xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 33023fb85a42b53bf778bc025f9667b582282be4)
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	   = dev->dev->caps.max_sq_sg;
521 	props->max_recv_sge	   = dev->dev->caps.max_rq_sg;
522 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
523 	props->max_cq		   = dev->dev->quotas.cq;
524 	props->max_cqe		   = dev->dev->caps.max_cqes;
525 	props->max_mr		   = dev->dev->quotas.mpt;
526 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
527 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
528 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
529 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
530 	props->max_srq		   = dev->dev->quotas.srq;
531 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
532 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
533 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
534 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
535 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
536 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
537 	props->masked_atomic_cap   = props->atomic_cap;
538 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
539 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
540 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
541 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
542 					   props->max_mcast_grp;
543 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
544 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
545 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
546 	props->max_ah = INT_MAX;
547 
548 	if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
549 	    mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
550 		if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
551 			props->rss_caps.max_rwq_indirection_tables =
552 				props->max_qp;
553 			props->rss_caps.max_rwq_indirection_table_size =
554 				dev->dev->caps.max_rss_tbl_sz;
555 			props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
556 			props->max_wq_type_rq = props->max_qp;
557 		}
558 
559 		if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
560 			props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
561 	}
562 
563 	props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
564 	props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
565 
566 	if (!mlx4_is_slave(dev->dev))
567 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
568 
569 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
570 		resp.response_length += sizeof(resp.hca_core_clock_offset);
571 		if (!err && !mlx4_is_slave(dev->dev)) {
572 			resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
573 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
574 		}
575 	}
576 
577 	if (uhw->outlen >= resp.response_length +
578 	    sizeof(resp.max_inl_recv_sz)) {
579 		resp.response_length += sizeof(resp.max_inl_recv_sz);
580 		resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
581 			sizeof(struct mlx4_wqe_data_seg);
582 	}
583 
584 	if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
585 		if (props->rss_caps.supported_qpts) {
586 			resp.rss_caps.rx_hash_function =
587 				MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
588 
589 			resp.rss_caps.rx_hash_fields_mask =
590 				MLX4_IB_RX_HASH_SRC_IPV4 |
591 				MLX4_IB_RX_HASH_DST_IPV4 |
592 				MLX4_IB_RX_HASH_SRC_IPV6 |
593 				MLX4_IB_RX_HASH_DST_IPV6 |
594 				MLX4_IB_RX_HASH_SRC_PORT_TCP |
595 				MLX4_IB_RX_HASH_DST_PORT_TCP |
596 				MLX4_IB_RX_HASH_SRC_PORT_UDP |
597 				MLX4_IB_RX_HASH_DST_PORT_UDP;
598 
599 			if (dev->dev->caps.tunnel_offload_mode ==
600 			    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
601 				resp.rss_caps.rx_hash_fields_mask |=
602 					MLX4_IB_RX_HASH_INNER;
603 		}
604 		resp.response_length = offsetof(typeof(resp), rss_caps) +
605 				       sizeof(resp.rss_caps);
606 	}
607 
608 	if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
609 		if (dev->dev->caps.max_gso_sz &&
610 		    ((mlx4_ib_port_link_layer(ibdev, 1) ==
611 		    IB_LINK_LAYER_ETHERNET) ||
612 		    (mlx4_ib_port_link_layer(ibdev, 2) ==
613 		    IB_LINK_LAYER_ETHERNET))) {
614 			resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
615 			resp.tso_caps.supported_qpts |=
616 				1 << IB_QPT_RAW_PACKET;
617 		}
618 		resp.response_length = offsetof(typeof(resp), tso_caps) +
619 				       sizeof(resp.tso_caps);
620 	}
621 
622 	if (uhw->outlen) {
623 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
624 		if (err)
625 			goto out;
626 	}
627 out:
628 	kfree(in_mad);
629 	kfree(out_mad);
630 
631 	return err;
632 }
633 
634 static enum rdma_link_layer
635 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
636 {
637 	struct mlx4_dev *dev = to_mdev(device)->dev;
638 
639 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
640 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
641 }
642 
643 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
644 			      struct ib_port_attr *props, int netw_view)
645 {
646 	struct ib_smp *in_mad  = NULL;
647 	struct ib_smp *out_mad = NULL;
648 	int ext_active_speed;
649 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
650 	int err = -ENOMEM;
651 
652 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
653 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
654 	if (!in_mad || !out_mad)
655 		goto out;
656 
657 	init_query_mad(in_mad);
658 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
659 	in_mad->attr_mod = cpu_to_be32(port);
660 
661 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
662 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
663 
664 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
665 				in_mad, out_mad);
666 	if (err)
667 		goto out;
668 
669 
670 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
671 	props->lmc		= out_mad->data[34] & 0x7;
672 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
673 	props->sm_sl		= out_mad->data[36] & 0xf;
674 	props->state		= out_mad->data[32] & 0xf;
675 	props->phys_state	= out_mad->data[33] >> 4;
676 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
677 	if (netw_view)
678 		props->gid_tbl_len = out_mad->data[50];
679 	else
680 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
681 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
682 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
683 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
684 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
685 	props->active_width	= out_mad->data[31] & 0xf;
686 	props->active_speed	= out_mad->data[35] >> 4;
687 	props->max_mtu		= out_mad->data[41] & 0xf;
688 	props->active_mtu	= out_mad->data[36] >> 4;
689 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
690 	props->max_vl_num	= out_mad->data[37] >> 4;
691 	props->init_type_reply	= out_mad->data[41] >> 4;
692 
693 	/* Check if extended speeds (EDR/FDR/...) are supported */
694 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
695 		ext_active_speed = out_mad->data[62] >> 4;
696 
697 		switch (ext_active_speed) {
698 		case 1:
699 			props->active_speed = IB_SPEED_FDR;
700 			break;
701 		case 2:
702 			props->active_speed = IB_SPEED_EDR;
703 			break;
704 		}
705 	}
706 
707 	/* If reported active speed is QDR, check if is FDR-10 */
708 	if (props->active_speed == IB_SPEED_QDR) {
709 		init_query_mad(in_mad);
710 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
711 		in_mad->attr_mod = cpu_to_be32(port);
712 
713 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
714 				   NULL, NULL, in_mad, out_mad);
715 		if (err)
716 			goto out;
717 
718 		/* Checking LinkSpeedActive for FDR-10 */
719 		if (out_mad->data[15] & 0x1)
720 			props->active_speed = IB_SPEED_FDR10;
721 	}
722 
723 	/* Avoid wrong speed value returned by FW if the IB link is down. */
724 	if (props->state == IB_PORT_DOWN)
725 		 props->active_speed = IB_SPEED_SDR;
726 
727 out:
728 	kfree(in_mad);
729 	kfree(out_mad);
730 	return err;
731 }
732 
733 static u8 state_to_phys_state(enum ib_port_state state)
734 {
735 	return state == IB_PORT_ACTIVE ? 5 : 3;
736 }
737 
738 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
739 			       struct ib_port_attr *props)
740 {
741 
742 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
743 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
744 	struct net_device *ndev;
745 	enum ib_mtu tmp;
746 	struct mlx4_cmd_mailbox *mailbox;
747 	int err = 0;
748 	int is_bonded = mlx4_is_bonded(mdev->dev);
749 
750 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
751 	if (IS_ERR(mailbox))
752 		return PTR_ERR(mailbox);
753 
754 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
755 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
756 			   MLX4_CMD_WRAPPED);
757 	if (err)
758 		goto out;
759 
760 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ||
761 				   (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
762 					   IB_WIDTH_4X : IB_WIDTH_1X;
763 	props->active_speed	=  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
764 					   IB_SPEED_FDR : IB_SPEED_QDR;
765 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
766 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
767 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
768 	props->pkey_tbl_len	= 1;
769 	props->max_mtu		= IB_MTU_4096;
770 	props->max_vl_num	= 2;
771 	props->state		= IB_PORT_DOWN;
772 	props->phys_state	= state_to_phys_state(props->state);
773 	props->active_mtu	= IB_MTU_256;
774 	spin_lock_bh(&iboe->lock);
775 	ndev = iboe->netdevs[port - 1];
776 	if (ndev && is_bonded) {
777 		rcu_read_lock(); /* required to get upper dev */
778 		ndev = netdev_master_upper_dev_get_rcu(ndev);
779 		rcu_read_unlock();
780 	}
781 	if (!ndev)
782 		goto out_unlock;
783 
784 	tmp = iboe_get_mtu(ndev->mtu);
785 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
786 
787 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
788 					IB_PORT_ACTIVE : IB_PORT_DOWN;
789 	props->phys_state	= state_to_phys_state(props->state);
790 out_unlock:
791 	spin_unlock_bh(&iboe->lock);
792 out:
793 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
794 	return err;
795 }
796 
797 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
798 			 struct ib_port_attr *props, int netw_view)
799 {
800 	int err;
801 
802 	/* props being zeroed by the caller, avoid zeroing it here */
803 
804 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
805 		ib_link_query_port(ibdev, port, props, netw_view) :
806 				eth_link_query_port(ibdev, port, props);
807 
808 	return err;
809 }
810 
811 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
812 			      struct ib_port_attr *props)
813 {
814 	/* returns host view */
815 	return __mlx4_ib_query_port(ibdev, port, props, 0);
816 }
817 
818 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
819 			union ib_gid *gid, int netw_view)
820 {
821 	struct ib_smp *in_mad  = NULL;
822 	struct ib_smp *out_mad = NULL;
823 	int err = -ENOMEM;
824 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
825 	int clear = 0;
826 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
827 
828 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
829 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
830 	if (!in_mad || !out_mad)
831 		goto out;
832 
833 	init_query_mad(in_mad);
834 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
835 	in_mad->attr_mod = cpu_to_be32(port);
836 
837 	if (mlx4_is_mfunc(dev->dev) && netw_view)
838 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
839 
840 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
841 	if (err)
842 		goto out;
843 
844 	memcpy(gid->raw, out_mad->data + 8, 8);
845 
846 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
847 		if (index) {
848 			/* For any index > 0, return the null guid */
849 			err = 0;
850 			clear = 1;
851 			goto out;
852 		}
853 	}
854 
855 	init_query_mad(in_mad);
856 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
857 	in_mad->attr_mod = cpu_to_be32(index / 8);
858 
859 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
860 			   NULL, NULL, in_mad, out_mad);
861 	if (err)
862 		goto out;
863 
864 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
865 
866 out:
867 	if (clear)
868 		memset(gid->raw + 8, 0, 8);
869 	kfree(in_mad);
870 	kfree(out_mad);
871 	return err;
872 }
873 
874 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
875 			     union ib_gid *gid)
876 {
877 	if (rdma_protocol_ib(ibdev, port))
878 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
879 	return 0;
880 }
881 
882 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
883 {
884 	union sl2vl_tbl_to_u64 sl2vl64;
885 	struct ib_smp *in_mad  = NULL;
886 	struct ib_smp *out_mad = NULL;
887 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
888 	int err = -ENOMEM;
889 	int jj;
890 
891 	if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
892 		*sl2vl_tbl = 0;
893 		return 0;
894 	}
895 
896 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
897 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
898 	if (!in_mad || !out_mad)
899 		goto out;
900 
901 	init_query_mad(in_mad);
902 	in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
903 	in_mad->attr_mod = 0;
904 
905 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
906 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
907 
908 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
909 			   in_mad, out_mad);
910 	if (err)
911 		goto out;
912 
913 	for (jj = 0; jj < 8; jj++)
914 		sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
915 	*sl2vl_tbl = sl2vl64.sl64;
916 
917 out:
918 	kfree(in_mad);
919 	kfree(out_mad);
920 	return err;
921 }
922 
923 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
924 {
925 	u64 sl2vl;
926 	int i;
927 	int err;
928 
929 	for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
930 		if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
931 			continue;
932 		err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
933 		if (err) {
934 			pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
935 			       i, err);
936 			sl2vl = 0;
937 		}
938 		atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
939 	}
940 }
941 
942 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
943 			 u16 *pkey, int netw_view)
944 {
945 	struct ib_smp *in_mad  = NULL;
946 	struct ib_smp *out_mad = NULL;
947 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
948 	int err = -ENOMEM;
949 
950 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
951 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
952 	if (!in_mad || !out_mad)
953 		goto out;
954 
955 	init_query_mad(in_mad);
956 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
957 	in_mad->attr_mod = cpu_to_be32(index / 32);
958 
959 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
960 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
961 
962 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
963 			   in_mad, out_mad);
964 	if (err)
965 		goto out;
966 
967 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
968 
969 out:
970 	kfree(in_mad);
971 	kfree(out_mad);
972 	return err;
973 }
974 
975 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
976 {
977 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
978 }
979 
980 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
981 				 struct ib_device_modify *props)
982 {
983 	struct mlx4_cmd_mailbox *mailbox;
984 	unsigned long flags;
985 
986 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
987 		return -EOPNOTSUPP;
988 
989 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
990 		return 0;
991 
992 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
993 		return -EOPNOTSUPP;
994 
995 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
996 	memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
997 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
998 
999 	/*
1000 	 * If possible, pass node desc to FW, so it can generate
1001 	 * a 144 trap.  If cmd fails, just ignore.
