xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 36d26912)
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 void free_gid_entry(struct gid_entry *entry)
250 {
251 	memset(&entry->gid, 0, sizeof(entry->gid));
252 	kfree(entry->ctx);
253 	entry->ctx = NULL;
254 }
255 
256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids = NULL;
266 	u16 vlan_id = 0xffff;
267 	u8 mac[ETH_ALEN];
268 
269 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
270 		return -EINVAL;
271 
272 	if (attr->port_num > MLX4_MAX_PORTS)
273 		return -EINVAL;
274 
275 	if (!context)
276 		return -EINVAL;
277 
278 	ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
279 	if (ret)
280 		return ret;
281 	port_gid_table = &iboe->gids[attr->port_num - 1];
282 	spin_lock_bh(&iboe->lock);
283 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
284 		if (!memcmp(&port_gid_table->gids[i].gid,
285 			    &attr->gid, sizeof(attr->gid)) &&
286 		    port_gid_table->gids[i].gid_type == attr->gid_type &&
287 		    port_gid_table->gids[i].vlan_id == vlan_id)  {
288 			found = i;
289 			break;
290 		}
291 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
292 			free = i; /* HW has space */
293 	}
294 
295 	if (found < 0) {
296 		if (free < 0) {
297 			ret = -ENOSPC;
298 		} else {
299 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
300 			if (!port_gid_table->gids[free].ctx) {
301 				ret = -ENOMEM;
302 			} else {
303 				*context = port_gid_table->gids[free].ctx;
304 				memcpy(&port_gid_table->gids[free].gid,
305 				       &attr->gid, sizeof(attr->gid));
306 				port_gid_table->gids[free].gid_type = attr->gid_type;
307 				port_gid_table->gids[free].vlan_id = vlan_id;
308 				port_gid_table->gids[free].ctx->real_index = free;
309 				port_gid_table->gids[free].ctx->refcount = 1;
310 				hw_update = 1;
311 			}
312 		}
313 	} else {
314 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
315 		*context = ctx;
316 		ctx->refcount++;
317 	}
318 	if (!ret && hw_update) {
319 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
320 				     GFP_ATOMIC);
321 		if (!gids) {
322 			ret = -ENOMEM;
323 			*context = NULL;
324 			free_gid_entry(&port_gid_table->gids[free]);
325 		} else {
326 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
327 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
328 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
329 			}
330 		}
331 	}
332 	spin_unlock_bh(&iboe->lock);
333 
334 	if (!ret && hw_update) {
335 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
336 		if (ret) {
337 			spin_lock_bh(&iboe->lock);
338 			*context = NULL;
339 			free_gid_entry(&port_gid_table->gids[free]);
340 			spin_unlock_bh(&iboe->lock);
341 		}
342 		kfree(gids);
343 	}
344 
345 	return ret;
346 }
347 
348 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
349 {
350 	struct gid_cache_context *ctx = *context;
351 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
352 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
353 	struct mlx4_port_gid_table   *port_gid_table;
354 	int ret = 0;
355 	int hw_update = 0;
356 	struct gid_entry *gids = NULL;
357 
358 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
359 		return -EINVAL;
360 
361 	if (attr->port_num > MLX4_MAX_PORTS)
362 		return -EINVAL;
363 
364 	port_gid_table = &iboe->gids[attr->port_num - 1];
365 	spin_lock_bh(&iboe->lock);
366 	if (ctx) {
367 		ctx->refcount--;
368 		if (!ctx->refcount) {
369 			unsigned int real_index = ctx->real_index;
370 
371 			free_gid_entry(&port_gid_table->gids[real_index]);
372 			hw_update = 1;
373 		}
374 	}
375 	if (!ret && hw_update) {
376 		int i;
377 
378 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
379 				     GFP_ATOMIC);
380 		if (!gids) {
381 			ret = -ENOMEM;
382 		} else {
383 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
384 				memcpy(&gids[i].gid,
385 				       &port_gid_table->gids[i].gid,
386 				       sizeof(union ib_gid));
387 				gids[i].gid_type =
388 				    port_gid_table->gids[i].gid_type;
389 			}
390 		}
391 	}
392 	spin_unlock_bh(&iboe->lock);
393 
394 	if (!ret && hw_update) {
395 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
396 		kfree(gids);
397 	}
398 	return ret;
399 }
400 
401 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
402 				    const struct ib_gid_attr *attr)
403 {
404 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
405 	struct gid_cache_context *ctx = NULL;
406 	struct mlx4_port_gid_table   *port_gid_table;
407 	int real_index = -EINVAL;
408 	int i;
409 	unsigned long flags;
410 	u8 port_num = attr->port_num;
411 
412 	if (port_num > MLX4_MAX_PORTS)
413 		return -EINVAL;
414 
415 	if (mlx4_is_bonded(ibdev->dev))
416 		port_num = 1;
417 
418 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
419 		return attr->index;
420 
421 	spin_lock_irqsave(&iboe->lock, flags);
422 	port_gid_table = &iboe->gids[port_num - 1];
423 
424 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
425 		if (!memcmp(&port_gid_table->gids[i].gid,
426 			    &attr->gid, sizeof(attr->gid)) &&
427 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
428 			ctx = port_gid_table->gids[i].ctx;
429 			break;
430 		}
431 	if (ctx)
432 		real_index = ctx->real_index;
433 	spin_unlock_irqrestore(&iboe->lock, flags);
434 	return real_index;
435 }
436 
437 static int mlx4_ib_query_device(struct ib_device *ibdev,
438 				struct ib_device_attr *props,
439 				struct ib_udata *uhw)
440 {
441 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
442 	struct ib_smp *in_mad  = NULL;
443 	struct ib_smp *out_mad = NULL;
444 	int err;
445 	int have_ib_ports;
446 	struct mlx4_uverbs_ex_query_device cmd;
447 	struct mlx4_uverbs_ex_query_device_resp resp = {};
448 	struct mlx4_clock_params clock_params;
449 
450 	if (uhw->inlen) {
451 		if (uhw->inlen < sizeof(cmd))
452 			return -EINVAL;
453 
454 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
455 		if (err)
456 			return err;
457 
458 		if (cmd.comp_mask)
459 			return -EINVAL;
460 
461 		if (cmd.reserved)
462 			return -EINVAL;
463 	}
464 
465 	resp.response_length = offsetof(typeof(resp), response_length) +
466 		sizeof(resp.response_length);
467 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
468 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
469 	err = -ENOMEM;
470 	if (!in_mad || !out_mad)
471 		goto out;
472 
473 	init_query_mad(in_mad);
474 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
475 
476 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
477 			   1, NULL, NULL, in_mad, out_mad);
478 	if (err)
479 		goto out;
480 
481 	memset(props, 0, sizeof *props);
482 
483 	have_ib_ports = num_ib_ports(dev->dev);
484 
485 	props->fw_ver = dev->dev->caps.fw_ver;
486 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
487 		IB_DEVICE_PORT_ACTIVE_EVENT		|
488 		IB_DEVICE_SYS_IMAGE_GUID		|
489 		IB_DEVICE_RC_RNR_NAK_GEN		|
490 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
492 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
493 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
494 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
496 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
497 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
498 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
499 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
500 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
501 	if (dev->dev->caps.max_gso_sz &&
502 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
503 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
504 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
505 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
506 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
507 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
508 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
509 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
510 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
511 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
512 		props->device_cap_flags |= IB_DEVICE_XRC;
513 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
514 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
515 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
516 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
517 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
518 		else
519 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
520 	}
521 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
522 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
523 
524 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
525 
526 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
527 		0xffffff;
528 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
529 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
530 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
531 
532 	props->max_mr_size	   = ~0ull;
533 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
534 	props->max_qp		   = dev->dev->quotas.qp;
535 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
536 	props->max_send_sge =
537 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
538 	props->max_recv_sge =
539 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
540 	props->max_sge_rd = MLX4_MAX_SGE_RD;
541 	props->max_cq		   = dev->dev->quotas.cq;
542 	props->max_cqe		   = dev->dev->caps.max_cqes;
543 	props->max_mr		   = dev->dev->quotas.mpt;
544 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
545 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
546 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
547 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
548 	props->max_srq		   = dev->dev->quotas.srq;
549 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
550 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
551 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
552 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
553 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
554 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
555 	props->masked_atomic_cap   = props->atomic_cap;
556 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
557 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
558 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
559 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
560 					   props->max_mcast_grp;
561 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
562 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
563 	props->max_ah = INT_MAX;
564 
565 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
566 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
567 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
568 			props->rss_caps.max_rwq_indirection_tables =
569 				props->max_qp;
570 			props->rss_caps.max_rwq_indirection_table_size =
571 				dev->dev->caps.max_rss_tbl_sz;
572 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
573 			props->max_wq_type_rq = props->max_qp;
574 		}
575 
576 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
577 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
578 	}
579 
580 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
581 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
582 
583 	if (!mlx4_is_slave(dev->dev))
584 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
585 
586 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
587 		resp.response_length += sizeof(resp.hca_core_clock_offset);
588 		if (!err && !mlx4_is_slave(dev->dev)) {
589 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
590 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
591 		}
592 	}
593 
594 	if (uhw->outlen >= resp.response_length +
595 	    sizeof(resp.max_inl_recv_sz)) {
596 		resp.response_length += sizeof(resp.max_inl_recv_sz);
597 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
598 			sizeof(struct mlx4_wqe_data_seg);
599 	}
600 
601 	if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
602 		if (props->rss_caps.supported_qpts) {
603 			resp.rss_caps.rx_hash_function =
604 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
605 
606 			resp.rss_caps.rx_hash_fields_mask =
607 				MLX4_IB_RX_HASH_SRC_IPV4 |
608 				MLX4_IB_RX_HASH_DST_IPV4 |
609 				MLX4_IB_RX_HASH_SRC_IPV6 |
610 				MLX4_IB_RX_HASH_DST_IPV6 |
611 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
612 				MLX4_IB_RX_HASH_DST_PORT_TCP |
613 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
614 				MLX4_IB_RX_HASH_DST_PORT_UDP;
615 
616 			if (dev->dev->caps.tunnel_offload_mode ==
617 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
618 				resp.rss_caps.rx_hash_fields_mask |=
619 					MLX4_IB_RX_HASH_INNER;
620 		}
621 		resp.response_length = offsetof(typeof(resp), rss_caps) +
622 				       sizeof(resp.rss_caps);
623 	}
624 
625 	if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
626 		if (dev->dev->caps.max_gso_sz &&
627 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
628 		    IB_LINK_LAYER_ETHERNET) ||
629 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
630 		    IB_LINK_LAYER_ETHERNET))) {
631 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
632 			resp.tso_caps.supported_qpts |=
633 				1 << IB_QPT_RAW_PACKET;
634 		}
635 		resp.response_length = offsetof(typeof(resp), tso_caps) +
636 				       sizeof(resp.tso_caps);
637 	}
638 
639 	if (uhw->outlen) {
