xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision b830f94f)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
250 {
251 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
252 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
253 	struct mlx4_port_gid_table   *port_gid_table;
254 	int free = -1, found = -1;
255 	int ret = 0;
256 	int hw_update = 0;
257 	int i;
258 	struct gid_entry *gids = NULL;
259 
260 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
261 		return -EINVAL;
262 
263 	if (attr->port_num > MLX4_MAX_PORTS)
264 		return -EINVAL;
265 
266 	if (!context)
267 		return -EINVAL;
268 
269 	port_gid_table = &iboe->gids[attr->port_num - 1];
270 	spin_lock_bh(&iboe->lock);
271 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
272 		if (!memcmp(&port_gid_table->gids[i].gid,
273 			    &attr->gid, sizeof(attr->gid)) &&
274 		    port_gid_table->gids[i].gid_type == attr->gid_type)  {
275 			found = i;
276 			break;
277 		}
278 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
279 			free = i; /* HW has space */
280 	}
281 
282 	if (found < 0) {
283 		if (free < 0) {
284 			ret = -ENOSPC;
285 		} else {
286 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
287 			if (!port_gid_table->gids[free].ctx) {
288 				ret = -ENOMEM;
289 			} else {
290 				*context = port_gid_table->gids[free].ctx;
291 				memcpy(&port_gid_table->gids[free].gid,
292 				       &attr->gid, sizeof(attr->gid));
293 				port_gid_table->gids[free].gid_type = attr->gid_type;
294 				port_gid_table->gids[free].ctx->real_index = free;
295 				port_gid_table->gids[free].ctx->refcount = 1;
296 				hw_update = 1;
297 			}
298 		}
299 	} else {
300 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
301 		*context = ctx;
302 		ctx->refcount++;
303 	}
304 	if (!ret && hw_update) {
305 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
306 				     GFP_ATOMIC);
307 		if (!gids) {
308 			ret = -ENOMEM;
309 		} else {
310 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
311 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
312 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
313 			}
314 		}
315 	}
316 	spin_unlock_bh(&iboe->lock);
317 
318 	if (!ret && hw_update) {
319 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
320 		kfree(gids);
321 	}
322 
323 	return ret;
324 }
325 
326 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
327 {
328 	struct gid_cache_context *ctx = *context;
329 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
330 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
331 	struct mlx4_port_gid_table   *port_gid_table;
332 	int ret = 0;
333 	int hw_update = 0;
334 	struct gid_entry *gids = NULL;
335 
336 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
337 		return -EINVAL;
338 
339 	if (attr->port_num > MLX4_MAX_PORTS)
340 		return -EINVAL;
341 
342 	port_gid_table = &iboe->gids[attr->port_num - 1];
343 	spin_lock_bh(&iboe->lock);
344 	if (ctx) {
345 		ctx->refcount--;
346 		if (!ctx->refcount) {
347 			unsigned int real_index = ctx->real_index;
348 
349 			memset(&port_gid_table->gids[real_index].gid, 0,
350 			       sizeof(port_gid_table->gids[real_index].gid));
351 			kfree(port_gid_table->gids[real_index].ctx);
352 			port_gid_table->gids[real_index].ctx = NULL;
353 			hw_update = 1;
354 		}
355 	}
356 	if (!ret && hw_update) {
357 		int i;
358 
359 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
360 				     GFP_ATOMIC);
361 		if (!gids) {
362 			ret = -ENOMEM;
363 		} else {
364 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
365 				memcpy(&gids[i].gid,
366 				       &port_gid_table->gids[i].gid,
367 				       sizeof(union ib_gid));
368 				gids[i].gid_type =
369 				    port_gid_table->gids[i].gid_type;
370 			}
371 		}
372 	}
373 	spin_unlock_bh(&iboe->lock);
374 
375 	if (!ret && hw_update) {
376 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
377 		kfree(gids);
378 	}
379 	return ret;
380 }
381 
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383 				    const struct ib_gid_attr *attr)
384 {
385 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386 	struct gid_cache_context *ctx = NULL;
387 	struct mlx4_port_gid_table   *port_gid_table;
388 	int real_index = -EINVAL;
389 	int i;
390 	unsigned long flags;
391 	u8 port_num = attr->port_num;
392 
393 	if (port_num > MLX4_MAX_PORTS)
394 		return -EINVAL;
395 
396 	if (mlx4_is_bonded(ibdev->dev))
397 		port_num = 1;
398 
399 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
400 		return attr->index;
401 
402 	spin_lock_irqsave(&iboe->lock, flags);
403 	port_gid_table = &iboe->gids[port_num - 1];
404 
405 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
406 		if (!memcmp(&port_gid_table->gids[i].gid,
407 			    &attr->gid, sizeof(attr->gid)) &&
408 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
409 			ctx = port_gid_table->gids[i].ctx;
410 			break;
411 		}
412 	if (ctx)
413 		real_index = ctx->real_index;
414 	spin_unlock_irqrestore(&iboe->lock, flags);
415 	return real_index;
416 }
417 
418 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
419 				    sizeof(((type *)0)->fld) <= (sz))
420 
421 static int mlx4_ib_query_device(struct ib_device *ibdev,
422 				struct ib_device_attr *props,
423 				struct ib_udata *uhw)
424 {
425 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
426 	struct ib_smp *in_mad  = NULL;
427 	struct ib_smp *out_mad = NULL;
428 	int err;
429 	int have_ib_ports;
430 	struct mlx4_uverbs_ex_query_device cmd;
431 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
432 	struct mlx4_clock_params clock_params;
433 
434 	if (uhw->inlen) {
435 		if (uhw->inlen < sizeof(cmd))
436 			return -EINVAL;
437 
438 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
439 		if (err)
440 			return err;
441 
442 		if (cmd.comp_mask)
443 			return -EINVAL;
444 
445 		if (cmd.reserved)
446 			return -EINVAL;
447 	}
448 
449 	resp.response_length = offsetof(typeof(resp), response_length) +
450 		sizeof(resp.response_length);
451 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
452 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
453 	err = -ENOMEM;
454 	if (!in_mad || !out_mad)
455 		goto out;
456 
457 	init_query_mad(in_mad);
458 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
459 
460 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
461 			   1, NULL, NULL, in_mad, out_mad);
462 	if (err)
463 		goto out;
464 
465 	memset(props, 0, sizeof *props);
466 
467 	have_ib_ports = num_ib_ports(dev->dev);
468 
469 	props->fw_ver = dev->dev->caps.fw_ver;
470 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
471 		IB_DEVICE_PORT_ACTIVE_EVENT		|
472 		IB_DEVICE_SYS_IMAGE_GUID		|
473 		IB_DEVICE_RC_RNR_NAK_GEN		|
474 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
475 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
476 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
477 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
478 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
479 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
480 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
481 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
482 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
484 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
485 	if (dev->dev->caps.max_gso_sz &&
486 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
487 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
488 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
489 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
490 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
491 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
492 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
493 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
494 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
496 		props->device_cap_flags |= IB_DEVICE_XRC;
497 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
498 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
499 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
500 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
501 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
502 		else
503 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
504 	}
505 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
506 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
507 
508 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
509 
510 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
511 		0xffffff;
512 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
513 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
514 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
515 
516 	props->max_mr_size	   = ~0ull;
517 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
518 	props->max_qp		   = dev->dev->quotas.qp;
519 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
520 	props->max_send_sge =
521 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
522 	props->max_recv_sge =
523 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
524 	props->max_sge_rd = MLX4_MAX_SGE_RD;
525 	props->max_cq		   = dev->dev->quotas.cq;
526 	props->max_cqe		   = dev->dev->caps.max_cqes;
527 	props->max_mr		   = dev->dev->quotas.mpt;
528 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
529 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
530 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
531 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
532 	props->max_srq		   = dev->dev->quotas.srq;
533 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
534 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
535 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
536 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
537 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
538 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
539 	props->masked_atomic_cap   = props->atomic_cap;
540 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
541 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
542 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
543 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
544 					   props->max_mcast_grp;
545 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
546 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
547 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
548 	props->max_ah = INT_MAX;
549 
550 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
551 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
552 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
553 			props->rss_caps.max_rwq_indirection_tables =
554 				props->max_qp;
555 			props->rss_caps.max_rwq_indirection_table_size =
556 				dev->dev->caps.max_rss_tbl_sz;
557 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
558 			props->max_wq_type_rq = props->max_qp;
559 		}
560 
561 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
562 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
563 	}
564 
565 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
566 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
567 
568 	if (!mlx4_is_slave(dev->dev))
569 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
570 
571 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
572 		resp.response_length += sizeof(resp.hca_core_clock_offset);
573 		if (!err && !mlx4_is_slave(dev->dev)) {
574 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
575 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
576 		}
577 	}
578 
579 	if (uhw->outlen >= resp.response_length +
580 	    sizeof(resp.max_inl_recv_sz)) {
581 		resp.response_length += sizeof(resp.max_inl_recv_sz);
582 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
583 			sizeof(struct mlx4_wqe_data_seg);
584 	}
585 
586 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
587 		if (props->rss_caps.supported_qpts) {
588 			resp.rss_caps.rx_hash_function =
589 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
590 
591 			resp.rss_caps.rx_hash_fields_mask =
592 				MLX4_IB_RX_HASH_SRC_IPV4 |
593 				MLX4_IB_RX_HASH_DST_IPV4 |
594 				MLX4_IB_RX_HASH_SRC_IPV6 |
595 				MLX4_IB_RX_HASH_DST_IPV6 |
596 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
597 				MLX4_IB_RX_HASH_DST_PORT_TCP |
598 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
599 				MLX4_IB_RX_HASH_DST_PORT_UDP;
600 
601 			if (dev->dev->caps.tunnel_offload_mode ==
602 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
603 				resp.rss_caps.rx_hash_fields_mask |=
604 					MLX4_IB_RX_HASH_INNER;
605 		}
606 		resp.response_length = offsetof(typeof(resp), rss_caps) +
607 				       sizeof(resp.rss_caps);
608 	}
609 
610 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
611 		if (dev->dev->caps.max_gso_sz &&
612 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
613 		    IB_LINK_LAYER_ETHERNET) ||
614 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
615 		    IB_LINK_LAYER_ETHERNET))) {
616 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
617 			resp.tso_caps.supported_qpts |=
618 				1 << IB_QPT_RAW_PACKET;
619 		}
620 		resp.response_length = offsetof(typeof(resp), tso_caps) +
621 				       sizeof(resp.tso_caps);
622 	}
623 
624 	if (uhw->outlen) {
