xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision a8fe58cec351c25e09c393bf46117c0c47b5a17c)
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 <net/ipv6.h>
43 #include <net/addrconf.h>
44 
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib_cache.h>
49 
50 #include <net/bonding.h>
51 
52 #include <linux/mlx4/driver.h>
53 #include <linux/mlx4/cmd.h>
54 #include <linux/mlx4/qp.h>
55 
56 #include "mlx4_ib.h"
57 #include "user.h"
58 
59 #define DRV_NAME	MLX4_IB_DRV_NAME
60 #define DRV_VERSION	"2.2-1"
61 #define DRV_RELDATE	"Feb 2014"
62 
63 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
64 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
65 #define MLX4_IB_CARD_REV_A0   0xA0
66 
67 MODULE_AUTHOR("Roland Dreier");
68 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
69 MODULE_LICENSE("Dual BSD/GPL");
70 MODULE_VERSION(DRV_VERSION);
71 
72 int mlx4_ib_sm_guid_assign = 0;
73 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
74 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
75 
76 static const char mlx4_ib_version[] =
77 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
78 	DRV_VERSION " (" DRV_RELDATE ")\n";
79 
80 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
81 
82 static struct workqueue_struct *wq;
83 
84 static void init_query_mad(struct ib_smp *mad)
85 {
86 	mad->base_version  = 1;
87 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
88 	mad->class_version = 1;
89 	mad->method	   = IB_MGMT_METHOD_GET;
90 }
91 
92 static int check_flow_steering_support(struct mlx4_dev *dev)
93 {
94 	int eth_num_ports = 0;
95 	int ib_num_ports = 0;
96 
97 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
98 
99 	if (dmfs) {
100 		int i;
101 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
102 			eth_num_ports++;
103 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
104 			ib_num_ports++;
105 		dmfs &= (!ib_num_ports ||
106 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
107 			(!eth_num_ports ||
108 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
109 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
110 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
111 			dmfs = 0;
112 		}
113 	}
114 	return dmfs;
115 }
116 
117 static int num_ib_ports(struct mlx4_dev *dev)
118 {
119 	int ib_ports = 0;
120 	int i;
121 
122 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
123 		ib_ports++;
124 
125 	return ib_ports;
126 }
127 
128 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
129 {
130 	struct mlx4_ib_dev *ibdev = to_mdev(device);
131 	struct net_device *dev;
132 
133 	rcu_read_lock();
134 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
135 
136 	if (dev) {
137 		if (mlx4_is_bonded(ibdev->dev)) {
138 			struct net_device *upper = NULL;
139 
140 			upper = netdev_master_upper_dev_get_rcu(dev);
141 			if (upper) {
142 				struct net_device *active;
143 
144 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
145 				if (active)
146 					dev = active;
147 			}
148 		}
149 	}
150 	if (dev)
151 		dev_hold(dev);
152 
153 	rcu_read_unlock();
154 	return dev;
155 }
156 
157 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
158 				  struct mlx4_ib_dev *ibdev,
159 				  u8 port_num)
160 {
161 	struct mlx4_cmd_mailbox *mailbox;
162 	int err;
163 	struct mlx4_dev *dev = ibdev->dev;
164 	int i;
165 	union ib_gid *gid_tbl;
166 
167 	mailbox = mlx4_alloc_cmd_mailbox(dev);
168 	if (IS_ERR(mailbox))
169 		return -ENOMEM;
170 
171 	gid_tbl = mailbox->buf;
172 
173 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
174 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
175 
176 	err = mlx4_cmd(dev, mailbox->dma,
177 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
178 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
179 		       MLX4_CMD_WRAPPED);
180 	if (mlx4_is_bonded(dev))
181 		err += mlx4_cmd(dev, mailbox->dma,
182 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
183 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
184 				MLX4_CMD_WRAPPED);
185 
186 	mlx4_free_cmd_mailbox(dev, mailbox);
187 	return err;
188 }
189 
190 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
191 				     struct mlx4_ib_dev *ibdev,
192 				     u8 port_num)
193 {
194 	struct mlx4_cmd_mailbox *mailbox;
195 	int err;
196 	struct mlx4_dev *dev = ibdev->dev;
197 	int i;
198 	struct {
199 		union ib_gid	gid;
200 		__be32		rsrvd1[2];
201 		__be16		rsrvd2;
202 		u8		type;
203 		u8		version;
204 		__be32		rsrvd3;
205 	} *gid_tbl;
206 
207 	mailbox = mlx4_alloc_cmd_mailbox(dev);
208 	if (IS_ERR(mailbox))
209 		return -ENOMEM;
210 
211 	gid_tbl = mailbox->buf;
212 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
213 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
214 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
215 			gid_tbl[i].version = 2;
216 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
217 				gid_tbl[i].type = 1;
218 			else
219 				memset(&gid_tbl[i].gid, 0, 12);
220 		}
221 	}
222 
223 	err = mlx4_cmd(dev, mailbox->dma,
224 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
225 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
226 		       MLX4_CMD_WRAPPED);
227 	if (mlx4_is_bonded(dev))
228 		err += mlx4_cmd(dev, mailbox->dma,
229 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
230 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
231 				MLX4_CMD_WRAPPED);
232 
233 	mlx4_free_cmd_mailbox(dev, mailbox);
234 	return err;
235 }
236 
237 static int mlx4_ib_update_gids(struct gid_entry *gids,
238 			       struct mlx4_ib_dev *ibdev,
239 			       u8 port_num)
240 {
241 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
242 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
243 
244 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
245 }
246 
247 static int mlx4_ib_add_gid(struct ib_device *device,
248 			   u8 port_num,
249 			   unsigned int index,
250 			   const union ib_gid *gid,
251 			   const struct ib_gid_attr *attr,
252 			   void **context)
253 {
254 	struct mlx4_ib_dev *ibdev = to_mdev(device);
255 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
256 	struct mlx4_port_gid_table   *port_gid_table;
257 	int free = -1, found = -1;
258 	int ret = 0;
259 	int hw_update = 0;
260 	int i;
261 	struct gid_entry *gids = NULL;
262 
263 	if (!rdma_cap_roce_gid_table(device, port_num))
264 		return -EINVAL;
265 
266 	if (port_num > MLX4_MAX_PORTS)
267 		return -EINVAL;
268 
269 	if (!context)
270 		return -EINVAL;
271 
272 	port_gid_table = &iboe->gids[port_num - 1];
273 	spin_lock_bh(&iboe->lock);
274 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
275 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
276 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
277 			found = i;
278 			break;
279 		}
280 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
281 			free = i; /* HW has space */
282 	}
283 
284 	if (found < 0) {
285 		if (free < 0) {
286 			ret = -ENOSPC;
287 		} else {
288 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
289 			if (!port_gid_table->gids[free].ctx) {
290 				ret = -ENOMEM;
291 			} else {
292 				*context = port_gid_table->gids[free].ctx;
293 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
294 				port_gid_table->gids[free].gid_type = attr->gid_type;
295 				port_gid_table->gids[free].ctx->real_index = free;
296 				port_gid_table->gids[free].ctx->refcount = 1;
297 				hw_update = 1;
298 			}
299 		}
300 	} else {
301 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
302 		*context = ctx;
303 		ctx->refcount++;
304 	}
305 	if (!ret && hw_update) {
306 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
307 		if (!gids) {
308 			ret = -ENOMEM;
309 		} else {
310 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
311 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
312 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
313 			}
314 		}
315 	}
316 	spin_unlock_bh(&iboe->lock);
317 
318 	if (!ret && hw_update) {
319 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
320 		kfree(gids);
321 	}
322 
323 	return ret;
324 }
325 
326 static int mlx4_ib_del_gid(struct ib_device *device,
327 			   u8 port_num,
328 			   unsigned int index,
329 			   void **context)
330 {
331 	struct gid_cache_context *ctx = *context;
332 	struct mlx4_ib_dev *ibdev = to_mdev(device);
333 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
334 	struct mlx4_port_gid_table   *port_gid_table;
335 	int ret = 0;
336 	int hw_update = 0;
337 	struct gid_entry *gids = NULL;
338 
339 	if (!rdma_cap_roce_gid_table(device, port_num))
340 		return -EINVAL;
341 
342 	if (port_num > MLX4_MAX_PORTS)
343 		return -EINVAL;
344 
345 	port_gid_table = &iboe->gids[port_num - 1];
346 	spin_lock_bh(&iboe->lock);
347 	if (ctx) {
348 		ctx->refcount--;
349 		if (!ctx->refcount) {
350 			unsigned int real_index = ctx->real_index;
351 
352 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
353 			kfree(port_gid_table->gids[real_index].ctx);
354 			port_gid_table->gids[real_index].ctx = NULL;
355 			hw_update = 1;
356 		}
357 	}
358 	if (!ret && hw_update) {
359 		int i;
360 
361 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
362 		if (!gids) {
363 			ret = -ENOMEM;
364 		} else {
365 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
366 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
367 		}
368 	}
369 	spin_unlock_bh(&iboe->lock);
370 
371 	if (!ret && hw_update) {
372 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
373 		kfree(gids);
374 	}
375 	return ret;
376 }
377 
378 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
379 				    u8 port_num, int index)
380 {
381 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
382 	struct gid_cache_context *ctx = NULL;
383 	union ib_gid gid;
384 	struct mlx4_port_gid_table   *port_gid_table;
385 	int real_index = -EINVAL;
386 	int i;
387 	int ret;
388 	unsigned long flags;
389 	struct ib_gid_attr attr;
390 
391 	if (port_num > MLX4_MAX_PORTS)
392 		return -EINVAL;
393 
394 	if (mlx4_is_bonded(ibdev->dev))
395 		port_num = 1;
396 
397 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
398 		return index;
399 
400 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
401 	if (ret)
402 		return ret;
403 
404 	if (attr.ndev)
405 		dev_put(attr.ndev);
406 
407 	if (!memcmp(&gid, &zgid, sizeof(gid)))
408 		return -EINVAL;
409 
410 	spin_lock_irqsave(&iboe->lock, flags);
411 	port_gid_table = &iboe->gids[port_num - 1];
412 
413 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
414 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
415 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
416 			ctx = port_gid_table->gids[i].ctx;
417 			break;
418 		}
419 	if (ctx)
420 		real_index = ctx->real_index;
421 	spin_unlock_irqrestore(&iboe->lock, flags);
422 	return real_index;
423 }
424 
425 static int mlx4_ib_query_device(struct ib_device *ibdev,
426 				struct ib_device_attr *props,
427 				struct ib_udata *uhw)
428 {
429 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
430 	struct ib_smp *in_mad  = NULL;
431 	struct ib_smp *out_mad = NULL;
432 	int err = -ENOMEM;
433 	int have_ib_ports;
434 	struct mlx4_uverbs_ex_query_device cmd;
435 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
436 	struct mlx4_clock_params clock_params;
437 
438 	if (uhw->inlen) {
439 		if (uhw->inlen < sizeof(cmd))
440 			return -EINVAL;
441 
442 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
443 		if (err)
444 			return err;
445 
446 		if (cmd.comp_mask)
447 			return -EINVAL;
448 
449 		if (cmd.reserved)
450 			return -EINVAL;
451 	}
452 
453 	resp.response_length = offsetof(typeof(resp), response_length) +
454 		sizeof(resp.response_length);
455 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
456 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
457 	if (!in_mad || !out_mad)
458 		goto out;
459 
460 	init_query_mad(in_mad);
461 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
462 
463 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
464 			   1, NULL, NULL, in_mad, out_mad);
465 	if (err)
466 		goto out;
467 
468 	memset(props, 0, sizeof *props);
469 
470 	have_ib_ports = num_ib_ports(dev->dev);
471 
472 	props->fw_ver = dev->dev->caps.fw_ver;
473 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
474 		IB_DEVICE_PORT_ACTIVE_EVENT		|
475 		IB_DEVICE_SYS_IMAGE_GUID		|
476 		IB_DEVICE_RC_RNR_NAK_GEN		|
477 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
478 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
479 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
480 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
481 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
482 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
483 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
484 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
485 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
486 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
487 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
488 	if (dev->dev->caps.max_gso_sz &&
489 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
490 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
491 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
492 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