640 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
641 		if (err)
642 			goto out;
643 	}
644 out:
645 	kfree(in_mad);
646 	kfree(out_mad);
647 
648 	return err;
649 }
650 
651 static enum rdma_link_layer
652 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
653 {
654 	struct mlx4_dev *dev = to_mdev(device)->dev;
655 
656 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
657 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
658 }
659 
660 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
661 			      struct ib_port_attr *props, int netw_view)
662 {
663 	struct ib_smp *in_mad  = NULL;
664 	struct ib_smp *out_mad = NULL;
665 	int ext_active_speed;
666 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
667 	int err = -ENOMEM;
668 
669 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
670 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
671 	if (!in_mad || !out_mad)
672 		goto out;
673 
674 	init_query_mad(in_mad);
675 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
676 	in_mad->attr_mod = cpu_to_be32(port);
677 
678 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
679 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
680 
681 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
682 				in_mad, out_mad);
683 	if (err)
684 		goto out;
685 
686 
687 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
688 	props->lmc		= out_mad->data[34] & 0x7;
689 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
690 	props->sm_sl		= out_mad->data[36] & 0xf;
691 	props->state		= out_mad->data[32] & 0xf;
692 	props->phys_state	= out_mad->data[33] >> 4;
693 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
694 	if (netw_view)
695 		props->gid_tbl_len = out_mad->data[50];
696 	else
697 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
698 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
699 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
700 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
701 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
702 	props->active_width	= out_mad->data[31] & 0xf;
703 	props->active_speed	= out_mad->data[35] >> 4;
704 	props->max_mtu		= out_mad->data[41] & 0xf;
705 	props->active_mtu	= out_mad->data[36] >> 4;
706 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
707 	props->max_vl_num	= out_mad->data[37] >> 4;
708 	props->init_type_reply	= out_mad->data[41] >> 4;
709 
710 	/* Check if extended speeds (EDR/FDR/...) are supported */
711 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
712 		ext_active_speed = out_mad->data[62] >> 4;
713 
714 		switch (ext_active_speed) {
715 		case 1:
716 			props->active_speed = IB_SPEED_FDR;
717 			break;
718 		case 2:
719 			props->active_speed = IB_SPEED_EDR;
720 			break;
721 		}
722 	}
723 
724 	/* If reported active speed is QDR, check if is FDR-10 */
725 	if (props->active_speed == IB_SPEED_QDR) {
726 		init_query_mad(in_mad);
727 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
728 		in_mad->attr_mod = cpu_to_be32(port);
729 
730 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
731 				   NULL, NULL, in_mad, out_mad);
732 		if (err)
733 			goto out;
734 
735 		/* Checking LinkSpeedActive for FDR-10 */
736 		if (out_mad->data[15] & 0x1)
737 			props->active_speed = IB_SPEED_FDR10;
738 	}
739 
740 	/* Avoid wrong speed value returned by FW if the IB link is down. */
741 	if (props->state == IB_PORT_DOWN)
742 		 props->active_speed = IB_SPEED_SDR;
743 
744 out:
745 	kfree(in_mad);
746 	kfree(out_mad);
747 	return err;
748 }
749 
750 static u8 state_to_phys_state(enum ib_port_state state)
751 {
752 	return state == IB_PORT_ACTIVE ?
753 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
754 }
755 
756 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
757 			       struct ib_port_attr *props)
758 {
759 
760 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
761 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
762 	struct net_device *ndev;
763 	enum ib_mtu tmp;
764 	struct mlx4_cmd_mailbox *mailbox;
765 	int err = 0;
766 	int is_bonded = mlx4_is_bonded(mdev->dev);
767 
768 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
769 	if (IS_ERR(mailbox))
770 		return PTR_ERR(mailbox);
771 
772 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
773 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
774 			   MLX4_CMD_WRAPPED);
775 	if (err)
776 		goto out;
777 
778 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
779 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
780 					   IB_WIDTH_4X : IB_WIDTH_1X;
781 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
782 					   IB_SPEED_FDR : IB_SPEED_QDR;
783 	props->port_cap_flags	= IB_PORT_CM_SUP;
784 	props->ip_gids = true;
785 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
786 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
787 	if (mdev->dev->caps.pkey_table_len[port])
788 		props->pkey_tbl_len = 1;
789 	props->max_mtu		= IB_MTU_4096;
790 	props->max_vl_num	= 2;
791 	props->state		= IB_PORT_DOWN;
792 	props->phys_state	= state_to_phys_state(props->state);
793 	props->active_mtu	= IB_MTU_256;
794 	spin_lock_bh(&iboe->lock);
795 	ndev = iboe->netdevs[port - 1];
796 	if (ndev && is_bonded) {
797 		rcu_read_lock(); /* required to get upper dev */
798 		ndev = netdev_master_upper_dev_get_rcu(ndev);
799 		rcu_read_unlock();
800 	}
801 	if (!ndev)
802 		goto out_unlock;
803 
804 	tmp = iboe_get_mtu(ndev->mtu);
805 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
806 
807 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
808 					IB_PORT_ACTIVE : IB_PORT_DOWN;
809 	props->phys_state	= state_to_phys_state(props->state);
810 out_unlock:
811 	spin_unlock_bh(&iboe->lock);
812 out:
813 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
814 	return err;
815 }
816 
817 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
818 			 struct ib_port_attr *props, int netw_view)
819 {
820 	int err;
821 
822 	/* props being zeroed by the caller, avoid zeroing it here */
823 
824 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
825 		ib_link_query_port(ibdev, port, props, netw_view) :
826 				eth_link_query_port(ibdev, port, props);
827 
828 	return err;
829 }
830 
831 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
832 			      struct ib_port_attr *props)
833 {
834 	/* returns host view */
835 	return __mlx4_ib_query_port(ibdev, port, props, 0);
836 }
837 
838 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
839 			union ib_gid *gid, int netw_view)
840 {
841 	struct ib_smp *in_mad  = NULL;
842 	struct ib_smp *out_mad = NULL;
843 	int err = -ENOMEM;
844 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
845 	int clear = 0;
846 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
847 
848 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
849 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
850 	if (!in_mad || !out_mad)
851 		goto out;
852 
853 	init_query_mad(in_mad);
854 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
855 	in_mad->attr_mod = cpu_to_be32(port);
856 
857 	if (mlx4_is_mfunc(dev->dev) && netw_view)
858 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
859 
860 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
861 	if (err)
862 		goto out;
863 
864 	memcpy(gid->raw, out_mad->data + 8, 8);
865 
866 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
867 		if (index) {
868 			/* For any index > 0, return the null guid */
869 			err = 0;
870 			clear = 1;
871 			goto out;
872 		}
873 	}
874 
875 	init_query_mad(in_mad);
876 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
877 	in_mad->attr_mod = cpu_to_be32(index / 8);
878 
879 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
880 			   NULL, NULL, in_mad, out_mad);
881 	if (err)
882 		goto out;
883 
884 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
885 
886 out:
887 	if (clear)
888 		memset(gid->raw + 8, 0, 8);
889 	kfree(in_mad);
890 	kfree(out_mad);
891 	return err;
892 }
893 
894 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
895 			     union ib_gid *gid)
896 {
897 	if (rdma_protocol_ib(ibdev, port))
898 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
899 	return 0;
900 }
901 
902 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
903 {
904 	union sl2vl_tbl_to_u64 sl2vl64;
905 	struct ib_smp *in_mad  = NULL;
906 	struct ib_smp *out_mad = NULL;
907 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
908 	int err = -ENOMEM;
909 	int jj;
910 
911 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
912 		*sl2vl_tbl = 0;
913 		return 0;
914 	}
915 
916 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
917 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
918 	if (!in_mad || !out_mad)
919 		goto out;
920 
921 	init_query_mad(in_mad);
922 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
923 	in_mad->attr_mod = 0;
924 
925 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
926 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
927 
928 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
929 			   in_mad, out_mad);
930 	if (err)
931 		goto out;
932 
933 	for (jj = 0; jj < 8; jj++)
934 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
935 	*sl2vl_tbl = sl2vl64.sl64;
936 
937 out:
938 	kfree(in_mad);
939 	kfree(out_mad);
940 	return err;
941 }
942 
943 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
944 {
945 	u64 sl2vl;
946 	int i;
947 	int err;
948 
949 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
950 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
951 			continue;
952 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
953 		if (err) {
954 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
955 			       i, err);
956 			sl2vl = 0;
957 		}
958 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
959 	}
960 }
961 
962 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
963 			 u16 *pkey, int netw_view)
964 {
965 	struct ib_smp *in_mad  = NULL;
966 	struct ib_smp *out_mad = NULL;
967 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
968 	int err = -ENOMEM;
969 
970 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
971 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
972 	if (!in_mad || !out_mad)
973 		goto out;
974 
975 	init_query_mad(in_mad);
976 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
977 	in_mad->attr_mod = cpu_to_be32(index / 32);
978 
979 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
980 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
981 
982 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
983 			   in_mad, out_mad);
984 	if (err)
985 		goto out;
986 
987 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
988 
989 out:
990 	kfree(in_mad);
991 	kfree(out_mad);
992 	return err;
993 }
994 
995 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
996 {
997 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
998 }
999 
1000 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
1001 				 struct ib_device_modify *props)
1002 {
1003 	struct mlx4_cmd_mailbox *mailbox;
1004 	unsigned long flags;
1005 
1006 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
1007 		return -EOPNOTSUPP;
1008 
1009 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1010 		return 0;
1011 
1012 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1013 		return -EOPNOTSUPP;
1014 
1015 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1016 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1017 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1018 
1019 	/*
1020 	 * If possible, pass node desc to FW, so it can generate
1021 	 * a 144 trap.  If cmd fails, just ignore.
1022 	 */
1023 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1024 	if (IS_ERR(mailbox))
1025 		return 0;
1026 
1027 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1028 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1029 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1030 
1031 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1032 
1033 	return 0;
1034 }
1035 
1036 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1037 			    u32 cap_mask)
1038 {
1039 	struct mlx4_cmd_mailbox *mailbox;
1040 	int err;
1041 
1042 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1043 	if (IS_ERR(mailbox))
1044 		return PTR_ERR(mailbox);
1045 
1046 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1047 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1048 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1049 	} else {
1050 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1051 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1052 	}
1053 
1054 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1055 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1056 		       MLX4_CMD_WRAPPED);
1057 
1058 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1059 	return err;
1060 }
1061 
1062 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1063 			       struct ib_port_modify *props)
1064 {
1065 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1066 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1067 	struct ib_port_attr attr;
1068 	u32 cap_mask;
1069 	int err;
1070 
1071 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1072 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1073 	 * violations and port capabilities are not meaningful.
