xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision faffb083)
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 						    u32 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static int check_flow_steering_support(struct mlx4_dev *dev)
89 {
90 	int eth_num_ports = 0;
91 	int ib_num_ports = 0;
92 
93 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
94 
95 	if (dmfs) {
96 		int i;
97 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
98 			eth_num_ports++;
99 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
100 			ib_num_ports++;
101 		dmfs &= (!ib_num_ports ||
102 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
103 			(!eth_num_ports ||
104 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
105 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
106 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
107 			dmfs = 0;
108 		}
109 	}
110 	return dmfs;
111 }
112 
113 static int num_ib_ports(struct mlx4_dev *dev)
114 {
115 	int ib_ports = 0;
116 	int i;
117 
118 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
119 		ib_ports++;
120 
121 	return ib_ports;
122 }
123 
124 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device,
125 					     u32 port_num)
126 {
127 	struct mlx4_ib_dev *ibdev = to_mdev(device);
128 	struct net_device *dev;
129 
130 	rcu_read_lock();
131 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
132 
133 	if (dev) {
134 		if (mlx4_is_bonded(ibdev->dev)) {
135 			struct net_device *upper = NULL;
136 
137 			upper = netdev_master_upper_dev_get_rcu(dev);
138 			if (upper) {
139 				struct net_device *active;
140 
141 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
142 				if (active)
143 					dev = active;
144 			}
145 		}
146 	}
147 	dev_hold(dev);
148 
149 	rcu_read_unlock();
150 	return dev;
151 }
152 
153 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
154 				  struct mlx4_ib_dev *ibdev,
155 				  u32 port_num)
156 {
157 	struct mlx4_cmd_mailbox *mailbox;
158 	int err;
159 	struct mlx4_dev *dev = ibdev->dev;
160 	int i;
161 	union ib_gid *gid_tbl;
162 
163 	mailbox = mlx4_alloc_cmd_mailbox(dev);
164 	if (IS_ERR(mailbox))
165 		return -ENOMEM;
166 
167 	gid_tbl = mailbox->buf;
168 
169 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
170 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
171 
172 	err = mlx4_cmd(dev, mailbox->dma,
173 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
174 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
175 		       MLX4_CMD_WRAPPED);
176 	if (mlx4_is_bonded(dev))
177 		err += mlx4_cmd(dev, mailbox->dma,
178 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
179 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
180 				MLX4_CMD_WRAPPED);
181 
182 	mlx4_free_cmd_mailbox(dev, mailbox);
183 	return err;
184 }
185 
186 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
187 				     struct mlx4_ib_dev *ibdev,
188 				     u32 port_num)
189 {
190 	struct mlx4_cmd_mailbox *mailbox;
191 	int err;
192 	struct mlx4_dev *dev = ibdev->dev;
193 	int i;
194 	struct {
195 		union ib_gid	gid;
196 		__be32		rsrvd1[2];
197 		__be16		rsrvd2;
198 		u8		type;
199 		u8		version;
200 		__be32		rsrvd3;
201 	} *gid_tbl;
202 
203 	mailbox = mlx4_alloc_cmd_mailbox(dev);
204 	if (IS_ERR(mailbox))
205 		return -ENOMEM;
206 
207 	gid_tbl = mailbox->buf;
208 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
209 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
210 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
211 			gid_tbl[i].version = 2;
212 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
213 				gid_tbl[i].type = 1;
214 		}
215 	}
216 
217 	err = mlx4_cmd(dev, mailbox->dma,
218 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
219 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
220 		       MLX4_CMD_WRAPPED);
221 	if (mlx4_is_bonded(dev))
222 		err += mlx4_cmd(dev, mailbox->dma,
223 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
224 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
225 				MLX4_CMD_WRAPPED);
226 
227 	mlx4_free_cmd_mailbox(dev, mailbox);
228 	return err;
229 }
230 
231 static int mlx4_ib_update_gids(struct gid_entry *gids,
232 			       struct mlx4_ib_dev *ibdev,
233 			       u32 port_num)
234 {
235 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
236 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
237 
238 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
239 }
240 
241 static void free_gid_entry(struct gid_entry *entry)
242 {
243 	memset(&entry->gid, 0, sizeof(entry->gid));
244 	kfree(entry->ctx);
245 	entry->ctx = NULL;
246 }
247 
248 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
249 {
250 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
251 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
252 	struct mlx4_port_gid_table   *port_gid_table;
253 	int free = -1, found = -1;
254 	int ret = 0;
255 	int hw_update = 0;
256 	int i;
257 	struct gid_entry *gids = NULL;
258 	u16 vlan_id = 0xffff;
259 	u8 mac[ETH_ALEN];
260 
261 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
262 		return -EINVAL;
263 
264 	if (attr->port_num > MLX4_MAX_PORTS)
265 		return -EINVAL;
266 
267 	if (!context)
268 		return -EINVAL;
269 
270 	ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
271 	if (ret)
272 		return ret;
273 	port_gid_table = &iboe->gids[attr->port_num - 1];
274 	spin_lock_bh(&iboe->lock);
275 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
276 		if (!memcmp(&port_gid_table->gids[i].gid,
277 			    &attr->gid, sizeof(attr->gid)) &&
278 		    port_gid_table->gids[i].gid_type == attr->gid_type &&
279 		    port_gid_table->gids[i].vlan_id == vlan_id)  {
280 			found = i;
281 			break;
282 		}
283 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
284 			free = i; /* HW has space */
285 	}
286 
287 	if (found < 0) {
288 		if (free < 0) {
289 			ret = -ENOSPC;
290 		} else {
291 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
292 			if (!port_gid_table->gids[free].ctx) {
293 				ret = -ENOMEM;
294 			} else {
295 				*context = port_gid_table->gids[free].ctx;
296 				memcpy(&port_gid_table->gids[free].gid,
297 				       &attr->gid, sizeof(attr->gid));
298 				port_gid_table->gids[free].gid_type = attr->gid_type;
299 				port_gid_table->gids[free].vlan_id = vlan_id;
300 				port_gid_table->gids[free].ctx->real_index = free;
301 				port_gid_table->gids[free].ctx->refcount = 1;
302 				hw_update = 1;
303 			}
304 		}
305 	} else {
306 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
307 		*context = ctx;
308 		ctx->refcount++;
309 	}
310 	if (!ret && hw_update) {
311 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
312 				     GFP_ATOMIC);
313 		if (!gids) {
314 			ret = -ENOMEM;
315 			*context = NULL;
316 			free_gid_entry(&port_gid_table->gids[free]);
317 		} else {
318 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
319 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
320 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
321 			}
322 		}
323 	}
324 	spin_unlock_bh(&iboe->lock);
325 
326 	if (!ret && hw_update) {
327 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
328 		if (ret) {
329 			spin_lock_bh(&iboe->lock);
330 			*context = NULL;
331 			free_gid_entry(&port_gid_table->gids[free]);
332 			spin_unlock_bh(&iboe->lock);
333 		}
334 		kfree(gids);
335 	}
336 
337 	return ret;
338 }
339 
340 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
341 {
342 	struct gid_cache_context *ctx = *context;
343 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
344 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
345 	struct mlx4_port_gid_table   *port_gid_table;
346 	int ret = 0;
347 	int hw_update = 0;
348 	struct gid_entry *gids = NULL;
349 
350 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
351 		return -EINVAL;
352 
353 	if (attr->port_num > MLX4_MAX_PORTS)
354 		return -EINVAL;
355 
356 	port_gid_table = &iboe->gids[attr->port_num - 1];
357 	spin_lock_bh(&iboe->lock);
358 	if (ctx) {
359 		ctx->refcount--;
360 		if (!ctx->refcount) {
361 			unsigned int real_index = ctx->real_index;
362 
363 			free_gid_entry(&port_gid_table->gids[real_index]);
364 			hw_update = 1;
365 		}
366 	}
367 	if (!ret && hw_update) {
368 		int i;
369 
370 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
371 				     GFP_ATOMIC);
372 		if (!gids) {
373 			ret = -ENOMEM;
374 		} else {
375 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
376 				memcpy(&gids[i].gid,
377 				       &port_gid_table->gids[i].gid,
378 				       sizeof(union ib_gid));
379 				gids[i].gid_type =
380 				    port_gid_table->gids[i].gid_type;
381 			}
382 		}
383 	}
384 	spin_unlock_bh(&iboe->lock);
385 
386 	if (!ret && hw_update) {
387 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
388 		kfree(gids);
389 	}
390 	return ret;
391 }
392 
393 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
394 				    const struct ib_gid_attr *attr)
395 {
396 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
397 	struct gid_cache_context *ctx = NULL;
398 	struct mlx4_port_gid_table   *port_gid_table;
399 	int real_index = -EINVAL;
400 	int i;
401 	unsigned long flags;
402 	u32 port_num = attr->port_num;
403 
404 	if (port_num > MLX4_MAX_PORTS)
405 		return -EINVAL;
406 
407 	if (mlx4_is_bonded(ibdev->dev))
408 		port_num = 1;
409 
410 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
411 		return attr->index;
412 
413 	spin_lock_irqsave(&iboe->lock, flags);
414 	port_gid_table = &iboe->gids[port_num - 1];
415 
416 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
417 		if (!memcmp(&port_gid_table->gids[i].gid,
418 			    &attr->gid, sizeof(attr->gid)) &&
419 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
420 			ctx = port_gid_table->gids[i].ctx;
421 			break;
422 		}
423 	if (ctx)
424 		real_index = ctx->real_index;
425 	spin_unlock_irqrestore(&iboe->lock, flags);
426 	return real_index;
427 }
428 
429 static int mlx4_ib_query_device(struct ib_device *ibdev,
430 				struct ib_device_attr *props,
431 				struct ib_udata *uhw)
432 {
433 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
434 	struct ib_smp *in_mad  = NULL;
435 	struct ib_smp *out_mad = NULL;
436 	int err;
437 	int have_ib_ports;
438 	struct mlx4_uverbs_ex_query_device cmd;
439 	struct mlx4_uverbs_ex_query_device_resp resp = {};
440 	struct mlx4_clock_params clock_params;
441 
442 	if (uhw->inlen) {
443 		if (uhw->inlen < sizeof(cmd))
444 			return -EINVAL;
445 
446 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
447 		if (err)
448 			return err;
449 
450 		if (cmd.comp_mask)
451 			return -EINVAL;
452 
453 		if (cmd.reserved)
454 			return -EINVAL;
455 	}
456 
457 	resp.response_length = offsetof(typeof(resp), response_length) +
458 		sizeof(resp.response_length);
459 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
460 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
461 	err = -ENOMEM;
462 	if (!in_mad || !out_mad)
463 		goto out;
464 
465 	ib_init_query_mad(in_mad);
466 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
467 
468 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
469 			   1, NULL, NULL, in_mad, out_mad);
470 	if (err)
471 		goto out;
472 
473 	memset(props, 0, sizeof *props);
474 
475 	have_ib_ports = num_ib_ports(dev->dev);
476 
477 	props->fw_ver = dev->dev->caps.fw_ver;
478 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
479 		IB_DEVICE_PORT_ACTIVE_EVENT		|
480 		IB_DEVICE_SYS_IMAGE_GUID		|
481 		IB_DEVICE_RC_RNR_NAK_GEN;
482 	props->kernel_cap_flags = IBK_BLOCK_MULTICAST_LOOPBACK;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
484 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
485 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
486 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
488 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
490 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
492 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
493 	if (dev->dev->caps.max_gso_sz &&
494 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
495 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
496 		props->kernel_cap_flags |= IBK_UD_TSO;
497 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
498 		props->kernel_cap_flags |= IBK_LOCAL_DMA_LKEY;
499 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
500 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
501 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
502 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
503 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
504 		props->device_cap_flags |= IB_DEVICE_XRC;
505 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
506 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
507 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
508 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
509 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
510 		else
511 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
512 	}
513 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
514 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
515 
516 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
517 
518 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
519 		0xffffff;
520 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
521 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
522 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
523 
524 	props->max_mr_size	   = ~0ull;
525 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
526 	props->max_qp		   = dev->dev->quotas.qp;
527 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
528 	props->max_send_sge =
529 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
530 	props->max_recv_sge =
531 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
532 	props->max_sge_rd = MLX4_MAX_SGE_RD;
533 	props->max_cq		   = dev->dev->quotas.cq;
534 	props->max_cqe		   = dev->dev->caps.max_cqes;
535 	props->max_mr		   = dev->dev->quotas.mpt;
536 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
537 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
538 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
539 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
540 	props->max_srq		   = dev->dev->quotas.srq;
541 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
542 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
543 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
544 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
545 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
546 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
547 	props->masked_atomic_cap   = props->atomic_cap;
548 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
549 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
550 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
