xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 9a29f5fc)
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 	if (dev)
148 		dev_hold(dev);
149 
150 	rcu_read_unlock();
151 	return dev;
152 }
153 
154 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
155 				  struct mlx4_ib_dev *ibdev,
156 				  u32 port_num)
157 {
158 	struct mlx4_cmd_mailbox *mailbox;
159 	int err;
160 	struct mlx4_dev *dev = ibdev->dev;
161 	int i;
162 	union ib_gid *gid_tbl;
163 
164 	mailbox = mlx4_alloc_cmd_mailbox(dev);
165 	if (IS_ERR(mailbox))
166 		return -ENOMEM;
167 
168 	gid_tbl = mailbox->buf;
169 
170 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
171 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
172 
173 	err = mlx4_cmd(dev, mailbox->dma,
174 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
175 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
176 		       MLX4_CMD_WRAPPED);
177 	if (mlx4_is_bonded(dev))
178 		err += mlx4_cmd(dev, mailbox->dma,
179 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
180 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
181 				MLX4_CMD_WRAPPED);
182 
183 	mlx4_free_cmd_mailbox(dev, mailbox);
184 	return err;
185 }
186 
187 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
188 				     struct mlx4_ib_dev *ibdev,
189 				     u32 port_num)
190 {
191 	struct mlx4_cmd_mailbox *mailbox;
192 	int err;
193 	struct mlx4_dev *dev = ibdev->dev;
194 	int i;
195 	struct {
196 		union ib_gid	gid;
197 		__be32		rsrvd1[2];
198 		__be16		rsrvd2;
199 		u8		type;
200 		u8		version;
201 		__be32		rsrvd3;
202 	} *gid_tbl;
203 
204 	mailbox = mlx4_alloc_cmd_mailbox(dev);
205 	if (IS_ERR(mailbox))
206 		return -ENOMEM;
207 
208 	gid_tbl = mailbox->buf;
209 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
210 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
211 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
212 			gid_tbl[i].version = 2;
213 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
214 				gid_tbl[i].type = 1;
215 		}
216 	}
217 
218 	err = mlx4_cmd(dev, mailbox->dma,
219 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
220 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
221 		       MLX4_CMD_WRAPPED);
222 	if (mlx4_is_bonded(dev))
223 		err += mlx4_cmd(dev, mailbox->dma,
224 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
225 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
226 				MLX4_CMD_WRAPPED);
227 
228 	mlx4_free_cmd_mailbox(dev, mailbox);
229 	return err;
230 }
231 
232 static int mlx4_ib_update_gids(struct gid_entry *gids,
233 			       struct mlx4_ib_dev *ibdev,
234 			       u32 port_num)
235 {
236 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
237 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
238 
239 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
240 }
241 
242 static void free_gid_entry(struct gid_entry *entry)
243 {
244 	memset(&entry->gid, 0, sizeof(entry->gid));
245 	kfree(entry->ctx);
246 	entry->ctx = NULL;
247 }
248 
249 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
250 {
251 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
252 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
253 	struct mlx4_port_gid_table   *port_gid_table;
254 	int free = -1, found = -1;
255 	int ret = 0;
256 	int hw_update = 0;
257 	int i;
258 	struct gid_entry *gids = NULL;
259 	u16 vlan_id = 0xffff;
260 	u8 mac[ETH_ALEN];
261 
262 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
263 		return -EINVAL;
264 
265 	if (attr->port_num > MLX4_MAX_PORTS)
266 		return -EINVAL;
267 
268 	if (!context)
269 		return -EINVAL;
270 
271 	ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
272 	if (ret)
273 		return ret;
274 	port_gid_table = &iboe->gids[attr->port_num - 1];
275 	spin_lock_bh(&iboe->lock);
276 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
277 		if (!memcmp(&port_gid_table->gids[i].gid,
278 			    &attr->gid, sizeof(attr->gid)) &&
279 		    port_gid_table->gids[i].gid_type == attr->gid_type &&
280 		    port_gid_table->gids[i].vlan_id == vlan_id)  {
281 			found = i;
282 			break;
283 		}
284 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
285 			free = i; /* HW has space */
286 	}
287 
288 	if (found < 0) {
289 		if (free < 0) {
290 			ret = -ENOSPC;
291 		} else {
292 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
293 			if (!port_gid_table->gids[free].ctx) {
294 				ret = -ENOMEM;
295 			} else {
296 				*context = port_gid_table->gids[free].ctx;
297 				memcpy(&port_gid_table->gids[free].gid,
298 				       &attr->gid, sizeof(attr->gid));
299 				port_gid_table->gids[free].gid_type = attr->gid_type;
300 				port_gid_table->gids[free].vlan_id = vlan_id;
301 				port_gid_table->gids[free].ctx->real_index = free;
302 				port_gid_table->gids[free].ctx->refcount = 1;
303 				hw_update = 1;
304 			}
305 		}
306 	} else {
307 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
308 		*context = ctx;
309 		ctx->refcount++;
310 	}
311 	if (!ret && hw_update) {
312 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
313 				     GFP_ATOMIC);
314 		if (!gids) {
315 			ret = -ENOMEM;
316 			*context = NULL;
317 			free_gid_entry(&port_gid_table->gids[free]);
318 		} else {
319 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
320 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
321 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
322 			}
323 		}
324 	}
325 	spin_unlock_bh(&iboe->lock);
326 
327 	if (!ret && hw_update) {
328 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
329 		if (ret) {
330 			spin_lock_bh(&iboe->lock);
331 			*context = NULL;
332 			free_gid_entry(&port_gid_table->gids[free]);
333 			spin_unlock_bh(&iboe->lock);
334 		}
335 		kfree(gids);
336 	}
337 
338 	return ret;
339 }
340 
341 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
342 {
343 	struct gid_cache_context *ctx = *context;
344 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
345 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
346 	struct mlx4_port_gid_table   *port_gid_table;
347 	int ret = 0;
348 	int hw_update = 0;
349 	struct gid_entry *gids = NULL;
350 
351 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
352 		return -EINVAL;
353 
354 	if (attr->port_num > MLX4_MAX_PORTS)
355 		return -EINVAL;
356 
357 	port_gid_table = &iboe->gids[attr->port_num - 1];
358 	spin_lock_bh(&iboe->lock);
359 	if (ctx) {
360 		ctx->refcount--;
361 		if (!ctx->refcount) {
362 			unsigned int real_index = ctx->real_index;
363 
364 			free_gid_entry(&port_gid_table->gids[real_index]);
365 			hw_update = 1;
366 		}
367 	}
368 	if (!ret && hw_update) {
369 		int i;
370 
371 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
372 				     GFP_ATOMIC);
373 		if (!gids) {
374 			ret = -ENOMEM;
375 		} else {
376 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
377 				memcpy(&gids[i].gid,
378 				       &port_gid_table->gids[i].gid,
379 				       sizeof(union ib_gid));
380 				gids[i].gid_type =
381 				    port_gid_table->gids[i].gid_type;
382 			}
383 		}
384 	}
385 	spin_unlock_bh(&iboe->lock);
386 
387 	if (!ret && hw_update) {
388 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
389 		kfree(gids);
390 	}
391 	return ret;
392 }
393 
394 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
395 				    const struct ib_gid_attr *attr)
396 {
397 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
398 	struct gid_cache_context *ctx = NULL;
399 	struct mlx4_port_gid_table   *port_gid_table;
400 	int real_index = -EINVAL;
401 	int i;
402 	unsigned long flags;
403 	u32 port_num = attr->port_num;
404 
405 	if (port_num > MLX4_MAX_PORTS)
406 		return -EINVAL;
407 
408 	if (mlx4_is_bonded(ibdev->dev))
409 		port_num = 1;
410 
411 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
412 		return attr->index;
413 
414 	spin_lock_irqsave(&iboe->lock, flags);
415 	port_gid_table = &iboe->gids[port_num - 1];
416 
417 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
418 		if (!memcmp(&port_gid_table->gids[i].gid,
419 			    &attr->gid, sizeof(attr->gid)) &&
420 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
421 			ctx = port_gid_table->gids[i].ctx;
422 			break;
423 		}
424 	if (ctx)
425 		real_index = ctx->real_index;
426 	spin_unlock_irqrestore(&iboe->lock, flags);
427 	return real_index;
428 }
429 
430 static int mlx4_ib_query_device(struct ib_device *ibdev,
431 				struct ib_device_attr *props,
432 				struct ib_udata *uhw)
433 {
434 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
435 	struct ib_smp *in_mad  = NULL;
436 	struct ib_smp *out_mad = NULL;
437 	int err;
438 	int have_ib_ports;
439 	struct mlx4_uverbs_ex_query_device cmd;
440 	struct mlx4_uverbs_ex_query_device_resp resp = {};
441 	struct mlx4_clock_params clock_params;
442 
443 	if (uhw->inlen) {
444 		if (uhw->inlen < sizeof(cmd))
445 			return -EINVAL;
446 
447 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
448 		if (err)
449 			return err;
450 
451 		if (cmd.comp_mask)
452 			return -EINVAL;
453 
454 		if (cmd.reserved)
455 			return -EINVAL;
456 	}
457 
458 	resp.response_length = offsetof(typeof(resp), response_length) +
459 		sizeof(resp.response_length);
460 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
461 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
462 	err = -ENOMEM;
463 	if (!in_mad || !out_mad)
464 		goto out;
465 
466 	ib_init_query_mad(in_mad);
467 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
468 
469 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
470 			   1, NULL, NULL, in_mad, out_mad);
471 	if (err)
472 		goto out;
473 
474 	memset(props, 0, sizeof *props);
475 
476 	have_ib_ports = num_ib_ports(dev->dev);
477 
478 	props->fw_ver = dev->dev->caps.fw_ver;
479 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
480 		IB_DEVICE_PORT_ACTIVE_EVENT		|
481 		IB_DEVICE_SYS_IMAGE_GUID		|
482 		IB_DEVICE_RC_RNR_NAK_GEN;
483 	props->kernel_cap_flags = IBK_BLOCK_MULTICAST_LOOPBACK;
484 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
485 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
486 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
487 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
488 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
489 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
490 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
491 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
492 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
493 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
494 	if (dev->dev->caps.max_gso_sz &&
495 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
496 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
497 		props->kernel_cap_flags |= IBK_UD_TSO;
498 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
499 		props->kernel_cap_flags |= IBK_LOCAL_DMA_LKEY;
500 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
501 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
502 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
503 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
504 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
505 		props->device_cap_flags |= IB_DEVICE_XRC;
506 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
507 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
508 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
509 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
510 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
511 		else
512 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
513 	}
514 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
515 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
516 
517 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
518 
519 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
520 		0xffffff;
521 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
522 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
523 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
524 
525 	props->max_mr_size	   = ~0ull;
526 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
527 	props->max_qp		   = dev->dev->quotas.qp;
528 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
529 	props->max_send_sge =
530 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
531 	props->max_recv_sge =
532 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
533 	props->max_sge_rd = MLX4_MAX_SGE_RD;
534 	props->max_cq		   = dev->dev->quotas.cq;
535 	props->max_cqe		   = dev->dev->caps.max_cqes;
536 	props->max_mr		   = dev->dev->quotas.mpt;
537 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
538 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
539 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
540 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
541 	props->max_srq		   = dev->dev->quotas.srq;
542 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
543 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
544 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
545 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
546 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
547 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
548 	props->masked_atomic_cap   = props->atomic_cap;
549 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
550 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
551 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
552 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
553 					   props->max_mcast_grp;
554 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
555 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
556 	props->max_ah = INT_MAX;
557 
558 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
559 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
560 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
561 			props->rss_caps.max_rwq_indirection_tables =
562 				props->max_qp;
563 			props->rss_caps.max_rwq_indirection_table_size =
564 				dev->dev->caps.max_rss_tbl_sz;
565 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
566 			props->max_wq_type_rq = props->max_qp;