1074 	 */
1075 	if (is_eth)
1076 		return 0;
1077 
1078 	mutex_lock(&mdev->cap_mask_mutex);
1079 
1080 	err = ib_query_port(ibdev, port, &attr);
1081 	if (err)
1082 		goto out;
1083 
1084 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1085 		~props->clr_port_cap_mask;
1086 
1087 	err = mlx4_ib_SET_PORT(mdev, port,
1088 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1089 			       cap_mask);
1090 
1091 out:
1092 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1093 	return err;
1094 }
1095 
1096 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1097 				  struct ib_udata *udata)
1098 {
1099 	struct ib_device *ibdev = uctx->device;
1100 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1101 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1102 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1103 	struct mlx4_ib_alloc_ucontext_resp resp;
1104 	int err;
1105 
1106 	if (!dev->ib_active)
1107 		return -EAGAIN;
1108 
1109 	if (ibdev->ops.uverbs_abi_ver ==
1110 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1111 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1112 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1113 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1114 	} else {
1115 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1116 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1117 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1118 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1119 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1120 	}
1121 
1122 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1123 	if (err)
1124 		return err;
1125 
1126 	INIT_LIST_HEAD(&context->db_page_list);
1127 	mutex_init(&context->db_page_mutex);
1128 
1129 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1130 	mutex_init(&context->wqn_ranges_mutex);
1131 
1132 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1133 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1134 	else
1135 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1136 
1137 	if (err) {
1138 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1139 		return -EFAULT;
1140 	}
1141 
1142 	return err;
1143 }
1144 
1145 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1146 {
1147 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1148 
1149 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1150 }
1151 
1152 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1153 {
1154 }
1155 
1156 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1157 {
1158 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1159 
1160 	switch (vma->vm_pgoff) {
1161 	case 0:
1162 		return rdma_user_mmap_io(context, vma,
1163 					 to_mucontext(context)->uar.pfn,
1164 					 PAGE_SIZE,
1165 					 pgprot_noncached(vma->vm_page_prot),
1166 					 NULL);
1167 
1168 	case 1:
1169 		if (dev->dev->caps.bf_reg_size == 0)
1170 			return -EINVAL;
1171 		return rdma_user_mmap_io(
1172 			context, vma,
1173 			to_mucontext(context)->uar.pfn +
1174 				dev->dev->caps.num_uars,
1175 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1176 			NULL);
1177 
1178 	case 3: {
1179 		struct mlx4_clock_params params;
1180 		int ret;
1181 
1182 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1183 		if (ret)
1184 			return ret;
1185 
1186 		return rdma_user_mmap_io(
1187 			context, vma,
1188 			(pci_resource_start(dev->dev->persist->pdev,
1189 					    params.bar) +
1190 			 params.offset) >>
1191 				PAGE_SHIFT,
1192 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1193 			NULL);
1194 	}
1195 
1196 	default:
1197 		return -EINVAL;
1198 	}
1199 }
1200 
1201 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1202 {
1203 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1204 	struct ib_device *ibdev = ibpd->device;
1205 	int err;
1206 
1207 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1208 	if (err)
1209 		return err;
1210 
1211 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1212 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1213 		return -EFAULT;
1214 	}
1215 	return 0;
1216 }
1217 
1218 static void mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1219 {
1220 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1221 }
1222 
1223 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1224 {
1225 	struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1226 	struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1227 	struct ib_cq_init_attr cq_attr = {};
1228 	int err;
1229 
1230 	if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1231 		return -EOPNOTSUPP;
1232 
1233 	err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1234 	if (err)
1235 		return err;
1236 
1237 	xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1238 	if (IS_ERR(xrcd->pd)) {
1239 		err = PTR_ERR(xrcd->pd);
1240 		goto err2;
1241 	}
1242 
1243 	cq_attr.cqe = 1;
1244 	xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1245 	if (IS_ERR(xrcd->cq)) {
1246 		err = PTR_ERR(xrcd->cq);
1247 		goto err3;
1248 	}
1249 
1250 	return 0;
1251 
1252 err3:
1253 	ib_dealloc_pd(xrcd->pd);
1254 err2:
1255 	mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1256 	return err;
1257 }
1258 
1259 static void mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1260 {
1261 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1262 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1263 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1264 }
1265 
1266 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1267 {
1268 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1269 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1270 	struct mlx4_ib_gid_entry *ge;
1271 
1272 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1273 	if (!ge)
1274 		return -ENOMEM;
1275 
1276 	ge->gid = *gid;
1277 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1278 		ge->port = mqp->port;
1279 		ge->added = 1;
1280 	}
1281 
1282 	mutex_lock(&mqp->mutex);
1283 	list_add_tail(&ge->list, &mqp->gid_list);
1284 	mutex_unlock(&mqp->mutex);
1285 
1286 	return 0;
1287 }
1288 
1289 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1290 					  struct mlx4_ib_counters *ctr_table)
1291 {
1292 	struct counter_index *counter, *tmp_count;
1293 
1294 	mutex_lock(&ctr_table->mutex);
1295 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1296 				 list) {
1297 		if (counter->allocated)
1298 			mlx4_counter_free(ibdev->dev, counter->index);
1299 		list_del(&counter->list);
1300 		kfree(counter);
1301 	}
1302 	mutex_unlock(&ctr_table->mutex);
1303 }
1304 
1305 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1306 		   union ib_gid *gid)
1307 {
1308 	struct net_device *ndev;
1309 	int ret = 0;
1310 
1311 	if (!mqp->port)
1312 		return 0;
1313 
1314 	spin_lock_bh(&mdev->iboe.lock);
1315 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1316 	if (ndev)
1317 		dev_hold(ndev);
1318 	spin_unlock_bh(&mdev->iboe.lock);
1319 
1320 	if (ndev) {
1321 		ret = 1;
1322 		dev_put(ndev);
1323 	}
1324 
1325 	return ret;
1326 }
1327 
1328 struct mlx4_ib_steering {
1329 	struct list_head list;
1330 	struct mlx4_flow_reg_id reg_id;
1331 	union ib_gid gid;
1332 };
1333 
1334 #define LAST_ETH_FIELD vlan_tag
1335 #define LAST_IB_FIELD sl
1336 #define LAST_IPV4_FIELD dst_ip
1337 #define LAST_TCP_UDP_FIELD src_port
1338 
1339 /* Field is the last supported field */
1340 #define FIELDS_NOT_SUPPORTED(filter, field)\
1341 	memchr_inv((void *)&filter.field  +\
1342 		   sizeof(filter.field), 0,\
1343 		   sizeof(filter) -\
1344 		   offsetof(typeof(filter), field) -\
1345 		   sizeof(filter.field))
1346 
1347 static int parse_flow_attr(struct mlx4_dev *dev,
1348 			   u32 qp_num,
1349 			   union ib_flow_spec *ib_spec,
1350 			   struct _rule_hw *mlx4_spec)
1351 {
1352 	enum mlx4_net_trans_rule_id type;
1353 
1354 	switch (ib_spec->type) {
1355 	case IB_FLOW_SPEC_ETH:
1356 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1357 			return -ENOTSUPP;
1358 
1359 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1360 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1361 		       ETH_ALEN);
1362 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1363 		       ETH_ALEN);
1364 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1365 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1366 		break;
1367 	case IB_FLOW_SPEC_IB:
1368 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1369 			return -ENOTSUPP;
1370 
1371 		type = MLX4_NET_TRANS_RULE_ID_IB;
1372 		mlx4_spec->ib.l3_qpn =
1373 			cpu_to_be32(qp_num);
1374 		mlx4_spec->ib.qpn_mask =
1375 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1376 		break;
1377 
1378 
1379 	case IB_FLOW_SPEC_IPV4:
1380 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1381 			return -ENOTSUPP;
1382 
1383 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1384 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1385 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1386 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1387 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1388 		break;
1389 
1390 	case IB_FLOW_SPEC_TCP:
1391 	case IB_FLOW_SPEC_UDP:
1392 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1393 			return -ENOTSUPP;
1394 
1395 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1396 					MLX4_NET_TRANS_RULE_ID_TCP :
1397 					MLX4_NET_TRANS_RULE_ID_UDP;
1398 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1399 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1400 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1401 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1402 		break;
1403 
1404 	default:
1405 		return -EINVAL;
1406 	}
1407 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1408 	    mlx4_hw_rule_sz(dev, type) < 0)
1409 		return -EINVAL;
1410 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1411 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1412 	return mlx4_hw_rule_sz(dev, type);
1413 }
1414 
1415 struct default_rules {
1416 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1417 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1418 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1419 	__u8  link_layer;
1420 };
1421 static const struct default_rules default_table[] = {
1422 	{
1423 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1424 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1425 		.rules_create_list = {IB_FLOW_SPEC_IB},
1426 		.link_layer = IB_LINK_LAYER_INFINIBAND
1427 	}
1428 };
1429 
1430 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1431 					 struct ib_flow_attr *flow_attr)
1432 {
1433 	int i, j, k;
1434 	void *ib_flow;
1435 	const struct default_rules *pdefault_rules = default_table;
1436 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1437 
1438 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1439 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1440 		memset(&field_types, 0, sizeof(field_types));
1441 
1442 		if (link_layer != pdefault_rules->link_layer)
1443 			continue;
1444 
1445 		ib_flow = flow_attr + 1;
1446 		/* we assume the specs are sorted */
1447 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1448 		     j < flow_attr->num_of_specs; k++) {
1449 			union ib_flow_spec *current_flow =
1450 				(union ib_flow_spec *)ib_flow;
1451 
1452 			/* same layer but different type */
1453 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1454 			     (pdefault_rules->mandatory_fields[k] &
1455 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1456 			    (current_flow->type !=
1457 			     pdefault_rules->mandatory_fields[k]))
1458 				goto out;
1459 
1460 			/* same layer, try match next one */
1461 			if (current_flow->type ==
1462 			    pdefault_rules->mandatory_fields[k]) {
1463 				j++;
1464 				ib_flow +=
1465 					((union ib_flow_spec *)ib_flow)->size;
1466 			}
1467 		}
1468 
1469 		ib_flow = flow_attr + 1;
1470 		for (j = 0; j < flow_attr->num_of_specs;
