xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision 604caebc7f069aef602bc75c5f0e6cf7a3ec456b)
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 #include <rdma/uverbs_ioctl.h>
54 
55 #include <net/bonding.h>
56 
57 #include <linux/mlx4/driver.h>
58 #include <linux/mlx4/cmd.h>
59 #include <linux/mlx4/qp.h>
60 
61 #include "mlx4_ib.h"
62 #include <rdma/mlx4-abi.h>
63 
64 #define DRV_NAME	MLX4_IB_DRV_NAME
65 #define DRV_VERSION	"4.0-0"
66 
67 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
68 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
69 #define MLX4_IB_CARD_REV_A0   0xA0
70 
71 MODULE_AUTHOR("Roland Dreier");
72 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
73 MODULE_LICENSE("Dual BSD/GPL");
74 
75 int mlx4_ib_sm_guid_assign = 0;
76 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
77 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
78 
79 static const char mlx4_ib_version[] =
80 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
81 	DRV_VERSION "\n";
82 
83 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
84 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
85 						    u32 port_num);
86 static int mlx4_ib_event(struct notifier_block *this, unsigned long event,
87 			 void *param);
88 
89 static struct workqueue_struct *wq;
90 
91 static int check_flow_steering_support(struct mlx4_dev *dev)
92 {
93 	int eth_num_ports = 0;
94 	int ib_num_ports = 0;
95 
96 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
97 
98 	if (dmfs) {
99 		int i;
100 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
101 			eth_num_ports++;
102 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
103 			ib_num_ports++;
104 		dmfs &= (!ib_num_ports ||
105 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
106 			(!eth_num_ports ||
107 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
108 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
109 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
110 			dmfs = 0;
111 		}
112 	}
113 	return dmfs;
114 }
115 
116 static int num_ib_ports(struct mlx4_dev *dev)
117 {
118 	int ib_ports = 0;
119 	int i;
120 
121 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
122 		ib_ports++;
123 
124 	return ib_ports;
125 }
126 
127 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device,
128 					     u32 port_num)
129 {
130 	struct mlx4_ib_dev *ibdev = to_mdev(device);
131 	struct net_device *dev, *ret = NULL;
132 
133 	rcu_read_lock();
134 	for_each_netdev_rcu(&init_net, dev) {
135 		if (dev->dev.parent != ibdev->ib_dev.dev.parent ||
136 		    dev->dev_port + 1 != port_num)
137 			continue;
138 
139 		if (mlx4_is_bonded(ibdev->dev)) {
140 			struct net_device *upper;
141 
142 			upper = netdev_master_upper_dev_get_rcu(dev);
143 			if (upper) {
144 				struct net_device *active;
145 
146 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
147 				if (active)
148 					dev = active;
149 			}
150 		}
151 
152 		dev_hold(dev);
153 		ret = dev;
154 		break;
155 	}
156 
157 	rcu_read_unlock();
158 	return ret;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u32 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 				     u32 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 			       u32 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static void free_gid_entry(struct gid_entry *entry)
250 {
251 	memset(&entry->gid, 0, sizeof(entry->gid));
252 	kfree(entry->ctx);
253 	entry->ctx = NULL;
254 }
255 
256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids;
266 	u16 vlan_id = 0xffff;
267 	u8 mac[ETH_ALEN];
268 
269 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
270 		return -EINVAL;
271 
272 	if (attr->port_num > MLX4_MAX_PORTS)
273 		return -EINVAL;
274 
275 	if (!context)
276 		return -EINVAL;
277 
278 	ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
279 	if (ret)
280 		return ret;
281 	port_gid_table = &iboe->gids[attr->port_num - 1];
282 	spin_lock_bh(&iboe->lock);
283 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
284 		if (!memcmp(&port_gid_table->gids[i].gid,
285 			    &attr->gid, sizeof(attr->gid)) &&
286 		    port_gid_table->gids[i].gid_type == attr->gid_type &&
287 		    port_gid_table->gids[i].vlan_id == vlan_id)  {
288 			found = i;
289 			break;
290 		}
291 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
292 			free = i; /* HW has space */
293 	}
294 
295 	if (found < 0) {
296 		if (free < 0) {
297 			ret = -ENOSPC;
298 		} else {
299 			port_gid_table->gids[free].ctx = kmalloc_obj(*port_gid_table->gids[free].ctx,
300 								     GFP_ATOMIC);
301 			if (!port_gid_table->gids[free].ctx) {
302 				ret = -ENOMEM;
303 			} else {
304 				*context = port_gid_table->gids[free].ctx;
305 				port_gid_table->gids[free].gid = attr->gid;
306 				port_gid_table->gids[free].gid_type = attr->gid_type;
307 				port_gid_table->gids[free].vlan_id = vlan_id;
308 				port_gid_table->gids[free].ctx->real_index = free;
309 				port_gid_table->gids[free].ctx->refcount = 1;
310 				hw_update = 1;
311 			}
312 		}
313 	} else {
314 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
315 		*context = ctx;
316 		ctx->refcount++;
317 	}
318 	if (!ret && hw_update) {
319 		gids = kmalloc_objs(*gids, MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
320 		if (!gids) {
321 			ret = -ENOMEM;
322 			*context = NULL;
323 			free_gid_entry(&port_gid_table->gids[free]);
324 		} else {
325 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
326 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
327 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
328 			}
329 		}
330 	}
331 	spin_unlock_bh(&iboe->lock);
332 
333 	if (!ret && hw_update) {
334 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
335 		if (ret) {
336 			spin_lock_bh(&iboe->lock);
337 			*context = NULL;
338 			free_gid_entry(&port_gid_table->gids[free]);
339 			spin_unlock_bh(&iboe->lock);
340 		}
341 		kfree(gids);
342 	}
343 
344 	return ret;
345 }
346 
347 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
348 {
349 	struct gid_cache_context *ctx = *context;
350 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
351 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
352 	struct mlx4_port_gid_table   *port_gid_table;
353 	int ret = 0;
354 	int hw_update = 0;
355 	struct gid_entry *gids = NULL;
356 
357 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
358 		return -EINVAL;
359 
360 	if (attr->port_num > MLX4_MAX_PORTS)
361 		return -EINVAL;
362 
363 	port_gid_table = &iboe->gids[attr->port_num - 1];
364 	spin_lock_bh(&iboe->lock);
365 	if (ctx) {
366 		ctx->refcount--;
367 		if (!ctx->refcount) {
368 			unsigned int real_index = ctx->real_index;
369 
370 			free_gid_entry(&port_gid_table->gids[real_index]);
371 			hw_update = 1;
372 		}
373 	}
374 	if (!ret && hw_update) {
375 		int i;
376 
377 		gids = kmalloc_objs(*gids, MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
378 		if (!gids) {
379 			ret = -ENOMEM;
380 		} else {
381 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
382 				memcpy(&gids[i].gid,
383 				       &port_gid_table->gids[i].gid,
384 				       sizeof(union ib_gid));
385 				gids[i].gid_type =
386 				    port_gid_table->gids[i].gid_type;
387 			}
388 		}
389 	}
390 	spin_unlock_bh(&iboe->lock);
391 
392 	if (gids)
393 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
394 
395 	kfree(gids);
396 	return ret;
397 }
398 
399 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
400 				    const struct ib_gid_attr *attr)
401 {
402 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
403 	struct gid_cache_context *ctx = NULL;
404 	struct mlx4_port_gid_table   *port_gid_table;
405 	int real_index = -EINVAL;
406 	int i;
407 	unsigned long flags;
408 	u32 port_num = attr->port_num;
409 
410 	if (port_num > MLX4_MAX_PORTS)
411 		return -EINVAL;
412 
413 	if (mlx4_is_bonded(ibdev->dev))
414 		port_num = 1;
415 
416 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
417 		return attr->index;
418 
419 	spin_lock_irqsave(&iboe->lock, flags);
420 	port_gid_table = &iboe->gids[port_num - 1];
421 
422 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
423 		if (!memcmp(&port_gid_table->gids[i].gid,
424 			    &attr->gid, sizeof(attr->gid)) &&
425 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
426 			ctx = port_gid_table->gids[i].ctx;
427 			break;
428 		}
429 	if (ctx)
430 		real_index = ctx->real_index;
431 	spin_unlock_irqrestore(&iboe->lock, flags);
432 	return real_index;
433 }
434 
435 static int mlx4_ib_query_device(struct ib_device *ibdev,
436 				struct ib_device_attr *props,
437 				struct ib_udata *uhw)
438 {
439 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
440 	struct ib_smp *in_mad;
441 	struct ib_smp *out_mad;
442 	int err;
443 	int have_ib_ports;
444 	struct mlx4_uverbs_ex_query_device cmd;
445 	struct mlx4_uverbs_ex_query_device_resp resp = {};
446 	struct mlx4_clock_params clock_params;
447 
448 	if (uhw->inlen) {
449 		err = ib_copy_validate_udata_in_cm(uhw, cmd, reserved, 0);
450 		if (err)
451 			return err;
452 
453 		if (cmd.reserved)
454 			return -EINVAL;
455 	}
456 
457 	resp.response_length = offsetof(typeof(resp), response_length) +
458 		sizeof(resp.response_length);
459 	in_mad = kzalloc_obj(*in_mad);
460 	out_mad = kmalloc_obj(*out_mad);
461 	err = -ENOMEM;
462 	if (!in_mad || !out_mad)
463 		goto out;
464 
465 	ib_init_query_mad(in_mad);
466 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
467 
468 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
469 			   1, NULL, NULL, in_mad, out_mad);
470 	if (err)
471 		goto out;
472 
473 	memset(props, 0, sizeof *props);
474 
475 	have_ib_ports = num_ib_ports(dev->dev);
476 
477 	props->fw_ver = dev->dev->caps.fw_ver;
478 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
479 		IB_DEVICE_PORT_ACTIVE_EVENT		|
480 		IB_DEVICE_SYS_IMAGE_GUID		|
481 		IB_DEVICE_RC_RNR_NAK_GEN;
482 	props->kernel_cap_flags = IBK_BLOCK_MULTICAST_LOOPBACK;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
484 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
485 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
486 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
488 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
490 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
492 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
493 	if (dev->dev->caps.max_gso_sz &&
494 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
495 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
496 		props->kernel_cap_flags |= IBK_UD_TSO;
497 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
498 		props->kernel_cap_flags |= IBK_LOCAL_DMA_LKEY;
499 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
500 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
501 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
502 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
503 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
504 		props->device_cap_flags |= IB_DEVICE_XRC;
505 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
506 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
507 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
508 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
509 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
510 		else
511 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
512 	}
513 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
514 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
515 
516 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
517 
518 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
519 		0xffffff;
520 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
521 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
522 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
523 
524 	props->max_mr_size	   = ~0ull;
525 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
526 	props->max_qp		   = dev->dev->quotas.qp;
527 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
528 	props->max_send_sge =
529 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
530 	props->max_recv_sge =
531 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
532 	props->max_sge_rd = MLX4_MAX_SGE_RD;
533 	props->max_cq		   = dev->dev->quotas.cq;
534 	props->max_cqe		   = dev->dev->caps.max_cqes;
535 	props->max_mr		   = dev->dev->quotas.mpt;
536 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
537 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
538 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
539 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
540 	props->max_srq		   = dev->dev->quotas.srq;
541 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
542 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
543 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
544 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
545 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
546 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
547 	props->masked_atomic_cap   = props->atomic_cap;
548 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
549 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
550 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
551 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
552 					   props->max_mcast_grp;
553 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
554 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
555 	props->max_ah = INT_MAX;
556 
557 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
558 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
559 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
560 			props->rss_caps.max_rwq_indirection_tables =
561 				props->max_qp;
562 			props->rss_caps.max_rwq_indirection_table_size =
563 				dev->dev->caps.max_rss_tbl_sz;
564 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
565 			props->max_wq_type_rq = props->max_qp;
