xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision f8e17c17)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static void free_gid_entry(struct gid_entry *entry)
250 {
251 	memset(&entry->gid, 0, sizeof(entry->gid));
252 	kfree(entry->ctx);
253 	entry->ctx = NULL;
254 }
255 
256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids = NULL;
266 	u16 vlan_id = 0xffff;
267 	u8 mac[ETH_ALEN];
268 
269 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
270 		return -EINVAL;
271 
272 	if (attr->port_num > MLX4_MAX_PORTS)
273 		return -EINVAL;
274 
275 	if (!context)
276 		return -EINVAL;
277 
278 	ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
279 	if (ret)
280 		return ret;
281 	port_gid_table = &iboe->gids[attr->port_num - 1];
282 	spin_lock_bh(&iboe->lock);
283 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
284 		if (!memcmp(&port_gid_table->gids[i].gid,
285 			    &attr->gid, sizeof(attr->gid)) &&
286 		    port_gid_table->gids[i].gid_type == attr->gid_type &&
287 		    port_gid_table->gids[i].vlan_id == vlan_id)  {
288 			found = i;
289 			break;
290 		}
291 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
292 			free = i; /* HW has space */
293 	}
294 
295 	if (found < 0) {
296 		if (free < 0) {
297 			ret = -ENOSPC;
298 		} else {
299 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
300 			if (!port_gid_table->gids[free].ctx) {
301 				ret = -ENOMEM;
302 			} else {
303 				*context = port_gid_table->gids[free].ctx;
304 				memcpy(&port_gid_table->gids[free].gid,
305 				       &attr->gid, sizeof(attr->gid));
306 				port_gid_table->gids[free].gid_type = attr->gid_type;
307 				port_gid_table->gids[free].vlan_id = vlan_id;
308 				port_gid_table->gids[free].ctx->real_index = free;
309 				port_gid_table->gids[free].ctx->refcount = 1;
310 				hw_update = 1;
311 			}
312 		}
313 	} else {
314 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
315 		*context = ctx;
316 		ctx->refcount++;
317 	}
318 	if (!ret && hw_update) {
319 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
320 				     GFP_ATOMIC);
321 		if (!gids) {
322 			ret = -ENOMEM;
323 			*context = NULL;
324 			free_gid_entry(&port_gid_table->gids[free]);
325 		} else {
326 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
327 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
328 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
329 			}
330 		}
331 	}
332 	spin_unlock_bh(&iboe->lock);
333 
334 	if (!ret && hw_update) {
335 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
336 		if (ret) {
337 			spin_lock_bh(&iboe->lock);
338 			*context = NULL;
339 			free_gid_entry(&port_gid_table->gids[free]);
340 			spin_unlock_bh(&iboe->lock);
341 		}
342 		kfree(gids);
343 	}
344 
345 	return ret;
346 }
347 
348 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
349 {
350 	struct gid_cache_context *ctx = *context;
351 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
352 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
353 	struct mlx4_port_gid_table   *port_gid_table;
354 	int ret = 0;
355 	int hw_update = 0;
356 	struct gid_entry *gids = NULL;
357 
358 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
359 		return -EINVAL;
360 
361 	if (attr->port_num > MLX4_MAX_PORTS)
362 		return -EINVAL;
363 
364 	port_gid_table = &iboe->gids[attr->port_num - 1];
365 	spin_lock_bh(&iboe->lock);
366 	if (ctx) {
367 		ctx->refcount--;
368 		if (!ctx->refcount) {
369 			unsigned int real_index = ctx->real_index;
370 
371 			free_gid_entry(&port_gid_table->gids[real_index]);
372 			hw_update = 1;
373 		}
374 	}
375 	if (!ret && hw_update) {
376 		int i;
377 
378 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
379 				     GFP_ATOMIC);
380 		if (!gids) {
381 			ret = -ENOMEM;
382 		} else {
383 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
384 				memcpy(&gids[i].gid,
385 				       &port_gid_table->gids[i].gid,
386 				       sizeof(union ib_gid));
387 				gids[i].gid_type =
388 				    port_gid_table->gids[i].gid_type;
389 			}
390 		}
391 	}
392 	spin_unlock_bh(&iboe->lock);
393 
394 	if (!ret && hw_update) {
395 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
396 		kfree(gids);
397 	}
398 	return ret;
399 }
400 
401 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
402 				    const struct ib_gid_attr *attr)
403 {
404 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
405 	struct gid_cache_context *ctx = NULL;
406 	struct mlx4_port_gid_table   *port_gid_table;
407 	int real_index = -EINVAL;
408 	int i;
409 	unsigned long flags;
410 	u8 port_num = attr->port_num;
411 
412 	if (port_num > MLX4_MAX_PORTS)
413 		return -EINVAL;
414 
415 	if (mlx4_is_bonded(ibdev->dev))
416 		port_num = 1;
417 
418 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
419 		return attr->index;
420 
421 	spin_lock_irqsave(&iboe->lock, flags);
422 	port_gid_table = &iboe->gids[port_num - 1];
423 
424 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
425 		if (!memcmp(&port_gid_table->gids[i].gid,
426 			    &attr->gid, sizeof(attr->gid)) &&
427 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
428 			ctx = port_gid_table->gids[i].ctx;
429 			break;
430 		}
431 	if (ctx)
432 		real_index = ctx->real_index;
433 	spin_unlock_irqrestore(&iboe->lock, flags);
434 	return real_index;
435 }
436 
437 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
438 				    sizeof(((type *)0)->fld) <= (sz))
439 
440 static int mlx4_ib_query_device(struct ib_device *ibdev,
441 				struct ib_device_attr *props,
442 				struct ib_udata *uhw)
443 {
444 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
445 	struct ib_smp *in_mad  = NULL;
446 	struct ib_smp *out_mad = NULL;
447 	int err;
448 	int have_ib_ports;
449 	struct mlx4_uverbs_ex_query_device cmd;
450 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
451 	struct mlx4_clock_params clock_params;
452 
453 	if (uhw->inlen) {
454 		if (uhw->inlen < sizeof(cmd))
455 			return -EINVAL;
456 
457 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
458 		if (err)
459 			return err;
460 
461 		if (cmd.comp_mask)
462 			return -EINVAL;
463 
464 		if (cmd.reserved)
465 			return -EINVAL;
466 	}
467 
468 	resp.response_length = offsetof(typeof(resp), response_length) +
469 		sizeof(resp.response_length);
470 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
471 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
472 	err = -ENOMEM;
473 	if (!in_mad || !out_mad)
474 		goto out;
475 
476 	init_query_mad(in_mad);
477 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
478 
479 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
480 			   1, NULL, NULL, in_mad, out_mad);
481 	if (err)
482 		goto out;
483 
484 	memset(props, 0, sizeof *props);
485 
486 	have_ib_ports = num_ib_ports(dev->dev);
487 
488 	props->fw_ver = dev->dev->caps.fw_ver;
489 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
490 		IB_DEVICE_PORT_ACTIVE_EVENT		|
491 		IB_DEVICE_SYS_IMAGE_GUID		|
492 		IB_DEVICE_RC_RNR_NAK_GEN		|
493 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
494 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
495 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
496 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
497 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
498 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
499 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
500 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
501 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
502 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
503 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
504 	if (dev->dev->caps.max_gso_sz &&
505 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
506 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
507 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
508 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
509 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
510 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
511 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
512 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
513 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
514 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
515 		props->device_cap_flags |= IB_DEVICE_XRC;
516 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
517 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
518 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
519 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
520 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
521 		else
522 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
523 	}
524 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
525 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
526 
527 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
528 
529 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
530 		0xffffff;
531 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
532 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
533 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
534 
535 	props->max_mr_size	   = ~0ull;
536 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
537 	props->max_qp		   = dev->dev->quotas.qp;
538 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
539 	props->max_send_sge =
540 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
541 	props->max_recv_sge =
542 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
543 	props->max_sge_rd = MLX4_MAX_SGE_RD;
544 	props->max_cq		   = dev->dev->quotas.cq;
545 	props->max_cqe		   = dev->dev->caps.max_cqes;
546 	props->max_mr		   = dev->dev->quotas.mpt;
547 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
548 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
549 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
550 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
551 	props->max_srq		   = dev->dev->quotas.srq;
552 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
553 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
554 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
555 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
556 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
557 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
558 	props->masked_atomic_cap   = props->atomic_cap;
559 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
560 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
561 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
562 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
563 					   props->max_mcast_grp;
564 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
565 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
566 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
567 	props->max_ah = INT_MAX;
568 
569 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
570 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
571 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
572 			props->rss_caps.max_rwq_indirection_tables =
573 				props->max_qp;
574 			props->rss_caps.max_rwq_indirection_table_size =
575 				dev->dev->caps.max_rss_tbl_sz;
576 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
577 			props->max_wq_type_rq = props->max_qp;
578 		}
579 
580 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
581 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
582 	}
583 
584 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
585 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
586 
587 	if (!mlx4_is_slave(dev->dev))
588 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
589 
590 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
591 		resp.response_length += sizeof(resp.hca_core_clock_offset);
592 		if (!err && !mlx4_is_slave(dev->dev)) {
593 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
594 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
595 		}
596 	}
597 
598 	if (uhw->outlen >= resp.response_length +
599 	    sizeof(resp.max_inl_recv_sz)) {
600 		resp.response_length += sizeof(resp.max_inl_recv_sz);
601 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
602 			sizeof(struct mlx4_wqe_data_seg);
603 	}
604 
605 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
606 		if (props->rss_caps.supported_qpts) {
607 			resp.rss_caps.rx_hash_function =
608 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
609 
610 			resp.rss_caps.rx_hash_fields_mask =
611 				MLX4_IB_RX_HASH_SRC_IPV4 |
612 				MLX4_IB_RX_HASH_DST_IPV4 |
613 				MLX4_IB_RX_HASH_SRC_IPV6 |
614 				MLX4_IB_RX_HASH_DST_IPV6 |
615 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
616 				MLX4_IB_RX_HASH_DST_PORT_TCP |
617 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
618 				MLX4_IB_RX_HASH_DST_PORT_UDP;
619 
620 			if (dev->dev->caps.tunnel_offload_mode ==
621 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
622 				resp.rss_caps.rx_hash_fields_mask |=
623 					MLX4_IB_RX_HASH_INNER;
624 		}
625 		resp.response_length = offsetof(typeof(resp), rss_caps) +
626 				       sizeof(resp.rss_caps);
627 	}
628 
629 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
630 		if (dev->dev->caps.max_gso_sz &&
631 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
632 		    IB_LINK_LAYER_ETHERNET) ||
633 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
634 		    IB_LINK_LAYER_ETHERNET))) {
635 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
636 			resp.tso_caps.supported_qpts |=
637 				1 << IB_QPT_RAW_PACKET;
638 		}
639 		resp.response_length = offsetof(typeof(resp), tso_caps) +
640 				       sizeof(resp.tso_caps);
641 	}
642 
643 	if (uhw->outlen) {
644 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
645 		if (err)
646 			goto out;
647 	}
648 out:
649 	kfree(in_mad);
650 	kfree(out_mad);
651 
652 	return err;
653 }
654 
655 static enum rdma_link_layer
656 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
657 {
658 	struct mlx4_dev *dev = to_mdev(device)->dev;
659 
660 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
661 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
662 }
663 
664 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
665 			      struct ib_port_attr *props, int netw_view)
666 {
667 	struct ib_smp *in_mad  = NULL;
668 	struct ib_smp *out_mad = NULL;
669 	int ext_active_speed;
670 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
671 	int err = -ENOMEM;
672 
673 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
674 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
675 	if (!in_mad || !out_mad)
676 		goto out;
677 
678 	init_query_mad(in_mad);
679 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
680 	in_mad->attr_mod = cpu_to_be32(port);
681 
682 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
683 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
684 
685 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
686 				in_mad, out_mad);
687 	if (err)
688 		goto out;
689 
690 
691 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
692 	props->lmc		= out_mad->data[34] & 0x7;
693 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
694 	props->sm_sl		= out_mad->data[36] & 0xf;
695 	props->state		= out_mad->data[32] & 0xf;
696 	props->phys_state	= out_mad->data[33] >> 4;
697 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
698 	if (netw_view)
699 		props->gid_tbl_len = out_mad->data[50];
700 	else
701 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
702 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
703 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
704 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
705 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
706 	props->active_width	= out_mad->data[31] & 0xf;
707 	props->active_speed	= out_mad->data[35] >> 4;
708 	props->max_mtu		= out_mad->data[41] & 0xf;
709 	props->active_mtu	= out_mad->data[36] >> 4;
710 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
711 	props->max_vl_num	= out_mad->data[37] >> 4;
712 	props->init_type_reply	= out_mad->data[41] >> 4;
713 
714 	/* Check if extended speeds (EDR/FDR/...) are supported */
715 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
716 		ext_active_speed = out_mad->data[62] >> 4;
717 
718 		switch (ext_active_speed) {
719 		case 1:
720 			props->active_speed = IB_SPEED_FDR;
721 			break;
722 		case 2:
723 			props->active_speed = IB_SPEED_EDR;
724 			break;
725 		}
726 	}
727 
728 	/* If reported active speed is QDR, check if is FDR-10 */
729 	if (props->active_speed == IB_SPEED_QDR) {
730 		init_query_mad(in_mad);
731 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
732 		in_mad->attr_mod = cpu_to_be32(port);
733 
734 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
735 				   NULL, NULL, in_mad, out_mad);
736 		if (err)
737 			goto out;
738 
739 		/* Checking LinkSpeedActive for FDR-10 */
740 		if (out_mad->data[15] & 0x1)
741 			props->active_speed = IB_SPEED_FDR10;
742 	}
743 
744 	/* Avoid wrong speed value returned by FW if the IB link is down. */
745 	if (props->state == IB_PORT_DOWN)
746 		 props->active_speed = IB_SPEED_SDR;
747 
748 out:
749 	kfree(in_mad);
750 	kfree(out_mad);
751 	return err;
752 }
753 
754 static u8 state_to_phys_state(enum ib_port_state state)
755 {
756 	return state == IB_PORT_ACTIVE ?