567 		}
568 
569 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
570 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
571 	}
572 
573 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
574 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
575 
576 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
577 		resp.response_length += sizeof(resp.hca_core_clock_offset);
578 		if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
579 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
580 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
581 		}
582 	}
583 
584 	if (uhw->outlen >= resp.response_length +
585 	    sizeof(resp.max_inl_recv_sz)) {
586 		resp.response_length += sizeof(resp.max_inl_recv_sz);
587 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
588 			sizeof(struct mlx4_wqe_data_seg);
589 	}
590 
591 	if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
592 		if (props->rss_caps.supported_qpts) {
593 			resp.rss_caps.rx_hash_function =
594 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
595 
596 			resp.rss_caps.rx_hash_fields_mask =
597 				MLX4_IB_RX_HASH_SRC_IPV4 |
598 				MLX4_IB_RX_HASH_DST_IPV4 |
599 				MLX4_IB_RX_HASH_SRC_IPV6 |
600 				MLX4_IB_RX_HASH_DST_IPV6 |
601 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
602 				MLX4_IB_RX_HASH_DST_PORT_TCP |
603 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
604 				MLX4_IB_RX_HASH_DST_PORT_UDP;
605 
606 			if (dev->dev->caps.tunnel_offload_mode ==
607 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
608 				resp.rss_caps.rx_hash_fields_mask |=
609 					MLX4_IB_RX_HASH_INNER;
610 		}
611 		resp.response_length = offsetof(typeof(resp), rss_caps) +
612 				       sizeof(resp.rss_caps);
613 	}
614 
615 	if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
616 		if (dev->dev->caps.max_gso_sz &&
617 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
618 		    IB_LINK_LAYER_ETHERNET) ||
619 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
620 		    IB_LINK_LAYER_ETHERNET))) {
621 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
622 			resp.tso_caps.supported_qpts |=
623 				1 << IB_QPT_RAW_PACKET;
624 		}
625 		resp.response_length = offsetof(typeof(resp), tso_caps) +
626 				       sizeof(resp.tso_caps);
627 	}
628 
629 	if (uhw->outlen) {
630 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
631 		if (err)
632 			goto out;
633 	}
634 out:
635 	kfree(in_mad);
636 	kfree(out_mad);
637 
638 	return err;
639 }
640 
641 static enum rdma_link_layer
642 mlx4_ib_port_link_layer(struct ib_device *device, u32 port_num)
643 {
644 	struct mlx4_dev *dev = to_mdev(device)->dev;
645 
646 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
647 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
648 }
649 
650 static int ib_link_query_port(struct ib_device *ibdev, u32 port,
651 			      struct ib_port_attr *props, int netw_view)
652 {
653 	struct ib_smp *in_mad  = NULL;
654 	struct ib_smp *out_mad = NULL;
655 	int ext_active_speed;
656 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
657 	int err = -ENOMEM;
658 
659 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
660 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
661 	if (!in_mad || !out_mad)
662 		goto out;
663 
664 	ib_init_query_mad(in_mad);
665 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
666 	in_mad->attr_mod = cpu_to_be32(port);
667 
668 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
669 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
670 
671 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
672 				in_mad, out_mad);
673 	if (err)
674 		goto out;
675 
676 
677 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
678 	props->lmc		= out_mad->data[34] & 0x7;
679 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
680 	props->sm_sl		= out_mad->data[36] & 0xf;
681 	props->state		= out_mad->data[32] & 0xf;
682 	props->phys_state	= out_mad->data[33] >> 4;
683 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
684 	if (netw_view)
685 		props->gid_tbl_len = out_mad->data[50];
686 	else
687 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
688 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
689 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
690 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
691 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
692 	props->active_width	= out_mad->data[31] & 0xf;
693 	props->active_speed	= out_mad->data[35] >> 4;
694 	props->max_mtu		= out_mad->data[41] & 0xf;
695 	props->active_mtu	= out_mad->data[36] >> 4;
696 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
697 	props->max_vl_num	= out_mad->data[37] >> 4;
698 	props->init_type_reply	= out_mad->data[41] >> 4;
699 
700 	/* Check if extended speeds (EDR/FDR/...) are supported */
701 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
702 		ext_active_speed = out_mad->data[62] >> 4;
703 
704 		switch (ext_active_speed) {
705 		case 1:
706 			props->active_speed = IB_SPEED_FDR;
707 			break;
708 		case 2:
709 			props->active_speed = IB_SPEED_EDR;
710 			break;
711 		}
712 	}
713 
714 	/* If reported active speed is QDR, check if is FDR-10 */
715 	if (props->active_speed == IB_SPEED_QDR) {
716 		ib_init_query_mad(in_mad);
717 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
718 		in_mad->attr_mod = cpu_to_be32(port);
719 
720 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
721 				   NULL, NULL, in_mad, out_mad);
722 		if (err)
723 			goto out;
724 
725 		/* Checking LinkSpeedActive for FDR-10 */
726 		if (out_mad->data[15] & 0x1)
727 			props->active_speed = IB_SPEED_FDR10;
728 	}
729 
730 	/* Avoid wrong speed value returned by FW if the IB link is down. */
731 	if (props->state == IB_PORT_DOWN)
732 		 props->active_speed = IB_SPEED_SDR;
733 
734 out:
735 	kfree(in_mad);
736 	kfree(out_mad);
737 	return err;
738 }
739 
740 static u8 state_to_phys_state(enum ib_port_state state)
741 {
742 	return state == IB_PORT_ACTIVE ?
743 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
744 }
745 
746 static int eth_link_query_port(struct ib_device *ibdev, u32 port,
747 			       struct ib_port_attr *props)
748 {
749 
750 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
751 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
752 	struct net_device *ndev;
753 	enum ib_mtu tmp;
754 	struct mlx4_cmd_mailbox *mailbox;
755 	int err = 0;
756 	int is_bonded = mlx4_is_bonded(mdev->dev);
757 
758 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
759 	if (IS_ERR(mailbox))
760 		return PTR_ERR(mailbox);
761 
762 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
763 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
764 			   MLX4_CMD_WRAPPED);
765 	if (err)
766 		goto out;
767 
768 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
769 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
770 					   IB_WIDTH_4X : IB_WIDTH_1X;
771 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
772 					   IB_SPEED_FDR : IB_SPEED_QDR;
773 	props->port_cap_flags	= IB_PORT_CM_SUP;
774 	props->ip_gids = true;
775 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
776 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
777 	if (mdev->dev->caps.pkey_table_len[port])
778 		props->pkey_tbl_len = 1;
779 	props->max_mtu		= IB_MTU_4096;
780 	props->max_vl_num	= 2;
781 	props->state		= IB_PORT_DOWN;
782 	props->phys_state	= state_to_phys_state(props->state);
783 	props->active_mtu	= IB_MTU_256;
784 	spin_lock_bh(&iboe->lock);
785 	ndev = iboe->netdevs[port - 1];
786 	if (ndev && is_bonded) {
787 		rcu_read_lock(); /* required to get upper dev */
788 		ndev = netdev_master_upper_dev_get_rcu(ndev);
789 		rcu_read_unlock();
790 	}
791 	if (!ndev)
792 		goto out_unlock;
793 
794 	tmp = iboe_get_mtu(ndev->mtu);
795 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
796 
797 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
798 					IB_PORT_ACTIVE : IB_PORT_DOWN;
799 	props->phys_state	= state_to_phys_state(props->state);
800 out_unlock:
801 	spin_unlock_bh(&iboe->lock);
802 out:
803 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
804 	return err;
805 }
806 
807 int __mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
808 			 struct ib_port_attr *props, int netw_view)
809 {
810 	int err;
811 
812 	/* props being zeroed by the caller, avoid zeroing it here */
813 
814 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
815 		ib_link_query_port(ibdev, port, props, netw_view) :
816 				eth_link_query_port(ibdev, port, props);
817 
818 	return err;
819 }
820 
821 static int mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
822 			      struct ib_port_attr *props)
823 {
824 	/* returns host view */
825 	return __mlx4_ib_query_port(ibdev, port, props, 0);
826 }
827 
828 int __mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
829 			union ib_gid *gid, int netw_view)
830 {
831 	struct ib_smp *in_mad  = NULL;
832 	struct ib_smp *out_mad = NULL;
833 	int err = -ENOMEM;
834 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
835 	int clear = 0;
836 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
837 
838 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
839 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
840 	if (!in_mad || !out_mad)
841 		goto out;
842 
843 	ib_init_query_mad(in_mad);
844 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
845 	in_mad->attr_mod = cpu_to_be32(port);
846 
847 	if (mlx4_is_mfunc(dev->dev) && netw_view)
848 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
849 
850 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
851 	if (err)
852 		goto out;
853 
854 	memcpy(gid->raw, out_mad->data + 8, 8);
855 
856 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
857 		if (index) {
858 			/* For any index > 0, return the null guid */
859 			err = 0;
860 			clear = 1;
861 			goto out;
862 		}
863 	}
864 
865 	ib_init_query_mad(in_mad);
866 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
867 	in_mad->attr_mod = cpu_to_be32(index / 8);
868 
869 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
870 			   NULL, NULL, in_mad, out_mad);
871 	if (err)
872 		goto out;
873 
874 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
875 
876 out:
877 	if (clear)
878 		memset(gid->raw + 8, 0, 8);
879 	kfree(in_mad);
880 	kfree(out_mad);
881 	return err;
882 }
883 
884 static int mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
885 			     union ib_gid *gid)
886 {
887 	if (rdma_protocol_ib(ibdev, port))
888 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
889 	return 0;
890 }
891 
892 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u32 port,
893 			       u64 *sl2vl_tbl)
894 {
895 	union sl2vl_tbl_to_u64 sl2vl64;
896 	struct ib_smp *in_mad  = NULL;
897 	struct ib_smp *out_mad = NULL;
898 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
899 	int err = -ENOMEM;
900 	int jj;
901 
902 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
903 		*sl2vl_tbl = 0;
904 		return 0;
905 	}
906 
907 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
908 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
909 	if (!in_mad || !out_mad)
910 		goto out;
911 
912 	ib_init_query_mad(in_mad);
913 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
914 	in_mad->attr_mod = 0;
915 
916 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
917 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
918 
919 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
920 			   in_mad, out_mad);
921 	if (err)
922 		goto out;
923 
924 	for (jj = 0; jj < 8; jj++)
925 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
926 	*sl2vl_tbl = sl2vl64.sl64;
927 
928 out:
929 	kfree(in_mad);
930 	kfree(out_mad);
931 	return err;
932 }
933 
934 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
935 {
936 	u64 sl2vl;
937 	int i;
938 	int err;
939 
940 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
941 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
942 			continue;
943 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
944 		if (err) {
945 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
946 			       i, err);
947 			sl2vl = 0;
948 		}
949 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
950 	}
951 }
952 
953 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
954 			 u16 *pkey, int netw_view)
955 {
956 	struct ib_smp *in_mad  = NULL;
957 	struct ib_smp *out_mad = NULL;
958 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
959 	int err = -ENOMEM;
960 
961 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
962 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
963 	if (!in_mad || !out_mad)
964 		goto out;
965 
966 	ib_init_query_mad(in_mad);
967 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
968 	in_mad->attr_mod = cpu_to_be32(index / 32);
969 
970 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
971 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
972 
973 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
974 			   in_mad, out_mad);
975 	if (err)
976 		goto out;
977 
978 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
979 
980 out:
981 	kfree(in_mad);
982 	kfree(out_mad);
983 	return err;
984 }
985 
986 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
987 			      u16 *pkey)
988 {
989 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
990 }
991 
992 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
993 				 struct ib_device_modify *props)
994 {
995 	struct mlx4_cmd_mailbox *mailbox;
996 	unsigned long flags;
997 
998 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
999 		return -EOPNOTSUPP;
1000 
1001 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1002 		return 0;
1003 
1004 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1005 		return -EOPNOTSUPP;
1006 
1007 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1008 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1009 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1010 
1011 	/*
1012 	 * If possible, pass node desc to FW, so it can generate
1013 	 * a 144 trap.  If cmd fails, just ignore.