1002 	 */
1003 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1004 	if (IS_ERR(mailbox))
1005 		return 0;
1006 
1007 	memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1008 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1009 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1010 
1011 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1012 
1013 	return 0;
1014 }
1015 
1016 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1017 			    u32 cap_mask)
1018 {
1019 	struct mlx4_cmd_mailbox *mailbox;
1020 	int err;
1021 
1022 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1023 	if (IS_ERR(mailbox))
1024 		return PTR_ERR(mailbox);
1025 
1026 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1027 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
1028 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1029 	} else {
1030 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1031 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1032 	}
1033 
1034 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1035 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1036 		       MLX4_CMD_WRAPPED);
1037 
1038 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
1039 	return err;
1040 }
1041 
1042 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1043 			       struct ib_port_modify *props)
1044 {
1045 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1046 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1047 	struct ib_port_attr attr;
1048 	u32 cap_mask;
1049 	int err;
1050 
1051 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
1052 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
1053 	 * violations and port capabilities are not meaningful.
1054 	 */
1055 	if (is_eth)
1056 		return 0;
1057 
1058 	mutex_lock(&mdev->cap_mask_mutex);
1059 
1060 	err = ib_query_port(ibdev, port, &attr);
1061 	if (err)
1062 		goto out;
1063 
1064 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1065 		~props->clr_port_cap_mask;
1066 
1067 	err = mlx4_ib_SET_PORT(mdev, port,
1068 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
1069 			       cap_mask);
1070 
1071 out:
1072 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1073 	return err;
1074 }
1075 
1076 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1077 						  struct ib_udata *udata)
1078 {
1079 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
1080 	struct mlx4_ib_ucontext *context;
1081 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1082 	struct mlx4_ib_alloc_ucontext_resp resp;
1083 	int err;
1084 
1085 	if (!dev->ib_active)
1086 		return ERR_PTR(-EAGAIN);
1087 
1088 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1089 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1090 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1091 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1092 	} else {
1093 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
1094 		resp.qp_tab_size      = dev->dev->caps.num_qps;
1095 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1096 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1097 		resp.cqe_size	      = dev->dev->caps.cqe_size;
1098 	}
1099 
1100 	context = kzalloc(sizeof(*context), GFP_KERNEL);
1101 	if (!context)
1102 		return ERR_PTR(-ENOMEM);
1103 
1104 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1105 	if (err) {
1106 		kfree(context);
1107 		return ERR_PTR(err);
1108 	}
1109 
1110 	INIT_LIST_HEAD(&context->db_page_list);
1111 	mutex_init(&context->db_page_mutex);
1112 
1113 	INIT_LIST_HEAD(&context->wqn_ranges_list);
1114 	mutex_init(&context->wqn_ranges_mutex);
1115 
1116 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1117 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1118 	else
1119 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1120 
1121 	if (err) {
1122 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1123 		kfree(context);
1124 		return ERR_PTR(-EFAULT);
1125 	}
1126 
1127 	return &context->ibucontext;
1128 }
1129 
1130 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1131 {
1132 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1133 
1134 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1135 	kfree(context);
1136 
1137 	return 0;
1138 }
1139 
1140 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1141 {
1142 	/* vma_open is called when a new VMA is created on top of our VMA.
1143 	 * This is done through either mremap flow or split_vma (usually due
1144 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
1145 	 * vma, as this VMA is strongly hardware related. Therefore we set the
1146 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1147 	 * calling us again and trying to do incorrect actions. We assume that
1148 	 * the original vma size is exactly a single page that there will be no
1149 	 * "splitting" operations on.
1150 	 */
1151 	area->vm_ops = NULL;
1152 }
1153 
1154 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1155 {
1156 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1157 
1158 	/* It's guaranteed that all VMAs opened on a FD are closed before the
1159 	 * file itself is closed, therefore no sync is needed with the regular
1160 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1161 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1162 	 * The close operation is usually called under mm->mmap_sem except when
1163 	 * process is exiting.  The exiting case is handled explicitly as part
1164 	 * of mlx4_ib_disassociate_ucontext.
1165 	 */
1166 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1167 				area->vm_private_data;
1168 
1169 	/* set the vma context pointer to null in the mlx4_ib driver's private
1170 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1171 	 */
1172 	mlx4_ib_vma_priv_data->vma = NULL;
1173 }
1174 
1175 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1176 	.open = mlx4_ib_vma_open,
1177 	.close = mlx4_ib_vma_close
1178 };
1179 
1180 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1181 {
1182 	int i;
1183 	struct vm_area_struct *vma;
1184 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1185 
1186 	/* need to protect from a race on closing the vma as part of
1187 	 * mlx4_ib_vma_close().
