xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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 
49 #include <linux/mlx4/driver.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/qp.h>
52 
53 #include "mlx4_ib.h"
54 #include "user.h"
55 
56 #define DRV_NAME	MLX4_IB_DRV_NAME
57 #define DRV_VERSION	"2.2-1"
58 #define DRV_RELDATE	"Feb 2014"
59 
60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62 #define MLX4_IB_CARD_REV_A0   0xA0
63 
64 MODULE_AUTHOR("Roland Dreier");
65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
66 MODULE_LICENSE("Dual BSD/GPL");
67 MODULE_VERSION(DRV_VERSION);
68 
69 int mlx4_ib_sm_guid_assign = 1;
70 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
71 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
72 
73 static const char mlx4_ib_version[] =
74 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
75 	DRV_VERSION " (" DRV_RELDATE ")\n";
76 
77 struct update_gid_work {
78 	struct work_struct	work;
79 	union ib_gid		gids[128];
80 	struct mlx4_ib_dev     *dev;
81 	int			port;
82 };
83 
84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
85 
86 static struct workqueue_struct *wq;
87 
88 static void init_query_mad(struct ib_smp *mad)
89 {
90 	mad->base_version  = 1;
91 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 	mad->class_version = 1;
93 	mad->method	   = IB_MGMT_METHOD_GET;
94 }
95 
96 static union ib_gid zgid;
97 
98 static int check_flow_steering_support(struct mlx4_dev *dev)
99 {
100 	int eth_num_ports = 0;
101 	int ib_num_ports = 0;
102 
103 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
104 
105 	if (dmfs) {
106 		int i;
107 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
108 			eth_num_ports++;
109 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
110 			ib_num_ports++;
111 		dmfs &= (!ib_num_ports ||
112 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
113 			(!eth_num_ports ||
114 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
115 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
116 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
117 			dmfs = 0;
118 		}
119 	}
120 	return dmfs;
121 }
122 
123 static int num_ib_ports(struct mlx4_dev *dev)
124 {
125 	int ib_ports = 0;
126 	int i;
127 
128 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
129 		ib_ports++;
130 
131 	return ib_ports;
132 }
133 
134 static int mlx4_ib_query_device(struct ib_device *ibdev,
135 				struct ib_device_attr *props)
136 {
137 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
138 	struct ib_smp *in_mad  = NULL;
139 	struct ib_smp *out_mad = NULL;
140 	int err = -ENOMEM;
141 	int have_ib_ports;
142 
143 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
144 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
145 	if (!in_mad || !out_mad)
146 		goto out;
147 
148 	init_query_mad(in_mad);
149 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
150 
151 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
152 			   1, NULL, NULL, in_mad, out_mad);
153 	if (err)
154 		goto out;
155 
156 	memset(props, 0, sizeof *props);
157 
158 	have_ib_ports = num_ib_ports(dev->dev);
159 
160 	props->fw_ver = dev->dev->caps.fw_ver;
161 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
162 		IB_DEVICE_PORT_ACTIVE_EVENT		|
163 		IB_DEVICE_SYS_IMAGE_GUID		|
164 		IB_DEVICE_RC_RNR_NAK_GEN		|
165 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
166 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
167 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
168 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
169 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
170 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
171 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
172 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
173 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
174 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
175 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
176 	if (dev->dev->caps.max_gso_sz &&
177 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
178 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
179 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
180 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
181 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
182 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
183 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
184 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
185 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
186 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
187 		props->device_cap_flags |= IB_DEVICE_XRC;
188 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
189 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
190 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
191 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
192 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
193 		else
194 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
195 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
196 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
197 	}
198 
199 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
200 		0xffffff;
201 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
202 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
203 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
204 
205 	props->max_mr_size	   = ~0ull;
206 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
207 	props->max_qp		   = dev->dev->quotas.qp;
208 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
209 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
210 					 dev->dev->caps.max_rq_sg);
211 	props->max_cq		   = dev->dev->quotas.cq;
212 	props->max_cqe		   = dev->dev->caps.max_cqes;
213 	props->max_mr		   = dev->dev->quotas.mpt;
214 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
215 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
216 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
217 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
218 	props->max_srq		   = dev->dev->quotas.srq;
219 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
220 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
221 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
222 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
223 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
224 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
225 	props->masked_atomic_cap   = props->atomic_cap;
226 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
227 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
228 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
229 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
230 					   props->max_mcast_grp;
231 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
232 
233 out:
234 	kfree(in_mad);
235 	kfree(out_mad);
236 
237 	return err;
238 }
239 
240 static enum rdma_link_layer
241 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
242 {
243 	struct mlx4_dev *dev = to_mdev(device)->dev;
244 
245 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
246 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
247 }
248 
249 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
250 			      struct ib_port_attr *props, int netw_view)
251 {
252 	struct ib_smp *in_mad  = NULL;
253 	struct ib_smp *out_mad = NULL;
254 	int ext_active_speed;
255 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
256 	int err = -ENOMEM;
257 
258 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
259 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
260 	if (!in_mad || !out_mad)
261 		goto out;
262 
263 	init_query_mad(in_mad);
264 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
265 	in_mad->attr_mod = cpu_to_be32(port);
266 
267 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
268 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
269 
270 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
271 				in_mad, out_mad);
272 	if (err)
273 		goto out;
274 
275 
276 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
277 	props->lmc		= out_mad->data[34] & 0x7;
278 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
279 	props->sm_sl		= out_mad->data[36] & 0xf;
280 	props->state		= out_mad->data[32] & 0xf;
281 	props->phys_state	= out_mad->data[33] >> 4;
282 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
283 	if (netw_view)
284 		props->gid_tbl_len = out_mad->data[50];
285 	else
286 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
287 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
288 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
289 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
290 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
291 	props->active_width	= out_mad->data[31] & 0xf;
292 	props->active_speed	= out_mad->data[35] >> 4;
293 	props->max_mtu		= out_mad->data[41] & 0xf;
294 	props->active_mtu	= out_mad->data[36] >> 4;
295 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
296 	props->max_vl_num	= out_mad->data[37] >> 4;
297 	props->init_type_reply	= out_mad->data[41] >> 4;
298 
299 	/* Check if extended speeds (EDR/FDR/...) are supported */
