xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 2c684d89)
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, NULL);
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->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
446 
447 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
448 		0xffffff;
449 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
450 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
451 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
452 
453 	props->max_mr_size	   = ~0ull;
454 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
455 	props->max_qp		   = dev->dev->quotas.qp;
456 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
457 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
458 					 dev->dev->caps.max_rq_sg);
459 	props->max_sge_rd	   = MLX4_MAX_SGE_RD;
460 	props->max_cq		   = dev->dev->quotas.cq;
461 	props->max_cqe		   = dev->dev->caps.max_cqes;
462 	props->max_mr		   = dev->dev->quotas.mpt;
463 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
464 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
465 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
466 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
467 	props->max_srq		   = dev->dev->quotas.srq;
468 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
469 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
470 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
471 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
472 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
473 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
474 	props->masked_atomic_cap   = props->atomic_cap;
475 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
476 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
477 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
478 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
479 					   props->max_mcast_grp;
480 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
481 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
482 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
483 
484 	if (!mlx4_is_slave(dev->dev))
485 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
486 
487 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
488 		resp.response_length += sizeof(resp.hca_core_clock_offset);
489 		if (!err && !mlx4_is_slave(dev->dev)) {
490 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
491 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
492 		}
493 	}
494 
495 	if (uhw->outlen) {
496 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
497 		if (err)
498 			goto out;
499 	}
500 out:
501 	kfree(in_mad);
502 	kfree(out_mad);
503 
504 	return err;
505 }
506 
507 static enum rdma_link_layer
508 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
509 {
510 	struct mlx4_dev *dev = to_mdev(device)->dev;
511 
512 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
513 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
514 }
515 
516 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
517 			      struct ib_port_attr *props, int netw_view)
518 {
519 	struct ib_smp *in_mad  = NULL;
520 	struct ib_smp *out_mad = NULL;
521 	int ext_active_speed;
522 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
523 	int err = -ENOMEM;
524 
525 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
526 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
527 	if (!in_mad || !out_mad)
528 		goto out;
529 
530 	init_query_mad(in_mad);
531 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
532 	in_mad->attr_mod = cpu_to_be32(port);
533 
534 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
535 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
536 
537 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
538 				in_mad, out_mad);
539 	if (err)
540 		goto out;
541 
542 
543 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
544 	props->lmc		= out_mad->data[34] & 0x7;
545 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
546 	props->sm_sl		= out_mad->data[36] & 0xf;
547 	props->state		= out_mad->data[32] & 0xf;
548 	props->phys_state	= out_mad->data[33] >> 4;
549 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
550 	if (netw_view)
551 		props->gid_tbl_len = out_mad->data[50];
552 	else
553 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
554 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
555 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
556 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
557 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
558 	props->active_width	= out_mad->data[31] & 0xf;
559 	props->active_speed	= out_mad->data[35] >> 4;
560 	props->max_mtu		= out_mad->data[41] & 0xf;
561 	props->active_mtu	= out_mad->data[36] >> 4;
562 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
563 	props->max_vl_num	= out_mad->data[37] >> 4;
564 	props->init_type_reply	= out_mad->data[41] >> 4;
565 
566 	/* Check if extended speeds (EDR/FDR/...) are supported */
567 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
568 		ext_active_speed = out_mad->data[62] >> 4;
569 
570 		switch (ext_active_speed) {
571 		case 1:
572 			props->active_speed = IB_SPEED_FDR;
573 			break;
574 		case 2:
575 			props->active_speed = IB_SPEED_EDR;
576 			break;
577 		}
578 	}
579 
580 	/* If reported active speed is QDR, check if is FDR-10 */
581 	if (props->active_speed == IB_SPEED_QDR) {
582 		init_query_mad(in_mad);
583 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
584 		in_mad->attr_mod = cpu_to_be32(port);
585 
586 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
587 				   NULL, NULL, in_mad, out_mad);
588 		if (err)
589 			goto out;
590 
591 		/* Checking LinkSpeedActive for FDR-10 */
592 		if (out_mad->data[15] & 0x1)
593 			props->active_speed = IB_SPEED_FDR10;
594 	}
595 
596 	/* Avoid wrong speed value returned by FW if the IB link is down. */
597 	if (props->state == IB_PORT_DOWN)
598 		 props->active_speed = IB_SPEED_SDR;
599 
600 out:
601 	kfree(in_mad);
602 	kfree(out_mad);
603 	return err;
604 }
605 
606 static u8 state_to_phys_state(enum ib_port_state state)
607 {
608 	return state == IB_PORT_ACTIVE ? 5 : 3;
609 }
610 
611 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
612 			       struct ib_port_attr *props, int netw_view)
613 {
614 
615 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
616 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
617 	struct net_device *ndev;
618 	enum ib_mtu tmp;
619 	struct mlx4_cmd_mailbox *mailbox;
620 	int err = 0;
621 	int is_bonded = mlx4_is_bonded(mdev->dev);
622 
623 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
624 	if (IS_ERR(mailbox))
625 		return PTR_ERR(mailbox);
626 
627 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
628 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
629 			   MLX4_CMD_WRAPPED);
630 	if (err)
631 		goto out;
632 
633 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
634 						IB_WIDTH_4X : IB_WIDTH_1X;
635 	props->active_speed	= IB_SPEED_QDR;
636 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
637 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
638 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
639 	props->pkey_tbl_len	= 1;
640 	props->max_mtu		= IB_MTU_4096;
641 	props->max_vl_num	= 2;
642 	props->state		= IB_PORT_DOWN;
643 	props->phys_state	= state_to_phys_state(props->state);
644 	props->active_mtu	= IB_MTU_256;
645 	spin_lock_bh(&iboe->lock);
646 	ndev = iboe->netdevs[port - 1];
647 	if (ndev && is_bonded) {
648 		rcu_read_lock(); /* required to get upper dev */
649 		ndev = netdev_master_upper_dev_get_rcu(ndev);
650 		rcu_read_unlock();
651 	}
652 	if (!ndev)
653 		goto out_unlock;
654 
655 	tmp = iboe_get_mtu(ndev->mtu);
656 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
657 
658 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
659 					IB_PORT_ACTIVE : IB_PORT_DOWN;
660 	props->phys_state	= state_to_phys_state(props->state);
661 out_unlock:
662 	spin_unlock_bh(&iboe->lock);
663 out:
664 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
665 	return err;
666 }
667 
668 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
669 			 struct ib_port_attr *props, int netw_view)
670 {
671 	int err;
672 
673 	memset(props, 0, sizeof *props);
674 
675 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
676 		ib_link_query_port(ibdev, port, props, netw_view) :
677 				eth_link_query_port(ibdev, port, props, netw_view);
678 
679 	return err;
680 }
681 
682 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
683 			      struct ib_port_attr *props)
684 {
685 	/* returns host view */
686 	return __mlx4_ib_query_port(ibdev, port, props, 0);
687 }
688 
689 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
690 			union ib_gid *gid, int netw_view)
691 {
692 	struct ib_smp *in_mad  = NULL;
693 	struct ib_smp *out_mad = NULL;
694 	int err = -ENOMEM;
695 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
696 	int clear = 0;
697 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
698 
699 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
700 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
701 	if (!in_mad || !out_mad)
702 		goto out;
703 
704 	init_query_mad(in_mad);
705 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
706 	in_mad->attr_mod = cpu_to_be32(port);
707 
708 	if (mlx4_is_mfunc(dev->dev) && netw_view)
709 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
710 
711 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
712 	if (err)
713 		goto out;
714 
715 	memcpy(gid->raw, out_mad->data + 8, 8);
716 
717 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
718 		if (index) {
719 			/* For any index > 0, return the null guid */
720 			err = 0;
721 			clear = 1;
722 			goto out;
723 		}
724 	}
725 
726 	init_query_mad(in_mad);
727 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
728 	in_mad->attr_mod = cpu_to_be32(index / 8);
729 
730 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
731 			   NULL, NULL, in_mad, out_mad);
732 	if (err)
733 		goto out;
734 
735 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
736 
737 out:
738 	if (clear)
739 		memset(gid->raw + 8, 0, 8);
740 	kfree(in_mad);
741 	kfree(out_mad);
742 	return err;
743 }
744 
745 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
746 			     union ib_gid *gid)
747 {
748 	int ret;
749 
750 	if (rdma_protocol_ib(ibdev, port))
751 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
752 
753 	if (!rdma_protocol_roce(ibdev, port))
754 		return -ENODEV;
755 
756 	if (!rdma_cap_roce_gid_table(ibdev, port))
757 		return -ENODEV;
758 
759 	ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
760 	if (ret == -EAGAIN) {
761 		memcpy(gid, &zgid, sizeof(*gid));
762 		return 0;
763 	}
764 
765 	return ret;
766 }
767 
768 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
769 			 u16 *pkey, int netw_view)
770 {
771 	struct ib_smp *in_mad  = NULL;
772 	struct ib_smp *out_mad = NULL;
773 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
774 	int err = -ENOMEM;
775 
776 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
777 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
778 	if (!in_mad || !out_mad)
779 		goto out;
780 
781 	init_query_mad(in_mad);
782 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
783 	in_mad->attr_mod = cpu_to_be32(index / 32);
784 
785 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
786 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
787 
788 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
789 			   in_mad, out_mad);
790 	if (err)
791 		goto out;
792 
793 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
794 
795 out:
796 	kfree(in_mad);
797 	kfree(out_mad);
798 	return err;
799 }
800 
801 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
802 {
803 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
804 }
805 
806 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
807 				 struct ib_device_modify *props)
808 {
809 	struct mlx4_cmd_mailbox *mailbox;
810 	unsigned long flags;
811 
812 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
813 		return -EOPNOTSUPP;
814 
815 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
816 		return 0;
817 
818 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
819 		return -EOPNOTSUPP;
820 
821 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
822 	memcpy(ibdev->node_desc, props->node_desc, 64);
823 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
824 
825 	/*
826 	 * If possible, pass node desc to FW, so it can generate
827 	 * a 144 trap.  If cmd fails, just ignore.
