xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 2359ccdd)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static int mlx4_ib_add_gid(const union ib_gid *gid,
250 			   const struct ib_gid_attr *attr,
251 			   void **context)
252 {
253 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
254 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
255 	struct mlx4_port_gid_table   *port_gid_table;
256 	int free = -1, found = -1;
257 	int ret = 0;
258 	int hw_update = 0;
259 	int i;
260 	struct gid_entry *gids = NULL;
261 
262 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
263 		return -EINVAL;
264 
265 	if (attr->port_num > MLX4_MAX_PORTS)
266 		return -EINVAL;
267 
268 	if (!context)
269 		return -EINVAL;
270 
271 	port_gid_table = &iboe->gids[attr->port_num - 1];
272 	spin_lock_bh(&iboe->lock);
273 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
274 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
275 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
276 			found = i;
277 			break;
278 		}
279 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
280 			free = i; /* HW has space */
281 	}
282 
283 	if (found < 0) {
284 		if (free < 0) {
285 			ret = -ENOSPC;
286 		} else {
287 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
288 			if (!port_gid_table->gids[free].ctx) {
289 				ret = -ENOMEM;
290 			} else {
291 				*context = port_gid_table->gids[free].ctx;
292 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
293 				port_gid_table->gids[free].gid_type = attr->gid_type;
294 				port_gid_table->gids[free].ctx->real_index = free;
295 				port_gid_table->gids[free].ctx->refcount = 1;
296 				hw_update = 1;
297 			}
298 		}
299 	} else {
300 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
301 		*context = ctx;
302 		ctx->refcount++;
303 	}
304 	if (!ret && hw_update) {
305 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
306 		if (!gids) {
307 			ret = -ENOMEM;
308 		} else {
309 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
310 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
311 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
312 			}
313 		}
314 	}
315 	spin_unlock_bh(&iboe->lock);
316 
317 	if (!ret && hw_update) {
318 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
319 		kfree(gids);
320 	}
321 
322 	return ret;
323 }
324 
325 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
326 {
327 	struct gid_cache_context *ctx = *context;
328 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
329 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
330 	struct mlx4_port_gid_table   *port_gid_table;
331 	int ret = 0;
332 	int hw_update = 0;
333 	struct gid_entry *gids = NULL;
334 
335 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
336 		return -EINVAL;
337 
338 	if (attr->port_num > MLX4_MAX_PORTS)
339 		return -EINVAL;
340 
341 	port_gid_table = &iboe->gids[attr->port_num - 1];
342 	spin_lock_bh(&iboe->lock);
343 	if (ctx) {
344 		ctx->refcount--;
345 		if (!ctx->refcount) {
346 			unsigned int real_index = ctx->real_index;
347 
348 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
349 			kfree(port_gid_table->gids[real_index].ctx);
350 			port_gid_table->gids[real_index].ctx = NULL;
351 			hw_update = 1;
352 		}
353 	}
354 	if (!ret && hw_update) {
355 		int i;
356 
357 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
358 		if (!gids) {
359 			ret = -ENOMEM;
360 		} else {
361 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
362 				memcpy(&gids[i].gid,
363 				       &port_gid_table->gids[i].gid,
364 				       sizeof(union ib_gid));
365 				gids[i].gid_type =
366 				    port_gid_table->gids[i].gid_type;
367 			}
368 		}
369 	}
370 	spin_unlock_bh(&iboe->lock);
371 
372 	if (!ret && hw_update) {
373 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
374 		kfree(gids);
375 	}
376 	return ret;
377 }
378 
379 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
380 				    u8 port_num, int index)
381 {
382 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
383 	struct gid_cache_context *ctx = NULL;
384 	union ib_gid gid;
385 	struct mlx4_port_gid_table   *port_gid_table;
386 	int real_index = -EINVAL;
387 	int i;
388 	int ret;
389 	unsigned long flags;
390 	struct ib_gid_attr attr;
391 
392 	if (port_num > MLX4_MAX_PORTS)
393 		return -EINVAL;
394 
395 	if (mlx4_is_bonded(ibdev->dev))
396 		port_num = 1;
397 
398 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
399 		return index;
400 
401 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
402 	if (ret)
403 		return ret;
404 
405 	if (attr.ndev)
406 		dev_put(attr.ndev);
407 
408 	spin_lock_irqsave(&iboe->lock, flags);
409 	port_gid_table = &iboe->gids[port_num - 1];
410 
411 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
412 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
413 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
414 			ctx = port_gid_table->gids[i].ctx;
415 			break;
416 		}
417 	if (ctx)
418 		real_index = ctx->real_index;
419 	spin_unlock_irqrestore(&iboe->lock, flags);
420 	return real_index;
421 }
422 
423 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
424 				    sizeof(((type *)0)->fld) <= (sz))
425 
426 static int mlx4_ib_query_device(struct ib_device *ibdev,
427 				struct ib_device_attr *props,
428 				struct ib_udata *uhw)
429 {
430 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
431 	struct ib_smp *in_mad  = NULL;
432 	struct ib_smp *out_mad = NULL;
433 	int err;
434 	int have_ib_ports;
435 	struct mlx4_uverbs_ex_query_device cmd;
436 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
437 	struct mlx4_clock_params clock_params;
438 
439 	if (uhw->inlen) {
440 		if (uhw->inlen < sizeof(cmd))
441 			return -EINVAL;
442 
443 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
444 		if (err)
445 			return err;
446 
447 		if (cmd.comp_mask)
448 			return -EINVAL;
449 
450 		if (cmd.reserved)
451 			return -EINVAL;
452 	}
453 
454 	resp.response_length = offsetof(typeof(resp), response_length) +
455 		sizeof(resp.response_length);
456 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
457 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
458 	err = -ENOMEM;
459 	if (!in_mad || !out_mad)
460 		goto out;
461 
462 	init_query_mad(in_mad);
463 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
464 
465 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
466 			   1, NULL, NULL, in_mad, out_mad);
467 	if (err)
468 		goto out;
469 
470 	memset(props, 0, sizeof *props);
471 
472 	have_ib_ports = num_ib_ports(dev->dev);
473 
474 	props->fw_ver = dev->dev->caps.fw_ver;
475 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
476 		IB_DEVICE_PORT_ACTIVE_EVENT		|
477 		IB_DEVICE_SYS_IMAGE_GUID		|
478 		IB_DEVICE_RC_RNR_NAK_GEN		|
479 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
480 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
481 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
482 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
483 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
484 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
485 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
486 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
487 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
488 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
489 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
490 	if (dev->dev->caps.max_gso_sz &&
491 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
492 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
493 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
494 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
495 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
496 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
497 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
498 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
499 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
500 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
501 		props->device_cap_flags |= IB_DEVICE_XRC;
502 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
503 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
504 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
505 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
506 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
507 		else
508 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
509 	}
510 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
511 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
512 
513 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
514 
515 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
516 		0xffffff;
517 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
518 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
519 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
520 
521 	props->max_mr_size	   = ~0ull;
522 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
523 	props->max_qp		   = dev->dev->quotas.qp;
524 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
525 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
526 					 dev->dev->caps.max_rq_sg);
527 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
528 	props->max_cq		   = dev->dev->quotas.cq;
529 	props->max_cqe		   = dev->dev->caps.max_cqes;
530 	props->max_mr		   = dev->dev->quotas.mpt;
531 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
532 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
533 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
534 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
535 	props->max_srq		   = dev->dev->quotas.srq;
536 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
537 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
538 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
539 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
540 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
541 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
542 	props->masked_atomic_cap   = props->atomic_cap;
543 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
544 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
545 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
546 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
547 					   props->max_mcast_grp;
548 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
549 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
550 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
551 	props->max_ah = INT_MAX;
552 
553 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
554 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
555 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
556 			props->rss_caps.max_rwq_indirection_tables =
557 				props->max_qp;
558 			props->rss_caps.max_rwq_indirection_table_size =
559 				dev->dev->caps.max_rss_tbl_sz;
560 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
561 			props->max_wq_type_rq = props->max_qp;
562 		}
563 
564 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
565 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
566 	}
567 
568 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
569 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
570 
571 	if (!mlx4_is_slave(dev->dev))
572 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
573 
574 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
575 		resp.response_length += sizeof(resp.hca_core_clock_offset);
576 		if (!err && !mlx4_is_slave(dev->dev)) {
577 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
578 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
579 		}
580 	}
581 
582 	if (uhw->outlen >= resp.response_length +
583 	    sizeof(resp.max_inl_recv_sz)) {
584 		resp.response_length += sizeof(resp.max_inl_recv_sz);
585 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
586 			sizeof(struct mlx4_wqe_data_seg);
587 	}
588 
589 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
590 		if (props->rss_caps.supported_qpts) {
591 			resp.rss_caps.rx_hash_function =
592 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
593 
594 			resp.rss_caps.rx_hash_fields_mask =
595 				MLX4_IB_RX_HASH_SRC_IPV4 |
596 				MLX4_IB_RX_HASH_DST_IPV4 |
597 				MLX4_IB_RX_HASH_SRC_IPV6 |
598 				MLX4_IB_RX_HASH_DST_IPV6 |
599 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
600 				MLX4_IB_RX_HASH_DST_PORT_TCP |
601 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
602 				MLX4_IB_RX_HASH_DST_PORT_UDP;
603 
604 			if (dev->dev->caps.tunnel_offload_mode ==
605 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
606 				resp.rss_caps.rx_hash_fields_mask |=
607 					MLX4_IB_RX_HASH_INNER;
608 		}
609 		resp.response_length = offsetof(typeof(resp), rss_caps) +
610 				       sizeof(resp.rss_caps);
611 	}
612 
613 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
614 		if (dev->dev->caps.max_gso_sz &&
615 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
616 		    IB_LINK_LAYER_ETHERNET) ||
617 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
618 		    IB_LINK_LAYER_ETHERNET))) {
619 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
620 			resp.tso_caps.supported_qpts |=
621 				1 << IB_QPT_RAW_PACKET;
622 		}
623 		resp.response_length = offsetof(typeof(resp), tso_caps) +
624 				       sizeof(resp.tso_caps);
625 	}
626 
627 	if (uhw->outlen) {
628 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
629 		if (err)
630 			goto out;
631 	}
632 out:
633 	kfree(in_mad);
634 	kfree(out_mad);
635 
636 	return err;
637 }
638 
639 static enum rdma_link_layer
640 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
641 {
642 	struct mlx4_dev *dev = to_mdev(device)->dev;
643 
644 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
645 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
646 }
647 
648 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
649 			      struct ib_port_attr *props, int netw_view)
650 {
651 	struct ib_smp *in_mad  = NULL;
652 	struct ib_smp *out_mad = NULL;
653 	int ext_active_speed;
654 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
655 	int err = -ENOMEM;
656 
657 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
658 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
659 	if (!in_mad || !out_mad)
660 		goto out;
661 
662 	init_query_mad(in_mad);
663 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
664 	in_mad->attr_mod = cpu_to_be32(port);
665 
666 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
667 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
668 
669 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
670 				in_mad, out_mad);