551 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
552 					   props->max_mcast_grp;
553 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
554 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
555 	props->max_ah = INT_MAX;
556 
557 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
558 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
559 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
560 			props->rss_caps.max_rwq_indirection_tables =
561 				props->max_qp;
562 			props->rss_caps.max_rwq_indirection_table_size =
563 				dev->dev->caps.max_rss_tbl_sz;
564 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
565 			props->max_wq_type_rq = props->max_qp;
566 		}
567 
568 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
569 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
570 	}
571 
572 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
573 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
574 
575 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
576 		resp.response_length += sizeof(resp.hca_core_clock_offset);
577 		if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
578 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
579 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
580 		}
581 	}
582 
583 	if (uhw->outlen >= resp.response_length +
584 	    sizeof(resp.max_inl_recv_sz)) {
585 		resp.response_length += sizeof(resp.max_inl_recv_sz);
586 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
587 			sizeof(struct mlx4_wqe_data_seg);
588 	}
589 
590 	if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
591 		if (props->rss_caps.supported_qpts) {
592 			resp.rss_caps.rx_hash_function =
593 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
594 
595 			resp.rss_caps.rx_hash_fields_mask =
596 				MLX4_IB_RX_HASH_SRC_IPV4 |
597 				MLX4_IB_RX_HASH_DST_IPV4 |
598 				MLX4_IB_RX_HASH_SRC_IPV6 |
599 				MLX4_IB_RX_HASH_DST_IPV6 |
600 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
601 				MLX4_IB_RX_HASH_DST_PORT_TCP |
602 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
603 				MLX4_IB_RX_HASH_DST_PORT_UDP;
604 
605 			if (dev->dev->caps.tunnel_offload_mode ==
606 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
607 				resp.rss_caps.rx_hash_fields_mask |=
608 					MLX4_IB_RX_HASH_INNER;
609 		}
610 		resp.response_length = offsetof(typeof(resp), rss_caps) +
611 				       sizeof(resp.rss_caps);
612 	}
613 
614 	if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
615 		if (dev->dev->caps.max_gso_sz &&
616 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
617 		    IB_LINK_LAYER_ETHERNET) ||
618 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
619 		    IB_LINK_LAYER_ETHERNET))) {
620 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
621 			resp.tso_caps.supported_qpts |=
622 				1 << IB_QPT_RAW_PACKET;
623 		}
624 		resp.response_length = offsetof(typeof(resp), tso_caps) +
625 				       sizeof(resp.tso_caps);
626 	}
627 
628 	if (uhw->outlen) {
629 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
630 		if (err)
631 			goto out;
632 	}
633 out:
634 	kfree(in_mad);
635 	kfree(out_mad);
636 
637 	return err;
638 }
639 
640 static enum rdma_link_layer
641 mlx4_ib_port_link_layer(struct ib_device *device, u32 port_num)
642 {
643 	struct mlx4_dev *dev = to_mdev(device)->dev;
644 
645 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
646 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
647 }
648 
649 static int ib_link_query_port(struct ib_device *ibdev, u32 port,
650 			      struct ib_port_attr *props, int netw_view)
651 {
652 	struct ib_smp *in_mad  = NULL;
653 	struct ib_smp *out_mad = NULL;
654 	int ext_active_speed;
655 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
656 	int err = -ENOMEM;
657 
658 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
659 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
660 	if (!in_mad || !out_mad)
661 		goto out;
662 
663 	ib_init_query_mad(in_mad);
664 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
665 	in_mad->attr_mod = cpu_to_be32(port);
666 
667 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
668 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
669 
670 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
671 				in_mad, out_mad);
672 	if (err)
673 		goto out;
674 
675 
676 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
677 	props->lmc		= out_mad->data[34] & 0x7;
678 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
679 	props->sm_sl		= out_mad->data[36] & 0xf;
680 	props->state		= out_mad->data[32] & 0xf;
681 	props->phys_state	= out_mad->data[33] >> 4;
682 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
683 	if (netw_view)
684 		props->gid_tbl_len = out_mad->data[50];
685 	else
686 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
687 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
688 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
689 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
690 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
691 	props->active_width	= out_mad->data[31] & 0xf;
692 	props->active_speed	= out_mad->data[35] >> 4;
693 	props->max_mtu		= out_mad->data[41] & 0xf;
694 	props->active_mtu	= out_mad->data[36] >> 4;
695 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
696 	props->max_vl_num	= out_mad->data[37] >> 4;
697 	props->init_type_reply	= out_mad->data[41] >> 4;
698 
699 	/* Check if extended speeds (EDR/FDR/...) are supported */
700 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
701 		ext_active_speed = out_mad->data[62] >> 4;
702 
703 		switch (ext_active_speed) {
704 		case 1:
705 			props->active_speed = IB_SPEED_FDR;
706 			break;
707 		case 2:
708 			props->active_speed = IB_SPEED_EDR;
709 			break;
710 		}
711 	}
712 
713 	/* If reported active speed is QDR, check if is FDR-10 */
714 	if (props->active_speed == IB_SPEED_QDR) {
715 		ib_init_query_mad(in_mad);
716 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
717 		in_mad->attr_mod = cpu_to_be32(port);
718 
719 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
720 				   NULL, NULL, in_mad, out_mad);
721 		if (err)
722 			goto out;
723 
724 		/* Checking LinkSpeedActive for FDR-10 */
725 		if (out_mad->data[15] & 0x1)
726 			props->active_speed = IB_SPEED_FDR10;
727 	}
728 
729 	/* Avoid wrong speed value returned by FW if the IB link is down. */
730 	if (props->state == IB_PORT_DOWN)
731 		 props->active_speed = IB_SPEED_SDR;
732 
733 out:
734 	kfree(in_mad);
735 	kfree(out_mad);
736 	return err;
737 }
738 
739 static u8 state_to_phys_state(enum ib_port_state state)
740 {
741 	return state == IB_PORT_ACTIVE ?
742 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
743 }
744 
745 static int eth_link_query_port(struct ib_device *ibdev, u32 port,
746 			       struct ib_port_attr *props)
747 {
748 
749 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
750 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
751 	struct net_device *ndev;
752 	enum ib_mtu tmp;
753 	struct mlx4_cmd_mailbox *mailbox;
754 	int err = 0;
755 	int is_bonded = mlx4_is_bonded(mdev->dev);
756 
757 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
758 	if (IS_ERR(mailbox))
759 		return PTR_ERR(mailbox);
760 
761 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
762 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
763 			   MLX4_CMD_WRAPPED);
764 	if (err)
765 		goto out;
766 
767 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
768 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
769 					   IB_WIDTH_4X : IB_WIDTH_1X;
770 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
771 					   IB_SPEED_FDR : IB_SPEED_QDR;
772 	props->port_cap_flags	= IB_PORT_CM_SUP;
773 	props->ip_gids = true;
774 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
775 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
776 	if (mdev->dev->caps.pkey_table_len[port])
777 		props->pkey_tbl_len = 1;
778 	props->max_mtu		= IB_MTU_4096;
779 	props->max_vl_num	= 2;
780 	props->state		= IB_PORT_DOWN;
781 	props->phys_state	= state_to_phys_state(props->state);
782 	props->active_mtu	= IB_MTU_256;
783 	spin_lock_bh(&iboe->lock);
784 	ndev = iboe->netdevs[port - 1];
785 	if (ndev && is_bonded) {
786 		rcu_read_lock(); /* required to get upper dev */
787 		ndev = netdev_master_upper_dev_get_rcu(ndev);
788 		rcu_read_unlock();
789 	}
790 	if (!ndev)
791 		goto out_unlock;
792 
793 	tmp = iboe_get_mtu(ndev->mtu);
794 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
795 
796 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
797 					IB_PORT_ACTIVE : IB_PORT_DOWN;
798 	props->phys_state	= state_to_phys_state(props->state);
799 out_unlock:
800 	spin_unlock_bh(&iboe->lock);
801 out:
802 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
803 	return err;
804 }
805 
806 int __mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
807 			 struct ib_port_attr *props, int netw_view)
808 {
809 	int err;
810 
811 	/* props being zeroed by the caller, avoid zeroing it here */
812 
813 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
814 		ib_link_query_port(ibdev, port, props, netw_view) :
815 				eth_link_query_port(ibdev, port, props);
816 
817 	return err;
818 }
819 
820 static int mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
821 			      struct ib_port_attr *props)
822 {
823 	/* returns host view */
824 	return __mlx4_ib_query_port(ibdev, port, props, 0);
825 }
826 
827 int __mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
828 			union ib_gid *gid, int netw_view)
829 {
830 	struct ib_smp *in_mad  = NULL;
831 	struct ib_smp *out_mad = NULL;
832 	int err = -ENOMEM;
833 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
834 	int clear = 0;
835 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
836 
837 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
838 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
839 	if (!in_mad || !out_mad)
840 		goto out;
841 
842 	ib_init_query_mad(in_mad);
843 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
844 	in_mad->attr_mod = cpu_to_be32(port);
845 
846 	if (mlx4_is_mfunc(dev->dev) && netw_view)
847 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
848 
849 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
850 	if (err)
851 		goto out;
852 
853 	memcpy(gid->raw, out_mad->data + 8, 8);
854 
855 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
856 		if (index) {
857 			/* For any index > 0, return the null guid */
858 			err = 0;
859 			clear = 1;
860 			goto out;
861 		}
862 	}
863 
864 	ib_init_query_mad(in_mad);
865 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
866 	in_mad->attr_mod = cpu_to_be32(index / 8);
867 
868 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
869 			   NULL, NULL, in_mad, out_mad);
870 	if (err)
871 		goto out;
872 
873 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
874 
875 out:
876 	if (clear)
877 		memset(gid->raw + 8, 0, 8);
878 	kfree(in_mad);
879 	kfree(out_mad);
880 	return err;
881 }
882 
883 static int mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
884 			     union ib_gid *gid)
885 {
886 	if (rdma_protocol_ib(ibdev, port))
887 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
888 	return 0;
889 }
890 
891 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u32 port,
892 			       u64 *sl2vl_tbl)
893 {
894 	union sl2vl_tbl_to_u64 sl2vl64;
895 	struct ib_smp *in_mad  = NULL;
896 	struct ib_smp *out_mad = NULL;
897 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
898 	int err = -ENOMEM;
899 	int jj;
900 
901 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
902 		*sl2vl_tbl = 0;
903 		return 0;
904 	}
905 
906 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
907 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
908 	if (!in_mad || !out_mad)
909 		goto out;
910 
911 	ib_init_query_mad(in_mad);
912 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
913 	in_mad->attr_mod = 0;
914 
915 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
916 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
917 
918 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
919 			   in_mad, out_mad);
920 	if (err)
921 		goto out;
922 
923 	for (jj = 0; jj < 8; jj++)
924 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
925 	*sl2vl_tbl = sl2vl64.sl64;
926 
927 out:
928 	kfree(in_mad);
929 	kfree(out_mad);
930 	return err;
931 }
932 
933 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
934 {
935 	u64 sl2vl;
936 	int i;
937 	int err;
938 
939 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
940 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
941 			continue;
942 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
943 		if (err) {
944 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
945 			       i, err);
946 			sl2vl = 0;
947 		}
948 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
949 	}
950 }
951 
952 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
953 			 u16 *pkey, int netw_view)
954 {
955 	struct ib_smp *in_mad  = NULL;
956 	struct ib_smp *out_mad = NULL;
957 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
958 	int err = -ENOMEM;
959 
960 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
961 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
962 	if (!in_mad || !out_mad)
963 		goto out;
964 
965 	ib_init_query_mad(in_mad);
966 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
967 	in_mad->attr_mod = cpu_to_be32(index / 32);
968 
969 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
970 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
971 
972 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
973 			   in_mad, out_mad);
974 	if (err)
975 		goto out;
976 
977 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
978 
979 out:
980 	kfree(in_mad);
981 	kfree(out_mad);
982 	return err;
983 }
984 
985 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
986 			      u16 *pkey)
987 {
988 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
989 }
990 
991 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
992 				 struct ib_device_modify *props)
993 {
994 	struct mlx4_cmd_mailbox *mailbox;
995 	unsigned long flags;
996 
997 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
998 		return -EOPNOTSUPP;
999 
1000 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1001 		return 0;
1002 
1003 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1004 		return -EOPNOTSUPP;
1005 
1006 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1007 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1008 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1009 
1010 	/*
1011 	 * If possible, pass node desc to FW, so it can generate
1012 	 * a 144 trap.  If cmd fails, just ignore.