1014 	 */
1015 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1016 	if (IS_ERR(mailbox))
1017 		return 0;
1018 
1019 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1020 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1021 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1022 
1023 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1024 
1025 	return 0;
1026 }
1027 
1028 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u32 port,
1029 			    int reset_qkey_viols, u32 cap_mask)
1030 {
1031 	struct mlx4_cmd_mailbox *mailbox;
1032 	int err;
1033 
1034 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1035 	if (IS_ERR(mailbox))
1036 		return PTR_ERR(mailbox);
1037 
1038 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1039 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1040 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1041 	} else {
1042 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1043 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1044 	}
1045 
1046 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1047 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1048 		       MLX4_CMD_WRAPPED);
1049 
1050 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1051 	return err;
1052 }
1053 
1054 static int mlx4_ib_modify_port(struct ib_device *ibdev, u32 port, int mask,
1055 			       struct ib_port_modify *props)
1056 {
1057 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1058 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1059 	struct ib_port_attr attr;
1060 	u32 cap_mask;
1061 	int err;
1062 
1063 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1064 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1065 	 * violations and port capabilities are not meaningful.
1066 	 */
1067 	if (is_eth)
1068 		return 0;
1069 
1070 	mutex_lock(&mdev->cap_mask_mutex);
1071 
1072 	err = ib_query_port(ibdev, port, &attr);
1073 	if (err)
1074 		goto out;
1075 
1076 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1077 		~props->clr_port_cap_mask;
1078 
1079 	err = mlx4_ib_SET_PORT(mdev, port,
1080 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1081 			       cap_mask);
1082 
1083 out:
1084 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1085 	return err;
1086 }
1087 
1088 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1089 				  struct ib_udata *udata)
1090 {
1091 	struct ib_device *ibdev = uctx->device;
1092 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1093 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1094 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1095 	struct mlx4_ib_alloc_ucontext_resp resp;
1096 	int err;
1097 
1098 	if (!dev->ib_active)
1099 		return -EAGAIN;
1100 
1101 	if (ibdev->ops.uverbs_abi_ver ==
1102 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1103 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1104 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1105 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1106 	} else {
1107 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1108 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1109 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1110 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1111 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1112 	}
1113 
1114 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1115 	if (err)
1116 		return err;
1117 
1118 	INIT_LIST_HEAD(&context->db_page_list);
1119 	mutex_init(&context->db_page_mutex);
1120 
1121 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1122 	mutex_init(&context->wqn_ranges_mutex);
1123 
1124 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1125 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1126 	else
1127 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1128 
1129 	if (err) {
1130 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1131 		return -EFAULT;
1132 	}
1133 
1134 	return err;
1135 }
1136 
1137 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1138 {
1139 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1140 
1141 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1142 }
1143 
1144 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1145 {
1146 }
1147 
1148 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1149 {
1150 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1151 
1152 	switch (vma->vm_pgoff) {
1153 	case 0:
1154 		return rdma_user_mmap_io(context, vma,
1155 					 to_mucontext(context)->uar.pfn,
1156 					 PAGE_SIZE,
1157 					 pgprot_noncached(vma->vm_page_prot),
1158 					 NULL);
1159 
1160 	case 1:
1161 		if (dev->dev->caps.bf_reg_size == 0)
1162 			return -EINVAL;
1163 		return rdma_user_mmap_io(
1164 			context, vma,
1165 			to_mucontext(context)->uar.pfn +
1166 				dev->dev->caps.num_uars,
1167 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1168 			NULL);
1169 
1170 	case 3: {
1171 		struct mlx4_clock_params params;
1172 		int ret;
1173 
1174 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1175 		if (ret)
1176 			return ret;
1177 
1178 		return rdma_user_mmap_io(
1179 			context, vma,
1180 			(pci_resource_start(dev->dev->persist->pdev,
1181 					    params.bar) +
1182 			 params.offset) >>
1183 				PAGE_SHIFT,
1184 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1185 			NULL);
1186 	}
1187 
1188 	default:
1189 		return -EINVAL;
1190 	}
1191 }
1192 
1193 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1194 {
1195 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1196 	struct ib_device *ibdev = ibpd->device;
1197 	int err;
1198 
1199 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1200 	if (err)
1201 		return err;
1202 
1203 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1204 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1205 		return -EFAULT;
1206 	}
1207 	return 0;
1208 }
1209 
1210 static int mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1211 {
1212 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1213 	return 0;
1214 }
1215 
1216 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1217 {
1218 	struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1219 	struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1220 	struct ib_cq_init_attr cq_attr = {};
1221 	int err;
1222 
1223 	if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1224 		return -EOPNOTSUPP;
1225 
1226 	err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1227 	if (err)
1228 		return err;
1229 
1230 	xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1231 	if (IS_ERR(xrcd->pd)) {
1232 		err = PTR_ERR(xrcd->pd);
1233 		goto err2;
1234 	}
1235 
1236 	cq_attr.cqe = 1;
1237 	xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1238 	if (IS_ERR(xrcd->cq)) {
1239 		err = PTR_ERR(xrcd->cq);
1240 		goto err3;
1241 	}
1242 
1243 	return 0;
1244 
1245 err3:
1246 	ib_dealloc_pd(xrcd->pd);
1247 err2:
1248 	mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1249 	return err;
1250 }
1251 
1252 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1253 {
1254 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1255 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1256 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1257 	return 0;
1258 }
1259 
1260 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1261 {
1262 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1263 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1264 	struct mlx4_ib_gid_entry *ge;
1265 
1266 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1267 	if (!ge)
1268 		return -ENOMEM;
1269 
1270 	ge->gid = *gid;
1271 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1272 		ge->port = mqp->port;
1273 		ge->added = 1;
1274 	}
1275 
1276 	mutex_lock(&mqp->mutex);
1277 	list_add_tail(&ge->list, &mqp->gid_list);
1278 	mutex_unlock(&mqp->mutex);
1279 
1280 	return 0;
1281 }
1282 
1283 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1284 					  struct mlx4_ib_counters *ctr_table)
1285 {
1286 	struct counter_index *counter, *tmp_count;
1287 
1288 	mutex_lock(&ctr_table->mutex);
1289 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1290 				 list) {
1291 		if (counter->allocated)
1292 			mlx4_counter_free(ibdev->dev, counter->index);
1293 		list_del(&counter->list);
1294 		kfree(counter);
1295 	}
1296 	mutex_unlock(&ctr_table->mutex);
1297 }
1298 
1299 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1300 		   union ib_gid *gid)
1301 {
1302 	struct net_device *ndev;
1303 	int ret = 0;
1304 
1305 	if (!mqp->port)
1306 		return 0;
1307 
1308 	spin_lock_bh(&mdev->iboe.lock);
1309 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1310 	if (ndev)
1311 		dev_hold(ndev);
1312 	spin_unlock_bh(&mdev->iboe.lock);
1313 
1314 	if (ndev) {
1315 		ret = 1;
1316 		dev_put(ndev);
1317 	}
1318 
1319 	return ret;
1320 }
1321 
1322 struct mlx4_ib_steering {
1323 	struct list_head list;
1324 	struct mlx4_flow_reg_id reg_id;
1325 	union ib_gid gid;
1326 };
1327 
1328 #define LAST_ETH_FIELD vlan_tag
1329 #define LAST_IB_FIELD sl
1330 #define LAST_IPV4_FIELD dst_ip
1331 #define LAST_TCP_UDP_FIELD src_port
1332 
1333 /* Field is the last supported field */
1334 #define FIELDS_NOT_SUPPORTED(filter, field)\
1335 	memchr_inv((void *)&filter.field  +\
1336 		   sizeof(filter.field), 0,\
1337 		   sizeof(filter) -\
1338 		   offsetof(typeof(filter), field) -\
1339 		   sizeof(filter.field))
1340 
1341 static int parse_flow_attr(struct mlx4_dev *dev,
1342 			   u32 qp_num,
1343 			   union ib_flow_spec *ib_spec,
1344 			   struct _rule_hw *mlx4_spec)
1345 {
1346 	enum mlx4_net_trans_rule_id type;
1347 
1348 	switch (ib_spec->type) {
1349 	case IB_FLOW_SPEC_ETH:
1350 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1351 			return -ENOTSUPP;
1352 
1353 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1354 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1355 		       ETH_ALEN);
1356 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1357 		       ETH_ALEN);
1358 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1359 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1360 		break;
1361 	case IB_FLOW_SPEC_IB:
1362 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1363 			return -ENOTSUPP;
1364 
1365 		type = MLX4_NET_TRANS_RULE_ID_IB;
1366 		mlx4_spec->ib.l3_qpn =
1367 			cpu_to_be32(qp_num);
1368 		mlx4_spec->ib.qpn_mask =
1369 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1370 		break;
1371 
1372 
1373 	case IB_FLOW_SPEC_IPV4:
1374 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1375 			return -ENOTSUPP;
1376 
1377 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1378 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1379 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1380 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1381 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1382 		break;
1383 
1384 	case IB_FLOW_SPEC_TCP:
1385 	case IB_FLOW_SPEC_UDP:
1386 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1387 			return -ENOTSUPP;
1388 
1389 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1390 					MLX4_NET_TRANS_RULE_ID_TCP :
1391 					MLX4_NET_TRANS_RULE_ID_UDP;
1392 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1393 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1394 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1395 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1396 		break;
1397 
1398 	default:
1399 		return -EINVAL;
1400 	}
1401 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1402 	    mlx4_hw_rule_sz(dev, type) < 0)
1403 		return -EINVAL;
1404 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1405 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1406 	return mlx4_hw_rule_sz(dev, type);
1407 }
1408 
1409 struct default_rules {
1410 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1411 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1412 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1413 	__u8  link_layer;
1414 };
1415 static const struct default_rules default_table[] = {
1416 	{
1417 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1418 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1419 		.rules_create_list = {IB_FLOW_SPEC_IB},
1420 		.link_layer = IB_LINK_LAYER_INFINIBAND
1421 	}
1422 };
1423 
