xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision ccb01374)
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(const struct ib_gid_attr *attr, void **context)
250 {
251 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
252 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
253 	struct mlx4_port_gid_table   *port_gid_table;
254 	int free = -1, found = -1;
255 	int ret = 0;
256 	int hw_update = 0;
257 	int i;
258 	struct gid_entry *gids = NULL;
259 
260 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
261 		return -EINVAL;
262 
263 	if (attr->port_num > MLX4_MAX_PORTS)
264 		return -EINVAL;
265 
266 	if (!context)
267 		return -EINVAL;
268 
269 	port_gid_table = &iboe->gids[attr->port_num - 1];
270 	spin_lock_bh(&iboe->lock);
271 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
272 		if (!memcmp(&port_gid_table->gids[i].gid,
273 			    &attr->gid, sizeof(attr->gid)) &&
274 		    port_gid_table->gids[i].gid_type == attr->gid_type)  {
275 			found = i;
276 			break;
277 		}
278 		if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
279 			free = i; /* HW has space */
280 	}
281 
282 	if (found < 0) {
283 		if (free < 0) {
284 			ret = -ENOSPC;
285 		} else {
286 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
287 			if (!port_gid_table->gids[free].ctx) {
288 				ret = -ENOMEM;
289 			} else {
290 				*context = port_gid_table->gids[free].ctx;
291 				memcpy(&port_gid_table->gids[free].gid,
292 				       &attr->gid, sizeof(attr->gid));
293 				port_gid_table->gids[free].gid_type = attr->gid_type;
294 				port_gid_table->gids[free].ctx->real_index = free;
295 				port_gid_table->gids[free].ctx->refcount = 1;
296 				hw_update = 1;
297 			}
298 		}
299 	} else {
300 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
301 		*context = ctx;
302 		ctx->refcount++;
303 	}
304 	if (!ret && hw_update) {
305 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
306 				     GFP_ATOMIC);
307 		if (!gids) {
308 			ret = -ENOMEM;
309 		} else {
310 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
311 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
312 				gids[i].gid_type = port_gid_table->gids[i].gid_type;
313 			}
314 		}
315 	}
316 	spin_unlock_bh(&iboe->lock);
317 
318 	if (!ret && hw_update) {
319 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
320 		kfree(gids);
321 	}
322 
323 	return ret;
324 }
325 
326 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
327 {
328 	struct gid_cache_context *ctx = *context;
329 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
330 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
331 	struct mlx4_port_gid_table   *port_gid_table;
332 	int ret = 0;
333 	int hw_update = 0;
334 	struct gid_entry *gids = NULL;
335 
336 	if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
337 		return -EINVAL;
338 
339 	if (attr->port_num > MLX4_MAX_PORTS)
340 		return -EINVAL;
341 
342 	port_gid_table = &iboe->gids[attr->port_num - 1];
343 	spin_lock_bh(&iboe->lock);
344 	if (ctx) {
345 		ctx->refcount--;
346 		if (!ctx->refcount) {
347 			unsigned int real_index = ctx->real_index;
348 
349 			memset(&port_gid_table->gids[real_index].gid, 0,
350 			       sizeof(port_gid_table->gids[real_index].gid));
351 			kfree(port_gid_table->gids[real_index].ctx);
352 			port_gid_table->gids[real_index].ctx = NULL;
353 			hw_update = 1;
354 		}
355 	}
356 	if (!ret && hw_update) {
357 		int i;
358 
359 		gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
360 				     GFP_ATOMIC);
361 		if (!gids) {
362 			ret = -ENOMEM;
363 		} else {
364 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
365 				memcpy(&gids[i].gid,
366 				       &port_gid_table->gids[i].gid,
367 				       sizeof(union ib_gid));
368 				gids[i].gid_type =
369 				    port_gid_table->gids[i].gid_type;
370 			}
371 		}
372 	}
373 	spin_unlock_bh(&iboe->lock);
374 
375 	if (!ret && hw_update) {
376 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
377 		kfree(gids);
378 	}
379 	return ret;
380 }
381 
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383 				    const struct ib_gid_attr *attr)
384 {
385 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386 	struct gid_cache_context *ctx = NULL;
387 	struct mlx4_port_gid_table   *port_gid_table;
388 	int real_index = -EINVAL;
389 	int i;
390 	unsigned long flags;
391 	u8 port_num = attr->port_num;
392 
393 	if (port_num > MLX4_MAX_PORTS)
394 		return -EINVAL;
395 
396 	if (mlx4_is_bonded(ibdev->dev))
397 		port_num = 1;
398 
399 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
400 		return attr->index;
401 
402 	spin_lock_irqsave(&iboe->lock, flags);
403 	port_gid_table = &iboe->gids[port_num - 1];
404 
405 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
406 		if (!memcmp(&port_gid_table->gids[i].gid,
407 			    &attr->gid, sizeof(attr->gid)) &&
408 		    attr->gid_type == port_gid_table->gids[i].gid_type) {
409 			ctx = port_gid_table->gids[i].ctx;
410 			break;
411 		}
412 	if (ctx)
413 		real_index = ctx->real_index;
414 	spin_unlock_irqrestore(&iboe->lock, flags);
415 	return real_index;
416 }
417 
418 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
419 				    sizeof(((type *)0)->fld) <= (sz))
420 
421 static int mlx4_ib_query_device(struct ib_device *ibdev,
422 				struct ib_device_attr *props,
423 				struct ib_udata *uhw)
424 {
425 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
426 	struct ib_smp *in_mad  = NULL;
427 	struct ib_smp *out_mad = NULL;
428 	int err;
429 	int have_ib_ports;
430 	struct mlx4_uverbs_ex_query_device cmd;
431 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
432 	struct mlx4_clock_params clock_params;
433 
434 	if (uhw->inlen) {
435 		if (uhw->inlen < sizeof(cmd))
436 			return -EINVAL;
437 
438 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
439 		if (err)
440 			return err;
441 
442 		if (cmd.comp_mask)
443 			return -EINVAL;
444 
445 		if (cmd.reserved)
446 			return -EINVAL;
447 	}
448 
449 	resp.response_length = offsetof(typeof(resp), response_length) +
450 		sizeof(resp.response_length);
451 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
452 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
453 	err = -ENOMEM;
454 	if (!in_mad || !out_mad)
455 		goto out;
456 
457 	init_query_mad(in_mad);
458 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
459 
460 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
461 			   1, NULL, NULL, in_mad, out_mad);
462 	if (err)
463 		goto out;
464 
465 	memset(props, 0, sizeof *props);
466 
467 	have_ib_ports = num_ib_ports(dev->dev);
468 
469 	props->fw_ver = dev->dev->caps.fw_ver;
470 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
471 		IB_DEVICE_PORT_ACTIVE_EVENT		|
472 		IB_DEVICE_SYS_IMAGE_GUID		|
473 		IB_DEVICE_RC_RNR_NAK_GEN		|
474 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
475 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
476 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
477 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
478 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
479 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
480 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
481 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
482 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
483 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
484 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
485 	if (dev->dev->caps.max_gso_sz &&
486 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
487 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
488 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
489 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
490 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
491 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
492 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
493 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
494 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
495 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
496 		props->device_cap_flags |= IB_DEVICE_XRC;
497 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
498 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
499 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
500 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
501 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
502 		else
503 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
504 	}
505 	if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
506 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
507 
508 	props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
509 
510 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
511 		0xffffff;
512 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
513 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
514 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
515 
516 	props->max_mr_size	   = ~0ull;
517 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
518 	props->max_qp		   = dev->dev->quotas.qp;
519 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
520 	props->max_send_sge =
521 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
522 	props->max_recv_sge =
523 		min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
524 	props->max_sge_rd = MLX4_MAX_SGE_RD;
525 	props->max_cq		   = dev->dev->quotas.cq;
526 	props->max_cqe		   = dev->dev->caps.max_cqes;
527 	props->max_mr		   = dev->dev->quotas.mpt;
528 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
529 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
530 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
531 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
532 	props->max_srq		   = dev->dev->quotas.srq;
533 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
534 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
535 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
536 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
537 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
538 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
539 	props->masked_atomic_cap   = props->atomic_cap;
540 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
541 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
542 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
543 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
544 					   props->max_mcast_grp;
545 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
546 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
547 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
548 	props->max_ah = INT_MAX;
549 
550 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
551 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
552 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
553 			props->rss_caps.max_rwq_indirection_tables =
554 				props->max_qp;
555 			props->rss_caps.max_rwq_indirection_table_size =
556 				dev->dev->caps.max_rss_tbl_sz;
557 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
558 			props->max_wq_type_rq = props->max_qp;
559 		}
560 
561 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
562 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
563 	}
564 
565 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
566 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
567 
568 	if (!mlx4_is_slave(dev->dev))
569 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
570 
571 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
572 		resp.response_length += sizeof(resp.hca_core_clock_offset);
573 		if (!err && !mlx4_is_slave(dev->dev)) {
574 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
575 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
576 		}
577 	}
578 
579 	if (uhw->outlen >= resp.response_length +
580 	    sizeof(resp.max_inl_recv_sz)) {
581 		resp.response_length += sizeof(resp.max_inl_recv_sz);
582 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
583 			sizeof(struct mlx4_wqe_data_seg);
584 	}
585 
586 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
587 		if (props->rss_caps.supported_qpts) {
588 			resp.rss_caps.rx_hash_function =
589 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
590 
591 			resp.rss_caps.rx_hash_fields_mask =
592 				MLX4_IB_RX_HASH_SRC_IPV4 |
593 				MLX4_IB_RX_HASH_DST_IPV4 |
594 				MLX4_IB_RX_HASH_SRC_IPV6 |
595 				MLX4_IB_RX_HASH_DST_IPV6 |
596 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
597 				MLX4_IB_RX_HASH_DST_PORT_TCP |
598 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
599 				MLX4_IB_RX_HASH_DST_PORT_UDP;
600 
601 			if (dev->dev->caps.tunnel_offload_mode ==
602 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
603 				resp.rss_caps.rx_hash_fields_mask |=
604 					MLX4_IB_RX_HASH_INNER;
605 		}
606 		resp.response_length = offsetof(typeof(resp), rss_caps) +
607 				       sizeof(resp.rss_caps);
608 	}
609 
610 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
611 		if (dev->dev->caps.max_gso_sz &&
612 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
613 		    IB_LINK_LAYER_ETHERNET) ||
614 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
615 		    IB_LINK_LAYER_ETHERNET))) {
616 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
617 			resp.tso_caps.supported_qpts |=
618 				1 << IB_QPT_RAW_PACKET;
619 		}
