xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 4f6cce39)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"2.2-1"
65 #define DRV_RELDATE	"Feb 2014"
66 
67 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
68 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
69 #define MLX4_IB_CARD_REV_A0   0xA0
70 
71 MODULE_AUTHOR("Roland Dreier");
72 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
73 MODULE_LICENSE("Dual BSD/GPL");
74 MODULE_VERSION(DRV_VERSION);
75 
76 int mlx4_ib_sm_guid_assign = 0;
77 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
78 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
79 
80 static const char mlx4_ib_version[] =
81 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
82 	DRV_VERSION " (" DRV_RELDATE ")\n";
83 
84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 			else
223 				memset(&gid_tbl[i].gid, 0, 12);
224 		}
225 	}
226 
227 	err = mlx4_cmd(dev, mailbox->dma,
228 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
229 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
230 		       MLX4_CMD_WRAPPED);
231 	if (mlx4_is_bonded(dev))
232 		err += mlx4_cmd(dev, mailbox->dma,
233 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
234 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
235 				MLX4_CMD_WRAPPED);
236 
237 	mlx4_free_cmd_mailbox(dev, mailbox);
238 	return err;
239 }
240 
241 static int mlx4_ib_update_gids(struct gid_entry *gids,
242 			       struct mlx4_ib_dev *ibdev,
243 			       u8 port_num)
244 {
245 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
246 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
247 
248 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
249 }
250 
251 static int mlx4_ib_add_gid(struct ib_device *device,
252 			   u8 port_num,
253 			   unsigned int index,
254 			   const union ib_gid *gid,
255 			   const struct ib_gid_attr *attr,
256 			   void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids = NULL;
266 
267 	if (!rdma_cap_roce_gid_table(device, port_num))
268 		return -EINVAL;
269 
270 	if (port_num > MLX4_MAX_PORTS)
271 		return -EINVAL;
272 
273 	if (!context)
274 		return -EINVAL;
275 
276 	port_gid_table = &iboe->gids[port_num - 1];
277 	spin_lock_bh(&iboe->lock);
278 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
279 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
280 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
281 			found = i;
282 			break;
283 		}
284 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
285 			free = i; /* HW has space */
286 	}
287 
288 	if (found < 0) {
289 		if (free < 0) {
290 			ret = -ENOSPC;
291 		} else {
292 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
293 			if (!port_gid_table->gids[free].ctx) {
294 				ret = -ENOMEM;
295 			} else {
296 				*context = port_gid_table->gids[free].ctx;
297 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
298 				port_gid_table->gids[free].gid_type = attr->gid_type;
299 				port_gid_table->gids[free].ctx->real_index = free;
300 				port_gid_table->gids[free].ctx->refcount = 1;
301 				hw_update = 1;
302 			}
303 		}
304 	} else {
305 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
306 		*context = ctx;
307 		ctx->refcount++;
308 	}
309 	if (!ret && hw_update) {
310 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
311 		if (!gids) {
312 			ret = -ENOMEM;
313 		} else {
314 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
315 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
316 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
317 			}
318 		}
319 	}
320 	spin_unlock_bh(&iboe->lock);
321 
322 	if (!ret && hw_update) {
323 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
324 		kfree(gids);
325 	}
326 
327 	return ret;
328 }
329 
330 static int mlx4_ib_del_gid(struct ib_device *device,
331 			   u8 port_num,
332 			   unsigned int index,
333 			   void **context)
334 {
335 	struct gid_cache_context *ctx = *context;
336 	struct mlx4_ib_dev *ibdev = to_mdev(device);
337 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
338 	struct mlx4_port_gid_table   *port_gid_table;
339 	int ret = 0;
340 	int hw_update = 0;
341 	struct gid_entry *gids = NULL;
342 
343 	if (!rdma_cap_roce_gid_table(device, port_num))
344 		return -EINVAL;
345 
346 	if (port_num > MLX4_MAX_PORTS)
347 		return -EINVAL;
348 
349 	port_gid_table = &iboe->gids[port_num - 1];
350 	spin_lock_bh(&iboe->lock);
351 	if (ctx) {
352 		ctx->refcount--;
353 		if (!ctx->refcount) {
354 			unsigned int real_index = ctx->real_index;
355 
356 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
357 			kfree(port_gid_table->gids[real_index].ctx);
358 			port_gid_table->gids[real_index].ctx = NULL;
359 			hw_update = 1;
360 		}
361 	}
362 	if (!ret && hw_update) {
363 		int i;
364 
365 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
366 		if (!gids) {
367 			ret = -ENOMEM;
368 		} else {
369 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
370 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
371 		}
372 	}
373 	spin_unlock_bh(&iboe->lock);
374 
375 	if (!ret && hw_update) {
376 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
377 		kfree(gids);
378 	}
379 	return ret;
380 }
381 
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383 				    u8 port_num, int index)
384 {
385 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386 	struct gid_cache_context *ctx = NULL;
387 	union ib_gid gid;
388 	struct mlx4_port_gid_table   *port_gid_table;
389 	int real_index = -EINVAL;
390 	int i;
391 	int ret;
392 	unsigned long flags;
393 	struct ib_gid_attr attr;
394 
395 	if (port_num > MLX4_MAX_PORTS)
396 		return -EINVAL;
397 
398 	if (mlx4_is_bonded(ibdev->dev))
399 		port_num = 1;
400 
401 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
402 		return index;
403 
404 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
405 	if (ret)
406 		return ret;
407 
408 	if (attr.ndev)
409 		dev_put(attr.ndev);
410 
411 	if (!memcmp(&gid, &zgid, sizeof(gid)))
412 		return -EINVAL;
413 
414 	spin_lock_irqsave(&iboe->lock, flags);
415 	port_gid_table = &iboe->gids[port_num - 1];
416 
417 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
418 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
419 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
420 			ctx = port_gid_table->gids[i].ctx;
421 			break;
422 		}
423 	if (ctx)
424 		real_index = ctx->real_index;
425 	spin_unlock_irqrestore(&iboe->lock, flags);
426 	return real_index;
427 }
428 
429 static int mlx4_ib_query_device(struct ib_device *ibdev,
430 				struct ib_device_attr *props,
431 				struct ib_udata *uhw)
432 {
433 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
434 	struct ib_smp *in_mad  = NULL;
435 	struct ib_smp *out_mad = NULL;
436 	int err;
437 	int have_ib_ports;
438 	struct mlx4_uverbs_ex_query_device cmd;
439 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
440 	struct mlx4_clock_params clock_params;
441 
442 	if (uhw->inlen) {
443 		if (uhw->inlen < sizeof(cmd))
444 			return -EINVAL;
445 
446 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
447 		if (err)
448 			return err;
449 
450 		if (cmd.comp_mask)
451 			return -EINVAL;
452 
453 		if (cmd.reserved)
454 			return -EINVAL;
455 	}
456 
457 	resp.response_length = offsetof(typeof(resp), response_length) +
458 		sizeof(resp.response_length);
459 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
460 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
461 	err = -ENOMEM;
462 	if (!in_mad || !out_mad)
463 		goto out;
464 
465 	init_query_mad(in_mad);
466 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
467 
468 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
469 			   1, NULL, NULL, in_mad, out_mad);
470 	if (err)
471 		goto out;
472 
473 	memset(props, 0, sizeof *props);
474 
475 	have_ib_ports = num_ib_ports(dev->dev);
476 
477 	props->fw_ver = dev->dev->caps.fw_ver;
478 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
479 		IB_DEVICE_PORT_ACTIVE_EVENT		|
480 		IB_DEVICE_SYS_IMAGE_GUID		|
481 		IB_DEVICE_RC_RNR_NAK_GEN		|
482 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
484 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
485 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
486 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
488 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
490 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
492 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
493 	if (dev->dev->caps.max_gso_sz &&
494 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
495 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
496 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
497 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
498 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
499 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
500 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
501 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
502 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
503 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
504 		props->device_cap_flags |= IB_DEVICE_XRC;
505 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
506 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
507 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
508 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
509 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
510 		else
511 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
512 	}
513 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
514 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
515 
516 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
517 
518 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
519 		0xffffff;
520 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
521 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
522 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
523 
524 	props->max_mr_size	   = ~0ull;
525 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
526 	props->max_qp		   = dev->dev->quotas.qp;
527 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
528 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
529 					 dev->dev->caps.max_rq_sg);
530 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
531 	props->max_cq		   = dev->dev->quotas.cq;
532 	props->max_cqe		   = dev->dev->caps.max_cqes;
533 	props->max_mr		   = dev->dev->quotas.mpt;
534 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
535 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
536 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
537 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
538 	props->max_srq		   = dev->dev->quotas.srq;
539 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
540 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
541 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
542 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
543 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
544 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
545 	props->masked_atomic_cap   = props->atomic_cap;
546 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
547 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
548 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
549 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
550 					   props->max_mcast_grp;
551 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
552 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
553 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
554 	props->max_ah = INT_MAX;
555 
556 	if (!mlx4_is_slave(dev->dev))
557 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
558 
559 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
560 		resp.response_length += sizeof(resp.hca_core_clock_offset);
561 		if (!err && !mlx4_is_slave(dev->dev)) {
562 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
563 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
564 		}
565 	}
566 
567 	if (uhw->outlen) {
568 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
569 		if (err)
570 			goto out;
571 	}
572 out:
573 	kfree(in_mad);
574 	kfree(out_mad);
575 
576 	return err;
577 }
578 
579 static enum rdma_link_layer
580 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
581 {
582 	struct mlx4_dev *dev = to_mdev(device)->dev;
583 
584 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
585 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
586 }
587 
588 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
589 			      struct ib_port_attr *props, int netw_view)
590 {
591 	struct ib_smp *in_mad  = NULL;
592 	struct ib_smp *out_mad = NULL;
593 	int ext_active_speed;
594 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
595 	int err = -ENOMEM;
596 
597 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
598 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
599 	if (!in_mad || !out_mad)
600 		goto out;
601 
602 	init_query_mad(in_mad);
603 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
604 	in_mad->attr_mod = cpu_to_be32(port);
605 
606 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
607 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
608 
609 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
610 				in_mad, out_mad);
611 	if (err)
612 		goto out;
613 
614 
615 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
616 	props->lmc		= out_mad->data[34] & 0x7;
617 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
618 	props->sm_sl		= out_mad->data[36] & 0xf;
619 	props->state		= out_mad->data[32] & 0xf;
620 	props->phys_state	= out_mad->data[33] >> 4;
621 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
622 	if (netw_view)
623 		props->gid_tbl_len = out_mad->data[50];
624 	else
625 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
626 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
627 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
628 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
629 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