671 	if (err)
672 		goto out;
673 
674 
675 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
676 	props->lmc		= out_mad->data[34] & 0x7;
677 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
678 	props->sm_sl		= out_mad->data[36] & 0xf;
679 	props->state		= out_mad->data[32] & 0xf;
680 	props->phys_state	= out_mad->data[33] >> 4;
681 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
682 	if (netw_view)
683 		props->gid_tbl_len = out_mad->data[50];
684 	else
685 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
686 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
687 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
688 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
689 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
690 	props->active_width	= out_mad->data[31] & 0xf;
691 	props->active_speed	= out_mad->data[35] >> 4;
692 	props->max_mtu		= out_mad->data[41] & 0xf;
693 	props->active_mtu	= out_mad->data[36] >> 4;
694 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
695 	props->max_vl_num	= out_mad->data[37] >> 4;
696 	props->init_type_reply	= out_mad->data[41] >> 4;
697 
698 	/* Check if extended speeds (EDR/FDR/...) are supported */
699 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
700 		ext_active_speed = out_mad->data[62] >> 4;
701 
702 		switch (ext_active_speed) {
703 		case 1:
704 			props->active_speed = IB_SPEED_FDR;
705 			break;
706 		case 2:
707 			props->active_speed = IB_SPEED_EDR;
708 			break;
709 		}
710 	}
711 
712 	/* If reported active speed is QDR, check if is FDR-10 */
713 	if (props->active_speed == IB_SPEED_QDR) {
714 		init_query_mad(in_mad);
715 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
716 		in_mad->attr_mod = cpu_to_be32(port);
717 
718 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
719 				   NULL, NULL, in_mad, out_mad);
720 		if (err)
721 			goto out;
722 
723 		/* Checking LinkSpeedActive for FDR-10 */
724 		if (out_mad->data[15] & 0x1)
725 			props->active_speed = IB_SPEED_FDR10;
726 	}
727 
728 	/* Avoid wrong speed value returned by FW if the IB link is down. */
729 	if (props->state == IB_PORT_DOWN)
730 		 props->active_speed = IB_SPEED_SDR;
731 
732 out:
733 	kfree(in_mad);
734 	kfree(out_mad);
735 	return err;
736 }
737 
738 static u8 state_to_phys_state(enum ib_port_state state)
739 {
740 	return state == IB_PORT_ACTIVE ? 5 : 3;
741 }
742 
743 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
744 			       struct ib_port_attr *props)
745 {
746 
747 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
748 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
749 	struct net_device *ndev;
750 	enum ib_mtu tmp;
751 	struct mlx4_cmd_mailbox *mailbox;
752 	int err = 0;
753 	int is_bonded = mlx4_is_bonded(mdev->dev);
754 
755 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
756 	if (IS_ERR(mailbox))
757 		return PTR_ERR(mailbox);
758 
759 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
760 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
761 			   MLX4_CMD_WRAPPED);
762 	if (err)
763 		goto out;
764 
765 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
766 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
767 					   IB_WIDTH_4X : IB_WIDTH_1X;
768 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
769 					   IB_SPEED_FDR : IB_SPEED_QDR;
770 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
771 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
772 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
773 	props->pkey_tbl_len	= 1;
774 	props->max_mtu		= IB_MTU_4096;
775 	props->max_vl_num	= 2;
776 	props->state		= IB_PORT_DOWN;
777 	props->phys_state	= state_to_phys_state(props->state);
778 	props->active_mtu	= IB_MTU_256;
779 	spin_lock_bh(&iboe->lock);
780 	ndev = iboe->netdevs[port - 1];
781 	if (ndev && is_bonded) {
782 		rcu_read_lock(); /* required to get upper dev */
783 		ndev = netdev_master_upper_dev_get_rcu(ndev);
784 		rcu_read_unlock();
785 	}
786 	if (!ndev)
787 		goto out_unlock;
788 
789 	tmp = iboe_get_mtu(ndev->mtu);
790 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
791 
792 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
793 					IB_PORT_ACTIVE : IB_PORT_DOWN;
794 	props->phys_state	= state_to_phys_state(props->state);
795 out_unlock:
796 	spin_unlock_bh(&iboe->lock);
797 out:
798 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
799 	return err;
800 }
801 
802 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
803 			 struct ib_port_attr *props, int netw_view)
804 {
805 	int err;
806 
807 	/* props being zeroed by the caller, avoid zeroing it here */
808 
809 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
810 		ib_link_query_port(ibdev, port, props, netw_view) :
811 				eth_link_query_port(ibdev, port, props);
812 
813 	return err;
814 }
815 
816 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
817 			      struct ib_port_attr *props)
818 {
819 	/* returns host view */
820 	return __mlx4_ib_query_port(ibdev, port, props, 0);
821 }
822 
823 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
824 			union ib_gid *gid, int netw_view)
825 {
826 	struct ib_smp *in_mad  = NULL;
827 	struct ib_smp *out_mad = NULL;
828 	int err = -ENOMEM;
829 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
830 	int clear = 0;
831 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
832 
833 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
834 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
835 	if (!in_mad || !out_mad)
836 		goto out;
837 
838 	init_query_mad(in_mad);
839 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
840 	in_mad->attr_mod = cpu_to_be32(port);
841 
842 	if (mlx4_is_mfunc(dev->dev) && netw_view)
843 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
844 
845 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
846 	if (err)
847 		goto out;
848 
849 	memcpy(gid->raw, out_mad->data + 8, 8);
850 
851 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
852 		if (index) {
853 			/* For any index > 0, return the null guid */
854 			err = 0;
855 			clear = 1;
856 			goto out;
857 		}
858 	}
859 
860 	init_query_mad(in_mad);
861 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
862 	in_mad->attr_mod = cpu_to_be32(index / 8);
863 
864 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
865 			   NULL, NULL, in_mad, out_mad);
866 	if (err)
867 		goto out;
868 
869 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
870 
871 out:
872 	if (clear)
873 		memset(gid->raw + 8, 0, 8);
874 	kfree(in_mad);
875 	kfree(out_mad);
876 	return err;
877 }
878 
879 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
880 			     union ib_gid *gid)
881 {
882 	if (rdma_protocol_ib(ibdev, port))
883 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
884 	return 0;
885 }
886 
887 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
888 {
889 	union sl2vl_tbl_to_u64 sl2vl64;
890 	struct ib_smp *in_mad  = NULL;
891 	struct ib_smp *out_mad = NULL;
892 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
893 	int err = -ENOMEM;
894 	int jj;
895 
896 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
897 		*sl2vl_tbl = 0;
898 		return 0;
899 	}
900 
901 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
902 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
903 	if (!in_mad || !out_mad)
904 		goto out;
905 
906 	init_query_mad(in_mad);
907 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
908 	in_mad->attr_mod = 0;
909 
910 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
911 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
912 
913 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
914 			   in_mad, out_mad);
915 	if (err)
916 		goto out;
917 
918 	for (jj = 0; jj < 8; jj++)
919 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
920 	*sl2vl_tbl = sl2vl64.sl64;
921 
922 out:
923 	kfree(in_mad);
924 	kfree(out_mad);
925 	return err;
926 }
927 
928 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
929 {
930 	u64 sl2vl;
931 	int i;
932 	int err;
933 
934 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
935 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
936 			continue;
937 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
938 		if (err) {
939 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
940 			       i, err);
941 			sl2vl = 0;
942 		}
943 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
944 	}
945 }
946 
947 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
948 			 u16 *pkey, int netw_view)
949 {
950 	struct ib_smp *in_mad  = NULL;
951 	struct ib_smp *out_mad = NULL;
952 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
953 	int err = -ENOMEM;
954 
955 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
956 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
957 	if (!in_mad || !out_mad)
958 		goto out;
959 
960 	init_query_mad(in_mad);
961 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
962 	in_mad->attr_mod = cpu_to_be32(index / 32);
963 
964 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
965 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
966 
967 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
968 			   in_mad, out_mad);
969 	if (err)
970 		goto out;
971 
972 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
973 
974 out:
975 	kfree(in_mad);
976 	kfree(out_mad);
977 	return err;
978 }
979 
980 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
981 {
982 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
983 }
984 
985 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
986 				 struct ib_device_modify *props)
987 {
988 	struct mlx4_cmd_mailbox *mailbox;
989 	unsigned long flags;
990 
991 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
992 		return -EOPNOTSUPP;
993 
994 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
995 		return 0;
996 
997 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
998 		return -EOPNOTSUPP;
999 
1000 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1001 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1002 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1003 
1004 	/*
1005 	 * If possible, pass node desc to FW, so it can generate
1006 	 * a 144 trap.  If cmd fails, just ignore.
1007 	 */
1008 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1009 	if (IS_ERR(mailbox))
1010 		return 0;
1011 
1012 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1013 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1014 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1015 
1016 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1017 
1018 	return 0;
1019 }
1020 
1021 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1022 			    u32 cap_mask)
1023 {
1024 	struct mlx4_cmd_mailbox *mailbox;
1025 	int err;
1026 
1027 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1028 	if (IS_ERR(mailbox))
1029 		return PTR_ERR(mailbox);
1030 
1031 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1032 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1033 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1034 	} else {
1035 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1036 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1037 	}
1038 
1039 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1040 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1041 		       MLX4_CMD_WRAPPED);
1042 
1043 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1044 	return err;
1045 }
1046 
1047 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1048 			       struct ib_port_modify *props)
1049 {
1050 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1051 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1052 	struct ib_port_attr attr;
1053 	u32 cap_mask;
1054 	int err;
1055 
1056 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1057 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1058 	 * violations and port capabilities are not meaningful.
1059 	 */
1060 	if (is_eth)
1061 		return 0;
1062 
1063 	mutex_lock(&mdev->cap_mask_mutex);
1064 
1065 	err = ib_query_port(ibdev, port, &attr);
1066 	if (err)
1067 		goto out;
1068 
1069 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1070 		~props->clr_port_cap_mask;
1071 
1072 	err = mlx4_ib_SET_PORT(mdev, port,
1073 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1074 			       cap_mask);
1075 
1076 out:
1077 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1078 	return err;
1079 }
1080 
1081 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1082 						  struct ib_udata *udata)
1083 {
1084 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1085 	struct mlx4_ib_ucontext *context;
1086 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1087 	struct mlx4_ib_alloc_ucontext_resp resp;
1088 	int err;
1089 
1090 	if (!dev->ib_active)
1091 		return ERR_PTR(-EAGAIN);
1092 
1093 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1094 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1095 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1096 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1097 	} else {
1098 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1099 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1100 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1101 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1102 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1103 	}
1104 
1105 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1106 	if (!context)
1107 		return ERR_PTR(-ENOMEM);
1108 
1109 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1110 	if (err) {
1111 		kfree(context);
1112 		return ERR_PTR(err);
1113 	}
1114 
1115 	INIT_LIST_HEAD(&context->db_page_list);
1116 	mutex_init(&context->db_page_mutex);
1117 
1118 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1119 	mutex_init(&context->wqn_ranges_mutex);
1120 
1121 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1122 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1123 	else
1124 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1125 
1126 	if (err) {
1127 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1128 		kfree(context);
1129 		return ERR_PTR(-EFAULT);
1130 	}
1131 
1132 	return &context->ibucontext;
1133 }
1134 
1135 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1136 {
1137 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1138 
1139 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1140 	kfree(context);
1141 
1142 	return 0;
1143 }
1144 
1145 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1146 {
1147 	/* vma_open is called when a new VMA is created on top of our VMA.