566 		}
567 
568 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
569 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
570 	}
571 
572 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
573 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
574 
575 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
576 		resp.response_length += sizeof(resp.hca_core_clock_offset);
577 		if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
578 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
579 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
580 		}
581 	}
582 
583 	if (uhw->outlen >= resp.response_length +
584 	    sizeof(resp.max_inl_recv_sz)) {
585 		resp.response_length += sizeof(resp.max_inl_recv_sz);
586 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
587 			sizeof(struct mlx4_wqe_data_seg);
588 	}
589 
590 	if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
591 		if (props->rss_caps.supported_qpts) {
592 			resp.rss_caps.rx_hash_function =
593 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
594 
595 			resp.rss_caps.rx_hash_fields_mask =
596 				MLX4_IB_RX_HASH_SRC_IPV4 |
597 				MLX4_IB_RX_HASH_DST_IPV4 |
598 				MLX4_IB_RX_HASH_SRC_IPV6 |
599 				MLX4_IB_RX_HASH_DST_IPV6 |
600 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
601 				MLX4_IB_RX_HASH_DST_PORT_TCP |
602 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
603 				MLX4_IB_RX_HASH_DST_PORT_UDP;
604 
605 			if (dev->dev->caps.tunnel_offload_mode ==
606 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
607 				resp.rss_caps.rx_hash_fields_mask |=
608 					MLX4_IB_RX_HASH_INNER;
609 		}
610 		resp.response_length = offsetof(typeof(resp), rss_caps) +
611 				       sizeof(resp.rss_caps);
612 	}
613 
614 	if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
615 		if (dev->dev->caps.max_gso_sz &&
616 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
617 		    IB_LINK_LAYER_ETHERNET) ||
618 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
619 		    IB_LINK_LAYER_ETHERNET))) {
620 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
621 			resp.tso_caps.supported_qpts |=
622 				1 << IB_QPT_RAW_PACKET;
623 		}
624 		resp.response_length = offsetof(typeof(resp), tso_caps) +
625 				       sizeof(resp.tso_caps);
626 	}
627 
628 	if (uhw->outlen) {
629 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
630 		if (err)
631 			goto out;
632 	}
633 out:
634 	kfree(in_mad);
635 	kfree(out_mad);
636 
637 	return err;
638 }
639 
640 static enum rdma_link_layer
641 mlx4_ib_port_link_layer(struct ib_device *device, u32 port_num)
642 {
643 	struct mlx4_dev *dev = to_mdev(device)->dev;
644 
645 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
646 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
647 }
648 
649 static int ib_link_query_port(struct ib_device *ibdev, u32 port,
650 			      struct ib_port_attr *props, int netw_view)
651 {
652 	struct ib_smp *in_mad;
653 	struct ib_smp *out_mad;
654 	int ext_active_speed;
655 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
656 	int err = -ENOMEM;
657 
658 	in_mad = kzalloc_obj(*in_mad);
659 	out_mad = kmalloc_obj(*out_mad);
660 	if (!in_mad || !out_mad)
661 		goto out;
662 
663 	ib_init_query_mad(in_mad);
664 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
665 	in_mad->attr_mod = cpu_to_be32(port);
666 
667 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
668 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
669 
670 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
671 				in_mad, out_mad);
672 	if (err)
673 		goto out;
674 
675 
676 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
677 	props->lmc		= out_mad->data[34] & 0x7;
678 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
679 	props->sm_sl		= out_mad->data[36] & 0xf;
680 	props->state		= out_mad->data[32] & 0xf;
681 	props->phys_state	= out_mad->data[33] >> 4;
682 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
683 	if (netw_view)
684 		props->gid_tbl_len = out_mad->data[50];
685 	else
686 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
687 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
688 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
689 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
690 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
691 	props->active_width	= out_mad->data[31] & 0xf;
692 	props->active_speed	= out_mad->data[35] >> 4;
693 	props->max_mtu		= out_mad->data[41] & 0xf;
694 	props->active_mtu	= out_mad->data[36] >> 4;
695 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
696 	props->max_vl_num	= out_mad->data[37] >> 4;
697 	props->init_type_reply	= out_mad->data[41] >> 4;
698 
699 	/* Check if extended speeds (EDR/FDR/...) are supported */
700 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
701 		ext_active_speed = out_mad->data[62] >> 4;
702 
703 		switch (ext_active_speed) {
704 		case 1:
705 			props->active_speed = IB_SPEED_FDR;
706 			break;
707 		case 2:
708 			props->active_speed = IB_SPEED_EDR;
709 			break;
710 		}
711 	}
712 
713 	/* If reported active speed is QDR, check if is FDR-10 */
714 	if (props->active_speed == IB_SPEED_QDR) {
715 		ib_init_query_mad(in_mad);
716 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
717 		in_mad->attr_mod = cpu_to_be32(port);
718 
719 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
720 				   NULL, NULL, in_mad, out_mad);
721 		if (err)
722 			goto out;
723 
724 		/* Checking LinkSpeedActive for FDR-10 */
725 		if (out_mad->data[15] & 0x1)
726 			props->active_speed = IB_SPEED_FDR10;
727 	}
728 
729 	/* Avoid wrong speed value returned by FW if the IB link is down. */
730 	if (props->state == IB_PORT_DOWN)
731 		 props->active_speed = IB_SPEED_SDR;
732 
733 out:
734 	kfree(in_mad);
735 	kfree(out_mad);
736 	return err;
737 }
738 
739 static u8 state_to_phys_state(enum ib_port_state state)
740 {
741 	return state == IB_PORT_ACTIVE ?
742 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
743 }
744 
745 static int eth_link_query_port(struct ib_device *ibdev, u32 port,
746 			       struct ib_port_attr *props)
747 {
748 
749 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
750 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
751 	struct net_device *ndev;
752 	enum ib_mtu tmp;
753 	struct mlx4_cmd_mailbox *mailbox;
754 	int err = 0;
755 	int is_bonded = mlx4_is_bonded(mdev->dev);
756 
757 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
758 	if (IS_ERR(mailbox))
759 		return PTR_ERR(mailbox);
760 
761 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
762 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
763 			   MLX4_CMD_WRAPPED);
764 	if (err)
765 		goto out;
766 
767 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
768 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
769 					   IB_WIDTH_4X : IB_WIDTH_1X;
770 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
771 					   IB_SPEED_FDR : IB_SPEED_QDR;
772 	props->port_cap_flags	= IB_PORT_CM_SUP;
773 	props->ip_gids = true;
774 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
775 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
776 	if (mdev->dev->caps.pkey_table_len[port])
777 		props->pkey_tbl_len = 1;
778 	props->max_mtu		= IB_MTU_4096;
779 	props->max_vl_num	= 2;
780 	props->state		= IB_PORT_DOWN;
781 	props->phys_state	= state_to_phys_state(props->state);
782 	props->active_mtu	= IB_MTU_256;
783 	spin_lock_bh(&iboe->lock);
784 	ndev = iboe->netdevs[port - 1];
785 	if (ndev && is_bonded) {
786 		rcu_read_lock(); /* required to get upper dev */
787 		ndev = netdev_master_upper_dev_get_rcu(ndev);
788 		rcu_read_unlock();
789 	}
790 	if (!ndev)
791 		goto out_unlock;
792 
793 	tmp = iboe_get_mtu(ndev->mtu);
794 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
795 
796 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
797 					IB_PORT_ACTIVE : IB_PORT_DOWN;
798 	props->phys_state	= state_to_phys_state(props->state);
799 out_unlock:
800 	spin_unlock_bh(&iboe->lock);
801 out:
802 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
803 	return err;
804 }
805 
806 int __mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
807 			 struct ib_port_attr *props, int netw_view)
808 {
809 	int err;
810 
811 	/* props being zeroed by the caller, avoid zeroing it here */
812 
813 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
814 		ib_link_query_port(ibdev, port, props, netw_view) :
815 				eth_link_query_port(ibdev, port, props);
816 
817 	return err;
818 }
819 
820 static int mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
821 			      struct ib_port_attr *props)
822 {
823 	/* returns host view */
824 	return __mlx4_ib_query_port(ibdev, port, props, 0);
825 }
826 
827 int __mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
828 			union ib_gid *gid, int netw_view)
829 {
830 	struct ib_smp *in_mad;
831 	struct ib_smp *out_mad;
832 	int err = -ENOMEM;
833 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
834 	int clear = 0;
835 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
836 
837 	in_mad = kzalloc_obj(*in_mad);
838 	out_mad = kmalloc_obj(*out_mad);
839 	if (!in_mad || !out_mad)
840 		goto out;
841 
842 	ib_init_query_mad(in_mad);
843 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
844 	in_mad->attr_mod = cpu_to_be32(port);
845 
846 	if (mlx4_is_mfunc(dev->dev) && netw_view)
847 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
848 
849 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
850 	if (err)
851 		goto out;
852 
853 	memcpy(gid->raw, out_mad->data + 8, 8);
854 
855 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
856 		if (index) {
857 			/* For any index > 0, return the null guid */
858 			err = 0;
859 			clear = 1;
860 			goto out;
861 		}
862 	}
863 
864 	ib_init_query_mad(in_mad);
865 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
866 	in_mad->attr_mod = cpu_to_be32(index / 8);
867 
868 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
869 			   NULL, NULL, in_mad, out_mad);
870 	if (err)
871 		goto out;
872 
873 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
874 
875 out:
876 	if (clear)
877 		memset(gid->raw + 8, 0, 8);
878 	kfree(in_mad);
879 	kfree(out_mad);
880 	return err;
881 }
882 
883 static int mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
884 			     union ib_gid *gid)
885 {
886 	if (rdma_protocol_ib(ibdev, port))
887 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
888 	return 0;
889 }
890 
891 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u32 port,
892 			       u64 *sl2vl_tbl)
893 {
894 	union sl2vl_tbl_to_u64 sl2vl64;
895 	struct ib_smp *in_mad;
896 	struct ib_smp *out_mad;
897 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
898 	int err = -ENOMEM;
899 	int jj;
900 
901 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
902 		*sl2vl_tbl = 0;
903 		return 0;
904 	}
905 
906 	in_mad = kzalloc_obj(*in_mad);
907 	out_mad = kmalloc_obj(*out_mad);
908 	if (!in_mad || !out_mad)
909 		goto out;
910 
911 	ib_init_query_mad(in_mad);
912 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
913 	in_mad->attr_mod = 0;
914 
915 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
916 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
917 
918 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
919 			   in_mad, out_mad);
920 	if (err)
921 		goto out;
922 
923 	for (jj = 0; jj < 8; jj++)
924 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
925 	*sl2vl_tbl = sl2vl64.sl64;
926 
927 out:
928 	kfree(in_mad);
929 	kfree(out_mad);
930 	return err;
931 }
932 
933 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
934 {
935 	u64 sl2vl;
936 	int i;
937 	int err;
938 
939 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
940 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
941 			continue;
942 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
943 		if (err) {
944 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
945 			       i, err);
946 			sl2vl = 0;
947 		}
948 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
949 	}
950 }
951 
952 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
953 			 u16 *pkey, int netw_view)
954 {
955 	struct ib_smp *in_mad;
956 	struct ib_smp *out_mad;
957 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
958 	int err = -ENOMEM;
959 
960 	in_mad = kzalloc_obj(*in_mad);
961 	out_mad = kmalloc_obj(*out_mad);
962 	if (!in_mad || !out_mad)
963 		goto out;
964 
965 	ib_init_query_mad(in_mad);
966 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
967 	in_mad->attr_mod = cpu_to_be32(index / 32);
968 
969 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
970 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
971 
972 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
973 			   in_mad, out_mad);
974 	if (err)
975 		goto out;
976 
977 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
978 
979 out:
980 	kfree(in_mad);
981 	kfree(out_mad);
982 	return err;
983 }
984 
985 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
986 			      u16 *pkey)
987 {
988 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
989 }
990 
991 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
992 				 struct ib_device_modify *props)
993 {
994 	struct mlx4_cmd_mailbox *mailbox;
995 	unsigned long flags;
996 
997 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
998 		return -EOPNOTSUPP;
999 
1000 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1001 		return 0;
1002 
1003 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1004 		return -EOPNOTSUPP;
1005 
1006 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1007 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1008 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1009 
1010 	/*
1011 	 * If possible, pass node desc to FW, so it can generate
1012 	 * a 144 trap.  If cmd fails, just ignore.