300 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
301 		ext_active_speed = out_mad->data[62] >> 4;
302 
303 		switch (ext_active_speed) {
304 		case 1:
305 			props->active_speed = IB_SPEED_FDR;
306 			break;
307 		case 2:
308 			props->active_speed = IB_SPEED_EDR;
309 			break;
310 		}
311 	}
312 
313 	/* If reported active speed is QDR, check if is FDR-10 */
314 	if (props->active_speed == IB_SPEED_QDR) {
315 		init_query_mad(in_mad);
316 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
317 		in_mad->attr_mod = cpu_to_be32(port);
318 
319 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
320 				   NULL, NULL, in_mad, out_mad);
321 		if (err)
322 			goto out;
323 
324 		/* Checking LinkSpeedActive for FDR-10 */
325 		if (out_mad->data[15] & 0x1)
326 			props->active_speed = IB_SPEED_FDR10;
327 	}
328 
329 	/* Avoid wrong speed value returned by FW if the IB link is down. */
330 	if (props->state == IB_PORT_DOWN)
331 		 props->active_speed = IB_SPEED_SDR;
332 
333 out:
334 	kfree(in_mad);
335 	kfree(out_mad);
336 	return err;
337 }
338 
339 static u8 state_to_phys_state(enum ib_port_state state)
340 {
341 	return state == IB_PORT_ACTIVE ? 5 : 3;
342 }
343 
344 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
345 			       struct ib_port_attr *props, int netw_view)
346 {
347 
348 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
349 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
350 	struct net_device *ndev;
351 	enum ib_mtu tmp;
352 	struct mlx4_cmd_mailbox *mailbox;
353 	int err = 0;
354 	int is_bonded = mlx4_is_bonded(mdev->dev);
355 
356 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
357 	if (IS_ERR(mailbox))
358 		return PTR_ERR(mailbox);
359 
360 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
361 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
362 			   MLX4_CMD_WRAPPED);
363 	if (err)
364 		goto out;
365 
366 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
367 						IB_WIDTH_4X : IB_WIDTH_1X;
368 	props->active_speed	= IB_SPEED_QDR;
369 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
370 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
371 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
372 	props->pkey_tbl_len	= 1;
373 	props->max_mtu		= IB_MTU_4096;
374 	props->max_vl_num	= 2;
375 	props->state		= IB_PORT_DOWN;
376 	props->phys_state	= state_to_phys_state(props->state);
377 	props->active_mtu	= IB_MTU_256;
378 	if (is_bonded)
379 		rtnl_lock(); /* required to get upper dev */
380 	spin_lock_bh(&iboe->lock);
381 	ndev = iboe->netdevs[port - 1];
382 	if (ndev && is_bonded)
383 		ndev = netdev_master_upper_dev_get(ndev);
384 	if (!ndev)
385 		goto out_unlock;
386 
387 	tmp = iboe_get_mtu(ndev->mtu);
388 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
389 
390 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
391 					IB_PORT_ACTIVE : IB_PORT_DOWN;
392 	props->phys_state	= state_to_phys_state(props->state);
393 out_unlock:
394 	spin_unlock_bh(&iboe->lock);
395 	if (is_bonded)
396 		rtnl_unlock();
397 out:
398 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
399 	return err;
400 }
401 
402 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
403 			 struct ib_port_attr *props, int netw_view)
404 {
405 	int err;
406 
407 	memset(props, 0, sizeof *props);
408 
409 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
410 		ib_link_query_port(ibdev, port, props, netw_view) :
411 				eth_link_query_port(ibdev, port, props, netw_view);
412 
413 	return err;
414 }
415 
416 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
417 			      struct ib_port_attr *props)
418 {
419 	/* returns host view */
420 	return __mlx4_ib_query_port(ibdev, port, props, 0);
421 }
422 
423 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
424 			union ib_gid *gid, int netw_view)
425 {
426 	struct ib_smp *in_mad  = NULL;
427 	struct ib_smp *out_mad = NULL;
428 	int err = -ENOMEM;
429 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
430 	int clear = 0;
431 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
432 
433 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
434 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
435 	if (!in_mad || !out_mad)
436 		goto out;
437 
438 	init_query_mad(in_mad);
439 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
440 	in_mad->attr_mod = cpu_to_be32(port);
441 
442 	if (mlx4_is_mfunc(dev->dev) && netw_view)
443 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
444 
445 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
446 	if (err)
447 		goto out;
448 
449 	memcpy(gid->raw, out_mad->data + 8, 8);
450 
451 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
452 		if (index) {
453 			/* For any index > 0, return the null guid */
454 			err = 0;
455 			clear = 1;
456 			goto out;
457 		}
458 	}
459 
460 	init_query_mad(in_mad);
461 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
462 	in_mad->attr_mod = cpu_to_be32(index / 8);
463 
464 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
465 			   NULL, NULL, in_mad, out_mad);
466 	if (err)
467 		goto out;
468 
469 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
470 
471 out:
472 	if (clear)
473 		memset(gid->raw + 8, 0, 8);
474 	kfree(in_mad);
475 	kfree(out_mad);
476 	return err;
477 }
478 
479 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
480 			  union ib_gid *gid)
481 {
482 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
483 
484 	*gid = dev->iboe.gid_table[port - 1][index];
485 
486 	return 0;
487 }
488 
489 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
490 			     union ib_gid *gid)
491 {
492 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
493 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
494 	else
495 		return iboe_query_gid(ibdev, port, index, gid);
496 }
497 
498 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
499 			 u16 *pkey, int netw_view)
500 {
501 	struct ib_smp *in_mad  = NULL;
502 	struct ib_smp *out_mad = NULL;
503 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
504 	int err = -ENOMEM;
505 
506 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
507 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
508 	if (!in_mad || !out_mad)
509 		goto out;
510 
511 	init_query_mad(in_mad);
512 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
513 	in_mad->attr_mod = cpu_to_be32(index / 32);
514 
515 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
516 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
517 
518 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
519 			   in_mad, out_mad);
520 	if (err)
521 		goto out;
522 
523 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
524 
525 out:
526 	kfree(in_mad);
527 	kfree(out_mad);
528 	return err;
529 }
530 
531 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
532 {
533 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
534 }
535 
536 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
537 				 struct ib_device_modify *props)
538 {
539 	struct mlx4_cmd_mailbox *mailbox;
540 	unsigned long flags;
541 
542 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
543 		return -EOPNOTSUPP;
544 
545 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
546 		return 0;
547 
548 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
549 		return -EOPNOTSUPP;
550 
551 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
552 	memcpy(ibdev->node_desc, props->node_desc, 64);
553 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
554 
555 	/*
556 	 * If possible, pass node desc to FW, so it can generate
557 	 * a 144 trap.  If cmd fails, just ignore.
558 	 */
559 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
560 	if (IS_ERR(mailbox))
561 		return 0;
562 
563 	memcpy(mailbox->buf, props->node_desc, 64);
564 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
565 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
566 
567 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
568 
569 	return 0;
570 }
571 
572 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
573 			    u32 cap_mask)
574 {
575 	struct mlx4_cmd_mailbox *mailbox;
576 	int err;
577 
578 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
579 	if (IS_ERR(mailbox))
580 		return PTR_ERR(mailbox);
581 
582 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
583 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
584 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
585 	} else {
586 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
587 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
588 	}
589 
590 	err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
591 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
592 
593 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
594 	return err;
595 }
596 
597 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
598 			       struct ib_port_modify *props)
599 {
600 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
601 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
602 	struct ib_port_attr attr;
603 	u32 cap_mask;
604 	int err;
605 
606 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
607 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
608 	 * violations and port capabilities are not meaningful.