625 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
626 		if (err)
627 			goto out;
628 	}
629 out:
630 	kfree(in_mad);
631 	kfree(out_mad);
632 
633 	return err;
634 }
635 
636 static enum rdma_link_layer
637 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
638 {
639 	struct mlx4_dev *dev = to_mdev(device)->dev;
640 
641 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
642 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
643 }
644 
645 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
646 			      struct ib_port_attr *props, int netw_view)
647 {
648 	struct ib_smp *in_mad  = NULL;
649 	struct ib_smp *out_mad = NULL;
650 	int ext_active_speed;
651 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
652 	int err = -ENOMEM;
653 
654 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
655 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
656 	if (!in_mad || !out_mad)
657 		goto out;
658 
659 	init_query_mad(in_mad);
660 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
661 	in_mad->attr_mod = cpu_to_be32(port);
662 
663 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
664 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
665 
666 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
667 				in_mad, out_mad);
668 	if (err)
669 		goto out;
670 
671 
672 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
673 	props->lmc		= out_mad->data[34] & 0x7;
674 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
675 	props->sm_sl		= out_mad->data[36] & 0xf;
676 	props->state		= out_mad->data[32] & 0xf;
677 	props->phys_state	= out_mad->data[33] >> 4;
678 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
679 	if (netw_view)
680 		props->gid_tbl_len = out_mad->data[50];
681 	else
682 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
683 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
684 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
685 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
686 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
687 	props->active_width	= out_mad->data[31] & 0xf;
688 	props->active_speed	= out_mad->data[35] >> 4;
689 	props->max_mtu		= out_mad->data[41] & 0xf;
690 	props->active_mtu	= out_mad->data[36] >> 4;
691 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
692 	props->max_vl_num	= out_mad->data[37] >> 4;
693 	props->init_type_reply	= out_mad->data[41] >> 4;
694 
695 	/* Check if extended speeds (EDR/FDR/...) are supported */
696 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
697 		ext_active_speed = out_mad->data[62] >> 4;
698 
699 		switch (ext_active_speed) {
700 		case 1:
701 			props->active_speed = IB_SPEED_FDR;
702 			break;
703 		case 2:
704 			props->active_speed = IB_SPEED_EDR;
705 			break;
706 		}
707 	}
708 
709 	/* If reported active speed is QDR, check if is FDR-10 */
710 	if (props->active_speed == IB_SPEED_QDR) {
711 		init_query_mad(in_mad);
712 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
713 		in_mad->attr_mod = cpu_to_be32(port);
714 
715 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
716 				   NULL, NULL, in_mad, out_mad);
717 		if (err)
718 			goto out;
719 
720 		/* Checking LinkSpeedActive for FDR-10 */
721 		if (out_mad->data[15] & 0x1)
722 			props->active_speed = IB_SPEED_FDR10;
723 	}
724 
725 	/* Avoid wrong speed value returned by FW if the IB link is down. */
726 	if (props->state == IB_PORT_DOWN)
727 		 props->active_speed = IB_SPEED_SDR;
728 
729 out:
730 	kfree(in_mad);
731 	kfree(out_mad);
732 	return err;
733 }
734 
735 static u8 state_to_phys_state(enum ib_port_state state)
736 {
737 	return state == IB_PORT_ACTIVE ? 5 : 3;
738 }
739 
740 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
741 			       struct ib_port_attr *props)
742 {
743 
744 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
745 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
746 	struct net_device *ndev;
747 	enum ib_mtu tmp;
748 	struct mlx4_cmd_mailbox *mailbox;
749 	int err = 0;
750 	int is_bonded = mlx4_is_bonded(mdev->dev);
751 
752 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
753 	if (IS_ERR(mailbox))
754 		return PTR_ERR(mailbox);
755 
756 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
757 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
758 			   MLX4_CMD_WRAPPED);
759 	if (err)
760 		goto out;
761 
762 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
763 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
764 					   IB_WIDTH_4X : IB_WIDTH_1X;
765 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
766 					   IB_SPEED_FDR : IB_SPEED_QDR;
767 	props->port_cap_flags	= IB_PORT_CM_SUP;
768 	props->ip_gids = true;
769 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
770 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
771 	props->pkey_tbl_len	= 1;
772 	props->max_mtu		= IB_MTU_4096;
773 	props->max_vl_num	= 2;
774 	props->state		= IB_PORT_DOWN;
775 	props->phys_state	= state_to_phys_state(props->state);
776 	props->active_mtu	= IB_MTU_256;
777 	spin_lock_bh(&iboe->lock);
778 	ndev = iboe->netdevs[port - 1];
779 	if (ndev && is_bonded) {
780 		rcu_read_lock(); /* required to get upper dev */
781 		ndev = netdev_master_upper_dev_get_rcu(ndev);
782 		rcu_read_unlock();
783 	}
784 	if (!ndev)
785 		goto out_unlock;
786 
787 	tmp = iboe_get_mtu(ndev->mtu);
788 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
789 
790 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
791 					IB_PORT_ACTIVE : IB_PORT_DOWN;
792 	props->phys_state	= state_to_phys_state(props->state);
793 out_unlock:
794 	spin_unlock_bh(&iboe->lock);
795 out:
796 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
797 	return err;
798 }
799 
800 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
801 			 struct ib_port_attr *props, int netw_view)
802 {
803 	int err;
804 
805 	/* props being zeroed by the caller, avoid zeroing it here */
806 
807 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
808 		ib_link_query_port(ibdev, port, props, netw_view) :
809 				eth_link_query_port(ibdev, port, props);
810 
811 	return err;
812 }
813 
814 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
815 			      struct ib_port_attr *props)
816 {
817 	/* returns host view */
818 	return __mlx4_ib_query_port(ibdev, port, props, 0);
819 }
820 
821 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
822 			union ib_gid *gid, int netw_view)
823 {
824 	struct ib_smp *in_mad  = NULL;
825 	struct ib_smp *out_mad = NULL;
826 	int err = -ENOMEM;
827 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
828 	int clear = 0;
829 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
830 
831 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
832 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
833 	if (!in_mad || !out_mad)
834 		goto out;
835 
836 	init_query_mad(in_mad);
837 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
838 	in_mad->attr_mod = cpu_to_be32(port);
839 
840 	if (mlx4_is_mfunc(dev->dev) && netw_view)
841 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
842 
843 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
844 	if (err)
845 		goto out;
846 
847 	memcpy(gid->raw, out_mad->data + 8, 8);
848 
849 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
850 		if (index) {
851 			/* For any index > 0, return the null guid */
852 			err = 0;
853 			clear = 1;
854 			goto out;
855 		}
856 	}
857 
858 	init_query_mad(in_mad);
859 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
860 	in_mad->attr_mod = cpu_to_be32(index / 8);
861 
862 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
863 			   NULL, NULL, in_mad, out_mad);
864 	if (err)
865 		goto out;
866 
867 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
868 
869 out:
870 	if (clear)
871 		memset(gid->raw + 8, 0, 8);
872 	kfree(in_mad);
873 	kfree(out_mad);
874 	return err;
875 }
876 
877 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
878 			     union ib_gid *gid)
879 {
880 	if (rdma_protocol_ib(ibdev, port))
881 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
882 	return 0;
883 }
884 
885 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
886 {
887 	union sl2vl_tbl_to_u64 sl2vl64;
888 	struct ib_smp *in_mad  = NULL;
889 	struct ib_smp *out_mad = NULL;
890 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
891 	int err = -ENOMEM;
892 	int jj;
893 
894 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
895 		*sl2vl_tbl = 0;
896 		return 0;
897 	}
898 
899 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
900 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
901 	if (!in_mad || !out_mad)
902 		goto out;
903 
904 	init_query_mad(in_mad);
905 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
906 	in_mad->attr_mod = 0;
907 
908 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
909 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
910 
911 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
912 			   in_mad, out_mad);
913 	if (err)
914 		goto out;
915 
916 	for (jj = 0; jj < 8; jj++)
917 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
918 	*sl2vl_tbl = sl2vl64.sl64;
919 
920 out:
921 	kfree(in_mad);
922 	kfree(out_mad);
923 	return err;
924 }
925 
926 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
927 {
928 	u64 sl2vl;
929 	int i;
930 	int err;
931 
932 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
933 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
934 			continue;
935 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
936 		if (err) {
937 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
938 			       i, err);
939 			sl2vl = 0;
940 		}
941 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
942 	}
943 }
944 
945 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
946 			 u16 *pkey, int netw_view)
947 {
948 	struct ib_smp *in_mad  = NULL;
949 	struct ib_smp *out_mad = NULL;
950 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
951 	int err = -ENOMEM;
952 
953 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
954 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
955 	if (!in_mad || !out_mad)
956 		goto out;
957 
958 	init_query_mad(in_mad);
959 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
960 	in_mad->attr_mod = cpu_to_be32(index / 32);
961 
962 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
963 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
964 
965 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
966 			   in_mad, out_mad);
967 	if (err)
968 		goto out;
969 
970 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
971 
972 out:
973 	kfree(in_mad);
974 	kfree(out_mad);
975 	return err;
976 }
977 
978 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
979 {
980 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
981 }
982 
983 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
984 				 struct ib_device_modify *props)
985 {
986 	struct mlx4_cmd_mailbox *mailbox;
987 	unsigned long flags;
988 
989 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
990 		return -EOPNOTSUPP;
991 
992 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
993 		return 0;
994 
995 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
996 		return -EOPNOTSUPP;
997 
998 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
999 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1000 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1001 
1002 	/*
1003 	 * If possible, pass node desc to FW, so it can generate
1004 	 * a 144 trap.  If cmd fails, just ignore.
1005 	 */
1006 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1007 	if (IS_ERR(mailbox))
1008 		return 0;
1009 
1010 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1011 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1012 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1013 
1014 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1015 
1016 	return 0;
1017 }
1018 
1019 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1020 			    u32 cap_mask)
1021 {
1022 	struct mlx4_cmd_mailbox *mailbox;
1023 	int err;
1024 
1025 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1026 	if (IS_ERR(mailbox))
1027 		return PTR_ERR(mailbox);
1028 
1029 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1030 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1031 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1032 	} else {
1033 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1034 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1035 	}
1036 
1037 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1038 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1039 		       MLX4_CMD_WRAPPED);
1040 
1041 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1042 	return err;
1043 }
1044 
1045 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1046 			       struct ib_port_modify *props)
1047 {
1048 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1049 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1050 	struct ib_port_attr attr;
1051 	u32 cap_mask;
1052 	int err;
1053 
1054 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1055 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1056 	 * violations and port capabilities are not meaningful.
