xref: /freebsd/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c (revision 47dd1d1b619cc035b82b49a91a25544309ff95ae)
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 #define	LINUXKPI_PARAM_PREFIX mlx4_
35 
36 #include <linux/module.h>
37 #include <linux/slab.h>
38 #include <linux/errno.h>
39 #include <linux/etherdevice.h>
40 #include <linux/netdevice.h>
41 #include <linux/inetdevice.h>
42 #include <linux/if_vlan.h>
43 #include <linux/fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/notifier.h>
46 #include <linux/delay.h>
47 
48 #include <net/ipv6.h>
49 
50 #include <rdma/ib_smi.h>
51 #include <rdma/ib_user_verbs.h>
52 #include <rdma/ib_addr.h>
53 #include <rdma/ib_cache.h>
54 
55 #include <dev/mlx4/driver.h>
56 #include <dev/mlx4/cmd.h>
57 #include <dev/mlx4/qp.h>
58 #include <linux/sched.h>
59 #include <linux/page.h>
60 #include <linux/printk.h>
61 #include "mlx4_ib.h"
62 #include <rdma/mlx4-abi.h>
63 #include "wc.h"
64 
65 #define DRV_NAME	MLX4_IB_DRV_NAME
66 #ifndef DRV_VERSION
67 #define DRV_VERSION	"3.4.1"
68 #endif
69 #define DRV_RELDATE	"February 2018"
70 
71 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
72 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
73 #define MLX4_IB_CARD_REV_A0   0xA0
74 
75 MODULE_AUTHOR("Roland Dreier");
76 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
77 MODULE_LICENSE("Dual BSD/GPL");
78 
79 int mlx4_ib_sm_guid_assign = 0;
80 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
81 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
82 
83 static const char mlx4_ib_version[] =
84 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
85 	DRV_VERSION " (" DRV_RELDATE ")\n";
86 
87 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
88 
89 static struct workqueue_struct *wq;
90 
91 static void init_query_mad(struct ib_smp *mad)
92 {
93 	mad->base_version  = 1;
94 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
95 	mad->class_version = 1;
96 	mad->method	   = IB_MGMT_METHOD_GET;
97 }
98 
99 static int check_flow_steering_support(struct mlx4_dev *dev)
100 {
101 	int eth_num_ports = 0;
102 	int ib_num_ports = 0;
103 
104 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
105 
106 	if (dmfs) {
107 		int i;
108 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
109 			eth_num_ports++;
110 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
111 			ib_num_ports++;
112 		dmfs &= (!ib_num_ports ||
113 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
114 			(!eth_num_ports ||
115 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
116 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
117 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
118 			dmfs = 0;
119 		}
120 	}
121 	return dmfs;
122 }
123 
124 static int num_ib_ports(struct mlx4_dev *dev)
125 {
126 	int ib_ports = 0;
127 	int i;
128 
129 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
130 		ib_ports++;
131 
132 	return ib_ports;
133 }
134 
135 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
136 {
137 	struct mlx4_ib_dev *ibdev = to_mdev(device);
138 	struct net_device *dev;
139 
140 	rcu_read_lock();
141 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
142 
143 #if 0
144 	if (dev) {
145 		if (mlx4_is_bonded(ibdev->dev)) {
146 			struct net_device *upper = NULL;
147 
148 			upper = netdev_master_upper_dev_get_rcu(dev);
149 			if (upper) {
150 				struct net_device *active;
151 
152 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
153 				if (active)
154 					dev = active;
155 			}
156 		}
157 	}
158 #endif
159 	if (dev)
160 		dev_hold(dev);
161 
162 	rcu_read_unlock();
163 	return dev;
164 }
165 
166 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
167 				  struct mlx4_ib_dev *ibdev,
168 				  u8 port_num)
169 {
170 	struct mlx4_cmd_mailbox *mailbox;
171 	int err;
172 	struct mlx4_dev *dev = ibdev->dev;
173 	int i;
174 	union ib_gid *gid_tbl;
175 
176 	mailbox = mlx4_alloc_cmd_mailbox(dev);
177 	if (IS_ERR(mailbox))
178 		return -ENOMEM;
179 
180 	gid_tbl = mailbox->buf;
181 
182 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
183 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
184 
185 	err = mlx4_cmd(dev, mailbox->dma,
186 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
187 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 		       MLX4_CMD_WRAPPED);
189 	if (mlx4_is_bonded(dev))
190 		err += mlx4_cmd(dev, mailbox->dma,
191 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
192 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
193 				MLX4_CMD_WRAPPED);
194 
195 	mlx4_free_cmd_mailbox(dev, mailbox);
196 	return err;
197 }
198 
199 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
200 				     struct mlx4_ib_dev *ibdev,
201 				     u8 port_num)
202 {
203 	struct mlx4_cmd_mailbox *mailbox;
204 	int err;
205 	struct mlx4_dev *dev = ibdev->dev;
206 	int i;
207 	struct {
208 		union ib_gid	gid;
209 		__be32		rsrvd1[2];
210 		__be16		rsrvd2;
211 		u8		type;
212 		u8		version;
213 		__be32		rsrvd3;
214 	} *gid_tbl;
215 
216 	mailbox = mlx4_alloc_cmd_mailbox(dev);
217 	if (IS_ERR(mailbox))
218 		return -ENOMEM;
219 
220 	gid_tbl = mailbox->buf;
221 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
222 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
223 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
224 			gid_tbl[i].version = 2;
225 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
226 				gid_tbl[i].type = 1;
227 			else
228 				memset(&gid_tbl[i].gid, 0, 12);
229 		}
230 	}
231 
232 	err = mlx4_cmd(dev, mailbox->dma,
233 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
234 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
235 		       MLX4_CMD_WRAPPED);
236 	if (mlx4_is_bonded(dev))
237 		err += mlx4_cmd(dev, mailbox->dma,
238 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
239 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
240 				MLX4_CMD_WRAPPED);
241 
242 	mlx4_free_cmd_mailbox(dev, mailbox);
243 	return err;
244 }
245 
246 static int mlx4_ib_update_gids(struct gid_entry *gids,
247 			       struct mlx4_ib_dev *ibdev,
248 			       u8 port_num)
249 {
250 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
251 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
252 
253 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
254 }
255 
256 static int mlx4_ib_add_gid(struct ib_device *device,
257 			   u8 port_num,
258 			   unsigned int index,
259 			   const union ib_gid *gid,
260 			   const struct ib_gid_attr *attr,
261 			   void **context)
262 {
263 	struct mlx4_ib_dev *ibdev = to_mdev(device);
264 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
265 	struct mlx4_port_gid_table   *port_gid_table;
266 	int free = -1, found = -1;
267 	int ret = 0;
268 	int hw_update = 0;
269 	int i;
270 	struct gid_entry *gids = NULL;
271 
272 	if (!rdma_cap_roce_gid_table(device, port_num))
273 		return -EINVAL;
274 
275 	if (port_num > MLX4_MAX_PORTS)
276 		return -EINVAL;
277 
278 	if (!context)
279 		return -EINVAL;
280 
281 	port_gid_table = &iboe->gids[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, gid, sizeof(*gid)) &&
285 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
286 			found = i;
287 			break;
288 		}
289 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
290 			free = i; /* HW has space */
291 	}
292 
293 	if (found < 0) {
294 		if (free < 0) {
295 			ret = -ENOSPC;
296 		} else {
297 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
298 			if (!port_gid_table->gids[free].ctx) {
299 				ret = -ENOMEM;
300 			} else {
301 				*context = port_gid_table->gids[free].ctx;
302 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
303 				port_gid_table->gids[free].gid_type = attr->gid_type;
304 				port_gid_table->gids[free].ctx->real_index = free;
305 				port_gid_table->gids[free].ctx->refcount = 1;
306 				hw_update = 1;
307 			}
308 		}
309 	} else {
310 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
311 		*context = ctx;
312 		ctx->refcount++;
313 	}
314 	if (!ret && hw_update) {
315 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
316 		if (!gids) {
317 			ret = -ENOMEM;
318 		} else {
319 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
320 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
321 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
322 			}
323 		}
324 	}
325 	spin_unlock_bh(&iboe->lock);
326 
327 	if (!ret && hw_update) {
328 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
329 		kfree(gids);
330 	}
331 
332 	return ret;
333 }
334 
335 static int mlx4_ib_del_gid(struct ib_device *device,
336 			   u8 port_num,
337 			   unsigned int index,
338 			   void **context)
339 {
340 	struct gid_cache_context *ctx = *context;
341 	struct mlx4_ib_dev *ibdev = to_mdev(device);
342 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
343 	struct mlx4_port_gid_table   *port_gid_table;
344 	int ret = 0;
345 	int hw_update = 0;
346 	struct gid_entry *gids = NULL;
347 
348 	if (!rdma_cap_roce_gid_table(device, port_num))
349 		return -EINVAL;
350 
351 	if (port_num > MLX4_MAX_PORTS)
352 		return -EINVAL;
353 
354 	port_gid_table = &iboe->gids[port_num - 1];
355 	spin_lock_bh(&iboe->lock);
356 	if (ctx) {
357 		ctx->refcount--;
358 		if (!ctx->refcount) {
359 			unsigned int real_index = ctx->real_index;
360 
361 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
362 			kfree(port_gid_table->gids[real_index].ctx);
363 			port_gid_table->gids[real_index].ctx = NULL;
364 			hw_update = 1;
365 		}
366 	}
367 	if (!ret && hw_update) {
368 		int i;
369 
370 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
371 		if (!gids) {
372 			ret = -ENOMEM;
373 		} else {
374 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
375 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
376 		}
377 	}
378 	spin_unlock_bh(&iboe->lock);
379 
380 	if (!ret && hw_update) {
381 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
382 		kfree(gids);
383 	}
384 	return ret;
385 }
386 
387 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
388 				    u8 port_num, int index)
389 {
390 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
391 	struct gid_cache_context *ctx = NULL;
392 	union ib_gid gid;
393 	struct mlx4_port_gid_table   *port_gid_table;
394 	int real_index = -EINVAL;
395 	int i;
396 	int ret;
397 	unsigned long flags;
398 	struct ib_gid_attr attr;
399 
400 	if (port_num > MLX4_MAX_PORTS)
401 		return -EINVAL;
402 
403 	if (mlx4_is_bonded(ibdev->dev))
404 		port_num = 1;
405 
406 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
407 		return index;
408 
409 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
410 	if (ret)
411 		return ret;
412 
413 	if (attr.ndev)
414 		dev_put(attr.ndev);
415 
416 	if (!memcmp(&gid, &zgid, sizeof(gid)))
417 		return -EINVAL;
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, &gid, sizeof(gid)) &&
424 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
425 			ctx = port_gid_table->gids[i].ctx;
426 			break;
427 		}
428 	if (ctx)
429 		real_index = ctx->real_index;
430 	spin_unlock_irqrestore(&iboe->lock, flags);
431 	return real_index;
432 }
433 
434 static int mlx4_ib_query_device(struct ib_device *ibdev,
435 				struct ib_device_attr *props,
436 				struct ib_udata *uhw)
437 {
438 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
439 	struct ib_smp *in_mad  = NULL;
440 	struct ib_smp *out_mad = NULL;
441 	int err = -ENOMEM;
442 	int have_ib_ports;
443 	struct mlx4_uverbs_ex_query_device cmd;
444 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
445 	struct mlx4_clock_params clock_params;
446 
447 	if (uhw->inlen) {
448 		if (uhw->inlen < sizeof(cmd))
449 			return -EINVAL;
450 
451 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
452 		if (err)
453 			return err;
454 
455 		if (cmd.comp_mask)
456 			return -EINVAL;
457 
458 		if (cmd.reserved)
459 			return -EINVAL;
460 	}
461 
462 	resp.response_length = offsetof(typeof(resp), response_length) +
463 		sizeof(resp.response_length);
464 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
465 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
466 	if (!in_mad || !out_mad)
467 		goto out;
468 
469 	init_query_mad(in_mad);
470 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
471 
472 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
473 			   1, NULL, NULL, in_mad, out_mad);
474 	if (err)
475 		goto out;
476 
477 	memset(props, 0, sizeof *props);
478 
479 	have_ib_ports = num_ib_ports(dev->dev);
480 
481 	props->fw_ver = dev->dev->caps.fw_ver;
482 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
483 		IB_DEVICE_PORT_ACTIVE_EVENT		|
484 		IB_DEVICE_SYS_IMAGE_GUID		|
485 		IB_DEVICE_RC_RNR_NAK_GEN		|
486 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
488 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
490 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
492 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
493 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
494 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
496 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
497 	if (dev->dev->caps.max_gso_sz &&
498 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
499 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
500 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
501 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
502 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
503 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
504 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
505 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
506 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
507 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
508 		props->device_cap_flags |= IB_DEVICE_XRC;
509 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
510 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
511 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
512 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
513 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
514 		else
515 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
516 	}
517 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