620 		resp.response_length = offsetof(typeof(resp), tso_caps) +
621 				       sizeof(resp.tso_caps);
622 	}
623 
624 	if (uhw->outlen) {
625 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
626 		if (err)
627 			goto out;
628 	}
629 out:
630 	kfree(in_mad);
631 	kfree(out_mad);
632 
633 	return err;
634 }
635 
636 static enum rdma_link_layer
637 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
638 {
639 	struct mlx4_dev *dev = to_mdev(device)->dev;
640 
641 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
642 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
643 }
644 
645 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
646 			      struct ib_port_attr *props, int netw_view)
647 {
648 	struct ib_smp *in_mad  = NULL;
649 	struct ib_smp *out_mad = NULL;
650 	int ext_active_speed;
651 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
652 	int err = -ENOMEM;
653 
654 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
655 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
656 	if (!in_mad || !out_mad)
657 		goto out;
658 
659 	init_query_mad(in_mad);
660 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
661 	in_mad->attr_mod = cpu_to_be32(port);
662 
663 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
664 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
665 
666 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
667 				in_mad, out_mad);
668 	if (err)
669 		goto out;
670 
671 
672 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
673 	props->lmc		= out_mad->data[34] & 0x7;
674 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
675 	props->sm_sl		= out_mad->data[36] & 0xf;
676 	props->state		= out_mad->data[32] & 0xf;
677 	props->phys_state	= out_mad->data[33] >> 4;
678 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
679 	if (netw_view)
680 		props->gid_tbl_len = out_mad->data[50];
681 	else
682 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
683 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
684 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
685 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
686 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
687 	props->active_width	= out_mad->data[31] & 0xf;
688 	props->active_speed	= out_mad->data[35] >> 4;
689 	props->max_mtu		= out_mad->data[41] & 0xf;
690 	props->active_mtu	= out_mad->data[36] >> 4;
691 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
692 	props->max_vl_num	= out_mad->data[37] >> 4;
693 	props->init_type_reply	= out_mad->data[41] >> 4;
694 
695 	/* Check if extended speeds (EDR/FDR/...) are supported */
696 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
697 		ext_active_speed = out_mad->data[62] >> 4;
698 
699 		switch (ext_active_speed) {
700 		case 1:
701 			props->active_speed = IB_SPEED_FDR;
702 			break;
703 		case 2:
704 			props->active_speed = IB_SPEED_EDR;
705 			break;
706 		}
707 	}
708 
709 	/* If reported active speed is QDR, check if is FDR-10 */
710 	if (props->active_speed == IB_SPEED_QDR) {
711 		init_query_mad(in_mad);
712 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
713 		in_mad->attr_mod = cpu_to_be32(port);
714 
715 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
716 				   NULL, NULL, in_mad, out_mad);
717 		if (err)
718 			goto out;
719 
720 		/* Checking LinkSpeedActive for FDR-10 */
721 		if (out_mad->data[15] & 0x1)
722 			props->active_speed = IB_SPEED_FDR10;
723 	}
724 
725 	/* Avoid wrong speed value returned by FW if the IB link is down. */
726 	if (props->state == IB_PORT_DOWN)
727 		 props->active_speed = IB_SPEED_SDR;
728 
729 out:
730 	kfree(in_mad);
731 	kfree(out_mad);
732 	return err;
733 }
734 
735 static u8 state_to_phys_state(enum ib_port_state state)
736 {
737 	return state == IB_PORT_ACTIVE ? 5 : 3;
738 }
739 
740 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
741 			       struct ib_port_attr *props)
742 {
743 
744 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
745 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
746 	struct net_device *ndev;
747 	enum ib_mtu tmp;
748 	struct mlx4_cmd_mailbox *mailbox;
749 	int err = 0;
750 	int is_bonded = mlx4_is_bonded(mdev->dev);
751 
752 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
753 	if (IS_ERR(mailbox))
754 		return PTR_ERR(mailbox);
755 
756 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
757 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
758 			   MLX4_CMD_WRAPPED);
759 	if (err)
760 		goto out;
761 
762 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
763 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
764 					   IB_WIDTH_4X : IB_WIDTH_1X;
765 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
766 					   IB_SPEED_FDR : IB_SPEED_QDR;
767 	props->port_cap_flags	= IB_PORT_CM_SUP;
768 	props->ip_gids = true;
769 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
770 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
771 	props->pkey_tbl_len	= 1;
772 	props->max_mtu		= IB_MTU_4096;
773 	props->max_vl_num	= 2;
774 	props->state		= IB_PORT_DOWN;
775 	props->phys_state	= state_to_phys_state(props->state);
776 	props->active_mtu	= IB_MTU_256;
777 	spin_lock_bh(&iboe->lock);
778 	ndev = iboe->netdevs[port - 1];
779 	if (ndev && is_bonded) {
780 		rcu_read_lock(); /* required to get upper dev */
781 		ndev = netdev_master_upper_dev_get_rcu(ndev);
782 		rcu_read_unlock();
783 	}
784 	if (!ndev)
785 		goto out_unlock;
786 
787 	tmp = iboe_get_mtu(ndev->mtu);
788 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
789 
790 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
791 					IB_PORT_ACTIVE : IB_PORT_DOWN;
792 	props->phys_state	= state_to_phys_state(props->state);
793 out_unlock:
794 	spin_unlock_bh(&iboe->lock);
795 out:
796 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
797 	return err;
798 }
799 
800 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
801 			 struct ib_port_attr *props, int netw_view)
802 {
803 	int err;
804 
805 	/* props being zeroed by the caller, avoid zeroing it here */
806 
807 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
808 		ib_link_query_port(ibdev, port, props, netw_view) :
809 				eth_link_query_port(ibdev, port, props);
810 
811 	return err;
812 }
813 
814 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
815 			      struct ib_port_attr *props)
816 {
817 	/* returns host view */
818 	return __mlx4_ib_query_port(ibdev, port, props, 0);
819 }
820 
821 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
822 			union ib_gid *gid, int netw_view)
823 {
824 	struct ib_smp *in_mad  = NULL;
825 	struct ib_smp *out_mad = NULL;
826 	int err = -ENOMEM;
827 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
828 	int clear = 0;
829 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
830 
831 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
832 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
833 	if (!in_mad || !out_mad)
834 		goto out;
835 
836 	init_query_mad(in_mad);
837 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
838 	in_mad->attr_mod = cpu_to_be32(port);
839 
840 	if (mlx4_is_mfunc(dev->dev) && netw_view)
841 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
842 
843 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
844 	if (err)
845 		goto out;
846 
847 	memcpy(gid->raw, out_mad->data + 8, 8);
848 
849 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
850 		if (index) {
851 			/* For any index > 0, return the null guid */
852 			err = 0;
853 			clear = 1;
854 			goto out;
855 		}
856 	}
857 
858 	init_query_mad(in_mad);
859 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
860 	in_mad->attr_mod = cpu_to_be32(index / 8);
861 
862 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
863 			   NULL, NULL, in_mad, out_mad);
864 	if (err)
865 		goto out;
866 
867 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
868 
869 out:
870 	if (clear)
871 		memset(gid->raw + 8, 0, 8);
872 	kfree(in_mad);
873 	kfree(out_mad);
874 	return err;
875 }
876 
877 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
878 			     union ib_gid *gid)
879 {
880 	if (rdma_protocol_ib(ibdev, port))
881 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
882 	return 0;
883 }
884 
885 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
886 {
887 	union sl2vl_tbl_to_u64 sl2vl64;
888 	struct ib_smp *in_mad  = NULL;
889 	struct ib_smp *out_mad = NULL;
890 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
891 	int err = -ENOMEM;
892 	int jj;
893 
894 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
895 		*sl2vl_tbl = 0;
896 		return 0;
897 	}
898 
899 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
900 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
901 	if (!in_mad || !out_mad)
902 		goto out;
903 
904 	init_query_mad(in_mad);
905 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
906 	in_mad->attr_mod = 0;
907 
908 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
909 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
910 
911 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
912 			   in_mad, out_mad);
913 	if (err)
914 		goto out;
915 
916 	for (jj = 0; jj < 8; jj++)
917 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
918 	*sl2vl_tbl = sl2vl64.sl64;
919 
920 out:
921 	kfree(in_mad);
922 	kfree(out_mad);
923 	return err;
924 }
925 
926 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
927 {
928 	u64 sl2vl;
929 	int i;
930 	int err;
931 
932 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
933 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
934 			continue;
935 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
936 		if (err) {
937 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
938 			       i, err);
939 			sl2vl = 0;
940 		}
941 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
942 	}
943 }
944 
945 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
946 			 u16 *pkey, int netw_view)
947 {
948 	struct ib_smp *in_mad  = NULL;
949 	struct ib_smp *out_mad = NULL;
950 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
951 	int err = -ENOMEM;
952 
953 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
954 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
955 	if (!in_mad || !out_mad)
956 		goto out;
957 
958 	init_query_mad(in_mad);
959 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
960 	in_mad->attr_mod = cpu_to_be32(index / 32);
961 
962 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
963 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
964 
965 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
966 			   in_mad, out_mad);
967 	if (err)
968 		goto out;
969 
970 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
971 
972 out:
973 	kfree(in_mad);
974 	kfree(out_mad);
975 	return err;
976 }
977 
978 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
979 {
980 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
981 }
982 
983 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
984 				 struct ib_device_modify *props)
985 {
986 	struct mlx4_cmd_mailbox *mailbox;
987 	unsigned long flags;
988 
989 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
990 		return -EOPNOTSUPP;
991 
992 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
993 		return 0;
994 
995 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
996 		return -EOPNOTSUPP;
997 
998 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
999 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1000 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1001 
1002 	/*
1003 	 * If possible, pass node desc to FW, so it can generate
1004 	 * a 144 trap.  If cmd fails, just ignore.
1005 	 */
1006 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1007 	if (IS_ERR(mailbox))
1008 		return 0;
1009 
1010 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1011 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1012 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1013 
1014 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1015 
1016 	return 0;
1017 }
1018 
1019 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1020 			    u32 cap_mask)
1021 {
1022 	struct mlx4_cmd_mailbox *mailbox;
1023 	int err;
1024 
1025 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1026 	if (IS_ERR(mailbox))
1027 		return PTR_ERR(mailbox);
1028 
1029 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1030 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1031 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1032 	} else {
1033 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1034 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1035 	}
1036 
1037 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1038 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1039 		       MLX4_CMD_WRAPPED);
1040 
1041 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1042 	return err;
1043 }
1044 
1045 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1046 			       struct ib_port_modify *props)
1047 {
1048 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1049 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1050 	struct ib_port_attr attr;
1051 	u32 cap_mask;
1052 	int err;
1053 
1054 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1055 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1056 	 * violations and port capabilities are not meaningful.