1057 	 */
1058 	if (is_eth)
1059 		return 0;
1060 
1061 	mutex_lock(&mdev->cap_mask_mutex);
1062 
1063 	err = ib_query_port(ibdev, port, &attr);
1064 	if (err)
1065 		goto out;
1066 
1067 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1068 		~props->clr_port_cap_mask;
1069 
1070 	err = mlx4_ib_SET_PORT(mdev, port,
1071 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1072 			       cap_mask);
1073 
1074 out:
1075 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1076 	return err;
1077 }
1078 
1079 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1080 				  struct ib_udata *udata)
1081 {
1082 	struct ib_device *ibdev = uctx->device;
1083 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1084 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1085 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1086 	struct mlx4_ib_alloc_ucontext_resp resp;
1087 	int err;
1088 
1089 	if (!dev->ib_active)
1090 		return -EAGAIN;
1091 
1092 	if (ibdev->ops.uverbs_abi_ver ==
1093 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1094 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1095 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1096 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1097 	} else {
1098 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1099 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1100 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1101 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1102 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1103 	}
1104 
1105 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1106 	if (err)
1107 		return err;
1108 
1109 	INIT_LIST_HEAD(&context->db_page_list);
1110 	mutex_init(&context->db_page_mutex);
1111 
1112 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1113 	mutex_init(&context->wqn_ranges_mutex);
1114 
1115 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1116 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1117 	else
1118 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1119 
1120 	if (err) {
1121 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1122 		return -EFAULT;
1123 	}
1124 
1125 	return err;
1126 }
1127 
1128 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1129 {
1130 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1131 
1132 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1133 }
1134 
1135 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1136 {
1137 }
1138 
1139 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1140 {
1141 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1142 
1143 	switch (vma->vm_pgoff) {
1144 	case 0:
1145 		return rdma_user_mmap_io(context, vma,
1146 					 to_mucontext(context)->uar.pfn,
1147 					 PAGE_SIZE,
1148 					 pgprot_noncached(vma->vm_page_prot));
1149 
1150 	case 1:
1151 		if (dev->dev->caps.bf_reg_size == 0)
1152 			return -EINVAL;
1153 		return rdma_user_mmap_io(
1154 			context, vma,
1155 			to_mucontext(context)->uar.pfn +
1156 				dev->dev->caps.num_uars,
1157 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot));
1158 
1159 	case 3: {
1160 		struct mlx4_clock_params params;
1161 		int ret;
1162 
1163 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1164 		if (ret)
1165 			return ret;
1166 
1167 		return rdma_user_mmap_io(
1168 			context, vma,
1169 			(pci_resource_start(dev->dev->persist->pdev,
1170 					    params.bar) +
1171 			 params.offset) >>
1172 				PAGE_SHIFT,
1173 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot));
1174 	}
1175 
1176 	default:
1177 		return -EINVAL;
1178 	}
1179 }
1180 
1181 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1182 {
1183 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1184 	struct ib_device *ibdev = ibpd->device;
1185 	int err;
1186 
1187 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1188 	if (err)
1189 		return err;
1190 
1191 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1192 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1193 		return -EFAULT;
1194 	}
1195 	return 0;
1196 }
1197 
1198 static void mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1199 {
1200 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1201 }
1202 
1203 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1204 					  struct ib_udata *udata)
1205 {
1206 	struct mlx4_ib_xrcd *xrcd;
1207 	struct ib_cq_init_attr cq_attr = {};
1208 	int err;
1209 
1210 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1211 		return ERR_PTR(-ENOSYS);
1212 
1213 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1214 	if (!xrcd)
1215 		return ERR_PTR(-ENOMEM);
1216 
1217 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1218 	if (err)
1219 		goto err1;
1220 
1221 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1222 	if (IS_ERR(xrcd->pd)) {
1223 		err = PTR_ERR(xrcd->pd);
1224 		goto err2;
1225 	}
1226 
1227 	cq_attr.cqe = 1;
1228 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1229 	if (IS_ERR(xrcd->cq)) {
1230 		err = PTR_ERR(xrcd->cq);
1231 		goto err3;
1232 	}
1233 
1234 	return &xrcd->ibxrcd;
1235 
1236 err3:
1237 	ib_dealloc_pd(xrcd->pd);
1238 err2:
1239 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1240 err1:
1241 	kfree(xrcd);
1242 	return ERR_PTR(err);
1243 }
1244 
1245 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1246 {
1247 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1248 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1249 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1250 	kfree(xrcd);
1251 
1252 	return 0;
1253 }
1254 
1255 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1256 {
1257 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1258 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1259 	struct mlx4_ib_gid_entry *ge;
1260 
1261 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1262 	if (!ge)
1263 		return -ENOMEM;
1264 
1265 	ge->gid = *gid;
1266 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1267 		ge->port = mqp->port;
1268 		ge->added = 1;
1269 	}
1270 
1271 	mutex_lock(&mqp->mutex);
1272 	list_add_tail(&ge->list, &mqp->gid_list);
1273 	mutex_unlock(&mqp->mutex);
1274 
1275 	return 0;
1276 }
1277 
1278 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1279 					  struct mlx4_ib_counters *ctr_table)
1280 {
1281 	struct counter_index *counter, *tmp_count;
1282 
1283 	mutex_lock(&ctr_table->mutex);
1284 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1285 				 list) {
1286 		if (counter->allocated)
1287 			mlx4_counter_free(ibdev->dev, counter->index);
1288 		list_del(&counter->list);
1289 		kfree(counter);
1290 	}
1291 	mutex_unlock(&ctr_table->mutex);
1292 }
1293 
1294 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1295 		   union ib_gid *gid)
1296 {
1297 	struct net_device *ndev;
1298 	int ret = 0;
1299 
1300 	if (!mqp->port)
1301 		return 0;
1302 
1303 	spin_lock_bh(&mdev->iboe.lock);
1304 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1305 	if (ndev)
1306 		dev_hold(ndev);
1307 	spin_unlock_bh(&mdev->iboe.lock);
1308 
1309 	if (ndev) {
1310 		ret = 1;
1311 		dev_put(ndev);
1312 	}
1313 
1314 	return ret;
1315 }
1316 
1317 struct mlx4_ib_steering {
1318 	struct list_head list;
1319 	struct mlx4_flow_reg_id reg_id;
1320 	union ib_gid gid;
1321 };
1322 
1323 #define LAST_ETH_FIELD vlan_tag
1324 #define LAST_IB_FIELD sl
1325 #define LAST_IPV4_FIELD dst_ip
1326 #define LAST_TCP_UDP_FIELD src_port
1327 
1328 /* Field is the last supported field */
1329 #define FIELDS_NOT_SUPPORTED(filter, field)\
1330 	memchr_inv((void *)&filter.field  +\
1331 		   sizeof(filter.field), 0,\
1332 		   sizeof(filter) -\
1333 		   offsetof(typeof(filter), field) -\
1334 		   sizeof(filter.field))
1335 
1336 static int parse_flow_attr(struct mlx4_dev *dev,
1337 			   u32 qp_num,
1338 			   union ib_flow_spec *ib_spec,
1339 			   struct _rule_hw *mlx4_spec)
1340 {
1341 	enum mlx4_net_trans_rule_id type;
1342 
1343 	switch (ib_spec->type) {
1344 	case IB_FLOW_SPEC_ETH:
1345 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1346 			return -ENOTSUPP;
1347 
1348 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1349 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1350 		       ETH_ALEN);
1351 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1352 		       ETH_ALEN);
1353 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1354 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1355 		break;
1356 	case IB_FLOW_SPEC_IB:
1357 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1358 			return -ENOTSUPP;
1359 
1360 		type = MLX4_NET_TRANS_RULE_ID_IB;
1361 		mlx4_spec->ib.l3_qpn =
1362 			cpu_to_be32(qp_num);
1363 		mlx4_spec->ib.qpn_mask =
1364 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1365 		break;
1366 
1367 
1368 	case IB_FLOW_SPEC_IPV4:
1369 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1370 			return -ENOTSUPP;
1371 
1372 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1373 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1374 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1375 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1376 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1377 		break;
1378 
1379 	case IB_FLOW_SPEC_TCP:
1380 	case IB_FLOW_SPEC_UDP:
1381 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1382 			return -ENOTSUPP;
1383 
1384 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1385 					MLX4_NET_TRANS_RULE_ID_TCP :
1386 					MLX4_NET_TRANS_RULE_ID_UDP;
1387 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1388 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1389 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1390 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1391 		break;
1392 
1393 	default:
1394 		return -EINVAL;
1395 	}
1396 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1397 	    mlx4_hw_rule_sz(dev, type) < 0)
1398 		return -EINVAL;
1399 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1400 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1401 	return mlx4_hw_rule_sz(dev, type);
1402 }
1403 
1404 struct default_rules {
1405 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1406 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1407 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1408 	__u8  link_layer;
1409 };
1410 static const struct default_rules default_table[] = {
1411 	{
1412 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1413 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1414 		.rules_create_list = {IB_FLOW_SPEC_IB},
1415 		.link_layer = IB_LINK_LAYER_INFINIBAND
1416 	}
1417 };
1418 
1419 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1420 					 struct ib_flow_attr *flow_attr)
1421 {
1422 	int i, j, k;
1423 	void *ib_flow;
1424 	const struct default_rules *pdefault_rules = default_table;
1425 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1426 
1427 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1428 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1429 		memset(&field_types, 0, sizeof(field_types));
1430 
1431 		if (link_layer != pdefault_rules->link_layer)
1432 			continue;
1433 
1434 		ib_flow = flow_attr + 1;
1435 		/* we assume the specs are sorted */
1436 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1437 		     j < flow_attr->num_of_specs; k++) {
1438 			union ib_flow_spec *current_flow =
1439 				(union ib_flow_spec *)ib_flow;
1440 
1441 			/* same layer but different type */
1442 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1443 			     (pdefault_rules->mandatory_fields[k] &
1444 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1445 			    (current_flow->type !=
1446 			     pdefault_rules->mandatory_fields[k]))
1447 				goto out;
1448 
1449 			/* same layer, try match next one */
1450 			if (current_flow->type ==
1451 			    pdefault_rules->mandatory_fields[k]) {
1452 				j++;
1453 				ib_flow +=
1454 					((union ib_flow_spec *)ib_flow)->size;
1455 			}
1456 		}
1457 
1458 		ib_flow = flow_attr + 1;
1459 		for (j = 0; j < flow_attr->num_of_specs;