518 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
519 
520 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
521 
522 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
523 		0xffffff;
524 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
525 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
526 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
527 
528 	props->max_mr_size	   = ~0ull;
529 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
530 	props->max_qp		   = dev->dev->quotas.qp;
531 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
532 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
533 					 dev->dev->caps.max_rq_sg);
534 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
535 	props->max_cq		   = dev->dev->quotas.cq;
536 	props->max_cqe		   = dev->dev->caps.max_cqes;
537 	props->max_mr		   = dev->dev->quotas.mpt;
538 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
539 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
540 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
541 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
542 	props->max_srq		   = dev->dev->quotas.srq;
543 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
544 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
545 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
546 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
547 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
548 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
549 	props->masked_atomic_cap   = props->atomic_cap;
550 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
551 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
552 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
553 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
554 					   props->max_mcast_grp;
555 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
556 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
557 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
558 
559 	if (!mlx4_is_slave(dev->dev))
560 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
561 
562 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
563 		resp.response_length += sizeof(resp.hca_core_clock_offset);
564 		if (!err && !mlx4_is_slave(dev->dev)) {
565 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
566 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
567 		}
568 	}
569 
570 	if (uhw->outlen) {
571 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
572 		if (err)
573 			goto out;
574 	}
575 out:
576 	kfree(in_mad);
577 	kfree(out_mad);
578 
579 	return err;
580 }
581 
582 static enum rdma_link_layer
583 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
584 {
585 	struct mlx4_dev *dev = to_mdev(device)->dev;
586 
587 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
588 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
589 }
590 
591 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
592 			      struct ib_port_attr *props, int netw_view)
593 {
594 	struct ib_smp *in_mad  = NULL;
595 	struct ib_smp *out_mad = NULL;
596 	int ext_active_speed;
597 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
598 	int err = -ENOMEM;
599 
600 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
601 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
602 	if (!in_mad || !out_mad)
603 		goto out;
604 
605 	init_query_mad(in_mad);
606 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
607 	in_mad->attr_mod = cpu_to_be32(port);
608 
609 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
610 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
611 
612 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
613 				in_mad, out_mad);
614 	if (err)
615 		goto out;
616 
617 
618 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
619 	props->lmc		= out_mad->data[34] & 0x7;
620 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
621 	props->sm_sl		= out_mad->data[36] & 0xf;
622 	props->state		= out_mad->data[32] & 0xf;
623 	props->phys_state	= out_mad->data[33] >> 4;
624 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
625 	if (netw_view)
626 		props->gid_tbl_len = out_mad->data[50];
627 	else
628 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
629 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
630 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
631 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
632 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
633 	props->active_width	= out_mad->data[31] & 0xf;
634 	props->active_speed	= out_mad->data[35] >> 4;
635 	props->max_mtu		= out_mad->data[41] & 0xf;
636 	props->active_mtu	= out_mad->data[36] >> 4;
637 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
638 	props->max_vl_num	= out_mad->data[37] >> 4;
639 	props->init_type_reply	= out_mad->data[41] >> 4;
640 
641 	/* Check if extended speeds (EDR/FDR/...) are supported */
642 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
643 		ext_active_speed = out_mad->data[62] >> 4;
644 
645 		switch (ext_active_speed) {
646 		case 1:
647 			props->active_speed = IB_SPEED_FDR;
648 			break;
649 		case 2:
650 			props->active_speed = IB_SPEED_EDR;
651 			break;
652 		}
653 	}
654 
655 	/* If reported active speed is QDR, check if is FDR-10 */
656 	if (props->active_speed == IB_SPEED_QDR) {
657 		init_query_mad(in_mad);
658 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
659 		in_mad->attr_mod = cpu_to_be32(port);
660 
661 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
662 				   NULL, NULL, in_mad, out_mad);
663 		if (err)
664 			goto out;
665 
666 		/* Checking LinkSpeedActive for FDR-10 */
667 		if (out_mad->data[15] & 0x1)
668 			props->active_speed = IB_SPEED_FDR10;
669 	}
670 
671 	/* Avoid wrong speed value returned by FW if the IB link is down. */
672 	if (props->state == IB_PORT_DOWN)
673 		 props->active_speed = IB_SPEED_SDR;
674 
675 out:
676 	kfree(in_mad);
677 	kfree(out_mad);
678 	return err;
679 }
680 
681 static u8 state_to_phys_state(enum ib_port_state state)
682 {
683 	return state == IB_PORT_ACTIVE ? 5 : 3;
684 }
685 
686 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
687 			       struct ib_port_attr *props, int netw_view)
688 {
689 
690 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
691 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
692 	struct net_device *ndev;
693 	enum ib_mtu tmp;
694 	struct mlx4_cmd_mailbox *mailbox;
695 	int err = 0;
696 	int is_bonded = mlx4_is_bonded(mdev->dev);
697 
698 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
699 	if (IS_ERR(mailbox))
700 		return PTR_ERR(mailbox);
701 
702 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
703 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
704 			   MLX4_CMD_WRAPPED);
705 	if (err)
706 		goto out;
707 
708 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
709 						IB_WIDTH_4X : IB_WIDTH_1X;
710 	props->active_speed	= IB_SPEED_QDR;
711 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
712 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
713 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
714 	props->pkey_tbl_len	= 1;
715 	props->max_mtu		= IB_MTU_4096;
716 	props->max_vl_num	= 2;
717 	props->state		= IB_PORT_DOWN;
718 	props->phys_state	= state_to_phys_state(props->state);
719 	props->active_mtu	= IB_MTU_256;
720 	spin_lock_bh(&iboe->lock);
721 	ndev = iboe->netdevs[port - 1];
722 	if (ndev && is_bonded) {
723 #if 0
724 		rcu_read_lock(); /* required to get upper dev */
725 		ndev = netdev_master_upper_dev_get_rcu(ndev);
726 		rcu_read_unlock();
727 #endif
728 	}
729 	if (!ndev)
730 		goto out_unlock;
731 
732 	tmp = iboe_get_mtu(ndev->if_mtu);
733 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
734 
735 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
736 					IB_PORT_ACTIVE : IB_PORT_DOWN;
737 	props->phys_state	= state_to_phys_state(props->state);
738 out_unlock:
739 	spin_unlock_bh(&iboe->lock);
740 out:
741 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
742 	return err;
743 }
744 
745 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
746 			 struct ib_port_attr *props, int netw_view)
747 {
748 	int err;
749 
750 	memset(props, 0, sizeof *props);
751 
752 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
753 		ib_link_query_port(ibdev, port, props, netw_view) :
754 				eth_link_query_port(ibdev, port, props, netw_view);
755 
756 	return err;
757 }
758 
759 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
760 			      struct ib_port_attr *props)
761 {
762 	/* returns host view */
763 	return __mlx4_ib_query_port(ibdev, port, props, 0);
764 }
765 
766 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
767 			union ib_gid *gid, int netw_view)
768 {
769 	struct ib_smp *in_mad  = NULL;
770 	struct ib_smp *out_mad = NULL;
771 	int err = -ENOMEM;
772 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
773 	int clear = 0;
774 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
775 
776 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
777 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
778 	if (!in_mad || !out_mad)
779 		goto out;
780 
781 	init_query_mad(in_mad);
782 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
783 	in_mad->attr_mod = cpu_to_be32(port);
784 
785 	if (mlx4_is_mfunc(dev->dev) && netw_view)
786 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
787 
788 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
789 	if (err)
790 		goto out;
791 
792 	memcpy(gid->raw, out_mad->data + 8, 8);
793 
794 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
795 		if (index) {
796 			/* For any index > 0, return the null guid */
797 			err = 0;
798 			clear = 1;
799 			goto out;
800 		}
801 	}
802 
803 	init_query_mad(in_mad);
804 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
805 	in_mad->attr_mod = cpu_to_be32(index / 8);
806 
807 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
808 			   NULL, NULL, in_mad, out_mad);
809 	if (err)
810 		goto out;
811 
812 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
813 
814 out:
815 	if (clear)
816 		memset(gid->raw + 8, 0, 8);
817 	kfree(in_mad);
818 	kfree(out_mad);
819 	return err;
820 }
821 
822 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
823 			     union ib_gid *gid)
824 {
825 	int ret;
826 
827 	if (rdma_protocol_ib(ibdev, port))
828 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
829 
830 	if (!rdma_protocol_roce(ibdev, port))
831 		return -ENODEV;
832 
833 	if (!rdma_cap_roce_gid_table(ibdev, port))
834 		return -ENODEV;
835 
836 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
837 	if (ret == -EAGAIN) {
838 		memcpy(gid, &zgid, sizeof(*gid));
839 		return 0;
840 	}
841 
842 	return ret;
843 }
844 
845 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
846 {
847 	union sl2vl_tbl_to_u64 sl2vl64;
848 	struct ib_smp *in_mad  = NULL;
849 	struct ib_smp *out_mad = NULL;
850 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
851 	int err = -ENOMEM;
852 	int jj;
853 
854 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
855 		*sl2vl_tbl = 0;
856 		return 0;
857 	}
858 
859 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
860 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
861 	if (!in_mad || !out_mad)
862 		goto out;
863 
864 	init_query_mad(in_mad);
865 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
866 	in_mad->attr_mod = 0;
867 
868 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
869 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
870 
871 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
872 			   in_mad, out_mad);
873 	if (err)
874 		goto out;
875 
876 	for (jj = 0; jj < 8; jj++)
877 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
878 	*sl2vl_tbl = sl2vl64.sl64;
879 
880 out:
881 	kfree(in_mad);
882 	kfree(out_mad);
883 	return err;
884 }
885 
886 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
887 {
888 	u64 sl2vl;
889 	int i;
890 	int err;
891 
892 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
893 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
894 			continue;
895 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
896 		if (err) {
897 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
898 			       i, err);
899 			sl2vl = 0;
900 		}
901 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
902 	}
903 }
904 
905 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
906 			 u16 *pkey, int netw_view)
907 {
908 	struct ib_smp *in_mad  = NULL;
909 	struct ib_smp *out_mad = NULL;
910 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
911 	int err = -ENOMEM;
912 
913 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
914 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
915 	if (!in_mad || !out_mad)
916 		goto out;
917 
918 	init_query_mad(in_mad);
919 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
920 	in_mad->attr_mod = cpu_to_be32(index / 32);
921 
922 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
923 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
924 
925 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
926 			   in_mad, out_mad);
927 	if (err)
928 		goto out;
929 
930 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
931 
932 out:
933 	kfree(in_mad);
934 	kfree(out_mad);
935 	return err;
936 }
937 
938 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
939 {
940 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
941 }
942 
943 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
944 				 struct ib_device_modify *props)
945 {
946 	struct mlx4_cmd_mailbox *mailbox;
947 	unsigned long flags;
948 
949 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
950 		return -EOPNOTSUPP;
951 
952 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
953 		return 0;
954 
955 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
956 		return -EOPNOTSUPP;
957 
958 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
959 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
960 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
961 
962 	/*
963 	 * If possible, pass node desc to FW, so it can generate
964 	 * a 144 trap.  If cmd fails, just ignore.