1013 	 */
1014 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1015 	if (IS_ERR(mailbox))
1016 		return 0;
1017 
1018 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1019 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1020 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1021 
1022 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1023 
1024 	return 0;
1025 }
1026 
1027 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u32 port,
1028 			    int reset_qkey_viols, u32 cap_mask)
1029 {
1030 	struct mlx4_cmd_mailbox *mailbox;
1031 	int err;
1032 
1033 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1034 	if (IS_ERR(mailbox))
1035 		return PTR_ERR(mailbox);
1036 
1037 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1038 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1039 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1040 	} else {
1041 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1042 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1043 	}
1044 
1045 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1046 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1047 		       MLX4_CMD_WRAPPED);
1048 
1049 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1050 	return err;
1051 }
1052 
1053 static int mlx4_ib_modify_port(struct ib_device *ibdev, u32 port, int mask,
1054 			       struct ib_port_modify *props)
1055 {
1056 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1057 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1058 	struct ib_port_attr attr;
1059 	u32 cap_mask;
1060 	int err;
1061 
1062 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1063 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1064 	 * violations and port capabilities are not meaningful.
1065 	 */
1066 	if (is_eth)
1067 		return 0;
1068 
1069 	mutex_lock(&mdev->cap_mask_mutex);
1070 
1071 	err = ib_query_port(ibdev, port, &attr);
1072 	if (err)
1073 		goto out;
1074 
1075 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1076 		~props->clr_port_cap_mask;
1077 
1078 	err = mlx4_ib_SET_PORT(mdev, port,
1079 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1080 			       cap_mask);
1081 
1082 out:
1083 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1084 	return err;
1085 }
1086 
1087 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1088 				  struct ib_udata *udata)
1089 {
1090 	struct ib_device *ibdev = uctx->device;
1091 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1092 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1093 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1094 	struct mlx4_ib_alloc_ucontext_resp resp;
1095 	int err;
1096 
1097 	if (!dev->ib_active)
1098 		return -EAGAIN;
1099 
1100 	if (ibdev->ops.uverbs_abi_ver ==
1101 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1102 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1103 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1104 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1105 	} else {
1106 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1107 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1108 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1109 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1110 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1111 	}
1112 
1113 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1114 	if (err)
1115 		return err;
1116 
1117 	INIT_LIST_HEAD(&context->db_page_list);
1118 	mutex_init(&context->db_page_mutex);
1119 
1120 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1121 	mutex_init(&context->wqn_ranges_mutex);
1122 
1123 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1124 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1125 	else
1126 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1127 
1128 	if (err) {
1129 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1130 		return -EFAULT;
1131 	}
1132 
1133 	return err;
1134 }
1135 
1136 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1137 {
1138 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1139 
1140 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1141 }
1142 
1143 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1144 {
1145 }
1146 
1147 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1148 {
1149 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1150 
1151 	switch (vma->vm_pgoff) {
1152 	case 0:
1153 		return rdma_user_mmap_io(context, vma,
1154 					 to_mucontext(context)->uar.pfn,
1155 					 PAGE_SIZE,
1156 					 pgprot_noncached(vma->vm_page_prot),
1157 					 NULL);
1158 
1159 	case 1:
1160 		if (dev->dev->caps.bf_reg_size == 0)
1161 			return -EINVAL;
1162 		return rdma_user_mmap_io(
1163 			context, vma,
1164 			to_mucontext(context)->uar.pfn +
1165 				dev->dev->caps.num_uars,
1166 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1167 			NULL);
1168 
1169 	case 3: {
1170 		struct mlx4_clock_params params;
1171 		int ret;
1172 
1173 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1174 		if (ret)
1175 			return ret;
1176 
1177 		return rdma_user_mmap_io(
1178 			context, vma,
1179 			(pci_resource_start(dev->dev->persist->pdev,
1180 					    params.bar) +
1181 			 params.offset) >>
1182 				PAGE_SHIFT,
1183 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1184 			NULL);
1185 	}
1186 
1187 	default:
1188 		return -EINVAL;
1189 	}
1190 }
1191 
1192 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1193 {
1194 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1195 	struct ib_device *ibdev = ibpd->device;
1196 	int err;
1197 
1198 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1199 	if (err)
1200 		return err;
1201 
1202 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1203 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1204 		return -EFAULT;
1205 	}
1206 	return 0;
1207 }
1208 
1209 static int mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1210 {
1211 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1212 	return 0;
1213 }
1214 
1215 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1216 {
1217 	struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1218 	struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1219 	struct ib_cq_init_attr cq_attr = {};
1220 	int err;
1221 
1222 	if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1223 		return -EOPNOTSUPP;
1224 
1225 	err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1226 	if (err)
1227 		return err;
1228 
1229 	xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1230 	if (IS_ERR(xrcd->pd)) {
1231 		err = PTR_ERR(xrcd->pd);
1232 		goto err2;
1233 	}
1234 
1235 	cq_attr.cqe = 1;
1236 	xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1237 	if (IS_ERR(xrcd->cq)) {
1238 		err = PTR_ERR(xrcd->cq);
1239 		goto err3;
1240 	}
1241 
1242 	return 0;
1243 
1244 err3:
1245 	ib_dealloc_pd(xrcd->pd);
1246 err2:
1247 	mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1248 	return err;
1249 }
1250 
1251 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1252 {
1253 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1254 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1255 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1256 	return 0;
1257 }
1258 
1259 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1260 {
1261 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1262 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1263 	struct mlx4_ib_gid_entry *ge;
1264 
1265 	ge = kzalloc_obj(*ge);
1266 	if (!ge)
1267 		return -ENOMEM;
1268 
1269 	ge->gid = *gid;
1270 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1271 		ge->port = mqp->port;
1272 		ge->added = 1;
1273 	}
1274 
1275 	mutex_lock(&mqp->mutex);
1276 	list_add_tail(&ge->list, &mqp->gid_list);
1277 	mutex_unlock(&mqp->mutex);
1278 
1279 	return 0;
1280 }
1281 
1282 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1283 					  struct mlx4_ib_counters *ctr_table)
1284 {
1285 	struct counter_index *counter, *tmp_count;
1286 
1287 	mutex_lock(&ctr_table->mutex);
1288 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1289 				 list) {
1290 		if (counter->allocated)
1291 			mlx4_counter_free(ibdev->dev, counter->index);
1292 		list_del(&counter->list);
1293 		kfree(counter);
1294 	}
1295 	mutex_unlock(&ctr_table->mutex);
1296 }
1297 
1298 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1299 		   union ib_gid *gid)
1300 {
1301 	struct net_device *ndev;
1302 	int ret = 0;
1303 
1304 	if (!mqp->port)
1305 		return 0;
1306 
1307 	spin_lock_bh(&mdev->iboe.lock);
1308 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1309 	dev_hold(ndev);
1310 	spin_unlock_bh(&mdev->iboe.lock);
1311 
1312 	if (ndev) {
1313 		ret = 1;
1314 		dev_put(ndev);
1315 	}
1316 
1317 	return ret;
1318 }
1319 
1320 struct mlx4_ib_steering {
1321 	struct list_head list;
1322 	struct mlx4_flow_reg_id reg_id;
1323 	union ib_gid gid;
1324 };
1325 
1326 #define LAST_ETH_FIELD vlan_tag
1327 #define LAST_IB_FIELD sl
1328 #define LAST_IPV4_FIELD dst_ip
1329 #define LAST_TCP_UDP_FIELD src_port
1330 
1331 /* Field is the last supported field */
1332 #define FIELDS_NOT_SUPPORTED(filter, field)\
1333 	memchr_inv((void *)&filter.field  +\
1334 		   sizeof(filter.field), 0,\
1335 		   sizeof(filter) -\
1336 		   offsetof(typeof(filter), field) -\
1337 		   sizeof(filter.field))
1338 
1339 static int parse_flow_attr(struct mlx4_dev *dev,
1340 			   u32 qp_num,
1341 			   union ib_flow_spec *ib_spec,
1342 			   struct _rule_hw *mlx4_spec)
1343 {
1344 	enum mlx4_net_trans_rule_id type;
1345 
1346 	switch (ib_spec->type) {
1347 	case IB_FLOW_SPEC_ETH:
1348 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1349 			return -ENOTSUPP;
1350 
1351 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1352 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1353 		       ETH_ALEN);
1354 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1355 		       ETH_ALEN);
1356 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1357 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1358 		break;
1359 	case IB_FLOW_SPEC_IB:
1360 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1361 			return -ENOTSUPP;
1362 
1363 		type = MLX4_NET_TRANS_RULE_ID_IB;
1364 		mlx4_spec->ib.l3_qpn =
1365 			cpu_to_be32(qp_num);
1366 		mlx4_spec->ib.qpn_mask =
1367 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1368 		break;
1369 
1370 
1371 	case IB_FLOW_SPEC_IPV4:
1372 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1373 			return -ENOTSUPP;
1374 
1375 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1376 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1377 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1378 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1379 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1380 		break;
1381 
1382 	case IB_FLOW_SPEC_TCP:
1383 	case IB_FLOW_SPEC_UDP:
1384 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1385 			return -ENOTSUPP;
1386 
1387 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1388 					MLX4_NET_TRANS_RULE_ID_TCP :
1389 					MLX4_NET_TRANS_RULE_ID_UDP;
1390 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1391 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1392 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1393 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1394 		break;
1395 
1396 	default:
1397 		return -EINVAL;
1398 	}
1399 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1400 	    mlx4_hw_rule_sz(dev, type) < 0)
1401 		return -EINVAL;
1402 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1403 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1404 	return mlx4_hw_rule_sz(dev, type);
1405 }
1406 
1407 struct default_rules {
1408 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1409 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1410 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1411 	__u8  link_layer;
1412 };
1413 static const struct default_rules default_table[] = {
1414 	{
1415 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1416 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1417 		.rules_create_list = {IB_FLOW_SPEC_IB},
