2510 		 (int) dev->dev->caps.fw_ver & 0xffff);
2511 }
2512 
2513 static const struct ib_device_ops mlx4_ib_dev_ops = {
2514 	.add_gid = mlx4_ib_add_gid,
2515 	.alloc_mr = mlx4_ib_alloc_mr,
2516 	.alloc_pd = mlx4_ib_alloc_pd,
2517 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2518 	.attach_mcast = mlx4_ib_mcg_attach,
2519 	.create_ah = mlx4_ib_create_ah,
2520 	.create_cq = mlx4_ib_create_cq,
2521 	.create_qp = mlx4_ib_create_qp,
2522 	.create_srq = mlx4_ib_create_srq,
2523 	.dealloc_pd = mlx4_ib_dealloc_pd,
2524 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2525 	.del_gid = mlx4_ib_del_gid,
2526 	.dereg_mr = mlx4_ib_dereg_mr,
2527 	.destroy_ah = mlx4_ib_destroy_ah,
2528 	.destroy_cq = mlx4_ib_destroy_cq,
2529 	.destroy_qp = mlx4_ib_destroy_qp,
2530 	.destroy_srq = mlx4_ib_destroy_srq,
2531 	.detach_mcast = mlx4_ib_mcg_detach,
2532 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2533 	.drain_rq = mlx4_ib_drain_rq,
2534 	.drain_sq = mlx4_ib_drain_sq,
2535 	.get_dev_fw_str = get_fw_ver_str,
2536 	.get_dma_mr = mlx4_ib_get_dma_mr,
2537 	.get_link_layer = mlx4_ib_port_link_layer,
2538 	.get_netdev = mlx4_ib_get_netdev,
2539 	.get_port_immutable = mlx4_port_immutable,
2540 	.map_mr_sg = mlx4_ib_map_mr_sg,
2541 	.mmap = mlx4_ib_mmap,
2542 	.modify_cq = mlx4_ib_modify_cq,
2543 	.modify_device = mlx4_ib_modify_device,
2544 	.modify_port = mlx4_ib_modify_port,
2545 	.modify_qp = mlx4_ib_modify_qp,
2546 	.modify_srq = mlx4_ib_modify_srq,
2547 	.poll_cq = mlx4_ib_poll_cq,
2548 	.post_recv = mlx4_ib_post_recv,
2549 	.post_send = mlx4_ib_post_send,
2550 	.post_srq_recv = mlx4_ib_post_srq_recv,
2551 	.process_mad = mlx4_ib_process_mad,
2552 	.query_ah = mlx4_ib_query_ah,
2553 	.query_device = mlx4_ib_query_device,
2554 	.query_gid = mlx4_ib_query_gid,
2555 	.query_pkey = mlx4_ib_query_pkey,
2556 	.query_port = mlx4_ib_query_port,
2557 	.query_qp = mlx4_ib_query_qp,
2558 	.query_srq = mlx4_ib_query_srq,
2559 	.reg_user_mr = mlx4_ib_reg_user_mr,
2560 	.req_notify_cq = mlx4_ib_arm_cq,
2561 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2562 	.resize_cq = mlx4_ib_resize_cq,
2563 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2564 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2565 };
2566 
2567 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2568 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2569 	.create_wq = mlx4_ib_create_wq,
2570 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2571 	.destroy_wq = mlx4_ib_destroy_wq,
2572 	.modify_wq = mlx4_ib_modify_wq,
2573 };
2574 
2575 static const struct ib_device_ops mlx4_ib_dev_fmr_ops = {
2576 	.alloc_fmr = mlx4_ib_fmr_alloc,
2577 	.dealloc_fmr = mlx4_ib_fmr_dealloc,
2578 	.map_phys_fmr = mlx4_ib_map_phys_fmr,
2579 	.unmap_fmr = mlx4_ib_unmap_fmr,
2580 };
2581 
2582 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2583 	.alloc_mw = mlx4_ib_alloc_mw,
2584 	.dealloc_mw = mlx4_ib_dealloc_mw,
2585 };
2586 
2587 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2588 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2589 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2590 };
2591 
2592 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2593 	.create_flow = mlx4_ib_create_flow,
2594 	.destroy_flow = mlx4_ib_destroy_flow,
2595 };
2596 
2597 static void *mlx4_ib_add(struct mlx4_dev *dev)
2598 {
2599 	struct mlx4_ib_dev *ibdev;
2600 	int num_ports = 0;
2601 	int i, j;
2602 	int err;
2603 	struct mlx4_ib_iboe *iboe;
2604 	int ib_num_ports = 0;
2605 	int num_req_counters;
2606 	int allocated;
2607 	u32 counter_index;
2608 	struct counter_index *new_counter_index = NULL;
2609 
2610 	pr_info_once("%s", mlx4_ib_version);
2611 
2612 	num_ports = 0;
2613 	mlx4_foreach_ib_transport_port(i, dev)
2614 		num_ports++;
2615 
2616 	/* No point in registering a device with no ports... */
2617 	if (num_ports == 0)
2618 		return NULL;
2619 
2620 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2621 	if (!ibdev) {
2622 		dev_err(&dev->persist->pdev->dev,
2623 			"Device struct alloc failed\n");
2624 		return NULL;
2625 	}
