xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 60772e48)
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 			else
223 				memset(&gid_tbl[i].gid, 0, 12);
224 		}
225 	}
226 
227 	err = mlx4_cmd(dev, mailbox->dma,
228 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
229 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
230 		       MLX4_CMD_WRAPPED);
231 	if (mlx4_is_bonded(dev))
232 		err += mlx4_cmd(dev, mailbox->dma,
233 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
234 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
235 				MLX4_CMD_WRAPPED);
236 
237 	mlx4_free_cmd_mailbox(dev, mailbox);
238 	return err;
239 }
240 
241 static int mlx4_ib_update_gids(struct gid_entry *gids,
242 			       struct mlx4_ib_dev *ibdev,
243 			       u8 port_num)
244 {
245 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
246 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
247 
248 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
249 }
250 
251 static int mlx4_ib_add_gid(struct ib_device *device,
252 			   u8 port_num,
253 			   unsigned int index,
254 			   const union ib_gid *gid,
255 			   const struct ib_gid_attr *attr,
256 			   void **context)
257 {
258 	struct mlx4_ib_dev *ibdev = to_mdev(device);
259 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
260 	struct mlx4_port_gid_table   *port_gid_table;
261 	int free = -1, found = -1;
262 	int ret = 0;
263 	int hw_update = 0;
264 	int i;
265 	struct gid_entry *gids = NULL;
266 
267 	if (!rdma_cap_roce_gid_table(device, port_num))
268 		return -EINVAL;
269 
270 	if (port_num > MLX4_MAX_PORTS)
271 		return -EINVAL;
272 
273 	if (!context)
274 		return -EINVAL;
275 
276 	port_gid_table = &iboe->gids[port_num - 1];
277 	spin_lock_bh(&iboe->lock);
278 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
279 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
280 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
281 			found = i;
282 			break;
283 		}
284 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
285 			free = i; /* HW has space */
286 	}
287 
288 	if (found < 0) {
289 		if (free < 0) {
290 			ret = -ENOSPC;
291 		} else {
292 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
293 			if (!port_gid_table->gids[free].ctx) {
294 				ret = -ENOMEM;
295 			} else {
296 				*context = port_gid_table->gids[free].ctx;
297 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
298 				port_gid_table->gids[free].gid_type = attr->gid_type;
299 				port_gid_table->gids[free].ctx->real_index = free;
300 				port_gid_table->gids[free].ctx->refcount = 1;
301 				hw_update = 1;
302 			}
303 		}
304 	} else {
305 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
306 		*context = ctx;
307 		ctx->refcount++;
308 	}
309 	if (!ret && hw_update) {
310 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
311 		if (!gids) {
312 			ret = -ENOMEM;
313 		} else {
314 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
315 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
316 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
317 			}
318 		}
319 	}
320 	spin_unlock_bh(&iboe->lock);
321 
322 	if (!ret && hw_update) {
323 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
324 		kfree(gids);
325 	}
326 
327 	return ret;
328 }
329 
330 static int mlx4_ib_del_gid(struct ib_device *device,
331 			   u8 port_num,
332 			   unsigned int index,
333 			   void **context)
334 {
335 	struct gid_cache_context *ctx = *context;
336 	struct mlx4_ib_dev *ibdev = to_mdev(device);
337 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
338 	struct mlx4_port_gid_table   *port_gid_table;
339 	int ret = 0;
340 	int hw_update = 0;
341 	struct gid_entry *gids = NULL;
342 
343 	if (!rdma_cap_roce_gid_table(device, port_num))
344 		return -EINVAL;
345 
346 	if (port_num > MLX4_MAX_PORTS)
347 		return -EINVAL;
348 
349 	port_gid_table = &iboe->gids[port_num - 1];
350 	spin_lock_bh(&iboe->lock);
351 	if (ctx) {
352 		ctx->refcount--;
353 		if (!ctx->refcount) {
354 			unsigned int real_index = ctx->real_index;
355 
356 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
357 			kfree(port_gid_table->gids[real_index].ctx);
358 			port_gid_table->gids[real_index].ctx = NULL;
359 			hw_update = 1;
360 		}
361 	}
362 	if (!ret && hw_update) {
363 		int i;
364 
365 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
366 		if (!gids) {
367 			ret = -ENOMEM;
368 		} else {
369 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
370 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
371 		}
372 	}
373 	spin_unlock_bh(&iboe->lock);
374 
375 	if (!ret && hw_update) {
376 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
377 		kfree(gids);
378 	}
379 	return ret;
380 }
381 
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383 				    u8 port_num, int index)
384 {
385 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386 	struct gid_cache_context *ctx = NULL;
387 	union ib_gid gid;
388 	struct mlx4_port_gid_table   *port_gid_table;
389 	int real_index = -EINVAL;
390 	int i;
391 	int ret;
392 	unsigned long flags;
393 	struct ib_gid_attr attr;
394 
395 	if (port_num > MLX4_MAX_PORTS)
396 		return -EINVAL;
397 
398 	if (mlx4_is_bonded(ibdev->dev))
399 		port_num = 1;
400 
401 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
402 		return index;
403 
404 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
405 	if (ret)
406 		return ret;
407 
408 	if (attr.ndev)
409 		dev_put(attr.ndev);
410 
411 	if (!memcmp(&gid, &zgid, sizeof(gid)))
412 		return -EINVAL;
413 
414 	spin_lock_irqsave(&iboe->lock, flags);
415 	port_gid_table = &iboe->gids[port_num - 1];
416 
417 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
418 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
419 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
420 			ctx = port_gid_table->gids[i].ctx;
421 			break;
422 		}
423 	if (ctx)
424 		real_index = ctx->real_index;
425 	spin_unlock_irqrestore(&iboe->lock, flags);
426 	return real_index;
427 }
428 
429 static int mlx4_ib_query_device(struct ib_device *ibdev,
430 				struct ib_device_attr *props,
431 				struct ib_udata *uhw)
432 {
433 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
434 	struct ib_smp *in_mad  = NULL;
435 	struct ib_smp *out_mad = NULL;
436 	int err;
437 	int have_ib_ports;
438 	struct mlx4_uverbs_ex_query_device cmd;
439 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
440 	struct mlx4_clock_params clock_params;
441 
442 	if (uhw->inlen) {
443 		if (uhw->inlen < sizeof(cmd))
444 			return -EINVAL;
445 
446 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
447 		if (err)
448 			return err;
449 
450 		if (cmd.comp_mask)
451 			return -EINVAL;
452 
453 		if (cmd.reserved)
454 			return -EINVAL;
455 	}
456 
457 	resp.response_length = offsetof(typeof(resp), response_length) +
458 		sizeof(resp.response_length);
459 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
460 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
461 	err = -ENOMEM;
462 	if (!in_mad || !out_mad)
463 		goto out;
464 
465 	init_query_mad(in_mad);
466 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
467 
468 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
469 			   1, NULL, NULL, in_mad, out_mad);
470 	if (err)
471 		goto out;
472 
473 	memset(props, 0, sizeof *props);
474 
475 	have_ib_ports = num_ib_ports(dev->dev);
476 
477 	props->fw_ver = dev->dev->caps.fw_ver;
478 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
479 		IB_DEVICE_PORT_ACTIVE_EVENT		|
480 		IB_DEVICE_SYS_IMAGE_GUID		|
481 		IB_DEVICE_RC_RNR_NAK_GEN		|
482 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
484 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
485 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
486 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
487 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
488 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
489 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
490 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
491 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
492 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
493 	if (dev->dev->caps.max_gso_sz &&
494 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
495 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
496 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
497 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
498 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
499 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
500 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
501 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
502 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
503 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
504 		props->device_cap_flags |= IB_DEVICE_XRC;
505 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
506 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
507 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
508 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
509 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
510 		else
511 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
512 	}
513 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
514 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
515 
516 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
517 
518 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
519 		0xffffff;
520 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
521 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
522 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
523 
524 	props->max_mr_size	   = ~0ull;
525 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
526 	props->max_qp		   = dev->dev->quotas.qp;
527 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
528 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
529 					 dev->dev->caps.max_rq_sg);
530 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
531 	props->max_cq		   = dev->dev->quotas.cq;
532 	props->max_cqe		   = dev->dev->caps.max_cqes;
533 	props->max_mr		   = dev->dev->quotas.mpt;
534 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
535 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
536 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
537 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
538 	props->max_srq		   = dev->dev->quotas.srq;
539 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
540 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
541 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
542 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
543 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
544 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
545 	props->masked_atomic_cap   = props->atomic_cap;
546 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
547 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
548 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
549 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
550 					   props->max_mcast_grp;
551 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
552 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
553 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
554 	props->max_ah = INT_MAX;
555 
556 	if ((dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
557 	    (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
558 	     mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET)) {
559 		props->rss_caps.max_rwq_indirection_tables = props->max_qp;
560 		props->rss_caps.max_rwq_indirection_table_size =
561 			dev->dev->caps.max_rss_tbl_sz;
562 		props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
563 		props->max_wq_type_rq = props->max_qp;
564 	}
565 
566 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
567 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
568 
569 	if (!mlx4_is_slave(dev->dev))
570 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
571 
572 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
573 		resp.response_length += sizeof(resp.hca_core_clock_offset);
574 		if (!err && !mlx4_is_slave(dev->dev)) {
575 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
576 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
577 		}
578 	}
579 
580 	if (uhw->outlen >= resp.response_length +
581 	    sizeof(resp.max_inl_recv_sz)) {
582 		resp.response_length += sizeof(resp.max_inl_recv_sz);
583 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
584 			sizeof(struct mlx4_wqe_data_seg);
585 	}
586 
587 	if (uhw->outlen >= resp.response_length + sizeof(resp.rss_caps)) {
588 		resp.response_length += sizeof(resp.rss_caps);
589 		if (props->rss_caps.supported_qpts) {
590 			resp.rss_caps.rx_hash_function =
591 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
592 
593 			resp.rss_caps.rx_hash_fields_mask =
594 				MLX4_IB_RX_HASH_SRC_IPV4 |
595 				MLX4_IB_RX_HASH_DST_IPV4 |
596 				MLX4_IB_RX_HASH_SRC_IPV6 |
597 				MLX4_IB_RX_HASH_DST_IPV6 |
598 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
599 				MLX4_IB_RX_HASH_DST_PORT_TCP |
600 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
601 				MLX4_IB_RX_HASH_DST_PORT_UDP;
602 
603 			if (dev->dev->caps.tunnel_offload_mode ==
604 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
605 				resp.rss_caps.rx_hash_fields_mask |=
606 					MLX4_IB_RX_HASH_INNER;
607 		}
608 	}
609 
610 	if (uhw->outlen) {
611 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