1471 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1472 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1473 				/* same layer and same type */
1474 				if (((union ib_flow_spec *)ib_flow)->type ==
1475 				    pdefault_rules->mandatory_not_fields[k])
1476 					goto out;
1477 
1478 		return i;
1479 	}
1480 out:
1481 	return -1;
1482 }
1483 
1484 static int __mlx4_ib_create_default_rules(
1485 		struct mlx4_ib_dev *mdev,
1486 		struct ib_qp *qp,
1487 		const struct default_rules *pdefault_rules,
1488 		struct _rule_hw *mlx4_spec) {
1489 	int size = 0;
1490 	int i;
1491 
1492 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1493 		union ib_flow_spec ib_spec = {};
1494 		int ret;
1495 
1496 		switch (pdefault_rules->rules_create_list[i]) {
1497 		case 0:
1498 			/* no rule */
1499 			continue;
1500 		case IB_FLOW_SPEC_IB:
1501 			ib_spec.type = IB_FLOW_SPEC_IB;
1502 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1503 
1504 			break;
1505 		default:
1506 			/* invalid rule */
1507 			return -EINVAL;
1508 		}
1509 		/* We must put empty rule, qpn is being ignored */
1510 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1511 				      mlx4_spec);
1512 		if (ret < 0) {
1513 			pr_info("invalid parsing\n");
1514 			return -EINVAL;
1515 		}
1516 
1517 		mlx4_spec = (void *)mlx4_spec + ret;
1518 		size += ret;
1519 	}
1520 	return size;
1521 }
1522 
1523 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1524 			  int domain,
1525 			  enum mlx4_net_trans_promisc_mode flow_type,
1526 			  u64 *reg_id)
1527 {
1528 	int ret, i;
1529 	int size = 0;
1530 	void *ib_flow;
1531 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1532 	struct mlx4_cmd_mailbox *mailbox;
1533 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1534 	int default_flow;
1535 
1536 	static const u16 __mlx4_domain[] = {
1537 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1538 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1539 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1540 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1541 	};
1542 
1543 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1544 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1545 		return -EINVAL;
1546 	}
1547 
1548 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1549 		pr_err("Invalid domain value %d\n", domain);
1550 		return -EINVAL;
1551 	}
1552 
1553 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1554 		return -EINVAL;
1555 
1556 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1557 	if (IS_ERR(mailbox))
1558 		return PTR_ERR(mailbox);
1559 	ctrl = mailbox->buf;
1560 
1561 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1562 				 flow_attr->priority);
1563 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1564 	ctrl->port = flow_attr->port;
1565 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1566 
1567 	ib_flow = flow_attr + 1;
1568 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1569 	/* Add default flows */
1570 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1571 	if (default_flow >= 0) {
1572 		ret = __mlx4_ib_create_default_rules(
1573 				mdev, qp, default_table + default_flow,
1574 				mailbox->buf + size);
1575 		if (ret < 0) {
1576 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1577 			return -EINVAL;
1578 		}
1579 		size += ret;
1580 	}
1581 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1582 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1583 				      mailbox->buf + size);
1584 		if (ret < 0) {
1585 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1586 			return -EINVAL;
1587 		}
1588 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1589 		size += ret;
1590 	}
1591 
1592 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1593 	    flow_attr->num_of_specs == 1) {
1594 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1595 		enum ib_flow_spec_type header_spec =
1596 			((union ib_flow_spec *)(flow_attr + 1))->type;
1597 
1598 		if (header_spec == IB_FLOW_SPEC_ETH)
1599 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1600 	}
1601 
1602 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1603 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1604 			   MLX4_CMD_NATIVE);
1605 	if (ret == -ENOMEM)
1606 		pr_err("mcg table is full. Fail to register network rule.\n");
1607 	else if (ret == -ENXIO)
1608 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1609 	else if (ret)
1610 		pr_err("Invalid argument. Fail to register network rule.\n");
1611 
1612 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1613 	return ret;
1614 }
1615 
1616 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1617 {
1618 	int err;
1619 	err = mlx4_cmd(dev, reg_id, 0, 0,
1620 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1621 		       MLX4_CMD_NATIVE);
1622 	if (err)
1623 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1624 		       reg_id);
1625 	return err;
1626 }
1627 
1628 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1629 				    u64 *reg_id)
1630 {
1631 	void *ib_flow;
1632 	union ib_flow_spec *ib_spec;
1633 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1634 	int err = 0;
1635 
1636 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1637 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1638 		return 0; /* do nothing */
1639 
1640 	ib_flow = flow_attr + 1;
1641 	ib_spec = (union ib_flow_spec *)ib_flow;
1642 
1643 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1644 		return 0; /* do nothing */
1645 
1646 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1647 				    flow_attr->port, qp->qp_num,
1648 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1649 				    reg_id);
1650 	return err;
1651 }
1652 
1653 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1654 				      struct ib_flow_attr *flow_attr,
1655 				      enum mlx4_net_trans_promisc_mode *type)
1656 {
1657 	int err = 0;
1658 
1659 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1660 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1661 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1662 		return -EOPNOTSUPP;
1663 	}
1664 
1665 	if (flow_attr->num_of_specs == 0) {
1666 		type[0] = MLX4_FS_MC_SNIFFER;
1667 		type[1] = MLX4_FS_UC_SNIFFER;
1668 	} else {
1669 		union ib_flow_spec *ib_spec;
1670 
1671 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1672 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1673 			return -EINVAL;
1674 
1675 		/* if all is zero than MC and UC */
1676 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1677 			type[0] = MLX4_FS_MC_SNIFFER;
1678 			type[1] = MLX4_FS_UC_SNIFFER;
1679 		} else {
1680 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1681 					    ib_spec->eth.mask.dst_mac[1],
1682 					    ib_spec->eth.mask.dst_mac[2],
1683 					    ib_spec->eth.mask.dst_mac[3],
1684 					    ib_spec->eth.mask.dst_mac[4],
1685 					    ib_spec->eth.mask.dst_mac[5]};
1686 
1687 			/* Above xor was only on MC bit, non empty mask is valid
1688 			 * only if this bit is set and rest are zero.
1689 			 */
1690 			if (!is_zero_ether_addr(&mac[0]))
1691 				return -EINVAL;
1692 
1693 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1694 				type[0] = MLX4_FS_MC_SNIFFER;
1695 			else
1696 				type[0] = MLX4_FS_UC_SNIFFER;
1697 		}
1698 	}
1699 
1700 	return err;
1701 }
1702 
1703 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1704 				    struct ib_flow_attr *flow_attr,
1705 				    int domain, struct ib_udata *udata)
1706 {
1707 	int err = 0, i = 0, j = 0;
1708 	struct mlx4_ib_flow *mflow;
1709 	enum mlx4_net_trans_promisc_mode type[2];
1710 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1711 	int is_bonded = mlx4_is_bonded(dev);
1712 
1713 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1714 		return ERR_PTR(-EINVAL);
1715 
1716 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1717 		return ERR_PTR(-EOPNOTSUPP);
1718 
1719 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1720 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1721 		return ERR_PTR(-EOPNOTSUPP);
1722 
1723 	if (udata &&
1724 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1725 		return ERR_PTR(-EOPNOTSUPP);
1726 
1727 	memset(type, 0, sizeof(type));
1728 
1729 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1730 	if (!mflow) {
1731 		err = -ENOMEM;
1732 		goto err_free;
1733 	}
1734 
1735 	switch (flow_attr->type) {
1736 	case IB_FLOW_ATTR_NORMAL:
1737 		/* If dont trap flag (continue match) is set, under specific
1738 		 * condition traffic be replicated to given qp,
1739 		 * without stealing it
1740 		 */
1741 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1742 			err = mlx4_ib_add_dont_trap_rule(dev,
1743 							 flow_attr,
1744 							 type);
1745 			if (err)
1746 				goto err_free;
1747 		} else {
1748 			type[0] = MLX4_FS_REGULAR;
1749 		}
1750 		break;
1751 
1752 	case IB_FLOW_ATTR_ALL_DEFAULT:
1753 		type[0] = MLX4_FS_ALL_DEFAULT;
1754 		break;
1755 
1756 	case IB_FLOW_ATTR_MC_DEFAULT:
1757 		type[0] = MLX4_FS_MC_DEFAULT;
1758 		break;
1759 
1760 	case IB_FLOW_ATTR_SNIFFER:
1761 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1762 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1763 		break;
1764 
1765 	default:
1766 		err = -EINVAL;
1767 		goto err_free;
1768 	}
1769 
1770 	while (i < ARRAY_SIZE(type) && type[i]) {
1771 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1772 					    &mflow->reg_id[i].id);
1773 		if (err)
1774 			goto err_create_flow;
1775 		if (is_bonded) {
1776 			/* Application always sees one port so the mirror rule
1777 			 * must be on port #2
1778 			 */
1779 			flow_attr->port = 2;
1780 			err = __mlx4_ib_create_flow(qp, flow_attr,
1781 						    domain, type[j],
1782 						    &mflow->reg_id[j].mirror);
1783 			flow_attr->port = 1;
1784 			if (err)
1785 				goto err_create_flow;
1786 			j++;
1787 		}
1788 
1789 		i++;
1790 	}
1791 
1792 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1793 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1794 					       &mflow->reg_id[i].id);
1795 		if (err)
1796 			goto err_create_flow;
1797 
1798 		if (is_bonded) {
1799 			flow_attr->port = 2;
1800 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1801 						       &mflow->reg_id[j].mirror);
1802 			flow_attr->port = 1;
1803 			if (err)
1804 				goto err_create_flow;
1805 			j++;
1806 		}
1807 		/* function to create mirror rule */
1808 		i++;
1809 	}
1810 
1811 	return &mflow->ibflow;
1812 
1813 err_create_flow:
1814 	while (i) {
1815 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1816 					     mflow->reg_id[i].id);
1817 		i--;
1818 	}
1819 
1820 	while (j) {
1821 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1822 					     mflow->reg_id[j].mirror);
1823 		j--;
1824 	}
1825 err_free:
1826 	kfree(mflow);
1827 	return ERR_PTR(err);
1828 }
1829 
1830 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1831 {
1832 	int err, ret = 0;
1833 	int i = 0;
1834 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1835 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1836 
1837 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1838 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1839 		if (err)
1840 			ret = err;
1841 		if (mflow->reg_id[i].mirror) {
1842 			err = __mlx4_ib_destroy_flow(mdev->dev,
1843 						     mflow->reg_id[i].mirror);
1844 			if (err)
1845 				ret = err;
1846 		}
1847 		i++;
1848 	}
1849 
1850 	kfree(mflow);
1851 	return ret;
1852 }
1853 
1854 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1855 {
1856 	int err;
1857 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1858 	struct mlx4_dev	*dev = mdev->dev;
1859 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1860 	struct mlx4_ib_steering *ib_steering = NULL;
1861 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1862 	struct mlx4_flow_reg_id	reg_id;
1863 
1864 	if (mdev->dev->caps.steering_mode ==
1865 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1866 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1867 		if (!ib_steering)
1868 			return -ENOMEM;
