xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <net/ipv6.h>
43 #include <net/addrconf.h>
44 
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib_cache.h>
49 
50 #include <net/bonding.h>
51 
52 #include <linux/mlx4/driver.h>
53 #include <linux/mlx4/cmd.h>
54 #include <linux/mlx4/qp.h>
55 
56 #include "mlx4_ib.h"
57 #include "user.h"
58 
59 #define DRV_NAME	MLX4_IB_DRV_NAME
60 #define DRV_VERSION	"2.2-1"
61 #define DRV_RELDATE	"Feb 2014"
62 
63 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
64 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
65 #define MLX4_IB_CARD_REV_A0   0xA0
66 
67 MODULE_AUTHOR("Roland Dreier");
68 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
69 MODULE_LICENSE("Dual BSD/GPL");
70 MODULE_VERSION(DRV_VERSION);
71 
72 int mlx4_ib_sm_guid_assign = 0;
73 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
74 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
75 
76 static const char mlx4_ib_version[] =
77 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
78 	DRV_VERSION " (" DRV_RELDATE ")\n";
79 
80 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
81 
82 static struct workqueue_struct *wq;
83 
84 static void init_query_mad(struct ib_smp *mad)
85 {
86 	mad->base_version  = 1;
87 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
88 	mad->class_version = 1;
89 	mad->method	   = IB_MGMT_METHOD_GET;
90 }
91 
92 static int check_flow_steering_support(struct mlx4_dev *dev)
93 {
94 	int eth_num_ports = 0;
95 	int ib_num_ports = 0;
96 
97 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
98 
99 	if (dmfs) {
100 		int i;
101 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
102 			eth_num_ports++;
103 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
104 			ib_num_ports++;
105 		dmfs &= (!ib_num_ports ||
106 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
107 			(!eth_num_ports ||
108 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
109 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
110 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
111 			dmfs = 0;
112 		}
113 	}
114 	return dmfs;
115 }
116 
117 static int num_ib_ports(struct mlx4_dev *dev)
118 {
119 	int ib_ports = 0;
120 	int i;
121 
122 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
123 		ib_ports++;
124 
125 	return ib_ports;
126 }
127 
128 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
129 {
130 	struct mlx4_ib_dev *ibdev = to_mdev(device);
131 	struct net_device *dev;
132 
133 	rcu_read_lock();
134 	dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
135 
136 	if (dev) {
137 		if (mlx4_is_bonded(ibdev->dev)) {
138 			struct net_device *upper = NULL;
139 
140 			upper = netdev_master_upper_dev_get_rcu(dev);
141 			if (upper) {
142 				struct net_device *active;
143 
144 				active = bond_option_active_slave_get_rcu(netdev_priv(upper));
145 				if (active)
146 					dev = active;
147 			}
148 		}
149 	}
150 	if (dev)
151 		dev_hold(dev);
152 
153 	rcu_read_unlock();
154 	return dev;
155 }
156 
157 static int mlx4_ib_update_gids(struct gid_entry *gids,
158 			       struct mlx4_ib_dev *ibdev,
159 			       u8 port_num)
160 {
161 	struct mlx4_cmd_mailbox *mailbox;
162 	int err;
163 	struct mlx4_dev *dev = ibdev->dev;
164 	int i;
165 	union ib_gid *gid_tbl;
166 
167 	mailbox = mlx4_alloc_cmd_mailbox(dev);
168 	if (IS_ERR(mailbox))
169 		return -ENOMEM;
170 
171 	gid_tbl = mailbox->buf;
172 
173 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
174 		memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
175 
176 	err = mlx4_cmd(dev, mailbox->dma,
177 		       MLX4_SET_PORT_GID_TABLE << 8 | port_num,
178 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
179 		       MLX4_CMD_WRAPPED);
180 	if (mlx4_is_bonded(dev))
181 		err += mlx4_cmd(dev, mailbox->dma,
182 				MLX4_SET_PORT_GID_TABLE << 8 | 2,
183 				1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
184 				MLX4_CMD_WRAPPED);
185 
186 	mlx4_free_cmd_mailbox(dev, mailbox);
187 	return err;
188 }
189 
190 static int mlx4_ib_add_gid(struct ib_device *device,
191 			   u8 port_num,
192 			   unsigned int index,
193 			   const union ib_gid *gid,
194 			   const struct ib_gid_attr *attr,
195 			   void **context)
196 {
197 	struct mlx4_ib_dev *ibdev = to_mdev(device);
198 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
199 	struct mlx4_port_gid_table   *port_gid_table;
200 	int free = -1, found = -1;
201 	int ret = 0;
202 	int hw_update = 0;
203 	int i;
204 	struct gid_entry *gids = NULL;
205 
206 	if (!rdma_cap_roce_gid_table(device, port_num))
207 		return -EINVAL;
208 
209 	if (port_num > MLX4_MAX_PORTS)
210 		return -EINVAL;
211 
212 	if (!context)
213 		return -EINVAL;
214 
215 	port_gid_table = &iboe->gids[port_num - 1];
216 	spin_lock_bh(&iboe->lock);
217 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
218 		if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid))) {
219 			found = i;
220 			break;
221 		}
222 		if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
223 			free = i; /* HW has space */
224 	}
225 
226 	if (found < 0) {
227 		if (free < 0) {
228 			ret = -ENOSPC;
229 		} else {
230 			port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
231 			if (!port_gid_table->gids[free].ctx) {
232 				ret = -ENOMEM;
233 			} else {
234 				*context = port_gid_table->gids[free].ctx;
235 				memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
236 				port_gid_table->gids[free].ctx->real_index = free;
237 				port_gid_table->gids[free].ctx->refcount = 1;
238 				hw_update = 1;
239 			}
240 		}
241 	} else {
242 		struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
243 		*context = ctx;
244 		ctx->refcount++;
245 	}
246 	if (!ret && hw_update) {
247 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
248 		if (!gids) {
249 			ret = -ENOMEM;
250 		} else {
251 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
252 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
253 		}
254 	}
255 	spin_unlock_bh(&iboe->lock);
256 
257 	if (!ret && hw_update) {
258 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
259 		kfree(gids);
260 	}
261 
262 	return ret;
263 }
264 
265 static int mlx4_ib_del_gid(struct ib_device *device,
266 			   u8 port_num,
267 			   unsigned int index,
268 			   void **context)
269 {
270 	struct gid_cache_context *ctx = *context;
271 	struct mlx4_ib_dev *ibdev = to_mdev(device);
272 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
273 	struct mlx4_port_gid_table   *port_gid_table;
274 	int ret = 0;
275 	int hw_update = 0;
276 	struct gid_entry *gids = NULL;
277 
278 	if (!rdma_cap_roce_gid_table(device, port_num))
279 		return -EINVAL;
280 
281 	if (port_num > MLX4_MAX_PORTS)
282 		return -EINVAL;
283 
284 	port_gid_table = &iboe->gids[port_num - 1];
285 	spin_lock_bh(&iboe->lock);
286 	if (ctx) {
287 		ctx->refcount--;
288 		if (!ctx->refcount) {
289 			unsigned int real_index = ctx->real_index;
290 
291 			memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
292 			kfree(port_gid_table->gids[real_index].ctx);
293 			port_gid_table->gids[real_index].ctx = NULL;
294 			hw_update = 1;
295 		}
296 	}
297 	if (!ret && hw_update) {
298 		int i;
299 
300 		gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
301 		if (!gids) {
302 			ret = -ENOMEM;
303 		} else {
304 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
305 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
306 		}
307 	}
308 	spin_unlock_bh(&iboe->lock);
309 
310 	if (!ret && hw_update) {
311 		ret = mlx4_ib_update_gids(gids, ibdev, port_num);
312 		kfree(gids);
313 	}
314 	return ret;
315 }
316 
317 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
318 				    u8 port_num, int index)
319 {
320 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
321 	struct gid_cache_context *ctx = NULL;
322 	union ib_gid gid;
323 	struct mlx4_port_gid_table   *port_gid_table;
324 	int real_index = -EINVAL;
325 	int i;
326 	int ret;
327 	unsigned long flags;
328 
329 	if (port_num > MLX4_MAX_PORTS)
330 		return -EINVAL;
331 
332 	if (mlx4_is_bonded(ibdev->dev))
333 		port_num = 1;
334 
335 	if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
336 		return index;
337 
338 	ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid);
339 	if (ret)
340 		return ret;
341 
342 	if (!memcmp(&gid, &zgid, sizeof(gid)))
343 		return -EINVAL;
344 
345 	spin_lock_irqsave(&iboe->lock, flags);
346 	port_gid_table = &iboe->gids[port_num - 1];
347 
348 	for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
349 		if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid))) {
350 			ctx = port_gid_table->gids[i].ctx;
351 			break;
352 		}
353 	if (ctx)
354 		real_index = ctx->real_index;
355 	spin_unlock_irqrestore(&iboe->lock, flags);
356 	return real_index;
357 }
358 
359 static int mlx4_ib_query_device(struct ib_device *ibdev,
360 				struct ib_device_attr *props,
361 				struct ib_udata *uhw)
362 {
363 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
364 	struct ib_smp *in_mad  = NULL;
365 	struct ib_smp *out_mad = NULL;
366 	int err = -ENOMEM;
367 	int have_ib_ports;
368 	struct mlx4_uverbs_ex_query_device cmd;
369 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
370 	struct mlx4_clock_params clock_params;
371 
372 	if (uhw->inlen) {
373 		if (uhw->inlen < sizeof(cmd))
374 			return -EINVAL;
375 
376 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
377 		if (err)
378 			return err;
379 
380 		if (cmd.comp_mask)
381 			return -EINVAL;
382 
383 		if (cmd.reserved)
384 			return -EINVAL;
385 	}
386 
387 	resp.response_length = offsetof(typeof(resp), response_length) +
388 		sizeof(resp.response_length);
389 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
390 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
391 	if (!in_mad || !out_mad)
392 		goto out;
393 
394 	init_query_mad(in_mad);
395 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
396 
397 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
398 			   1, NULL, NULL, in_mad, out_mad);
399 	if (err)
400 		goto out;
401 
402 	memset(props, 0, sizeof *props);
403 
404 	have_ib_ports = num_ib_ports(dev->dev);
405 
406 	props->fw_ver = dev->dev->caps.fw_ver;
407 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
408 		IB_DEVICE_PORT_ACTIVE_EVENT		|
409 		IB_DEVICE_SYS_IMAGE_GUID		|
410 		IB_DEVICE_RC_RNR_NAK_GEN		|
411 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
412 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
413 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
414 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
415 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
416 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
417 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
418 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
419 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
420 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
421 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
422 	if (dev->dev->caps.max_gso_sz &&
423 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
424 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
425 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
426 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
427 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
428 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
429 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
430 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
431 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
432 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
433 		props->device_cap_flags |= IB_DEVICE_XRC;
434 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
435 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
436 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
437 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
438 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
439 		else
440 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
441 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
442 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
443 	}
444 
445 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
446 		0xffffff;
447 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