2626 
2627 	iboe = &ibdev->iboe;
2628 
2629 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2630 		goto err_dealloc;
2631 
2632 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2633 		goto err_pd;
2634 
2635 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2636 				 PAGE_SIZE);
2637 	if (!ibdev->uar_map)
2638 		goto err_uar;
2639 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2640 
2641 	ibdev->dev = dev;
2642 	ibdev->bond_next_port	= 0;
2643 
2644 	ibdev->ib_dev.owner		= THIS_MODULE;
2645 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2646 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2647 	ibdev->num_ports		= num_ports;
2648 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2649 						1 : ibdev->num_ports;
2650 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2651 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2652 
2653 	if (dev->caps.userspace_caps)
2654 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2655 	else
2656 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2657 
2658 	ibdev->ib_dev.uverbs_cmd_mask	=
2659 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2660 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2661 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2662 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2663 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2664 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2665 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2666 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2667 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2668 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2669 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2670 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2671 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2672 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2673 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2674 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2675 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2676 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2677 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2678 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2679 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2680 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2681 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2682 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2683 
2684 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2685 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2686 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2687 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2688 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2689 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2690 
2691 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2692 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2693 	    IB_LINK_LAYER_ETHERNET) ||
2694 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2695 	    IB_LINK_LAYER_ETHERNET))) {
2696 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2697 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2698 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2699 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2700 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2701 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2702 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2703 	}
2704 
2705 	if (!mlx4_is_slave(ibdev->dev))
2706 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fmr_ops);
2707 
2708 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2709 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2710 		ibdev->ib_dev.uverbs_cmd_mask |=
2711 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2712 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2713 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2714 	}
2715 
2716 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2717 		ibdev->ib_dev.uverbs_cmd_mask |=