1013 	 */
1014 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1015 	if (IS_ERR(mailbox))
1016 		return 0;
1017 
1018 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1019 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1020 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1021 
1022 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1023 
1024 	return 0;
1025 }
1026 
1027 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u32 port,
1028 			    int reset_qkey_viols, u32 cap_mask)
1029 {
1030 	struct mlx4_cmd_mailbox *mailbox;
1031 	int err;
1032 
1033 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1034 	if (IS_ERR(mailbox))
1035 		return PTR_ERR(mailbox);
1036 
1037 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1038 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1039 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1040 	} else {
1041 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1042 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1043 	}
1044 
1045 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1046 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1047 		       MLX4_CMD_WRAPPED);
1048 
1049 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1050 	return err;
1051 }
1052 
1053 static int mlx4_ib_modify_port(struct ib_device *ibdev, u32 port, int mask,
1054 			       struct ib_port_modify *props)
1055 {
1056 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1057 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1058 	struct ib_port_attr attr;
1059 	u32 cap_mask;
1060 	int err;
1061 
1062 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1063 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1064 	 * violations and port capabilities are not meaningful.
1065 	 */
1066 	if (is_eth)
1067 		return 0;
1068 
1069 	mutex_lock(&mdev->cap_mask_mutex);
1070 
1071 	err = ib_query_port(ibdev, port, &attr);
1072 	if (err)
1073 		goto out;
1074 
1075 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1076 		~props->clr_port_cap_mask;
1077 
1078 	err = mlx4_ib_SET_PORT(mdev, port,
1079 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1080 			       cap_mask);
1081 
1082 out:
1083 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1084 	return err;
1085 }
1086 
1087 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1088 				  struct ib_udata *udata)
1089 {
1090 	struct ib_device *ibdev = uctx->device;
1091 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1092 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1093 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1094 	struct mlx4_ib_alloc_ucontext_resp resp;
1095 	int err;
1096 
1097 	if (!dev->ib_active)
1098 		return -EAGAIN;
1099 
1100 	if (ibdev->ops.uverbs_abi_ver ==
1101 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1102 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1103 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1104 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1105 	} else {
1106 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1107 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1108 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1109 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1110 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1111 	}
1112 
1113 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1114 	if (err)
1115 		return err;
1116 
1117 	INIT_LIST_HEAD(&context->db_page_list);
1118 	mutex_init(&context->db_page_mutex);
1119 
1120 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1121 	mutex_init(&context->wqn_ranges_mutex);
1122 
1123 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1124 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1125 	else
1126 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1127 
1128 	if (err) {
1129 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1130 		return -EFAULT;
1131 	}
1132 
1133 	return err;
1134 }
1135 
1136 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1137 {
1138 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1139 
1140 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1141 }
1142 
1143 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1144 {
1145 }
1146 
1147 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1148 {
1149 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1150 
1151 	switch (vma->vm_pgoff) {
1152 	case 0:
1153 		return rdma_user_mmap_io(context, vma,
1154 					 to_mucontext(context)->uar.pfn,
1155 					 PAGE_SIZE,
1156 					 pgprot_noncached(vma->vm_page_prot),
1157 					 NULL);
1158 
1159 	case 1:
1160 		if (dev->dev->caps.bf_reg_size == 0)
1161 			return -EINVAL;
1162 		return rdma_user_mmap_io(
1163 			context, vma,
1164 			to_mucontext(context)->uar.pfn +
1165 				dev->dev->caps.num_uars,
1166 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1167 			NULL);
1168 
1169 	case 3: {
1170 		struct mlx4_clock_params params;
1171 		int ret;
1172 
1173 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1174 		if (ret)
1175 			return ret;
1176 
1177 		return rdma_user_mmap_io(
1178 			context, vma,
1179 			(pci_resource_start(dev->dev->persist->pdev,
1180 					    params.bar) +
1181 			 params.offset) >>
1182 				PAGE_SHIFT,
1183 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1184 			NULL);
1185 	}
1186 
1187 	default:
1188 		return -EINVAL;
1189 	}
1190 }
1191 
1192 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1193 {
1194 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1195 	struct ib_device *ibdev = ibpd->device;
1196 	int err;
1197 
1198 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1199 	if (err)
1200 		return err;
1201 
1202 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1203 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1204 		return -EFAULT;
1205 	}
1206 	return 0;
1207 }
1208 
1209 static int mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1210 {
1211 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1212 	return 0;
1213 }
1214 
1215 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1216 {
1217 	struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1218 	struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1219 	struct ib_cq_init_attr cq_attr = {};
1220 	int err;
1221 
1222 	if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1223 		return -EOPNOTSUPP;
1224 
1225 	err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1226 	if (err)
1227 		return err;
1228 
1229 	xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1230 	if (IS_ERR(xrcd->pd)) {
1231 		err = PTR_ERR(xrcd->pd);
1232 		goto err2;
1233 	}
1234 
1235 	cq_attr.cqe = 1;
1236 	xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1237 	if (IS_ERR(xrcd->cq)) {
1238 		err = PTR_ERR(xrcd->cq);
1239 		goto err3;
1240 	}
1241 
1242 	return 0;
1243 
1244 err3:
1245 	ib_dealloc_pd(xrcd->pd);
1246 err2:
1247 	mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1248 	return err;
1249 }
1250 
1251 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1252 {
1253 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1254 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1255 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1256 	return 0;
1257 }
1258 
1259 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1260 {
1261 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1262 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1263 	struct mlx4_ib_gid_entry *ge;
1264 
1265 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1266 	if (!ge)
1267 		return -ENOMEM;
1268 
1269 	ge->gid = *gid;
1270 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1271 		ge->port = mqp->port;
1272 		ge->added = 1;
1273 	}
1274 
1275 	mutex_lock(&mqp->mutex);
1276 	list_add_tail(&ge->list, &mqp->gid_list);
1277 	mutex_unlock(&mqp->mutex);
1278 
1279 	return 0;
1280 }
1281 
1282 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1283 					  struct mlx4_ib_counters *ctr_table)
1284 {
1285 	struct counter_index *counter, *tmp_count;
1286 
1287 	mutex_lock(&ctr_table->mutex);
1288 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1289 				 list) {
1290 		if (counter->allocated)
1291 			mlx4_counter_free(ibdev->dev, counter->index);
1292 		list_del(&counter->list);
1293 		kfree(counter);
1294 	}
1295 	mutex_unlock(&ctr_table->mutex);
1296 }
1297 
1298 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1299 		   union ib_gid *gid)
1300 {
1301 	struct net_device *ndev;
1302 	int ret = 0;
1303 
1304 	if (!mqp->port)
1305 		return 0;
1306 
1307 	spin_lock_bh(&mdev->iboe.lock);
1308 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1309 	dev_hold(ndev);
1310 	spin_unlock_bh(&mdev->iboe.lock);
1311 
1312 	if (ndev) {
1313 		ret = 1;
1314 		dev_put(ndev);
1315 	}
1316 
1317 	return ret;
1318 }
1319 
1320 struct mlx4_ib_steering {
1321 	struct list_head list;
1322 	struct mlx4_flow_reg_id reg_id;
1323 	union ib_gid gid;
1324 };
1325 
1326 #define LAST_ETH_FIELD vlan_tag
1327 #define LAST_IB_FIELD sl
1328 #define LAST_IPV4_FIELD dst_ip
1329 #define LAST_TCP_UDP_FIELD src_port
1330 
1331 /* Field is the last supported field */
1332 #define FIELDS_NOT_SUPPORTED(filter, field)\
1333 	memchr_inv((void *)&filter.field  +\
1334 		   sizeof(filter.field), 0,\
1335 		   sizeof(filter) -\
1336 		   offsetof(typeof(filter), field) -\
1337 		   sizeof(filter.field))
1338 
1339 static int parse_flow_attr(struct mlx4_dev *dev,
1340 			   u32 qp_num,
1341 			   union ib_flow_spec *ib_spec,
1342 			   struct _rule_hw *mlx4_spec)
1343 {
1344 	enum mlx4_net_trans_rule_id type;
1345 
1346 	switch (ib_spec->type) {
1347 	case IB_FLOW_SPEC_ETH:
1348 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1349 			return -ENOTSUPP;
1350 
1351 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1352 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1353 		       ETH_ALEN);
1354 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1355 		       ETH_ALEN);
1356 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1357 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1358 		break;
1359 	case IB_FLOW_SPEC_IB:
1360 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1361 			return -ENOTSUPP;
1362 
1363 		type = MLX4_NET_TRANS_RULE_ID_IB;
1364 		mlx4_spec->ib.l3_qpn =
1365 			cpu_to_be32(qp_num);
1366 		mlx4_spec->ib.qpn_mask =
1367 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1368 		break;
1369 
1370 
1371 	case IB_FLOW_SPEC_IPV4:
1372 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1373 			return -ENOTSUPP;
1374 
1375 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1376 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1377 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1378 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1379 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1380 		break;
1381 
1382 	case IB_FLOW_SPEC_TCP:
1383 	case IB_FLOW_SPEC_UDP:
1384 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1385 			return -ENOTSUPP;
1386 
1387 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1388 					MLX4_NET_TRANS_RULE_ID_TCP :
1389 					MLX4_NET_TRANS_RULE_ID_UDP;
1390 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1391 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1392 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1393 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1394 		break;
1395 
1396 	default:
1397 		return -EINVAL;
1398 	}
1399 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1400 	    mlx4_hw_rule_sz(dev, type) < 0)
1401 		return -EINVAL;
1402 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1403 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1404 	return mlx4_hw_rule_sz(dev, type);
1405 }
1406 
1407 struct default_rules {
1408 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1409 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1410 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1411 	__u8  link_layer;
1412 };
1413 static const struct default_rules default_table[] = {
1414 	{
1415 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1416 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1417 		.rules_create_list = {IB_FLOW_SPEC_IB},
1418 		.link_layer = IB_LINK_LAYER_INFINIBAND
1419 	}
1420 };
1421 