448 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
449 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
450 
451 	props->max_mr_size	   = ~0ull;
452 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
453 	props->max_qp		   = dev->dev->quotas.qp;
454 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
455 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
456 					 dev->dev->caps.max_rq_sg);
457 	props->max_sge_rd = props->max_sge;
458 	props->max_cq		   = dev->dev->quotas.cq;
459 	props->max_cqe		   = dev->dev->caps.max_cqes;
460 	props->max_mr		   = dev->dev->quotas.mpt;
461 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
462 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
463 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
464 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
465 	props->max_srq		   = dev->dev->quotas.srq;
466 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
467 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
468 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
469 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
470 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
471 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
472 	props->masked_atomic_cap   = props->atomic_cap;
473 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
474 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
475 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
476 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
477 					   props->max_mcast_grp;
478 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
479 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
480 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
481 
482 	if (!mlx4_is_slave(dev->dev))
483 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
484 
485 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
486 		resp.response_length += sizeof(resp.hca_core_clock_offset);
487 		if (!err && !mlx4_is_slave(dev->dev)) {
488 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
489 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
490 		}
491 	}
492 
493 	if (uhw->outlen) {
494 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
495 		if (err)
496 			goto out;
497 	}
498 out:
499 	kfree(in_mad);
500 	kfree(out_mad);
501 
502 	return err;
503 }
504 
505 static enum rdma_link_layer
506 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
507 {
508 	struct mlx4_dev *dev = to_mdev(device)->dev;
509 
510 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
511 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
512 }
513 
514 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
515 			      struct ib_port_attr *props, int netw_view)
516 {
517 	struct ib_smp *in_mad  = NULL;
518 	struct ib_smp *out_mad = NULL;
519 	int ext_active_speed;
520 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
521 	int err = -ENOMEM;
522 
523 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
524 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
525 	if (!in_mad || !out_mad)
526 		goto out;
527 
528 	init_query_mad(in_mad);
529 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
530 	in_mad->attr_mod = cpu_to_be32(port);
531 
532 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
533 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
534 
535 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
536 				in_mad, out_mad);
537 	if (err)
538 		goto out;
539 
540 
541 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
542 	props->lmc		= out_mad->data[34] & 0x7;
543 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
544 	props->sm_sl		= out_mad->data[36] & 0xf;
545 	props->state		= out_mad->data[32] & 0xf;
546 	props->phys_state	= out_mad->data[33] >> 4;
547 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
548 	if (netw_view)
549 		props->gid_tbl_len = out_mad->data[50];
550 	else
551 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
552 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
553 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
554 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
555 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
556 	props->active_width	= out_mad->data[31] & 0xf;
557 	props->active_speed	= out_mad->data[35] >> 4;
558 	props->max_mtu		= out_mad->data[41] & 0xf;
559 	props->active_mtu	= out_mad->data[36] >> 4;
560 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
561 	props->max_vl_num	= out_mad->data[37] >> 4;
562 	props->init_type_reply	= out_mad->data[41] >> 4;
563 
564 	/* Check if extended speeds (EDR/FDR/...) are supported */
565 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
566 		ext_active_speed = out_mad->data[62] >> 4;
567 
568 		switch (ext_active_speed) {
569 		case 1:
570 			props->active_speed = IB_SPEED_FDR;
571 			break;
572 		case 2:
573 			props->active_speed = IB_SPEED_EDR;
574 			break;
575 		}
576 	}
577 
578 	/* If reported active speed is QDR, check if is FDR-10 */
579 	if (props->active_speed == IB_SPEED_QDR) {
580 		init_query_mad(in_mad);
581 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
582 		in_mad->attr_mod = cpu_to_be32(port);
583 
584 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
585 				   NULL, NULL, in_mad, out_mad);
586 		if (err)
587 			goto out;
588 
589 		/* Checking LinkSpeedActive for FDR-10 */
590 		if (out_mad->data[15] & 0x1)
591 			props->active_speed = IB_SPEED_FDR10;
592 	}
593 
594 	/* Avoid wrong speed value returned by FW if the IB link is down. */
595 	if (props->state == IB_PORT_DOWN)
596 		 props->active_speed = IB_SPEED_SDR;
597 
598 out:
599 	kfree(in_mad);
600 	kfree(out_mad);
601 	return err;
602 }
603 
604 static u8 state_to_phys_state(enum ib_port_state state)
605 {
606 	return state == IB_PORT_ACTIVE ? 5 : 3;
607 }
608 
609 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
610 			       struct ib_port_attr *props, int netw_view)
611 {
612 
613 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
614 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
615 	struct net_device *ndev;
616 	enum ib_mtu tmp;
617 	struct mlx4_cmd_mailbox *mailbox;
618 	int err = 0;
619 	int is_bonded = mlx4_is_bonded(mdev->dev);
620 
621 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
622 	if (IS_ERR(mailbox))
623 		return PTR_ERR(mailbox);
624 
625 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
626 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
627 			   MLX4_CMD_WRAPPED);
628 	if (err)
629 		goto out;
630 
631 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
632 						IB_WIDTH_4X : IB_WIDTH_1X;
633 	props->active_speed	= IB_SPEED_QDR;
634 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
635 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
636 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
637 	props->pkey_tbl_len	= 1;
638 	props->max_mtu		= IB_MTU_4096;
639 	props->max_vl_num	= 2;
640 	props->state		= IB_PORT_DOWN;
641 	props->phys_state	= state_to_phys_state(props->state);
642 	props->active_mtu	= IB_MTU_256;
643 	spin_lock_bh(&iboe->lock);
644 	ndev = iboe->netdevs[port - 1];
645 	if (ndev && is_bonded) {
646 		rcu_read_lock(); /* required to get upper dev */
647 		ndev = netdev_master_upper_dev_get_rcu(ndev);
648 		rcu_read_unlock();
649 	}
650 	if (!ndev)
651 		goto out_unlock;
652 
653 	tmp = iboe_get_mtu(ndev->mtu);
654 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
655 
656 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
657 					IB_PORT_ACTIVE : IB_PORT_DOWN;
658 	props->phys_state	= state_to_phys_state(props->state);
659 out_unlock:
660 	spin_unlock_bh(&iboe->lock);
661 out:
662 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
663 	return err;
664 }
665 
666 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
667 			 struct ib_port_attr *props, int netw_view)
668 {
669 	int err;
670 
671 	memset(props, 0, sizeof *props);
672 
673 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
674 		ib_link_query_port(ibdev, port, props, netw_view) :
675 				eth_link_query_port(ibdev, port, props, netw_view);
676 
677 	return err;
678 }
679 
680 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
681 			      struct ib_port_attr *props)
682 {
683 	/* returns host view */
684 	return __mlx4_ib_query_port(ibdev, port, props, 0);
685 }
686 
687 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
688 			union ib_gid *gid, int netw_view)
689 {
690 	struct ib_smp *in_mad  = NULL;
691 	struct ib_smp *out_mad = NULL;
692 	int err = -ENOMEM;
693 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
694 	int clear = 0;
695 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
696 
697 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
698 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
699 	if (!in_mad || !out_mad)
700 		goto out;
701 
702 	init_query_mad(in_mad);
703 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
704 	in_mad->attr_mod = cpu_to_be32(port);
705 
706 	if (mlx4_is_mfunc(dev->dev) && netw_view)
707 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
708 
709 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
710 	if (err)
711 		goto out;
712 
713 	memcpy(gid->raw, out_mad->data + 8, 8);
714 
715 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
716 		if (index) {
717 			/* For any index > 0, return the null guid */
718 			err = 0;
719 			clear = 1;
720 			goto out;
721 		}
722 	}
723 
724 	init_query_mad(in_mad);
725 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
726 	in_mad->attr_mod = cpu_to_be32(index / 8);
727 
728 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
729 			   NULL, NULL, in_mad, out_mad);
730 	if (err)
731 		goto out;
732 
733 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
734 
735 out:
736 	if (clear)
737 		memset(gid->raw + 8, 0, 8);
738 	kfree(in_mad);
739 	kfree(out_mad);
740 	return err;
741 }
742 
743 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
744 			     union ib_gid *gid)
745 {
746 	int ret;
747 
748 	if (rdma_protocol_ib(ibdev, port))
749 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
750 
751 	if (!rdma_protocol_roce(ibdev, port))
752 		return -ENODEV;
753 
754 	if (!rdma_cap_roce_gid_table(ibdev, port))
755 		return -ENODEV;
756 
757 	ret = ib_get_cached_gid(ibdev, port, index, gid);
758 	if (ret == -EAGAIN) {
759 		memcpy(gid, &zgid, sizeof(*gid));
760 		return 0;
761 	}
762 
763 	return ret;
764 }
765 
766 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
767 			 u16 *pkey, int netw_view)
768 {
769 	struct ib_smp *in_mad  = NULL;
770 	struct ib_smp *out_mad = NULL;
771 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
772 	int err = -ENOMEM;
773 
774 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
775 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
776 	if (!in_mad || !out_mad)
777 		goto out;
778 
779 	init_query_mad(in_mad);
780 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
781 	in_mad->attr_mod = cpu_to_be32(index / 32);
782 
783 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
784 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
785 
786 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
787 			   in_mad, out_mad);
788 	if (err)
789 		goto out;
790 
791 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
792 
793 out:
794 	kfree(in_mad);
795 	kfree(out_mad);
796 	return err;
797 }
798 
799 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
800 {
801 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
802 }
803 
804 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
805 				 struct ib_device_modify *props)
806 {
807 	struct mlx4_cmd_mailbox *mailbox;
808 	unsigned long flags;
809 
810 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
811 		return -EOPNOTSUPP;
812 
813 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
814 		return 0;
815 
816 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
817 		return -EOPNOTSUPP;
818 
819 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
820 	memcpy(ibdev->node_desc, props->node_desc, 64);
821 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
822 
823 	/*
824 	 * If possible, pass node desc to FW, so it can generate
825 	 * a 144 trap.  If cmd fails, just ignore.