2718 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2719 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2720 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2721 	}
2722 
2723 	if (check_flow_steering_support(dev)) {
2724 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2725 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2726 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2727 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2728 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2729 	}
2730 
2731 	mlx4_ib_alloc_eqs(dev, ibdev);
2732 
2733 	spin_lock_init(&iboe->lock);
2734 
2735 	if (init_node_data(ibdev))
2736 		goto err_map;
2737 	mlx4_init_sl2vl_tbl(ibdev);
2738 
2739 	for (i = 0; i < ibdev->num_ports; ++i) {
2740 		mutex_init(&ibdev->counters_table[i].mutex);
2741 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2742 		iboe->last_port_state[i] = IB_PORT_DOWN;
2743 	}
2744 
2745 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2746 	for (i = 0; i < num_req_counters; ++i) {
2747 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2748 		allocated = 0;
2749 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2750 						IB_LINK_LAYER_ETHERNET) {
2751 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2752 						 MLX4_RES_USAGE_DRIVER);
2753 			/* if failed to allocate a new counter, use default */
2754 			if (err)
2755 				counter_index =
2756 					mlx4_get_default_counter_index(dev,
2757 								       i + 1);
2758 			else
2759 				allocated = 1;
2760 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2761 			counter_index = mlx4_get_default_counter_index(dev,
2762 								       i + 1);
2763 		}
2764 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2765 					    GFP_KERNEL);
2766 		if (!new_counter_index) {
2767 			if (allocated)
2768 				mlx4_counter_free(ibdev->dev, counter_index);
2769 			goto err_counter;
2770 		}
2771 		new_counter_index->index = counter_index;
2772 		new_counter_index->allocated = allocated;
2773 		list_add_tail(&new_counter_index->list,
2774 			      &ibdev->counters_table[i].counters_list);
2775 		ibdev->counters_table[i].default_counter = counter_index;
2776 		pr_info("counter index %d for port %d allocated %d\n",
2777 			counter_index, i + 1, allocated);
2778 	}
2779 	if (mlx4_is_bonded(dev))
2780 		for (i = 1; i < ibdev->num_ports ; ++i) {
2781 			new_counter_index =
2782 					kmalloc(sizeof(struct counter_index),
2783 						GFP_KERNEL);
2784 			if (!new_counter_index)
2785 				goto err_counter;
2786 			new_counter_index->index = counter_index;
2787 			new_counter_index->allocated = 0;
2788 			list_add_tail(&new_counter_index->list,
2789 				      &ibdev->counters_table[i].counters_list);
2790 			ibdev->counters_table[i].default_counter =
2791 								counter_index;
2792 		}
2793 
2794 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2795 		ib_num_ports++;
2796 
2797 	spin_lock_init(&ibdev->sm_lock);
2798 	mutex_init(&ibdev->cap_mask_mutex);
2799 	INIT_LIST_HEAD(&ibdev->qp_list);
2800 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2801 
2802 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2803 	    ib_num_ports) {
2804 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2805 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2806 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2807 					    &ibdev->steer_qpn_base, 0,
2808 					    MLX4_RES_USAGE_DRIVER);
2809 		if (err)
2810 			goto err_counter;
2811 
2812 		ibdev->ib_uc_qpns_bitmap =
2813 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2814 				      sizeof(long),
2815 				      GFP_KERNEL);
2816 		if (!ibdev->ib_uc_qpns_bitmap)
2817 			goto err_steer_qp_release;
2818 
2819 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2820 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2821 				    ibdev->steer_qpn_count);
2822 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2823 					dev, ibdev->steer_qpn_base,
2824 					ibdev->steer_qpn_base +
2825 					ibdev->steer_qpn_count - 1);
2826 			if (err)
2827 				goto err_steer_free_bitmap;