1460 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1461 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1462 				/* same layer and same type */
1463 				if (((union ib_flow_spec *)ib_flow)->type ==
1464 				    pdefault_rules->mandatory_not_fields[k])
1465 					goto out;
1466 
1467 		return i;
1468 	}
1469 out:
1470 	return -1;
1471 }
1472 
1473 static int __mlx4_ib_create_default_rules(
1474 		struct mlx4_ib_dev *mdev,
1475 		struct ib_qp *qp,
1476 		const struct default_rules *pdefault_rules,
1477 		struct _rule_hw *mlx4_spec) {
1478 	int size = 0;
1479 	int i;
1480 
1481 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1482 		int ret;
1483 		union ib_flow_spec ib_spec;
1484 		switch (pdefault_rules->rules_create_list[i]) {
1485 		case 0:
1486 			/* no rule */
1487 			continue;
1488 		case IB_FLOW_SPEC_IB:
1489 			ib_spec.type = IB_FLOW_SPEC_IB;
1490 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1491 
1492 			break;
1493 		default:
1494 			/* invalid rule */
1495 			return -EINVAL;
1496 		}
1497 		/* We must put empty rule, qpn is being ignored */
1498 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1499 				      mlx4_spec);
1500 		if (ret < 0) {
1501 			pr_info("invalid parsing\n");
1502 			return -EINVAL;
1503 		}
1504 
1505 		mlx4_spec = (void *)mlx4_spec + ret;
1506 		size += ret;
1507 	}
1508 	return size;
1509 }
1510 
1511 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1512 			  int domain,
1513 			  enum mlx4_net_trans_promisc_mode flow_type,
1514 			  u64 *reg_id)
1515 {
1516 	int ret, i;
1517 	int size = 0;
1518 	void *ib_flow;
1519 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1520 	struct mlx4_cmd_mailbox *mailbox;
1521 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1522 	int default_flow;
1523 
1524 	static const u16 __mlx4_domain[] = {
1525 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1526 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1527 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1528 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1529 	};
1530 
1531 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1532 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1533 		return -EINVAL;
1534 	}
1535 
1536 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1537 		pr_err("Invalid domain value %d\n", domain);
1538 		return -EINVAL;
1539 	}
1540 
1541 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1542 		return -EINVAL;
1543 
1544 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1545 	if (IS_ERR(mailbox))
1546 		return PTR_ERR(mailbox);
1547 	ctrl = mailbox->buf;
1548 
1549 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1550 				 flow_attr->priority);
1551 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1552 	ctrl->port = flow_attr->port;
1553 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1554 
1555 	ib_flow = flow_attr + 1;
1556 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1557 	/* Add default flows */
1558 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1559 	if (default_flow >= 0) {
1560 		ret = __mlx4_ib_create_default_rules(
1561 				mdev, qp, default_table + default_flow,
1562 				mailbox->buf + size);
1563 		if (ret < 0) {
1564 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1565 			return -EINVAL;
1566 		}
1567 		size += ret;
1568 	}
1569 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1570 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1571 				      mailbox->buf + size);
1572 		if (ret < 0) {
1573 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1574 			return -EINVAL;
1575 		}
1576 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1577 		size += ret;
1578 	}
1579 
1580 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1581 	    flow_attr->num_of_specs == 1) {
1582 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1583 		enum ib_flow_spec_type header_spec =
1584 			((union ib_flow_spec *)(flow_attr + 1))->type;
1585 
1586 		if (header_spec == IB_FLOW_SPEC_ETH)
1587 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1588 	}
1589 
1590 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1591 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1592 			   MLX4_CMD_NATIVE);
1593 	if (ret == -ENOMEM)
1594 		pr_err("mcg table is full. Fail to register network rule.\n");
1595 	else if (ret == -ENXIO)
1596 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1597 	else if (ret)
1598 		pr_err("Invalid argument. Fail to register network rule.\n");
1599 
1600 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1601 	return ret;
1602 }
1603 
1604 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1605 {
1606 	int err;
1607 	err = mlx4_cmd(dev, reg_id, 0, 0,
1608 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1609 		       MLX4_CMD_NATIVE);
1610 	if (err)
1611 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1612 		       reg_id);
1613 	return err;
1614 }
1615 
1616 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1617 				    u64 *reg_id)
1618 {
1619 	void *ib_flow;
1620 	union ib_flow_spec *ib_spec;
1621 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1622 	int err = 0;
1623 
1624 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1625 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1626 		return 0; /* do nothing */
1627 
1628 	ib_flow = flow_attr + 1;
1629 	ib_spec = (union ib_flow_spec *)ib_flow;
1630 
1631 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1632 		return 0; /* do nothing */
1633 
1634 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1635 				    flow_attr->port, qp->qp_num,
1636 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1637 				    reg_id);
1638 	return err;
1639 }
1640 
1641 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1642 				      struct ib_flow_attr *flow_attr,
1643 				      enum mlx4_net_trans_promisc_mode *type)
1644 {
1645 	int err = 0;
1646 
1647 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1648 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1649 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1650 		return -EOPNOTSUPP;
1651 	}
1652 
1653 	if (flow_attr->num_of_specs == 0) {
1654 		type[0] = MLX4_FS_MC_SNIFFER;
1655 		type[1] = MLX4_FS_UC_SNIFFER;
1656 	} else {
1657 		union ib_flow_spec *ib_spec;
1658 
1659 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1660 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1661 			return -EINVAL;
1662 
1663 		/* if all is zero than MC and UC */
1664 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1665 			type[0] = MLX4_FS_MC_SNIFFER;
1666 			type[1] = MLX4_FS_UC_SNIFFER;
1667 		} else {
1668 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1669 					    ib_spec->eth.mask.dst_mac[1],
1670 					    ib_spec->eth.mask.dst_mac[2],
1671 					    ib_spec->eth.mask.dst_mac[3],
1672 					    ib_spec->eth.mask.dst_mac[4],
1673 					    ib_spec->eth.mask.dst_mac[5]};
1674 
1675 			/* Above xor was only on MC bit, non empty mask is valid
1676 			 * only if this bit is set and rest are zero.
1677 			 */
1678 			if (!is_zero_ether_addr(&mac[0]))
1679 				return -EINVAL;
1680 
1681 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1682 				type[0] = MLX4_FS_MC_SNIFFER;
1683 			else
1684 				type[0] = MLX4_FS_UC_SNIFFER;
1685 		}
1686 	}
1687 
1688 	return err;
1689 }
1690 
1691 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1692 				    struct ib_flow_attr *flow_attr,
1693 				    int domain, struct ib_udata *udata)
1694 {
1695 	int err = 0, i = 0, j = 0;
1696 	struct mlx4_ib_flow *mflow;
1697 	enum mlx4_net_trans_promisc_mode type[2];
1698 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1699 	int is_bonded = mlx4_is_bonded(dev);
1700 
1701 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1702 		return ERR_PTR(-EINVAL);
1703 
1704 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1705 		return ERR_PTR(-EOPNOTSUPP);
1706 
1707 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1708 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1709 		return ERR_PTR(-EOPNOTSUPP);
1710 
1711 	if (udata &&
1712 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1713 		return ERR_PTR(-EOPNOTSUPP);
1714 
1715 	memset(type, 0, sizeof(type));
1716 
1717 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1718 	if (!mflow) {
1719 		err = -ENOMEM;
1720 		goto err_free;
1721 	}
1722 
1723 	switch (flow_attr->type) {
1724 	case IB_FLOW_ATTR_NORMAL:
1725 		/* If dont trap flag (continue match) is set, under specific
1726 		 * condition traffic be replicated to given qp,
1727 		 * without stealing it
1728 		 */
1729 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1730 			err = mlx4_ib_add_dont_trap_rule(dev,
1731 							 flow_attr,
1732 							 type);
1733 			if (err)
1734 				goto err_free;
1735 		} else {
1736 			type[0] = MLX4_FS_REGULAR;
1737 		}
1738 		break;
1739 
1740 	case IB_FLOW_ATTR_ALL_DEFAULT:
1741 		type[0] = MLX4_FS_ALL_DEFAULT;
1742 		break;
1743 
1744 	case IB_FLOW_ATTR_MC_DEFAULT:
1745 		type[0] = MLX4_FS_MC_DEFAULT;
1746 		break;
1747 
1748 	case IB_FLOW_ATTR_SNIFFER:
1749 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1750 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1751 		break;
1752 
1753 	default:
1754 		err = -EINVAL;
1755 		goto err_free;
1756 	}
1757 
1758 	while (i < ARRAY_SIZE(type) && type[i]) {
1759 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1760 					    &mflow->reg_id[i].id);
1761 		if (err)
1762 			goto err_create_flow;
1763 		if (is_bonded) {
1764 			/* Application always sees one port so the mirror rule
1765 			 * must be on port #2
1766 			 */
1767 			flow_attr->port = 2;
1768 			err = __mlx4_ib_create_flow(qp, flow_attr,
1769 						    domain, type[j],
1770 						    &mflow->reg_id[j].mirror);
1771 			flow_attr->port = 1;
1772 			if (err)
1773 				goto err_create_flow;
1774 			j++;
1775 		}
1776 
1777 		i++;
1778 	}
1779 
1780 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1781 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1782 					       &mflow->reg_id[i].id);
1783 		if (err)
1784 			goto err_create_flow;
1785 
1786 		if (is_bonded) {
1787 			flow_attr->port = 2;
1788 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1789 						       &mflow->reg_id[j].mirror);
1790 			flow_attr->port = 1;
1791 			if (err)
1792 				goto err_create_flow;
1793 			j++;
1794 		}
1795 		/* function to create mirror rule */
1796 		i++;
1797 	}
1798 
1799 	return &mflow->ibflow;
1800 
1801 err_create_flow:
1802 	while (i) {
1803 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1804 					     mflow->reg_id[i].id);
1805 		i--;
1806 	}
1807 
1808 	while (j) {
1809 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1810 					     mflow->reg_id[j].mirror);
1811 		j--;
1812 	}
1813 err_free:
1814 	kfree(mflow);
1815 	return ERR_PTR(err);
1816 }
1817 
1818 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1819 {
1820 	int err, ret = 0;
1821 	int i = 0;
1822 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1823 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1824 
1825 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1826 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1827 		if (err)
1828 			ret = err;
1829 		if (mflow->reg_id[i].mirror) {
1830 			err = __mlx4_ib_destroy_flow(mdev->dev,
1831 						     mflow->reg_id[i].mirror);
1832 			if (err)
1833 				ret = err;
1834 		}
1835 		i++;
1836 	}
1837 
1838 	kfree(mflow);
1839 	return ret;
1840 }
1841 
1842 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1843 {
1844 	int err;
1845 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1846 	struct mlx4_dev	*dev = mdev->dev;
1847 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1848 	struct mlx4_ib_steering *ib_steering = NULL;
1849 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1850 	struct mlx4_flow_reg_id	reg_id;
1851 
1852 	if (mdev->dev->caps.steering_mode ==
1853 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1854 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1855 		if (!ib_steering)
1856 			return -ENOMEM;
1857 	}
1858 