493 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
494 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
495 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
496 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
497 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
498 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
499 		props->device_cap_flags |= IB_DEVICE_XRC;
500 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
501 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
502 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
503 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
504 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
505 		else
506 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
507 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
508 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
509 	}
510 
511 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
512 
513 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
514 		0xffffff;
515 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
516 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
517 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
518 
519 	props->max_mr_size	   = ~0ull;
520 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
521 	props->max_qp		   = dev->dev->quotas.qp;
522 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
523 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
524 					 dev->dev->caps.max_rq_sg);
525 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
526 	props->max_cq		   = dev->dev->quotas.cq;
527 	props->max_cqe		   = dev->dev->caps.max_cqes;
528 	props->max_mr		   = dev->dev->quotas.mpt;
529 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
530 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
531 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
532 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
533 	props->max_srq		   = dev->dev->quotas.srq;
534 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
535 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
536 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
537 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
538 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
539 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
540 	props->masked_atomic_cap   = props->atomic_cap;
541 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
542 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
543 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
544 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
545 					   props->max_mcast_grp;
546 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
547 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
548 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
549 
550 	if (!mlx4_is_slave(dev->dev))
551 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
552 
553 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
554 		resp.response_length += sizeof(resp.hca_core_clock_offset);
555 		if (!err && !mlx4_is_slave(dev->dev)) {
556 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
557 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
558 		}
559 	}
560 
561 	if (uhw->outlen) {
562 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
563 		if (err)
564 			goto out;
565 	}
566 out:
567 	kfree(in_mad);
568 	kfree(out_mad);
569 
570 	return err;
571 }
572 
573 static enum rdma_link_layer
574 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
575 {
576 	struct mlx4_dev *dev = to_mdev(device)->dev;
577 
578 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
579 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
580 }
581 
582 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
583 			      struct ib_port_attr *props, int netw_view)
584 {
585 	struct ib_smp *in_mad  = NULL;
586 	struct ib_smp *out_mad = NULL;
587 	int ext_active_speed;
588 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
589 	int err = -ENOMEM;
590 
591 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
592 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
593 	if (!in_mad || !out_mad)
594 		goto out;
595 
596 	init_query_mad(in_mad);
597 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
598 	in_mad->attr_mod = cpu_to_be32(port);
599 
600 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
601 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
602 
603 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
604 				in_mad, out_mad);
605 	if (err)
606 		goto out;
607 
608 
609 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
610 	props->lmc		= out_mad->data[34] & 0x7;
611 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
612 	props->sm_sl		= out_mad->data[36] & 0xf;
613 	props->state		= out_mad->data[32] & 0xf;
614 	props->phys_state	= out_mad->data[33] >> 4;
615 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
616 	if (netw_view)
617 		props->gid_tbl_len = out_mad->data[50];
618 	else
619 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
620 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
621 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
622 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
623 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
624 	props->active_width	= out_mad->data[31] & 0xf;
625 	props->active_speed	= out_mad->data[35] >> 4;
626 	props->max_mtu		= out_mad->data[41] & 0xf;
627 	props->active_mtu	= out_mad->data[36] >> 4;
628 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
629 	props->max_vl_num	= out_mad->data[37] >> 4;
630 	props->init_type_reply	= out_mad->data[41] >> 4;
631 
632 	/* Check if extended speeds (EDR/FDR/...) are supported */
633 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
634 		ext_active_speed = out_mad->data[62] >> 4;
635 
636 		switch (ext_active_speed) {
637 		case 1:
638 			props->active_speed = IB_SPEED_FDR;
639 			break;
640 		case 2:
641 			props->active_speed = IB_SPEED_EDR;
642 			break;
643 		}
644 	}
645 
646 	/* If reported active speed is QDR, check if is FDR-10 */
647 	if (props->active_speed == IB_SPEED_QDR) {
648 		init_query_mad(in_mad);
649 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
650 		in_mad->attr_mod = cpu_to_be32(port);
651 
652 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
653 				   NULL, NULL, in_mad, out_mad);
654 		if (err)
655 			goto out;
656 
657 		/* Checking LinkSpeedActive for FDR-10 */
658 		if (out_mad->data[15] & 0x1)
659 			props->active_speed = IB_SPEED_FDR10;
660 	}
661 
662 	/* Avoid wrong speed value returned by FW if the IB link is down. */
663 	if (props->state == IB_PORT_DOWN)
664 		 props->active_speed = IB_SPEED_SDR;
665 
666 out:
667 	kfree(in_mad);
668 	kfree(out_mad);
669 	return err;
670 }
671 
672 static u8 state_to_phys_state(enum ib_port_state state)
673 {
674 	return state == IB_PORT_ACTIVE ? 5 : 3;
675 }
676 
677 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
678 			       struct ib_port_attr *props, int netw_view)
679 {
680 
681 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
682 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
683 	struct net_device *ndev;
684 	enum ib_mtu tmp;
685 	struct mlx4_cmd_mailbox *mailbox;
686 	int err = 0;
687 	int is_bonded = mlx4_is_bonded(mdev->dev);
688 
689 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
690 	if (IS_ERR(mailbox))
691 		return PTR_ERR(mailbox);
692 
693 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
694 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
695 			   MLX4_CMD_WRAPPED);
696 	if (err)
697 		goto out;
698 
699 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
700 						IB_WIDTH_4X : IB_WIDTH_1X;
701 	props->active_speed	= IB_SPEED_QDR;
702 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
703 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
704 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
705 	props->pkey_tbl_len	= 1;
706 	props->max_mtu		= IB_MTU_4096;
707 	props->max_vl_num	= 2;
708 	props->state		= IB_PORT_DOWN;
709 	props->phys_state	= state_to_phys_state(props->state);
710 	props->active_mtu	= IB_MTU_256;
711 	spin_lock_bh(&iboe->lock);
712 	ndev = iboe->netdevs[port - 1];
713 	if (ndev && is_bonded) {
714 		rcu_read_lock(); /* required to get upper dev */
715 		ndev = netdev_master_upper_dev_get_rcu(ndev);
716 		rcu_read_unlock();
717 	}
718 	if (!ndev)
719 		goto out_unlock;
720 
721 	tmp = iboe_get_mtu(ndev->mtu);
722 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
723 
724 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
725 					IB_PORT_ACTIVE : IB_PORT_DOWN;
726 	props->phys_state	= state_to_phys_state(props->state);
727 out_unlock:
728 	spin_unlock_bh(&iboe->lock);
729 out:
730 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
731 	return err;
732 }
733 
734 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
735 			 struct ib_port_attr *props, int netw_view)
736 {
737 	int err;
738 
739 	memset(props, 0, sizeof *props);
740 
741 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
742 		ib_link_query_port(ibdev, port, props, netw_view) :
743 				eth_link_query_port(ibdev, port, props, netw_view);
744 
745 	return err;
746 }
747 
748 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
749 			      struct ib_port_attr *props)
750 {
751 	/* returns host view */
752 	return __mlx4_ib_query_port(ibdev, port, props, 0);
753 }
754 
755 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
756 			union ib_gid *gid, int netw_view)
757 {
758 	struct ib_smp *in_mad  = NULL;
759 	struct ib_smp *out_mad = NULL;
760 	int err = -ENOMEM;
761 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
762 	int clear = 0;
763 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
764 
765 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
766 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
767 	if (!in_mad || !out_mad)
768 		goto out;
769 
770 	init_query_mad(in_mad);
771 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
772 	in_mad->attr_mod = cpu_to_be32(port);
773 
774 	if (mlx4_is_mfunc(dev->dev) && netw_view)
775 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
776 
777 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
778 	if (err)
779 		goto out;
780 
781 	memcpy(gid->raw, out_mad->data + 8, 8);
782 
783 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
784 		if (index) {
785 			/* For any index > 0, return the null guid */
786 			err = 0;
787 			clear = 1;
788 			goto out;
789 		}
790 	}
791 
792 	init_query_mad(in_mad);
793 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
794 	in_mad->attr_mod = cpu_to_be32(index / 8);
795 
796 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
797 			   NULL, NULL, in_mad, out_mad);
798 	if (err)
799 		goto out;
800 
801 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
802 
803 out:
804 	if (clear)
805 		memset(gid->raw + 8, 0, 8);
806 	kfree(in_mad);
807 	kfree(out_mad);
808 	return err;
809 }
810 
811 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
812 			     union ib_gid *gid)
813 {
814 	int ret;
815 
816 	if (rdma_protocol_ib(ibdev, port))
817 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
818 
819 	if (!rdma_protocol_roce(ibdev, port))
820 		return -ENODEV;
821 
822 	if (!rdma_cap_roce_gid_table(ibdev, port))
823 		return -ENODEV;
824 
825 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
826 	if (ret == -EAGAIN) {
827 		memcpy(gid, &zgid, sizeof(*gid));
828 		return 0;
829 	}
830 
831 	return ret;
832 }
833 
834 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
835 			 u16 *pkey, int netw_view)
836 {
837 	struct ib_smp *in_mad  = NULL;
838 	struct ib_smp *out_mad = NULL;
839 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
840 	int err = -ENOMEM;
841 
842 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
843 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
844 	if (!in_mad || !out_mad)
845 		goto out;
846 
847 	init_query_mad(in_mad);
848 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
849 	in_mad->attr_mod = cpu_to_be32(index / 32);
850 
851 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
852 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
853 
854 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
855 			   in_mad, out_mad);
856 	if (err)
857 		goto out;
858 
859 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
860 
861 out:
862 	kfree(in_mad);
863 	kfree(out_mad);
864 	return err;
865 }
866 
867 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
868 {
869 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
870 }
871 
872 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
873 				 struct ib_device_modify *props)
874 {
875 	struct mlx4_cmd_mailbox *mailbox;
876 	unsigned long flags;
877 
878 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
879 		return -EOPNOTSUPP;
880 
881 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
882 		return 0;
883 
884 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
885 		return -EOPNOTSUPP;
886 
887 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
888 	memcpy(ibdev->node_desc, props->node_desc, 64);
889 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
890 
891 	/*
892 	 * If possible, pass node desc to FW, so it can generate
893 	 * a 144 trap.  If cmd fails, just ignore.