630 	props->active_width	= out_mad->data[31] & 0xf;
631 	props->active_speed	= out_mad->data[35] >> 4;
632 	props->max_mtu		= out_mad->data[41] & 0xf;
633 	props->active_mtu	= out_mad->data[36] >> 4;
634 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
635 	props->max_vl_num	= out_mad->data[37] >> 4;
636 	props->init_type_reply	= out_mad->data[41] >> 4;
637 
638 	/* Check if extended speeds (EDR/FDR/...) are supported */
639 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
640 		ext_active_speed = out_mad->data[62] >> 4;
641 
642 		switch (ext_active_speed) {
643 		case 1:
644 			props->active_speed = IB_SPEED_FDR;
645 			break;
646 		case 2:
647 			props->active_speed = IB_SPEED_EDR;
648 			break;
649 		}
650 	}
651 
652 	/* If reported active speed is QDR, check if is FDR-10 */
653 	if (props->active_speed == IB_SPEED_QDR) {
654 		init_query_mad(in_mad);
655 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
656 		in_mad->attr_mod = cpu_to_be32(port);
657 
658 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
659 				   NULL, NULL, in_mad, out_mad);
660 		if (err)
661 			goto out;
662 
663 		/* Checking LinkSpeedActive for FDR-10 */
664 		if (out_mad->data[15] & 0x1)
665 			props->active_speed = IB_SPEED_FDR10;
666 	}
667 
668 	/* Avoid wrong speed value returned by FW if the IB link is down. */
669 	if (props->state == IB_PORT_DOWN)
670 		 props->active_speed = IB_SPEED_SDR;
671 
672 out:
673 	kfree(in_mad);
674 	kfree(out_mad);
675 	return err;
676 }
677 
678 static u8 state_to_phys_state(enum ib_port_state state)
679 {
680 	return state == IB_PORT_ACTIVE ? 5 : 3;
681 }
682 
683 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
684 			       struct ib_port_attr *props)
685 {
686 
687 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
688 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
689 	struct net_device *ndev;
690 	enum ib_mtu tmp;
691 	struct mlx4_cmd_mailbox *mailbox;
692 	int err = 0;
693 	int is_bonded = mlx4_is_bonded(mdev->dev);
694 
695 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
696 	if (IS_ERR(mailbox))
697 		return PTR_ERR(mailbox);
698 
699 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
700 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
701 			   MLX4_CMD_WRAPPED);
702 	if (err)
703 		goto out;
704 
705 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
706 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
707 					   IB_WIDTH_4X : IB_WIDTH_1X;
708 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
709 					   IB_SPEED_FDR : IB_SPEED_QDR;
710 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
711 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
712 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
713 	props->pkey_tbl_len	= 1;
714 	props->max_mtu		= IB_MTU_4096;
715 	props->max_vl_num	= 2;
716 	props->state		= IB_PORT_DOWN;
717 	props->phys_state	= state_to_phys_state(props->state);
718 	props->active_mtu	= IB_MTU_256;
719 	spin_lock_bh(&iboe->lock);
720 	ndev = iboe->netdevs[port - 1];
721 	if (ndev && is_bonded) {
722 		rcu_read_lock(); /* required to get upper dev */
723 		ndev = netdev_master_upper_dev_get_rcu(ndev);
724 		rcu_read_unlock();
725 	}
726 	if (!ndev)
727 		goto out_unlock;
728 
729 	tmp = iboe_get_mtu(ndev->mtu);
730 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
731 
732 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
733 					IB_PORT_ACTIVE : IB_PORT_DOWN;
734 	props->phys_state	= state_to_phys_state(props->state);
735 out_unlock:
736 	spin_unlock_bh(&iboe->lock);
737 out:
738 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
739 	return err;
740 }
741 
742 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
743 			 struct ib_port_attr *props, int netw_view)
744 {
745 	int err;
746 
747 	/* props being zeroed by the caller, avoid zeroing it here */
748 
749 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
750 		ib_link_query_port(ibdev, port, props, netw_view) :
751 				eth_link_query_port(ibdev, port, props);
752 
753 	return err;
754 }
755 
756 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
757 			      struct ib_port_attr *props)
758 {
759 	/* returns host view */
760 	return __mlx4_ib_query_port(ibdev, port, props, 0);
761 }
762 
763 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
764 			union ib_gid *gid, int netw_view)
765 {
766 	struct ib_smp *in_mad  = NULL;
767 	struct ib_smp *out_mad = NULL;
768 	int err = -ENOMEM;
769 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
770 	int clear = 0;
771 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
772 
773 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
774 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
775 	if (!in_mad || !out_mad)
776 		goto out;
777 
778 	init_query_mad(in_mad);
779 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
780 	in_mad->attr_mod = cpu_to_be32(port);
781 
782 	if (mlx4_is_mfunc(dev->dev) && netw_view)
783 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
784 
785 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
786 	if (err)
787 		goto out;
788 
789 	memcpy(gid->raw, out_mad->data + 8, 8);
790 
791 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
792 		if (index) {
793 			/* For any index > 0, return the null guid */
794 			err = 0;
795 			clear = 1;
796 			goto out;
797 		}
798 	}
799 
800 	init_query_mad(in_mad);
801 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
802 	in_mad->attr_mod = cpu_to_be32(index / 8);
803 
804 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
805 			   NULL, NULL, in_mad, out_mad);
806 	if (err)
807 		goto out;
808 
809 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
810 
811 out:
812 	if (clear)
813 		memset(gid->raw + 8, 0, 8);
814 	kfree(in_mad);
815 	kfree(out_mad);
816 	return err;
817 }
818 
819 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
820 			     union ib_gid *gid)
821 {
822 	int ret;
823 
824 	if (rdma_protocol_ib(ibdev, port))
825 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
826 
827 	if (!rdma_protocol_roce(ibdev, port))
828 		return -ENODEV;
829 
830 	if (!rdma_cap_roce_gid_table(ibdev, port))
831 		return -ENODEV;
832 
833 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
834 	if (ret == -EAGAIN) {
835 		memcpy(gid, &zgid, sizeof(*gid));
836 		return 0;
837 	}
838 
839 	return ret;
840 }
841 
842 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
843 {
844 	union sl2vl_tbl_to_u64 sl2vl64;
845 	struct ib_smp *in_mad  = NULL;
846 	struct ib_smp *out_mad = NULL;
847 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
848 	int err = -ENOMEM;
849 	int jj;
850 
851 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
852 		*sl2vl_tbl = 0;
853 		return 0;
854 	}
855 
856 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
857 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
858 	if (!in_mad || !out_mad)
859 		goto out;
860 
861 	init_query_mad(in_mad);
862 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
863 	in_mad->attr_mod = 0;
864 
865 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
866 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
867 
868 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
869 			   in_mad, out_mad);
870 	if (err)
871 		goto out;
872 
873 	for (jj = 0; jj < 8; jj++)
874 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
875 	*sl2vl_tbl = sl2vl64.sl64;
876 
877 out:
878 	kfree(in_mad);
879 	kfree(out_mad);
880 	return err;
881 }
882 
883 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
884 {
885 	u64 sl2vl;
886 	int i;
887 	int err;
888 
889 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
890 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
891 			continue;
892 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
893 		if (err) {
894 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
895 			       i, err);
896 			sl2vl = 0;
897 		}
898 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
899 	}
900 }
901 
902 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
903 			 u16 *pkey, int netw_view)
904 {
905 	struct ib_smp *in_mad  = NULL;
906 	struct ib_smp *out_mad = NULL;
907 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
908 	int err = -ENOMEM;
909 
910 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
911 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
912 	if (!in_mad || !out_mad)
913 		goto out;
914 
915 	init_query_mad(in_mad);
916 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
917 	in_mad->attr_mod = cpu_to_be32(index / 32);
918 
919 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
920 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
921 
922 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
923 			   in_mad, out_mad);
924 	if (err)
925 		goto out;
926 
927 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
928 
929 out:
930 	kfree(in_mad);
931 	kfree(out_mad);
932 	return err;
933 }
934 
935 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
936 {
937 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
938 }
939 
940 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
941 				 struct ib_device_modify *props)
942 {
943 	struct mlx4_cmd_mailbox *mailbox;
944 	unsigned long flags;
945 
946 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
947 		return -EOPNOTSUPP;
948 
949 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
950 		return 0;
951 
952 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
953 		return -EOPNOTSUPP;
954 
955 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
956 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
957 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
958 
959 	/*
960 	 * If possible, pass node desc to FW, so it can generate
961 	 * a 144 trap.  If cmd fails, just ignore.
962 	 */
963 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
964 	if (IS_ERR(mailbox))
965 		return 0;
966 
967 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
968 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
969 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
970 
971 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
972 
973 	return 0;
974 }
975 
976 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
977 			    u32 cap_mask)
978 {
979 	struct mlx4_cmd_mailbox *mailbox;
980 	int err;
981 
982 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
983 	if (IS_ERR(mailbox))
984 		return PTR_ERR(mailbox);
985 
986 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
987 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
988 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
989 	} else {
990 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
991 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
992 	}
993 
994 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
995 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
996 		       MLX4_CMD_WRAPPED);
997 
998 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
999 	return err;
1000 }
1001 
1002 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1003 			       struct ib_port_modify *props)
1004 {
1005 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1006 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1007 	struct ib_port_attr attr;
1008 	u32 cap_mask;
1009 	int err;
1010 
1011 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1012 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1013 	 * violations and port capabilities are not meaningful.
1014 	 */
1015 	if (is_eth)
1016 		return 0;
1017 
1018 	mutex_lock(&mdev->cap_mask_mutex);
1019 
1020 	err = ib_query_port(ibdev, port, &attr);
1021 	if (err)
1022 		goto out;
1023 
1024 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1025 		~props->clr_port_cap_mask;
1026 
1027 	err = mlx4_ib_SET_PORT(mdev, port,
1028 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1029 			       cap_mask);
1030 
1031 out:
1032 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1033 	return err;
1034 }
1035 
1036 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1037 						  struct ib_udata *udata)
1038 {
1039 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1040 	struct mlx4_ib_ucontext *context;
1041 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1042 	struct mlx4_ib_alloc_ucontext_resp resp;
1043 	int err;
1044 
1045 	if (!dev->ib_active)
1046 		return ERR_PTR(-EAGAIN);
1047 
1048 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1049 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1050 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1051 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1052 	} else {
1053 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1054 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1055 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1056 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1057 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1058 	}
1059 
1060 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1061 	if (!context)
1062 		return ERR_PTR(-ENOMEM);
1063 
1064 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1065 	if (err) {
1066 		kfree(context);
1067 		return ERR_PTR(err);
1068 	}
1069 
1070 	INIT_LIST_HEAD(&context->db_page_list);
1071 	mutex_init(&context->db_page_mutex);
1072 
1073 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1074 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1075 	else
1076 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1077 
1078 	if (err) {
1079 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1080 		kfree(context);
1081 		return ERR_PTR(-EFAULT);
1082 	}
1083 
1084 	return &context->ibucontext;
1085 }
1086 
1087 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1088 {
1089 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1090 
1091 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1092 	kfree(context);
1093 
1094 	return 0;
1095 }
1096 
1097 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1098 {
1099 	/* vma_open is called when a new VMA is created on top of our VMA.