757 		IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
758 }
759 
760 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
761 			       struct ib_port_attr *props)
762 {
763 
764 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
765 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
766 	struct net_device *ndev;
767 	enum ib_mtu tmp;
768 	struct mlx4_cmd_mailbox *mailbox;
769 	int err = 0;
770 	int is_bonded = mlx4_is_bonded(mdev->dev);
771 
772 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
773 	if (IS_ERR(mailbox))
774 		return PTR_ERR(mailbox);
775 
776 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
777 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
778 			   MLX4_CMD_WRAPPED);
779 	if (err)
780 		goto out;
781 
782 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
783 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
784 					   IB_WIDTH_4X : IB_WIDTH_1X;
785 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
786 					   IB_SPEED_FDR : IB_SPEED_QDR;
787 	props->port_cap_flags	= IB_PORT_CM_SUP;
788 	props->ip_gids = true;
789 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
790 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
791 	props->pkey_tbl_len	= 1;
792 	props->max_mtu		= IB_MTU_4096;
793 	props->max_vl_num	= 2;
794 	props->state		= IB_PORT_DOWN;
795 	props->phys_state	= state_to_phys_state(props->state);
796 	props->active_mtu	= IB_MTU_256;
797 	spin_lock_bh(&iboe->lock);
798 	ndev = iboe->netdevs[port - 1];
799 	if (ndev && is_bonded) {
800 		rcu_read_lock(); /* required to get upper dev */
801 		ndev = netdev_master_upper_dev_get_rcu(ndev);
802 		rcu_read_unlock();
803 	}
804 	if (!ndev)
805 		goto out_unlock;
806 
807 	tmp = iboe_get_mtu(ndev->mtu);
808 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
809 
810 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
811 					IB_PORT_ACTIVE : IB_PORT_DOWN;
812 	props->phys_state	= state_to_phys_state(props->state);
813 out_unlock:
814 	spin_unlock_bh(&iboe->lock);
815 out:
816 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
817 	return err;
818 }
819 
820 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
821 			 struct ib_port_attr *props, int netw_view)
822 {
823 	int err;
824 
825 	/* props being zeroed by the caller, avoid zeroing it here */
826 
827 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
828 		ib_link_query_port(ibdev, port, props, netw_view) :
829 				eth_link_query_port(ibdev, port, props);
830 
831 	return err;
832 }
833 
834 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
835 			      struct ib_port_attr *props)
836 {
837 	/* returns host view */
838 	return __mlx4_ib_query_port(ibdev, port, props, 0);
839 }
840 
841 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
842 			union ib_gid *gid, int netw_view)
843 {
844 	struct ib_smp *in_mad  = NULL;
845 	struct ib_smp *out_mad = NULL;
846 	int err = -ENOMEM;
847 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
848 	int clear = 0;
849 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
850 
851 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
852 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
853 	if (!in_mad || !out_mad)
854 		goto out;
855 
856 	init_query_mad(in_mad);
857 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
858 	in_mad->attr_mod = cpu_to_be32(port);
859 
860 	if (mlx4_is_mfunc(dev->dev) && netw_view)
861 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
862 
863 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
864 	if (err)
865 		goto out;
866 
867 	memcpy(gid->raw, out_mad->data + 8, 8);
868 
869 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
870 		if (index) {
871 			/* For any index > 0, return the null guid */
872 			err = 0;
873 			clear = 1;
874 			goto out;
875 		}
876 	}
877 
878 	init_query_mad(in_mad);
879 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
880 	in_mad->attr_mod = cpu_to_be32(index / 8);
881 
882 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
883 			   NULL, NULL, in_mad, out_mad);
884 	if (err)
885 		goto out;
886 
887 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
888 
889 out:
890 	if (clear)
891 		memset(gid->raw + 8, 0, 8);
892 	kfree(in_mad);
893 	kfree(out_mad);
894 	return err;
895 }
896 
897 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
898 			     union ib_gid *gid)
899 {
900 	if (rdma_protocol_ib(ibdev, port))
901 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
902 	return 0;
903 }
904 
905 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
906 {
907 	union sl2vl_tbl_to_u64 sl2vl64;
908 	struct ib_smp *in_mad  = NULL;
909 	struct ib_smp *out_mad = NULL;
910 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
911 	int err = -ENOMEM;
912 	int jj;
913 
914 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
915 		*sl2vl_tbl = 0;
916 		return 0;
917 	}
918 
919 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
920 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
921 	if (!in_mad || !out_mad)
922 		goto out;
923 
924 	init_query_mad(in_mad);
925 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
926 	in_mad->attr_mod = 0;
927 
928 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
929 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
930 
931 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
932 			   in_mad, out_mad);
933 	if (err)
934 		goto out;
935 
936 	for (jj = 0; jj < 8; jj++)
937 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
938 	*sl2vl_tbl = sl2vl64.sl64;
939 
940 out:
941 	kfree(in_mad);
942 	kfree(out_mad);
943 	return err;
944 }
945 
946 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
947 {
948 	u64 sl2vl;
949 	int i;
950 	int err;
951 
952 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
953 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
954 			continue;
955 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
956 		if (err) {
957 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
958 			       i, err);
959 			sl2vl = 0;
960 		}
961 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
962 	}
963 }
964 
965 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
966 			 u16 *pkey, int netw_view)
967 {
968 	struct ib_smp *in_mad  = NULL;
969 	struct ib_smp *out_mad = NULL;
970 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
971 	int err = -ENOMEM;
972 
973 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
974 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
975 	if (!in_mad || !out_mad)
976 		goto out;
977 
978 	init_query_mad(in_mad);
979 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
980 	in_mad->attr_mod = cpu_to_be32(index / 32);
981 
982 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
983 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
984 
985 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
986 			   in_mad, out_mad);
987 	if (err)
988 		goto out;
989 
990 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
991 
992 out:
993 	kfree(in_mad);
994 	kfree(out_mad);
995 	return err;
996 }
997 
998 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
999 {
1000 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
1001 }
1002 
1003 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
1004 				 struct ib_device_modify *props)
1005 {
1006 	struct mlx4_cmd_mailbox *mailbox;
1007 	unsigned long flags;
1008 
1009 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
1010 		return -EOPNOTSUPP;
1011 
1012 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1013 		return 0;
1014 
1015 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
1016 		return -EOPNOTSUPP;
1017 
1018 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1019 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1020 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1021 
1022 	/*
1023 	 * If possible, pass node desc to FW, so it can generate
1024 	 * a 144 trap.  If cmd fails, just ignore.
1025 	 */
1026 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1027 	if (IS_ERR(mailbox))
1028 		return 0;
1029 
1030 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1031 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1032 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1033 
1034 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1035 
1036 	return 0;
1037 }
1038 
1039 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1040 			    u32 cap_mask)
1041 {
1042 	struct mlx4_cmd_mailbox *mailbox;
1043 	int err;
1044 
1045 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1046 	if (IS_ERR(mailbox))
1047 		return PTR_ERR(mailbox);
1048 
1049 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1050 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1051 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1052 	} else {
1053 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1054 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1055 	}
1056 
1057 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1058 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1059 		       MLX4_CMD_WRAPPED);
1060 
1061 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1062 	return err;
1063 }
1064 
1065 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1066 			       struct ib_port_modify *props)
1067 {
1068 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1069 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1070 	struct ib_port_attr attr;
1071 	u32 cap_mask;
1072 	int err;
1073 
1074 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1075 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1076 	 * violations and port capabilities are not meaningful.