1869 	}
1870 
1871 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1872 				    !!(mqp->flags &
1873 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1874 				    prot, &reg_id.id);
1875 	if (err) {
1876 		pr_err("multicast attach op failed, err %d\n", err);
1877 		goto err_malloc;
1878 	}
1879 
1880 	reg_id.mirror = 0;
1881 	if (mlx4_is_bonded(dev)) {
1882 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1883 					    (mqp->port == 1) ? 2 : 1,
1884 					    !!(mqp->flags &
1885 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1886 					    prot, &reg_id.mirror);
1887 		if (err)
1888 			goto err_add;
1889 	}
1890 
1891 	err = add_gid_entry(ibqp, gid);
1892 	if (err)
1893 		goto err_add;
1894 
1895 	if (ib_steering) {
1896 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1897 		ib_steering->reg_id = reg_id;
1898 		mutex_lock(&mqp->mutex);
1899 		list_add(&ib_steering->list, &mqp->steering_rules);
1900 		mutex_unlock(&mqp->mutex);
1901 	}
1902 	return 0;
1903 
1904 err_add:
1905 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1906 			      prot, reg_id.id);
1907 	if (reg_id.mirror)
1908 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1909 				      prot, reg_id.mirror);
1910 err_malloc:
1911 	kfree(ib_steering);
1912 
1913 	return err;
1914 }
1915 
1916 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1917 {
1918 	struct mlx4_ib_gid_entry *ge;
1919 	struct mlx4_ib_gid_entry *tmp;
1920 	struct mlx4_ib_gid_entry *ret = NULL;
1921 
1922 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1923 		if (!memcmp(raw, ge->gid.raw, 16)) {
1924 			ret = ge;
1925 			break;
1926 		}
1927 	}
1928 
1929 	return ret;
1930 }
1931 
1932 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1933 {
1934 	int err;
1935 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1936 	struct mlx4_dev *dev = mdev->dev;
1937 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1938 	struct net_device *ndev;
1939 	struct mlx4_ib_gid_entry *ge;
1940 	struct mlx4_flow_reg_id reg_id = {0, 0};
1941 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1942 
1943 	if (mdev->dev->caps.steering_mode ==
1944 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1945 		struct mlx4_ib_steering *ib_steering;
1946 
1947 		mutex_lock(&mqp->mutex);
1948 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1949 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1950 				list_del(&ib_steering->list);
1951 				break;
1952 			}
1953 		}
1954 		mutex_unlock(&mqp->mutex);
1955 		if (&ib_steering->list == &mqp->steering_rules) {
1956 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1957 			return -EINVAL;
1958 		}
1959 		reg_id = ib_steering->reg_id;
1960 		kfree(ib_steering);
1961 	}
1962 
1963 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1964 				    prot, reg_id.id);
1965 	if (err)
1966 		return err;
1967 
1968 	if (mlx4_is_bonded(dev)) {
1969 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1970 					    prot, reg_id.mirror);
1971 		if (err)
1972 			return err;
1973 	}
1974 
1975 	mutex_lock(&mqp->mutex);
1976 	ge = find_gid_entry(mqp, gid->raw);
1977 	if (ge) {
1978 		spin_lock_bh(&mdev->iboe.lock);
1979 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1980 		if (ndev)
1981 			dev_hold(ndev);
1982 		spin_unlock_bh(&mdev->iboe.lock);
1983 		if (ndev)
1984 			dev_put(ndev);
1985 		list_del(&ge->list);
1986 		kfree(ge);
1987 	} else
1988 		pr_warn("could not find mgid entry\n");
1989 
1990 	mutex_unlock(&mqp->mutex);
1991 
1992 	return 0;
1993 }
1994 
1995 static int init_node_data(struct mlx4_ib_dev *dev)
1996 {
1997 	struct ib_smp *in_mad  = NULL;
1998 	struct ib_smp *out_mad = NULL;
1999 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2000 	int err = -ENOMEM;
2001 
2002 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2003 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2004 	if (!in_mad || !out_mad)
2005 		goto out;
2006 
2007 	init_query_mad(in_mad);
2008 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2009 	if (mlx4_is_master(dev->dev))
2010 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2011 
2012 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2013 	if (err)
2014 		goto out;
2015 
2016 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2017 
2018 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2019 
2020 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2021 	if (err)
2022 		goto out;
2023 
2024 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2025 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2026 
2027 out:
2028 	kfree(in_mad);
2029 	kfree(out_mad);
2030 	return err;
2031 }
2032 
2033 static ssize_t hca_type_show(struct device *device,
2034 			     struct device_attribute *attr, char *buf)
2035 {
2036 	struct mlx4_ib_dev *dev =
2037 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2038 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2039 }
2040 static DEVICE_ATTR_RO(hca_type);
2041 
2042 static ssize_t hw_rev_show(struct device *device,
2043 			   struct device_attribute *attr, char *buf)
2044 {
2045 	struct mlx4_ib_dev *dev =
2046 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2047 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2048 }
2049 static DEVICE_ATTR_RO(hw_rev);
2050 
2051 static ssize_t board_id_show(struct device *device,
2052 			     struct device_attribute *attr, char *buf)
2053 {
2054 	struct mlx4_ib_dev *dev =
2055 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2056 
2057 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2058 		       dev->dev->board_id);
2059 }
2060 static DEVICE_ATTR_RO(board_id);
2061 
2062 static struct attribute *mlx4_class_attributes[] = {
2063 	&dev_attr_hw_rev.attr,
2064 	&dev_attr_hca_type.attr,
2065 	&dev_attr_board_id.attr,
2066 	NULL
2067 };
2068 
2069 static const struct attribute_group mlx4_attr_group = {
2070 	.attrs = mlx4_class_attributes,
2071 };
2072 
2073 struct diag_counter {
2074 	const char *name;
2075 	u32 offset;
2076 };
2077 
2078 #define DIAG_COUNTER(_name, _offset)			\
2079 	{ .name = #_name, .offset = _offset }
2080 
2081 static const struct diag_counter diag_basic[] = {
2082 	DIAG_COUNTER(rq_num_lle, 0x00),
2083 	DIAG_COUNTER(sq_num_lle, 0x04),
2084 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2085 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2086 	DIAG_COUNTER(rq_num_lpe, 0x18),
2087 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2088 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2089 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2090 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2091 	DIAG_COUNTER(sq_num_bre, 0x34),
2092 	DIAG_COUNTER(sq_num_rire, 0x44),
2093 	DIAG_COUNTER(rq_num_rire, 0x48),
2094 	DIAG_COUNTER(sq_num_rae, 0x4C),
2095 	DIAG_COUNTER(rq_num_rae, 0x50),
2096 	DIAG_COUNTER(sq_num_roe, 0x54),
2097 	DIAG_COUNTER(sq_num_tree, 0x5C),
2098 	DIAG_COUNTER(sq_num_rree, 0x64),
2099 	DIAG_COUNTER(rq_num_rnr, 0x68),
2100 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2101 	DIAG_COUNTER(rq_num_oos, 0x100),
2102 	DIAG_COUNTER(sq_num_oos, 0x104),
2103 };
2104 
2105 static const struct diag_counter diag_ext[] = {
2106 	DIAG_COUNTER(rq_num_dup, 0x130),
2107 	DIAG_COUNTER(sq_num_to, 0x134),
2108 };
2109 
2110 static const struct diag_counter diag_device_only[] = {
2111 	DIAG_COUNTER(num_cqovf, 0x1A0),
2112 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2113 };
2114 
2115 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2116 						    u8 port_num)
2117 {
2118 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2119 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2120 
2121 	if (!diag[!!port_num].name)
2122 		return NULL;
2123 
2124 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2125 					  diag[!!port_num].num_counters,
2126 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2127 }
2128 
2129 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2130 				struct rdma_hw_stats *stats,
2131 				u8 port, int index)
2132 {
2133 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2134 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2135 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2136 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2137 	int ret;
2138 	int i;
2139 
2140 	ret = mlx4_query_diag_counters(dev->dev,
2141 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2142 				       diag[!!port].offset, hw_value,
2143 				       diag[!!port].num_counters, port);
2144 
2145 	if (ret)
2146 		return ret;
2147 
2148 	for (i = 0; i < diag[!!port].num_counters; i++)
2149 		stats->value[i] = hw_value[i];
2150 
2151 	return diag[!!port].num_counters;
2152 }
2153 
2154 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2155 					 const char ***name,
2156 					 u32 **offset,
2157 					 u32 *num,
2158 					 bool port)
2159 {
2160 	u32 num_counters;
2161 
2162 	num_counters = ARRAY_SIZE(diag_basic);
2163 
2164 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2165 		num_counters += ARRAY_SIZE(diag_ext);
2166 
2167 	if (!port)
2168 		num_counters += ARRAY_SIZE(diag_device_only);
2169 
2170 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2171 	if (!*name)
2172 		return -ENOMEM;
2173 
2174 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2175 	if (!*offset)
2176 		goto err_name;
2177 
2178 	*num = num_counters;
2179 
2180 	return 0;
2181 
2182 err_name:
2183 	kfree(*name);
2184 	return -ENOMEM;
2185 }
2186 
2187 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2188 				       const char **name,
2189 				       u32 *offset,
2190 				       bool port)
2191 {
2192 	int i;
2193 	int j;
2194 
2195 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2196 		name[i] = diag_basic[i].name;
2197 		offset[i] = diag_basic[i].offset;
2198 	}
2199 
2200 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2201 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2202 			name[j] = diag_ext[i].name;
2203 			offset[j] = diag_ext[i].offset;
2204 		}
2205 	}
2206 
2207 	if (!port) {
2208 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2209 			name[j] = diag_device_only[i].name;
2210 			offset[j] = diag_device_only[i].offset;
2211 		}
2212 	}
2213 }
2214 
2215 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2216 	.alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2217 	.get_hw_stats = mlx4_ib_get_hw_stats,
2218 };
2219 
2220 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2221 {
2222 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2223 	int i;
2224 	int ret;
2225 	bool per_port = !!(ibdev->dev->caps.flags2 &
2226 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2227 
2228 	if (mlx4_is_slave(ibdev->dev))
2229 		return 0;
2230 
2231 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2232 		/* i == 1 means we are building port counters */
2233 		if (i && !per_port)
2234 			continue;
2235 
2236 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2237 						    &diag[i].offset,
2238 						    &diag[i].num_counters, i);
2239 		if (ret)
2240 			goto err_alloc;
2241 
2242 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2243 					   diag[i].offset, i);
2244 	}
2245 
2246 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2247 
2248 	return 0;
2249 
2250 err_alloc:
2251 	if (i) {
2252 		kfree(diag[i - 1].name);
2253 		kfree(diag[i - 1].offset);
2254 	}
2255 
2256 	return ret;
2257 }
2258 
2259 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2260 {
2261 	int i;
2262 
2263 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2264 		kfree(ibdev->diag_counters[i].offset);
2265 		kfree(ibdev->diag_counters[i].name);
2266 	}
2267 }
2268 
2269 #define MLX4_IB_INVALID_MAC	((u64)-1)
2270 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2271 			       struct net_device *dev,
2272 			       int port)
2273 {
2274 	u64 new_smac = 0;
2275 	u64 release_mac = MLX4_IB_INVALID_MAC;
2276 	struct mlx4_ib_qp *qp;
2277 
2278 	read_lock(&dev_base_lock);
2279 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2280 	read_unlock(&dev_base_lock);