826 	 */
827 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
828 	if (IS_ERR(mailbox))
829 		return 0;
830 
831 	memcpy(mailbox->buf, props->node_desc, 64);
832 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
833 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
834 
835 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
836 
837 	return 0;
838 }
839 
840 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
841 			    u32 cap_mask)
842 {
843 	struct mlx4_cmd_mailbox *mailbox;
844 	int err;
845 
846 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
847 	if (IS_ERR(mailbox))
848 		return PTR_ERR(mailbox);
849 
850 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
851 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
852 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
853 	} else {
854 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
855 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
856 	}
857 
858 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
859 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
860 		       MLX4_CMD_WRAPPED);
861 
862 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
863 	return err;
864 }
865 
866 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
867 			       struct ib_port_modify *props)
868 {
869 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
870 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
871 	struct ib_port_attr attr;
872 	u32 cap_mask;
873 	int err;
874 
875 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
876 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
877 	 * violations and port capabilities are not meaningful.
878 	 */
879 	if (is_eth)
880 		return 0;
881 
882 	mutex_lock(&mdev->cap_mask_mutex);
883 
884 	err = mlx4_ib_query_port(ibdev, port, &attr);
885 	if (err)
886 		goto out;
887 
888 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
889 		~props->clr_port_cap_mask;
890 
891 	err = mlx4_ib_SET_PORT(mdev, port,
892 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
893 			       cap_mask);
894 
895 out:
896 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
897 	return err;
898 }
899 
900 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
901 						  struct ib_udata *udata)
902 {
903 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
904 	struct mlx4_ib_ucontext *context;
905 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
906 	struct mlx4_ib_alloc_ucontext_resp resp;
907 	int err;
908 
909 	if (!dev->ib_active)
910 		return ERR_PTR(-EAGAIN);
911 
912 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
913 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
914 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
915 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
916 	} else {
917 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
918 		resp.qp_tab_size      = dev->dev->caps.num_qps;
919 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
920 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
921 		resp.cqe_size	      = dev->dev->caps.cqe_size;
922 	}
923 
924 	context = kzalloc(sizeof(*context), GFP_KERNEL);
925 	if (!context)
926 		return ERR_PTR(-ENOMEM);
927 
928 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
929 	if (err) {
930 		kfree(context);
931 		return ERR_PTR(err);
932 	}
933 
934 	INIT_LIST_HEAD(&context->db_page_list);
935 	mutex_init(&context->db_page_mutex);
936 
937 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
938 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
939 	else
940 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
941 
942 	if (err) {
943 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
944 		kfree(context);
945 		return ERR_PTR(-EFAULT);
946 	}
947 
948 	return &context->ibucontext;
949 }
950 
951 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
952 {
953 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
954 
955 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
956 	kfree(context);
957 
958 	return 0;
959 }
960 
961 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
962 {
963 	/* vma_open is called when a new VMA is created on top of our VMA.
964 	 * This is done through either mremap flow or split_vma (usually due
965 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
966 	 * vma, as this VMA is strongly hardware related. Therefore we set the
967 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
968 	 * calling us again and trying to do incorrect actions. We assume that
969 	 * the original vma size is exactly a single page that there will be no
970 	 * "splitting" operations on.
971 	 */
972 	area->vm_ops = NULL;
973 }
974 
975 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
976 {
977 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
978 
979 	/* It's guaranteed that all VMAs opened on a FD are closed before the
980 	 * file itself is closed, therefore no sync is needed with the regular
981 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
982 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
983 	 * The close operation is usually called under mm->mmap_sem except when
984 	 * process is exiting.  The exiting case is handled explicitly as part
985 	 * of mlx4_ib_disassociate_ucontext.
986 	 */
987 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
988 				area->vm_private_data;
989 
990 	/* set the vma context pointer to null in the mlx4_ib driver's private
991 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
992 	 */
993 	mlx4_ib_vma_priv_data->vma = NULL;
994 }
995 
996 static const struct vm_operations_struct mlx4_ib_vm_ops = {
997 	.open = mlx4_ib_vma_open,
998 	.close = mlx4_ib_vma_close
999 };
1000 
1001 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1002 {
1003 	int i;
1004 	int ret = 0;
1005 	struct vm_area_struct *vma;
1006 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1007 	struct task_struct *owning_process  = NULL;
1008 	struct mm_struct   *owning_mm       = NULL;
1009 
1010 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1011 	if (!owning_process)
1012 		return;
1013 
1014 	owning_mm = get_task_mm(owning_process);
1015 	if (!owning_mm) {
1016 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1017 		while (1) {
1018 			/* make sure that task is dead before returning, it may
1019 			 * prevent a rare case of module down in parallel to a
1020 			 * call to mlx4_ib_vma_close.
1021 			 */
1022 			put_task_struct(owning_process);
1023 			msleep(1);
1024 			owning_process = get_pid_task(ibcontext->tgid,
1025 						      PIDTYPE_PID);
1026 			if (!owning_process ||
1027 			    owning_process->state == TASK_DEAD) {
1028 				pr_info("disassociate ucontext done, task was terminated\n");
1029 				/* in case task was dead need to release the task struct */
1030 				if (owning_process)
1031 					put_task_struct(owning_process);
1032 				return;
1033 			}
1034 		}
1035 	}
1036 
1037 	/* need to protect from a race on closing the vma as part of
1038 	 * mlx4_ib_vma_close().