2828 		} else {
2829 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2830 				    ibdev->steer_qpn_count);
2831 		}
2832 	}
2833 
2834 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2835 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2836 
2837 	if (mlx4_ib_alloc_diag_counters(ibdev))
2838 		goto err_steer_free_bitmap;
2839 
2840 	rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2841 	ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2842 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d"))
2843 		goto err_diag_counters;
2844 
2845 	if (mlx4_ib_mad_init(ibdev))
2846 		goto err_reg;
2847 
2848 	if (mlx4_ib_init_sriov(ibdev))
2849 		goto err_mad;
2850 
2851 	if (!iboe->nb.notifier_call) {
2852 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2853 		err = register_netdevice_notifier(&iboe->nb);
2854 		if (err) {
2855 			iboe->nb.notifier_call = NULL;
2856 			goto err_notif;
2857 		}
2858 	}
2859 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2860 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2861 		if (err)
2862 			goto err_notif;
2863 	}
2864 
2865 	ibdev->ib_active = true;
2866 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2867 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2868 					 &ibdev->ib_dev);
2869 
2870 	if (mlx4_is_mfunc(ibdev->dev))
2871 		init_pkeys(ibdev);
2872 
2873 	/* create paravirt contexts for any VFs which are active */
2874 	if (mlx4_is_master(ibdev->dev)) {
2875 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2876 			if (j == mlx4_master_func_num(ibdev->dev))
2877 				continue;
2878 			if (mlx4_is_slave_active(ibdev->dev, j))
2879 				do_slave_init(ibdev, j, 1);
2880 		}
2881 	}
2882 	return ibdev;
2883 
2884 err_notif:
2885 	if (ibdev->iboe.nb.notifier_call) {
2886 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2887 			pr_warn("failure unregistering notifier\n");
2888 		ibdev->iboe.nb.notifier_call = NULL;
2889 	}
2890 	flush_workqueue(wq);
2891 
2892 	mlx4_ib_close_sriov(ibdev);
2893 
2894 err_mad:
2895 	mlx4_ib_mad_cleanup(ibdev);
2896 
2897 err_reg:
2898 	ib_unregister_device(&ibdev->ib_dev);
2899 
2900 err_diag_counters:
2901 	mlx4_ib_diag_cleanup(ibdev);
2902 
2903 err_steer_free_bitmap:
2904 	kfree(ibdev->ib_uc_qpns_bitmap);
2905 
2906 err_steer_qp_release:
2907 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2908 			      ibdev->steer_qpn_count);
2909 err_counter:
2910 	for (i = 0; i < ibdev->num_ports; ++i)
2911 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2912 
2913 err_map:
2914 	mlx4_ib_free_eqs(dev, ibdev);
2915 	iounmap(ibdev->uar_map);
2916 
2917 err_uar:
2918 	mlx4_uar_free(dev, &ibdev->priv_uar);
2919 
2920 err_pd:
2921 	mlx4_pd_free(dev, ibdev->priv_pdn);
2922 
2923 err_dealloc:
2924 	ib_dealloc_device(&ibdev->ib_dev);
2925 
2926 	return NULL;
2927 }
2928 
2929 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2930 {
2931 	int offset;
2932 
2933 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2934 
2935 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2936 					 dev->steer_qpn_count,
2937 					 get_count_order(count));
2938 	if (offset < 0)
2939 		return offset;
2940 
2941 	*qpn = dev->steer_qpn_base + offset;
2942 	return 0;
2943 }
2944 
2945 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2946 {
2947 	if (!qpn ||
2948 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2949 		return;
2950 
2951 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2952 		 qpn, dev->steer_qpn_base))
2953 		/* not supposed to be here */
2954 		return;
2955 
2956 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2957 			      qpn - dev->steer_qpn_base,
2958 			      get_count_order(count));
2959 }
2960 
2961 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2962 			 int is_attach)
2963 {
2964 	int err;
2965 	size_t flow_size;
2966 	struct ib_flow_attr *flow = NULL;
2967 	struct ib_flow_spec_ib *ib_spec;
2968 
2969 	if (is_attach) {
2970 		flow_size = sizeof(struct ib_flow_attr) +
2971 			    sizeof(struct ib_flow_spec_ib);
2972 		flow = kzalloc(flow_size, GFP_KERNEL);