894 	 */
895 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
896 	if (IS_ERR(mailbox))
897 		return 0;
898 
899 	memcpy(mailbox->buf, props->node_desc, 64);
900 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
901 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
902 
903 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
904 
905 	return 0;
906 }
907 
908 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
909 			    u32 cap_mask)
910 {
911 	struct mlx4_cmd_mailbox *mailbox;
912 	int err;
913 
914 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
915 	if (IS_ERR(mailbox))
916 		return PTR_ERR(mailbox);
917 
918 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
919 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
920 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
921 	} else {
922 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
923 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
924 	}
925 
926 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
927 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
928 		       MLX4_CMD_WRAPPED);
929 
930 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
931 	return err;
932 }
933 
934 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
935 			       struct ib_port_modify *props)
936 {
937 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
938 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
939 	struct ib_port_attr attr;
940 	u32 cap_mask;
941 	int err;
942 
943 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
944 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
945 	 * violations and port capabilities are not meaningful.
946 	 */
947 	if (is_eth)
948 		return 0;
949 
950 	mutex_lock(&mdev->cap_mask_mutex);
951 
952 	err = mlx4_ib_query_port(ibdev, port, &attr);
953 	if (err)
954 		goto out;
955 
956 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
957 		~props->clr_port_cap_mask;
958 
959 	err = mlx4_ib_SET_PORT(mdev, port,
960 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
961 			       cap_mask);
962 
963 out:
964 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
965 	return err;
966 }
967 
968 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
969 						  struct ib_udata *udata)
970 {
971 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
972 	struct mlx4_ib_ucontext *context;
973 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
974 	struct mlx4_ib_alloc_ucontext_resp resp;
975 	int err;
976 
977 	if (!dev->ib_active)
978 		return ERR_PTR(-EAGAIN);
979 
980 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
981 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
982 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
983 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
984 	} else {
985 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
986 		resp.qp_tab_size      = dev->dev->caps.num_qps;
987 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
988 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
989 		resp.cqe_size	      = dev->dev->caps.cqe_size;
990 	}
991 
992 	context = kzalloc(sizeof(*context), GFP_KERNEL);
993 	if (!context)
994 		return ERR_PTR(-ENOMEM);
995 
996 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
997 	if (err) {
998 		kfree(context);
999 		return ERR_PTR(err);
1000 	}
1001 
1002 	INIT_LIST_HEAD(&context->db_page_list);
1003 	mutex_init(&context->db_page_mutex);
1004 
1005 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1006 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1007 	else
1008 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1009 
1010 	if (err) {
1011 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1012 		kfree(context);
1013 		return ERR_PTR(-EFAULT);
1014 	}
1015 
1016 	return &context->ibucontext;
1017 }
1018 
1019 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1020 {
1021 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1022 
1023 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1024 	kfree(context);
1025 
1026 	return 0;
1027 }
1028 
1029 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1030 {
1031 	/* vma_open is called when a new VMA is created on top of our VMA.
1032 	 * This is done through either mremap flow or split_vma (usually due
1033 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1034 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1035 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1036 	 * calling us again and trying to do incorrect actions. We assume that
1037 	 * the original vma size is exactly a single page that there will be no
1038 	 * "splitting" operations on.
1039 	 */
1040 	area->vm_ops = NULL;
1041 }
1042 
1043 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1044 {
1045 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1046 
1047 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1048 	 * file itself is closed, therefore no sync is needed with the regular
1049 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1050 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1051 	 * The close operation is usually called under mm->mmap_sem except when
1052 	 * process is exiting.  The exiting case is handled explicitly as part
1053 	 * of mlx4_ib_disassociate_ucontext.
1054 	 */
1055 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1056 				area->vm_private_data;
1057 
1058 	/* set the vma context pointer to null in the mlx4_ib driver's private
1059 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1060 	 */
1061 	mlx4_ib_vma_priv_data->vma = NULL;
1062 }
1063 
1064 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1065 	.open = mlx4_ib_vma_open,
1066 	.close = mlx4_ib_vma_close
1067 };
1068 
1069 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1070 {
1071 	int i;
1072 	int ret = 0;
1073 	struct vm_area_struct *vma;
1074 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1075 	struct task_struct *owning_process  = NULL;
1076 	struct mm_struct   *owning_mm       = NULL;
1077 
1078 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1079 	if (!owning_process)
1080 		return;
1081 
1082 	owning_mm = get_task_mm(owning_process);
1083 	if (!owning_mm) {
1084 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1085 		while (1) {
1086 			/* make sure that task is dead before returning, it may
1087 			 * prevent a rare case of module down in parallel to a
1088 			 * call to mlx4_ib_vma_close.
1089 			 */
1090 			put_task_struct(owning_process);
1091 			msleep(1);
1092 			owning_process = get_pid_task(ibcontext->tgid,
1093 						      PIDTYPE_PID);
1094 			if (!owning_process ||
1095 			    owning_process->state == TASK_DEAD) {
1096 				pr_info("disassociate ucontext done, task was terminated\n");
1097 				/* in case task was dead need to release the task struct */
1098 				if (owning_process)
1099 					put_task_struct(owning_process);
1100 				return;
1101 			}
1102 		}
1103 	}
1104 
1105 	/* need to protect from a race on closing the vma as part of
1106 	 * mlx4_ib_vma_close().