1424 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1425 					 struct ib_flow_attr *flow_attr)
1426 {
1427 	int i, j, k;
1428 	void *ib_flow;
1429 	const struct default_rules *pdefault_rules = default_table;
1430 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1431 
1432 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1433 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1434 		memset(&field_types, 0, sizeof(field_types));
1435 
1436 		if (link_layer != pdefault_rules->link_layer)
1437 			continue;
1438 
1439 		ib_flow = flow_attr + 1;
1440 		/* we assume the specs are sorted */
1441 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1442 		     j < flow_attr->num_of_specs; k++) {
1443 			union ib_flow_spec *current_flow =
1444 				(union ib_flow_spec *)ib_flow;
1445 
1446 			/* same layer but different type */
1447 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1448 			     (pdefault_rules->mandatory_fields[k] &
1449 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1450 			    (current_flow->type !=
1451 			     pdefault_rules->mandatory_fields[k]))
1452 				goto out;
1453 
1454 			/* same layer, try match next one */
1455 			if (current_flow->type ==
1456 			    pdefault_rules->mandatory_fields[k]) {
1457 				j++;
1458 				ib_flow +=
1459 					((union ib_flow_spec *)ib_flow)->size;
1460 			}
1461 		}
1462 
1463 		ib_flow = flow_attr + 1;
1464 		for (j = 0; j < flow_attr->num_of_specs;
1465 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1466 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1467 				/* same layer and same type */
1468 				if (((union ib_flow_spec *)ib_flow)->type ==
1469 				    pdefault_rules->mandatory_not_fields[k])
1470 					goto out;
1471 
1472 		return i;
1473 	}
1474 out:
1475 	return -1;
1476 }
1477 
1478 static int __mlx4_ib_create_default_rules(
1479 		struct mlx4_ib_dev *mdev,
1480 		struct ib_qp *qp,
1481 		const struct default_rules *pdefault_rules,
1482 		struct _rule_hw *mlx4_spec) {
1483 	int size = 0;
1484 	int i;
1485 
1486 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1487 		union ib_flow_spec ib_spec = {};
1488 		int ret;
1489 
1490 		switch (pdefault_rules->rules_create_list[i]) {
1491 		case 0:
1492 			/* no rule */
1493 			continue;
1494 		case IB_FLOW_SPEC_IB:
1495 			ib_spec.type = IB_FLOW_SPEC_IB;
1496 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1497 
1498 			break;
1499 		default:
1500 			/* invalid rule */
1501 			return -EINVAL;
1502 		}
1503 		/* We must put empty rule, qpn is being ignored */
1504 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1505 				      mlx4_spec);
1506 		if (ret < 0) {
1507 			pr_info("invalid parsing\n");
1508 			return -EINVAL;
1509 		}
1510 
1511 		mlx4_spec = (void *)mlx4_spec + ret;
1512 		size += ret;
1513 	}
1514 	return size;
1515 }
1516 
1517 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1518 			  int domain,
1519 			  enum mlx4_net_trans_promisc_mode flow_type,
1520 			  u64 *reg_id)
1521 {
1522 	int ret, i;
1523 	int size = 0;
1524 	void *ib_flow;
1525 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1526 	struct mlx4_cmd_mailbox *mailbox;
1527 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1528 	int default_flow;
1529 
1530 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1531 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1532 		return -EINVAL;
1533 	}
1534 
1535 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1536 		return -EINVAL;
1537 
1538 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1539 	if (IS_ERR(mailbox))
1540 		return PTR_ERR(mailbox);
1541 	ctrl = mailbox->buf;
1542 
1543 	ctrl->prio = cpu_to_be16(domain | flow_attr->priority);
1544 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1545 	ctrl->port = flow_attr->port;
1546 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1547 
1548 	ib_flow = flow_attr + 1;
1549 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1550 	/* Add default flows */
1551 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1552 	if (default_flow >= 0) {
1553 		ret = __mlx4_ib_create_default_rules(
1554 				mdev, qp, default_table + default_flow,
1555 				mailbox->buf + size);
1556 		if (ret < 0) {
1557 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1558 			return -EINVAL;
1559 		}
1560 		size += ret;
1561 	}
1562 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1563 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1564 				      mailbox->buf + size);
1565 		if (ret < 0) {
1566 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1567 			return -EINVAL;
1568 		}
1569 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1570 		size += ret;
1571 	}
1572 
1573 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1574 	    flow_attr->num_of_specs == 1) {
1575 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1576 		enum ib_flow_spec_type header_spec =
1577 			((union ib_flow_spec *)(flow_attr + 1))->type;
1578 
1579 		if (header_spec == IB_FLOW_SPEC_ETH)
1580 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1581 	}
1582 
1583 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1584 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1585 			   MLX4_CMD_NATIVE);
1586 	if (ret == -ENOMEM)
1587 		pr_err("mcg table is full. Fail to register network rule.\n");
1588 	else if (ret == -ENXIO)
1589 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1590 	else if (ret)
1591 		pr_err("Invalid argument. Fail to register network rule.\n");
1592 
1593 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1594 	return ret;
1595 }
1596 
1597 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1598 {
1599 	int err;
1600 	err = mlx4_cmd(dev, reg_id, 0, 0,
1601 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1602 		       MLX4_CMD_NATIVE);
1603 	if (err)
1604 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1605 		       reg_id);
1606 	return err;
1607 }
1608 
1609 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1610 				    u64 *reg_id)
1611 {
1612 	void *ib_flow;
1613 	union ib_flow_spec *ib_spec;
1614 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1615 	int err = 0;
1616 
1617 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1618 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1619 		return 0; /* do nothing */
1620 
1621 	ib_flow = flow_attr + 1;
1622 	ib_spec = (union ib_flow_spec *)ib_flow;
1623 
1624 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1625 		return 0; /* do nothing */
1626 
1627 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1628 				    flow_attr->port, qp->qp_num,
1629 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1630 				    reg_id);
1631 	return err;
1632 }
1633 
1634 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1635 				      struct ib_flow_attr *flow_attr,
1636 				      enum mlx4_net_trans_promisc_mode *type)
1637 {
1638 	int err = 0;
1639 
1640 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1641 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1642 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1643 		return -EOPNOTSUPP;
1644 	}
1645 
1646 	if (flow_attr->num_of_specs == 0) {
1647 		type[0] = MLX4_FS_MC_SNIFFER;
1648 		type[1] = MLX4_FS_UC_SNIFFER;
1649 	} else {
1650 		union ib_flow_spec *ib_spec;
1651 
1652 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1653 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1654 			return -EINVAL;
1655 
1656 		/* if all is zero than MC and UC */
1657 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1658 			type[0] = MLX4_FS_MC_SNIFFER;
1659 			type[1] = MLX4_FS_UC_SNIFFER;
1660 		} else {
1661 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1662 					    ib_spec->eth.mask.dst_mac[1],
1663 					    ib_spec->eth.mask.dst_mac[2],
1664 					    ib_spec->eth.mask.dst_mac[3],
1665 					    ib_spec->eth.mask.dst_mac[4],
1666 					    ib_spec->eth.mask.dst_mac[5]};
1667 
1668 			/* Above xor was only on MC bit, non empty mask is valid
1669 			 * only if this bit is set and rest are zero.
1670 			 */
1671 			if (!is_zero_ether_addr(&mac[0]))
1672 				return -EINVAL;
1673 
1674 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1675 				type[0] = MLX4_FS_MC_SNIFFER;
1676 			else
1677 				type[0] = MLX4_FS_UC_SNIFFER;
1678 		}
1679 	}
1680 
1681 	return err;
1682 }
1683 
1684 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1685 					   struct ib_flow_attr *flow_attr,
1686 					   struct ib_udata *udata)
1687 {
1688 	int err = 0, i = 0, j = 0;
1689 	struct mlx4_ib_flow *mflow;
1690 	enum mlx4_net_trans_promisc_mode type[2];
1691 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1692 	int is_bonded = mlx4_is_bonded(dev);
1693 
1694 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1695 		return ERR_PTR(-EOPNOTSUPP);
1696 
1697 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1698 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1699 		return ERR_PTR(-EOPNOTSUPP);
1700 
1701 	if (udata &&
1702 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1703 		return ERR_PTR(-EOPNOTSUPP);
1704 
1705 	memset(type, 0, sizeof(type));
1706 
1707 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1708 	if (!mflow) {
1709 		err = -ENOMEM;
1710 		goto err_free;
1711 	}
1712 
1713 	switch (flow_attr->type) {
1714 	case IB_FLOW_ATTR_NORMAL:
1715 		/* If dont trap flag (continue match) is set, under specific
1716 		 * condition traffic be replicated to given qp,
1717 		 * without stealing it
1718 		 */
1719 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1720 			err = mlx4_ib_add_dont_trap_rule(dev,
1721 							 flow_attr,
1722 							 type);
1723 			if (err)
1724 				goto err_free;
1725 		} else {
1726 			type[0] = MLX4_FS_REGULAR;
1727 		}
1728 		break;
1729 
1730 	case IB_FLOW_ATTR_ALL_DEFAULT:
1731 		type[0] = MLX4_FS_ALL_DEFAULT;
1732 		break;
1733 
1734 	case IB_FLOW_ATTR_MC_DEFAULT:
1735 		type[0] = MLX4_FS_MC_DEFAULT;
1736 		break;
1737 
1738 	case IB_FLOW_ATTR_SNIFFER:
1739 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1740 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1741 		break;
1742 
1743 	default:
1744 		err = -EINVAL;
1745 		goto err_free;
1746 	}
1747 
1748 	while (i < ARRAY_SIZE(type) && type[i]) {
1749 		err = __mlx4_ib_create_flow(qp, flow_attr, MLX4_DOMAIN_UVERBS,
1750 					    type[i], &mflow->reg_id[i].id);
1751 		if (err)
1752 			goto err_create_flow;
1753 		if (is_bonded) {
1754 			/* Application always sees one port so the mirror rule
1755 			 * must be on port #2
1756 			 */
1757 			flow_attr->port = 2;
1758 			err = __mlx4_ib_create_flow(qp, flow_attr,
1759 						    MLX4_DOMAIN_UVERBS, type[j],
1760 						    &mflow->reg_id[j].mirror);
1761 			flow_attr->port = 1;
1762 			if (err)
1763 				goto err_create_flow;
1764 			j++;
1765 		}
1766 
1767 		i++;
1768 	}
1769 
1770 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1771 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1772 					       &mflow->reg_id[i].id);
1773 		if (err)
1774 			goto err_create_flow;
1775 
1776 		if (is_bonded) {
1777 			flow_attr->port = 2;
1778 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1779 						       &mflow->reg_id[j].mirror);
1780 			flow_attr->port = 1;
1781 			if (err)
1782 				goto err_create_flow;
1783 			j++;
1784 		}
1785 		/* function to create mirror rule */
1786 		i++;
1787 	}
1788 
1789 	return &mflow->ibflow;
1790 
1791 err_create_flow:
1792 	while (i) {
1793 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1794 					     mflow->reg_id[i].id);
1795 		i--;
1796 	}
1797 
1798 	while (j) {
1799 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1800 					     mflow->reg_id[j].mirror);
1801 		j--;
1802 	}
1803 err_free:
1804 	kfree(mflow);
1805 	return ERR_PTR(err);
1806 }
1807 
1808 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1809 {
1810 	int err, ret = 0;
1811 	int i = 0;
1812 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1813 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1814 
1815 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1816 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1817 		if (err)
1818 			ret = err;
1819 		if (mflow->reg_id[i].mirror) {
1820 			err = __mlx4_ib_destroy_flow(mdev->dev,
1821 						     mflow->reg_id[i].mirror);
1822 			if (err)