965 	 */
966 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
967 	if (IS_ERR(mailbox))
968 		return 0;
969 
970 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
971 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
972 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
973 
974 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
975 
976 	return 0;
977 }
978 
979 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
980 			    u32 cap_mask)
981 {
982 	struct mlx4_cmd_mailbox *mailbox;
983 	int err;
984 
985 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
986 	if (IS_ERR(mailbox))
987 		return PTR_ERR(mailbox);
988 
989 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
990 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
991 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
992 	} else {
993 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
994 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
995 	}
996 
997 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
998 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
999 		       MLX4_CMD_WRAPPED);
1000 
1001 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1002 	return err;
1003 }
1004 
1005 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1006 			       struct ib_port_modify *props)
1007 {
1008 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1009 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1010 	struct ib_port_attr attr;
1011 	u32 cap_mask;
1012 	int err;
1013 
1014 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1015 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1016 	 * violations and port capabilities are not meaningful.
1017 	 */
1018 	if (is_eth)
1019 		return 0;
1020 
1021 	mutex_lock(&mdev->cap_mask_mutex);
1022 
1023 	err = mlx4_ib_query_port(ibdev, port, &attr);
1024 	if (err)
1025 		goto out;
1026 
1027 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1028 		~props->clr_port_cap_mask;
1029 
1030 	err = mlx4_ib_SET_PORT(mdev, port,
1031 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1032 			       cap_mask);
1033 
1034 out:
1035 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1036 	return err;
1037 }
1038 
1039 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1040 						  struct ib_udata *udata)
1041 {
1042 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1043 	struct mlx4_ib_ucontext *context;
1044 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1045 	struct mlx4_ib_alloc_ucontext_resp resp;
1046 	int err;
1047 
1048 	if (!dev->ib_active)
1049 		return ERR_PTR(-EAGAIN);
1050 
1051 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1052 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1053 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1054 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1055 	} else {
1056 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1057 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1058 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1059 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1060 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1061 	}
1062 
1063 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1064 	if (!context)
1065 		return ERR_PTR(-ENOMEM);
1066 
1067 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1068 	if (err) {
1069 		kfree(context);
1070 		return ERR_PTR(err);
1071 	}
1072 
1073 	INIT_LIST_HEAD(&context->db_page_list);
1074 	mutex_init(&context->db_page_mutex);
1075 
1076 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1077 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1078 	else
1079 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1080 
1081 	if (err) {
1082 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1083 		kfree(context);
1084 		return ERR_PTR(-EFAULT);
1085 	}
1086 
1087 	return &context->ibucontext;
1088 }
1089 
1090 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1091 {
1092 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1093 
1094 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1095 	kfree(context);
1096 
1097 	return 0;
1098 }
1099 
1100 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1101 {
1102 	/* vma_open is called when a new VMA is created on top of our VMA.
1103 	 * This is done through either mremap flow or split_vma (usually due
1104 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1105 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1106 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1107 	 * calling us again and trying to do incorrect actions. We assume that
1108 	 * the original vma size is exactly a single page that there will be no
1109 	 * "splitting" operations on.
1110 	 */
1111 	area->vm_ops = NULL;
1112 }
1113 
1114 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1115 {
1116 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1117 
1118 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1119 	 * file itself is closed, therefore no sync is needed with the regular
1120 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1121 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1122 	 * The close operation is usually called under mm->mmap_sem except when
1123 	 * process is exiting.  The exiting case is handled explicitly as part
1124 	 * of mlx4_ib_disassociate_ucontext.
1125 	 */
1126 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1127 				area->vm_private_data;
1128 
1129 	/* set the vma context pointer to null in the mlx4_ib driver's private
1130 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1131 	 */
1132 	mlx4_ib_vma_priv_data->vma = NULL;
1133 }
1134 
1135 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1136 	.open = mlx4_ib_vma_open,
1137 	.close = mlx4_ib_vma_close
1138 };
1139 
1140 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1141 				 struct mlx4_ib_vma_private_data *vma_private_data)
1142 {
1143 	vma_private_data->vma = vma;
1144 	vma->vm_private_data = vma_private_data;
1145 	vma->vm_ops =  &mlx4_ib_vm_ops;
1146 }
1147 
1148 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1149 {
1150 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1151 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1152 
1153 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1154 		return -EINVAL;
1155 
1156 	if (vma->vm_pgoff == 0) {
1157 		/* We prevent double mmaping on same context */
1158 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1159 			return -EINVAL;
1160 
1161 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1162 
1163 		if (io_remap_pfn_range(vma, vma->vm_start,
1164 				       to_mucontext(context)->uar.pfn,
1165 				       PAGE_SIZE, vma->vm_page_prot))
1166 			return -EAGAIN;
1167 
1168 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1169 
1170 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1171 		/* We prevent double mmaping on same context */
1172 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1173 			return -EINVAL;
1174 
1175 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1176 
1177 		if (io_remap_pfn_range(vma, vma->vm_start,
1178 				       to_mucontext(context)->uar.pfn +
1179 				       dev->dev->caps.num_uars,
1180 				       PAGE_SIZE, vma->vm_page_prot))
1181 			return -EAGAIN;
1182 
1183 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1184 
1185 	} else if (vma->vm_pgoff == 3) {
1186 		struct mlx4_clock_params params;
1187 		int ret;
1188 
1189 		/* We prevent double mmaping on same context */
1190 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1191 			return -EINVAL;
1192 
1193 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1194 
1195 		if (ret)
1196 			return ret;
1197 
1198 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1199 		if (io_remap_pfn_range(vma, vma->vm_start,
1200 				       (pci_resource_start(dev->dev->persist->pdev,
1201 							   params.bar) +
1202 					params.offset)
1203 				       >> PAGE_SHIFT,
1204 				       PAGE_SIZE, vma->vm_page_prot))
1205 			return -EAGAIN;
1206 
1207 		mlx4_ib_set_vma_data(vma,
1208 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1209 	} else {
1210 		return -EINVAL;
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1217 				      struct ib_ucontext *context,
1218 				      struct ib_udata *udata)
1219 {
1220 	struct mlx4_ib_pd *pd;
1221 	int err;
1222 
1223 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1224 	if (!pd)
1225 		return ERR_PTR(-ENOMEM);
1226 
1227 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1228 	if (err) {
1229 		kfree(pd);
1230 		return ERR_PTR(err);
1231 	}
1232 
1233 	if (context)
1234 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1235 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1236 			kfree(pd);
1237 			return ERR_PTR(-EFAULT);
1238 		}
1239 
1240 	return &pd->ibpd;
1241 }
1242 
1243 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1244 {
1245 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1246 	kfree(pd);
1247 
1248 	return 0;
1249 }
1250 
1251 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1252 					  struct ib_ucontext *context,
1253 					  struct ib_udata *udata)
1254 {
1255 	struct mlx4_ib_xrcd *xrcd;
1256 	struct ib_cq_init_attr cq_attr = {};
1257 	int err;
1258 
1259 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1260 		return ERR_PTR(-ENOSYS);
1261 
1262 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1263 	if (!xrcd)
1264 		return ERR_PTR(-ENOMEM);
1265 
1266 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1267 	if (err)
1268 		goto err1;
1269 
1270 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1271 	if (IS_ERR(xrcd->pd)) {
1272 		err = PTR_ERR(xrcd->pd);
1273 		goto err2;
1274 	}
1275 
1276 	cq_attr.cqe = 1;
1277 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1278 	if (IS_ERR(xrcd->cq)) {
1279 		err = PTR_ERR(xrcd->cq);
1280 		goto err3;
1281 	}
1282 
1283 	return &xrcd->ibxrcd;
1284 
1285 err3:
1286 	ib_dealloc_pd(xrcd->pd);
1287 err2:
1288 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1289 err1:
1290 	kfree(xrcd);
1291 	return ERR_PTR(err);
1292 }
1293 
1294 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1295 {
1296 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1297 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1298 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1299 	kfree(xrcd);
1300 
1301 	return 0;
1302 }
1303 
1304 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1305 {
1306 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1307 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1308 	struct mlx4_ib_gid_entry *ge;
1309 
1310 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1311 	if (!ge)
1312 		return -ENOMEM;
1313 
1314 	ge->gid = *gid;
1315 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1316 		ge->port = mqp->port;
1317 		ge->added = 1;
1318 	}
1319 
1320 	mutex_lock(&mqp->mutex);
1321 	list_add_tail(&ge->list, &mqp->gid_list);
1322 	mutex_unlock(&mqp->mutex);
1323 
1324 	return 0;
1325 }
1326 
1327 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1328 					  struct mlx4_ib_counters *ctr_table)
1329 {
1330 	struct counter_index *counter, *tmp_count;
1331 
1332 	mutex_lock(&ctr_table->mutex);
1333 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1334 				 list) {
1335 		if (counter->allocated)
1336 			mlx4_counter_free(ibdev->dev, counter->index);
1337 		list_del(&counter->list);
1338 		kfree(counter);
1339 	}
1340 	mutex_unlock(&ctr_table->mutex);
1341 }
1342 
1343 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1344 		   union ib_gid *gid)
1345 {
1346 	struct net_device *ndev;
1347 	int ret = 0;
1348 
1349 	if (!mqp->port)
1350 		return 0;
1351 
1352 	spin_lock_bh(&mdev->iboe.lock);
1353 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1354 	if (ndev)