609 	 */
610 	if (is_eth)
611 		return 0;
612 
613 	mutex_lock(&mdev->cap_mask_mutex);
614 
615 	err = mlx4_ib_query_port(ibdev, port, &attr);
616 	if (err)
617 		goto out;
618 
619 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
620 		~props->clr_port_cap_mask;
621 
622 	err = mlx4_ib_SET_PORT(mdev, port,
623 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
624 			       cap_mask);
625 
626 out:
627 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
628 	return err;
629 }
630 
631 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
632 						  struct ib_udata *udata)
633 {
634 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
635 	struct mlx4_ib_ucontext *context;
636 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
637 	struct mlx4_ib_alloc_ucontext_resp resp;
638 	int err;
639 
640 	if (!dev->ib_active)
641 		return ERR_PTR(-EAGAIN);
642 
643 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
644 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
645 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
646 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
647 	} else {
648 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
649 		resp.qp_tab_size      = dev->dev->caps.num_qps;
650 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
651 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
652 		resp.cqe_size	      = dev->dev->caps.cqe_size;
653 	}
654 
655 	context = kmalloc(sizeof *context, GFP_KERNEL);
656 	if (!context)
657 		return ERR_PTR(-ENOMEM);
658 
659 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
660 	if (err) {
661 		kfree(context);
662 		return ERR_PTR(err);
663 	}
664 
665 	INIT_LIST_HEAD(&context->db_page_list);
666 	mutex_init(&context->db_page_mutex);
667 
668 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
669 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
670 	else
671 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
672 
673 	if (err) {
674 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
675 		kfree(context);
676 		return ERR_PTR(-EFAULT);
677 	}
678 
679 	return &context->ibucontext;
680 }
681 
682 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
683 {
684 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
685 
686 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
687 	kfree(context);
688 
689 	return 0;
690 }
691 
692 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
693 {
694 	struct mlx4_ib_dev *dev = to_mdev(context->device);
695 
696 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
697 		return -EINVAL;
698 
699 	if (vma->vm_pgoff == 0) {
700 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
701 
702 		if (io_remap_pfn_range(vma, vma->vm_start,
703 				       to_mucontext(context)->uar.pfn,
704 				       PAGE_SIZE, vma->vm_page_prot))
705 			return -EAGAIN;
706 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
707 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
708 
709 		if (io_remap_pfn_range(vma, vma->vm_start,
710 				       to_mucontext(context)->uar.pfn +
711 				       dev->dev->caps.num_uars,
712 				       PAGE_SIZE, vma->vm_page_prot))
713 			return -EAGAIN;
714 	} else
715 		return -EINVAL;
716 
717 	return 0;
718 }
719 
720 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
721 				      struct ib_ucontext *context,
722 				      struct ib_udata *udata)
723 {
724 	struct mlx4_ib_pd *pd;
725 	int err;
726 
727 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
728 	if (!pd)
729 		return ERR_PTR(-ENOMEM);
730 
731 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
732 	if (err) {
733 		kfree(pd);
734 		return ERR_PTR(err);
735 	}
736 
737 	if (context)
738 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
739 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
740 			kfree(pd);
741 			return ERR_PTR(-EFAULT);
742 		}
743 
744 	return &pd->ibpd;
745 }
746 
747 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
748 {
749 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
750 	kfree(pd);
751 
752 	return 0;
753 }
754 
755 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
756 					  struct ib_ucontext *context,
757 					  struct ib_udata *udata)
758 {
759 	struct mlx4_ib_xrcd *xrcd;
760 	int err;
761 
762 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
763 		return ERR_PTR(-ENOSYS);
764 
765 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
766 	if (!xrcd)
767 		return ERR_PTR(-ENOMEM);
768 
769 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
770 	if (err)
771 		goto err1;
772 
773 	xrcd->pd = ib_alloc_pd(ibdev);
774 	if (IS_ERR(xrcd->pd)) {
775 		err = PTR_ERR(xrcd->pd);
776 		goto err2;
777 	}
778 
779 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
780 	if (IS_ERR(xrcd->cq)) {
781 		err = PTR_ERR(xrcd->cq);
782 		goto err3;
783 	}
784 
785 	return &xrcd->ibxrcd;
786 
787 err3:
788 	ib_dealloc_pd(xrcd->pd);
789 err2:
790 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
791 err1:
792 	kfree(xrcd);
793 	return ERR_PTR(err);
794 }
795 
796 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
797 {
798 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
799 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
800 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
801 	kfree(xrcd);
802 
803 	return 0;
804 }
805 
806 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
807 {
808 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
809 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
810 	struct mlx4_ib_gid_entry *ge;
811 
812 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
813 	if (!ge)
814 		return -ENOMEM;
815 
816 	ge->gid = *gid;
817 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
818 		ge->port = mqp->port;
819 		ge->added = 1;
820 	}
821 
822 	mutex_lock(&mqp->mutex);
823 	list_add_tail(&ge->list, &mqp->gid_list);
824 	mutex_unlock(&mqp->mutex);
825 
826 	return 0;
827 }
828 
829 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
830 		   union ib_gid *gid)
831 {
832 	struct net_device *ndev;
833 	int ret = 0;
834 
835 	if (!mqp->port)
836 		return 0;
837 
838 	spin_lock_bh(&mdev->iboe.lock);
839 	ndev = mdev->iboe.netdevs[mqp->port - 1];
840 	if (ndev)
841 		dev_hold(ndev);
842 	spin_unlock_bh(&mdev->iboe.lock);
843 
844 	if (ndev) {
845 		ret = 1;
846 		dev_put(ndev);
847 	}
848 
849 	return ret;
850 }
851 
852 struct mlx4_ib_steering {
853 	struct list_head list;
854 	struct mlx4_flow_reg_id reg_id;
855 	union ib_gid gid;
856 };
857 
858 static int parse_flow_attr(struct mlx4_dev *dev,
859 			   u32 qp_num,
860 			   union ib_flow_spec *ib_spec,
861 			   struct _rule_hw *mlx4_spec)
862 {
863 	enum mlx4_net_trans_rule_id type;
864 
865 	switch (ib_spec->type) {
866 	case IB_FLOW_SPEC_ETH:
867 		type = MLX4_NET_TRANS_RULE_ID_ETH;
868 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
869 		       ETH_ALEN);
870 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
871 		       ETH_ALEN);
872 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
873 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
874 		break;
875 	case IB_FLOW_SPEC_IB:
876 		type = MLX4_NET_TRANS_RULE_ID_IB;
877 		mlx4_spec->ib.l3_qpn =
878 			cpu_to_be32(qp_num);
879 		mlx4_spec->ib.qpn_mask =
880 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
881 		break;
882 
883 
884 	case IB_FLOW_SPEC_IPV4:
885 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
886 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
887 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
888 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
889 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
890 		break;
891 
892 	case IB_FLOW_SPEC_TCP:
893 	case IB_FLOW_SPEC_UDP:
894 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
895 					MLX4_NET_TRANS_RULE_ID_TCP :
896 					MLX4_NET_TRANS_RULE_ID_UDP;
897 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
898 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
899 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
900 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
901 		break;
902 
903 	default:
904 		return -EINVAL;
905 	}
906 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
907 	    mlx4_hw_rule_sz(dev, type) < 0)
908 		return -EINVAL;
909 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
910 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
911 	return mlx4_hw_rule_sz(dev, type);
912 }
913 
914 struct default_rules {
915 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
916 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
917 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
918 	__u8  link_layer;
919 };
920 static const struct default_rules default_table[] = {
921 	{
922 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
923 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
924 		.rules_create_list = {IB_FLOW_SPEC_IB},
925 		.link_layer = IB_LINK_LAYER_INFINIBAND
926 	}
927 };
928 
929 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
930 					 struct ib_flow_attr *flow_attr)
931 {
932 	int i, j, k;
933 	void *ib_flow;
934 	const struct default_rules *pdefault_rules = default_table;
935 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