1422 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1423 					 struct ib_flow_attr *flow_attr)
1424 {
1425 	int i, j, k;
1426 	void *ib_flow;
1427 	const struct default_rules *pdefault_rules = default_table;
1428 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1429 
1430 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1431 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1432 		memset(&field_types, 0, sizeof(field_types));
1433 
1434 		if (link_layer != pdefault_rules->link_layer)
1435 			continue;
1436 
1437 		ib_flow = flow_attr + 1;
1438 		/* we assume the specs are sorted */
1439 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1440 		     j < flow_attr->num_of_specs; k++) {
1441 			union ib_flow_spec *current_flow =
1442 				(union ib_flow_spec *)ib_flow;
1443 
1444 			/* same layer but different type */
1445 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1446 			     (pdefault_rules->mandatory_fields[k] &
1447 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1448 			    (current_flow->type !=
1449 			     pdefault_rules->mandatory_fields[k]))
1450 				goto out;
1451 
1452 			/* same layer, try match next one */
1453 			if (current_flow->type ==
1454 			    pdefault_rules->mandatory_fields[k]) {
1455 				j++;
1456 				ib_flow +=
1457 					((union ib_flow_spec *)ib_flow)->size;
1458 			}
1459 		}
1460 
1461 		ib_flow = flow_attr + 1;
1462 		for (j = 0; j < flow_attr->num_of_specs;
1463 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1464 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1465 				/* same layer and same type */
1466 				if (((union ib_flow_spec *)ib_flow)->type ==
1467 				    pdefault_rules->mandatory_not_fields[k])
1468 					goto out;
1469 
1470 		return i;
1471 	}
1472 out:
1473 	return -1;
1474 }
1475 
1476 static int __mlx4_ib_create_default_rules(
1477 		struct mlx4_ib_dev *mdev,
1478 		struct ib_qp *qp,
1479 		const struct default_rules *pdefault_rules,
1480 		struct _rule_hw *mlx4_spec) {
1481 	int size = 0;
1482 	int i;
1483 
1484 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1485 		union ib_flow_spec ib_spec = {};
1486 		int ret;
1487 
1488 		switch (pdefault_rules->rules_create_list[i]) {
1489 		case 0:
1490 			/* no rule */
1491 			continue;
1492 		case IB_FLOW_SPEC_IB:
1493 			ib_spec.type = IB_FLOW_SPEC_IB;
1494 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1495 
1496 			break;
1497 		default:
1498 			/* invalid rule */
1499 			return -EINVAL;
1500 		}
1501 		/* We must put empty rule, qpn is being ignored */
1502 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1503 				      mlx4_spec);
1504 		if (ret < 0) {
1505 			pr_info("invalid parsing\n");
1506 			return -EINVAL;
1507 		}
1508 
1509 		mlx4_spec = (void *)mlx4_spec + ret;
1510 		size += ret;
1511 	}
1512 	return size;
1513 }
1514 
1515 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1516 			  int domain,
1517 			  enum mlx4_net_trans_promisc_mode flow_type,
1518 			  u64 *reg_id)
1519 {
1520 	int ret, i;
1521 	int size = 0;
1522 	void *ib_flow;
1523 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1524 	struct mlx4_cmd_mailbox *mailbox;
1525 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1526 	int default_flow;
1527 
1528 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1529 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1530 		return -EINVAL;
1531 	}
1532 
1533 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1534 		return -EINVAL;
1535 
1536 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1537 	if (IS_ERR(mailbox))
1538 		return PTR_ERR(mailbox);
1539 	ctrl = mailbox->buf;
1540 
1541 	ctrl->prio = cpu_to_be16(domain | flow_attr->priority);
1542 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1543 	ctrl->port = flow_attr->port;
1544 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1545 
1546 	ib_flow = flow_attr + 1;
1547 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1548 	/* Add default flows */
1549 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1550 	if (default_flow >= 0) {
1551 		ret = __mlx4_ib_create_default_rules(
1552 				mdev, qp, default_table + default_flow,
1553 				mailbox->buf + size);
1554 		if (ret < 0) {
1555 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1556 			return -EINVAL;
1557 		}
1558 		size += ret;
1559 	}
1560 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1561 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1562 				      mailbox->buf + size);
1563 		if (ret < 0) {
1564 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1565 			return -EINVAL;
1566 		}
1567 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1568 		size += ret;
1569 	}
1570 
1571 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1572 	    flow_attr->num_of_specs == 1) {
1573 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1574 		enum ib_flow_spec_type header_spec =
1575 			((union ib_flow_spec *)(flow_attr + 1))->type;
1576 
1577 		if (header_spec == IB_FLOW_SPEC_ETH)
1578 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1579 	}
1580 
1581 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1582 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1583 			   MLX4_CMD_NATIVE);
1584 	if (ret == -ENOMEM)
1585 		pr_err("mcg table is full. Fail to register network rule.\n");
1586 	else if (ret == -ENXIO)
1587 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1588 	else if (ret)
1589 		pr_err("Invalid argument. Fail to register network rule.\n");
1590 
1591 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1592 	return ret;
1593 }
1594 
1595 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1596 {
1597 	int err;
1598 	err = mlx4_cmd(dev, reg_id, 0, 0,
1599 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1600 		       MLX4_CMD_NATIVE);
1601 	if (err)
1602 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1603 		       reg_id);
1604 	return err;
1605 }
1606 
1607 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1608 				    u64 *reg_id)
1609 {
1610 	void *ib_flow;
1611 	union ib_flow_spec *ib_spec;
1612 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1613 	int err = 0;
1614 
1615 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1616 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1617 		return 0; /* do nothing */
1618 
1619 	ib_flow = flow_attr + 1;
1620 	ib_spec = (union ib_flow_spec *)ib_flow;
1621 
1622 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1623 		return 0; /* do nothing */
1624 
1625 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1626 				    flow_attr->port, qp->qp_num,
1627 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1628 				    reg_id);
1629 	return err;
1630 }
1631 
1632 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1633 				      struct ib_flow_attr *flow_attr,
1634 				      enum mlx4_net_trans_promisc_mode *type)
1635 {
1636 	int err = 0;
1637 
1638 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1639 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1640 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1641 		return -EOPNOTSUPP;
1642 	}
1643 
1644 	if (flow_attr->num_of_specs == 0) {
1645 		type[0] = MLX4_FS_MC_SNIFFER;
1646 		type[1] = MLX4_FS_UC_SNIFFER;
1647 	} else {
1648 		union ib_flow_spec *ib_spec;
1649 
1650 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1651 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1652 			return -EINVAL;
1653 
1654 		/* if all is zero than MC and UC */
1655 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1656 			type[0] = MLX4_FS_MC_SNIFFER;
1657 			type[1] = MLX4_FS_UC_SNIFFER;
1658 		} else {
1659 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1660 					    ib_spec->eth.mask.dst_mac[1],
1661 					    ib_spec->eth.mask.dst_mac[2],
1662 					    ib_spec->eth.mask.dst_mac[3],
1663 					    ib_spec->eth.mask.dst_mac[4],
1664 					    ib_spec->eth.mask.dst_mac[5]};
1665 
1666 			/* Above xor was only on MC bit, non empty mask is valid
1667 			 * only if this bit is set and rest are zero.
1668 			 */
1669 			if (!is_zero_ether_addr(&mac[0]))
1670 				return -EINVAL;
1671 
1672 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1673 				type[0] = MLX4_FS_MC_SNIFFER;
1674 			else
1675 				type[0] = MLX4_FS_UC_SNIFFER;
1676 		}
1677 	}
1678 
1679 	return err;
1680 }
1681 
1682 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1683 					   struct ib_flow_attr *flow_attr,
1684 					   struct ib_udata *udata)
1685 {
1686 	int err = 0, i = 0, j = 0;
1687 	struct mlx4_ib_flow *mflow;
1688 	enum mlx4_net_trans_promisc_mode type[2];
1689 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1690 	int is_bonded = mlx4_is_bonded(dev);
1691 
1692 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1693 		return ERR_PTR(-EOPNOTSUPP);
1694 
1695 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1696 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1697 		return ERR_PTR(-EOPNOTSUPP);
1698 
1699 	if (udata &&
1700 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1701 		return ERR_PTR(-EOPNOTSUPP);
1702 
1703 	memset(type, 0, sizeof(type));
1704 
1705 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1706 	if (!mflow) {
1707 		err = -ENOMEM;
1708 		goto err_free;
1709 	}
1710 
1711 	switch (flow_attr->type) {
1712 	case IB_FLOW_ATTR_NORMAL:
1713 		/* If dont trap flag (continue match) is set, under specific
1714 		 * condition traffic be replicated to given qp,
1715 		 * without stealing it
1716 		 */
1717 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1718 			err = mlx4_ib_add_dont_trap_rule(dev,
1719 							 flow_attr,
1720 							 type);
1721 			if (err)
1722 				goto err_free;
1723 		} else {
1724 			type[0] = MLX4_FS_REGULAR;
1725 		}
1726 		break;
1727 
1728 	case IB_FLOW_ATTR_ALL_DEFAULT:
1729 		type[0] = MLX4_FS_ALL_DEFAULT;
1730 		break;
1731 
1732 	case IB_FLOW_ATTR_MC_DEFAULT:
1733 		type[0] = MLX4_FS_MC_DEFAULT;
1734 		break;
1735 
1736 	case IB_FLOW_ATTR_SNIFFER:
1737 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1738 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1739 		break;
1740 
1741 	default:
1742 		err = -EINVAL;
1743 		goto err_free;
1744 	}
1745 
1746 	while (i < ARRAY_SIZE(type) && type[i]) {
1747 		err = __mlx4_ib_create_flow(qp, flow_attr, MLX4_DOMAIN_UVERBS,
1748 					    type[i], &mflow->reg_id[i].id);
1749 		if (err)
1750 			goto err_create_flow;
1751 		if (is_bonded) {
1752 			/* Application always sees one port so the mirror rule
1753 			 * must be on port #2
1754 			 */
1755 			flow_attr->port = 2;
1756 			err = __mlx4_ib_create_flow(qp, flow_attr,
1757 						    MLX4_DOMAIN_UVERBS, type[j],
1758 						    &mflow->reg_id[j].mirror);
1759 			flow_attr->port = 1;
1760 			if (err)
1761 				goto err_create_flow;
1762 			j++;
1763 		}
1764 
1765 		i++;
1766 	}
1767 
1768 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1769 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1770 					       &mflow->reg_id[i].id);
1771 		if (err)
1772 			goto err_create_flow;
1773 
1774 		if (is_bonded) {
1775 			flow_attr->port = 2;
1776 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1777 						       &mflow->reg_id[j].mirror);
1778 			flow_attr->port = 1;
1779 			if (err)
1780 				goto err_create_flow;
1781 			j++;
1782 		}
1783 		/* function to create mirror rule */
1784 		i++;
1785 	}
1786 
1787 	return &mflow->ibflow;
1788 
1789 err_create_flow:
1790 	while (i) {
1791 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1792 					     mflow->reg_id[i].id);
1793 		i--;
1794 	}
1795 
1796 	while (j) {
1797 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1798 					     mflow->reg_id[j].mirror);
1799 		j--;
1800 	}
1801 err_free:
1802 	kfree(mflow);
1803 	return ERR_PTR(err);
1804 }
1805 
1806 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1807 {
1808 	int err, ret = 0;
1809 	int i = 0;
1810 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1811 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1812 
1813 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1814 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1815 		if (err)
1816 			ret = err;
1817 		if (mflow->reg_id[i].mirror) {
1818 			err = __mlx4_ib_destroy_flow(mdev->dev,
1819 						     mflow->reg_id[i].mirror);
