xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision b240b419db5d624ce7a5a397d6f62a1a686009ec)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <linux/sched/mm.h>
43 #include <linux/sched/task.h>
44 
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/devlink.h>
48 
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53 
54 #include <net/bonding.h>
55 
56 #include <linux/mlx4/driver.h>
57 #include <linux/mlx4/cmd.h>
58 #include <linux/mlx4/qp.h>
59 
60 #include "mlx4_ib.h"
61 #include <rdma/mlx4-abi.h>
62 
63 #define DRV_NAME	MLX4_IB_DRV_NAME
64 #define DRV_VERSION	"4.0-0"
65 
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 #define MLX4_IB_CARD_REV_A0   0xA0
69 
70 MODULE_AUTHOR("Roland Dreier");
71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 
74 int mlx4_ib_sm_guid_assign = 0;
75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
77 
78 static const char mlx4_ib_version[] =
79 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
80 	DRV_VERSION "\n";
81 
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
84 						    u8 port_num);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static int check_flow_steering_support(struct mlx4_dev *dev)
97 {
98 	int eth_num_ports = 0;
99 	int ib_num_ports = 0;
100 
101 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
102 
103 	if (dmfs) {
104 		int i;
105 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
106 			eth_num_ports++;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
108 			ib_num_ports++;
109 		dmfs &= (!ib_num_ports ||
110 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
111 			(!eth_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
115 			dmfs = 0;
116 		}
117 	}
118 	return dmfs;
119 }
120 
121 static int num_ib_ports(struct mlx4_dev *dev)
122 {
123 	int ib_ports = 0;
124 	int i;
125 
126 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
127 		ib_ports++;
128 
129 	return ib_ports;
130 }
131 
132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
133 {
134 	struct mlx4_ib_dev *ibdev = to_mdev(device);
135 	struct net_device *dev;
136 
137 	rcu_read_lock();
138 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
139 
140 	if (dev) {
141 		if (mlx4_is_bonded(ibdev->dev)) {
142 			struct net_device *upper = NULL;
143 
144 			upper = netdev_master_upper_dev_get_rcu(dev);
145 			if (upper) {
146 				struct net_device *active;
147 
148 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
149 				if (active)
150 					dev = active;
151 			}
152 		}
153 	}
154 	if (dev)
155 		dev_hold(dev);
156 
157 	rcu_read_unlock();
158 	return dev;
159 }
160 
161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
162 				  struct mlx4_ib_dev *ibdev,
163 				  u8 port_num)
164 {
165 	struct mlx4_cmd_mailbox *mailbox;
166 	int err;
167 	struct mlx4_dev *dev = ibdev->dev;
168 	int i;
169 	union ib_gid *gid_tbl;
170 
171 	mailbox = mlx4_alloc_cmd_mailbox(dev);
172 	if (IS_ERR(mailbox))
173 		return -ENOMEM;
174 
175 	gid_tbl = mailbox->buf;
176 
177 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
178 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
179 
180 	err = mlx4_cmd(dev, mailbox->dma,
181 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
182 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
183 		       MLX4_CMD_WRAPPED);
184 	if (mlx4_is_bonded(dev))
185 		err += mlx4_cmd(dev, mailbox->dma,
186 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
187 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
188 				MLX4_CMD_WRAPPED);
189 
190 	mlx4_free_cmd_mailbox(dev, mailbox);
191 	return err;
192 }
193 
194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
195 				     struct mlx4_ib_dev *ibdev,
196 				     u8 port_num)
197 {
198 	struct mlx4_cmd_mailbox *mailbox;
199 	int err;
200 	struct mlx4_dev *dev = ibdev->dev;
201 	int i;
202 	struct {
203 		union ib_gid	gid;
204 		__be32		rsrvd1[2];
205 		__be16		rsrvd2;
206 		u8		type;
207 		u8		version;
208 		__be32		rsrvd3;
209 	} *gid_tbl;
210 
211 	mailbox = mlx4_alloc_cmd_mailbox(dev);
212 	if (IS_ERR(mailbox))
213 		return -ENOMEM;
214 
215 	gid_tbl = mailbox->buf;
216 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
217 		memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
218 		if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
219 			gid_tbl[i].version = 2;
220 			if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
221 				gid_tbl[i].type = 1;
222 		}
223 	}
224 
225 	err = mlx4_cmd(dev, mailbox->dma,
226 		       MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
227 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
228 		       MLX4_CMD_WRAPPED);
229 	if (mlx4_is_bonded(dev))
230 		err += mlx4_cmd(dev, mailbox->dma,
231 				MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
232 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
233 				MLX4_CMD_WRAPPED);
234 
235 	mlx4_free_cmd_mailbox(dev, mailbox);
236 	return err;
237 }
238 
239 static int mlx4_ib_update_gids(struct gid_entry *gids,
240 			       struct mlx4_ib_dev *ibdev,
241 			       u8 port_num)
242 {
243 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
244 		return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
245 
246 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
247 }
248 
249 static int mlx4_ib_add_gid(struct ib_device *device,
250 			   u8 port_num,
251 			   unsigned int index,
252 			   const union ib_gid *gid,
253 			   const struct ib_gid_attr *attr,
254 			   void **context)
255 {
256 	struct mlx4_ib_dev *ibdev = to_mdev(device);
257 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
258 	struct mlx4_port_gid_table   *port_gid_table;
259 	int free = -1, found = -1;
260 	int ret = 0;
261 	int hw_update = 0;
262 	int i;
263 	struct gid_entry *gids = NULL;
264 
265 	if (!rdma_cap_roce_gid_table(device, port_num))
266 		return -EINVAL;
267 
268 	if (port_num > MLX4_MAX_PORTS)
269 		return -EINVAL;
270 
271 	if (!context)
272 		return -EINVAL;
273 
274 	port_gid_table = &iboe->gids[port_num - 1];
275 	spin_lock_bh(&iboe->lock);
276 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
277 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
278 		    (port_gid_table->gids[i].gid_type == attr->gid_type))  {
279 			found = i;
280 			break;
281 		}
282 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
283 			free = i; /* HW has space */
284 	}
285 
286 	if (found < 0) {
287 		if (free < 0) {
288 			ret = -ENOSPC;
289 		} else {
290 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
291 			if (!port_gid_table->gids[free].ctx) {
292 				ret = -ENOMEM;
293 			} else {
294 				*context = port_gid_table->gids[free].ctx;
295 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
296 				port_gid_table->gids[free].gid_type = attr->gid_type;
297 				port_gid_table->gids[free].ctx->real_index = free;
298 				port_gid_table->gids[free].ctx->refcount = 1;
299 				hw_update = 1;
300 			}
301 		}
302 	} else {
303 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
304 		*context = ctx;
305 		ctx->refcount++;
306 	}
307 	if (!ret && hw_update) {
308 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
309 		if (!gids) {
310 			ret = -ENOMEM;
311 		} else {
312 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
313 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
314 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
315 			}
316 		}
317 	}
318 	spin_unlock_bh(&iboe->lock);
319 
320 	if (!ret && hw_update) {
321 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
322 		kfree(gids);
323 	}
324 
325 	return ret;
326 }
327 
328 static int mlx4_ib_del_gid(struct ib_device *device,
329 			   u8 port_num,
330 			   unsigned int index,
331 			   void **context)
332 {
333 	struct gid_cache_context *ctx = *context;
334 	struct mlx4_ib_dev *ibdev = to_mdev(device);
335 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
336 	struct mlx4_port_gid_table   *port_gid_table;
337 	int ret = 0;
338 	int hw_update = 0;
339 	struct gid_entry *gids = NULL;
340 
341 	if (!rdma_cap_roce_gid_table(device, port_num))
342 		return -EINVAL;
343 
344 	if (port_num > MLX4_MAX_PORTS)
345 		return -EINVAL;
346 
347 	port_gid_table = &iboe->gids[port_num - 1];
348 	spin_lock_bh(&iboe->lock);
349 	if (ctx) {
350 		ctx->refcount--;
351 		if (!ctx->refcount) {
352 			unsigned int real_index = ctx->real_index;
353 
354 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
355 			kfree(port_gid_table->gids[real_index].ctx);
356 			port_gid_table->gids[real_index].ctx = NULL;
357 			hw_update = 1;
358 		}
359 	}
360 	if (!ret && hw_update) {
361 		int i;
362 
363 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
364 		if (!gids) {
365 			ret = -ENOMEM;
366 		} else {
367 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
368 				memcpy(&gids[i].gid,
369 				       &port_gid_table->gids[i].gid,
370 				       sizeof(union ib_gid));
371 				gids[i].gid_type =
372 				    port_gid_table->gids[i].gid_type;
373 			}
374 		}
375 	}
376 	spin_unlock_bh(&iboe->lock);
377 
378 	if (!ret && hw_update) {
379 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
380 		kfree(gids);
381 	}
382 	return ret;
383 }
384 
385 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
386 				    u8 port_num, int index)
387 {
388 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
389 	struct gid_cache_context *ctx = NULL;
390 	union ib_gid gid;
391 	struct mlx4_port_gid_table   *port_gid_table;
392 	int real_index = -EINVAL;
393 	int i;
394 	int ret;
395 	unsigned long flags;
396 	struct ib_gid_attr attr;
397 
398 	if (port_num > MLX4_MAX_PORTS)
399 		return -EINVAL;
400 
401 	if (mlx4_is_bonded(ibdev->dev))
402 		port_num = 1;
403 
404 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
405 		return index;
406 
407 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
408 	if (ret)
409 		return ret;
410 
411 	if (attr.ndev)
412 		dev_put(attr.ndev);
413 
414 	if (!memcmp(&gid, &zgid, sizeof(gid)))
415 		return -EINVAL;
416 
417 	spin_lock_irqsave(&iboe->lock, flags);
418 	port_gid_table = &iboe->gids[port_num - 1];
419 
420 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
421 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
422 		    attr.gid_type == port_gid_table->gids[i].gid_type) {
423 			ctx = port_gid_table->gids[i].ctx;
424 			break;
425 		}
426 	if (ctx)
427 		real_index = ctx->real_index;
428 	spin_unlock_irqrestore(&iboe->lock, flags);
429 	return real_index;
430 }
431 
432 static int mlx4_ib_query_device(struct ib_device *ibdev,
433 				struct ib_device_attr *props,
434 				struct ib_udata *uhw)
435 {
436 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
437 	struct ib_smp *in_mad  = NULL;
438 	struct ib_smp *out_mad = NULL;
439 	int err;
440 	int have_ib_ports;
441 	struct mlx4_uverbs_ex_query_device cmd;
442 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
443 	struct mlx4_clock_params clock_params;
444 
445 	if (uhw->inlen) {
446 		if (uhw->inlen < sizeof(cmd))
447 			return -EINVAL;
448 
449 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
450 		if (err)
451 			return err;
452 
453 		if (cmd.comp_mask)
454 			return -EINVAL;
455 
456 		if (cmd.reserved)
457 			return -EINVAL;
458 	}
459 
460 	resp.response_length = offsetof(typeof(resp), response_length) +
461 		sizeof(resp.response_length);
462 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
463 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
464 	err = -ENOMEM;
465 	if (!in_mad || !out_mad)
466 		goto out;
467 
468 	init_query_mad(in_mad);
469 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
470 
471 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
472 			   1, NULL, NULL, in_mad, out_mad);
473 	if (err)
474 		goto out;
475 
476 	memset(props, 0, sizeof *props);
477 
478 	have_ib_ports = num_ib_ports(dev->dev);
479 
480 	props->fw_ver = dev->dev->caps.fw_ver;
481 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
482 		IB_DEVICE_PORT_ACTIVE_EVENT		|
483 		IB_DEVICE_SYS_IMAGE_GUID		|
484 		IB_DEVICE_RC_RNR_NAK_GEN		|
485 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
486 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
487 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
488 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
489 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
490 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
491 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
492 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
493 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
494 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
495 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
496 	if (dev->dev->caps.max_gso_sz &&
497 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
498 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
499 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
500 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
501 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
502 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
503 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
504 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
505 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
506 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
507 		props->device_cap_flags |= IB_DEVICE_XRC;
508 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
509 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
510 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
511 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
512 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
513 		else
514 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
515 	}
516 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
517 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
518 
519 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
520 
521 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
522 		0xffffff;
523 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
524 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
525 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
526 
527 	props->max_mr_size	   = ~0ull;
528 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
529 	props->max_qp		   = dev->dev->quotas.qp;
530 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
531 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
532 					 dev->dev->caps.max_rq_sg);
533 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
534 	props->max_cq		   = dev->dev->quotas.cq;
535 	props->max_cqe		   = dev->dev->caps.max_cqes;
536 	props->max_mr		   = dev->dev->quotas.mpt;
537 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
538 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
539 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
540 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