936 
937 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
938 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
939 		memset(&field_types, 0, sizeof(field_types));
940 
941 		if (link_layer != pdefault_rules->link_layer)
942 			continue;
943 
944 		ib_flow = flow_attr + 1;
945 		/* we assume the specs are sorted */
946 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
947 		     j < flow_attr->num_of_specs; k++) {
948 			union ib_flow_spec *current_flow =
949 				(union ib_flow_spec *)ib_flow;
950 
951 			/* same layer but different type */
952 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
953 			     (pdefault_rules->mandatory_fields[k] &
954 			      IB_FLOW_SPEC_LAYER_MASK)) &&
955 			    (current_flow->type !=
956 			     pdefault_rules->mandatory_fields[k]))
957 				goto out;
958 
959 			/* same layer, try match next one */
960 			if (current_flow->type ==
961 			    pdefault_rules->mandatory_fields[k]) {
962 				j++;
963 				ib_flow +=
964 					((union ib_flow_spec *)ib_flow)->size;
965 			}
966 		}
967 
968 		ib_flow = flow_attr + 1;
969 		for (j = 0; j < flow_attr->num_of_specs;
970 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
971 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
972 				/* same layer and same type */
973 				if (((union ib_flow_spec *)ib_flow)->type ==
974 				    pdefault_rules->mandatory_not_fields[k])
975 					goto out;
976 
977 		return i;
978 	}
979 out:
980 	return -1;
981 }
982 
983 static int __mlx4_ib_create_default_rules(
984 		struct mlx4_ib_dev *mdev,
985 		struct ib_qp *qp,
986 		const struct default_rules *pdefault_rules,
987 		struct _rule_hw *mlx4_spec) {
988 	int size = 0;
989 	int i;
990 
991 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
992 		int ret;
993 		union ib_flow_spec ib_spec;
994 		switch (pdefault_rules->rules_create_list[i]) {
995 		case 0:
996 			/* no rule */
997 			continue;
998 		case IB_FLOW_SPEC_IB:
999 			ib_spec.type = IB_FLOW_SPEC_IB;
1000 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1001 
1002 			break;
1003 		default:
1004 			/* invalid rule */
1005 			return -EINVAL;
1006 		}
1007 		/* We must put empty rule, qpn is being ignored */
1008 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1009 				      mlx4_spec);
1010 		if (ret < 0) {
1011 			pr_info("invalid parsing\n");
1012 			return -EINVAL;
1013 		}
1014 
1015 		mlx4_spec = (void *)mlx4_spec + ret;
1016 		size += ret;
1017 	}
1018 	return size;
1019 }
1020 
1021 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1022 			  int domain,
1023 			  enum mlx4_net_trans_promisc_mode flow_type,
1024 			  u64 *reg_id)
1025 {
1026 	int ret, i;
1027 	int size = 0;
1028 	void *ib_flow;
1029 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1030 	struct mlx4_cmd_mailbox *mailbox;
1031 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1032 	int default_flow;
1033 
1034 	static const u16 __mlx4_domain[] = {
1035 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1036 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1037 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1038 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1039 	};
1040 
1041 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1042 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1043 		return -EINVAL;
1044 	}
1045 
1046 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1047 		pr_err("Invalid domain value %d\n", domain);
1048 		return -EINVAL;
1049 	}
1050 
1051 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1052 		return -EINVAL;
1053 
1054 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1055 	if (IS_ERR(mailbox))
1056 		return PTR_ERR(mailbox);
1057 	ctrl = mailbox->buf;
1058 
1059 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1060 				 flow_attr->priority);
1061 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1062 	ctrl->port = flow_attr->port;
1063 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1064 
1065 	ib_flow = flow_attr + 1;
1066 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1067 	/* Add default flows */
1068 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1069 	if (default_flow >= 0) {
1070 		ret = __mlx4_ib_create_default_rules(
1071 				mdev, qp, default_table + default_flow,
1072 				mailbox->buf + size);
1073 		if (ret < 0) {
1074 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1075 			return -EINVAL;
1076 		}
1077 		size += ret;
1078 	}
1079 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1080 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1081 				      mailbox->buf + size);
1082 		if (ret < 0) {
1083 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1084 			return -EINVAL;
1085 		}
1086 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1087 		size += ret;
1088 	}
1089 
1090 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1091 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1092 			   MLX4_CMD_NATIVE);
1093 	if (ret == -ENOMEM)
1094 		pr_err("mcg table is full. Fail to register network rule.\n");
1095 	else if (ret == -ENXIO)
1096 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1097 	else if (ret)
1098 		pr_err("Invalid argumant. Fail to register network rule.\n");
1099 
1100 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1101 	return ret;
1102 }
1103 
1104 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1105 {
1106 	int err;
1107 	err = mlx4_cmd(dev, reg_id, 0, 0,
1108 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1109 		       MLX4_CMD_NATIVE);
1110 	if (err)
1111 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1112 		       reg_id);
1113 	return err;
1114 }
1115 
1116 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1117 				    u64 *reg_id)
1118 {
1119 	void *ib_flow;
1120 	union ib_flow_spec *ib_spec;
1121 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1122 	int err = 0;
1123 
1124 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1125 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1126 		return 0; /* do nothing */
1127 
1128 	ib_flow = flow_attr + 1;
1129 	ib_spec = (union ib_flow_spec *)ib_flow;
1130 
1131 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1132 		return 0; /* do nothing */
1133 
1134 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1135 				    flow_attr->port, qp->qp_num,
1136 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1137 				    reg_id);
1138 	return err;
1139 }
1140 
1141 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1142 				    struct ib_flow_attr *flow_attr,
1143 				    int domain)
1144 {
1145 	int err = 0, i = 0, j = 0;
1146 	struct mlx4_ib_flow *mflow;
1147 	enum mlx4_net_trans_promisc_mode type[2];
1148 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1149 	int is_bonded = mlx4_is_bonded(dev);
1150 
1151 	memset(type, 0, sizeof(type));
1152 
1153 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1154 	if (!mflow) {
1155 		err = -ENOMEM;
1156 		goto err_free;
1157 	}
1158 
1159 	switch (flow_attr->type) {
1160 	case IB_FLOW_ATTR_NORMAL:
1161 		type[0] = MLX4_FS_REGULAR;
1162 		break;
1163 
1164 	case IB_FLOW_ATTR_ALL_DEFAULT:
1165 		type[0] = MLX4_FS_ALL_DEFAULT;
1166 		break;
1167 
1168 	case IB_FLOW_ATTR_MC_DEFAULT:
1169 		type[0] = MLX4_FS_MC_DEFAULT;
1170 		break;
1171 
1172 	case IB_FLOW_ATTR_SNIFFER:
1173 		type[0] = MLX4_FS_UC_SNIFFER;
1174 		type[1] = MLX4_FS_MC_SNIFFER;
1175 		break;
1176 
1177 	default:
1178 		err = -EINVAL;
1179 		goto err_free;
1180 	}
1181 
1182 	while (i < ARRAY_SIZE(type) && type[i]) {
1183 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1184 					    &mflow->reg_id[i].id);
1185 		if (err)
1186 			goto err_create_flow;
1187 		i++;
1188 		if (is_bonded) {
1189 			/* Application always sees one port so the mirror rule
1190 			 * must be on port #2
1191 			 */
1192 			flow_attr->port = 2;
1193 			err = __mlx4_ib_create_flow(qp, flow_attr,
1194 						    domain, type[j],
1195 						    &mflow->reg_id[j].mirror);
1196 			flow_attr->port = 1;
1197 			if (err)
1198 				goto err_create_flow;
1199 			j++;
1200 		}
1201 
1202 	}
1203 
1204 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1205 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1206 					       &mflow->reg_id[i].id);
1207 		if (err)
1208 			goto err_create_flow;
1209 		i++;
1210 		if (is_bonded) {
1211 			flow_attr->port = 2;
1212 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1213 						       &mflow->reg_id[j].mirror);
1214 			flow_attr->port = 1;
1215 			if (err)
1216 				goto err_create_flow;
1217 			j++;
1218 		}
1219 		/* function to create mirror rule */
1220 	}
1221 
1222 	return &mflow->ibflow;
1223 
1224 err_create_flow:
1225 	while (i) {
1226 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1227 					     mflow->reg_id[i].id);
1228 		i--;
1229 	}
1230 
1231 	while (j) {
1232 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1233 					     mflow->reg_id[j].mirror);
1234 		j--;
1235 	}
1236 err_free:
1237 	kfree(mflow);
1238 	return ERR_PTR(err);
1239 }
1240 
1241 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1242 {
1243 	int err, ret = 0;
1244 	int i = 0;
1245 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1246 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1247 