828 	 */
829 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
830 	if (IS_ERR(mailbox))
831 		return 0;
832 
833 	memcpy(mailbox->buf, props->node_desc, 64);
834 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
835 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
836 
837 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
838 
839 	return 0;
840 }
841 
842 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
843 			    u32 cap_mask)
844 {
845 	struct mlx4_cmd_mailbox *mailbox;
846 	int err;
847 
848 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
849 	if (IS_ERR(mailbox))
850 		return PTR_ERR(mailbox);
851 
852 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
853 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
854 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
855 	} else {
856 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
857 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
858 	}
859 
860 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
861 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
862 		       MLX4_CMD_WRAPPED);
863 
864 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
865 	return err;
866 }
867 
868 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
869 			       struct ib_port_modify *props)
870 {
871 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
872 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
873 	struct ib_port_attr attr;
874 	u32 cap_mask;
875 	int err;
876 
877 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
878 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
879 	 * violations and port capabilities are not meaningful.
880 	 */
881 	if (is_eth)
882 		return 0;
883 
884 	mutex_lock(&mdev->cap_mask_mutex);
885 
886 	err = mlx4_ib_query_port(ibdev, port, &attr);
887 	if (err)
888 		goto out;
889 
890 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
891 		~props->clr_port_cap_mask;
892 
893 	err = mlx4_ib_SET_PORT(mdev, port,
894 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
895 			       cap_mask);
896 
897 out:
898 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
899 	return err;
900 }
901 
902 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
903 						  struct ib_udata *udata)
904 {
905 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
906 	struct mlx4_ib_ucontext *context;
907 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
908 	struct mlx4_ib_alloc_ucontext_resp resp;
909 	int err;
910 
911 	if (!dev->ib_active)
912 		return ERR_PTR(-EAGAIN);
913 
914 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
915 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
916 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
917 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
918 	} else {
919 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
920 		resp.qp_tab_size      = dev->dev->caps.num_qps;
921 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
922 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
923 		resp.cqe_size	      = dev->dev->caps.cqe_size;
924 	}
925 
926 	context = kzalloc(sizeof(*context), GFP_KERNEL);
927 	if (!context)
928 		return ERR_PTR(-ENOMEM);
929 
930 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
931 	if (err) {
932 		kfree(context);
933 		return ERR_PTR(err);
934 	}
935 
936 	INIT_LIST_HEAD(&context->db_page_list);
937 	mutex_init(&context->db_page_mutex);
938 
939 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
940 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
941 	else
942 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
943 
944 	if (err) {
945 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
946 		kfree(context);
947 		return ERR_PTR(-EFAULT);
948 	}
949 
950 	return &context->ibucontext;
951 }
952 
953 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
954 {
955 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
956 
957 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
958 	kfree(context);
959 
960 	return 0;
961 }
962 
963 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
964 {
965 	/* vma_open is called when a new VMA is created on top of our VMA.
966 	 * This is done through either mremap flow or split_vma (usually due
967 	 * to mlock, madvise, munmap, etc.). We do not support a clone of the
968 	 * vma, as this VMA is strongly hardware related. Therefore we set the
969 	 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
970 	 * calling us again and trying to do incorrect actions. We assume that
971 	 * the original vma size is exactly a single page that there will be no
972 	 * "splitting" operations on.
973 	 */
974 	area->vm_ops = NULL;
975 }
976 
977 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
978 {
979 	struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
980 
981 	/* It's guaranteed that all VMAs opened on a FD are closed before the
982 	 * file itself is closed, therefore no sync is needed with the regular
983 	 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
984 	 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
985 	 * The close operation is usually called under mm->mmap_sem except when
986 	 * process is exiting.  The exiting case is handled explicitly as part
987 	 * of mlx4_ib_disassociate_ucontext.
988 	 */
989 	mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
990 				area->vm_private_data;
991 
992 	/* set the vma context pointer to null in the mlx4_ib driver's private
993 	 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
994 	 */
995 	mlx4_ib_vma_priv_data->vma = NULL;
996 }
997 
998 static const struct vm_operations_struct mlx4_ib_vm_ops = {
999 	.open = mlx4_ib_vma_open,
1000 	.close = mlx4_ib_vma_close
1001 };
1002 
1003 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1004 {
1005 	int i;
1006 	int ret = 0;
1007 	struct vm_area_struct *vma;
1008 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1009 	struct task_struct *owning_process  = NULL;
1010 	struct mm_struct   *owning_mm       = NULL;
1011 
1012 	owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1013 	if (!owning_process)
1014 		return;
1015 
1016 	owning_mm = get_task_mm(owning_process);
1017 	if (!owning_mm) {
1018 		pr_info("no mm, disassociate ucontext is pending task termination\n");
1019 		while (1) {
1020 			/* make sure that task is dead before returning, it may
1021 			 * prevent a rare case of module down in parallel to a
1022 			 * call to mlx4_ib_vma_close.
1023 			 */
1024 			put_task_struct(owning_process);
1025 			msleep(1);
1026 			owning_process = get_pid_task(ibcontext->tgid,
1027 						      PIDTYPE_PID);
1028 			if (!owning_process ||
1029 			    owning_process->state == TASK_DEAD) {
1030 				pr_info("disassociate ucontext done, task was terminated\n");
1031 				/* in case task was dead need to release the task struct */
1032 				if (owning_process)
1033 					put_task_struct(owning_process);
1034 				return;
1035 			}
1036 		}
1037 	}
1038 
1039 	/* need to protect from a race on closing the vma as part of
1040 	 * mlx4_ib_vma_close().