1188 	 */
1189 	for (i = 0; i < HW_BAR_COUNT; i++) {
1190 		vma = context->hw_bar_info[i].vma;
1191 		if (!vma)
1192 			continue;
1193 
1194 		zap_vma_ptes(context->hw_bar_info[i].vma,
1195 			     context->hw_bar_info[i].vma->vm_start, PAGE_SIZE);
1196 
1197 		context->hw_bar_info[i].vma->vm_flags &=
1198 			~(VM_SHARED | VM_MAYSHARE);
1199 		/* context going to be destroyed, should not access ops any more */
1200 		context->hw_bar_info[i].vma->vm_ops = NULL;
1201 	}
1202 }
1203 
1204 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1205 				 struct mlx4_ib_vma_private_data *vma_private_data)
1206 {
1207 	vma_private_data->vma = vma;
1208 	vma->vm_private_data = vma_private_data;
1209 	vma->vm_ops =  &mlx4_ib_vm_ops;
1210 }
1211 
1212 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1213 {
1214 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1215 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1216 
1217 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1218 		return -EINVAL;
1219 
1220 	if (vma->vm_pgoff == 0) {
1221 		/* We prevent double mmaping on same context */
1222 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1223 			return -EINVAL;
1224 
1225 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1226 
1227 		if (io_remap_pfn_range(vma, vma->vm_start,
1228 				       to_mucontext(context)->uar.pfn,
1229 				       PAGE_SIZE, vma->vm_page_prot))
1230 			return -EAGAIN;
1231 
1232 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1233 
1234 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1235 		/* We prevent double mmaping on same context */
1236 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1237 			return -EINVAL;
1238 
1239 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1240 
1241 		if (io_remap_pfn_range(vma, vma->vm_start,
1242 				       to_mucontext(context)->uar.pfn +
1243 				       dev->dev->caps.num_uars,
1244 				       PAGE_SIZE, vma->vm_page_prot))
1245 			return -EAGAIN;
1246 
1247 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1248 
1249 	} else if (vma->vm_pgoff == 3) {
1250 		struct mlx4_clock_params params;
1251 		int ret;
1252 
1253 		/* We prevent double mmaping on same context */
1254 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1255 			return -EINVAL;
1256 
1257 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1258 
1259 		if (ret)
1260 			return ret;
1261 
1262 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1263 		if (io_remap_pfn_range(vma, vma->vm_start,
1264 				       (pci_resource_start(dev->dev->persist->pdev,
1265 							   params.bar) +
1266 					params.offset)
1267 				       >> PAGE_SHIFT,
1268 				       PAGE_SIZE, vma->vm_page_prot))
1269 			return -EAGAIN;
1270 
1271 		mlx4_ib_set_vma_data(vma,
1272 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1273 	} else {
1274 		return -EINVAL;
1275 	}
1276 
1277 	return 0;
1278 }
1279 
1280 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1281 				      struct ib_ucontext *context,
1282 				      struct ib_udata *udata)
1283 {
1284 	struct mlx4_ib_pd *pd;
1285 	int err;
1286 
1287 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1288 	if (!pd)
1289 		return ERR_PTR(-ENOMEM);
1290 
1291 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1292 	if (err) {
1293 		kfree(pd);
1294 		return ERR_PTR(err);
1295 	}
1296 
1297 	if (context)
1298 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1299 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1300 			kfree(pd);
1301 			return ERR_PTR(-EFAULT);
1302 		}
1303 	return &pd->ibpd;
1304 }
1305 
1306 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1307 {
1308 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1309 	kfree(pd);
1310 
1311 	return 0;
1312 }
1313 
1314 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1315 					  struct ib_ucontext *context,
1316 					  struct ib_udata *udata)
1317 {
1318 	struct mlx4_ib_xrcd *xrcd;
1319 	struct ib_cq_init_attr cq_attr = {};
1320 	int err;
1321 
1322 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1323 		return ERR_PTR(-ENOSYS);
1324 
1325 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1326 	if (!xrcd)
1327 		return ERR_PTR(-ENOMEM);
1328 
1329 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1330 	if (err)
1331 		goto err1;
1332 
1333 	xrcd->pd = ib_alloc_pd(ibdev, 0);
1334 	if (IS_ERR(xrcd->pd)) {
1335 		err = PTR_ERR(xrcd->pd);
1336 		goto err2;
1337 	}
1338 
1339 	cq_attr.cqe = 1;
1340 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1341 	if (IS_ERR(xrcd->cq)) {
1342 		err = PTR_ERR(xrcd->cq);
1343 		goto err3;
1344 	}
1345 
1346 	return &xrcd->ibxrcd;
1347 
1348 err3:
1349 	ib_dealloc_pd(xrcd->pd);
1350 err2:
1351 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1352 err1:
1353 	kfree(xrcd);
1354 	return ERR_PTR(err);
1355 }
1356 
1357 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1358 {
1359 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1360 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1361 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1362 	kfree(xrcd);
1363 
1364 	return 0;
1365 }
1366 
1367 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1368 {
1369 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1370 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1371 	struct mlx4_ib_gid_entry *ge;
1372 
1373 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1374 	if (!ge)
1375 		return -ENOMEM;
1376 
1377 	ge->gid = *gid;
1378 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1379 		ge->port = mqp->port;
1380 		ge->added = 1;
1381 	}
1382 
1383 	mutex_lock(&mqp->mutex);
1384 	list_add_tail(&ge->list, &mqp->gid_list);
1385 	mutex_unlock(&mqp->mutex);
1386 
1387 	return 0;
1388 }
1389 
1390 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1391 					  struct mlx4_ib_counters *ctr_table)
1392 {
1393 	struct counter_index *counter, *tmp_count;
1394 
1395 	mutex_lock(&ctr_table->mutex);
1396 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1397 				 list) {
1398 		if (counter->allocated)
1399 			mlx4_counter_free(ibdev->dev, counter->index);
1400 		list_del(&counter->list);
1401 		kfree(counter);
1402 	}
1403 	mutex_unlock(&ctr_table->mutex);
1404 }
1405 
1406 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1407 		   union ib_gid *gid)
1408 {
1409 	struct net_device *ndev;
1410 	int ret = 0;
1411 
1412 	if (!mqp->port)
1413 		return 0;
1414 
1415 	spin_lock_bh(&mdev->iboe.lock);
1416 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1417 	if (ndev)
1418 		dev_hold(ndev);
1419 	spin_unlock_bh(&mdev->iboe.lock);
1420 
1421 	if (ndev) {
1422 		ret = 1;
1423 		dev_put(ndev);
1424 	}
1425 
1426 	return ret;
1427 }
1428 
1429 struct mlx4_ib_steering {
1430 	struct list_head list;
1431 	struct mlx4_flow_reg_id reg_id;
1432 	union ib_gid gid;
1433 };
1434 
1435 #define LAST_ETH_FIELD vlan_tag
1436 #define LAST_IB_FIELD sl
1437 #define LAST_IPV4_FIELD dst_ip
1438 #define LAST_TCP_UDP_FIELD src_port
1439 
1440 /* Field is the last supported field */
1441 #define FIELDS_NOT_SUPPORTED(filter, field)\
1442 	memchr_inv((void *)&filter.field  +\
1443 		   sizeof(filter.field), 0,\
1444 		   sizeof(filter) -\
1445 		   offsetof(typeof(filter), field) -\
1446 		   sizeof(filter.field))
1447 
1448 static int parse_flow_attr(struct mlx4_dev *dev,
1449 			   u32 qp_num,
1450 			   union ib_flow_spec *ib_spec,
1451 			   struct _rule_hw *mlx4_spec)
1452 {
1453 	enum mlx4_net_trans_rule_id type;
1454 
1455 	switch (ib_spec->type) {
1456 	case IB_FLOW_SPEC_ETH:
1457 		if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1458 			return -ENOTSUPP;
1459 
1460 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1461 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1462 		       ETH_ALEN);
1463 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1464 		       ETH_ALEN);
1465 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1466 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1467 		break;
1468 	case IB_FLOW_SPEC_IB:
1469 		if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1470 			return -ENOTSUPP;
1471 
1472 		type = MLX4_NET_TRANS_RULE_ID_IB;
1473 		mlx4_spec->ib.l3_qpn =
1474 			cpu_to_be32(qp_num);
1475 		mlx4_spec->ib.qpn_mask =
1476 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1477 		break;
1478 
1479 
1480 	case IB_FLOW_SPEC_IPV4:
1481 		if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1482 			return -ENOTSUPP;
1483 
1484 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1485 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1486 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1487 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1488 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1489 		break;
1490 
1491 	case IB_FLOW_SPEC_TCP:
1492 	case IB_FLOW_SPEC_UDP:
1493 		if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1494 			return -ENOTSUPP;
1495 
1496 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1497 					MLX4_NET_TRANS_RULE_ID_TCP :
1498 					MLX4_NET_TRANS_RULE_ID_UDP;
1499 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1500 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1501 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1502 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1503 		break;
1504 
1505 	default:
1506 		return -EINVAL;
1507 	}
1508 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1509 	    mlx4_hw_rule_sz(dev, type) < 0)
1510 		return -EINVAL;
1511 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1512 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1513 	return mlx4_hw_rule_sz(dev, type);
1514 }
1515 
1516 struct default_rules {
1517 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1518 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1519 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1520 	__u8  link_layer;
1521 };
1522 static const struct default_rules default_table[] = {
1523 	{
1524 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1525 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1526 		.rules_create_list = {IB_FLOW_SPEC_IB},
1527 		.link_layer = IB_LINK_LAYER_INFINIBAND
1528 	}
1529 };
1530 
1531 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1532 					 struct ib_flow_attr *flow_attr)
1533 {
1534 	int i, j, k;
1535 	void *ib_flow;
1536 	const struct default_rules *pdefault_rules = default_table;
1537 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1538 
1539 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1540 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1541 		memset(&field_types, 0, sizeof(field_types));
1542 
1543 		if (link_layer != pdefault_rules->link_layer)
1544 			continue;
1545 
1546 		ib_flow = flow_attr + 1;
1547 		/* we assume the specs are sorted */
1548 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1549 		     j < flow_attr->num_of_specs; k++) {
1550 			union ib_flow_spec *current_flow =
1551 				(union ib_flow_spec *)ib_flow;
1552 
1553 			/* same layer but different type */
1554 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1555 			     (pdefault_rules->mandatory_fields[k] &
1556 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1557 			    (current_flow->type !=
1558 			     pdefault_rules->mandatory_fields[k]))
1559 				goto out;
1560 
1561 			/* same layer, try match next one */
1562 			if (current_flow->type ==
1563 			    pdefault_rules->mandatory_fields[k]) {
1564 				j++;
1565 				ib_flow +=
1566 					((union ib_flow_spec *)ib_flow)->size;
1567 			}
1568 		}
1569 
1570 		ib_flow = flow_attr + 1;
1571 		for (j = 0; j < flow_attr->num_of_specs;
1572 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1573 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1574 				/* same layer and same type */
1575 				if (((union ib_flow_spec *)ib_flow)->type ==