612 		if (err)
613 			goto out;
614 	}
615 out:
616 	kfree(in_mad);
617 	kfree(out_mad);
618 
619 	return err;
620 }
621 
622 static enum rdma_link_layer
623 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
624 {
625 	struct mlx4_dev *dev = to_mdev(device)->dev;
626 
627 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
628 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
629 }
630 
631 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
632 			      struct ib_port_attr *props, int netw_view)
633 {
634 	struct ib_smp *in_mad  = NULL;
635 	struct ib_smp *out_mad = NULL;
636 	int ext_active_speed;
637 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
638 	int err = -ENOMEM;
639 
640 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
641 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
642 	if (!in_mad || !out_mad)
643 		goto out;
644 
645 	init_query_mad(in_mad);
646 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
647 	in_mad->attr_mod = cpu_to_be32(port);
648 
649 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
650 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
651 
652 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
653 				in_mad, out_mad);
654 	if (err)
655 		goto out;
656 
657 
658 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
659 	props->lmc		= out_mad->data[34] & 0x7;
660 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
661 	props->sm_sl		= out_mad->data[36] & 0xf;
662 	props->state		= out_mad->data[32] & 0xf;
663 	props->phys_state	= out_mad->data[33] >> 4;
664 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
665 	if (netw_view)
666 		props->gid_tbl_len = out_mad->data[50];
667 	else
668 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
669 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
670 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
671 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
672 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
673 	props->active_width	= out_mad->data[31] & 0xf;
674 	props->active_speed	= out_mad->data[35] >> 4;
675 	props->max_mtu		= out_mad->data[41] & 0xf;
676 	props->active_mtu	= out_mad->data[36] >> 4;
677 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
678 	props->max_vl_num	= out_mad->data[37] >> 4;
679 	props->init_type_reply	= out_mad->data[41] >> 4;
680 
681 	/* Check if extended speeds (EDR/FDR/...) are supported */
682 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
683 		ext_active_speed = out_mad->data[62] >> 4;
684 
685 		switch (ext_active_speed) {
686 		case 1:
687 			props->active_speed = IB_SPEED_FDR;
688 			break;
689 		case 2:
690 			props->active_speed = IB_SPEED_EDR;
691 			break;
692 		}
693 	}
694 
695 	/* If reported active speed is QDR, check if is FDR-10 */
696 	if (props->active_speed == IB_SPEED_QDR) {
697 		init_query_mad(in_mad);
698 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
699 		in_mad->attr_mod = cpu_to_be32(port);
700 
701 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
702 				   NULL, NULL, in_mad, out_mad);
703 		if (err)
704 			goto out;
705 
706 		/* Checking LinkSpeedActive for FDR-10 */
707 		if (out_mad->data[15] & 0x1)
708 			props->active_speed = IB_SPEED_FDR10;
709 	}
710 
711 	/* Avoid wrong speed value returned by FW if the IB link is down. */
712 	if (props->state == IB_PORT_DOWN)
713 		 props->active_speed = IB_SPEED_SDR;
714 
715 out:
716 	kfree(in_mad);
717 	kfree(out_mad);
718 	return err;
719 }
720 
721 static u8 state_to_phys_state(enum ib_port_state state)
722 {
723 	return state == IB_PORT_ACTIVE ? 5 : 3;
724 }
725 
726 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
727 			       struct ib_port_attr *props)
728 {
729 
730 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
731 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
732 	struct net_device *ndev;
733 	enum ib_mtu tmp;
734 	struct mlx4_cmd_mailbox *mailbox;
735 	int err = 0;
736 	int is_bonded = mlx4_is_bonded(mdev->dev);
737 
738 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
739 	if (IS_ERR(mailbox))
740 		return PTR_ERR(mailbox);
741 
742 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
743 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
744 			   MLX4_CMD_WRAPPED);
745 	if (err)
746 		goto out;
747 
748 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
749 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
750 					   IB_WIDTH_4X : IB_WIDTH_1X;
751 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
752 					   IB_SPEED_FDR : IB_SPEED_QDR;
753 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
754 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
755 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
756 	props->pkey_tbl_len	= 1;
757 	props->max_mtu		= IB_MTU_4096;
758 	props->max_vl_num	= 2;
759 	props->state		= IB_PORT_DOWN;
760 	props->phys_state	= state_to_phys_state(props->state);
761 	props->active_mtu	= IB_MTU_256;
762 	spin_lock_bh(&iboe->lock);
763 	ndev = iboe->netdevs[port - 1];
764 	if (ndev && is_bonded) {
765 		rcu_read_lock(); /* required to get upper dev */
766 		ndev = netdev_master_upper_dev_get_rcu(ndev);
767 		rcu_read_unlock();
768 	}
769 	if (!ndev)
770 		goto out_unlock;
771 
772 	tmp = iboe_get_mtu(ndev->mtu);
773 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
774 
775 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
776 					IB_PORT_ACTIVE : IB_PORT_DOWN;
777 	props->phys_state	= state_to_phys_state(props->state);
778 out_unlock:
779 	spin_unlock_bh(&iboe->lock);
780 out:
781 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
782 	return err;
783 }
784 
785 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
786 			 struct ib_port_attr *props, int netw_view)
787 {
788 	int err;
789 
790 	/* props being zeroed by the caller, avoid zeroing it here */
791 
792 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
793 		ib_link_query_port(ibdev, port, props, netw_view) :
794 				eth_link_query_port(ibdev, port, props);
795 
796 	return err;
797 }
798 
799 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
800 			      struct ib_port_attr *props)
801 {
802 	/* returns host view */
803 	return __mlx4_ib_query_port(ibdev, port, props, 0);
804 }
805 
806 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
807 			union ib_gid *gid, int netw_view)
808 {
809 	struct ib_smp *in_mad  = NULL;
810 	struct ib_smp *out_mad = NULL;
811 	int err = -ENOMEM;
812 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
813 	int clear = 0;
814 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
815 
816 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
817 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
818 	if (!in_mad || !out_mad)
819 		goto out;
820 
821 	init_query_mad(in_mad);
822 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
823 	in_mad->attr_mod = cpu_to_be32(port);
824 
825 	if (mlx4_is_mfunc(dev->dev) && netw_view)
826 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
827 
828 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
829 	if (err)
830 		goto out;
831 
832 	memcpy(gid->raw, out_mad->data + 8, 8);
833 
834 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
835 		if (index) {
836 			/* For any index > 0, return the null guid */
837 			err = 0;
838 			clear = 1;
839 			goto out;
840 		}
841 	}
842 
843 	init_query_mad(in_mad);
844 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
845 	in_mad->attr_mod = cpu_to_be32(index / 8);
846 
847 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
848 			   NULL, NULL, in_mad, out_mad);
849 	if (err)
850 		goto out;
851 
852 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
853 
854 out:
855 	if (clear)
856 		memset(gid->raw + 8, 0, 8);
857 	kfree(in_mad);
858 	kfree(out_mad);
859 	return err;
860 }
861 
862 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
863 			     union ib_gid *gid)
864 {
865 	int ret;
866 
867 	if (rdma_protocol_ib(ibdev, port))
868 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
869 
870 	if (!rdma_protocol_roce(ibdev, port))
871 		return -ENODEV;
872 
873 	if (!rdma_cap_roce_gid_table(ibdev, port))
874 		return -ENODEV;
875 
876 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
877 	if (ret == -EAGAIN) {
878 		memcpy(gid, &zgid, sizeof(*gid));
879 		return 0;
880 	}
881 
882 	return ret;
883 }
884 
885 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
886 {
887 	union sl2vl_tbl_to_u64 sl2vl64;
888 	struct ib_smp *in_mad  = NULL;
889 	struct ib_smp *out_mad = NULL;
890 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
891 	int err = -ENOMEM;
892 	int jj;
893 
894 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
895 		*sl2vl_tbl = 0;
896 		return 0;
897 	}
898 
899 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
900 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
901 	if (!in_mad || !out_mad)
902 		goto out;
903 
904 	init_query_mad(in_mad);
905 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
906 	in_mad->attr_mod = 0;
907 
908 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
909 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
910 
911 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
912 			   in_mad, out_mad);
913 	if (err)
914 		goto out;
915 
916 	for (jj = 0; jj < 8; jj++)
917 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
918 	*sl2vl_tbl = sl2vl64.sl64;
919 
920 out:
921 	kfree(in_mad);
922 	kfree(out_mad);
923 	return err;
924 }
925 
926 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
927 {
928 	u64 sl2vl;
929 	int i;
930 	int err;
931 
932 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
933 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
934 			continue;
935 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
936 		if (err) {
937 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
938 			       i, err);
939 			sl2vl = 0;
940 		}
941 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
942 	}
943 }
944 
945 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
946 			 u16 *pkey, int netw_view)
947 {
948 	struct ib_smp *in_mad  = NULL;
949 	struct ib_smp *out_mad = NULL;
950 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
951 	int err = -ENOMEM;
952 
953 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
954 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
955 	if (!in_mad || !out_mad)
956 		goto out;
957 
958 	init_query_mad(in_mad);
959 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
960 	in_mad->attr_mod = cpu_to_be32(index / 32);
961 
962 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
963 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
964 
965 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
966 			   in_mad, out_mad);
967 	if (err)
968 		goto out;
969 
970 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
971 
972 out:
973 	kfree(in_mad);
974 	kfree(out_mad);
975 	return err;
976 }
977 
978 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
979 {
980 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
981 }
982 
983 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
984 				 struct ib_device_modify *props)
985 {
986 	struct mlx4_cmd_mailbox *mailbox;
987 	unsigned long flags;
988 
989 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
990 		return -EOPNOTSUPP;
991 
992 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
993 		return 0;
994 
995 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
996 		return -EOPNOTSUPP;
997 
998 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
999 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1000 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1001 
1002 	/*
1003 	 * If possible, pass node desc to FW, so it can generate
1004 	 * a 144 trap.  If cmd fails, just ignore.
1005 	 */
1006 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1007 	if (IS_ERR(mailbox))
1008 		return 0;
1009 
1010 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1011 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1012 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1013 
1014 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1015 
1016 	return 0;
1017 }
1018 
1019 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1020 			    u32 cap_mask)
1021 {
1022 	struct mlx4_cmd_mailbox *mailbox;
1023 	int err;
1024 
1025 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1026 	if (IS_ERR(mailbox))
1027 		return PTR_ERR(mailbox);
1028 
1029 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1030 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1031 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1032 	} else {
1033 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1034 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1035 	}
1036 
1037 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1038 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1039 		       MLX4_CMD_WRAPPED);
1040 
1041 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1042 	return err;
1043 }
1044 
1045 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1046 			       struct ib_port_modify *props)
1047 {
1048 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1049 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1050 	struct ib_port_attr attr;
1051 	u32 cap_mask;
1052 	int err;
1053 
1054 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1055 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1056 	 * violations and port capabilities are not meaningful.