1041 	 */
1042 	down_read(&owning_mm->mmap_sem);
1043 	for (i = 0; i < HW_BAR_COUNT; i++) {
1044 		vma = context->hw_bar_info[i].vma;
1045 		if (!vma)
1046 			continue;
1047 
1048 		ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1049 				   context->hw_bar_info[i].vma->vm_start,
1050 				   PAGE_SIZE);
1051 		if (ret) {
1052 			pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1053 			BUG_ON(1);
1054 		}
1055 
1056 		/* context going to be destroyed, should not access ops any more */
1057 		context->hw_bar_info[i].vma->vm_ops = NULL;
1058 	}
1059 
1060 	up_read(&owning_mm->mmap_sem);
1061 	mmput(owning_mm);
1062 	put_task_struct(owning_process);
1063 }
1064 
1065 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1066 				 struct mlx4_ib_vma_private_data *vma_private_data)
1067 {
1068 	vma_private_data->vma = vma;
1069 	vma->vm_private_data = vma_private_data;
1070 	vma->vm_ops =  &mlx4_ib_vm_ops;
1071 }
1072 
1073 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1074 {
1075 	struct mlx4_ib_dev *dev = to_mdev(context->device);
1076 	struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1077 
1078 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1079 		return -EINVAL;
1080 
1081 	if (vma->vm_pgoff == 0) {
1082 		/* We prevent double mmaping on same context */
1083 		if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1084 			return -EINVAL;
1085 
1086 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1087 
1088 		if (io_remap_pfn_range(vma, vma->vm_start,
1089 				       to_mucontext(context)->uar.pfn,
1090 				       PAGE_SIZE, vma->vm_page_prot))
1091 			return -EAGAIN;
1092 
1093 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1094 
1095 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1096 		/* We prevent double mmaping on same context */
1097 		if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1098 			return -EINVAL;
1099 
1100 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1101 
1102 		if (io_remap_pfn_range(vma, vma->vm_start,
1103 				       to_mucontext(context)->uar.pfn +
1104 				       dev->dev->caps.num_uars,
1105 				       PAGE_SIZE, vma->vm_page_prot))
1106 			return -EAGAIN;
1107 
1108 		mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1109 
1110 	} else if (vma->vm_pgoff == 3) {
1111 		struct mlx4_clock_params params;
1112 		int ret;
1113 
1114 		/* We prevent double mmaping on same context */
1115 		if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1116 			return -EINVAL;
1117 
1118 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
1119 
1120 		if (ret)
1121 			return ret;
1122 
1123 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1124 		if (io_remap_pfn_range(vma, vma->vm_start,
1125 				       (pci_resource_start(dev->dev->persist->pdev,
1126 							   params.bar) +
1127 					params.offset)
1128 				       >> PAGE_SHIFT,
1129 				       PAGE_SIZE, vma->vm_page_prot))
1130 			return -EAGAIN;
1131 
1132 		mlx4_ib_set_vma_data(vma,
1133 				     &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1134 	} else {
1135 		return -EINVAL;
1136 	}
1137 
1138 	return 0;
1139 }
1140 
1141 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1142 				      struct ib_ucontext *context,
1143 				      struct ib_udata *udata)
1144 {
1145 	struct mlx4_ib_pd *pd;
1146 	int err;
1147 
1148 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
1149 	if (!pd)
1150 		return ERR_PTR(-ENOMEM);
1151 
1152 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1153 	if (err) {
1154 		kfree(pd);
1155 		return ERR_PTR(err);
1156 	}
1157 
1158 	if (context)
1159 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1160 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1161 			kfree(pd);
1162 			return ERR_PTR(-EFAULT);
1163 		}
1164 
1165 	return &pd->ibpd;
1166 }
1167 
1168 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1169 {
1170 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1171 	kfree(pd);
1172 
1173 	return 0;
1174 }
1175 
1176 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1177 					  struct ib_ucontext *context,
1178 					  struct ib_udata *udata)
1179 {
1180 	struct mlx4_ib_xrcd *xrcd;
1181 	struct ib_cq_init_attr cq_attr = {};
1182 	int err;
1183 
1184 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1185 		return ERR_PTR(-ENOSYS);
1186 
1187 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1188 	if (!xrcd)
1189 		return ERR_PTR(-ENOMEM);
1190 
1191 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1192 	if (err)
1193 		goto err1;
1194 
1195 	xrcd->pd = ib_alloc_pd(ibdev);
1196 	if (IS_ERR(xrcd->pd)) {
1197 		err = PTR_ERR(xrcd->pd);
1198 		goto err2;
1199 	}
1200 
1201 	cq_attr.cqe = 1;
1202 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1203 	if (IS_ERR(xrcd->cq)) {
1204 		err = PTR_ERR(xrcd->cq);
1205 		goto err3;
1206 	}
1207 
1208 	return &xrcd->ibxrcd;
1209 
1210 err3:
1211 	ib_dealloc_pd(xrcd->pd);
1212 err2:
1213 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1214 err1:
1215 	kfree(xrcd);
1216 	return ERR_PTR(err);
1217 }
1218 
1219 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1220 {
1221 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
1222 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1223 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1224 	kfree(xrcd);
1225 
1226 	return 0;
1227 }
1228 
1229 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1230 {
1231 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1232 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1233 	struct mlx4_ib_gid_entry *ge;
1234 
1235 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
1236 	if (!ge)
1237 		return -ENOMEM;
1238 
1239 	ge->gid = *gid;
1240 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1241 		ge->port = mqp->port;
1242 		ge->added = 1;
1243 	}
1244 
1245 	mutex_lock(&mqp->mutex);
1246 	list_add_tail(&ge->list, &mqp->gid_list);
1247 	mutex_unlock(&mqp->mutex);
1248 
1249 	return 0;
1250 }
1251 
1252 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1253 					  struct mlx4_ib_counters *ctr_table)
1254 {
1255 	struct counter_index *counter, *tmp_count;
1256 
1257 	mutex_lock(&ctr_table->mutex);
1258 	list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1259 				 list) {
1260 		if (counter->allocated)
1261 			mlx4_counter_free(ibdev->dev, counter->index);
1262 		list_del(&counter->list);
1263 		kfree(counter);
1264 	}
1265 	mutex_unlock(&ctr_table->mutex);
1266 }
1267 
1268 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1269 		   union ib_gid *gid)
1270 {
1271 	struct net_device *ndev;
1272 	int ret = 0;
1273 
1274 	if (!mqp->port)
1275 		return 0;
1276 
1277 	spin_lock_bh(&mdev->iboe.lock);
1278 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1279 	if (ndev)
1280 		dev_hold(ndev);
1281 	spin_unlock_bh(&mdev->iboe.lock);
1282 
1283 	if (ndev) {
1284 		ret = 1;
1285 		dev_put(ndev);
1286 	}
1287 
1288 	return ret;
1289 }
1290 
1291 struct mlx4_ib_steering {
1292 	struct list_head list;
1293 	struct mlx4_flow_reg_id reg_id;
1294 	union ib_gid gid;
1295 };
1296 
1297 static int parse_flow_attr(struct mlx4_dev *dev,
1298 			   u32 qp_num,
1299 			   union ib_flow_spec *ib_spec,
1300 			   struct _rule_hw *mlx4_spec)
1301 {
1302 	enum mlx4_net_trans_rule_id type;
1303 
1304 	switch (ib_spec->type) {
1305 	case IB_FLOW_SPEC_ETH:
1306 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1307 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1308 		       ETH_ALEN);
1309 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1310 		       ETH_ALEN);
1311 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1312 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1313 		break;
1314 	case IB_FLOW_SPEC_IB:
1315 		type = MLX4_NET_TRANS_RULE_ID_IB;
1316 		mlx4_spec->ib.l3_qpn =
1317 			cpu_to_be32(qp_num);
1318 		mlx4_spec->ib.qpn_mask =
1319 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1320 		break;
1321 
1322 
1323 	case IB_FLOW_SPEC_IPV4:
1324 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1325 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1326 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1327 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1328 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1329 		break;
1330 
1331 	case IB_FLOW_SPEC_TCP:
1332 	case IB_FLOW_SPEC_UDP:
1333 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1334 					MLX4_NET_TRANS_RULE_ID_TCP :
1335 					MLX4_NET_TRANS_RULE_ID_UDP;
1336 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1337 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1338 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1339 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1340 		break;
1341 
1342 	default:
1343 		return -EINVAL;
1344 	}
1345 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1346 	    mlx4_hw_rule_sz(dev, type) < 0)
1347 		return -EINVAL;
1348 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1349 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1350 	return mlx4_hw_rule_sz(dev, type);
1351 }
1352 
1353 struct default_rules {
1354 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1355 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1356 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1357 	__u8  link_layer;
1358 };
1359 static const struct default_rules default_table[] = {
1360 	{
1361 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
1362 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1363 		.rules_create_list = {IB_FLOW_SPEC_IB},
1364 		.link_layer = IB_LINK_LAYER_INFINIBAND
1365 	}
1366 };
1367 
1368 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1369 					 struct ib_flow_attr *flow_attr)