1576 				    pdefault_rules->mandatory_not_fields[k])
1577 					goto out;
1578 
1579 		return i;
1580 	}
1581 out:
1582 	return -1;
1583 }
1584 
1585 static int __mlx4_ib_create_default_rules(
1586 		struct mlx4_ib_dev *mdev,
1587 		struct ib_qp *qp,
1588 		const struct default_rules *pdefault_rules,
1589 		struct _rule_hw *mlx4_spec) {
1590 	int size = 0;
1591 	int i;
1592 
1593 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1594 		int ret;
1595 		union ib_flow_spec ib_spec;
1596 		switch (pdefault_rules->rules_create_list[i]) {
1597 		case 0:
1598 			/* no rule */
1599 			continue;
1600 		case IB_FLOW_SPEC_IB:
1601 			ib_spec.type = IB_FLOW_SPEC_IB;
1602 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1603 
1604 			break;
1605 		default:
1606 			/* invalid rule */
1607 			return -EINVAL;
1608 		}
1609 		/* We must put empty rule, qpn is being ignored */
1610 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1611 				      mlx4_spec);
1612 		if (ret < 0) {
1613 			pr_info("invalid parsing\n");
1614 			return -EINVAL;
1615 		}
1616 
1617 		mlx4_spec = (void *)mlx4_spec + ret;
1618 		size += ret;
1619 	}
1620 	return size;
1621 }
1622 
1623 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1624 			  int domain,
1625 			  enum mlx4_net_trans_promisc_mode flow_type,
1626 			  u64 *reg_id)
1627 {
1628 	int ret, i;
1629 	int size = 0;
1630 	void *ib_flow;
1631 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1632 	struct mlx4_cmd_mailbox *mailbox;
1633 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1634 	int default_flow;
1635 
1636 	static const u16 __mlx4_domain[] = {
1637 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1638 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1639 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1640 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1641 	};
1642 
1643 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1644 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1645 		return -EINVAL;
1646 	}
1647 
1648 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1649 		pr_err("Invalid domain value %d\n", domain);
1650 		return -EINVAL;
1651 	}
1652 
1653 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1654 		return -EINVAL;
1655 
1656 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1657 	if (IS_ERR(mailbox))
1658 		return PTR_ERR(mailbox);
1659 	ctrl = mailbox->buf;
1660 
1661 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1662 				 flow_attr->priority);
1663 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1664 	ctrl->port = flow_attr->port;
1665 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1666 
1667 	ib_flow = flow_attr + 1;
1668 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1669 	/* Add default flows */
1670 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1671 	if (default_flow >= 0) {
1672 		ret = __mlx4_ib_create_default_rules(
1673 				mdev, qp, default_table + default_flow,
1674 				mailbox->buf + size);
1675 		if (ret < 0) {
1676 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1677 			return -EINVAL;
1678 		}
1679 		size += ret;
1680 	}
1681 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1682 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1683 				      mailbox->buf + size);
1684 		if (ret < 0) {
1685 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1686 			return -EINVAL;
1687 		}
1688 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1689 		size += ret;
1690 	}
1691 
1692 	if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1693 	    flow_attr->num_of_specs == 1) {
1694 		struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1695 		enum ib_flow_spec_type header_spec =
1696 			((union ib_flow_spec *)(flow_attr + 1))->type;
1697 
1698 		if (header_spec == IB_FLOW_SPEC_ETH)
1699 			mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1700 	}
1701 
1702 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1703 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1704 			   MLX4_CMD_NATIVE);
1705 	if (ret == -ENOMEM)
1706 		pr_err("mcg table is full. Fail to register network rule.\n");
1707 	else if (ret == -ENXIO)
1708 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1709 	else if (ret)
1710 		pr_err("Invalid argument. Fail to register network rule.\n");
1711 
1712 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1713 	return ret;
1714 }
1715 
1716 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1717 {
1718 	int err;
1719 	err = mlx4_cmd(dev, reg_id, 0, 0,
1720 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1721 		       MLX4_CMD_NATIVE);
1722 	if (err)
1723 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1724 		       reg_id);
1725 	return err;
1726 }
1727 
1728 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1729 				    u64 *reg_id)
1730 {
1731 	void *ib_flow;
1732 	union ib_flow_spec *ib_spec;
1733 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1734 	int err = 0;
1735 
1736 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1737 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1738 		return 0; /* do nothing */
1739 
1740 	ib_flow = flow_attr + 1;
1741 	ib_spec = (union ib_flow_spec *)ib_flow;
1742 
1743 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1744 		return 0; /* do nothing */
1745 
1746 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1747 				    flow_attr->port, qp->qp_num,
1748 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1749 				    reg_id);
1750 	return err;
1751 }
1752 
1753 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1754 				      struct ib_flow_attr *flow_attr,
1755 				      enum mlx4_net_trans_promisc_mode *type)
1756 {
1757 	int err = 0;
1758 
1759 	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1760 	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1761 	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1762 		return -EOPNOTSUPP;
1763 	}
1764 
1765 	if (flow_attr->num_of_specs == 0) {
1766 		type[0] = MLX4_FS_MC_SNIFFER;
1767 		type[1] = MLX4_FS_UC_SNIFFER;
1768 	} else {
1769 		union ib_flow_spec *ib_spec;
1770 
1771 		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1772 		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1773 			return -EINVAL;
1774 
1775 		/* if all is zero than MC and UC */
1776 		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1777 			type[0] = MLX4_FS_MC_SNIFFER;
1778 			type[1] = MLX4_FS_UC_SNIFFER;
1779 		} else {
1780 			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1781 					    ib_spec->eth.mask.dst_mac[1],
1782 					    ib_spec->eth.mask.dst_mac[2],
1783 					    ib_spec->eth.mask.dst_mac[3],
1784 					    ib_spec->eth.mask.dst_mac[4],
1785 					    ib_spec->eth.mask.dst_mac[5]};
1786 
1787 			/* Above xor was only on MC bit, non empty mask is valid
1788 			 * only if this bit is set and rest are zero.
1789 			 */
1790 			if (!is_zero_ether_addr(&mac[0]))
1791 				return -EINVAL;
1792 
1793 			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1794 				type[0] = MLX4_FS_MC_SNIFFER;
1795 			else
1796 				type[0] = MLX4_FS_UC_SNIFFER;
1797 		}
1798 	}
1799 
1800 	return err;
1801 }
1802 
1803 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1804 				    struct ib_flow_attr *flow_attr,
1805 				    int domain, struct ib_udata *udata)
1806 {
1807 	int err = 0, i = 0, j = 0;
1808 	struct mlx4_ib_flow *mflow;
1809 	enum mlx4_net_trans_promisc_mode type[2];
1810 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1811 	int is_bonded = mlx4_is_bonded(dev);
1812 
1813 	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1814 		return ERR_PTR(-EINVAL);
1815 
1816 	if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1817 		return ERR_PTR(-EOPNOTSUPP);
1818 
1819 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1820 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1821 		return ERR_PTR(-EOPNOTSUPP);
1822 
1823 	if (udata &&
1824 	    udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1825 		return ERR_PTR(-EOPNOTSUPP);
1826 
1827 	memset(type, 0, sizeof(type));
1828 
1829 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1830 	if (!mflow) {
1831 		err = -ENOMEM;
1832 		goto err_free;
1833 	}
1834 
1835 	switch (flow_attr->type) {
1836 	case IB_FLOW_ATTR_NORMAL:
1837 		/* If dont trap flag (continue match) is set, under specific
1838 		 * condition traffic be replicated to given qp,
1839 		 * without stealing it
1840 		 */
1841 		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1842 			err = mlx4_ib_add_dont_trap_rule(dev,
1843 							 flow_attr,
1844 							 type);
1845 			if (err)
1846 				goto err_free;
1847 		} else {
1848 			type[0] = MLX4_FS_REGULAR;
1849 		}
1850 		break;
1851 
1852 	case IB_FLOW_ATTR_ALL_DEFAULT:
1853 		type[0] = MLX4_FS_ALL_DEFAULT;
1854 		break;
1855 
1856 	case IB_FLOW_ATTR_MC_DEFAULT:
1857 		type[0] = MLX4_FS_MC_DEFAULT;
1858 		break;
1859 
1860 	case IB_FLOW_ATTR_SNIFFER:
1861 		type[0] = MLX4_FS_MIRROR_RX_PORT;
1862 		type[1] = MLX4_FS_MIRROR_SX_PORT;
1863 		break;
1864 
1865 	default:
1866 		err = -EINVAL;
1867 		goto err_free;
1868 	}
1869 
1870 	while (i < ARRAY_SIZE(type) && type[i]) {
1871 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1872 					    &mflow->reg_id[i].id);
1873 		if (err)
1874 			goto err_create_flow;
1875 		if (is_bonded) {
1876 			/* Application always sees one port so the mirror rule
1877 			 * must be on port #2
1878 			 */
1879 			flow_attr->port = 2;
1880 			err = __mlx4_ib_create_flow(qp, flow_attr,
1881 						    domain, type[j],
1882 						    &mflow->reg_id[j].mirror);
1883 			flow_attr->port = 1;
1884 			if (err)
1885 				goto err_create_flow;
1886 			j++;
1887 		}
1888 
1889 		i++;
1890 	}
1891 
1892 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1893 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1894 					       &mflow->reg_id[i].id);
1895 		if (err)
1896 			goto err_create_flow;
1897 
1898 		if (is_bonded) {
1899 			flow_attr->port = 2;
1900 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1901 						       &mflow->reg_id[j].mirror);
1902 			flow_attr->port = 1;
1903 			if (err)
1904 				goto err_create_flow;
1905 			j++;
1906 		}
1907 		/* function to create mirror rule */
1908 		i++;
1909 	}
1910 
1911 	return &mflow->ibflow;
1912 
1913 err_create_flow:
1914 	while (i) {
1915 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1916 					     mflow->reg_id[i].id);
1917 		i--;
1918 	}
1919 
1920 	while (j) {
1921 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1922 					     mflow->reg_id[j].mirror);
1923 		j--;
1924 	}
1925 err_free:
1926 	kfree(mflow);
1927 	return ERR_PTR(err);
1928 }
1929 
1930 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1931 {
1932 	int err, ret = 0;
1933 	int i = 0;
1934 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1935 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1936 
1937 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1938 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1939 		if (err)
1940 			ret = err;
1941 		if (mflow->reg_id[i].mirror) {
1942 			err = __mlx4_ib_destroy_flow(mdev->dev,
1943 						     mflow->reg_id[i].mirror);
1944 			if (err)
1945 				ret = err;
1946 		}
1947 		i++;
1948 	}
1949 
1950 	kfree(mflow);
1951 	return ret;
1952 }
1953 
1954 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1955 {
1956 	int err;
1957 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1958 	struct mlx4_dev	*dev = mdev->dev;
1959 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1960 	struct mlx4_ib_steering *ib_steering = NULL;
1961 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1962 	struct mlx4_flow_reg_id	reg_id;
1963 
1964 	if (mdev->dev->caps.steering_mode ==
1965 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1966 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1967 		if (!ib_steering)