1077 	 */
1078 	if (is_eth)
1079 		return 0;
1080 
1081 	mutex_lock(&mdev->cap_mask_mutex);
1082 
1083 	err = ib_query_port(ibdev, port, &attr);
1084 	if (err)
1085 		goto out;
1086 
1087 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1088 		~props->clr_port_cap_mask;
1089 
1090 	err = mlx4_ib_SET_PORT(mdev, port,
1091 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1092 			       cap_mask);
1093 
1094 out:
1095 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1096 	return err;
1097 }
1098 
1099 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1100 				  struct ib_udata *udata)
1101 {
1102 	struct ib_device *ibdev = uctx->device;
1103 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1104 	struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1105 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1106 	struct mlx4_ib_alloc_ucontext_resp resp;
1107 	int err;
1108 
1109 	if (!dev->ib_active)
1110 		return -EAGAIN;
1111 
1112 	if (ibdev->ops.uverbs_abi_ver ==
1113 	    MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1114 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1115 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1116 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1117 	} else {
1118 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1119 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1120 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1121 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1122 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1123 	}
1124 
1125 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1126 	if (err)
1127 		return err;
1128 
1129 	INIT_LIST_HEAD(&context->db_page_list);
1130 	mutex_init(&context->db_page_mutex);
1131 
1132 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1133 	mutex_init(&context->wqn_ranges_mutex);
1134 
1135 	if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1136 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1137 	else
1138 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1139 
1140 	if (err) {
1141 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1142 		return -EFAULT;
1143 	}
1144 
1145 	return err;
1146 }
1147 
1148 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1149 {
1150 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1151 
1152 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1153 }
1154 
1155 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1156 {
1157 }
1158 
1159 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1160 {
1161 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1162 
1163 	switch (vma->vm_pgoff) {
1164 	case 0:
1165 		return rdma_user_mmap_io(context, vma,
1166 					 to_mucontext(context)->uar.pfn,
1167 					 PAGE_SIZE,
1168 					 pgprot_noncached(vma->vm_page_prot),
1169 					 NULL);
1170 
1171 	case 1:
1172 		if (dev->dev->caps.bf_reg_size == 0)
1173 			return -EINVAL;
1174 		return rdma_user_mmap_io(
1175 			context, vma,
1176 			to_mucontext(context)->uar.pfn +
1177 				dev->dev->caps.num_uars,
1178 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1179 			NULL);
1180 
1181 	case 3: {
1182 		struct mlx4_clock_params params;
1183 		int ret;
1184 
1185 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1186 		if (ret)
1187 			return ret;
1188 
1189 		return rdma_user_mmap_io(
1190 			context, vma,
1191 			(pci_resource_start(dev->dev->persist->pdev,
1192 					    params.bar) +
1193 			 params.offset) >>
1194 				PAGE_SHIFT,
1195 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1196 			NULL);
1197 	}
1198 
1199 	default:
1200 		return -EINVAL;
1201 	}
1202 }
1203 
1204 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1205 {
1206 	struct mlx4_ib_pd *pd = to_mpd(ibpd);
1207 	struct ib_device *ibdev = ibpd->device;
1208 	int err;
1209 
1210 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1211 	if (err)
1212 		return err;
1213 
1214 	if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1215 		mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1216 		return -EFAULT;
1217 	}
1218 	return 0;
1219 }
1220 
1221 static void mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1222 {
1223 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1224 }
1225 
1226 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1227 					  struct ib_udata *udata)
1228 {
1229 	struct mlx4_ib_xrcd *xrcd;
1230 	struct ib_cq_init_attr cq_attr = {};
1231 	int err;
1232 
1233 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1234 		return ERR_PTR(-ENOSYS);
1235 
1236 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1237 	if (!xrcd)
1238 		return ERR_PTR(-ENOMEM);
1239 
1240 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1241 	if (err)
1242 		goto err1;
1243 
1244 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1245 	if (IS_ERR(xrcd->pd)) {
1246 		err = PTR_ERR(xrcd->pd);
1247 		goto err2;
1248 	}
1249 
1250 	cq_attr.cqe = 1;
1251 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1252 	if (IS_ERR(xrcd->cq)) {
1253 		err = PTR_ERR(xrcd->cq);
1254 		goto err3;
1255 	}
1256 
1257 	return &xrcd->ibxrcd;
1258 
1259 err3:
1260 	ib_dealloc_pd(xrcd->pd);
1261 err2:
1262 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1263 err1:
1264 	kfree(xrcd);
1265 	return ERR_PTR(err);
1266 }
1267 
1268 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1269 {
1270 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1271 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1272 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1273 	kfree(xrcd);
1274 
1275 	return 0;
1276 }
1277 
1278 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1279 {
1280 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1281 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1282 	struct mlx4_ib_gid_entry *ge;
1283 
1284 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1285 	if (!ge)
1286 		return -ENOMEM;
1287 
1288 	ge->gid = *gid;
1289 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1290 		ge->port = mqp->port;
1291 		ge->added = 1;
1292 	}
1293 
1294 	mutex_lock(&mqp->mutex);
1295 	list_add_tail(&ge->list, &mqp->gid_list);
1296 	mutex_unlock(&mqp->mutex);
1297 
1298 	return 0;
1299 }
1300 
1301 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1302 					  struct mlx4_ib_counters *ctr_table)
1303 {
1304 	struct counter_index *counter, *tmp_count;
1305 
1306 	mutex_lock(&ctr_table->mutex);
1307 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1308 				 list) {
1309 		if (counter->allocated)
1310 			mlx4_counter_free(ibdev->dev, counter->index);
1311 		list_del(&counter->list);
1312 		kfree(counter);
1313 	}
1314 	mutex_unlock(&ctr_table->mutex);
1315 }
1316 
1317 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1318 		   union ib_gid *gid)
1319 {
1320 	struct net_device *ndev;
1321 	int ret = 0;
1322 
1323 	if (!mqp->port)
1324 		return 0;
1325 
1326 	spin_lock_bh(&mdev->iboe.lock);
1327 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1328 	if (ndev)
1329 		dev_hold(ndev);
1330 	spin_unlock_bh(&mdev->iboe.lock);
1331 
1332 	if (ndev) {
1333 		ret = 1;
1334 		dev_put(ndev);
1335 	}
1336 
1337 	return ret;
1338 }
1339 
1340 struct mlx4_ib_steering {
1341 	struct list_head list;
1342 	struct mlx4_flow_reg_id reg_id;
1343 	union ib_gid gid;
1344 };
1345 
1346 #define LAST_ETH_FIELD vlan_tag
1347 #define LAST_IB_FIELD sl
1348 #define LAST_IPV4_FIELD dst_ip
1349 #define LAST_TCP_UDP_FIELD src_port
1350 
1351 /* Field is the last supported field */
1352 #define FIELDS_NOT_SUPPORTED(filter, field)\
1353 	memchr_inv((void *)&filter.field  +\
1354 		   sizeof(filter.field), 0,\
1355 		   sizeof(filter) -\
1356 		   offsetof(typeof(filter), field) -\
1357 		   sizeof(filter.field))
1358 
1359 static int parse_flow_attr(struct mlx4_dev *dev,
1360 			   u32 qp_num,
1361 			   union ib_flow_spec *ib_spec,
1362 			   struct _rule_hw *mlx4_spec)
1363 {
1364 	enum mlx4_net_trans_rule_id type;
1365 
1366 	switch (ib_spec->type) {
1367 	case IB_FLOW_SPEC_ETH:
1368 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1369 			return -ENOTSUPP;
1370 
1371 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1372 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1373 		       ETH_ALEN);
1374 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1375 		       ETH_ALEN);
1376 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1377 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1378 		break;
1379 	case IB_FLOW_SPEC_IB:
1380 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1381 			return -ENOTSUPP;
1382 
1383 		type = MLX4_NET_TRANS_RULE_ID_IB;
1384 		mlx4_spec->ib.l3_qpn =
1385 			cpu_to_be32(qp_num);
1386 		mlx4_spec->ib.qpn_mask =
1387 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1388 		break;
1389 
1390 
1391 	case IB_FLOW_SPEC_IPV4:
1392 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1393 			return -ENOTSUPP;
1394 
1395 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1396 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1397 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1398 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1399 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1400 		break;
1401 
1402 	case IB_FLOW_SPEC_TCP:
1403 	case IB_FLOW_SPEC_UDP:
1404 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1405 			return -ENOTSUPP;
1406 
1407 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1408 					MLX4_NET_TRANS_RULE_ID_TCP :
1409 					MLX4_NET_TRANS_RULE_ID_UDP;
1410 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1411 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1412 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1413 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1414 		break;
1415 
1416 	default:
1417 		return -EINVAL;
1418 	}
1419 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1420 	    mlx4_hw_rule_sz(dev, type) < 0)
1421 		return -EINVAL;
1422 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1423 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1424 	return mlx4_hw_rule_sz(dev, type);
1425 }
1426 
1427 struct default_rules {
1428 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1429 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1430 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1431 	__u8  link_layer;
1432 };
1433 static const struct default_rules default_table[] = {
1434 	{
1435 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1436 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1437 		.rules_create_list = {IB_FLOW_SPEC_IB},
1438 		.link_layer = IB_LINK_LAYER_INFINIBAND
1439 	}
1440 };
1441 
1442 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1443 					 struct ib_flow_attr *flow_attr)
1444 {
1445 	int i, j, k;
1446 	void *ib_flow;
1447 	const struct default_rules *pdefault_rules = default_table;
1448 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1449 
1450 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1451 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1452 		memset(&field_types, 0, sizeof(field_types));
1453 
1454 		if (link_layer != pdefault_rules->link_layer)
1455 			continue;
1456 
1457 		ib_flow = flow_attr + 1;
1458 		/* we assume the specs are sorted */
1459 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1460 		     j < flow_attr->num_of_specs; k++) {
1461 			union ib_flow_spec *current_flow =
1462 				(union ib_flow_spec *)ib_flow;
1463 
1464 			/* same layer but different type */
1465 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1466 			     (pdefault_rules->mandatory_fields[k] &
1467 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1468 			    (current_flow->type !=
1469 			     pdefault_rules->mandatory_fields[k]))
1470 				goto out;
1471 
1472 			/* same layer, try match next one */
1473 			if (current_flow->type ==
1474 			    pdefault_rules->mandatory_fields[k]) {
1475 				j++;
1476 				ib_flow +=
1477 					((union ib_flow_spec *)ib_flow)->size;
1478 			}
1479 		}
1480 
1481 		ib_flow = flow_attr + 1;
1482 		for (j = 0; j < flow_attr->num_of_specs;