1057 	 */
1058 	if (is_eth)
1059 		return 0;
1060 
1061 	mutex_lock(&mdev->cap_mask_mutex);
1062 
1063 	err = ib_query_port(ibdev, port, &attr);
1064 	if (err)
1065 		goto out;
1066 
1067 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1068 		~props->clr_port_cap_mask;
1069 
1070 	err = mlx4_ib_SET_PORT(mdev, port,
1071 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1072 			       cap_mask);
1073 
1074 out:
1075 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1076 	return err;
1077 }
1078 
1079 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1080 						  struct ib_udata *udata)
1081 {
1082 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1083 	struct mlx4_ib_ucontext *context;
1084 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1085 	struct mlx4_ib_alloc_ucontext_resp resp;
1086 	int err;
1087 
1088 	if (!dev->ib_active)
1089 		return ERR_PTR(-EAGAIN);
1090 
1091 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1092 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1093 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1094 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1095 	} else {
1096 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1097 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1098 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1099 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1100 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1101 	}
1102 
1103 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1104 	if (!context)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1108 	if (err) {
1109 		kfree(context);
1110 		return ERR_PTR(err);
1111 	}
1112 
1113 	INIT_LIST_HEAD(&context->db_page_list);
1114 	mutex_init(&context->db_page_mutex);
1115 
1116 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1117 	mutex_init(&context->wqn_ranges_mutex);
1118 
1119 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1120 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1121 	else
1122 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1123 
1124 	if (err) {
1125 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1126 		kfree(context);
1127 		return ERR_PTR(-EFAULT);
1128 	}
1129 
1130 	return &context->ibucontext;
1131 }
1132 
1133 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1134 {
1135 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1136 
1137 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1138 	kfree(context);
1139 
1140 	return 0;
1141 }
1142 
1143 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1144 {
1145 	/* vma_open is called when a new VMA is created on top of our VMA.
1146 	 * This is done through either mremap flow or split_vma (usually due
1147 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1148 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1149 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1150 	 * calling us again and trying to do incorrect actions. We assume that
1151 	 * the original vma size is exactly a single page that there will be no
1152 	 * "splitting" operations on.
1153 	 */
1154 	area->vm_ops = NULL;
1155 }
1156 
1157 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1158 {
1159 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1160 
1161 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1162 	 * file itself is closed, therefore no sync is needed with the regular
1163 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1164 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1165 	 * The close operation is usually called under mm->mmap_sem except when
1166 	 * process is exiting.  The exiting case is handled explicitly as part
1167 	 * of mlx4_ib_disassociate_ucontext.
1168 	 */
1169 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1170 				area->vm_private_data;
1171 
1172 	/* set the vma context pointer to null in the mlx4_ib driver's private
1173 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1174 	 */
1175 	mlx4_ib_vma_priv_data->vma = NULL;
1176 }
1177 
1178 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1179 	.open = mlx4_ib_vma_open,
1180 	.close = mlx4_ib_vma_close
1181 };
1182 
1183 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1184 {
1185 	int i;
1186 	int ret = 0;
1187 	struct vm_area_struct *vma;
1188 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1189 	struct task_struct *owning_process  = NULL;
1190 	struct mm_struct   *owning_mm       = NULL;
1191 
1192 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1193 	if (!owning_process)
1194 		return;
1195 
1196 	owning_mm = get_task_mm(owning_process);
1197 	if (!owning_mm) {
1198 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1199 		while (1) {
1200 			/* make sure that task is dead before returning, it may
1201 			 * prevent a rare case of module down in parallel to a
1202 			 * call to mlx4_ib_vma_close.
1203 			 */
1204 			put_task_struct(owning_process);
1205 			usleep_range(1000, 2000);
1206 			owning_process = get_pid_task(ibcontext->tgid,
1207 						      PIDTYPE_PID);
1208 			if (!owning_process ||
1209 			    owning_process->state == TASK_DEAD) {
1210 				pr_info("disassociate ucontext done, task was terminated\n");
1211 				/* in case task was dead need to release the task struct */
1212 				if (owning_process)
1213 					put_task_struct(owning_process);
1214 				return;
1215 			}
1216 		}
1217 	}
1218 
1219 	/* need to protect from a race on closing the vma as part of
1220 	 * mlx4_ib_vma_close().
1221 	 */
1222 	down_write(&owning_mm->mmap_sem);
1223 	for (i = 0; i < HW_BAR_COUNT; i++) {
1224 		vma = context->hw_bar_info[i].vma;
1225 		if (!vma)
1226 			continue;
1227 
1228 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1229 				   context->hw_bar_info[i].vma->vm_start,
1230 				   PAGE_SIZE);
1231 		if (ret) {
1232 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1233 			BUG_ON(1);
1234 		}
1235 
1236 		context->hw_bar_info[i].vma->vm_flags &=
1237 			~(VM_SHARED | VM_MAYSHARE);
1238 		/* context going to be destroyed, should not access ops any more */
1239 		context->hw_bar_info[i].vma->vm_ops = NULL;
1240 	}
1241 
1242 	up_write(&owning_mm->mmap_sem);
1243 	mmput(owning_mm);
1244 	put_task_struct(owning_process);
1245 }
1246 
1247 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1248 				 struct mlx4_ib_vma_private_data *vma_private_data)
1249 {
1250 	vma_private_data->vma = vma;
1251 	vma->vm_private_data = vma_private_data;
1252 	vma->vm_ops =  &mlx4_ib_vm_ops;
1253 }
1254 
1255 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1256 {
1257 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1258 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1259 
1260 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1261 		return -EINVAL;
1262 
1263 	if (vma->vm_pgoff == 0) {
1264 		/* We prevent double mmaping on same context */
1265 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1266 			return -EINVAL;
1267 
1268 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1269 
1270 		if (io_remap_pfn_range(vma, vma->vm_start,
1271 				       to_mucontext(context)->uar.pfn,
1272 				       PAGE_SIZE, vma->vm_page_prot))
1273 			return -EAGAIN;
1274 
1275 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1276 
1277 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1278 		/* We prevent double mmaping on same context */
1279 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1280 			return -EINVAL;
1281 
1282 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1283 
1284 		if (io_remap_pfn_range(vma, vma->vm_start,
1285 				       to_mucontext(context)->uar.pfn +
1286 				       dev->dev->caps.num_uars,
1287 				       PAGE_SIZE, vma->vm_page_prot))
1288 			return -EAGAIN;
1289 
1290 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1291 
1292 	} else if (vma->vm_pgoff == 3) {
1293 		struct mlx4_clock_params params;
1294 		int ret;
1295 
1296 		/* We prevent double mmaping on same context */
1297 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1298 			return -EINVAL;
1299 
1300 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1301 
1302 		if (ret)
1303 			return ret;
1304 
1305 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1306 		if (io_remap_pfn_range(vma, vma->vm_start,
1307 				       (pci_resource_start(dev->dev->persist->pdev,
1308 							   params.bar) +
1309 					params.offset)
1310 				       >> PAGE_SHIFT,
1311 				       PAGE_SIZE, vma->vm_page_prot))
1312 			return -EAGAIN;
1313 
1314 		mlx4_ib_set_vma_data(vma,
1315 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1316 	} else {
1317 		return -EINVAL;
1318 	}
1319 
1320 	return 0;
1321 }
1322 
1323 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1324 				      struct ib_ucontext *context,
1325 				      struct ib_udata *udata)
1326 {
1327 	struct mlx4_ib_pd *pd;
1328 	int err;
1329 
1330 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1331 	if (!pd)
1332 		return ERR_PTR(-ENOMEM);
1333 
1334 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1335 	if (err) {
1336 		kfree(pd);
1337 		return ERR_PTR(err);
1338 	}
1339 
1340 	if (context)
1341 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1342 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1343 			kfree(pd);
1344 			return ERR_PTR(-EFAULT);
1345 		}
1346 
1347 	return &pd->ibpd;
1348 }
1349 
1350 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1351 {
1352 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1353 	kfree(pd);
1354 
1355 	return 0;
1356 }
1357 
1358 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1359 					  struct ib_ucontext *context,
1360 					  struct ib_udata *udata)
1361 {
1362 	struct mlx4_ib_xrcd *xrcd;
1363 	struct ib_cq_init_attr cq_attr = {};
1364 	int err;
1365 
1366 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1367 		return ERR_PTR(-ENOSYS);
1368 
1369 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1370 	if (!xrcd)
1371 		return ERR_PTR(-ENOMEM);
1372 
1373 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1374 	if (err)
1375 		goto err1;
1376 
1377 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1378 	if (IS_ERR(xrcd->pd)) {
1379 		err = PTR_ERR(xrcd->pd);
1380 		goto err2;
1381 	}
1382 
1383 	cq_attr.cqe = 1;
1384 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1385 	if (IS_ERR(xrcd->cq)) {
1386 		err = PTR_ERR(xrcd->cq);
1387 		goto err3;
1388 	}
1389 
1390 	return &xrcd->ibxrcd;
1391 
1392 err3:
1393 	ib_dealloc_pd(xrcd->pd);
1394 err2:
1395 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1396 err1:
1397 	kfree(xrcd);
1398 	return ERR_PTR(err);
1399 }
1400 
1401 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1402 {
1403 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1404 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1405 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1406 	kfree(xrcd);
1407 
1408 	return 0;
1409 }
1410 
1411 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1412 {
1413 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1414 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1415 	struct mlx4_ib_gid_entry *ge;
1416 
1417 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1418 	if (!ge)
1419 		return -ENOMEM;
1420 
1421 	ge->gid = *gid;
1422 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1423 		ge->port = mqp->port;
1424 		ge->added = 1;
1425 	}
1426 
1427 	mutex_lock(&mqp->mutex);
1428 	list_add_tail(&ge->list, &mqp->gid_list);
1429 	mutex_unlock(&mqp->mutex);
1430 
1431 	return 0;
1432 }
1433 
1434 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1435 					  struct mlx4_ib_counters *ctr_table)
1436 {
1437 	struct counter_index *counter, *tmp_count;
1438 
1439 	mutex_lock(&ctr_table->mutex);
1440 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1441 				 list) {
1442 		if (counter->allocated)
1443 			mlx4_counter_free(ibdev->dev, counter->index);
1444 		list_del(&counter->list);
1445 		kfree(counter);
1446 	}
1447 	mutex_unlock(&ctr_table->mutex);
1448 }
1449 
1450 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1451 		   union ib_gid *gid)
1452 {
1453 	struct net_device *ndev;
1454 	int ret = 0;
1455 
1456 	if (!mqp->port)
1457 		return 0;
1458 