1355 		dev_hold(ndev);
1356 	spin_unlock_bh(&mdev->iboe.lock);
1357 
1358 	if (ndev) {
1359 		ret = 1;
1360 		dev_put(ndev);
1361 	}
1362 
1363 	return ret;
1364 }
1365 
1366 struct mlx4_ib_steering {
1367 	struct list_head list;
1368 	struct mlx4_flow_reg_id reg_id;
1369 	union ib_gid gid;
1370 };
1371 
1372 #define LAST_ETH_FIELD vlan_tag
1373 #define LAST_IB_FIELD sl
1374 #define LAST_IPV4_FIELD dst_ip
1375 #define LAST_TCP_UDP_FIELD src_port
1376 
1377 /* Field is the last supported field */
1378 #define FIELDS_NOT_SUPPORTED(filter, field)\
1379 	memchr_inv((void *)&filter.field  +\
1380 		   sizeof(filter.field), 0,\
1381 		   sizeof(filter) -\
1382 		   offsetof(typeof(filter), field) -\
1383 		   sizeof(filter.field))
1384 
1385 static int parse_flow_attr(struct mlx4_dev *dev,
1386 			   u32 qp_num,
1387 			   union ib_flow_spec *ib_spec,
1388 			   struct _rule_hw *mlx4_spec)
1389 {
1390 	enum mlx4_net_trans_rule_id type;
1391 
1392 	switch (ib_spec->type) {
1393 	case IB_FLOW_SPEC_ETH:
1394 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1395 			return -ENOTSUPP;
1396 
1397 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1398 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1399 		       ETH_ALEN);
1400 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1401 		       ETH_ALEN);
1402 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1403 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1404 		break;
1405 	case IB_FLOW_SPEC_IB:
1406 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1407 			return -ENOTSUPP;
1408 
1409 		type = MLX4_NET_TRANS_RULE_ID_IB;
1410 		mlx4_spec->ib.l3_qpn =
1411 			cpu_to_be32(qp_num);
1412 		mlx4_spec->ib.qpn_mask =
1413 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1414 		break;
1415 
1416 
1417 	case IB_FLOW_SPEC_IPV4:
1418 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1419 			return -ENOTSUPP;
1420 
1421 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1422 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1423 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1424 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1425 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1426 		break;
1427 
1428 	case IB_FLOW_SPEC_TCP:
1429 	case IB_FLOW_SPEC_UDP:
1430 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1431 			return -ENOTSUPP;
1432 
1433 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1434 					MLX4_NET_TRANS_RULE_ID_TCP :
1435 					MLX4_NET_TRANS_RULE_ID_UDP;
1436 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1437 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1438 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1439 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1440 		break;
1441 
1442 	default:
1443 		return -EINVAL;
1444 	}
1445 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1446 	    mlx4_hw_rule_sz(dev, type) < 0)
1447 		return -EINVAL;
1448 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1449 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1450 	return mlx4_hw_rule_sz(dev, type);
1451 }
1452 
1453 struct default_rules {
1454 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1455 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1456 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1457 	__u8  link_layer;
1458 };
1459 static const struct default_rules default_table[] = {
1460 	{
1461 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1462 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1463 		.rules_create_list = {IB_FLOW_SPEC_IB},
1464 		.link_layer = IB_LINK_LAYER_INFINIBAND
1465 	}
1466 };
1467 
1468 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1469 					 struct ib_flow_attr *flow_attr)
1470 {
1471 	int i, j, k;
1472 	void *ib_flow;
1473 	const struct default_rules *pdefault_rules = default_table;
1474 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1475 
1476 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1477 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1478 		memset(&field_types, 0, sizeof(field_types));
1479 
1480 		if (link_layer != pdefault_rules->link_layer)
1481 			continue;
1482 
1483 		ib_flow = flow_attr + 1;
1484 		/* we assume the specs are sorted */
1485 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1486 		     j < flow_attr->num_of_specs; k++) {
1487 			union ib_flow_spec *current_flow =
1488 				(union ib_flow_spec *)ib_flow;
1489 
1490 			/* same layer but different type */
1491 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1492 			     (pdefault_rules->mandatory_fields[k] &
1493 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1494 			    (current_flow->type !=
1495 			     pdefault_rules->mandatory_fields[k]))
1496 				goto out;
1497 
1498 			/* same layer, try match next one */
1499 			if (current_flow->type ==
1500 			    pdefault_rules->mandatory_fields[k]) {
1501 				j++;
1502 				ib_flow +=
1503 					((union ib_flow_spec *)ib_flow)->size;
1504 			}
1505 		}
1506 
1507 		ib_flow = flow_attr + 1;
1508 		for (j = 0; j < flow_attr->num_of_specs;
1509 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1510 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1511 				/* same layer and same type */
1512 				if (((union ib_flow_spec *)ib_flow)->type ==
1513 				    pdefault_rules->mandatory_not_fields[k])
1514 					goto out;
1515 
1516 		return i;
1517 	}
1518 out:
1519 	return -1;
1520 }
1521 
1522 static int __mlx4_ib_create_default_rules(
1523 		struct mlx4_ib_dev *mdev,
1524 		struct ib_qp *qp,
1525 		const struct default_rules *pdefault_rules,
1526 		struct _rule_hw *mlx4_spec) {
1527 	int size = 0;
1528 	int i;
1529 
1530 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1531 		int ret;
1532 		union ib_flow_spec ib_spec;
1533 		switch (pdefault_rules->rules_create_list[i]) {
1534 		case 0:
1535 			/* no rule */
1536 			continue;
1537 		case IB_FLOW_SPEC_IB:
1538 			ib_spec.type = IB_FLOW_SPEC_IB;
1539 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1540 
1541 			break;
1542 		default:
1543 			/* invalid rule */
1544 			return -EINVAL;
1545 		}
1546 		/* We must put empty rule, qpn is being ignored */
1547 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1548 				      mlx4_spec);
1549 		if (ret < 0) {
1550 			pr_info("invalid parsing\n");
1551 			return -EINVAL;
1552 		}
1553 
1554 		mlx4_spec = (void *)mlx4_spec + ret;
1555 		size += ret;
1556 	}
1557 	return size;
1558 }
1559 
1560 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1561 			  int domain,
1562 			  enum mlx4_net_trans_promisc_mode flow_type,
1563 			  u64 *reg_id)
1564 {
1565 	int ret, i;
1566 	int size = 0;
1567 	void *ib_flow;
1568 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1569 	struct mlx4_cmd_mailbox *mailbox;
1570 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1571 	int default_flow;
1572 
1573 	static const u16 __mlx4_domain[] = {
1574 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1575 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1576 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1577 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1578 	};
1579 
1580 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1581 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1582 		return -EINVAL;
1583 	}
1584 
1585 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1586 		pr_err("Invalid domain value %d\n", domain);
1587 		return -EINVAL;
1588 	}
1589 
1590 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1591 		return -EINVAL;
1592 
1593 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1594 	if (IS_ERR(mailbox))
1595 		return PTR_ERR(mailbox);
1596 	ctrl = mailbox->buf;
1597 
1598 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1599 				 flow_attr->priority);
1600 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1601 	ctrl->port = flow_attr->port;
1602 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1603 
1604 	ib_flow = flow_attr + 1;
1605 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1606 	/* Add default flows */
1607 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1608 	if (default_flow >= 0) {
1609 		ret = __mlx4_ib_create_default_rules(
1610 				mdev, qp, default_table + default_flow,
1611 				mailbox->buf + size);
1612 		if (ret < 0) {
1613 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1614 			return -EINVAL;
1615 		}
1616 		size += ret;
1617 	}
1618 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1619 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1620 				      mailbox->buf + size);
1621 		if (ret < 0) {
1622 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1623 			return -EINVAL;
1624 		}
1625 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1626 		size += ret;
1627 	}
1628 
1629 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1630 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1631 			   MLX4_CMD_WRAPPED);
1632 	if (ret == -ENOMEM)
1633 		pr_err("mcg table is full. Fail to register network rule.\n");
1634 	else if (ret == -ENXIO)
1635 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1636 	else if (ret)
1637 		pr_err("Invalid argument. Fail to register network rule.\n");
1638 
1639 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1640 	return ret;
1641 }
1642 
1643 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1644 {
1645 	int err;
1646 	err = mlx4_cmd(dev, reg_id, 0, 0,
1647 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1648 		       MLX4_CMD_WRAPPED);
1649 	if (err)
1650 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1651 		       (long long)reg_id);
1652 	return err;
1653 }
1654 
1655 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1656 				    u64 *reg_id)
1657 {
1658 	void *ib_flow;
1659 	union ib_flow_spec *ib_spec;
1660 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1661 	int err = 0;
1662 
1663 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1664 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1665 		return 0; /* do nothing */
1666 
1667 	ib_flow = flow_attr + 1;
1668 	ib_spec = (union ib_flow_spec *)ib_flow;
1669 
1670 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1671 		return 0; /* do nothing */
1672 
1673 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1674 				    flow_attr->port, qp->qp_num,
1675 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1676 				    reg_id);
1677 	return err;
1678 }
1679 
1680 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1681 				      struct ib_flow_attr *flow_attr,
1682 				      enum mlx4_net_trans_promisc_mode *type)
1683 {
1684 	int err = 0;
1685 
1686 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1687 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1688 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1689 		return -EOPNOTSUPP;
1690 	}
1691 
1692 	if (flow_attr->num_of_specs == 0) {
1693 		type[0] = MLX4_FS_MC_SNIFFER;
1694 		type[1] = MLX4_FS_UC_SNIFFER;
1695 	} else {
1696 		union ib_flow_spec *ib_spec;
1697 
1698 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1699 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1700 			return -EINVAL;
1701 
1702 		/* if all is zero than MC and UC */
1703 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1704 			type[0] = MLX4_FS_MC_SNIFFER;
1705 			type[1] = MLX4_FS_UC_SNIFFER;
1706 		} else {
1707 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1708 					    ib_spec->eth.mask.dst_mac[1],
1709 					    ib_spec->eth.mask.dst_mac[2],
1710 					    ib_spec->eth.mask.dst_mac[3],
1711 					    ib_spec->eth.mask.dst_mac[4],
1712 					    ib_spec->eth.mask.dst_mac[5]};
1713 
1714 			/* Above xor was only on MC bit, non empty mask is valid
1715 			 * only if this bit is set and rest are zero.