2973 		if (!flow)
2974 			return -ENOMEM;
2975 		flow->port = mqp->port;
2976 		flow->num_of_specs = 1;
2977 		flow->size = flow_size;
2978 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2979 		ib_spec->type = IB_FLOW_SPEC_IB;
2980 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2981 		/* Add an empty rule for IB L2 */
2982 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2983 
2984 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2985 					    IB_FLOW_DOMAIN_NIC,
2986 					    MLX4_FS_REGULAR,
2987 					    &mqp->reg_id);
2988 	} else {
2989 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2990 	}
2991 	kfree(flow);
2992 	return err;
2993 }
2994 
2995 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2996 {
2997 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2998 	int p;
2999 	int i;
3000 
3001 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3002 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3003 	ibdev->ib_active = false;
3004 	flush_workqueue(wq);
3005 
3006 	mlx4_ib_close_sriov(ibdev);
3007 	mlx4_ib_mad_cleanup(ibdev);
3008 	ib_unregister_device(&ibdev->ib_dev);
3009 	mlx4_ib_diag_cleanup(ibdev);
3010 	if (ibdev->iboe.nb.notifier_call) {
3011 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3012 			pr_warn("failure unregistering notifier\n");
3013 		ibdev->iboe.nb.notifier_call = NULL;
3014 	}
3015 
3016 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3017 			      ibdev->steer_qpn_count);
3018 	kfree(ibdev->ib_uc_qpns_bitmap);
3019 
3020 	iounmap(ibdev->uar_map);
3021 	for (p = 0; p < ibdev->num_ports; ++p)
3022 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3023 
3024 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3025 		mlx4_CLOSE_PORT(dev, p);
3026 
3027 	mlx4_ib_free_eqs(dev, ibdev);
3028 
3029 	mlx4_uar_free(dev, &ibdev->priv_uar);
3030 	mlx4_pd_free(dev, ibdev->priv_pdn);
3031 	ib_dealloc_device(&ibdev->ib_dev);
3032 }
3033 
3034 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3035 {
3036 	struct mlx4_ib_demux_work **dm = NULL;
3037 	struct mlx4_dev *dev = ibdev->dev;
3038 	int i;
3039 	unsigned long flags;
3040 	struct mlx4_active_ports actv_ports;
3041 	unsigned int ports;
3042 	unsigned int first_port;
3043 
3044 	if (!mlx4_is_master(dev))
3045 		return;
3046 
3047 	actv_ports = mlx4_get_active_ports(dev, slave);
3048 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3049 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3050 
3051 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3052 	if (!dm)
3053 		return;
3054 
3055 	for (i = 0; i < ports; i++) {
3056 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3057 		if (!dm[i]) {
3058 			while (--i >= 0)
3059 				kfree(dm[i]);
3060 			goto out;
3061 		}
3062 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3063 		dm[i]->port = first_port + i + 1;
3064 		dm[i]->slave = slave;
3065 		dm[i]->do_init = do_init;
3066 		dm[i]->dev = ibdev;
3067 	}
3068 	/* initialize or tear down tunnel QPs for the slave */
3069 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3070 	if (!ibdev->sriov.is_going_down) {
3071 		for (i = 0; i < ports; i++)
3072 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3073 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3074 	} else {
3075 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3076 		for (i = 0; i < ports; i++)
3077 			kfree(dm[i]);
3078 	}
3079 out:
3080 	kfree(dm);
3081 	return;
3082 }
3083 
3084 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3085 {
3086 	struct mlx4_ib_qp *mqp;
3087 	unsigned long flags_qp;
3088 	unsigned long flags_cq;
3089 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3090 	struct list_head    cq_notify_list;
3091 	struct mlx4_cq *mcq;
3092 	unsigned long flags;
3093 
3094 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3095 	INIT_LIST_HEAD(&cq_notify_list);
3096 
3097 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3098 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3099 
3100 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3101 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3102 		if (mqp->sq.tail != mqp->sq.head) {