1459 	spin_lock_bh(&mdev->iboe.lock);
1460 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1461 	if (ndev)
1462 		dev_hold(ndev);
1463 	spin_unlock_bh(&mdev->iboe.lock);
1464 
1465 	if (ndev) {
1466 		ret = 1;
1467 		dev_put(ndev);
1468 	}
1469 
1470 	return ret;
1471 }
1472 
1473 struct mlx4_ib_steering {
1474 	struct list_head list;
1475 	struct mlx4_flow_reg_id reg_id;
1476 	union ib_gid gid;
1477 };
1478 
1479 #define LAST_ETH_FIELD vlan_tag
1480 #define LAST_IB_FIELD sl
1481 #define LAST_IPV4_FIELD dst_ip
1482 #define LAST_TCP_UDP_FIELD src_port
1483 
1484 /* Field is the last supported field */
1485 #define FIELDS_NOT_SUPPORTED(filter, field)\
1486 	memchr_inv((void *)&filter.field  +\
1487 		   sizeof(filter.field), 0,\
1488 		   sizeof(filter) -\
1489 		   offsetof(typeof(filter), field) -\
1490 		   sizeof(filter.field))
1491 
1492 static int parse_flow_attr(struct mlx4_dev *dev,
1493 			   u32 qp_num,
1494 			   union ib_flow_spec *ib_spec,
1495 			   struct _rule_hw *mlx4_spec)
1496 {
1497 	enum mlx4_net_trans_rule_id type;
1498 
1499 	switch (ib_spec->type) {
1500 	case IB_FLOW_SPEC_ETH:
1501 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1502 			return -ENOTSUPP;
1503 
1504 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1505 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1506 		       ETH_ALEN);
1507 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1508 		       ETH_ALEN);
1509 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1510 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1511 		break;
1512 	case IB_FLOW_SPEC_IB:
1513 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1514 			return -ENOTSUPP;
1515 
1516 		type = MLX4_NET_TRANS_RULE_ID_IB;
1517 		mlx4_spec->ib.l3_qpn =
1518 			cpu_to_be32(qp_num);
1519 		mlx4_spec->ib.qpn_mask =
1520 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1521 		break;
1522 
1523 
1524 	case IB_FLOW_SPEC_IPV4:
1525 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1526 			return -ENOTSUPP;
1527 
1528 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1529 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1530 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1531 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1532 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1533 		break;
1534 
1535 	case IB_FLOW_SPEC_TCP:
1536 	case IB_FLOW_SPEC_UDP:
1537 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1538 			return -ENOTSUPP;
1539 
1540 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1541 					MLX4_NET_TRANS_RULE_ID_TCP :
1542 					MLX4_NET_TRANS_RULE_ID_UDP;
1543 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1544 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1545 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1546 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1547 		break;
1548 
1549 	default:
1550 		return -EINVAL;
1551 	}
1552 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1553 	    mlx4_hw_rule_sz(dev, type) < 0)
1554 		return -EINVAL;
1555 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1556 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1557 	return mlx4_hw_rule_sz(dev, type);
1558 }
1559 
1560 struct default_rules {
1561 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1562 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1563 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1564 	__u8  link_layer;
1565 };
1566 static const struct default_rules default_table[] = {
1567 	{
1568 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1569 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1570 		.rules_create_list = {IB_FLOW_SPEC_IB},
1571 		.link_layer = IB_LINK_LAYER_INFINIBAND
1572 	}
1573 };
1574 
1575 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1576 					 struct ib_flow_attr *flow_attr)
1577 {
1578 	int i, j, k;
1579 	void *ib_flow;
1580 	const struct default_rules *pdefault_rules = default_table;
1581 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1582 
1583 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1584 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1585 		memset(&field_types, 0, sizeof(field_types));
1586 
1587 		if (link_layer != pdefault_rules->link_layer)
1588 			continue;
1589 
1590 		ib_flow = flow_attr + 1;
1591 		/* we assume the specs are sorted */
1592 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1593 		     j < flow_attr->num_of_specs; k++) {
1594 			union ib_flow_spec *current_flow =
1595 				(union ib_flow_spec *)ib_flow;
1596 
1597 			/* same layer but different type */
1598 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1599 			     (pdefault_rules->mandatory_fields[k] &
1600 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1601 			    (current_flow->type !=
1602 			     pdefault_rules->mandatory_fields[k]))
1603 				goto out;
1604 
1605 			/* same layer, try match next one */
1606 			if (current_flow->type ==
1607 			    pdefault_rules->mandatory_fields[k]) {
1608 				j++;
1609 				ib_flow +=
1610 					((union ib_flow_spec *)ib_flow)->size;
1611 			}
1612 		}
1613 
1614 		ib_flow = flow_attr + 1;
1615 		for (j = 0; j < flow_attr->num_of_specs;
1616 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1617 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1618 				/* same layer and same type */
1619 				if (((union ib_flow_spec *)ib_flow)->type ==
1620 				    pdefault_rules->mandatory_not_fields[k])
1621 					goto out;
1622 
1623 		return i;
1624 	}
1625 out:
1626 	return -1;
1627 }
1628 
1629 static int __mlx4_ib_create_default_rules(
1630 		struct mlx4_ib_dev *mdev,
1631 		struct ib_qp *qp,
1632 		const struct default_rules *pdefault_rules,
1633 		struct _rule_hw *mlx4_spec) {
1634 	int size = 0;
1635 	int i;
1636 
1637 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1638 		int ret;
1639 		union ib_flow_spec ib_spec;
1640 		switch (pdefault_rules->rules_create_list[i]) {
1641 		case 0:
1642 			/* no rule */
1643 			continue;
1644 		case IB_FLOW_SPEC_IB:
1645 			ib_spec.type = IB_FLOW_SPEC_IB;
1646 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1647 
1648 			break;
1649 		default:
1650 			/* invalid rule */
1651 			return -EINVAL;
1652 		}
1653 		/* We must put empty rule, qpn is being ignored */
1654 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1655 				      mlx4_spec);
1656 		if (ret < 0) {
1657 			pr_info("invalid parsing\n");
1658 			return -EINVAL;
1659 		}
1660 
1661 		mlx4_spec = (void *)mlx4_spec + ret;
1662 		size += ret;
1663 	}
1664 	return size;
1665 }
1666 
1667 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1668 			  int domain,
1669 			  enum mlx4_net_trans_promisc_mode flow_type,
1670 			  u64 *reg_id)
1671 {
1672 	int ret, i;
1673 	int size = 0;
1674 	void *ib_flow;
1675 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1676 	struct mlx4_cmd_mailbox *mailbox;
1677 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1678 	int default_flow;
1679 
1680 	static const u16 __mlx4_domain[] = {
1681 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1682 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1683 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1684 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1685 	};
1686 
1687 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1688 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1689 		return -EINVAL;
1690 	}
1691 
1692 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1693 		pr_err("Invalid domain value %d\n", domain);
1694 		return -EINVAL;
1695 	}
1696 
1697 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1698 		return -EINVAL;
1699 
1700 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1701 	if (IS_ERR(mailbox))
1702 		return PTR_ERR(mailbox);
1703 	ctrl = mailbox->buf;
1704 
1705 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1706 				 flow_attr->priority);
1707 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1708 	ctrl->port = flow_attr->port;
1709 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1710 
1711 	ib_flow = flow_attr + 1;
1712 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1713 	/* Add default flows */
1714 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1715 	if (default_flow >= 0) {
1716 		ret = __mlx4_ib_create_default_rules(
1717 				mdev, qp, default_table + default_flow,
1718 				mailbox->buf + size);
1719 		if (ret < 0) {
1720 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1721 			return -EINVAL;
1722 		}
1723 		size += ret;
1724 	}
1725 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1726 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1727 				      mailbox->buf + size);
1728 		if (ret < 0) {
1729 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1730 			return -EINVAL;
1731 		}
1732 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1733 		size += ret;
1734 	}
1735 
1736 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1737 	    flow_attr->num_of_specs == 1) {
1738 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1739 		enum ib_flow_spec_type header_spec =
1740 			((union ib_flow_spec *)(flow_attr + 1))->type;
1741 
1742 		if (header_spec == IB_FLOW_SPEC_ETH)
1743 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1744 	}
1745 
1746 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1747 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1748 			   MLX4_CMD_NATIVE);
1749 	if (ret == -ENOMEM)
1750 		pr_err("mcg table is full. Fail to register network rule.\n");
1751 	else if (ret == -ENXIO)
1752 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1753 	else if (ret)
1754 		pr_err("Invalid argument. Fail to register network rule.\n");
1755 
1756 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1757 	return ret;
1758 }
1759 
1760 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1761 {
1762 	int err;
1763 	err = mlx4_cmd(dev, reg_id, 0, 0,
1764 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1765 		       MLX4_CMD_NATIVE);
1766 	if (err)
1767 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1768 		       reg_id);
1769 	return err;
1770 }
1771 
1772 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1773 				    u64 *reg_id)
1774 {
1775 	void *ib_flow;
1776 	union ib_flow_spec *ib_spec;
1777 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1778 	int err = 0;
1779 
1780 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1781 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1782 		return 0; /* do nothing */
1783 
1784 	ib_flow = flow_attr + 1;
1785 	ib_spec = (union ib_flow_spec *)ib_flow;
1786 
1787 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1788 		return 0; /* do nothing */
1789 
1790 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1791 				    flow_attr->port, qp->qp_num,
1792 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1793 				    reg_id);
1794 	return err;
1795 }
1796 
1797 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1798 				      struct ib_flow_attr *flow_attr,
1799 				      enum mlx4_net_trans_promisc_mode *type)
1800 {
1801 	int err = 0;
1802 
1803 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1804 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1805 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1806 		return -EOPNOTSUPP;
1807 	}
1808 
1809 	if (flow_attr->num_of_specs == 0) {
1810 		type[0] = MLX4_FS_MC_SNIFFER;
1811 		type[1] = MLX4_FS_UC_SNIFFER;
1812 	} else {
1813 		union ib_flow_spec *ib_spec;
1814 
1815 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1816 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1817 			return -EINVAL;
1818 
1819 		/* if all is zero than MC and UC */
1820 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1821 			type[0] = MLX4_FS_MC_SNIFFER;
1822 			type[1] = MLX4_FS_UC_SNIFFER;
1823 		} else {
1824 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1825 					    ib_spec->eth.mask.dst_mac[1],
1826 					    ib_spec->eth.mask.dst_mac[2],
1827 					    ib_spec->eth.mask.dst_mac[3],
1828 					    ib_spec->eth.mask.dst_mac[4],
1829 					    ib_spec->eth.mask.dst_mac[5]};
1830 
1831 			/* Above xor was only on MC bit, non empty mask is valid
1832 			 * only if this bit is set and rest are zero.