1148 	 * This is done through either mremap flow or split_vma (usually due
1149 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1150 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1151 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1152 	 * calling us again and trying to do incorrect actions. We assume that
1153 	 * the original vma size is exactly a single page that there will be no
1154 	 * "splitting" operations on.
1155 	 */
1156 	area->vm_ops = NULL;
1157 }
1158 
1159 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1160 {
1161 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1162 
1163 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1164 	 * file itself is closed, therefore no sync is needed with the regular
1165 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1166 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1167 	 * The close operation is usually called under mm->mmap_sem except when
1168 	 * process is exiting.  The exiting case is handled explicitly as part
1169 	 * of mlx4_ib_disassociate_ucontext.
1170 	 */
1171 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1172 				area->vm_private_data;
1173 
1174 	/* set the vma context pointer to null in the mlx4_ib driver's private
1175 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1176 	 */
1177 	mlx4_ib_vma_priv_data->vma = NULL;
1178 }
1179 
1180 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1181 	.open = mlx4_ib_vma_open,
1182 	.close = mlx4_ib_vma_close
1183 };
1184 
1185 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1186 {
1187 	int i;
1188 	int ret = 0;
1189 	struct vm_area_struct *vma;
1190 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1191 	struct task_struct *owning_process  = NULL;
1192 	struct mm_struct   *owning_mm       = NULL;
1193 
1194 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1195 	if (!owning_process)
1196 		return;
1197 
1198 	owning_mm = get_task_mm(owning_process);
1199 	if (!owning_mm) {
1200 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1201 		while (1) {
1202 			/* make sure that task is dead before returning, it may
1203 			 * prevent a rare case of module down in parallel to a
1204 			 * call to mlx4_ib_vma_close.
1205 			 */
1206 			put_task_struct(owning_process);
1207 			usleep_range(1000, 2000);
1208 			owning_process = get_pid_task(ibcontext->tgid,
1209 						      PIDTYPE_PID);
1210 			if (!owning_process ||
1211 			    owning_process->state == TASK_DEAD) {
1212 				pr_info("disassociate ucontext done, task was terminated\n");
1213 				/* in case task was dead need to release the task struct */
1214 				if (owning_process)
1215 					put_task_struct(owning_process);
1216 				return;
1217 			}
1218 		}
1219 	}
1220 
1221 	/* need to protect from a race on closing the vma as part of
1222 	 * mlx4_ib_vma_close().
1223 	 */
1224 	down_write(&owning_mm->mmap_sem);
1225 	for (i = 0; i < HW_BAR_COUNT; i++) {
1226 		vma = context->hw_bar_info[i].vma;
1227 		if (!vma)
1228 			continue;
1229 
1230 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1231 				   context->hw_bar_info[i].vma->vm_start,
1232 				   PAGE_SIZE);
1233 		if (ret) {
1234 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1235 			BUG_ON(1);
1236 		}
1237 
1238 		context->hw_bar_info[i].vma->vm_flags &=
1239 			~(VM_SHARED | VM_MAYSHARE);
1240 		/* context going to be destroyed, should not access ops any more */
1241 		context->hw_bar_info[i].vma->vm_ops = NULL;
1242 	}
1243 
1244 	up_write(&owning_mm->mmap_sem);
1245 	mmput(owning_mm);
1246 	put_task_struct(owning_process);
1247 }
1248 
1249 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1250 				 struct mlx4_ib_vma_private_data *vma_private_data)
1251 {
1252 	vma_private_data->vma = vma;
1253 	vma->vm_private_data = vma_private_data;
1254 	vma->vm_ops =  &mlx4_ib_vm_ops;
1255 }
1256 
1257 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1258 {
1259 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1260 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1261 
1262 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1263 		return -EINVAL;
1264 
1265 	if (vma->vm_pgoff == 0) {
1266 		/* We prevent double mmaping on same context */
1267 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1268 			return -EINVAL;
1269 
1270 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1271 
1272 		if (io_remap_pfn_range(vma, vma->vm_start,
1273 				       to_mucontext(context)->uar.pfn,
1274 				       PAGE_SIZE, vma->vm_page_prot))
1275 			return -EAGAIN;
1276 
1277 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1278 
1279 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1280 		/* We prevent double mmaping on same context */
1281 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1282 			return -EINVAL;
1283 
1284 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1285 
1286 		if (io_remap_pfn_range(vma, vma->vm_start,
1287 				       to_mucontext(context)->uar.pfn +
1288 				       dev->dev->caps.num_uars,
1289 				       PAGE_SIZE, vma->vm_page_prot))
1290 			return -EAGAIN;
1291 
1292 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1293 
1294 	} else if (vma->vm_pgoff == 3) {
1295 		struct mlx4_clock_params params;
1296 		int ret;
1297 
1298 		/* We prevent double mmaping on same context */
1299 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1300 			return -EINVAL;
1301 
1302 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1303 
1304 		if (ret)
1305 			return ret;
1306 
1307 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1308 		if (io_remap_pfn_range(vma, vma->vm_start,
1309 				       (pci_resource_start(dev->dev->persist->pdev,
1310 							   params.bar) +
1311 					params.offset)
1312 				       >> PAGE_SHIFT,
1313 				       PAGE_SIZE, vma->vm_page_prot))
1314 			return -EAGAIN;
1315 
1316 		mlx4_ib_set_vma_data(vma,
1317 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1318 	} else {
1319 		return -EINVAL;
1320 	}
1321 
1322 	return 0;
1323 }
1324 
1325 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1326 				      struct ib_ucontext *context,
1327 				      struct ib_udata *udata)
1328 {
1329 	struct mlx4_ib_pd *pd;
1330 	int err;
1331 
1332 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1333 	if (!pd)
1334 		return ERR_PTR(-ENOMEM);
1335 
1336 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1337 	if (err) {
1338 		kfree(pd);
1339 		return ERR_PTR(err);
1340 	}
1341 
1342 	if (context)
1343 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1344 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1345 			kfree(pd);
1346 			return ERR_PTR(-EFAULT);
1347 		}
1348 	return &pd->ibpd;
1349 }
1350 
1351 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1352 {
1353 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1354 	kfree(pd);
1355 
1356 	return 0;
1357 }
1358 
1359 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1360 					  struct ib_ucontext *context,
1361 					  struct ib_udata *udata)
1362 {
1363 	struct mlx4_ib_xrcd *xrcd;
1364 	struct ib_cq_init_attr cq_attr = {};
1365 	int err;
1366 
1367 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1368 		return ERR_PTR(-ENOSYS);
1369 
1370 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1371 	if (!xrcd)
1372 		return ERR_PTR(-ENOMEM);
1373 
1374 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1375 	if (err)
1376 		goto err1;
1377 
1378 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1379 	if (IS_ERR(xrcd->pd)) {
1380 		err = PTR_ERR(xrcd->pd);
1381 		goto err2;
1382 	}
1383 
1384 	cq_attr.cqe = 1;
1385 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1386 	if (IS_ERR(xrcd->cq)) {
1387 		err = PTR_ERR(xrcd->cq);
1388 		goto err3;
1389 	}
1390 
1391 	return &xrcd->ibxrcd;
1392 
1393 err3:
1394 	ib_dealloc_pd(xrcd->pd);
1395 err2:
1396 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1397 err1:
1398 	kfree(xrcd);
1399 	return ERR_PTR(err);
1400 }
1401 
1402 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1403 {
1404 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1405 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1406 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1407 	kfree(xrcd);
1408 
1409 	return 0;
1410 }
1411 
1412 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1413 {
1414 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1415 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1416 	struct mlx4_ib_gid_entry *ge;
1417 
1418 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1419 	if (!ge)
1420 		return -ENOMEM;
1421 
1422 	ge->gid = *gid;
1423 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1424 		ge->port = mqp->port;
1425 		ge->added = 1;
1426 	}
1427 
1428 	mutex_lock(&mqp->mutex);
1429 	list_add_tail(&ge->list, &mqp->gid_list);
1430 	mutex_unlock(&mqp->mutex);
1431 
1432 	return 0;
1433 }
1434 
1435 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1436 					  struct mlx4_ib_counters *ctr_table)
1437 {
1438 	struct counter_index *counter, *tmp_count;
1439 
1440 	mutex_lock(&ctr_table->mutex);
1441 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1442 				 list) {
1443 		if (counter->allocated)
1444 			mlx4_counter_free(ibdev->dev, counter->index);
1445 		list_del(&counter->list);
1446 		kfree(counter);
1447 	}
1448 	mutex_unlock(&ctr_table->mutex);
1449 }
1450 
1451 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1452 		   union ib_gid *gid)
1453 {
1454 	struct net_device *ndev;
1455 	int ret = 0;
1456 
1457 	if (!mqp->port)
1458 		return 0;
1459 
1460 	spin_lock_bh(&mdev->iboe.lock);
1461 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1462 	if (ndev)
1463 		dev_hold(ndev);
1464 	spin_unlock_bh(&mdev->iboe.lock);
1465 
1466 	if (ndev) {
1467 		ret = 1;
1468 		dev_put(ndev);
1469 	}
1470 
1471 	return ret;
1472 }
1473 
1474 struct mlx4_ib_steering {
1475 	struct list_head list;
1476 	struct mlx4_flow_reg_id reg_id;
1477 	union ib_gid gid;
1478 };
1479 
1480 #define LAST_ETH_FIELD vlan_tag
1481 #define LAST_IB_FIELD sl
1482 #define LAST_IPV4_FIELD dst_ip
1483 #define LAST_TCP_UDP_FIELD src_port
1484 
1485 /* Field is the last supported field */
1486 #define FIELDS_NOT_SUPPORTED(filter, field)\
1487 	memchr_inv((void *)&filter.field  +\
1488 		   sizeof(filter.field), 0,\
1489 		   sizeof(filter) -\
1490 		   offsetof(typeof(filter), field) -\
1491 		   sizeof(filter.field))
1492 
1493 static int parse_flow_attr(struct mlx4_dev *dev,
1494 			   u32 qp_num,
1495 			   union ib_flow_spec *ib_spec,
1496 			   struct _rule_hw *mlx4_spec)
1497 {
1498 	enum mlx4_net_trans_rule_id type;
1499 
1500 	switch (ib_spec->type) {
1501 	case IB_FLOW_SPEC_ETH:
1502 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1503 			return -ENOTSUPP;
1504 
1505 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1506 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1507 		       ETH_ALEN);