1418 		.link_layer = IB_LINK_LAYER_INFINIBAND
1419 	}
1420 };
1421 
1422 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1423 					 struct ib_flow_attr *flow_attr)
1424 {
1425 	int i, j, k;
1426 	void *ib_flow;
1427 	const struct default_rules *pdefault_rules = default_table;
1428 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1429 
1430 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1431 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1432 		memset(&field_types, 0, sizeof(field_types));
1433 
1434 		if (link_layer != pdefault_rules->link_layer)
1435 			continue;
1436 
1437 		ib_flow = flow_attr + 1;
1438 		/* we assume the specs are sorted */
1439 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1440 		     j < flow_attr->num_of_specs; k++) {
1441 			union ib_flow_spec *current_flow =
1442 				(union ib_flow_spec *)ib_flow;
1443 
1444 			/* same layer but different type */
1445 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1446 			     (pdefault_rules->mandatory_fields[k] &
1447 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1448 			    (current_flow->type !=
1449 			     pdefault_rules->mandatory_fields[k]))
1450 				goto out;
1451 
1452 			/* same layer, try match next one */
1453 			if (current_flow->type ==
1454 			    pdefault_rules->mandatory_fields[k]) {
1455 				j++;
1456 				ib_flow +=
1457 					((union ib_flow_spec *)ib_flow)->size;
1458 			}
1459 		}
1460 
1461 		ib_flow = flow_attr + 1;
1462 		for (j = 0; j < flow_attr->num_of_specs;
1463 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1464 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1465 				/* same layer and same type */
1466 				if (((union ib_flow_spec *)ib_flow)->type ==
1467 				    pdefault_rules->mandatory_not_fields[k])
1468 					goto out;
1469 
1470 		return i;
1471 	}
1472 out:
1473 	return -1;
1474 }
1475 
1476 static int __mlx4_ib_create_default_rules(
1477 		struct mlx4_ib_dev *mdev,
1478 		struct ib_qp *qp,
1479 		const struct default_rules *pdefault_rules,
1480 		struct _rule_hw *mlx4_spec) {
1481 	int size = 0;
1482 	int i;
1483 
1484 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1485 		union ib_flow_spec ib_spec = {};
1486 		int ret;
1487 
1488 		switch (pdefault_rules->rules_create_list[i]) {
1489 		case 0:
1490 			/* no rule */
1491 			continue;
1492 		case IB_FLOW_SPEC_IB:
1493 			ib_spec.type = IB_FLOW_SPEC_IB;
1494 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1495 
1496 			break;
1497 		default:
1498 			/* invalid rule */
1499 			return -EINVAL;
1500 		}
1501 		/* We must put empty rule, qpn is being ignored */
1502 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1503 				      mlx4_spec);
1504 		if (ret < 0) {
1505 			pr_info("invalid parsing\n");
1506 			return -EINVAL;
1507 		}
1508 
1509 		mlx4_spec = (void *)mlx4_spec + ret;
1510 		size += ret;
1511 	}
1512 	return size;
1513 }
1514 
1515 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1516 			  int domain,
1517 			  enum mlx4_net_trans_promisc_mode flow_type,
1518 			  u64 *reg_id)
1519 {
1520 	int ret, i;
1521 	int size = 0;
1522 	void *ib_flow;
1523 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1524 	struct mlx4_cmd_mailbox *mailbox;
1525 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1526 	int default_flow;
1527 
1528 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1529 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1530 		return -EINVAL;
1531 	}
1532 
1533 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1534 		return -EINVAL;
1535 
1536 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1537 	if (IS_ERR(mailbox))
1538 		return PTR_ERR(mailbox);
1539 	ctrl = mailbox->buf;
1540 
1541 	ctrl->prio = cpu_to_be16(domain | flow_attr->priority);
1542 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1543 	ctrl->port = flow_attr->port;
1544 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1545 
1546 	ib_flow = flow_attr + 1;
1547 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1548 	/* Add default flows */
1549 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1550 	if (default_flow >= 0) {
1551 		ret = __mlx4_ib_create_default_rules(
1552 				mdev, qp, default_table + default_flow,
1553 				mailbox->buf + size);
1554 		if (ret < 0) {
1555 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1556 			return -EINVAL;
1557 		}
1558 		size += ret;
1559 	}
1560 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1561 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1562 				      mailbox->buf + size);
1563 		if (ret < 0) {
1564 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1565 			return -EINVAL;
1566 		}
1567 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1568 		size += ret;
1569 	}
1570 
1571 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1572 	    flow_attr->num_of_specs == 1) {
1573 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1574 		enum ib_flow_spec_type header_spec =
1575 			((union ib_flow_spec *)(flow_attr + 1))->type;
1576 
1577 		if (header_spec == IB_FLOW_SPEC_ETH)
1578 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1579 	}
1580 
1581 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1582 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1583 			   MLX4_CMD_NATIVE);
1584 	if (ret == -ENOMEM)
1585 		pr_err("mcg table is full. Fail to register network rule.\n");
1586 	else if (ret == -ENXIO)
1587 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1588 	else if (ret)
1589 		pr_err("Invalid argument. Fail to register network rule.\n");
1590 
1591 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1592 	return ret;
1593 }
1594 
1595 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1596 {
1597 	int err;
1598 	err = mlx4_cmd(dev, reg_id, 0, 0,
1599 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1600 		       MLX4_CMD_NATIVE);
1601 	if (err)
1602 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1603 		       reg_id);
1604 	return err;
1605 }
1606 
1607 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1608 				    u64 *reg_id)
1609 {
1610 	void *ib_flow;
1611 	union ib_flow_spec *ib_spec;
1612 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1613 	int err = 0;
1614 
1615 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1616 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1617 		return 0; /* do nothing */
1618 
1619 	ib_flow = flow_attr + 1;
1620 	ib_spec = (union ib_flow_spec *)ib_flow;
1621 
1622 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1623 		return 0; /* do nothing */
1624 
1625 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1626 				    flow_attr->port, qp->qp_num,
1627 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1628 				    reg_id);
1629 	return err;
1630 }
1631 
1632 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1633 				      struct ib_flow_attr *flow_attr,
1634 				      enum mlx4_net_trans_promisc_mode *type)
1635 {
1636 	int err = 0;
1637 
1638 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1639 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1640 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1641 		return -EOPNOTSUPP;
1642 	}
1643 
1644 	if (flow_attr->num_of_specs == 0) {
1645 		type[0] = MLX4_FS_MC_SNIFFER;
1646 		type[1] = MLX4_FS_UC_SNIFFER;
1647 	} else {
1648 		union ib_flow_spec *ib_spec;
1649 
1650 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1651 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1652 			return -EINVAL;
1653 
1654 		/* if all is zero than MC and UC */
1655 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1656 			type[0] = MLX4_FS_MC_SNIFFER;
1657 			type[1] = MLX4_FS_UC_SNIFFER;
1658 		} else {
1659 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1660 					    ib_spec->eth.mask.dst_mac[1],
1661 					    ib_spec->eth.mask.dst_mac[2],
1662 					    ib_spec->eth.mask.dst_mac[3],
1663 					    ib_spec->eth.mask.dst_mac[4],
1664 					    ib_spec->eth.mask.dst_mac[5]};
1665 
1666 			/* Above xor was only on MC bit, non empty mask is valid
1667 			 * only if this bit is set and rest are zero.
1668 			 */
1669 			if (!is_zero_ether_addr(&mac[0]))
1670 				return -EINVAL;
1671 
1672 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1673 				type[0] = MLX4_FS_MC_SNIFFER;
1674 			else
1675 				type[0] = MLX4_FS_UC_SNIFFER;
1676 		}
1677 	}
1678 
1679 	return err;
1680 }
1681 
1682 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1683 					   struct ib_flow_attr *flow_attr,
1684 					   struct ib_udata *udata)
1685 {
1686 	int err = 0, i = 0, j = 0;
1687 	struct mlx4_ib_flow *mflow;
1688 	enum mlx4_net_trans_promisc_mode type[2];
1689 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1690 	int is_bonded = mlx4_is_bonded(dev);
1691 
1692 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1693 		return ERR_PTR(-EOPNOTSUPP);
1694 
1695 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1696 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1697 		return ERR_PTR(-EOPNOTSUPP);
1698 
1699 	if (udata &&
1700 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1701 		return ERR_PTR(-EOPNOTSUPP);
1702 
1703 	memset(type, 0, sizeof(type));
1704 
1705 	mflow = kzalloc_obj(*mflow);
1706 	if (!mflow) {
1707 		err = -ENOMEM;
1708 		goto err_free;
1709 	}
1710 
1711 	switch (flow_attr->type) {
1712 	case IB_FLOW_ATTR_NORMAL:
1713 		/* If dont trap flag (continue match) is set, under specific
1714 		 * condition traffic be replicated to given qp,
1715 		 * without stealing it
1716 		 */
1717 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1718 			err = mlx4_ib_add_dont_trap_rule(dev,
1719 							 flow_attr,
1720 							 type);
1721 			if (err)
1722 				goto err_free;
1723 		} else {
1724 			type[0] = MLX4_FS_REGULAR;
1725 		}
1726 		break;
1727 
1728 	case IB_FLOW_ATTR_ALL_DEFAULT:
1729 		type[0] = MLX4_FS_ALL_DEFAULT;
1730 		break;
1731 
1732 	case IB_FLOW_ATTR_MC_DEFAULT:
1733 		type[0] = MLX4_FS_MC_DEFAULT;
1734 		break;
1735 
1736 	case IB_FLOW_ATTR_SNIFFER:
1737 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1738 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1739 		break;
1740 
1741 	default:
1742 		err = -EINVAL;
1743 		goto err_free;
1744 	}
1745 
1746 	while (i < ARRAY_SIZE(type) && type[i]) {
1747 		err = __mlx4_ib_create_flow(qp, flow_attr, MLX4_DOMAIN_UVERBS,
1748 					    type[i], &mflow->reg_id[i].id);
1749 		if (err)
1750 			goto err_create_flow;
1751 		if (is_bonded) {
1752 			/* Application always sees one port so the mirror rule
1753 			 * must be on port #2
1754 			 */
1755 			flow_attr->port = 2;
1756 			err = __mlx4_ib_create_flow(qp, flow_attr,
1757 						    MLX4_DOMAIN_UVERBS, type[j],
1758 						    &mflow->reg_id[j].mirror);
1759 			flow_attr->port = 1;
1760 			if (err)
1761 				goto err_create_flow;
1762 			j++;
1763 		}
1764 
1765 		i++;
1766 	}
1767 
1768 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1769 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1770 					       &mflow->reg_id[i].id);
1771 		if (err)
1772 			goto err_create_flow;
1773 
1774 		if (is_bonded) {
1775 			flow_attr->port = 2;
1776 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1777 						       &mflow->reg_id[j].mirror);
1778 			flow_attr->port = 1;
1779 			if (err)
1780 				goto err_create_flow;
1781 			j++;
1782 		}
1783 		/* function to create mirror rule */
1784 		i++;
1785 	}
1786 
1787 	return &mflow->ibflow;
1788 
1789 err_create_flow:
1790 	while (i) {
1791 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1792 					     mflow->reg_id[i].id);
1793 		i--;
1794 	}
1795 
1796 	while (j) {
1797 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1798 					     mflow->reg_id[j].mirror);
1799 		j--;
1800 	}
1801 err_free:
1802 	kfree(mflow);
1803 	return ERR_PTR(err);
1804 }
1805 
1806 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1807 {
1808 	int err, ret = 0;
1809 	int i = 0;
1810 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1811 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1812 
1813 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1814 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1815 		if (err)
1816 			ret = err;