1833 			 */
1834 			if (!is_zero_ether_addr(&mac[0]))
1835 				return -EINVAL;
1836 
1837 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1838 				type[0] = MLX4_FS_MC_SNIFFER;
1839 			else
1840 				type[0] = MLX4_FS_UC_SNIFFER;
1841 		}
1842 	}
1843 
1844 	return err;
1845 }
1846 
1847 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1848 				    struct ib_flow_attr *flow_attr,
1849 				    int domain)
1850 {
1851 	int err = 0, i = 0, j = 0;
1852 	struct mlx4_ib_flow *mflow;
1853 	enum mlx4_net_trans_promisc_mode type[2];
1854 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1855 	int is_bonded = mlx4_is_bonded(dev);
1856 
1857 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1858 		return ERR_PTR(-EINVAL);
1859 
1860 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1861 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1862 		return ERR_PTR(-EOPNOTSUPP);
1863 
1864 	memset(type, 0, sizeof(type));
1865 
1866 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1867 	if (!mflow) {
1868 		err = -ENOMEM;
1869 		goto err_free;
1870 	}
1871 
1872 	switch (flow_attr->type) {
1873 	case IB_FLOW_ATTR_NORMAL:
1874 		/* If dont trap flag (continue match) is set, under specific
1875 		 * condition traffic be replicated to given qp,
1876 		 * without stealing it
1877 		 */
1878 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1879 			err = mlx4_ib_add_dont_trap_rule(dev,
1880 							 flow_attr,
1881 							 type);
1882 			if (err)
1883 				goto err_free;
1884 		} else {
1885 			type[0] = MLX4_FS_REGULAR;
1886 		}
1887 		break;
1888 
1889 	case IB_FLOW_ATTR_ALL_DEFAULT:
1890 		type[0] = MLX4_FS_ALL_DEFAULT;
1891 		break;
1892 
1893 	case IB_FLOW_ATTR_MC_DEFAULT:
1894 		type[0] = MLX4_FS_MC_DEFAULT;
1895 		break;
1896 
1897 	case IB_FLOW_ATTR_SNIFFER:
1898 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1899 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1900 		break;
1901 
1902 	default:
1903 		err = -EINVAL;
1904 		goto err_free;
1905 	}
1906 
1907 	while (i < ARRAY_SIZE(type) && type[i]) {
1908 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1909 					    &mflow->reg_id[i].id);
1910 		if (err)
1911 			goto err_create_flow;
1912 		if (is_bonded) {
1913 			/* Application always sees one port so the mirror rule
1914 			 * must be on port #2
1915 			 */
1916 			flow_attr->port = 2;
1917 			err = __mlx4_ib_create_flow(qp, flow_attr,
1918 						    domain, type[j],
1919 						    &mflow->reg_id[j].mirror);
1920 			flow_attr->port = 1;
1921 			if (err)
1922 				goto err_create_flow;
1923 			j++;
1924 		}
1925 
1926 		i++;
1927 	}
1928 
1929 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1930 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1931 					       &mflow->reg_id[i].id);
1932 		if (err)
1933 			goto err_create_flow;
1934 
1935 		if (is_bonded) {
1936 			flow_attr->port = 2;
1937 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1938 						       &mflow->reg_id[j].mirror);
1939 			flow_attr->port = 1;
1940 			if (err)
1941 				goto err_create_flow;
1942 			j++;
1943 		}
1944 		/* function to create mirror rule */
1945 		i++;
1946 	}
1947 
1948 	return &mflow->ibflow;
1949 
1950 err_create_flow:
1951 	while (i) {
1952 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1953 					     mflow->reg_id[i].id);
1954 		i--;
1955 	}
1956 
1957 	while (j) {
1958 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1959 					     mflow->reg_id[j].mirror);
1960 		j--;
1961 	}
1962 err_free:
1963 	kfree(mflow);
1964 	return ERR_PTR(err);
1965 }
1966 
1967 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1968 {
1969 	int err, ret = 0;
1970 	int i = 0;
1971 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1972 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1973 
1974 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1975 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1976 		if (err)
1977 			ret = err;
1978 		if (mflow->reg_id[i].mirror) {
1979 			err = __mlx4_ib_destroy_flow(mdev->dev,
1980 						     mflow->reg_id[i].mirror);
1981 			if (err)
1982 				ret = err;
1983 		}
1984 		i++;
1985 	}
1986 
1987 	kfree(mflow);
1988 	return ret;
1989 }
1990 
1991 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1992 {
1993 	int err;
1994 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1995 	struct mlx4_dev	*dev = mdev->dev;
1996 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1997 	struct mlx4_ib_steering *ib_steering = NULL;
1998 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1999 	struct mlx4_flow_reg_id	reg_id;
2000 
2001 	if (mdev->dev->caps.steering_mode ==
2002 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2003 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
2004 		if (!ib_steering)
2005 			return -ENOMEM;
2006 	}
2007 
2008 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
2009 				    !!(mqp->flags &
2010 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2011 				    prot, &reg_id.id);
2012 	if (err) {
2013 		pr_err("multicast attach op failed, err %d\n", err);
2014 		goto err_malloc;
2015 	}
2016 
2017 	reg_id.mirror = 0;
2018 	if (mlx4_is_bonded(dev)) {
2019 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
2020 					    (mqp->port == 1) ? 2 : 1,
2021 					    !!(mqp->flags &
2022 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2023 					    prot, &reg_id.mirror);
2024 		if (err)
2025 			goto err_add;
2026 	}
2027 
2028 	err = add_gid_entry(ibqp, gid);
2029 	if (err)
2030 		goto err_add;
2031 
2032 	if (ib_steering) {
2033 		memcpy(ib_steering->gid.raw, gid->raw, 16);
2034 		ib_steering->reg_id = reg_id;
2035 		mutex_lock(&mqp->mutex);
2036 		list_add(&ib_steering->list, &mqp->steering_rules);
2037 		mutex_unlock(&mqp->mutex);
2038 	}
2039 	return 0;
2040 
2041 err_add:
2042 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2043 			      prot, reg_id.id);
2044 	if (reg_id.mirror)
2045 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2046 				      prot, reg_id.mirror);
2047 err_malloc:
2048 	kfree(ib_steering);
2049 
2050 	return err;
2051 }
2052 
2053 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2054 {
2055 	struct mlx4_ib_gid_entry *ge;
2056 	struct mlx4_ib_gid_entry *tmp;
2057 	struct mlx4_ib_gid_entry *ret = NULL;
2058 
2059 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2060 		if (!memcmp(raw, ge->gid.raw, 16)) {
2061 			ret = ge;
2062 			break;
2063 		}
2064 	}
2065 
2066 	return ret;
2067 }
2068 
2069 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2070 {
2071 	int err;
2072 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2073 	struct mlx4_dev *dev = mdev->dev;
2074 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2075 	struct net_device *ndev;
2076 	struct mlx4_ib_gid_entry *ge;
2077 	struct mlx4_flow_reg_id reg_id = {0, 0};
2078 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2079 
2080 	if (mdev->dev->caps.steering_mode ==
2081 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2082 		struct mlx4_ib_steering *ib_steering;
2083 
2084 		mutex_lock(&mqp->mutex);
2085 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2086 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2087 				list_del(&ib_steering->list);
2088 				break;
2089 			}
2090 		}
2091 		mutex_unlock(&mqp->mutex);
2092 		if (&ib_steering->list == &mqp->steering_rules) {
2093 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2094 			return -EINVAL;
2095 		}
2096 		reg_id = ib_steering->reg_id;
2097 		kfree(ib_steering);
2098 	}
2099 
2100 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2101 				    prot, reg_id.id);
2102 	if (err)
2103 		return err;
2104 
2105 	if (mlx4_is_bonded(dev)) {
2106 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2107 					    prot, reg_id.mirror);
2108 		if (err)
2109 			return err;
2110 	}
2111 
2112 	mutex_lock(&mqp->mutex);
2113 	ge = find_gid_entry(mqp, gid->raw);
2114 	if (ge) {
2115 		spin_lock_bh(&mdev->iboe.lock);
2116 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2117 		if (ndev)
2118 			dev_hold(ndev);
2119 		spin_unlock_bh(&mdev->iboe.lock);
2120 		if (ndev)
2121 			dev_put(ndev);
2122 		list_del(&ge->list);
2123 		kfree(ge);
2124 	} else
2125 		pr_warn("could not find mgid entry\n");
2126 
2127 	mutex_unlock(&mqp->mutex);
2128 
2129 	return 0;
2130 }
2131 
2132 static int init_node_data(struct mlx4_ib_dev *dev)
2133 {
2134 	struct ib_smp *in_mad  = NULL;
2135 	struct ib_smp *out_mad = NULL;
2136 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2137 	int err = -ENOMEM;
2138 
2139 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2140 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2141 	if (!in_mad || !out_mad)
2142 		goto out;
2143 
2144 	init_query_mad(in_mad);
2145 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2146 	if (mlx4_is_master(dev->dev))
2147 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2148 
2149 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2150 	if (err)
2151 		goto out;
2152 
2153 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2154 
2155 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2156 
2157 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2158 	if (err)
2159 		goto out;
2160 
2161 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2162 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2163 
2164 out:
2165 	kfree(in_mad);
2166 	kfree(out_mad);
2167 	return err;
2168 }
2169 
2170 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2171 			char *buf)
2172 {
2173 	struct mlx4_ib_dev *dev =
2174 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2175 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2176 }
2177 
2178 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2179 			char *buf)
2180 {
2181 	struct mlx4_ib_dev *dev =
2182 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2183 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2184 }
2185 
2186 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2187 			  char *buf)
2188 {
2189 	struct mlx4_ib_dev *dev =
2190 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2191 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2192 		       dev->dev->board_id);
2193 }
2194 
2195 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2196 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2197 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2198 
2199 static struct device_attribute *mlx4_class_attributes[] = {
2200 	&dev_attr_hw_rev,
2201 	&dev_attr_hca_type,
2202 	&dev_attr_board_id
2203 };
2204 
2205 struct diag_counter {
2206 	const char *name;
2207 	u32 offset;
2208 };
2209 
2210 #define DIAG_COUNTER(_name, _offset)			\
2211 	{ .name = #_name, .offset = _offset }
2212 
2213 static const struct diag_counter diag_basic[] = {
2214 	DIAG_COUNTER(rq_num_lle, 0x00),
2215 	DIAG_COUNTER(sq_num_lle, 0x04),
2216 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2217 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2218 	DIAG_COUNTER(rq_num_lpe, 0x18),
2219 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2220 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2221 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2222 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2223 	DIAG_COUNTER(sq_num_bre, 0x34),
2224 	DIAG_COUNTER(sq_num_rire, 0x44),
2225 	DIAG_COUNTER(rq_num_rire, 0x48),
2226 	DIAG_COUNTER(sq_num_rae, 0x4C),
2227 	DIAG_COUNTER(rq_num_rae, 0x50),
2228 	DIAG_COUNTER(sq_num_roe, 0x54),
2229 	DIAG_COUNTER(sq_num_tree, 0x5C),
2230 	DIAG_COUNTER(sq_num_rree, 0x64),
2231 	DIAG_COUNTER(rq_num_rnr, 0x68),
2232 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2233 	DIAG_COUNTER(rq_num_oos, 0x100),
2234 	DIAG_COUNTER(sq_num_oos, 0x104),
2235 };
2236 
2237 static const struct diag_counter diag_ext[] = {
2238 	DIAG_COUNTER(rq_num_dup, 0x130),
2239 	DIAG_COUNTER(sq_num_to, 0x134),
2240 };
2241 
2242 static const struct diag_counter diag_device_only[] = {
2243 	DIAG_COUNTER(num_cqovf, 0x1A0),
2244 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2245 };
2246 
2247 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2248 						    u8 port_num)
2249 {
2250 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2251 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2252 
2253 	if (!diag[!!port_num].name)
2254 		return NULL;
2255 
2256 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2257 					  diag[!!port_num].num_counters,
2258 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2259 }
2260 
2261 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2262 				struct rdma_hw_stats *stats,
2263 				u8 port, int index)
2264 {
2265 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2266 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2267 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2268 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2269 	int ret;
2270 	int i;
2271 
2272 	ret = mlx4_query_diag_counters(dev->dev,