1859 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1860 				    !!(mqp->flags &
1861 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1862 				    prot, &reg_id.id);
1863 	if (err) {
1864 		pr_err("multicast attach op failed, err %d\n", err);
1865 		goto err_malloc;
1866 	}
1867 
1868 	reg_id.mirror = 0;
1869 	if (mlx4_is_bonded(dev)) {
1870 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1871 					    (mqp->port == 1) ? 2 : 1,
1872 					    !!(mqp->flags &
1873 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1874 					    prot, &reg_id.mirror);
1875 		if (err)
1876 			goto err_add;
1877 	}
1878 
1879 	err = add_gid_entry(ibqp, gid);
1880 	if (err)
1881 		goto err_add;
1882 
1883 	if (ib_steering) {
1884 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1885 		ib_steering->reg_id = reg_id;
1886 		mutex_lock(&mqp->mutex);
1887 		list_add(&ib_steering->list, &mqp->steering_rules);
1888 		mutex_unlock(&mqp->mutex);
1889 	}
1890 	return 0;
1891 
1892 err_add:
1893 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1894 			      prot, reg_id.id);
1895 	if (reg_id.mirror)
1896 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1897 				      prot, reg_id.mirror);
1898 err_malloc:
1899 	kfree(ib_steering);
1900 
1901 	return err;
1902 }
1903 
1904 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1905 {
1906 	struct mlx4_ib_gid_entry *ge;
1907 	struct mlx4_ib_gid_entry *tmp;
1908 	struct mlx4_ib_gid_entry *ret = NULL;
1909 
1910 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1911 		if (!memcmp(raw, ge->gid.raw, 16)) {
1912 			ret = ge;
1913 			break;
1914 		}
1915 	}
1916 
1917 	return ret;
1918 }
1919 
1920 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1921 {
1922 	int err;
1923 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1924 	struct mlx4_dev *dev = mdev->dev;
1925 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1926 	struct net_device *ndev;
1927 	struct mlx4_ib_gid_entry *ge;
1928 	struct mlx4_flow_reg_id reg_id = {0, 0};
1929 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1930 
1931 	if (mdev->dev->caps.steering_mode ==
1932 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1933 		struct mlx4_ib_steering *ib_steering;
1934 
1935 		mutex_lock(&mqp->mutex);
1936 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1937 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1938 				list_del(&ib_steering->list);
1939 				break;
1940 			}
1941 		}
1942 		mutex_unlock(&mqp->mutex);
1943 		if (&ib_steering->list == &mqp->steering_rules) {
1944 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1945 			return -EINVAL;
1946 		}
1947 		reg_id = ib_steering->reg_id;
1948 		kfree(ib_steering);
1949 	}
1950 
1951 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1952 				    prot, reg_id.id);
1953 	if (err)
1954 		return err;
1955 
1956 	if (mlx4_is_bonded(dev)) {
1957 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1958 					    prot, reg_id.mirror);
1959 		if (err)
1960 			return err;
1961 	}
1962 
1963 	mutex_lock(&mqp->mutex);
1964 	ge = find_gid_entry(mqp, gid->raw);
1965 	if (ge) {
1966 		spin_lock_bh(&mdev->iboe.lock);
1967 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1968 		if (ndev)
1969 			dev_hold(ndev);
1970 		spin_unlock_bh(&mdev->iboe.lock);
1971 		if (ndev)
1972 			dev_put(ndev);
1973 		list_del(&ge->list);
1974 		kfree(ge);
1975 	} else
1976 		pr_warn("could not find mgid entry\n");
1977 
1978 	mutex_unlock(&mqp->mutex);
1979 
1980 	return 0;
1981 }
1982 
1983 static int init_node_data(struct mlx4_ib_dev *dev)
1984 {
1985 	struct ib_smp *in_mad  = NULL;
1986 	struct ib_smp *out_mad = NULL;
1987 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1988 	int err = -ENOMEM;
1989 
1990 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1991 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1992 	if (!in_mad || !out_mad)
1993 		goto out;
1994 
1995 	init_query_mad(in_mad);
1996 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1997 	if (mlx4_is_master(dev->dev))
1998 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1999 
2000 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2001 	if (err)
2002 		goto out;
2003 
2004 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2005 
2006 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2007 
2008 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2009 	if (err)
2010 		goto out;
2011 
2012 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2013 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2014 
2015 out:
2016 	kfree(in_mad);
2017 	kfree(out_mad);
2018 	return err;
2019 }
2020 
2021 static ssize_t hca_type_show(struct device *device,
2022 			     struct device_attribute *attr, char *buf)
2023 {
2024 	struct mlx4_ib_dev *dev =
2025 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2026 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2027 }
2028 static DEVICE_ATTR_RO(hca_type);
2029 
2030 static ssize_t hw_rev_show(struct device *device,
2031 			   struct device_attribute *attr, char *buf)
2032 {
2033 	struct mlx4_ib_dev *dev =
2034 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2035 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2036 }
2037 static DEVICE_ATTR_RO(hw_rev);
2038 
2039 static ssize_t board_id_show(struct device *device,
2040 			     struct device_attribute *attr, char *buf)
2041 {
2042 	struct mlx4_ib_dev *dev =
2043 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2044 
2045 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2046 		       dev->dev->board_id);
2047 }
2048 static DEVICE_ATTR_RO(board_id);
2049 
2050 static struct attribute *mlx4_class_attributes[] = {
2051 	&dev_attr_hw_rev.attr,
2052 	&dev_attr_hca_type.attr,
2053 	&dev_attr_board_id.attr,
2054 	NULL
2055 };
2056 
2057 static const struct attribute_group mlx4_attr_group = {
2058 	.attrs = mlx4_class_attributes,
2059 };
2060 
2061 struct diag_counter {
2062 	const char *name;
2063 	u32 offset;
2064 };
2065 
2066 #define DIAG_COUNTER(_name, _offset)			\
2067 	{ .name = #_name, .offset = _offset }
2068 
2069 static const struct diag_counter diag_basic[] = {
2070 	DIAG_COUNTER(rq_num_lle, 0x00),
2071 	DIAG_COUNTER(sq_num_lle, 0x04),
2072 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2073 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2074 	DIAG_COUNTER(rq_num_lpe, 0x18),
2075 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2076 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2077 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2078 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2079 	DIAG_COUNTER(sq_num_bre, 0x34),
2080 	DIAG_COUNTER(sq_num_rire, 0x44),
2081 	DIAG_COUNTER(rq_num_rire, 0x48),
2082 	DIAG_COUNTER(sq_num_rae, 0x4C),
2083 	DIAG_COUNTER(rq_num_rae, 0x50),
2084 	DIAG_COUNTER(sq_num_roe, 0x54),
2085 	DIAG_COUNTER(sq_num_tree, 0x5C),
2086 	DIAG_COUNTER(sq_num_rree, 0x64),
2087 	DIAG_COUNTER(rq_num_rnr, 0x68),
2088 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2089 	DIAG_COUNTER(rq_num_oos, 0x100),
2090 	DIAG_COUNTER(sq_num_oos, 0x104),
2091 };
2092 
2093 static const struct diag_counter diag_ext[] = {
2094 	DIAG_COUNTER(rq_num_dup, 0x130),
2095 	DIAG_COUNTER(sq_num_to, 0x134),
2096 };
2097 
2098 static const struct diag_counter diag_device_only[] = {
2099 	DIAG_COUNTER(num_cqovf, 0x1A0),
2100 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2101 };
2102 
2103 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2104 						    u8 port_num)
2105 {
2106 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2107 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2108 
2109 	if (!diag[!!port_num].name)
2110 		return NULL;
2111 
2112 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2113 					  diag[!!port_num].num_counters,
2114 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2115 }
2116 
2117 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2118 				struct rdma_hw_stats *stats,
2119 				u8 port, int index)
2120 {
2121 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2122 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2123 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2124 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2125 	int ret;
2126 	int i;
2127 
2128 	ret = mlx4_query_diag_counters(dev->dev,
2129 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2130 				       diag[!!port].offset, hw_value,
2131 				       diag[!!port].num_counters, port);
2132 
2133 	if (ret)
2134 		return ret;
2135 
2136 	for (i = 0; i < diag[!!port].num_counters; i++)
2137 		stats->value[i] = hw_value[i];
2138 
2139 	return diag[!!port].num_counters;
2140 }
2141 
2142 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2143 					 const char ***name,
2144 					 u32 **offset,
2145 					 u32 *num,
2146 					 bool port)
2147 {
2148 	u32 num_counters;
2149 
2150 	num_counters = ARRAY_SIZE(diag_basic);
2151 
2152 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2153 		num_counters += ARRAY_SIZE(diag_ext);
2154 
2155 	if (!port)
2156 		num_counters += ARRAY_SIZE(diag_device_only);
2157 
2158 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2159 	if (!*name)
2160 		return -ENOMEM;
2161 
2162 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2163 	if (!*offset)
2164 		goto err_name;
2165 
2166 	*num = num_counters;
2167 
2168 	return 0;
2169 
2170 err_name:
2171 	kfree(*name);
2172 	return -ENOMEM;
2173 }
2174 
2175 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2176 				       const char **name,
2177 				       u32 *offset,
2178 				       bool port)
2179 {
2180 	int i;
2181 	int j;
2182 
2183 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2184 		name[i] = diag_basic[i].name;
2185 		offset[i] = diag_basic[i].offset;
2186 	}
2187 
2188 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2189 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2190 			name[j] = diag_ext[i].name;
2191 			offset[j] = diag_ext[i].offset;
2192 		}
2193 	}
2194 
2195 	if (!port) {
2196 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2197 			name[j] = diag_device_only[i].name;
2198 			offset[j] = diag_device_only[i].offset;
2199 		}
2200 	}
2201 }
2202 
2203 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2204 	.alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2205 	.get_hw_stats = mlx4_ib_get_hw_stats,
2206 };
2207 
2208 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2209 {
2210 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2211 	int i;
2212 	int ret;
2213 	bool per_port = !!(ibdev->dev->caps.flags2 &
2214 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2215 
2216 	if (mlx4_is_slave(ibdev->dev))
2217 		return 0;
2218 
2219 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2220 		/* i == 1 means we are building port counters */
2221 		if (i && !per_port)
2222 			continue;
2223 
2224 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2225 						    &diag[i].offset,
2226 						    &diag[i].num_counters, i);
2227 		if (ret)
2228 			goto err_alloc;
2229 
2230 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2231 					   diag[i].offset, i);
2232 	}
2233 
2234 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2235 
2236 	return 0;
2237 
2238 err_alloc:
2239 	if (i) {
2240 		kfree(diag[i - 1].name);
2241 		kfree(diag[i - 1].offset);
2242 	}
2243 
2244 	return ret;
2245 }
2246 
2247 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2248 {
2249 	int i;
2250 
2251 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2252 		kfree(ibdev->diag_counters[i].offset);
2253 		kfree(ibdev->diag_counters[i].name);
2254 	}
2255 }
2256 
2257 #define MLX4_IB_INVALID_MAC	((u64)-1)
2258 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2259 			       struct net_device *dev,
2260 			       int port)
2261 {
2262 	u64 new_smac = 0;
2263 	u64 release_mac = MLX4_IB_INVALID_MAC;
2264 	struct mlx4_ib_qp *qp;
2265 
2266 	read_lock(&dev_base_lock);
2267 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2268 	read_unlock(&dev_base_lock);
2269 
2270 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2271 