541 	props->max_srq		   = dev->dev->quotas.srq;
542 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
543 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
544 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
545 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
546 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
547 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
548 	props->masked_atomic_cap   = props->atomic_cap;
549 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
550 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
551 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
552 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
553 					   props->max_mcast_grp;
554 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
555 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
556 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
557 	props->max_ah = INT_MAX;
558 
559 	if ((dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
560 	    (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
561 	     mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET)) {
562 		props->rss_caps.max_rwq_indirection_tables = props->max_qp;
563 		props->rss_caps.max_rwq_indirection_table_size =
564 			dev->dev->caps.max_rss_tbl_sz;
565 		props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
566 		props->max_wq_type_rq = props->max_qp;
567 	}
568 
569 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
570 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
571 
572 	if (!mlx4_is_slave(dev->dev))
573 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
574 
575 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
576 		resp.response_length += sizeof(resp.hca_core_clock_offset);
577 		if (!err && !mlx4_is_slave(dev->dev)) {
578 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
579 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
580 		}
581 	}
582 
583 	if (uhw->outlen >= resp.response_length +
584 	    sizeof(resp.max_inl_recv_sz)) {
585 		resp.response_length += sizeof(resp.max_inl_recv_sz);
586 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
587 			sizeof(struct mlx4_wqe_data_seg);
588 	}
589 
590 	if (uhw->outlen >= resp.response_length + sizeof(resp.rss_caps)) {
591 		resp.response_length += sizeof(resp.rss_caps);
592 		if (props->rss_caps.supported_qpts) {
593 			resp.rss_caps.rx_hash_function =
594 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
595 
596 			resp.rss_caps.rx_hash_fields_mask =
597 				MLX4_IB_RX_HASH_SRC_IPV4 |
598 				MLX4_IB_RX_HASH_DST_IPV4 |
599 				MLX4_IB_RX_HASH_SRC_IPV6 |
600 				MLX4_IB_RX_HASH_DST_IPV6 |
601 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
602 				MLX4_IB_RX_HASH_DST_PORT_TCP |
603 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
604 				MLX4_IB_RX_HASH_DST_PORT_UDP;
605 
606 			if (dev->dev->caps.tunnel_offload_mode ==
607 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
608 				resp.rss_caps.rx_hash_fields_mask |=
609 					MLX4_IB_RX_HASH_INNER;
610 		}
611 	}
612 
613 	if (uhw->outlen) {
614 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
615 		if (err)
616 			goto out;
617 	}
618 out:
619 	kfree(in_mad);
620 	kfree(out_mad);
621 
622 	return err;
623 }
624 
625 static enum rdma_link_layer
626 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
627 {
628 	struct mlx4_dev *dev = to_mdev(device)->dev;
629 
630 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
631 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
632 }
633 
634 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
635 			      struct ib_port_attr *props, int netw_view)
636 {
637 	struct ib_smp *in_mad  = NULL;
638 	struct ib_smp *out_mad = NULL;
639 	int ext_active_speed;
640 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
641 	int err = -ENOMEM;
642 
643 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
644 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
645 	if (!in_mad || !out_mad)
646 		goto out;
647 
648 	init_query_mad(in_mad);
649 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
650 	in_mad->attr_mod = cpu_to_be32(port);
651 
652 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
653 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
654 
655 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
656 				in_mad, out_mad);
657 	if (err)
658 		goto out;
659 
660 
661 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
662 	props->lmc		= out_mad->data[34] & 0x7;
663 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
664 	props->sm_sl		= out_mad->data[36] & 0xf;
665 	props->state		= out_mad->data[32] & 0xf;
666 	props->phys_state	= out_mad->data[33] >> 4;
667 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
668 	if (netw_view)
669 		props->gid_tbl_len = out_mad->data[50];
670 	else
671 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
672 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
673 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
674 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
675 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
676 	props->active_width	= out_mad->data[31] & 0xf;
677 	props->active_speed	= out_mad->data[35] >> 4;
678 	props->max_mtu		= out_mad->data[41] & 0xf;
679 	props->active_mtu	= out_mad->data[36] >> 4;
680 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
681 	props->max_vl_num	= out_mad->data[37] >> 4;
682 	props->init_type_reply	= out_mad->data[41] >> 4;
683 
684 	/* Check if extended speeds (EDR/FDR/...) are supported */
685 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
686 		ext_active_speed = out_mad->data[62] >> 4;
687 
688 		switch (ext_active_speed) {
689 		case 1:
690 			props->active_speed = IB_SPEED_FDR;
691 			break;
692 		case 2:
693 			props->active_speed = IB_SPEED_EDR;
694 			break;
695 		}
696 	}
697 
698 	/* If reported active speed is QDR, check if is FDR-10 */
699 	if (props->active_speed == IB_SPEED_QDR) {
700 		init_query_mad(in_mad);
701 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
702 		in_mad->attr_mod = cpu_to_be32(port);
703 
704 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
705 				   NULL, NULL, in_mad, out_mad);
706 		if (err)
707 			goto out;
708 
709 		/* Checking LinkSpeedActive for FDR-10 */
710 		if (out_mad->data[15] & 0x1)
711 			props->active_speed = IB_SPEED_FDR10;
712 	}
713 
714 	/* Avoid wrong speed value returned by FW if the IB link is down. */
715 	if (props->state == IB_PORT_DOWN)
716 		 props->active_speed = IB_SPEED_SDR;
717 
718 out:
719 	kfree(in_mad);
720 	kfree(out_mad);
721 	return err;
722 }
723 
724 static u8 state_to_phys_state(enum ib_port_state state)
725 {
726 	return state == IB_PORT_ACTIVE ? 5 : 3;
727 }
728 
729 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
730 			       struct ib_port_attr *props)
731 {
732 
733 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
734 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
735 	struct net_device *ndev;
736 	enum ib_mtu tmp;
737 	struct mlx4_cmd_mailbox *mailbox;
738 	int err = 0;
739 	int is_bonded = mlx4_is_bonded(mdev->dev);
740 
741 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
742 	if (IS_ERR(mailbox))
743 		return PTR_ERR(mailbox);
744 
745 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
746 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
747 			   MLX4_CMD_WRAPPED);
748 	if (err)
749 		goto out;
750 
751 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
752 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
753 					   IB_WIDTH_4X : IB_WIDTH_1X;
754 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
755 					   IB_SPEED_FDR : IB_SPEED_QDR;
756 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
757 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
758 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
759 	props->pkey_tbl_len	= 1;
760 	props->max_mtu		= IB_MTU_4096;
761 	props->max_vl_num	= 2;
762 	props->state		= IB_PORT_DOWN;
763 	props->phys_state	= state_to_phys_state(props->state);
764 	props->active_mtu	= IB_MTU_256;
765 	spin_lock_bh(&iboe->lock);
766 	ndev = iboe->netdevs[port - 1];
767 	if (ndev && is_bonded) {
768 		rcu_read_lock(); /* required to get upper dev */
769 		ndev = netdev_master_upper_dev_get_rcu(ndev);
770 		rcu_read_unlock();
771 	}
772 	if (!ndev)
773 		goto out_unlock;
774 
775 	tmp = iboe_get_mtu(ndev->mtu);
776 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
777 
778 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
779 					IB_PORT_ACTIVE : IB_PORT_DOWN;
780 	props->phys_state	= state_to_phys_state(props->state);
781 out_unlock:
782 	spin_unlock_bh(&iboe->lock);
783 out:
784 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
785 	return err;
786 }
787 
788 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
789 			 struct ib_port_attr *props, int netw_view)
790 {
791 	int err;
792 
793 	/* props being zeroed by the caller, avoid zeroing it here */
794 
795 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
796 		ib_link_query_port(ibdev, port, props, netw_view) :
797 				eth_link_query_port(ibdev, port, props);
798 
799 	return err;
800 }
801 
802 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
803 			      struct ib_port_attr *props)
804 {
805 	/* returns host view */
806 	return __mlx4_ib_query_port(ibdev, port, props, 0);
807 }
808 
809 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
810 			union ib_gid *gid, int netw_view)
811 {
812 	struct ib_smp *in_mad  = NULL;
813 	struct ib_smp *out_mad = NULL;
814 	int err = -ENOMEM;
815 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
816 	int clear = 0;
817 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
818 
819 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
820 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
821 	if (!in_mad || !out_mad)
822 		goto out;
823 
824 	init_query_mad(in_mad);
825 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
826 	in_mad->attr_mod = cpu_to_be32(port);
827 
828 	if (mlx4_is_mfunc(dev->dev) && netw_view)
829 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
830 
831 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
832 	if (err)
833 		goto out;
834 
835 	memcpy(gid->raw, out_mad->data + 8, 8);
836 
837 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
838 		if (index) {
839 			/* For any index > 0, return the null guid */
840 			err = 0;
841 			clear = 1;
842 			goto out;
843 		}
844 	}
845 
846 	init_query_mad(in_mad);
847 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
848 	in_mad->attr_mod = cpu_to_be32(index / 8);
849 
850 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
851 			   NULL, NULL, in_mad, out_mad);
852 	if (err)
853 		goto out;
854 
855 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
856 
857 out:
858 	if (clear)
859 		memset(gid->raw + 8, 0, 8);
860 	kfree(in_mad);
861 	kfree(out_mad);
862 	return err;
863 }
864 
865 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
866 			     union ib_gid *gid)
867 {
868 	int ret;
869 
870 	if (rdma_protocol_ib(ibdev, port))
871 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
872 
873 	if (!rdma_protocol_roce(ibdev, port))
874 		return -ENODEV;
875 
876 	if (!rdma_cap_roce_gid_table(ibdev, port))
877 		return -ENODEV;
878 
879 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
880 	if (ret == -EAGAIN) {
881 		memcpy(gid, &zgid, sizeof(*gid));
882 		return 0;
883 	}
884 
885 	return ret;
886 }
887 
888 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
889 {
890 	union sl2vl_tbl_to_u64 sl2vl64;
891 	struct ib_smp *in_mad  = NULL;
892 	struct ib_smp *out_mad = NULL;
893 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
894 	int err = -ENOMEM;
895 	int jj;
896 
897 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
898 		*sl2vl_tbl = 0;
899 		return 0;
900 	}
901 
902 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
903 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
904 	if (!in_mad || !out_mad)
905 		goto out;
906 
907 	init_query_mad(in_mad);
908 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
909 	in_mad->attr_mod = 0;
910 
911 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
912 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
913 
914 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
915 			   in_mad, out_mad);
916 	if (err)
917 		goto out;
918 
919 	for (jj = 0; jj < 8; jj++)
920 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
921 	*sl2vl_tbl = sl2vl64.sl64;
922 
923 out:
924 	kfree(in_mad);
925 	kfree(out_mad);
926 	return err;
927 }
928 
929 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
930 {
931 	u64 sl2vl;
932 	int i;
933 	int err;
934 
935 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
936 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
937 			continue;
938 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
939 		if (err) {
940 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
941 			       i, err);
942 			sl2vl = 0;
943 		}
944 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
945 	}
946 }
947 
948 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
949 			 u16 *pkey, int netw_view)
950 {
951 	struct ib_smp *in_mad  = NULL;
952 	struct ib_smp *out_mad = NULL;
953 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
954 	int err = -ENOMEM;
955 
956 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
957 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
958 	if (!in_mad || !out_mad)
959 		goto out;
960 
961 	init_query_mad(in_mad);
962 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
963 	in_mad->attr_mod = cpu_to_be32(index / 32);
964 
965 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
966 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
967 
968 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
969 			   in_mad, out_mad);
970 	if (err)
971 		goto out;
972 
973 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
974 
975 out:
976 	kfree(in_mad);
977 	kfree(out_mad);
978 	return err;
979 }
980 
981 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
982 {
983 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
984 }
985 
986 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
987 				 struct ib_device_modify *props)
988 {
989 	struct mlx4_cmd_mailbox *mailbox;
990 	unsigned long flags;
991 
992 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
993 		return -EOPNOTSUPP;
994 
995 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
996 		return 0;
997 
998 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
999 		return -EOPNOTSUPP;
1000 
1001 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1002 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1003 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1004 
1005 	/*
1006 	 * If possible, pass node desc to FW, so it can generate
1007 	 * a 144 trap.  If cmd fails, just ignore.