1057 	 */
1058 	if (is_eth)
1059 		return 0;
1060 
1061 	mutex_lock(&mdev->cap_mask_mutex);
1062 
1063 	err = ib_query_port(ibdev, port, &attr);
1064 	if (err)
1065 		goto out;
1066 
1067 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1068 		~props->clr_port_cap_mask;
1069 
1070 	err = mlx4_ib_SET_PORT(mdev, port,
1071 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1072 			       cap_mask);
1073 
1074 out:
1075 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1076 	return err;
1077 }
1078 
1079 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1080 						  struct ib_udata *udata)
1081 {
1082 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1083 	struct mlx4_ib_ucontext *context;
1084 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1085 	struct mlx4_ib_alloc_ucontext_resp resp;
1086 	int err;
1087 
1088 	if (!dev->ib_active)
1089 		return ERR_PTR(-EAGAIN);
1090 
1091 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1092 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1093 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1094 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1095 	} else {
1096 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1097 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1098 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1099 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1100 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1101 	}
1102 
1103 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1104 	if (!context)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1108 	if (err) {
1109 		kfree(context);
1110 		return ERR_PTR(err);
1111 	}
1112 
1113 	INIT_LIST_HEAD(&context->db_page_list);
1114 	mutex_init(&context->db_page_mutex);
1115 
1116 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1117 	mutex_init(&context->wqn_ranges_mutex);
1118 
1119 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1120 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1121 	else
1122 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1123 
1124 	if (err) {
1125 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1126 		kfree(context);
1127 		return ERR_PTR(-EFAULT);
1128 	}
1129 
1130 	return &context->ibucontext;
1131 }
1132 
1133 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1134 {
1135 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1136 
1137 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1138 	kfree(context);
1139 
1140 	return 0;
1141 }
1142 
1143 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1144 {
1145 }
1146 
1147 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1148 {
1149 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1150 
1151 	switch (vma->vm_pgoff) {
1152 	case 0:
1153 		return rdma_user_mmap_io(context, vma,
1154 					 to_mucontext(context)->uar.pfn,
1155 					 PAGE_SIZE,
1156 					 pgprot_noncached(vma->vm_page_prot));
1157 
1158 	case 1:
1159 		if (dev->dev->caps.bf_reg_size == 0)
1160 			return -EINVAL;
1161 		return rdma_user_mmap_io(
1162 			context, vma,
1163 			to_mucontext(context)->uar.pfn +
1164 				dev->dev->caps.num_uars,
1165 			PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot));
1166 
1167 	case 3: {
1168 		struct mlx4_clock_params params;
1169 		int ret;
1170 
1171 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1172 		if (ret)
1173 			return ret;
1174 
1175 		return rdma_user_mmap_io(
1176 			context, vma,
1177 			(pci_resource_start(dev->dev->persist->pdev,
1178 					    params.bar) +
1179 			 params.offset) >>
1180 				PAGE_SHIFT,
1181 			PAGE_SIZE, pgprot_noncached(vma->vm_page_prot));
1182 	}
1183 
1184 	default:
1185 		return -EINVAL;
1186 	}
1187 }
1188 
1189 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1190 				      struct ib_ucontext *context,
1191 				      struct ib_udata *udata)
1192 {
1193 	struct mlx4_ib_pd *pd;
1194 	int err;
1195 
1196 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1197 	if (!pd)
1198 		return ERR_PTR(-ENOMEM);
1199 
1200 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1201 	if (err) {
1202 		kfree(pd);
1203 		return ERR_PTR(err);
1204 	}
1205 
1206 	if (context)
1207 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1208 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1209 			kfree(pd);
1210 			return ERR_PTR(-EFAULT);
1211 		}
1212 	return &pd->ibpd;
1213 }
1214 
1215 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1216 {
1217 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1218 	kfree(pd);
1219 
1220 	return 0;
1221 }
1222 
1223 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1224 					  struct ib_ucontext *context,
1225 					  struct ib_udata *udata)
1226 {
1227 	struct mlx4_ib_xrcd *xrcd;
1228 	struct ib_cq_init_attr cq_attr = {};
1229 	int err;
1230 
1231 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1232 		return ERR_PTR(-ENOSYS);
1233 
1234 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1235 	if (!xrcd)
1236 		return ERR_PTR(-ENOMEM);
1237 
1238 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1239 	if (err)
1240 		goto err1;
1241 
1242 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1243 	if (IS_ERR(xrcd->pd)) {
1244 		err = PTR_ERR(xrcd->pd);
1245 		goto err2;
1246 	}
1247 
1248 	cq_attr.cqe = 1;
1249 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1250 	if (IS_ERR(xrcd->cq)) {
1251 		err = PTR_ERR(xrcd->cq);
1252 		goto err3;
1253 	}
1254 
1255 	return &xrcd->ibxrcd;
1256 
1257 err3:
1258 	ib_dealloc_pd(xrcd->pd);
1259 err2:
1260 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1261 err1:
1262 	kfree(xrcd);
1263 	return ERR_PTR(err);
1264 }
1265 
1266 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1267 {
1268 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1269 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1270 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1271 	kfree(xrcd);
1272 
1273 	return 0;
1274 }
1275 
1276 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1277 {
1278 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1279 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1280 	struct mlx4_ib_gid_entry *ge;
1281 
1282 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1283 	if (!ge)
1284 		return -ENOMEM;
1285 
1286 	ge->gid = *gid;
1287 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1288 		ge->port = mqp->port;
1289 		ge->added = 1;
1290 	}
1291 
1292 	mutex_lock(&mqp->mutex);
1293 	list_add_tail(&ge->list, &mqp->gid_list);
1294 	mutex_unlock(&mqp->mutex);
1295 
1296 	return 0;
1297 }
1298 
1299 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1300 					  struct mlx4_ib_counters *ctr_table)
1301 {
1302 	struct counter_index *counter, *tmp_count;
1303 
1304 	mutex_lock(&ctr_table->mutex);
1305 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1306 				 list) {
1307 		if (counter->allocated)
1308 			mlx4_counter_free(ibdev->dev, counter->index);
1309 		list_del(&counter->list);
1310 		kfree(counter);
1311 	}
1312 	mutex_unlock(&ctr_table->mutex);
1313 }
1314 
1315 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1316 		   union ib_gid *gid)
1317 {
1318 	struct net_device *ndev;
1319 	int ret = 0;
1320 
1321 	if (!mqp->port)
1322 		return 0;
1323 
1324 	spin_lock_bh(&mdev->iboe.lock);
1325 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1326 	if (ndev)
1327 		dev_hold(ndev);
1328 	spin_unlock_bh(&mdev->iboe.lock);
1329 
1330 	if (ndev) {
1331 		ret = 1;
1332 		dev_put(ndev);
1333 	}
1334 
1335 	return ret;
1336 }
1337 
1338 struct mlx4_ib_steering {
1339 	struct list_head list;
1340 	struct mlx4_flow_reg_id reg_id;
1341 	union ib_gid gid;
1342 };
1343 
1344 #define LAST_ETH_FIELD vlan_tag
1345 #define LAST_IB_FIELD sl
1346 #define LAST_IPV4_FIELD dst_ip
1347 #define LAST_TCP_UDP_FIELD src_port
1348 
1349 /* Field is the last supported field */
1350 #define FIELDS_NOT_SUPPORTED(filter, field)\
1351 	memchr_inv((void *)&filter.field  +\
1352 		   sizeof(filter.field), 0,\
1353 		   sizeof(filter) -\
1354 		   offsetof(typeof(filter), field) -\
1355 		   sizeof(filter.field))
1356 
1357 static int parse_flow_attr(struct mlx4_dev *dev,
1358 			   u32 qp_num,
1359 			   union ib_flow_spec *ib_spec,
1360 			   struct _rule_hw *mlx4_spec)
1361 {
1362 	enum mlx4_net_trans_rule_id type;
1363 
1364 	switch (ib_spec->type) {
1365 	case IB_FLOW_SPEC_ETH:
1366 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1367 			return -ENOTSUPP;
1368 
1369 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1370 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1371 		       ETH_ALEN);
1372 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1373 		       ETH_ALEN);
1374 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1375 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1376 		break;
1377 	case IB_FLOW_SPEC_IB:
1378 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1379 			return -ENOTSUPP;
1380 
1381 		type = MLX4_NET_TRANS_RULE_ID_IB;
1382 		mlx4_spec->ib.l3_qpn =
1383 			cpu_to_be32(qp_num);
1384 		mlx4_spec->ib.qpn_mask =
1385 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1386 		break;
1387 
1388 
1389 	case IB_FLOW_SPEC_IPV4:
1390 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1391 			return -ENOTSUPP;
1392 
1393 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1394 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1395 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1396 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1397 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1398 		break;
1399 
1400 	case IB_FLOW_SPEC_TCP:
1401 	case IB_FLOW_SPEC_UDP:
1402 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1403 			return -ENOTSUPP;
1404 
1405 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1406 					MLX4_NET_TRANS_RULE_ID_TCP :
1407 					MLX4_NET_TRANS_RULE_ID_UDP;
1408 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1409 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1410 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1411 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1412 		break;
1413 
1414 	default:
1415 		return -EINVAL;
1416 	}
1417 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1418 	    mlx4_hw_rule_sz(dev, type) < 0)
1419 		return -EINVAL;
1420 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1421 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1422 	return mlx4_hw_rule_sz(dev, type);
1423 }
1424 
1425 struct default_rules {
1426 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1427 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1428 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1429 	__u8  link_layer;
1430 };
1431 static const struct default_rules default_table[] = {
1432 	{
1433 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1434 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1435 		.rules_create_list = {IB_FLOW_SPEC_IB},
1436 		.link_layer = IB_LINK_LAYER_INFINIBAND
1437 	}
1438 };
1439 
1440 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1441 					 struct ib_flow_attr *flow_attr)
1442 {
1443 	int i, j, k;
1444 	void *ib_flow;
1445 	const struct default_rules *pdefault_rules = default_table;
1446 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1447 
1448 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1449 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1450 		memset(&field_types, 0, sizeof(field_types));
1451 
1452 		if (link_layer != pdefault_rules->link_layer)
1453 			continue;
1454 
1455 		ib_flow = flow_attr + 1;
1456 		/* we assume the specs are sorted */
1457 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1458 		     j < flow_attr->num_of_specs; k++) {
1459 			union ib_flow_spec *current_flow =
1460 				(union ib_flow_spec *)ib_flow;
1461 
1462 			/* same layer but different type */
1463 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1464 			     (pdefault_rules->mandatory_fields[k] &
1465 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1466 			    (current_flow->type !=
1467 			     pdefault_rules->mandatory_fields[k]))