1100 	 * This is done through either mremap flow or split_vma (usually due
1101 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1102 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1103 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1104 	 * calling us again and trying to do incorrect actions. We assume that
1105 	 * the original vma size is exactly a single page that there will be no
1106 	 * "splitting" operations on.
1107 	 */
1108 	area->vm_ops = NULL;
1109 }
1110 
1111 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1112 {
1113 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1114 
1115 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1116 	 * file itself is closed, therefore no sync is needed with the regular
1117 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1118 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1119 	 * The close operation is usually called under mm->mmap_sem except when
1120 	 * process is exiting.  The exiting case is handled explicitly as part
1121 	 * of mlx4_ib_disassociate_ucontext.
1122 	 */
1123 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1124 				area->vm_private_data;
1125 
1126 	/* set the vma context pointer to null in the mlx4_ib driver's private
1127 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1128 	 */
1129 	mlx4_ib_vma_priv_data->vma = NULL;
1130 }
1131 
1132 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1133 	.open = mlx4_ib_vma_open,
1134 	.close = mlx4_ib_vma_close
1135 };
1136 
1137 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1138 {
1139 	int i;
1140 	int ret = 0;
1141 	struct vm_area_struct *vma;
1142 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1143 	struct task_struct *owning_process  = NULL;
1144 	struct mm_struct   *owning_mm       = NULL;
1145 
1146 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1147 	if (!owning_process)
1148 		return;
1149 
1150 	owning_mm = get_task_mm(owning_process);
1151 	if (!owning_mm) {
1152 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1153 		while (1) {
1154 			/* make sure that task is dead before returning, it may
1155 			 * prevent a rare case of module down in parallel to a
1156 			 * call to mlx4_ib_vma_close.
1157 			 */
1158 			put_task_struct(owning_process);
1159 			msleep(1);
1160 			owning_process = get_pid_task(ibcontext->tgid,
1161 						      PIDTYPE_PID);
1162 			if (!owning_process ||
1163 			    owning_process->state == TASK_DEAD) {
1164 				pr_info("disassociate ucontext done, task was terminated\n");
1165 				/* in case task was dead need to release the task struct */
1166 				if (owning_process)
1167 					put_task_struct(owning_process);
1168 				return;
1169 			}
1170 		}
1171 	}
1172 
1173 	/* need to protect from a race on closing the vma as part of
1174 	 * mlx4_ib_vma_close().
1175 	 */
1176 	down_read(&owning_mm->mmap_sem);
1177 	for (i = 0; i < HW_BAR_COUNT; i++) {
1178 		vma = context->hw_bar_info[i].vma;
1179 		if (!vma)
1180 			continue;
1181 
1182 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1183 				   context->hw_bar_info[i].vma->vm_start,
1184 				   PAGE_SIZE);
1185 		if (ret) {
1186 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1187 			BUG_ON(1);
1188 		}
1189 
1190 		/* context going to be destroyed, should not access ops any more */
1191 		context->hw_bar_info[i].vma->vm_ops = NULL;
1192 	}
1193 
1194 	up_read(&owning_mm->mmap_sem);
1195 	mmput(owning_mm);
1196 	put_task_struct(owning_process);
1197 }
1198 
1199 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1200 				 struct mlx4_ib_vma_private_data *vma_private_data)
1201 {
1202 	vma_private_data->vma = vma;
1203 	vma->vm_private_data = vma_private_data;
1204 	vma->vm_ops =  &mlx4_ib_vm_ops;
1205 }
1206 
1207 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1208 {
1209 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1210 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1211 
1212 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1213 		return -EINVAL;
1214 
1215 	if (vma->vm_pgoff == 0) {
1216 		/* We prevent double mmaping on same context */
1217 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1218 			return -EINVAL;
1219 
1220 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1221 
1222 		if (io_remap_pfn_range(vma, vma->vm_start,
1223 				       to_mucontext(context)->uar.pfn,
1224 				       PAGE_SIZE, vma->vm_page_prot))
1225 			return -EAGAIN;
1226 
1227 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1228 
1229 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1230 		/* We prevent double mmaping on same context */
1231 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1232 			return -EINVAL;
1233 
1234 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1235 
1236 		if (io_remap_pfn_range(vma, vma->vm_start,
1237 				       to_mucontext(context)->uar.pfn +
1238 				       dev->dev->caps.num_uars,
1239 				       PAGE_SIZE, vma->vm_page_prot))
1240 			return -EAGAIN;
1241 
1242 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1243 
1244 	} else if (vma->vm_pgoff == 3) {
1245 		struct mlx4_clock_params params;
1246 		int ret;
1247 
1248 		/* We prevent double mmaping on same context */
1249 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1250 			return -EINVAL;
1251 
1252 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1253 
1254 		if (ret)
1255 			return ret;
1256 
1257 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1258 		if (io_remap_pfn_range(vma, vma->vm_start,
1259 				       (pci_resource_start(dev->dev->persist->pdev,
1260 							   params.bar) +
1261 					params.offset)
1262 				       >> PAGE_SHIFT,
1263 				       PAGE_SIZE, vma->vm_page_prot))
1264 			return -EAGAIN;
1265 
1266 		mlx4_ib_set_vma_data(vma,
1267 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1268 	} else {
1269 		return -EINVAL;
1270 	}
1271 
1272 	return 0;
1273 }
1274 
1275 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1276 				      struct ib_ucontext *context,
1277 				      struct ib_udata *udata)
1278 {
1279 	struct mlx4_ib_pd *pd;
1280 	int err;
1281 
1282 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1283 	if (!pd)
1284 		return ERR_PTR(-ENOMEM);
1285 
1286 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1287 	if (err) {
1288 		kfree(pd);
1289 		return ERR_PTR(err);
1290 	}
1291 
1292 	if (context)
1293 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1294 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1295 			kfree(pd);
1296 			return ERR_PTR(-EFAULT);
1297 		}
1298 
1299 	return &pd->ibpd;
1300 }
1301 
1302 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1303 {
1304 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1305 	kfree(pd);
1306 
1307 	return 0;
1308 }
1309 
1310 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1311 					  struct ib_ucontext *context,
1312 					  struct ib_udata *udata)
1313 {
1314 	struct mlx4_ib_xrcd *xrcd;
1315 	struct ib_cq_init_attr cq_attr = {};
1316 	int err;
1317 
1318 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1319 		return ERR_PTR(-ENOSYS);
1320 
1321 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1322 	if (!xrcd)
1323 		return ERR_PTR(-ENOMEM);
1324 
1325 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1326 	if (err)
1327 		goto err1;
1328 
1329 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1330 	if (IS_ERR(xrcd->pd)) {
1331 		err = PTR_ERR(xrcd->pd);
1332 		goto err2;
1333 	}
1334 
1335 	cq_attr.cqe = 1;
1336 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1337 	if (IS_ERR(xrcd->cq)) {
1338 		err = PTR_ERR(xrcd->cq);
1339 		goto err3;
1340 	}
1341 
1342 	return &xrcd->ibxrcd;
1343 
1344 err3:
1345 	ib_dealloc_pd(xrcd->pd);
1346 err2:
1347 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1348 err1:
1349 	kfree(xrcd);
1350 	return ERR_PTR(err);
1351 }
1352 
1353 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1354 {
1355 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1356 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1357 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1358 	kfree(xrcd);
1359 
1360 	return 0;
1361 }
1362 
1363 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1364 {
1365 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1366 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1367 	struct mlx4_ib_gid_entry *ge;
1368 
1369 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1370 	if (!ge)
1371 		return -ENOMEM;
1372 
1373 	ge->gid = *gid;
1374 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1375 		ge->port = mqp->port;
1376 		ge->added = 1;
1377 	}
1378 
1379 	mutex_lock(&mqp->mutex);
1380 	list_add_tail(&ge->list, &mqp->gid_list);
1381 	mutex_unlock(&mqp->mutex);
1382 
1383 	return 0;
1384 }
1385 
1386 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1387 					  struct mlx4_ib_counters *ctr_table)
1388 {
1389 	struct counter_index *counter, *tmp_count;
1390 
1391 	mutex_lock(&ctr_table->mutex);
1392 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1393 				 list) {
1394 		if (counter->allocated)
1395 			mlx4_counter_free(ibdev->dev, counter->index);
1396 		list_del(&counter->list);
1397 		kfree(counter);
1398 	}
1399 	mutex_unlock(&ctr_table->mutex);
1400 }
1401 
1402 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1403 		   union ib_gid *gid)
1404 {
1405 	struct net_device *ndev;
1406 	int ret = 0;
1407 
1408 	if (!mqp->port)
1409 		return 0;
1410 
1411 	spin_lock_bh(&mdev->iboe.lock);
1412 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1413 	if (ndev)
1414 		dev_hold(ndev);
1415 	spin_unlock_bh(&mdev->iboe.lock);
1416 
1417 	if (ndev) {
1418 		ret = 1;
1419 		dev_put(ndev);
1420 	}
1421 
1422 	return ret;
1423 }
1424 
1425 struct mlx4_ib_steering {
1426 	struct list_head list;
1427 	struct mlx4_flow_reg_id reg_id;
1428 	union ib_gid gid;
1429 };
1430 
1431 #define LAST_ETH_FIELD vlan_tag
1432 #define LAST_IB_FIELD sl
1433 #define LAST_IPV4_FIELD dst_ip
1434 #define LAST_TCP_UDP_FIELD src_port
1435 
1436 /* Field is the last supported field */
1437 #define FIELDS_NOT_SUPPORTED(filter, field)\
1438 	memchr_inv((void *)&filter.field  +\
1439 		   sizeof(filter.field), 0,\
1440 		   sizeof(filter) -\
1441 		   offsetof(typeof(filter), field) -\
1442 		   sizeof(filter.field))
1443 
1444 static int parse_flow_attr(struct mlx4_dev *dev,
1445 			   u32 qp_num,
1446 			   union ib_flow_spec *ib_spec,
1447 			   struct _rule_hw *mlx4_spec)
1448 {
1449 	enum mlx4_net_trans_rule_id type;
1450 
1451 	switch (ib_spec->type) {
1452 	case IB_FLOW_SPEC_ETH:
1453 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1454 			return -ENOTSUPP;
1455 
1456 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1457 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1458 		       ETH_ALEN);