1039 	 */
1040 	down_read(&owning_mm->mmap_sem);
1041 	for (i = 0; i < HW_BAR_COUNT; i++) {
1042 		vma = context->hw_bar_info[i].vma;
1043 		if (!vma)
1044 			continue;
1045 
1046 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1047 				   context->hw_bar_info[i].vma->vm_start,
1048 				   PAGE_SIZE);
1049 		if (ret) {
1050 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1051 			BUG_ON(1);
1052 		}
1053 
1054 		/* context going to be destroyed, should not access ops any more */
1055 		context->hw_bar_info[i].vma->vm_ops = NULL;
1056 	}
1057 
1058 	up_read(&owning_mm->mmap_sem);
1059 	mmput(owning_mm);
1060 	put_task_struct(owning_process);
1061 }
1062 
1063 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1064 				 struct mlx4_ib_vma_private_data *vma_private_data)
1065 {
1066 	vma_private_data->vma = vma;
1067 	vma->vm_private_data = vma_private_data;
1068 	vma->vm_ops =  &mlx4_ib_vm_ops;
1069 }
1070 
1071 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1072 {
1073 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1074 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1075 
1076 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1077 		return -EINVAL;
1078 
1079 	if (vma->vm_pgoff == 0) {
1080 		/* We prevent double mmaping on same context */
1081 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1082 			return -EINVAL;
1083 
1084 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1085 
1086 		if (io_remap_pfn_range(vma, vma->vm_start,
1087 				       to_mucontext(context)->uar.pfn,
1088 				       PAGE_SIZE, vma->vm_page_prot))
1089 			return -EAGAIN;
1090 
1091 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1092 
1093 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1094 		/* We prevent double mmaping on same context */
1095 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1096 			return -EINVAL;
1097 
1098 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1099 
1100 		if (io_remap_pfn_range(vma, vma->vm_start,
1101 				       to_mucontext(context)->uar.pfn +
1102 				       dev->dev->caps.num_uars,
1103 				       PAGE_SIZE, vma->vm_page_prot))
1104 			return -EAGAIN;
1105 
1106 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1107 
1108 	} else if (vma->vm_pgoff == 3) {
1109 		struct mlx4_clock_params params;
1110 		int ret;
1111 
1112 		/* We prevent double mmaping on same context */
1113 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1114 			return -EINVAL;
1115 
1116 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1117 
1118 		if (ret)
1119 			return ret;
1120 
1121 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1122 		if (io_remap_pfn_range(vma, vma->vm_start,
1123 				       (pci_resource_start(dev->dev->persist->pdev,
1124 							   params.bar) +
1125 					params.offset)
1126 				       >> PAGE_SHIFT,
1127 				       PAGE_SIZE, vma->vm_page_prot))
1128 			return -EAGAIN;
1129 
1130 		mlx4_ib_set_vma_data(vma,
1131 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1132 	} else {
1133 		return -EINVAL;
1134 	}
1135 
1136 	return 0;
1137 }
1138 
1139 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1140 				      struct ib_ucontext *context,
1141 				      struct ib_udata *udata)
1142 {
1143 	struct mlx4_ib_pd *pd;
1144 	int err;
1145 
1146 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1147 	if (!pd)
1148 		return ERR_PTR(-ENOMEM);
1149 
1150 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1151 	if (err) {
1152 		kfree(pd);
1153 		return ERR_PTR(err);
1154 	}
1155 
1156 	if (context)
1157 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1158 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1159 			kfree(pd);
1160 			return ERR_PTR(-EFAULT);
1161 		}
1162 
1163 	return &pd->ibpd;
1164 }
1165 
1166 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1167 {
1168 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1169 	kfree(pd);
1170 
1171 	return 0;
1172 }
1173 
1174 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1175 					  struct ib_ucontext *context,
1176 					  struct ib_udata *udata)
1177 {
1178 	struct mlx4_ib_xrcd *xrcd;
1179 	struct ib_cq_init_attr cq_attr = {};
1180 	int err;
1181 
1182 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1183 		return ERR_PTR(-ENOSYS);
1184 
1185 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1186 	if (!xrcd)
1187 		return ERR_PTR(-ENOMEM);
1188 
1189 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1190 	if (err)
1191 		goto err1;
1192 
1193 	xrcd->pd = ib_alloc_pd(ibdev);
1194 	if (IS_ERR(xrcd->pd)) {
1195 		err = PTR_ERR(xrcd->pd);
1196 		goto err2;
1197 	}
1198 
1199 	cq_attr.cqe = 1;
1200 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1201 	if (IS_ERR(xrcd->cq)) {
1202 		err = PTR_ERR(xrcd->cq);
1203 		goto err3;
1204 	}
1205 
1206 	return &xrcd->ibxrcd;
1207 
1208 err3:
1209 	ib_dealloc_pd(xrcd->pd);
1210 err2:
1211 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1212 err1:
1213 	kfree(xrcd);
1214 	return ERR_PTR(err);
1215 }
1216 
1217 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1218 {
1219 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1220 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1221 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1222 	kfree(xrcd);
1223 
1224 	return 0;
1225 }
1226 
1227 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1228 {
1229 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1230 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1231 	struct mlx4_ib_gid_entry *ge;
1232 
1233 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1234 	if (!ge)
1235 		return -ENOMEM;
1236 
1237 	ge->gid = *gid;
1238 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1239 		ge->port = mqp->port;
1240 		ge->added = 1;
1241 	}
1242 
1243 	mutex_lock(&mqp->mutex);
1244 	list_add_tail(&ge->list, &mqp->gid_list);
1245 	mutex_unlock(&mqp->mutex);
1246 
1247 	return 0;
1248 }
1249 
1250 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1251 		   union ib_gid *gid)
1252 {
1253 	struct net_device *ndev;
1254 	int ret = 0;
1255 
1256 	if (!mqp->port)
1257 		return 0;
1258 
1259 	spin_lock_bh(&mdev->iboe.lock);
1260 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1261 	if (ndev)
1262 		dev_hold(ndev);
1263 	spin_unlock_bh(&mdev->iboe.lock);
1264 
1265 	if (ndev) {
1266 		ret = 1;
1267 		dev_put(ndev);
1268 	}
1269 
1270 	return ret;
1271 }
1272 
1273 struct mlx4_ib_steering {
1274 	struct list_head list;
1275 	struct mlx4_flow_reg_id reg_id;
1276 	union ib_gid gid;
1277 };
1278 
1279 static int parse_flow_attr(struct mlx4_dev *dev,
1280 			   u32 qp_num,
1281 			   union ib_flow_spec *ib_spec,
1282 			   struct _rule_hw *mlx4_spec)
1283 {
1284 	enum mlx4_net_trans_rule_id type;
1285 
1286 	switch (ib_spec->type) {
1287 	case IB_FLOW_SPEC_ETH:
1288 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1289 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1290 		       ETH_ALEN);
1291 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1292 		       ETH_ALEN);
1293 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1294 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1295 		break;
1296 	case IB_FLOW_SPEC_IB:
1297 		type = MLX4_NET_TRANS_RULE_ID_IB;
1298 		mlx4_spec->ib.l3_qpn =
1299 			cpu_to_be32(qp_num);
1300 		mlx4_spec->ib.qpn_mask =
1301 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1302 		break;
1303 
1304 
1305 	case IB_FLOW_SPEC_IPV4:
1306 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1307 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1308 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1309 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1310 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1311 		break;
1312 
1313 	case IB_FLOW_SPEC_TCP:
1314 	case IB_FLOW_SPEC_UDP:
1315 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1316 					MLX4_NET_TRANS_RULE_ID_TCP :
1317 					MLX4_NET_TRANS_RULE_ID_UDP;
1318 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1319 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1320 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1321 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1322 		break;
1323 
1324 	default:
1325 		return -EINVAL;
1326 	}
1327 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1328 	    mlx4_hw_rule_sz(dev, type) < 0)
1329 		return -EINVAL;
1330 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1331 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1332 	return mlx4_hw_rule_sz(dev, type);
1333 }
1334 
1335 struct default_rules {
1336 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1337 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1338 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1339 	__u8  link_layer;
1340 };
1341 static const struct default_rules default_table[] = {
1342 	{
1343 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1344 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1345 		.rules_create_list = {IB_FLOW_SPEC_IB},
1346 		.link_layer = IB_LINK_LAYER_INFINIBAND
1347 	}
1348 };
1349 
1350 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1351 					 struct ib_flow_attr *flow_attr)
1352 {
1353 	int i, j, k;
1354 	void *ib_flow;
1355 	const struct default_rules *pdefault_rules = default_table;
1356 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1357 
1358 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1359 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1360 		memset(&field_types, 0, sizeof(field_types));
1361 
1362 		if (link_layer != pdefault_rules->link_layer)
1363 			continue;
1364 
1365 		ib_flow = flow_attr + 1;
1366 		/* we assume the specs are sorted */
1367 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1368 		     j < flow_attr->num_of_specs; k++) {
1369 			union ib_flow_spec *current_flow =
1370 				(union ib_flow_spec *)ib_flow;
1371 
1372 			/* same layer but different type */
1373 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1374 			     (pdefault_rules->mandatory_fields[k] &
1375 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1376 			    (current_flow->type !=
1377 			     pdefault_rules->mandatory_fields[k]))
1378 				goto out;
1379 
1380 			/* same layer, try match next one */
1381 			if (current_flow->type ==
1382 			    pdefault_rules->mandatory_fields[k]) {
1383 				j++;
1384 				ib_flow +=
1385 					((union ib_flow_spec *)ib_flow)->size;
1386 			}
1387 		}
1388 
1389 		ib_flow = flow_attr + 1;
1390 		for (j = 0; j < flow_attr->num_of_specs;
1391 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1392 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1393 				/* same layer and same type */
1394 				if (((union ib_flow_spec *)ib_flow)->type ==
1395 				    pdefault_rules->mandatory_not_fields[k])
1396 					goto out;
1397 
1398 		return i;
1399 	}
1400 out:
1401 	return -1;
1402 }
1403 
1404 static int __mlx4_ib_create_default_rules(
1405 		struct mlx4_ib_dev *mdev,
1406 		struct ib_qp *qp,
1407 		const struct default_rules *pdefault_rules,
1408 		struct _rule_hw *mlx4_spec) {
1409 	int size = 0;
1410 	int i;
1411 
1412 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1413 		int ret;
1414 		union ib_flow_spec ib_spec;
1415 		switch (pdefault_rules->rules_create_list[i]) {
1416 		case 0:
1417 			/* no rule */
1418 			continue;
1419 		case IB_FLOW_SPEC_IB:
1420 			ib_spec.type = IB_FLOW_SPEC_IB;