1248 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1249 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1250 		if (err)
1251 			ret = err;
1252 		if (mflow->reg_id[i].mirror) {
1253 			err = __mlx4_ib_destroy_flow(mdev->dev,
1254 						     mflow->reg_id[i].mirror);
1255 			if (err)
1256 				ret = err;
1257 		}
1258 		i++;
1259 	}
1260 
1261 	kfree(mflow);
1262 	return ret;
1263 }
1264 
1265 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1266 {
1267 	int err;
1268 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1269 	struct mlx4_dev	*dev = mdev->dev;
1270 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1271 	struct mlx4_ib_steering *ib_steering = NULL;
1272 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1273 	struct mlx4_flow_reg_id	reg_id;
1274 
1275 	if (mdev->dev->caps.steering_mode ==
1276 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1277 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1278 		if (!ib_steering)
1279 			return -ENOMEM;
1280 	}
1281 
1282 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1283 				    !!(mqp->flags &
1284 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1285 				    prot, &reg_id.id);
1286 	if (err) {
1287 		pr_err("multicast attach op failed, err %d\n", err);
1288 		goto err_malloc;
1289 	}
1290 
1291 	reg_id.mirror = 0;
1292 	if (mlx4_is_bonded(dev)) {
1293 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1294 					    (mqp->port == 1) ? 2 : 1,
1295 					    !!(mqp->flags &
1296 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1297 					    prot, &reg_id.mirror);
1298 		if (err)
1299 			goto err_add;
1300 	}
1301 
1302 	err = add_gid_entry(ibqp, gid);
1303 	if (err)
1304 		goto err_add;
1305 
1306 	if (ib_steering) {
1307 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1308 		ib_steering->reg_id = reg_id;
1309 		mutex_lock(&mqp->mutex);
1310 		list_add(&ib_steering->list, &mqp->steering_rules);
1311 		mutex_unlock(&mqp->mutex);
1312 	}
1313 	return 0;
1314 
1315 err_add:
1316 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1317 			      prot, reg_id.id);
1318 	if (reg_id.mirror)
1319 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1320 				      prot, reg_id.mirror);
1321 err_malloc:
1322 	kfree(ib_steering);
1323 
1324 	return err;
1325 }
1326 
1327 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1328 {
1329 	struct mlx4_ib_gid_entry *ge;
1330 	struct mlx4_ib_gid_entry *tmp;
1331 	struct mlx4_ib_gid_entry *ret = NULL;
1332 
1333 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1334 		if (!memcmp(raw, ge->gid.raw, 16)) {
1335 			ret = ge;
1336 			break;
1337 		}
1338 	}
1339 
1340 	return ret;
1341 }
1342 
1343 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1344 {
1345 	int err;
1346 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1347 	struct mlx4_dev *dev = mdev->dev;
1348 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1349 	struct net_device *ndev;
1350 	struct mlx4_ib_gid_entry *ge;
1351 	struct mlx4_flow_reg_id reg_id = {0, 0};
1352 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1353 
1354 	if (mdev->dev->caps.steering_mode ==
1355 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1356 		struct mlx4_ib_steering *ib_steering;
1357 
1358 		mutex_lock(&mqp->mutex);
1359 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1360 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1361 				list_del(&ib_steering->list);
1362 				break;
1363 			}
1364 		}
1365 		mutex_unlock(&mqp->mutex);
1366 		if (&ib_steering->list == &mqp->steering_rules) {
1367 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1368 			return -EINVAL;
1369 		}
1370 		reg_id = ib_steering->reg_id;
1371 		kfree(ib_steering);
1372 	}
1373 
1374 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1375 				    prot, reg_id.id);
1376 	if (err)
1377 		return err;
1378 
1379 	if (mlx4_is_bonded(dev)) {
1380 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1381 					    prot, reg_id.mirror);
1382 		if (err)
1383 			return err;
1384 	}
1385 
1386 	mutex_lock(&mqp->mutex);
1387 	ge = find_gid_entry(mqp, gid->raw);
1388 	if (ge) {
1389 		spin_lock_bh(&mdev->iboe.lock);
1390 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1391 		if (ndev)
1392 			dev_hold(ndev);
1393 		spin_unlock_bh(&mdev->iboe.lock);
1394 		if (ndev)
1395 			dev_put(ndev);
1396 		list_del(&ge->list);
1397 		kfree(ge);
1398 	} else
1399 		pr_warn("could not find mgid entry\n");
1400 
1401 	mutex_unlock(&mqp->mutex);
1402 
1403 	return 0;
1404 }
1405 
1406 static int init_node_data(struct mlx4_ib_dev *dev)
1407 {
1408 	struct ib_smp *in_mad  = NULL;
1409 	struct ib_smp *out_mad = NULL;
1410 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1411 	int err = -ENOMEM;
1412 
1413 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1414 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1415 	if (!in_mad || !out_mad)
1416 		goto out;
1417 
1418 	init_query_mad(in_mad);
1419 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1420 	if (mlx4_is_master(dev->dev))
1421 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1422 
1423 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1424 	if (err)
1425 		goto out;
1426 
1427 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1428 
1429 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1430 
1431 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1432 	if (err)
1433 		goto out;
1434 
1435 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1436 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1437 
1438 out:
1439 	kfree(in_mad);
1440 	kfree(out_mad);
1441 	return err;
1442 }
1443 
1444 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1445 			char *buf)
1446 {
1447 	struct mlx4_ib_dev *dev =
1448 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1449 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1450 }
1451 
1452 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1453 			   char *buf)
1454 {
1455 	struct mlx4_ib_dev *dev =
1456 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1457 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1458 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1459 		       (int) dev->dev->caps.fw_ver & 0xffff);
1460 }
1461 
1462 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1463 			char *buf)
1464 {
1465 	struct mlx4_ib_dev *dev =
1466 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1467 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1468 }
1469 
1470 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1471 			  char *buf)
1472 {
1473 	struct mlx4_ib_dev *dev =
1474 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1475 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1476 		       dev->dev->board_id);
1477 }
1478 
1479 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1480 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1481 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1482 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1483 
1484 static struct device_attribute *mlx4_class_attributes[] = {
1485 	&dev_attr_hw_rev,
1486 	&dev_attr_fw_ver,
1487 	&dev_attr_hca_type,
1488 	&dev_attr_board_id
1489 };
1490 
1491 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1492 				     struct net_device *dev)
1493 {
1494 	memcpy(eui, dev->dev_addr, 3);
1495 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1496 	if (vlan_id < 0x1000) {
1497 		eui[3] = vlan_id >> 8;
1498 		eui[4] = vlan_id & 0xff;
1499 	} else {
1500 		eui[3] = 0xff;
1501 		eui[4] = 0xfe;
1502 	}
1503 	eui[0] ^= 2;
1504 }
1505 
1506 static void update_gids_task(struct work_struct *work)
1507 {
1508 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1509 	struct mlx4_cmd_mailbox *mailbox;
1510 	union ib_gid *gids;
1511 	int err;
1512 	struct mlx4_dev	*dev = gw->dev->dev;
1513 	int is_bonded = mlx4_is_bonded(dev);
1514 
1515 	if (!gw->dev->ib_active)
1516 		return;
1517 
1518 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1519 	if (IS_ERR(mailbox)) {
1520 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1521 		return;
1522 	}
1523 
1524 	gids = mailbox->buf;
1525 	memcpy(gids, gw->gids, sizeof gw->gids);
1526 
1527 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1528 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1529 		       MLX4_CMD_WRAPPED);
1530 	if (err)
1531 		pr_warn("set port command failed\n");
1532 	else
1533 		if ((gw->port == 1) || !is_bonded)
1534 			mlx4_ib_dispatch_event(gw->dev,
1535 					       is_bonded ? 1 : gw->port,
1536 					       IB_EVENT_GID_CHANGE);
1537 
1538 	mlx4_free_cmd_mailbox(dev, mailbox);
1539 	kfree(gw);
1540 }
1541 
1542 static void reset_gids_task(struct work_struct *work)
1543 {
1544 	struct update_gid_work *gw =
1545 			container_of(work, struct update_gid_work, work);
1546 	struct mlx4_cmd_mailbox *mailbox;
1547 	union ib_gid *gids;
1548 	int err;
1549 	struct mlx4_dev	*dev = gw->dev->dev;
1550 
1551 	if (!gw->dev->ib_active)
1552 		return;
1553 
1554 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1555 	if (IS_ERR(mailbox)) {
1556 		pr_warn("reset gid table failed\n");
1557 		goto free;
1558 	}
1559 
1560 	gids = mailbox->buf;