1008 	 */
1009 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1010 	if (IS_ERR(mailbox))
1011 		return 0;
1012 
1013 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1014 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1015 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1016 
1017 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1018 
1019 	return 0;
1020 }
1021 
1022 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1023 			    u32 cap_mask)
1024 {
1025 	struct mlx4_cmd_mailbox *mailbox;
1026 	int err;
1027 
1028 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1029 	if (IS_ERR(mailbox))
1030 		return PTR_ERR(mailbox);
1031 
1032 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1033 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1034 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1035 	} else {
1036 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1037 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1038 	}
1039 
1040 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1041 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1042 		       MLX4_CMD_WRAPPED);
1043 
1044 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1045 	return err;
1046 }
1047 
1048 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1049 			       struct ib_port_modify *props)
1050 {
1051 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1052 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1053 	struct ib_port_attr attr;
1054 	u32 cap_mask;
1055 	int err;
1056 
1057 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1058 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1059 	 * violations and port capabilities are not meaningful.
1060 	 */
1061 	if (is_eth)
1062 		return 0;
1063 
1064 	mutex_lock(&mdev->cap_mask_mutex);
1065 
1066 	err = ib_query_port(ibdev, port, &attr);
1067 	if (err)
1068 		goto out;
1069 
1070 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1071 		~props->clr_port_cap_mask;
1072 
1073 	err = mlx4_ib_SET_PORT(mdev, port,
1074 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1075 			       cap_mask);
1076 
1077 out:
1078 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1079 	return err;
1080 }
1081 
1082 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1083 						  struct ib_udata *udata)
1084 {
1085 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1086 	struct mlx4_ib_ucontext *context;
1087 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1088 	struct mlx4_ib_alloc_ucontext_resp resp;
1089 	int err;
1090 
1091 	if (!dev->ib_active)
1092 		return ERR_PTR(-EAGAIN);
1093 
1094 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1095 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1096 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1097 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1098 	} else {
1099 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1100 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1101 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1102 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1103 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1104 	}
1105 
1106 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1107 	if (!context)
1108 		return ERR_PTR(-ENOMEM);
1109 
1110 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1111 	if (err) {
1112 		kfree(context);
1113 		return ERR_PTR(err);
1114 	}
1115 
1116 	INIT_LIST_HEAD(&context->db_page_list);
1117 	mutex_init(&context->db_page_mutex);
1118 
1119 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1120 	mutex_init(&context->wqn_ranges_mutex);
1121 
1122 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1123 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1124 	else
1125 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1126 
1127 	if (err) {
1128 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1129 		kfree(context);
1130 		return ERR_PTR(-EFAULT);
1131 	}
1132 
1133 	return &context->ibucontext;
1134 }
1135 
1136 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1137 {
1138 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1139 
1140 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1141 	kfree(context);
1142 
1143 	return 0;
1144 }
1145 
1146 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1147 {
1148 	/* vma_open is called when a new VMA is created on top of our VMA.
1149 	 * This is done through either mremap flow or split_vma (usually due
1150 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1151 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1152 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1153 	 * calling us again and trying to do incorrect actions. We assume that
1154 	 * the original vma size is exactly a single page that there will be no
1155 	 * "splitting" operations on.
1156 	 */
1157 	area->vm_ops = NULL;
1158 }
1159 
1160 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1161 {
1162 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1163 
1164 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1165 	 * file itself is closed, therefore no sync is needed with the regular
1166 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1167 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1168 	 * The close operation is usually called under mm->mmap_sem except when
1169 	 * process is exiting.  The exiting case is handled explicitly as part
1170 	 * of mlx4_ib_disassociate_ucontext.
1171 	 */
1172 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1173 				area->vm_private_data;
1174 
1175 	/* set the vma context pointer to null in the mlx4_ib driver's private
1176 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1177 	 */
1178 	mlx4_ib_vma_priv_data->vma = NULL;
1179 }
1180 
1181 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1182 	.open = mlx4_ib_vma_open,
1183 	.close = mlx4_ib_vma_close
1184 };
1185 
1186 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1187 {
1188 	int i;
1189 	int ret = 0;
1190 	struct vm_area_struct *vma;
1191 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1192 	struct task_struct *owning_process  = NULL;
1193 	struct mm_struct   *owning_mm       = NULL;
1194 
1195 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1196 	if (!owning_process)
1197 		return;
1198 
1199 	owning_mm = get_task_mm(owning_process);
1200 	if (!owning_mm) {
1201 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1202 		while (1) {
1203 			/* make sure that task is dead before returning, it may
1204 			 * prevent a rare case of module down in parallel to a
1205 			 * call to mlx4_ib_vma_close.
1206 			 */
1207 			put_task_struct(owning_process);
1208 			usleep_range(1000, 2000);
1209 			owning_process = get_pid_task(ibcontext->tgid,
1210 						      PIDTYPE_PID);
1211 			if (!owning_process ||
1212 			    owning_process->state == TASK_DEAD) {
1213 				pr_info("disassociate ucontext done, task was terminated\n");
1214 				/* in case task was dead need to release the task struct */
1215 				if (owning_process)
1216 					put_task_struct(owning_process);
1217 				return;
1218 			}
1219 		}
1220 	}
1221 
1222 	/* need to protect from a race on closing the vma as part of
1223 	 * mlx4_ib_vma_close().
1224 	 */
1225 	down_write(&owning_mm->mmap_sem);
1226 	for (i = 0; i < HW_BAR_COUNT; i++) {
1227 		vma = context->hw_bar_info[i].vma;
1228 		if (!vma)
1229 			continue;
1230 
1231 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1232 				   context->hw_bar_info[i].vma->vm_start,
1233 				   PAGE_SIZE);
1234 		if (ret) {
1235 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1236 			BUG_ON(1);
1237 		}
1238 
1239 		context->hw_bar_info[i].vma->vm_flags &=
1240 			~(VM_SHARED | VM_MAYSHARE);
1241 		/* context going to be destroyed, should not access ops any more */
1242 		context->hw_bar_info[i].vma->vm_ops = NULL;
1243 	}
1244 
1245 	up_write(&owning_mm->mmap_sem);
1246 	mmput(owning_mm);
1247 	put_task_struct(owning_process);
1248 }
1249 
1250 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1251 				 struct mlx4_ib_vma_private_data *vma_private_data)
1252 {
1253 	vma_private_data->vma = vma;
1254 	vma->vm_private_data = vma_private_data;
1255 	vma->vm_ops =  &mlx4_ib_vm_ops;
1256 }
1257 
1258 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1259 {
1260 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1261 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1262 
1263 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1264 		return -EINVAL;
1265 
1266 	if (vma->vm_pgoff == 0) {
1267 		/* We prevent double mmaping on same context */
1268 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1269 			return -EINVAL;
1270 
1271 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1272 
1273 		if (io_remap_pfn_range(vma, vma->vm_start,
1274 				       to_mucontext(context)->uar.pfn,
1275 				       PAGE_SIZE, vma->vm_page_prot))
1276 			return -EAGAIN;
1277 
1278 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1279 
1280 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1281 		/* We prevent double mmaping on same context */
1282 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1283 			return -EINVAL;
1284 
1285 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1286 
1287 		if (io_remap_pfn_range(vma, vma->vm_start,
1288 				       to_mucontext(context)->uar.pfn +
1289 				       dev->dev->caps.num_uars,
1290 				       PAGE_SIZE, vma->vm_page_prot))
1291 			return -EAGAIN;
1292 
1293 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1294 
1295 	} else if (vma->vm_pgoff == 3) {
1296 		struct mlx4_clock_params params;
1297 		int ret;
1298 
1299 		/* We prevent double mmaping on same context */
1300 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1301 			return -EINVAL;
1302 
1303 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1304 
1305 		if (ret)
1306 			return ret;
1307 
1308 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1309 		if (io_remap_pfn_range(vma, vma->vm_start,
1310 				       (pci_resource_start(dev->dev->persist->pdev,
1311 							   params.bar) +
1312 					params.offset)
1313 				       >> PAGE_SHIFT,
1314 				       PAGE_SIZE, vma->vm_page_prot))
1315 			return -EAGAIN;
1316 
1317 		mlx4_ib_set_vma_data(vma,
1318 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1319 	} else {
1320 		return -EINVAL;
1321 	}
1322 
1323 	return 0;
1324 }
1325 
1326 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1327 				      struct ib_ucontext *context,
1328 				      struct ib_udata *udata)
1329 {
1330 	struct mlx4_ib_pd *pd;
1331 	int err;
1332 
1333 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1334 	if (!pd)
1335 		return ERR_PTR(-ENOMEM);
1336 
1337 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1338 	if (err) {
1339 		kfree(pd);
1340 		return ERR_PTR(err);
1341 	}
1342 
1343 	if (context)
1344 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1345 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1346 			kfree(pd);
1347 			return ERR_PTR(-EFAULT);
1348 		}
1349 
1350 	return &pd->ibpd;
1351 }
1352 
1353 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1354 {
1355 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1356 	kfree(pd);
1357 
1358 	return 0;
1359 }
1360 
1361 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1362 					  struct ib_ucontext *context,
1363 					  struct ib_udata *udata)
1364 {
1365 	struct mlx4_ib_xrcd *xrcd;
1366 	struct ib_cq_init_attr cq_attr = {};
1367 	int err;
1368 
1369 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1370 		return ERR_PTR(-ENOSYS);
1371 
1372 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1373 	if (!xrcd)
1374 		return ERR_PTR(-ENOMEM);
1375 
1376 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1377 	if (err)
1378 		goto err1;
1379 
1380 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1381 	if (IS_ERR(xrcd->pd)) {
1382 		err = PTR_ERR(xrcd->pd);
1383 		goto err2;
1384 	}
1385 
1386 	cq_attr.cqe = 1;
1387 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1388 	if (IS_ERR(xrcd->cq)) {
1389 		err = PTR_ERR(xrcd->cq);
1390 		goto err3;
1391 	}
1392 
1393 	return &xrcd->ibxrcd;
1394 
1395 err3:
1396 	ib_dealloc_pd(xrcd->pd);
1397 err2:
1398 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1399 err1:
1400 	kfree(xrcd);
1401 	return ERR_PTR(err);
1402 }
1403 
1404 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1405 {
1406 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1407 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1408 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1409 	kfree(xrcd);
1410 
1411 	return 0;
1412 }
1413 
1414 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1415 {
1416 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1417 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1418 	struct mlx4_ib_gid_entry *ge;
1419 
1420 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1421 	if (!ge)
1422 		return -ENOMEM;
1423 
1424 	ge->gid = *gid;
1425 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1426 		ge->port = mqp->port;