1716 			 */
1717 			if (!is_zero_ether_addr(&mac[0]))
1718 				return -EINVAL;
1719 
1720 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1721 				type[0] = MLX4_FS_MC_SNIFFER;
1722 			else
1723 				type[0] = MLX4_FS_UC_SNIFFER;
1724 		}
1725 	}
1726 
1727 	return err;
1728 }
1729 
1730 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1731 				    struct ib_flow_attr *flow_attr,
1732 				    int domain)
1733 {
1734 	int err = 0, i = 0, j = 0;
1735 	struct mlx4_ib_flow *mflow;
1736 	enum mlx4_net_trans_promisc_mode type[2];
1737 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1738 	int is_bonded = mlx4_is_bonded(dev);
1739 
1740 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1741 		return ERR_PTR(-EINVAL);
1742 
1743 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1744 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1745 		return ERR_PTR(-EOPNOTSUPP);
1746 
1747 	memset(type, 0, sizeof(type));
1748 
1749 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1750 	if (!mflow) {
1751 		err = -ENOMEM;
1752 		goto err_free;
1753 	}
1754 
1755 	switch (flow_attr->type) {
1756 	case IB_FLOW_ATTR_NORMAL:
1757 		/* If dont trap flag (continue match) is set, under specific
1758 		 * condition traffic be replicated to given qp,
1759 		 * without stealing it
1760 		 */
1761 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1762 			err = mlx4_ib_add_dont_trap_rule(dev,
1763 							 flow_attr,
1764 							 type);
1765 			if (err)
1766 				goto err_free;
1767 		} else {
1768 			type[0] = MLX4_FS_REGULAR;
1769 		}
1770 		break;
1771 
1772 	case IB_FLOW_ATTR_ALL_DEFAULT:
1773 		type[0] = MLX4_FS_ALL_DEFAULT;
1774 		break;
1775 
1776 	case IB_FLOW_ATTR_MC_DEFAULT:
1777 		type[0] = MLX4_FS_MC_DEFAULT;
1778 		break;
1779 
1780 	case IB_FLOW_ATTR_SNIFFER:
1781 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1782 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1783 		break;
1784 
1785 	default:
1786 		err = -EINVAL;
1787 		goto err_free;
1788 	}
1789 
1790 	while (i < ARRAY_SIZE(type) && type[i]) {
1791 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1792 					    &mflow->reg_id[i].id);
1793 		if (err)
1794 			goto err_create_flow;
1795 		if (is_bonded) {
1796 			/* Application always sees one port so the mirror rule
1797 			 * must be on port #2
1798 			 */
1799 			flow_attr->port = 2;
1800 			err = __mlx4_ib_create_flow(qp, flow_attr,
1801 						    domain, type[j],
1802 						    &mflow->reg_id[j].mirror);
1803 			flow_attr->port = 1;
1804 			if (err)
1805 				goto err_create_flow;
1806 			j++;
1807 		}
1808 
1809 		i++;
1810 	}
1811 
1812 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1813 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1814 					       &mflow->reg_id[i].id);
1815 		if (err)
1816 			goto err_create_flow;
1817 
1818 		if (is_bonded) {
1819 			flow_attr->port = 2;
1820 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1821 						       &mflow->reg_id[j].mirror);
1822 			flow_attr->port = 1;
1823 			if (err)
1824 				goto err_create_flow;
1825 			j++;
1826 		}
1827 		/* function to create mirror rule */
1828 		i++;
1829 	}
1830 
1831 	return &mflow->ibflow;
1832 
1833 err_create_flow:
1834 	while (i) {
1835 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1836 					     mflow->reg_id[i].id);
1837 		i--;
1838 	}
1839 
1840 	while (j) {
1841 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1842 					     mflow->reg_id[j].mirror);
1843 		j--;
1844 	}
1845 err_free:
1846 	kfree(mflow);
1847 	return ERR_PTR(err);
1848 }
1849 
1850 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1851 {
1852 	int err, ret = 0;
1853 	int i = 0;
1854 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1855 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1856 
1857 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1858 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1859 		if (err)
1860 			ret = err;
1861 		if (mflow->reg_id[i].mirror) {
1862 			err = __mlx4_ib_destroy_flow(mdev->dev,
1863 						     mflow->reg_id[i].mirror);
1864 			if (err)
1865 				ret = err;
1866 		}
1867 		i++;
1868 	}
1869 
1870 	kfree(mflow);
1871 	return ret;
1872 }
1873 
1874 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1875 {
1876 	int err;
1877 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1878 	struct mlx4_dev	*dev = mdev->dev;
1879 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1880 	struct mlx4_ib_steering *ib_steering = NULL;
1881 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1882 	struct mlx4_flow_reg_id	reg_id;
1883 
1884 	if (mdev->dev->caps.steering_mode ==
1885 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1886 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1887 		if (!ib_steering)
1888 			return -ENOMEM;
1889 	}
1890 
1891 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1892 				    !!(mqp->flags &
1893 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1894 				    prot, &reg_id.id);
1895 	if (err) {
1896 		pr_err("multicast attach op failed, err %d\n", err);
1897 		goto err_malloc;
1898 	}
1899 
1900 	reg_id.mirror = 0;
1901 	if (mlx4_is_bonded(dev)) {
1902 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1903 					    (mqp->port == 1) ? 2 : 1,
1904 					    !!(mqp->flags &
1905 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1906 					    prot, &reg_id.mirror);
1907 		if (err)
1908 			goto err_add;
1909 	}
1910 
1911 	err = add_gid_entry(ibqp, gid);
1912 	if (err)
1913 		goto err_add;
1914 
1915 	if (ib_steering) {
1916 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1917 		ib_steering->reg_id = reg_id;
1918 		mutex_lock(&mqp->mutex);
1919 		list_add(&ib_steering->list, &mqp->steering_rules);
1920 		mutex_unlock(&mqp->mutex);
1921 	}
1922 	return 0;
1923 
1924 err_add:
1925 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1926 			      prot, reg_id.id);
1927 	if (reg_id.mirror)
1928 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1929 				      prot, reg_id.mirror);
1930 err_malloc:
1931 	kfree(ib_steering);
1932 
1933 	return err;
1934 }
1935 
1936 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1937 {
1938 	struct mlx4_ib_gid_entry *ge;
1939 	struct mlx4_ib_gid_entry *tmp;
1940 	struct mlx4_ib_gid_entry *ret = NULL;
1941 
1942 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1943 		if (!memcmp(raw, ge->gid.raw, 16)) {
1944 			ret = ge;
1945 			break;
1946 		}
1947 	}
1948 
1949 	return ret;
1950 }
1951 
1952 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1953 {
1954 	int err;
1955 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1956 	struct mlx4_dev *dev = mdev->dev;
1957 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1958 	struct net_device *ndev;
1959 	struct mlx4_ib_gid_entry *ge;
1960 	struct mlx4_flow_reg_id reg_id = {0, 0};
1961 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1962 
1963 	if (mdev->dev->caps.steering_mode ==
1964 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1965 		struct mlx4_ib_steering *ib_steering;
1966 
1967 		mutex_lock(&mqp->mutex);
1968 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1969 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1970 				list_del(&ib_steering->list);
1971 				break;
1972 			}
1973 		}
1974 		mutex_unlock(&mqp->mutex);
1975 		if (&ib_steering->list == &mqp->steering_rules) {
1976 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1977 			return -EINVAL;
1978 		}
1979 		reg_id = ib_steering->reg_id;
1980 		kfree(ib_steering);
1981 	}
1982 
1983 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1984 				    prot, reg_id.id);
1985 	if (err)
1986 		return err;
1987 
1988 	if (mlx4_is_bonded(dev)) {
1989 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1990 					    prot, reg_id.mirror);
1991 		if (err)
1992 			return err;
1993 	}
1994 
1995 	mutex_lock(&mqp->mutex);
1996 	ge = find_gid_entry(mqp, gid->raw);
1997 	if (ge) {
1998 		spin_lock_bh(&mdev->iboe.lock);
1999 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2000 		if (ndev)
2001 			dev_hold(ndev);
2002 		spin_unlock_bh(&mdev->iboe.lock);
2003 		if (ndev)
2004 			dev_put(ndev);
2005 		list_del(&ge->list);
2006 		kfree(ge);
2007 	} else
2008 		pr_warn("could not find mgid entry\n");
2009 
2010 	mutex_unlock(&mqp->mutex);
2011 
2012 	return 0;
2013 }
2014 
2015 static int init_node_data(struct mlx4_ib_dev *dev)
2016 {
2017 	struct ib_smp *in_mad  = NULL;
2018 	struct ib_smp *out_mad = NULL;
2019 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2020 	int err = -ENOMEM;
2021 
2022 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2023 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2024 	if (!in_mad || !out_mad)
2025 		goto out;
2026 
2027 	init_query_mad(in_mad);
2028 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2029 	if (mlx4_is_master(dev->dev))
2030 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2031 
2032 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2033 	if (err)
2034 		goto out;
2035 
2036 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2037 
2038 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2039 
2040 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2041 	if (err)
2042 		goto out;
2043 
2044 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2045 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2046 
2047 out:
2048 	kfree(in_mad);
2049 	kfree(out_mad);
2050 	return err;
2051 }
2052 
2053 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2054 			char *buf)
2055 {
2056 	struct mlx4_ib_dev *dev =
2057 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2058 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2059 }
2060 
2061 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2062 			char *buf)
2063 {
2064 	struct mlx4_ib_dev *dev =
2065 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2066 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2067 }
2068 
2069 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2070 			  char *buf)
2071 {
2072 	struct mlx4_ib_dev *dev =
2073 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2074 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2075 		       dev->dev->board_id);
2076 }
2077 
2078 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2079 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2080 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2081 
2082 static struct device_attribute *mlx4_class_attributes[] = {
2083 	&dev_attr_hw_rev,
2084 	&dev_attr_hca_type,
2085 	&dev_attr_board_id
2086 };
2087 
2088 struct diag_counter {
2089 	const char *name;
2090 	u32 offset;
2091 };
2092 
2093 #define DIAG_COUNTER(_name, _offset)			\
2094 	{ .name = #_name, .offset = _offset }
2095 
2096 static const struct diag_counter diag_basic[] = {
2097 	DIAG_COUNTER(rq_num_lle, 0x00),
2098 	DIAG_COUNTER(sq_num_lle, 0x04),
2099 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2100 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2101 	DIAG_COUNTER(rq_num_lpe, 0x18),
2102 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2103 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2104 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2105 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2106 	DIAG_COUNTER(sq_num_bre, 0x34),
2107 	DIAG_COUNTER(sq_num_rire, 0x44),
2108 	DIAG_COUNTER(rq_num_rire, 0x48),
2109 	DIAG_COUNTER(sq_num_rae, 0x4C),
2110 	DIAG_COUNTER(rq_num_rae, 0x50),
2111 	DIAG_COUNTER(sq_num_roe, 0x54),
2112 	DIAG_COUNTER(sq_num_tree, 0x5C),
2113 	DIAG_COUNTER(sq_num_rree, 0x64),
2114 	DIAG_COUNTER(rq_num_rnr, 0x68),
2115 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2116 	DIAG_COUNTER(rq_num_oos, 0x100),
2117 	DIAG_COUNTER(sq_num_oos, 0x104),
2118 };
2119 
2120 static const struct diag_counter diag_ext[] = {
2121 	DIAG_COUNTER(rq_num_dup, 0x130),
2122 	DIAG_COUNTER(sq_num_to, 0x134),
2123 };
2124 
2125 static const struct diag_counter diag_device_only[] = {
2126 	DIAG_COUNTER(num_cqovf, 0x1A0),
2127 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2128 };
2129 
2130 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2131 						    u8 port_num)
2132 {
2133 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2134 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2135 
2136 	if (!diag[!!port_num].name)
2137 		return NULL;
2138 
2139 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2140 					  diag[!!port_num].num_counters,
2141 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2142 }
2143 
2144 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2145 				struct rdma_hw_stats *stats,
2146 				u8 port, int index)
2147 {