2272 	/* no need for update QP1 and mac registration in non-SRIOV */
2273 	if (!mlx4_is_mfunc(ibdev->dev))
2274 		return;
2275 
2276 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2277 	qp = ibdev->qp1_proxy[port - 1];
2278 	if (qp) {
2279 		int new_smac_index;
2280 		u64 old_smac;
2281 		struct mlx4_update_qp_params update_params;
2282 
2283 		mutex_lock(&qp->mutex);
2284 		old_smac = qp->pri.smac;
2285 		if (new_smac == old_smac)
2286 			goto unlock;
2287 
2288 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2289 
2290 		if (new_smac_index < 0)
2291 			goto unlock;
2292 
2293 		update_params.smac_index = new_smac_index;
2294 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2295 				   &update_params)) {
2296 			release_mac = new_smac;
2297 			goto unlock;
2298 		}
2299 		/* if old port was zero, no mac was yet registered for this QP */
2300 		if (qp->pri.smac_port)
2301 			release_mac = old_smac;
2302 		qp->pri.smac = new_smac;
2303 		qp->pri.smac_port = port;
2304 		qp->pri.smac_index = new_smac_index;
2305 	}
2306 
2307 unlock:
2308 	if (release_mac != MLX4_IB_INVALID_MAC)
2309 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2310 	if (qp)
2311 		mutex_unlock(&qp->mutex);
2312 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2313 }
2314 
2315 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2316 				 struct net_device *dev,
2317 				 unsigned long event)
2318 
2319 {
2320 	struct mlx4_ib_iboe *iboe;
2321 	int update_qps_port = -1;
2322 	int port;
2323 
2324 	ASSERT_RTNL();
2325 
2326 	iboe = &ibdev->iboe;
2327 
2328 	spin_lock_bh(&iboe->lock);
2329 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2330 
2331 		iboe->netdevs[port - 1] =
2332 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2333 
2334 		if (dev == iboe->netdevs[port - 1] &&
2335 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2336 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2337 			update_qps_port = port;
2338 
2339 		if (dev == iboe->netdevs[port - 1] &&
2340 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2341 			enum ib_port_state port_state;
2342 			struct ib_event ibev = { };
2343 
2344 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2345 						     &port_state))
2346 				continue;
2347 
2348 			if (event == NETDEV_UP &&
2349 			    (port_state != IB_PORT_ACTIVE ||
2350 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2351 				continue;
2352 			if (event == NETDEV_DOWN &&
2353 			    (port_state != IB_PORT_DOWN ||
2354 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2355 				continue;
2356 			iboe->last_port_state[port - 1] = port_state;
2357 
2358 			ibev.device = &ibdev->ib_dev;
2359 			ibev.element.port_num = port;
2360 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2361 							  IB_EVENT_PORT_ERR;
2362 			ib_dispatch_event(&ibev);
2363 		}
2364 
2365 	}
2366 	spin_unlock_bh(&iboe->lock);
2367 
2368 	if (update_qps_port > 0)
2369 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2370 }
2371 
2372 static int mlx4_ib_netdev_event(struct notifier_block *this,
2373 				unsigned long event, void *ptr)
2374 {
2375 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2376 	struct mlx4_ib_dev *ibdev;
2377 
2378 	if (!net_eq(dev_net(dev), &init_net))
2379 		return NOTIFY_DONE;
2380 
2381 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2382 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2383 
2384 	return NOTIFY_DONE;
2385 }
2386 
2387 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2388 {
2389 	int port;
2390 	int slave;
2391 	int i;
2392 
2393 	if (mlx4_is_master(ibdev->dev)) {
2394 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2395 		     ++slave) {
2396 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2397 				for (i = 0;
2398 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2399 				     ++i) {
2400 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2401 					/* master has the identity virt2phys pkey mapping */
2402 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2403 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2404 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2405 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2406 				}
2407 			}
2408 		}
2409 		/* initialize pkey cache */
2410 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2411 			for (i = 0;
2412 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2413 			     ++i)
2414 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2415 					(i) ? 0 : 0xFFFF;
2416 		}
2417 	}
2418 }
2419 
2420 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2421 {
2422 	int i, j, eq = 0, total_eqs = 0;
2423 
2424 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2425 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2426 	if (!ibdev->eq_table)
2427 		return;
2428 
2429 	for (i = 1; i <= dev->caps.num_ports; i++) {
2430 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2431 		     j++, total_eqs++) {
2432 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2433 				continue;
2434 			ibdev->eq_table[eq] = total_eqs;
2435 			if (!mlx4_assign_eq(dev, i,
2436 					    &ibdev->eq_table[eq]))
2437 				eq++;
2438 			else
2439 				ibdev->eq_table[eq] = -1;
2440 		}
2441 	}
2442 
2443 	for (i = eq; i < dev->caps.num_comp_vectors;
2444 	     ibdev->eq_table[i++] = -1)
2445 		;
2446 
2447 	/* Advertise the new number of EQs to clients */
2448 	ibdev->ib_dev.num_comp_vectors = eq;
2449 }
2450 
2451 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2452 {
2453 	int i;
2454 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2455 
2456 	/* no eqs were allocated */
2457 	if (!ibdev->eq_table)
2458 		return;
2459 
2460 	/* Reset the advertised EQ number */
2461 	ibdev->ib_dev.num_comp_vectors = 0;
2462 
2463 	for (i = 0; i < total_eqs; i++)
2464 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2465 
2466 	kfree(ibdev->eq_table);
2467 	ibdev->eq_table = NULL;
2468 }
2469 
2470 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2471 			       struct ib_port_immutable *immutable)
2472 {
2473 	struct ib_port_attr attr;
2474 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2475 	int err;
2476 
2477 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2478 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2479 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2480 	} else {
2481 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2482 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2483 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2484 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2485 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2486 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2487 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2488 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2489 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2490 	}
2491 
2492 	err = ib_query_port(ibdev, port_num, &attr);
2493 	if (err)
2494 		return err;
2495 
2496 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2497 	immutable->gid_tbl_len = attr.gid_tbl_len;
2498 
2499 	return 0;
2500 }
2501 
2502 static void get_fw_ver_str(struct ib_device *device, char *str)
2503 {
2504 	struct mlx4_ib_dev *dev =
2505 		container_of(device, struct mlx4_ib_dev, ib_dev);
2506 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2507 		 (int) (dev->dev->caps.fw_ver >> 32),
2508 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2509 		 (int) dev->dev->caps.fw_ver & 0xffff);
2510 }
2511 
2512 static const struct ib_device_ops mlx4_ib_dev_ops = {
2513 	.owner = THIS_MODULE,
2514 	.driver_id = RDMA_DRIVER_MLX4,
2515 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2516 
2517 	.add_gid = mlx4_ib_add_gid,
2518 	.alloc_mr = mlx4_ib_alloc_mr,
2519 	.alloc_pd = mlx4_ib_alloc_pd,
2520 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2521 	.attach_mcast = mlx4_ib_mcg_attach,
2522 	.create_ah = mlx4_ib_create_ah,
2523 	.create_cq = mlx4_ib_create_cq,
2524 	.create_qp = mlx4_ib_create_qp,
2525 	.create_srq = mlx4_ib_create_srq,
2526 	.dealloc_pd = mlx4_ib_dealloc_pd,
2527 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2528 	.del_gid = mlx4_ib_del_gid,
2529 	.dereg_mr = mlx4_ib_dereg_mr,
2530 	.destroy_ah = mlx4_ib_destroy_ah,
2531 	.destroy_cq = mlx4_ib_destroy_cq,
2532 	.destroy_qp = mlx4_ib_destroy_qp,
2533 	.destroy_srq = mlx4_ib_destroy_srq,
2534 	.detach_mcast = mlx4_ib_mcg_detach,
2535 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2536 	.drain_rq = mlx4_ib_drain_rq,
2537 	.drain_sq = mlx4_ib_drain_sq,
2538 	.get_dev_fw_str = get_fw_ver_str,
2539 	.get_dma_mr = mlx4_ib_get_dma_mr,
2540 	.get_link_layer = mlx4_ib_port_link_layer,
2541 	.get_netdev = mlx4_ib_get_netdev,
2542 	.get_port_immutable = mlx4_port_immutable,
2543 	.map_mr_sg = mlx4_ib_map_mr_sg,
2544 	.mmap = mlx4_ib_mmap,
2545 	.modify_cq = mlx4_ib_modify_cq,
2546 	.modify_device = mlx4_ib_modify_device,
2547 	.modify_port = mlx4_ib_modify_port,
2548 	.modify_qp = mlx4_ib_modify_qp,
2549 	.modify_srq = mlx4_ib_modify_srq,
2550 	.poll_cq = mlx4_ib_poll_cq,
2551 	.post_recv = mlx4_ib_post_recv,
2552 	.post_send = mlx4_ib_post_send,
2553 	.post_srq_recv = mlx4_ib_post_srq_recv,
2554 	.process_mad = mlx4_ib_process_mad,
2555 	.query_ah = mlx4_ib_query_ah,
2556 	.query_device = mlx4_ib_query_device,
2557 	.query_gid = mlx4_ib_query_gid,
2558 	.query_pkey = mlx4_ib_query_pkey,
2559 	.query_port = mlx4_ib_query_port,
2560 	.query_qp = mlx4_ib_query_qp,
2561 	.query_srq = mlx4_ib_query_srq,
2562 	.reg_user_mr = mlx4_ib_reg_user_mr,
2563 	.req_notify_cq = mlx4_ib_arm_cq,
2564 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2565 	.resize_cq = mlx4_ib_resize_cq,
2566 
2567 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2568 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2569 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2570 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2571 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2572 };
2573 
2574 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2575 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2576 	.create_wq = mlx4_ib_create_wq,
2577 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2578 	.destroy_wq = mlx4_ib_destroy_wq,
2579 	.modify_wq = mlx4_ib_modify_wq,
2580 };
2581 
2582 static const struct ib_device_ops mlx4_ib_dev_fmr_ops = {
2583 	.alloc_fmr = mlx4_ib_fmr_alloc,
2584 	.dealloc_fmr = mlx4_ib_fmr_dealloc,
2585 	.map_phys_fmr = mlx4_ib_map_phys_fmr,
2586 	.unmap_fmr = mlx4_ib_unmap_fmr,
2587 };
2588 
2589 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2590 	.alloc_mw = mlx4_ib_alloc_mw,
2591 	.dealloc_mw = mlx4_ib_dealloc_mw,
2592 };
2593 
2594 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2595 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2596 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2597 };
2598 
2599 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2600 	.create_flow = mlx4_ib_create_flow,
2601 	.destroy_flow = mlx4_ib_destroy_flow,
2602 };
2603 
2604 static void *mlx4_ib_add(struct mlx4_dev *dev)
2605 {
2606 	struct mlx4_ib_dev *ibdev;
2607 	int num_ports = 0;
2608 	int i, j;
2609 	int err;
2610 	struct mlx4_ib_iboe *iboe;
2611 	int ib_num_ports = 0;
2612 	int num_req_counters;
2613 	int allocated;
2614 	u32 counter_index;
2615 	struct counter_index *new_counter_index = NULL;
2616 
2617 	pr_info_once("%s", mlx4_ib_version);
2618 
2619 	num_ports = 0;
2620 	mlx4_foreach_ib_transport_port(i, dev)
2621 		num_ports++;
2622 
2623 	/* No point in registering a device with no ports... */
2624 	if (num_ports == 0)
2625 		return NULL;
2626 
2627 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2628 	if (!ibdev) {
2629 		dev_err(&dev->persist->pdev->dev,
2630 			"Device struct alloc failed\n");
2631 		return NULL;
2632 	}
2633 
2634 	iboe = &ibdev->iboe;
2635 
2636 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2637 		goto err_dealloc;
2638 
2639 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2640 		goto err_pd;
2641 