1370 {
1371 	int i, j, k;
1372 	void *ib_flow;
1373 	const struct default_rules *pdefault_rules = default_table;
1374 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1375 
1376 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1377 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1378 		memset(&field_types, 0, sizeof(field_types));
1379 
1380 		if (link_layer != pdefault_rules->link_layer)
1381 			continue;
1382 
1383 		ib_flow = flow_attr + 1;
1384 		/* we assume the specs are sorted */
1385 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1386 		     j < flow_attr->num_of_specs; k++) {
1387 			union ib_flow_spec *current_flow =
1388 				(union ib_flow_spec *)ib_flow;
1389 
1390 			/* same layer but different type */
1391 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1392 			     (pdefault_rules->mandatory_fields[k] &
1393 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1394 			    (current_flow->type !=
1395 			     pdefault_rules->mandatory_fields[k]))
1396 				goto out;
1397 
1398 			/* same layer, try match next one */
1399 			if (current_flow->type ==
1400 			    pdefault_rules->mandatory_fields[k]) {
1401 				j++;
1402 				ib_flow +=
1403 					((union ib_flow_spec *)ib_flow)->size;
1404 			}
1405 		}
1406 
1407 		ib_flow = flow_attr + 1;
1408 		for (j = 0; j < flow_attr->num_of_specs;
1409 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1410 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1411 				/* same layer and same type */
1412 				if (((union ib_flow_spec *)ib_flow)->type ==
1413 				    pdefault_rules->mandatory_not_fields[k])
1414 					goto out;
1415 
1416 		return i;
1417 	}
1418 out:
1419 	return -1;
1420 }
1421 
1422 static int __mlx4_ib_create_default_rules(
1423 		struct mlx4_ib_dev *mdev,
1424 		struct ib_qp *qp,
1425 		const struct default_rules *pdefault_rules,
1426 		struct _rule_hw *mlx4_spec) {
1427 	int size = 0;
1428 	int i;
1429 
1430 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1431 		int ret;
1432 		union ib_flow_spec ib_spec;
1433 		switch (pdefault_rules->rules_create_list[i]) {
1434 		case 0:
1435 			/* no rule */
1436 			continue;
1437 		case IB_FLOW_SPEC_IB:
1438 			ib_spec.type = IB_FLOW_SPEC_IB;
1439 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1440 
1441 			break;
1442 		default:
1443 			/* invalid rule */
1444 			return -EINVAL;
1445 		}
1446 		/* We must put empty rule, qpn is being ignored */
1447 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1448 				      mlx4_spec);
1449 		if (ret < 0) {
1450 			pr_info("invalid parsing\n");
1451 			return -EINVAL;
1452 		}
1453 
1454 		mlx4_spec = (void *)mlx4_spec + ret;
1455 		size += ret;
1456 	}
1457 	return size;
1458 }
1459 
1460 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1461 			  int domain,
1462 			  enum mlx4_net_trans_promisc_mode flow_type,
1463 			  u64 *reg_id)
1464 {
1465 	int ret, i;
1466 	int size = 0;
1467 	void *ib_flow;
1468 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1469 	struct mlx4_cmd_mailbox *mailbox;
1470 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1471 	int default_flow;
1472 
1473 	static const u16 __mlx4_domain[] = {
1474 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1475 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1476 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1477 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1478 	};
1479 
1480 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1481 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1482 		return -EINVAL;
1483 	}
1484 
1485 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1486 		pr_err("Invalid domain value %d\n", domain);
1487 		return -EINVAL;
1488 	}
1489 
1490 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1491 		return -EINVAL;
1492 
1493 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1494 	if (IS_ERR(mailbox))
1495 		return PTR_ERR(mailbox);
1496 	ctrl = mailbox->buf;
1497 
1498 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1499 				 flow_attr->priority);
1500 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1501 	ctrl->port = flow_attr->port;
1502 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1503 
1504 	ib_flow = flow_attr + 1;
1505 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1506 	/* Add default flows */
1507 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1508 	if (default_flow >= 0) {
1509 		ret = __mlx4_ib_create_default_rules(
1510 				mdev, qp, default_table + default_flow,
1511 				mailbox->buf + size);
1512 		if (ret < 0) {
1513 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1514 			return -EINVAL;
1515 		}
1516 		size += ret;
1517 	}
1518 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1519 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1520 				      mailbox->buf + size);
1521 		if (ret < 0) {
1522 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1523 			return -EINVAL;
1524 		}
1525 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1526 		size += ret;
1527 	}
1528 
1529 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1530 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1531 			   MLX4_CMD_WRAPPED);
1532 	if (ret == -ENOMEM)
1533 		pr_err("mcg table is full. Fail to register network rule.\n");
1534 	else if (ret == -ENXIO)
1535 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1536 	else if (ret)
1537 		pr_err("Invalid argumant. Fail to register network rule.\n");
1538 
1539 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1540 	return ret;
1541 }
1542 
1543 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1544 {
1545 	int err;
1546 	err = mlx4_cmd(dev, reg_id, 0, 0,
1547 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1548 		       MLX4_CMD_WRAPPED);
1549 	if (err)
1550 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1551 		       reg_id);
1552 	return err;
1553 }
1554 
1555 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1556 				    u64 *reg_id)
1557 {
1558 	void *ib_flow;
1559 	union ib_flow_spec *ib_spec;
1560 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1561 	int err = 0;
1562 
1563 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1564 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1565 		return 0; /* do nothing */
1566 
1567 	ib_flow = flow_attr + 1;
1568 	ib_spec = (union ib_flow_spec *)ib_flow;
1569 
1570 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1571 		return 0; /* do nothing */
1572 
1573 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1574 				    flow_attr->port, qp->qp_num,
1575 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1576 				    reg_id);
1577 	return err;
1578 }
1579 
1580 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1581 				    struct ib_flow_attr *flow_attr,
1582 				    int domain)
1583 {
1584 	int err = 0, i = 0, j = 0;
1585 	struct mlx4_ib_flow *mflow;
1586 	enum mlx4_net_trans_promisc_mode type[2];
1587 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1588 	int is_bonded = mlx4_is_bonded(dev);
1589 
1590 	memset(type, 0, sizeof(type));
1591 
1592 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1593 	if (!mflow) {
1594 		err = -ENOMEM;
1595 		goto err_free;
1596 	}
1597 
1598 	switch (flow_attr->type) {
1599 	case IB_FLOW_ATTR_NORMAL:
1600 		type[0] = MLX4_FS_REGULAR;
1601 		break;
1602 
1603 	case IB_FLOW_ATTR_ALL_DEFAULT:
1604 		type[0] = MLX4_FS_ALL_DEFAULT;
1605 		break;
1606 
1607 	case IB_FLOW_ATTR_MC_DEFAULT:
1608 		type[0] = MLX4_FS_MC_DEFAULT;
1609 		break;
1610 
1611 	case IB_FLOW_ATTR_SNIFFER:
1612 		type[0] = MLX4_FS_UC_SNIFFER;
1613 		type[1] = MLX4_FS_MC_SNIFFER;
1614 		break;
1615 
1616 	default:
1617 		err = -EINVAL;
1618 		goto err_free;
1619 	}
1620 
1621 	while (i < ARRAY_SIZE(type) && type[i]) {
1622 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1623 					    &mflow->reg_id[i].id);
1624 		if (err)
1625 			goto err_create_flow;
1626 		if (is_bonded) {
1627 			/* Application always sees one port so the mirror rule
1628 			 * must be on port #2
1629 			 */
1630 			flow_attr->port = 2;
1631 			err = __mlx4_ib_create_flow(qp, flow_attr,
1632 						    domain, type[j],
1633 						    &mflow->reg_id[j].mirror);
1634 			flow_attr->port = 1;
1635 			if (err)
1636 				goto err_create_flow;
1637 			j++;
1638 		}
1639 
1640 		i++;
1641 	}
1642 
1643 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1644 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1645 					       &mflow->reg_id[i].id);
1646 		if (err)
1647 			goto err_create_flow;
1648 
1649 		if (is_bonded) {
1650 			flow_attr->port = 2;
1651 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1652 						       &mflow->reg_id[j].mirror);
1653 			flow_attr->port = 1;
1654 			if (err)
1655 				goto err_create_flow;
1656 			j++;
1657 		}
1658 		/* function to create mirror rule */
1659 		i++;
1660 	}
1661 
1662 	return &mflow->ibflow;
1663 
1664 err_create_flow:
1665 	while (i) {
1666 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1667 					     mflow->reg_id[i].id);
1668 		i--;
1669 	}
1670 
1671 	while (j) {
1672 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1673 					     mflow->reg_id[j].mirror);
1674 		j--;
1675 	}
1676 err_free:
1677 	kfree(mflow);
1678 	return ERR_PTR(err);
1679 }
1680 
1681 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1682 {
1683 	int err, ret = 0;
1684 	int i = 0;
1685 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1686 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1687 
1688 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1689 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1690 		if (err)
1691 			ret = err;
1692 		if (mflow->reg_id[i].mirror) {