1468 				goto out;
1469 
1470 			/* same layer, try match next one */
1471 			if (current_flow->type ==
1472 			    pdefault_rules->mandatory_fields[k]) {
1473 				j++;
1474 				ib_flow +=
1475 					((union ib_flow_spec *)ib_flow)->size;
1476 			}
1477 		}
1478 
1479 		ib_flow = flow_attr + 1;
1480 		for (j = 0; j < flow_attr->num_of_specs;
1481 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1482 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1483 				/* same layer and same type */
1484 				if (((union ib_flow_spec *)ib_flow)->type ==
1485 				    pdefault_rules->mandatory_not_fields[k])
1486 					goto out;
1487 
1488 		return i;
1489 	}
1490 out:
1491 	return -1;
1492 }
1493 
1494 static int __mlx4_ib_create_default_rules(
1495 		struct mlx4_ib_dev *mdev,
1496 		struct ib_qp *qp,
1497 		const struct default_rules *pdefault_rules,
1498 		struct _rule_hw *mlx4_spec) {
1499 	int size = 0;
1500 	int i;
1501 
1502 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1503 		int ret;
1504 		union ib_flow_spec ib_spec;
1505 		switch (pdefault_rules->rules_create_list[i]) {
1506 		case 0:
1507 			/* no rule */
1508 			continue;
1509 		case IB_FLOW_SPEC_IB:
1510 			ib_spec.type = IB_FLOW_SPEC_IB;
1511 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1512 
1513 			break;
1514 		default:
1515 			/* invalid rule */
1516 			return -EINVAL;
1517 		}
1518 		/* We must put empty rule, qpn is being ignored */
1519 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1520 				      mlx4_spec);
1521 		if (ret < 0) {
1522 			pr_info("invalid parsing\n");
1523 			return -EINVAL;
1524 		}
1525 
1526 		mlx4_spec = (void *)mlx4_spec + ret;
1527 		size += ret;
1528 	}
1529 	return size;
1530 }
1531 
1532 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1533 			  int domain,
1534 			  enum mlx4_net_trans_promisc_mode flow_type,
1535 			  u64 *reg_id)
1536 {
1537 	int ret, i;
1538 	int size = 0;
1539 	void *ib_flow;
1540 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1541 	struct mlx4_cmd_mailbox *mailbox;
1542 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1543 	int default_flow;
1544 
1545 	static const u16 __mlx4_domain[] = {
1546 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1547 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1548 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1549 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1550 	};
1551 
1552 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1553 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1554 		return -EINVAL;
1555 	}
1556 
1557 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1558 		pr_err("Invalid domain value %d\n", domain);
1559 		return -EINVAL;
1560 	}
1561 
1562 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1563 		return -EINVAL;
1564 
1565 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1566 	if (IS_ERR(mailbox))
1567 		return PTR_ERR(mailbox);
1568 	ctrl = mailbox->buf;
1569 
1570 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1571 				 flow_attr->priority);
1572 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1573 	ctrl->port = flow_attr->port;
1574 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1575 
1576 	ib_flow = flow_attr + 1;
1577 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1578 	/* Add default flows */
1579 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1580 	if (default_flow >= 0) {
1581 		ret = __mlx4_ib_create_default_rules(
1582 				mdev, qp, default_table + default_flow,
1583 				mailbox->buf + size);
1584 		if (ret < 0) {
1585 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1586 			return -EINVAL;
1587 		}
1588 		size += ret;
1589 	}
1590 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1591 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1592 				      mailbox->buf + size);
1593 		if (ret < 0) {
1594 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1595 			return -EINVAL;
1596 		}
1597 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1598 		size += ret;
1599 	}
1600 
1601 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1602 	    flow_attr->num_of_specs == 1) {
1603 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1604 		enum ib_flow_spec_type header_spec =
1605 			((union ib_flow_spec *)(flow_attr + 1))->type;
1606 
1607 		if (header_spec == IB_FLOW_SPEC_ETH)
1608 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1609 	}
1610 
1611 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1612 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1613 			   MLX4_CMD_NATIVE);
1614 	if (ret == -ENOMEM)
1615 		pr_err("mcg table is full. Fail to register network rule.\n");
1616 	else if (ret == -ENXIO)
1617 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1618 	else if (ret)
1619 		pr_err("Invalid argument. Fail to register network rule.\n");
1620 
1621 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1622 	return ret;
1623 }
1624 
1625 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1626 {
1627 	int err;
1628 	err = mlx4_cmd(dev, reg_id, 0, 0,
1629 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1630 		       MLX4_CMD_NATIVE);
1631 	if (err)
1632 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1633 		       reg_id);
1634 	return err;
1635 }
1636 
1637 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1638 				    u64 *reg_id)
1639 {
1640 	void *ib_flow;
1641 	union ib_flow_spec *ib_spec;
1642 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1643 	int err = 0;
1644 
1645 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1646 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1647 		return 0; /* do nothing */
1648 
1649 	ib_flow = flow_attr + 1;
1650 	ib_spec = (union ib_flow_spec *)ib_flow;
1651 
1652 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1653 		return 0; /* do nothing */
1654 
1655 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1656 				    flow_attr->port, qp->qp_num,
1657 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1658 				    reg_id);
1659 	return err;
1660 }
1661 
1662 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1663 				      struct ib_flow_attr *flow_attr,
1664 				      enum mlx4_net_trans_promisc_mode *type)
1665 {
1666 	int err = 0;
1667 
1668 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1669 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1670 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1671 		return -EOPNOTSUPP;
1672 	}
1673 
1674 	if (flow_attr->num_of_specs == 0) {
1675 		type[0] = MLX4_FS_MC_SNIFFER;
1676 		type[1] = MLX4_FS_UC_SNIFFER;
1677 	} else {
1678 		union ib_flow_spec *ib_spec;
1679 
1680 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1681 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1682 			return -EINVAL;
1683 
1684 		/* if all is zero than MC and UC */
1685 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1686 			type[0] = MLX4_FS_MC_SNIFFER;
1687 			type[1] = MLX4_FS_UC_SNIFFER;
1688 		} else {
1689 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1690 					    ib_spec->eth.mask.dst_mac[1],
1691 					    ib_spec->eth.mask.dst_mac[2],
1692 					    ib_spec->eth.mask.dst_mac[3],
1693 					    ib_spec->eth.mask.dst_mac[4],
1694 					    ib_spec->eth.mask.dst_mac[5]};
1695 
1696 			/* Above xor was only on MC bit, non empty mask is valid
1697 			 * only if this bit is set and rest are zero.
1698 			 */
1699 			if (!is_zero_ether_addr(&mac[0]))
1700 				return -EINVAL;
1701 
1702 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1703 				type[0] = MLX4_FS_MC_SNIFFER;
1704 			else
1705 				type[0] = MLX4_FS_UC_SNIFFER;
1706 		}
1707 	}
1708 
1709 	return err;
1710 }
1711 
1712 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1713 				    struct ib_flow_attr *flow_attr,
1714 				    int domain, struct ib_udata *udata)
1715 {
1716 	int err = 0, i = 0, j = 0;
1717 	struct mlx4_ib_flow *mflow;
1718 	enum mlx4_net_trans_promisc_mode type[2];
1719 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1720 	int is_bonded = mlx4_is_bonded(dev);
1721 
1722 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1723 		return ERR_PTR(-EINVAL);
1724 
1725 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1726 		return ERR_PTR(-EOPNOTSUPP);
1727 
1728 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1729 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1730 		return ERR_PTR(-EOPNOTSUPP);
1731 
1732 	if (udata &&
1733 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1734 		return ERR_PTR(-EOPNOTSUPP);
1735 
1736 	memset(type, 0, sizeof(type));
1737 
1738 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1739 	if (!mflow) {
1740 		err = -ENOMEM;
1741 		goto err_free;
1742 	}
1743 
1744 	switch (flow_attr->type) {
1745 	case IB_FLOW_ATTR_NORMAL:
1746 		/* If dont trap flag (continue match) is set, under specific
1747 		 * condition traffic be replicated to given qp,
1748 		 * without stealing it
1749 		 */
1750 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1751 			err = mlx4_ib_add_dont_trap_rule(dev,
1752 							 flow_attr,
1753 							 type);
1754 			if (err)
1755 				goto err_free;
1756 		} else {
1757 			type[0] = MLX4_FS_REGULAR;
1758 		}
1759 		break;
1760 
1761 	case IB_FLOW_ATTR_ALL_DEFAULT:
1762 		type[0] = MLX4_FS_ALL_DEFAULT;
1763 		break;
1764 
1765 	case IB_FLOW_ATTR_MC_DEFAULT:
1766 		type[0] = MLX4_FS_MC_DEFAULT;
1767 		break;
1768 
1769 	case IB_FLOW_ATTR_SNIFFER:
1770 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1771 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1772 		break;
1773 
1774 	default:
1775 		err = -EINVAL;
1776 		goto err_free;
1777 	}
1778 
1779 	while (i < ARRAY_SIZE(type) && type[i]) {
1780 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1781 					    &mflow->reg_id[i].id);
1782 		if (err)
1783 			goto err_create_flow;
1784 		if (is_bonded) {
1785 			/* Application always sees one port so the mirror rule
1786 			 * must be on port #2
1787 			 */
1788 			flow_attr->port = 2;
1789 			err = __mlx4_ib_create_flow(qp, flow_attr,
1790 						    domain, type[j],
1791 						    &mflow->reg_id[j].mirror);
1792 			flow_attr->port = 1;
1793 			if (err)
1794 				goto err_create_flow;
1795 			j++;
1796 		}
1797 
1798 		i++;
1799 	}
1800 
1801 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1802 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1803 					       &mflow->reg_id[i].id);
1804 		if (err)
1805 			goto err_create_flow;
1806 
1807 		if (is_bonded) {
1808 			flow_attr->port = 2;
1809 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1810 						       &mflow->reg_id[j].mirror);
1811 			flow_attr->port = 1;
1812 			if (err)
1813 				goto err_create_flow;
1814 			j++;
1815 		}
1816 		/* function to create mirror rule */
1817 		i++;
1818 	}
1819 
1820 	return &mflow->ibflow;
1821 
1822 err_create_flow:
1823 	while (i) {
1824 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1825 					     mflow->reg_id[i].id);
1826 		i--;
1827 	}
1828 
1829 	while (j) {
1830 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1831 					     mflow->reg_id[j].mirror);
1832 		j--;
1833 	}
1834 err_free:
1835 	kfree(mflow);
1836 	return ERR_PTR(err);
1837 }
1838 
1839 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1840 {
1841 	int err, ret = 0;
1842 	int i = 0;
1843 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1844 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1845 
1846 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1847 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1848 		if (err)
1849 			ret = err;
1850 		if (mflow->reg_id[i].mirror) {
1851 			err = __mlx4_ib_destroy_flow(mdev->dev,
1852 						     mflow->reg_id[i].mirror);
1853 			if (err)
1854 				ret = err;
1855 		}
1856 		i++;
1857 	}
1858 
1859 	kfree(mflow);
1860 	return ret;
1861 }
1862 
1863 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1864 {
1865 	int err;
1866 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1867 	struct mlx4_dev	*dev = mdev->dev;
1868 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1869 	struct mlx4_ib_steering *ib_steering = NULL;