1968 			return -ENOMEM;
1969 	}
1970 
1971 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1972 				    !!(mqp->flags &
1973 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1974 				    prot, &reg_id.id);
1975 	if (err) {
1976 		pr_err("multicast attach op failed, err %d\n", err);
1977 		goto err_malloc;
1978 	}
1979 
1980 	reg_id.mirror = 0;
1981 	if (mlx4_is_bonded(dev)) {
1982 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1983 					    (mqp->port == 1) ? 2 : 1,
1984 					    !!(mqp->flags &
1985 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1986 					    prot, &reg_id.mirror);
1987 		if (err)
1988 			goto err_add;
1989 	}
1990 
1991 	err = add_gid_entry(ibqp, gid);
1992 	if (err)
1993 		goto err_add;
1994 
1995 	if (ib_steering) {
1996 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1997 		ib_steering->reg_id = reg_id;
1998 		mutex_lock(&mqp->mutex);
1999 		list_add(&ib_steering->list, &mqp->steering_rules);
2000 		mutex_unlock(&mqp->mutex);
2001 	}
2002 	return 0;
2003 
2004 err_add:
2005 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2006 			      prot, reg_id.id);
2007 	if (reg_id.mirror)
2008 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2009 				      prot, reg_id.mirror);
2010 err_malloc:
2011 	kfree(ib_steering);
2012 
2013 	return err;
2014 }
2015 
2016 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2017 {
2018 	struct mlx4_ib_gid_entry *ge;
2019 	struct mlx4_ib_gid_entry *tmp;
2020 	struct mlx4_ib_gid_entry *ret = NULL;
2021 
2022 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2023 		if (!memcmp(raw, ge->gid.raw, 16)) {
2024 			ret = ge;
2025 			break;
2026 		}
2027 	}
2028 
2029 	return ret;
2030 }
2031 
2032 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2033 {
2034 	int err;
2035 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2036 	struct mlx4_dev *dev = mdev->dev;
2037 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2038 	struct net_device *ndev;
2039 	struct mlx4_ib_gid_entry *ge;
2040 	struct mlx4_flow_reg_id reg_id = {0, 0};
2041 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2042 
2043 	if (mdev->dev->caps.steering_mode ==
2044 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
2045 		struct mlx4_ib_steering *ib_steering;
2046 
2047 		mutex_lock(&mqp->mutex);
2048 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2049 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2050 				list_del(&ib_steering->list);
2051 				break;
2052 			}
2053 		}
2054 		mutex_unlock(&mqp->mutex);
2055 		if (&ib_steering->list == &mqp->steering_rules) {
2056 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2057 			return -EINVAL;
2058 		}
2059 		reg_id = ib_steering->reg_id;
2060 		kfree(ib_steering);
2061 	}
2062 
2063 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2064 				    prot, reg_id.id);
2065 	if (err)
2066 		return err;
2067 
2068 	if (mlx4_is_bonded(dev)) {
2069 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2070 					    prot, reg_id.mirror);
2071 		if (err)
2072 			return err;
2073 	}
2074 
2075 	mutex_lock(&mqp->mutex);
2076 	ge = find_gid_entry(mqp, gid->raw);
2077 	if (ge) {
2078 		spin_lock_bh(&mdev->iboe.lock);
2079 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2080 		if (ndev)
2081 			dev_hold(ndev);
2082 		spin_unlock_bh(&mdev->iboe.lock);
2083 		if (ndev)
2084 			dev_put(ndev);
2085 		list_del(&ge->list);
2086 		kfree(ge);
2087 	} else
2088 		pr_warn("could not find mgid entry\n");
2089 
2090 	mutex_unlock(&mqp->mutex);
2091 
2092 	return 0;
2093 }
2094 
2095 static int init_node_data(struct mlx4_ib_dev *dev)
2096 {
2097 	struct ib_smp *in_mad  = NULL;
2098 	struct ib_smp *out_mad = NULL;
2099 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2100 	int err = -ENOMEM;
2101 
2102 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2103 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2104 	if (!in_mad || !out_mad)
2105 		goto out;
2106 
2107 	init_query_mad(in_mad);
2108 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2109 	if (mlx4_is_master(dev->dev))
2110 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2111 
2112 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2113 	if (err)
2114 		goto out;
2115 
2116 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2117 
2118 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2119 
2120 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2121 	if (err)
2122 		goto out;
2123 
2124 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2125 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2126 
2127 out:
2128 	kfree(in_mad);
2129 	kfree(out_mad);
2130 	return err;
2131 }
2132 
2133 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2134 			char *buf)
2135 {
2136 	struct mlx4_ib_dev *dev =
2137 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2138 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2139 }
2140 
2141 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2142 			char *buf)
2143 {
2144 	struct mlx4_ib_dev *dev =
2145 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2146 	return sprintf(buf, "%x\n", dev->dev->rev_id);
2147 }
2148 
2149 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2150 			  char *buf)
2151 {
2152 	struct mlx4_ib_dev *dev =
2153 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2154 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2155 		       dev->dev->board_id);
2156 }
2157 
2158 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2159 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2160 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2161 
2162 static struct device_attribute *mlx4_class_attributes[] = {
2163 	&dev_attr_hw_rev,
2164 	&dev_attr_hca_type,
2165 	&dev_attr_board_id
2166 };
2167 
2168 struct diag_counter {
2169 	const char *name;
2170 	u32 offset;
2171 };
2172 
2173 #define DIAG_COUNTER(_name, _offset)			\
2174 	{ .name = #_name, .offset = _offset }
2175 
2176 static const struct diag_counter diag_basic[] = {
2177 	DIAG_COUNTER(rq_num_lle, 0x00),
2178 	DIAG_COUNTER(sq_num_lle, 0x04),
2179 	DIAG_COUNTER(rq_num_lqpoe, 0x08),
2180 	DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2181 	DIAG_COUNTER(rq_num_lpe, 0x18),
2182 	DIAG_COUNTER(sq_num_lpe, 0x1C),
2183 	DIAG_COUNTER(rq_num_wrfe, 0x20),
2184 	DIAG_COUNTER(sq_num_wrfe, 0x24),
2185 	DIAG_COUNTER(sq_num_mwbe, 0x2C),
2186 	DIAG_COUNTER(sq_num_bre, 0x34),
2187 	DIAG_COUNTER(sq_num_rire, 0x44),
2188 	DIAG_COUNTER(rq_num_rire, 0x48),
2189 	DIAG_COUNTER(sq_num_rae, 0x4C),
2190 	DIAG_COUNTER(rq_num_rae, 0x50),
2191 	DIAG_COUNTER(sq_num_roe, 0x54),
2192 	DIAG_COUNTER(sq_num_tree, 0x5C),
2193 	DIAG_COUNTER(sq_num_rree, 0x64),
2194 	DIAG_COUNTER(rq_num_rnr, 0x68),
2195 	DIAG_COUNTER(sq_num_rnr, 0x6C),
2196 	DIAG_COUNTER(rq_num_oos, 0x100),
2197 	DIAG_COUNTER(sq_num_oos, 0x104),
2198 };
2199 
2200 static const struct diag_counter diag_ext[] = {
2201 	DIAG_COUNTER(rq_num_dup, 0x130),
2202 	DIAG_COUNTER(sq_num_to, 0x134),
2203 };
2204 
2205 static const struct diag_counter diag_device_only[] = {
2206 	DIAG_COUNTER(num_cqovf, 0x1A0),
2207 	DIAG_COUNTER(rq_num_udsdprd, 0x118),
2208 };
2209 
2210 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2211 						    u8 port_num)
2212 {
2213 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2214 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2215 
2216 	if (!diag[!!port_num].name)
2217 		return NULL;
2218 
2219 	return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2220 					  diag[!!port_num].num_counters,
2221 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
2222 }
2223 
2224 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2225 				struct rdma_hw_stats *stats,
2226 				u8 port, int index)
2227 {
2228 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
2229 	struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2230 	u32 hw_value[ARRAY_SIZE(diag_device_only) +
2231 		ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2232 	int ret;
2233 	int i;
2234 
2235 	ret = mlx4_query_diag_counters(dev->dev,
2236 				       MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2237 				       diag[!!port].offset, hw_value,
2238 				       diag[!!port].num_counters, port);
2239 
2240 	if (ret)
2241 		return ret;
2242 
2243 	for (i = 0; i < diag[!!port].num_counters; i++)
2244 		stats->value[i] = hw_value[i];
2245 
2246 	return diag[!!port].num_counters;
2247 }
2248 
2249 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2250 					 const char ***name,
2251 					 u32 **offset,
2252 					 u32 *num,
2253 					 bool port)
2254 {
2255 	u32 num_counters;
2256 
2257 	num_counters = ARRAY_SIZE(diag_basic);
2258 
2259 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2260 		num_counters += ARRAY_SIZE(diag_ext);
2261 
2262 	if (!port)
2263 		num_counters += ARRAY_SIZE(diag_device_only);
2264 
2265 	*name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2266 	if (!*name)
2267 		return -ENOMEM;
2268 
2269 	*offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2270 	if (!*offset)
2271 		goto err_name;
2272 
2273 	*num = num_counters;
2274 
2275 	return 0;
2276 
2277 err_name:
2278 	kfree(*name);
2279 	return -ENOMEM;
2280 }
2281 
2282 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2283 				       const char **name,
2284 				       u32 *offset,
2285 				       bool port)
2286 {
2287 	int i;
2288 	int j;
2289 
2290 	for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2291 		name[i] = diag_basic[i].name;
2292 		offset[i] = diag_basic[i].offset;
2293 	}
2294 
2295 	if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2296 		for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2297 			name[j] = diag_ext[i].name;
2298 			offset[j] = diag_ext[i].offset;
2299 		}
2300 	}
2301 
2302 	if (!port) {
2303 		for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2304 			name[j] = diag_device_only[i].name;
2305 			offset[j] = diag_device_only[i].offset;
2306 		}
2307 	}
2308 }
2309 
2310 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2311 {
2312 	struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2313 	int i;
2314 	int ret;
2315 	bool per_port = !!(ibdev->dev->caps.flags2 &
2316 		MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2317 
2318 	if (mlx4_is_slave(ibdev->dev))
2319 		return 0;
2320 
2321 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2322 		/* i == 1 means we are building port counters */
2323 		if (i && !per_port)
2324 			continue;
2325 
2326 		ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2327 						    &diag[i].offset,
2328 						    &diag[i].num_counters, i);
2329 		if (ret)
2330 			goto err_alloc;
2331 
2332 		mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2333 					   diag[i].offset, i);
2334 	}
2335 
2336 	ibdev->ib_dev.get_hw_stats	= mlx4_ib_get_hw_stats;
2337 	ibdev->ib_dev.alloc_hw_stats	= mlx4_ib_alloc_hw_stats;
2338 
2339 	return 0;
2340 
2341 err_alloc:
2342 	if (i) {
2343 		kfree(diag[i - 1].name);
2344 		kfree(diag[i - 1].offset);
2345 	}
2346 
2347 	return ret;
2348 }
2349 
2350 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2351 {
2352 	int i;
2353 
2354 	for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2355 		kfree(ibdev->diag_counters[i].offset);
2356 		kfree(ibdev->diag_counters[i].name);
2357 	}
2358 }
2359 
2360 #define MLX4_IB_INVALID_MAC	((u64)-1)
2361 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2362 			       struct net_device *dev,
2363 			       int port)
2364 {
2365 	u64 new_smac = 0;
2366 	u64 release_mac = MLX4_IB_INVALID_MAC;
2367 	struct mlx4_ib_qp *qp;
2368 
2369 	read_lock(&dev_base_lock);
2370 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
2371 	read_unlock(&dev_base_lock);
2372 
2373 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2374 