1693 			err = __mlx4_ib_destroy_flow(mdev->dev,
1694 						     mflow->reg_id[i].mirror);
1695 			if (err)
1696 				ret = err;
1697 		}
1698 		i++;
1699 	}
1700 
1701 	kfree(mflow);
1702 	return ret;
1703 }
1704 
1705 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1706 {
1707 	int err;
1708 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1709 	struct mlx4_dev	*dev = mdev->dev;
1710 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1711 	struct mlx4_ib_steering *ib_steering = NULL;
1712 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1713 	struct mlx4_flow_reg_id	reg_id;
1714 
1715 	if (mdev->dev->caps.steering_mode ==
1716 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1717 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1718 		if (!ib_steering)
1719 			return -ENOMEM;
1720 	}
1721 
1722 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1723 				    !!(mqp->flags &
1724 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1725 				    prot, &reg_id.id);
1726 	if (err) {
1727 		pr_err("multicast attach op failed, err %d\n", err);
1728 		goto err_malloc;
1729 	}
1730 
1731 	reg_id.mirror = 0;
1732 	if (mlx4_is_bonded(dev)) {
1733 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1734 					    (mqp->port == 1) ? 2 : 1,
1735 					    !!(mqp->flags &
1736 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1737 					    prot, &reg_id.mirror);
1738 		if (err)
1739 			goto err_add;
1740 	}
1741 
1742 	err = add_gid_entry(ibqp, gid);
1743 	if (err)
1744 		goto err_add;
1745 
1746 	if (ib_steering) {
1747 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1748 		ib_steering->reg_id = reg_id;
1749 		mutex_lock(&mqp->mutex);
1750 		list_add(&ib_steering->list, &mqp->steering_rules);
1751 		mutex_unlock(&mqp->mutex);
1752 	}
1753 	return 0;
1754 
1755 err_add:
1756 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1757 			      prot, reg_id.id);
1758 	if (reg_id.mirror)
1759 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1760 				      prot, reg_id.mirror);
1761 err_malloc:
1762 	kfree(ib_steering);
1763 
1764 	return err;
1765 }
1766 
1767 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1768 {
1769 	struct mlx4_ib_gid_entry *ge;
1770 	struct mlx4_ib_gid_entry *tmp;
1771 	struct mlx4_ib_gid_entry *ret = NULL;
1772 
1773 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1774 		if (!memcmp(raw, ge->gid.raw, 16)) {
1775 			ret = ge;
1776 			break;
1777 		}
1778 	}
1779 
1780 	return ret;
1781 }
1782 
1783 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1784 {
1785 	int err;
1786 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1787 	struct mlx4_dev *dev = mdev->dev;
1788 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1789 	struct net_device *ndev;
1790 	struct mlx4_ib_gid_entry *ge;
1791 	struct mlx4_flow_reg_id reg_id = {0, 0};
1792 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1793 
1794 	if (mdev->dev->caps.steering_mode ==
1795 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1796 		struct mlx4_ib_steering *ib_steering;
1797 
1798 		mutex_lock(&mqp->mutex);
1799 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1800 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1801 				list_del(&ib_steering->list);
1802 				break;
1803 			}
1804 		}
1805 		mutex_unlock(&mqp->mutex);
1806 		if (&ib_steering->list == &mqp->steering_rules) {
1807 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1808 			return -EINVAL;
1809 		}
1810 		reg_id = ib_steering->reg_id;
1811 		kfree(ib_steering);
1812 	}
1813 
1814 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1815 				    prot, reg_id.id);
1816 	if (err)
1817 		return err;
1818 
1819 	if (mlx4_is_bonded(dev)) {
1820 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1821 					    prot, reg_id.mirror);
1822 		if (err)
1823 			return err;
1824 	}
1825 
1826 	mutex_lock(&mqp->mutex);
1827 	ge = find_gid_entry(mqp, gid->raw);
1828 	if (ge) {
1829 		spin_lock_bh(&mdev->iboe.lock);
1830 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1831 		if (ndev)
1832 			dev_hold(ndev);
1833 		spin_unlock_bh(&mdev->iboe.lock);
1834 		if (ndev)
1835 			dev_put(ndev);
1836 		list_del(&ge->list);
1837 		kfree(ge);
1838 	} else
1839 		pr_warn("could not find mgid entry\n");
1840 
1841 	mutex_unlock(&mqp->mutex);
1842 
1843 	return 0;
1844 }
1845 
1846 static int init_node_data(struct mlx4_ib_dev *dev)
1847 {
1848 	struct ib_smp *in_mad  = NULL;
1849 	struct ib_smp *out_mad = NULL;
1850 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1851 	int err = -ENOMEM;
1852 
1853 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1854 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1855 	if (!in_mad || !out_mad)
1856 		goto out;
1857 
1858 	init_query_mad(in_mad);
1859 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1860 	if (mlx4_is_master(dev->dev))
1861 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1862 
1863 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1864 	if (err)
1865 		goto out;
1866 
1867 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1868 
1869 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1870 
1871 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1872 	if (err)
1873 		goto out;
1874 
1875 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1876 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1877 
1878 out:
1879 	kfree(in_mad);
1880 	kfree(out_mad);
1881 	return err;
1882 }
1883 
1884 static ssize_t show_hca(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, "MT%d\n", dev->dev->persist->pdev->device);
1890 }
1891 
1892 static ssize_t show_fw_ver(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, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1898 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1899 		       (int) dev->dev->caps.fw_ver & 0xffff);
1900 }
1901 
1902 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1903 			char *buf)
1904 {
1905 	struct mlx4_ib_dev *dev =
1906 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1907 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1908 }
1909 
1910 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1911 			  char *buf)
1912 {
1913 	struct mlx4_ib_dev *dev =
1914 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1915 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1916 		       dev->dev->board_id);
1917 }
1918 
1919 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1920 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1921 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1922 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1923 
1924 static struct device_attribute *mlx4_class_attributes[] = {
1925 	&dev_attr_hw_rev,
1926 	&dev_attr_fw_ver,
1927 	&dev_attr_hca_type,
1928 	&dev_attr_board_id
1929 };
1930 
1931 #define MLX4_IB_INVALID_MAC	((u64)-1)
1932 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1933 			       struct net_device *dev,
1934 			       int port)
1935 {
1936 	u64 new_smac = 0;
1937 	u64 release_mac = MLX4_IB_INVALID_MAC;
1938 	struct mlx4_ib_qp *qp;
1939 
1940 	read_lock(&dev_base_lock);
1941 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1942 	read_unlock(&dev_base_lock);
1943 
1944 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1945 
1946 	/* no need for update QP1 and mac registration in non-SRIOV */
1947 	if (!mlx4_is_mfunc(ibdev->dev))
1948 		return;
1949 
1950 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1951 	qp = ibdev->qp1_proxy[port - 1];
1952 	if (qp) {
1953 		int new_smac_index;
1954 		u64 old_smac;
1955 		struct mlx4_update_qp_params update_params;
1956 
1957 		mutex_lock(&qp->mutex);
1958 		old_smac = qp->pri.smac;
1959 		if (new_smac == old_smac)
1960 			goto unlock;
1961 
1962 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1963 
1964 		if (new_smac_index < 0)
1965 			goto unlock;
1966 
1967 		update_params.smac_index = new_smac_index;
1968 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1969 				   &update_params)) {
1970 			release_mac = new_smac;
1971 			goto unlock;
1972 		}
1973 		/* if old port was zero, no mac was yet registered for this QP */
1974 		if (qp->pri.smac_port)
1975 			release_mac = old_smac;
1976 		qp->pri.smac = new_smac;
1977 		qp->pri.smac_port = port;
1978 		qp->pri.smac_index = new_smac_index;
1979 	}
1980 
1981 unlock:
1982 	if (release_mac != MLX4_IB_INVALID_MAC)
1983 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1984 	if (qp)
1985 		mutex_unlock(&qp->mutex);
1986 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1987 }
1988 
1989 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1990 				 struct net_device *dev,
1991 				 unsigned long event)
1992 
1993 {
1994 	struct mlx4_ib_iboe *iboe;
1995 	int update_qps_port = -1;
1996 	int port;
1997 
1998 	ASSERT_RTNL();
1999 
2000 	iboe = &ibdev->iboe;
2001 
2002 	spin_lock_bh(&iboe->lock);
2003 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2004 
2005 		iboe->netdevs[port - 1] =
2006 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2007 
2008 		if (dev == iboe->netdevs[port - 1] &&
2009 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2010 		     event == NETDEV_UP || event == NETDEV_CHANGE))
2011 			update_qps_port = port;
2012 
2013 	}
2014 	spin_unlock_bh(&iboe->lock);
2015 
2016 	if (update_qps_port > 0)
2017 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2018 }
2019 
2020 static int mlx4_ib_netdev_event(struct notifier_block *this,
2021 				unsigned long event, void *ptr)
2022 {
2023 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2024 	struct mlx4_ib_dev *ibdev;
2025 
2026 	if (!net_eq(dev_net(dev), &init_net))
2027 		return NOTIFY_DONE;
2028 