1107 	 */
1108 	down_read(&owning_mm->mmap_sem);
1109 	for (i = 0; i < HW_BAR_COUNT; i++) {
1110 		vma = context->hw_bar_info[i].vma;
1111 		if (!vma)
1112 			continue;
1113 
1114 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1115 				   context->hw_bar_info[i].vma->vm_start,
1116 				   PAGE_SIZE);
1117 		if (ret) {
1118 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1119 			BUG_ON(1);
1120 		}
1121 
1122 		/* context going to be destroyed, should not access ops any more */
1123 		context->hw_bar_info[i].vma->vm_ops = NULL;
1124 	}
1125 
1126 	up_read(&owning_mm->mmap_sem);
1127 	mmput(owning_mm);
1128 	put_task_struct(owning_process);
1129 }
1130 
1131 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1132 				 struct mlx4_ib_vma_private_data *vma_private_data)
1133 {
1134 	vma_private_data->vma = vma;
1135 	vma->vm_private_data = vma_private_data;
1136 	vma->vm_ops =  &mlx4_ib_vm_ops;
1137 }
1138 
1139 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1140 {
1141 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1142 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1143 
1144 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1145 		return -EINVAL;
1146 
1147 	if (vma->vm_pgoff == 0) {
1148 		/* We prevent double mmaping on same context */
1149 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1150 			return -EINVAL;
1151 
1152 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1153 
1154 		if (io_remap_pfn_range(vma, vma->vm_start,
1155 				       to_mucontext(context)->uar.pfn,
1156 				       PAGE_SIZE, vma->vm_page_prot))
1157 			return -EAGAIN;
1158 
1159 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1160 
1161 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1162 		/* We prevent double mmaping on same context */
1163 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1164 			return -EINVAL;
1165 
1166 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1167 
1168 		if (io_remap_pfn_range(vma, vma->vm_start,
1169 				       to_mucontext(context)->uar.pfn +
1170 				       dev->dev->caps.num_uars,
1171 				       PAGE_SIZE, vma->vm_page_prot))
1172 			return -EAGAIN;
1173 
1174 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1175 
1176 	} else if (vma->vm_pgoff == 3) {
1177 		struct mlx4_clock_params params;
1178 		int ret;
1179 
1180 		/* We prevent double mmaping on same context */
1181 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1182 			return -EINVAL;
1183 
1184 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1185 
1186 		if (ret)
1187 			return ret;
1188 
1189 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1190 		if (io_remap_pfn_range(vma, vma->vm_start,
1191 				       (pci_resource_start(dev->dev->persist->pdev,
1192 							   params.bar) +
1193 					params.offset)
1194 				       >> PAGE_SHIFT,
1195 				       PAGE_SIZE, vma->vm_page_prot))
1196 			return -EAGAIN;
1197 
1198 		mlx4_ib_set_vma_data(vma,
1199 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1200 	} else {
1201 		return -EINVAL;
1202 	}
1203 
1204 	return 0;
1205 }
1206 
1207 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1208 				      struct ib_ucontext *context,
1209 				      struct ib_udata *udata)
1210 {
1211 	struct mlx4_ib_pd *pd;
1212 	int err;
1213 
1214 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1215 	if (!pd)
1216 		return ERR_PTR(-ENOMEM);
1217 
1218 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1219 	if (err) {
1220 		kfree(pd);
1221 		return ERR_PTR(err);
1222 	}
1223 
1224 	if (context)
1225 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1226 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1227 			kfree(pd);
1228 			return ERR_PTR(-EFAULT);
1229 		}
1230 
1231 	return &pd->ibpd;
1232 }
1233 
1234 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1235 {
1236 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1237 	kfree(pd);
1238 
1239 	return 0;
1240 }
1241 
1242 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1243 					  struct ib_ucontext *context,
1244 					  struct ib_udata *udata)
1245 {
1246 	struct mlx4_ib_xrcd *xrcd;
1247 	struct ib_cq_init_attr cq_attr = {};
1248 	int err;
1249 
1250 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1251 		return ERR_PTR(-ENOSYS);
1252 
1253 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1254 	if (!xrcd)
1255 		return ERR_PTR(-ENOMEM);
1256 
1257 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1258 	if (err)
1259 		goto err1;
1260 
1261 	xrcd->pd = ib_alloc_pd(ibdev);
1262 	if (IS_ERR(xrcd->pd)) {
1263 		err = PTR_ERR(xrcd->pd);
1264 		goto err2;
1265 	}
1266 
1267 	cq_attr.cqe = 1;
1268 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1269 	if (IS_ERR(xrcd->cq)) {
1270 		err = PTR_ERR(xrcd->cq);
1271 		goto err3;
1272 	}
1273 
1274 	return &xrcd->ibxrcd;
1275 
1276 err3:
1277 	ib_dealloc_pd(xrcd->pd);
1278 err2:
1279 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1280 err1:
1281 	kfree(xrcd);
1282 	return ERR_PTR(err);
1283 }
1284 
1285 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1286 {
1287 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1288 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1289 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1290 	kfree(xrcd);
1291 
1292 	return 0;
1293 }
1294 
1295 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1296 {
1297 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1298 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1299 	struct mlx4_ib_gid_entry *ge;
1300 
1301 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1302 	if (!ge)
1303 		return -ENOMEM;
1304 
1305 	ge->gid = *gid;
1306 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1307 		ge->port = mqp->port;
1308 		ge->added = 1;
1309 	}
1310 
1311 	mutex_lock(&mqp->mutex);
1312 	list_add_tail(&ge->list, &mqp->gid_list);
1313 	mutex_unlock(&mqp->mutex);
1314 
1315 	return 0;
1316 }
1317 
1318 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1319 					  struct mlx4_ib_counters *ctr_table)
1320 {
1321 	struct counter_index *counter, *tmp_count;
1322 
1323 	mutex_lock(&ctr_table->mutex);
1324 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1325 				 list) {
1326 		if (counter->allocated)
1327 			mlx4_counter_free(ibdev->dev, counter->index);
1328 		list_del(&counter->list);
1329 		kfree(counter);
1330 	}
1331 	mutex_unlock(&ctr_table->mutex);
1332 }
1333 
1334 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1335 		   union ib_gid *gid)
1336 {
1337 	struct net_device *ndev;
1338 	int ret = 0;
1339 
1340 	if (!mqp->port)
1341 		return 0;
1342 
1343 	spin_lock_bh(&mdev->iboe.lock);
1344 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1345 	if (ndev)
1346 		dev_hold(ndev);
1347 	spin_unlock_bh(&mdev->iboe.lock);
1348 
1349 	if (ndev) {
1350 		ret = 1;
1351 		dev_put(ndev);
1352 	}
1353 
1354 	return ret;
1355 }
1356 
1357 struct mlx4_ib_steering {
1358 	struct list_head list;
1359 	struct mlx4_flow_reg_id reg_id;
1360 	union ib_gid gid;
1361 };
1362 
1363 static int parse_flow_attr(struct mlx4_dev *dev,
1364 			   u32 qp_num,
1365 			   union ib_flow_spec *ib_spec,
1366 			   struct _rule_hw *mlx4_spec)
1367 {
1368 	enum mlx4_net_trans_rule_id type;
1369 
1370 	switch (ib_spec->type) {
1371 	case IB_FLOW_SPEC_ETH:
1372 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1373 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1374 		       ETH_ALEN);
1375 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1376 		       ETH_ALEN);
1377 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1378 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1379 		break;
1380 	case IB_FLOW_SPEC_IB:
1381 		type = MLX4_NET_TRANS_RULE_ID_IB;
1382 		mlx4_spec->ib.l3_qpn =
1383 			cpu_to_be32(qp_num);
1384 		mlx4_spec->ib.qpn_mask =
1385 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1386 		break;
1387 
1388 
1389 	case IB_FLOW_SPEC_IPV4:
1390 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1391 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1392 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1393 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1394 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1395 		break;
1396 
1397 	case IB_FLOW_SPEC_TCP:
1398 	case IB_FLOW_SPEC_UDP:
1399 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1400 					MLX4_NET_TRANS_RULE_ID_TCP :
1401 					MLX4_NET_TRANS_RULE_ID_UDP;
1402 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1403 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1404 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1405 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1406 		break;
1407 
1408 	default:
1409 		return -EINVAL;
1410 	}
1411 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1412 	    mlx4_hw_rule_sz(dev, type) < 0)
1413 		return -EINVAL;
1414 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1415 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1416 	return mlx4_hw_rule_sz(dev, type);
1417 }
1418 
1419 struct default_rules {
1420 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1421 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1422 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1423 	__u8  link_layer;
1424 };
1425 static const struct default_rules default_table[] = {
1426 	{
1427 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1428 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1429 		.rules_create_list = {IB_FLOW_SPEC_IB},
1430 		.link_layer = IB_LINK_LAYER_INFINIBAND
1431 	}
1432 };
1433 
1434 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1435 					 struct ib_flow_attr *flow_attr)
1436 {
1437 	int i, j, k;
1438 	void *ib_flow;
1439 	const struct default_rules *pdefault_rules = default_table;
1440 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1441 
1442 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1443 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1444 		memset(&field_types, 0, sizeof(field_types));
1445 
1446 		if (link_layer != pdefault_rules->link_layer)
1447 			continue;
1448 
1449 		ib_flow = flow_attr + 1;
1450 		/* we assume the specs are sorted */
1451 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1452 		     j < flow_attr->num_of_specs; k++) {
1453 			union ib_flow_spec *current_flow =
1454 				(union ib_flow_spec *)ib_flow;
1455 
1456 			/* same layer but different type */
1457 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1458 			     (pdefault_rules->mandatory_fields[k] &
1459 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1460 			    (current_flow->type !=
1461 			     pdefault_rules->mandatory_fields[k]))
1462 				goto out;
1463 
1464 			/* same layer, try match next one */
1465 			if (current_flow->type ==
1466 			    pdefault_rules->mandatory_fields[k]) {
1467 				j++;
1468 				ib_flow +=
1469 					((union ib_flow_spec *)ib_flow)->size;
1470 			}
1471 		}
1472 
1473 		ib_flow = flow_attr + 1;
1474 		for (j = 0; j < flow_attr->num_of_specs;
1475 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1476 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1477 				/* same layer and same type */
1478 				if (((union ib_flow_spec *)ib_flow)->type ==
1479 				    pdefault_rules->mandatory_not_fields[k])
1480 					goto out;
1481 
1482 		return i;
1483 	}
1484 out:
1485 	return -1;
1486 }
1487 
1488 static int __mlx4_ib_create_default_rules(
1489 		struct mlx4_ib_dev *mdev,
1490 		struct ib_qp *qp,
1491 		const struct default_rules *pdefault_rules,
1492 		struct _rule_hw *mlx4_spec) {
1493 	int size = 0;
1494 	int i;
1495 
1496 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1497 		int ret;
1498 		union ib_flow_spec ib_spec;
1499 		switch (pdefault_rules->rules_create_list[i]) {
1500 		case 0:
1501 			/* no rule */
1502 			continue;
1503 		case IB_FLOW_SPEC_IB:
1504 			ib_spec.type = IB_FLOW_SPEC_IB;
1505 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1506 
1507 			break;
1508 		default:
1509 			/* invalid rule */
1510 			return -EINVAL;
1511 		}
1512 		/* We must put empty rule, qpn is being ignored */
1513 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1514 				      mlx4_spec);
1515 		if (ret < 0) {
1516 			pr_info("invalid parsing\n");
1517 			return -EINVAL;
1518 		}
1519 
1520 		mlx4_spec = (void *)mlx4_spec + ret;
1521 		size += ret;
1522 	}
1523 	return size;
1524 }
1525 
1526 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1527 			  int domain,
1528 			  enum mlx4_net_trans_promisc_mode flow_type,
1529 			  u64 *reg_id)
1530 {
1531 	int ret, i;
1532 	int size = 0;
1533 	void *ib_flow;
1534 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1535 	struct mlx4_cmd_mailbox *mailbox;
1536 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1537 	int default_flow;
1538 
1539 	static const u16 __mlx4_domain[] = {
1540 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1541 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1542 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1543 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1544 	};
1545 
1546 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1547 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1548 		return -EINVAL;
1549 	}
1550 
1551 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1552 		pr_err("Invalid domain value %d\n", domain);
1553 		return -EINVAL;
1554 	}
1555 
1556 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1557 		return -EINVAL;
1558 
1559 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1560 	if (IS_ERR(mailbox))
1561 		return PTR_ERR(mailbox);
1562 	ctrl = mailbox->buf;
1563 
1564 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1565 				 flow_attr->priority);
1566 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1567 	ctrl->port = flow_attr->port;
1568 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1569 
1570 	ib_flow = flow_attr + 1;
1571 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1572 	/* Add default flows */
1573 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1574 	if (default_flow >= 0) {
1575 		ret = __mlx4_ib_create_default_rules(
1576 				mdev, qp, default_table + default_flow,
1577 				mailbox->buf + size);
1578 		if (ret < 0) {
1579 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1580 			return -EINVAL;
1581 		}
1582 		size += ret;
1583 	}
1584 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1585 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1586 				      mailbox->buf + size);
1587 		if (ret < 0) {
1588 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1589 			return -EINVAL;
1590 		}
1591 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1592 		size += ret;
1593 	}
1594 
1595 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1596 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1597 			   MLX4_CMD_WRAPPED);
1598 	if (ret == -ENOMEM)
1599 		pr_err("mcg table is full. Fail to register network rule.\n");
1600 	else if (ret == -ENXIO)
1601 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1602 	else if (ret)
1603 		pr_err("Invalid argumant. Fail to register network rule.\n");
1604 
1605 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1606 	return ret;
1607 }
1608 
1609 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1610 {
1611 	int err;
1612 	err = mlx4_cmd(dev, reg_id, 0, 0,
1613 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1614 		       MLX4_CMD_WRAPPED);
1615 	if (err)
1616 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1617 		       reg_id);
1618 	return err;
1619 }
1620 
1621 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1622 				    u64 *reg_id)
1623 {
1624 	void *ib_flow;
1625 	union ib_flow_spec *ib_spec;
1626 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1627 	int err = 0;
1628 
1629 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1630 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1631 		return 0; /* do nothing */
1632 
1633 	ib_flow = flow_attr + 1;
1634 	ib_spec = (union ib_flow_spec *)ib_flow;
1635 
1636 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1637 		return 0; /* do nothing */
1638 
1639 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1640 				    flow_attr->port, qp->qp_num,
1641 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1642 				    reg_id);
1643 	return err;
1644 }
1645 
1646 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1647 				    struct ib_flow_attr *flow_attr,
1648 				    int domain)
1649 {
1650 	int err = 0, i = 0, j = 0;
1651 	struct mlx4_ib_flow *mflow;
1652 	enum mlx4_net_trans_promisc_mode type[2];
1653 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1654 	int is_bonded = mlx4_is_bonded(dev);
1655 
1656 	memset(type, 0, sizeof(type));
1657 
1658 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1659 	if (!mflow) {
1660 		err = -ENOMEM;
1661 		goto err_free;
1662 	}
1663 
1664 	switch (flow_attr->type) {
1665 	case IB_FLOW_ATTR_NORMAL:
1666 		type[0] = MLX4_FS_REGULAR;
1667 		break;
1668 
1669 	case IB_FLOW_ATTR_ALL_DEFAULT:
1670 		type[0] = MLX4_FS_ALL_DEFAULT;
1671 		break;
1672 
1673 	case IB_FLOW_ATTR_MC_DEFAULT:
1674 		type[0] = MLX4_FS_MC_DEFAULT;
1675 		break;
1676 
1677 	case IB_FLOW_ATTR_SNIFFER:
1678 		type[0] = MLX4_FS_UC_SNIFFER;
1679 		type[1] = MLX4_FS_MC_SNIFFER;
1680 		break;
1681 
1682 	default:
1683 		err = -EINVAL;
1684 		goto err_free;
1685 	}
1686 
1687 	while (i < ARRAY_SIZE(type) && type[i]) {
1688 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1689 					    &mflow->reg_id[i].id);
1690 		if (err)
1691 			goto err_create_flow;
1692 		if (is_bonded) {
1693 			/* Application always sees one port so the mirror rule
1694 			 * must be on port #2
1695 			 */
1696 			flow_attr->port = 2;
1697 			err = __mlx4_ib_create_flow(qp, flow_attr,
1698 						    domain, type[j],
1699 						    &mflow->reg_id[j].mirror);
1700 			flow_attr->port = 1;
1701 			if (err)
1702 				goto err_create_flow;
1703 			j++;
1704 		}
1705 
1706 		i++;
1707 	}
1708 
1709 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1710 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1711 					       &mflow->reg_id[i].id);
1712 		if (err)
1713 			goto err_create_flow;
1714 
1715 		if (is_bonded) {
1716 			flow_attr->port = 2;
1717 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1718 						       &mflow->reg_id[j].mirror);
1719 			flow_attr->port = 1;
1720 			if (err)
1721 				goto err_create_flow;
1722 			j++;
1723 		}
1724 		/* function to create mirror rule */
1725 		i++;
1726 	}
1727 
1728 	return &mflow->ibflow;
1729 
1730 err_create_flow:
1731 	while (i) {
1732 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1733 					     mflow->reg_id[i].id);
1734 		i--;
1735 	}
1736 
1737 	while (j) {
1738 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1739 					     mflow->reg_id[j].mirror);
1740 		j--;
1741 	}
1742 err_free:
1743 	kfree(mflow);
1744 	return ERR_PTR(err);
1745 }
1746 
1747 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1748 {
1749 	int err, ret = 0;
1750 	int i = 0;
1751 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1752 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1753 
1754 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1755 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1756 		if (err)
1757 			ret = err;
1758 		if (mflow->reg_id[i].mirror) {
1759 			err = __mlx4_ib_destroy_flow(mdev->dev,
1760 						     mflow->reg_id[i].mirror);
1761 			if (err)
1762 				ret = err;
1763 		}
1764 		i++;
1765 	}
1766 
1767 	kfree(mflow);
1768 	return ret;
1769 }
1770 
1771 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1772 {
1773 	int err;
1774 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1775 	struct mlx4_dev	*dev = mdev->dev;
1776 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1777 	struct mlx4_ib_steering *ib_steering = NULL;
1778 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1779 	struct mlx4_flow_reg_id	reg_id;
1780 
1781 	if (mdev->dev->caps.steering_mode ==
1782 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1783 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1784 		if (!ib_steering)
1785 			return -ENOMEM;
1786 	}
1787 
1788 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1789 				    !!(mqp->flags &
1790 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1791 				    prot, &reg_id.id);
1792 	if (err) {
1793 		pr_err("multicast attach op failed, err %d\n", err);
1794 		goto err_malloc;
1795 	}
1796 
1797 	reg_id.mirror = 0;
1798 	if (mlx4_is_bonded(dev)) {
1799 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1800 					    (mqp->port == 1) ? 2 : 1,
1801 					    !!(mqp->flags &
1802 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1803 					    prot, &reg_id.mirror);
1804 		if (err)
1805 			goto err_add;
1806 	}
1807 
1808 	err = add_gid_entry(ibqp, gid);
1809 	if (err)
1810 		goto err_add;
1811 
1812 	if (ib_steering) {
1813 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1814 		ib_steering->reg_id = reg_id;
1815 		mutex_lock(&mqp->mutex);
1816 		list_add(&ib_steering->list, &mqp->steering_rules);
1817 		mutex_unlock(&mqp->mutex);
1818 	}
1819 	return 0;
1820 
1821 err_add:
1822 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1823 			      prot, reg_id.id);
1824 	if (reg_id.mirror)
1825 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1826 				      prot, reg_id.mirror);
1827 err_malloc:
1828 	kfree(ib_steering);
1829 
1830 	return err;
1831 }
1832 
1833 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1834 {
1835 	struct mlx4_ib_gid_entry *ge;
1836 	struct mlx4_ib_gid_entry *tmp;
1837 	struct mlx4_ib_gid_entry *ret = NULL;
1838 
1839 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1840 		if (!memcmp(raw, ge->gid.raw, 16)) {
1841 			ret = ge;
1842 			break;
1843 		}
1844 	}
1845 
1846 	return ret;
1847 }
1848 
1849 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1850 {
1851 	int err;
1852 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1853 	struct mlx4_dev *dev = mdev->dev;
1854 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1855 	struct net_device *ndev;
1856 	struct mlx4_ib_gid_entry *ge;
1857 	struct mlx4_flow_reg_id reg_id = {0, 0};
1858 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1859 
1860 	if (mdev->dev->caps.steering_mode ==
1861 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1862 		struct mlx4_ib_steering *ib_steering;
1863 
1864 		mutex_lock(&mqp->mutex);
1865 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1866 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1867 				list_del(&ib_steering->list);
1868 				break;
1869 			}
1870 		}
1871 		mutex_unlock(&mqp->mutex);
1872 		if (&ib_steering->list == &mqp->steering_rules) {
1873 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1874 			return -EINVAL;
1875 		}
1876 		reg_id = ib_steering->reg_id;
1877 		kfree(ib_steering);
1878 	}
1879 
1880 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1881 				    prot, reg_id.id);
1882 	if (err)
1883 		return err;
1884 
1885 	if (mlx4_is_bonded(dev)) {
1886 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1887 					    prot, reg_id.mirror);
1888 		if (err)
1889 			return err;
1890 	}
1891 
1892 	mutex_lock(&mqp->mutex);
1893 	ge = find_gid_entry(mqp, gid->raw);
1894 	if (ge) {
1895 		spin_lock_bh(&mdev->iboe.lock);
1896 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1897 		if (ndev)
1898 			dev_hold(ndev);
1899 		spin_unlock_bh(&mdev->iboe.lock);
1900 		if (ndev)
1901 			dev_put(ndev);
1902 		list_del(&ge->list);