1427 		ge->added = 1;
1428 	}
1429 
1430 	mutex_lock(&mqp->mutex);
1431 	list_add_tail(&ge->list, &mqp->gid_list);
1432 	mutex_unlock(&mqp->mutex);
1433 
1434 	return 0;
1435 }
1436 
1437 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1438 					  struct mlx4_ib_counters *ctr_table)
1439 {
1440 	struct counter_index *counter, *tmp_count;
1441 
1442 	mutex_lock(&ctr_table->mutex);
1443 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1444 				 list) {
1445 		if (counter->allocated)
1446 			mlx4_counter_free(ibdev->dev, counter->index);
1447 		list_del(&counter->list);
1448 		kfree(counter);
1449 	}
1450 	mutex_unlock(&ctr_table->mutex);
1451 }
1452 
1453 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1454 		   union ib_gid *gid)
1455 {
1456 	struct net_device *ndev;
1457 	int ret = 0;
1458 
1459 	if (!mqp->port)
1460 		return 0;
1461 
1462 	spin_lock_bh(&mdev->iboe.lock);
1463 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1464 	if (ndev)
1465 		dev_hold(ndev);
1466 	spin_unlock_bh(&mdev->iboe.lock);
1467 
1468 	if (ndev) {
1469 		ret = 1;
1470 		dev_put(ndev);
1471 	}
1472 
1473 	return ret;
1474 }
1475 
1476 struct mlx4_ib_steering {
1477 	struct list_head list;
1478 	struct mlx4_flow_reg_id reg_id;
1479 	union ib_gid gid;
1480 };
1481 
1482 #define LAST_ETH_FIELD vlan_tag
1483 #define LAST_IB_FIELD sl
1484 #define LAST_IPV4_FIELD dst_ip
1485 #define LAST_TCP_UDP_FIELD src_port
1486 
1487 /* Field is the last supported field */
1488 #define FIELDS_NOT_SUPPORTED(filter, field)\
1489 	memchr_inv((void *)&filter.field  +\
1490 		   sizeof(filter.field), 0,\
1491 		   sizeof(filter) -\
1492 		   offsetof(typeof(filter), field) -\
1493 		   sizeof(filter.field))
1494 
1495 static int parse_flow_attr(struct mlx4_dev *dev,
1496 			   u32 qp_num,
1497 			   union ib_flow_spec *ib_spec,
1498 			   struct _rule_hw *mlx4_spec)
1499 {
1500 	enum mlx4_net_trans_rule_id type;
1501 
1502 	switch (ib_spec->type) {
1503 	case IB_FLOW_SPEC_ETH:
1504 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1505 			return -ENOTSUPP;
1506 
1507 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1508 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1509 		       ETH_ALEN);
1510 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1511 		       ETH_ALEN);
1512 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1513 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1514 		break;
1515 	case IB_FLOW_SPEC_IB:
1516 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1517 			return -ENOTSUPP;
1518 
1519 		type = MLX4_NET_TRANS_RULE_ID_IB;
1520 		mlx4_spec->ib.l3_qpn =
1521 			cpu_to_be32(qp_num);
1522 		mlx4_spec->ib.qpn_mask =
1523 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1524 		break;
1525 
1526 
1527 	case IB_FLOW_SPEC_IPV4:
1528 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1529 			return -ENOTSUPP;
1530 
1531 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1532 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1533 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1534 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1535 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1536 		break;
1537 
1538 	case IB_FLOW_SPEC_TCP:
1539 	case IB_FLOW_SPEC_UDP:
1540 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1541 			return -ENOTSUPP;
1542 
1543 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1544 					MLX4_NET_TRANS_RULE_ID_TCP :
1545 					MLX4_NET_TRANS_RULE_ID_UDP;
1546 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1547 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1548 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1549 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1550 		break;
1551 
1552 	default:
1553 		return -EINVAL;
1554 	}
1555 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1556 	    mlx4_hw_rule_sz(dev, type) < 0)
1557 		return -EINVAL;
1558 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1559 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1560 	return mlx4_hw_rule_sz(dev, type);
1561 }
1562 
1563 struct default_rules {
1564 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1565 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1566 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1567 	__u8  link_layer;
1568 };
1569 static const struct default_rules default_table[] = {
1570 	{
1571 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1572 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1573 		.rules_create_list = {IB_FLOW_SPEC_IB},
1574 		.link_layer = IB_LINK_LAYER_INFINIBAND
1575 	}
1576 };
1577 
1578 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1579 					 struct ib_flow_attr *flow_attr)
1580 {
1581 	int i, j, k;
1582 	void *ib_flow;
1583 	const struct default_rules *pdefault_rules = default_table;
1584 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1585 
1586 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1587 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1588 		memset(&field_types, 0, sizeof(field_types));
1589 
1590 		if (link_layer != pdefault_rules->link_layer)
1591 			continue;
1592 
1593 		ib_flow = flow_attr + 1;
1594 		/* we assume the specs are sorted */
1595 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1596 		     j < flow_attr->num_of_specs; k++) {
1597 			union ib_flow_spec *current_flow =
1598 				(union ib_flow_spec *)ib_flow;
1599 
1600 			/* same layer but different type */
1601 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1602 			     (pdefault_rules->mandatory_fields[k] &
1603 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1604 			    (current_flow->type !=
1605 			     pdefault_rules->mandatory_fields[k]))
1606 				goto out;
1607 
1608 			/* same layer, try match next one */
1609 			if (current_flow->type ==
1610 			    pdefault_rules->mandatory_fields[k]) {
1611 				j++;
1612 				ib_flow +=
1613 					((union ib_flow_spec *)ib_flow)->size;
1614 			}
1615 		}
1616 
1617 		ib_flow = flow_attr + 1;
1618 		for (j = 0; j < flow_attr->num_of_specs;
1619 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1620 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1621 				/* same layer and same type */
1622 				if (((union ib_flow_spec *)ib_flow)->type ==
1623 				    pdefault_rules->mandatory_not_fields[k])
1624 					goto out;
1625 
1626 		return i;
1627 	}
1628 out:
1629 	return -1;
1630 }
1631 
1632 static int __mlx4_ib_create_default_rules(
1633 		struct mlx4_ib_dev *mdev,
1634 		struct ib_qp *qp,
1635 		const struct default_rules *pdefault_rules,
1636 		struct _rule_hw *mlx4_spec) {
1637 	int size = 0;
1638 	int i;
1639 
1640 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1641 		int ret;
1642 		union ib_flow_spec ib_spec;
1643 		switch (pdefault_rules->rules_create_list[i]) {
1644 		case 0:
1645 			/* no rule */
1646 			continue;
1647 		case IB_FLOW_SPEC_IB:
1648 			ib_spec.type = IB_FLOW_SPEC_IB;
1649 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1650 
1651 			break;
1652 		default:
1653 			/* invalid rule */
1654 			return -EINVAL;
1655 		}
1656 		/* We must put empty rule, qpn is being ignored */
1657 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1658 				      mlx4_spec);
1659 		if (ret < 0) {
1660 			pr_info("invalid parsing\n");
1661 			return -EINVAL;
1662 		}
1663 
1664 		mlx4_spec = (void *)mlx4_spec + ret;
1665 		size += ret;
1666 	}
1667 	return size;
1668 }
1669 
1670 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1671 			  int domain,
1672 			  enum mlx4_net_trans_promisc_mode flow_type,
1673 			  u64 *reg_id)
1674 {
1675 	int ret, i;
1676 	int size = 0;
1677 	void *ib_flow;
1678 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1679 	struct mlx4_cmd_mailbox *mailbox;
1680 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1681 	int default_flow;
1682 
1683 	static const u16 __mlx4_domain[] = {
1684 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1685 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1686 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1687 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1688 	};
1689 
1690 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1691 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1692 		return -EINVAL;
1693 	}
1694 
1695 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1696 		pr_err("Invalid domain value %d\n", domain);
1697 		return -EINVAL;
1698 	}
1699 
1700 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1701 		return -EINVAL;
1702 
1703 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1704 	if (IS_ERR(mailbox))
1705 		return PTR_ERR(mailbox);
1706 	ctrl = mailbox->buf;
1707 
1708 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1709 				 flow_attr->priority);
1710 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1711 	ctrl->port = flow_attr->port;
1712 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1713 
1714 	ib_flow = flow_attr + 1;
1715 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1716 	/* Add default flows */
1717 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1718 	if (default_flow >= 0) {
1719 		ret = __mlx4_ib_create_default_rules(
1720 				mdev, qp, default_table + default_flow,
1721 				mailbox->buf + size);
1722 		if (ret < 0) {
1723 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1724 			return -EINVAL;
1725 		}
1726 		size += ret;
1727 	}
1728 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1729 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1730 				      mailbox->buf + size);
1731 		if (ret < 0) {
1732 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1733 			return -EINVAL;
1734 		}
1735 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1736 		size += ret;
1737 	}
1738 
1739 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1740 	    flow_attr->num_of_specs == 1) {
1741 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1742 		enum ib_flow_spec_type header_spec =
1743 			((union ib_flow_spec *)(flow_attr + 1))->type;
1744 
1745 		if (header_spec == IB_FLOW_SPEC_ETH)
1746 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1747 	}
1748 
1749 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1750 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1751 			   MLX4_CMD_NATIVE);
1752 	if (ret == -ENOMEM)
1753 		pr_err("mcg table is full. Fail to register network rule.\n");
1754 	else if (ret == -ENXIO)
1755 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1756 	else if (ret)
1757 		pr_err("Invalid argument. Fail to register network rule.\n");
1758 
1759 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1760 	return ret;
1761 }
1762 
1763 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1764 {
1765 	int err;
1766 	err = mlx4_cmd(dev, reg_id, 0, 0,
1767 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1768 		       MLX4_CMD_NATIVE);
1769 	if (err)
1770 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1771 		       reg_id);
1772 	return err;
1773 }
1774 
1775 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1776 				    u64 *reg_id)
1777 {
1778 	void *ib_flow;
1779 	union ib_flow_spec *ib_spec;
1780 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1781 	int err = 0;
1782 
1783 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1784 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1785 		return 0; /* do nothing */
1786 
1787 	ib_flow = flow_attr + 1;
1788 	ib_spec = (union ib_flow_spec *)ib_flow;
1789 
1790 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1791 		return 0; /* do nothing */
1792 
1793 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1794 				    flow_attr->port, qp->qp_num,
1795 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1796 				    reg_id);
1797 	return err;
1798 }
1799 
1800 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1801 				      struct ib_flow_attr *flow_attr,
1802 				      enum mlx4_net_trans_promisc_mode *type)
1803 {
1804 	int err = 0;
1805 
1806 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1807 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1808 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1809 		return -EOPNOTSUPP;
1810 	}
1811 
1812 	if (flow_attr->num_of_specs == 0) {
1813 		type[0] = MLX4_FS_MC_SNIFFER;
1814 		type[1] = MLX4_FS_UC_SNIFFER;
1815 	} else {
1816 		union ib_flow_spec *ib_spec;
1817 
1818 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1819 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1820 			return -EINVAL;
1821 
1822 		/* if all is zero than MC and UC */
1823 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1824 			type[0] = MLX4_FS_MC_SNIFFER;
1825 			type[1] = MLX4_FS_UC_SNIFFER;
1826 		} else {
1827 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1828 					    ib_spec->eth.mask.dst_mac[1],
1829 					    ib_spec->eth.mask.dst_mac[2],
1830 					    ib_spec->eth.mask.dst_mac[3],
1831 					    ib_spec->eth.mask.dst_mac[4],
1832 					    ib_spec->eth.mask.dst_mac[5]};
1833 
1834 			/* Above xor was only on MC bit, non empty mask is valid
1835 			 * only if this bit is set and rest are zero.