2029 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2030 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2031 
2032 	return NOTIFY_DONE;
2033 }
2034 
2035 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2036 {
2037 	int port;
2038 	int slave;
2039 	int i;
2040 
2041 	if (mlx4_is_master(ibdev->dev)) {
2042 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2043 		     ++slave) {
2044 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2045 				for (i = 0;
2046 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2047 				     ++i) {
2048 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2049 					/* master has the identity virt2phys pkey mapping */
2050 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2051 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2052 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2053 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2054 				}
2055 			}
2056 		}
2057 		/* initialize pkey cache */
2058 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2059 			for (i = 0;
2060 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2061 			     ++i)
2062 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2063 					(i) ? 0 : 0xFFFF;
2064 		}
2065 	}
2066 }
2067 
2068 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2069 {
2070 	int i, j, eq = 0, total_eqs = 0;
2071 
2072 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2073 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2074 	if (!ibdev->eq_table)
2075 		return;
2076 
2077 	for (i = 1; i <= dev->caps.num_ports; i++) {
2078 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2079 		     j++, total_eqs++) {
2080 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2081 				continue;
2082 			ibdev->eq_table[eq] = total_eqs;
2083 			if (!mlx4_assign_eq(dev, i,
2084 					    &ibdev->eq_table[eq]))
2085 				eq++;
2086 			else
2087 				ibdev->eq_table[eq] = -1;
2088 		}
2089 	}
2090 
2091 	for (i = eq; i < dev->caps.num_comp_vectors;
2092 	     ibdev->eq_table[i++] = -1)
2093 		;
2094 
2095 	/* Advertise the new number of EQs to clients */
2096 	ibdev->ib_dev.num_comp_vectors = eq;
2097 }
2098 
2099 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2100 {
2101 	int i;
2102 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2103 
2104 	/* no eqs were allocated */
2105 	if (!ibdev->eq_table)
2106 		return;
2107 
2108 	/* Reset the advertised EQ number */
2109 	ibdev->ib_dev.num_comp_vectors = 0;
2110 
2111 	for (i = 0; i < total_eqs; i++)
2112 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2113 
2114 	kfree(ibdev->eq_table);
2115 	ibdev->eq_table = NULL;
2116 }
2117 
2118 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2119 			       struct ib_port_immutable *immutable)
2120 {
2121 	struct ib_port_attr attr;
2122 	int err;
2123 
2124 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2125 	if (err)
2126 		return err;
2127 
2128 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2129 	immutable->gid_tbl_len = attr.gid_tbl_len;
2130 
2131 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2132 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2133 	else
2134 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2135 
2136 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2137 
2138 	return 0;
2139 }
2140 
2141 static void *mlx4_ib_add(struct mlx4_dev *dev)
2142 {
2143 	struct mlx4_ib_dev *ibdev;
2144 	int num_ports = 0;
2145 	int i, j;
2146 	int err;
2147 	struct mlx4_ib_iboe *iboe;
2148 	int ib_num_ports = 0;
2149 	int num_req_counters;
2150 	int allocated;
2151 	u32 counter_index;
2152 	struct counter_index *new_counter_index = NULL;
2153 
2154 	pr_info_once("%s", mlx4_ib_version);
2155 
2156 	num_ports = 0;
2157 	mlx4_foreach_ib_transport_port(i, dev)
2158 		num_ports++;
2159 
2160 	/* No point in registering a device with no ports... */
2161 	if (num_ports == 0)
2162 		return NULL;
2163 
2164 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2165 	if (!ibdev) {
2166 		dev_err(&dev->persist->pdev->dev,
2167 			"Device struct alloc failed\n");
2168 		return NULL;
2169 	}
2170 
2171 	iboe = &ibdev->iboe;
2172 
2173 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2174 		goto err_dealloc;
2175 
2176 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2177 		goto err_pd;
2178 
2179 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2180 				 PAGE_SIZE);
2181 	if (!ibdev->uar_map)
2182 		goto err_uar;
2183 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2184 
2185 	ibdev->dev = dev;
2186 	ibdev->bond_next_port	= 0;
2187 
2188 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2189 	ibdev->ib_dev.owner		= THIS_MODULE;
2190 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2191 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2192 	ibdev->num_ports		= num_ports;
2193 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2194 						1 : ibdev->num_ports;
2195 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2196 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2197 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2198 	ibdev->ib_dev.add_gid		= mlx4_ib_add_gid;
2199 	ibdev->ib_dev.del_gid		= mlx4_ib_del_gid;
2200 
2201 	if (dev->caps.userspace_caps)
2202 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2203 	else
2204 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2205 
2206 	ibdev->ib_dev.uverbs_cmd_mask	=
2207 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2208 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2209 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2210 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2211 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2212 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2213 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2214 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2215 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2216 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2217 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2218 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2219 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2220 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2221 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2222 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2223 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2224 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2225 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2226 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2227 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2228 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2229 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2230 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2231 
2232 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2233 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2234 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2235 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2236 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2237 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2238 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2239 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2240 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2241 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2242 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2243 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2244 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2245 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2246 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2247 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2248 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2249 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2250 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2251 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2252 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2253 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2254 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2255 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2256 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2257 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2258 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2259 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2260 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2261 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2262 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2263 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2264 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2265 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2266 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2267 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2268 	ibdev->ib_dev.alloc_mr		= mlx4_ib_alloc_mr;
2269 	ibdev->ib_dev.map_mr_sg		= mlx4_ib_map_mr_sg;
2270 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2271 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2272 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2273 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2274 	ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2275 
2276 	if (!mlx4_is_slave(ibdev->dev)) {
2277 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2278 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2279 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2280 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2281 	}
2282 