2281 
2282 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2283 
2284 	/* no need for update QP1 and mac registration in non-SRIOV */
2285 	if (!mlx4_is_mfunc(ibdev->dev))
2286 		return;
2287 
2288 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2289 	qp = ibdev->qp1_proxy[port - 1];
2290 	if (qp) {
2291 		int new_smac_index;
2292 		u64 old_smac;
2293 		struct mlx4_update_qp_params update_params;
2294 
2295 		mutex_lock(&qp->mutex);
2296 		old_smac = qp->pri.smac;
2297 		if (new_smac == old_smac)
2298 			goto unlock;
2299 
2300 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2301 
2302 		if (new_smac_index < 0)
2303 			goto unlock;
2304 
2305 		update_params.smac_index = new_smac_index;
2306 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2307 				   &update_params)) {
2308 			release_mac = new_smac;
2309 			goto unlock;
2310 		}
2311 		/* if old port was zero, no mac was yet registered for this QP */
2312 		if (qp->pri.smac_port)
2313 			release_mac = old_smac;
2314 		qp->pri.smac = new_smac;
2315 		qp->pri.smac_port = port;
2316 		qp->pri.smac_index = new_smac_index;
2317 	}
2318 
2319 unlock:
2320 	if (release_mac != MLX4_IB_INVALID_MAC)
2321 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2322 	if (qp)
2323 		mutex_unlock(&qp->mutex);
2324 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2325 }
2326 
2327 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2328 				 struct net_device *dev,
2329 				 unsigned long event)
2330 
2331 {
2332 	struct mlx4_ib_iboe *iboe;
2333 	int update_qps_port = -1;
2334 	int port;
2335 
2336 	ASSERT_RTNL();
2337 
2338 	iboe = &ibdev->iboe;
2339 
2340 	spin_lock_bh(&iboe->lock);
2341 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2342 
2343 		iboe->netdevs[port - 1] =
2344 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2345 
2346 		if (dev == iboe->netdevs[port - 1] &&
2347 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2348 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2349 			update_qps_port = port;
2350 
2351 		if (dev == iboe->netdevs[port - 1] &&
2352 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2353 			enum ib_port_state port_state;
2354 			struct ib_event ibev = { };
2355 
2356 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2357 						     &port_state))
2358 				continue;
2359 
2360 			if (event == NETDEV_UP &&
2361 			    (port_state != IB_PORT_ACTIVE ||
2362 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2363 				continue;
2364 			if (event == NETDEV_DOWN &&
2365 			    (port_state != IB_PORT_DOWN ||
2366 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2367 				continue;
2368 			iboe->last_port_state[port - 1] = port_state;
2369 
2370 			ibev.device = &ibdev->ib_dev;
2371 			ibev.element.port_num = port;
2372 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2373 							  IB_EVENT_PORT_ERR;
2374 			ib_dispatch_event(&ibev);
2375 		}
2376 
2377 	}
2378 	spin_unlock_bh(&iboe->lock);
2379 
2380 	if (update_qps_port > 0)
2381 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2382 }
2383 
2384 static int mlx4_ib_netdev_event(struct notifier_block *this,
2385 				unsigned long event, void *ptr)
2386 {
2387 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2388 	struct mlx4_ib_dev *ibdev;
2389 
2390 	if (!net_eq(dev_net(dev), &init_net))
2391 		return NOTIFY_DONE;
2392 
2393 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2394 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2395 
2396 	return NOTIFY_DONE;
2397 }
2398 
2399 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2400 {
2401 	int port;
2402 	int slave;
2403 	int i;
2404 
2405 	if (mlx4_is_master(ibdev->dev)) {
2406 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2407 		     ++slave) {
2408 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2409 				for (i = 0;
2410 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2411 				     ++i) {
2412 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2413 					/* master has the identity virt2phys pkey mapping */
2414 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2415 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2416 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2417 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2418 				}
2419 			}
2420 		}
2421 		/* initialize pkey cache */
2422 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2423 			for (i = 0;
2424 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2425 			     ++i)
2426 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2427 					(i) ? 0 : 0xFFFF;
2428 		}
2429 	}
2430 }
2431 
2432 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2433 {
2434 	int i, j, eq = 0, total_eqs = 0;
2435 
2436 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2437 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2438 	if (!ibdev->eq_table)
2439 		return;
2440 
2441 	for (i = 1; i <= dev->caps.num_ports; i++) {
2442 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2443 		     j++, total_eqs++) {
2444 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2445 				continue;
2446 			ibdev->eq_table[eq] = total_eqs;
2447 			if (!mlx4_assign_eq(dev, i,
2448 					    &ibdev->eq_table[eq]))
2449 				eq++;
2450 			else
2451 				ibdev->eq_table[eq] = -1;
2452 		}
2453 	}
2454 
2455 	for (i = eq; i < dev->caps.num_comp_vectors;
2456 	     ibdev->eq_table[i++] = -1)
2457 		;
2458 
2459 	/* Advertise the new number of EQs to clients */
2460 	ibdev->ib_dev.num_comp_vectors = eq;
2461 }
2462 
2463 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2464 {
2465 	int i;
2466 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2467 
2468 	/* no eqs were allocated */
2469 	if (!ibdev->eq_table)
2470 		return;
2471 
2472 	/* Reset the advertised EQ number */
2473 	ibdev->ib_dev.num_comp_vectors = 0;
2474 
2475 	for (i = 0; i < total_eqs; i++)
2476 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2477 
2478 	kfree(ibdev->eq_table);
2479 	ibdev->eq_table = NULL;
2480 }
2481 
2482 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2483 			       struct ib_port_immutable *immutable)
2484 {
2485 	struct ib_port_attr attr;
2486 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2487 	int err;
2488 
2489 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2490 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2491 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2492 	} else {
2493 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2494 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2495 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2496 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2497 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2498 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2499 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2500 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2501 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2502 	}
2503 
2504 	err = ib_query_port(ibdev, port_num, &attr);
2505 	if (err)
2506 		return err;
2507 
2508 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2509 	immutable->gid_tbl_len = attr.gid_tbl_len;
2510 
2511 	return 0;
2512 }
2513 
2514 static void get_fw_ver_str(struct ib_device *device, char *str)
2515 {
2516 	struct mlx4_ib_dev *dev =
2517 		container_of(device, struct mlx4_ib_dev, ib_dev);
2518 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2519 		 (int) (dev->dev->caps.fw_ver >> 32),
2520 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2521 		 (int) dev->dev->caps.fw_ver & 0xffff);
2522 }
2523 
2524 static const struct ib_device_ops mlx4_ib_dev_ops = {
2525 	.owner = THIS_MODULE,
2526 	.driver_id = RDMA_DRIVER_MLX4,
2527 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2528 
2529 	.add_gid = mlx4_ib_add_gid,
2530 	.alloc_mr = mlx4_ib_alloc_mr,
2531 	.alloc_pd = mlx4_ib_alloc_pd,
2532 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2533 	.attach_mcast = mlx4_ib_mcg_attach,
2534 	.create_ah = mlx4_ib_create_ah,
2535 	.create_cq = mlx4_ib_create_cq,
2536 	.create_qp = mlx4_ib_create_qp,
2537 	.create_srq = mlx4_ib_create_srq,
2538 	.dealloc_pd = mlx4_ib_dealloc_pd,
2539 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2540 	.del_gid = mlx4_ib_del_gid,
2541 	.dereg_mr = mlx4_ib_dereg_mr,
2542 	.destroy_ah = mlx4_ib_destroy_ah,
2543 	.destroy_cq = mlx4_ib_destroy_cq,
2544 	.destroy_qp = mlx4_ib_destroy_qp,
2545 	.destroy_srq = mlx4_ib_destroy_srq,
2546 	.detach_mcast = mlx4_ib_mcg_detach,
2547 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2548 	.drain_rq = mlx4_ib_drain_rq,
2549 	.drain_sq = mlx4_ib_drain_sq,
2550 	.get_dev_fw_str = get_fw_ver_str,
2551 	.get_dma_mr = mlx4_ib_get_dma_mr,
2552 	.get_link_layer = mlx4_ib_port_link_layer,
2553 	.get_netdev = mlx4_ib_get_netdev,
2554 	.get_port_immutable = mlx4_port_immutable,
2555 	.map_mr_sg = mlx4_ib_map_mr_sg,
2556 	.mmap = mlx4_ib_mmap,
2557 	.modify_cq = mlx4_ib_modify_cq,
2558 	.modify_device = mlx4_ib_modify_device,
2559 	.modify_port = mlx4_ib_modify_port,
2560 	.modify_qp = mlx4_ib_modify_qp,
2561 	.modify_srq = mlx4_ib_modify_srq,
2562 	.poll_cq = mlx4_ib_poll_cq,
2563 	.post_recv = mlx4_ib_post_recv,
2564 	.post_send = mlx4_ib_post_send,
2565 	.post_srq_recv = mlx4_ib_post_srq_recv,
2566 	.process_mad = mlx4_ib_process_mad,
2567 	.query_ah = mlx4_ib_query_ah,
2568 	.query_device = mlx4_ib_query_device,
2569 	.query_gid = mlx4_ib_query_gid,
2570 	.query_pkey = mlx4_ib_query_pkey,
2571 	.query_port = mlx4_ib_query_port,
2572 	.query_qp = mlx4_ib_query_qp,
2573 	.query_srq = mlx4_ib_query_srq,
2574 	.reg_user_mr = mlx4_ib_reg_user_mr,
2575 	.req_notify_cq = mlx4_ib_arm_cq,
2576 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2577 	.resize_cq = mlx4_ib_resize_cq,
2578 
2579 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2580 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2581 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2582 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2583 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2584 };
2585 
2586 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2587 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2588 	.create_wq = mlx4_ib_create_wq,
2589 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2590 	.destroy_wq = mlx4_ib_destroy_wq,
2591 	.modify_wq = mlx4_ib_modify_wq,
2592 };
2593 
2594 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2595 	.alloc_mw = mlx4_ib_alloc_mw,
2596 	.dealloc_mw = mlx4_ib_dealloc_mw,
2597 };
2598 
2599 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2600 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2601 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2602 
2603 	INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2604 };
2605 
2606 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2607 	.create_flow = mlx4_ib_create_flow,
2608 	.destroy_flow = mlx4_ib_destroy_flow,
2609 };
2610 
2611 static void *mlx4_ib_add(struct mlx4_dev *dev)
2612 {
2613 	struct mlx4_ib_dev *ibdev;
2614 	int num_ports = 0;
2615 	int i, j;
2616 	int err;
2617 	struct mlx4_ib_iboe *iboe;
2618 	int ib_num_ports = 0;
2619 	int num_req_counters;
2620 	int allocated;
2621 	u32 counter_index;
2622 	struct counter_index *new_counter_index = NULL;
2623 
2624 	pr_info_once("%s", mlx4_ib_version);
2625 
2626 	num_ports = 0;
2627 	mlx4_foreach_ib_transport_port(i, dev)
2628 		num_ports++;
2629 
2630 	/* No point in registering a device with no ports... */
2631 	if (num_ports == 0)
2632 		return NULL;
2633 
2634 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2635 	if (!ibdev) {
2636 		dev_err(&dev->persist->pdev->dev,
2637 			"Device struct alloc failed\n");
2638 		return NULL;
2639 	}
2640 
2641 	iboe = &ibdev->iboe;
2642 
2643 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2644 		goto err_dealloc;
2645 
2646 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2647 		goto err_pd;
2648 