1459 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1460 		       ETH_ALEN);
1461 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1462 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1463 		break;
1464 	case IB_FLOW_SPEC_IB:
1465 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1466 			return -ENOTSUPP;
1467 
1468 		type = MLX4_NET_TRANS_RULE_ID_IB;
1469 		mlx4_spec->ib.l3_qpn =
1470 			cpu_to_be32(qp_num);
1471 		mlx4_spec->ib.qpn_mask =
1472 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1473 		break;
1474 
1475 
1476 	case IB_FLOW_SPEC_IPV4:
1477 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1478 			return -ENOTSUPP;
1479 
1480 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1481 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1482 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1483 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1484 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1485 		break;
1486 
1487 	case IB_FLOW_SPEC_TCP:
1488 	case IB_FLOW_SPEC_UDP:
1489 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1490 			return -ENOTSUPP;
1491 
1492 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1493 					MLX4_NET_TRANS_RULE_ID_TCP :
1494 					MLX4_NET_TRANS_RULE_ID_UDP;
1495 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1496 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1497 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1498 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1499 		break;
1500 
1501 	default:
1502 		return -EINVAL;
1503 	}
1504 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1505 	    mlx4_hw_rule_sz(dev, type) < 0)
1506 		return -EINVAL;
1507 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1508 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1509 	return mlx4_hw_rule_sz(dev, type);
1510 }
1511 
1512 struct default_rules {
1513 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1514 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1515 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1516 	__u8  link_layer;
1517 };
1518 static const struct default_rules default_table[] = {
1519 	{
1520 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1521 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1522 		.rules_create_list = {IB_FLOW_SPEC_IB},
1523 		.link_layer = IB_LINK_LAYER_INFINIBAND
1524 	}
1525 };
1526 
1527 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1528 					 struct ib_flow_attr *flow_attr)
1529 {
1530 	int i, j, k;
1531 	void *ib_flow;
1532 	const struct default_rules *pdefault_rules = default_table;
1533 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1534 
1535 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1536 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1537 		memset(&field_types, 0, sizeof(field_types));
1538 
1539 		if (link_layer != pdefault_rules->link_layer)
1540 			continue;
1541 
1542 		ib_flow = flow_attr + 1;
1543 		/* we assume the specs are sorted */
1544 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1545 		     j < flow_attr->num_of_specs; k++) {
1546 			union ib_flow_spec *current_flow =
1547 				(union ib_flow_spec *)ib_flow;
1548 
1549 			/* same layer but different type */
1550 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1551 			     (pdefault_rules->mandatory_fields[k] &
1552 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1553 			    (current_flow->type !=
1554 			     pdefault_rules->mandatory_fields[k]))
1555 				goto out;
1556 
1557 			/* same layer, try match next one */
1558 			if (current_flow->type ==
1559 			    pdefault_rules->mandatory_fields[k]) {
1560 				j++;
1561 				ib_flow +=
1562 					((union ib_flow_spec *)ib_flow)->size;
1563 			}
1564 		}
1565 
1566 		ib_flow = flow_attr + 1;
1567 		for (j = 0; j < flow_attr->num_of_specs;
1568 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1569 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1570 				/* same layer and same type */
1571 				if (((union ib_flow_spec *)ib_flow)->type ==
1572 				    pdefault_rules->mandatory_not_fields[k])
1573 					goto out;
1574 
1575 		return i;
1576 	}
1577 out:
1578 	return -1;
1579 }
1580 
1581 static int __mlx4_ib_create_default_rules(
1582 		struct mlx4_ib_dev *mdev,
1583 		struct ib_qp *qp,
1584 		const struct default_rules *pdefault_rules,
1585 		struct _rule_hw *mlx4_spec) {
1586 	int size = 0;
1587 	int i;
1588 
1589 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1590 		int ret;
1591 		union ib_flow_spec ib_spec;
1592 		switch (pdefault_rules->rules_create_list[i]) {
1593 		case 0:
1594 			/* no rule */
1595 			continue;
1596 		case IB_FLOW_SPEC_IB:
1597 			ib_spec.type = IB_FLOW_SPEC_IB;
1598 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1599 
1600 			break;
1601 		default:
1602 			/* invalid rule */
1603 			return -EINVAL;
1604 		}
1605 		/* We must put empty rule, qpn is being ignored */
1606 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1607 				      mlx4_spec);
1608 		if (ret < 0) {
1609 			pr_info("invalid parsing\n");
1610 			return -EINVAL;
1611 		}
1612 
1613 		mlx4_spec = (void *)mlx4_spec + ret;
1614 		size += ret;
1615 	}
1616 	return size;
1617 }
1618 
1619 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1620 			  int domain,
1621 			  enum mlx4_net_trans_promisc_mode flow_type,
1622 			  u64 *reg_id)
1623 {
1624 	int ret, i;
1625 	int size = 0;
1626 	void *ib_flow;
1627 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1628 	struct mlx4_cmd_mailbox *mailbox;
1629 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1630 	int default_flow;
1631 
1632 	static const u16 __mlx4_domain[] = {
1633 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1634 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1635 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1636 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1637 	};
1638 
1639 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1640 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1641 		return -EINVAL;
1642 	}
1643 
1644 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1645 		pr_err("Invalid domain value %d\n", domain);
1646 		return -EINVAL;
1647 	}
1648 
1649 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1650 		return -EINVAL;
1651 
1652 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1653 	if (IS_ERR(mailbox))
1654 		return PTR_ERR(mailbox);
1655 	ctrl = mailbox->buf;
1656 
1657 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1658 				 flow_attr->priority);
1659 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1660 	ctrl->port = flow_attr->port;
1661 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1662 
1663 	ib_flow = flow_attr + 1;
1664 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1665 	/* Add default flows */
1666 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1667 	if (default_flow >= 0) {
1668 		ret = __mlx4_ib_create_default_rules(
1669 				mdev, qp, default_table + default_flow,
1670 				mailbox->buf + size);
1671 		if (ret < 0) {
1672 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1673 			return -EINVAL;
1674 		}
1675 		size += ret;
1676 	}
1677 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1678 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1679 				      mailbox->buf + size);
1680 		if (ret < 0) {
1681 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1682 			return -EINVAL;
1683 		}
1684 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1685 		size += ret;
1686 	}
1687 
1688 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1689 	    flow_attr->num_of_specs == 1) {
1690 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1691 		enum ib_flow_spec_type header_spec =
1692 			((union ib_flow_spec *)(flow_attr + 1))->type;
1693 
1694 		if (header_spec == IB_FLOW_SPEC_ETH)
1695 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1696 	}
1697 
1698 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1699 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1700 			   MLX4_CMD_NATIVE);
1701 	if (ret == -ENOMEM)
1702 		pr_err("mcg table is full. Fail to register network rule.\n");
1703 	else if (ret == -ENXIO)
1704 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1705 	else if (ret)
1706 		pr_err("Invalid argument. Fail to register network rule.\n");
1707 
1708 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1709 	return ret;
1710 }
1711 
1712 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1713 {
1714 	int err;
1715 	err = mlx4_cmd(dev, reg_id, 0, 0,
1716 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1717 		       MLX4_CMD_NATIVE);
1718 	if (err)
1719 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1720 		       reg_id);
1721 	return err;
1722 }
1723 
1724 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1725 				    u64 *reg_id)
1726 {
1727 	void *ib_flow;
1728 	union ib_flow_spec *ib_spec;
1729 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1730 	int err = 0;
1731 
1732 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1733 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1734 		return 0; /* do nothing */
1735 
1736 	ib_flow = flow_attr + 1;
1737 	ib_spec = (union ib_flow_spec *)ib_flow;
1738 
1739 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1740 		return 0; /* do nothing */
1741 
1742 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1743 				    flow_attr->port, qp->qp_num,
1744 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1745 				    reg_id);
1746 	return err;
1747 }
1748 
1749 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1750 				      struct ib_flow_attr *flow_attr,
1751 				      enum mlx4_net_trans_promisc_mode *type)
1752 {
1753 	int err = 0;
1754 
1755 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1756 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1757 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1758 		return -EOPNOTSUPP;
1759 	}
1760 
1761 	if (flow_attr->num_of_specs == 0) {
1762 		type[0] = MLX4_FS_MC_SNIFFER;
1763 		type[1] = MLX4_FS_UC_SNIFFER;
1764 	} else {
1765 		union ib_flow_spec *ib_spec;
1766 
1767 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1768 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1769 			return -EINVAL;
1770 
1771 		/* if all is zero than MC and UC */
1772 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1773 			type[0] = MLX4_FS_MC_SNIFFER;
1774 			type[1] = MLX4_FS_UC_SNIFFER;
1775 		} else {
1776 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1777 					    ib_spec->eth.mask.dst_mac[1],
1778 					    ib_spec->eth.mask.dst_mac[2],
1779 					    ib_spec->eth.mask.dst_mac[3],
1780 					    ib_spec->eth.mask.dst_mac[4],
1781 					    ib_spec->eth.mask.dst_mac[5]};
1782 
1783 			/* Above xor was only on MC bit, non empty mask is valid
1784 			 * only if this bit is set and rest are zero.