1508 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1509 		       ETH_ALEN);
1510 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1511 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1512 		break;
1513 	case IB_FLOW_SPEC_IB:
1514 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1515 			return -ENOTSUPP;
1516 
1517 		type = MLX4_NET_TRANS_RULE_ID_IB;
1518 		mlx4_spec->ib.l3_qpn =
1519 			cpu_to_be32(qp_num);
1520 		mlx4_spec->ib.qpn_mask =
1521 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1522 		break;
1523 
1524 
1525 	case IB_FLOW_SPEC_IPV4:
1526 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1527 			return -ENOTSUPP;
1528 
1529 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1530 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1531 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1532 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1533 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1534 		break;
1535 
1536 	case IB_FLOW_SPEC_TCP:
1537 	case IB_FLOW_SPEC_UDP:
1538 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1539 			return -ENOTSUPP;
1540 
1541 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1542 					MLX4_NET_TRANS_RULE_ID_TCP :
1543 					MLX4_NET_TRANS_RULE_ID_UDP;
1544 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1545 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1546 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1547 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1548 		break;
1549 
1550 	default:
1551 		return -EINVAL;
1552 	}
1553 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1554 	    mlx4_hw_rule_sz(dev, type) < 0)
1555 		return -EINVAL;
1556 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1557 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1558 	return mlx4_hw_rule_sz(dev, type);
1559 }
1560 
1561 struct default_rules {
1562 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1563 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1564 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1565 	__u8  link_layer;
1566 };
1567 static const struct default_rules default_table[] = {
1568 	{
1569 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1570 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1571 		.rules_create_list = {IB_FLOW_SPEC_IB},
1572 		.link_layer = IB_LINK_LAYER_INFINIBAND
1573 	}
1574 };
1575 
1576 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1577 					 struct ib_flow_attr *flow_attr)
1578 {
1579 	int i, j, k;
1580 	void *ib_flow;
1581 	const struct default_rules *pdefault_rules = default_table;
1582 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1583 
1584 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1585 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1586 		memset(&field_types, 0, sizeof(field_types));
1587 
1588 		if (link_layer != pdefault_rules->link_layer)
1589 			continue;
1590 
1591 		ib_flow = flow_attr + 1;
1592 		/* we assume the specs are sorted */
1593 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1594 		     j < flow_attr->num_of_specs; k++) {
1595 			union ib_flow_spec *current_flow =
1596 				(union ib_flow_spec *)ib_flow;
1597 
1598 			/* same layer but different type */
1599 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1600 			     (pdefault_rules->mandatory_fields[k] &
1601 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1602 			    (current_flow->type !=
1603 			     pdefault_rules->mandatory_fields[k]))
1604 				goto out;
1605 
1606 			/* same layer, try match next one */
1607 			if (current_flow->type ==
1608 			    pdefault_rules->mandatory_fields[k]) {
1609 				j++;
1610 				ib_flow +=
1611 					((union ib_flow_spec *)ib_flow)->size;
1612 			}
1613 		}
1614 
1615 		ib_flow = flow_attr + 1;
1616 		for (j = 0; j < flow_attr->num_of_specs;
1617 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1618 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1619 				/* same layer and same type */
1620 				if (((union ib_flow_spec *)ib_flow)->type ==
1621 				    pdefault_rules->mandatory_not_fields[k])
1622 					goto out;
1623 
1624 		return i;
1625 	}
1626 out:
1627 	return -1;
1628 }
1629 
1630 static int __mlx4_ib_create_default_rules(
1631 		struct mlx4_ib_dev *mdev,
1632 		struct ib_qp *qp,
1633 		const struct default_rules *pdefault_rules,
1634 		struct _rule_hw *mlx4_spec) {
1635 	int size = 0;
1636 	int i;
1637 
1638 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1639 		int ret;
1640 		union ib_flow_spec ib_spec;
1641 		switch (pdefault_rules->rules_create_list[i]) {
1642 		case 0:
1643 			/* no rule */
1644 			continue;
1645 		case IB_FLOW_SPEC_IB:
1646 			ib_spec.type = IB_FLOW_SPEC_IB;
1647 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1648 
1649 			break;
1650 		default:
1651 			/* invalid rule */
1652 			return -EINVAL;
1653 		}
1654 		/* We must put empty rule, qpn is being ignored */
1655 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1656 				      mlx4_spec);
1657 		if (ret < 0) {
1658 			pr_info("invalid parsing\n");
1659 			return -EINVAL;
1660 		}
1661 
1662 		mlx4_spec = (void *)mlx4_spec + ret;
1663 		size += ret;
1664 	}
1665 	return size;
1666 }
1667 
1668 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1669 			  int domain,
1670 			  enum mlx4_net_trans_promisc_mode flow_type,
1671 			  u64 *reg_id)
1672 {
1673 	int ret, i;
1674 	int size = 0;
1675 	void *ib_flow;
1676 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1677 	struct mlx4_cmd_mailbox *mailbox;
1678 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1679 	int default_flow;
1680 
1681 	static const u16 __mlx4_domain[] = {
1682 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1683 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1684 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1685 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1686 	};
1687 
1688 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1689 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1690 		return -EINVAL;
1691 	}
1692 
1693 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1694 		pr_err("Invalid domain value %d\n", domain);
1695 		return -EINVAL;
1696 	}
1697 
1698 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1699 		return -EINVAL;
1700 
1701 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1702 	if (IS_ERR(mailbox))
1703 		return PTR_ERR(mailbox);
1704 	ctrl = mailbox->buf;
1705 
1706 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1707 				 flow_attr->priority);
1708 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1709 	ctrl->port = flow_attr->port;
1710 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1711 
1712 	ib_flow = flow_attr + 1;
1713 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1714 	/* Add default flows */
1715 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1716 	if (default_flow >= 0) {
1717 		ret = __mlx4_ib_create_default_rules(
1718 				mdev, qp, default_table + default_flow,
1719 				mailbox->buf + size);
1720 		if (ret < 0) {
1721 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1722 			return -EINVAL;
1723 		}
1724 		size += ret;
1725 	}
1726 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1727 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1728 				      mailbox->buf + size);
1729 		if (ret < 0) {
1730 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1731 			return -EINVAL;
1732 		}
1733 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1734 		size += ret;
1735 	}
1736 
1737 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1738 	    flow_attr->num_of_specs == 1) {
1739 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1740 		enum ib_flow_spec_type header_spec =
1741 			((union ib_flow_spec *)(flow_attr + 1))->type;
1742 
1743 		if (header_spec == IB_FLOW_SPEC_ETH)
1744 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1745 	}
1746 
1747 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1748 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1749 			   MLX4_CMD_NATIVE);
1750 	if (ret == -ENOMEM)
1751 		pr_err("mcg table is full. Fail to register network rule.\n");
1752 	else if (ret == -ENXIO)
1753 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1754 	else if (ret)
1755 		pr_err("Invalid argument. Fail to register network rule.\n");
1756 
1757 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1758 	return ret;
1759 }
1760 
1761 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1762 {
1763 	int err;
1764 	err = mlx4_cmd(dev, reg_id, 0, 0,
1765 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1766 		       MLX4_CMD_NATIVE);
1767 	if (err)
1768 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1769 		       reg_id);
1770 	return err;
1771 }
1772 
1773 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1774 				    u64 *reg_id)
1775 {
1776 	void *ib_flow;
1777 	union ib_flow_spec *ib_spec;
1778 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1779 	int err = 0;
1780 
1781 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1782 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1783 		return 0; /* do nothing */
1784 
1785 	ib_flow = flow_attr + 1;
1786 	ib_spec = (union ib_flow_spec *)ib_flow;
1787 
1788 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1789 		return 0; /* do nothing */
1790 
1791 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1792 				    flow_attr->port, qp->qp_num,
1793 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1794 				    reg_id);
1795 	return err;
1796 }
1797 
1798 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1799 				      struct ib_flow_attr *flow_attr,
1800 				      enum mlx4_net_trans_promisc_mode *type)
1801 {
1802 	int err = 0;
1803 
1804 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1805 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1806 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1807 		return -EOPNOTSUPP;
1808 	}
1809 
1810 	if (flow_attr->num_of_specs == 0) {
1811 		type[0] = MLX4_FS_MC_SNIFFER;
1812 		type[1] = MLX4_FS_UC_SNIFFER;
1813 	} else {
1814 		union ib_flow_spec *ib_spec;
1815 
1816 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1817 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1818 			return -EINVAL;
1819 
1820 		/* if all is zero than MC and UC */
1821 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1822 			type[0] = MLX4_FS_MC_SNIFFER;
1823 			type[1] = MLX4_FS_UC_SNIFFER;
1824 		} else {
1825 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1826 					    ib_spec->eth.mask.dst_mac[1],
1827 					    ib_spec->eth.mask.dst_mac[2],
1828 					    ib_spec->eth.mask.dst_mac[3],
1829 					    ib_spec->eth.mask.dst_mac[4],
1830 					    ib_spec->eth.mask.dst_mac[5]};
1831 
1832 			/* Above xor was only on MC bit, non empty mask is valid
1833 			 * only if this bit is set and rest are zero.