2283 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2284 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2285 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2286 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2287 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2288 
2289 		ibdev->ib_dev.uverbs_cmd_mask |=
2290 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2291 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2292 	}
2293 
2294 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2295 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2296 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2297 		ibdev->ib_dev.uverbs_cmd_mask |=
2298 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2299 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2300 	}
2301 
2302 	if (check_flow_steering_support(dev)) {
2303 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2304 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2305 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2306 
2307 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2308 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2309 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2310 	}
2311 
2312 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2313 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2314 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2315 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2316 
2317 	mlx4_ib_alloc_eqs(dev, ibdev);
2318 
2319 	spin_lock_init(&iboe->lock);
2320 
2321 	if (init_node_data(ibdev))
2322 		goto err_map;
2323 
2324 	for (i = 0; i < ibdev->num_ports; ++i) {
2325 		mutex_init(&ibdev->counters_table[i].mutex);
2326 		INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2327 	}
2328 
2329 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2330 	for (i = 0; i < num_req_counters; ++i) {
2331 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2332 		allocated = 0;
2333 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2334 						IB_LINK_LAYER_ETHERNET) {
2335 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2336 			/* if failed to allocate a new counter, use default */
2337 			if (err)
2338 				counter_index =
2339 					mlx4_get_default_counter_index(dev,
2340 								       i + 1);
2341 			else
2342 				allocated = 1;
2343 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2344 			counter_index = mlx4_get_default_counter_index(dev,
2345 								       i + 1);
2346 		}
2347 		new_counter_index = kmalloc(sizeof(*new_counter_index),
2348 					    GFP_KERNEL);
2349 		if (!new_counter_index) {
2350 			if (allocated)
2351 				mlx4_counter_free(ibdev->dev, counter_index);
2352 			goto err_counter;
2353 		}
2354 		new_counter_index->index = counter_index;
2355 		new_counter_index->allocated = allocated;
2356 		list_add_tail(&new_counter_index->list,
2357 			      &ibdev->counters_table[i].counters_list);
2358 		ibdev->counters_table[i].default_counter = counter_index;
2359 		pr_info("counter index %d for port %d allocated %d\n",
2360 			counter_index, i + 1, allocated);
2361 	}
2362 	if (mlx4_is_bonded(dev))
2363 		for (i = 1; i < ibdev->num_ports ; ++i) {
2364 			new_counter_index =
2365 					kmalloc(sizeof(struct counter_index),
2366 						GFP_KERNEL);
2367 			if (!new_counter_index)
2368 				goto err_counter;
2369 			new_counter_index->index = counter_index;
2370 			new_counter_index->allocated = 0;
2371 			list_add_tail(&new_counter_index->list,
2372 				      &ibdev->counters_table[i].counters_list);
2373 			ibdev->counters_table[i].default_counter =
2374 								counter_index;
2375 		}
2376 
2377 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2378 		ib_num_ports++;
2379 
2380 	spin_lock_init(&ibdev->sm_lock);
2381 	mutex_init(&ibdev->cap_mask_mutex);
2382 	INIT_LIST_HEAD(&ibdev->qp_list);
2383 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2384 
2385 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2386 	    ib_num_ports) {
2387 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2388 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2389 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2390 					    &ibdev->steer_qpn_base, 0);
2391 		if (err)
2392 			goto err_counter;
2393 
2394 		ibdev->ib_uc_qpns_bitmap =
2395 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2396 				sizeof(long),
2397 				GFP_KERNEL);
2398 		if (!ibdev->ib_uc_qpns_bitmap) {
2399 			dev_err(&dev->persist->pdev->dev,
2400 				"bit map alloc failed\n");
2401 			goto err_steer_qp_release;
2402 		}
2403 
2404 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2405 
2406 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2407 				dev, ibdev->steer_qpn_base,
2408 				ibdev->steer_qpn_base +
2409 				ibdev->steer_qpn_count - 1);
2410 		if (err)
2411 			goto err_steer_free_bitmap;
2412 	}
2413 
2414 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2415 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2416 
2417 	if (ib_register_device(&ibdev->ib_dev, NULL))
2418 		goto err_steer_free_bitmap;
2419 
2420 	if (mlx4_ib_mad_init(ibdev))
2421 		goto err_reg;
2422 
2423 	if (mlx4_ib_init_sriov(ibdev))
2424 		goto err_mad;
2425 
2426 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2427 		if (!iboe->nb.notifier_call) {
2428 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2429 			err = register_netdevice_notifier(&iboe->nb);
2430 			if (err) {
2431 				iboe->nb.notifier_call = NULL;
2432 				goto err_notif;
2433 			}
2434 		}
2435 	}
2436 
2437 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2438 		if (device_create_file(&ibdev->ib_dev.dev,
2439 				       mlx4_class_attributes[j]))
2440 			goto err_notif;
2441 	}
2442 
2443 	ibdev->ib_active = true;
2444 
2445 	if (mlx4_is_mfunc(ibdev->dev))
2446 		init_pkeys(ibdev);
2447 
2448 	/* create paravirt contexts for any VFs which are active */
2449 	if (mlx4_is_master(ibdev->dev)) {
2450 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2451 			if (j == mlx4_master_func_num(ibdev->dev))
2452 				continue;
2453 			if (mlx4_is_slave_active(ibdev->dev, j))
2454 				do_slave_init(ibdev, j, 1);
2455 		}
2456 	}
2457 	return ibdev;
2458 
2459 err_notif:
2460 	if (ibdev->iboe.nb.notifier_call) {
2461 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2462 			pr_warn("failure unregistering notifier\n");
2463 		ibdev->iboe.nb.notifier_call = NULL;
2464 	}
2465 	flush_workqueue(wq);
2466 
2467 	mlx4_ib_close_sriov(ibdev);
2468 
2469 err_mad:
2470 	mlx4_ib_mad_cleanup(ibdev);
2471 
2472 err_reg:
2473 	ib_unregister_device(&ibdev->ib_dev);
2474 
2475 err_steer_free_bitmap:
2476 	kfree(ibdev->ib_uc_qpns_bitmap);
2477 
2478 err_steer_qp_release:
2479 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2480 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2481 				      ibdev->steer_qpn_count);
2482 err_counter:
2483 	for (i = 0; i < ibdev->num_ports; ++i)
2484 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2485 
2486 err_map:
2487 	iounmap(ibdev->uar_map);
2488 
2489 err_uar:
2490 	mlx4_uar_free(dev, &ibdev->priv_uar);
2491 
2492 err_pd:
2493 	mlx4_pd_free(dev, ibdev->priv_pdn);
2494 
2495 err_dealloc:
2496 	ib_dealloc_device(&ibdev->ib_dev);
2497 
2498 	return NULL;
2499 }
2500 
2501 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2502 {
2503 	int offset;
2504 
2505 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2506 
2507 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2508 					 dev->steer_qpn_count,
2509 					 get_count_order(count));
2510 	if (offset < 0)
2511 		return offset;
2512 
2513 	*qpn = dev->steer_qpn_base + offset;
2514 	return 0;
2515 }
2516 
2517 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2518 {
2519 	if (!qpn ||
2520 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2521 		return;
2522 
2523 	BUG_ON(qpn < dev->steer_qpn_base);
2524 
2525 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2526 			      qpn - dev->steer_qpn_base,
2527 			      get_count_order(count));
2528 }
2529 
2530 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2531 			 int is_attach)
2532 {
2533 	int err;
2534 	size_t flow_size;
2535 	struct ib_flow_attr *flow = NULL;
2536 	struct ib_flow_spec_ib *ib_spec;
2537 
2538 	if (is_attach) {
2539 		flow_size = sizeof(struct ib_flow_attr) +
2540 			    sizeof(struct ib_flow_spec_ib);
2541 		flow = kzalloc(flow_size, GFP_KERNEL);
2542 		if (!flow)
2543 			return -ENOMEM;
2544 		flow->port = mqp->port;
2545 		flow->num_of_specs = 1;
2546 		flow->size = flow_size;
2547 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2548 		ib_spec->type = IB_FLOW_SPEC_IB;
2549 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2550 		/* Add an empty rule for IB L2 */
2551 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2552 
2553 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2554 					    IB_FLOW_DOMAIN_NIC,
2555 					    MLX4_FS_REGULAR,
2556 					    &mqp->reg_id);
2557 	} else {
2558 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2559 	}
2560 	kfree(flow);
2561 	return err;
2562 }
2563 
2564 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2565 {
2566 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2567 	int p;
2568 
2569 	ibdev->ib_active = false;
2570 	flush_workqueue(wq);
2571 
2572 	mlx4_ib_close_sriov(ibdev);
2573 	mlx4_ib_mad_cleanup(ibdev);
2574 	ib_unregister_device(&ibdev->ib_dev);
2575 	if (ibdev->iboe.nb.notifier_call) {
2576 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2577 			pr_warn("failure unregistering notifier\n");
2578 		ibdev->iboe.nb.notifier_call = NULL;
2579 	}
2580 
2581 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2582 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2583 				      ibdev->steer_qpn_count);
2584 		kfree(ibdev->ib_uc_qpns_bitmap);
2585 	}
2586 