1483 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1484 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1485 				/* same layer and same type */
1486 				if (((union ib_flow_spec *)ib_flow)->type ==
1487 				    pdefault_rules->mandatory_not_fields[k])
1488 					goto out;
1489 
1490 		return i;
1491 	}
1492 out:
1493 	return -1;
1494 }
1495 
1496 static int __mlx4_ib_create_default_rules(
1497 		struct mlx4_ib_dev *mdev,
1498 		struct ib_qp *qp,
1499 		const struct default_rules *pdefault_rules,
1500 		struct _rule_hw *mlx4_spec) {
1501 	int size = 0;
1502 	int i;
1503 
1504 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1505 		int ret;
1506 		union ib_flow_spec ib_spec;
1507 		switch (pdefault_rules->rules_create_list[i]) {
1508 		case 0:
1509 			/* no rule */
1510 			continue;
1511 		case IB_FLOW_SPEC_IB:
1512 			ib_spec.type = IB_FLOW_SPEC_IB;
1513 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1514 
1515 			break;
1516 		default:
1517 			/* invalid rule */
1518 			return -EINVAL;
1519 		}
1520 		/* We must put empty rule, qpn is being ignored */
1521 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1522 				      mlx4_spec);
1523 		if (ret < 0) {
1524 			pr_info("invalid parsing\n");
1525 			return -EINVAL;
1526 		}
1527 
1528 		mlx4_spec = (void *)mlx4_spec + ret;
1529 		size += ret;
1530 	}
1531 	return size;
1532 }
1533 
1534 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1535 			  int domain,
1536 			  enum mlx4_net_trans_promisc_mode flow_type,
1537 			  u64 *reg_id)
1538 {
1539 	int ret, i;
1540 	int size = 0;
1541 	void *ib_flow;
1542 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1543 	struct mlx4_cmd_mailbox *mailbox;
1544 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1545 	int default_flow;
1546 
1547 	static const u16 __mlx4_domain[] = {
1548 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1549 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1550 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1551 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1552 	};
1553 
1554 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1555 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1556 		return -EINVAL;
1557 	}
1558 
1559 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1560 		pr_err("Invalid domain value %d\n", domain);
1561 		return -EINVAL;
1562 	}
1563 
1564 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1565 		return -EINVAL;
1566 
1567 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1568 	if (IS_ERR(mailbox))
1569 		return PTR_ERR(mailbox);
1570 	ctrl = mailbox->buf;
1571 
1572 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1573 				 flow_attr->priority);
1574 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1575 	ctrl->port = flow_attr->port;
1576 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1577 
1578 	ib_flow = flow_attr + 1;
1579 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1580 	/* Add default flows */
1581 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1582 	if (default_flow >= 0) {
1583 		ret = __mlx4_ib_create_default_rules(
1584 				mdev, qp, default_table + default_flow,
1585 				mailbox->buf + size);
1586 		if (ret < 0) {
1587 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1588 			return -EINVAL;
1589 		}
1590 		size += ret;
1591 	}
1592 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1593 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1594 				      mailbox->buf + size);
1595 		if (ret < 0) {
1596 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1597 			return -EINVAL;
1598 		}
1599 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1600 		size += ret;
1601 	}
1602 
1603 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1604 	    flow_attr->num_of_specs == 1) {
1605 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1606 		enum ib_flow_spec_type header_spec =
1607 			((union ib_flow_spec *)(flow_attr + 1))->type;
1608 
1609 		if (header_spec == IB_FLOW_SPEC_ETH)
1610 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1611 	}
1612 
1613 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1614 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1615 			   MLX4_CMD_NATIVE);
1616 	if (ret == -ENOMEM)
1617 		pr_err("mcg table is full. Fail to register network rule.\n");
1618 	else if (ret == -ENXIO)
1619 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1620 	else if (ret)
1621 		pr_err("Invalid argument. Fail to register network rule.\n");
1622 
1623 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1624 	return ret;
1625 }
1626 
1627 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1628 {
1629 	int err;
1630 	err = mlx4_cmd(dev, reg_id, 0, 0,
1631 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1632 		       MLX4_CMD_NATIVE);
1633 	if (err)
1634 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1635 		       reg_id);
1636 	return err;
1637 }
1638 
1639 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1640 				    u64 *reg_id)
1641 {
1642 	void *ib_flow;
1643 	union ib_flow_spec *ib_spec;
1644 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1645 	int err = 0;
1646 
1647 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1648 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1649 		return 0; /* do nothing */
1650 
1651 	ib_flow = flow_attr + 1;
1652 	ib_spec = (union ib_flow_spec *)ib_flow;
1653 
1654 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1655 		return 0; /* do nothing */
1656 
1657 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1658 				    flow_attr->port, qp->qp_num,
1659 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1660 				    reg_id);
1661 	return err;
1662 }
1663 
1664 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1665 				      struct ib_flow_attr *flow_attr,
1666 				      enum mlx4_net_trans_promisc_mode *type)
1667 {
1668 	int err = 0;
1669 
1670 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1671 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1672 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1673 		return -EOPNOTSUPP;
1674 	}
1675 
1676 	if (flow_attr->num_of_specs == 0) {
1677 		type[0] = MLX4_FS_MC_SNIFFER;
1678 		type[1] = MLX4_FS_UC_SNIFFER;
1679 	} else {
1680 		union ib_flow_spec *ib_spec;
1681 
1682 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1683 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1684 			return -EINVAL;
1685 
1686 		/* if all is zero than MC and UC */
1687 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1688 			type[0] = MLX4_FS_MC_SNIFFER;
1689 			type[1] = MLX4_FS_UC_SNIFFER;
1690 		} else {
1691 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1692 					    ib_spec->eth.mask.dst_mac[1],
1693 					    ib_spec->eth.mask.dst_mac[2],
1694 					    ib_spec->eth.mask.dst_mac[3],
1695 					    ib_spec->eth.mask.dst_mac[4],
1696 					    ib_spec->eth.mask.dst_mac[5]};
1697 
1698 			/* Above xor was only on MC bit, non empty mask is valid
1699 			 * only if this bit is set and rest are zero.
1700 			 */
1701 			if (!is_zero_ether_addr(&mac[0]))
1702 				return -EINVAL;
1703 
1704 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1705 				type[0] = MLX4_FS_MC_SNIFFER;
1706 			else
1707 				type[0] = MLX4_FS_UC_SNIFFER;
1708 		}
1709 	}
1710 
1711 	return err;
1712 }
1713 
1714 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1715 				    struct ib_flow_attr *flow_attr,
1716 				    int domain, struct ib_udata *udata)
1717 {
1718 	int err = 0, i = 0, j = 0;
1719 	struct mlx4_ib_flow *mflow;
1720 	enum mlx4_net_trans_promisc_mode type[2];
1721 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1722 	int is_bonded = mlx4_is_bonded(dev);
1723 
1724 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1725 		return ERR_PTR(-EINVAL);
1726 
1727 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1728 		return ERR_PTR(-EOPNOTSUPP);
1729 
1730 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1731 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1732 		return ERR_PTR(-EOPNOTSUPP);
1733 
1734 	if (udata &&
1735 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1736 		return ERR_PTR(-EOPNOTSUPP);
1737 
1738 	memset(type, 0, sizeof(type));
1739 
1740 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1741 	if (!mflow) {
1742 		err = -ENOMEM;
1743 		goto err_free;
1744 	}
1745 
1746 	switch (flow_attr->type) {
1747 	case IB_FLOW_ATTR_NORMAL:
1748 		/* If dont trap flag (continue match) is set, under specific
1749 		 * condition traffic be replicated to given qp,
1750 		 * without stealing it
1751 		 */
1752 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1753 			err = mlx4_ib_add_dont_trap_rule(dev,
1754 							 flow_attr,
1755 							 type);
1756 			if (err)
1757 				goto err_free;
1758 		} else {
1759 			type[0] = MLX4_FS_REGULAR;
1760 		}
1761 		break;
1762 
1763 	case IB_FLOW_ATTR_ALL_DEFAULT:
1764 		type[0] = MLX4_FS_ALL_DEFAULT;
1765 		break;
1766 
1767 	case IB_FLOW_ATTR_MC_DEFAULT:
1768 		type[0] = MLX4_FS_MC_DEFAULT;
1769 		break;
1770 
1771 	case IB_FLOW_ATTR_SNIFFER:
1772 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1773 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1774 		break;
1775 
1776 	default:
1777 		err = -EINVAL;
1778 		goto err_free;
1779 	}
1780 
1781 	while (i < ARRAY_SIZE(type) && type[i]) {
1782 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1783 					    &mflow->reg_id[i].id);
1784 		if (err)
1785 			goto err_create_flow;
1786 		if (is_bonded) {
1787 			/* Application always sees one port so the mirror rule
1788 			 * must be on port #2
1789 			 */
1790 			flow_attr->port = 2;
1791 			err = __mlx4_ib_create_flow(qp, flow_attr,
1792 						    domain, type[j],
1793 						    &mflow->reg_id[j].mirror);
1794 			flow_attr->port = 1;
1795 			if (err)
1796 				goto err_create_flow;
1797 			j++;
1798 		}
1799 
1800 		i++;
1801 	}
1802 
1803 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1804 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1805 					       &mflow->reg_id[i].id);
1806 		if (err)
1807 			goto err_create_flow;
1808 
1809 		if (is_bonded) {
1810 			flow_attr->port = 2;
1811 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1812 						       &mflow->reg_id[j].mirror);
1813 			flow_attr->port = 1;
1814 			if (err)
1815 				goto err_create_flow;
1816 			j++;
1817 		}
1818 		/* function to create mirror rule */
1819 		i++;
1820 	}
1821 
1822 	return &mflow->ibflow;
1823 
1824 err_create_flow:
1825 	while (i) {
1826 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1827 					     mflow->reg_id[i].id);
1828 		i--;
1829 	}
1830 
1831 	while (j) {
1832 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1833 					     mflow->reg_id[j].mirror);
1834 		j--;
1835 	}
1836 err_free:
1837 	kfree(mflow);
1838 	return ERR_PTR(err);
1839 }
1840 
1841 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1842 {
1843 	int err, ret = 0;
1844 	int i = 0;
1845 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1846 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1847 
1848 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1849 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1850 		if (err)
1851 			ret = err;
1852 		if (mflow->reg_id[i].mirror) {
1853 			err = __mlx4_ib_destroy_flow(mdev->dev,
1854 						     mflow->reg_id[i].mirror);
1855 			if (err)
1856 				ret = err;
1857 		}
1858 		i++;
1859 	}
1860 
1861 	kfree(mflow);
1862 	return ret;
1863 }
1864 
1865 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1866 {
1867 	int err;
1868 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1869 	struct mlx4_dev	*dev = mdev->dev;
1870 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1871 	struct mlx4_ib_steering *ib_steering = NULL;
1872 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1873 	struct mlx4_flow_reg_id	reg_id;
1874 
1875 	if (mdev->dev->caps.steering_mode ==
1876 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1877 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1878 		if (!ib_steering)
1879 			return -ENOMEM;
1880 	}
1881 