1785 			 */
1786 			if (!is_zero_ether_addr(&mac[0]))
1787 				return -EINVAL;
1788 
1789 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1790 				type[0] = MLX4_FS_MC_SNIFFER;
1791 			else
1792 				type[0] = MLX4_FS_UC_SNIFFER;
1793 		}
1794 	}
1795 
1796 	return err;
1797 }
1798 
1799 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1800 				    struct ib_flow_attr *flow_attr,
1801 				    int domain)
1802 {
1803 	int err = 0, i = 0, j = 0;
1804 	struct mlx4_ib_flow *mflow;
1805 	enum mlx4_net_trans_promisc_mode type[2];
1806 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1807 	int is_bonded = mlx4_is_bonded(dev);
1808 
1809 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1810 		return ERR_PTR(-EINVAL);
1811 
1812 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1813 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1814 		return ERR_PTR(-EOPNOTSUPP);
1815 
1816 	memset(type, 0, sizeof(type));
1817 
1818 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1819 	if (!mflow) {
1820 		err = -ENOMEM;
1821 		goto err_free;
1822 	}
1823 
1824 	switch (flow_attr->type) {
1825 	case IB_FLOW_ATTR_NORMAL:
1826 		/* If dont trap flag (continue match) is set, under specific
1827 		 * condition traffic be replicated to given qp,
1828 		 * without stealing it
1829 		 */
1830 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1831 			err = mlx4_ib_add_dont_trap_rule(dev,
1832 							 flow_attr,
1833 							 type);
1834 			if (err)
1835 				goto err_free;
1836 		} else {
1837 			type[0] = MLX4_FS_REGULAR;
1838 		}
1839 		break;
1840 
1841 	case IB_FLOW_ATTR_ALL_DEFAULT:
1842 		type[0] = MLX4_FS_ALL_DEFAULT;
1843 		break;
1844 
1845 	case IB_FLOW_ATTR_MC_DEFAULT:
1846 		type[0] = MLX4_FS_MC_DEFAULT;
1847 		break;
1848 
1849 	case IB_FLOW_ATTR_SNIFFER:
1850 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1851 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1852 		break;
1853 
1854 	default:
1855 		err = -EINVAL;
1856 		goto err_free;
1857 	}
1858 
1859 	while (i < ARRAY_SIZE(type) && type[i]) {
1860 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1861 					    &mflow->reg_id[i].id);
1862 		if (err)
1863 			goto err_create_flow;
1864 		if (is_bonded) {
1865 			/* Application always sees one port so the mirror rule
1866 			 * must be on port #2
1867 			 */
1868 			flow_attr->port = 2;
1869 			err = __mlx4_ib_create_flow(qp, flow_attr,
1870 						    domain, type[j],
1871 						    &mflow->reg_id[j].mirror);
1872 			flow_attr->port = 1;
1873 			if (err)
1874 				goto err_create_flow;
1875 			j++;
1876 		}
1877 
1878 		i++;
1879 	}
1880 
1881 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1882 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1883 					       &mflow->reg_id[i].id);
1884 		if (err)
1885 			goto err_create_flow;
1886 
1887 		if (is_bonded) {
1888 			flow_attr->port = 2;
1889 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1890 						       &mflow->reg_id[j].mirror);
1891 			flow_attr->port = 1;
1892 			if (err)
1893 				goto err_create_flow;
1894 			j++;
1895 		}
1896 		/* function to create mirror rule */
1897 		i++;
1898 	}
1899 
1900 	return &mflow->ibflow;
1901 
1902 err_create_flow:
1903 	while (i) {
1904 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1905 					     mflow->reg_id[i].id);
1906 		i--;
1907 	}
1908 
1909 	while (j) {
1910 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1911 					     mflow->reg_id[j].mirror);
1912 		j--;
1913 	}
1914 err_free:
1915 	kfree(mflow);
1916 	return ERR_PTR(err);
1917 }
1918 
1919 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1920 {
1921 	int err, ret = 0;
1922 	int i = 0;
1923 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1924 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1925 
1926 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1927 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1928 		if (err)
1929 			ret = err;
1930 		if (mflow->reg_id[i].mirror) {
1931 			err = __mlx4_ib_destroy_flow(mdev->dev,
1932 						     mflow->reg_id[i].mirror);
1933 			if (err)
1934 				ret = err;
1935 		}
1936 		i++;
1937 	}
1938 
1939 	kfree(mflow);
1940 	return ret;
1941 }
1942 
1943 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1944 {
1945 	int err;
1946 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1947 	struct mlx4_dev	*dev = mdev->dev;
1948 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1949 	struct mlx4_ib_steering *ib_steering = NULL;
1950 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1951 	struct mlx4_flow_reg_id	reg_id;
1952 
1953 	if (mdev->dev->caps.steering_mode ==
1954 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1955 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1956 		if (!ib_steering)
1957 			return -ENOMEM;
1958 	}
1959 
1960 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1961 				    !!(mqp->flags &
1962 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1963 				    prot, &reg_id.id);
1964 	if (err) {
1965 		pr_err("multicast attach op failed, err %d\n", err);
1966 		goto err_malloc;
1967 	}
1968 
1969 	reg_id.mirror = 0;
1970 	if (mlx4_is_bonded(dev)) {
1971 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1972 					    (mqp->port == 1) ? 2 : 1,
1973 					    !!(mqp->flags &
1974 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1975 					    prot, &reg_id.mirror);
1976 		if (err)
1977 			goto err_add;
1978 	}
1979 
1980 	err = add_gid_entry(ibqp, gid);
1981 	if (err)
1982 		goto err_add;
1983 
1984 	if (ib_steering) {
1985 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1986 		ib_steering->reg_id = reg_id;
1987 		mutex_lock(&mqp->mutex);
1988 		list_add(&ib_steering->list, &mqp->steering_rules);
1989 		mutex_unlock(&mqp->mutex);
1990 	}
1991 	return 0;
1992 
1993 err_add:
1994 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1995 			      prot, reg_id.id);
1996 	if (reg_id.mirror)
1997 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1998 				      prot, reg_id.mirror);
1999 err_malloc:
2000 	kfree(ib_steering);
2001 
2002 	return err;
2003 }
2004 
2005 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2006 {
2007 	struct mlx4_ib_gid_entry *ge;
2008 	struct mlx4_ib_gid_entry *tmp;
2009 	struct mlx4_ib_gid_entry *ret = NULL;
2010 
2011 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2012 		if (!memcmp(raw, ge->gid.raw, 16)) {
2013 			ret = ge;
2014 			break;
2015 		}
2016 	}
2017 
2018 	return ret;
2019 }
2020 
2021 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2022 {
2023 	int err;
2024 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2025 	struct mlx4_dev *dev = mdev->dev;
2026 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2027 	struct net_device *ndev;
2028 	struct mlx4_ib_gid_entry *ge;
2029 	struct mlx4_flow_reg_id reg_id = {0, 0};
2030 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2031 
2032 	if (mdev->dev->caps.steering_mode ==
2033 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2034 		struct mlx4_ib_steering *ib_steering;
2035 
2036 		mutex_lock(&mqp->mutex);
2037 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2038 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2039 				list_del(&ib_steering->list);
2040 				break;
2041 			}
2042 		}
2043 		mutex_unlock(&mqp->mutex);
2044 		if (&ib_steering->list == &mqp->steering_rules) {
2045 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2046 			return -EINVAL;
2047 		}
2048 		reg_id = ib_steering->reg_id;
2049 		kfree(ib_steering);
2050 	}
2051 
2052 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2053 				    prot, reg_id.id);
2054 	if (err)
2055 		return err;
2056 
2057 	if (mlx4_is_bonded(dev)) {
2058 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2059 					    prot, reg_id.mirror);
2060 		if (err)
2061 			return err;
2062 	}
2063 
2064 	mutex_lock(&mqp->mutex);
2065 	ge = find_gid_entry(mqp, gid->raw);
2066 	if (ge) {
2067 		spin_lock_bh(&mdev->iboe.lock);
2068 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2069 		if (ndev)
2070 			dev_hold(ndev);
2071 		spin_unlock_bh(&mdev->iboe.lock);
2072 		if (ndev)
2073 			dev_put(ndev);
2074 		list_del(&ge->list);
2075 		kfree(ge);
2076 	} else
2077 		pr_warn("could not find mgid entry\n");
2078 
2079 	mutex_unlock(&mqp->mutex);
2080 
2081 	return 0;
2082 }
2083 
2084 static int init_node_data(struct mlx4_ib_dev *dev)
2085 {
2086 	struct ib_smp *in_mad  = NULL;
2087 	struct ib_smp *out_mad = NULL;
2088 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2089 	int err = -ENOMEM;
2090 
2091 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2092 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2093 	if (!in_mad || !out_mad)
2094 		goto out;
2095 
2096 	init_query_mad(in_mad);
2097 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2098 	if (mlx4_is_master(dev->dev))
2099 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2100 
2101 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2102 	if (err)
2103 		goto out;
2104 
2105 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2106 
2107 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2108 
2109 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2110 	if (err)
2111 		goto out;
2112 
2113 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2114 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2115 
2116 out:
2117 	kfree(in_mad);
2118 	kfree(out_mad);
2119 	return err;
2120 }
2121 
2122 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2123 			char *buf)
2124 {
2125 	struct mlx4_ib_dev *dev =
2126 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2127 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2128 }
2129 
2130 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2131 			char *buf)
2132 {
2133 	struct mlx4_ib_dev *dev =
2134 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2135 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2136 }
2137 
2138 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2139 			  char *buf)
2140 {
2141 	struct mlx4_ib_dev *dev =
2142 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2143 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2144 		       dev->dev->board_id);
2145 }
2146 
2147 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2148 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2149 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2150 
2151 static struct device_attribute *mlx4_class_attributes[] = {
2152 	&dev_attr_hw_rev,
2153 	&dev_attr_hca_type,
2154 	&dev_attr_board_id
2155 };
2156 
2157 struct diag_counter {
2158 	const char *name;
2159 	u32 offset;
2160 };
2161 
2162 #define DIAG_COUNTER(_name, _offset)			\
2163 	{ .name = #_name, .offset = _offset }
2164 
2165 static const struct diag_counter diag_basic[] = {
2166 	DIAG_COUNTER(rq_num_lle, 0x00),
2167 	DIAG_COUNTER(sq_num_lle, 0x04),
2168 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2169 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2170 	DIAG_COUNTER(rq_num_lpe, 0x18),
2171 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2172 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2173 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2174 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2175 	DIAG_COUNTER(sq_num_bre, 0x34),
2176 	DIAG_COUNTER(sq_num_rire, 0x44),
2177 	DIAG_COUNTER(rq_num_rire, 0x48),
2178 	DIAG_COUNTER(sq_num_rae, 0x4C),
2179 	DIAG_COUNTER(rq_num_rae, 0x50),
2180 	DIAG_COUNTER(sq_num_roe, 0x54),
2181 	DIAG_COUNTER(sq_num_tree, 0x5C),
2182 	DIAG_COUNTER(sq_num_rree, 0x64),
2183 	DIAG_COUNTER(rq_num_rnr, 0x68),
2184 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2185 	DIAG_COUNTER(rq_num_oos, 0x100),
2186 	DIAG_COUNTER(sq_num_oos, 0x104),
2187 };
2188 
2189 static const struct diag_counter diag_ext[] = {
2190 	DIAG_COUNTER(rq_num_dup, 0x130),
2191 	DIAG_COUNTER(sq_num_to, 0x134),
2192 };
2193 
2194 static const struct diag_counter diag_device_only[] = {
2195 	DIAG_COUNTER(num_cqovf, 0x1A0),
2196 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2197 };
2198 
2199 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2200 						    u8 port_num)
2201 {
2202 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2203 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2204 
2205 	if (!diag[!!port_num].name)
2206 		return NULL;
2207 
2208 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2209 					  diag[!!port_num].num_counters,
2210 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2211 }
2212 
2213 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2214 				struct rdma_hw_stats *stats,
2215 				u8 port, int index)
2216 {