1421 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1422 
1423 			break;
1424 		default:
1425 			/* invalid rule */
1426 			return -EINVAL;
1427 		}
1428 		/* We must put empty rule, qpn is being ignored */
1429 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1430 				      mlx4_spec);
1431 		if (ret < 0) {
1432 			pr_info("invalid parsing\n");
1433 			return -EINVAL;
1434 		}
1435 
1436 		mlx4_spec = (void *)mlx4_spec + ret;
1437 		size += ret;
1438 	}
1439 	return size;
1440 }
1441 
1442 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1443 			  int domain,
1444 			  enum mlx4_net_trans_promisc_mode flow_type,
1445 			  u64 *reg_id)
1446 {
1447 	int ret, i;
1448 	int size = 0;
1449 	void *ib_flow;
1450 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1451 	struct mlx4_cmd_mailbox *mailbox;
1452 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1453 	int default_flow;
1454 
1455 	static const u16 __mlx4_domain[] = {
1456 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1457 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1458 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1459 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1460 	};
1461 
1462 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1463 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1464 		return -EINVAL;
1465 	}
1466 
1467 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1468 		pr_err("Invalid domain value %d\n", domain);
1469 		return -EINVAL;
1470 	}
1471 
1472 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1473 		return -EINVAL;
1474 
1475 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1476 	if (IS_ERR(mailbox))
1477 		return PTR_ERR(mailbox);
1478 	ctrl = mailbox->buf;
1479 
1480 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1481 				 flow_attr->priority);
1482 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1483 	ctrl->port = flow_attr->port;
1484 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1485 
1486 	ib_flow = flow_attr + 1;
1487 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1488 	/* Add default flows */
1489 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1490 	if (default_flow >= 0) {
1491 		ret = __mlx4_ib_create_default_rules(
1492 				mdev, qp, default_table + default_flow,
1493 				mailbox->buf + size);
1494 		if (ret < 0) {
1495 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1496 			return -EINVAL;
1497 		}
1498 		size += ret;
1499 	}
1500 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1501 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1502 				      mailbox->buf + size);
1503 		if (ret < 0) {
1504 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1505 			return -EINVAL;
1506 		}
1507 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1508 		size += ret;
1509 	}
1510 
1511 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1512 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1513 			   MLX4_CMD_WRAPPED);
1514 	if (ret == -ENOMEM)
1515 		pr_err("mcg table is full. Fail to register network rule.\n");
1516 	else if (ret == -ENXIO)
1517 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1518 	else if (ret)
1519 		pr_err("Invalid argumant. Fail to register network rule.\n");
1520 
1521 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1522 	return ret;
1523 }
1524 
1525 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1526 {
1527 	int err;
1528 	err = mlx4_cmd(dev, reg_id, 0, 0,
1529 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1530 		       MLX4_CMD_WRAPPED);
1531 	if (err)
1532 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1533 		       reg_id);
1534 	return err;
1535 }
1536 
1537 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1538 				    u64 *reg_id)
1539 {
1540 	void *ib_flow;
1541 	union ib_flow_spec *ib_spec;
1542 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1543 	int err = 0;
1544 
1545 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1546 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1547 		return 0; /* do nothing */
1548 
1549 	ib_flow = flow_attr + 1;
1550 	ib_spec = (union ib_flow_spec *)ib_flow;
1551 
1552 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1553 		return 0; /* do nothing */
1554 
1555 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1556 				    flow_attr->port, qp->qp_num,
1557 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1558 				    reg_id);
1559 	return err;
1560 }
1561 
1562 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1563 				    struct ib_flow_attr *flow_attr,
1564 				    int domain)
1565 {
1566 	int err = 0, i = 0, j = 0;
1567 	struct mlx4_ib_flow *mflow;
1568 	enum mlx4_net_trans_promisc_mode type[2];
1569 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1570 	int is_bonded = mlx4_is_bonded(dev);
1571 
1572 	memset(type, 0, sizeof(type));
1573 
1574 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1575 	if (!mflow) {
1576 		err = -ENOMEM;
1577 		goto err_free;
1578 	}
1579 
1580 	switch (flow_attr->type) {
1581 	case IB_FLOW_ATTR_NORMAL:
1582 		type[0] = MLX4_FS_REGULAR;
1583 		break;
1584 
1585 	case IB_FLOW_ATTR_ALL_DEFAULT:
1586 		type[0] = MLX4_FS_ALL_DEFAULT;
1587 		break;
1588 
1589 	case IB_FLOW_ATTR_MC_DEFAULT:
1590 		type[0] = MLX4_FS_MC_DEFAULT;
1591 		break;
1592 
1593 	case IB_FLOW_ATTR_SNIFFER:
1594 		type[0] = MLX4_FS_UC_SNIFFER;
1595 		type[1] = MLX4_FS_MC_SNIFFER;
1596 		break;
1597 
1598 	default:
1599 		err = -EINVAL;
1600 		goto err_free;
1601 	}
1602 
1603 	while (i < ARRAY_SIZE(type) && type[i]) {
1604 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1605 					    &mflow->reg_id[i].id);
1606 		if (err)
1607 			goto err_create_flow;
1608 		if (is_bonded) {
1609 			/* Application always sees one port so the mirror rule
1610 			 * must be on port #2
1611 			 */
1612 			flow_attr->port = 2;
1613 			err = __mlx4_ib_create_flow(qp, flow_attr,
1614 						    domain, type[j],
1615 						    &mflow->reg_id[j].mirror);
1616 			flow_attr->port = 1;
1617 			if (err)
1618 				goto err_create_flow;
1619 			j++;
1620 		}
1621 
1622 		i++;
1623 	}
1624 
1625 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1626 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1627 					       &mflow->reg_id[i].id);
1628 		if (err)
1629 			goto err_create_flow;
1630 
1631 		if (is_bonded) {
1632 			flow_attr->port = 2;
1633 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1634 						       &mflow->reg_id[j].mirror);
1635 			flow_attr->port = 1;
1636 			if (err)
1637 				goto err_create_flow;
1638 			j++;
1639 		}
1640 		/* function to create mirror rule */
1641 		i++;
1642 	}
1643 
1644 	return &mflow->ibflow;
1645 
1646 err_create_flow:
1647 	while (i) {
1648 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1649 					     mflow->reg_id[i].id);
1650 		i--;
1651 	}
1652 
1653 	while (j) {
1654 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1655 					     mflow->reg_id[j].mirror);
1656 		j--;
1657 	}
1658 err_free:
1659 	kfree(mflow);
1660 	return ERR_PTR(err);
1661 }
1662 
1663 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1664 {
1665 	int err, ret = 0;
1666 	int i = 0;
1667 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1668 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1669 
1670 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1671 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1672 		if (err)
1673 			ret = err;
1674 		if (mflow->reg_id[i].mirror) {
1675 			err = __mlx4_ib_destroy_flow(mdev->dev,
1676 						     mflow->reg_id[i].mirror);
1677 			if (err)
1678 				ret = err;
1679 		}
1680 		i++;
1681 	}
1682 
1683 	kfree(mflow);
1684 	return ret;
1685 }
1686 
1687 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1688 {
1689 	int err;
1690 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1691 	struct mlx4_dev	*dev = mdev->dev;
1692 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1693 	struct mlx4_ib_steering *ib_steering = NULL;
1694 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1695 	struct mlx4_flow_reg_id	reg_id;
1696 
1697 	if (mdev->dev->caps.steering_mode ==
1698 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1699 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1700 		if (!ib_steering)
1701 			return -ENOMEM;
1702 	}
1703 
1704 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1705 				    !!(mqp->flags &
1706 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1707 				    prot, &reg_id.id);
1708 	if (err) {
1709 		pr_err("multicast attach op failed, err %d\n", err);
1710 		goto err_malloc;
1711 	}
1712 
1713 	reg_id.mirror = 0;
1714 	if (mlx4_is_bonded(dev)) {
1715 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1716 					    (mqp->port == 1) ? 2 : 1,
1717 					    !!(mqp->flags &
1718 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1719 					    prot, &reg_id.mirror);
1720 		if (err)
1721 			goto err_add;
1722 	}
1723 
1724 	err = add_gid_entry(ibqp, gid);
1725 	if (err)
1726 		goto err_add;
1727 
1728 	if (ib_steering) {
1729 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1730 		ib_steering->reg_id = reg_id;
1731 		mutex_lock(&mqp->mutex);
1732 		list_add(&ib_steering->list, &mqp->steering_rules);
1733 		mutex_unlock(&mqp->mutex);
1734 	}
1735 	return 0;
1736 
1737 err_add:
1738 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1739 			      prot, reg_id.id);
1740 	if (reg_id.mirror)
1741 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1742 				      prot, reg_id.mirror);
1743 err_malloc:
1744 	kfree(ib_steering);
1745 
1746 	return err;
1747 }
1748 
1749 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1750 {
1751 	struct mlx4_ib_gid_entry *ge;
1752 	struct mlx4_ib_gid_entry *tmp;
1753 	struct mlx4_ib_gid_entry *ret = NULL;
1754 
1755 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1756 		if (!memcmp(raw, ge->gid.raw, 16)) {
1757 			ret = ge;
1758 			break;
1759 		}
1760 	}
1761 
1762 	return ret;
1763 }
1764 
1765 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1766 {
1767 	int err;
1768 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1769 	struct mlx4_dev *dev = mdev->dev;
1770 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1771 	struct net_device *ndev;
1772 	struct mlx4_ib_gid_entry *ge;
1773 	struct mlx4_flow_reg_id reg_id = {0, 0};
1774 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1775 
1776 	if (mdev->dev->caps.steering_mode ==
1777 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1778 		struct mlx4_ib_steering *ib_steering;
1779 
1780 		mutex_lock(&mqp->mutex);
1781 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1782 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1783 				list_del(&ib_steering->list);
1784 				break;
1785 			}
1786 		}
1787 		mutex_unlock(&mqp->mutex);
1788 		if (&ib_steering->list == &mqp->steering_rules) {
1789 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1790 			return -EINVAL;
1791 		}
1792 		reg_id = ib_steering->reg_id;
1793 		kfree(ib_steering);
1794 	}
1795 
1796 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1797 				    prot, reg_id.id);
1798 	if (err)
1799 		return err;
1800 
1801 	if (mlx4_is_bonded(dev)) {
1802 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1803 					    prot, reg_id.mirror);
1804 		if (err)
1805 			return err;
1806 	}
1807 
1808 	mutex_lock(&mqp->mutex);
1809 	ge = find_gid_entry(mqp, gid->raw);
1810 	if (ge) {
1811 		spin_lock_bh(&mdev->iboe.lock);