1882 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1883 				    !!(mqp->flags &
1884 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1885 				    prot, &reg_id.id);
1886 	if (err) {
1887 		pr_err("multicast attach op failed, err %d\n", err);
1888 		goto err_malloc;
1889 	}
1890 
1891 	reg_id.mirror = 0;
1892 	if (mlx4_is_bonded(dev)) {
1893 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1894 					    (mqp->port == 1) ? 2 : 1,
1895 					    !!(mqp->flags &
1896 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1897 					    prot, &reg_id.mirror);
1898 		if (err)
1899 			goto err_add;
1900 	}
1901 
1902 	err = add_gid_entry(ibqp, gid);
1903 	if (err)
1904 		goto err_add;
1905 
1906 	if (ib_steering) {
1907 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1908 		ib_steering->reg_id = reg_id;
1909 		mutex_lock(&mqp->mutex);
1910 		list_add(&ib_steering->list, &mqp->steering_rules);
1911 		mutex_unlock(&mqp->mutex);
1912 	}
1913 	return 0;
1914 
1915 err_add:
1916 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1917 			      prot, reg_id.id);
1918 	if (reg_id.mirror)
1919 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1920 				      prot, reg_id.mirror);
1921 err_malloc:
1922 	kfree(ib_steering);
1923 
1924 	return err;
1925 }
1926 
1927 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1928 {
1929 	struct mlx4_ib_gid_entry *ge;
1930 	struct mlx4_ib_gid_entry *tmp;
1931 	struct mlx4_ib_gid_entry *ret = NULL;
1932 
1933 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1934 		if (!memcmp(raw, ge->gid.raw, 16)) {
1935 			ret = ge;
1936 			break;
1937 		}
1938 	}
1939 
1940 	return ret;
1941 }
1942 
1943 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1944 {
1945 	int err;
1946 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1947 	struct mlx4_dev *dev = mdev->dev;
1948 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1949 	struct net_device *ndev;
1950 	struct mlx4_ib_gid_entry *ge;
1951 	struct mlx4_flow_reg_id reg_id = {0, 0};
1952 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1953 
1954 	if (mdev->dev->caps.steering_mode ==
1955 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1956 		struct mlx4_ib_steering *ib_steering;
1957 
1958 		mutex_lock(&mqp->mutex);
1959 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1960 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1961 				list_del(&ib_steering->list);
1962 				break;
1963 			}
1964 		}
1965 		mutex_unlock(&mqp->mutex);
1966 		if (&ib_steering->list == &mqp->steering_rules) {
1967 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1968 			return -EINVAL;
1969 		}
1970 		reg_id = ib_steering->reg_id;
1971 		kfree(ib_steering);
1972 	}
1973 
1974 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1975 				    prot, reg_id.id);
1976 	if (err)
1977 		return err;
1978 
1979 	if (mlx4_is_bonded(dev)) {
1980 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1981 					    prot, reg_id.mirror);
1982 		if (err)
1983 			return err;
1984 	}
1985 
1986 	mutex_lock(&mqp->mutex);
1987 	ge = find_gid_entry(mqp, gid->raw);
1988 	if (ge) {
1989 		spin_lock_bh(&mdev->iboe.lock);
1990 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1991 		if (ndev)
1992 			dev_hold(ndev);
1993 		spin_unlock_bh(&mdev->iboe.lock);
1994 		if (ndev)
1995 			dev_put(ndev);
1996 		list_del(&ge->list);
1997 		kfree(ge);
1998 	} else
1999 		pr_warn("could not find mgid entry\n");
2000 
2001 	mutex_unlock(&mqp->mutex);
2002 
2003 	return 0;
2004 }
2005 
2006 static int init_node_data(struct mlx4_ib_dev *dev)
2007 {
2008 	struct ib_smp *in_mad  = NULL;
2009 	struct ib_smp *out_mad = NULL;
2010 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2011 	int err = -ENOMEM;
2012 
2013 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2014 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2015 	if (!in_mad || !out_mad)
2016 		goto out;
2017 
2018 	init_query_mad(in_mad);
2019 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2020 	if (mlx4_is_master(dev->dev))
2021 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2022 
2023 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2024 	if (err)
2025 		goto out;
2026 
2027 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2028 
2029 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2030 
2031 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2032 	if (err)
2033 		goto out;
2034 
2035 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2036 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2037 
2038 out:
2039 	kfree(in_mad);
2040 	kfree(out_mad);
2041 	return err;
2042 }
2043 
2044 static ssize_t hca_type_show(struct device *device,
2045 			     struct device_attribute *attr, char *buf)
2046 {
2047 	struct mlx4_ib_dev *dev =
2048 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2049 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2050 }
2051 static DEVICE_ATTR_RO(hca_type);
2052 
2053 static ssize_t hw_rev_show(struct device *device,
2054 			   struct device_attribute *attr, char *buf)
2055 {
2056 	struct mlx4_ib_dev *dev =
2057 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2058 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2059 }
2060 static DEVICE_ATTR_RO(hw_rev);
2061 
2062 static ssize_t board_id_show(struct device *device,
2063 			     struct device_attribute *attr, char *buf)
2064 {
2065 	struct mlx4_ib_dev *dev =
2066 		rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2067 
2068 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2069 		       dev->dev->board_id);
2070 }
2071 static DEVICE_ATTR_RO(board_id);
2072 
2073 static struct attribute *mlx4_class_attributes[] = {
2074 	&dev_attr_hw_rev.attr,
2075 	&dev_attr_hca_type.attr,
2076 	&dev_attr_board_id.attr,
2077 	NULL
2078 };
2079 
2080 static const struct attribute_group mlx4_attr_group = {
2081 	.attrs = mlx4_class_attributes,
2082 };
2083 
2084 struct diag_counter {
2085 	const char *name;
2086 	u32 offset;
2087 };
2088 
2089 #define DIAG_COUNTER(_name, _offset)			\
2090 	{ .name = #_name, .offset = _offset }
2091 
2092 static const struct diag_counter diag_basic[] = {
2093 	DIAG_COUNTER(rq_num_lle, 0x00),
2094 	DIAG_COUNTER(sq_num_lle, 0x04),
2095 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2096 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2097 	DIAG_COUNTER(rq_num_lpe, 0x18),
2098 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2099 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2100 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2101 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2102 	DIAG_COUNTER(sq_num_bre, 0x34),
2103 	DIAG_COUNTER(sq_num_rire, 0x44),
2104 	DIAG_COUNTER(rq_num_rire, 0x48),
2105 	DIAG_COUNTER(sq_num_rae, 0x4C),
2106 	DIAG_COUNTER(rq_num_rae, 0x50),
2107 	DIAG_COUNTER(sq_num_roe, 0x54),
2108 	DIAG_COUNTER(sq_num_tree, 0x5C),
2109 	DIAG_COUNTER(sq_num_rree, 0x64),
2110 	DIAG_COUNTER(rq_num_rnr, 0x68),
2111 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2112 	DIAG_COUNTER(rq_num_oos, 0x100),
2113 	DIAG_COUNTER(sq_num_oos, 0x104),
2114 };
2115 
2116 static const struct diag_counter diag_ext[] = {
2117 	DIAG_COUNTER(rq_num_dup, 0x130),
2118 	DIAG_COUNTER(sq_num_to, 0x134),
2119 };
2120 
2121 static const struct diag_counter diag_device_only[] = {
2122 	DIAG_COUNTER(num_cqovf, 0x1A0),
2123 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2124 };
2125 
2126 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2127 						    u8 port_num)
2128 {
2129 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2130 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2131 
2132 	if (!diag[!!port_num].name)
2133 		return NULL;
2134 
2135 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2136 					  diag[!!port_num].num_counters,
2137 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2138 }
2139 
2140 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2141 				struct rdma_hw_stats *stats,
2142 				u8 port, int index)
2143 {
2144 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2145 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2146 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2147 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2148 	int ret;
2149 	int i;
2150 
2151 	ret = mlx4_query_diag_counters(dev->dev,
2152 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2153 				       diag[!!port].offset, hw_value,
2154 				       diag[!!port].num_counters, port);
2155 
2156 	if (ret)
2157 		return ret;
2158 
2159 	for (i = 0; i < diag[!!port].num_counters; i++)
2160 		stats->value[i] = hw_value[i];
2161 
2162 	return diag[!!port].num_counters;
2163 }
2164 
2165 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2166 					 const char ***name,
2167 					 u32 **offset,
2168 					 u32 *num,
2169 					 bool port)
2170 {
2171 	u32 num_counters;
2172 
2173 	num_counters = ARRAY_SIZE(diag_basic);
2174 
2175 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2176 		num_counters += ARRAY_SIZE(diag_ext);
2177 
2178 	if (!port)
2179 		num_counters += ARRAY_SIZE(diag_device_only);
2180 
2181 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2182 	if (!*name)
2183 		return -ENOMEM;
2184 
2185 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2186 	if (!*offset)
2187 		goto err_name;
2188 
2189 	*num = num_counters;
2190 
2191 	return 0;
2192 
2193 err_name:
2194 	kfree(*name);
2195 	return -ENOMEM;
2196 }
2197 
2198 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2199 				       const char **name,
2200 				       u32 *offset,
2201 				       bool port)
2202 {
2203 	int i;
2204 	int j;
2205 
2206 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2207 		name[i] = diag_basic[i].name;
2208 		offset[i] = diag_basic[i].offset;
2209 	}
2210 
2211 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2212 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2213 			name[j] = diag_ext[i].name;
2214 			offset[j] = diag_ext[i].offset;
2215 		}
2216 	}
2217 
2218 	if (!port) {
2219 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2220 			name[j] = diag_device_only[i].name;
2221 			offset[j] = diag_device_only[i].offset;
2222 		}
2223 	}
2224 }
2225 
2226 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2227 	.alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2228 	.get_hw_stats = mlx4_ib_get_hw_stats,
2229 };
2230 
2231 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2232 {
2233 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2234 	int i;
2235 	int ret;
2236 	bool per_port = !!(ibdev->dev->caps.flags2 &
2237 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2238 
2239 	if (mlx4_is_slave(ibdev->dev))
2240 		return 0;
2241 
2242 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2243 		/* i == 1 means we are building port counters */
2244 		if (i && !per_port)
2245 			continue;
2246 
2247 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2248 						    &diag[i].offset,
2249 						    &diag[i].num_counters, i);
2250 		if (ret)
2251 			goto err_alloc;
2252 
2253 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2254 					   diag[i].offset, i);
2255 	}
2256 
2257 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2258 
2259 	return 0;
2260 
2261 err_alloc:
2262 	if (i) {
2263 		kfree(diag[i - 1].name);
2264 		kfree(diag[i - 1].offset);
2265 	}
2266 
2267 	return ret;
2268 }
2269 
2270 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2271 {
2272 	int i;
2273 
2274 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2275 		kfree(ibdev->diag_counters[i].offset);
2276 		kfree(ibdev->diag_counters[i].name);
2277 	}
2278 }
2279 
2280 #define MLX4_IB_INVALID_MAC	((u64)-1)
2281 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2282 			       struct net_device *dev,
2283 			       int port)
2284 {
2285 	u64 new_smac = 0;
2286 	u64 release_mac = MLX4_IB_INVALID_MAC;
2287 	struct mlx4_ib_qp *qp;
2288 
2289 	read_lock(&dev_base_lock);
2290 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2291 	read_unlock(&dev_base_lock);
2292 
2293 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2294 