1834 			 */
1835 			if (!is_zero_ether_addr(&mac[0]))
1836 				return -EINVAL;
1837 
1838 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1839 				type[0] = MLX4_FS_MC_SNIFFER;
1840 			else
1841 				type[0] = MLX4_FS_UC_SNIFFER;
1842 		}
1843 	}
1844 
1845 	return err;
1846 }
1847 
1848 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1849 				    struct ib_flow_attr *flow_attr,
1850 				    int domain)
1851 {
1852 	int err = 0, i = 0, j = 0;
1853 	struct mlx4_ib_flow *mflow;
1854 	enum mlx4_net_trans_promisc_mode type[2];
1855 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1856 	int is_bonded = mlx4_is_bonded(dev);
1857 
1858 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1859 		return ERR_PTR(-EINVAL);
1860 
1861 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1862 		return ERR_PTR(-EOPNOTSUPP);
1863 
1864 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1865 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1866 		return ERR_PTR(-EOPNOTSUPP);
1867 
1868 	memset(type, 0, sizeof(type));
1869 
1870 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1871 	if (!mflow) {
1872 		err = -ENOMEM;
1873 		goto err_free;
1874 	}
1875 
1876 	switch (flow_attr->type) {
1877 	case IB_FLOW_ATTR_NORMAL:
1878 		/* If dont trap flag (continue match) is set, under specific
1879 		 * condition traffic be replicated to given qp,
1880 		 * without stealing it
1881 		 */
1882 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1883 			err = mlx4_ib_add_dont_trap_rule(dev,
1884 							 flow_attr,
1885 							 type);
1886 			if (err)
1887 				goto err_free;
1888 		} else {
1889 			type[0] = MLX4_FS_REGULAR;
1890 		}
1891 		break;
1892 
1893 	case IB_FLOW_ATTR_ALL_DEFAULT:
1894 		type[0] = MLX4_FS_ALL_DEFAULT;
1895 		break;
1896 
1897 	case IB_FLOW_ATTR_MC_DEFAULT:
1898 		type[0] = MLX4_FS_MC_DEFAULT;
1899 		break;
1900 
1901 	case IB_FLOW_ATTR_SNIFFER:
1902 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1903 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1904 		break;
1905 
1906 	default:
1907 		err = -EINVAL;
1908 		goto err_free;
1909 	}
1910 
1911 	while (i < ARRAY_SIZE(type) && type[i]) {
1912 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1913 					    &mflow->reg_id[i].id);
1914 		if (err)
1915 			goto err_create_flow;
1916 		if (is_bonded) {
1917 			/* Application always sees one port so the mirror rule
1918 			 * must be on port #2
1919 			 */
1920 			flow_attr->port = 2;
1921 			err = __mlx4_ib_create_flow(qp, flow_attr,
1922 						    domain, type[j],
1923 						    &mflow->reg_id[j].mirror);
1924 			flow_attr->port = 1;
1925 			if (err)
1926 				goto err_create_flow;
1927 			j++;
1928 		}
1929 
1930 		i++;
1931 	}
1932 
1933 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1934 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1935 					       &mflow->reg_id[i].id);
1936 		if (err)
1937 			goto err_create_flow;
1938 
1939 		if (is_bonded) {
1940 			flow_attr->port = 2;
1941 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1942 						       &mflow->reg_id[j].mirror);
1943 			flow_attr->port = 1;
1944 			if (err)
1945 				goto err_create_flow;
1946 			j++;
1947 		}
1948 		/* function to create mirror rule */
1949 		i++;
1950 	}
1951 
1952 	return &mflow->ibflow;
1953 
1954 err_create_flow:
1955 	while (i) {
1956 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1957 					     mflow->reg_id[i].id);
1958 		i--;
1959 	}
1960 
1961 	while (j) {
1962 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1963 					     mflow->reg_id[j].mirror);
1964 		j--;
1965 	}
1966 err_free:
1967 	kfree(mflow);
1968 	return ERR_PTR(err);
1969 }
1970 
1971 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1972 {
1973 	int err, ret = 0;
1974 	int i = 0;
1975 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1976 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1977 
1978 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1979 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1980 		if (err)
1981 			ret = err;
1982 		if (mflow->reg_id[i].mirror) {
1983 			err = __mlx4_ib_destroy_flow(mdev->dev,
1984 						     mflow->reg_id[i].mirror);
1985 			if (err)
1986 				ret = err;
1987 		}
1988 		i++;
1989 	}
1990 
1991 	kfree(mflow);
1992 	return ret;
1993 }
1994 
1995 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1996 {
1997 	int err;
1998 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1999 	struct mlx4_dev	*dev = mdev->dev;
2000 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2001 	struct mlx4_ib_steering *ib_steering = NULL;
2002 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
2003 	struct mlx4_flow_reg_id	reg_id;
2004 
2005 	if (mdev->dev->caps.steering_mode ==
2006 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2007 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
2008 		if (!ib_steering)
2009 			return -ENOMEM;
2010 	}
2011 
2012 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
2013 				    !!(mqp->flags &
2014 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2015 				    prot, &reg_id.id);
2016 	if (err) {
2017 		pr_err("multicast attach op failed, err %d\n", err);
2018 		goto err_malloc;
2019 	}
2020 
2021 	reg_id.mirror = 0;
2022 	if (mlx4_is_bonded(dev)) {
2023 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
2024 					    (mqp->port == 1) ? 2 : 1,
2025 					    !!(mqp->flags &
2026 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2027 					    prot, &reg_id.mirror);
2028 		if (err)
2029 			goto err_add;
2030 	}
2031 
2032 	err = add_gid_entry(ibqp, gid);
2033 	if (err)
2034 		goto err_add;
2035 
2036 	if (ib_steering) {
2037 		memcpy(ib_steering->gid.raw, gid->raw, 16);
2038 		ib_steering->reg_id = reg_id;
2039 		mutex_lock(&mqp->mutex);
2040 		list_add(&ib_steering->list, &mqp->steering_rules);
2041 		mutex_unlock(&mqp->mutex);
2042 	}
2043 	return 0;
2044 
2045 err_add:
2046 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2047 			      prot, reg_id.id);
2048 	if (reg_id.mirror)
2049 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2050 				      prot, reg_id.mirror);
2051 err_malloc:
2052 	kfree(ib_steering);
2053 
2054 	return err;
2055 }
2056 
2057 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2058 {
2059 	struct mlx4_ib_gid_entry *ge;
2060 	struct mlx4_ib_gid_entry *tmp;
2061 	struct mlx4_ib_gid_entry *ret = NULL;
2062 
2063 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2064 		if (!memcmp(raw, ge->gid.raw, 16)) {
2065 			ret = ge;
2066 			break;
2067 		}
2068 	}
2069 
2070 	return ret;
2071 }
2072 
2073 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2074 {
2075 	int err;
2076 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2077 	struct mlx4_dev *dev = mdev->dev;
2078 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2079 	struct net_device *ndev;
2080 	struct mlx4_ib_gid_entry *ge;
2081 	struct mlx4_flow_reg_id reg_id = {0, 0};
2082 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2083 
2084 	if (mdev->dev->caps.steering_mode ==
2085 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2086 		struct mlx4_ib_steering *ib_steering;
2087 
2088 		mutex_lock(&mqp->mutex);
2089 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2090 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2091 				list_del(&ib_steering->list);
2092 				break;
2093 			}
2094 		}
2095 		mutex_unlock(&mqp->mutex);
2096 		if (&ib_steering->list == &mqp->steering_rules) {
2097 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2098 			return -EINVAL;
2099 		}
2100 		reg_id = ib_steering->reg_id;
2101 		kfree(ib_steering);
2102 	}
2103 
2104 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2105 				    prot, reg_id.id);
2106 	if (err)
2107 		return err;
2108 
2109 	if (mlx4_is_bonded(dev)) {
2110 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2111 					    prot, reg_id.mirror);
2112 		if (err)
2113 			return err;
2114 	}
2115 
2116 	mutex_lock(&mqp->mutex);
2117 	ge = find_gid_entry(mqp, gid->raw);
2118 	if (ge) {
2119 		spin_lock_bh(&mdev->iboe.lock);
2120 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2121 		if (ndev)
2122 			dev_hold(ndev);
2123 		spin_unlock_bh(&mdev->iboe.lock);
2124 		if (ndev)
2125 			dev_put(ndev);
2126 		list_del(&ge->list);
2127 		kfree(ge);
2128 	} else
2129 		pr_warn("could not find mgid entry\n");
2130 
2131 	mutex_unlock(&mqp->mutex);
2132 
2133 	return 0;
2134 }
2135 
2136 static int init_node_data(struct mlx4_ib_dev *dev)
2137 {
2138 	struct ib_smp *in_mad  = NULL;
2139 	struct ib_smp *out_mad = NULL;
2140 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2141 	int err = -ENOMEM;
2142 
2143 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2144 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2145 	if (!in_mad || !out_mad)
2146 		goto out;
2147 
2148 	init_query_mad(in_mad);
2149 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2150 	if (mlx4_is_master(dev->dev))
2151 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2152 
2153 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2154 	if (err)
2155 		goto out;
2156 
2157 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2158 
2159 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2160 
2161 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2162 	if (err)
2163 		goto out;
2164 
2165 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2166 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2167 
2168 out:
2169 	kfree(in_mad);
2170 	kfree(out_mad);
2171 	return err;
2172 }
2173 
2174 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2175 			char *buf)
2176 {
2177 	struct mlx4_ib_dev *dev =
2178 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2179 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2180 }
2181 
2182 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2183 			char *buf)
2184 {
2185 	struct mlx4_ib_dev *dev =
2186 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2187 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2188 }
2189 
2190 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2191 			  char *buf)
2192 {
2193 	struct mlx4_ib_dev *dev =
2194 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2195 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2196 		       dev->dev->board_id);
2197 }
2198 
2199 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2200 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2201 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2202 
2203 static struct device_attribute *mlx4_class_attributes[] = {
2204 	&dev_attr_hw_rev,
2205 	&dev_attr_hca_type,
2206 	&dev_attr_board_id
2207 };
2208 
2209 struct diag_counter {
2210 	const char *name;
2211 	u32 offset;
2212 };
2213 
2214 #define DIAG_COUNTER(_name, _offset)			\
2215 	{ .name = #_name, .offset = _offset }
2216 
2217 static const struct diag_counter diag_basic[] = {
2218 	DIAG_COUNTER(rq_num_lle, 0x00),
2219 	DIAG_COUNTER(sq_num_lle, 0x04),
2220 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2221 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2222 	DIAG_COUNTER(rq_num_lpe, 0x18),
2223 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2224 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2225 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2226 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2227 	DIAG_COUNTER(sq_num_bre, 0x34),
2228 	DIAG_COUNTER(sq_num_rire, 0x44),
2229 	DIAG_COUNTER(rq_num_rire, 0x48),
2230 	DIAG_COUNTER(sq_num_rae, 0x4C),
2231 	DIAG_COUNTER(rq_num_rae, 0x50),
2232 	DIAG_COUNTER(sq_num_roe, 0x54),
2233 	DIAG_COUNTER(sq_num_tree, 0x5C),
2234 	DIAG_COUNTER(sq_num_rree, 0x64),
2235 	DIAG_COUNTER(rq_num_rnr, 0x68),
2236 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2237 	DIAG_COUNTER(rq_num_oos, 0x100),
2238 	DIAG_COUNTER(sq_num_oos, 0x104),
2239 };
2240 
2241 static const struct diag_counter diag_ext[] = {
2242 	DIAG_COUNTER(rq_num_dup, 0x130),
2243 	DIAG_COUNTER(sq_num_to, 0x134),
2244 };
2245 
2246 static const struct diag_counter diag_device_only[] = {
2247 	DIAG_COUNTER(num_cqovf, 0x1A0),
2248 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2249 };
2250 
2251 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2252 						    u8 port_num)
2253 {
2254 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2255 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2256 
2257 	if (!diag[!!port_num].name)
2258 		return NULL;
2259 
2260 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2261 					  diag[!!port_num].num_counters,
2262 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2263 }
2264 
2265 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2266 				struct rdma_hw_stats *stats,
2267 				u8 port, int index)
2268 {
2269 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2270 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2271 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2272 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2273 	int ret;
2274 	int i;
2275 