1561 	memcpy(gids, gw->gids, sizeof(gw->gids));
1562 
1563 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1564 				    IB_LINK_LAYER_ETHERNET) {
1565 		err = mlx4_cmd(dev, mailbox->dma,
1566 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1567 			       1, MLX4_CMD_SET_PORT,
1568 			       MLX4_CMD_TIME_CLASS_B,
1569 			       MLX4_CMD_WRAPPED);
1570 		if (err)
1571 			pr_warn(KERN_WARNING
1572 				"set port %d command failed\n", gw->port);
1573 	}
1574 
1575 	mlx4_free_cmd_mailbox(dev, mailbox);
1576 free:
1577 	kfree(gw);
1578 }
1579 
1580 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1581 			    union ib_gid *gid, int clear,
1582 			    int default_gid)
1583 {
1584 	struct update_gid_work *work;
1585 	int i;
1586 	int need_update = 0;
1587 	int free = -1;
1588 	int found = -1;
1589 	int max_gids;
1590 
1591 	if (default_gid) {
1592 		free = 0;
1593 	} else {
1594 		max_gids = dev->dev->caps.gid_table_len[port];
1595 		for (i = 1; i < max_gids; ++i) {
1596 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1597 				    sizeof(*gid)))
1598 				found = i;
1599 
1600 			if (clear) {
1601 				if (found >= 0) {
1602 					need_update = 1;
1603 					dev->iboe.gid_table[port - 1][found] =
1604 						zgid;
1605 					break;
1606 				}
1607 			} else {
1608 				if (found >= 0)
1609 					break;
1610 
1611 				if (free < 0 &&
1612 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1613 					    &zgid, sizeof(*gid)))
1614 					free = i;
1615 			}
1616 		}
1617 	}
1618 
1619 	if (found == -1 && !clear && free >= 0) {
1620 		dev->iboe.gid_table[port - 1][free] = *gid;
1621 		need_update = 1;
1622 	}
1623 
1624 	if (!need_update)
1625 		return 0;
1626 
1627 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1628 	if (!work)
1629 		return -ENOMEM;
1630 
1631 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1632 	INIT_WORK(&work->work, update_gids_task);
1633 	work->port = port;
1634 	work->dev = dev;
1635 	queue_work(wq, &work->work);
1636 
1637 	return 0;
1638 }
1639 
1640 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1641 {
1642 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1643 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1644 }
1645 
1646 
1647 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1648 {
1649 	struct update_gid_work *work;
1650 
1651 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1652 	if (!work)
1653 		return -ENOMEM;
1654 
1655 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1656 	memset(work->gids, 0, sizeof(work->gids));
1657 	INIT_WORK(&work->work, reset_gids_task);
1658 	work->dev = dev;
1659 	work->port = port;
1660 	queue_work(wq, &work->work);
1661 	return 0;
1662 }
1663 
1664 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1665 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1666 {
1667 	struct mlx4_ib_iboe *iboe;
1668 	int port = 0;
1669 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1670 				rdma_vlan_dev_real_dev(event_netdev) :
1671 				event_netdev;
1672 	union ib_gid default_gid;
1673 
1674 	mlx4_make_default_gid(real_dev, &default_gid);
1675 
1676 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1677 		return 0;
1678 
1679 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1680 		return 0;
1681 
1682 	if ((real_dev != event_netdev) &&
1683 	    (event == NETDEV_DOWN) &&
1684 	    rdma_link_local_addr((struct in6_addr *)gid))
1685 		return 0;
1686 
1687 	iboe = &ibdev->iboe;
1688 	spin_lock_bh(&iboe->lock);
1689 
1690 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1691 		if ((netif_is_bond_master(real_dev) &&
1692 		     (real_dev == iboe->masters[port - 1])) ||
1693 		     (!netif_is_bond_master(real_dev) &&
1694 		     (real_dev == iboe->netdevs[port - 1])))
1695 			update_gid_table(ibdev, port, gid,
1696 					 event == NETDEV_DOWN, 0);
1697 
1698 	spin_unlock_bh(&iboe->lock);
1699 	return 0;
1700 
1701 }
1702 
1703 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1704 			       struct mlx4_ib_dev *ibdev)
1705 {
1706 	u8 port = 0;
1707 	struct mlx4_ib_iboe *iboe;
1708 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1709 				rdma_vlan_dev_real_dev(dev) : dev;
1710 
1711 	iboe = &ibdev->iboe;
1712 
1713 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1714 		if ((netif_is_bond_master(real_dev) &&
1715 		     (real_dev == iboe->masters[port - 1])) ||
1716 		     (!netif_is_bond_master(real_dev) &&
1717 		     (real_dev == iboe->netdevs[port - 1])))
1718 			break;
1719 
1720 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1721 		return 0;
1722 	else
1723 		return port;
1724 }
1725 
1726 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1727 				void *ptr)
1728 {
1729 	struct mlx4_ib_dev *ibdev;
1730 	struct in_ifaddr *ifa = ptr;
1731 	union ib_gid gid;
1732 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1733 
1734 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1735 
1736 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1737 
1738 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1739 	return NOTIFY_DONE;
1740 }
1741 
1742 #if IS_ENABLED(CONFIG_IPV6)
1743 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1744 				void *ptr)
1745 {
1746 	struct mlx4_ib_dev *ibdev;
1747 	struct inet6_ifaddr *ifa = ptr;
1748 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1749 	struct net_device *event_netdev = ifa->idev->dev;
1750 
1751 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1752 
1753 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1754 	return NOTIFY_DONE;
1755 }
1756 #endif
1757 
1758 #define MLX4_IB_INVALID_MAC	((u64)-1)
1759 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1760 			       struct net_device *dev,
1761 			       int port)
1762 {
1763 	u64 new_smac = 0;
1764 	u64 release_mac = MLX4_IB_INVALID_MAC;
1765 	struct mlx4_ib_qp *qp;
1766 
1767 	read_lock(&dev_base_lock);
1768 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1769 	read_unlock(&dev_base_lock);
1770 
1771 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1772 
1773 	/* no need for update QP1 and mac registration in non-SRIOV */
1774 	if (!mlx4_is_mfunc(ibdev->dev))
1775 		return;
1776 
1777 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1778 	qp = ibdev->qp1_proxy[port - 1];
1779 	if (qp) {
1780 		int new_smac_index;
1781 		u64 old_smac;
1782 		struct mlx4_update_qp_params update_params;
1783 
1784 		mutex_lock(&qp->mutex);
1785 		old_smac = qp->pri.smac;
1786 		if (new_smac == old_smac)
1787 			goto unlock;
1788 
1789 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1790 
1791 		if (new_smac_index < 0)
1792 			goto unlock;
1793 
1794 		update_params.smac_index = new_smac_index;
1795 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1796 				   &update_params)) {
1797 			release_mac = new_smac;
1798 			goto unlock;
1799 		}
1800 		/* if old port was zero, no mac was yet registered for this QP */
1801 		if (qp->pri.smac_port)
1802 			release_mac = old_smac;
1803 		qp->pri.smac = new_smac;
1804 		qp->pri.smac_port = port;
1805 		qp->pri.smac_index = new_smac_index;
1806 	}
1807 
1808 unlock:
1809 	if (release_mac != MLX4_IB_INVALID_MAC)
1810 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1811 	if (qp)
1812 		mutex_unlock(&qp->mutex);
1813 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1814 }
1815 
1816 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1817 				 struct mlx4_ib_dev *ibdev, u8 port)
1818 {
1819 	struct in_device *in_dev;
1820 #if IS_ENABLED(CONFIG_IPV6)
1821 	struct inet6_dev *in6_dev;
1822 	union ib_gid  *pgid;
1823 	struct inet6_ifaddr *ifp;
1824 	union ib_gid default_gid;
1825 #endif
1826 	union ib_gid gid;
1827 
1828 
1829 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1830 		return;
1831 
1832 	/* IPv4 gids */
1833 	in_dev = in_dev_get(dev);
1834 	if (in_dev) {
1835 		for_ifa(in_dev) {
1836 			/*ifa->ifa_address;*/
1837 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1838 					       (struct in6_addr *)&gid);
1839 			update_gid_table(ibdev, port, &gid, 0, 0);
1840 		}
1841 		endfor_ifa(in_dev);
1842 		in_dev_put(in_dev);
1843 	}
1844 #if IS_ENABLED(CONFIG_IPV6)
1845 	mlx4_make_default_gid(dev, &default_gid);
1846 	/* IPv6 gids */
1847 	in6_dev = in6_dev_get(dev);
1848 	if (in6_dev) {
1849 		read_lock_bh(&in6_dev->lock);
1850 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1851 			pgid = (union ib_gid *)&ifp->addr;
1852 			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1853 				continue;
1854 			update_gid_table(ibdev, port, pgid, 0, 0);
1855 		}
1856 		read_unlock_bh(&in6_dev->lock);
1857 		in6_dev_put(in6_dev);
1858 	}
1859 #endif
1860 }
1861 
1862 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1863 				 struct  net_device *dev, u8 port)
1864 {
1865 	union ib_gid gid;
1866 	mlx4_make_default_gid(dev, &gid);
1867 	update_gid_table(ibdev, port, &gid, 0, 1);
1868 }
1869 
1870 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1871 {
1872 	struct	net_device *dev;
1873 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1874 	int i;
1875 	int err = 0;
1876 
1877 	for (i = 1; i <= ibdev->num_ports; ++i) {
1878 		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1879 		    IB_LINK_LAYER_ETHERNET) {