2148 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2149 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2150 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2151 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2152 	int ret;
2153 	int i;
2154 
2155 	ret = mlx4_query_diag_counters(dev->dev,
2156 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2157 				       diag[!!port].offset, hw_value,
2158 				       diag[!!port].num_counters, port);
2159 
2160 	if (ret)
2161 		return ret;
2162 
2163 	for (i = 0; i < diag[!!port].num_counters; i++)
2164 		stats->value[i] = hw_value[i];
2165 
2166 	return diag[!!port].num_counters;
2167 }
2168 
2169 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2170 					 const char ***name,
2171 					 u32 **offset,
2172 					 u32 *num,
2173 					 bool port)
2174 {
2175 	u32 num_counters;
2176 
2177 	num_counters = ARRAY_SIZE(diag_basic);
2178 
2179 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2180 		num_counters += ARRAY_SIZE(diag_ext);
2181 
2182 	if (!port)
2183 		num_counters += ARRAY_SIZE(diag_device_only);
2184 
2185 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2186 	if (!*name)
2187 		return -ENOMEM;
2188 
2189 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2190 	if (!*offset)
2191 		goto err_name;
2192 
2193 	*num = num_counters;
2194 
2195 	return 0;
2196 
2197 err_name:
2198 	kfree(*name);
2199 	return -ENOMEM;
2200 }
2201 
2202 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2203 				       const char **name,
2204 				       u32 *offset,
2205 				       bool port)
2206 {
2207 	int i;
2208 	int j;
2209 
2210 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2211 		name[i] = diag_basic[i].name;
2212 		offset[i] = diag_basic[i].offset;
2213 	}
2214 
2215 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2216 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2217 			name[j] = diag_ext[i].name;
2218 			offset[j] = diag_ext[i].offset;
2219 		}
2220 	}
2221 
2222 	if (!port) {
2223 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2224 			name[j] = diag_device_only[i].name;
2225 			offset[j] = diag_device_only[i].offset;
2226 		}
2227 	}
2228 }
2229 
2230 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2231 {
2232 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2233 	int i;
2234 	int ret;
2235 	bool per_port = !!(ibdev->dev->caps.flags2 &
2236 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2237 
2238 	if (mlx4_is_slave(ibdev->dev))
2239 		return 0;
2240 
2241 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2242 		/* i == 1 means we are building port counters */
2243 		if (i && !per_port)
2244 			continue;
2245 
2246 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2247 						    &diag[i].offset,
2248 						    &diag[i].num_counters, i);
2249 		if (ret)
2250 			goto err_alloc;
2251 
2252 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2253 					   diag[i].offset, i);
2254 	}
2255 
2256 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2257 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2258 
2259 	return 0;
2260 
2261 err_alloc:
2262 	if (i) {
2263 		kfree(diag[i - 1].name);
2264 		kfree(diag[i - 1].offset);
2265 	}
2266 
2267 	return ret;
2268 }
2269 
2270 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2271 {
2272 	int i;
2273 
2274 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2275 		kfree(ibdev->diag_counters[i].offset);
2276 		kfree(ibdev->diag_counters[i].name);
2277 	}
2278 }
2279 
2280 #define MLX4_IB_INVALID_MAC	((u64)-1)
2281 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2282 			       struct net_device *dev,
2283 			       int port)
2284 {
2285 	u64 new_smac = 0;
2286 	u64 release_mac = MLX4_IB_INVALID_MAC;
2287 	struct mlx4_ib_qp *qp;
2288 
2289 	new_smac = mlx4_mac_to_u64(IF_LLADDR(dev));
2290 
2291 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2292 
2293 	/* no need for update QP1 and mac registration in non-SRIOV */
2294 	if (!mlx4_is_mfunc(ibdev->dev))
2295 		return;
2296 
2297 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2298 	qp = ibdev->qp1_proxy[port - 1];
2299 	if (qp) {
2300 		int new_smac_index;
2301 		u64 old_smac;
2302 		struct mlx4_update_qp_params update_params;
2303 
2304 		mutex_lock(&qp->mutex);
2305 		old_smac = qp->pri.smac;
2306 		if (new_smac == old_smac)
2307 			goto unlock;
2308 
2309 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2310 
2311 		if (new_smac_index < 0)
2312 			goto unlock;
2313 
2314 		update_params.smac_index = new_smac_index;
2315 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2316 				   &update_params)) {
2317 			release_mac = new_smac;
2318 			goto unlock;
2319 		}
2320 		/* if old port was zero, no mac was yet registered for this QP */
2321 		if (qp->pri.smac_port)
2322 			release_mac = old_smac;
2323 		qp->pri.smac = new_smac;
2324 		qp->pri.smac_port = port;
2325 		qp->pri.smac_index = new_smac_index;
2326 	}
2327 
2328 unlock:
2329 	if (release_mac != MLX4_IB_INVALID_MAC)
2330 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2331 	if (qp)
2332 		mutex_unlock(&qp->mutex);
2333 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2334 }
2335 
2336 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2337 				 struct net_device *dev,
2338 				 unsigned long event)
2339 
2340 {
2341 	struct mlx4_ib_iboe *iboe;
2342 	int update_qps_port = -1;
2343 	int port;
2344 
2345 	iboe = &ibdev->iboe;
2346 
2347 	spin_lock_bh(&iboe->lock);
2348 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2349 
2350 		iboe->netdevs[port - 1] =
2351 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2352 
2353 		if (dev == iboe->netdevs[port - 1] &&
2354 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2355 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2356 			update_qps_port = port;
2357 
2358 	}
2359 	spin_unlock_bh(&iboe->lock);
2360 
2361 	if (update_qps_port > 0)
2362 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2363 }
2364 
2365 static int mlx4_ib_netdev_event(struct notifier_block *this,
2366 				unsigned long event, void *ptr)
2367 {
2368 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2369 	struct mlx4_ib_dev *ibdev;
2370 
2371 	if (!net_eq(dev_net(dev), &init_net))
2372 		return NOTIFY_DONE;
2373 
2374 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2375 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2376 
2377 	return NOTIFY_DONE;
2378 }
2379 
2380 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2381 {
2382 	int port;
2383 	int slave;
2384 	int i;
2385 
2386 	if (mlx4_is_master(ibdev->dev)) {
2387 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2388 		     ++slave) {
2389 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2390 				for (i = 0;
2391 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2392 				     ++i) {
2393 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2394 					/* master has the identity virt2phys pkey mapping */
2395 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2396 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2397 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2398 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2399 				}
2400 			}
2401 		}
2402 		/* initialize pkey cache */
2403 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2404 			for (i = 0;
2405 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2406 			     ++i)
2407 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2408 					(i) ? 0 : 0xFFFF;
2409 		}
2410 	}
2411 }
2412 
2413 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2414 {
2415 	int i, j, eq = 0, total_eqs = 0;
2416 
2417 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2418 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2419 	if (!ibdev->eq_table)
2420 		return;
2421 
2422 	for (i = 1; i <= dev->caps.num_ports; i++) {
2423 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2424 		     j++, total_eqs++) {
2425 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2426 				continue;
2427 			ibdev->eq_table[eq] = total_eqs;
2428 			if (!mlx4_assign_eq(dev, i,
2429 					    &ibdev->eq_table[eq]))
2430 				eq++;
2431 			else
2432 				ibdev->eq_table[eq] = -1;
2433 		}
2434 	}
2435 
2436 	for (i = eq; i < dev->caps.num_comp_vectors;
2437 	     ibdev->eq_table[i++] = -1)
2438 		;
2439 
2440 	/* Advertise the new number of EQs to clients */
2441 	ibdev->ib_dev.num_comp_vectors = eq;
2442 }
2443 
2444 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2445 {
2446 	int i;
2447 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2448 
2449 	/* no eqs were allocated */
2450 	if (!ibdev->eq_table)
2451 		return;
2452 
2453 	/* Reset the advertised EQ number */
2454 	ibdev->ib_dev.num_comp_vectors = 0;
2455 
2456 	for (i = 0; i < total_eqs; i++)
2457 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2458 
2459 	kfree(ibdev->eq_table);
2460 	ibdev->eq_table = NULL;
2461 }
2462 
2463 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2464 			       struct ib_port_immutable *immutable)
2465 {
2466 	struct ib_port_attr attr;
2467 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2468 	int err;
2469 
2470 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2471 	if (err)
2472 		return err;
2473 
2474 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2475 	immutable->gid_tbl_len = attr.gid_tbl_len;
2476 
2477 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2478 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2479 	} 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 	}
2486 
2487 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2488 
2489 	return 0;
2490 }
2491 
2492 static void get_fw_ver_str(struct ib_device *device, char *str,
2493 			   size_t str_len)
2494 {
2495 	struct mlx4_ib_dev *dev =
2496 		container_of(device, struct mlx4_ib_dev, ib_dev);
2497 	snprintf(str, str_len, "%d.%d.%d",
2498 		 (int) (dev->dev->caps.fw_ver >> 32),
2499 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2500 		 (int) dev->dev->caps.fw_ver & 0xffff);
2501 }
2502 
2503 static void *mlx4_ib_add(struct mlx4_dev *dev)
2504 {
2505 	struct mlx4_ib_dev *ibdev;
2506 	int num_ports;
2507 	int i, j;
2508 	int err;
2509 	struct mlx4_ib_iboe *iboe;
2510 	int ib_num_ports = 0;
2511 	int num_req_counters;
2512 	int allocated;
2513 	u32 counter_index;
2514 	struct counter_index *new_counter_index = NULL;
2515 
2516 	pr_info_once("%s", mlx4_ib_version);
2517 
2518 	num_ports = 0;
2519 	mlx4_foreach_ib_transport_port(i, dev)
2520 		num_ports++;
2521 
2522 	/* No point in registering a device with no ports... */
2523 	if (num_ports == 0)
2524 		return NULL;
2525 
2526 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2527 	if (!ibdev) {
2528 		dev_err(&dev->persist->pdev->dev,
2529 			"Device struct alloc failed\n");
2530 		return NULL;
2531 	}
2532 
2533 	iboe = &ibdev->iboe;
2534 
2535 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2536 		goto err_dealloc;
2537 
2538 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2539 		goto err_pd;
2540 
2541 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2542 				 PAGE_SIZE);
2543 	if (!ibdev->uar_map)
2544 		goto err_uar;
2545 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2546 
2547 	ibdev->dev = dev;
2548 	ibdev->bond_next_port	= 0;
2549 
2550 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2551 	ibdev->ib_dev.owner		= THIS_MODULE;
2552 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2553 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2554 	ibdev->num_ports		= num_ports;
2555 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2556 						1 : ibdev->num_ports;
2557 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2558 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2559 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2560 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2561 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2562 
2563 	if (dev->caps.userspace_caps)
2564 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2565 	else
2566 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2567 
2568 	ibdev->ib_dev.uverbs_cmd_mask	=