2273 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2274 				       diag[!!port].offset, hw_value,
2275 				       diag[!!port].num_counters, port);
2276 
2277 	if (ret)
2278 		return ret;
2279 
2280 	for (i = 0; i < diag[!!port].num_counters; i++)
2281 		stats->value[i] = hw_value[i];
2282 
2283 	return diag[!!port].num_counters;
2284 }
2285 
2286 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2287 					 const char ***name,
2288 					 u32 **offset,
2289 					 u32 *num,
2290 					 bool port)
2291 {
2292 	u32 num_counters;
2293 
2294 	num_counters = ARRAY_SIZE(diag_basic);
2295 
2296 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2297 		num_counters += ARRAY_SIZE(diag_ext);
2298 
2299 	if (!port)
2300 		num_counters += ARRAY_SIZE(diag_device_only);
2301 
2302 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2303 	if (!*name)
2304 		return -ENOMEM;
2305 
2306 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2307 	if (!*offset)
2308 		goto err_name;
2309 
2310 	*num = num_counters;
2311 
2312 	return 0;
2313 
2314 err_name:
2315 	kfree(*name);
2316 	return -ENOMEM;
2317 }
2318 
2319 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2320 				       const char **name,
2321 				       u32 *offset,
2322 				       bool port)
2323 {
2324 	int i;
2325 	int j;
2326 
2327 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2328 		name[i] = diag_basic[i].name;
2329 		offset[i] = diag_basic[i].offset;
2330 	}
2331 
2332 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2333 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2334 			name[j] = diag_ext[i].name;
2335 			offset[j] = diag_ext[i].offset;
2336 		}
2337 	}
2338 
2339 	if (!port) {
2340 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2341 			name[j] = diag_device_only[i].name;
2342 			offset[j] = diag_device_only[i].offset;
2343 		}
2344 	}
2345 }
2346 
2347 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2348 {
2349 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2350 	int i;
2351 	int ret;
2352 	bool per_port = !!(ibdev->dev->caps.flags2 &
2353 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2354 
2355 	if (mlx4_is_slave(ibdev->dev))
2356 		return 0;
2357 
2358 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2359 		/* i == 1 means we are building port counters */
2360 		if (i && !per_port)
2361 			continue;
2362 
2363 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2364 						    &diag[i].offset,
2365 						    &diag[i].num_counters, i);
2366 		if (ret)
2367 			goto err_alloc;
2368 
2369 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2370 					   diag[i].offset, i);
2371 	}
2372 
2373 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2374 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2375 
2376 	return 0;
2377 
2378 err_alloc:
2379 	if (i) {
2380 		kfree(diag[i - 1].name);
2381 		kfree(diag[i - 1].offset);
2382 	}
2383 
2384 	return ret;
2385 }
2386 
2387 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2388 {
2389 	int i;
2390 
2391 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2392 		kfree(ibdev->diag_counters[i].offset);
2393 		kfree(ibdev->diag_counters[i].name);
2394 	}
2395 }
2396 
2397 #define MLX4_IB_INVALID_MAC	((u64)-1)
2398 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2399 			       struct net_device *dev,
2400 			       int port)
2401 {
2402 	u64 new_smac = 0;
2403 	u64 release_mac = MLX4_IB_INVALID_MAC;
2404 	struct mlx4_ib_qp *qp;
2405 
2406 	read_lock(&dev_base_lock);
2407 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2408 	read_unlock(&dev_base_lock);
2409 
2410 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2411 
2412 	/* no need for update QP1 and mac registration in non-SRIOV */
2413 	if (!mlx4_is_mfunc(ibdev->dev))
2414 		return;
2415 
2416 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2417 	qp = ibdev->qp1_proxy[port - 1];
2418 	if (qp) {
2419 		int new_smac_index;
2420 		u64 old_smac;
2421 		struct mlx4_update_qp_params update_params;
2422 
2423 		mutex_lock(&qp->mutex);
2424 		old_smac = qp->pri.smac;
2425 		if (new_smac == old_smac)
2426 			goto unlock;
2427 
2428 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2429 
2430 		if (new_smac_index < 0)
2431 			goto unlock;
2432 
2433 		update_params.smac_index = new_smac_index;
2434 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2435 				   &update_params)) {
2436 			release_mac = new_smac;
2437 			goto unlock;
2438 		}
2439 		/* if old port was zero, no mac was yet registered for this QP */
2440 		if (qp->pri.smac_port)
2441 			release_mac = old_smac;
2442 		qp->pri.smac = new_smac;
2443 		qp->pri.smac_port = port;
2444 		qp->pri.smac_index = new_smac_index;
2445 	}
2446 
2447 unlock:
2448 	if (release_mac != MLX4_IB_INVALID_MAC)
2449 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2450 	if (qp)
2451 		mutex_unlock(&qp->mutex);
2452 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2453 }
2454 
2455 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2456 				 struct net_device *dev,
2457 				 unsigned long event)
2458 
2459 {
2460 	struct mlx4_ib_iboe *iboe;
2461 	int update_qps_port = -1;
2462 	int port;
2463 
2464 	ASSERT_RTNL();
2465 
2466 	iboe = &ibdev->iboe;
2467 
2468 	spin_lock_bh(&iboe->lock);
2469 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2470 
2471 		iboe->netdevs[port - 1] =
2472 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2473 
2474 		if (dev == iboe->netdevs[port - 1] &&
2475 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2476 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2477 			update_qps_port = port;
2478 
2479 	}
2480 	spin_unlock_bh(&iboe->lock);
2481 
2482 	if (update_qps_port > 0)
2483 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2484 }
2485 
2486 static int mlx4_ib_netdev_event(struct notifier_block *this,
2487 				unsigned long event, void *ptr)
2488 {
2489 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2490 	struct mlx4_ib_dev *ibdev;
2491 
2492 	if (!net_eq(dev_net(dev), &init_net))
2493 		return NOTIFY_DONE;
2494 
2495 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2496 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2497 
2498 	return NOTIFY_DONE;
2499 }
2500 
2501 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2502 {
2503 	int port;
2504 	int slave;
2505 	int i;
2506 
2507 	if (mlx4_is_master(ibdev->dev)) {
2508 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2509 		     ++slave) {
2510 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2511 				for (i = 0;
2512 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2513 				     ++i) {
2514 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2515 					/* master has the identity virt2phys pkey mapping */
2516 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2517 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2518 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2519 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2520 				}
2521 			}
2522 		}
2523 		/* initialize pkey cache */
2524 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2525 			for (i = 0;
2526 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2527 			     ++i)
2528 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2529 					(i) ? 0 : 0xFFFF;
2530 		}
2531 	}
2532 }
2533 
2534 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2535 {
2536 	int i, j, eq = 0, total_eqs = 0;
2537 
2538 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2539 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2540 	if (!ibdev->eq_table)
2541 		return;
2542 
2543 	for (i = 1; i <= dev->caps.num_ports; i++) {
2544 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2545 		     j++, total_eqs++) {
2546 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2547 				continue;
2548 			ibdev->eq_table[eq] = total_eqs;
2549 			if (!mlx4_assign_eq(dev, i,
2550 					    &ibdev->eq_table[eq]))
2551 				eq++;
2552 			else
2553 				ibdev->eq_table[eq] = -1;
2554 		}
2555 	}
2556 
2557 	for (i = eq; i < dev->caps.num_comp_vectors;
2558 	     ibdev->eq_table[i++] = -1)
2559 		;
2560 
2561 	/* Advertise the new number of EQs to clients */
2562 	ibdev->ib_dev.num_comp_vectors = eq;
2563 }
2564 
2565 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2566 {
2567 	int i;
2568 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2569 
2570 	/* no eqs were allocated */
2571 	if (!ibdev->eq_table)
2572 		return;
2573 
2574 	/* Reset the advertised EQ number */
2575 	ibdev->ib_dev.num_comp_vectors = 0;
2576 
2577 	for (i = 0; i < total_eqs; i++)
2578 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2579 
2580 	kfree(ibdev->eq_table);
2581 	ibdev->eq_table = NULL;
2582 }
2583 
2584 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2585 			       struct ib_port_immutable *immutable)
2586 {
2587 	struct ib_port_attr attr;
2588 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2589 	int err;
2590 
2591 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2592 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2593 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2594 	} else {
2595 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2596 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2597 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2598 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2599 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2600 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2601 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2602 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2603 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2604 	}
2605 
2606 	err = ib_query_port(ibdev, port_num, &attr);
2607 	if (err)
2608 		return err;
2609 
2610 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2611 	immutable->gid_tbl_len = attr.gid_tbl_len;
2612 
2613 	return 0;
2614 }
2615 
2616 static void get_fw_ver_str(struct ib_device *device, char *str)
2617 {
2618 	struct mlx4_ib_dev *dev =
2619 		container_of(device, struct mlx4_ib_dev, ib_dev);
2620 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2621 		 (int) (dev->dev->caps.fw_ver >> 32),
2622 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2623 		 (int) dev->dev->caps.fw_ver & 0xffff);
2624 }
2625 
2626 static void *mlx4_ib_add(struct mlx4_dev *dev)
2627 {
2628 	struct mlx4_ib_dev *ibdev;
2629 	int num_ports = 0;
2630 	int i, j;
2631 	int err;
2632 	struct mlx4_ib_iboe *iboe;
2633 	int ib_num_ports = 0;
2634 	int num_req_counters;
2635 	int allocated;
2636 	u32 counter_index;
2637 	struct counter_index *new_counter_index = NULL;
2638 
2639 	pr_info_once("%s", mlx4_ib_version);
2640 
2641 	num_ports = 0;
2642 	mlx4_foreach_ib_transport_port(i, dev)
2643 		num_ports++;
2644 
2645 	/* No point in registering a device with no ports... */
2646 	if (num_ports == 0)
2647 		return NULL;
2648 
2649 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2650 	if (!ibdev) {
2651 		dev_err(&dev->persist->pdev->dev,
2652 			"Device struct alloc failed\n");
2653 		return NULL;
2654 	}
2655 
2656 	iboe = &ibdev->iboe;
2657 
2658 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2659 		goto err_dealloc;
2660 
2661 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2662 		goto err_pd;
2663 
2664 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2665 				 PAGE_SIZE);
2666 	if (!ibdev->uar_map)
2667 		goto err_uar;
2668 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2669 
2670 	ibdev->dev = dev;
2671 	ibdev->bond_next_port	= 0;
2672 
2673 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2674 	ibdev->ib_dev.owner		= THIS_MODULE;
2675 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2676 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2677 	ibdev->num_ports		= num_ports;
2678 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2679 						1 : ibdev->num_ports;
2680 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2681 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2682 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2683 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2684 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2685 
2686 	if (dev->caps.userspace_caps)
2687 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2688 	else
2689 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2690 
2691 	ibdev->ib_dev.uverbs_cmd_mask	=