1836 			 */
1837 			if (!is_zero_ether_addr(&mac[0]))
1838 				return -EINVAL;
1839 
1840 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1841 				type[0] = MLX4_FS_MC_SNIFFER;
1842 			else
1843 				type[0] = MLX4_FS_UC_SNIFFER;
1844 		}
1845 	}
1846 
1847 	return err;
1848 }
1849 
1850 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1851 				    struct ib_flow_attr *flow_attr,
1852 				    int domain)
1853 {
1854 	int err = 0, i = 0, j = 0;
1855 	struct mlx4_ib_flow *mflow;
1856 	enum mlx4_net_trans_promisc_mode type[2];
1857 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1858 	int is_bonded = mlx4_is_bonded(dev);
1859 
1860 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1861 		return ERR_PTR(-EINVAL);
1862 
1863 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1864 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1865 		return ERR_PTR(-EOPNOTSUPP);
1866 
1867 	memset(type, 0, sizeof(type));
1868 
1869 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1870 	if (!mflow) {
1871 		err = -ENOMEM;
1872 		goto err_free;
1873 	}
1874 
1875 	switch (flow_attr->type) {
1876 	case IB_FLOW_ATTR_NORMAL:
1877 		/* If dont trap flag (continue match) is set, under specific
1878 		 * condition traffic be replicated to given qp,
1879 		 * without stealing it
1880 		 */
1881 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1882 			err = mlx4_ib_add_dont_trap_rule(dev,
1883 							 flow_attr,
1884 							 type);
1885 			if (err)
1886 				goto err_free;
1887 		} else {
1888 			type[0] = MLX4_FS_REGULAR;
1889 		}
1890 		break;
1891 
1892 	case IB_FLOW_ATTR_ALL_DEFAULT:
1893 		type[0] = MLX4_FS_ALL_DEFAULT;
1894 		break;
1895 
1896 	case IB_FLOW_ATTR_MC_DEFAULT:
1897 		type[0] = MLX4_FS_MC_DEFAULT;
1898 		break;
1899 
1900 	case IB_FLOW_ATTR_SNIFFER:
1901 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1902 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1903 		break;
1904 
1905 	default:
1906 		err = -EINVAL;
1907 		goto err_free;
1908 	}
1909 
1910 	while (i < ARRAY_SIZE(type) && type[i]) {
1911 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1912 					    &mflow->reg_id[i].id);
1913 		if (err)
1914 			goto err_create_flow;
1915 		if (is_bonded) {
1916 			/* Application always sees one port so the mirror rule
1917 			 * must be on port #2
1918 			 */
1919 			flow_attr->port = 2;
1920 			err = __mlx4_ib_create_flow(qp, flow_attr,
1921 						    domain, type[j],
1922 						    &mflow->reg_id[j].mirror);
1923 			flow_attr->port = 1;
1924 			if (err)
1925 				goto err_create_flow;
1926 			j++;
1927 		}
1928 
1929 		i++;
1930 	}
1931 
1932 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1933 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1934 					       &mflow->reg_id[i].id);
1935 		if (err)
1936 			goto err_create_flow;
1937 
1938 		if (is_bonded) {
1939 			flow_attr->port = 2;
1940 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1941 						       &mflow->reg_id[j].mirror);
1942 			flow_attr->port = 1;
1943 			if (err)
1944 				goto err_create_flow;
1945 			j++;
1946 		}
1947 		/* function to create mirror rule */
1948 		i++;
1949 	}
1950 
1951 	return &mflow->ibflow;
1952 
1953 err_create_flow:
1954 	while (i) {
1955 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1956 					     mflow->reg_id[i].id);
1957 		i--;
1958 	}
1959 
1960 	while (j) {
1961 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1962 					     mflow->reg_id[j].mirror);
1963 		j--;
1964 	}
1965 err_free:
1966 	kfree(mflow);
1967 	return ERR_PTR(err);
1968 }
1969 
1970 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1971 {
1972 	int err, ret = 0;
1973 	int i = 0;
1974 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1975 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1976 
1977 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1978 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1979 		if (err)
1980 			ret = err;
1981 		if (mflow->reg_id[i].mirror) {
1982 			err = __mlx4_ib_destroy_flow(mdev->dev,
1983 						     mflow->reg_id[i].mirror);
1984 			if (err)
1985 				ret = err;
1986 		}
1987 		i++;
1988 	}
1989 
1990 	kfree(mflow);
1991 	return ret;
1992 }
1993 
1994 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1995 {
1996 	int err;
1997 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1998 	struct mlx4_dev	*dev = mdev->dev;
1999 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2000 	struct mlx4_ib_steering *ib_steering = NULL;
2001 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
2002 	struct mlx4_flow_reg_id	reg_id;
2003 
2004 	if (mdev->dev->caps.steering_mode ==
2005 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2006 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
2007 		if (!ib_steering)
2008 			return -ENOMEM;
2009 	}
2010 
2011 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
2012 				    !!(mqp->flags &
2013 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2014 				    prot, &reg_id.id);
2015 	if (err) {
2016 		pr_err("multicast attach op failed, err %d\n", err);
2017 		goto err_malloc;
2018 	}
2019 
2020 	reg_id.mirror = 0;
2021 	if (mlx4_is_bonded(dev)) {
2022 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
2023 					    (mqp->port == 1) ? 2 : 1,
2024 					    !!(mqp->flags &
2025 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2026 					    prot, &reg_id.mirror);
2027 		if (err)
2028 			goto err_add;
2029 	}
2030 
2031 	err = add_gid_entry(ibqp, gid);
2032 	if (err)
2033 		goto err_add;
2034 
2035 	if (ib_steering) {
2036 		memcpy(ib_steering->gid.raw, gid->raw, 16);
2037 		ib_steering->reg_id = reg_id;
2038 		mutex_lock(&mqp->mutex);
2039 		list_add(&ib_steering->list, &mqp->steering_rules);
2040 		mutex_unlock(&mqp->mutex);
2041 	}
2042 	return 0;
2043 
2044 err_add:
2045 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2046 			      prot, reg_id.id);
2047 	if (reg_id.mirror)
2048 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2049 				      prot, reg_id.mirror);
2050 err_malloc:
2051 	kfree(ib_steering);
2052 
2053 	return err;
2054 }
2055 
2056 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2057 {
2058 	struct mlx4_ib_gid_entry *ge;
2059 	struct mlx4_ib_gid_entry *tmp;
2060 	struct mlx4_ib_gid_entry *ret = NULL;
2061 
2062 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2063 		if (!memcmp(raw, ge->gid.raw, 16)) {
2064 			ret = ge;
2065 			break;
2066 		}
2067 	}
2068 
2069 	return ret;
2070 }
2071 
2072 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2073 {
2074 	int err;
2075 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2076 	struct mlx4_dev *dev = mdev->dev;
2077 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2078 	struct net_device *ndev;
2079 	struct mlx4_ib_gid_entry *ge;
2080 	struct mlx4_flow_reg_id reg_id = {0, 0};
2081 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2082 
2083 	if (mdev->dev->caps.steering_mode ==
2084 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2085 		struct mlx4_ib_steering *ib_steering;
2086 
2087 		mutex_lock(&mqp->mutex);
2088 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2089 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2090 				list_del(&ib_steering->list);
2091 				break;
2092 			}
2093 		}
2094 		mutex_unlock(&mqp->mutex);
2095 		if (&ib_steering->list == &mqp->steering_rules) {
2096 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2097 			return -EINVAL;
2098 		}
2099 		reg_id = ib_steering->reg_id;
2100 		kfree(ib_steering);
2101 	}
2102 
2103 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2104 				    prot, reg_id.id);
2105 	if (err)
2106 		return err;
2107 
2108 	if (mlx4_is_bonded(dev)) {
2109 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2110 					    prot, reg_id.mirror);
2111 		if (err)
2112 			return err;
2113 	}
2114 
2115 	mutex_lock(&mqp->mutex);
2116 	ge = find_gid_entry(mqp, gid->raw);
2117 	if (ge) {
2118 		spin_lock_bh(&mdev->iboe.lock);
2119 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2120 		if (ndev)
2121 			dev_hold(ndev);
2122 		spin_unlock_bh(&mdev->iboe.lock);
2123 		if (ndev)
2124 			dev_put(ndev);
2125 		list_del(&ge->list);
2126 		kfree(ge);
2127 	} else
2128 		pr_warn("could not find mgid entry\n");
2129 
2130 	mutex_unlock(&mqp->mutex);
2131 
2132 	return 0;
2133 }
2134 
2135 static int init_node_data(struct mlx4_ib_dev *dev)
2136 {
2137 	struct ib_smp *in_mad  = NULL;
2138 	struct ib_smp *out_mad = NULL;
2139 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2140 	int err = -ENOMEM;
2141 
2142 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2143 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2144 	if (!in_mad || !out_mad)
2145 		goto out;
2146 
2147 	init_query_mad(in_mad);
2148 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2149 	if (mlx4_is_master(dev->dev))
2150 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2151 
2152 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2153 	if (err)
2154 		goto out;
2155 
2156 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2157 
2158 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2159 
2160 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2161 	if (err)
2162 		goto out;
2163 
2164 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2165 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2166 
2167 out:
2168 	kfree(in_mad);
2169 	kfree(out_mad);
2170 	return err;
2171 }
2172 
2173 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2174 			char *buf)
2175 {
2176 	struct mlx4_ib_dev *dev =
2177 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2178 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2179 }
2180 
2181 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2182 			char *buf)
2183 {
2184 	struct mlx4_ib_dev *dev =
2185 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2186 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2187 }
2188 
2189 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2190 			  char *buf)
2191 {
2192 	struct mlx4_ib_dev *dev =
2193 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2194 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2195 		       dev->dev->board_id);
2196 }
2197 
2198 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2199 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2200 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2201 
2202 static struct device_attribute *mlx4_class_attributes[] = {
2203 	&dev_attr_hw_rev,
2204 	&dev_attr_hca_type,
2205 	&dev_attr_board_id
2206 };
2207 
2208 struct diag_counter {
2209 	const char *name;
2210 	u32 offset;
2211 };
2212 
2213 #define DIAG_COUNTER(_name, _offset)			\
2214 	{ .name = #_name, .offset = _offset }
2215 
2216 static const struct diag_counter diag_basic[] = {
2217 	DIAG_COUNTER(rq_num_lle, 0x00),
2218 	DIAG_COUNTER(sq_num_lle, 0x04),
2219 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2220 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2221 	DIAG_COUNTER(rq_num_lpe, 0x18),
2222 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2223 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2224 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2225 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2226 	DIAG_COUNTER(sq_num_bre, 0x34),
2227 	DIAG_COUNTER(sq_num_rire, 0x44),
2228 	DIAG_COUNTER(rq_num_rire, 0x48),
2229 	DIAG_COUNTER(sq_num_rae, 0x4C),
2230 	DIAG_COUNTER(rq_num_rae, 0x50),
2231 	DIAG_COUNTER(sq_num_roe, 0x54),
2232 	DIAG_COUNTER(sq_num_tree, 0x5C),
2233 	DIAG_COUNTER(sq_num_rree, 0x64),
2234 	DIAG_COUNTER(rq_num_rnr, 0x68),
2235 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2236 	DIAG_COUNTER(rq_num_oos, 0x100),
2237 	DIAG_COUNTER(sq_num_oos, 0x104),
2238 };
2239 
2240 static const struct diag_counter diag_ext[] = {
2241 	DIAG_COUNTER(rq_num_dup, 0x130),
2242 	DIAG_COUNTER(sq_num_to, 0x134),
2243 };
2244 
2245 static const struct diag_counter diag_device_only[] = {
2246 	DIAG_COUNTER(num_cqovf, 0x1A0),
2247 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2248 };
2249 
2250 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2251 						    u8 port_num)
2252 {
2253 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2254 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2255 
2256 	if (!diag[!!port_num].name)
2257 		return NULL;
2258 
2259 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2260 					  diag[!!port_num].num_counters,
2261 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2262 }
2263 
2264 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2265 				struct rdma_hw_stats *stats,
2266 				u8 port, int index)
2267 {
2268 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2269 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2270 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2271 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2272 	int ret;
2273 	int i;
2274 
2275 	ret = mlx4_query_diag_counters(dev->dev,