2569 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2570 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2571 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2572 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2573 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2574 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2575 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2576 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2577 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2578 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2579 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2580 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2581 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2582 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2583 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2584 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2585 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2586 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2587 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2588 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2589 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2590 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2591 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2592 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2593 
2594 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2595 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2596 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2597 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2598 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2599 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2600 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2601 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2602 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2603 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2604 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2605 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2606 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2607 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2608 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2609 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2610 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2611 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2612 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2613 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2614 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2615 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2616 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2617 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2618 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2619 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2620 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2621 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2622 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2623 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2624 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2625 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2626 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2627 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2628 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2629 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2630 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2631 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2632 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2633 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2634 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2635 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2636 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2637 
2638 	if (!mlx4_is_slave(ibdev->dev)) {
2639 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2640 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2641 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2642 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2643 	}
2644 
2645 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2646 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2647 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2648 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2649 
2650 		ibdev->ib_dev.uverbs_cmd_mask |=
2651 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2652 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2653 	}
2654 
2655 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2656 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2657 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2658 		ibdev->ib_dev.uverbs_cmd_mask |=
2659 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2660 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2661 	}
2662 
2663 	if (check_flow_steering_support(dev)) {
2664 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2665 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2666 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2667 
2668 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2669 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2670 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2671 	}
2672 
2673 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2674 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2675 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2676 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2677 
2678 	mlx4_ib_alloc_eqs(dev, ibdev);
2679 
2680 	spin_lock_init(&iboe->lock);
2681 
2682 	if (init_node_data(ibdev))
2683 		goto err_map;
2684 	mlx4_init_sl2vl_tbl(ibdev);
2685 
2686 	for (i = 0; i < ibdev->num_ports; ++i) {
2687 		mutex_init(&ibdev->counters_table[i].mutex);
2688 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2689 	}
2690 
2691 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2692 	for (i = 0; i < num_req_counters; ++i) {
2693 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2694 		allocated = 0;
2695 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2696 						IB_LINK_LAYER_ETHERNET) {
2697 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2698 			/* if failed to allocate a new counter, use default */
2699 			if (err)
2700 				counter_index =
2701 					mlx4_get_default_counter_index(dev,
2702 								       i + 1);
2703 			else
2704 				allocated = 1;
2705 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2706 			counter_index = mlx4_get_default_counter_index(dev,
2707 								       i + 1);
2708 		}
2709 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2710 					    GFP_KERNEL);
2711 		if (!new_counter_index) {
2712 			if (allocated)
2713 				mlx4_counter_free(ibdev->dev, counter_index);
2714 			goto err_counter;
2715 		}
2716 		new_counter_index->index = counter_index;
2717 		new_counter_index->allocated = allocated;
2718 		list_add_tail(&new_counter_index->list,
2719 			      &ibdev->counters_table[i].counters_list);
2720 		ibdev->counters_table[i].default_counter = counter_index;
2721 		pr_info("counter index %d for port %d allocated %d\n",
2722 			counter_index, i + 1, allocated);
2723 	}
2724 	if (mlx4_is_bonded(dev))
2725 		for (i = 1; i < ibdev->num_ports ; ++i) {
2726 			new_counter_index =
2727 					kmalloc(sizeof(struct counter_index),
2728 						GFP_KERNEL);
2729 			if (!new_counter_index)
2730 				goto err_counter;
2731 			new_counter_index->index = counter_index;
2732 			new_counter_index->allocated = 0;
2733 			list_add_tail(&new_counter_index->list,
2734 				      &ibdev->counters_table[i].counters_list);
2735 			ibdev->counters_table[i].default_counter =
2736 								counter_index;
2737 		}
2738 
2739 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2740 		ib_num_ports++;
2741 
2742 	spin_lock_init(&ibdev->sm_lock);
2743 	mutex_init(&ibdev->cap_mask_mutex);
2744 	INIT_LIST_HEAD(&ibdev->qp_list);
2745 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2746 
2747 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2748 	    ib_num_ports) {
2749 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2750 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2751 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2752 					    &ibdev->steer_qpn_base, 0);
2753 		if (err)
2754 			goto err_counter;
2755 
2756 		ibdev->ib_uc_qpns_bitmap =
2757 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2758 				sizeof(long),
2759 				GFP_KERNEL);
2760 		if (!ibdev->ib_uc_qpns_bitmap) {
2761 			dev_err(&dev->persist->pdev->dev,
2762 				"bit map alloc failed\n");
2763 			goto err_steer_qp_release;
2764 		}
2765 
2766 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2767 
2768 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2769 				dev, ibdev->steer_qpn_base,
2770 				ibdev->steer_qpn_base +
2771 				ibdev->steer_qpn_count - 1);
2772 		if (err)
2773 			goto err_steer_free_bitmap;
2774 	}
2775 
2776 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2777 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2778 
2779 	if (mlx4_ib_alloc_diag_counters(ibdev))
2780 		goto err_steer_free_bitmap;
2781 
2782 	if (ib_register_device(&ibdev->ib_dev, NULL))
2783 		goto err_diag_counters;
2784 
2785 	if (mlx4_ib_mad_init(ibdev))
2786 		goto err_reg;
2787 
2788 	if (mlx4_ib_init_sriov(ibdev))
2789 		goto err_mad;
2790 
2791 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2792 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2793 		if (!iboe->nb.notifier_call) {
2794 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2795 			err = register_netdevice_notifier(&iboe->nb);
2796 			if (err) {
2797 				iboe->nb.notifier_call = NULL;
2798 				goto err_notif;
2799 			}
2800 		}
2801 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2802 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2803 			if (err) {
2804 				goto err_notif;
2805 			}
2806 		}
2807 	}
2808 
2809 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2810 		if (device_create_file(&ibdev->ib_dev.dev,
2811 				       mlx4_class_attributes[j]))
2812 			goto err_notif;
2813 	}
2814 
2815 	ibdev->ib_active = true;
2816 
2817 	if (mlx4_is_mfunc(ibdev->dev))
2818 		init_pkeys(ibdev);
2819 
2820 	/* create paravirt contexts for any VFs which are active */
2821 	if (mlx4_is_master(ibdev->dev)) {
2822 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2823 			if (j == mlx4_master_func_num(ibdev->dev))
2824 				continue;
2825 			if (mlx4_is_slave_active(ibdev->dev, j))
2826 				do_slave_init(ibdev, j, 1);
2827 		}
2828 	}
2829 	return ibdev;
2830 
2831 err_notif:
2832 	if (ibdev->iboe.nb.notifier_call) {
2833 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2834 			pr_warn("failure unregistering notifier\n");
2835 		ibdev->iboe.nb.notifier_call = NULL;
2836 	}
2837 	flush_workqueue(wq);
2838 
2839 	mlx4_ib_close_sriov(ibdev);
2840 
2841 err_mad:
2842 	mlx4_ib_mad_cleanup(ibdev);
2843 
2844 err_reg:
2845 	ib_unregister_device(&ibdev->ib_dev);
2846 
2847 err_diag_counters:
2848 	mlx4_ib_diag_cleanup(ibdev);
2849 
2850 err_steer_free_bitmap:
2851 	kfree(ibdev->ib_uc_qpns_bitmap);
2852 
2853 err_steer_qp_release:
2854 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2855 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2856 				      ibdev->steer_qpn_count);
2857 err_counter:
2858 	for (i = 0; i < ibdev->num_ports; ++i)
2859 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2860 
2861 err_map:
2862 	iounmap(ibdev->uar_map);
2863 
2864 err_uar:
2865 	mlx4_uar_free(dev, &ibdev->priv_uar);
2866 
2867 err_pd:
2868 	mlx4_pd_free(dev, ibdev->priv_pdn);
2869 
2870 err_dealloc:
2871 	ib_dealloc_device(&ibdev->ib_dev);
2872 
2873 	return NULL;
2874 }
2875 
2876 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2877 {
2878 	int offset;
2879 
2880 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2881 
2882 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2883 					 dev->steer_qpn_count,
2884 					 get_count_order(count));
2885 	if (offset < 0)
2886 		return offset;
2887 
2888 	*qpn = dev->steer_qpn_base + offset;
2889 	return 0;
2890 }
2891 
2892 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2893 {
2894 	if (!qpn ||
2895 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2896 		return;
2897 
2898 	BUG_ON(qpn < dev->steer_qpn_base);
2899 
2900 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2901 			      qpn - dev->steer_qpn_base,
2902 			      get_count_order(count));
2903 }
2904 
2905 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2906 			 int is_attach)
2907 {
2908 	int err;
2909 	size_t flow_size;
2910 	struct ib_flow_attr *flow = NULL;
2911 	struct ib_flow_spec_ib *ib_spec;
2912 
2913 	if (is_attach) {
2914 		flow_size = sizeof(struct ib_flow_attr) +
2915 			    sizeof(struct ib_flow_spec_ib);
2916 		flow = kzalloc(flow_size, GFP_KERNEL);
2917 		if (!flow)
2918 			return -ENOMEM;
2919 		flow->port = mqp->port;
2920 		flow->num_of_specs = 1;
2921 		flow->size = flow_size;
2922 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2923 		ib_spec->type = IB_FLOW_SPEC_IB;
2924 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2925 		/* Add an empty rule for IB L2 */