1820 			if (err)
1821 				ret = err;
1822 		}
1823 		i++;
1824 	}
1825 
1826 	kfree(mflow);
1827 	return ret;
1828 }
1829 
1830 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1831 {
1832 	int err;
1833 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1834 	struct mlx4_dev	*dev = mdev->dev;
1835 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1836 	struct mlx4_ib_steering *ib_steering = NULL;
1837 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1838 	struct mlx4_flow_reg_id	reg_id;
1839 
1840 	if (mdev->dev->caps.steering_mode ==
1841 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1842 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1843 		if (!ib_steering)
1844 			return -ENOMEM;
1845 	}
1846 
1847 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1848 				    !!(mqp->flags &
1849 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1850 				    prot, &reg_id.id);
1851 	if (err) {
1852 		pr_err("multicast attach op failed, err %d\n", err);
1853 		goto err_malloc;
1854 	}
1855 
1856 	reg_id.mirror = 0;
1857 	if (mlx4_is_bonded(dev)) {
1858 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1859 					    (mqp->port == 1) ? 2 : 1,
1860 					    !!(mqp->flags &
1861 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1862 					    prot, &reg_id.mirror);
1863 		if (err)
1864 			goto err_add;
1865 	}
1866 
1867 	err = add_gid_entry(ibqp, gid);
1868 	if (err)
1869 		goto err_add;
1870 
1871 	if (ib_steering) {
1872 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1873 		ib_steering->reg_id = reg_id;
1874 		mutex_lock(&mqp->mutex);
1875 		list_add(&ib_steering->list, &mqp->steering_rules);
1876 		mutex_unlock(&mqp->mutex);
1877 	}
1878 	return 0;
1879 
1880 err_add:
1881 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1882 			      prot, reg_id.id);
1883 	if (reg_id.mirror)
1884 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1885 				      prot, reg_id.mirror);
1886 err_malloc:
1887 	kfree(ib_steering);
1888 
1889 	return err;
1890 }
1891 
1892 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1893 {
1894 	struct mlx4_ib_gid_entry *ge;
1895 	struct mlx4_ib_gid_entry *tmp;
1896 	struct mlx4_ib_gid_entry *ret = NULL;
1897 
1898 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1899 		if (!memcmp(raw, ge->gid.raw, 16)) {
1900 			ret = ge;
1901 			break;
1902 		}
1903 	}
1904 
1905 	return ret;
1906 }
1907 
1908 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1909 {
1910 	int err;
1911 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1912 	struct mlx4_dev *dev = mdev->dev;
1913 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1914 	struct net_device *ndev;
1915 	struct mlx4_ib_gid_entry *ge;
1916 	struct mlx4_flow_reg_id reg_id = {0, 0};
1917 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1918 
1919 	if (mdev->dev->caps.steering_mode ==
1920 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1921 		struct mlx4_ib_steering *ib_steering;
1922 
1923 		mutex_lock(&mqp->mutex);
1924 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1925 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1926 				list_del(&ib_steering->list);
1927 				break;
1928 			}
1929 		}
1930 		mutex_unlock(&mqp->mutex);
1931 		if (&ib_steering->list == &mqp->steering_rules) {
1932 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1933 			return -EINVAL;
1934 		}
1935 		reg_id = ib_steering->reg_id;
1936 		kfree(ib_steering);
1937 	}
1938 
1939 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1940 				    prot, reg_id.id);
1941 	if (err)
1942 		return err;
1943 
1944 	if (mlx4_is_bonded(dev)) {
1945 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1946 					    prot, reg_id.mirror);
1947 		if (err)
1948 			return err;
1949 	}
1950 
1951 	mutex_lock(&mqp->mutex);
1952 	ge = find_gid_entry(mqp, gid->raw);
1953 	if (ge) {
1954 		spin_lock_bh(&mdev->iboe.lock);
1955 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1956 		dev_hold(ndev);
1957 		spin_unlock_bh(&mdev->iboe.lock);
1958 		dev_put(ndev);
1959 		list_del(&ge->list);
1960 		kfree(ge);
1961 	} else
1962 		pr_warn("could not find mgid entry\n");
1963 
1964 	mutex_unlock(&mqp->mutex);
1965 
1966 	return 0;
1967 }
1968 
1969 static int init_node_data(struct mlx4_ib_dev *dev)
1970 {
1971 	struct ib_smp *in_mad  = NULL;
1972 	struct ib_smp *out_mad = NULL;
1973 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1974 	int err = -ENOMEM;
1975 
1976 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1977 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1978 	if (!in_mad || !out_mad)
1979 		goto out;
1980 
1981 	ib_init_query_mad(in_mad);
1982 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1983 	if (mlx4_is_master(dev->dev))
1984 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1985 
1986 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1987 	if (err)
1988 		goto out;
1989 
1990 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1991 
1992 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1993 
1994 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1995 	if (err)
1996 		goto out;
1997 
1998 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1999 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2000 
2001 out:
2002 	kfree(in_mad);
2003 	kfree(out_mad);
2004 	return err;
2005 }
2006 
2007 static ssize_t hca_type_show(struct device *device,
2008 			     struct device_attribute *attr, char *buf)
2009 {
2010 	struct mlx4_ib_dev *dev =
2011 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2012 
2013 	return sysfs_emit(buf, "MT%d\n", dev->dev->persist->pdev->device);
2014 }
2015 static DEVICE_ATTR_RO(hca_type);
2016 
2017 static ssize_t hw_rev_show(struct device *device,
2018 			   struct device_attribute *attr, char *buf)
2019 {
2020 	struct mlx4_ib_dev *dev =
2021 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2022 
2023 	return sysfs_emit(buf, "%x\n", dev->dev->rev_id);
2024 }
2025 static DEVICE_ATTR_RO(hw_rev);
2026 
2027 static ssize_t board_id_show(struct device *device,
2028 			     struct device_attribute *attr, char *buf)
2029 {
2030 	struct mlx4_ib_dev *dev =
2031 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2032 
2033 	return sysfs_emit(buf, "%.*s\n", MLX4_BOARD_ID_LEN, dev->dev->board_id);
2034 }
2035 static DEVICE_ATTR_RO(board_id);
2036 
2037 static struct attribute *mlx4_class_attributes[] = {
2038 	&dev_attr_hw_rev.attr,
2039 	&dev_attr_hca_type.attr,
2040 	&dev_attr_board_id.attr,
2041 	NULL
2042 };
2043 
2044 static const struct attribute_group mlx4_attr_group = {
2045 	.attrs = mlx4_class_attributes,
2046 };
2047 
2048 struct diag_counter {
2049 	const char *name;
2050 	u32 offset;
2051 };
2052 
2053 #define DIAG_COUNTER(_name, _offset)			\
2054 	{ .name = #_name, .offset = _offset }
2055 
2056 static const struct diag_counter diag_basic[] = {
2057 	DIAG_COUNTER(rq_num_lle, 0x00),
2058 	DIAG_COUNTER(sq_num_lle, 0x04),
2059 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2060 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2061 	DIAG_COUNTER(rq_num_lpe, 0x18),
2062 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2063 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2064 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2065 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2066 	DIAG_COUNTER(sq_num_bre, 0x34),
2067 	DIAG_COUNTER(sq_num_rire, 0x44),
2068 	DIAG_COUNTER(rq_num_rire, 0x48),
2069 	DIAG_COUNTER(sq_num_rae, 0x4C),
2070 	DIAG_COUNTER(rq_num_rae, 0x50),
2071 	DIAG_COUNTER(sq_num_roe, 0x54),
2072 	DIAG_COUNTER(sq_num_tree, 0x5C),
2073 	DIAG_COUNTER(sq_num_rree, 0x64),
2074 	DIAG_COUNTER(rq_num_rnr, 0x68),
2075 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2076 	DIAG_COUNTER(rq_num_oos, 0x100),
2077 	DIAG_COUNTER(sq_num_oos, 0x104),
2078 };
2079 
2080 static const struct diag_counter diag_ext[] = {
2081 	DIAG_COUNTER(rq_num_dup, 0x130),
2082 	DIAG_COUNTER(sq_num_to, 0x134),
2083 };
2084 
2085 static const struct diag_counter diag_device_only[] = {
2086 	DIAG_COUNTER(num_cqovf, 0x1A0),
2087 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2088 };
2089 
2090 static struct rdma_hw_stats *
2091 mlx4_ib_alloc_hw_device_stats(struct ib_device *ibdev)
2092 {
2093 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2094 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2095 
2096 	if (!diag[0].descs)
2097 		return NULL;
2098 
2099 	return rdma_alloc_hw_stats_struct(diag[0].descs, diag[0].num_counters,
2100 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2101 }
2102 
2103 static struct rdma_hw_stats *
2104 mlx4_ib_alloc_hw_port_stats(struct ib_device *ibdev, u32 port_num)
2105 {
2106 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2107 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2108 
2109 	if (!diag[1].descs)
2110 		return NULL;
2111 
2112 	return rdma_alloc_hw_stats_struct(diag[1].descs, diag[1].num_counters,
2113 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2114 }
2115 
2116 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2117 				struct rdma_hw_stats *stats,
2118 				u32 port, int index)
2119 {
2120 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2121 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2122 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2123 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2124 	int ret;
2125 	int i;
2126 
2127 	ret = mlx4_query_diag_counters(dev->dev,
2128 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2129 				       diag[!!port].offset, hw_value,
2130 				       diag[!!port].num_counters, port);
2131 
2132 	if (ret)
2133 		return ret;
2134 
2135 	for (i = 0; i < diag[!!port].num_counters; i++)
2136 		stats->value[i] = hw_value[i];
2137 
2138 	return diag[!!port].num_counters;
2139 }
2140 
2141 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2142 					 struct rdma_stat_desc **pdescs,
2143 					 u32 **offset, u32 *num, bool port)
2144 {
2145 	u32 num_counters;
2146 
2147 	num_counters = ARRAY_SIZE(diag_basic);
2148 
2149 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2150 		num_counters += ARRAY_SIZE(diag_ext);
2151 
2152 	if (!port)
2153 		num_counters += ARRAY_SIZE(diag_device_only);
2154 
2155 	*pdescs = kcalloc(num_counters, sizeof(struct rdma_stat_desc),
2156 			  GFP_KERNEL);
2157 	if (!*pdescs)
2158 		return -ENOMEM;
2159 
2160 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2161 	if (!*offset)
2162 		goto err;
2163 
2164 	*num = num_counters;
2165 
2166 	return 0;
2167 
2168 err:
2169 	kfree(*pdescs);
2170 	return -ENOMEM;
2171 }
2172 
2173 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2174 				       struct rdma_stat_desc *descs,
2175 				       u32 *offset, bool port)
2176 {
2177 	int i;
2178 	int j;
2179 
2180 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2181 		descs[i].name = diag_basic[i].name;
2182 		offset[i] = diag_basic[i].offset;
2183 	}
2184 
2185 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2186 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2187 			descs[j].name = diag_ext[i].name;
2188 			offset[j] = diag_ext[i].offset;
2189 		}
2190 	}
2191 
2192 	if (!port) {
2193 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2194 			descs[j].name = diag_device_only[i].name;
2195 			offset[j] = diag_device_only[i].offset;
2196 		}
2197 	}
2198 }
2199 
2200 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2201 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2202 	.alloc_hw_port_stats = mlx4_ib_alloc_hw_port_stats,
2203 	.get_hw_stats = mlx4_ib_get_hw_stats,
2204 };
2205 
2206 static const struct ib_device_ops mlx4_ib_hw_stats_ops1 = {
2207 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2208 	.get_hw_stats = mlx4_ib_get_hw_stats,
2209 };
2210 
2211 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2212 {
2213 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2214 	int i;
2215 	int ret;
2216 	bool per_port = !!(ibdev->dev->caps.flags2 &
2217 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2218 
2219 	if (mlx4_is_slave(ibdev->dev))
2220 		return 0;
2221 
2222 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2223 		/*
2224 		 * i == 1 means we are building port counters, set a different
2225 		 * stats ops without port stats callback.