1880 			err = reset_gid_table(ibdev, i);
1881 			if (err)
1882 				goto out;
1883 		}
1884 	}
1885 
1886 	read_lock(&dev_base_lock);
1887 	spin_lock_bh(&iboe->lock);
1888 
1889 	for_each_netdev(&init_net, dev) {
1890 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1891 		/* port will be non-zero only for ETH ports */
1892 		if (port) {
1893 			mlx4_ib_set_default_gid(ibdev, dev, port);
1894 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1895 		}
1896 	}
1897 
1898 	spin_unlock_bh(&iboe->lock);
1899 	read_unlock(&dev_base_lock);
1900 out:
1901 	return err;
1902 }
1903 
1904 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1905 				 struct net_device *dev,
1906 				 unsigned long event)
1907 
1908 {
1909 	struct mlx4_ib_iboe *iboe;
1910 	int update_qps_port = -1;
1911 	int port;
1912 
1913 	iboe = &ibdev->iboe;
1914 
1915 	spin_lock_bh(&iboe->lock);
1916 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1917 		enum ib_port_state	port_state = IB_PORT_NOP;
1918 		struct net_device *old_master = iboe->masters[port - 1];
1919 		struct net_device *curr_netdev;
1920 		struct net_device *curr_master;
1921 
1922 		iboe->netdevs[port - 1] =
1923 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1924 		if (iboe->netdevs[port - 1])
1925 			mlx4_ib_set_default_gid(ibdev,
1926 						iboe->netdevs[port - 1], port);
1927 		curr_netdev = iboe->netdevs[port - 1];
1928 
1929 		if (iboe->netdevs[port - 1] &&
1930 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1931 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1932 				iboe->netdevs[port - 1]);
1933 		} else {
1934 			iboe->masters[port - 1] = NULL;
1935 		}
1936 		curr_master = iboe->masters[port - 1];
1937 
1938 		if (dev == iboe->netdevs[port - 1] &&
1939 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1940 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1941 			update_qps_port = port;
1942 
1943 		if (curr_netdev) {
1944 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1945 						IB_PORT_ACTIVE : IB_PORT_DOWN;
1946 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1947 			if (curr_master) {
1948 				/* if using bonding/team and a slave port is down, we
1949 				 * don't want the bond IP based gids in the table since
1950 				 * flows that select port by gid may get the down port.
1951 				*/
1952 				if (port_state == IB_PORT_DOWN &&
1953 				    !mlx4_is_bonded(ibdev->dev)) {
1954 					reset_gid_table(ibdev, port);
1955 					mlx4_ib_set_default_gid(ibdev,
1956 								curr_netdev,
1957 								port);
1958 				} else {
1959 					/* gids from the upper dev (bond/team)
1960 					 * should appear in port's gid table
1961 					*/
1962 					mlx4_ib_get_dev_addr(curr_master,
1963 							     ibdev, port);
1964 				}
1965 			}
1966 			/* if bonding is used it is possible that we add it to
1967 			 * masters only after IP address is assigned to the
1968 			 * net bonding interface.
1969 			*/
1970 			if (curr_master && (old_master != curr_master)) {
1971 				reset_gid_table(ibdev, port);
1972 				mlx4_ib_set_default_gid(ibdev,
1973 							curr_netdev, port);
1974 				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1975 			}
1976 
1977 			if (!curr_master && (old_master != curr_master)) {
1978 				reset_gid_table(ibdev, port);
1979 				mlx4_ib_set_default_gid(ibdev,
1980 							curr_netdev, port);
1981 				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1982 			}
1983 		} else {
1984 			reset_gid_table(ibdev, port);
1985 		}
1986 	}
1987 
1988 	spin_unlock_bh(&iboe->lock);
1989 
1990 	if (update_qps_port > 0)
1991 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1992 }
1993 
1994 static int mlx4_ib_netdev_event(struct notifier_block *this,
1995 				unsigned long event, void *ptr)
1996 {
1997 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1998 	struct mlx4_ib_dev *ibdev;
1999 
2000 	if (!net_eq(dev_net(dev), &init_net))
2001 		return NOTIFY_DONE;
2002 
2003 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2004 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2005 
2006 	return NOTIFY_DONE;
2007 }
2008 
2009 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2010 {
2011 	int port;
2012 	int slave;
2013 	int i;
2014 
2015 	if (mlx4_is_master(ibdev->dev)) {
2016 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2017 		     ++slave) {
2018 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2019 				for (i = 0;
2020 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2021 				     ++i) {
2022 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2023 					/* master has the identity virt2phys pkey mapping */
2024 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2025 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2026 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2027 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2028 				}
2029 			}
2030 		}
2031 		/* initialize pkey cache */
2032 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2033 			for (i = 0;
2034 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2035 			     ++i)
2036 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2037 					(i) ? 0 : 0xFFFF;
2038 		}
2039 	}
2040 }
2041 
2042 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2043 {
2044 	char name[80];
2045 	int eq_per_port = 0;
2046 	int added_eqs = 0;
2047 	int total_eqs = 0;
2048 	int i, j, eq;
2049 
2050 	/* Legacy mode or comp_pool is not large enough */
2051 	if (dev->caps.comp_pool == 0 ||
2052 	    dev->caps.num_ports > dev->caps.comp_pool)
2053 		return;
2054 
2055 	eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
2056 
2057 	/* Init eq table */
2058 	added_eqs = 0;
2059 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2060 		added_eqs += eq_per_port;
2061 
2062 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
2063 
2064 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2065 	if (!ibdev->eq_table)
2066 		return;
2067 
2068 	ibdev->eq_added = added_eqs;
2069 
2070 	eq = 0;
2071 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2072 		for (j = 0; j < eq_per_port; j++) {
2073 			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
2074 				 i, j, dev->persist->pdev->bus->name);
2075 			/* Set IRQ for specific name (per ring) */
2076 			if (mlx4_assign_eq(dev, name, NULL,
2077 					   &ibdev->eq_table[eq])) {
2078 				/* Use legacy (same as mlx4_en driver) */
2079 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2080 				ibdev->eq_table[eq] =
2081 					(eq % dev->caps.num_comp_vectors);
2082 			}
2083 			eq++;
2084 		}
2085 	}
2086 
2087 	/* Fill the reset of the vector with legacy EQ */
2088 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2089 		ibdev->eq_table[eq++] = i;
2090 
2091 	/* Advertise the new number of EQs to clients */
2092 	ibdev->ib_dev.num_comp_vectors = total_eqs;
2093 }
2094 
2095 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2096 {
2097 	int i;
2098 
2099 	/* no additional eqs were added */
2100 	if (!ibdev->eq_table)
2101 		return;
2102 
2103 	/* Reset the advertised EQ number */
2104 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2105 
2106 	/* Free only the added eqs */
2107 	for (i = 0; i < ibdev->eq_added; i++) {
2108 		/* Don't free legacy eqs if used */
2109 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2110 			continue;
2111 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2112 	}
2113 
2114 	kfree(ibdev->eq_table);
2115 }
2116 
2117 static void *mlx4_ib_add(struct mlx4_dev *dev)
2118 {
2119 	struct mlx4_ib_dev *ibdev;
2120 	int num_ports = 0;
2121 	int i, j;
2122 	int err;
2123 	struct mlx4_ib_iboe *iboe;
2124 	int ib_num_ports = 0;
2125 	int num_req_counters;
2126 
2127 	pr_info_once("%s", mlx4_ib_version);
2128 
2129 	num_ports = 0;
2130 	mlx4_foreach_ib_transport_port(i, dev)
2131 		num_ports++;
2132 
2133 	/* No point in registering a device with no ports... */
2134 	if (num_ports == 0)
2135 		return NULL;
2136 
2137 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2138 	if (!ibdev) {
2139 		dev_err(&dev->persist->pdev->dev,
2140 			"Device struct alloc failed\n");
2141 		return NULL;
2142 	}
2143 
2144 	iboe = &ibdev->iboe;
2145 
2146 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2147 		goto err_dealloc;
2148 
2149 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2150 		goto err_pd;
2151 
2152 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2153 				 PAGE_SIZE);
2154 	if (!ibdev->uar_map)
2155 		goto err_uar;
2156 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2157 
2158 	ibdev->dev = dev;
2159 	ibdev->bond_next_port	= 0;
2160 
2161 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2162 	ibdev->ib_dev.owner		= THIS_MODULE;
2163 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2164 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2165 	ibdev->num_ports		= num_ports;
2166 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2167 						1 : ibdev->num_ports;
2168 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2169 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2170 
2171 	if (dev->caps.userspace_caps)
2172 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2173 	else
2174 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2175 
2176 	ibdev->ib_dev.uverbs_cmd_mask	=