2649 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2650 				 PAGE_SIZE);
2651 	if (!ibdev->uar_map)
2652 		goto err_uar;
2653 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2654 
2655 	ibdev->dev = dev;
2656 	ibdev->bond_next_port	= 0;
2657 
2658 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2659 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2660 	ibdev->num_ports		= num_ports;
2661 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2662 						1 : ibdev->num_ports;
2663 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2664 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2665 
2666 	ibdev->ib_dev.uverbs_cmd_mask	=
2667 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2668 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2669 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2670 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2671 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2672 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2673 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2674 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2675 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2676 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2677 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2678 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2679 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2680 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2681 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2682 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2683 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2684 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2685 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2686 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2687 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2688 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2689 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2690 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2691 
2692 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2693 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2694 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2695 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2696 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2697 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2698 
2699 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2700 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2701 	    IB_LINK_LAYER_ETHERNET) ||
2702 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2703 	    IB_LINK_LAYER_ETHERNET))) {
2704 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2705 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2706 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2707 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2708 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2709 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2710 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2711 	}
2712 
2713 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2714 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2715 		ibdev->ib_dev.uverbs_cmd_mask |=
2716 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2717 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2718 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2719 	}
2720 
2721 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2722 		ibdev->ib_dev.uverbs_cmd_mask |=
2723 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2724 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2725 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2726 	}
2727 
2728 	if (check_flow_steering_support(dev)) {
2729 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2730 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2731 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2732 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2733 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2734 	}
2735 
2736 	if (!dev->caps.userspace_caps)
2737 		ibdev->ib_dev.ops.uverbs_abi_ver =
2738 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2739 
2740 	mlx4_ib_alloc_eqs(dev, ibdev);
2741 
2742 	spin_lock_init(&iboe->lock);
2743 
2744 	if (init_node_data(ibdev))
2745 		goto err_map;
2746 	mlx4_init_sl2vl_tbl(ibdev);
2747 
2748 	for (i = 0; i < ibdev->num_ports; ++i) {
2749 		mutex_init(&ibdev->counters_table[i].mutex);
2750 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2751 		iboe->last_port_state[i] = IB_PORT_DOWN;
2752 	}
2753 
2754 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2755 	for (i = 0; i < num_req_counters; ++i) {
2756 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2757 		allocated = 0;
2758 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2759 						IB_LINK_LAYER_ETHERNET) {
2760 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2761 						 MLX4_RES_USAGE_DRIVER);
2762 			/* if failed to allocate a new counter, use default */
2763 			if (err)
2764 				counter_index =
2765 					mlx4_get_default_counter_index(dev,
2766 								       i + 1);
2767 			else
2768 				allocated = 1;
2769 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2770 			counter_index = mlx4_get_default_counter_index(dev,
2771 								       i + 1);
2772 		}
2773 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2774 					    GFP_KERNEL);
2775 		if (!new_counter_index) {
2776 			if (allocated)
2777 				mlx4_counter_free(ibdev->dev, counter_index);
2778 			goto err_counter;
2779 		}
2780 		new_counter_index->index = counter_index;
2781 		new_counter_index->allocated = allocated;
2782 		list_add_tail(&new_counter_index->list,
2783 			      &ibdev->counters_table[i].counters_list);
2784 		ibdev->counters_table[i].default_counter = counter_index;
2785 		pr_info("counter index %d for port %d allocated %d\n",
2786 			counter_index, i + 1, allocated);
2787 	}
2788 	if (mlx4_is_bonded(dev))
2789 		for (i = 1; i < ibdev->num_ports ; ++i) {
2790 			new_counter_index =
2791 					kmalloc(sizeof(struct counter_index),
2792 						GFP_KERNEL);
2793 			if (!new_counter_index)
2794 				goto err_counter;
2795 			new_counter_index->index = counter_index;
2796 			new_counter_index->allocated = 0;
2797 			list_add_tail(&new_counter_index->list,
2798 				      &ibdev->counters_table[i].counters_list);
2799 			ibdev->counters_table[i].default_counter =
2800 								counter_index;
2801 		}
2802 
2803 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2804 		ib_num_ports++;
2805 
2806 	spin_lock_init(&ibdev->sm_lock);
2807 	mutex_init(&ibdev->cap_mask_mutex);
2808 	INIT_LIST_HEAD(&ibdev->qp_list);
2809 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2810 
2811 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2812 	    ib_num_ports) {
2813 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2814 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2815 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2816 					    &ibdev->steer_qpn_base, 0,
2817 					    MLX4_RES_USAGE_DRIVER);
2818 		if (err)
2819 			goto err_counter;
2820 
2821 		ibdev->ib_uc_qpns_bitmap =
2822 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2823 				      sizeof(long),
2824 				      GFP_KERNEL);
2825 		if (!ibdev->ib_uc_qpns_bitmap)
2826 			goto err_steer_qp_release;
2827 
2828 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2829 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2830 				    ibdev->steer_qpn_count);
2831 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2832 					dev, ibdev->steer_qpn_base,
2833 					ibdev->steer_qpn_base +
2834 					ibdev->steer_qpn_count - 1);
2835 			if (err)
2836 				goto err_steer_free_bitmap;
2837 		} else {
2838 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2839 				    ibdev->steer_qpn_count);
2840 		}
2841 	}
2842 
2843 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2844 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2845 
2846 	if (mlx4_ib_alloc_diag_counters(ibdev))
2847 		goto err_steer_free_bitmap;
2848 
2849 	rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2850 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d"))
2851 		goto err_diag_counters;
2852 
2853 	if (mlx4_ib_mad_init(ibdev))
2854 		goto err_reg;
2855 
2856 	if (mlx4_ib_init_sriov(ibdev))
2857 		goto err_mad;
2858 
2859 	if (!iboe->nb.notifier_call) {
2860 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2861 		err = register_netdevice_notifier(&iboe->nb);
2862 		if (err) {
2863 			iboe->nb.notifier_call = NULL;
2864 			goto err_notif;
2865 		}
2866 	}
2867 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2868 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2869 		if (err)
2870 			goto err_notif;
2871 	}
2872 
2873 	ibdev->ib_active = true;
2874 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2875 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2876 					 &ibdev->ib_dev);
2877 
2878 	if (mlx4_is_mfunc(ibdev->dev))
2879 		init_pkeys(ibdev);
2880 
2881 	/* create paravirt contexts for any VFs which are active */
2882 	if (mlx4_is_master(ibdev->dev)) {
2883 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2884 			if (j == mlx4_master_func_num(ibdev->dev))
2885 				continue;
2886 			if (mlx4_is_slave_active(ibdev->dev, j))
2887 				do_slave_init(ibdev, j, 1);
2888 		}
2889 	}
2890 	return ibdev;
2891 
2892 err_notif:
2893 	if (ibdev->iboe.nb.notifier_call) {
2894 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2895 			pr_warn("failure unregistering notifier\n");
2896 		ibdev->iboe.nb.notifier_call = NULL;
2897 	}
2898 	flush_workqueue(wq);
2899 
2900 	mlx4_ib_close_sriov(ibdev);
2901 
2902 err_mad:
2903 	mlx4_ib_mad_cleanup(ibdev);
2904 
2905 err_reg:
2906 	ib_unregister_device(&ibdev->ib_dev);
2907 
2908 err_diag_counters:
2909 	mlx4_ib_diag_cleanup(ibdev);
2910 
2911 err_steer_free_bitmap:
2912 	kfree(ibdev->ib_uc_qpns_bitmap);
2913 
2914 err_steer_qp_release:
2915 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2916 			      ibdev->steer_qpn_count);
2917 err_counter:
2918 	for (i = 0; i < ibdev->num_ports; ++i)
2919 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2920 
2921 err_map:
2922 	mlx4_ib_free_eqs(dev, ibdev);
2923 	iounmap(ibdev->uar_map);
2924 
2925 err_uar:
2926 	mlx4_uar_free(dev, &ibdev->priv_uar);
2927 
2928 err_pd:
2929 	mlx4_pd_free(dev, ibdev->priv_pdn);
2930 
2931 err_dealloc:
2932 	ib_dealloc_device(&ibdev->ib_dev);
2933 
2934 	return NULL;
2935 }
2936 
2937 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2938 {
2939 	int offset;
2940 
2941 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2942 
2943 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2944 					 dev->steer_qpn_count,
2945 					 get_count_order(count));
2946 	if (offset < 0)
2947 		return offset;
2948 
2949 	*qpn = dev->steer_qpn_base + offset;
2950 	return 0;
2951 }
2952 
2953 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2954 {
2955 	if (!qpn ||
2956 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2957 		return;
2958 
2959 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2960 		 qpn, dev->steer_qpn_base))
2961 		/* not supposed to be here */
2962 		return;
2963 
2964 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2965 			      qpn - dev->steer_qpn_base,
2966 			      get_count_order(count));
2967 }
2968 
2969 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2970 			 int is_attach)
2971 {
2972 	int err;
2973 	size_t flow_size;
2974 	struct ib_flow_attr *flow = NULL;
2975 	struct ib_flow_spec_ib *ib_spec;
2976 
2977 	if (is_attach) {
2978 		flow_size = sizeof(struct ib_flow_attr) +
2979 			    sizeof(struct ib_flow_spec_ib);
2980 		flow = kzalloc(flow_size, GFP_KERNEL);
2981 		if (!flow)
2982 			return -ENOMEM;
2983 		flow->port = mqp->port;
2984 		flow->num_of_specs = 1;
2985 		flow->size = flow_size;
2986 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2987 		ib_spec->type = IB_FLOW_SPEC_IB;