3103 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3104 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3105 			if (send_mcq->mcq.comp &&
3106 			    mqp->ibqp.send_cq->comp_handler) {
3107 				if (!send_mcq->mcq.reset_notify_added) {
3108 					send_mcq->mcq.reset_notify_added = 1;
3109 					list_add_tail(&send_mcq->mcq.reset_notify,
3110 						      &cq_notify_list);
3111 				}
3112 			}
3113 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3114 		}
3115 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3116 		/* Now, handle the QP's receive queue */
3117 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3118 		/* no handling is needed for SRQ */
3119 		if (!mqp->ibqp.srq) {
3120 			if (mqp->rq.tail != mqp->rq.head) {
3121 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3122 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3123 				if (recv_mcq->mcq.comp &&
3124 				    mqp->ibqp.recv_cq->comp_handler) {
3125 					if (!recv_mcq->mcq.reset_notify_added) {
3126 						recv_mcq->mcq.reset_notify_added = 1;
3127 						list_add_tail(&recv_mcq->mcq.reset_notify,
3128 							      &cq_notify_list);
3129 					}
3130 				}
3131 				spin_unlock_irqrestore(&recv_mcq->lock,
3132 						       flags_cq);
3133 			}
3134 		}
3135 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3136 	}
3137 
3138 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3139 		mcq->comp(mcq);
3140 	}
3141 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3142 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3143 }
3144 
3145 static void handle_bonded_port_state_event(struct work_struct *work)
3146 {
3147 	struct ib_event_work *ew =
3148 		container_of(work, struct ib_event_work, work);
3149 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3150 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3151 	int i;
3152 	struct ib_event ibev;
3153 
3154 	kfree(ew);
3155 	spin_lock_bh(&ibdev->iboe.lock);
3156 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3157 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3158 		enum ib_port_state curr_port_state;
3159 
3160 		if (!curr_netdev)
3161 			continue;
3162 
3163 		curr_port_state =
3164 			(netif_running(curr_netdev) &&
3165 			 netif_carrier_ok(curr_netdev)) ?
3166 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3167 
3168 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3169 			curr_port_state : IB_PORT_ACTIVE;
3170 	}
3171 	spin_unlock_bh(&ibdev->iboe.lock);
3172 
3173 	ibev.device = &ibdev->ib_dev;
3174 	ibev.element.port_num = 1;
3175 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3176 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3177 
3178 	ib_dispatch_event(&ibev);
3179 }
3180 
3181 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3182 {
3183 	u64 sl2vl;
3184 	int err;
3185 
3186 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3187 	if (err) {
3188 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3189 		       port, err);
3190 		sl2vl = 0;
3191 	}
3192 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3193 }
3194 
3195 static void ib_sl2vl_update_work(struct work_struct *work)
3196 {
3197 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3198 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3199 	int port = ew->port;
3200 
3201 	mlx4_ib_sl2vl_update(mdev, port);
3202 
3203 	kfree(ew);
3204 }
3205 
3206 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3207 				     int port)
3208 {
3209 	struct ib_event_work *ew;
3210 
3211 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3212 	if (ew) {
3213 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3214 		ew->port = port;
3215 		ew->ib_dev = ibdev;
3216 		queue_work(wq, &ew->work);
3217 	}
3218 }
3219 
3220 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3221 			  enum mlx4_dev_event event, unsigned long param)
3222 {
3223 	struct ib_event ibev;
3224 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3225 	struct mlx4_eqe *eqe = NULL;
3226 	struct ib_event_work *ew;