2375 	/* no need for update QP1 and mac registration in non-SRIOV */
2376 	if (!mlx4_is_mfunc(ibdev->dev))
2377 		return;
2378 
2379 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2380 	qp = ibdev->qp1_proxy[port - 1];
2381 	if (qp) {
2382 		int new_smac_index;
2383 		u64 old_smac;
2384 		struct mlx4_update_qp_params update_params;
2385 
2386 		mutex_lock(&qp->mutex);
2387 		old_smac = qp->pri.smac;
2388 		if (new_smac == old_smac)
2389 			goto unlock;
2390 
2391 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2392 
2393 		if (new_smac_index < 0)
2394 			goto unlock;
2395 
2396 		update_params.smac_index = new_smac_index;
2397 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2398 				   &update_params)) {
2399 			release_mac = new_smac;
2400 			goto unlock;
2401 		}
2402 		/* if old port was zero, no mac was yet registered for this QP */
2403 		if (qp->pri.smac_port)
2404 			release_mac = old_smac;
2405 		qp->pri.smac = new_smac;
2406 		qp->pri.smac_port = port;
2407 		qp->pri.smac_index = new_smac_index;
2408 	}
2409 
2410 unlock:
2411 	if (release_mac != MLX4_IB_INVALID_MAC)
2412 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
2413 	if (qp)
2414 		mutex_unlock(&qp->mutex);
2415 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2416 }
2417 
2418 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2419 				 struct net_device *dev,
2420 				 unsigned long event)
2421 
2422 {
2423 	struct mlx4_ib_iboe *iboe;
2424 	int update_qps_port = -1;
2425 	int port;
2426 
2427 	ASSERT_RTNL();
2428 
2429 	iboe = &ibdev->iboe;
2430 
2431 	spin_lock_bh(&iboe->lock);
2432 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2433 
2434 		iboe->netdevs[port - 1] =
2435 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2436 
2437 		if (dev == iboe->netdevs[port - 1] &&
2438 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2439 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2440 			update_qps_port = port;
2441 
2442 	}
2443 	spin_unlock_bh(&iboe->lock);
2444 
2445 	if (update_qps_port > 0)
2446 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2447 }
2448 
2449 static int mlx4_ib_netdev_event(struct notifier_block *this,
2450 				unsigned long event, void *ptr)
2451 {
2452 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2453 	struct mlx4_ib_dev *ibdev;
2454 
2455 	if (!net_eq(dev_net(dev), &init_net))
2456 		return NOTIFY_DONE;
2457 
2458 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2459 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2460 
2461 	return NOTIFY_DONE;
2462 }
2463 
2464 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2465 {
2466 	int port;
2467 	int slave;
2468 	int i;
2469 
2470 	if (mlx4_is_master(ibdev->dev)) {
2471 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2472 		     ++slave) {
2473 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2474 				for (i = 0;
2475 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2476 				     ++i) {
2477 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2478 					/* master has the identity virt2phys pkey mapping */
2479 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2480 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2481 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2482 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2483 				}
2484 			}
2485 		}
2486 		/* initialize pkey cache */
2487 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2488 			for (i = 0;
2489 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2490 			     ++i)
2491 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2492 					(i) ? 0 : 0xFFFF;
2493 		}
2494 	}
2495 }
2496 
2497 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2498 {
2499 	int i, j, eq = 0, total_eqs = 0;
2500 
2501 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2502 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2503 	if (!ibdev->eq_table)
2504 		return;
2505 
2506 	for (i = 1; i <= dev->caps.num_ports; i++) {
2507 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2508 		     j++, total_eqs++) {
2509 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2510 				continue;
2511 			ibdev->eq_table[eq] = total_eqs;
2512 			if (!mlx4_assign_eq(dev, i,
2513 					    &ibdev->eq_table[eq]))
2514 				eq++;
2515 			else
2516 				ibdev->eq_table[eq] = -1;
2517 		}
2518 	}
2519 
2520 	for (i = eq; i < dev->caps.num_comp_vectors;
2521 	     ibdev->eq_table[i++] = -1)
2522 		;
2523 
2524 	/* Advertise the new number of EQs to clients */
2525 	ibdev->ib_dev.num_comp_vectors = eq;
2526 }
2527 
2528 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2529 {
2530 	int i;
2531 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2532 
2533 	/* no eqs were allocated */
2534 	if (!ibdev->eq_table)
2535 		return;
2536 
2537 	/* Reset the advertised EQ number */
2538 	ibdev->ib_dev.num_comp_vectors = 0;
2539 
2540 	for (i = 0; i < total_eqs; i++)
2541 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2542 
2543 	kfree(ibdev->eq_table);
2544 	ibdev->eq_table = NULL;
2545 }
2546 
2547 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2548 			       struct ib_port_immutable *immutable)
2549 {
2550 	struct ib_port_attr attr;
2551 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2552 	int err;
2553 
2554 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2555 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2556 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2557 	} else {
2558 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2559 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2560 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2561 			immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2562 				RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2563 		immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2564 		if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2565 		    RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2566 			immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2567 	}
2568 
2569 	err = ib_query_port(ibdev, port_num, &attr);
2570 	if (err)
2571 		return err;
2572 
2573 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2574 	immutable->gid_tbl_len = attr.gid_tbl_len;
2575 
2576 	return 0;
2577 }
2578 
2579 static void get_fw_ver_str(struct ib_device *device, char *str)
2580 {
2581 	struct mlx4_ib_dev *dev =
2582 		container_of(device, struct mlx4_ib_dev, ib_dev);
2583 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2584 		 (int) (dev->dev->caps.fw_ver >> 32),
2585 		 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2586 		 (int) dev->dev->caps.fw_ver & 0xffff);
2587 }
2588 
2589 static void *mlx4_ib_add(struct mlx4_dev *dev)
2590 {
2591 	struct mlx4_ib_dev *ibdev;
2592 	int num_ports = 0;
2593 	int i, j;
2594 	int err;
2595 	struct mlx4_ib_iboe *iboe;
2596 	int ib_num_ports = 0;
2597 	int num_req_counters;
2598 	int allocated;
2599 	u32 counter_index;
2600 	struct counter_index *new_counter_index = NULL;
2601 
2602 	pr_info_once("%s", mlx4_ib_version);
2603 
2604 	num_ports = 0;
2605 	mlx4_foreach_ib_transport_port(i, dev)
2606 		num_ports++;
2607 
2608 	/* No point in registering a device with no ports... */
2609 	if (num_ports == 0)
2610 		return NULL;
2611 
2612 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2613 	if (!ibdev) {
2614 		dev_err(&dev->persist->pdev->dev,
2615 			"Device struct alloc failed\n");
2616 		return NULL;
2617 	}
2618 
2619 	iboe = &ibdev->iboe;
2620 
2621 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2622 		goto err_dealloc;
2623 
2624 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2625 		goto err_pd;
2626 
2627 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2628 				 PAGE_SIZE);
2629 	if (!ibdev->uar_map)
2630 		goto err_uar;
2631 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2632 
2633 	ibdev->dev = dev;
2634 	ibdev->bond_next_port	= 0;
2635 
2636 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2637 	ibdev->ib_dev.owner		= THIS_MODULE;
2638 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2639 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2640 	ibdev->num_ports		= num_ports;
2641 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2642 						1 : ibdev->num_ports;
2643 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2644 	ibdev->ib_dev.dev.parent	= &dev->persist->pdev->dev;
2645 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2646 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2647 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2648 
2649 	if (dev->caps.userspace_caps)
2650 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2651 	else
2652 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2653 
2654 	ibdev->ib_dev.uverbs_cmd_mask	=
2655 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2656 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2657 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2658 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2659 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2660 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2661 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2662 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2663 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2664 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2665 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2666 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2667 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2668 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2669 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2670 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2671 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2672 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2673 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2674 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2675 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2676 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2677 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2678 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2679 
2680 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2681 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2682 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2683 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2684 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2685 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2686 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2687 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2688 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2689 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2690 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2691 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2692 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2693 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2694 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2695 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2696 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2697 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2698 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2699 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2700 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2701 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2702 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2703 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2704 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2705 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2706 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2707 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2708 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2709 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2710 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2711 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2712 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2713 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2714 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2715 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2716 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2717 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2718 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2719 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2720 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2721 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2722 	ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2723 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2724 