2642 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2643 				 PAGE_SIZE);
2644 	if (!ibdev->uar_map)
2645 		goto err_uar;
2646 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2647 
2648 	ibdev->dev = dev;
2649 	ibdev->bond_next_port	= 0;
2650 
2651 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2652 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2653 	ibdev->num_ports		= num_ports;
2654 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2655 						1 : ibdev->num_ports;
2656 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2657 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2658 
2659 	ibdev->ib_dev.uverbs_cmd_mask	=
2660 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2661 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2662 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2663 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2664 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2665 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2666 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2667 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2668 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2669 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2670 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2671 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2672 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2673 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2674 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2675 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2676 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2677 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2678 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2679 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2680 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2681 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2682 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2683 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2684 
2685 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2686 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2687 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2688 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2689 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2690 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2691 
2692 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2693 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2694 	    IB_LINK_LAYER_ETHERNET) ||
2695 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2696 	    IB_LINK_LAYER_ETHERNET))) {
2697 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2698 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2699 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2700 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2701 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2702 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2703 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2704 	}
2705 
2706 	if (!mlx4_is_slave(ibdev->dev))
2707 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fmr_ops);
2708 
2709 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2710 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2711 		ibdev->ib_dev.uverbs_cmd_mask |=
2712 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2713 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2714 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2715 	}
2716 
2717 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2718 		ibdev->ib_dev.uverbs_cmd_mask |=
2719 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2720 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2721 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2722 	}
2723 
2724 	if (check_flow_steering_support(dev)) {
2725 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2726 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2727 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2728 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2729 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2730 	}
2731 
2732 	if (!dev->caps.userspace_caps)
2733 		ibdev->ib_dev.ops.uverbs_abi_ver =
2734 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2735 
2736 	mlx4_ib_alloc_eqs(dev, ibdev);
2737 
2738 	spin_lock_init(&iboe->lock);
2739 
2740 	if (init_node_data(ibdev))
2741 		goto err_map;
2742 	mlx4_init_sl2vl_tbl(ibdev);
2743 
2744 	for (i = 0; i < ibdev->num_ports; ++i) {
2745 		mutex_init(&ibdev->counters_table[i].mutex);
2746 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2747 		iboe->last_port_state[i] = IB_PORT_DOWN;
2748 	}
2749 
2750 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2751 	for (i = 0; i < num_req_counters; ++i) {
2752 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2753 		allocated = 0;
2754 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2755 						IB_LINK_LAYER_ETHERNET) {
2756 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2757 						 MLX4_RES_USAGE_DRIVER);
2758 			/* if failed to allocate a new counter, use default */
2759 			if (err)
2760 				counter_index =
2761 					mlx4_get_default_counter_index(dev,
2762 								       i + 1);
2763 			else
2764 				allocated = 1;
2765 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2766 			counter_index = mlx4_get_default_counter_index(dev,
2767 								       i + 1);
2768 		}
2769 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2770 					    GFP_KERNEL);
2771 		if (!new_counter_index) {
2772 			if (allocated)
2773 				mlx4_counter_free(ibdev->dev, counter_index);
2774 			goto err_counter;
2775 		}
2776 		new_counter_index->index = counter_index;
2777 		new_counter_index->allocated = allocated;
2778 		list_add_tail(&new_counter_index->list,
2779 			      &ibdev->counters_table[i].counters_list);
2780 		ibdev->counters_table[i].default_counter = counter_index;
2781 		pr_info("counter index %d for port %d allocated %d\n",
2782 			counter_index, i + 1, allocated);
2783 	}
2784 	if (mlx4_is_bonded(dev))
2785 		for (i = 1; i < ibdev->num_ports ; ++i) {
2786 			new_counter_index =
2787 					kmalloc(sizeof(struct counter_index),
2788 						GFP_KERNEL);
2789 			if (!new_counter_index)
2790 				goto err_counter;
2791 			new_counter_index->index = counter_index;
2792 			new_counter_index->allocated = 0;
2793 			list_add_tail(&new_counter_index->list,
2794 				      &ibdev->counters_table[i].counters_list);
2795 			ibdev->counters_table[i].default_counter =
2796 								counter_index;
2797 		}
2798 
2799 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2800 		ib_num_ports++;
2801 
2802 	spin_lock_init(&ibdev->sm_lock);
2803 	mutex_init(&ibdev->cap_mask_mutex);
2804 	INIT_LIST_HEAD(&ibdev->qp_list);
2805 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2806 
2807 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2808 	    ib_num_ports) {
2809 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2810 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2811 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2812 					    &ibdev->steer_qpn_base, 0,
2813 					    MLX4_RES_USAGE_DRIVER);
2814 		if (err)
2815 			goto err_counter;
2816 
2817 		ibdev->ib_uc_qpns_bitmap =
2818 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2819 				      sizeof(long),
2820 				      GFP_KERNEL);
2821 		if (!ibdev->ib_uc_qpns_bitmap)
2822 			goto err_steer_qp_release;
2823 
2824 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2825 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2826 				    ibdev->steer_qpn_count);
2827 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2828 					dev, ibdev->steer_qpn_base,
2829 					ibdev->steer_qpn_base +
2830 					ibdev->steer_qpn_count - 1);
2831 			if (err)
2832 				goto err_steer_free_bitmap;
2833 		} else {
2834 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2835 				    ibdev->steer_qpn_count);
2836 		}
2837 	}
2838 
2839 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2840 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2841 
2842 	if (mlx4_ib_alloc_diag_counters(ibdev))
2843 		goto err_steer_free_bitmap;
2844 
2845 	rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2846 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d"))
2847 		goto err_diag_counters;
2848 
2849 	if (mlx4_ib_mad_init(ibdev))
2850 		goto err_reg;
2851 
2852 	if (mlx4_ib_init_sriov(ibdev))
2853 		goto err_mad;
2854 
2855 	if (!iboe->nb.notifier_call) {
2856 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2857 		err = register_netdevice_notifier(&iboe->nb);
2858 		if (err) {
2859 			iboe->nb.notifier_call = NULL;
2860 			goto err_notif;
2861 		}
2862 	}
2863 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2864 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2865 		if (err)
2866 			goto err_notif;
2867 	}
2868 
2869 	ibdev->ib_active = true;
2870 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2871 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2872 					 &ibdev->ib_dev);
2873 
2874 	if (mlx4_is_mfunc(ibdev->dev))
2875 		init_pkeys(ibdev);
2876 
2877 	/* create paravirt contexts for any VFs which are active */
2878 	if (mlx4_is_master(ibdev->dev)) {
2879 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2880 			if (j == mlx4_master_func_num(ibdev->dev))
2881 				continue;
2882 			if (mlx4_is_slave_active(ibdev->dev, j))
2883 				do_slave_init(ibdev, j, 1);
2884 		}
2885 	}
2886 	return ibdev;
2887 
2888 err_notif:
2889 	if (ibdev->iboe.nb.notifier_call) {
2890 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2891 			pr_warn("failure unregistering notifier\n");
2892 		ibdev->iboe.nb.notifier_call = NULL;
2893 	}
2894 	flush_workqueue(wq);
2895 
2896 	mlx4_ib_close_sriov(ibdev);
2897 
2898 err_mad:
2899 	mlx4_ib_mad_cleanup(ibdev);
2900 
2901 err_reg:
2902 	ib_unregister_device(&ibdev->ib_dev);
2903 
2904 err_diag_counters:
2905 	mlx4_ib_diag_cleanup(ibdev);
2906 
2907 err_steer_free_bitmap:
2908 	kfree(ibdev->ib_uc_qpns_bitmap);
2909 
2910 err_steer_qp_release:
2911 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2912 			      ibdev->steer_qpn_count);
2913 err_counter:
2914 	for (i = 0; i < ibdev->num_ports; ++i)
2915 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2916 
2917 err_map:
2918 	mlx4_ib_free_eqs(dev, ibdev);
2919 	iounmap(ibdev->uar_map);
2920 
2921 err_uar:
2922 	mlx4_uar_free(dev, &ibdev->priv_uar);
2923 
2924 err_pd:
2925 	mlx4_pd_free(dev, ibdev->priv_pdn);
2926 
2927 err_dealloc:
2928 	ib_dealloc_device(&ibdev->ib_dev);
2929 
2930 	return NULL;
2931 }
2932 
2933 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2934 {
2935 	int offset;
2936 
2937 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2938 
2939 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2940 					 dev->steer_qpn_count,
2941 					 get_count_order(count));
2942 	if (offset < 0)
2943 		return offset;
2944 
2945 	*qpn = dev->steer_qpn_base + offset;
2946 	return 0;
2947 }
2948 
2949 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2950 {
2951 	if (!qpn ||
2952 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2953 		return;
2954 
2955 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2956 		 qpn, dev->steer_qpn_base))
2957 		/* not supposed to be here */
2958 		return;
2959 
2960 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2961 			      qpn - dev->steer_qpn_base,
2962 			      get_count_order(count));
2963 }
2964 
2965 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2966 			 int is_attach)
2967 {
2968 	int err;
2969 	size_t flow_size;
2970 	struct ib_flow_attr *flow = NULL;
2971 	struct ib_flow_spec_ib *ib_spec;
2972 
2973 	if (is_attach) {
2974 		flow_size = sizeof(struct ib_flow_attr) +
2975 			    sizeof(struct ib_flow_spec_ib);
2976 		flow = kzalloc(flow_size, GFP_KERNEL);
2977 		if (!flow)
2978 			return -ENOMEM;
2979 		flow->port = mqp->port;
2980 		flow->num_of_specs = 1;
2981 		flow->size = flow_size;