1817 		if (mflow->reg_id[i].mirror) {
1818 			err = __mlx4_ib_destroy_flow(mdev->dev,
1819 						     mflow->reg_id[i].mirror);
1820 			if (err)
1821 				ret = err;
1822 		}
1823 		i++;
1824 	}
1825 
1826 	kfree(mflow);
1827 	return ret;
1828 }
1829 
1830 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1831 {
1832 	int err;
1833 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1834 	struct mlx4_dev	*dev = mdev->dev;
1835 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1836 	struct mlx4_ib_steering *ib_steering = NULL;
1837 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1838 	struct mlx4_flow_reg_id	reg_id;
1839 
1840 	if (mdev->dev->caps.steering_mode ==
1841 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1842 		ib_steering = kmalloc_obj(*ib_steering);
1843 		if (!ib_steering)
1844 			return -ENOMEM;
1845 	}
1846 
1847 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1848 				    !!(mqp->flags &
1849 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1850 				    prot, &reg_id.id);
1851 	if (err) {
1852 		pr_err("multicast attach op failed, err %d\n", err);
1853 		goto err_malloc;
1854 	}
1855 
1856 	reg_id.mirror = 0;
1857 	if (mlx4_is_bonded(dev)) {
1858 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1859 					    (mqp->port == 1) ? 2 : 1,
1860 					    !!(mqp->flags &
1861 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1862 					    prot, &reg_id.mirror);
1863 		if (err)
1864 			goto err_add;
1865 	}
1866 
1867 	err = add_gid_entry(ibqp, gid);
1868 	if (err)
1869 		goto err_add;
1870 
1871 	if (ib_steering) {
1872 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1873 		ib_steering->reg_id = reg_id;
1874 		mutex_lock(&mqp->mutex);
1875 		list_add(&ib_steering->list, &mqp->steering_rules);
1876 		mutex_unlock(&mqp->mutex);
1877 	}
1878 	return 0;
1879 
1880 err_add:
1881 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1882 			      prot, reg_id.id);
1883 	if (reg_id.mirror)
1884 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1885 				      prot, reg_id.mirror);
1886 err_malloc:
1887 	kfree(ib_steering);
1888 
1889 	return err;
1890 }
1891 
1892 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1893 {
1894 	struct mlx4_ib_gid_entry *ge;
1895 	struct mlx4_ib_gid_entry *tmp;
1896 	struct mlx4_ib_gid_entry *ret = NULL;
1897 
1898 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1899 		if (!memcmp(raw, ge->gid.raw, 16)) {
1900 			ret = ge;
1901 			break;
1902 		}
1903 	}
1904 
1905 	return ret;
1906 }
1907 
1908 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1909 {
1910 	int err;
1911 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1912 	struct mlx4_dev *dev = mdev->dev;
1913 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1914 	struct net_device *ndev;
1915 	struct mlx4_ib_gid_entry *ge;
1916 	struct mlx4_flow_reg_id reg_id = {0, 0};
1917 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1918 
1919 	if (mdev->dev->caps.steering_mode ==
1920 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1921 		struct mlx4_ib_steering *ib_steering;
1922 
1923 		mutex_lock(&mqp->mutex);
1924 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1925 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1926 				list_del(&ib_steering->list);
1927 				break;
1928 			}
1929 		}
1930 		mutex_unlock(&mqp->mutex);
1931 		if (&ib_steering->list == &mqp->steering_rules) {
1932 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1933 			return -EINVAL;
1934 		}
1935 		reg_id = ib_steering->reg_id;
1936 		kfree(ib_steering);
1937 	}
1938 
1939 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1940 				    prot, reg_id.id);
1941 	if (err)
1942 		return err;
1943 
1944 	if (mlx4_is_bonded(dev)) {
1945 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1946 					    prot, reg_id.mirror);
1947 		if (err)
1948 			return err;
1949 	}
1950 
1951 	mutex_lock(&mqp->mutex);
1952 	ge = find_gid_entry(mqp, gid->raw);
1953 	if (ge) {
1954 		spin_lock_bh(&mdev->iboe.lock);
1955 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1956 		dev_hold(ndev);
1957 		spin_unlock_bh(&mdev->iboe.lock);
1958 		dev_put(ndev);
1959 		list_del(&ge->list);
1960 		kfree(ge);
1961 	} else
1962 		pr_warn("could not find mgid entry\n");
1963 
1964 	mutex_unlock(&mqp->mutex);
1965 
1966 	return 0;
1967 }
1968 
1969 static int init_node_data(struct mlx4_ib_dev *dev)
1970 {
1971 	struct ib_smp *in_mad;
1972 	struct ib_smp *out_mad;
1973 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1974 	int err = -ENOMEM;
1975 
1976 	in_mad = kzalloc_obj(*in_mad);
1977 	out_mad = kmalloc_obj(*out_mad);
1978 	if (!in_mad || !out_mad)
1979 		goto out;
1980 
1981 	ib_init_query_mad(in_mad);
1982 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1983 	if (mlx4_is_master(dev->dev))
1984 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1985 
1986 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1987 	if (err)
1988 		goto out;
1989 
1990 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1991 
1992 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1993 
1994 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1995 	if (err)
1996 		goto out;
1997 
1998 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1999 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2000 
2001 out:
2002 	kfree(in_mad);
2003 	kfree(out_mad);
2004 	return err;
2005 }
2006 
2007 static ssize_t hca_type_show(struct device *device,
2008 			     struct device_attribute *attr, char *buf)
2009 {
2010 	struct mlx4_ib_dev *dev =
2011 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2012 
2013 	return sysfs_emit(buf, "MT%d\n", dev->dev->persist->pdev->device);
2014 }
2015 static DEVICE_ATTR_RO(hca_type);
2016 
2017 static ssize_t hw_rev_show(struct device *device,
2018 			   struct device_attribute *attr, char *buf)
2019 {
2020 	struct mlx4_ib_dev *dev =
2021 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2022 
2023 	return sysfs_emit(buf, "%x\n", dev->dev->rev_id);
2024 }
2025 static DEVICE_ATTR_RO(hw_rev);
2026 
2027 static ssize_t board_id_show(struct device *device,
2028 			     struct device_attribute *attr, char *buf)
2029 {
2030 	struct mlx4_ib_dev *dev =
2031 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2032 
2033 	return sysfs_emit(buf, "%.*s\n", MLX4_BOARD_ID_LEN, dev->dev->board_id);
2034 }
2035 static DEVICE_ATTR_RO(board_id);
2036 
2037 static struct attribute *mlx4_class_attributes[] = {
2038 	&dev_attr_hw_rev.attr,
2039 	&dev_attr_hca_type.attr,
2040 	&dev_attr_board_id.attr,
2041 	NULL
2042 };
2043 
2044 static const struct attribute_group mlx4_attr_group = {
2045 	.attrs = mlx4_class_attributes,
2046 };
2047 
2048 struct diag_counter {
2049 	const char *name;
2050 	u32 offset;
2051 };
2052 
2053 #define DIAG_COUNTER(_name, _offset)			\
2054 	{ .name = #_name, .offset = _offset }
2055 
2056 static const struct diag_counter diag_basic[] = {
2057 	DIAG_COUNTER(rq_num_lle, 0x00),
2058 	DIAG_COUNTER(sq_num_lle, 0x04),
2059 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2060 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2061 	DIAG_COUNTER(rq_num_lpe, 0x18),
2062 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2063 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2064 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2065 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2066 	DIAG_COUNTER(sq_num_bre, 0x34),
2067 	DIAG_COUNTER(sq_num_rire, 0x44),
2068 	DIAG_COUNTER(rq_num_rire, 0x48),
2069 	DIAG_COUNTER(sq_num_rae, 0x4C),
2070 	DIAG_COUNTER(rq_num_rae, 0x50),
2071 	DIAG_COUNTER(sq_num_roe, 0x54),
2072 	DIAG_COUNTER(sq_num_tree, 0x5C),
2073 	DIAG_COUNTER(sq_num_rree, 0x64),
2074 	DIAG_COUNTER(rq_num_rnr, 0x68),
2075 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2076 	DIAG_COUNTER(rq_num_oos, 0x100),
2077 	DIAG_COUNTER(sq_num_oos, 0x104),
2078 };
2079 
2080 static const struct diag_counter diag_ext[] = {
2081 	DIAG_COUNTER(rq_num_dup, 0x130),
2082 	DIAG_COUNTER(sq_num_to, 0x134),
2083 };
2084 
2085 static const struct diag_counter diag_device_only[] = {
2086 	DIAG_COUNTER(num_cqovf, 0x1A0),
2087 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2088 };
2089 
2090 static struct rdma_hw_stats *
2091 mlx4_ib_alloc_hw_device_stats(struct ib_device *ibdev)
2092 {
2093 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2094 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2095 
2096 	if (!diag[0].descs)
2097 		return NULL;
2098 
2099 	return rdma_alloc_hw_stats_struct(diag[0].descs, diag[0].num_counters,
2100 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2101 }
2102 
2103 static struct rdma_hw_stats *
2104 mlx4_ib_alloc_hw_port_stats(struct ib_device *ibdev, u32 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[1].descs)
2110 		return NULL;
2111 
2112 	return rdma_alloc_hw_stats_struct(diag[1].descs, diag[1].num_counters,
2113 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2114 }
2115 
2116 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2117 				struct rdma_hw_stats *stats,
2118 				u32 port, int index)
2119 {
2120 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2121 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2122 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2123 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2124 	int ret;
2125 	int i;
2126 
2127 	ret = mlx4_query_diag_counters(dev->dev,
2128 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2129 				       diag[!!port].offset, hw_value,
2130 				       diag[!!port].num_counters, port);
2131 
2132 	if (ret)
2133 		return ret;
2134 
2135 	for (i = 0; i < diag[!!port].num_counters; i++)
2136 		stats->value[i] = hw_value[i];
2137 
2138 	return diag[!!port].num_counters;
2139 }
2140 
2141 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2142 					 struct rdma_stat_desc **pdescs,
2143 					 u32 **offset, u32 *num, bool port)
2144 {
2145 	u32 num_counters;
2146 
2147 	num_counters = ARRAY_SIZE(diag_basic);
2148 
2149 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2150 		num_counters += ARRAY_SIZE(diag_ext);
2151 
2152 	if (!port)
2153 		num_counters += ARRAY_SIZE(diag_device_only);
2154 
2155 	*pdescs = kzalloc_objs(struct rdma_stat_desc, num_counters);
2156 	if (!*pdescs)
2157 		return -ENOMEM;
2158 
2159 	*offset = kzalloc_objs(**offset, num_counters);
2160 	if (!*offset)
2161 		goto err;
2162 
2163 	*num = num_counters;
2164 
2165 	return 0;
2166 
2167 err:
2168 	kfree(*pdescs);
2169 	return -ENOMEM;
2170 }
2171 
2172 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2173 				       struct rdma_stat_desc *descs,
2174 				       u32 *offset, bool port)
2175 {
2176 	int i;
2177 	int j;
2178 
2179 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2180 		descs[i].name = diag_basic[i].name;
2181 		offset[i] = diag_basic[i].offset;
2182 	}
2183 
2184 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2185 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2186 			descs[j].name = diag_ext[i].name;
2187 			offset[j] = diag_ext[i].offset;
2188 		}
2189 	}
2190 
2191 	if (!port) {
2192 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2193 			descs[j].name = diag_device_only[i].name;
2194 			offset[j] = diag_device_only[i].offset;
2195 		}
2196 	}
2197 }
2198 
2199 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2200 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2201 	.alloc_hw_port_stats = mlx4_ib_alloc_hw_port_stats,
2202 	.get_hw_stats = mlx4_ib_get_hw_stats,
2203 };
2204 
2205 static const struct ib_device_ops mlx4_ib_hw_stats_ops1 = {
2206 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2207 	.get_hw_stats = mlx4_ib_get_hw_stats,
2208 };
2209 
2210 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2211 {
2212 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2213 	int i;
2214 	int ret;
2215 	bool per_port = !!(ibdev->dev->caps.flags2 &
2216 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2217 
2218 	if (mlx4_is_slave(ibdev->dev))
2219 		return 0;
2220 
2221 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2222 		/*
2223 		 * i == 1 means we are building port counters, set a different
2224 		 * stats ops without port stats callback.