2276 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2277 				       diag[!!port].offset, hw_value,
2278 				       diag[!!port].num_counters, port);
2279 
2280 	if (ret)
2281 		return ret;
2282 
2283 	for (i = 0; i < diag[!!port].num_counters; i++)
2284 		stats->value[i] = hw_value[i];
2285 
2286 	return diag[!!port].num_counters;
2287 }
2288 
2289 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2290 					 const char ***name,
2291 					 u32 **offset,
2292 					 u32 *num,
2293 					 bool port)
2294 {
2295 	u32 num_counters;
2296 
2297 	num_counters = ARRAY_SIZE(diag_basic);
2298 
2299 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2300 		num_counters += ARRAY_SIZE(diag_ext);
2301 
2302 	if (!port)
2303 		num_counters += ARRAY_SIZE(diag_device_only);
2304 
2305 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2306 	if (!*name)
2307 		return -ENOMEM;
2308 
2309 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2310 	if (!*offset)
2311 		goto err_name;
2312 
2313 	*num = num_counters;
2314 
2315 	return 0;
2316 
2317 err_name:
2318 	kfree(*name);
2319 	return -ENOMEM;
2320 }
2321 
2322 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2323 				       const char **name,
2324 				       u32 *offset,
2325 				       bool port)
2326 {
2327 	int i;
2328 	int j;
2329 
2330 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2331 		name[i] = diag_basic[i].name;
2332 		offset[i] = diag_basic[i].offset;
2333 	}
2334 
2335 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2336 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2337 			name[j] = diag_ext[i].name;
2338 			offset[j] = diag_ext[i].offset;
2339 		}
2340 	}
2341 
2342 	if (!port) {
2343 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2344 			name[j] = diag_device_only[i].name;
2345 			offset[j] = diag_device_only[i].offset;
2346 		}
2347 	}
2348 }
2349 
2350 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2351 {
2352 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2353 	int i;
2354 	int ret;
2355 	bool per_port = !!(ibdev->dev->caps.flags2 &
2356 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2357 
2358 	if (mlx4_is_slave(ibdev->dev))
2359 		return 0;
2360 
2361 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2362 		/* i == 1 means we are building port counters */
2363 		if (i && !per_port)
2364 			continue;
2365 
2366 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2367 						    &diag[i].offset,
2368 						    &diag[i].num_counters, i);
2369 		if (ret)
2370 			goto err_alloc;
2371 
2372 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2373 					   diag[i].offset, i);
2374 	}
2375 
2376 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2377 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2378 
2379 	return 0;
2380 
2381 err_alloc:
2382 	if (i) {
2383 		kfree(diag[i - 1].name);
2384 		kfree(diag[i - 1].offset);
2385 	}
2386 
2387 	return ret;
2388 }
2389 
2390 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2391 {
2392 	int i;
2393 
2394 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2395 		kfree(ibdev->diag_counters[i].offset);
2396 		kfree(ibdev->diag_counters[i].name);
2397 	}
2398 }
2399 
2400 #define MLX4_IB_INVALID_MAC	((u64)-1)
2401 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2402 			       struct net_device *dev,
2403 			       int port)
2404 {
2405 	u64 new_smac = 0;
2406 	u64 release_mac = MLX4_IB_INVALID_MAC;
2407 	struct mlx4_ib_qp *qp;
2408 
2409 	read_lock(&dev_base_lock);
2410 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2411 	read_unlock(&dev_base_lock);
2412 
2413 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2414 
2415 	/* no need for update QP1 and mac registration in non-SRIOV */
2416 	if (!mlx4_is_mfunc(ibdev->dev))
2417 		return;
2418 
2419 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2420 	qp = ibdev->qp1_proxy[port - 1];
2421 	if (qp) {
2422 		int new_smac_index;
2423 		u64 old_smac;
2424 		struct mlx4_update_qp_params update_params;
2425 
2426 		mutex_lock(&qp->mutex);
2427 		old_smac = qp->pri.smac;
2428 		if (new_smac == old_smac)
2429 			goto unlock;
2430 
2431 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2432 
2433 		if (new_smac_index < 0)
2434 			goto unlock;
2435 
2436 		update_params.smac_index = new_smac_index;
2437 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2438 				   &update_params)) {
2439 			release_mac = new_smac;
2440 			goto unlock;
2441 		}
2442 		/* if old port was zero, no mac was yet registered for this QP */
2443 		if (qp->pri.smac_port)
2444 			release_mac = old_smac;
2445 		qp->pri.smac = new_smac;
2446 		qp->pri.smac_port = port;
2447 		qp->pri.smac_index = new_smac_index;
2448 	}
2449 
2450 unlock:
2451 	if (release_mac != MLX4_IB_INVALID_MAC)
2452 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2453 	if (qp)
2454 		mutex_unlock(&qp->mutex);
2455 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2456 }
2457 
2458 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2459 				 struct net_device *dev,
2460 				 unsigned long event)
2461 
2462 {
2463 	struct mlx4_ib_iboe *iboe;
2464 	int update_qps_port = -1;
2465 	int port;
2466 
2467 	ASSERT_RTNL();
2468 
2469 	iboe = &ibdev->iboe;
2470 
2471 	spin_lock_bh(&iboe->lock);
2472 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2473 
2474 		iboe->netdevs[port - 1] =
2475 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2476 
2477 		if (dev == iboe->netdevs[port - 1] &&
2478 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2479 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2480 			update_qps_port = port;
2481 
2482 	}
2483 	spin_unlock_bh(&iboe->lock);
2484 
2485 	if (update_qps_port > 0)
2486 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2487 }
2488 
2489 static int mlx4_ib_netdev_event(struct notifier_block *this,
2490 				unsigned long event, void *ptr)
2491 {
2492 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2493 	struct mlx4_ib_dev *ibdev;
2494 
2495 	if (!net_eq(dev_net(dev), &init_net))
2496 		return NOTIFY_DONE;
2497 
2498 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2499 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2500 
2501 	return NOTIFY_DONE;
2502 }
2503 
2504 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2505 {
2506 	int port;
2507 	int slave;
2508 	int i;
2509 
2510 	if (mlx4_is_master(ibdev->dev)) {
2511 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2512 		     ++slave) {
2513 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2514 				for (i = 0;
2515 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2516 				     ++i) {
2517 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2518 					/* master has the identity virt2phys pkey mapping */
2519 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2520 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2521 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2522 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2523 				}
2524 			}
2525 		}
2526 		/* initialize pkey cache */
2527 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2528 			for (i = 0;
2529 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2530 			     ++i)
2531 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2532 					(i) ? 0 : 0xFFFF;
2533 		}
2534 	}
2535 }
2536 
2537 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2538 {
2539 	int i, j, eq = 0, total_eqs = 0;
2540 
2541 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2542 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2543 	if (!ibdev->eq_table)
2544 		return;
2545 
2546 	for (i = 1; i <= dev->caps.num_ports; i++) {
2547 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2548 		     j++, total_eqs++) {
2549 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2550 				continue;
2551 			ibdev->eq_table[eq] = total_eqs;
2552 			if (!mlx4_assign_eq(dev, i,
2553 					    &ibdev->eq_table[eq]))
2554 				eq++;
2555 			else
2556 				ibdev->eq_table[eq] = -1;
2557 		}
2558 	}
2559 
2560 	for (i = eq; i < dev->caps.num_comp_vectors;
2561 	     ibdev->eq_table[i++] = -1)
2562 		;
2563 
2564 	/* Advertise the new number of EQs to clients */
2565 	ibdev->ib_dev.num_comp_vectors = eq;
2566 }
2567 
2568 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2569 {
2570 	int i;
2571 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2572 
2573 	/* no eqs were allocated */
2574 	if (!ibdev->eq_table)
2575 		return;
2576 
2577 	/* Reset the advertised EQ number */
2578 	ibdev->ib_dev.num_comp_vectors = 0;
2579 
2580 	for (i = 0; i < total_eqs; i++)
2581 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2582 
2583 	kfree(ibdev->eq_table);
2584 	ibdev->eq_table = NULL;
2585 }
2586 
2587 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2588 			       struct ib_port_immutable *immutable)
2589 {
2590 	struct ib_port_attr attr;
2591 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2592 	int err;
2593 
2594 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2595 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2596 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2597 	} else {
2598 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2599 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2600 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2601 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2602 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2603 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2604 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2605 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2606 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2607 	}
2608 
2609 	err = ib_query_port(ibdev, port_num, &attr);
2610 	if (err)
2611 		return err;
2612 
2613 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2614 	immutable->gid_tbl_len = attr.gid_tbl_len;
2615 
2616 	return 0;
2617 }
2618 
2619 static void get_fw_ver_str(struct ib_device *device, char *str)
2620 {
2621 	struct mlx4_ib_dev *dev =
2622 		container_of(device, struct mlx4_ib_dev, ib_dev);
2623 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2624 		 (int) (dev->dev->caps.fw_ver >> 32),
2625 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2626 		 (int) dev->dev->caps.fw_ver & 0xffff);
2627 }
2628 
2629 static void *mlx4_ib_add(struct mlx4_dev *dev)
2630 {
2631 	struct mlx4_ib_dev *ibdev;
2632 	int num_ports = 0;
2633 	int i, j;
2634 	int err;
2635 	struct mlx4_ib_iboe *iboe;
2636 	int ib_num_ports = 0;
2637 	int num_req_counters;
2638 	int allocated;
2639 	u32 counter_index;
2640 	struct counter_index *new_counter_index = NULL;
2641 
2642 	pr_info_once("%s", mlx4_ib_version);
2643 
2644 	num_ports = 0;
2645 	mlx4_foreach_ib_transport_port(i, dev)
2646 		num_ports++;
2647 
2648 	/* No point in registering a device with no ports... */
2649 	if (num_ports == 0)
2650 		return NULL;
2651 
2652 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2653 	if (!ibdev) {
2654 		dev_err(&dev->persist->pdev->dev,
2655 			"Device struct alloc failed\n");
2656 		return NULL;
2657 	}
2658 
2659 	iboe = &ibdev->iboe;
2660 
2661 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2662 		goto err_dealloc;
2663 
2664 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2665 		goto err_pd;
2666 
2667 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2668 				 PAGE_SIZE);
2669 	if (!ibdev->uar_map)
2670 		goto err_uar;
2671 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2672 
2673 	ibdev->dev = dev;
2674 	ibdev->bond_next_port	= 0;
2675 
2676 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2677 	ibdev->ib_dev.owner		= THIS_MODULE;
2678 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2679 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2680 	ibdev->num_ports		= num_ports;
2681 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2682 						1 : ibdev->num_ports;
2683 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2684 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2685 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2686 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2687 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2688 
2689 	if (dev->caps.userspace_caps)
2690 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2691 	else
2692 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2693 
2694 	ibdev->ib_dev.uverbs_cmd_mask	=