2692 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2693 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2694 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2695 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2696 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2697 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2698 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2699 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2700 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2701 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2702 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2703 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2704 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2705 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2706 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2707 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2708 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2709 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2710 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2711 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2712 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2713 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2714 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2715 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2716 
2717 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2718 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2719 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2720 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2721 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2722 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2723 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2724 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2725 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2726 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2727 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2728 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2729 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2730 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2731 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2732 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2733 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2734 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2735 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2736 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2737 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2738 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2739 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2740 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2741 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2742 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2743 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2744 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2745 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2746 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2747 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2748 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2749 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2750 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2751 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2752 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2753 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2754 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2755 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2756 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2757 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2758 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2759 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2760 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2761 
2762 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2763 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2764 
2765 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2766 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2767 	    IB_LINK_LAYER_ETHERNET) ||
2768 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2769 	    IB_LINK_LAYER_ETHERNET))) {
2770 		ibdev->ib_dev.create_wq		= mlx4_ib_create_wq;
2771 		ibdev->ib_dev.modify_wq		= mlx4_ib_modify_wq;
2772 		ibdev->ib_dev.destroy_wq	= mlx4_ib_destroy_wq;
2773 		ibdev->ib_dev.create_rwq_ind_table  =
2774 			mlx4_ib_create_rwq_ind_table;
2775 		ibdev->ib_dev.destroy_rwq_ind_table =
2776 			mlx4_ib_destroy_rwq_ind_table;
2777 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2778 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2779 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2780 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2781 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2782 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2783 	}
2784 
2785 	if (!mlx4_is_slave(ibdev->dev)) {
2786 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2787 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2788 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2789 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2790 	}
2791 
2792 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2793 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2794 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2795 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2796 
2797 		ibdev->ib_dev.uverbs_cmd_mask |=
2798 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2799 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2800 	}
2801 
2802 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2803 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2804 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2805 		ibdev->ib_dev.uverbs_cmd_mask |=
2806 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2807 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2808 	}
2809 
2810 	if (check_flow_steering_support(dev)) {
2811 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2812 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2813 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2814 
2815 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2816 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2817 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2818 	}
2819 
2820 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2821 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2822 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2823 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2824 
2825 	mlx4_ib_alloc_eqs(dev, ibdev);
2826 
2827 	spin_lock_init(&iboe->lock);
2828 
2829 	if (init_node_data(ibdev))
2830 		goto err_map;
2831 	mlx4_init_sl2vl_tbl(ibdev);
2832 
2833 	for (i = 0; i < ibdev->num_ports; ++i) {
2834 		mutex_init(&ibdev->counters_table[i].mutex);
2835 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2836 	}
2837 
2838 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2839 	for (i = 0; i < num_req_counters; ++i) {
2840 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2841 		allocated = 0;
2842 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2843 						IB_LINK_LAYER_ETHERNET) {
2844 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2845 						 MLX4_RES_USAGE_DRIVER);
2846 			/* if failed to allocate a new counter, use default */
2847 			if (err)
2848 				counter_index =
2849 					mlx4_get_default_counter_index(dev,
2850 								       i + 1);
2851 			else
2852 				allocated = 1;
2853 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2854 			counter_index = mlx4_get_default_counter_index(dev,
2855 								       i + 1);
2856 		}
2857 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2858 					    GFP_KERNEL);
2859 		if (!new_counter_index) {
2860 			if (allocated)
2861 				mlx4_counter_free(ibdev->dev, counter_index);
2862 			goto err_counter;
2863 		}
2864 		new_counter_index->index = counter_index;
2865 		new_counter_index->allocated = allocated;
2866 		list_add_tail(&new_counter_index->list,
2867 			      &ibdev->counters_table[i].counters_list);
2868 		ibdev->counters_table[i].default_counter = counter_index;
2869 		pr_info("counter index %d for port %d allocated %d\n",
2870 			counter_index, i + 1, allocated);
2871 	}
2872 	if (mlx4_is_bonded(dev))
2873 		for (i = 1; i < ibdev->num_ports ; ++i) {
2874 			new_counter_index =
2875 					kmalloc(sizeof(struct counter_index),
2876 						GFP_KERNEL);
2877 			if (!new_counter_index)
2878 				goto err_counter;
2879 			new_counter_index->index = counter_index;
2880 			new_counter_index->allocated = 0;
2881 			list_add_tail(&new_counter_index->list,
2882 				      &ibdev->counters_table[i].counters_list);
2883 			ibdev->counters_table[i].default_counter =
2884 								counter_index;
2885 		}
2886 
2887 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2888 		ib_num_ports++;
2889 
2890 	spin_lock_init(&ibdev->sm_lock);
2891 	mutex_init(&ibdev->cap_mask_mutex);
2892 	INIT_LIST_HEAD(&ibdev->qp_list);
2893 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2894 
2895 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2896 	    ib_num_ports) {
2897 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2898 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2899 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2900 					    &ibdev->steer_qpn_base, 0,
2901 					    MLX4_RES_USAGE_DRIVER);
2902 		if (err)
2903 			goto err_counter;
2904 
2905 		ibdev->ib_uc_qpns_bitmap =
2906 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2907 				sizeof(long),
2908 				GFP_KERNEL);
2909 		if (!ibdev->ib_uc_qpns_bitmap)
2910 			goto err_steer_qp_release;
2911 
2912 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2913 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2914 				    ibdev->steer_qpn_count);
2915 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2916 					dev, ibdev->steer_qpn_base,
2917 					ibdev->steer_qpn_base +
2918 					ibdev->steer_qpn_count - 1);
2919 			if (err)
2920 				goto err_steer_free_bitmap;
2921 		} else {
2922 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2923 				    ibdev->steer_qpn_count);
2924 		}
2925 	}
2926 
2927 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2928 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2929 
2930 	if (mlx4_ib_alloc_diag_counters(ibdev))
2931 		goto err_steer_free_bitmap;
2932 
2933 	if (ib_register_device(&ibdev->ib_dev, NULL))
2934 		goto err_diag_counters;
2935 
2936 	if (mlx4_ib_mad_init(ibdev))
2937 		goto err_reg;
2938 
2939 	if (mlx4_ib_init_sriov(ibdev))
2940 		goto err_mad;
2941 
2942 	if (!iboe->nb.notifier_call) {
2943 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2944 		err = register_netdevice_notifier(&iboe->nb);
2945 		if (err) {
2946 			iboe->nb.notifier_call = NULL;
2947 			goto err_notif;
2948 		}
2949 	}
2950 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2951 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2952 		if (err)
2953 			goto err_notif;
2954 	}
2955 
2956 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2957 		if (device_create_file(&ibdev->ib_dev.dev,
2958 				       mlx4_class_attributes[j]))
2959 			goto err_notif;
2960 	}
2961 
2962 	ibdev->ib_active = true;
2963 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2964 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2965 					 &ibdev->ib_dev);
2966 
2967 	if (mlx4_is_mfunc(ibdev->dev))
2968 		init_pkeys(ibdev);
2969 
2970 	/* create paravirt contexts for any VFs which are active */
2971 	if (mlx4_is_master(ibdev->dev)) {
2972 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2973 			if (j == mlx4_master_func_num(ibdev->dev))
2974 				continue;
2975 			if (mlx4_is_slave_active(ibdev->dev, j))
2976 				do_slave_init(ibdev, j, 1);
2977 		}
2978 	}
2979 	return ibdev;
2980 
2981 err_notif:
2982 	if (ibdev->iboe.nb.notifier_call) {
2983 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2984 			pr_warn("failure unregistering notifier\n");
2985 		ibdev->iboe.nb.notifier_call = NULL;
2986 	}
2987 	flush_workqueue(wq);
2988 
2989 	mlx4_ib_close_sriov(ibdev);
2990 
2991 err_mad:
2992 	mlx4_ib_mad_cleanup(ibdev);
2993 
2994 err_reg:
2995 	ib_unregister_device(&ibdev->ib_dev);
2996 
2997 err_diag_counters:
2998 	mlx4_ib_diag_cleanup(ibdev);
2999 
3000 err_steer_free_bitmap:
3001 	kfree(ibdev->ib_uc_qpns_bitmap);
3002 
3003 err_steer_qp_release:
3004 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3005 			      ibdev->steer_qpn_count);
3006 err_counter:
3007 	for (i = 0; i < ibdev->num_ports; ++i)
3008 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
3009 
3010 err_map:
3011 	mlx4_ib_free_eqs(dev, ibdev);
3012 	iounmap(ibdev->uar_map);
3013 
3014 err_uar:
3015 	mlx4_uar_free(dev, &ibdev->priv_uar);
3016 
3017 err_pd:
3018 	mlx4_pd_free(dev, ibdev->priv_pdn);
3019 
3020 err_dealloc:
3021 	ib_dealloc_device(&ibdev->ib_dev);
3022 
3023 	return NULL;
3024 }
3025 
3026 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
3027 {
3028 	int offset;
3029 
3030 	WARN_ON(!dev->ib_uc_qpns_bitmap);
3031 
3032 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3033 					 dev->steer_qpn_count,
3034 					 get_count_order(count));
3035 	if (offset < 0)
3036 		return offset;
3037 
3038 	*qpn = dev->steer_qpn_base + offset;
3039 	return 0;
3040 }
3041 