2276 	ret = mlx4_query_diag_counters(dev->dev,
2277 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2278 				       diag[!!port].offset, hw_value,
2279 				       diag[!!port].num_counters, port);
2280 
2281 	if (ret)
2282 		return ret;
2283 
2284 	for (i = 0; i < diag[!!port].num_counters; i++)
2285 		stats->value[i] = hw_value[i];
2286 
2287 	return diag[!!port].num_counters;
2288 }
2289 
2290 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2291 					 const char ***name,
2292 					 u32 **offset,
2293 					 u32 *num,
2294 					 bool port)
2295 {
2296 	u32 num_counters;
2297 
2298 	num_counters = ARRAY_SIZE(diag_basic);
2299 
2300 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2301 		num_counters += ARRAY_SIZE(diag_ext);
2302 
2303 	if (!port)
2304 		num_counters += ARRAY_SIZE(diag_device_only);
2305 
2306 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2307 	if (!*name)
2308 		return -ENOMEM;
2309 
2310 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2311 	if (!*offset)
2312 		goto err_name;
2313 
2314 	*num = num_counters;
2315 
2316 	return 0;
2317 
2318 err_name:
2319 	kfree(*name);
2320 	return -ENOMEM;
2321 }
2322 
2323 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2324 				       const char **name,
2325 				       u32 *offset,
2326 				       bool port)
2327 {
2328 	int i;
2329 	int j;
2330 
2331 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2332 		name[i] = diag_basic[i].name;
2333 		offset[i] = diag_basic[i].offset;
2334 	}
2335 
2336 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2337 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2338 			name[j] = diag_ext[i].name;
2339 			offset[j] = diag_ext[i].offset;
2340 		}
2341 	}
2342 
2343 	if (!port) {
2344 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2345 			name[j] = diag_device_only[i].name;
2346 			offset[j] = diag_device_only[i].offset;
2347 		}
2348 	}
2349 }
2350 
2351 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2352 {
2353 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2354 	int i;
2355 	int ret;
2356 	bool per_port = !!(ibdev->dev->caps.flags2 &
2357 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2358 
2359 	if (mlx4_is_slave(ibdev->dev))
2360 		return 0;
2361 
2362 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2363 		/* i == 1 means we are building port counters */
2364 		if (i && !per_port)
2365 			continue;
2366 
2367 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2368 						    &diag[i].offset,
2369 						    &diag[i].num_counters, i);
2370 		if (ret)
2371 			goto err_alloc;
2372 
2373 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2374 					   diag[i].offset, i);
2375 	}
2376 
2377 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2378 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2379 
2380 	return 0;
2381 
2382 err_alloc:
2383 	if (i) {
2384 		kfree(diag[i - 1].name);
2385 		kfree(diag[i - 1].offset);
2386 	}
2387 
2388 	return ret;
2389 }
2390 
2391 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2392 {
2393 	int i;
2394 
2395 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2396 		kfree(ibdev->diag_counters[i].offset);
2397 		kfree(ibdev->diag_counters[i].name);
2398 	}
2399 }
2400 
2401 #define MLX4_IB_INVALID_MAC	((u64)-1)
2402 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2403 			       struct net_device *dev,
2404 			       int port)
2405 {
2406 	u64 new_smac = 0;
2407 	u64 release_mac = MLX4_IB_INVALID_MAC;
2408 	struct mlx4_ib_qp *qp;
2409 
2410 	read_lock(&dev_base_lock);
2411 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2412 	read_unlock(&dev_base_lock);
2413 
2414 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2415 
2416 	/* no need for update QP1 and mac registration in non-SRIOV */
2417 	if (!mlx4_is_mfunc(ibdev->dev))
2418 		return;
2419 
2420 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2421 	qp = ibdev->qp1_proxy[port - 1];
2422 	if (qp) {
2423 		int new_smac_index;
2424 		u64 old_smac;
2425 		struct mlx4_update_qp_params update_params;
2426 
2427 		mutex_lock(&qp->mutex);
2428 		old_smac = qp->pri.smac;
2429 		if (new_smac == old_smac)
2430 			goto unlock;
2431 
2432 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2433 
2434 		if (new_smac_index < 0)
2435 			goto unlock;
2436 
2437 		update_params.smac_index = new_smac_index;
2438 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2439 				   &update_params)) {
2440 			release_mac = new_smac;
2441 			goto unlock;
2442 		}
2443 		/* if old port was zero, no mac was yet registered for this QP */
2444 		if (qp->pri.smac_port)
2445 			release_mac = old_smac;
2446 		qp->pri.smac = new_smac;
2447 		qp->pri.smac_port = port;
2448 		qp->pri.smac_index = new_smac_index;
2449 	}
2450 
2451 unlock:
2452 	if (release_mac != MLX4_IB_INVALID_MAC)
2453 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2454 	if (qp)
2455 		mutex_unlock(&qp->mutex);
2456 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2457 }
2458 
2459 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2460 				 struct net_device *dev,
2461 				 unsigned long event)
2462 
2463 {
2464 	struct mlx4_ib_iboe *iboe;
2465 	int update_qps_port = -1;
2466 	int port;
2467 
2468 	ASSERT_RTNL();
2469 
2470 	iboe = &ibdev->iboe;
2471 
2472 	spin_lock_bh(&iboe->lock);
2473 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2474 
2475 		iboe->netdevs[port - 1] =
2476 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2477 
2478 		if (dev == iboe->netdevs[port - 1] &&
2479 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2480 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2481 			update_qps_port = port;
2482 
2483 	}
2484 	spin_unlock_bh(&iboe->lock);
2485 
2486 	if (update_qps_port > 0)
2487 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2488 }
2489 
2490 static int mlx4_ib_netdev_event(struct notifier_block *this,
2491 				unsigned long event, void *ptr)
2492 {
2493 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2494 	struct mlx4_ib_dev *ibdev;
2495 
2496 	if (!net_eq(dev_net(dev), &init_net))
2497 		return NOTIFY_DONE;
2498 
2499 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2500 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2501 
2502 	return NOTIFY_DONE;
2503 }
2504 
2505 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2506 {
2507 	int port;
2508 	int slave;
2509 	int i;
2510 
2511 	if (mlx4_is_master(ibdev->dev)) {
2512 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2513 		     ++slave) {
2514 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2515 				for (i = 0;
2516 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2517 				     ++i) {
2518 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2519 					/* master has the identity virt2phys pkey mapping */
2520 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2521 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2522 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2523 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2524 				}
2525 			}
2526 		}
2527 		/* initialize pkey cache */
2528 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2529 			for (i = 0;
2530 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2531 			     ++i)
2532 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2533 					(i) ? 0 : 0xFFFF;
2534 		}
2535 	}
2536 }
2537 
2538 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2539 {
2540 	int i, j, eq = 0, total_eqs = 0;
2541 
2542 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2543 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2544 	if (!ibdev->eq_table)
2545 		return;
2546 
2547 	for (i = 1; i <= dev->caps.num_ports; i++) {
2548 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2549 		     j++, total_eqs++) {
2550 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2551 				continue;
2552 			ibdev->eq_table[eq] = total_eqs;
2553 			if (!mlx4_assign_eq(dev, i,
2554 					    &ibdev->eq_table[eq]))
2555 				eq++;
2556 			else
2557 				ibdev->eq_table[eq] = -1;
2558 		}
2559 	}
2560 
2561 	for (i = eq; i < dev->caps.num_comp_vectors;
2562 	     ibdev->eq_table[i++] = -1)
2563 		;
2564 
2565 	/* Advertise the new number of EQs to clients */
2566 	ibdev->ib_dev.num_comp_vectors = eq;
2567 }
2568 
2569 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2570 {
2571 	int i;
2572 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2573 
2574 	/* no eqs were allocated */
2575 	if (!ibdev->eq_table)
2576 		return;
2577 
2578 	/* Reset the advertised EQ number */
2579 	ibdev->ib_dev.num_comp_vectors = 0;
2580 
2581 	for (i = 0; i < total_eqs; i++)
2582 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2583 
2584 	kfree(ibdev->eq_table);
2585 	ibdev->eq_table = NULL;
2586 }
2587 
2588 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2589 			       struct ib_port_immutable *immutable)
2590 {
2591 	struct ib_port_attr attr;
2592 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2593 	int err;
2594 
2595 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2596 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2597 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2598 	} else {
2599 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2600 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2601 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2602 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2603 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2604 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2605 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2606 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2607 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2608 	}
2609 
2610 	err = ib_query_port(ibdev, port_num, &attr);
2611 	if (err)
2612 		return err;
2613 
2614 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2615 	immutable->gid_tbl_len = attr.gid_tbl_len;
2616 
2617 	return 0;
2618 }
2619 
2620 static void get_fw_ver_str(struct ib_device *device, char *str)
2621 {
2622 	struct mlx4_ib_dev *dev =
2623 		container_of(device, struct mlx4_ib_dev, ib_dev);
2624 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2625 		 (int) (dev->dev->caps.fw_ver >> 32),
2626 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2627 		 (int) dev->dev->caps.fw_ver & 0xffff);
2628 }
2629 
2630 static void *mlx4_ib_add(struct mlx4_dev *dev)
2631 {
2632 	struct mlx4_ib_dev *ibdev;
2633 	int num_ports = 0;
2634 	int i, j;
2635 	int err;
2636 	struct mlx4_ib_iboe *iboe;
2637 	int ib_num_ports = 0;
2638 	int num_req_counters;
2639 	int allocated;
2640 	u32 counter_index;
2641 	struct counter_index *new_counter_index = NULL;
2642 
2643 	pr_info_once("%s", mlx4_ib_version);
2644 
2645 	num_ports = 0;
2646 	mlx4_foreach_ib_transport_port(i, dev)
2647 		num_ports++;
2648 
2649 	/* No point in registering a device with no ports... */
2650 	if (num_ports == 0)
2651 		return NULL;
2652 
2653 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2654 	if (!ibdev) {
2655 		dev_err(&dev->persist->pdev->dev,
2656 			"Device struct alloc failed\n");
2657 		return NULL;
2658 	}
2659 
2660 	iboe = &ibdev->iboe;
2661 
2662 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2663 		goto err_dealloc;
2664 
2665 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2666 		goto err_pd;
2667 
2668 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2669 				 PAGE_SIZE);
2670 	if (!ibdev->uar_map)
2671 		goto err_uar;
2672 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2673 
2674 	ibdev->dev = dev;
2675 	ibdev->bond_next_port	= 0;
2676 
2677 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2678 	ibdev->ib_dev.owner		= THIS_MODULE;
2679 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2680 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2681 	ibdev->num_ports		= num_ports;
2682 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2683 						1 : ibdev->num_ports;
2684 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2685 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2686 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2687 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2688 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2689 
2690 	if (dev->caps.userspace_caps)
2691 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2692 	else
2693 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2694 
2695 	ibdev->ib_dev.uverbs_cmd_mask	=