2225 		 */
2226 		if (i && !per_port) {
2227 			ib_set_device_ops(&ibdev->ib_dev,
2228 					  &mlx4_ib_hw_stats_ops1);
2229 
2230 			return 0;
2231 		}
2232 
2233 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].descs,
2234 						    &diag[i].offset,
2235 						    &diag[i].num_counters, i);
2236 		if (ret)
2237 			goto err_alloc;
2238 
2239 		mlx4_ib_fill_diag_counters(ibdev, diag[i].descs,
2240 					   diag[i].offset, i);
2241 	}
2242 
2243 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2244 
2245 	return 0;
2246 
2247 err_alloc:
2248 	if (i) {
2249 		kfree(diag[i - 1].descs);
2250 		kfree(diag[i - 1].offset);
2251 	}
2252 
2253 	return ret;
2254 }
2255 
2256 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2257 {
2258 	int i;
2259 
2260 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2261 		kfree(ibdev->diag_counters[i].offset);
2262 		kfree(ibdev->diag_counters[i].descs);
2263 	}
2264 }
2265 
2266 #define MLX4_IB_INVALID_MAC	((u64)-1)
2267 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2268 			       struct net_device *dev,
2269 			       int port)
2270 {
2271 	u64 new_smac = 0;
2272 	u64 release_mac = MLX4_IB_INVALID_MAC;
2273 	struct mlx4_ib_qp *qp;
2274 
2275 	new_smac = ether_addr_to_u64(dev->dev_addr);
2276 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2277 
2278 	/* no need for update QP1 and mac registration in non-SRIOV */
2279 	if (!mlx4_is_mfunc(ibdev->dev))
2280 		return;
2281 
2282 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2283 	qp = ibdev->qp1_proxy[port - 1];
2284 	if (qp) {
2285 		int new_smac_index;
2286 		u64 old_smac;
2287 		struct mlx4_update_qp_params update_params;
2288 
2289 		mutex_lock(&qp->mutex);
2290 		old_smac = qp->pri.smac;
2291 		if (new_smac == old_smac)
2292 			goto unlock;
2293 
2294 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2295 
2296 		if (new_smac_index < 0)
2297 			goto unlock;
2298 
2299 		update_params.smac_index = new_smac_index;
2300 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2301 				   &update_params)) {
2302 			release_mac = new_smac;
2303 			goto unlock;
2304 		}
2305 		/* if old port was zero, no mac was yet registered for this QP */
2306 		if (qp->pri.smac_port)
2307 			release_mac = old_smac;
2308 		qp->pri.smac = new_smac;
2309 		qp->pri.smac_port = port;
2310 		qp->pri.smac_index = new_smac_index;
2311 	}
2312 
2313 unlock:
2314 	if (release_mac != MLX4_IB_INVALID_MAC)
2315 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2316 	if (qp)
2317 		mutex_unlock(&qp->mutex);
2318 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2319 }
2320 
2321 static void mlx4_ib_scan_netdev(struct mlx4_ib_dev *ibdev,
2322 				struct net_device *dev,
2323 				unsigned long event)
2324 
2325 {
2326 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
2327 
2328 	ASSERT_RTNL();
2329 
2330 	if (dev->dev.parent != ibdev->ib_dev.dev.parent)
2331 		return;
2332 
2333 	spin_lock_bh(&iboe->lock);
2334 
2335 	iboe->netdevs[dev->dev_port] = event != NETDEV_UNREGISTER ? dev : NULL;
2336 
2337 	spin_unlock_bh(&iboe->lock);
2338 
2339 	if (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER)
2340 		mlx4_ib_update_qps(ibdev, dev, dev->dev_port + 1);
2341 }
2342 
2343 static void mlx4_ib_port_event(struct ib_device *ibdev, struct net_device *ndev,
2344 			       unsigned long event)
2345 {
2346 	struct mlx4_ib_dev *mlx4_ibdev =
2347 		container_of(ibdev, struct mlx4_ib_dev, ib_dev);
2348 	struct mlx4_ib_iboe *iboe = &mlx4_ibdev->iboe;
2349 
2350 	if (!net_eq(dev_net(ndev), &init_net))
2351 		return;
2352 
2353 	ASSERT_RTNL();
2354 
2355 	if (ndev->dev.parent != mlx4_ibdev->ib_dev.dev.parent)
2356 		return;
2357 
2358 	spin_lock_bh(&iboe->lock);
2359 
2360 	iboe->netdevs[ndev->dev_port] = event != NETDEV_UNREGISTER ? ndev : NULL;
2361 
2362 	if (event == NETDEV_UP || event == NETDEV_DOWN)
2363 		ib_dispatch_port_state_event(&mlx4_ibdev->ib_dev, ndev);
2364 
2365 	spin_unlock_bh(&iboe->lock);
2366 
2367 	if (event == NETDEV_UP || event == NETDEV_CHANGE)
2368 		mlx4_ib_update_qps(mlx4_ibdev, ndev, ndev->dev_port + 1);
2369 }
2370 
2371 static int mlx4_ib_netdev_event(struct notifier_block *this,
2372 				unsigned long event, void *ptr)
2373 {
2374 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2375 	struct mlx4_ib_dev *ibdev;
2376 
2377 	if (!net_eq(dev_net(dev), &init_net))
2378 		return NOTIFY_DONE;
2379 
2380 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2381 	mlx4_ib_scan_netdev(ibdev, dev, event);
2382 
2383 	return NOTIFY_DONE;
2384 }
2385 
2386 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2387 {
2388 	int port;
2389 	int slave;
2390 	int i;
2391 
2392 	if (mlx4_is_master(ibdev->dev)) {
2393 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2394 		     ++slave) {
2395 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2396 				for (i = 0;
2397 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2398 				     ++i) {
2399 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2400 					/* master has the identity virt2phys pkey mapping */
2401 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2402 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2403 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2404 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2405 				}
2406 			}
2407 		}
2408 		/* initialize pkey cache */
2409 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2410 			for (i = 0;
2411 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2412 			     ++i)
2413 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2414 					(i) ? 0 : 0xFFFF;
2415 		}
2416 	}
2417 }
2418 
2419 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2420 {
2421 	int i, j, eq = 0, total_eqs = 0;
2422 
2423 	ibdev->eq_table = kzalloc_objs(ibdev->eq_table[0],
2424 				       dev->caps.num_comp_vectors);
2425 	if (!ibdev->eq_table)
2426 		return;
2427 
2428 	for (i = 1; i <= dev->caps.num_ports; i++) {
2429 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2430 		     j++, total_eqs++) {
2431 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2432 				continue;
2433 			ibdev->eq_table[eq] = total_eqs;
2434 			if (!mlx4_assign_eq(dev, i,
2435 					    &ibdev->eq_table[eq]))
2436 				eq++;
2437 			else
2438 				ibdev->eq_table[eq] = -1;
2439 		}
2440 	}
2441 
2442 	for (i = eq; i < dev->caps.num_comp_vectors;
2443 	     ibdev->eq_table[i++] = -1)
2444 		;
2445 
2446 	/* Advertise the new number of EQs to clients */
2447 	ibdev->ib_dev.num_comp_vectors = eq;
2448 }
2449 
2450 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2451 {
2452 	int i;
2453 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2454 
2455 	/* no eqs were allocated */
2456 	if (!ibdev->eq_table)
2457 		return;
2458 
2459 	/* Reset the advertised EQ number */
2460 	ibdev->ib_dev.num_comp_vectors = 0;
2461 
2462 	for (i = 0; i < total_eqs; i++)
2463 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2464 
2465 	kfree(ibdev->eq_table);
2466 	ibdev->eq_table = NULL;
2467 }
2468 
2469 static int mlx4_port_immutable(struct ib_device *ibdev, u32 port_num,
2470 			       struct ib_port_immutable *immutable)
2471 {
2472 	struct ib_port_attr attr;
2473 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2474 	int err;
2475 
2476 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2477 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2478 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2479 	} else {
2480 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2481 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2482 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2483 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2484 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2485 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2486 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2487 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2488 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2489 	}
2490 
2491 	err = ib_query_port(ibdev, port_num, &attr);
2492 	if (err)
2493 		return err;
2494 
2495 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2496 	immutable->gid_tbl_len = attr.gid_tbl_len;
2497 
2498 	return 0;
2499 }
2500 
2501 static void get_fw_ver_str(struct ib_device *device, char *str)
2502 {
2503 	struct mlx4_ib_dev *dev =
2504 		container_of(device, struct mlx4_ib_dev, ib_dev);
2505 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2506 		 (int) (dev->dev->caps.fw_ver >> 32),
2507 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2508 		 (int) dev->dev->caps.fw_ver & 0xffff);
2509 }
2510 
2511 static const struct ib_device_ops mlx4_ib_dev_ops = {
2512 	.owner = THIS_MODULE,
2513 	.driver_id = RDMA_DRIVER_MLX4,
2514 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2515 
2516 	.add_gid = mlx4_ib_add_gid,
2517 	.alloc_mr = mlx4_ib_alloc_mr,
2518 	.alloc_pd = mlx4_ib_alloc_pd,
2519 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2520 	.attach_mcast = mlx4_ib_mcg_attach,
2521 	.create_ah = mlx4_ib_create_ah,
2522 	.create_cq = mlx4_ib_create_cq,
2523 	.create_user_cq = mlx4_ib_create_user_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 	.device_group = &mlx4_attr_group,
2536 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2537 	.drain_rq = mlx4_ib_drain_rq,
2538 	.drain_sq = mlx4_ib_drain_sq,
2539 	.get_dev_fw_str = get_fw_ver_str,
2540 	.get_dma_mr = mlx4_ib_get_dma_mr,
2541 	.get_link_layer = mlx4_ib_port_link_layer,
2542 	.get_netdev = mlx4_ib_get_netdev,
2543 	.get_port_immutable = mlx4_port_immutable,
2544 	.map_mr_sg = mlx4_ib_map_mr_sg,
2545 	.mmap = mlx4_ib_mmap,
2546 	.modify_cq = mlx4_ib_modify_cq,
2547 	.modify_device = mlx4_ib_modify_device,
2548 	.modify_port = mlx4_ib_modify_port,
2549 	.modify_qp = mlx4_ib_modify_qp,
2550 	.modify_srq = mlx4_ib_modify_srq,
2551 	.poll_cq = mlx4_ib_poll_cq,
2552 	.post_recv = mlx4_ib_post_recv,
2553 	.post_send = mlx4_ib_post_send,
2554 	.post_srq_recv = mlx4_ib_post_srq_recv,
2555 	.process_mad = mlx4_ib_process_mad,
2556 	.query_ah = mlx4_ib_query_ah,
2557 	.query_device = mlx4_ib_query_device,
2558 	.query_gid = mlx4_ib_query_gid,
2559 	.query_pkey = mlx4_ib_query_pkey,
2560 	.query_port = mlx4_ib_query_port,
2561 	.query_qp = mlx4_ib_query_qp,
2562 	.query_srq = mlx4_ib_query_srq,
2563 	.reg_user_mr = mlx4_ib_reg_user_mr,
2564 	.req_notify_cq = mlx4_ib_arm_cq,
2565 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2566 	.resize_user_cq = mlx4_ib_resize_cq,
2567 	.report_port_event = mlx4_ib_port_event,
2568 
2569 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2570 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2571 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2572 	INIT_RDMA_OBJ_SIZE(ib_qp, mlx4_ib_qp, ibqp),
2573 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2574 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2575 };
2576 
2577 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2578 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2579 	.create_wq = mlx4_ib_create_wq,
2580 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2581 	.destroy_wq = mlx4_ib_destroy_wq,
2582 	.modify_wq = mlx4_ib_modify_wq,
2583 
2584 	INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx4_ib_rwq_ind_table,
2585 			   ib_rwq_ind_tbl),
2586 };
2587 
2588 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2589 	.alloc_mw = mlx4_ib_alloc_mw,
2590 	.dealloc_mw = mlx4_ib_dealloc_mw,
2591 
2592 	INIT_RDMA_OBJ_SIZE(ib_mw, mlx4_ib_mw, ibmw),
2593 };
2594 
2595 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2596 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2597 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2598 
2599 	INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2600 };
2601 
2602 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2603 	.create_flow = mlx4_ib_create_flow,
2604 	.destroy_flow = mlx4_ib_destroy_flow,
2605 };
2606 
2607 static int mlx4_ib_probe(struct auxiliary_device *adev,