1870 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1871 	struct mlx4_flow_reg_id	reg_id;
1872 
1873 	if (mdev->dev->caps.steering_mode ==
1874 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1875 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1876 		if (!ib_steering)
1877 			return -ENOMEM;
1878 	}
1879 
1880 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1881 				    !!(mqp->flags &
1882 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1883 				    prot, &reg_id.id);
1884 	if (err) {
1885 		pr_err("multicast attach op failed, err %d\n", err);
1886 		goto err_malloc;
1887 	}
1888 
1889 	reg_id.mirror = 0;
1890 	if (mlx4_is_bonded(dev)) {
1891 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1892 					    (mqp->port == 1) ? 2 : 1,
1893 					    !!(mqp->flags &
1894 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1895 					    prot, &reg_id.mirror);
1896 		if (err)
1897 			goto err_add;
1898 	}
1899 
1900 	err = add_gid_entry(ibqp, gid);
1901 	if (err)
1902 		goto err_add;
1903 
1904 	if (ib_steering) {
1905 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1906 		ib_steering->reg_id = reg_id;
1907 		mutex_lock(&mqp->mutex);
1908 		list_add(&ib_steering->list, &mqp->steering_rules);
1909 		mutex_unlock(&mqp->mutex);
1910 	}
1911 	return 0;
1912 
1913 err_add:
1914 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1915 			      prot, reg_id.id);
1916 	if (reg_id.mirror)
1917 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1918 				      prot, reg_id.mirror);
1919 err_malloc:
1920 	kfree(ib_steering);
1921 
1922 	return err;
1923 }
1924 
1925 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1926 {
1927 	struct mlx4_ib_gid_entry *ge;
1928 	struct mlx4_ib_gid_entry *tmp;
1929 	struct mlx4_ib_gid_entry *ret = NULL;
1930 
1931 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1932 		if (!memcmp(raw, ge->gid.raw, 16)) {
1933 			ret = ge;
1934 			break;
1935 		}
1936 	}
1937 
1938 	return ret;
1939 }
1940 
1941 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1942 {
1943 	int err;
1944 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1945 	struct mlx4_dev *dev = mdev->dev;
1946 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1947 	struct net_device *ndev;
1948 	struct mlx4_ib_gid_entry *ge;
1949 	struct mlx4_flow_reg_id reg_id = {0, 0};
1950 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1951 
1952 	if (mdev->dev->caps.steering_mode ==
1953 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1954 		struct mlx4_ib_steering *ib_steering;
1955 
1956 		mutex_lock(&mqp->mutex);
1957 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1958 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1959 				list_del(&ib_steering->list);
1960 				break;
1961 			}
1962 		}
1963 		mutex_unlock(&mqp->mutex);
1964 		if (&ib_steering->list == &mqp->steering_rules) {
1965 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1966 			return -EINVAL;
1967 		}
1968 		reg_id = ib_steering->reg_id;
1969 		kfree(ib_steering);
1970 	}
1971 
1972 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1973 				    prot, reg_id.id);
1974 	if (err)
1975 		return err;
1976 
1977 	if (mlx4_is_bonded(dev)) {
1978 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1979 					    prot, reg_id.mirror);
1980 		if (err)
1981 			return err;
1982 	}
1983 
1984 	mutex_lock(&mqp->mutex);
1985 	ge = find_gid_entry(mqp, gid->raw);
1986 	if (ge) {
1987 		spin_lock_bh(&mdev->iboe.lock);
1988 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1989 		if (ndev)
1990 			dev_hold(ndev);
1991 		spin_unlock_bh(&mdev->iboe.lock);
1992 		if (ndev)
1993 			dev_put(ndev);
1994 		list_del(&ge->list);
1995 		kfree(ge);
1996 	} else
1997 		pr_warn("could not find mgid entry\n");
1998 
1999 	mutex_unlock(&mqp->mutex);
2000 
2001 	return 0;
2002 }
2003 
2004 static int init_node_data(struct mlx4_ib_dev *dev)
2005 {
2006 	struct ib_smp *in_mad  = NULL;
2007 	struct ib_smp *out_mad = NULL;
2008 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2009 	int err = -ENOMEM;
2010 
2011 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2012 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2013 	if (!in_mad || !out_mad)
2014 		goto out;
2015 
2016 	init_query_mad(in_mad);
2017 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2018 	if (mlx4_is_master(dev->dev))
2019 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2020 
2021 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2022 	if (err)
2023 		goto out;
2024 
2025 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2026 
2027 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2028 
2029 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2030 	if (err)
2031 		goto out;
2032 
2033 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2034 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2035 
2036 out:
2037 	kfree(in_mad);
2038 	kfree(out_mad);
2039 	return err;
2040 }
2041 
2042 static ssize_t hca_type_show(struct device *device,
2043 			     struct device_attribute *attr, char *buf)
2044 {
2045 	struct mlx4_ib_dev *dev =
2046 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2047 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2048 }
2049 static DEVICE_ATTR_RO(hca_type);
2050 
2051 static ssize_t hw_rev_show(struct device *device,
2052 			   struct device_attribute *attr, char *buf)
2053 {
2054 	struct mlx4_ib_dev *dev =
2055 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2056 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2057 }
2058 static DEVICE_ATTR_RO(hw_rev);
2059 
2060 static ssize_t board_id_show(struct device *device,
2061 			     struct device_attribute *attr, char *buf)
2062 {
2063 	struct mlx4_ib_dev *dev =
2064 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2065 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2066 		       dev->dev->board_id);
2067 }
2068 static DEVICE_ATTR_RO(board_id);
2069 
2070 static struct attribute *mlx4_class_attributes[] = {
2071 	&dev_attr_hw_rev.attr,
2072 	&dev_attr_hca_type.attr,
2073 	&dev_attr_board_id.attr,
2074 	NULL
2075 };
2076 
2077 static const struct attribute_group mlx4_attr_group = {
2078 	.attrs = mlx4_class_attributes,
2079 };
2080 
2081 struct diag_counter {
2082 	const char *name;
2083 	u32 offset;
2084 };
2085 
2086 #define DIAG_COUNTER(_name, _offset)			\
2087 	{ .name = #_name, .offset = _offset }
2088 
2089 static const struct diag_counter diag_basic[] = {
2090 	DIAG_COUNTER(rq_num_lle, 0x00),
2091 	DIAG_COUNTER(sq_num_lle, 0x04),
2092 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2093 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2094 	DIAG_COUNTER(rq_num_lpe, 0x18),
2095 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2096 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2097 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2098 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2099 	DIAG_COUNTER(sq_num_bre, 0x34),
2100 	DIAG_COUNTER(sq_num_rire, 0x44),
2101 	DIAG_COUNTER(rq_num_rire, 0x48),
2102 	DIAG_COUNTER(sq_num_rae, 0x4C),
2103 	DIAG_COUNTER(rq_num_rae, 0x50),
2104 	DIAG_COUNTER(sq_num_roe, 0x54),
2105 	DIAG_COUNTER(sq_num_tree, 0x5C),
2106 	DIAG_COUNTER(sq_num_rree, 0x64),
2107 	DIAG_COUNTER(rq_num_rnr, 0x68),
2108 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2109 	DIAG_COUNTER(rq_num_oos, 0x100),
2110 	DIAG_COUNTER(sq_num_oos, 0x104),
2111 };
2112 
2113 static const struct diag_counter diag_ext[] = {
2114 	DIAG_COUNTER(rq_num_dup, 0x130),
2115 	DIAG_COUNTER(sq_num_to, 0x134),
2116 };
2117 
2118 static const struct diag_counter diag_device_only[] = {
2119 	DIAG_COUNTER(num_cqovf, 0x1A0),
2120 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2121 };
2122 
2123 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2124 						    u8 port_num)
2125 {
2126 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2127 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2128 
2129 	if (!diag[!!port_num].name)
2130 		return NULL;
2131 
2132 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2133 					  diag[!!port_num].num_counters,
2134 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2135 }
2136 
2137 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2138 				struct rdma_hw_stats *stats,
2139 				u8 port, int index)
2140 {
2141 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2142 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2143 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2144 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2145 	int ret;
2146 	int i;
2147 
2148 	ret = mlx4_query_diag_counters(dev->dev,
2149 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2150 				       diag[!!port].offset, hw_value,
2151 				       diag[!!port].num_counters, port);
2152 
2153 	if (ret)
2154 		return ret;
2155 
2156 	for (i = 0; i < diag[!!port].num_counters; i++)
2157 		stats->value[i] = hw_value[i];
2158 
2159 	return diag[!!port].num_counters;
2160 }
2161 
2162 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2163 					 const char ***name,
2164 					 u32 **offset,
2165 					 u32 *num,
2166 					 bool port)
2167 {
2168 	u32 num_counters;
2169 
2170 	num_counters = ARRAY_SIZE(diag_basic);
2171 
2172 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2173 		num_counters += ARRAY_SIZE(diag_ext);
2174 
2175 	if (!port)
2176 		num_counters += ARRAY_SIZE(diag_device_only);
2177 
2178 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2179 	if (!*name)
2180 		return -ENOMEM;
2181 
2182 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2183 	if (!*offset)
2184 		goto err_name;
2185 
2186 	*num = num_counters;
2187 
2188 	return 0;
2189 
2190 err_name:
2191 	kfree(*name);
2192 	return -ENOMEM;
2193 }
2194 
2195 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2196 				       const char **name,
2197 				       u32 *offset,
2198 				       bool port)
2199 {
2200 	int i;
2201 	int j;
2202 
2203 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2204 		name[i] = diag_basic[i].name;
2205 		offset[i] = diag_basic[i].offset;
2206 	}
2207 
2208 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2209 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2210 			name[j] = diag_ext[i].name;
2211 			offset[j] = diag_ext[i].offset;
2212 		}
2213 	}
2214 
2215 	if (!port) {
2216 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2217 			name[j] = diag_device_only[i].name;
2218 			offset[j] = diag_device_only[i].offset;
2219 		}
2220 	}
2221 }
2222 
2223 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2224 	.alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2225 	.get_hw_stats = mlx4_ib_get_hw_stats,
2226 };
2227 
2228 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2229 {
2230 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2231 	int i;
2232 	int ret;
2233 	bool per_port = !!(ibdev->dev->caps.flags2 &
2234 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2235 
2236 	if (mlx4_is_slave(ibdev->dev))
2237 		return 0;
2238 
2239 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2240 		/* i == 1 means we are building port counters */
2241 		if (i && !per_port)
2242 			continue;
2243 
2244 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2245 						    &diag[i].offset,
2246 						    &diag[i].num_counters, i);
2247 		if (ret)
2248 			goto err_alloc;
2249 
2250 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2251 					   diag[i].offset, i);
2252 	}
2253 
2254 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2255 
2256 	return 0;
2257 
2258 err_alloc:
2259 	if (i) {
2260 		kfree(diag[i - 1].name);
2261 		kfree(diag[i - 1].offset);
2262 	}
2263 
2264 	return ret;
2265 }
2266 
2267 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2268 {
2269 	int i;
2270 
2271 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2272 		kfree(ibdev->diag_counters[i].offset);
2273 		kfree(ibdev->diag_counters[i].name);
2274 	}
2275 }
2276 
2277 #define MLX4_IB_INVALID_MAC	((u64)-1)
2278 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2279 			       struct net_device *dev,