1823 				ret = err;
1824 		}
1825 		i++;
1826 	}
1827 
1828 	kfree(mflow);
1829 	return ret;
1830 }
1831 
1832 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1833 {
1834 	int err;
1835 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1836 	struct mlx4_dev	*dev = mdev->dev;
1837 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1838 	struct mlx4_ib_steering *ib_steering = NULL;
1839 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1840 	struct mlx4_flow_reg_id	reg_id;
1841 
1842 	if (mdev->dev->caps.steering_mode ==
1843 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1844 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1845 		if (!ib_steering)
1846 			return -ENOMEM;
1847 	}
1848 
1849 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1850 				    !!(mqp->flags &
1851 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1852 				    prot, &reg_id.id);
1853 	if (err) {
1854 		pr_err("multicast attach op failed, err %d\n", err);
1855 		goto err_malloc;
1856 	}
1857 
1858 	reg_id.mirror = 0;
1859 	if (mlx4_is_bonded(dev)) {
1860 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1861 					    (mqp->port == 1) ? 2 : 1,
1862 					    !!(mqp->flags &
1863 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1864 					    prot, &reg_id.mirror);
1865 		if (err)
1866 			goto err_add;
1867 	}
1868 
1869 	err = add_gid_entry(ibqp, gid);
1870 	if (err)
1871 		goto err_add;
1872 
1873 	if (ib_steering) {
1874 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1875 		ib_steering->reg_id = reg_id;
1876 		mutex_lock(&mqp->mutex);
1877 		list_add(&ib_steering->list, &mqp->steering_rules);
1878 		mutex_unlock(&mqp->mutex);
1879 	}
1880 	return 0;
1881 
1882 err_add:
1883 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1884 			      prot, reg_id.id);
1885 	if (reg_id.mirror)
1886 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1887 				      prot, reg_id.mirror);
1888 err_malloc:
1889 	kfree(ib_steering);
1890 
1891 	return err;
1892 }
1893 
1894 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1895 {
1896 	struct mlx4_ib_gid_entry *ge;
1897 	struct mlx4_ib_gid_entry *tmp;
1898 	struct mlx4_ib_gid_entry *ret = NULL;
1899 
1900 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1901 		if (!memcmp(raw, ge->gid.raw, 16)) {
1902 			ret = ge;
1903 			break;
1904 		}
1905 	}
1906 
1907 	return ret;
1908 }
1909 
1910 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1911 {
1912 	int err;
1913 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1914 	struct mlx4_dev *dev = mdev->dev;
1915 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1916 	struct net_device *ndev;
1917 	struct mlx4_ib_gid_entry *ge;
1918 	struct mlx4_flow_reg_id reg_id = {0, 0};
1919 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1920 
1921 	if (mdev->dev->caps.steering_mode ==
1922 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1923 		struct mlx4_ib_steering *ib_steering;
1924 
1925 		mutex_lock(&mqp->mutex);
1926 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1927 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1928 				list_del(&ib_steering->list);
1929 				break;
1930 			}
1931 		}
1932 		mutex_unlock(&mqp->mutex);
1933 		if (&ib_steering->list == &mqp->steering_rules) {
1934 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1935 			return -EINVAL;
1936 		}
1937 		reg_id = ib_steering->reg_id;
1938 		kfree(ib_steering);
1939 	}
1940 
1941 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1942 				    prot, reg_id.id);
1943 	if (err)
1944 		return err;
1945 
1946 	if (mlx4_is_bonded(dev)) {
1947 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1948 					    prot, reg_id.mirror);
1949 		if (err)
1950 			return err;
1951 	}
1952 
1953 	mutex_lock(&mqp->mutex);
1954 	ge = find_gid_entry(mqp, gid->raw);
1955 	if (ge) {
1956 		spin_lock_bh(&mdev->iboe.lock);
1957 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1958 		if (ndev)
1959 			dev_hold(ndev);
1960 		spin_unlock_bh(&mdev->iboe.lock);
1961 		if (ndev)
1962 			dev_put(ndev);
1963 		list_del(&ge->list);
1964 		kfree(ge);
1965 	} else
1966 		pr_warn("could not find mgid entry\n");
1967 
1968 	mutex_unlock(&mqp->mutex);
1969 
1970 	return 0;
1971 }
1972 
1973 static int init_node_data(struct mlx4_ib_dev *dev)
1974 {
1975 	struct ib_smp *in_mad  = NULL;
1976 	struct ib_smp *out_mad = NULL;
1977 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1978 	int err = -ENOMEM;
1979 
1980 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1981 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1982 	if (!in_mad || !out_mad)
1983 		goto out;
1984 
1985 	ib_init_query_mad(in_mad);
1986 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1987 	if (mlx4_is_master(dev->dev))
1988 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1989 
1990 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1991 	if (err)
1992 		goto out;
1993 
1994 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1995 
1996 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1997 
1998 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1999 	if (err)
2000 		goto out;
2001 
2002 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2003 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2004 
2005 out:
2006 	kfree(in_mad);
2007 	kfree(out_mad);
2008 	return err;
2009 }
2010 
2011 static ssize_t hca_type_show(struct device *device,
2012 			     struct device_attribute *attr, char *buf)
2013 {
2014 	struct mlx4_ib_dev *dev =
2015 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2016 
2017 	return sysfs_emit(buf, "MT%d\n", dev->dev->persist->pdev->device);
2018 }
2019 static DEVICE_ATTR_RO(hca_type);
2020 
2021 static ssize_t hw_rev_show(struct device *device,
2022 			   struct device_attribute *attr, char *buf)
2023 {
2024 	struct mlx4_ib_dev *dev =
2025 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2026 
2027 	return sysfs_emit(buf, "%x\n", dev->dev->rev_id);
2028 }
2029 static DEVICE_ATTR_RO(hw_rev);
2030 
2031 static ssize_t board_id_show(struct device *device,
2032 			     struct device_attribute *attr, char *buf)
2033 {
2034 	struct mlx4_ib_dev *dev =
2035 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2036 
2037 	return sysfs_emit(buf, "%.*s\n", MLX4_BOARD_ID_LEN, dev->dev->board_id);
2038 }
2039 static DEVICE_ATTR_RO(board_id);
2040 
2041 static struct attribute *mlx4_class_attributes[] = {
2042 	&dev_attr_hw_rev.attr,
2043 	&dev_attr_hca_type.attr,
2044 	&dev_attr_board_id.attr,
2045 	NULL
2046 };
2047 
2048 static const struct attribute_group mlx4_attr_group = {
2049 	.attrs = mlx4_class_attributes,
2050 };
2051 
2052 struct diag_counter {
2053 	const char *name;
2054 	u32 offset;
2055 };
2056 
2057 #define DIAG_COUNTER(_name, _offset)			\
2058 	{ .name = #_name, .offset = _offset }
2059 
2060 static const struct diag_counter diag_basic[] = {
2061 	DIAG_COUNTER(rq_num_lle, 0x00),
2062 	DIAG_COUNTER(sq_num_lle, 0x04),
2063 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2064 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2065 	DIAG_COUNTER(rq_num_lpe, 0x18),
2066 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2067 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2068 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2069 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2070 	DIAG_COUNTER(sq_num_bre, 0x34),
2071 	DIAG_COUNTER(sq_num_rire, 0x44),
2072 	DIAG_COUNTER(rq_num_rire, 0x48),
2073 	DIAG_COUNTER(sq_num_rae, 0x4C),
2074 	DIAG_COUNTER(rq_num_rae, 0x50),
2075 	DIAG_COUNTER(sq_num_roe, 0x54),
2076 	DIAG_COUNTER(sq_num_tree, 0x5C),
2077 	DIAG_COUNTER(sq_num_rree, 0x64),
2078 	DIAG_COUNTER(rq_num_rnr, 0x68),
2079 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2080 	DIAG_COUNTER(rq_num_oos, 0x100),
2081 	DIAG_COUNTER(sq_num_oos, 0x104),
2082 };
2083 
2084 static const struct diag_counter diag_ext[] = {
2085 	DIAG_COUNTER(rq_num_dup, 0x130),
2086 	DIAG_COUNTER(sq_num_to, 0x134),
2087 };
2088 
2089 static const struct diag_counter diag_device_only[] = {
2090 	DIAG_COUNTER(num_cqovf, 0x1A0),
2091 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2092 };
2093 
2094 static struct rdma_hw_stats *
2095 mlx4_ib_alloc_hw_device_stats(struct ib_device *ibdev)
2096 {
2097 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2098 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2099 
2100 	if (!diag[0].descs)
2101 		return NULL;
2102 
2103 	return rdma_alloc_hw_stats_struct(diag[0].descs, diag[0].num_counters,
2104 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2105 }
2106 
2107 static struct rdma_hw_stats *
2108 mlx4_ib_alloc_hw_port_stats(struct ib_device *ibdev, u32 port_num)
2109 {
2110 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2111 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2112 
2113 	if (!diag[1].descs)
2114 		return NULL;
2115 
2116 	return rdma_alloc_hw_stats_struct(diag[1].descs, diag[1].num_counters,
2117 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2118 }
2119 
2120 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2121 				struct rdma_hw_stats *stats,
2122 				u32 port, int index)
2123 {
2124 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2125 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2126 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2127 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2128 	int ret;
2129 	int i;
2130 
2131 	ret = mlx4_query_diag_counters(dev->dev,
2132 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2133 				       diag[!!port].offset, hw_value,
2134 				       diag[!!port].num_counters, port);
2135 
2136 	if (ret)
2137 		return ret;
2138 
2139 	for (i = 0; i < diag[!!port].num_counters; i++)
2140 		stats->value[i] = hw_value[i];
2141 
2142 	return diag[!!port].num_counters;
2143 }
2144 
2145 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2146 					 struct rdma_stat_desc **pdescs,
2147 					 u32 **offset, u32 *num, bool port)
2148 {
2149 	u32 num_counters;
2150 
2151 	num_counters = ARRAY_SIZE(diag_basic);
2152 
2153 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2154 		num_counters += ARRAY_SIZE(diag_ext);
2155 
2156 	if (!port)
2157 		num_counters += ARRAY_SIZE(diag_device_only);
2158 
2159 	*pdescs = kcalloc(num_counters, sizeof(struct rdma_stat_desc),
2160 			  GFP_KERNEL);
2161 	if (!*pdescs)
2162 		return -ENOMEM;
2163 
2164 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2165 	if (!*offset)
2166 		goto err;
2167 
2168 	*num = num_counters;
2169 
2170 	return 0;
2171 
2172 err:
2173 	kfree(*pdescs);
2174 	return -ENOMEM;
2175 }
2176 
2177 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2178 				       struct rdma_stat_desc *descs,
2179 				       u32 *offset, bool port)
2180 {
2181 	int i;
2182 	int j;
2183 
2184 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2185 		descs[i].name = diag_basic[i].name;
2186 		offset[i] = diag_basic[i].offset;
2187 	}
2188 
2189 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2190 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2191 			descs[j].name = diag_ext[i].name;
2192 			offset[j] = diag_ext[i].offset;
2193 		}
2194 	}
2195 
2196 	if (!port) {
2197 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2198 			descs[j].name = diag_device_only[i].name;
2199 			offset[j] = diag_device_only[i].offset;
2200 		}
2201 	}
2202 }
2203 
2204 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2205 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2206 	.alloc_hw_port_stats = mlx4_ib_alloc_hw_port_stats,
2207 	.get_hw_stats = mlx4_ib_get_hw_stats,
2208 };
2209 
2210 static const struct ib_device_ops mlx4_ib_hw_stats_ops1 = {
2211 	.alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2212 	.get_hw_stats = mlx4_ib_get_hw_stats,
2213 };
2214 
2215 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2216 {
2217 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2218 	int i;
2219 	int ret;
2220 	bool per_port = !!(ibdev->dev->caps.flags2 &
2221 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2222 
2223 	if (mlx4_is_slave(ibdev->dev))
2224 		return 0;
2225 
2226 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2227 		/*
2228 		 * i == 1 means we are building port counters, set a different
2229 		 * stats ops without port stats callback.