2926 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2927 
2928 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2929 					    IB_FLOW_DOMAIN_NIC,
2930 					    MLX4_FS_REGULAR,
2931 					    &mqp->reg_id);
2932 	} else {
2933 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2934 	}
2935 	kfree(flow);
2936 	return err;
2937 }
2938 
2939 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2940 {
2941 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2942 	int p;
2943 
2944 	ibdev->ib_active = false;
2945 	flush_workqueue(wq);
2946 
2947 	mlx4_ib_close_sriov(ibdev);
2948 	mlx4_ib_mad_cleanup(ibdev);
2949 	ib_unregister_device(&ibdev->ib_dev);
2950 	mlx4_ib_diag_cleanup(ibdev);
2951 	if (ibdev->iboe.nb.notifier_call) {
2952 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2953 			pr_warn("failure unregistering notifier\n");
2954 		ibdev->iboe.nb.notifier_call = NULL;
2955 	}
2956 
2957 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2958 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2959 				      ibdev->steer_qpn_count);
2960 		kfree(ibdev->ib_uc_qpns_bitmap);
2961 	}
2962 
2963 	iounmap(ibdev->uar_map);
2964 	for (p = 0; p < ibdev->num_ports; ++p)
2965 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2966 
2967 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2968 		mlx4_CLOSE_PORT(dev, p);
2969 
2970 	mlx4_ib_free_eqs(dev, ibdev);
2971 
2972 	mlx4_uar_free(dev, &ibdev->priv_uar);
2973 	mlx4_pd_free(dev, ibdev->priv_pdn);
2974 	ib_dealloc_device(&ibdev->ib_dev);
2975 }
2976 
2977 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2978 {
2979 	struct mlx4_ib_demux_work **dm = NULL;
2980 	struct mlx4_dev *dev = ibdev->dev;
2981 	int i;
2982 	unsigned long flags;
2983 	struct mlx4_active_ports actv_ports;
2984 	unsigned int ports;
2985 	unsigned int first_port;
2986 
2987 	if (!mlx4_is_master(dev))
2988 		return;
2989 
2990 	actv_ports = mlx4_get_active_ports(dev, slave);
2991 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2992 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2993 
2994 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2995 	if (!dm) {
2996 		pr_err("failed to allocate memory for tunneling qp update\n");
2997 		return;
2998 	}
2999 
3000 	for (i = 0; i < ports; i++) {
3001 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3002 		if (!dm[i]) {
3003 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
3004 			while (--i >= 0)
3005 				kfree(dm[i]);
3006 			goto out;
3007 		}
3008 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3009 		dm[i]->port = first_port + i + 1;
3010 		dm[i]->slave = slave;
3011 		dm[i]->do_init = do_init;
3012 		dm[i]->dev = ibdev;
3013 	}
3014 	/* initialize or tear down tunnel QPs for the slave */
3015 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3016 	if (!ibdev->sriov.is_going_down) {
3017 		for (i = 0; i < ports; i++)
3018 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3019 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3020 	} else {
3021 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3022 		for (i = 0; i < ports; i++)
3023 			kfree(dm[i]);
3024 	}
3025 out:
3026 	kfree(dm);
3027 	return;
3028 }
3029 
3030 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3031 {
3032 	struct mlx4_ib_qp *mqp;
3033 	unsigned long flags_qp;
3034 	unsigned long flags_cq;
3035 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3036 	struct list_head    cq_notify_list;
3037 	struct mlx4_cq *mcq;
3038 	unsigned long flags;
3039 
3040 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3041 	INIT_LIST_HEAD(&cq_notify_list);
3042 
3043 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3044 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3045 
3046 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3047 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3048 		if (mqp->sq.tail != mqp->sq.head) {
3049 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3050 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3051 			if (send_mcq->mcq.comp &&
3052 			    mqp->ibqp.send_cq->comp_handler) {
3053 				if (!send_mcq->mcq.reset_notify_added) {
3054 					send_mcq->mcq.reset_notify_added = 1;
3055 					list_add_tail(&send_mcq->mcq.reset_notify,
3056 						      &cq_notify_list);
3057 				}
3058 			}
3059 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3060 		}
3061 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3062 		/* Now, handle the QP's receive queue */
3063 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3064 		/* no handling is needed for SRQ */
3065 		if (!mqp->ibqp.srq) {
3066 			if (mqp->rq.tail != mqp->rq.head) {
3067 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3068 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3069 				if (recv_mcq->mcq.comp &&
3070 				    mqp->ibqp.recv_cq->comp_handler) {
3071 					if (!recv_mcq->mcq.reset_notify_added) {
3072 						recv_mcq->mcq.reset_notify_added = 1;
3073 						list_add_tail(&recv_mcq->mcq.reset_notify,
3074 							      &cq_notify_list);
3075 					}
3076 				}
3077 				spin_unlock_irqrestore(&recv_mcq->lock,
3078 						       flags_cq);
3079 			}
3080 		}
3081 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3082 	}
3083 
3084 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3085 		mcq->comp(mcq);
3086 	}
3087 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3088 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3089 }
3090 
3091 static void handle_bonded_port_state_event(struct work_struct *work)
3092 {
3093 	struct ib_event_work *ew =
3094 		container_of(work, struct ib_event_work, work);
3095 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3096 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3097 	int i;
3098 	struct ib_event ibev;
3099 
3100 	kfree(ew);
3101 	spin_lock_bh(&ibdev->iboe.lock);
3102 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3103 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3104 		enum ib_port_state curr_port_state;
3105 
3106 		if (!curr_netdev)
3107 			continue;
3108 
3109 		curr_port_state =
3110 			(netif_running(curr_netdev) &&
3111 			 netif_carrier_ok(curr_netdev)) ?
3112 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3113 
3114 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3115 			curr_port_state : IB_PORT_ACTIVE;
3116 	}
3117 	spin_unlock_bh(&ibdev->iboe.lock);
3118 
3119 	ibev.device = &ibdev->ib_dev;
3120 	ibev.element.port_num = 1;
3121 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3122 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3123 
3124 	ib_dispatch_event(&ibev);
3125 }
3126 
3127 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3128 {
3129 	u64 sl2vl;
3130 	int err;
3131 
3132 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3133 	if (err) {
3134 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3135 		       port, err);
3136 		sl2vl = 0;
3137 	}
3138 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3139 }
3140 
3141 static void ib_sl2vl_update_work(struct work_struct *work)
3142 {
3143 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3144 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3145 	int port = ew->port;
3146 
3147 	mlx4_ib_sl2vl_update(mdev, port);
3148 
3149 	kfree(ew);
3150 }
3151 
3152 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3153 				     int port)
3154 {
3155 	struct ib_event_work *ew;
3156 
3157 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3158 	if (ew) {
3159 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3160 		ew->port = port;
3161 		ew->ib_dev = ibdev;
3162 		queue_work(wq, &ew->work);
3163 	} else {
3164 		pr_err("failed to allocate memory for sl2vl update work\n");
3165 	}
3166 }
3167 
3168 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3169 			  enum mlx4_dev_event event, unsigned long param)
3170 {
3171 	struct ib_event ibev;
3172 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3173 	struct mlx4_eqe *eqe = NULL;
3174 	struct ib_event_work *ew;
3175 	int p = 0;
3176 
3177 	if (mlx4_is_bonded(dev) &&
3178 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3179 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3180 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3181 		if (!ew)
3182 			return;
3183 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3184 		ew->ib_dev = ibdev;
3185 		queue_work(wq, &ew->work);
3186 		return;
3187 	}
3188 
3189 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3190 		eqe = (struct mlx4_eqe *)param;
3191 	else
3192 		p = (int) param;
3193 
3194 	switch (event) {
3195 	case MLX4_DEV_EVENT_PORT_UP:
3196 		if (p > ibdev->num_ports)
3197 			return;
3198 		if (!mlx4_is_slave(dev) &&
3199 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3200 			IB_LINK_LAYER_INFINIBAND) {
3201 			if (mlx4_is_master(dev))
3202 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3203 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3204 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3205 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3206 		}
3207 		ibev.event = IB_EVENT_PORT_ACTIVE;
3208 		break;
3209 
3210 	case MLX4_DEV_EVENT_PORT_DOWN:
3211 		if (p > ibdev->num_ports)
3212 			return;
3213 		ibev.event = IB_EVENT_PORT_ERR;
3214 		break;
3215 
3216 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3217 		ibdev->ib_active = false;
3218 		ibev.event = IB_EVENT_DEVICE_FATAL;
3219 		mlx4_ib_handle_catas_error(ibdev);
3220 		break;
3221 
3222 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3223 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3224 		if (!ew) {
3225 			pr_err("failed to allocate memory for events work\n");
3226 			break;
3227 		}
3228 
3229 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3230 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3231 		ew->ib_dev = ibdev;
3232 		/* need to queue only for port owner, which uses GEN_EQE */
3233 		if (mlx4_is_master(dev))
3234 			queue_work(wq, &ew->work);
3235 		else
3236 			handle_port_mgmt_change_event(&ew->work);
3237 		return;
3238 
3239 	case MLX4_DEV_EVENT_SLAVE_INIT:
3240 		/* here, p is the slave id */
3241 		do_slave_init(ibdev, p, 1);
3242 		if (mlx4_is_master(dev)) {
3243 			int i;
3244 
3245 			for (i = 1; i <= ibdev->num_ports; i++) {
3246 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3247 					== IB_LINK_LAYER_INFINIBAND)
3248 					mlx4_ib_slave_alias_guid_event(ibdev,
3249 								       p, i,
3250 								       1);
3251 			}
3252 		}
3253 		return;
3254 
3255 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3256 		if (mlx4_is_master(dev)) {
3257 			int i;
3258 
3259 			for (i = 1; i <= ibdev->num_ports; i++) {
3260 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3261 					== IB_LINK_LAYER_INFINIBAND)
3262 					mlx4_ib_slave_alias_guid_event(ibdev,
3263 								       p, i,
3264 								       0);
3265 			}
3266 		}
3267 		/* here, p is the slave id */
3268 		do_slave_init(ibdev, p, 0);
3269 		return;
3270 
3271 	default:
3272 		return;
3273 	}
3274 
3275 	ibev.device	      = ibdev_ptr;
3276 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3277 
3278 	ib_dispatch_event(&ibev);
3279 }
3280 
3281 static struct mlx4_interface mlx4_ib_interface = {
3282 	.add		= mlx4_ib_add,
3283 	.remove		= mlx4_ib_remove,
3284 	.event		= mlx4_ib_event,
3285 	.protocol	= MLX4_PROT_IB_IPV6,
3286 	.flags		= MLX4_INTFF_BONDING
3287 };
3288 
3289 static int __init mlx4_ib_init(void)
3290 {
3291 	int err;
3292 
3293 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3294 	if (!wq)
3295 		return -ENOMEM;
3296 
3297 	err = mlx4_ib_mcg_init();
3298 	if (err)
3299 		goto clean_wq;
3300 
3301 	err = mlx4_register_interface(&mlx4_ib_interface);
3302 	if (err)
3303 		goto clean_mcg;
3304 
3305 	return 0;
3306 
3307 clean_mcg:
3308 	mlx4_ib_mcg_destroy();
3309 
3310 clean_wq:
3311 	destroy_workqueue(wq);
3312 	return err;
3313 }
3314 
3315 static void __exit mlx4_ib_cleanup(void)
3316 {
3317 	mlx4_unregister_interface(&mlx4_ib_interface);
3318 	mlx4_ib_mcg_destroy();
3319 	destroy_workqueue(wq);
3320 }
3321 
3322 module_init_order(mlx4_ib_init, SI_ORDER_THIRD);
3323 module_exit(mlx4_ib_cleanup);
3324 
3325 static int
3326 mlx4ib_evhand(module_t mod, int event, void *arg)
3327 {
3328 	return (0);
3329 }
3330 
3331 static moduledata_t mlx4ib_mod = {
3332 	.name = "mlx4ib",
3333 	.evhand = mlx4ib_evhand,
3334 };
3335 
3336 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_LAST, SI_ORDER_ANY);
3337 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
3338 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);
3339 MODULE_DEPEND(mlx4ib, linuxkpi, 1, 1, 1);
3340