2280 			       int port)
2281 {
2282 	u64 new_smac = 0;
2283 	u64 release_mac = MLX4_IB_INVALID_MAC;
2284 	struct mlx4_ib_qp *qp;
2285 
2286 	read_lock(&dev_base_lock);
2287 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2288 	read_unlock(&dev_base_lock);
2289 
2290 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2291 
2292 	/* no need for update QP1 and mac registration in non-SRIOV */
2293 	if (!mlx4_is_mfunc(ibdev->dev))
2294 		return;
2295 
2296 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2297 	qp = ibdev->qp1_proxy[port - 1];
2298 	if (qp) {
2299 		int new_smac_index;
2300 		u64 old_smac;
2301 		struct mlx4_update_qp_params update_params;
2302 
2303 		mutex_lock(&qp->mutex);
2304 		old_smac = qp->pri.smac;
2305 		if (new_smac == old_smac)
2306 			goto unlock;
2307 
2308 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2309 
2310 		if (new_smac_index < 0)
2311 			goto unlock;
2312 
2313 		update_params.smac_index = new_smac_index;
2314 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2315 				   &update_params)) {
2316 			release_mac = new_smac;
2317 			goto unlock;
2318 		}
2319 		/* if old port was zero, no mac was yet registered for this QP */
2320 		if (qp->pri.smac_port)
2321 			release_mac = old_smac;
2322 		qp->pri.smac = new_smac;
2323 		qp->pri.smac_port = port;
2324 		qp->pri.smac_index = new_smac_index;
2325 	}
2326 
2327 unlock:
2328 	if (release_mac != MLX4_IB_INVALID_MAC)
2329 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2330 	if (qp)
2331 		mutex_unlock(&qp->mutex);
2332 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2333 }
2334 
2335 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2336 				 struct net_device *dev,
2337 				 unsigned long event)
2338 
2339 {
2340 	struct mlx4_ib_iboe *iboe;
2341 	int update_qps_port = -1;
2342 	int port;
2343 
2344 	ASSERT_RTNL();
2345 
2346 	iboe = &ibdev->iboe;
2347 
2348 	spin_lock_bh(&iboe->lock);
2349 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2350 
2351 		iboe->netdevs[port - 1] =
2352 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2353 
2354 		if (dev == iboe->netdevs[port - 1] &&
2355 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2356 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2357 			update_qps_port = port;
2358 
2359 		if (dev == iboe->netdevs[port - 1] &&
2360 		    (event == NETDEV_UP || event == NETDEV_DOWN)) {
2361 			enum ib_port_state port_state;
2362 			struct ib_event ibev = { };
2363 
2364 			if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2365 						     &port_state))
2366 				continue;
2367 
2368 			if (event == NETDEV_UP &&
2369 			    (port_state != IB_PORT_ACTIVE ||
2370 			     iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2371 				continue;
2372 			if (event == NETDEV_DOWN &&
2373 			    (port_state != IB_PORT_DOWN ||
2374 			     iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2375 				continue;
2376 			iboe->last_port_state[port - 1] = port_state;
2377 
2378 			ibev.device = &ibdev->ib_dev;
2379 			ibev.element.port_num = port;
2380 			ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2381 							  IB_EVENT_PORT_ERR;
2382 			ib_dispatch_event(&ibev);
2383 		}
2384 
2385 	}
2386 	spin_unlock_bh(&iboe->lock);
2387 
2388 	if (update_qps_port > 0)
2389 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2390 }
2391 
2392 static int mlx4_ib_netdev_event(struct notifier_block *this,
2393 				unsigned long event, void *ptr)
2394 {
2395 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2396 	struct mlx4_ib_dev *ibdev;
2397 
2398 	if (!net_eq(dev_net(dev), &init_net))
2399 		return NOTIFY_DONE;
2400 
2401 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2402 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2403 
2404 	return NOTIFY_DONE;
2405 }
2406 
2407 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2408 {
2409 	int port;
2410 	int slave;
2411 	int i;
2412 
2413 	if (mlx4_is_master(ibdev->dev)) {
2414 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2415 		     ++slave) {
2416 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2417 				for (i = 0;
2418 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2419 				     ++i) {
2420 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2421 					/* master has the identity virt2phys pkey mapping */
2422 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2423 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2424 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2425 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2426 				}
2427 			}
2428 		}
2429 		/* initialize pkey cache */
2430 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2431 			for (i = 0;
2432 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2433 			     ++i)
2434 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2435 					(i) ? 0 : 0xFFFF;
2436 		}
2437 	}
2438 }
2439 
2440 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2441 {
2442 	int i, j, eq = 0, total_eqs = 0;
2443 
2444 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2445 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2446 	if (!ibdev->eq_table)
2447 		return;
2448 
2449 	for (i = 1; i <= dev->caps.num_ports; i++) {
2450 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2451 		     j++, total_eqs++) {
2452 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2453 				continue;
2454 			ibdev->eq_table[eq] = total_eqs;
2455 			if (!mlx4_assign_eq(dev, i,
2456 					    &ibdev->eq_table[eq]))
2457 				eq++;
2458 			else
2459 				ibdev->eq_table[eq] = -1;
2460 		}
2461 	}
2462 
2463 	for (i = eq; i < dev->caps.num_comp_vectors;
2464 	     ibdev->eq_table[i++] = -1)
2465 		;
2466 
2467 	/* Advertise the new number of EQs to clients */
2468 	ibdev->ib_dev.num_comp_vectors = eq;
2469 }
2470 
2471 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2472 {
2473 	int i;
2474 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2475 
2476 	/* no eqs were allocated */
2477 	if (!ibdev->eq_table)
2478 		return;
2479 
2480 	/* Reset the advertised EQ number */
2481 	ibdev->ib_dev.num_comp_vectors = 0;
2482 
2483 	for (i = 0; i < total_eqs; i++)
2484 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2485 
2486 	kfree(ibdev->eq_table);
2487 	ibdev->eq_table = NULL;
2488 }
2489 
2490 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2491 			       struct ib_port_immutable *immutable)
2492 {
2493 	struct ib_port_attr attr;
2494 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2495 	int err;
2496 
2497 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2498 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2499 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2500 	} else {
2501 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2502 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2503 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2504 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2505 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2506 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2507 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2508 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2509 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2510 	}
2511 
2512 	err = ib_query_port(ibdev, port_num, &attr);
2513 	if (err)
2514 		return err;
2515 
2516 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2517 	immutable->gid_tbl_len = attr.gid_tbl_len;
2518 
2519 	return 0;
2520 }
2521 
2522 static void get_fw_ver_str(struct ib_device *device, char *str)
2523 {
2524 	struct mlx4_ib_dev *dev =
2525 		container_of(device, struct mlx4_ib_dev, ib_dev);
2526 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2527 		 (int) (dev->dev->caps.fw_ver >> 32),
2528 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2529 		 (int) dev->dev->caps.fw_ver & 0xffff);
2530 }
2531 
2532 static const struct ib_device_ops mlx4_ib_dev_ops = {
2533 	.add_gid = mlx4_ib_add_gid,
2534 	.alloc_mr = mlx4_ib_alloc_mr,
2535 	.alloc_pd = mlx4_ib_alloc_pd,
2536 	.alloc_ucontext = mlx4_ib_alloc_ucontext,
2537 	.attach_mcast = mlx4_ib_mcg_attach,
2538 	.create_ah = mlx4_ib_create_ah,
2539 	.create_cq = mlx4_ib_create_cq,
2540 	.create_qp = mlx4_ib_create_qp,
2541 	.create_srq = mlx4_ib_create_srq,
2542 	.dealloc_pd = mlx4_ib_dealloc_pd,
2543 	.dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2544 	.del_gid = mlx4_ib_del_gid,
2545 	.dereg_mr = mlx4_ib_dereg_mr,
2546 	.destroy_ah = mlx4_ib_destroy_ah,
2547 	.destroy_cq = mlx4_ib_destroy_cq,
2548 	.destroy_qp = mlx4_ib_destroy_qp,
2549 	.destroy_srq = mlx4_ib_destroy_srq,
2550 	.detach_mcast = mlx4_ib_mcg_detach,
2551 	.disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2552 	.drain_rq = mlx4_ib_drain_rq,
2553 	.drain_sq = mlx4_ib_drain_sq,
2554 	.get_dev_fw_str = get_fw_ver_str,
2555 	.get_dma_mr = mlx4_ib_get_dma_mr,
2556 	.get_link_layer = mlx4_ib_port_link_layer,
2557 	.get_netdev = mlx4_ib_get_netdev,
2558 	.get_port_immutable = mlx4_port_immutable,
2559 	.map_mr_sg = mlx4_ib_map_mr_sg,
2560 	.mmap = mlx4_ib_mmap,
2561 	.modify_cq = mlx4_ib_modify_cq,
2562 	.modify_device = mlx4_ib_modify_device,
2563 	.modify_port = mlx4_ib_modify_port,
2564 	.modify_qp = mlx4_ib_modify_qp,
2565 	.modify_srq = mlx4_ib_modify_srq,
2566 	.poll_cq = mlx4_ib_poll_cq,
2567 	.post_recv = mlx4_ib_post_recv,
2568 	.post_send = mlx4_ib_post_send,
2569 	.post_srq_recv = mlx4_ib_post_srq_recv,
2570 	.process_mad = mlx4_ib_process_mad,
2571 	.query_ah = mlx4_ib_query_ah,
2572 	.query_device = mlx4_ib_query_device,
2573 	.query_gid = mlx4_ib_query_gid,
2574 	.query_pkey = mlx4_ib_query_pkey,
2575 	.query_port = mlx4_ib_query_port,
2576 	.query_qp = mlx4_ib_query_qp,
2577 	.query_srq = mlx4_ib_query_srq,
2578 	.reg_user_mr = mlx4_ib_reg_user_mr,
2579 	.req_notify_cq = mlx4_ib_arm_cq,
2580 	.rereg_user_mr = mlx4_ib_rereg_user_mr,
2581 	.resize_cq = mlx4_ib_resize_cq,
2582 };
2583 
2584 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2585 	.create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2586 	.create_wq = mlx4_ib_create_wq,
2587 	.destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2588 	.destroy_wq = mlx4_ib_destroy_wq,
2589 	.modify_wq = mlx4_ib_modify_wq,
2590 };
2591 
2592 static const struct ib_device_ops mlx4_ib_dev_fmr_ops = {
2593 	.alloc_fmr = mlx4_ib_fmr_alloc,
2594 	.dealloc_fmr = mlx4_ib_fmr_dealloc,
2595 	.map_phys_fmr = mlx4_ib_map_phys_fmr,
2596 	.unmap_fmr = mlx4_ib_unmap_fmr,
2597 };
2598 
2599 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2600 	.alloc_mw = mlx4_ib_alloc_mw,
2601 	.dealloc_mw = mlx4_ib_dealloc_mw,
2602 };
2603 
2604 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2605 	.alloc_xrcd = mlx4_ib_alloc_xrcd,
2606 	.dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2607 };
2608 
2609 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2610 	.create_flow = mlx4_ib_create_flow,
2611 	.destroy_flow = mlx4_ib_destroy_flow,
2612 };
2613 
2614 static void *mlx4_ib_add(struct mlx4_dev *dev)
2615 {
2616 	struct mlx4_ib_dev *ibdev;
2617 	int num_ports = 0;
2618 	int i, j;
2619 	int err;
2620 	struct mlx4_ib_iboe *iboe;
2621 	int ib_num_ports = 0;
2622 	int num_req_counters;
2623 	int allocated;
2624 	u32 counter_index;
2625 	struct counter_index *new_counter_index = NULL;
2626 
2627 	pr_info_once("%s", mlx4_ib_version);
2628 
2629 	num_ports = 0;
2630 	mlx4_foreach_ib_transport_port(i, dev)
2631 		num_ports++;
2632 
2633 	/* No point in registering a device with no ports... */
2634 	if (num_ports == 0)
2635 		return NULL;
2636 
2637 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2638 	if (!ibdev) {
2639 		dev_err(&dev->persist->pdev->dev,
2640 			"Device struct alloc failed\n");
2641 		return NULL;
2642 	}
2643 
2644 	iboe = &ibdev->iboe;
2645 
2646 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2647 		goto err_dealloc;
2648 
2649 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2650 		goto err_pd;
2651 
2652 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2653 				 PAGE_SIZE);
2654 	if (!ibdev->uar_map)