2696 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2697 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2698 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2699 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2700 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2701 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2702 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2703 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2704 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2705 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2706 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2707 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2708 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2709 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2710 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2711 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2712 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2713 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2714 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2715 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2716 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2717 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2718 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2719 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2720 
2721 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2722 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2723 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2724 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2725 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2726 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2727 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2728 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2729 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2730 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2731 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2732 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2733 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2734 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2735 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2736 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2737 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2738 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2739 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2740 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2741 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2742 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2743 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2744 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2745 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2746 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2747 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2748 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2749 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2750 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2751 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2752 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2753 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2754 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2755 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2756 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2757 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2758 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2759 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2760 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2761 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2762 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2763 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2764 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2765 
2766 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2767 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2768 
2769 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2770 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2771 	    IB_LINK_LAYER_ETHERNET) ||
2772 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2773 	    IB_LINK_LAYER_ETHERNET))) {
2774 		ibdev->ib_dev.create_wq		= mlx4_ib_create_wq;
2775 		ibdev->ib_dev.modify_wq		= mlx4_ib_modify_wq;
2776 		ibdev->ib_dev.destroy_wq	= mlx4_ib_destroy_wq;
2777 		ibdev->ib_dev.create_rwq_ind_table  =
2778 			mlx4_ib_create_rwq_ind_table;
2779 		ibdev->ib_dev.destroy_rwq_ind_table =
2780 			mlx4_ib_destroy_rwq_ind_table;
2781 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2782 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2783 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2784 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2785 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2786 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2787 	}
2788 
2789 	if (!mlx4_is_slave(ibdev->dev)) {
2790 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2791 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2792 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2793 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2794 	}
2795 
2796 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2797 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2798 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2799 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2800 
2801 		ibdev->ib_dev.uverbs_cmd_mask |=
2802 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2803 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2804 	}
2805 
2806 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2807 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2808 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2809 		ibdev->ib_dev.uverbs_cmd_mask |=
2810 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2811 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2812 	}
2813 
2814 	if (check_flow_steering_support(dev)) {
2815 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2816 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2817 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2818 
2819 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2820 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2821 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2822 	}
2823 
2824 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2825 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2826 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2827 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2828 
2829 	mlx4_ib_alloc_eqs(dev, ibdev);
2830 
2831 	spin_lock_init(&iboe->lock);
2832 
2833 	if (init_node_data(ibdev))
2834 		goto err_map;
2835 	mlx4_init_sl2vl_tbl(ibdev);
2836 
2837 	for (i = 0; i < ibdev->num_ports; ++i) {
2838 		mutex_init(&ibdev->counters_table[i].mutex);
2839 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2840 	}
2841 
2842 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2843 	for (i = 0; i < num_req_counters; ++i) {
2844 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2845 		allocated = 0;
2846 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2847 						IB_LINK_LAYER_ETHERNET) {
2848 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2849 						 MLX4_RES_USAGE_DRIVER);
2850 			/* if failed to allocate a new counter, use default */
2851 			if (err)
2852 				counter_index =
2853 					mlx4_get_default_counter_index(dev,
2854 								       i + 1);
2855 			else
2856 				allocated = 1;
2857 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2858 			counter_index = mlx4_get_default_counter_index(dev,
2859 								       i + 1);
2860 		}
2861 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2862 					    GFP_KERNEL);
2863 		if (!new_counter_index) {
2864 			if (allocated)
2865 				mlx4_counter_free(ibdev->dev, counter_index);
2866 			goto err_counter;
2867 		}
2868 		new_counter_index->index = counter_index;
2869 		new_counter_index->allocated = allocated;
2870 		list_add_tail(&new_counter_index->list,
2871 			      &ibdev->counters_table[i].counters_list);
2872 		ibdev->counters_table[i].default_counter = counter_index;
2873 		pr_info("counter index %d for port %d allocated %d\n",
2874 			counter_index, i + 1, allocated);
2875 	}
2876 	if (mlx4_is_bonded(dev))
2877 		for (i = 1; i < ibdev->num_ports ; ++i) {
2878 			new_counter_index =
2879 					kmalloc(sizeof(struct counter_index),
2880 						GFP_KERNEL);
2881 			if (!new_counter_index)
2882 				goto err_counter;
2883 			new_counter_index->index = counter_index;
2884 			new_counter_index->allocated = 0;
2885 			list_add_tail(&new_counter_index->list,
2886 				      &ibdev->counters_table[i].counters_list);
2887 			ibdev->counters_table[i].default_counter =
2888 								counter_index;
2889 		}
2890 
2891 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2892 		ib_num_ports++;
2893 
2894 	spin_lock_init(&ibdev->sm_lock);
2895 	mutex_init(&ibdev->cap_mask_mutex);
2896 	INIT_LIST_HEAD(&ibdev->qp_list);
2897 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2898 
2899 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2900 	    ib_num_ports) {
2901 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2902 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2903 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2904 					    &ibdev->steer_qpn_base, 0,
2905 					    MLX4_RES_USAGE_DRIVER);
2906 		if (err)
2907 			goto err_counter;
2908 
2909 		ibdev->ib_uc_qpns_bitmap =
2910 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2911 				sizeof(long),
2912 				GFP_KERNEL);
2913 		if (!ibdev->ib_uc_qpns_bitmap)
2914 			goto err_steer_qp_release;
2915 
2916 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2917 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2918 				    ibdev->steer_qpn_count);
2919 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2920 					dev, ibdev->steer_qpn_base,
2921 					ibdev->steer_qpn_base +
2922 					ibdev->steer_qpn_count - 1);
2923 			if (err)
2924 				goto err_steer_free_bitmap;
2925 		} else {
2926 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2927 				    ibdev->steer_qpn_count);
2928 		}
2929 	}
2930 
2931 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2932 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2933 
2934 	if (mlx4_ib_alloc_diag_counters(ibdev))
2935 		goto err_steer_free_bitmap;
2936 
2937 	ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2938 	if (ib_register_device(&ibdev->ib_dev, NULL))
2939 		goto err_diag_counters;
2940 
2941 	if (mlx4_ib_mad_init(ibdev))
2942 		goto err_reg;
2943 
2944 	if (mlx4_ib_init_sriov(ibdev))
2945 		goto err_mad;
2946 
2947 	if (!iboe->nb.notifier_call) {
2948 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2949 		err = register_netdevice_notifier(&iboe->nb);
2950 		if (err) {
2951 			iboe->nb.notifier_call = NULL;
2952 			goto err_notif;
2953 		}
2954 	}
2955 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2956 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2957 		if (err)
2958 			goto err_notif;
2959 	}
2960 
2961 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2962 		if (device_create_file(&ibdev->ib_dev.dev,
2963 				       mlx4_class_attributes[j]))
2964 			goto err_notif;
2965 	}
2966 
2967 	ibdev->ib_active = true;
2968 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2969 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2970 					 &ibdev->ib_dev);
2971 
2972 	if (mlx4_is_mfunc(ibdev->dev))
2973 		init_pkeys(ibdev);
2974 
2975 	/* create paravirt contexts for any VFs which are active */
2976 	if (mlx4_is_master(ibdev->dev)) {
2977 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2978 			if (j == mlx4_master_func_num(ibdev->dev))
2979 				continue;
2980 			if (mlx4_is_slave_active(ibdev->dev, j))
2981 				do_slave_init(ibdev, j, 1);
2982 		}
2983 	}
2984 	return ibdev;
2985 
2986 err_notif:
2987 	if (ibdev->iboe.nb.notifier_call) {
2988 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2989 			pr_warn("failure unregistering notifier\n");
2990 		ibdev->iboe.nb.notifier_call = NULL;
2991 	}
2992 	flush_workqueue(wq);
2993 
2994 	mlx4_ib_close_sriov(ibdev);
2995 
2996 err_mad:
2997 	mlx4_ib_mad_cleanup(ibdev);
2998 
2999 err_reg:
3000 	ib_unregister_device(&ibdev->ib_dev);
3001 
3002 err_diag_counters:
3003 	mlx4_ib_diag_cleanup(ibdev);
3004 
3005 err_steer_free_bitmap:
3006 	kfree(ibdev->ib_uc_qpns_bitmap);
3007 
3008 err_steer_qp_release:
3009 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3010 			      ibdev->steer_qpn_count);
3011 err_counter:
3012 	for (i = 0; i < ibdev->num_ports; ++i)
3013 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
3014 
3015 err_map:
3016 	mlx4_ib_free_eqs(dev, ibdev);
3017 	iounmap(ibdev->uar_map);
3018 
3019 err_uar:
3020 	mlx4_uar_free(dev, &ibdev->priv_uar);
3021 
3022 err_pd:
3023 	mlx4_pd_free(dev, ibdev->priv_pdn);
3024 
3025 err_dealloc:
3026 	ib_dealloc_device(&ibdev->ib_dev);
3027 
3028 	return NULL;
3029 }
3030 
3031 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
3032 {
3033 	int offset;
3034 
3035 	WARN_ON(!dev->ib_uc_qpns_bitmap);
3036 
3037 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3038 					 dev->steer_qpn_count,
3039 					 get_count_order(count));
3040 	if (offset < 0)
3041 		return offset;
3042 
3043 	*qpn = dev->steer_qpn_base + offset;
3044 	return 0;