2217 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2218 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2219 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2220 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2221 	int ret;
2222 	int i;
2223 
2224 	ret = mlx4_query_diag_counters(dev->dev,
2225 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2226 				       diag[!!port].offset, hw_value,
2227 				       diag[!!port].num_counters, port);
2228 
2229 	if (ret)
2230 		return ret;
2231 
2232 	for (i = 0; i < diag[!!port].num_counters; i++)
2233 		stats->value[i] = hw_value[i];
2234 
2235 	return diag[!!port].num_counters;
2236 }
2237 
2238 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2239 					 const char ***name,
2240 					 u32 **offset,
2241 					 u32 *num,
2242 					 bool port)
2243 {
2244 	u32 num_counters;
2245 
2246 	num_counters = ARRAY_SIZE(diag_basic);
2247 
2248 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2249 		num_counters += ARRAY_SIZE(diag_ext);
2250 
2251 	if (!port)
2252 		num_counters += ARRAY_SIZE(diag_device_only);
2253 
2254 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2255 	if (!*name)
2256 		return -ENOMEM;
2257 
2258 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2259 	if (!*offset)
2260 		goto err_name;
2261 
2262 	*num = num_counters;
2263 
2264 	return 0;
2265 
2266 err_name:
2267 	kfree(*name);
2268 	return -ENOMEM;
2269 }
2270 
2271 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2272 				       const char **name,
2273 				       u32 *offset,
2274 				       bool port)
2275 {
2276 	int i;
2277 	int j;
2278 
2279 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2280 		name[i] = diag_basic[i].name;
2281 		offset[i] = diag_basic[i].offset;
2282 	}
2283 
2284 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2285 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2286 			name[j] = diag_ext[i].name;
2287 			offset[j] = diag_ext[i].offset;
2288 		}
2289 	}
2290 
2291 	if (!port) {
2292 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2293 			name[j] = diag_device_only[i].name;
2294 			offset[j] = diag_device_only[i].offset;
2295 		}
2296 	}
2297 }
2298 
2299 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2300 {
2301 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2302 	int i;
2303 	int ret;
2304 	bool per_port = !!(ibdev->dev->caps.flags2 &
2305 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2306 
2307 	if (mlx4_is_slave(ibdev->dev))
2308 		return 0;
2309 
2310 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2311 		/* i == 1 means we are building port counters */
2312 		if (i && !per_port)
2313 			continue;
2314 
2315 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2316 						    &diag[i].offset,
2317 						    &diag[i].num_counters, i);
2318 		if (ret)
2319 			goto err_alloc;
2320 
2321 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2322 					   diag[i].offset, i);
2323 	}
2324 
2325 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2326 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2327 
2328 	return 0;
2329 
2330 err_alloc:
2331 	if (i) {
2332 		kfree(diag[i - 1].name);
2333 		kfree(diag[i - 1].offset);
2334 	}
2335 
2336 	return ret;
2337 }
2338 
2339 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2340 {
2341 	int i;
2342 
2343 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2344 		kfree(ibdev->diag_counters[i].offset);
2345 		kfree(ibdev->diag_counters[i].name);
2346 	}
2347 }
2348 
2349 #define MLX4_IB_INVALID_MAC	((u64)-1)
2350 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2351 			       struct net_device *dev,
2352 			       int port)
2353 {
2354 	u64 new_smac = 0;
2355 	u64 release_mac = MLX4_IB_INVALID_MAC;
2356 	struct mlx4_ib_qp *qp;
2357 
2358 	read_lock(&dev_base_lock);
2359 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2360 	read_unlock(&dev_base_lock);
2361 
2362 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2363 
2364 	/* no need for update QP1 and mac registration in non-SRIOV */
2365 	if (!mlx4_is_mfunc(ibdev->dev))
2366 		return;
2367 
2368 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2369 	qp = ibdev->qp1_proxy[port - 1];
2370 	if (qp) {
2371 		int new_smac_index;
2372 		u64 old_smac;
2373 		struct mlx4_update_qp_params update_params;
2374 
2375 		mutex_lock(&qp->mutex);
2376 		old_smac = qp->pri.smac;
2377 		if (new_smac == old_smac)
2378 			goto unlock;
2379 
2380 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2381 
2382 		if (new_smac_index < 0)
2383 			goto unlock;
2384 
2385 		update_params.smac_index = new_smac_index;
2386 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2387 				   &update_params)) {
2388 			release_mac = new_smac;
2389 			goto unlock;
2390 		}
2391 		/* if old port was zero, no mac was yet registered for this QP */
2392 		if (qp->pri.smac_port)
2393 			release_mac = old_smac;
2394 		qp->pri.smac = new_smac;
2395 		qp->pri.smac_port = port;
2396 		qp->pri.smac_index = new_smac_index;
2397 	}
2398 
2399 unlock:
2400 	if (release_mac != MLX4_IB_INVALID_MAC)
2401 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2402 	if (qp)
2403 		mutex_unlock(&qp->mutex);
2404 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2405 }
2406 
2407 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2408 				 struct net_device *dev,
2409 				 unsigned long event)
2410 
2411 {
2412 	struct mlx4_ib_iboe *iboe;
2413 	int update_qps_port = -1;
2414 	int port;
2415 
2416 	ASSERT_RTNL();
2417 
2418 	iboe = &ibdev->iboe;
2419 
2420 	spin_lock_bh(&iboe->lock);
2421 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2422 
2423 		iboe->netdevs[port - 1] =
2424 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2425 
2426 		if (dev == iboe->netdevs[port - 1] &&
2427 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2428 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2429 			update_qps_port = port;
2430 
2431 	}
2432 	spin_unlock_bh(&iboe->lock);
2433 
2434 	if (update_qps_port > 0)
2435 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2436 }
2437 
2438 static int mlx4_ib_netdev_event(struct notifier_block *this,
2439 				unsigned long event, void *ptr)
2440 {
2441 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2442 	struct mlx4_ib_dev *ibdev;
2443 
2444 	if (!net_eq(dev_net(dev), &init_net))
2445 		return NOTIFY_DONE;
2446 
2447 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2448 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2449 
2450 	return NOTIFY_DONE;
2451 }
2452 
2453 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2454 {
2455 	int port;
2456 	int slave;
2457 	int i;
2458 
2459 	if (mlx4_is_master(ibdev->dev)) {
2460 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2461 		     ++slave) {
2462 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2463 				for (i = 0;
2464 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2465 				     ++i) {
2466 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2467 					/* master has the identity virt2phys pkey mapping */
2468 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2469 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2470 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2471 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2472 				}
2473 			}
2474 		}
2475 		/* initialize pkey cache */
2476 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2477 			for (i = 0;
2478 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2479 			     ++i)
2480 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2481 					(i) ? 0 : 0xFFFF;
2482 		}
2483 	}
2484 }
2485 
2486 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2487 {
2488 	int i, j, eq = 0, total_eqs = 0;
2489 
2490 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2491 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2492 	if (!ibdev->eq_table)
2493 		return;
2494 
2495 	for (i = 1; i <= dev->caps.num_ports; i++) {
2496 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2497 		     j++, total_eqs++) {
2498 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2499 				continue;
2500 			ibdev->eq_table[eq] = total_eqs;
2501 			if (!mlx4_assign_eq(dev, i,
2502 					    &ibdev->eq_table[eq]))
2503 				eq++;
2504 			else
2505 				ibdev->eq_table[eq] = -1;
2506 		}
2507 	}
2508 
2509 	for (i = eq; i < dev->caps.num_comp_vectors;
2510 	     ibdev->eq_table[i++] = -1)
2511 		;
2512 
2513 	/* Advertise the new number of EQs to clients */
2514 	ibdev->ib_dev.num_comp_vectors = eq;
2515 }
2516 
2517 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2518 {
2519 	int i;
2520 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2521 
2522 	/* no eqs were allocated */
2523 	if (!ibdev->eq_table)
2524 		return;
2525 
2526 	/* Reset the advertised EQ number */
2527 	ibdev->ib_dev.num_comp_vectors = 0;
2528 
2529 	for (i = 0; i < total_eqs; i++)
2530 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2531 
2532 	kfree(ibdev->eq_table);
2533 	ibdev->eq_table = NULL;
2534 }
2535 
2536 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2537 			       struct ib_port_immutable *immutable)
2538 {
2539 	struct ib_port_attr attr;
2540 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2541 	int err;
2542 
2543 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2544 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2545 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2546 	} else {
2547 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2548 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2549 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2550 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2551 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2552 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2553 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2554 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2555 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2556 	}
2557 
2558 	err = ib_query_port(ibdev, port_num, &attr);
2559 	if (err)
2560 		return err;
2561 
2562 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2563 	immutable->gid_tbl_len = attr.gid_tbl_len;
2564 
2565 	return 0;
2566 }
2567 
2568 static void get_fw_ver_str(struct ib_device *device, char *str,
2569 			   size_t str_len)
2570 {
2571 	struct mlx4_ib_dev *dev =
2572 		container_of(device, struct mlx4_ib_dev, ib_dev);
2573 	snprintf(str, str_len, "%d.%d.%d",
2574 		 (int) (dev->dev->caps.fw_ver >> 32),
2575 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2576 		 (int) dev->dev->caps.fw_ver & 0xffff);
2577 }
2578 
2579 static void *mlx4_ib_add(struct mlx4_dev *dev)
2580 {
2581 	struct mlx4_ib_dev *ibdev;
2582 	int num_ports = 0;
2583 	int i, j;
2584 	int err;
2585 	struct mlx4_ib_iboe *iboe;
2586 	int ib_num_ports = 0;
2587 	int num_req_counters;
2588 	int allocated;
2589 	u32 counter_index;
2590 	struct counter_index *new_counter_index = NULL;
2591 
2592 	pr_info_once("%s", mlx4_ib_version);
2593 
2594 	num_ports = 0;
2595 	mlx4_foreach_ib_transport_port(i, dev)
2596 		num_ports++;
2597 
2598 	/* No point in registering a device with no ports... */
2599 	if (num_ports == 0)
2600 		return NULL;
2601 
2602 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2603 	if (!ibdev) {
2604 		dev_err(&dev->persist->pdev->dev,
2605 			"Device struct alloc failed\n");
2606 		return NULL;
2607 	}
2608 
2609 	iboe = &ibdev->iboe;
2610 
2611 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2612 		goto err_dealloc;
2613 
2614 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2615 		goto err_pd;
2616 
2617 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2618 				 PAGE_SIZE);
2619 	if (!ibdev->uar_map)
2620 		goto err_uar;
2621 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2622 
2623 	ibdev->dev = dev;
2624 	ibdev->bond_next_port	= 0;
2625 
2626 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2627 	ibdev->ib_dev.owner		= THIS_MODULE;
2628 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2629 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2630 	ibdev->num_ports		= num_ports;
2631 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2632 						1 : ibdev->num_ports;
2633 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2634 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2635 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2636 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2637 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2638 
2639 	if (dev->caps.userspace_caps)
2640 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2641 	else
2642 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2643 
2644 	ibdev->ib_dev.uverbs_cmd_mask	=