1812 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1813 		if (ndev)
1814 			dev_hold(ndev);
1815 		spin_unlock_bh(&mdev->iboe.lock);
1816 		if (ndev)
1817 			dev_put(ndev);
1818 		list_del(&ge->list);
1819 		kfree(ge);
1820 	} else
1821 		pr_warn("could not find mgid entry\n");
1822 
1823 	mutex_unlock(&mqp->mutex);
1824 
1825 	return 0;
1826 }
1827 
1828 static int init_node_data(struct mlx4_ib_dev *dev)
1829 {
1830 	struct ib_smp *in_mad  = NULL;
1831 	struct ib_smp *out_mad = NULL;
1832 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1833 	int err = -ENOMEM;
1834 
1835 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1836 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1837 	if (!in_mad || !out_mad)
1838 		goto out;
1839 
1840 	init_query_mad(in_mad);
1841 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1842 	if (mlx4_is_master(dev->dev))
1843 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1844 
1845 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1846 	if (err)
1847 		goto out;
1848 
1849 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1850 
1851 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1852 
1853 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1854 	if (err)
1855 		goto out;
1856 
1857 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1858 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1859 
1860 out:
1861 	kfree(in_mad);
1862 	kfree(out_mad);
1863 	return err;
1864 }
1865 
1866 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1867 			char *buf)
1868 {
1869 	struct mlx4_ib_dev *dev =
1870 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1871 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1872 }
1873 
1874 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1875 			   char *buf)
1876 {
1877 	struct mlx4_ib_dev *dev =
1878 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1879 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1880 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1881 		       (int) dev->dev->caps.fw_ver & 0xffff);
1882 }
1883 
1884 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1885 			char *buf)
1886 {
1887 	struct mlx4_ib_dev *dev =
1888 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1889 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1890 }
1891 
1892 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1893 			  char *buf)
1894 {
1895 	struct mlx4_ib_dev *dev =
1896 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1897 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1898 		       dev->dev->board_id);
1899 }
1900 
1901 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1902 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1903 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1904 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1905 
1906 static struct device_attribute *mlx4_class_attributes[] = {
1907 	&dev_attr_hw_rev,
1908 	&dev_attr_fw_ver,
1909 	&dev_attr_hca_type,
1910 	&dev_attr_board_id
1911 };
1912 
1913 #define MLX4_IB_INVALID_MAC	((u64)-1)
1914 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1915 			       struct net_device *dev,
1916 			       int port)
1917 {
1918 	u64 new_smac = 0;
1919 	u64 release_mac = MLX4_IB_INVALID_MAC;
1920 	struct mlx4_ib_qp *qp;
1921 
1922 	read_lock(&dev_base_lock);
1923 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1924 	read_unlock(&dev_base_lock);
1925 
1926 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1927 
1928 	/* no need for update QP1 and mac registration in non-SRIOV */
1929 	if (!mlx4_is_mfunc(ibdev->dev))
1930 		return;
1931 
1932 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1933 	qp = ibdev->qp1_proxy[port - 1];
1934 	if (qp) {
1935 		int new_smac_index;
1936 		u64 old_smac;
1937 		struct mlx4_update_qp_params update_params;
1938 
1939 		mutex_lock(&qp->mutex);
1940 		old_smac = qp->pri.smac;
1941 		if (new_smac == old_smac)
1942 			goto unlock;
1943 
1944 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1945 
1946 		if (new_smac_index < 0)
1947 			goto unlock;
1948 
1949 		update_params.smac_index = new_smac_index;
1950 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1951 				   &update_params)) {
1952 			release_mac = new_smac;
1953 			goto unlock;
1954 		}
1955 		/* if old port was zero, no mac was yet registered for this QP */
1956 		if (qp->pri.smac_port)
1957 			release_mac = old_smac;
1958 		qp->pri.smac = new_smac;
1959 		qp->pri.smac_port = port;
1960 		qp->pri.smac_index = new_smac_index;
1961 	}
1962 
1963 unlock:
1964 	if (release_mac != MLX4_IB_INVALID_MAC)
1965 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1966 	if (qp)
1967 		mutex_unlock(&qp->mutex);
1968 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1969 }
1970 
1971 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1972 				 struct net_device *dev,
1973 				 unsigned long event)
1974 
1975 {
1976 	struct mlx4_ib_iboe *iboe;
1977 	int update_qps_port = -1;
1978 	int port;
1979 
1980 	ASSERT_RTNL();
1981 
1982 	iboe = &ibdev->iboe;
1983 
1984 	spin_lock_bh(&iboe->lock);
1985 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1986 
1987 		iboe->netdevs[port - 1] =
1988 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1989 
1990 		if (dev == iboe->netdevs[port - 1] &&
1991 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1992 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1993 			update_qps_port = port;
1994 
1995 	}
1996 	spin_unlock_bh(&iboe->lock);
1997 
1998 	if (update_qps_port > 0)
1999 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2000 }
2001 
2002 static int mlx4_ib_netdev_event(struct notifier_block *this,
2003 				unsigned long event, void *ptr)
2004 {
2005 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2006 	struct mlx4_ib_dev *ibdev;
2007 
2008 	if (!net_eq(dev_net(dev), &init_net))
2009 		return NOTIFY_DONE;
2010 
2011 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2012 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2013 
2014 	return NOTIFY_DONE;
2015 }
2016 
2017 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2018 {
2019 	int port;
2020 	int slave;
2021 	int i;
2022 
2023 	if (mlx4_is_master(ibdev->dev)) {
2024 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2025 		     ++slave) {
2026 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2027 				for (i = 0;
2028 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2029 				     ++i) {
2030 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2031 					/* master has the identity virt2phys pkey mapping */
2032 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2033 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2034 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2035 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2036 				}
2037 			}
2038 		}
2039 		/* initialize pkey cache */
2040 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2041 			for (i = 0;
2042 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2043 			     ++i)
2044 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2045 					(i) ? 0 : 0xFFFF;
2046 		}
2047 	}
2048 }
2049 
2050 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2051 {
2052 	int i, j, eq = 0, total_eqs = 0;
2053 
2054 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2055 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2056 	if (!ibdev->eq_table)
2057 		return;
2058 
2059 	for (i = 1; i <= dev->caps.num_ports; i++) {
2060 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2061 		     j++, total_eqs++) {
2062 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2063 				continue;
2064 			ibdev->eq_table[eq] = total_eqs;
2065 			if (!mlx4_assign_eq(dev, i,
2066 					    &ibdev->eq_table[eq]))
2067 				eq++;
2068 			else
2069 				ibdev->eq_table[eq] = -1;
2070 		}
2071 	}
2072 
2073 	for (i = eq; i < dev->caps.num_comp_vectors;
2074 	     ibdev->eq_table[i++] = -1)
2075 		;
2076 
2077 	/* Advertise the new number of EQs to clients */
2078 	ibdev->ib_dev.num_comp_vectors = eq;
2079 }
2080 
2081 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2082 {
2083 	int i;
2084 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2085 
2086 	/* no eqs were allocated */
2087 	if (!ibdev->eq_table)
2088 		return;
2089 
2090 	/* Reset the advertised EQ number */
2091 	ibdev->ib_dev.num_comp_vectors = 0;
2092 
2093 	for (i = 0; i < total_eqs; i++)
2094 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2095 
2096 	kfree(ibdev->eq_table);
2097 	ibdev->eq_table = NULL;
2098 }
2099 
2100 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2101 			       struct ib_port_immutable *immutable)
2102 {
2103 	struct ib_port_attr attr;
2104 	int err;
2105 
2106 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2107 	if (err)
2108 		return err;
2109 
2110 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2111 	immutable->gid_tbl_len = attr.gid_tbl_len;
2112 
2113 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2114 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2115 	else
2116 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2117 
2118 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2119 
2120 	return 0;
2121 }
2122 
2123 static void *mlx4_ib_add(struct mlx4_dev *dev)
2124 {
2125 	struct mlx4_ib_dev *ibdev;
2126 	int num_ports = 0;
2127 	int i, j;
2128 	int err;
2129 	struct mlx4_ib_iboe *iboe;
2130 	int ib_num_ports = 0;
2131 	int num_req_counters;
2132 	int allocated;
2133 	u32 counter_index;
2134 
2135 	pr_info_once("%s", mlx4_ib_version);
2136 
2137 	num_ports = 0;
2138 	mlx4_foreach_ib_transport_port(i, dev)
2139 		num_ports++;
2140 
2141 	/* No point in registering a device with no ports... */
2142 	if (num_ports == 0)
2143 		return NULL;
2144 
2145 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2146 	if (!ibdev) {
2147 		dev_err(&dev->persist->pdev->dev,
2148 			"Device struct alloc failed\n");
2149 		return NULL;
2150 	}
2151 
2152 	iboe = &ibdev->iboe;
2153 
2154 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2155 		goto err_dealloc;
2156 
2157 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2158 		goto err_pd;
2159 
2160 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2161 				 PAGE_SIZE);
2162 	if (!ibdev->uar_map)
2163 		goto err_uar;
2164 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2165 
2166 	ibdev->dev = dev;
2167 	ibdev->bond_next_port	= 0;
2168 
2169 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2170 	ibdev->ib_dev.owner		= THIS_MODULE;
2171 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2172 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2173 	ibdev->num_ports		= num_ports;
2174 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2175 						1 : ibdev->num_ports;
2176 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2177 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2178 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2179 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2180 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2181 
2182 	if (dev->caps.userspace_caps)
2183 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2184 	else
2185 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2186 
2187 	ibdev->ib_dev.uverbs_cmd_mask	=