2295 	/* no need for update QP1 and mac registration in non-SRIOV */
2296 	if (!mlx4_is_mfunc(ibdev->dev))
2297 		return;
2298 
2299 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2300 	qp = ibdev->qp1_proxy[port - 1];
2301 	if (qp) {
2302 		int new_smac_index;
2303 		u64 old_smac;
2304 		struct mlx4_update_qp_params update_params;
2305 
2306 		mutex_lock(&qp->mutex);
2307 		old_smac = qp->pri.smac;
2308 		if (new_smac == old_smac)
2309 			goto unlock;
2310 
2311 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2312 
2313 		if (new_smac_index < 0)
2314 			goto unlock;
2315 
2316 		update_params.smac_index = new_smac_index;
2317 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2318 				   &update_params)) {
2319 			release_mac = new_smac;
2320 			goto unlock;
2321 		}
2322 		/* if old port was zero, no mac was yet registered for this QP */
2323 		if (qp->pri.smac_port)
2324 			release_mac = old_smac;
2325 		qp->pri.smac = new_smac;
2326 		qp->pri.smac_port = port;
2327 		qp->pri.smac_index = new_smac_index;
2328 	}
2329 
2330 unlock:
2331 	if (release_mac != MLX4_IB_INVALID_MAC)
2332 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2333 	if (qp)
2334 		mutex_unlock(&qp->mutex);
2335 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2336 }
2337 
2338 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2339 				 struct net_device *dev,
2340 				 unsigned long event)
2341 
2342 {
2343 	struct mlx4_ib_iboe *iboe;
2344 	int update_qps_port = -1;
2345 	int port;
2346 
2347 	ASSERT_RTNL();
2348 
2349 	iboe = &ibdev->iboe;
2350 
2351 	spin_lock_bh(&iboe->lock);
2352 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2353 
2354 		iboe->netdevs[port - 1] =
2355 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2356 
2357 		if (dev == iboe->netdevs[port - 1] &&
2358 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2359 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2360 			update_qps_port = port;
2361 
2362 		if (dev == iboe->netdevs[port - 1] &&
2363 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2364 			enum ib_port_state port_state;
2365 			struct ib_event ibev = { };
2366 
2367 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2368 						     &port_state))
2369 				continue;
2370 
2371 			if (event == NETDEV_UP &&
2372 			    (port_state != IB_PORT_ACTIVE ||
2373 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2374 				continue;
2375 			if (event == NETDEV_DOWN &&
2376 			    (port_state != IB_PORT_DOWN ||
2377 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2378 				continue;
2379 			iboe->last_port_state[port - 1] = port_state;
2380 
2381 			ibev.device = &ibdev->ib_dev;
2382 			ibev.element.port_num = port;
2383 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2384 							  IB_EVENT_PORT_ERR;
2385 			ib_dispatch_event(&ibev);
2386 		}
2387 
2388 	}
2389 	spin_unlock_bh(&iboe->lock);
2390 
2391 	if (update_qps_port > 0)
2392 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2393 }
2394 
2395 static int mlx4_ib_netdev_event(struct notifier_block *this,
2396 				unsigned long event, void *ptr)
2397 {
2398 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2399 	struct mlx4_ib_dev *ibdev;
2400 
2401 	if (!net_eq(dev_net(dev), &init_net))
2402 		return NOTIFY_DONE;
2403 
2404 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2405 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2406 
2407 	return NOTIFY_DONE;
2408 }
2409 
2410 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2411 {
2412 	int port;
2413 	int slave;
2414 	int i;
2415 
2416 	if (mlx4_is_master(ibdev->dev)) {
2417 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2418 		     ++slave) {
2419 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2420 				for (i = 0;
2421 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2422 				     ++i) {
2423 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2424 					/* master has the identity virt2phys pkey mapping */
2425 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2426 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2427 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2428 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2429 				}
2430 			}
2431 		}
2432 		/* initialize pkey cache */
2433 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2434 			for (i = 0;
2435 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2436 			     ++i)
2437 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2438 					(i) ? 0 : 0xFFFF;
2439 		}
2440 	}
2441 }
2442 
2443 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2444 {
2445 	int i, j, eq = 0, total_eqs = 0;
2446 
2447 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2448 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2449 	if (!ibdev->eq_table)
2450 		return;
2451 
2452 	for (i = 1; i <= dev->caps.num_ports; i++) {
2453 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2454 		     j++, total_eqs++) {
2455 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2456 				continue;
2457 			ibdev->eq_table[eq] = total_eqs;
2458 			if (!mlx4_assign_eq(dev, i,
2459 					    &ibdev->eq_table[eq]))
2460 				eq++;
2461 			else
2462 				ibdev->eq_table[eq] = -1;
2463 		}
2464 	}
2465 
2466 	for (i = eq; i < dev->caps.num_comp_vectors;
2467 	     ibdev->eq_table[i++] = -1)
2468 		;
2469 
2470 	/* Advertise the new number of EQs to clients */
2471 	ibdev->ib_dev.num_comp_vectors = eq;
2472 }
2473 
2474 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2475 {
2476 	int i;
2477 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2478 
2479 	/* no eqs were allocated */
2480 	if (!ibdev->eq_table)
2481 		return;
2482 
2483 	/* Reset the advertised EQ number */
2484 	ibdev->ib_dev.num_comp_vectors = 0;
2485 
2486 	for (i = 0; i < total_eqs; i++)
2487 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2488 
2489 	kfree(ibdev->eq_table);
2490 	ibdev->eq_table = NULL;
2491 }
2492 
2493 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2494 			       struct ib_port_immutable *immutable)
2495 {
2496 	struct ib_port_attr attr;
2497 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2498 	int err;
2499 
2500 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2501 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2502 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2503 	} else {
2504 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2505 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2506 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2507 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2508 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2509 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2510 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2511 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2512 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2513 	}
2514 
2515 	err = ib_query_port(ibdev, port_num, &attr);
2516 	if (err)
2517 		return err;
2518 
2519 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2520 	immutable->gid_tbl_len = attr.gid_tbl_len;
2521 
2522 	return 0;
2523 }
2524 
2525 static void get_fw_ver_str(struct ib_device *device, char *str)
2526 {
2527 	struct mlx4_ib_dev *dev =
2528 		container_of(device, struct mlx4_ib_dev, ib_dev);
2529 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2530 		 (int) (dev->dev->caps.fw_ver >> 32),
2531 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2532 		 (int) dev->dev->caps.fw_ver & 0xffff);
2533 }
2534 
2535 static const struct ib_device_ops mlx4_ib_dev_ops = {
2536 	.owner = THIS_MODULE,
2537 	.driver_id = RDMA_DRIVER_MLX4,
2538 	.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2539 
2540 	.add_gid = mlx4_ib_add_gid,
2541 	.alloc_mr = mlx4_ib_alloc_mr,
2542 	.alloc_pd = mlx4_ib_alloc_pd,
2543 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2544 	.attach_mcast = mlx4_ib_mcg_attach,
2545 	.create_ah = mlx4_ib_create_ah,
2546 	.create_cq = mlx4_ib_create_cq,
2547 	.create_qp = mlx4_ib_create_qp,
2548 	.create_srq = mlx4_ib_create_srq,
2549 	.dealloc_pd = mlx4_ib_dealloc_pd,
2550 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2551 	.del_gid = mlx4_ib_del_gid,
2552 	.dereg_mr = mlx4_ib_dereg_mr,
2553 	.destroy_ah = mlx4_ib_destroy_ah,
2554 	.destroy_cq = mlx4_ib_destroy_cq,
2555 	.destroy_qp = mlx4_ib_destroy_qp,
2556 	.destroy_srq = mlx4_ib_destroy_srq,
2557 	.detach_mcast = mlx4_ib_mcg_detach,
2558 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2559 	.drain_rq = mlx4_ib_drain_rq,
2560 	.drain_sq = mlx4_ib_drain_sq,
2561 	.get_dev_fw_str = get_fw_ver_str,
2562 	.get_dma_mr = mlx4_ib_get_dma_mr,
2563 	.get_link_layer = mlx4_ib_port_link_layer,
2564 	.get_netdev = mlx4_ib_get_netdev,
2565 	.get_port_immutable = mlx4_port_immutable,
2566 	.map_mr_sg = mlx4_ib_map_mr_sg,
2567 	.mmap = mlx4_ib_mmap,
2568 	.modify_cq = mlx4_ib_modify_cq,
2569 	.modify_device = mlx4_ib_modify_device,
2570 	.modify_port = mlx4_ib_modify_port,
2571 	.modify_qp = mlx4_ib_modify_qp,
2572 	.modify_srq = mlx4_ib_modify_srq,
2573 	.poll_cq = mlx4_ib_poll_cq,
2574 	.post_recv = mlx4_ib_post_recv,
2575 	.post_send = mlx4_ib_post_send,
2576 	.post_srq_recv = mlx4_ib_post_srq_recv,
2577 	.process_mad = mlx4_ib_process_mad,
2578 	.query_ah = mlx4_ib_query_ah,
2579 	.query_device = mlx4_ib_query_device,
2580 	.query_gid = mlx4_ib_query_gid,
2581 	.query_pkey = mlx4_ib_query_pkey,
2582 	.query_port = mlx4_ib_query_port,
2583 	.query_qp = mlx4_ib_query_qp,
2584 	.query_srq = mlx4_ib_query_srq,
2585 	.reg_user_mr = mlx4_ib_reg_user_mr,
2586 	.req_notify_cq = mlx4_ib_arm_cq,
2587 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2588 	.resize_cq = mlx4_ib_resize_cq,
2589 
2590 	INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2591 	INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2592 	INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2593 	INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2594 	INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2595 };
2596 
2597 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2598 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2599 	.create_wq = mlx4_ib_create_wq,
2600 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2601 	.destroy_wq = mlx4_ib_destroy_wq,
2602 	.modify_wq = mlx4_ib_modify_wq,
2603 };
2604 
2605 static const struct ib_device_ops mlx4_ib_dev_fmr_ops = {
2606 	.alloc_fmr = mlx4_ib_fmr_alloc,
2607 	.dealloc_fmr = mlx4_ib_fmr_dealloc,
2608 	.map_phys_fmr = mlx4_ib_map_phys_fmr,
2609 	.unmap_fmr = mlx4_ib_unmap_fmr,
2610 };
2611 
2612 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2613 	.alloc_mw = mlx4_ib_alloc_mw,
2614 	.dealloc_mw = mlx4_ib_dealloc_mw,
2615 };
2616 
2617 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2618 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2619 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2620 };
2621 
2622 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2623 	.create_flow = mlx4_ib_create_flow,
2624 	.destroy_flow = mlx4_ib_destroy_flow,
2625 };
2626 
2627 static void *mlx4_ib_add(struct mlx4_dev *dev)
2628 {
2629 	struct mlx4_ib_dev *ibdev;
2630 	int num_ports = 0;
2631 	int i, j;
2632 	int err;
2633 	struct mlx4_ib_iboe *iboe;
2634 	int ib_num_ports = 0;
2635 	int num_req_counters;
2636 	int allocated;
2637 	u32 counter_index;
2638 	struct counter_index *new_counter_index = NULL;
2639 
2640 	pr_info_once("%s", mlx4_ib_version);
2641 
2642 	num_ports = 0;
2643 	mlx4_foreach_ib_transport_port(i, dev)
2644 		num_ports++;
2645 
2646 	/* No point in registering a device with no ports... */
2647 	if (num_ports == 0)
2648 		return NULL;
2649 
2650 	ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2651 	if (!ibdev) {
2652 		dev_err(&dev->persist->pdev->dev,
2653 			"Device struct alloc failed\n");
2654 		return NULL;
2655 	}
2656 
2657 	iboe = &ibdev->iboe;
2658 
2659 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2660 		goto err_dealloc;
2661 
2662 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2663 		goto err_pd;