3227 	int p = 0;
3228 
3229 	if (mlx4_is_bonded(dev) &&
3230 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3231 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3232 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3233 		if (!ew)
3234 			return;
3235 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3236 		ew->ib_dev = ibdev;
3237 		queue_work(wq, &ew->work);
3238 		return;
3239 	}
3240 
3241 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3242 		eqe = (struct mlx4_eqe *)param;
3243 	else
3244 		p = (int) param;
3245 
3246 	switch (event) {
3247 	case MLX4_DEV_EVENT_PORT_UP:
3248 		if (p > ibdev->num_ports)
3249 			return;
3250 		if (!mlx4_is_slave(dev) &&
3251 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3252 			IB_LINK_LAYER_INFINIBAND) {
3253 			if (mlx4_is_master(dev))
3254 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3255 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3256 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3257 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3258 		}
3259 		ibev.event = IB_EVENT_PORT_ACTIVE;
3260 		break;
3261 
3262 	case MLX4_DEV_EVENT_PORT_DOWN:
3263 		if (p > ibdev->num_ports)
3264 			return;
3265 		ibev.event = IB_EVENT_PORT_ERR;
3266 		break;
3267 
3268 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3269 		ibdev->ib_active = false;
3270 		ibev.event = IB_EVENT_DEVICE_FATAL;
3271 		mlx4_ib_handle_catas_error(ibdev);
3272 		break;
3273 
3274 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3275 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3276 		if (!ew)
3277 			break;
3278 
3279 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3280 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3281 		ew->ib_dev = ibdev;
3282 		/* need to queue only for port owner, which uses GEN_EQE */
3283 		if (mlx4_is_master(dev))
3284 			queue_work(wq, &ew->work);
3285 		else
3286 			handle_port_mgmt_change_event(&ew->work);
3287 		return;
3288 
3289 	case MLX4_DEV_EVENT_SLAVE_INIT:
3290 		/* here, p is the slave id */
3291 		do_slave_init(ibdev, p, 1);
3292 		if (mlx4_is_master(dev)) {
3293 			int i;
3294 
3295 			for (i = 1; i <= ibdev->num_ports; i++) {
3296 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3297 					== IB_LINK_LAYER_INFINIBAND)
3298 					mlx4_ib_slave_alias_guid_event(ibdev,
3299 								       p, i,
3300 								       1);
3301 			}
3302 		}
3303 		return;
3304 
3305 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3306 		if (mlx4_is_master(dev)) {
3307 			int i;
3308 
3309 			for (i = 1; i <= ibdev->num_ports; i++) {
3310 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3311 					== IB_LINK_LAYER_INFINIBAND)
3312 					mlx4_ib_slave_alias_guid_event(ibdev,
3313 								       p, i,
3314 								       0);
3315 			}
3316 		}
3317 		/* here, p is the slave id */
3318 		do_slave_init(ibdev, p, 0);
3319 		return;
3320 
3321 	default:
3322 		return;
3323 	}
3324 
3325 	ibev.device	      = ibdev_ptr;
3326 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3327 
3328 	ib_dispatch_event(&ibev);
3329 }
3330 
3331 static struct mlx4_interface mlx4_ib_interface = {
3332 	.add		= mlx4_ib_add,
3333 	.remove		= mlx4_ib_remove,
3334 	.event		= mlx4_ib_event,
3335 	.protocol	= MLX4_PROT_IB_IPV6,
3336 	.flags		= MLX4_INTFF_BONDING
3337 };
3338 
3339 static int __init mlx4_ib_init(void)
3340 {
3341 	int err;
3342 
3343 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3344 	if (!wq)
3345 		return -ENOMEM;
3346 
3347 	err = mlx4_ib_mcg_init();
3348 	if (err)
3349 		goto clean_wq;
3350 
3351 	err = mlx4_register_interface(&mlx4_ib_interface);
3352 	if (err)
3353 		goto clean_mcg;
3354 
3355 	return 0;
3356 
3357 clean_mcg:
3358 	mlx4_ib_mcg_destroy();
3359 
3360 clean_wq:
3361 	destroy_workqueue(wq);
3362 	return err;
3363 }
3364 
3365 static void __exit mlx4_ib_cleanup(void)
3366 {
3367 	mlx4_unregister_interface(&mlx4_ib_interface);
3368 	mlx4_ib_mcg_destroy();
3369 	destroy_workqueue(wq);
3370 }
3371 
3372 module_init(mlx4_ib_init);
3373 module_exit(mlx4_ib_cleanup);
3374