2725 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2726 		(1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2727 
2728 	if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2729 	    ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2730 	    IB_LINK_LAYER_ETHERNET) ||
2731 	    (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2732 	    IB_LINK_LAYER_ETHERNET))) {
2733 		ibdev->ib_dev.create_wq		= mlx4_ib_create_wq;
2734 		ibdev->ib_dev.modify_wq		= mlx4_ib_modify_wq;
2735 		ibdev->ib_dev.destroy_wq	= mlx4_ib_destroy_wq;
2736 		ibdev->ib_dev.create_rwq_ind_table  =
2737 			mlx4_ib_create_rwq_ind_table;
2738 		ibdev->ib_dev.destroy_rwq_ind_table =
2739 			mlx4_ib_destroy_rwq_ind_table;
2740 		ibdev->ib_dev.uverbs_ex_cmd_mask |=
2741 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)	  |
2742 			(1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)	  |
2743 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)	  |
2744 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2745 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2746 	}
2747 
2748 	if (!mlx4_is_slave(ibdev->dev)) {
2749 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2750 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2751 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2752 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2753 	}
2754 
2755 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2756 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2757 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2758 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2759 
2760 		ibdev->ib_dev.uverbs_cmd_mask |=
2761 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2762 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2763 	}
2764 
2765 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2766 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2767 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2768 		ibdev->ib_dev.uverbs_cmd_mask |=
2769 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2770 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2771 	}
2772 
2773 	if (check_flow_steering_support(dev)) {
2774 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2775 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2776 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2777 
2778 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2779 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2780 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2781 	}
2782 
2783 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2784 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2785 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2786 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2787 
2788 	mlx4_ib_alloc_eqs(dev, ibdev);
2789 
2790 	spin_lock_init(&iboe->lock);
2791 
2792 	if (init_node_data(ibdev))
2793 		goto err_map;
2794 	mlx4_init_sl2vl_tbl(ibdev);
2795 
2796 	for (i = 0; i < ibdev->num_ports; ++i) {
2797 		mutex_init(&ibdev->counters_table[i].mutex);
2798 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2799 	}
2800 
2801 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2802 	for (i = 0; i < num_req_counters; ++i) {
2803 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2804 		allocated = 0;
2805 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2806 						IB_LINK_LAYER_ETHERNET) {
2807 			err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2808 						 MLX4_RES_USAGE_DRIVER);
2809 			/* if failed to allocate a new counter, use default */
2810 			if (err)
2811 				counter_index =
2812 					mlx4_get_default_counter_index(dev,
2813 								       i + 1);
2814 			else
2815 				allocated = 1;
2816 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2817 			counter_index = mlx4_get_default_counter_index(dev,
2818 								       i + 1);
2819 		}
2820 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2821 					    GFP_KERNEL);
2822 		if (!new_counter_index) {
2823 			if (allocated)
2824 				mlx4_counter_free(ibdev->dev, counter_index);
2825 			goto err_counter;
2826 		}
2827 		new_counter_index->index = counter_index;
2828 		new_counter_index->allocated = allocated;
2829 		list_add_tail(&new_counter_index->list,
2830 			      &ibdev->counters_table[i].counters_list);
2831 		ibdev->counters_table[i].default_counter = counter_index;
2832 		pr_info("counter index %d for port %d allocated %d\n",
2833 			counter_index, i + 1, allocated);
2834 	}
2835 	if (mlx4_is_bonded(dev))
2836 		for (i = 1; i < ibdev->num_ports ; ++i) {
2837 			new_counter_index =
2838 					kmalloc(sizeof(struct counter_index),
2839 						GFP_KERNEL);
2840 			if (!new_counter_index)
2841 				goto err_counter;
2842 			new_counter_index->index = counter_index;
2843 			new_counter_index->allocated = 0;
2844 			list_add_tail(&new_counter_index->list,
2845 				      &ibdev->counters_table[i].counters_list);
2846 			ibdev->counters_table[i].default_counter =
2847 								counter_index;
2848 		}
2849 
2850 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2851 		ib_num_ports++;
2852 
2853 	spin_lock_init(&ibdev->sm_lock);
2854 	mutex_init(&ibdev->cap_mask_mutex);
2855 	INIT_LIST_HEAD(&ibdev->qp_list);
2856 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2857 
2858 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2859 	    ib_num_ports) {
2860 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2861 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2862 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2863 					    &ibdev->steer_qpn_base, 0,
2864 					    MLX4_RES_USAGE_DRIVER);
2865 		if (err)
2866 			goto err_counter;
2867 
2868 		ibdev->ib_uc_qpns_bitmap =
2869 			kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2870 				      sizeof(long),
2871 				      GFP_KERNEL);
2872 		if (!ibdev->ib_uc_qpns_bitmap)
2873 			goto err_steer_qp_release;
2874 
2875 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2876 			bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2877 				    ibdev->steer_qpn_count);
2878 			err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2879 					dev, ibdev->steer_qpn_base,
2880 					ibdev->steer_qpn_base +
2881 					ibdev->steer_qpn_count - 1);
2882 			if (err)
2883 				goto err_steer_free_bitmap;
2884 		} else {
2885 			bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2886 				    ibdev->steer_qpn_count);
2887 		}
2888 	}
2889 
2890 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2891 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2892 
2893 	if (mlx4_ib_alloc_diag_counters(ibdev))
2894 		goto err_steer_free_bitmap;
2895 
2896 	ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2897 	if (ib_register_device(&ibdev->ib_dev, NULL))
2898 		goto err_diag_counters;
2899 
2900 	if (mlx4_ib_mad_init(ibdev))
2901 		goto err_reg;
2902 
2903 	if (mlx4_ib_init_sriov(ibdev))
2904 		goto err_mad;
2905 
2906 	if (!iboe->nb.notifier_call) {
2907 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
2908 		err = register_netdevice_notifier(&iboe->nb);
2909 		if (err) {
2910 			iboe->nb.notifier_call = NULL;
2911 			goto err_notif;
2912 		}
2913 	}
2914 	if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2915 		err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2916 		if (err)
2917 			goto err_notif;
2918 	}
2919 
2920 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2921 		if (device_create_file(&ibdev->ib_dev.dev,
2922 				       mlx4_class_attributes[j]))
2923 			goto err_notif;
2924 	}
2925 
2926 	ibdev->ib_active = true;
2927 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2928 		devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2929 					 &ibdev->ib_dev);
2930 
2931 	if (mlx4_is_mfunc(ibdev->dev))
2932 		init_pkeys(ibdev);
2933 
2934 	/* create paravirt contexts for any VFs which are active */
2935 	if (mlx4_is_master(ibdev->dev)) {
2936 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2937 			if (j == mlx4_master_func_num(ibdev->dev))
2938 				continue;
2939 			if (mlx4_is_slave_active(ibdev->dev, j))
2940 				do_slave_init(ibdev, j, 1);
2941 		}
2942 	}
2943 	return ibdev;
2944 
2945 err_notif:
2946 	if (ibdev->iboe.nb.notifier_call) {
2947 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2948 			pr_warn("failure unregistering notifier\n");
2949 		ibdev->iboe.nb.notifier_call = NULL;
2950 	}
2951 	flush_workqueue(wq);
2952 
2953 	mlx4_ib_close_sriov(ibdev);
2954 
2955 err_mad:
2956 	mlx4_ib_mad_cleanup(ibdev);
2957 
2958 err_reg:
2959 	ib_unregister_device(&ibdev->ib_dev);
2960 
2961 err_diag_counters:
2962 	mlx4_ib_diag_cleanup(ibdev);
2963 
2964 err_steer_free_bitmap:
2965 	kfree(ibdev->ib_uc_qpns_bitmap);
2966 
2967 err_steer_qp_release:
2968 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2969 			      ibdev->steer_qpn_count);
2970 err_counter:
2971 	for (i = 0; i < ibdev->num_ports; ++i)
2972 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2973 
2974 err_map:
2975 	mlx4_ib_free_eqs(dev, ibdev);
2976 	iounmap(ibdev->uar_map);
2977 
2978 err_uar:
2979 	mlx4_uar_free(dev, &ibdev->priv_uar);
2980 
2981 err_pd:
2982 	mlx4_pd_free(dev, ibdev->priv_pdn);
2983 
2984 err_dealloc:
2985 	ib_dealloc_device(&ibdev->ib_dev);
2986 
2987 	return NULL;
2988 }
2989 
2990 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2991 {
2992 	int offset;
2993 
2994 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2995 
2996 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2997 					 dev->steer_qpn_count,
2998 					 get_count_order(count));
2999 	if (offset < 0)
3000 		return offset;
3001 
3002 	*qpn = dev->steer_qpn_base + offset;
3003 	return 0;
3004 }
3005 
3006 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3007 {
3008 	if (!qpn ||
3009 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3010 		return;
3011 
3012 	if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
3013 		 qpn, dev->steer_qpn_base))
3014 		/* not supposed to be here */
3015 		return;
3016 
3017 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
3018 			      qpn - dev->steer_qpn_base,
3019 			      get_count_order(count));
3020 }
3021 
3022 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3023 			 int is_attach)
3024 {
3025 	int err;
3026 	size_t flow_size;
3027 	struct ib_flow_attr *flow = NULL;
3028 	struct ib_flow_spec_ib *ib_spec;
3029 
3030 	if (is_attach) {
3031 		flow_size = sizeof(struct ib_flow_attr) +
3032 			    sizeof(struct ib_flow_spec_ib);
3033 		flow = kzalloc(flow_size, GFP_KERNEL);
3034 		if (!flow)
3035 			return -ENOMEM;
3036 		flow->port = mqp->port;
3037 		flow->num_of_specs = 1;
3038 		flow->size = flow_size;
3039 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3040 		ib_spec->type = IB_FLOW_SPEC_IB;
3041 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
3042 		/* Add an empty rule for IB L2 */
3043 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3044 