2188 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2189 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2190 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2191 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2192 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2193 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2194 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2195 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2196 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2197 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2198 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2199 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2200 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2201 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2202 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2203 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2204 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2205 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2206 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2207 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2208 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2209 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2210 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2211 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2212 
2213 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2214 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2215 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2216 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2217 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2218 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2219 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2220 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2221 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2222 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2223 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2224 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2225 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2226 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2227 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2228 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2229 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2230 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2231 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2232 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2233 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2234 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2235 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2236 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2237 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2238 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2239 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2240 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2241 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2242 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2243 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2244 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2245 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2246 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2247 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2248 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2249 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2250 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2251 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2252 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2253 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2254 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2255 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2256 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2257 
2258 	if (!mlx4_is_slave(ibdev->dev)) {
2259 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2260 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2261 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2262 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2263 	}
2264 
2265 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2266 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2267 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2268 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2269 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2270 
2271 		ibdev->ib_dev.uverbs_cmd_mask |=
2272 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2273 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2274 	}
2275 
2276 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2277 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2278 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2279 		ibdev->ib_dev.uverbs_cmd_mask |=
2280 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2281 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2282 	}
2283 
2284 	if (check_flow_steering_support(dev)) {
2285 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2286 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2287 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2288 
2289 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2290 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2291 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2292 	}
2293 
2294 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2295 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2296 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ);
2297 
2298 	mlx4_ib_alloc_eqs(dev, ibdev);
2299 
2300 	spin_lock_init(&iboe->lock);
2301 
2302 	if (init_node_data(ibdev))
2303 		goto err_map;
2304 
2305 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2306 	for (i = 0; i < num_req_counters; ++i) {
2307 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2308 		allocated = 0;
2309 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2310 						IB_LINK_LAYER_ETHERNET) {
2311 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2312 			/* if failed to allocate a new counter, use default */
2313 			if (err)
2314 				counter_index =
2315 					mlx4_get_default_counter_index(dev,
2316 								       i + 1);
2317 			else
2318 				allocated = 1;
2319 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2320 			counter_index = mlx4_get_default_counter_index(dev,
2321 								       i + 1);
2322 		}
2323 		ibdev->counters[i].index = counter_index;
2324 		ibdev->counters[i].allocated = allocated;
2325 		pr_info("counter index %d for port %d allocated %d\n",
2326 			counter_index, i + 1, allocated);
2327 	}
2328 	if (mlx4_is_bonded(dev))
2329 		for (i = 1; i < ibdev->num_ports ; ++i) {
2330 			ibdev->counters[i].index = ibdev->counters[0].index;
2331 			ibdev->counters[i].allocated = 0;
2332 		}
2333 
2334 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2335 		ib_num_ports++;
2336 
2337 	spin_lock_init(&ibdev->sm_lock);
2338 	mutex_init(&ibdev->cap_mask_mutex);
2339 	INIT_LIST_HEAD(&ibdev->qp_list);
2340 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2341 
2342 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2343 	    ib_num_ports) {
2344 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2345 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2346 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2347 					    &ibdev->steer_qpn_base, 0);
2348 		if (err)
2349 			goto err_counter;
2350 
2351 		ibdev->ib_uc_qpns_bitmap =
2352 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2353 				sizeof(long),
2354 				GFP_KERNEL);
2355 		if (!ibdev->ib_uc_qpns_bitmap) {
2356 			dev_err(&dev->persist->pdev->dev,
2357 				"bit map alloc failed\n");
2358 			goto err_steer_qp_release;
2359 		}
2360 
2361 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2362 
2363 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2364 				dev, ibdev->steer_qpn_base,
2365 				ibdev->steer_qpn_base +
2366 				ibdev->steer_qpn_count - 1);
2367 		if (err)
2368 			goto err_steer_free_bitmap;
2369 	}
2370 
2371 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2372 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2373 
2374 	if (ib_register_device(&ibdev->ib_dev, NULL))
2375 		goto err_steer_free_bitmap;
2376 
2377 	if (mlx4_ib_mad_init(ibdev))
2378 		goto err_reg;
2379 
2380 	if (mlx4_ib_init_sriov(ibdev))
2381 		goto err_mad;
2382 
2383 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2384 		if (!iboe->nb.notifier_call) {
2385 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2386 			err = register_netdevice_notifier(&iboe->nb);
2387 			if (err) {
2388 				iboe->nb.notifier_call = NULL;
2389 				goto err_notif;
2390 			}
2391 		}
2392 	}
2393 
2394 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2395 		if (device_create_file(&ibdev->ib_dev.dev,
2396 				       mlx4_class_attributes[j]))
2397 			goto err_notif;
2398 	}
2399 
2400 	ibdev->ib_active = true;
2401 
2402 	if (mlx4_is_mfunc(ibdev->dev))
2403 		init_pkeys(ibdev);
2404 
2405 	/* create paravirt contexts for any VFs which are active */
2406 	if (mlx4_is_master(ibdev->dev)) {
2407 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2408 			if (j == mlx4_master_func_num(ibdev->dev))
2409 				continue;
2410 			if (mlx4_is_slave_active(ibdev->dev, j))
2411 				do_slave_init(ibdev, j, 1);
2412 		}
2413 	}
2414 	return ibdev;
2415 
2416 err_notif:
2417 	if (ibdev->iboe.nb.notifier_call) {
2418 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2419 			pr_warn("failure unregistering notifier\n");
2420 		ibdev->iboe.nb.notifier_call = NULL;
2421 	}
2422 	flush_workqueue(wq);
2423 
2424 	mlx4_ib_close_sriov(ibdev);
2425 
2426 err_mad:
2427 	mlx4_ib_mad_cleanup(ibdev);
2428 
2429 err_reg:
2430 	ib_unregister_device(&ibdev->ib_dev);
2431 
2432 err_steer_free_bitmap:
2433 	kfree(ibdev->ib_uc_qpns_bitmap);
2434 
2435 err_steer_qp_release:
2436 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2437 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2438 				      ibdev->steer_qpn_count);
2439 err_counter:
2440 	for (i = 0; i < ibdev->num_ports; ++i) {
2441 		if (ibdev->counters[i].index != -1 &&
2442 		    ibdev->counters[i].allocated)
2443 			mlx4_counter_free(ibdev->dev,
2444 					  ibdev->counters[i].index);
2445 	}
2446 err_map:
2447 	iounmap(ibdev->uar_map);
2448 
2449 err_uar:
2450 	mlx4_uar_free(dev, &ibdev->priv_uar);
2451 
2452 err_pd:
2453 	mlx4_pd_free(dev, ibdev->priv_pdn);
2454 
2455 err_dealloc:
2456 	ib_dealloc_device(&ibdev->ib_dev);
2457 
2458 	return NULL;
2459 }
2460 
2461 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2462 {
2463 	int offset;
2464 
2465 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2466 
2467 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2468 					 dev->steer_qpn_count,
2469 					 get_count_order(count));
2470 	if (offset < 0)
2471 		return offset;
2472 
2473 	*qpn = dev->steer_qpn_base + offset;
2474 	return 0;
2475 }
2476 
2477 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2478 {
2479 	if (!qpn ||
2480 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2481 		return;
2482 
2483 	BUG_ON(qpn < dev->steer_qpn_base);
2484 
2485 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2486 			      qpn - dev->steer_qpn_base,
2487 			      get_count_order(count));
2488 }
2489 
2490 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2491 			 int is_attach)
2492 {
2493 	int err;
2494 	size_t flow_size;
2495 	struct ib_flow_attr *flow = NULL;