2226 		 */
2227 		if (i && !per_port) {
2228 			ib_set_device_ops(&ibdev->ib_dev,
2229 					  &mlx4_ib_hw_stats_ops1);
2230 
2231 			return 0;
2232 		}
2233 
2234 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].descs,
2235 						    &diag[i].offset,
2236 						    &diag[i].num_counters, i);
2237 		if (ret)
2238 			goto err_alloc;
2239 
2240 		mlx4_ib_fill_diag_counters(ibdev, diag[i].descs,
2241 					   diag[i].offset, i);
2242 	}
2243 
2244 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2245 
2246 	return 0;
2247 
2248 err_alloc:
2249 	if (i) {
2250 		kfree(diag[i - 1].descs);
2251 		kfree(diag[i - 1].offset);
2252 	}
2253 
2254 	return ret;
2255 }
2256 
2257 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2258 {
2259 	int i;
2260 
2261 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2262 		kfree(ibdev->diag_counters[i].offset);
2263 		kfree(ibdev->diag_counters[i].descs);
2264 	}
2265 }
2266 
2267 #define MLX4_IB_INVALID_MAC	((u64)-1)
2268 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2269 			       struct net_device *dev,
2270 			       int port)
2271 {
2272 	u64 new_smac = 0;
2273 	u64 release_mac = MLX4_IB_INVALID_MAC;
2274 	struct mlx4_ib_qp *qp;
2275 
2276 	new_smac = ether_addr_to_u64(dev->dev_addr);
2277 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2278 
2279 	/* no need for update QP1 and mac registration in non-SRIOV */
2280 	if (!mlx4_is_mfunc(ibdev->dev))
2281 		return;
2282 
2283 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2284 	qp = ibdev->qp1_proxy[port - 1];
2285 	if (qp) {
2286 		int new_smac_index;
2287 		u64 old_smac;
2288 		struct mlx4_update_qp_params update_params;
2289 
2290 		mutex_lock(&qp->mutex);
2291 		old_smac = qp->pri.smac;
2292 		if (new_smac == old_smac)
2293 			goto unlock;
2294 
2295 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2296 
2297 		if (new_smac_index < 0)
2298 			goto unlock;
2299 
2300 		update_params.smac_index = new_smac_index;
2301 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2302 				   &update_params)) {
2303 			release_mac = new_smac;
2304 			goto unlock;
2305 		}
2306 		/* if old port was zero, no mac was yet registered for this QP */
2307 		if (qp->pri.smac_port)
2308 			release_mac = old_smac;
2309 		qp->pri.smac = new_smac;
2310 		qp->pri.smac_port = port;
2311 		qp->pri.smac_index = new_smac_index;
2312 	}
2313 
2314 unlock:
2315 	if (release_mac != MLX4_IB_INVALID_MAC)
2316 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2317 	if (qp)
2318 		mutex_unlock(&qp->mutex);
2319 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2320 }
2321 
2322 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2323 				 struct net_device *dev,
2324 				 unsigned long event)
2325 
2326 {
2327 	struct mlx4_ib_iboe *iboe;
2328 	int update_qps_port = -1;
2329 	int port;
2330 
2331 	ASSERT_RTNL();
2332 
2333 	iboe = &ibdev->iboe;
2334 
2335 	spin_lock_bh(&iboe->lock);
2336 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2337 
2338 		iboe->netdevs[port - 1] =
2339 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2340 
2341 		if (dev == iboe->netdevs[port - 1] &&
2342 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2343 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2344 			update_qps_port = port;
2345 
2346 		if (dev == iboe->netdevs[port - 1] &&
2347 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2348 			enum ib_port_state port_state;
2349 			struct ib_event ibev = { };
2350 
2351 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2352 						     &port_state))
2353 				continue;
2354 
2355 			if (event == NETDEV_UP &&
2356 			    (port_state != IB_PORT_ACTIVE ||
2357 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2358 				continue;
2359 			if (event == NETDEV_DOWN &&
2360 			    (port_state != IB_PORT_DOWN ||
2361 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2362 				continue;
2363 			iboe->last_port_state[port - 1] = port_state;
2364 
2365 			ibev.device = &ibdev->ib_dev;
2366 			ibev.element.port_num = port;
2367 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2368 							  IB_EVENT_PORT_ERR;
2369 			ib_dispatch_event(&ibev);
2370 		}
2371 
2372 	}
2373 	spin_unlock_bh(&iboe->lock);
2374 
2375 	if (update_qps_port > 0)
2376 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2377 }
2378 
2379 static int mlx4_ib_netdev_event(struct notifier_block *this,
2380 				unsigned long event, void *ptr)
2381 {
2382 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2383 	struct mlx4_ib_dev *ibdev;
2384 
2385 	if (!net_eq(dev_net(dev), &init_net))
2386 		return NOTIFY_DONE;
2387 
2388 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2389 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2390 
2391 	return NOTIFY_DONE;
2392 }
2393 
2394 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2395 {
2396 	int port;
2397 	int slave;
2398 	int i;
2399 
2400 	if (mlx4_is_master(ibdev->dev)) {
2401 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2402 		     ++slave) {
2403 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2404 				for (i = 0;
2405 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2406 				     ++i) {
2407 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2408 					/* master has the identity virt2phys pkey mapping */
2409 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2410 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2411 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2412 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2413 				}
2414 			}
2415 		}
2416 		/* initialize pkey cache */
2417 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2418 			for (i = 0;
2419 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2420 			     ++i)
2421 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2422 					(i) ? 0 : 0xFFFF;
2423 		}
2424 	}
2425 }
2426 
2427 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2428 {
2429 	int i, j, eq = 0, total_eqs = 0;
2430 
2431 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2432 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2433 	if (!ibdev->eq_table)
2434 		return;
2435 
2436 	for (i = 1; i <= dev->caps.num_ports; i++) {
2437 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2438 		     j++, total_eqs++) {
2439 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2440 				continue;
2441 			ibdev->eq_table[eq] = total_eqs;
2442 			if (!mlx4_assign_eq(dev, i,
2443 					    &ibdev->eq_table[eq]))
2444 				eq++;
2445 			else
2446 				ibdev->eq_table[eq] = -1;
2447 		}
2448 	}
2449 
2450 	for (i = eq; i < dev->caps.num_comp_vectors;
2451 	     ibdev->eq_table[i++] = -1)
2452 		;
2453 
2454 	/* Advertise the new number of EQs to clients */
2455 	ibdev->ib_dev.num_comp_vectors = eq;
2456 }
2457 
2458 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2459 {
2460 	int i;
2461 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2462 
2463 	/* no eqs were allocated */
2464 	if (!ibdev->eq_table)
2465 		return;
2466 
2467 	/* Reset the advertised EQ number */
2468 	ibdev->ib_dev.num_comp_vectors = 0;
2469 
2470 	for (i = 0; i < total_eqs; i++)
2471 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2472 
2473 	kfree(ibdev->eq_table);
2474 	ibdev->eq_table = NULL;
2475 }
2476 
2477 static int mlx4_port_immutable(struct ib_device *ibdev, u32 port_num,
2478 			       struct ib_port_immutable *immutable)
2479 {
2480 	struct ib_port_attr attr;
2481 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2482 	int err;
2483 
2484 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2485 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2486 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2487 	} else {
2488 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2489 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2490 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2491 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2492 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2493 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2494 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2495 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2496 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2497 	}
2498 
2499 	err = ib_query_port(ibdev, port_num, &attr);
2500 	if (err)
2501 		return err;
2502 
2503 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2504 	immutable->gid_tbl_len = attr.gid_tbl_len;
2505 
2506 	return 0;
2507 }
2508 
2509 static void get_fw_ver_str(struct ib_device *device, char *str)
2510 {
2511 	struct mlx4_ib_dev *dev =
2512 		container_of(device, struct mlx4_ib_dev, ib_dev);
2513 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2514 		 (int) (dev->dev->caps.fw_ver >> 32),
2515 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2516 		 (int) dev->dev->caps.fw_ver & 0xffff);
2517 }
2518 
2519 static const struct ib_device_ops mlx4_ib_dev_ops = {
2520 	.owner = THIS_MODULE,
2521 	.driver_id = RDMA_DRIVER_MLX4,
2522 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2523 
2524 	.add_gid = mlx4_ib_add_gid,
2525 	.alloc_mr = mlx4_ib_alloc_mr,
2526 	.alloc_pd = mlx4_ib_alloc_pd,
2527 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2528 	.attach_mcast = mlx4_ib_mcg_attach,
2529 	.create_ah = mlx4_ib_create_ah,
2530 	.create_cq = mlx4_ib_create_cq,
2531 	.create_qp = mlx4_ib_create_qp,
2532 	.create_srq = mlx4_ib_create_srq,
2533 	.dealloc_pd = mlx4_ib_dealloc_pd,
2534 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2535 	.del_gid = mlx4_ib_del_gid,
2536 	.dereg_mr = mlx4_ib_dereg_mr,
2537 	.destroy_ah = mlx4_ib_destroy_ah,
2538 	.destroy_cq = mlx4_ib_destroy_cq,
2539 	.destroy_qp = mlx4_ib_destroy_qp,
2540 	.destroy_srq = mlx4_ib_destroy_srq,
2541 	.detach_mcast = mlx4_ib_mcg_detach,
2542 	.device_group = &mlx4_attr_group,
2543 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2544 	.drain_rq = mlx4_ib_drain_rq,
2545 	.drain_sq = mlx4_ib_drain_sq,
2546 	.get_dev_fw_str = get_fw_ver_str,
2547 	.get_dma_mr = mlx4_ib_get_dma_mr,
2548 	.get_link_layer = mlx4_ib_port_link_layer,
2549 	.get_netdev = mlx4_ib_get_netdev,
2550 	.get_port_immutable = mlx4_port_immutable,
2551 	.map_mr_sg = mlx4_ib_map_mr_sg,
2552 	.mmap = mlx4_ib_mmap,
2553 	.modify_cq = mlx4_ib_modify_cq,
2554 	.modify_device = mlx4_ib_modify_device,
2555 	.modify_port = mlx4_ib_modify_port,
2556 	.modify_qp = mlx4_ib_modify_qp,
2557 	.modify_srq = mlx4_ib_modify_srq,
2558 	.poll_cq = mlx4_ib_poll_cq,
2559 	.post_recv = mlx4_ib_post_recv,
2560 	.post_send = mlx4_ib_post_send,
2561 	.post_srq_recv = mlx4_ib_post_srq_recv,
2562 	.process_mad = mlx4_ib_process_mad,
2563 	.query_ah = mlx4_ib_query_ah,
2564 	.query_device = mlx4_ib_query_device,
2565 	.query_gid = mlx4_ib_query_gid,
2566 	.query_pkey = mlx4_ib_query_pkey,
2567 	.query_port = mlx4_ib_query_port,
2568 	.query_qp = mlx4_ib_query_qp,
2569 	.query_srq = mlx4_ib_query_srq,
2570 	.reg_user_mr = mlx4_ib_reg_user_mr,
2571 	.req_notify_cq = mlx4_ib_arm_cq,
2572 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2573 	.resize_cq = mlx4_ib_resize_cq,
2574 
2575 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2576 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2577 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2578 	INIT_RDMA_OBJ_SIZE(ib_qp, mlx4_ib_qp, ibqp),
2579 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2580 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2581 };
2582 
2583 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2584 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2585 	.create_wq = mlx4_ib_create_wq,
2586 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2587 	.destroy_wq = mlx4_ib_destroy_wq,