2655 		goto err_uar;
2656 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2657 
2658 	ibdev->dev = dev;
2659 	ibdev->bond_next_port	= 0;
2660 
2661 	ibdev->ib_dev.owner		= THIS_MODULE;
2662 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2663 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2664 	ibdev->num_ports		= num_ports;
2665 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2666 						1 : ibdev->num_ports;
2667 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2668 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2669 
2670 	if (dev->caps.userspace_caps)
2671 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2672 	else
2673 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2674 
2675 	ibdev->ib_dev.uverbs_cmd_mask	=
2676 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2677 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2678 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2679 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2680 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2681 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2682 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2683 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2684 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2685 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2686 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2687 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2688 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2689 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2690 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2691 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2692 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2693 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2694 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2695 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2696 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2697 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2698 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2699 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2700 
2701 	ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2702 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2703 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2704 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2705 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2706 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2707 
2708 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2709 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2710 	    IB_LINK_LAYER_ETHERNET) ||
2711 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2712 	    IB_LINK_LAYER_ETHERNET))) {
2713 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2714 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2715 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2716 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2717 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2718 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2719 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2720 	}
2721 
2722 	if (!mlx4_is_slave(ibdev->dev))
2723 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fmr_ops);
2724 
2725 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2726 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2727 		ibdev->ib_dev.uverbs_cmd_mask |=
2728 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2729 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2730 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2731 	}
2732 
2733 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2734 		ibdev->ib_dev.uverbs_cmd_mask |=
2735 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2736 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2737 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2738 	}
2739 
2740 	if (check_flow_steering_support(dev)) {
2741 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2742 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2743 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2744 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2745 		ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2746 	}
2747 
2748 	mlx4_ib_alloc_eqs(dev, ibdev);
2749 
2750 	spin_lock_init(&iboe->lock);
2751 
2752 	if (init_node_data(ibdev))
2753 		goto err_map;
2754 	mlx4_init_sl2vl_tbl(ibdev);
2755 
2756 	for (i = 0; i < ibdev->num_ports; ++i) {
2757 		mutex_init(&ibdev->counters_table[i].mutex);
2758 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2759 		iboe->last_port_state[i] = IB_PORT_DOWN;
2760 	}
2761 
2762 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2763 	for (i = 0; i < num_req_counters; ++i) {
2764 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2765 		allocated = 0;
2766 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2767 						IB_LINK_LAYER_ETHERNET) {
2768 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2769 						 MLX4_RES_USAGE_DRIVER);
2770 			/* if failed to allocate a new counter, use default */
2771 			if (err)
2772 				counter_index =
2773 					mlx4_get_default_counter_index(dev,
2774 								       i + 1);
2775 			else
2776 				allocated = 1;
2777 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2778 			counter_index = mlx4_get_default_counter_index(dev,
2779 								       i + 1);
2780 		}
2781 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2782 					    GFP_KERNEL);
2783 		if (!new_counter_index) {
2784 			if (allocated)
2785 				mlx4_counter_free(ibdev->dev, counter_index);
2786 			goto err_counter;
2787 		}
2788 		new_counter_index->index = counter_index;
2789 		new_counter_index->allocated = allocated;
2790 		list_add_tail(&new_counter_index->list,
2791 			      &ibdev->counters_table[i].counters_list);
2792 		ibdev->counters_table[i].default_counter = counter_index;
2793 		pr_info("counter index %d for port %d allocated %d\n",
2794 			counter_index, i + 1, allocated);
2795 	}
2796 	if (mlx4_is_bonded(dev))
2797 		for (i = 1; i < ibdev->num_ports ; ++i) {
2798 			new_counter_index =
2799 					kmalloc(sizeof(struct counter_index),
2800 						GFP_KERNEL);
2801 			if (!new_counter_index)
2802 				goto err_counter;
2803 			new_counter_index->index = counter_index;
2804 			new_counter_index->allocated = 0;
2805 			list_add_tail(&new_counter_index->list,
2806 				      &ibdev->counters_table[i].counters_list);
2807 			ibdev->counters_table[i].default_counter =
2808 								counter_index;
2809 		}
2810 
2811 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2812 		ib_num_ports++;
2813 
2814 	spin_lock_init(&ibdev->sm_lock);
2815 	mutex_init(&ibdev->cap_mask_mutex);
2816 	INIT_LIST_HEAD(&ibdev->qp_list);
2817 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2818 
2819 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2820 	    ib_num_ports) {
2821 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2822 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2823 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2824 					    &ibdev->steer_qpn_base, 0,
2825 					    MLX4_RES_USAGE_DRIVER);
2826 		if (err)
2827 			goto err_counter;
2828 
2829 		ibdev->ib_uc_qpns_bitmap =
2830 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2831 				      sizeof(long),
2832 				      GFP_KERNEL);
2833 		if (!ibdev->ib_uc_qpns_bitmap)
2834 			goto err_steer_qp_release;
2835 
2836 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2837 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2838 				    ibdev->steer_qpn_count);
2839 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2840 					dev, ibdev->steer_qpn_base,
2841 					ibdev->steer_qpn_base +
2842 					ibdev->steer_qpn_count - 1);
2843 			if (err)
2844 				goto err_steer_free_bitmap;
2845 		} else {
2846 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2847 				    ibdev->steer_qpn_count);
2848 		}
2849 	}
2850 
2851 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2852 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2853 
2854 	if (mlx4_ib_alloc_diag_counters(ibdev))
2855 		goto err_steer_free_bitmap;
2856 
2857 	rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2858 	ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2859 	if (ib_register_device(&ibdev->ib_dev, "mlx4_%d", NULL))
2860 		goto err_diag_counters;
2861 
2862 	if (mlx4_ib_mad_init(ibdev))
2863 		goto err_reg;
2864 
2865 	if (mlx4_ib_init_sriov(ibdev))
2866 		goto err_mad;
2867 
2868 	if (!iboe->nb.notifier_call) {
2869 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2870 		err = register_netdevice_notifier(&iboe->nb);
2871 		if (err) {
2872 			iboe->nb.notifier_call = NULL;
2873 			goto err_notif;
2874 		}
2875 	}
2876 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2877 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2878 		if (err)
2879 			goto err_notif;
2880 	}
2881 
2882 	ibdev->ib_active = true;
2883 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2884 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2885 					 &ibdev->ib_dev);
2886 
2887 	if (mlx4_is_mfunc(ibdev->dev))
2888 		init_pkeys(ibdev);
2889 
2890 	/* create paravirt contexts for any VFs which are active */
2891 	if (mlx4_is_master(ibdev->dev)) {
2892 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2893 			if (j == mlx4_master_func_num(ibdev->dev))
2894 				continue;
2895 			if (mlx4_is_slave_active(ibdev->dev, j))
2896 				do_slave_init(ibdev, j, 1);
2897 		}
2898 	}
2899 	return ibdev;
2900 
2901 err_notif:
2902 	if (ibdev->iboe.nb.notifier_call) {
2903 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2904 			pr_warn("failure unregistering notifier\n");
2905 		ibdev->iboe.nb.notifier_call = NULL;
2906 	}
2907 	flush_workqueue(wq);
2908 
2909 	mlx4_ib_close_sriov(ibdev);
2910 
2911 err_mad:
2912 	mlx4_ib_mad_cleanup(ibdev);
2913 
2914 err_reg:
2915 	ib_unregister_device(&ibdev->ib_dev);
2916 
2917 err_diag_counters:
2918 	mlx4_ib_diag_cleanup(ibdev);
2919 
2920 err_steer_free_bitmap:
2921 	kfree(ibdev->ib_uc_qpns_bitmap);
2922 
2923 err_steer_qp_release:
2924 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2925 			      ibdev->steer_qpn_count);
2926 err_counter:
2927 	for (i = 0; i < ibdev->num_ports; ++i)
2928 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2929 
2930 err_map:
2931 	mlx4_ib_free_eqs(dev, ibdev);
2932 	iounmap(ibdev->uar_map);
2933 
2934 err_uar:
2935 	mlx4_uar_free(dev, &ibdev->priv_uar);
2936 
2937 err_pd:
2938 	mlx4_pd_free(dev, ibdev->priv_pdn);
2939 
2940 err_dealloc:
2941 	ib_dealloc_device(&ibdev->ib_dev);
2942 
2943 	return NULL;
2944 }
2945 
2946 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2947 {
2948 	int offset;
2949 
2950 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2951 
2952 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2953 					 dev->steer_qpn_count,
2954 					 get_count_order(count));
2955 	if (offset < 0)
2956 		return offset;
2957 
2958 	*qpn = dev->steer_qpn_base + offset;
2959 	return 0;
2960 }
2961 
2962 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2963 {
2964 	if (!qpn ||
2965 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2966 		return;
2967 
2968 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2969 		 qpn, dev->steer_qpn_base))
2970 		/* not supposed to be here */
2971 		return;
2972 
2973 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2974 			      qpn - dev->steer_qpn_base,
2975 			      get_count_order(count));
2976 }
2977 
2978 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2979 			 int is_attach)
2980 {
2981 	int err;
2982 	size_t flow_size;
2983 	struct ib_flow_attr *flow = NULL;
2984 	struct ib_flow_spec_ib *ib_spec;
2985 
2986 	if (is_attach) {
2987 		flow_size = sizeof(struct ib_flow_attr) +
2988 			    sizeof(struct ib_flow_spec_ib);
2989 		flow = kzalloc(flow_size, GFP_KERNEL);
2990 		if (!flow)
2991 			return -ENOMEM;
2992 		flow->port = mqp->port;
2993 		flow->num_of_specs = 1;
2994 		flow->size = flow_size;