2645 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2646 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2647 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2648 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2649 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2650 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2651 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2652 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2653 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2654 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2655 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2656 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2657 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2658 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2659 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2660 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2661 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2662 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2663 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2664 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2665 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2666 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2667 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2668 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2669 
2670 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2671 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2672 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2673 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2674 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2675 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2676 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2677 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2678 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2679 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2680 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2681 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2682 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2683 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2684 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2685 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2686 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2687 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2688 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2689 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2690 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2691 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2692 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2693 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2694 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2695 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2696 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2697 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2698 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2699 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2700 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2701 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2702 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2703 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2704 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2705 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2706 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2707 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2708 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2709 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2710 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2711 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2712 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2713 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2714 
2715 	if (!mlx4_is_slave(ibdev->dev)) {
2716 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2717 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2718 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2719 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2720 	}
2721 
2722 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2723 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2724 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2725 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2726 
2727 		ibdev->ib_dev.uverbs_cmd_mask |=
2728 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2729 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2730 	}
2731 
2732 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2733 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2734 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2735 		ibdev->ib_dev.uverbs_cmd_mask |=
2736 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2737 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2738 	}
2739 
2740 	if (check_flow_steering_support(dev)) {
2741 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2742 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2743 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2744 
2745 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2746 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2747 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2748 	}
2749 
2750 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2751 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2752 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2753 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2754 
2755 	mlx4_ib_alloc_eqs(dev, ibdev);
2756 
2757 	spin_lock_init(&iboe->lock);
2758 
2759 	if (init_node_data(ibdev))
2760 		goto err_map;
2761 	mlx4_init_sl2vl_tbl(ibdev);
2762 
2763 	for (i = 0; i < ibdev->num_ports; ++i) {
2764 		mutex_init(&ibdev->counters_table[i].mutex);
2765 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2766 	}
2767 
2768 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2769 	for (i = 0; i < num_req_counters; ++i) {
2770 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2771 		allocated = 0;
2772 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2773 						IB_LINK_LAYER_ETHERNET) {
2774 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2775 			/* if failed to allocate a new counter, use default */
2776 			if (err)
2777 				counter_index =
2778 					mlx4_get_default_counter_index(dev,
2779 								       i + 1);
2780 			else
2781 				allocated = 1;
2782 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2783 			counter_index = mlx4_get_default_counter_index(dev,
2784 								       i + 1);
2785 		}
2786 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2787 					    GFP_KERNEL);
2788 		if (!new_counter_index) {
2789 			if (allocated)
2790 				mlx4_counter_free(ibdev->dev, counter_index);
2791 			goto err_counter;
2792 		}
2793 		new_counter_index->index = counter_index;
2794 		new_counter_index->allocated = allocated;
2795 		list_add_tail(&new_counter_index->list,
2796 			      &ibdev->counters_table[i].counters_list);
2797 		ibdev->counters_table[i].default_counter = counter_index;
2798 		pr_info("counter index %d for port %d allocated %d\n",
2799 			counter_index, i + 1, allocated);
2800 	}
2801 	if (mlx4_is_bonded(dev))
2802 		for (i = 1; i < ibdev->num_ports ; ++i) {
2803 			new_counter_index =
2804 					kmalloc(sizeof(struct counter_index),
2805 						GFP_KERNEL);
2806 			if (!new_counter_index)
2807 				goto err_counter;
2808 			new_counter_index->index = counter_index;
2809 			new_counter_index->allocated = 0;
2810 			list_add_tail(&new_counter_index->list,
2811 				      &ibdev->counters_table[i].counters_list);
2812 			ibdev->counters_table[i].default_counter =
2813 								counter_index;
2814 		}
2815 
2816 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2817 		ib_num_ports++;
2818 
2819 	spin_lock_init(&ibdev->sm_lock);
2820 	mutex_init(&ibdev->cap_mask_mutex);
2821 	INIT_LIST_HEAD(&ibdev->qp_list);
2822 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2823 
2824 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2825 	    ib_num_ports) {
2826 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2827 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2828 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2829 					    &ibdev->steer_qpn_base, 0);
2830 		if (err)
2831 			goto err_counter;
2832 
2833 		ibdev->ib_uc_qpns_bitmap =
2834 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2835 				sizeof(long),
2836 				GFP_KERNEL);
2837 		if (!ibdev->ib_uc_qpns_bitmap)
2838 			goto err_steer_qp_release;
2839 
2840 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2841 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2842 				    ibdev->steer_qpn_count);
2843 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2844 					dev, ibdev->steer_qpn_base,
2845 					ibdev->steer_qpn_base +
2846 					ibdev->steer_qpn_count - 1);
2847 			if (err)
2848 				goto err_steer_free_bitmap;
2849 		} else {
2850 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2851 				    ibdev->steer_qpn_count);
2852 		}
2853 	}
2854 
2855 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2856 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2857 
2858 	if (mlx4_ib_alloc_diag_counters(ibdev))
2859 		goto err_steer_free_bitmap;
2860 
2861 	if (ib_register_device(&ibdev->ib_dev, NULL))
2862 		goto err_diag_counters;
2863 
2864 	if (mlx4_ib_mad_init(ibdev))
2865 		goto err_reg;
2866 
2867 	if (mlx4_ib_init_sriov(ibdev))
2868 		goto err_mad;
2869 
2870 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2871 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2872 		if (!iboe->nb.notifier_call) {
2873 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2874 			err = register_netdevice_notifier(&iboe->nb);
2875 			if (err) {
2876 				iboe->nb.notifier_call = NULL;
2877 				goto err_notif;
2878 			}
2879 		}
2880 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2881 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2882 			if (err) {
2883 				goto err_notif;
2884 			}
2885 		}
2886 	}
2887 
2888 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2889 		if (device_create_file(&ibdev->ib_dev.dev,
2890 				       mlx4_class_attributes[j]))
2891 			goto err_notif;
2892 	}
2893 
2894 	ibdev->ib_active = true;
2895 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2896 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2897 					 &ibdev->ib_dev);
2898 
2899 	if (mlx4_is_mfunc(ibdev->dev))
2900 		init_pkeys(ibdev);
2901 
2902 	/* create paravirt contexts for any VFs which are active */
2903 	if (mlx4_is_master(ibdev->dev)) {
2904 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2905 			if (j == mlx4_master_func_num(ibdev->dev))
2906 				continue;
2907 			if (mlx4_is_slave_active(ibdev->dev, j))
2908 				do_slave_init(ibdev, j, 1);
2909 		}
2910 	}
2911 	return ibdev;
2912 
2913 err_notif:
2914 	if (ibdev->iboe.nb.notifier_call) {
2915 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2916 			pr_warn("failure unregistering notifier\n");
2917 		ibdev->iboe.nb.notifier_call = NULL;
2918 	}
2919 	flush_workqueue(wq);
2920 
2921 	mlx4_ib_close_sriov(ibdev);
2922 
2923 err_mad:
2924 	mlx4_ib_mad_cleanup(ibdev);
2925 
2926 err_reg:
2927 	ib_unregister_device(&ibdev->ib_dev);
2928 
2929 err_diag_counters:
2930 	mlx4_ib_diag_cleanup(ibdev);
2931 
2932 err_steer_free_bitmap:
2933 	kfree(ibdev->ib_uc_qpns_bitmap);
2934 
2935 err_steer_qp_release:
2936 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2937 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2938 				      ibdev->steer_qpn_count);
2939 err_counter:
2940 	for (i = 0; i < ibdev->num_ports; ++i)
2941 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2942 
2943 err_map:
2944 	iounmap(ibdev->uar_map);
2945 
2946 err_uar:
2947 	mlx4_uar_free(dev, &ibdev->priv_uar);
2948 
2949 err_pd:
2950 	mlx4_pd_free(dev, ibdev->priv_pdn);
2951 
2952 err_dealloc:
2953 	ib_dealloc_device(&ibdev->ib_dev);
2954 
2955 	return NULL;
2956 }
2957 
2958 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2959 {
2960 	int offset;
2961 
2962 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2963 
2964 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2965 					 dev->steer_qpn_count,
2966 					 get_count_order(count));
2967 	if (offset < 0)
2968 		return offset;
2969 
2970 	*qpn = dev->steer_qpn_base + offset;
2971 	return 0;
2972 }
2973 
2974 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2975 {
2976 	if (!qpn ||
2977 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2978 		return;
2979 
2980 	BUG_ON(qpn < dev->steer_qpn_base);
2981 
2982 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2983 			      qpn - dev->steer_qpn_base,