2588 	.modify_wq = mlx4_ib_modify_wq,
2589 
2590 	INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx4_ib_rwq_ind_table,
2591 			   ib_rwq_ind_tbl),
2592 };
2593 
2594 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2595 	.alloc_mw = mlx4_ib_alloc_mw,
2596 	.dealloc_mw = mlx4_ib_dealloc_mw,
2597 
2598 	INIT_RDMA_OBJ_SIZE(ib_mw, mlx4_ib_mw, ibmw),
2599 };
2600 
2601 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2602 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2603 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2604 
2605 	INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2606 };
2607 
2608 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2609 	.create_flow = mlx4_ib_create_flow,
2610 	.destroy_flow = mlx4_ib_destroy_flow,
2611 };
2612 
2613 static void *mlx4_ib_add(struct mlx4_dev *dev)
2614 {
2615 	struct mlx4_ib_dev *ibdev;
2616 	int num_ports = 0;
2617 	int i, j;
2618 	int err;
2619 	struct mlx4_ib_iboe *iboe;
2620 	int ib_num_ports = 0;
2621 	int num_req_counters;
2622 	int allocated;
2623 	u32 counter_index;
2624 	struct counter_index *new_counter_index = NULL;
2625 
2626 	pr_info_once("%s", mlx4_ib_version);
2627 
2628 	num_ports = 0;
2629 	mlx4_foreach_ib_transport_port(i, dev)
2630 		num_ports++;
2631 
2632 	/* No point in registering a device with no ports... */
2633 	if (num_ports == 0)
2634 		return NULL;
2635 
2636 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2637 	if (!ibdev) {
2638 		dev_err(&dev->persist->pdev->dev,
2639 			"Device struct alloc failed\n");
2640 		return NULL;
2641 	}
2642 
2643 	iboe = &ibdev->iboe;
2644 
2645 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2646 		goto err_dealloc;
2647 
2648 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2649 		goto err_pd;
2650 
2651 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2652 				 PAGE_SIZE);
2653 	if (!ibdev->uar_map)
2654 		goto err_uar;
2655 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2656 
2657 	ibdev->dev = dev;
2658 	ibdev->bond_next_port	= 0;
2659 
2660 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2661 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2662 	ibdev->num_ports		= num_ports;
2663 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2664 						1 : ibdev->num_ports;
2665 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2666 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2667 
2668 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2669 
2670 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2671 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2672 	    IB_LINK_LAYER_ETHERNET) ||
2673 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2674 	    IB_LINK_LAYER_ETHERNET)))
2675 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2676 
2677 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2678 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2679 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2680 
2681 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2682 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2683 	}
2684 
2685 	if (check_flow_steering_support(dev)) {
2686 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2687 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2688 	}
2689 
2690 	if (!dev->caps.userspace_caps)
2691 		ibdev->ib_dev.ops.uverbs_abi_ver =
2692 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2693 
2694 	mlx4_ib_alloc_eqs(dev, ibdev);
2695 
2696 	spin_lock_init(&iboe->lock);
2697 
2698 	if (init_node_data(ibdev))
2699 		goto err_map;
2700 	mlx4_init_sl2vl_tbl(ibdev);
2701 
2702 	for (i = 0; i < ibdev->num_ports; ++i) {
2703 		mutex_init(&ibdev->counters_table[i].mutex);
2704 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2705 		iboe->last_port_state[i] = IB_PORT_DOWN;
2706 	}
2707 
2708 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2709 	for (i = 0; i < num_req_counters; ++i) {
2710 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2711 		allocated = 0;
2712 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2713 						IB_LINK_LAYER_ETHERNET) {
2714 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2715 						 MLX4_RES_USAGE_DRIVER);
2716 			/* if failed to allocate a new counter, use default */
2717 			if (err)
2718 				counter_index =
2719 					mlx4_get_default_counter_index(dev,
2720 								       i + 1);
2721 			else
2722 				allocated = 1;
2723 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2724 			counter_index = mlx4_get_default_counter_index(dev,
2725 								       i + 1);
2726 		}
2727 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2728 					    GFP_KERNEL);
2729 		if (!new_counter_index) {
2730 			if (allocated)
2731 				mlx4_counter_free(ibdev->dev, counter_index);
2732 			goto err_counter;
2733 		}
2734 		new_counter_index->index = counter_index;
2735 		new_counter_index->allocated = allocated;
2736 		list_add_tail(&new_counter_index->list,
2737 			      &ibdev->counters_table[i].counters_list);
2738 		ibdev->counters_table[i].default_counter = counter_index;
2739 		pr_info("counter index %d for port %d allocated %d\n",
2740 			counter_index, i + 1, allocated);
2741 	}
2742 	if (mlx4_is_bonded(dev))
2743 		for (i = 1; i < ibdev->num_ports ; ++i) {
2744 			new_counter_index =
2745 					kmalloc(sizeof(struct counter_index),
2746 						GFP_KERNEL);
2747 			if (!new_counter_index)
2748 				goto err_counter;
2749 			new_counter_index->index = counter_index;
2750 			new_counter_index->allocated = 0;
2751 			list_add_tail(&new_counter_index->list,
2752 				      &ibdev->counters_table[i].counters_list);
2753 			ibdev->counters_table[i].default_counter =
2754 								counter_index;
2755 		}
2756 
2757 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2758 		ib_num_ports++;
2759 
2760 	spin_lock_init(&ibdev->sm_lock);
2761 	mutex_init(&ibdev->cap_mask_mutex);
2762 	INIT_LIST_HEAD(&ibdev->qp_list);
2763 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2764 
2765 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2766 	    ib_num_ports) {
2767 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2768 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2769 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2770 					    &ibdev->steer_qpn_base, 0,
2771 					    MLX4_RES_USAGE_DRIVER);
2772 		if (err)
2773 			goto err_counter;
2774 
2775 		ibdev->ib_uc_qpns_bitmap = bitmap_alloc(ibdev->steer_qpn_count,
2776 							GFP_KERNEL);
2777 		if (!ibdev->ib_uc_qpns_bitmap)
2778 			goto err_steer_qp_release;
2779 
2780 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2781 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2782 				    ibdev->steer_qpn_count);
2783 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2784 					dev, ibdev->steer_qpn_base,
2785 					ibdev->steer_qpn_base +
2786 					ibdev->steer_qpn_count - 1);
2787 			if (err)
2788 				goto err_steer_free_bitmap;
2789 		} else {
2790 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2791 				    ibdev->steer_qpn_count);
2792 		}
2793 	}
2794 
2795 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2796 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2797 
2798 	if (mlx4_ib_alloc_diag_counters(ibdev))
2799 		goto err_steer_free_bitmap;
2800 
2801 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d",
2802 			       &dev->persist->pdev->dev))
2803 		goto err_diag_counters;
2804 
2805 	if (mlx4_ib_mad_init(ibdev))
2806 		goto err_reg;
2807 
2808 	if (mlx4_ib_init_sriov(ibdev))
2809 		goto err_mad;
2810 
2811 	if (!iboe->nb.notifier_call) {
2812 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2813 		err = register_netdevice_notifier(&iboe->nb);
2814 		if (err) {
2815 			iboe->nb.notifier_call = NULL;
2816 			goto err_notif;
2817 		}
2818 	}
2819 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2820 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2821 		if (err)
2822 			goto err_notif;
2823 	}
2824 
2825 	ibdev->ib_active = true;
2826 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2827 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2828 					 &ibdev->ib_dev);
2829 
2830 	if (mlx4_is_mfunc(ibdev->dev))
2831 		init_pkeys(ibdev);
2832 
2833 	/* create paravirt contexts for any VFs which are active */
2834 	if (mlx4_is_master(ibdev->dev)) {
2835 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2836 			if (j == mlx4_master_func_num(ibdev->dev))
2837 				continue;
2838 			if (mlx4_is_slave_active(ibdev->dev, j))
2839 				do_slave_init(ibdev, j, 1);
2840 		}
2841 	}
2842 	return ibdev;
2843 
2844 err_notif:
2845 	if (ibdev->iboe.nb.notifier_call) {
2846 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2847 			pr_warn("failure unregistering notifier\n");
2848 		ibdev->iboe.nb.notifier_call = NULL;
2849 	}
2850 	flush_workqueue(wq);
2851 
2852 	mlx4_ib_close_sriov(ibdev);
2853 
2854 err_mad:
2855 	mlx4_ib_mad_cleanup(ibdev);
2856 
2857 err_reg:
2858 	ib_unregister_device(&ibdev->ib_dev);
2859 
2860 err_diag_counters:
2861 	mlx4_ib_diag_cleanup(ibdev);
2862 
2863 err_steer_free_bitmap:
2864 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2865 
2866 err_steer_qp_release:
2867 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2868 			      ibdev->steer_qpn_count);
2869 err_counter:
2870 	for (i = 0; i < ibdev->num_ports; ++i)
2871 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2872 
2873 err_map:
2874 	mlx4_ib_free_eqs(dev, ibdev);
2875 	iounmap(ibdev->uar_map);
2876 
2877 err_uar:
2878 	mlx4_uar_free(dev, &ibdev->priv_uar);
2879 
2880 err_pd:
2881 	mlx4_pd_free(dev, ibdev->priv_pdn);
2882 
2883 err_dealloc:
2884 	ib_dealloc_device(&ibdev->ib_dev);
2885 
2886 	return NULL;
2887 }
2888 
2889 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2890 {
2891 	int offset;
2892 
2893 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2894 
2895 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2896 					 dev->steer_qpn_count,
2897 					 get_count_order(count));
2898 	if (offset < 0)
2899 		return offset;
2900 
2901 	*qpn = dev->steer_qpn_base + offset;
2902 	return 0;
2903 }
2904 
2905 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2906 {
2907 	if (!qpn ||
2908 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2909 		return;
2910 
2911 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2912 		 qpn, dev->steer_qpn_base))
2913 		/* not supposed to be here */
2914 		return;
2915 
2916 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2917 			      qpn - dev->steer_qpn_base,
2918 			      get_count_order(count));
2919 }
2920 
2921 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2922 			 int is_attach)
2923 {
2924 	int err;
2925 	size_t flow_size;
2926 	struct ib_flow_attr *flow = NULL;
2927 	struct ib_flow_spec_ib *ib_spec;
2928 
2929 	if (is_attach) {
2930 		flow_size = sizeof(struct ib_flow_attr) +
2931 			    sizeof(struct ib_flow_spec_ib);
2932 		flow = kzalloc(flow_size, GFP_KERNEL);
2933 		if (!flow)
2934 			return -ENOMEM;
2935 		flow->port = mqp->port;
2936 		flow->num_of_specs = 1;
2937 		flow->size = flow_size;
2938 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2939 		ib_spec->type = IB_FLOW_SPEC_IB;
2940 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2941 		/* Add an empty rule for IB L2 */
2942 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2943 
2944 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow, MLX4_DOMAIN_NIC,
2945 					    MLX4_FS_REGULAR, &mqp->reg_id);
2946 	} else {
2947 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2948 	}
2949 	kfree(flow);
2950 	return err;
2951 }
2952 
2953 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2954 {