2587 	iounmap(ibdev->uar_map);
2588 	for (p = 0; p < ibdev->num_ports; ++p)
2589 		mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2590 
2591 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2592 		mlx4_CLOSE_PORT(dev, p);
2593 
2594 	mlx4_ib_free_eqs(dev, ibdev);
2595 
2596 	mlx4_uar_free(dev, &ibdev->priv_uar);
2597 	mlx4_pd_free(dev, ibdev->priv_pdn);
2598 	ib_dealloc_device(&ibdev->ib_dev);
2599 }
2600 
2601 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2602 {
2603 	struct mlx4_ib_demux_work **dm = NULL;
2604 	struct mlx4_dev *dev = ibdev->dev;
2605 	int i;
2606 	unsigned long flags;
2607 	struct mlx4_active_ports actv_ports;
2608 	unsigned int ports;
2609 	unsigned int first_port;
2610 
2611 	if (!mlx4_is_master(dev))
2612 		return;
2613 
2614 	actv_ports = mlx4_get_active_ports(dev, slave);
2615 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2616 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2617 
2618 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2619 	if (!dm) {
2620 		pr_err("failed to allocate memory for tunneling qp update\n");
2621 		return;
2622 	}
2623 
2624 	for (i = 0; i < ports; i++) {
2625 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2626 		if (!dm[i]) {
2627 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2628 			while (--i >= 0)
2629 				kfree(dm[i]);
2630 			goto out;
2631 		}
2632 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2633 		dm[i]->port = first_port + i + 1;
2634 		dm[i]->slave = slave;
2635 		dm[i]->do_init = do_init;
2636 		dm[i]->dev = ibdev;
2637 	}
2638 	/* initialize or tear down tunnel QPs for the slave */
2639 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2640 	if (!ibdev->sriov.is_going_down) {
2641 		for (i = 0; i < ports; i++)
2642 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2643 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2644 	} else {
2645 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2646 		for (i = 0; i < ports; i++)
2647 			kfree(dm[i]);
2648 	}
2649 out:
2650 	kfree(dm);
2651 	return;
2652 }
2653 
2654 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2655 {
2656 	struct mlx4_ib_qp *mqp;
2657 	unsigned long flags_qp;
2658 	unsigned long flags_cq;
2659 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2660 	struct list_head    cq_notify_list;
2661 	struct mlx4_cq *mcq;
2662 	unsigned long flags;
2663 
2664 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2665 	INIT_LIST_HEAD(&cq_notify_list);
2666 
2667 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2668 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2669 
2670 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2671 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2672 		if (mqp->sq.tail != mqp->sq.head) {
2673 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2674 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2675 			if (send_mcq->mcq.comp &&
2676 			    mqp->ibqp.send_cq->comp_handler) {
2677 				if (!send_mcq->mcq.reset_notify_added) {
2678 					send_mcq->mcq.reset_notify_added = 1;
2679 					list_add_tail(&send_mcq->mcq.reset_notify,
2680 						      &cq_notify_list);
2681 				}
2682 			}
2683 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2684 		}
2685 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2686 		/* Now, handle the QP's receive queue */
2687 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2688 		/* no handling is needed for SRQ */
2689 		if (!mqp->ibqp.srq) {
2690 			if (mqp->rq.tail != mqp->rq.head) {
2691 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2692 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2693 				if (recv_mcq->mcq.comp &&
2694 				    mqp->ibqp.recv_cq->comp_handler) {
2695 					if (!recv_mcq->mcq.reset_notify_added) {
2696 						recv_mcq->mcq.reset_notify_added = 1;
2697 						list_add_tail(&recv_mcq->mcq.reset_notify,
2698 							      &cq_notify_list);
2699 					}
2700 				}
2701 				spin_unlock_irqrestore(&recv_mcq->lock,
2702 						       flags_cq);
2703 			}
2704 		}
2705 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2706 	}
2707 
2708 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2709 		mcq->comp(mcq);
2710 	}
2711 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2712 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2713 }
2714 
2715 static void handle_bonded_port_state_event(struct work_struct *work)
2716 {
2717 	struct ib_event_work *ew =
2718 		container_of(work, struct ib_event_work, work);
2719 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2720 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2721 	int i;
2722 	struct ib_event ibev;
2723 
2724 	kfree(ew);
2725 	spin_lock_bh(&ibdev->iboe.lock);
2726 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2727 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2728 		enum ib_port_state curr_port_state;
2729 
2730 		if (!curr_netdev)
2731 			continue;
2732 
2733 		curr_port_state =
2734 			(netif_running(curr_netdev) &&
2735 			 netif_carrier_ok(curr_netdev)) ?
2736 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2737 
2738 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2739 			curr_port_state : IB_PORT_ACTIVE;
2740 	}
2741 	spin_unlock_bh(&ibdev->iboe.lock);
2742 
2743 	ibev.device = &ibdev->ib_dev;
2744 	ibev.element.port_num = 1;
2745 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2746 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2747 
2748 	ib_dispatch_event(&ibev);
2749 }
2750 
2751 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2752 			  enum mlx4_dev_event event, unsigned long param)
2753 {
2754 	struct ib_event ibev;
2755 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2756 	struct mlx4_eqe *eqe = NULL;
2757 	struct ib_event_work *ew;
2758 	int p = 0;
2759 
2760 	if (mlx4_is_bonded(dev) &&
2761 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2762 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2763 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2764 		if (!ew)
2765 			return;
2766 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2767 		ew->ib_dev = ibdev;
2768 		queue_work(wq, &ew->work);
2769 		return;
2770 	}
2771 
2772 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2773 		eqe = (struct mlx4_eqe *)param;
2774 	else
2775 		p = (int) param;
2776 
2777 	switch (event) {
2778 	case MLX4_DEV_EVENT_PORT_UP:
2779 		if (p > ibdev->num_ports)
2780 			return;
2781 		if (mlx4_is_master(dev) &&
2782 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2783 			IB_LINK_LAYER_INFINIBAND) {
2784 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2785 		}
2786 		ibev.event = IB_EVENT_PORT_ACTIVE;
2787 		break;
2788 
2789 	case MLX4_DEV_EVENT_PORT_DOWN:
2790 		if (p > ibdev->num_ports)
2791 			return;
2792 		ibev.event = IB_EVENT_PORT_ERR;
2793 		break;
2794 
2795 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2796 		ibdev->ib_active = false;
2797 		ibev.event = IB_EVENT_DEVICE_FATAL;
2798 		mlx4_ib_handle_catas_error(ibdev);
2799 		break;
2800 
2801 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2802 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2803 		if (!ew) {
2804 			pr_err("failed to allocate memory for events work\n");
2805 			break;
2806 		}
2807 
2808 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2809 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2810 		ew->ib_dev = ibdev;
2811 		/* need to queue only for port owner, which uses GEN_EQE */
2812 		if (mlx4_is_master(dev))
2813 			queue_work(wq, &ew->work);
2814 		else
2815 			handle_port_mgmt_change_event(&ew->work);
2816 		return;
2817 
2818 	case MLX4_DEV_EVENT_SLAVE_INIT:
2819 		/* here, p is the slave id */
2820 		do_slave_init(ibdev, p, 1);
2821 		if (mlx4_is_master(dev)) {
2822 			int i;
2823 
2824 			for (i = 1; i <= ibdev->num_ports; i++) {
2825 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2826 					== IB_LINK_LAYER_INFINIBAND)
2827 					mlx4_ib_slave_alias_guid_event(ibdev,
2828 								       p, i,
2829 								       1);
2830 			}
2831 		}
2832 		return;
2833 
2834 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2835 		if (mlx4_is_master(dev)) {
2836 			int i;
2837 
2838 			for (i = 1; i <= ibdev->num_ports; i++) {
2839 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2840 					== IB_LINK_LAYER_INFINIBAND)
2841 					mlx4_ib_slave_alias_guid_event(ibdev,
2842 								       p, i,
2843 								       0);
2844 			}
2845 		}
2846 		/* here, p is the slave id */
2847 		do_slave_init(ibdev, p, 0);
2848 		return;
2849 
2850 	default:
2851 		return;
2852 	}
2853 
2854 	ibev.device	      = ibdev_ptr;
2855 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2856 
2857 	ib_dispatch_event(&ibev);
2858 }
2859 
2860 static struct mlx4_interface mlx4_ib_interface = {
2861 	.add		= mlx4_ib_add,
2862 	.remove		= mlx4_ib_remove,
2863 	.event		= mlx4_ib_event,
2864 	.protocol	= MLX4_PROT_IB_IPV6,
2865 	.flags		= MLX4_INTFF_BONDING
2866 };
2867 
2868 static int __init mlx4_ib_init(void)
2869 {
2870 	int err;
2871 
2872 	wq = create_singlethread_workqueue("mlx4_ib");
2873 	if (!wq)
2874 		return -ENOMEM;
2875 
2876 	err = mlx4_ib_mcg_init();
2877 	if (err)
2878 		goto clean_wq;
2879 
2880 	err = mlx4_register_interface(&mlx4_ib_interface);
2881 	if (err)
2882 		goto clean_mcg;
2883 
2884 	return 0;
2885 
2886 clean_mcg:
2887 	mlx4_ib_mcg_destroy();
2888 
2889 clean_wq:
2890 	destroy_workqueue(wq);
2891 	return err;
2892 }
2893 
2894 static void __exit mlx4_ib_cleanup(void)
2895 {
2896 	mlx4_unregister_interface(&mlx4_ib_interface);
2897 	mlx4_ib_mcg_destroy();
2898 	destroy_workqueue(wq);
2899 }
2900 
2901 module_init(mlx4_ib_init);
2902 module_exit(mlx4_ib_cleanup);
2903