3042 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3043 {
3044 	if (!qpn ||
3045 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3046 		return;
3047 
3048 	BUG_ON(qpn < dev->steer_qpn_base);
3049 
3050 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
3051 			      qpn - dev->steer_qpn_base,
3052 			      get_count_order(count));
3053 }
3054 
3055 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3056 			 int is_attach)
3057 {
3058 	int err;
3059 	size_t flow_size;
3060 	struct ib_flow_attr *flow = NULL;
3061 	struct ib_flow_spec_ib *ib_spec;
3062 
3063 	if (is_attach) {
3064 		flow_size = sizeof(struct ib_flow_attr) +
3065 			    sizeof(struct ib_flow_spec_ib);
3066 		flow = kzalloc(flow_size, GFP_KERNEL);
3067 		if (!flow)
3068 			return -ENOMEM;
3069 		flow->port = mqp->port;
3070 		flow->num_of_specs = 1;
3071 		flow->size = flow_size;
3072 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3073 		ib_spec->type = IB_FLOW_SPEC_IB;
3074 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3075 		/* Add an empty rule for IB L2 */
3076 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3077 
3078 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3079 					    IB_FLOW_DOMAIN_NIC,
3080 					    MLX4_FS_REGULAR,
3081 					    &mqp->reg_id);
3082 	} else {
3083 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3084 	}
3085 	kfree(flow);
3086 	return err;
3087 }
3088 
3089 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3090 {
3091 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3092 	int p;
3093 	int i;
3094 
3095 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3096 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3097 	ibdev->ib_active = false;
3098 	flush_workqueue(wq);
3099 
3100 	mlx4_ib_close_sriov(ibdev);
3101 	mlx4_ib_mad_cleanup(ibdev);
3102 	ib_unregister_device(&ibdev->ib_dev);
3103 	mlx4_ib_diag_cleanup(ibdev);
3104 	if (ibdev->iboe.nb.notifier_call) {
3105 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3106 			pr_warn("failure unregistering notifier\n");
3107 		ibdev->iboe.nb.notifier_call = NULL;
3108 	}
3109 
3110 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3111 			      ibdev->steer_qpn_count);
3112 	kfree(ibdev->ib_uc_qpns_bitmap);
3113 
3114 	iounmap(ibdev->uar_map);
3115 	for (p = 0; p < ibdev->num_ports; ++p)
3116 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3117 
3118 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3119 		mlx4_CLOSE_PORT(dev, p);
3120 
3121 	mlx4_ib_free_eqs(dev, ibdev);
3122 
3123 	mlx4_uar_free(dev, &ibdev->priv_uar);
3124 	mlx4_pd_free(dev, ibdev->priv_pdn);
3125 	ib_dealloc_device(&ibdev->ib_dev);
3126 }
3127 
3128 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3129 {
3130 	struct mlx4_ib_demux_work **dm = NULL;
3131 	struct mlx4_dev *dev = ibdev->dev;
3132 	int i;
3133 	unsigned long flags;
3134 	struct mlx4_active_ports actv_ports;
3135 	unsigned int ports;
3136 	unsigned int first_port;
3137 
3138 	if (!mlx4_is_master(dev))
3139 		return;
3140 
3141 	actv_ports = mlx4_get_active_ports(dev, slave);
3142 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3143 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3144 
3145 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3146 	if (!dm)
3147 		return;
3148 
3149 	for (i = 0; i < ports; i++) {
3150 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3151 		if (!dm[i]) {
3152 			while (--i >= 0)
3153 				kfree(dm[i]);
3154 			goto out;
3155 		}
3156 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3157 		dm[i]->port = first_port + i + 1;
3158 		dm[i]->slave = slave;
3159 		dm[i]->do_init = do_init;
3160 		dm[i]->dev = ibdev;
3161 	}
3162 	/* initialize or tear down tunnel QPs for the slave */
3163 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3164 	if (!ibdev->sriov.is_going_down) {
3165 		for (i = 0; i < ports; i++)
3166 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3167 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3168 	} else {
3169 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3170 		for (i = 0; i < ports; i++)
3171 			kfree(dm[i]);
3172 	}
3173 out:
3174 	kfree(dm);
3175 	return;
3176 }
3177 
3178 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3179 {
3180 	struct mlx4_ib_qp *mqp;
3181 	unsigned long flags_qp;
3182 	unsigned long flags_cq;
3183 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3184 	struct list_head    cq_notify_list;
3185 	struct mlx4_cq *mcq;
3186 	unsigned long flags;
3187 
3188 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3189 	INIT_LIST_HEAD(&cq_notify_list);
3190 
3191 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3192 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3193 
3194 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3195 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3196 		if (mqp->sq.tail != mqp->sq.head) {
3197 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3198 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3199 			if (send_mcq->mcq.comp &&
3200 			    mqp->ibqp.send_cq->comp_handler) {
3201 				if (!send_mcq->mcq.reset_notify_added) {
3202 					send_mcq->mcq.reset_notify_added = 1;
3203 					list_add_tail(&send_mcq->mcq.reset_notify,
3204 						      &cq_notify_list);
3205 				}
3206 			}
3207 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3208 		}
3209 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3210 		/* Now, handle the QP's receive queue */
3211 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3212 		/* no handling is needed for SRQ */
3213 		if (!mqp->ibqp.srq) {
3214 			if (mqp->rq.tail != mqp->rq.head) {
3215 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3216 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3217 				if (recv_mcq->mcq.comp &&
3218 				    mqp->ibqp.recv_cq->comp_handler) {
3219 					if (!recv_mcq->mcq.reset_notify_added) {
3220 						recv_mcq->mcq.reset_notify_added = 1;
3221 						list_add_tail(&recv_mcq->mcq.reset_notify,
3222 							      &cq_notify_list);
3223 					}
3224 				}
3225 				spin_unlock_irqrestore(&recv_mcq->lock,
3226 						       flags_cq);
3227 			}
3228 		}
3229 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3230 	}
3231 
3232 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3233 		mcq->comp(mcq);
3234 	}
3235 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3236 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3237 }
3238 
3239 static void handle_bonded_port_state_event(struct work_struct *work)
3240 {
3241 	struct ib_event_work *ew =
3242 		container_of(work, struct ib_event_work, work);
3243 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3244 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3245 	int i;
3246 	struct ib_event ibev;
3247 
3248 	kfree(ew);
3249 	spin_lock_bh(&ibdev->iboe.lock);
3250 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3251 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3252 		enum ib_port_state curr_port_state;
3253 
3254 		if (!curr_netdev)
3255 			continue;
3256 
3257 		curr_port_state =
3258 			(netif_running(curr_netdev) &&
3259 			 netif_carrier_ok(curr_netdev)) ?
3260 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3261 
3262 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3263 			curr_port_state : IB_PORT_ACTIVE;
3264 	}
3265 	spin_unlock_bh(&ibdev->iboe.lock);
3266 
3267 	ibev.device = &ibdev->ib_dev;
3268 	ibev.element.port_num = 1;
3269 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3270 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3271 
3272 	ib_dispatch_event(&ibev);
3273 }
3274 
3275 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3276 {
3277 	u64 sl2vl;
3278 	int err;
3279 
3280 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3281 	if (err) {
3282 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3283 		       port, err);
3284 		sl2vl = 0;
3285 	}
3286 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3287 }
3288 
3289 static void ib_sl2vl_update_work(struct work_struct *work)
3290 {
3291 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3292 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3293 	int port = ew->port;
3294 
3295 	mlx4_ib_sl2vl_update(mdev, port);
3296 
3297 	kfree(ew);
3298 }
3299 
3300 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3301 				     int port)
3302 {
3303 	struct ib_event_work *ew;
3304 
3305 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3306 	if (ew) {
3307 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3308 		ew->port = port;
3309 		ew->ib_dev = ibdev;
3310 		queue_work(wq, &ew->work);
3311 	}
3312 }
3313 
3314 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3315 			  enum mlx4_dev_event event, unsigned long param)
3316 {
3317 	struct ib_event ibev;
3318 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3319 	struct mlx4_eqe *eqe = NULL;
3320 	struct ib_event_work *ew;
3321 	int p = 0;
3322 
3323 	if (mlx4_is_bonded(dev) &&
3324 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3325 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3326 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3327 		if (!ew)
3328 			return;
3329 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3330 		ew->ib_dev = ibdev;
3331 		queue_work(wq, &ew->work);
3332 		return;
3333 	}
3334 
3335 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3336 		eqe = (struct mlx4_eqe *)param;
3337 	else
3338 		p = (int) param;
3339 
3340 	switch (event) {
3341 	case MLX4_DEV_EVENT_PORT_UP:
3342 		if (p > ibdev->num_ports)
3343 			return;
3344 		if (!mlx4_is_slave(dev) &&
3345 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3346 			IB_LINK_LAYER_INFINIBAND) {
3347 			if (mlx4_is_master(dev))
3348 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3349 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3350 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3351 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3352 		}
3353 		ibev.event = IB_EVENT_PORT_ACTIVE;
3354 		break;
3355 
3356 	case MLX4_DEV_EVENT_PORT_DOWN:
3357 		if (p > ibdev->num_ports)
3358 			return;
3359 		ibev.event = IB_EVENT_PORT_ERR;
3360 		break;
3361 
3362 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3363 		ibdev->ib_active = false;
3364 		ibev.event = IB_EVENT_DEVICE_FATAL;
3365 		mlx4_ib_handle_catas_error(ibdev);
3366 		break;
3367 
3368 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3369 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3370 		if (!ew)
3371 			break;
3372 
3373 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3374 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3375 		ew->ib_dev = ibdev;
3376 		/* need to queue only for port owner, which uses GEN_EQE */
3377 		if (mlx4_is_master(dev))
3378 			queue_work(wq, &ew->work);
3379 		else
3380 			handle_port_mgmt_change_event(&ew->work);
3381 		return;
3382 
3383 	case MLX4_DEV_EVENT_SLAVE_INIT:
3384 		/* here, p is the slave id */
3385 		do_slave_init(ibdev, p, 1);
3386 		if (mlx4_is_master(dev)) {
3387 			int i;
3388 
3389 			for (i = 1; i <= ibdev->num_ports; i++) {
3390 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3391 					== IB_LINK_LAYER_INFINIBAND)
3392 					mlx4_ib_slave_alias_guid_event(ibdev,
3393 								       p, i,
3394 								       1);
3395 			}
3396 		}
3397 		return;
3398 
3399 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3400 		if (mlx4_is_master(dev)) {
3401 			int i;
3402 
3403 			for (i = 1; i <= ibdev->num_ports; i++) {
3404 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3405 					== IB_LINK_LAYER_INFINIBAND)
3406 					mlx4_ib_slave_alias_guid_event(ibdev,
3407 								       p, i,
3408 								       0);
3409 			}
3410 		}
3411 		/* here, p is the slave id */
3412 		do_slave_init(ibdev, p, 0);
3413 		return;
3414 
3415 	default:
3416 		return;
3417 	}
3418 
3419 	ibev.device	      = ibdev_ptr;
3420 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3421 
3422 	ib_dispatch_event(&ibev);
3423 }
3424 
3425 static struct mlx4_interface mlx4_ib_interface = {
3426 	.add		= mlx4_ib_add,
3427 	.remove		= mlx4_ib_remove,
3428 	.event		= mlx4_ib_event,
3429 	.protocol	= MLX4_PROT_IB_IPV6,
3430 	.flags		= MLX4_INTFF_BONDING
3431 };
3432 
3433 static int __init mlx4_ib_init(void)
3434 {
3435 	int err;
3436 
3437 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3438 	if (!wq)
3439 		return -ENOMEM;
3440 
3441 	err = mlx4_ib_mcg_init();
3442 	if (err)
3443 		goto clean_wq;
3444 
3445 	err = mlx4_register_interface(&mlx4_ib_interface);
3446 	if (err)
3447 		goto clean_mcg;
3448 
3449 	return 0;
3450 
3451 clean_mcg:
3452 	mlx4_ib_mcg_destroy();
3453 
3454 clean_wq:
3455 	destroy_workqueue(wq);
3456 	return err;
3457 }
3458 
3459 static void __exit mlx4_ib_cleanup(void)
3460 {
3461 	mlx4_unregister_interface(&mlx4_ib_interface);
3462 	mlx4_ib_mcg_destroy();
3463 	destroy_workqueue(wq);
3464 }
3465 
3466 module_init(mlx4_ib_init);
3467 module_exit(mlx4_ib_cleanup);
3468