1903 		kfree(ge);
1904 	} else
1905 		pr_warn("could not find mgid entry\n");
1906 
1907 	mutex_unlock(&mqp->mutex);
1908 
1909 	return 0;
1910 }
1911 
1912 static int init_node_data(struct mlx4_ib_dev *dev)
1913 {
1914 	struct ib_smp *in_mad  = NULL;
1915 	struct ib_smp *out_mad = NULL;
1916 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1917 	int err = -ENOMEM;
1918 
1919 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1920 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1921 	if (!in_mad || !out_mad)
1922 		goto out;
1923 
1924 	init_query_mad(in_mad);
1925 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1926 	if (mlx4_is_master(dev->dev))
1927 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1928 
1929 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1930 	if (err)
1931 		goto out;
1932 
1933 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1934 
1935 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1936 
1937 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1938 	if (err)
1939 		goto out;
1940 
1941 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1942 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1943 
1944 out:
1945 	kfree(in_mad);
1946 	kfree(out_mad);
1947 	return err;
1948 }
1949 
1950 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1951 			char *buf)
1952 {
1953 	struct mlx4_ib_dev *dev =
1954 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1955 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1956 }
1957 
1958 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1959 			   char *buf)
1960 {
1961 	struct mlx4_ib_dev *dev =
1962 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1963 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1964 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1965 		       (int) dev->dev->caps.fw_ver & 0xffff);
1966 }
1967 
1968 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1969 			char *buf)
1970 {
1971 	struct mlx4_ib_dev *dev =
1972 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1973 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1974 }
1975 
1976 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1977 			  char *buf)
1978 {
1979 	struct mlx4_ib_dev *dev =
1980 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1981 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1982 		       dev->dev->board_id);
1983 }
1984 
1985 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1986 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1987 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1988 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1989 
1990 static struct device_attribute *mlx4_class_attributes[] = {
1991 	&dev_attr_hw_rev,
1992 	&dev_attr_fw_ver,
1993 	&dev_attr_hca_type,
1994 	&dev_attr_board_id
1995 };
1996 
1997 #define MLX4_IB_INVALID_MAC	((u64)-1)
1998 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1999 			       struct net_device *dev,
2000 			       int port)
2001 {
2002 	u64 new_smac = 0;
2003 	u64 release_mac = MLX4_IB_INVALID_MAC;
2004 	struct mlx4_ib_qp *qp;
2005 
2006 	read_lock(&dev_base_lock);
2007 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2008 	read_unlock(&dev_base_lock);
2009 
2010 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2011 
2012 	/* no need for update QP1 and mac registration in non-SRIOV */
2013 	if (!mlx4_is_mfunc(ibdev->dev))
2014 		return;
2015 
2016 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2017 	qp = ibdev->qp1_proxy[port - 1];
2018 	if (qp) {
2019 		int new_smac_index;
2020 		u64 old_smac;
2021 		struct mlx4_update_qp_params update_params;
2022 
2023 		mutex_lock(&qp->mutex);
2024 		old_smac = qp->pri.smac;
2025 		if (new_smac == old_smac)
2026 			goto unlock;
2027 
2028 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2029 
2030 		if (new_smac_index < 0)
2031 			goto unlock;
2032 
2033 		update_params.smac_index = new_smac_index;
2034 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2035 				   &update_params)) {
2036 			release_mac = new_smac;
2037 			goto unlock;
2038 		}
2039 		/* if old port was zero, no mac was yet registered for this QP */
2040 		if (qp->pri.smac_port)
2041 			release_mac = old_smac;
2042 		qp->pri.smac = new_smac;
2043 		qp->pri.smac_port = port;
2044 		qp->pri.smac_index = new_smac_index;
2045 	}
2046 
2047 unlock:
2048 	if (release_mac != MLX4_IB_INVALID_MAC)
2049 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2050 	if (qp)
2051 		mutex_unlock(&qp->mutex);
2052 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2053 }
2054 
2055 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2056 				 struct net_device *dev,
2057 				 unsigned long event)
2058 
2059 {
2060 	struct mlx4_ib_iboe *iboe;
2061 	int update_qps_port = -1;
2062 	int port;
2063 
2064 	ASSERT_RTNL();
2065 
2066 	iboe = &ibdev->iboe;
2067 
2068 	spin_lock_bh(&iboe->lock);
2069 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2070 
2071 		iboe->netdevs[port - 1] =
2072 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2073 
2074 		if (dev == iboe->netdevs[port - 1] &&
2075 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2076 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2077 			update_qps_port = port;
2078 
2079 	}
2080 	spin_unlock_bh(&iboe->lock);
2081 
2082 	if (update_qps_port > 0)
2083 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2084 }
2085 
2086 static int mlx4_ib_netdev_event(struct notifier_block *this,
2087 				unsigned long event, void *ptr)
2088 {
2089 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2090 	struct mlx4_ib_dev *ibdev;
2091 
2092 	if (!net_eq(dev_net(dev), &init_net))
2093 		return NOTIFY_DONE;
2094 
2095 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2096 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2097 
2098 	return NOTIFY_DONE;
2099 }
2100 
2101 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2102 {
2103 	int port;
2104 	int slave;
2105 	int i;
2106 
2107 	if (mlx4_is_master(ibdev->dev)) {
2108 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2109 		     ++slave) {
2110 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2111 				for (i = 0;
2112 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2113 				     ++i) {
2114 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2115 					/* master has the identity virt2phys pkey mapping */
2116 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2117 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2118 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2119 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2120 				}
2121 			}
2122 		}
2123 		/* initialize pkey cache */
2124 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2125 			for (i = 0;
2126 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2127 			     ++i)
2128 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2129 					(i) ? 0 : 0xFFFF;
2130 		}
2131 	}
2132 }
2133 
2134 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2135 {
2136 	int i, j, eq = 0, total_eqs = 0;
2137 
2138 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2139 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2140 	if (!ibdev->eq_table)
2141 		return;
2142 
2143 	for (i = 1; i <= dev->caps.num_ports; i++) {
2144 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2145 		     j++, total_eqs++) {
2146 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2147 				continue;
2148 			ibdev->eq_table[eq] = total_eqs;
2149 			if (!mlx4_assign_eq(dev, i,
2150 					    &ibdev->eq_table[eq]))
2151 				eq++;
2152 			else
2153 				ibdev->eq_table[eq] = -1;
2154 		}
2155 	}
2156 
2157 	for (i = eq; i < dev->caps.num_comp_vectors;
2158 	     ibdev->eq_table[i++] = -1)
2159 		;
2160 
2161 	/* Advertise the new number of EQs to clients */
2162 	ibdev->ib_dev.num_comp_vectors = eq;
2163 }
2164 
2165 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2166 {
2167 	int i;
2168 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2169 
2170 	/* no eqs were allocated */
2171 	if (!ibdev->eq_table)
2172 		return;
2173 
2174 	/* Reset the advertised EQ number */
2175 	ibdev->ib_dev.num_comp_vectors = 0;
2176 
2177 	for (i = 0; i < total_eqs; i++)
2178 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2179 
2180 	kfree(ibdev->eq_table);
2181 	ibdev->eq_table = NULL;
2182 }
2183 
2184 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2185 			       struct ib_port_immutable *immutable)
2186 {
2187 	struct ib_port_attr attr;
2188 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2189 	int err;
2190 
2191 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2192 	if (err)
2193 		return err;
2194 
2195 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2196 	immutable->gid_tbl_len = attr.gid_tbl_len;
2197 
2198 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2199 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2200 	} else {
2201 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2202 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2203 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2204 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2205 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2206 	}
2207 
2208 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2209 
2210 	return 0;
2211 }
2212 
2213 static void *mlx4_ib_add(struct mlx4_dev *dev)
2214 {
2215 	struct mlx4_ib_dev *ibdev;
2216 	int num_ports = 0;
2217 	int i, j;
2218 	int err;
2219 	struct mlx4_ib_iboe *iboe;
2220 	int ib_num_ports = 0;
2221 	int num_req_counters;
2222 	int allocated;
2223 	u32 counter_index;
2224 	struct counter_index *new_counter_index = NULL;
2225 
2226 	pr_info_once("%s", mlx4_ib_version);
2227 
2228 	num_ports = 0;
2229 	mlx4_foreach_ib_transport_port(i, dev)
2230 		num_ports++;
2231 
2232 	/* No point in registering a device with no ports... */
2233 	if (num_ports == 0)
2234 		return NULL;
2235 
2236 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2237 	if (!ibdev) {
2238 		dev_err(&dev->persist->pdev->dev,
2239 			"Device struct alloc failed\n");
2240 		return NULL;
2241 	}
2242 
2243 	iboe = &ibdev->iboe;
2244 
2245 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2246 		goto err_dealloc;
2247 
2248 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2249 		goto err_pd;
2250 
2251 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2252 				 PAGE_SIZE);
2253 	if (!ibdev->uar_map)
2254 		goto err_uar;
2255 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2256 
2257 	ibdev->dev = dev;
2258 	ibdev->bond_next_port	= 0;
2259 
2260 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2261 	ibdev->ib_dev.owner		= THIS_MODULE;
2262 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2263 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2264 	ibdev->num_ports		= num_ports;
2265 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2266 						1 : ibdev->num_ports;
2267 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2268 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2269 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2270 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2271 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2272 
2273 	if (dev->caps.userspace_caps)
2274 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2275 	else
2276 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2277 
2278 	ibdev->ib_dev.uverbs_cmd_mask	=