3045 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3046 					    IB_FLOW_DOMAIN_NIC,
3047 					    MLX4_FS_REGULAR,
3048 					    &mqp->reg_id);
3049 	} else {
3050 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3051 	}
3052 	kfree(flow);
3053 	return err;
3054 }
3055 
3056 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3057 {
3058 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
3059 	int p;
3060 	int i;
3061 
3062 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3063 		devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3064 	ibdev->ib_active = false;
3065 	flush_workqueue(wq);
3066 
3067 	mlx4_ib_close_sriov(ibdev);
3068 	mlx4_ib_mad_cleanup(ibdev);
3069 	ib_unregister_device(&ibdev->ib_dev);
3070 	mlx4_ib_diag_cleanup(ibdev);
3071 	if (ibdev->iboe.nb.notifier_call) {
3072 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3073 			pr_warn("failure unregistering notifier\n");
3074 		ibdev->iboe.nb.notifier_call = NULL;
3075 	}
3076 
3077 	mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3078 			      ibdev->steer_qpn_count);
3079 	kfree(ibdev->ib_uc_qpns_bitmap);
3080 
3081 	iounmap(ibdev->uar_map);
3082 	for (p = 0; p < ibdev->num_ports; ++p)
3083 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3084 
3085 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3086 		mlx4_CLOSE_PORT(dev, p);
3087 
3088 	mlx4_ib_free_eqs(dev, ibdev);
3089 
3090 	mlx4_uar_free(dev, &ibdev->priv_uar);
3091 	mlx4_pd_free(dev, ibdev->priv_pdn);
3092 	ib_dealloc_device(&ibdev->ib_dev);
3093 }
3094 
3095 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3096 {
3097 	struct mlx4_ib_demux_work **dm = NULL;
3098 	struct mlx4_dev *dev = ibdev->dev;
3099 	int i;
3100 	unsigned long flags;
3101 	struct mlx4_active_ports actv_ports;
3102 	unsigned int ports;
3103 	unsigned int first_port;
3104 
3105 	if (!mlx4_is_master(dev))
3106 		return;
3107 
3108 	actv_ports = mlx4_get_active_ports(dev, slave);
3109 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3110 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3111 
3112 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3113 	if (!dm)
3114 		return;
3115 
3116 	for (i = 0; i < ports; i++) {
3117 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3118 		if (!dm[i]) {
3119 			while (--i >= 0)
3120 				kfree(dm[i]);
3121 			goto out;
3122 		}
3123 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3124 		dm[i]->port = first_port + i + 1;
3125 		dm[i]->slave = slave;
3126 		dm[i]->do_init = do_init;
3127 		dm[i]->dev = ibdev;
3128 	}
3129 	/* initialize or tear down tunnel QPs for the slave */
3130 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3131 	if (!ibdev->sriov.is_going_down) {
3132 		for (i = 0; i < ports; i++)
3133 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3134 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3135 	} else {
3136 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3137 		for (i = 0; i < ports; i++)
3138 			kfree(dm[i]);
3139 	}
3140 out:
3141 	kfree(dm);
3142 	return;
3143 }
3144 
3145 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3146 {
3147 	struct mlx4_ib_qp *mqp;
3148 	unsigned long flags_qp;
3149 	unsigned long flags_cq;
3150 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
3151 	struct list_head    cq_notify_list;
3152 	struct mlx4_cq *mcq;
3153 	unsigned long flags;
3154 
3155 	pr_warn("mlx4_ib_handle_catas_error was started\n");
3156 	INIT_LIST_HEAD(&cq_notify_list);
3157 
3158 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3159 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3160 
3161 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3162 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3163 		if (mqp->sq.tail != mqp->sq.head) {
3164 			send_mcq = to_mcq(mqp->ibqp.send_cq);
3165 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
3166 			if (send_mcq->mcq.comp &&
3167 			    mqp->ibqp.send_cq->comp_handler) {
3168 				if (!send_mcq->mcq.reset_notify_added) {
3169 					send_mcq->mcq.reset_notify_added = 1;
3170 					list_add_tail(&send_mcq->mcq.reset_notify,
3171 						      &cq_notify_list);
3172 				}
3173 			}
3174 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3175 		}
3176 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3177 		/* Now, handle the QP's receive queue */
3178 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3179 		/* no handling is needed for SRQ */
3180 		if (!mqp->ibqp.srq) {
3181 			if (mqp->rq.tail != mqp->rq.head) {
3182 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3183 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3184 				if (recv_mcq->mcq.comp &&
3185 				    mqp->ibqp.recv_cq->comp_handler) {
3186 					if (!recv_mcq->mcq.reset_notify_added) {
3187 						recv_mcq->mcq.reset_notify_added = 1;
3188 						list_add_tail(&recv_mcq->mcq.reset_notify,
3189 							      &cq_notify_list);
3190 					}
3191 				}
3192 				spin_unlock_irqrestore(&recv_mcq->lock,
3193 						       flags_cq);
3194 			}
3195 		}
3196 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3197 	}
3198 
3199 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3200 		mcq->comp(mcq);
3201 	}
3202 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3203 	pr_warn("mlx4_ib_handle_catas_error ended\n");
3204 }
3205 
3206 static void handle_bonded_port_state_event(struct work_struct *work)
3207 {
3208 	struct ib_event_work *ew =
3209 		container_of(work, struct ib_event_work, work);
3210 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
3211 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
3212 	int i;
3213 	struct ib_event ibev;
3214 
3215 	kfree(ew);
3216 	spin_lock_bh(&ibdev->iboe.lock);
3217 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3218 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3219 		enum ib_port_state curr_port_state;
3220 
3221 		if (!curr_netdev)
3222 			continue;
3223 
3224 		curr_port_state =
3225 			(netif_running(curr_netdev) &&
3226 			 netif_carrier_ok(curr_netdev)) ?
3227 			IB_PORT_ACTIVE : IB_PORT_DOWN;
3228 
3229 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3230 			curr_port_state : IB_PORT_ACTIVE;
3231 	}
3232 	spin_unlock_bh(&ibdev->iboe.lock);
3233 
3234 	ibev.device = &ibdev->ib_dev;
3235 	ibev.element.port_num = 1;
3236 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3237 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3238 
3239 	ib_dispatch_event(&ibev);
3240 }
3241 
3242 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3243 {
3244 	u64 sl2vl;
3245 	int err;
3246 
3247 	err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3248 	if (err) {
3249 		pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3250 		       port, err);
3251 		sl2vl = 0;
3252 	}
3253 	atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3254 }
3255 
3256 static void ib_sl2vl_update_work(struct work_struct *work)
3257 {
3258 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3259 	struct mlx4_ib_dev *mdev = ew->ib_dev;
3260 	int port = ew->port;
3261 
3262 	mlx4_ib_sl2vl_update(mdev, port);
3263 
3264 	kfree(ew);
3265 }
3266 
3267 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3268 				     int port)
3269 {
3270 	struct ib_event_work *ew;
3271 
3272 	ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3273 	if (ew) {
3274 		INIT_WORK(&ew->work, ib_sl2vl_update_work);
3275 		ew->port = port;
3276 		ew->ib_dev = ibdev;
3277 		queue_work(wq, &ew->work);
3278 	}
3279 }
3280 
3281 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3282 			  enum mlx4_dev_event event, unsigned long param)
3283 {
3284 	struct ib_event ibev;
3285 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3286 	struct mlx4_eqe *eqe = NULL;
3287 	struct ib_event_work *ew;
3288 	int p = 0;
3289 
3290 	if (mlx4_is_bonded(dev) &&
3291 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
3292 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3293 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3294 		if (!ew)
3295 			return;
3296 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
3297 		ew->ib_dev = ibdev;
3298 		queue_work(wq, &ew->work);
3299 		return;
3300 	}
3301 
3302 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3303 		eqe = (struct mlx4_eqe *)param;
3304 	else
3305 		p = (int) param;
3306 
3307 	switch (event) {
3308 	case MLX4_DEV_EVENT_PORT_UP:
3309 		if (p > ibdev->num_ports)
3310 			return;
3311 		if (!mlx4_is_slave(dev) &&
3312 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3313 			IB_LINK_LAYER_INFINIBAND) {
3314 			if (mlx4_is_master(dev))
3315 				mlx4_ib_invalidate_all_guid_record(ibdev, p);
3316 			if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3317 			    !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3318 				mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3319 		}
3320 		ibev.event = IB_EVENT_PORT_ACTIVE;
3321 		break;
3322 
3323 	case MLX4_DEV_EVENT_PORT_DOWN:
3324 		if (p > ibdev->num_ports)
3325 			return;
3326 		ibev.event = IB_EVENT_PORT_ERR;
3327 		break;
3328 
3329 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3330 		ibdev->ib_active = false;
3331 		ibev.event = IB_EVENT_DEVICE_FATAL;
3332 		mlx4_ib_handle_catas_error(ibdev);
3333 		break;
3334 
3335 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3336 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3337 		if (!ew)
3338 			break;
3339 
3340 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3341 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3342 		ew->ib_dev = ibdev;
3343 		/* need to queue only for port owner, which uses GEN_EQE */
3344 		if (mlx4_is_master(dev))
3345 			queue_work(wq, &ew->work);
3346 		else
3347 			handle_port_mgmt_change_event(&ew->work);
3348 		return;
3349 
3350 	case MLX4_DEV_EVENT_SLAVE_INIT:
3351 		/* here, p is the slave id */
3352 		do_slave_init(ibdev, p, 1);
3353 		if (mlx4_is_master(dev)) {
3354 			int i;
3355 
3356 			for (i = 1; i <= ibdev->num_ports; i++) {
3357 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3358 					== IB_LINK_LAYER_INFINIBAND)
3359 					mlx4_ib_slave_alias_guid_event(ibdev,
3360 								       p, i,
3361 								       1);
3362 			}
3363 		}
3364 		return;
3365 
3366 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3367 		if (mlx4_is_master(dev)) {
3368 			int i;
3369 
3370 			for (i = 1; i <= ibdev->num_ports; i++) {
3371 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3372 					== IB_LINK_LAYER_INFINIBAND)
3373 					mlx4_ib_slave_alias_guid_event(ibdev,
3374 								       p, i,
3375 								       0);
3376 			}
3377 		}
3378 		/* here, p is the slave id */
3379 		do_slave_init(ibdev, p, 0);
3380 		return;
3381 
3382 	default:
3383 		return;
3384 	}
3385 
3386 	ibev.device	      = ibdev_ptr;
3387 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3388 
3389 	ib_dispatch_event(&ibev);
3390 }
3391 
3392 static struct mlx4_interface mlx4_ib_interface = {
3393 	.add		= mlx4_ib_add,
3394 	.remove		= mlx4_ib_remove,
3395 	.event		= mlx4_ib_event,
3396 	.protocol	= MLX4_PROT_IB_IPV6,
3397 	.flags		= MLX4_INTFF_BONDING
3398 };
3399 
3400 static int __init mlx4_ib_init(void)
3401 {
3402 	int err;
3403 
3404 	wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3405 	if (!wq)
3406 		return -ENOMEM;
3407 
3408 	err = mlx4_ib_mcg_init();
3409 	if (err)
3410 		goto clean_wq;
3411 
3412 	err = mlx4_register_interface(&mlx4_ib_interface);
3413 	if (err)
3414 		goto clean_mcg;
3415 
3416 	return 0;
3417 
3418 clean_mcg:
3419 	mlx4_ib_mcg_destroy();
3420 
3421 clean_wq:
3422 	destroy_workqueue(wq);
3423 	return err;
3424 }
3425 
3426 static void __exit mlx4_ib_cleanup(void)
3427 {
3428 	mlx4_unregister_interface(&mlx4_ib_interface);
3429 	mlx4_ib_mcg_destroy();
3430 	destroy_workqueue(wq);
3431 }
3432 
3433 module_init(mlx4_ib_init);
3434 module_exit(mlx4_ib_cleanup);
3435