2230 		 */
2231 		if (i && !per_port) {
2232 			ib_set_device_ops(&ibdev->ib_dev,
2233 					  &mlx4_ib_hw_stats_ops1);
2234 
2235 			return 0;
2236 		}
2237 
2238 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].descs,
2239 						    &diag[i].offset,
2240 						    &diag[i].num_counters, i);
2241 		if (ret)
2242 			goto err_alloc;
2243 
2244 		mlx4_ib_fill_diag_counters(ibdev, diag[i].descs,
2245 					   diag[i].offset, i);
2246 	}
2247 
2248 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2249 
2250 	return 0;
2251 
2252 err_alloc:
2253 	if (i) {
2254 		kfree(diag[i - 1].descs);
2255 		kfree(diag[i - 1].offset);
2256 	}
2257 
2258 	return ret;
2259 }
2260 
2261 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2262 {
2263 	int i;
2264 
2265 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2266 		kfree(ibdev->diag_counters[i].offset);
2267 		kfree(ibdev->diag_counters[i].descs);
2268 	}
2269 }
2270 
2271 #define MLX4_IB_INVALID_MAC	((u64)-1)
2272 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2273 			       struct net_device *dev,
2274 			       int port)
2275 {
2276 	u64 new_smac = 0;
2277 	u64 release_mac = MLX4_IB_INVALID_MAC;
2278 	struct mlx4_ib_qp *qp;
2279 
2280 	new_smac = ether_addr_to_u64(dev->dev_addr);
2281 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2282 
2283 	/* no need for update QP1 and mac registration in non-SRIOV */
2284 	if (!mlx4_is_mfunc(ibdev->dev))
2285 		return;
2286 
2287 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2288 	qp = ibdev->qp1_proxy[port - 1];
2289 	if (qp) {
2290 		int new_smac_index;
2291 		u64 old_smac;
2292 		struct mlx4_update_qp_params update_params;
2293 
2294 		mutex_lock(&qp->mutex);
2295 		old_smac = qp->pri.smac;
2296 		if (new_smac == old_smac)
2297 			goto unlock;
2298 
2299 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2300 
2301 		if (new_smac_index < 0)
2302 			goto unlock;
2303 
2304 		update_params.smac_index = new_smac_index;
2305 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2306 				   &update_params)) {
2307 			release_mac = new_smac;
2308 			goto unlock;
2309 		}
2310 		/* if old port was zero, no mac was yet registered for this QP */
2311 		if (qp->pri.smac_port)
2312 			release_mac = old_smac;
2313 		qp->pri.smac = new_smac;
2314 		qp->pri.smac_port = port;
2315 		qp->pri.smac_index = new_smac_index;
2316 	}
2317 
2318 unlock:
2319 	if (release_mac != MLX4_IB_INVALID_MAC)
2320 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2321 	if (qp)
2322 		mutex_unlock(&qp->mutex);
2323 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2324 }
2325 
2326 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2327 				 struct net_device *dev,
2328 				 unsigned long event)
2329 
2330 {
2331 	struct mlx4_ib_iboe *iboe;
2332 	int update_qps_port = -1;
2333 	int port;
2334 
2335 	ASSERT_RTNL();
2336 
2337 	iboe = &ibdev->iboe;
2338 
2339 	spin_lock_bh(&iboe->lock);
2340 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2341 
2342 		iboe->netdevs[port - 1] =
2343 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2344 
2345 		if (dev == iboe->netdevs[port - 1] &&
2346 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2347 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2348 			update_qps_port = port;
2349 
2350 		if (dev == iboe->netdevs[port - 1] &&
2351 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2352 			enum ib_port_state port_state;
2353 			struct ib_event ibev = { };
2354 
2355 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2356 						     &port_state))
2357 				continue;
2358 
2359 			if (event == NETDEV_UP &&
2360 			    (port_state != IB_PORT_ACTIVE ||
2361 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2362 				continue;
2363 			if (event == NETDEV_DOWN &&
2364 			    (port_state != IB_PORT_DOWN ||
2365 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2366 				continue;
2367 			iboe->last_port_state[port - 1] = port_state;
2368 
2369 			ibev.device = &ibdev->ib_dev;
2370 			ibev.element.port_num = port;
2371 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2372 							  IB_EVENT_PORT_ERR;
2373 			ib_dispatch_event(&ibev);
2374 		}
2375 
2376 	}
2377 	spin_unlock_bh(&iboe->lock);
2378 
2379 	if (update_qps_port > 0)
2380 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2381 }
2382 
2383 static int mlx4_ib_netdev_event(struct notifier_block *this,
2384 				unsigned long event, void *ptr)
2385 {
2386 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2387 	struct mlx4_ib_dev *ibdev;
2388 
2389 	if (!net_eq(dev_net(dev), &init_net))
2390 		return NOTIFY_DONE;
2391 
2392 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2393 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2394 
2395 	return NOTIFY_DONE;
2396 }
2397 
2398 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2399 {
2400 	int port;
2401 	int slave;
2402 	int i;
2403 
2404 	if (mlx4_is_master(ibdev->dev)) {
2405 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2406 		     ++slave) {
2407 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2408 				for (i = 0;
2409 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2410 				     ++i) {
2411 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2412 					/* master has the identity virt2phys pkey mapping */
2413 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2414 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2415 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2416 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2417 				}
2418 			}
2419 		}
2420 		/* initialize pkey cache */
2421 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2422 			for (i = 0;
2423 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2424 			     ++i)
2425 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2426 					(i) ? 0 : 0xFFFF;
2427 		}
2428 	}
2429 }
2430 
2431 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2432 {
2433 	int i, j, eq = 0, total_eqs = 0;
2434 
2435 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2436 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2437 	if (!ibdev->eq_table)
2438 		return;
2439 
2440 	for (i = 1; i <= dev->caps.num_ports; i++) {
2441 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2442 		     j++, total_eqs++) {
2443 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2444 				continue;
2445 			ibdev->eq_table[eq] = total_eqs;
2446 			if (!mlx4_assign_eq(dev, i,
2447 					    &ibdev->eq_table[eq]))
2448 				eq++;
2449 			else
2450 				ibdev->eq_table[eq] = -1;
2451 		}
2452 	}
2453 
2454 	for (i = eq; i < dev->caps.num_comp_vectors;
2455 	     ibdev->eq_table[i++] = -1)
2456 		;
2457 
2458 	/* Advertise the new number of EQs to clients */
2459 	ibdev->ib_dev.num_comp_vectors = eq;
2460 }
2461 
2462 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2463 {
2464 	int i;
2465 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2466 
2467 	/* no eqs were allocated */
2468 	if (!ibdev->eq_table)
2469 		return;
2470 
2471 	/* Reset the advertised EQ number */
2472 	ibdev->ib_dev.num_comp_vectors = 0;
2473 
2474 	for (i = 0; i < total_eqs; i++)
2475 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2476 
2477 	kfree(ibdev->eq_table);
2478 	ibdev->eq_table = NULL;
2479 }
2480 
2481 static int mlx4_port_immutable(struct ib_device *ibdev, u32 port_num,
2482 			       struct ib_port_immutable *immutable)
2483 {
2484 	struct ib_port_attr attr;
2485 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2486 	int err;
2487 
2488 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2489 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2490 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2491 	} else {
2492 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2493 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2494 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2495 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2496 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2497 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2498 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2499 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2500 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2501 	}
2502 
2503 	err = ib_query_port(ibdev, port_num, &attr);
2504 	if (err)
2505 		return err;
2506 
2507 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2508 	immutable->gid_tbl_len = attr.gid_tbl_len;
2509 
2510 	return 0;
2511 }
2512 
2513 static void get_fw_ver_str(struct ib_device *device, char *str)
2514 {
2515 	struct mlx4_ib_dev *dev =
2516 		container_of(device, struct mlx4_ib_dev, ib_dev);
2517 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2518 		 (int) (dev->dev->caps.fw_ver >> 32),
2519 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2520 		 (int) dev->dev->caps.fw_ver & 0xffff);
2521 }
2522 
2523 static const struct ib_device_ops mlx4_ib_dev_ops = {
2524 	.owner = THIS_MODULE,
2525 	.driver_id = RDMA_DRIVER_MLX4,
2526 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2527 
2528 	.add_gid = mlx4_ib_add_gid,
2529 	.alloc_mr = mlx4_ib_alloc_mr,
2530 	.alloc_pd = mlx4_ib_alloc_pd,
2531 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2532 	.attach_mcast = mlx4_ib_mcg_attach,
2533 	.create_ah = mlx4_ib_create_ah,
2534 	.create_cq = mlx4_ib_create_cq,
2535 	.create_qp = mlx4_ib_create_qp,
2536 	.create_srq = mlx4_ib_create_srq,
2537 	.dealloc_pd = mlx4_ib_dealloc_pd,
2538 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2539 	.del_gid = mlx4_ib_del_gid,
2540 	.dereg_mr = mlx4_ib_dereg_mr,
2541 	.destroy_ah = mlx4_ib_destroy_ah,
2542 	.destroy_cq = mlx4_ib_destroy_cq,
2543 	.destroy_qp = mlx4_ib_destroy_qp,
2544 	.destroy_srq = mlx4_ib_destroy_srq,
2545 	.detach_mcast = mlx4_ib_mcg_detach,
2546 	.device_group = &mlx4_attr_group,
2547 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2548 	.drain_rq = mlx4_ib_drain_rq,
2549 	.drain_sq = mlx4_ib_drain_sq,
2550 	.get_dev_fw_str = get_fw_ver_str,
2551 	.get_dma_mr = mlx4_ib_get_dma_mr,
2552 	.get_link_layer = mlx4_ib_port_link_layer,
2553 	.get_netdev = mlx4_ib_get_netdev,
2554 	.get_port_immutable = mlx4_port_immutable,
2555 	.map_mr_sg = mlx4_ib_map_mr_sg,
2556 	.mmap = mlx4_ib_mmap,
2557 	.modify_cq = mlx4_ib_modify_cq,
2558 	.modify_device = mlx4_ib_modify_device,
2559 	.modify_port = mlx4_ib_modify_port,
2560 	.modify_qp = mlx4_ib_modify_qp,
2561 	.modify_srq = mlx4_ib_modify_srq,
2562 	.poll_cq = mlx4_ib_poll_cq,
2563 	.post_recv = mlx4_ib_post_recv,
2564 	.post_send = mlx4_ib_post_send,
2565 	.post_srq_recv = mlx4_ib_post_srq_recv,
2566 	.process_mad = mlx4_ib_process_mad,
2567 	.query_ah = mlx4_ib_query_ah,
2568 	.query_device = mlx4_ib_query_device,
2569 	.query_gid = mlx4_ib_query_gid,
2570 	.query_pkey = mlx4_ib_query_pkey,
2571 	.query_port = mlx4_ib_query_port,
2572 	.query_qp = mlx4_ib_query_qp,
2573 	.query_srq = mlx4_ib_query_srq,
2574 	.reg_user_mr = mlx4_ib_reg_user_mr,
2575 	.req_notify_cq = mlx4_ib_arm_cq,
2576 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2577 	.resize_cq = mlx4_ib_resize_cq,
2578 
2579 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2580 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2581 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2582 	INIT_RDMA_OBJ_SIZE(ib_qp, mlx4_ib_qp, ibqp),
2583 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2584 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2585 };
2586 
2587 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2588 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2589 	.create_wq = mlx4_ib_create_wq,
2590 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2591 	.destroy_wq = mlx4_ib_destroy_wq,