2695 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2696 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2697 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2698 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2699 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2700 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2701 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2702 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2703 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2704 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2705 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2706 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2707 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2708 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2709 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2710 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2711 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2712 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2713 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2714 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2715 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2716 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2717 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2718 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2719 
2720 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2721 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2722 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2723 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2724 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2725 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2726 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2727 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2728 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2729 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2730 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2731 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2732 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2733 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2734 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2735 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2736 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2737 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2738 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2739 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2740 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2741 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2742 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2743 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2744 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2745 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2746 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2747 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2748 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2749 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2750 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2751 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2752 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2753 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2754 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2755 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2756 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2757 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2758 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2759 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2760 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2761 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2762 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2763 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2764 
2765 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2766 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2767 
2768 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2769 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2770 	    IB_LINK_LAYER_ETHERNET) ||
2771 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2772 	    IB_LINK_LAYER_ETHERNET))) {
2773 		ibdev->ib_dev.create_wq		= mlx4_ib_create_wq;
2774 		ibdev->ib_dev.modify_wq		= mlx4_ib_modify_wq;
2775 		ibdev->ib_dev.destroy_wq	= mlx4_ib_destroy_wq;
2776 		ibdev->ib_dev.create_rwq_ind_table  =
2777 			mlx4_ib_create_rwq_ind_table;
2778 		ibdev->ib_dev.destroy_rwq_ind_table =
2779 			mlx4_ib_destroy_rwq_ind_table;
2780 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2781 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2782 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2783 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2784 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2785 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2786 	}
2787 
2788 	if (!mlx4_is_slave(ibdev->dev)) {
2789 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2790 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2791 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2792 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2793 	}
2794 
2795 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2796 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2797 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2798 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2799 
2800 		ibdev->ib_dev.uverbs_cmd_mask |=
2801 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2802 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2803 	}
2804 
2805 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2806 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2807 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2808 		ibdev->ib_dev.uverbs_cmd_mask |=
2809 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2810 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2811 	}
2812 
2813 	if (check_flow_steering_support(dev)) {
2814 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2815 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2816 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2817 
2818 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2819 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2820 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2821 	}
2822 
2823 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2824 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2825 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2826 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2827 
2828 	mlx4_ib_alloc_eqs(dev, ibdev);
2829 
2830 	spin_lock_init(&iboe->lock);
2831 
2832 	if (init_node_data(ibdev))
2833 		goto err_map;
2834 	mlx4_init_sl2vl_tbl(ibdev);
2835 
2836 	for (i = 0; i < ibdev->num_ports; ++i) {
2837 		mutex_init(&ibdev->counters_table[i].mutex);
2838 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2839 	}
2840 
2841 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2842 	for (i = 0; i < num_req_counters; ++i) {
2843 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2844 		allocated = 0;
2845 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2846 						IB_LINK_LAYER_ETHERNET) {
2847 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2848 						 MLX4_RES_USAGE_DRIVER);
2849 			/* if failed to allocate a new counter, use default */
2850 			if (err)
2851 				counter_index =
2852 					mlx4_get_default_counter_index(dev,
2853 								       i + 1);
2854 			else
2855 				allocated = 1;
2856 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2857 			counter_index = mlx4_get_default_counter_index(dev,
2858 								       i + 1);
2859 		}
2860 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2861 					    GFP_KERNEL);
2862 		if (!new_counter_index) {
2863 			if (allocated)
2864 				mlx4_counter_free(ibdev->dev, counter_index);
2865 			goto err_counter;
2866 		}
2867 		new_counter_index->index = counter_index;
2868 		new_counter_index->allocated = allocated;
2869 		list_add_tail(&new_counter_index->list,
2870 			      &ibdev->counters_table[i].counters_list);
2871 		ibdev->counters_table[i].default_counter = counter_index;
2872 		pr_info("counter index %d for port %d allocated %d\n",
2873 			counter_index, i + 1, allocated);
2874 	}
2875 	if (mlx4_is_bonded(dev))
2876 		for (i = 1; i < ibdev->num_ports ; ++i) {
2877 			new_counter_index =
2878 					kmalloc(sizeof(struct counter_index),
2879 						GFP_KERNEL);
2880 			if (!new_counter_index)
2881 				goto err_counter;
2882 			new_counter_index->index = counter_index;
2883 			new_counter_index->allocated = 0;
2884 			list_add_tail(&new_counter_index->list,
2885 				      &ibdev->counters_table[i].counters_list);
2886 			ibdev->counters_table[i].default_counter =
2887 								counter_index;
2888 		}
2889 
2890 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2891 		ib_num_ports++;
2892 
2893 	spin_lock_init(&ibdev->sm_lock);
2894 	mutex_init(&ibdev->cap_mask_mutex);
2895 	INIT_LIST_HEAD(&ibdev->qp_list);
2896 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2897 
2898 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2899 	    ib_num_ports) {
2900 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2901 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2902 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2903 					    &ibdev->steer_qpn_base, 0,
2904 					    MLX4_RES_USAGE_DRIVER);
2905 		if (err)
2906 			goto err_counter;
2907 
2908 		ibdev->ib_uc_qpns_bitmap =
2909 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2910 				sizeof(long),
2911 				GFP_KERNEL);
2912 		if (!ibdev->ib_uc_qpns_bitmap)
2913 			goto err_steer_qp_release;
2914 
2915 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2916 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2917 				    ibdev->steer_qpn_count);
2918 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2919 					dev, ibdev->steer_qpn_base,
2920 					ibdev->steer_qpn_base +
2921 					ibdev->steer_qpn_count - 1);
2922 			if (err)
2923 				goto err_steer_free_bitmap;
2924 		} else {
2925 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2926 				    ibdev->steer_qpn_count);
2927 		}
2928 	}
2929 
2930 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2931 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2932 
2933 	if (mlx4_ib_alloc_diag_counters(ibdev))
2934 		goto err_steer_free_bitmap;
2935 
2936 	if (ib_register_device(&ibdev->ib_dev, NULL))
2937 		goto err_diag_counters;
2938 
2939 	if (mlx4_ib_mad_init(ibdev))
2940 		goto err_reg;
2941 
2942 	if (mlx4_ib_init_sriov(ibdev))
2943 		goto err_mad;
2944 
2945 	if (!iboe->nb.notifier_call) {
2946 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2947 		err = register_netdevice_notifier(&iboe->nb);
2948 		if (err) {
2949 			iboe->nb.notifier_call = NULL;
2950 			goto err_notif;
2951 		}
2952 	}
2953 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2954 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2955 		if (err)
2956 			goto err_notif;
2957 	}
2958 
2959 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2960 		if (device_create_file(&ibdev->ib_dev.dev,
2961 				       mlx4_class_attributes[j]))
2962 			goto err_notif;
2963 	}
2964 
2965 	ibdev->ib_active = true;
2966 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2967 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2968 					 &ibdev->ib_dev);
2969 
2970 	if (mlx4_is_mfunc(ibdev->dev))
2971 		init_pkeys(ibdev);
2972 
2973 	/* create paravirt contexts for any VFs which are active */
2974 	if (mlx4_is_master(ibdev->dev)) {
2975 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2976 			if (j == mlx4_master_func_num(ibdev->dev))
2977 				continue;
2978 			if (mlx4_is_slave_active(ibdev->dev, j))
2979 				do_slave_init(ibdev, j, 1);
2980 		}
2981 	}
2982 	return ibdev;
2983 
2984 err_notif:
2985 	if (ibdev->iboe.nb.notifier_call) {
2986 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2987 			pr_warn("failure unregistering notifier\n");
2988 		ibdev->iboe.nb.notifier_call = NULL;
2989 	}
2990 	flush_workqueue(wq);
2991 
2992 	mlx4_ib_close_sriov(ibdev);
2993 
2994 err_mad:
2995 	mlx4_ib_mad_cleanup(ibdev);
2996 
2997 err_reg:
2998 	ib_unregister_device(&ibdev->ib_dev);
2999 
3000 err_diag_counters:
3001 	mlx4_ib_diag_cleanup(ibdev);
3002 
3003 err_steer_free_bitmap:
3004 	kfree(ibdev->ib_uc_qpns_bitmap);
3005 
3006 err_steer_qp_release:
3007 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3008 			      ibdev->steer_qpn_count);
3009 err_counter:
3010 	for (i = 0; i < ibdev->num_ports; ++i)
3011 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
3012 
3013 err_map:
3014 	mlx4_ib_free_eqs(dev, ibdev);
3015 	iounmap(ibdev->uar_map);
3016 
3017 err_uar:
3018 	mlx4_uar_free(dev, &ibdev->priv_uar);
3019 
3020 err_pd:
3021 	mlx4_pd_free(dev, ibdev->priv_pdn);
3022 
3023 err_dealloc:
3024 	ib_dealloc_device(&ibdev->ib_dev);
3025 
3026 	return NULL;
3027 }
3028 
3029 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
3030 {
3031 	int offset;
3032 
3033 	WARN_ON(!dev->ib_uc_qpns_bitmap);
3034 
3035 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3036 					 dev->steer_qpn_count,
3037 					 get_count_order(count));
3038 	if (offset < 0)
3039 		return offset;
3040 
3041 	*qpn = dev->steer_qpn_base + offset;
3042 	return 0;
3043 }
3044 