2995 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2996 		ib_spec->type = IB_FLOW_SPEC_IB;
2997 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2998 		/* Add an empty rule for IB L2 */
2999 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3000 
3001 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3002 					    IB_FLOW_DOMAIN_NIC,
3003 					    MLX4_FS_REGULAR,
3004 					    &mqp->reg_id);
3005 	} else {
3006 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3007 	}
3008 	kfree(flow);
3009 	return err;
3010 }
3011 
3012 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3013 {
3014 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3015 	int p;
3016 	int i;
3017 
3018 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3019 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3020 	ibdev->ib_active = false;
3021 	flush_workqueue(wq);
3022 
3023 	mlx4_ib_close_sriov(ibdev);
3024 	mlx4_ib_mad_cleanup(ibdev);
3025 	ib_unregister_device(&ibdev->ib_dev);
3026 	mlx4_ib_diag_cleanup(ibdev);
3027 	if (ibdev->iboe.nb.notifier_call) {
3028 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3029 			pr_warn("failure unregistering notifier\n");
3030 		ibdev->iboe.nb.notifier_call = NULL;
3031 	}
3032 
3033 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3034 			      ibdev->steer_qpn_count);
3035 	kfree(ibdev->ib_uc_qpns_bitmap);
3036 
3037 	iounmap(ibdev->uar_map);
3038 	for (p = 0; p < ibdev->num_ports; ++p)
3039 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3040 
3041 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3042 		mlx4_CLOSE_PORT(dev, p);
3043 
3044 	mlx4_ib_free_eqs(dev, ibdev);
3045 
3046 	mlx4_uar_free(dev, &ibdev->priv_uar);
3047 	mlx4_pd_free(dev, ibdev->priv_pdn);
3048 	ib_dealloc_device(&ibdev->ib_dev);
3049 }
3050 
3051 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3052 {
3053 	struct mlx4_ib_demux_work **dm = NULL;
3054 	struct mlx4_dev *dev = ibdev->dev;
3055 	int i;
3056 	unsigned long flags;
3057 	struct mlx4_active_ports actv_ports;
3058 	unsigned int ports;
3059 	unsigned int first_port;
3060 
3061 	if (!mlx4_is_master(dev))
3062 		return;
3063 
3064 	actv_ports = mlx4_get_active_ports(dev, slave);
3065 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3066 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3067 
3068 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3069 	if (!dm)
3070 		return;
3071 
3072 	for (i = 0; i < ports; i++) {
3073 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3074 		if (!dm[i]) {
3075 			while (--i >= 0)
3076 				kfree(dm[i]);
3077 			goto out;
3078 		}
3079 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3080 		dm[i]->port = first_port + i + 1;
3081 		dm[i]->slave = slave;
3082 		dm[i]->do_init = do_init;
3083 		dm[i]->dev = ibdev;
3084 	}
3085 	/* initialize or tear down tunnel QPs for the slave */
3086 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3087 	if (!ibdev->sriov.is_going_down) {
3088 		for (i = 0; i < ports; i++)
3089 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3090 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3091 	} else {
3092 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3093 		for (i = 0; i < ports; i++)
3094 			kfree(dm[i]);
3095 	}
3096 out:
3097 	kfree(dm);
3098 	return;
3099 }
3100 
3101 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3102 {
3103 	struct mlx4_ib_qp *mqp;
3104 	unsigned long flags_qp;
3105 	unsigned long flags_cq;
3106 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3107 	struct list_head    cq_notify_list;
3108 	struct mlx4_cq *mcq;
3109 	unsigned long flags;
3110 
3111 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3112 	INIT_LIST_HEAD(&cq_notify_list);
3113 
3114 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3115 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3116 
3117 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3118 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3119 		if (mqp->sq.tail != mqp->sq.head) {
3120 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3121 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3122 			if (send_mcq->mcq.comp &&
3123 			    mqp->ibqp.send_cq->comp_handler) {
3124 				if (!send_mcq->mcq.reset_notify_added) {
3125 					send_mcq->mcq.reset_notify_added = 1;
3126 					list_add_tail(&send_mcq->mcq.reset_notify,
3127 						      &cq_notify_list);
3128 				}
3129 			}
3130 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3131 		}
3132 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3133 		/* Now, handle the QP's receive queue */
3134 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3135 		/* no handling is needed for SRQ */
3136 		if (!mqp->ibqp.srq) {
3137 			if (mqp->rq.tail != mqp->rq.head) {
3138 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3139 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3140 				if (recv_mcq->mcq.comp &&
3141 				    mqp->ibqp.recv_cq->comp_handler) {
3142 					if (!recv_mcq->mcq.reset_notify_added) {
3143 						recv_mcq->mcq.reset_notify_added = 1;
3144 						list_add_tail(&recv_mcq->mcq.reset_notify,
3145 							      &cq_notify_list);
3146 					}
3147 				}
3148 				spin_unlock_irqrestore(&recv_mcq->lock,
3149 						       flags_cq);
3150 			}
3151 		}
3152 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3153 	}
3154 
3155 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3156 		mcq->comp(mcq);
3157 	}
3158 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3159 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3160 }
3161 
3162 static void handle_bonded_port_state_event(struct work_struct *work)
3163 {
3164 	struct ib_event_work *ew =
3165 		container_of(work, struct ib_event_work, work);
3166 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3167 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3168 	int i;
3169 	struct ib_event ibev;
3170 
3171 	kfree(ew);
3172 	spin_lock_bh(&ibdev->iboe.lock);
3173 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3174 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3175 		enum ib_port_state curr_port_state;
3176 
3177 		if (!curr_netdev)
3178 			continue;
3179 
3180 		curr_port_state =
3181 			(netif_running(curr_netdev) &&
3182 			 netif_carrier_ok(curr_netdev)) ?
3183 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3184 
3185 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3186 			curr_port_state : IB_PORT_ACTIVE;
3187 	}
3188 	spin_unlock_bh(&ibdev->iboe.lock);
3189 
3190 	ibev.device = &ibdev->ib_dev;
3191 	ibev.element.port_num = 1;
3192 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3193 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3194 
3195 	ib_dispatch_event(&ibev);
3196 }
3197 
3198 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3199 {
3200 	u64 sl2vl;
3201 	int err;
3202 
3203 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3204 	if (err) {
3205 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3206 		       port, err);
3207 		sl2vl = 0;
3208 	}
3209 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3210 }
3211 
3212 static void ib_sl2vl_update_work(struct work_struct *work)
3213 {
3214 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3215 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3216 	int port = ew->port;
3217 
3218 	mlx4_ib_sl2vl_update(mdev, port);
3219 
3220 	kfree(ew);
3221 }
3222 
3223 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3224 				     int port)
3225 {
3226 	struct ib_event_work *ew;
3227 
3228 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3229 	if (ew) {
3230 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3231 		ew->port = port;
3232 		ew->ib_dev = ibdev;
3233 		queue_work(wq, &ew->work);
3234 	}
3235 }
3236 
3237 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3238 			  enum mlx4_dev_event event, unsigned long param)
3239 {
3240 	struct ib_event ibev;
3241 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3242 	struct mlx4_eqe *eqe = NULL;
3243 	struct ib_event_work *ew;
3244 	int p = 0;
3245 
3246 	if (mlx4_is_bonded(dev) &&
3247 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3248 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3249 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3250 		if (!ew)
3251 			return;
3252 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3253 		ew->ib_dev = ibdev;
3254 		queue_work(wq, &ew->work);
3255 		return;
3256 	}
3257 
3258 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3259 		eqe = (struct mlx4_eqe *)param;
3260 	else
3261 		p = (int) param;
3262 
3263 	switch (event) {
3264 	case MLX4_DEV_EVENT_PORT_UP:
3265 		if (p > ibdev->num_ports)
3266 			return;
3267 		if (!mlx4_is_slave(dev) &&
3268 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3269 			IB_LINK_LAYER_INFINIBAND) {
3270 			if (mlx4_is_master(dev))
3271 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3272 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3273 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3274 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3275 		}
3276 		ibev.event = IB_EVENT_PORT_ACTIVE;
3277 		break;
3278 
3279 	case MLX4_DEV_EVENT_PORT_DOWN:
3280 		if (p > ibdev->num_ports)
3281 			return;
3282 		ibev.event = IB_EVENT_PORT_ERR;
3283 		break;
3284 
3285 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3286 		ibdev->ib_active = false;
3287 		ibev.event = IB_EVENT_DEVICE_FATAL;
3288 		mlx4_ib_handle_catas_error(ibdev);
3289 		break;
3290 
3291 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3292 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3293 		if (!ew)
3294 			break;
3295 
3296 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3297 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3298 		ew->ib_dev = ibdev;
3299 		/* need to queue only for port owner, which uses GEN_EQE */
3300 		if (mlx4_is_master(dev))
3301 			queue_work(wq, &ew->work);
3302 		else
3303 			handle_port_mgmt_change_event(&ew->work);
3304 		return;
3305 
3306 	case MLX4_DEV_EVENT_SLAVE_INIT:
3307 		/* here, p is the slave id */
3308 		do_slave_init(ibdev, p, 1);
3309 		if (mlx4_is_master(dev)) {
3310 			int i;
3311 
3312 			for (i = 1; i <= ibdev->num_ports; i++) {
3313 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3314 					== IB_LINK_LAYER_INFINIBAND)
3315 					mlx4_ib_slave_alias_guid_event(ibdev,
3316 								       p, i,
3317 								       1);
3318 			}
3319 		}
3320 		return;
3321 
3322 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3323 		if (mlx4_is_master(dev)) {
3324 			int i;
3325 
3326 			for (i = 1; i <= ibdev->num_ports; i++) {
3327 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3328 					== IB_LINK_LAYER_INFINIBAND)
3329 					mlx4_ib_slave_alias_guid_event(ibdev,
3330 								       p, i,
3331 								       0);
3332 			}
3333 		}
3334 		/* here, p is the slave id */
3335 		do_slave_init(ibdev, p, 0);
3336 		return;
3337 
3338 	default:
3339 		return;
3340 	}
3341 
3342 	ibev.device	      = ibdev_ptr;
3343 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3344 
3345 	ib_dispatch_event(&ibev);
3346 }
3347 
3348 static struct mlx4_interface mlx4_ib_interface = {
3349 	.add		= mlx4_ib_add,
3350 	.remove		= mlx4_ib_remove,
3351 	.event		= mlx4_ib_event,
3352 	.protocol	= MLX4_PROT_IB_IPV6,
3353 	.flags		= MLX4_INTFF_BONDING
3354 };
3355 
3356 static int __init mlx4_ib_init(void)
3357 {
3358 	int err;
3359 
3360 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3361 	if (!wq)
3362 		return -ENOMEM;
3363 
3364 	err = mlx4_ib_mcg_init();
3365 	if (err)
3366 		goto clean_wq;
3367 
3368 	err = mlx4_register_interface(&mlx4_ib_interface);
3369 	if (err)
3370 		goto clean_mcg;
3371 
3372 	return 0;
3373 
3374 clean_mcg:
3375 	mlx4_ib_mcg_destroy();
3376 
3377 clean_wq:
3378 	destroy_workqueue(wq);
3379 	return err;
3380 }
3381 
3382 static void __exit mlx4_ib_cleanup(void)
3383 {
3384 	mlx4_unregister_interface(&mlx4_ib_interface);
3385 	mlx4_ib_mcg_destroy();
3386 	destroy_workqueue(wq);
3387 }
3388 
3389 module_init(mlx4_ib_init);
3390 module_exit(mlx4_ib_cleanup);
3391