2279 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2280 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2281 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2282 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2283 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2284 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2285 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2286 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2287 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2288 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2289 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2290 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2291 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2292 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2293 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2294 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2295 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2296 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2297 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2298 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2299 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2300 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2301 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2302 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2303 
2304 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2305 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2306 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2307 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2308 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2309 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2310 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2311 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2312 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2313 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2314 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2315 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2316 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2317 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2318 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2319 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2320 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2321 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2322 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2323 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2324 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2325 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2326 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2327 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2328 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2329 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2330 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2331 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2332 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2333 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2334 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2335 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2336 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2337 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2338 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2339 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2340 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2341 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2342 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2343 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2344 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2345 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2346 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2347 
2348 	if (!mlx4_is_slave(ibdev->dev)) {
2349 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2350 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2351 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2352 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2353 	}
2354 
2355 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2356 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2357 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2358 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2359 
2360 		ibdev->ib_dev.uverbs_cmd_mask |=
2361 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2362 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2363 	}
2364 
2365 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2366 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2367 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2368 		ibdev->ib_dev.uverbs_cmd_mask |=
2369 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2370 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2371 	}
2372 
2373 	if (check_flow_steering_support(dev)) {
2374 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2375 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2376 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2377 
2378 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2379 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2380 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2381 	}
2382 
2383 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2384 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2385 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2386 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2387 
2388 	mlx4_ib_alloc_eqs(dev, ibdev);
2389 
2390 	spin_lock_init(&iboe->lock);
2391 
2392 	if (init_node_data(ibdev))
2393 		goto err_map;
2394 
2395 	for (i = 0; i < ibdev->num_ports; ++i) {
2396 		mutex_init(&ibdev->counters_table[i].mutex);
2397 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2398 	}
2399 
2400 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2401 	for (i = 0; i < num_req_counters; ++i) {
2402 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2403 		allocated = 0;
2404 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2405 						IB_LINK_LAYER_ETHERNET) {
2406 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2407 			/* if failed to allocate a new counter, use default */
2408 			if (err)
2409 				counter_index =
2410 					mlx4_get_default_counter_index(dev,
2411 								       i + 1);
2412 			else
2413 				allocated = 1;
2414 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2415 			counter_index = mlx4_get_default_counter_index(dev,
2416 								       i + 1);
2417 		}
2418 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2419 					    GFP_KERNEL);
2420 		if (!new_counter_index) {
2421 			if (allocated)
2422 				mlx4_counter_free(ibdev->dev, counter_index);
2423 			goto err_counter;
2424 		}
2425 		new_counter_index->index = counter_index;
2426 		new_counter_index->allocated = allocated;
2427 		list_add_tail(&new_counter_index->list,
2428 			      &ibdev->counters_table[i].counters_list);
2429 		ibdev->counters_table[i].default_counter = counter_index;
2430 		pr_info("counter index %d for port %d allocated %d\n",
2431 			counter_index, i + 1, allocated);
2432 	}
2433 	if (mlx4_is_bonded(dev))
2434 		for (i = 1; i < ibdev->num_ports ; ++i) {
2435 			new_counter_index =
2436 					kmalloc(sizeof(struct counter_index),
2437 						GFP_KERNEL);
2438 			if (!new_counter_index)
2439 				goto err_counter;
2440 			new_counter_index->index = counter_index;
2441 			new_counter_index->allocated = 0;
2442 			list_add_tail(&new_counter_index->list,
2443 				      &ibdev->counters_table[i].counters_list);
2444 			ibdev->counters_table[i].default_counter =
2445 								counter_index;
2446 		}
2447 
2448 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2449 		ib_num_ports++;
2450 
2451 	spin_lock_init(&ibdev->sm_lock);
2452 	mutex_init(&ibdev->cap_mask_mutex);
2453 	INIT_LIST_HEAD(&ibdev->qp_list);
2454 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2455 
2456 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2457 	    ib_num_ports) {
2458 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2459 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2460 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2461 					    &ibdev->steer_qpn_base, 0);
2462 		if (err)
2463 			goto err_counter;
2464 
2465 		ibdev->ib_uc_qpns_bitmap =
2466 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2467 				sizeof(long),
2468 				GFP_KERNEL);
2469 		if (!ibdev->ib_uc_qpns_bitmap) {
2470 			dev_err(&dev->persist->pdev->dev,
2471 				"bit map alloc failed\n");
2472 			goto err_steer_qp_release;
2473 		}
2474 
2475 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2476 
2477 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2478 				dev, ibdev->steer_qpn_base,
2479 				ibdev->steer_qpn_base +
2480 				ibdev->steer_qpn_count - 1);
2481 		if (err)
2482 			goto err_steer_free_bitmap;
2483 	}
2484 
2485 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2486 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2487 
2488 	if (ib_register_device(&ibdev->ib_dev, NULL))
2489 		goto err_steer_free_bitmap;
2490 
2491 	if (mlx4_ib_mad_init(ibdev))
2492 		goto err_reg;
2493 
2494 	if (mlx4_ib_init_sriov(ibdev))
2495 		goto err_mad;
2496 
2497 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2498 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2499 		if (!iboe->nb.notifier_call) {
2500 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2501 			err = register_netdevice_notifier(&iboe->nb);
2502 			if (err) {
2503 				iboe->nb.notifier_call = NULL;
2504 				goto err_notif;
2505 			}
2506 		}
2507 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2508 			err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2509 			if (err) {
2510 				goto err_notif;
2511 			}
2512 		}
2513 	}
2514 
2515 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2516 		if (device_create_file(&ibdev->ib_dev.dev,
2517 				       mlx4_class_attributes[j]))
2518 			goto err_notif;
2519 	}
2520 
2521 	ibdev->ib_active = true;
2522 
2523 	if (mlx4_is_mfunc(ibdev->dev))
2524 		init_pkeys(ibdev);
2525 
2526 	/* create paravirt contexts for any VFs which are active */
2527 	if (mlx4_is_master(ibdev->dev)) {
2528 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2529 			if (j == mlx4_master_func_num(ibdev->dev))
2530 				continue;
2531 			if (mlx4_is_slave_active(ibdev->dev, j))
2532 				do_slave_init(ibdev, j, 1);
2533 		}
2534 	}
2535 	return ibdev;
2536 
2537 err_notif:
2538 	if (ibdev->iboe.nb.notifier_call) {
2539 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2540 			pr_warn("failure unregistering notifier\n");
2541 		ibdev->iboe.nb.notifier_call = NULL;
2542 	}
2543 	flush_workqueue(wq);
2544 
2545 	mlx4_ib_close_sriov(ibdev);
2546 
2547 err_mad:
2548 	mlx4_ib_mad_cleanup(ibdev);
2549 
2550 err_reg:
2551 	ib_unregister_device(&ibdev->ib_dev);
2552 
2553 err_steer_free_bitmap:
2554 	kfree(ibdev->ib_uc_qpns_bitmap);
2555 
2556 err_steer_qp_release:
2557 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2558 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2559 				      ibdev->steer_qpn_count);
2560 err_counter:
2561 	for (i = 0; i < ibdev->num_ports; ++i)
2562 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2563 
2564 err_map:
2565 	iounmap(ibdev->uar_map);
2566 
2567 err_uar:
2568 	mlx4_uar_free(dev, &ibdev->priv_uar);
2569 
2570 err_pd:
2571 	mlx4_pd_free(dev, ibdev->priv_pdn);
2572 
2573 err_dealloc:
2574 	ib_dealloc_device(&ibdev->ib_dev);
2575 
2576 	return NULL;
2577 }
2578 
2579 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2580 {
2581 	int offset;
2582 
2583 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2584 
2585 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2586 					 dev->steer_qpn_count,
2587 					 get_count_order(count));
2588 	if (offset < 0)
2589 		return offset;
2590 
2591 	*qpn = dev->steer_qpn_base + offset;
2592 	return 0;
2593 }
2594 
2595 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2596 {
2597 	if (!qpn ||