2608 			 const struct auxiliary_device_id *id)
2609 {
2610 	struct mlx4_adev *madev = container_of(adev, struct mlx4_adev, adev);
2611 	struct mlx4_dev *dev = madev->mdev;
2612 	struct mlx4_ib_dev *ibdev;
2613 	int num_ports = 0;
2614 	int i, j;
2615 	int err;
2616 	struct mlx4_ib_iboe *iboe;
2617 	int ib_num_ports = 0;
2618 	int num_req_counters;
2619 	int allocated;
2620 	u32 counter_index;
2621 	struct counter_index *new_counter_index;
2622 
2623 	pr_info_once("%s", mlx4_ib_version);
2624 
2625 	num_ports = 0;
2626 	mlx4_foreach_ib_transport_port(i, dev)
2627 		num_ports++;
2628 
2629 	/* No point in registering a device with no ports... */
2630 	if (num_ports == 0)
2631 		return -ENODEV;
2632 
2633 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2634 	if (!ibdev) {
2635 		dev_err(&dev->persist->pdev->dev,
2636 			"Device struct alloc failed\n");
2637 		return -ENOMEM;
2638 	}
2639 
2640 	iboe = &ibdev->iboe;
2641 
2642 	err = mlx4_pd_alloc(dev, &ibdev->priv_pdn);
2643 	if (err)
2644 		goto err_dealloc;
2645 
2646 	err = mlx4_uar_alloc(dev, &ibdev->priv_uar);
2647 	if (err)
2648 		goto err_pd;
2649 
2650 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2651 				 PAGE_SIZE);
2652 	if (!ibdev->uar_map) {
2653 		err = -ENOMEM;
2654 		goto err_uar;
2655 	}
2656 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2657 
2658 	ibdev->dev = dev;
2659 	ibdev->bond_next_port	= 0;
2660 
2661 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2662 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2663 	ibdev->num_ports		= num_ports;
2664 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2665 						1 : ibdev->num_ports;
2666 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2667 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2668 
2669 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2670 
2671 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2672 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2673 	    IB_LINK_LAYER_ETHERNET) ||
2674 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2675 	    IB_LINK_LAYER_ETHERNET)))
2676 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2677 
2678 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2679 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2680 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2681 
2682 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2683 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2684 	}
2685 
2686 	if (check_flow_steering_support(dev)) {
2687 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2688 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2689 	}
2690 
2691 	if (!dev->caps.userspace_caps)
2692 		ibdev->ib_dev.ops.uverbs_abi_ver =
2693 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2694 
2695 	mlx4_ib_alloc_eqs(dev, ibdev);
2696 
2697 	spin_lock_init(&iboe->lock);
2698 
2699 	err = init_node_data(ibdev);
2700 	if (err)
2701 		goto err_map;
2702 	mlx4_init_sl2vl_tbl(ibdev);
2703 
2704 	for (i = 0; i < ibdev->num_ports; ++i) {
2705 		mutex_init(&ibdev->counters_table[i].mutex);
2706 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2707 		iboe->last_port_state[i] = IB_PORT_DOWN;
2708 	}
2709 
2710 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2711 	for (i = 0; i < num_req_counters; ++i) {
2712 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2713 		allocated = 0;
2714 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2715 						IB_LINK_LAYER_ETHERNET) {
2716 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2717 						 MLX4_RES_USAGE_DRIVER);
2718 			/* if failed to allocate a new counter, use default */
2719 			if (err)
2720 				counter_index =
2721 					mlx4_get_default_counter_index(dev,
2722 								       i + 1);
2723 			else
2724 				allocated = 1;
2725 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2726 			counter_index = mlx4_get_default_counter_index(dev,
2727 								       i + 1);
2728 		}
2729 		new_counter_index = kmalloc_obj(*new_counter_index);
2730 		if (!new_counter_index) {
2731 			err = -ENOMEM;
2732 			if (allocated)
2733 				mlx4_counter_free(ibdev->dev, counter_index);
2734 			goto err_counter;
2735 		}
2736 		new_counter_index->index = counter_index;
2737 		new_counter_index->allocated = allocated;
2738 		list_add_tail(&new_counter_index->list,
2739 			      &ibdev->counters_table[i].counters_list);
2740 		ibdev->counters_table[i].default_counter = counter_index;
2741 		pr_info("counter index %d for port %d allocated %d\n",
2742 			counter_index, i + 1, allocated);
2743 	}
2744 	if (mlx4_is_bonded(dev))
2745 		for (i = 1; i < ibdev->num_ports ; ++i) {
2746 			new_counter_index =
2747 					kmalloc_obj(struct counter_index);
2748 			if (!new_counter_index) {
2749 				err = -ENOMEM;
2750 				goto err_counter;
2751 			}
2752 			new_counter_index->index = counter_index;
2753 			new_counter_index->allocated = 0;
2754 			list_add_tail(&new_counter_index->list,
2755 				      &ibdev->counters_table[i].counters_list);
2756 			ibdev->counters_table[i].default_counter =
2757 								counter_index;
2758 		}
2759 
2760 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2761 		ib_num_ports++;
2762 
2763 	spin_lock_init(&ibdev->sm_lock);
2764 	mutex_init(&ibdev->cap_mask_mutex);
2765 	INIT_LIST_HEAD(&ibdev->qp_list);
2766 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2767 
2768 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2769 	    ib_num_ports) {
2770 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2771 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2772 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2773 					    &ibdev->steer_qpn_base, 0,
2774 					    MLX4_RES_USAGE_DRIVER);
2775 		if (err)
2776 			goto err_counter;
2777 
2778 		ibdev->ib_uc_qpns_bitmap = bitmap_alloc(ibdev->steer_qpn_count,
2779 							GFP_KERNEL);
2780 		if (!ibdev->ib_uc_qpns_bitmap) {
2781 			err = -ENOMEM;
2782 			goto err_steer_qp_release;
2783 		}
2784 
2785 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2786 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2787 				    ibdev->steer_qpn_count);
2788 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2789 					dev, ibdev->steer_qpn_base,
2790 					ibdev->steer_qpn_base +
2791 					ibdev->steer_qpn_count - 1);
2792 			if (err)
2793 				goto err_steer_free_bitmap;
2794 		} else {
2795 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2796 				    ibdev->steer_qpn_count);
2797 		}
2798 	}
2799 
2800 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2801 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2802 
2803 	err = mlx4_ib_alloc_diag_counters(ibdev);
2804 	if (err)
2805 		goto err_steer_free_bitmap;
2806 
2807 	err = ib_register_device(&ibdev->ib_dev, "mlx4_%d",
2808 				 &dev->persist->pdev->dev);
2809 	if (err)
2810 		goto err_diag_counters;
2811 
2812 	err = mlx4_ib_mad_init(ibdev);
2813 	if (err)
2814 		goto err_reg;
2815 
2816 	err = mlx4_ib_init_sriov(ibdev);
2817 	if (err)
2818 		goto err_mad;
2819 
2820 	if (!iboe->nb.notifier_call) {
2821 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2822 		err = register_netdevice_notifier(&iboe->nb);
2823 		if (err) {
2824 			iboe->nb.notifier_call = NULL;
2825 			goto err_notif;
2826 		}
2827 	}
2828 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2829 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2830 		if (err)
2831 			goto err_notif;
2832 	}
2833 
2834 	ibdev->ib_active = true;
2835 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2836 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2837 					 &ibdev->ib_dev);
2838 
2839 	if (mlx4_is_mfunc(ibdev->dev))
2840 		init_pkeys(ibdev);
2841 
2842 	/* create paravirt contexts for any VFs which are active */
2843 	if (mlx4_is_master(ibdev->dev)) {
2844 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2845 			if (j == mlx4_master_func_num(ibdev->dev))
2846 				continue;
2847 			if (mlx4_is_slave_active(ibdev->dev, j))
2848 				do_slave_init(ibdev, j, 1);
2849 		}
2850 	}
2851 
2852 	/* register mlx4 core notifier */
2853 	ibdev->mlx_nb.notifier_call = mlx4_ib_event;
2854 	err = mlx4_register_event_notifier(dev, &ibdev->mlx_nb);
2855 	WARN(err, "failed to register mlx4 event notifier (%d)", err);
2856 
2857 	auxiliary_set_drvdata(adev, ibdev);
2858 	return 0;
2859 
2860 err_notif:
2861 	if (ibdev->iboe.nb.notifier_call) {
2862 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2863 			pr_warn("failure unregistering notifier\n");
2864 		ibdev->iboe.nb.notifier_call = NULL;
2865 	}
2866 	flush_workqueue(wq);
2867 
2868 	mlx4_ib_close_sriov(ibdev);
2869 
2870 err_mad:
2871 	mlx4_ib_mad_cleanup(ibdev);
2872 
2873 err_reg:
2874 	ib_unregister_device(&ibdev->ib_dev);
2875 
2876 err_diag_counters:
2877 	mlx4_ib_diag_cleanup(ibdev);
2878 
2879 err_steer_free_bitmap:
2880 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2881 
2882 err_steer_qp_release:
2883 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2884 			      ibdev->steer_qpn_count);
2885 err_counter:
2886 	for (i = 0; i < ibdev->num_ports; ++i)
2887 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2888 
2889 err_map:
2890 	mlx4_ib_free_eqs(dev, ibdev);
2891 	iounmap(ibdev->uar_map);
2892 
2893 err_uar:
2894 	mlx4_uar_free(dev, &ibdev->priv_uar);
2895 
2896 err_pd:
2897 	mlx4_pd_free(dev, ibdev->priv_pdn);
2898 
2899 err_dealloc:
2900 	ib_dealloc_device(&ibdev->ib_dev);
2901 
2902 	return err;
2903 }
2904 
2905 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2906 {
2907 	int offset;
2908 
2909 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2910 
2911 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2912 					 dev->steer_qpn_count,
2913 					 get_count_order(count));
2914 	if (offset < 0)
2915 		return offset;
2916 
2917 	*qpn = dev->steer_qpn_base + offset;
2918 	return 0;
2919 }
2920 
2921 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2922 {
2923 	if (!qpn ||
2924 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2925 		return;
2926 
2927 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2928 		 qpn, dev->steer_qpn_base))
2929 		/* not supposed to be here */
2930 		return;
2931 
2932 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2933 			      qpn - dev->steer_qpn_base,
2934 			      get_count_order(count));
2935 }
2936 
2937 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2938 			 int is_attach)
2939 {
2940 	int err;
2941 	size_t flow_size;
2942 	struct ib_flow_attr *flow;
2943 	struct ib_flow_spec_ib *ib_spec;
2944 
2945 	if (is_attach) {
2946 		flow_size = sizeof(struct ib_flow_attr) +
2947 			    sizeof(struct ib_flow_spec_ib);
2948 		flow = kzalloc(flow_size, GFP_KERNEL);
2949 		if (!flow)
2950 			return -ENOMEM;
2951 		flow->port = mqp->port;
2952 		flow->num_of_specs = 1;
2953 		flow->size = flow_size;
2954 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2955 		ib_spec->type = IB_FLOW_SPEC_IB;
2956 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2957 		/* Add an empty rule for IB L2 */
2958 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2959 
2960 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow, MLX4_DOMAIN_NIC,
2961 					    MLX4_FS_REGULAR, &mqp->reg_id);
2962 		kfree(flow);
2963 		return err;
2964 	}
2965 
2966 	return __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2967 }
2968 
2969 static void mlx4_ib_remove(struct auxiliary_device *adev)
2970 {
2971 	struct mlx4_adev *madev = container_of(adev, struct mlx4_adev, adev);
2972 	struct mlx4_dev *dev = madev->mdev;
2973 	struct mlx4_ib_dev *ibdev = auxiliary_get_drvdata(adev);
2974 	int p;
2975 	int i;
2976 
2977 	mlx4_unregister_event_notifier(dev, &ibdev->mlx_nb);
2978 
2979 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2980 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2981 	ibdev->ib_active = false;
2982 	flush_workqueue(wq);
2983 
2984 	if (ibdev->iboe.nb.notifier_call) {
2985 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2986 			pr_warn("failure unregistering notifier\n");