2592 	.modify_wq = mlx4_ib_modify_wq,
2593 
2594 	INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx4_ib_rwq_ind_table,
2595 			   ib_rwq_ind_tbl),
2596 };
2597 
2598 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2599 	.alloc_mw = mlx4_ib_alloc_mw,
2600 	.dealloc_mw = mlx4_ib_dealloc_mw,
2601 
2602 	INIT_RDMA_OBJ_SIZE(ib_mw, mlx4_ib_mw, ibmw),
2603 };
2604 
2605 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2606 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2607 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2608 
2609 	INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2610 };
2611 
2612 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2613 	.create_flow = mlx4_ib_create_flow,
2614 	.destroy_flow = mlx4_ib_destroy_flow,
2615 };
2616 
2617 static void *mlx4_ib_add(struct mlx4_dev *dev)
2618 {
2619 	struct mlx4_ib_dev *ibdev;
2620 	int num_ports = 0;
2621 	int i, j;
2622 	int err;
2623 	struct mlx4_ib_iboe *iboe;
2624 	int ib_num_ports = 0;
2625 	int num_req_counters;
2626 	int allocated;
2627 	u32 counter_index;
2628 	struct counter_index *new_counter_index = NULL;
2629 
2630 	pr_info_once("%s", mlx4_ib_version);
2631 
2632 	num_ports = 0;
2633 	mlx4_foreach_ib_transport_port(i, dev)
2634 		num_ports++;
2635 
2636 	/* No point in registering a device with no ports... */
2637 	if (num_ports == 0)
2638 		return NULL;
2639 
2640 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2641 	if (!ibdev) {
2642 		dev_err(&dev->persist->pdev->dev,
2643 			"Device struct alloc failed\n");
2644 		return NULL;
2645 	}
2646 
2647 	iboe = &ibdev->iboe;
2648 
2649 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2650 		goto err_dealloc;
2651 
2652 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2653 		goto err_pd;
2654 
2655 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2656 				 PAGE_SIZE);
2657 	if (!ibdev->uar_map)
2658 		goto err_uar;
2659 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2660 
2661 	ibdev->dev = dev;
2662 	ibdev->bond_next_port	= 0;
2663 
2664 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2665 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2666 	ibdev->num_ports		= num_ports;
2667 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2668 						1 : ibdev->num_ports;
2669 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2670 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2671 
2672 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2673 
2674 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2675 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2676 	    IB_LINK_LAYER_ETHERNET) ||
2677 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2678 	    IB_LINK_LAYER_ETHERNET)))
2679 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2680 
2681 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2682 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2683 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2684 
2685 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2686 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2687 	}
2688 
2689 	if (check_flow_steering_support(dev)) {
2690 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2691 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2692 	}
2693 
2694 	if (!dev->caps.userspace_caps)
2695 		ibdev->ib_dev.ops.uverbs_abi_ver =
2696 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2697 
2698 	mlx4_ib_alloc_eqs(dev, ibdev);
2699 
2700 	spin_lock_init(&iboe->lock);
2701 
2702 	if (init_node_data(ibdev))
2703 		goto err_map;
2704 	mlx4_init_sl2vl_tbl(ibdev);
2705 
2706 	for (i = 0; i < ibdev->num_ports; ++i) {
2707 		mutex_init(&ibdev->counters_table[i].mutex);
2708 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2709 		iboe->last_port_state[i] = IB_PORT_DOWN;
2710 	}
2711 
2712 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2713 	for (i = 0; i < num_req_counters; ++i) {
2714 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2715 		allocated = 0;
2716 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2717 						IB_LINK_LAYER_ETHERNET) {
2718 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2719 						 MLX4_RES_USAGE_DRIVER);
2720 			/* if failed to allocate a new counter, use default */
2721 			if (err)
2722 				counter_index =
2723 					mlx4_get_default_counter_index(dev,
2724 								       i + 1);
2725 			else
2726 				allocated = 1;
2727 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2728 			counter_index = mlx4_get_default_counter_index(dev,
2729 								       i + 1);
2730 		}
2731 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2732 					    GFP_KERNEL);
2733 		if (!new_counter_index) {
2734 			if (allocated)
2735 				mlx4_counter_free(ibdev->dev, counter_index);
2736 			goto err_counter;
2737 		}
2738 		new_counter_index->index = counter_index;
2739 		new_counter_index->allocated = allocated;
2740 		list_add_tail(&new_counter_index->list,
2741 			      &ibdev->counters_table[i].counters_list);
2742 		ibdev->counters_table[i].default_counter = counter_index;
2743 		pr_info("counter index %d for port %d allocated %d\n",
2744 			counter_index, i + 1, allocated);
2745 	}
2746 	if (mlx4_is_bonded(dev))
2747 		for (i = 1; i < ibdev->num_ports ; ++i) {
2748 			new_counter_index =
2749 					kmalloc(sizeof(struct counter_index),
2750 						GFP_KERNEL);
2751 			if (!new_counter_index)
2752 				goto err_counter;
2753 			new_counter_index->index = counter_index;
2754 			new_counter_index->allocated = 0;
2755 			list_add_tail(&new_counter_index->list,
2756 				      &ibdev->counters_table[i].counters_list);
2757 			ibdev->counters_table[i].default_counter =
2758 								counter_index;
2759 		}
2760 
2761 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2762 		ib_num_ports++;
2763 
2764 	spin_lock_init(&ibdev->sm_lock);
2765 	mutex_init(&ibdev->cap_mask_mutex);
2766 	INIT_LIST_HEAD(&ibdev->qp_list);
2767 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2768 
2769 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2770 	    ib_num_ports) {
2771 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2772 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2773 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2774 					    &ibdev->steer_qpn_base, 0,
2775 					    MLX4_RES_USAGE_DRIVER);
2776 		if (err)
2777 			goto err_counter;
2778 
2779 		ibdev->ib_uc_qpns_bitmap = bitmap_alloc(ibdev->steer_qpn_count,
2780 							GFP_KERNEL);
2781 		if (!ibdev->ib_uc_qpns_bitmap)
2782 			goto err_steer_qp_release;
2783 
2784 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2785 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2786 				    ibdev->steer_qpn_count);
2787 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2788 					dev, ibdev->steer_qpn_base,
2789 					ibdev->steer_qpn_base +
2790 					ibdev->steer_qpn_count - 1);
2791 			if (err)
2792 				goto err_steer_free_bitmap;
2793 		} else {
2794 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2795 				    ibdev->steer_qpn_count);
2796 		}
2797 	}
2798 
2799 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2800 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2801 
2802 	if (mlx4_ib_alloc_diag_counters(ibdev))
2803 		goto err_steer_free_bitmap;
2804 
2805 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d",
2806 			       &dev->persist->pdev->dev))
2807 		goto err_diag_counters;
2808 
2809 	if (mlx4_ib_mad_init(ibdev))
2810 		goto err_reg;
2811 
2812 	if (mlx4_ib_init_sriov(ibdev))
2813 		goto err_mad;
2814 
2815 	if (!iboe->nb.notifier_call) {
2816 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2817 		err = register_netdevice_notifier(&iboe->nb);
2818 		if (err) {
2819 			iboe->nb.notifier_call = NULL;
2820 			goto err_notif;
2821 		}
2822 	}
2823 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2824 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2825 		if (err)
2826 			goto err_notif;
2827 	}
2828 
2829 	ibdev->ib_active = true;
2830 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2831 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2832 					 &ibdev->ib_dev);
2833 
2834 	if (mlx4_is_mfunc(ibdev->dev))
2835 		init_pkeys(ibdev);
2836 
2837 	/* create paravirt contexts for any VFs which are active */
2838 	if (mlx4_is_master(ibdev->dev)) {
2839 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2840 			if (j == mlx4_master_func_num(ibdev->dev))
2841 				continue;
2842 			if (mlx4_is_slave_active(ibdev->dev, j))
2843 				do_slave_init(ibdev, j, 1);
2844 		}
2845 	}
2846 	return ibdev;
2847 
2848 err_notif:
2849 	if (ibdev->iboe.nb.notifier_call) {
2850 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2851 			pr_warn("failure unregistering notifier\n");
2852 		ibdev->iboe.nb.notifier_call = NULL;
2853 	}
2854 	flush_workqueue(wq);
2855 
2856 	mlx4_ib_close_sriov(ibdev);
2857 
2858 err_mad:
2859 	mlx4_ib_mad_cleanup(ibdev);
2860 
2861 err_reg:
2862 	ib_unregister_device(&ibdev->ib_dev);
2863 
2864 err_diag_counters:
2865 	mlx4_ib_diag_cleanup(ibdev);
2866 
2867 err_steer_free_bitmap:
2868 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2869 
2870 err_steer_qp_release:
2871 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2872 			      ibdev->steer_qpn_count);
2873 err_counter:
2874 	for (i = 0; i < ibdev->num_ports; ++i)
2875 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2876 
2877 err_map:
2878 	mlx4_ib_free_eqs(dev, ibdev);
2879 	iounmap(ibdev->uar_map);
2880 
2881 err_uar:
2882 	mlx4_uar_free(dev, &ibdev->priv_uar);
2883 
2884 err_pd:
2885 	mlx4_pd_free(dev, ibdev->priv_pdn);
2886 
2887 err_dealloc:
2888 	ib_dealloc_device(&ibdev->ib_dev);
2889 
2890 	return NULL;
2891 }
2892 
2893 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2894 {
2895 	int offset;
2896 
2897 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2898 
2899 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2900 					 dev->steer_qpn_count,
2901 					 get_count_order(count));
2902 	if (offset < 0)
2903 		return offset;
2904 
2905 	*qpn = dev->steer_qpn_base + offset;
2906 	return 0;
2907 }
2908 
2909 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2910 {
2911 	if (!qpn ||
2912 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2913 		return;
2914 
2915 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2916 		 qpn, dev->steer_qpn_base))
2917 		/* not supposed to be here */
2918 		return;
2919 
2920 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2921 			      qpn - dev->steer_qpn_base,
2922 			      get_count_order(count));
2923 }
2924 
2925 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2926 			 int is_attach)
2927 {
2928 	int err;
2929 	size_t flow_size;
2930 	struct ib_flow_attr *flow = NULL;
2931 	struct ib_flow_spec_ib *ib_spec;
2932 
2933 	if (is_attach) {
2934 		flow_size = sizeof(struct ib_flow_attr) +
2935 			    sizeof(struct ib_flow_spec_ib);
2936 		flow = kzalloc(flow_size, GFP_KERNEL);
2937 		if (!flow)
2938 			return -ENOMEM;
2939 		flow->port = mqp->port;
2940 		flow->num_of_specs = 1;
2941 		flow->size = flow_size;
2942 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2943 		ib_spec->type = IB_FLOW_SPEC_IB;
2944 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2945 		/* Add an empty rule for IB L2 */
2946 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2947 
2948 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow, MLX4_DOMAIN_NIC,
2949 					    MLX4_FS_REGULAR, &mqp->reg_id);
2950 	} else {
2951 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2952 	}
2953 	kfree(flow);
2954 	return err;
2955 }
2956 
2957 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2958 {