3045 }
3046 
3047 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3048 {
3049 	if (!qpn ||
3050 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3051 		return;
3052 
3053 	BUG_ON(qpn < dev->steer_qpn_base);
3054 
3055 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
3056 			      qpn - dev->steer_qpn_base,
3057 			      get_count_order(count));
3058 }
3059 
3060 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3061 			 int is_attach)
3062 {
3063 	int err;
3064 	size_t flow_size;
3065 	struct ib_flow_attr *flow = NULL;
3066 	struct ib_flow_spec_ib *ib_spec;
3067 
3068 	if (is_attach) {
3069 		flow_size = sizeof(struct ib_flow_attr) +
3070 			    sizeof(struct ib_flow_spec_ib);
3071 		flow = kzalloc(flow_size, GFP_KERNEL);
3072 		if (!flow)
3073 			return -ENOMEM;
3074 		flow->port = mqp->port;
3075 		flow->num_of_specs = 1;
3076 		flow->size = flow_size;
3077 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3078 		ib_spec->type = IB_FLOW_SPEC_IB;
3079 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3080 		/* Add an empty rule for IB L2 */
3081 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3082 
3083 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3084 					    IB_FLOW_DOMAIN_NIC,
3085 					    MLX4_FS_REGULAR,
3086 					    &mqp->reg_id);
3087 	} else {
3088 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3089 	}
3090 	kfree(flow);
3091 	return err;
3092 }
3093 
3094 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3095 {
3096 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3097 	int p;
3098 	int i;
3099 
3100 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3101 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3102 	ibdev->ib_active = false;
3103 	flush_workqueue(wq);
3104 
3105 	mlx4_ib_close_sriov(ibdev);
3106 	mlx4_ib_mad_cleanup(ibdev);
3107 	ib_unregister_device(&ibdev->ib_dev);
3108 	mlx4_ib_diag_cleanup(ibdev);
3109 	if (ibdev->iboe.nb.notifier_call) {
3110 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3111 			pr_warn("failure unregistering notifier\n");
3112 		ibdev->iboe.nb.notifier_call = NULL;
3113 	}
3114 
3115 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3116 			      ibdev->steer_qpn_count);
3117 	kfree(ibdev->ib_uc_qpns_bitmap);
3118 
3119 	iounmap(ibdev->uar_map);
3120 	for (p = 0; p < ibdev->num_ports; ++p)
3121 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3122 
3123 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3124 		mlx4_CLOSE_PORT(dev, p);
3125 
3126 	mlx4_ib_free_eqs(dev, ibdev);
3127 
3128 	mlx4_uar_free(dev, &ibdev->priv_uar);
3129 	mlx4_pd_free(dev, ibdev->priv_pdn);
3130 	ib_dealloc_device(&ibdev->ib_dev);
3131 }
3132 
3133 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3134 {
3135 	struct mlx4_ib_demux_work **dm = NULL;
3136 	struct mlx4_dev *dev = ibdev->dev;
3137 	int i;
3138 	unsigned long flags;
3139 	struct mlx4_active_ports actv_ports;
3140 	unsigned int ports;
3141 	unsigned int first_port;
3142 
3143 	if (!mlx4_is_master(dev))
3144 		return;
3145 
3146 	actv_ports = mlx4_get_active_ports(dev, slave);
3147 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3148 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3149 
3150 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3151 	if (!dm)
3152 		return;
3153 
3154 	for (i = 0; i < ports; i++) {
3155 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3156 		if (!dm[i]) {
3157 			while (--i >= 0)
3158 				kfree(dm[i]);
3159 			goto out;
3160 		}
3161 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3162 		dm[i]->port = first_port + i + 1;
3163 		dm[i]->slave = slave;
3164 		dm[i]->do_init = do_init;
3165 		dm[i]->dev = ibdev;
3166 	}
3167 	/* initialize or tear down tunnel QPs for the slave */
3168 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3169 	if (!ibdev->sriov.is_going_down) {
3170 		for (i = 0; i < ports; i++)
3171 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3172 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3173 	} else {
3174 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3175 		for (i = 0; i < ports; i++)
3176 			kfree(dm[i]);
3177 	}
3178 out:
3179 	kfree(dm);
3180 	return;
3181 }
3182 
3183 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3184 {
3185 	struct mlx4_ib_qp *mqp;
3186 	unsigned long flags_qp;
3187 	unsigned long flags_cq;
3188 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3189 	struct list_head    cq_notify_list;
3190 	struct mlx4_cq *mcq;
3191 	unsigned long flags;
3192 
3193 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3194 	INIT_LIST_HEAD(&cq_notify_list);
3195 
3196 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3197 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3198 
3199 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3200 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3201 		if (mqp->sq.tail != mqp->sq.head) {
3202 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3203 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3204 			if (send_mcq->mcq.comp &&
3205 			    mqp->ibqp.send_cq->comp_handler) {
3206 				if (!send_mcq->mcq.reset_notify_added) {
3207 					send_mcq->mcq.reset_notify_added = 1;
3208 					list_add_tail(&send_mcq->mcq.reset_notify,
3209 						      &cq_notify_list);
3210 				}
3211 			}
3212 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3213 		}
3214 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3215 		/* Now, handle the QP's receive queue */
3216 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3217 		/* no handling is needed for SRQ */
3218 		if (!mqp->ibqp.srq) {
3219 			if (mqp->rq.tail != mqp->rq.head) {
3220 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3221 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3222 				if (recv_mcq->mcq.comp &&
3223 				    mqp->ibqp.recv_cq->comp_handler) {
3224 					if (!recv_mcq->mcq.reset_notify_added) {
3225 						recv_mcq->mcq.reset_notify_added = 1;
3226 						list_add_tail(&recv_mcq->mcq.reset_notify,
3227 							      &cq_notify_list);
3228 					}
3229 				}
3230 				spin_unlock_irqrestore(&recv_mcq->lock,
3231 						       flags_cq);
3232 			}
3233 		}
3234 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3235 	}
3236 
3237 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3238 		mcq->comp(mcq);
3239 	}
3240 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3241 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3242 }
3243 
3244 static void handle_bonded_port_state_event(struct work_struct *work)
3245 {
3246 	struct ib_event_work *ew =
3247 		container_of(work, struct ib_event_work, work);
3248 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3249 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3250 	int i;
3251 	struct ib_event ibev;
3252 
3253 	kfree(ew);
3254 	spin_lock_bh(&ibdev->iboe.lock);
3255 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3256 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3257 		enum ib_port_state curr_port_state;
3258 
3259 		if (!curr_netdev)
3260 			continue;
3261 
3262 		curr_port_state =
3263 			(netif_running(curr_netdev) &&
3264 			 netif_carrier_ok(curr_netdev)) ?
3265 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3266 
3267 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3268 			curr_port_state : IB_PORT_ACTIVE;
3269 	}
3270 	spin_unlock_bh(&ibdev->iboe.lock);
3271 
3272 	ibev.device = &ibdev->ib_dev;
3273 	ibev.element.port_num = 1;
3274 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3275 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3276 
3277 	ib_dispatch_event(&ibev);
3278 }
3279 
3280 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3281 {
3282 	u64 sl2vl;
3283 	int err;
3284 
3285 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3286 	if (err) {
3287 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3288 		       port, err);
3289 		sl2vl = 0;
3290 	}
3291 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3292 }
3293 
3294 static void ib_sl2vl_update_work(struct work_struct *work)
3295 {
3296 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3297 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3298 	int port = ew->port;
3299 
3300 	mlx4_ib_sl2vl_update(mdev, port);
3301 
3302 	kfree(ew);
3303 }
3304 
3305 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3306 				     int port)
3307 {
3308 	struct ib_event_work *ew;
3309 
3310 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3311 	if (ew) {
3312 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3313 		ew->port = port;
3314 		ew->ib_dev = ibdev;
3315 		queue_work(wq, &ew->work);
3316 	}
3317 }
3318 
3319 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3320 			  enum mlx4_dev_event event, unsigned long param)
3321 {
3322 	struct ib_event ibev;
3323 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3324 	struct mlx4_eqe *eqe = NULL;
3325 	struct ib_event_work *ew;
3326 	int p = 0;
3327 
3328 	if (mlx4_is_bonded(dev) &&
3329 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3330 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3331 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3332 		if (!ew)
3333 			return;
3334 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3335 		ew->ib_dev = ibdev;
3336 		queue_work(wq, &ew->work);
3337 		return;
3338 	}
3339 
3340 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3341 		eqe = (struct mlx4_eqe *)param;
3342 	else
3343 		p = (int) param;
3344 
3345 	switch (event) {
3346 	case MLX4_DEV_EVENT_PORT_UP:
3347 		if (p > ibdev->num_ports)
3348 			return;
3349 		if (!mlx4_is_slave(dev) &&
3350 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3351 			IB_LINK_LAYER_INFINIBAND) {
3352 			if (mlx4_is_master(dev))
3353 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3354 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3355 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3356 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3357 		}
3358 		ibev.event = IB_EVENT_PORT_ACTIVE;
3359 		break;
3360 
3361 	case MLX4_DEV_EVENT_PORT_DOWN:
3362 		if (p > ibdev->num_ports)
3363 			return;
3364 		ibev.event = IB_EVENT_PORT_ERR;
3365 		break;
3366 
3367 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3368 		ibdev->ib_active = false;
3369 		ibev.event = IB_EVENT_DEVICE_FATAL;
3370 		mlx4_ib_handle_catas_error(ibdev);
3371 		break;
3372 
3373 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3374 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3375 		if (!ew)
3376 			break;
3377 
3378 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3379 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3380 		ew->ib_dev = ibdev;
3381 		/* need to queue only for port owner, which uses GEN_EQE */
3382 		if (mlx4_is_master(dev))
3383 			queue_work(wq, &ew->work);
3384 		else
3385 			handle_port_mgmt_change_event(&ew->work);
3386 		return;
3387 
3388 	case MLX4_DEV_EVENT_SLAVE_INIT:
3389 		/* here, p is the slave id */
3390 		do_slave_init(ibdev, p, 1);
3391 		if (mlx4_is_master(dev)) {
3392 			int i;
3393 
3394 			for (i = 1; i <= ibdev->num_ports; i++) {
3395 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3396 					== IB_LINK_LAYER_INFINIBAND)
3397 					mlx4_ib_slave_alias_guid_event(ibdev,
3398 								       p, i,
3399 								       1);
3400 			}
3401 		}
3402 		return;
3403 
3404 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3405 		if (mlx4_is_master(dev)) {
3406 			int i;
3407 
3408 			for (i = 1; i <= ibdev->num_ports; i++) {
3409 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3410 					== IB_LINK_LAYER_INFINIBAND)
3411 					mlx4_ib_slave_alias_guid_event(ibdev,
3412 								       p, i,
3413 								       0);
3414 			}
3415 		}
3416 		/* here, p is the slave id */
3417 		do_slave_init(ibdev, p, 0);
3418 		return;
3419 
3420 	default:
3421 		return;
3422 	}
3423 
3424 	ibev.device	      = ibdev_ptr;
3425 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3426 
3427 	ib_dispatch_event(&ibev);
3428 }
3429 
3430 static struct mlx4_interface mlx4_ib_interface = {
3431 	.add		= mlx4_ib_add,
3432 	.remove		= mlx4_ib_remove,
3433 	.event		= mlx4_ib_event,
3434 	.protocol	= MLX4_PROT_IB_IPV6,
3435 	.flags		= MLX4_INTFF_BONDING
3436 };
3437 
3438 static int __init mlx4_ib_init(void)
3439 {
3440 	int err;
3441 
3442 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3443 	if (!wq)
3444 		return -ENOMEM;
3445 
3446 	err = mlx4_ib_mcg_init();
3447 	if (err)
3448 		goto clean_wq;
3449 
3450 	err = mlx4_register_interface(&mlx4_ib_interface);
3451 	if (err)
3452 		goto clean_mcg;
3453 
3454 	return 0;
3455 
3456 clean_mcg:
3457 	mlx4_ib_mcg_destroy();
3458 
3459 clean_wq:
3460 	destroy_workqueue(wq);
3461 	return err;
3462 }
3463 
3464 static void __exit mlx4_ib_cleanup(void)
3465 {
3466 	mlx4_unregister_interface(&mlx4_ib_interface);
3467 	mlx4_ib_mcg_destroy();
3468 	destroy_workqueue(wq);
3469 }
3470 
3471 module_init(mlx4_ib_init);
3472 module_exit(mlx4_ib_cleanup);
3473