3045 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3046 {
3047 	if (!qpn ||
3048 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3049 		return;
3050 
3051 	BUG_ON(qpn < dev->steer_qpn_base);
3052 
3053 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
3054 			      qpn - dev->steer_qpn_base,
3055 			      get_count_order(count));
3056 }
3057 
3058 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3059 			 int is_attach)
3060 {
3061 	int err;
3062 	size_t flow_size;
3063 	struct ib_flow_attr *flow = NULL;
3064 	struct ib_flow_spec_ib *ib_spec;
3065 
3066 	if (is_attach) {
3067 		flow_size = sizeof(struct ib_flow_attr) +
3068 			    sizeof(struct ib_flow_spec_ib);
3069 		flow = kzalloc(flow_size, GFP_KERNEL);
3070 		if (!flow)
3071 			return -ENOMEM;
3072 		flow->port = mqp->port;
3073 		flow->num_of_specs = 1;
3074 		flow->size = flow_size;
3075 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3076 		ib_spec->type = IB_FLOW_SPEC_IB;
3077 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3078 		/* Add an empty rule for IB L2 */
3079 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3080 
3081 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3082 					    IB_FLOW_DOMAIN_NIC,
3083 					    MLX4_FS_REGULAR,
3084 					    &mqp->reg_id);
3085 	} else {
3086 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3087 	}
3088 	kfree(flow);
3089 	return err;
3090 }
3091 
3092 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3093 {
3094 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3095 	int p;
3096 	int i;
3097 
3098 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3099 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3100 	ibdev->ib_active = false;
3101 	flush_workqueue(wq);
3102 
3103 	mlx4_ib_close_sriov(ibdev);
3104 	mlx4_ib_mad_cleanup(ibdev);
3105 	ib_unregister_device(&ibdev->ib_dev);
3106 	mlx4_ib_diag_cleanup(ibdev);
3107 	if (ibdev->iboe.nb.notifier_call) {
3108 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3109 			pr_warn("failure unregistering notifier\n");
3110 		ibdev->iboe.nb.notifier_call = NULL;
3111 	}
3112 
3113 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3114 			      ibdev->steer_qpn_count);
3115 	kfree(ibdev->ib_uc_qpns_bitmap);
3116 
3117 	iounmap(ibdev->uar_map);
3118 	for (p = 0; p < ibdev->num_ports; ++p)
3119 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3120 
3121 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3122 		mlx4_CLOSE_PORT(dev, p);
3123 
3124 	mlx4_ib_free_eqs(dev, ibdev);
3125 
3126 	mlx4_uar_free(dev, &ibdev->priv_uar);
3127 	mlx4_pd_free(dev, ibdev->priv_pdn);
3128 	ib_dealloc_device(&ibdev->ib_dev);
3129 }
3130 
3131 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3132 {
3133 	struct mlx4_ib_demux_work **dm = NULL;
3134 	struct mlx4_dev *dev = ibdev->dev;
3135 	int i;
3136 	unsigned long flags;
3137 	struct mlx4_active_ports actv_ports;
3138 	unsigned int ports;
3139 	unsigned int first_port;
3140 
3141 	if (!mlx4_is_master(dev))
3142 		return;
3143 
3144 	actv_ports = mlx4_get_active_ports(dev, slave);
3145 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3146 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3147 
3148 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3149 	if (!dm)
3150 		return;
3151 
3152 	for (i = 0; i < ports; i++) {
3153 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3154 		if (!dm[i]) {
3155 			while (--i >= 0)
3156 				kfree(dm[i]);
3157 			goto out;
3158 		}
3159 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3160 		dm[i]->port = first_port + i + 1;
3161 		dm[i]->slave = slave;
3162 		dm[i]->do_init = do_init;
3163 		dm[i]->dev = ibdev;
3164 	}
3165 	/* initialize or tear down tunnel QPs for the slave */
3166 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3167 	if (!ibdev->sriov.is_going_down) {
3168 		for (i = 0; i < ports; i++)
3169 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3170 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3171 	} else {
3172 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3173 		for (i = 0; i < ports; i++)
3174 			kfree(dm[i]);
3175 	}
3176 out:
3177 	kfree(dm);
3178 	return;
3179 }
3180 
3181 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3182 {
3183 	struct mlx4_ib_qp *mqp;
3184 	unsigned long flags_qp;
3185 	unsigned long flags_cq;
3186 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3187 	struct list_head    cq_notify_list;
3188 	struct mlx4_cq *mcq;
3189 	unsigned long flags;
3190 
3191 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3192 	INIT_LIST_HEAD(&cq_notify_list);
3193 
3194 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3195 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3196 
3197 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3198 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3199 		if (mqp->sq.tail != mqp->sq.head) {
3200 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3201 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3202 			if (send_mcq->mcq.comp &&
3203 			    mqp->ibqp.send_cq->comp_handler) {
3204 				if (!send_mcq->mcq.reset_notify_added) {
3205 					send_mcq->mcq.reset_notify_added = 1;
3206 					list_add_tail(&send_mcq->mcq.reset_notify,
3207 						      &cq_notify_list);
3208 				}
3209 			}
3210 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3211 		}
3212 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3213 		/* Now, handle the QP's receive queue */
3214 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3215 		/* no handling is needed for SRQ */
3216 		if (!mqp->ibqp.srq) {
3217 			if (mqp->rq.tail != mqp->rq.head) {
3218 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3219 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3220 				if (recv_mcq->mcq.comp &&
3221 				    mqp->ibqp.recv_cq->comp_handler) {
3222 					if (!recv_mcq->mcq.reset_notify_added) {
3223 						recv_mcq->mcq.reset_notify_added = 1;
3224 						list_add_tail(&recv_mcq->mcq.reset_notify,
3225 							      &cq_notify_list);
3226 					}
3227 				}
3228 				spin_unlock_irqrestore(&recv_mcq->lock,
3229 						       flags_cq);
3230 			}
3231 		}
3232 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3233 	}
3234 
3235 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3236 		mcq->comp(mcq);
3237 	}
3238 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3239 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3240 }
3241 
3242 static void handle_bonded_port_state_event(struct work_struct *work)
3243 {
3244 	struct ib_event_work *ew =
3245 		container_of(work, struct ib_event_work, work);
3246 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3247 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3248 	int i;
3249 	struct ib_event ibev;
3250 
3251 	kfree(ew);
3252 	spin_lock_bh(&ibdev->iboe.lock);
3253 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3254 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3255 		enum ib_port_state curr_port_state;
3256 
3257 		if (!curr_netdev)
3258 			continue;
3259 
3260 		curr_port_state =
3261 			(netif_running(curr_netdev) &&
3262 			 netif_carrier_ok(curr_netdev)) ?
3263 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3264 
3265 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3266 			curr_port_state : IB_PORT_ACTIVE;
3267 	}
3268 	spin_unlock_bh(&ibdev->iboe.lock);
3269 
3270 	ibev.device = &ibdev->ib_dev;
3271 	ibev.element.port_num = 1;
3272 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3273 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3274 
3275 	ib_dispatch_event(&ibev);
3276 }
3277 
3278 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3279 {
3280 	u64 sl2vl;
3281 	int err;
3282 
3283 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3284 	if (err) {
3285 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3286 		       port, err);
3287 		sl2vl = 0;
3288 	}
3289 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3290 }
3291 
3292 static void ib_sl2vl_update_work(struct work_struct *work)
3293 {
3294 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3295 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3296 	int port = ew->port;
3297 
3298 	mlx4_ib_sl2vl_update(mdev, port);
3299 
3300 	kfree(ew);
3301 }
3302 
3303 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3304 				     int port)
3305 {
3306 	struct ib_event_work *ew;
3307 
3308 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3309 	if (ew) {
3310 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3311 		ew->port = port;
3312 		ew->ib_dev = ibdev;
3313 		queue_work(wq, &ew->work);
3314 	}
3315 }
3316 
3317 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3318 			  enum mlx4_dev_event event, unsigned long param)
3319 {
3320 	struct ib_event ibev;
3321 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3322 	struct mlx4_eqe *eqe = NULL;
3323 	struct ib_event_work *ew;
3324 	int p = 0;
3325 
3326 	if (mlx4_is_bonded(dev) &&
3327 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3328 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3329 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3330 		if (!ew)
3331 			return;
3332 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3333 		ew->ib_dev = ibdev;
3334 		queue_work(wq, &ew->work);
3335 		return;
3336 	}
3337 
3338 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3339 		eqe = (struct mlx4_eqe *)param;
3340 	else
3341 		p = (int) param;
3342 
3343 	switch (event) {
3344 	case MLX4_DEV_EVENT_PORT_UP:
3345 		if (p > ibdev->num_ports)
3346 			return;
3347 		if (!mlx4_is_slave(dev) &&
3348 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3349 			IB_LINK_LAYER_INFINIBAND) {
3350 			if (mlx4_is_master(dev))
3351 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3352 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3353 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3354 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3355 		}
3356 		ibev.event = IB_EVENT_PORT_ACTIVE;
3357 		break;
3358 
3359 	case MLX4_DEV_EVENT_PORT_DOWN:
3360 		if (p > ibdev->num_ports)
3361 			return;
3362 		ibev.event = IB_EVENT_PORT_ERR;
3363 		break;
3364 
3365 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3366 		ibdev->ib_active = false;
3367 		ibev.event = IB_EVENT_DEVICE_FATAL;
3368 		mlx4_ib_handle_catas_error(ibdev);
3369 		break;
3370 
3371 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3372 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3373 		if (!ew)
3374 			break;
3375 
3376 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3377 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3378 		ew->ib_dev = ibdev;
3379 		/* need to queue only for port owner, which uses GEN_EQE */
3380 		if (mlx4_is_master(dev))
3381 			queue_work(wq, &ew->work);
3382 		else
3383 			handle_port_mgmt_change_event(&ew->work);
3384 		return;
3385 
3386 	case MLX4_DEV_EVENT_SLAVE_INIT:
3387 		/* here, p is the slave id */
3388 		do_slave_init(ibdev, p, 1);
3389 		if (mlx4_is_master(dev)) {
3390 			int i;
3391 
3392 			for (i = 1; i <= ibdev->num_ports; i++) {
3393 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3394 					== IB_LINK_LAYER_INFINIBAND)
3395 					mlx4_ib_slave_alias_guid_event(ibdev,
3396 								       p, i,
3397 								       1);
3398 			}
3399 		}
3400 		return;
3401 
3402 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3403 		if (mlx4_is_master(dev)) {
3404 			int i;
3405 
3406 			for (i = 1; i <= ibdev->num_ports; i++) {
3407 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3408 					== IB_LINK_LAYER_INFINIBAND)
3409 					mlx4_ib_slave_alias_guid_event(ibdev,
3410 								       p, i,
3411 								       0);
3412 			}
3413 		}
3414 		/* here, p is the slave id */
3415 		do_slave_init(ibdev, p, 0);
3416 		return;
3417 
3418 	default:
3419 		return;
3420 	}
3421 
3422 	ibev.device	      = ibdev_ptr;
3423 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3424 
3425 	ib_dispatch_event(&ibev);
3426 }
3427 
3428 static struct mlx4_interface mlx4_ib_interface = {
3429 	.add		= mlx4_ib_add,
3430 	.remove		= mlx4_ib_remove,
3431 	.event		= mlx4_ib_event,
3432 	.protocol	= MLX4_PROT_IB_IPV6,
3433 	.flags		= MLX4_INTFF_BONDING
3434 };
3435 
3436 static int __init mlx4_ib_init(void)
3437 {
3438 	int err;
3439 
3440 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3441 	if (!wq)
3442 		return -ENOMEM;
3443 
3444 	err = mlx4_ib_mcg_init();
3445 	if (err)
3446 		goto clean_wq;
3447 
3448 	err = mlx4_register_interface(&mlx4_ib_interface);
3449 	if (err)
3450 		goto clean_mcg;
3451 
3452 	return 0;
3453 
3454 clean_mcg:
3455 	mlx4_ib_mcg_destroy();
3456 
3457 clean_wq:
3458 	destroy_workqueue(wq);
3459 	return err;
3460 }
3461 
3462 static void __exit mlx4_ib_cleanup(void)
3463 {
3464 	mlx4_unregister_interface(&mlx4_ib_interface);
3465 	mlx4_ib_mcg_destroy();
3466 	destroy_workqueue(wq);
3467 }
3468 
3469 module_init(mlx4_ib_init);
3470 module_exit(mlx4_ib_cleanup);
3471