2959 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2960 	int p;
2961 	int i;
2962 
2963 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2964 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2965 	ibdev->ib_active = false;
2966 	flush_workqueue(wq);
2967 
2968 	if (ibdev->iboe.nb.notifier_call) {
2969 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2970 			pr_warn("failure unregistering notifier\n");
2971 		ibdev->iboe.nb.notifier_call = NULL;
2972 	}
2973 
2974 	mlx4_ib_close_sriov(ibdev);
2975 	mlx4_ib_mad_cleanup(ibdev);
2976 	ib_unregister_device(&ibdev->ib_dev);
2977 	mlx4_ib_diag_cleanup(ibdev);
2978 
2979 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2980 			      ibdev->steer_qpn_count);
2981 	bitmap_free(ibdev->ib_uc_qpns_bitmap);
2982 
2983 	iounmap(ibdev->uar_map);
2984 	for (p = 0; p < ibdev->num_ports; ++p)
2985 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2986 
2987 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2988 		mlx4_CLOSE_PORT(dev, p);
2989 
2990 	mlx4_ib_free_eqs(dev, ibdev);
2991 
2992 	mlx4_uar_free(dev, &ibdev->priv_uar);
2993 	mlx4_pd_free(dev, ibdev->priv_pdn);
2994 	ib_dealloc_device(&ibdev->ib_dev);
2995 }
2996 
2997 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2998 {
2999 	struct mlx4_ib_demux_work **dm = NULL;
3000 	struct mlx4_dev *dev = ibdev->dev;
3001 	int i;
3002 	unsigned long flags;
3003 	struct mlx4_active_ports actv_ports;
3004 	unsigned int ports;
3005 	unsigned int first_port;
3006 
3007 	if (!mlx4_is_master(dev))
3008 		return;
3009 
3010 	actv_ports = mlx4_get_active_ports(dev, slave);
3011 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3012 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3013 
3014 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3015 	if (!dm)
3016 		return;
3017 
3018 	for (i = 0; i < ports; i++) {
3019 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3020 		if (!dm[i]) {
3021 			while (--i >= 0)
3022 				kfree(dm[i]);
3023 			goto out;
3024 		}
3025 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3026 		dm[i]->port = first_port + i + 1;
3027 		dm[i]->slave = slave;
3028 		dm[i]->do_init = do_init;
3029 		dm[i]->dev = ibdev;
3030 	}
3031 	/* initialize or tear down tunnel QPs for the slave */
3032 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3033 	if (!ibdev->sriov.is_going_down) {
3034 		for (i = 0; i < ports; i++)
3035 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3036 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3037 	} else {
3038 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3039 		for (i = 0; i < ports; i++)
3040 			kfree(dm[i]);
3041 	}
3042 out:
3043 	kfree(dm);
3044 	return;
3045 }
3046 
3047 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3048 {
3049 	struct mlx4_ib_qp *mqp;
3050 	unsigned long flags_qp;
3051 	unsigned long flags_cq;
3052 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3053 	struct list_head    cq_notify_list;
3054 	struct mlx4_cq *mcq;
3055 	unsigned long flags;
3056 
3057 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3058 	INIT_LIST_HEAD(&cq_notify_list);
3059 
3060 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3061 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3062 
3063 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3064 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3065 		if (mqp->sq.tail != mqp->sq.head) {
3066 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3067 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3068 			if (send_mcq->mcq.comp &&
3069 			    mqp->ibqp.send_cq->comp_handler) {
3070 				if (!send_mcq->mcq.reset_notify_added) {
3071 					send_mcq->mcq.reset_notify_added = 1;
3072 					list_add_tail(&send_mcq->mcq.reset_notify,
3073 						      &cq_notify_list);
3074 				}
3075 			}
3076 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3077 		}
3078 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3079 		/* Now, handle the QP's receive queue */
3080 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3081 		/* no handling is needed for SRQ */
3082 		if (!mqp->ibqp.srq) {
3083 			if (mqp->rq.tail != mqp->rq.head) {
3084 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3085 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3086 				if (recv_mcq->mcq.comp &&
3087 				    mqp->ibqp.recv_cq->comp_handler) {
3088 					if (!recv_mcq->mcq.reset_notify_added) {
3089 						recv_mcq->mcq.reset_notify_added = 1;
3090 						list_add_tail(&recv_mcq->mcq.reset_notify,
3091 							      &cq_notify_list);
3092 					}
3093 				}
3094 				spin_unlock_irqrestore(&recv_mcq->lock,
3095 						       flags_cq);
3096 			}
3097 		}
3098 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3099 	}
3100 
3101 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3102 		mcq->comp(mcq);
3103 	}
3104 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3105 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3106 }
3107 
3108 static void handle_bonded_port_state_event(struct work_struct *work)
3109 {
3110 	struct ib_event_work *ew =
3111 		container_of(work, struct ib_event_work, work);
3112 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3113 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3114 	int i;
3115 	struct ib_event ibev;
3116 
3117 	kfree(ew);
3118 	spin_lock_bh(&ibdev->iboe.lock);
3119 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3120 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3121 		enum ib_port_state curr_port_state;
3122 
3123 		if (!curr_netdev)
3124 			continue;
3125 
3126 		curr_port_state =
3127 			(netif_running(curr_netdev) &&
3128 			 netif_carrier_ok(curr_netdev)) ?
3129 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3130 
3131 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3132 			curr_port_state : IB_PORT_ACTIVE;
3133 	}
3134 	spin_unlock_bh(&ibdev->iboe.lock);
3135 
3136 	ibev.device = &ibdev->ib_dev;
3137 	ibev.element.port_num = 1;
3138 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3139 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3140 
3141 	ib_dispatch_event(&ibev);
3142 }
3143 
3144 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3145 {
3146 	u64 sl2vl;
3147 	int err;
3148 
3149 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3150 	if (err) {
3151 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3152 		       port, err);
3153 		sl2vl = 0;
3154 	}
3155 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3156 }
3157 
3158 static void ib_sl2vl_update_work(struct work_struct *work)
3159 {
3160 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3161 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3162 	int port = ew->port;
3163 
3164 	mlx4_ib_sl2vl_update(mdev, port);
3165 
3166 	kfree(ew);
3167 }
3168 
3169 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3170 				     int port)
3171 {
3172 	struct ib_event_work *ew;
3173 
3174 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3175 	if (ew) {
3176 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3177 		ew->port = port;
3178 		ew->ib_dev = ibdev;
3179 		queue_work(wq, &ew->work);
3180 	}
3181 }
3182 
3183 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3184 			  enum mlx4_dev_event event, unsigned long param)
3185 {
3186 	struct ib_event ibev;
3187 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3188 	struct mlx4_eqe *eqe = NULL;
3189 	struct ib_event_work *ew;
3190 	int p = 0;
3191 
3192 	if (mlx4_is_bonded(dev) &&
3193 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3194 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3195 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3196 		if (!ew)
3197 			return;
3198 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3199 		ew->ib_dev = ibdev;
3200 		queue_work(wq, &ew->work);
3201 		return;
3202 	}
3203 
3204 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3205 		eqe = (struct mlx4_eqe *)param;
3206 	else
3207 		p = (int) param;
3208 
3209 	switch (event) {
3210 	case MLX4_DEV_EVENT_PORT_UP:
3211 		if (p > ibdev->num_ports)
3212 			return;
3213 		if (!mlx4_is_slave(dev) &&
3214 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3215 			IB_LINK_LAYER_INFINIBAND) {
3216 			if (mlx4_is_master(dev))
3217 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3218 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3219 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3220 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3221 		}
3222 		ibev.event = IB_EVENT_PORT_ACTIVE;
3223 		break;
3224 
3225 	case MLX4_DEV_EVENT_PORT_DOWN:
3226 		if (p > ibdev->num_ports)
3227 			return;
3228 		ibev.event = IB_EVENT_PORT_ERR;
3229 		break;
3230 
3231 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3232 		ibdev->ib_active = false;
3233 		ibev.event = IB_EVENT_DEVICE_FATAL;
3234 		mlx4_ib_handle_catas_error(ibdev);
3235 		break;
3236 
3237 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3238 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3239 		if (!ew)
3240 			return;
3241 
3242 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3243 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3244 		ew->ib_dev = ibdev;
3245 		/* need to queue only for port owner, which uses GEN_EQE */
3246 		if (mlx4_is_master(dev))
3247 			queue_work(wq, &ew->work);
3248 		else
3249 			handle_port_mgmt_change_event(&ew->work);
3250 		return;
3251 
3252 	case MLX4_DEV_EVENT_SLAVE_INIT:
3253 		/* here, p is the slave id */
3254 		do_slave_init(ibdev, p, 1);
3255 		if (mlx4_is_master(dev)) {
3256 			int i;
3257 
3258 			for (i = 1; i <= ibdev->num_ports; i++) {
3259 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3260 					== IB_LINK_LAYER_INFINIBAND)
3261 					mlx4_ib_slave_alias_guid_event(ibdev,
3262 								       p, i,
3263 								       1);
3264 			}
3265 		}
3266 		return;
3267 
3268 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3269 		if (mlx4_is_master(dev)) {
3270 			int i;
3271 
3272 			for (i = 1; i <= ibdev->num_ports; i++) {
3273 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3274 					== IB_LINK_LAYER_INFINIBAND)
3275 					mlx4_ib_slave_alias_guid_event(ibdev,
3276 								       p, i,
3277 								       0);
3278 			}
3279 		}
3280 		/* here, p is the slave id */
3281 		do_slave_init(ibdev, p, 0);
3282 		return;
3283 
3284 	default:
3285 		return;
3286 	}
3287 
3288 	ibev.device	      = ibdev_ptr;
3289 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3290 
3291 	ib_dispatch_event(&ibev);
3292 }
3293 
3294 static struct mlx4_interface mlx4_ib_interface = {
3295 	.add		= mlx4_ib_add,
3296 	.remove		= mlx4_ib_remove,
3297 	.event		= mlx4_ib_event,
3298 	.protocol	= MLX4_PROT_IB_IPV6,
3299 	.flags		= MLX4_INTFF_BONDING
3300 };
3301 
3302 static int __init mlx4_ib_init(void)
3303 {
3304 	int err;
3305 
3306 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3307 	if (!wq)
3308 		return -ENOMEM;
3309 
3310 	err = mlx4_ib_cm_init();
3311 	if (err)
3312 		goto clean_wq;
3313 
3314 	err = mlx4_ib_mcg_init();
3315 	if (err)
3316 		goto clean_cm;
3317 
3318 	err = mlx4_register_interface(&mlx4_ib_interface);
3319 	if (err)
3320 		goto clean_mcg;
3321 
3322 	return 0;
3323 
3324 clean_mcg:
3325 	mlx4_ib_mcg_destroy();
3326 
3327 clean_cm:
3328 	mlx4_ib_cm_destroy();
3329 
3330 clean_wq:
3331 	destroy_workqueue(wq);
3332 	return err;
3333 }
3334 
3335 static void __exit mlx4_ib_cleanup(void)
3336 {
3337 	mlx4_unregister_interface(&mlx4_ib_interface);
3338 	mlx4_ib_mcg_destroy();
3339 	mlx4_ib_cm_destroy();
3340 	destroy_workqueue(wq);
3341 }
3342 
3343 module_init(mlx4_ib_init);
3344 module_exit(mlx4_ib_cleanup);
3345