2987 		ibdev->iboe.nb.notifier_call = NULL;
2988 	}
2989 
2990 	mlx4_ib_close_sriov(ibdev);
2991 	mlx4_ib_mad_cleanup(ibdev);
2992 	ib_unregister_device(&ibdev->ib_dev);
2993 	mlx4_ib_diag_cleanup(ibdev);
2994 
2995 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2996 			      ibdev->steer_qpn_count);
2997 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2998 
2999 	iounmap(ibdev->uar_map);
3000 	for (p = 0; p < ibdev->num_ports; ++p)
3001 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3002 
3003 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3004 		mlx4_CLOSE_PORT(dev, p);
3005 
3006 	mlx4_ib_free_eqs(dev, ibdev);
3007 
3008 	mlx4_uar_free(dev, &ibdev->priv_uar);
3009 	mlx4_pd_free(dev, ibdev->priv_pdn);
3010 	ib_dealloc_device(&ibdev->ib_dev);
3011 }
3012 
3013 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3014 {
3015 	struct mlx4_ib_demux_work **dm;
3016 	struct mlx4_dev *dev = ibdev->dev;
3017 	int i;
3018 	unsigned long flags;
3019 	struct mlx4_active_ports actv_ports;
3020 	unsigned int ports;
3021 	unsigned int first_port;
3022 
3023 	if (!mlx4_is_master(dev))
3024 		return;
3025 
3026 	actv_ports = mlx4_get_active_ports(dev, slave);
3027 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3028 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3029 
3030 	dm = kzalloc_objs(*dm, ports, GFP_ATOMIC);
3031 	if (!dm)
3032 		return;
3033 
3034 	for (i = 0; i < ports; i++) {
3035 		dm[i] = kmalloc_obj(struct mlx4_ib_demux_work, GFP_ATOMIC);
3036 		if (!dm[i]) {
3037 			while (--i >= 0)
3038 				kfree(dm[i]);
3039 			goto out;
3040 		}
3041 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3042 		dm[i]->port = first_port + i + 1;
3043 		dm[i]->slave = slave;
3044 		dm[i]->do_init = do_init;
3045 		dm[i]->dev = ibdev;
3046 	}
3047 	/* initialize or tear down tunnel QPs for the slave */
3048 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3049 	if (!ibdev->sriov.is_going_down) {
3050 		for (i = 0; i < ports; i++)
3051 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3052 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3053 	} else {
3054 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3055 		for (i = 0; i < ports; i++)
3056 			kfree(dm[i]);
3057 	}
3058 out:
3059 	kfree(dm);
3060 	return;
3061 }
3062 
3063 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3064 {
3065 	struct mlx4_ib_qp *mqp;
3066 	unsigned long flags_qp;
3067 	unsigned long flags_cq;
3068 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3069 	struct list_head    cq_notify_list;
3070 	struct mlx4_cq *mcq;
3071 	unsigned long flags;
3072 
3073 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3074 	INIT_LIST_HEAD(&cq_notify_list);
3075 
3076 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3077 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3078 
3079 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3080 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3081 		if (mqp->sq.tail != mqp->sq.head) {
3082 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3083 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3084 			if (send_mcq->mcq.comp &&
3085 			    mqp->ibqp.send_cq->comp_handler) {
3086 				if (!send_mcq->mcq.reset_notify_added) {
3087 					send_mcq->mcq.reset_notify_added = 1;
3088 					list_add_tail(&send_mcq->mcq.reset_notify,
3089 						      &cq_notify_list);
3090 				}
3091 			}
3092 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3093 		}
3094 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3095 		/* Now, handle the QP's receive queue */
3096 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3097 		/* no handling is needed for SRQ */
3098 		if (!mqp->ibqp.srq) {
3099 			if (mqp->rq.tail != mqp->rq.head) {
3100 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3101 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3102 				if (recv_mcq->mcq.comp &&
3103 				    mqp->ibqp.recv_cq->comp_handler) {
3104 					if (!recv_mcq->mcq.reset_notify_added) {
3105 						recv_mcq->mcq.reset_notify_added = 1;
3106 						list_add_tail(&recv_mcq->mcq.reset_notify,
3107 							      &cq_notify_list);
3108 					}
3109 				}
3110 				spin_unlock_irqrestore(&recv_mcq->lock,
3111 						       flags_cq);
3112 			}
3113 		}
3114 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3115 	}
3116 
3117 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3118 		mcq->comp(mcq);
3119 	}
3120 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3121 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3122 }
3123 
3124 static void handle_bonded_port_state_event(struct work_struct *work)
3125 {
3126 	struct ib_event_work *ew =
3127 		container_of(work, struct ib_event_work, work);
3128 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3129 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3130 	int i;
3131 	struct ib_event ibev;
3132 
3133 	kfree(ew);
3134 	spin_lock_bh(&ibdev->iboe.lock);
3135 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3136 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3137 		enum ib_port_state curr_port_state;
3138 
3139 		if (!curr_netdev)
3140 			continue;
3141 
3142 		curr_port_state =
3143 			(netif_running(curr_netdev) &&
3144 			 netif_carrier_ok(curr_netdev)) ?
3145 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3146 
3147 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3148 			curr_port_state : IB_PORT_ACTIVE;
3149 	}
3150 	spin_unlock_bh(&ibdev->iboe.lock);
3151 
3152 	ibev.device = &ibdev->ib_dev;
3153 	ibev.element.port_num = 1;
3154 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3155 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3156 
3157 	ib_dispatch_event(&ibev);
3158 }
3159 
3160 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3161 {
3162 	u64 sl2vl;
3163 	int err;
3164 
3165 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3166 	if (err) {
3167 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3168 		       port, err);
3169 		sl2vl = 0;
3170 	}
3171 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3172 }
3173 
3174 static void ib_sl2vl_update_work(struct work_struct *work)
3175 {
3176 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3177 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3178 	int port = ew->port;
3179 
3180 	mlx4_ib_sl2vl_update(mdev, port);
3181 
3182 	kfree(ew);
3183 }
3184 
3185 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3186 				     int port)
3187 {
3188 	struct ib_event_work *ew;
3189 
3190 	ew = kmalloc_obj(*ew, GFP_ATOMIC);
3191 	if (ew) {
3192 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3193 		ew->port = port;
3194 		ew->ib_dev = ibdev;
3195 		queue_work(wq, &ew->work);
3196 	}
3197 }
3198 
3199 static int mlx4_ib_event(struct notifier_block *this, unsigned long event,
3200 			 void *param)
3201 {
3202 	struct mlx4_ib_dev *ibdev =
3203 		container_of(this, struct mlx4_ib_dev, mlx_nb);
3204 	struct mlx4_dev *dev = ibdev->dev;
3205 	struct ib_event ibev;
3206 	struct mlx4_eqe *eqe = NULL;
3207 	struct ib_event_work *ew;
3208 	int p = 0;
3209 
3210 	if (mlx4_is_bonded(dev) &&
3211 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3212 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3213 		ew = kmalloc_obj(*ew, GFP_ATOMIC);
3214 		if (!ew)
3215 			return NOTIFY_DONE;
3216 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3217 		ew->ib_dev = ibdev;
3218 		queue_work(wq, &ew->work);
3219 		return NOTIFY_DONE;
3220 	}
3221 
3222 	switch (event) {
3223 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3224 		break;
3225 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3226 		eqe = (struct mlx4_eqe *)param;
3227 		break;
3228 	default:
3229 		p = *(int *)param;
3230 		break;
3231 	}
3232 
3233 	switch (event) {
3234 	case MLX4_DEV_EVENT_PORT_UP:
3235 		if (p > ibdev->num_ports)
3236 			return NOTIFY_DONE;
3237 		if (!mlx4_is_slave(dev) &&
3238 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3239 			IB_LINK_LAYER_INFINIBAND) {
3240 			if (mlx4_is_master(dev))
3241 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3242 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3243 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3244 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3245 		}
3246 		ibev.event = IB_EVENT_PORT_ACTIVE;
3247 		break;
3248 
3249 	case MLX4_DEV_EVENT_PORT_DOWN:
3250 		if (p > ibdev->num_ports)
3251 			return NOTIFY_DONE;
3252 		ibev.event = IB_EVENT_PORT_ERR;
3253 		break;
3254 
3255 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3256 		ibdev->ib_active = false;
3257 		ibev.event = IB_EVENT_DEVICE_FATAL;
3258 		mlx4_ib_handle_catas_error(ibdev);
3259 		break;
3260 
3261 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3262 		ew = kmalloc_obj(*ew, GFP_ATOMIC);
3263 		if (!ew)
3264 			return NOTIFY_DONE;
3265 
3266 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3267 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3268 		ew->ib_dev = ibdev;
3269 		/* need to queue only for port owner, which uses GEN_EQE */
3270 		if (mlx4_is_master(dev))
3271 			queue_work(wq, &ew->work);
3272 		else
3273 			handle_port_mgmt_change_event(&ew->work);
3274 		return NOTIFY_DONE;
3275 
3276 	case MLX4_DEV_EVENT_SLAVE_INIT:
3277 		/* here, p is the slave id */
3278 		do_slave_init(ibdev, p, 1);
3279 		if (mlx4_is_master(dev)) {
3280 			int i;
3281 
3282 			for (i = 1; i <= ibdev->num_ports; i++) {
3283 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3284 					== IB_LINK_LAYER_INFINIBAND)
3285 					mlx4_ib_slave_alias_guid_event(ibdev,
3286 								       p, i,
3287 								       1);
3288 			}
3289 		}
3290 		return NOTIFY_DONE;
3291 
3292 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3293 		if (mlx4_is_master(dev)) {
3294 			int i;
3295 
3296 			for (i = 1; i <= ibdev->num_ports; i++) {
3297 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3298 					== IB_LINK_LAYER_INFINIBAND)
3299 					mlx4_ib_slave_alias_guid_event(ibdev,
3300 								       p, i,
3301 								       0);
3302 			}
3303 		}
3304 		/* here, p is the slave id */
3305 		do_slave_init(ibdev, p, 0);
3306 		return NOTIFY_DONE;
3307 
3308 	default:
3309 		return NOTIFY_DONE;
3310 	}
3311 
3312 	ibev.device	      = &ibdev->ib_dev;
3313 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3314 
3315 	ib_dispatch_event(&ibev);
3316 	return NOTIFY_DONE;
3317 }
3318 
3319 static const struct auxiliary_device_id mlx4_ib_id_table[] = {
3320 	{ .name = MLX4_ADEV_NAME ".ib" },
3321 	{},
3322 };
3323 
3324 MODULE_DEVICE_TABLE(auxiliary, mlx4_ib_id_table);
3325 
3326 static struct mlx4_adrv mlx4_ib_adrv = {
3327 	.adrv = {
3328 		.name	= "ib",
3329 		.probe	= mlx4_ib_probe,
3330 		.remove	= mlx4_ib_remove,
3331 		.id_table = mlx4_ib_id_table,
3332 	},
3333 	.protocol	= MLX4_PROT_IB_IPV6,
3334 	.flags		= MLX4_INTFF_BONDING
3335 };
3336 
3337 static int __init mlx4_ib_init(void)
3338 {
3339 	int err;
3340 
3341 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3342 	if (!wq)
3343 		return -ENOMEM;
3344 
3345 	err = mlx4_ib_qp_event_init();
3346 	if (err)
3347 		goto clean_qp_event;
3348 
3349 	err = mlx4_ib_cm_init();
3350 	if (err)
3351 		goto clean_wq;
3352 
3353 	err = mlx4_ib_mcg_init();
3354 	if (err)
3355 		goto clean_cm;
3356 
3357 	err = mlx4_register_auxiliary_driver(&mlx4_ib_adrv);
3358 	if (err)
3359 		goto clean_mcg;
3360 
3361 	return 0;
3362 
3363 clean_mcg:
3364 	mlx4_ib_mcg_destroy();
3365 
3366 clean_cm:
3367 	mlx4_ib_cm_destroy();
3368 
3369 clean_wq:
3370 	mlx4_ib_qp_event_cleanup();
3371 
3372 clean_qp_event:
3373 	destroy_workqueue(wq);
3374 	return err;
3375 }
3376 
3377 static void __exit mlx4_ib_cleanup(void)
3378 {
3379 	mlx4_unregister_auxiliary_driver(&mlx4_ib_adrv);
3380 	mlx4_ib_mcg_destroy();
3381 	mlx4_ib_cm_destroy();
3382 	mlx4_ib_qp_event_cleanup();
3383 	destroy_workqueue(wq);
3384 }
3385 
3386 module_init(mlx4_ib_init);
3387 module_exit(mlx4_ib_cleanup);
3388