2598 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2599 		return;
2600 
2601 	BUG_ON(qpn < dev->steer_qpn_base);
2602 
2603 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2604 			      qpn - dev->steer_qpn_base,
2605 			      get_count_order(count));
2606 }
2607 
2608 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2609 			 int is_attach)
2610 {
2611 	int err;
2612 	size_t flow_size;
2613 	struct ib_flow_attr *flow = NULL;
2614 	struct ib_flow_spec_ib *ib_spec;
2615 
2616 	if (is_attach) {
2617 		flow_size = sizeof(struct ib_flow_attr) +
2618 			    sizeof(struct ib_flow_spec_ib);
2619 		flow = kzalloc(flow_size, GFP_KERNEL);
2620 		if (!flow)
2621 			return -ENOMEM;
2622 		flow->port = mqp->port;
2623 		flow->num_of_specs = 1;
2624 		flow->size = flow_size;
2625 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2626 		ib_spec->type = IB_FLOW_SPEC_IB;
2627 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2628 		/* Add an empty rule for IB L2 */
2629 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2630 
2631 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2632 					    IB_FLOW_DOMAIN_NIC,
2633 					    MLX4_FS_REGULAR,
2634 					    &mqp->reg_id);
2635 	} else {
2636 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2637 	}
2638 	kfree(flow);
2639 	return err;
2640 }
2641 
2642 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2643 {
2644 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2645 	int p;
2646 
2647 	ibdev->ib_active = false;
2648 	flush_workqueue(wq);
2649 
2650 	mlx4_ib_close_sriov(ibdev);
2651 	mlx4_ib_mad_cleanup(ibdev);
2652 	ib_unregister_device(&ibdev->ib_dev);
2653 	if (ibdev->iboe.nb.notifier_call) {
2654 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2655 			pr_warn("failure unregistering notifier\n");
2656 		ibdev->iboe.nb.notifier_call = NULL;
2657 	}
2658 
2659 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2660 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2661 				      ibdev->steer_qpn_count);
2662 		kfree(ibdev->ib_uc_qpns_bitmap);
2663 	}
2664 
2665 	iounmap(ibdev->uar_map);
2666 	for (p = 0; p < ibdev->num_ports; ++p)
2667 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2668 
2669 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2670 		mlx4_CLOSE_PORT(dev, p);
2671 
2672 	mlx4_ib_free_eqs(dev, ibdev);
2673 
2674 	mlx4_uar_free(dev, &ibdev->priv_uar);
2675 	mlx4_pd_free(dev, ibdev->priv_pdn);
2676 	ib_dealloc_device(&ibdev->ib_dev);
2677 }
2678 
2679 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2680 {
2681 	struct mlx4_ib_demux_work **dm = NULL;
2682 	struct mlx4_dev *dev = ibdev->dev;
2683 	int i;
2684 	unsigned long flags;
2685 	struct mlx4_active_ports actv_ports;
2686 	unsigned int ports;
2687 	unsigned int first_port;
2688 
2689 	if (!mlx4_is_master(dev))
2690 		return;
2691 
2692 	actv_ports = mlx4_get_active_ports(dev, slave);
2693 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2694 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2695 
2696 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2697 	if (!dm) {
2698 		pr_err("failed to allocate memory for tunneling qp update\n");
2699 		return;
2700 	}
2701 
2702 	for (i = 0; i < ports; i++) {
2703 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2704 		if (!dm[i]) {
2705 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2706 			while (--i >= 0)
2707 				kfree(dm[i]);
2708 			goto out;
2709 		}
2710 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2711 		dm[i]->port = first_port + i + 1;
2712 		dm[i]->slave = slave;
2713 		dm[i]->do_init = do_init;
2714 		dm[i]->dev = ibdev;
2715 	}
2716 	/* initialize or tear down tunnel QPs for the slave */
2717 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2718 	if (!ibdev->sriov.is_going_down) {
2719 		for (i = 0; i < ports; i++)
2720 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2721 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2722 	} else {
2723 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2724 		for (i = 0; i < ports; i++)
2725 			kfree(dm[i]);
2726 	}
2727 out:
2728 	kfree(dm);
2729 	return;
2730 }
2731 
2732 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2733 {
2734 	struct mlx4_ib_qp *mqp;
2735 	unsigned long flags_qp;
2736 	unsigned long flags_cq;
2737 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2738 	struct list_head    cq_notify_list;
2739 	struct mlx4_cq *mcq;
2740 	unsigned long flags;
2741 
2742 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2743 	INIT_LIST_HEAD(&cq_notify_list);
2744 
2745 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2746 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2747 
2748 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2749 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2750 		if (mqp->sq.tail != mqp->sq.head) {
2751 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2752 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2753 			if (send_mcq->mcq.comp &&
2754 			    mqp->ibqp.send_cq->comp_handler) {
2755 				if (!send_mcq->mcq.reset_notify_added) {
2756 					send_mcq->mcq.reset_notify_added = 1;
2757 					list_add_tail(&send_mcq->mcq.reset_notify,
2758 						      &cq_notify_list);
2759 				}
2760 			}
2761 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2762 		}
2763 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2764 		/* Now, handle the QP's receive queue */
2765 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2766 		/* no handling is needed for SRQ */
2767 		if (!mqp->ibqp.srq) {
2768 			if (mqp->rq.tail != mqp->rq.head) {
2769 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2770 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2771 				if (recv_mcq->mcq.comp &&
2772 				    mqp->ibqp.recv_cq->comp_handler) {
2773 					if (!recv_mcq->mcq.reset_notify_added) {
2774 						recv_mcq->mcq.reset_notify_added = 1;
2775 						list_add_tail(&recv_mcq->mcq.reset_notify,
2776 							      &cq_notify_list);
2777 					}
2778 				}
2779 				spin_unlock_irqrestore(&recv_mcq->lock,
2780 						       flags_cq);
2781 			}
2782 		}
2783 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2784 	}
2785 
2786 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2787 		mcq->comp(mcq);
2788 	}
2789 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2790 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2791 }
2792 
2793 static void handle_bonded_port_state_event(struct work_struct *work)
2794 {
2795 	struct ib_event_work *ew =
2796 		container_of(work, struct ib_event_work, work);
2797 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2798 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2799 	int i;
2800 	struct ib_event ibev;
2801 
2802 	kfree(ew);
2803 	spin_lock_bh(&ibdev->iboe.lock);
2804 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2805 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2806 		enum ib_port_state curr_port_state;
2807 
2808 		if (!curr_netdev)
2809 			continue;
2810 
2811 		curr_port_state =
2812 			(netif_running(curr_netdev) &&
2813 			 netif_carrier_ok(curr_netdev)) ?
2814 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2815 
2816 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2817 			curr_port_state : IB_PORT_ACTIVE;
2818 	}
2819 	spin_unlock_bh(&ibdev->iboe.lock);
2820 
2821 	ibev.device = &ibdev->ib_dev;
2822 	ibev.element.port_num = 1;
2823 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2824 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2825 
2826 	ib_dispatch_event(&ibev);
2827 }
2828 
2829 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2830 			  enum mlx4_dev_event event, unsigned long param)
2831 {
2832 	struct ib_event ibev;
2833 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2834 	struct mlx4_eqe *eqe = NULL;
2835 	struct ib_event_work *ew;
2836 	int p = 0;
2837 
2838 	if (mlx4_is_bonded(dev) &&
2839 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2840 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2841 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2842 		if (!ew)
2843 			return;
2844 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2845 		ew->ib_dev = ibdev;
2846 		queue_work(wq, &ew->work);
2847 		return;
2848 	}
2849 
2850 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2851 		eqe = (struct mlx4_eqe *)param;
2852 	else
2853 		p = (int) param;
2854 
2855 	switch (event) {
2856 	case MLX4_DEV_EVENT_PORT_UP:
2857 		if (p > ibdev->num_ports)
2858 			return;
2859 		if (mlx4_is_master(dev) &&
2860 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2861 			IB_LINK_LAYER_INFINIBAND) {
2862 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2863 		}
2864 		ibev.event = IB_EVENT_PORT_ACTIVE;
2865 		break;
2866 
2867 	case MLX4_DEV_EVENT_PORT_DOWN:
2868 		if (p > ibdev->num_ports)
2869 			return;
2870 		ibev.event = IB_EVENT_PORT_ERR;
2871 		break;
2872 
2873 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2874 		ibdev->ib_active = false;
2875 		ibev.event = IB_EVENT_DEVICE_FATAL;
2876 		mlx4_ib_handle_catas_error(ibdev);
2877 		break;
2878 
2879 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2880 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2881 		if (!ew) {
2882 			pr_err("failed to allocate memory for events work\n");
2883 			break;
2884 		}
2885 
2886 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2887 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2888 		ew->ib_dev = ibdev;
2889 		/* need to queue only for port owner, which uses GEN_EQE */
2890 		if (mlx4_is_master(dev))
2891 			queue_work(wq, &ew->work);
2892 		else
2893 			handle_port_mgmt_change_event(&ew->work);
2894 		return;
2895 
2896 	case MLX4_DEV_EVENT_SLAVE_INIT:
2897 		/* here, p is the slave id */
2898 		do_slave_init(ibdev, p, 1);
2899 		if (mlx4_is_master(dev)) {
2900 			int i;
2901 
2902 			for (i = 1; i <= ibdev->num_ports; i++) {
2903 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2904 					== IB_LINK_LAYER_INFINIBAND)
2905 					mlx4_ib_slave_alias_guid_event(ibdev,
2906 								       p, i,
2907 								       1);
2908 			}
2909 		}
2910 		return;
2911 
2912 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2913 		if (mlx4_is_master(dev)) {
2914 			int i;
2915 
2916 			for (i = 1; i <= ibdev->num_ports; i++) {
2917 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2918 					== IB_LINK_LAYER_INFINIBAND)
2919 					mlx4_ib_slave_alias_guid_event(ibdev,
2920 								       p, i,
2921 								       0);
2922 			}
2923 		}
2924 		/* here, p is the slave id */
2925 		do_slave_init(ibdev, p, 0);
2926 		return;
2927 
2928 	default:
2929 		return;
2930 	}
2931 
2932 	ibev.device	      = ibdev_ptr;
2933 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2934 
2935 	ib_dispatch_event(&ibev);
2936 }
2937 
2938 static struct mlx4_interface mlx4_ib_interface = {
2939 	.add		= mlx4_ib_add,
2940 	.remove		= mlx4_ib_remove,
2941 	.event		= mlx4_ib_event,
2942 	.protocol	= MLX4_PROT_IB_IPV6,
2943 	.flags		= MLX4_INTFF_BONDING
2944 };
2945 
2946 static int __init mlx4_ib_init(void)
2947 {
2948 	int err;
2949 
2950 	wq = create_singlethread_workqueue("mlx4_ib");
2951 	if (!wq)
2952 		return -ENOMEM;
2953 
2954 	err = mlx4_ib_mcg_init();
2955 	if (err)
2956 		goto clean_wq;
2957 
2958 	err = mlx4_register_interface(&mlx4_ib_interface);
2959 	if (err)
2960 		goto clean_mcg;
2961 
2962 	return 0;
2963 
2964 clean_mcg:
2965 	mlx4_ib_mcg_destroy();
2966 
2967 clean_wq:
2968 	destroy_workqueue(wq);
2969 	return err;
2970 }
2971 
2972 static void __exit mlx4_ib_cleanup(void)
2973 {
2974 	mlx4_unregister_interface(&mlx4_ib_interface);
2975 	mlx4_ib_mcg_destroy();
2976 	destroy_workqueue(wq);
2977 }
2978 
2979 module_init(mlx4_ib_init);
2980 module_exit(mlx4_ib_cleanup);
2981