2664 
2665 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2666 				 PAGE_SIZE);
2667 	if (!ibdev->uar_map)
2668 		goto err_uar;
2669 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2670 
2671 	ibdev->dev = dev;
2672 	ibdev->bond_next_port	= 0;
2673 
2674 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2675 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2676 	ibdev->num_ports		= num_ports;
2677 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2678 						1 : ibdev->num_ports;
2679 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2680 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2681 
2682 	ibdev->ib_dev.uverbs_cmd_mask	=
2683 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2684 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2685 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2686 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2687 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2688 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2689 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2690 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2691 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2692 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2693 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2694 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2695 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2696 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2697 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2698 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2699 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2700 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2701 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2702 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2703 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2704 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2705 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2706 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2707 
2708 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2709 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2710 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2711 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2712 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2713 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2714 
2715 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2716 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2717 	    IB_LINK_LAYER_ETHERNET) ||
2718 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2719 	    IB_LINK_LAYER_ETHERNET))) {
2720 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2721 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2722 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2723 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2724 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2725 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2726 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2727 	}
2728 
2729 	if (!mlx4_is_slave(ibdev->dev))
2730 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fmr_ops);
2731 
2732 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2733 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2734 		ibdev->ib_dev.uverbs_cmd_mask |=
2735 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2736 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2737 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2738 	}
2739 
2740 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2741 		ibdev->ib_dev.uverbs_cmd_mask |=
2742 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2743 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2744 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2745 	}
2746 
2747 	if (check_flow_steering_support(dev)) {
2748 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2749 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2750 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2751 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2752 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2753 	}
2754 
2755 	if (!dev->caps.userspace_caps)
2756 		ibdev->ib_dev.ops.uverbs_abi_ver =
2757 			MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2758 
2759 	mlx4_ib_alloc_eqs(dev, ibdev);
2760 
2761 	spin_lock_init(&iboe->lock);
2762 
2763 	if (init_node_data(ibdev))
2764 		goto err_map;
2765 	mlx4_init_sl2vl_tbl(ibdev);
2766 
2767 	for (i = 0; i < ibdev->num_ports; ++i) {
2768 		mutex_init(&ibdev->counters_table[i].mutex);
2769 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2770 		iboe->last_port_state[i] = IB_PORT_DOWN;
2771 	}
2772 
2773 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2774 	for (i = 0; i < num_req_counters; ++i) {
2775 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2776 		allocated = 0;
2777 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2778 						IB_LINK_LAYER_ETHERNET) {
2779 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2780 						 MLX4_RES_USAGE_DRIVER);
2781 			/* if failed to allocate a new counter, use default */
2782 			if (err)
2783 				counter_index =
2784 					mlx4_get_default_counter_index(dev,
2785 								       i + 1);
2786 			else
2787 				allocated = 1;
2788 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2789 			counter_index = mlx4_get_default_counter_index(dev,
2790 								       i + 1);
2791 		}
2792 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2793 					    GFP_KERNEL);
2794 		if (!new_counter_index) {
2795 			if (allocated)
2796 				mlx4_counter_free(ibdev->dev, counter_index);
2797 			goto err_counter;
2798 		}
2799 		new_counter_index->index = counter_index;
2800 		new_counter_index->allocated = allocated;
2801 		list_add_tail(&new_counter_index->list,
2802 			      &ibdev->counters_table[i].counters_list);
2803 		ibdev->counters_table[i].default_counter = counter_index;
2804 		pr_info("counter index %d for port %d allocated %d\n",
2805 			counter_index, i + 1, allocated);
2806 	}
2807 	if (mlx4_is_bonded(dev))
2808 		for (i = 1; i < ibdev->num_ports ; ++i) {
2809 			new_counter_index =
2810 					kmalloc(sizeof(struct counter_index),
2811 						GFP_KERNEL);
2812 			if (!new_counter_index)
2813 				goto err_counter;
2814 			new_counter_index->index = counter_index;
2815 			new_counter_index->allocated = 0;
2816 			list_add_tail(&new_counter_index->list,
2817 				      &ibdev->counters_table[i].counters_list);
2818 			ibdev->counters_table[i].default_counter =
2819 								counter_index;
2820 		}
2821 
2822 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2823 		ib_num_ports++;
2824 
2825 	spin_lock_init(&ibdev->sm_lock);
2826 	mutex_init(&ibdev->cap_mask_mutex);
2827 	INIT_LIST_HEAD(&ibdev->qp_list);
2828 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2829 
2830 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2831 	    ib_num_ports) {
2832 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2833 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2834 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2835 					    &ibdev->steer_qpn_base, 0,
2836 					    MLX4_RES_USAGE_DRIVER);
2837 		if (err)
2838 			goto err_counter;
2839 
2840 		ibdev->ib_uc_qpns_bitmap =
2841 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2842 				      sizeof(long),
2843 				      GFP_KERNEL);
2844 		if (!ibdev->ib_uc_qpns_bitmap)
2845 			goto err_steer_qp_release;
2846 
2847 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2848 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2849 				    ibdev->steer_qpn_count);
2850 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2851 					dev, ibdev->steer_qpn_base,
2852 					ibdev->steer_qpn_base +
2853 					ibdev->steer_qpn_count - 1);
2854 			if (err)
2855 				goto err_steer_free_bitmap;
2856 		} else {
2857 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2858 				    ibdev->steer_qpn_count);
2859 		}
2860 	}
2861 
2862 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2863 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2864 
2865 	if (mlx4_ib_alloc_diag_counters(ibdev))
2866 		goto err_steer_free_bitmap;
2867 
2868 	rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2869 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d"))
2870 		goto err_diag_counters;
2871 
2872 	if (mlx4_ib_mad_init(ibdev))
2873 		goto err_reg;
2874 
2875 	if (mlx4_ib_init_sriov(ibdev))
2876 		goto err_mad;
2877 
2878 	if (!iboe->nb.notifier_call) {
2879 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2880 		err = register_netdevice_notifier(&iboe->nb);
2881 		if (err) {
2882 			iboe->nb.notifier_call = NULL;
2883 			goto err_notif;
2884 		}
2885 	}
2886 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2887 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2888 		if (err)
2889 			goto err_notif;
2890 	}
2891 
2892 	ibdev->ib_active = true;
2893 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2894 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2895 					 &ibdev->ib_dev);
2896 
2897 	if (mlx4_is_mfunc(ibdev->dev))
2898 		init_pkeys(ibdev);
2899 
2900 	/* create paravirt contexts for any VFs which are active */
2901 	if (mlx4_is_master(ibdev->dev)) {
2902 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2903 			if (j == mlx4_master_func_num(ibdev->dev))
2904 				continue;
2905 			if (mlx4_is_slave_active(ibdev->dev, j))
2906 				do_slave_init(ibdev, j, 1);
2907 		}
2908 	}
2909 	return ibdev;
2910 
2911 err_notif:
2912 	if (ibdev->iboe.nb.notifier_call) {
2913 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2914 			pr_warn("failure unregistering notifier\n");
2915 		ibdev->iboe.nb.notifier_call = NULL;
2916 	}
2917 	flush_workqueue(wq);
2918 
2919 	mlx4_ib_close_sriov(ibdev);
2920 
2921 err_mad:
2922 	mlx4_ib_mad_cleanup(ibdev);
2923 
2924 err_reg:
2925 	ib_unregister_device(&ibdev->ib_dev);
2926 
2927 err_diag_counters:
2928 	mlx4_ib_diag_cleanup(ibdev);
2929 
2930 err_steer_free_bitmap:
2931 	kfree(ibdev->ib_uc_qpns_bitmap);
2932 
2933 err_steer_qp_release:
2934 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2935 			      ibdev->steer_qpn_count);
2936 err_counter:
2937 	for (i = 0; i < ibdev->num_ports; ++i)
2938 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2939 
2940 err_map:
2941 	mlx4_ib_free_eqs(dev, ibdev);
2942 	iounmap(ibdev->uar_map);
2943 
2944 err_uar:
2945 	mlx4_uar_free(dev, &ibdev->priv_uar);
2946 
2947 err_pd:
2948 	mlx4_pd_free(dev, ibdev->priv_pdn);
2949 
2950 err_dealloc:
2951 	ib_dealloc_device(&ibdev->ib_dev);
2952 
2953 	return NULL;
2954 }
2955 
2956 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2957 {
2958 	int offset;
2959 
2960 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2961 
2962 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2963 					 dev->steer_qpn_count,
2964 					 get_count_order(count));
2965 	if (offset < 0)
2966 		return offset;
2967 
2968 	*qpn = dev->steer_qpn_base + offset;
2969 	return 0;
2970 }
2971 
2972 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2973 {
2974 	if (!qpn ||
2975 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2976 		return;
2977 
2978 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2979 		 qpn, dev->steer_qpn_base))
2980 		/* not supposed to be here */
2981 		return;
2982 
2983 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2984 			      qpn - dev->steer_qpn_base,
2985 			      get_count_order(count));
2986 }
2987 
2988 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2989 			 int is_attach)
2990 {
2991 	int err;
2992 	size_t flow_size;
2993 	struct ib_flow_attr *flow = NULL;
2994 	struct ib_flow_spec_ib *ib_spec;
2995 
2996 	if (is_attach) {
2997 		flow_size = sizeof(struct ib_flow_attr) +
2998 			    sizeof(struct ib_flow_spec_ib);
2999 		flow = kzalloc(flow_size, GFP_KERNEL);
3000 		if (!flow)
3001 			return -ENOMEM;
3002 		flow->port = mqp->port;
3003 		flow->num_of_specs = 1;
3004 		flow->size = flow_size;