2988 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2989 		/* Add an empty rule for IB L2 */
2990 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2991 
2992 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2993 					    IB_FLOW_DOMAIN_NIC,
2994 					    MLX4_FS_REGULAR,
2995 					    &mqp->reg_id);
2996 	} else {
2997 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2998 	}
2999 	kfree(flow);
3000 	return err;
3001 }
3002 
3003 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3004 {
3005 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3006 	int p;
3007 	int i;
3008 
3009 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3010 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3011 	ibdev->ib_active = false;
3012 	flush_workqueue(wq);
3013 
3014 	if (ibdev->iboe.nb.notifier_call) {
3015 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3016 			pr_warn("failure unregistering notifier\n");
3017 		ibdev->iboe.nb.notifier_call = NULL;
3018 	}
3019 
3020 	mlx4_ib_close_sriov(ibdev);
3021 	mlx4_ib_mad_cleanup(ibdev);
3022 	ib_unregister_device(&ibdev->ib_dev);
3023 	mlx4_ib_diag_cleanup(ibdev);
3024 
3025 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3026 			      ibdev->steer_qpn_count);
3027 	kfree(ibdev->ib_uc_qpns_bitmap);
3028 
3029 	iounmap(ibdev->uar_map);
3030 	for (p = 0; p < ibdev->num_ports; ++p)
3031 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3032 
3033 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3034 		mlx4_CLOSE_PORT(dev, p);
3035 
3036 	mlx4_ib_free_eqs(dev, ibdev);
3037 
3038 	mlx4_uar_free(dev, &ibdev->priv_uar);
3039 	mlx4_pd_free(dev, ibdev->priv_pdn);
3040 	ib_dealloc_device(&ibdev->ib_dev);
3041 }
3042 
3043 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3044 {
3045 	struct mlx4_ib_demux_work **dm = NULL;
3046 	struct mlx4_dev *dev = ibdev->dev;
3047 	int i;
3048 	unsigned long flags;
3049 	struct mlx4_active_ports actv_ports;
3050 	unsigned int ports;
3051 	unsigned int first_port;
3052 
3053 	if (!mlx4_is_master(dev))
3054 		return;
3055 
3056 	actv_ports = mlx4_get_active_ports(dev, slave);
3057 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3058 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3059 
3060 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3061 	if (!dm)
3062 		return;
3063 
3064 	for (i = 0; i < ports; i++) {
3065 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3066 		if (!dm[i]) {
3067 			while (--i >= 0)
3068 				kfree(dm[i]);
3069 			goto out;
3070 		}
3071 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3072 		dm[i]->port = first_port + i + 1;
3073 		dm[i]->slave = slave;
3074 		dm[i]->do_init = do_init;
3075 		dm[i]->dev = ibdev;
3076 	}
3077 	/* initialize or tear down tunnel QPs for the slave */
3078 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3079 	if (!ibdev->sriov.is_going_down) {
3080 		for (i = 0; i < ports; i++)
3081 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3082 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3083 	} else {
3084 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3085 		for (i = 0; i < ports; i++)
3086 			kfree(dm[i]);
3087 	}
3088 out:
3089 	kfree(dm);
3090 	return;
3091 }
3092 
3093 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3094 {
3095 	struct mlx4_ib_qp *mqp;
3096 	unsigned long flags_qp;
3097 	unsigned long flags_cq;
3098 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3099 	struct list_head    cq_notify_list;
3100 	struct mlx4_cq *mcq;
3101 	unsigned long flags;
3102 
3103 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3104 	INIT_LIST_HEAD(&cq_notify_list);
3105 
3106 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3107 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3108 
3109 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3110 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3111 		if (mqp->sq.tail != mqp->sq.head) {
3112 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3113 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3114 			if (send_mcq->mcq.comp &&
3115 			    mqp->ibqp.send_cq->comp_handler) {
3116 				if (!send_mcq->mcq.reset_notify_added) {
3117 					send_mcq->mcq.reset_notify_added = 1;
3118 					list_add_tail(&send_mcq->mcq.reset_notify,
3119 						      &cq_notify_list);
3120 				}
3121 			}
3122 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3123 		}
3124 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3125 		/* Now, handle the QP's receive queue */
3126 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3127 		/* no handling is needed for SRQ */
3128 		if (!mqp->ibqp.srq) {
3129 			if (mqp->rq.tail != mqp->rq.head) {
3130 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3131 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3132 				if (recv_mcq->mcq.comp &&
3133 				    mqp->ibqp.recv_cq->comp_handler) {
3134 					if (!recv_mcq->mcq.reset_notify_added) {
3135 						recv_mcq->mcq.reset_notify_added = 1;
3136 						list_add_tail(&recv_mcq->mcq.reset_notify,
3137 							      &cq_notify_list);
3138 					}
3139 				}
3140 				spin_unlock_irqrestore(&recv_mcq->lock,
3141 						       flags_cq);
3142 			}
3143 		}
3144 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3145 	}
3146 
3147 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3148 		mcq->comp(mcq);
3149 	}
3150 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3151 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3152 }
3153 
3154 static void handle_bonded_port_state_event(struct work_struct *work)
3155 {
3156 	struct ib_event_work *ew =
3157 		container_of(work, struct ib_event_work, work);
3158 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3159 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3160 	int i;
3161 	struct ib_event ibev;
3162 
3163 	kfree(ew);
3164 	spin_lock_bh(&ibdev->iboe.lock);
3165 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3166 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3167 		enum ib_port_state curr_port_state;
3168 
3169 		if (!curr_netdev)
3170 			continue;
3171 
3172 		curr_port_state =
3173 			(netif_running(curr_netdev) &&
3174 			 netif_carrier_ok(curr_netdev)) ?
3175 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3176 
3177 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3178 			curr_port_state : IB_PORT_ACTIVE;
3179 	}
3180 	spin_unlock_bh(&ibdev->iboe.lock);
3181 
3182 	ibev.device = &ibdev->ib_dev;
3183 	ibev.element.port_num = 1;
3184 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3185 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3186 
3187 	ib_dispatch_event(&ibev);
3188 }
3189 
3190 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3191 {
3192 	u64 sl2vl;
3193 	int err;
3194 
3195 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3196 	if (err) {
3197 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3198 		       port, err);
3199 		sl2vl = 0;
3200 	}
3201 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3202 }
3203 
3204 static void ib_sl2vl_update_work(struct work_struct *work)
3205 {
3206 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3207 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3208 	int port = ew->port;
3209 
3210 	mlx4_ib_sl2vl_update(mdev, port);
3211 
3212 	kfree(ew);
3213 }
3214 
3215 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3216 				     int port)
3217 {
3218 	struct ib_event_work *ew;
3219 
3220 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3221 	if (ew) {
3222 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3223 		ew->port = port;
3224 		ew->ib_dev = ibdev;
3225 		queue_work(wq, &ew->work);
3226 	}
3227 }
3228 
3229 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3230 			  enum mlx4_dev_event event, unsigned long param)
3231 {
3232 	struct ib_event ibev;
3233 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3234 	struct mlx4_eqe *eqe = NULL;
3235 	struct ib_event_work *ew;
3236 	int p = 0;
3237 
3238 	if (mlx4_is_bonded(dev) &&
3239 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3240 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3241 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3242 		if (!ew)
3243 			return;
3244 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3245 		ew->ib_dev = ibdev;
3246 		queue_work(wq, &ew->work);
3247 		return;
3248 	}
3249 
3250 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3251 		eqe = (struct mlx4_eqe *)param;
3252 	else
3253 		p = (int) param;
3254 
3255 	switch (event) {
3256 	case MLX4_DEV_EVENT_PORT_UP:
3257 		if (p > ibdev->num_ports)
3258 			return;
3259 		if (!mlx4_is_slave(dev) &&
3260 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3261 			IB_LINK_LAYER_INFINIBAND) {
3262 			if (mlx4_is_master(dev))
3263 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3264 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3265 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3266 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3267 		}
3268 		ibev.event = IB_EVENT_PORT_ACTIVE;
3269 		break;
3270 
3271 	case MLX4_DEV_EVENT_PORT_DOWN:
3272 		if (p > ibdev->num_ports)
3273 			return;
3274 		ibev.event = IB_EVENT_PORT_ERR;
3275 		break;
3276 
3277 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3278 		ibdev->ib_active = false;
3279 		ibev.event = IB_EVENT_DEVICE_FATAL;
3280 		mlx4_ib_handle_catas_error(ibdev);
3281 		break;
3282 
3283 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3284 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3285 		if (!ew)
3286 			break;
3287 
3288 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3289 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3290 		ew->ib_dev = ibdev;
3291 		/* need to queue only for port owner, which uses GEN_EQE */
3292 		if (mlx4_is_master(dev))
3293 			queue_work(wq, &ew->work);
3294 		else
3295 			handle_port_mgmt_change_event(&ew->work);
3296 		return;
3297 
3298 	case MLX4_DEV_EVENT_SLAVE_INIT:
3299 		/* here, p is the slave id */
3300 		do_slave_init(ibdev, p, 1);
3301 		if (mlx4_is_master(dev)) {
3302 			int i;
3303 
3304 			for (i = 1; i <= ibdev->num_ports; i++) {
3305 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3306 					== IB_LINK_LAYER_INFINIBAND)
3307 					mlx4_ib_slave_alias_guid_event(ibdev,
3308 								       p, i,
3309 								       1);
3310 			}
3311 		}
3312 		return;
3313 
3314 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3315 		if (mlx4_is_master(dev)) {
3316 			int i;
3317 
3318 			for (i = 1; i <= ibdev->num_ports; i++) {
3319 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3320 					== IB_LINK_LAYER_INFINIBAND)
3321 					mlx4_ib_slave_alias_guid_event(ibdev,
3322 								       p, i,
3323 								       0);
3324 			}
3325 		}
3326 		/* here, p is the slave id */
3327 		do_slave_init(ibdev, p, 0);
3328 		return;
3329 
3330 	default:
3331 		return;
3332 	}
3333 
3334 	ibev.device	      = ibdev_ptr;
3335 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3336 
3337 	ib_dispatch_event(&ibev);
3338 }
3339 
3340 static struct mlx4_interface mlx4_ib_interface = {
3341 	.add		= mlx4_ib_add,
3342 	.remove		= mlx4_ib_remove,
3343 	.event		= mlx4_ib_event,
3344 	.protocol	= MLX4_PROT_IB_IPV6,
3345 	.flags		= MLX4_INTFF_BONDING
3346 };
3347 
3348 static int __init mlx4_ib_init(void)
3349 {
3350 	int err;
3351 
3352 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3353 	if (!wq)
3354 		return -ENOMEM;
3355 
3356 	err = mlx4_ib_mcg_init();
3357 	if (err)
3358 		goto clean_wq;
3359 
3360 	err = mlx4_register_interface(&mlx4_ib_interface);
3361 	if (err)
3362 		goto clean_mcg;
3363 
3364 	return 0;
3365 
3366 clean_mcg:
3367 	mlx4_ib_mcg_destroy();
3368 
3369 clean_wq:
3370 	destroy_workqueue(wq);
3371 	return err;
3372 }
3373 
3374 static void __exit mlx4_ib_cleanup(void)
3375 {
3376 	mlx4_unregister_interface(&mlx4_ib_interface);
3377 	mlx4_ib_mcg_destroy();
3378 	destroy_workqueue(wq);
3379 }
3380 
3381 module_init(mlx4_ib_init);
3382 module_exit(mlx4_ib_cleanup);
3383