2982 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2983 		ib_spec->type = IB_FLOW_SPEC_IB;
2984 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2985 		/* Add an empty rule for IB L2 */
2986 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2987 
2988 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2989 					    IB_FLOW_DOMAIN_NIC,
2990 					    MLX4_FS_REGULAR,
2991 					    &mqp->reg_id);
2992 	} else {
2993 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2994 	}
2995 	kfree(flow);
2996 	return err;
2997 }
2998 
2999 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3000 {
3001 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3002 	int p;
3003 	int i;
3004 
3005 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3006 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3007 	ibdev->ib_active = false;
3008 	flush_workqueue(wq);
3009 
3010 	mlx4_ib_close_sriov(ibdev);
3011 	mlx4_ib_mad_cleanup(ibdev);
3012 	ib_unregister_device(&ibdev->ib_dev);
3013 	mlx4_ib_diag_cleanup(ibdev);
3014 	if (ibdev->iboe.nb.notifier_call) {
3015 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3016 			pr_warn("failure unregistering notifier\n");
3017 		ibdev->iboe.nb.notifier_call = NULL;
3018 	}
3019 
3020 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3021 			      ibdev->steer_qpn_count);
3022 	kfree(ibdev->ib_uc_qpns_bitmap);
3023 
3024 	iounmap(ibdev->uar_map);
3025 	for (p = 0; p < ibdev->num_ports; ++p)
3026 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3027 
3028 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3029 		mlx4_CLOSE_PORT(dev, p);
3030 
3031 	mlx4_ib_free_eqs(dev, ibdev);
3032 
3033 	mlx4_uar_free(dev, &ibdev->priv_uar);
3034 	mlx4_pd_free(dev, ibdev->priv_pdn);
3035 	ib_dealloc_device(&ibdev->ib_dev);
3036 }
3037 
3038 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3039 {
3040 	struct mlx4_ib_demux_work **dm = NULL;
3041 	struct mlx4_dev *dev = ibdev->dev;
3042 	int i;
3043 	unsigned long flags;
3044 	struct mlx4_active_ports actv_ports;
3045 	unsigned int ports;
3046 	unsigned int first_port;
3047 
3048 	if (!mlx4_is_master(dev))
3049 		return;
3050 
3051 	actv_ports = mlx4_get_active_ports(dev, slave);
3052 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3053 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3054 
3055 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3056 	if (!dm)
3057 		return;
3058 
3059 	for (i = 0; i < ports; i++) {
3060 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3061 		if (!dm[i]) {
3062 			while (--i >= 0)
3063 				kfree(dm[i]);
3064 			goto out;
3065 		}
3066 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3067 		dm[i]->port = first_port + i + 1;
3068 		dm[i]->slave = slave;
3069 		dm[i]->do_init = do_init;
3070 		dm[i]->dev = ibdev;
3071 	}
3072 	/* initialize or tear down tunnel QPs for the slave */
3073 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3074 	if (!ibdev->sriov.is_going_down) {
3075 		for (i = 0; i < ports; i++)
3076 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3077 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3078 	} else {
3079 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3080 		for (i = 0; i < ports; i++)
3081 			kfree(dm[i]);
3082 	}
3083 out:
3084 	kfree(dm);
3085 	return;
3086 }
3087 
3088 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3089 {
3090 	struct mlx4_ib_qp *mqp;
3091 	unsigned long flags_qp;
3092 	unsigned long flags_cq;
3093 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3094 	struct list_head    cq_notify_list;
3095 	struct mlx4_cq *mcq;
3096 	unsigned long flags;
3097 
3098 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3099 	INIT_LIST_HEAD(&cq_notify_list);
3100 
3101 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3102 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3103 
3104 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3105 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3106 		if (mqp->sq.tail != mqp->sq.head) {
3107 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3108 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3109 			if (send_mcq->mcq.comp &&
3110 			    mqp->ibqp.send_cq->comp_handler) {
3111 				if (!send_mcq->mcq.reset_notify_added) {
3112 					send_mcq->mcq.reset_notify_added = 1;
3113 					list_add_tail(&send_mcq->mcq.reset_notify,
3114 						      &cq_notify_list);
3115 				}
3116 			}
3117 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3118 		}
3119 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3120 		/* Now, handle the QP's receive queue */
3121 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3122 		/* no handling is needed for SRQ */
3123 		if (!mqp->ibqp.srq) {
3124 			if (mqp->rq.tail != mqp->rq.head) {
3125 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3126 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3127 				if (recv_mcq->mcq.comp &&
3128 				    mqp->ibqp.recv_cq->comp_handler) {
3129 					if (!recv_mcq->mcq.reset_notify_added) {
3130 						recv_mcq->mcq.reset_notify_added = 1;
3131 						list_add_tail(&recv_mcq->mcq.reset_notify,
3132 							      &cq_notify_list);
3133 					}
3134 				}
3135 				spin_unlock_irqrestore(&recv_mcq->lock,
3136 						       flags_cq);
3137 			}
3138 		}
3139 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3140 	}
3141 
3142 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3143 		mcq->comp(mcq);
3144 	}
3145 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3146 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3147 }
3148 
3149 static void handle_bonded_port_state_event(struct work_struct *work)
3150 {
3151 	struct ib_event_work *ew =
3152 		container_of(work, struct ib_event_work, work);
3153 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3154 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3155 	int i;
3156 	struct ib_event ibev;
3157 
3158 	kfree(ew);
3159 	spin_lock_bh(&ibdev->iboe.lock);
3160 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3161 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3162 		enum ib_port_state curr_port_state;
3163 
3164 		if (!curr_netdev)
3165 			continue;
3166 
3167 		curr_port_state =
3168 			(netif_running(curr_netdev) &&
3169 			 netif_carrier_ok(curr_netdev)) ?
3170 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3171 
3172 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3173 			curr_port_state : IB_PORT_ACTIVE;
3174 	}
3175 	spin_unlock_bh(&ibdev->iboe.lock);
3176 
3177 	ibev.device = &ibdev->ib_dev;
3178 	ibev.element.port_num = 1;
3179 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3180 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3181 
3182 	ib_dispatch_event(&ibev);
3183 }
3184 
3185 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3186 {
3187 	u64 sl2vl;
3188 	int err;
3189 
3190 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3191 	if (err) {
3192 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3193 		       port, err);
3194 		sl2vl = 0;
3195 	}
3196 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3197 }
3198 
3199 static void ib_sl2vl_update_work(struct work_struct *work)
3200 {
3201 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3202 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3203 	int port = ew->port;
3204 
3205 	mlx4_ib_sl2vl_update(mdev, port);
3206 
3207 	kfree(ew);
3208 }
3209 
3210 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3211 				     int port)
3212 {
3213 	struct ib_event_work *ew;
3214 
3215 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3216 	if (ew) {
3217 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3218 		ew->port = port;
3219 		ew->ib_dev = ibdev;
3220 		queue_work(wq, &ew->work);
3221 	}
3222 }
3223 
3224 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3225 			  enum mlx4_dev_event event, unsigned long param)
3226 {
3227 	struct ib_event ibev;
3228 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3229 	struct mlx4_eqe *eqe = NULL;
3230 	struct ib_event_work *ew;
3231 	int p = 0;
3232 
3233 	if (mlx4_is_bonded(dev) &&
3234 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3235 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3236 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3237 		if (!ew)
3238 			return;
3239 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3240 		ew->ib_dev = ibdev;
3241 		queue_work(wq, &ew->work);
3242 		return;
3243 	}
3244 
3245 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3246 		eqe = (struct mlx4_eqe *)param;
3247 	else
3248 		p = (int) param;
3249 
3250 	switch (event) {
3251 	case MLX4_DEV_EVENT_PORT_UP:
3252 		if (p > ibdev->num_ports)
3253 			return;
3254 		if (!mlx4_is_slave(dev) &&
3255 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3256 			IB_LINK_LAYER_INFINIBAND) {
3257 			if (mlx4_is_master(dev))
3258 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3259 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3260 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3261 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3262 		}
3263 		ibev.event = IB_EVENT_PORT_ACTIVE;
3264 		break;
3265 
3266 	case MLX4_DEV_EVENT_PORT_DOWN:
3267 		if (p > ibdev->num_ports)
3268 			return;
3269 		ibev.event = IB_EVENT_PORT_ERR;
3270 		break;
3271 
3272 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3273 		ibdev->ib_active = false;
3274 		ibev.event = IB_EVENT_DEVICE_FATAL;
3275 		mlx4_ib_handle_catas_error(ibdev);
3276 		break;
3277 
3278 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3279 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3280 		if (!ew)
3281 			break;
3282 
3283 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3284 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3285 		ew->ib_dev = ibdev;
3286 		/* need to queue only for port owner, which uses GEN_EQE */
3287 		if (mlx4_is_master(dev))
3288 			queue_work(wq, &ew->work);
3289 		else
3290 			handle_port_mgmt_change_event(&ew->work);
3291 		return;
3292 
3293 	case MLX4_DEV_EVENT_SLAVE_INIT:
3294 		/* here, p is the slave id */
3295 		do_slave_init(ibdev, p, 1);
3296 		if (mlx4_is_master(dev)) {
3297 			int i;
3298 
3299 			for (i = 1; i <= ibdev->num_ports; i++) {
3300 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3301 					== IB_LINK_LAYER_INFINIBAND)
3302 					mlx4_ib_slave_alias_guid_event(ibdev,
3303 								       p, i,
3304 								       1);
3305 			}
3306 		}
3307 		return;
3308 
3309 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3310 		if (mlx4_is_master(dev)) {
3311 			int i;
3312 
3313 			for (i = 1; i <= ibdev->num_ports; i++) {
3314 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3315 					== IB_LINK_LAYER_INFINIBAND)
3316 					mlx4_ib_slave_alias_guid_event(ibdev,
3317 								       p, i,
3318 								       0);
3319 			}
3320 		}
3321 		/* here, p is the slave id */
3322 		do_slave_init(ibdev, p, 0);
3323 		return;
3324 
3325 	default:
3326 		return;
3327 	}
3328 
3329 	ibev.device	      = ibdev_ptr;
3330 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3331 
3332 	ib_dispatch_event(&ibev);
3333 }
3334 
3335 static struct mlx4_interface mlx4_ib_interface = {
3336 	.add		= mlx4_ib_add,
3337 	.remove		= mlx4_ib_remove,
3338 	.event		= mlx4_ib_event,
3339 	.protocol	= MLX4_PROT_IB_IPV6,
3340 	.flags		= MLX4_INTFF_BONDING
3341 };
3342 
3343 static int __init mlx4_ib_init(void)
3344 {
3345 	int err;
3346 
3347 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3348 	if (!wq)
3349 		return -ENOMEM;
3350 
3351 	err = mlx4_ib_mcg_init();
3352 	if (err)
3353 		goto clean_wq;
3354 
3355 	err = mlx4_register_interface(&mlx4_ib_interface);
3356 	if (err)
3357 		goto clean_mcg;
3358 
3359 	return 0;
3360 
3361 clean_mcg:
3362 	mlx4_ib_mcg_destroy();
3363 
3364 clean_wq:
3365 	destroy_workqueue(wq);
3366 	return err;
3367 }
3368 
3369 static void __exit mlx4_ib_cleanup(void)
3370 {
3371 	mlx4_unregister_interface(&mlx4_ib_interface);
3372 	mlx4_ib_mcg_destroy();
3373 	destroy_workqueue(wq);
3374 }
3375 
3376 module_init(mlx4_ib_init);
3377 module_exit(mlx4_ib_cleanup);
3378