3005 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3006 		ib_spec->type = IB_FLOW_SPEC_IB;
3007 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3008 		/* Add an empty rule for IB L2 */
3009 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3010 
3011 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3012 					    IB_FLOW_DOMAIN_NIC,
3013 					    MLX4_FS_REGULAR,
3014 					    &mqp->reg_id);
3015 	} else {
3016 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3017 	}
3018 	kfree(flow);
3019 	return err;
3020 }
3021 
3022 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3023 {
3024 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3025 	int p;
3026 	int i;
3027 
3028 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3029 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3030 	ibdev->ib_active = false;
3031 	flush_workqueue(wq);
3032 
3033 	if (ibdev->iboe.nb.notifier_call) {
3034 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3035 			pr_warn("failure unregistering notifier\n");
3036 		ibdev->iboe.nb.notifier_call = NULL;
3037 	}
3038 
3039 	mlx4_ib_close_sriov(ibdev);
3040 	mlx4_ib_mad_cleanup(ibdev);
3041 	ib_unregister_device(&ibdev->ib_dev);
3042 	mlx4_ib_diag_cleanup(ibdev);
3043 
3044 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3045 			      ibdev->steer_qpn_count);
3046 	kfree(ibdev->ib_uc_qpns_bitmap);
3047 
3048 	iounmap(ibdev->uar_map);
3049 	for (p = 0; p < ibdev->num_ports; ++p)
3050 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3051 
3052 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3053 		mlx4_CLOSE_PORT(dev, p);
3054 
3055 	mlx4_ib_free_eqs(dev, ibdev);
3056 
3057 	mlx4_uar_free(dev, &ibdev->priv_uar);
3058 	mlx4_pd_free(dev, ibdev->priv_pdn);
3059 	ib_dealloc_device(&ibdev->ib_dev);
3060 }
3061 
3062 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3063 {
3064 	struct mlx4_ib_demux_work **dm = NULL;
3065 	struct mlx4_dev *dev = ibdev->dev;
3066 	int i;
3067 	unsigned long flags;
3068 	struct mlx4_active_ports actv_ports;
3069 	unsigned int ports;
3070 	unsigned int first_port;
3071 
3072 	if (!mlx4_is_master(dev))
3073 		return;
3074 
3075 	actv_ports = mlx4_get_active_ports(dev, slave);
3076 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3077 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3078 
3079 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3080 	if (!dm)
3081 		return;
3082 
3083 	for (i = 0; i < ports; i++) {
3084 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3085 		if (!dm[i]) {
3086 			while (--i >= 0)
3087 				kfree(dm[i]);
3088 			goto out;
3089 		}
3090 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3091 		dm[i]->port = first_port + i + 1;
3092 		dm[i]->slave = slave;
3093 		dm[i]->do_init = do_init;
3094 		dm[i]->dev = ibdev;
3095 	}
3096 	/* initialize or tear down tunnel QPs for the slave */
3097 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3098 	if (!ibdev->sriov.is_going_down) {
3099 		for (i = 0; i < ports; i++)
3100 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3101 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3102 	} else {
3103 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3104 		for (i = 0; i < ports; i++)
3105 			kfree(dm[i]);
3106 	}
3107 out:
3108 	kfree(dm);
3109 	return;
3110 }
3111 
3112 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3113 {
3114 	struct mlx4_ib_qp *mqp;
3115 	unsigned long flags_qp;
3116 	unsigned long flags_cq;
3117 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3118 	struct list_head    cq_notify_list;
3119 	struct mlx4_cq *mcq;
3120 	unsigned long flags;
3121 
3122 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3123 	INIT_LIST_HEAD(&cq_notify_list);
3124 
3125 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3126 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3127 
3128 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3129 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3130 		if (mqp->sq.tail != mqp->sq.head) {
3131 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3132 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3133 			if (send_mcq->mcq.comp &&
3134 			    mqp->ibqp.send_cq->comp_handler) {
3135 				if (!send_mcq->mcq.reset_notify_added) {
3136 					send_mcq->mcq.reset_notify_added = 1;
3137 					list_add_tail(&send_mcq->mcq.reset_notify,
3138 						      &cq_notify_list);
3139 				}
3140 			}
3141 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3142 		}
3143 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3144 		/* Now, handle the QP's receive queue */
3145 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3146 		/* no handling is needed for SRQ */
3147 		if (!mqp->ibqp.srq) {
3148 			if (mqp->rq.tail != mqp->rq.head) {
3149 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3150 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3151 				if (recv_mcq->mcq.comp &&
3152 				    mqp->ibqp.recv_cq->comp_handler) {
3153 					if (!recv_mcq->mcq.reset_notify_added) {
3154 						recv_mcq->mcq.reset_notify_added = 1;
3155 						list_add_tail(&recv_mcq->mcq.reset_notify,
3156 							      &cq_notify_list);
3157 					}
3158 				}
3159 				spin_unlock_irqrestore(&recv_mcq->lock,
3160 						       flags_cq);
3161 			}
3162 		}
3163 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3164 	}
3165 
3166 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3167 		mcq->comp(mcq);
3168 	}
3169 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3170 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3171 }
3172 
3173 static void handle_bonded_port_state_event(struct work_struct *work)
3174 {
3175 	struct ib_event_work *ew =
3176 		container_of(work, struct ib_event_work, work);
3177 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3178 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3179 	int i;
3180 	struct ib_event ibev;
3181 
3182 	kfree(ew);
3183 	spin_lock_bh(&ibdev->iboe.lock);
3184 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3185 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3186 		enum ib_port_state curr_port_state;
3187 
3188 		if (!curr_netdev)
3189 			continue;
3190 
3191 		curr_port_state =
3192 			(netif_running(curr_netdev) &&
3193 			 netif_carrier_ok(curr_netdev)) ?
3194 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3195 
3196 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3197 			curr_port_state : IB_PORT_ACTIVE;
3198 	}
3199 	spin_unlock_bh(&ibdev->iboe.lock);
3200 
3201 	ibev.device = &ibdev->ib_dev;
3202 	ibev.element.port_num = 1;
3203 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3204 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3205 
3206 	ib_dispatch_event(&ibev);
3207 }
3208 
3209 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3210 {
3211 	u64 sl2vl;
3212 	int err;
3213 
3214 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3215 	if (err) {
3216 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3217 		       port, err);
3218 		sl2vl = 0;
3219 	}
3220 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3221 }
3222 
3223 static void ib_sl2vl_update_work(struct work_struct *work)
3224 {
3225 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3226 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3227 	int port = ew->port;
3228 
3229 	mlx4_ib_sl2vl_update(mdev, port);
3230 
3231 	kfree(ew);
3232 }
3233 
3234 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3235 				     int port)
3236 {
3237 	struct ib_event_work *ew;
3238 
3239 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3240 	if (ew) {
3241 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3242 		ew->port = port;
3243 		ew->ib_dev = ibdev;
3244 		queue_work(wq, &ew->work);
3245 	}
3246 }
3247 
3248 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3249 			  enum mlx4_dev_event event, unsigned long param)
3250 {
3251 	struct ib_event ibev;
3252 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3253 	struct mlx4_eqe *eqe = NULL;
3254 	struct ib_event_work *ew;
3255 	int p = 0;
3256 
3257 	if (mlx4_is_bonded(dev) &&
3258 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3259 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3260 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3261 		if (!ew)
3262 			return;
3263 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3264 		ew->ib_dev = ibdev;
3265 		queue_work(wq, &ew->work);
3266 		return;
3267 	}
3268 
3269 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3270 		eqe = (struct mlx4_eqe *)param;
3271 	else
3272 		p = (int) param;
3273 
3274 	switch (event) {
3275 	case MLX4_DEV_EVENT_PORT_UP:
3276 		if (p > ibdev->num_ports)
3277 			return;
3278 		if (!mlx4_is_slave(dev) &&
3279 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3280 			IB_LINK_LAYER_INFINIBAND) {
3281 			if (mlx4_is_master(dev))
3282 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3283 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3284 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3285 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3286 		}
3287 		ibev.event = IB_EVENT_PORT_ACTIVE;
3288 		break;
3289 
3290 	case MLX4_DEV_EVENT_PORT_DOWN:
3291 		if (p > ibdev->num_ports)
3292 			return;
3293 		ibev.event = IB_EVENT_PORT_ERR;
3294 		break;
3295 
3296 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3297 		ibdev->ib_active = false;
3298 		ibev.event = IB_EVENT_DEVICE_FATAL;
3299 		mlx4_ib_handle_catas_error(ibdev);
3300 		break;
3301 
3302 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3303 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3304 		if (!ew)
3305 			break;
3306 
3307 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3308 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3309 		ew->ib_dev = ibdev;
3310 		/* need to queue only for port owner, which uses GEN_EQE */
3311 		if (mlx4_is_master(dev))
3312 			queue_work(wq, &ew->work);
3313 		else
3314 			handle_port_mgmt_change_event(&ew->work);
3315 		return;
3316 
3317 	case MLX4_DEV_EVENT_SLAVE_INIT:
3318 		/* here, p is the slave id */
3319 		do_slave_init(ibdev, p, 1);
3320 		if (mlx4_is_master(dev)) {
3321 			int i;
3322 
3323 			for (i = 1; i <= ibdev->num_ports; i++) {
3324 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3325 					== IB_LINK_LAYER_INFINIBAND)
3326 					mlx4_ib_slave_alias_guid_event(ibdev,
3327 								       p, i,
3328 								       1);
3329 			}
3330 		}
3331 		return;
3332 
3333 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3334 		if (mlx4_is_master(dev)) {
3335 			int i;
3336 
3337 			for (i = 1; i <= ibdev->num_ports; i++) {
3338 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3339 					== IB_LINK_LAYER_INFINIBAND)
3340 					mlx4_ib_slave_alias_guid_event(ibdev,
3341 								       p, i,
3342 								       0);
3343 			}
3344 		}
3345 		/* here, p is the slave id */
3346 		do_slave_init(ibdev, p, 0);
3347 		return;
3348 
3349 	default:
3350 		return;
3351 	}
3352 
3353 	ibev.device	      = ibdev_ptr;
3354 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3355 
3356 	ib_dispatch_event(&ibev);
3357 }
3358 
3359 static struct mlx4_interface mlx4_ib_interface = {
3360 	.add		= mlx4_ib_add,
3361 	.remove		= mlx4_ib_remove,
3362 	.event		= mlx4_ib_event,
3363 	.protocol	= MLX4_PROT_IB_IPV6,
3364 	.flags		= MLX4_INTFF_BONDING
3365 };
3366 
3367 static int __init mlx4_ib_init(void)
3368 {
3369 	int err;
3370 
3371 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3372 	if (!wq)
3373 		return -ENOMEM;
3374 
3375 	err = mlx4_ib_mcg_init();
3376 	if (err)
3377 		goto clean_wq;
3378 
3379 	err = mlx4_register_interface(&mlx4_ib_interface);
3380 	if (err)
3381 		goto clean_mcg;
3382 
3383 	return 0;
3384 
3385 clean_mcg:
3386 	mlx4_ib_mcg_destroy();
3387 
3388 clean_wq:
3389 	destroy_workqueue(wq);
3390 	return err;
3391 }
3392 
3393 static void __exit mlx4_ib_cleanup(void)
3394 {
3395 	mlx4_unregister_interface(&mlx4_ib_interface);
3396 	mlx4_ib_mcg_destroy();
3397 	destroy_workqueue(wq);
3398 }
3399 
3400 module_init(mlx4_ib_init);
3401 module_exit(mlx4_ib_cleanup);
3402