2984 			      get_count_order(count));
2985 }
2986 
2987 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2988 			 int is_attach)
2989 {
2990 	int err;
2991 	size_t flow_size;
2992 	struct ib_flow_attr *flow = NULL;
2993 	struct ib_flow_spec_ib *ib_spec;
2994 
2995 	if (is_attach) {
2996 		flow_size = sizeof(struct ib_flow_attr) +
2997 			    sizeof(struct ib_flow_spec_ib);
2998 		flow = kzalloc(flow_size, GFP_KERNEL);
2999 		if (!flow)
3000 			return -ENOMEM;
3001 		flow->port = mqp->port;
3002 		flow->num_of_specs = 1;
3003 		flow->size = flow_size;
3004 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3005 		ib_spec->type = IB_FLOW_SPEC_IB;
3006 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3007 		/* Add an empty rule for IB L2 */
3008 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3009 
3010 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3011 					    IB_FLOW_DOMAIN_NIC,
3012 					    MLX4_FS_REGULAR,
3013 					    &mqp->reg_id);
3014 	} else {
3015 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3016 	}
3017 	kfree(flow);
3018 	return err;
3019 }
3020 
3021 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3022 {
3023 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3024 	int p;
3025 	int i;
3026 
3027 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3028 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3029 	ibdev->ib_active = false;
3030 	flush_workqueue(wq);
3031 
3032 	mlx4_ib_close_sriov(ibdev);
3033 	mlx4_ib_mad_cleanup(ibdev);
3034 	ib_unregister_device(&ibdev->ib_dev);
3035 	mlx4_ib_diag_cleanup(ibdev);
3036 	if (ibdev->iboe.nb.notifier_call) {
3037 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3038 			pr_warn("failure unregistering notifier\n");
3039 		ibdev->iboe.nb.notifier_call = NULL;
3040 	}
3041 
3042 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
3043 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3044 				      ibdev->steer_qpn_count);
3045 		kfree(ibdev->ib_uc_qpns_bitmap);
3046 	}
3047 
3048 	iounmap(ibdev->uar_map);
3049 	for (p = 0; p < ibdev->num_ports; ++p)
3050 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3051 
3052 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3053 		mlx4_CLOSE_PORT(dev, p);
3054 
3055 	mlx4_ib_free_eqs(dev, ibdev);
3056 
3057 	mlx4_uar_free(dev, &ibdev->priv_uar);
3058 	mlx4_pd_free(dev, ibdev->priv_pdn);
3059 	ib_dealloc_device(&ibdev->ib_dev);
3060 }
3061 
3062 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3063 {
3064 	struct mlx4_ib_demux_work **dm = NULL;
3065 	struct mlx4_dev *dev = ibdev->dev;
3066 	int i;
3067 	unsigned long flags;
3068 	struct mlx4_active_ports actv_ports;
3069 	unsigned int ports;
3070 	unsigned int first_port;
3071 
3072 	if (!mlx4_is_master(dev))
3073 		return;
3074 
3075 	actv_ports = mlx4_get_active_ports(dev, slave);
3076 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3077 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3078 
3079 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3080 	if (!dm)
3081 		return;
3082 
3083 	for (i = 0; i < ports; i++) {
3084 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3085 		if (!dm[i]) {
3086 			while (--i >= 0)
3087 				kfree(dm[i]);
3088 			goto out;
3089 		}
3090 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3091 		dm[i]->port = first_port + i + 1;
3092 		dm[i]->slave = slave;
3093 		dm[i]->do_init = do_init;
3094 		dm[i]->dev = ibdev;
3095 	}
3096 	/* initialize or tear down tunnel QPs for the slave */
3097 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3098 	if (!ibdev->sriov.is_going_down) {
3099 		for (i = 0; i < ports; i++)
3100 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3101 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3102 	} else {
3103 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3104 		for (i = 0; i < ports; i++)
3105 			kfree(dm[i]);
3106 	}
3107 out:
3108 	kfree(dm);
3109 	return;
3110 }
3111 
3112 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3113 {
3114 	struct mlx4_ib_qp *mqp;
3115 	unsigned long flags_qp;
3116 	unsigned long flags_cq;
3117 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3118 	struct list_head    cq_notify_list;
3119 	struct mlx4_cq *mcq;
3120 	unsigned long flags;
3121 
3122 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3123 	INIT_LIST_HEAD(&cq_notify_list);
3124 
3125 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3126 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3127 
3128 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3129 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3130 		if (mqp->sq.tail != mqp->sq.head) {
3131 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3132 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3133 			if (send_mcq->mcq.comp &&
3134 			    mqp->ibqp.send_cq->comp_handler) {
3135 				if (!send_mcq->mcq.reset_notify_added) {
3136 					send_mcq->mcq.reset_notify_added = 1;
3137 					list_add_tail(&send_mcq->mcq.reset_notify,
3138 						      &cq_notify_list);
3139 				}
3140 			}
3141 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3142 		}
3143 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3144 		/* Now, handle the QP's receive queue */
3145 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3146 		/* no handling is needed for SRQ */
3147 		if (!mqp->ibqp.srq) {
3148 			if (mqp->rq.tail != mqp->rq.head) {
3149 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3150 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3151 				if (recv_mcq->mcq.comp &&
3152 				    mqp->ibqp.recv_cq->comp_handler) {
3153 					if (!recv_mcq->mcq.reset_notify_added) {
3154 						recv_mcq->mcq.reset_notify_added = 1;
3155 						list_add_tail(&recv_mcq->mcq.reset_notify,
3156 							      &cq_notify_list);
3157 					}
3158 				}
3159 				spin_unlock_irqrestore(&recv_mcq->lock,
3160 						       flags_cq);
3161 			}
3162 		}
3163 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3164 	}
3165 
3166 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3167 		mcq->comp(mcq);
3168 	}
3169 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3170 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3171 }
3172 
3173 static void handle_bonded_port_state_event(struct work_struct *work)
3174 {
3175 	struct ib_event_work *ew =
3176 		container_of(work, struct ib_event_work, work);
3177 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3178 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3179 	int i;
3180 	struct ib_event ibev;
3181 
3182 	kfree(ew);
3183 	spin_lock_bh(&ibdev->iboe.lock);
3184 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3185 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3186 		enum ib_port_state curr_port_state;
3187 
3188 		if (!curr_netdev)
3189 			continue;
3190 
3191 		curr_port_state =
3192 			(netif_running(curr_netdev) &&
3193 			 netif_carrier_ok(curr_netdev)) ?
3194 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3195 
3196 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3197 			curr_port_state : IB_PORT_ACTIVE;
3198 	}
3199 	spin_unlock_bh(&ibdev->iboe.lock);
3200 
3201 	ibev.device = &ibdev->ib_dev;
3202 	ibev.element.port_num = 1;
3203 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3204 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3205 
3206 	ib_dispatch_event(&ibev);
3207 }
3208 
3209 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3210 {
3211 	u64 sl2vl;
3212 	int err;
3213 
3214 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3215 	if (err) {
3216 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3217 		       port, err);
3218 		sl2vl = 0;
3219 	}
3220 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3221 }
3222 
3223 static void ib_sl2vl_update_work(struct work_struct *work)
3224 {
3225 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3226 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3227 	int port = ew->port;
3228 
3229 	mlx4_ib_sl2vl_update(mdev, port);
3230 
3231 	kfree(ew);
3232 }
3233 
3234 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3235 				     int port)
3236 {
3237 	struct ib_event_work *ew;
3238 
3239 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3240 	if (ew) {
3241 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3242 		ew->port = port;
3243 		ew->ib_dev = ibdev;
3244 		queue_work(wq, &ew->work);
3245 	}
3246 }
3247 
3248 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3249 			  enum mlx4_dev_event event, unsigned long param)
3250 {
3251 	struct ib_event ibev;
3252 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3253 	struct mlx4_eqe *eqe = NULL;
3254 	struct ib_event_work *ew;
3255 	int p = 0;
3256 
3257 	if (mlx4_is_bonded(dev) &&
3258 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3259 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3260 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3261 		if (!ew)
3262 			return;
3263 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3264 		ew->ib_dev = ibdev;
3265 		queue_work(wq, &ew->work);
3266 		return;
3267 	}
3268 
3269 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3270 		eqe = (struct mlx4_eqe *)param;
3271 	else
3272 		p = (int) param;
3273 
3274 	switch (event) {
3275 	case MLX4_DEV_EVENT_PORT_UP:
3276 		if (p > ibdev->num_ports)
3277 			return;
3278 		if (!mlx4_is_slave(dev) &&
3279 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3280 			IB_LINK_LAYER_INFINIBAND) {
3281 			if (mlx4_is_master(dev))
3282 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3283 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3284 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3285 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3286 		}
3287 		ibev.event = IB_EVENT_PORT_ACTIVE;
3288 		break;
3289 
3290 	case MLX4_DEV_EVENT_PORT_DOWN:
3291 		if (p > ibdev->num_ports)
3292 			return;
3293 		ibev.event = IB_EVENT_PORT_ERR;
3294 		break;
3295 
3296 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3297 		ibdev->ib_active = false;
3298 		ibev.event = IB_EVENT_DEVICE_FATAL;
3299 		mlx4_ib_handle_catas_error(ibdev);
3300 		break;
3301 
3302 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3303 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3304 		if (!ew)
3305 			break;
3306 
3307 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3308 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3309 		ew->ib_dev = ibdev;
3310 		/* need to queue only for port owner, which uses GEN_EQE */
3311 		if (mlx4_is_master(dev))
3312 			queue_work(wq, &ew->work);
3313 		else
3314 			handle_port_mgmt_change_event(&ew->work);
3315 		return;
3316 
3317 	case MLX4_DEV_EVENT_SLAVE_INIT:
3318 		/* here, p is the slave id */
3319 		do_slave_init(ibdev, p, 1);
3320 		if (mlx4_is_master(dev)) {
3321 			int i;
3322 
3323 			for (i = 1; i <= ibdev->num_ports; i++) {
3324 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3325 					== IB_LINK_LAYER_INFINIBAND)
3326 					mlx4_ib_slave_alias_guid_event(ibdev,
3327 								       p, i,
3328 								       1);
3329 			}
3330 		}
3331 		return;
3332 
3333 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3334 		if (mlx4_is_master(dev)) {
3335 			int i;
3336 
3337 			for (i = 1; i <= ibdev->num_ports; i++) {
3338 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3339 					== IB_LINK_LAYER_INFINIBAND)
3340 					mlx4_ib_slave_alias_guid_event(ibdev,
3341 								       p, i,
3342 								       0);
3343 			}
3344 		}
3345 		/* here, p is the slave id */
3346 		do_slave_init(ibdev, p, 0);
3347 		return;
3348 
3349 	default:
3350 		return;
3351 	}
3352 
3353 	ibev.device	      = ibdev_ptr;
3354 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3355 
3356 	ib_dispatch_event(&ibev);
3357 }
3358 
3359 static struct mlx4_interface mlx4_ib_interface = {
3360 	.add		= mlx4_ib_add,
3361 	.remove		= mlx4_ib_remove,
3362 	.event		= mlx4_ib_event,
3363 	.protocol	= MLX4_PROT_IB_IPV6,
3364 	.flags		= MLX4_INTFF_BONDING
3365 };
3366 
3367 static int __init mlx4_ib_init(void)
3368 {
3369 	int err;
3370 
3371 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3372 	if (!wq)
3373 		return -ENOMEM;
3374 
3375 	err = mlx4_ib_mcg_init();
3376 	if (err)
3377 		goto clean_wq;
3378 
3379 	err = mlx4_register_interface(&mlx4_ib_interface);
3380 	if (err)
3381 		goto clean_mcg;
3382 
3383 	return 0;
3384 
3385 clean_mcg:
3386 	mlx4_ib_mcg_destroy();
3387 
3388 clean_wq:
3389 	destroy_workqueue(wq);
3390 	return err;
3391 }
3392 
3393 static void __exit mlx4_ib_cleanup(void)
3394 {
3395 	mlx4_unregister_interface(&mlx4_ib_interface);
3396 	mlx4_ib_mcg_destroy();
3397 	destroy_workqueue(wq);
3398 }
3399 
3400 module_init(mlx4_ib_init);
3401 module_exit(mlx4_ib_cleanup);
3402