2955 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2956 	int p;
2957 	int i;
2958 
2959 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2960 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2961 	ibdev->ib_active = false;
2962 	flush_workqueue(wq);
2963 
2964 	if (ibdev->iboe.nb.notifier_call) {
2965 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2966 			pr_warn("failure unregistering notifier\n");
2967 		ibdev->iboe.nb.notifier_call = NULL;
2968 	}
2969 
2970 	mlx4_ib_close_sriov(ibdev);
2971 	mlx4_ib_mad_cleanup(ibdev);
2972 	ib_unregister_device(&ibdev->ib_dev);
2973 	mlx4_ib_diag_cleanup(ibdev);
2974 
2975 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2976 			      ibdev->steer_qpn_count);
2977 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2978 
2979 	iounmap(ibdev->uar_map);
2980 	for (p = 0; p < ibdev->num_ports; ++p)
2981 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2982 
2983 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2984 		mlx4_CLOSE_PORT(dev, p);
2985 
2986 	mlx4_ib_free_eqs(dev, ibdev);
2987 
2988 	mlx4_uar_free(dev, &ibdev->priv_uar);
2989 	mlx4_pd_free(dev, ibdev->priv_pdn);
2990 	ib_dealloc_device(&ibdev->ib_dev);
2991 }
2992 
2993 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2994 {
2995 	struct mlx4_ib_demux_work **dm = NULL;
2996 	struct mlx4_dev *dev = ibdev->dev;
2997 	int i;
2998 	unsigned long flags;
2999 	struct mlx4_active_ports actv_ports;
3000 	unsigned int ports;
3001 	unsigned int first_port;
3002 
3003 	if (!mlx4_is_master(dev))
3004 		return;
3005 
3006 	actv_ports = mlx4_get_active_ports(dev, slave);
3007 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3008 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3009 
3010 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3011 	if (!dm)
3012 		return;
3013 
3014 	for (i = 0; i < ports; i++) {
3015 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3016 		if (!dm[i]) {
3017 			while (--i >= 0)
3018 				kfree(dm[i]);
3019 			goto out;
3020 		}
3021 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3022 		dm[i]->port = first_port + i + 1;
3023 		dm[i]->slave = slave;
3024 		dm[i]->do_init = do_init;
3025 		dm[i]->dev = ibdev;
3026 	}
3027 	/* initialize or tear down tunnel QPs for the slave */
3028 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3029 	if (!ibdev->sriov.is_going_down) {
3030 		for (i = 0; i < ports; i++)
3031 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3032 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3033 	} else {
3034 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3035 		for (i = 0; i < ports; i++)
3036 			kfree(dm[i]);
3037 	}
3038 out:
3039 	kfree(dm);
3040 	return;
3041 }
3042 
3043 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3044 {
3045 	struct mlx4_ib_qp *mqp;
3046 	unsigned long flags_qp;
3047 	unsigned long flags_cq;
3048 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3049 	struct list_head    cq_notify_list;
3050 	struct mlx4_cq *mcq;
3051 	unsigned long flags;
3052 
3053 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3054 	INIT_LIST_HEAD(&cq_notify_list);
3055 
3056 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3057 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3058 
3059 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3060 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3061 		if (mqp->sq.tail != mqp->sq.head) {
3062 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3063 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3064 			if (send_mcq->mcq.comp &&
3065 			    mqp->ibqp.send_cq->comp_handler) {
3066 				if (!send_mcq->mcq.reset_notify_added) {
3067 					send_mcq->mcq.reset_notify_added = 1;
3068 					list_add_tail(&send_mcq->mcq.reset_notify,
3069 						      &cq_notify_list);
3070 				}
3071 			}
3072 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3073 		}
3074 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3075 		/* Now, handle the QP's receive queue */
3076 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3077 		/* no handling is needed for SRQ */
3078 		if (!mqp->ibqp.srq) {
3079 			if (mqp->rq.tail != mqp->rq.head) {
3080 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3081 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3082 				if (recv_mcq->mcq.comp &&
3083 				    mqp->ibqp.recv_cq->comp_handler) {
3084 					if (!recv_mcq->mcq.reset_notify_added) {
3085 						recv_mcq->mcq.reset_notify_added = 1;
3086 						list_add_tail(&recv_mcq->mcq.reset_notify,
3087 							      &cq_notify_list);
3088 					}
3089 				}
3090 				spin_unlock_irqrestore(&recv_mcq->lock,
3091 						       flags_cq);
3092 			}
3093 		}
3094 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3095 	}
3096 
3097 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3098 		mcq->comp(mcq);
3099 	}
3100 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3101 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3102 }
3103 
3104 static void handle_bonded_port_state_event(struct work_struct *work)
3105 {
3106 	struct ib_event_work *ew =
3107 		container_of(work, struct ib_event_work, work);
3108 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3109 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3110 	int i;
3111 	struct ib_event ibev;
3112 
3113 	kfree(ew);
3114 	spin_lock_bh(&ibdev->iboe.lock);
3115 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3116 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3117 		enum ib_port_state curr_port_state;
3118 
3119 		if (!curr_netdev)
3120 			continue;
3121 
3122 		curr_port_state =
3123 			(netif_running(curr_netdev) &&
3124 			 netif_carrier_ok(curr_netdev)) ?
3125 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3126 
3127 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3128 			curr_port_state : IB_PORT_ACTIVE;
3129 	}
3130 	spin_unlock_bh(&ibdev->iboe.lock);
3131 
3132 	ibev.device = &ibdev->ib_dev;
3133 	ibev.element.port_num = 1;
3134 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3135 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3136 
3137 	ib_dispatch_event(&ibev);
3138 }
3139 
3140 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3141 {
3142 	u64 sl2vl;
3143 	int err;
3144 
3145 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3146 	if (err) {
3147 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3148 		       port, err);
3149 		sl2vl = 0;
3150 	}
3151 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3152 }
3153 
3154 static void ib_sl2vl_update_work(struct work_struct *work)
3155 {
3156 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3157 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3158 	int port = ew->port;
3159 
3160 	mlx4_ib_sl2vl_update(mdev, port);
3161 
3162 	kfree(ew);
3163 }
3164 
3165 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3166 				     int port)
3167 {
3168 	struct ib_event_work *ew;
3169 
3170 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3171 	if (ew) {
3172 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3173 		ew->port = port;
3174 		ew->ib_dev = ibdev;
3175 		queue_work(wq, &ew->work);
3176 	}
3177 }
3178 
3179 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3180 			  enum mlx4_dev_event event, unsigned long param)
3181 {
3182 	struct ib_event ibev;
3183 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3184 	struct mlx4_eqe *eqe = NULL;
3185 	struct ib_event_work *ew;
3186 	int p = 0;
3187 
3188 	if (mlx4_is_bonded(dev) &&
3189 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3190 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3191 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3192 		if (!ew)
3193 			return;
3194 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3195 		ew->ib_dev = ibdev;
3196 		queue_work(wq, &ew->work);
3197 		return;
3198 	}
3199 
3200 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3201 		eqe = (struct mlx4_eqe *)param;
3202 	else
3203 		p = (int) param;
3204 
3205 	switch (event) {
3206 	case MLX4_DEV_EVENT_PORT_UP:
3207 		if (p > ibdev->num_ports)
3208 			return;
3209 		if (!mlx4_is_slave(dev) &&
3210 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3211 			IB_LINK_LAYER_INFINIBAND) {
3212 			if (mlx4_is_master(dev))
3213 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3214 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3215 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3216 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3217 		}
3218 		ibev.event = IB_EVENT_PORT_ACTIVE;
3219 		break;
3220 
3221 	case MLX4_DEV_EVENT_PORT_DOWN:
3222 		if (p > ibdev->num_ports)
3223 			return;
3224 		ibev.event = IB_EVENT_PORT_ERR;
3225 		break;
3226 
3227 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3228 		ibdev->ib_active = false;
3229 		ibev.event = IB_EVENT_DEVICE_FATAL;
3230 		mlx4_ib_handle_catas_error(ibdev);
3231 		break;
3232 
3233 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3234 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3235 		if (!ew)
3236 			return;
3237 
3238 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3239 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3240 		ew->ib_dev = ibdev;
3241 		/* need to queue only for port owner, which uses GEN_EQE */
3242 		if (mlx4_is_master(dev))
3243 			queue_work(wq, &ew->work);
3244 		else
3245 			handle_port_mgmt_change_event(&ew->work);
3246 		return;
3247 
3248 	case MLX4_DEV_EVENT_SLAVE_INIT:
3249 		/* here, p is the slave id */
3250 		do_slave_init(ibdev, p, 1);
3251 		if (mlx4_is_master(dev)) {
3252 			int i;
3253 
3254 			for (i = 1; i <= ibdev->num_ports; i++) {
3255 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3256 					== IB_LINK_LAYER_INFINIBAND)
3257 					mlx4_ib_slave_alias_guid_event(ibdev,
3258 								       p, i,
3259 								       1);
3260 			}
3261 		}
3262 		return;
3263 
3264 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3265 		if (mlx4_is_master(dev)) {
3266 			int i;
3267 
3268 			for (i = 1; i <= ibdev->num_ports; i++) {
3269 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3270 					== IB_LINK_LAYER_INFINIBAND)
3271 					mlx4_ib_slave_alias_guid_event(ibdev,
3272 								       p, i,
3273 								       0);
3274 			}
3275 		}
3276 		/* here, p is the slave id */
3277 		do_slave_init(ibdev, p, 0);
3278 		return;
3279 
3280 	default:
3281 		return;
3282 	}
3283 
3284 	ibev.device	      = ibdev_ptr;
3285 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3286 
3287 	ib_dispatch_event(&ibev);
3288 }
3289 
3290 static struct mlx4_interface mlx4_ib_interface = {
3291 	.add		= mlx4_ib_add,
3292 	.remove		= mlx4_ib_remove,
3293 	.event		= mlx4_ib_event,
3294 	.protocol	= MLX4_PROT_IB_IPV6,
3295 	.flags		= MLX4_INTFF_BONDING
3296 };
3297 
3298 static int __init mlx4_ib_init(void)
3299 {
3300 	int err;
3301 
3302 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3303 	if (!wq)
3304 		return -ENOMEM;
3305 
3306 	err = mlx4_ib_cm_init();
3307 	if (err)
3308 		goto clean_wq;
3309 
3310 	err = mlx4_ib_mcg_init();
3311 	if (err)
3312 		goto clean_cm;
3313 
3314 	err = mlx4_register_interface(&mlx4_ib_interface);
3315 	if (err)
3316 		goto clean_mcg;
3317 
3318 	return 0;
3319 
3320 clean_mcg:
3321 	mlx4_ib_mcg_destroy();
3322 
3323 clean_cm:
3324 	mlx4_ib_cm_destroy();
3325 
3326 clean_wq:
3327 	destroy_workqueue(wq);
3328 	return err;
3329 }
3330 
3331 static void __exit mlx4_ib_cleanup(void)
3332 {
3333 	mlx4_unregister_interface(&mlx4_ib_interface);
3334 	mlx4_ib_mcg_destroy();
3335 	mlx4_ib_cm_destroy();
3336 	destroy_workqueue(wq);
3337 }
3338 
3339 module_init(mlx4_ib_init);
3340 module_exit(mlx4_ib_cleanup);
3341