2496 	struct ib_flow_spec_ib *ib_spec;
2497 
2498 	if (is_attach) {
2499 		flow_size = sizeof(struct ib_flow_attr) +
2500 			    sizeof(struct ib_flow_spec_ib);
2501 		flow = kzalloc(flow_size, GFP_KERNEL);
2502 		if (!flow)
2503 			return -ENOMEM;
2504 		flow->port = mqp->port;
2505 		flow->num_of_specs = 1;
2506 		flow->size = flow_size;
2507 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2508 		ib_spec->type = IB_FLOW_SPEC_IB;
2509 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2510 		/* Add an empty rule for IB L2 */
2511 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2512 
2513 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2514 					    IB_FLOW_DOMAIN_NIC,
2515 					    MLX4_FS_REGULAR,
2516 					    &mqp->reg_id);
2517 	} else {
2518 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2519 	}
2520 	kfree(flow);
2521 	return err;
2522 }
2523 
2524 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2525 {
2526 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2527 	int p;
2528 
2529 	ibdev->ib_active = false;
2530 	flush_workqueue(wq);
2531 
2532 	mlx4_ib_close_sriov(ibdev);
2533 	mlx4_ib_mad_cleanup(ibdev);
2534 	ib_unregister_device(&ibdev->ib_dev);
2535 	if (ibdev->iboe.nb.notifier_call) {
2536 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2537 			pr_warn("failure unregistering notifier\n");
2538 		ibdev->iboe.nb.notifier_call = NULL;
2539 	}
2540 
2541 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2542 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2543 				      ibdev->steer_qpn_count);
2544 		kfree(ibdev->ib_uc_qpns_bitmap);
2545 	}
2546 
2547 	iounmap(ibdev->uar_map);
2548 	for (p = 0; p < ibdev->num_ports; ++p)
2549 		if (ibdev->counters[p].index != -1 &&
2550 		    ibdev->counters[p].allocated)
2551 			mlx4_counter_free(ibdev->dev, ibdev->counters[p].index);
2552 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2553 		mlx4_CLOSE_PORT(dev, p);
2554 
2555 	mlx4_ib_free_eqs(dev, ibdev);
2556 
2557 	mlx4_uar_free(dev, &ibdev->priv_uar);
2558 	mlx4_pd_free(dev, ibdev->priv_pdn);
2559 	ib_dealloc_device(&ibdev->ib_dev);
2560 }
2561 
2562 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2563 {
2564 	struct mlx4_ib_demux_work **dm = NULL;
2565 	struct mlx4_dev *dev = ibdev->dev;
2566 	int i;
2567 	unsigned long flags;
2568 	struct mlx4_active_ports actv_ports;
2569 	unsigned int ports;
2570 	unsigned int first_port;
2571 
2572 	if (!mlx4_is_master(dev))
2573 		return;
2574 
2575 	actv_ports = mlx4_get_active_ports(dev, slave);
2576 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2577 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2578 
2579 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2580 	if (!dm) {
2581 		pr_err("failed to allocate memory for tunneling qp update\n");
2582 		return;
2583 	}
2584 
2585 	for (i = 0; i < ports; i++) {
2586 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2587 		if (!dm[i]) {
2588 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2589 			while (--i >= 0)
2590 				kfree(dm[i]);
2591 			goto out;
2592 		}
2593 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2594 		dm[i]->port = first_port + i + 1;
2595 		dm[i]->slave = slave;
2596 		dm[i]->do_init = do_init;
2597 		dm[i]->dev = ibdev;
2598 	}
2599 	/* initialize or tear down tunnel QPs for the slave */
2600 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2601 	if (!ibdev->sriov.is_going_down) {
2602 		for (i = 0; i < ports; i++)
2603 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2604 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2605 	} else {
2606 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2607 		for (i = 0; i < ports; i++)
2608 			kfree(dm[i]);
2609 	}
2610 out:
2611 	kfree(dm);
2612 	return;
2613 }
2614 
2615 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2616 {
2617 	struct mlx4_ib_qp *mqp;
2618 	unsigned long flags_qp;
2619 	unsigned long flags_cq;
2620 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2621 	struct list_head    cq_notify_list;
2622 	struct mlx4_cq *mcq;
2623 	unsigned long flags;
2624 
2625 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2626 	INIT_LIST_HEAD(&cq_notify_list);
2627 
2628 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2629 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2630 
2631 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2632 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2633 		if (mqp->sq.tail != mqp->sq.head) {
2634 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2635 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2636 			if (send_mcq->mcq.comp &&
2637 			    mqp->ibqp.send_cq->comp_handler) {
2638 				if (!send_mcq->mcq.reset_notify_added) {
2639 					send_mcq->mcq.reset_notify_added = 1;
2640 					list_add_tail(&send_mcq->mcq.reset_notify,
2641 						      &cq_notify_list);
2642 				}
2643 			}
2644 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2645 		}
2646 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2647 		/* Now, handle the QP's receive queue */
2648 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2649 		/* no handling is needed for SRQ */
2650 		if (!mqp->ibqp.srq) {
2651 			if (mqp->rq.tail != mqp->rq.head) {
2652 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2653 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2654 				if (recv_mcq->mcq.comp &&
2655 				    mqp->ibqp.recv_cq->comp_handler) {
2656 					if (!recv_mcq->mcq.reset_notify_added) {
2657 						recv_mcq->mcq.reset_notify_added = 1;
2658 						list_add_tail(&recv_mcq->mcq.reset_notify,
2659 							      &cq_notify_list);
2660 					}
2661 				}
2662 				spin_unlock_irqrestore(&recv_mcq->lock,
2663 						       flags_cq);
2664 			}
2665 		}
2666 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2667 	}
2668 
2669 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2670 		mcq->comp(mcq);
2671 	}
2672 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2673 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2674 }
2675 
2676 static void handle_bonded_port_state_event(struct work_struct *work)
2677 {
2678 	struct ib_event_work *ew =
2679 		container_of(work, struct ib_event_work, work);
2680 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2681 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2682 	int i;
2683 	struct ib_event ibev;
2684 
2685 	kfree(ew);
2686 	spin_lock_bh(&ibdev->iboe.lock);
2687 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2688 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2689 		enum ib_port_state curr_port_state;
2690 
2691 		if (!curr_netdev)
2692 			continue;
2693 
2694 		curr_port_state =
2695 			(netif_running(curr_netdev) &&
2696 			 netif_carrier_ok(curr_netdev)) ?
2697 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2698 
2699 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2700 			curr_port_state : IB_PORT_ACTIVE;
2701 	}
2702 	spin_unlock_bh(&ibdev->iboe.lock);
2703 
2704 	ibev.device = &ibdev->ib_dev;
2705 	ibev.element.port_num = 1;
2706 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2707 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2708 
2709 	ib_dispatch_event(&ibev);
2710 }
2711 
2712 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2713 			  enum mlx4_dev_event event, unsigned long param)
2714 {
2715 	struct ib_event ibev;
2716 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2717 	struct mlx4_eqe *eqe = NULL;
2718 	struct ib_event_work *ew;
2719 	int p = 0;
2720 
2721 	if (mlx4_is_bonded(dev) &&
2722 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2723 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2724 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2725 		if (!ew)
2726 			return;
2727 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2728 		ew->ib_dev = ibdev;
2729 		queue_work(wq, &ew->work);
2730 		return;
2731 	}
2732 
2733 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2734 		eqe = (struct mlx4_eqe *)param;
2735 	else
2736 		p = (int) param;
2737 
2738 	switch (event) {
2739 	case MLX4_DEV_EVENT_PORT_UP:
2740 		if (p > ibdev->num_ports)
2741 			return;
2742 		if (mlx4_is_master(dev) &&
2743 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2744 			IB_LINK_LAYER_INFINIBAND) {
2745 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2746 		}
2747 		ibev.event = IB_EVENT_PORT_ACTIVE;
2748 		break;
2749 
2750 	case MLX4_DEV_EVENT_PORT_DOWN:
2751 		if (p > ibdev->num_ports)
2752 			return;
2753 		ibev.event = IB_EVENT_PORT_ERR;
2754 		break;
2755 
2756 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2757 		ibdev->ib_active = false;
2758 		ibev.event = IB_EVENT_DEVICE_FATAL;
2759 		mlx4_ib_handle_catas_error(ibdev);
2760 		break;
2761 
2762 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2763 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2764 		if (!ew) {
2765 			pr_err("failed to allocate memory for events work\n");
2766 			break;
2767 		}
2768 
2769 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2770 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2771 		ew->ib_dev = ibdev;
2772 		/* need to queue only for port owner, which uses GEN_EQE */
2773 		if (mlx4_is_master(dev))
2774 			queue_work(wq, &ew->work);
2775 		else
2776 			handle_port_mgmt_change_event(&ew->work);
2777 		return;
2778 
2779 	case MLX4_DEV_EVENT_SLAVE_INIT:
2780 		/* here, p is the slave id */
2781 		do_slave_init(ibdev, p, 1);
2782 		if (mlx4_is_master(dev)) {
2783 			int i;
2784 
2785 			for (i = 1; i <= ibdev->num_ports; i++) {
2786 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2787 					== IB_LINK_LAYER_INFINIBAND)
2788 					mlx4_ib_slave_alias_guid_event(ibdev,
2789 								       p, i,
2790 								       1);
2791 			}
2792 		}
2793 		return;
2794 
2795 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2796 		if (mlx4_is_master(dev)) {
2797 			int i;
2798 
2799 			for (i = 1; i <= ibdev->num_ports; i++) {
2800 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2801 					== IB_LINK_LAYER_INFINIBAND)
2802 					mlx4_ib_slave_alias_guid_event(ibdev,
2803 								       p, i,
2804 								       0);
2805 			}
2806 		}
2807 		/* here, p is the slave id */
2808 		do_slave_init(ibdev, p, 0);
2809 		return;
2810 
2811 	default:
2812 		return;
2813 	}
2814 
2815 	ibev.device	      = ibdev_ptr;
2816 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2817 
2818 	ib_dispatch_event(&ibev);
2819 }
2820 
2821 static struct mlx4_interface mlx4_ib_interface = {
2822 	.add		= mlx4_ib_add,
2823 	.remove		= mlx4_ib_remove,
2824 	.event		= mlx4_ib_event,
2825 	.protocol	= MLX4_PROT_IB_IPV6,
2826 	.flags		= MLX4_INTFF_BONDING
2827 };
2828 
2829 static int __init mlx4_ib_init(void)
2830 {
2831 	int err;
2832 
2833 	wq = create_singlethread_workqueue("mlx4_ib");
2834 	if (!wq)
2835 		return -ENOMEM;
2836 
2837 	err = mlx4_ib_mcg_init();
2838 	if (err)
2839 		goto clean_wq;
2840 
2841 	err = mlx4_register_interface(&mlx4_ib_interface);
2842 	if (err)
2843 		goto clean_mcg;
2844 
2845 	return 0;
2846 
2847 clean_mcg:
2848 	mlx4_ib_mcg_destroy();
2849 
2850 clean_wq:
2851 	destroy_workqueue(wq);
2852 	return err;
2853 }
2854 
2855 static void __exit mlx4_ib_cleanup(void)
2856 {
2857 	mlx4_unregister_interface(&mlx4_ib_interface);
2858 	mlx4_ib_mcg_destroy();
2859 	destroy_workqueue(wq);
2860 }
2861 
2862 module_init(mlx4_ib_init);
2863 module_exit(mlx4_ib_cleanup);
2864