2177 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2178 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2179 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2180 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2181 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2182 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2183 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2184 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2185 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2186 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2187 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2188 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2189 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2190 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2191 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2192 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2193 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2194 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2195 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2196 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2197 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2198 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2199 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2200 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2201 
2202 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2203 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2204 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2205 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2206 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2207 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2208 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2209 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2210 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2211 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2212 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2213 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2214 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2215 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2216 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2217 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2218 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2219 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2220 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2221 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2222 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2223 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2224 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2225 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2226 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2227 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2228 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2229 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2230 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2231 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2232 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2233 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2234 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2235 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2236 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2237 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2238 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2239 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2240 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2241 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2242 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2243 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2244 
2245 	if (!mlx4_is_slave(ibdev->dev)) {
2246 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2247 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2248 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2249 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2250 	}
2251 
2252 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2253 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2254 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2255 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2256 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2257 
2258 		ibdev->ib_dev.uverbs_cmd_mask |=
2259 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2260 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2261 	}
2262 
2263 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2264 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2265 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2266 		ibdev->ib_dev.uverbs_cmd_mask |=
2267 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2268 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2269 	}
2270 
2271 	if (check_flow_steering_support(dev)) {
2272 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2273 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2274 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2275 
2276 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2277 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2278 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2279 	}
2280 
2281 	mlx4_ib_alloc_eqs(dev, ibdev);
2282 
2283 	spin_lock_init(&iboe->lock);
2284 
2285 	if (init_node_data(ibdev))
2286 		goto err_map;
2287 
2288 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2289 	for (i = 0; i < num_req_counters; ++i) {
2290 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2291 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2292 						IB_LINK_LAYER_ETHERNET) {
2293 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2294 			if (err)
2295 				ibdev->counters[i] = -1;
2296 		} else {
2297 			ibdev->counters[i] = -1;
2298 		}
2299 	}
2300 	if (mlx4_is_bonded(dev))
2301 		for (i = 1; i < ibdev->num_ports ; ++i)
2302 			ibdev->counters[i] = ibdev->counters[0];
2303 
2304 
2305 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2306 		ib_num_ports++;
2307 
2308 	spin_lock_init(&ibdev->sm_lock);
2309 	mutex_init(&ibdev->cap_mask_mutex);
2310 	INIT_LIST_HEAD(&ibdev->qp_list);
2311 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2312 
2313 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2314 	    ib_num_ports) {
2315 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2316 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2317 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2318 					    &ibdev->steer_qpn_base, 0);
2319 		if (err)
2320 			goto err_counter;
2321 
2322 		ibdev->ib_uc_qpns_bitmap =
2323 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2324 				sizeof(long),
2325 				GFP_KERNEL);
2326 		if (!ibdev->ib_uc_qpns_bitmap) {
2327 			dev_err(&dev->persist->pdev->dev,
2328 				"bit map alloc failed\n");
2329 			goto err_steer_qp_release;
2330 		}
2331 
2332 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2333 
2334 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2335 				dev, ibdev->steer_qpn_base,
2336 				ibdev->steer_qpn_base +
2337 				ibdev->steer_qpn_count - 1);
2338 		if (err)
2339 			goto err_steer_free_bitmap;
2340 	}
2341 
2342 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2343 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2344 
2345 	if (ib_register_device(&ibdev->ib_dev, NULL))
2346 		goto err_steer_free_bitmap;
2347 
2348 	if (mlx4_ib_mad_init(ibdev))
2349 		goto err_reg;
2350 
2351 	if (mlx4_ib_init_sriov(ibdev))
2352 		goto err_mad;
2353 
2354 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2355 		if (!iboe->nb.notifier_call) {
2356 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2357 			err = register_netdevice_notifier(&iboe->nb);
2358 			if (err) {
2359 				iboe->nb.notifier_call = NULL;
2360 				goto err_notif;
2361 			}
2362 		}
2363 		if (!iboe->nb_inet.notifier_call) {
2364 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2365 			err = register_inetaddr_notifier(&iboe->nb_inet);
2366 			if (err) {
2367 				iboe->nb_inet.notifier_call = NULL;
2368 				goto err_notif;
2369 			}
2370 		}
2371 #if IS_ENABLED(CONFIG_IPV6)
2372 		if (!iboe->nb_inet6.notifier_call) {
2373 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2374 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2375 			if (err) {
2376 				iboe->nb_inet6.notifier_call = NULL;
2377 				goto err_notif;
2378 			}
2379 		}
2380 #endif
2381 		if (mlx4_ib_init_gid_table(ibdev))
2382 			goto err_notif;
2383 	}
2384 
2385 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2386 		if (device_create_file(&ibdev->ib_dev.dev,
2387 				       mlx4_class_attributes[j]))
2388 			goto err_notif;
2389 	}
2390 
2391 	ibdev->ib_active = true;
2392 
2393 	if (mlx4_is_mfunc(ibdev->dev))
2394 		init_pkeys(ibdev);
2395 
2396 	/* create paravirt contexts for any VFs which are active */
2397 	if (mlx4_is_master(ibdev->dev)) {
2398 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2399 			if (j == mlx4_master_func_num(ibdev->dev))
2400 				continue;
2401 			if (mlx4_is_slave_active(ibdev->dev, j))
2402 				do_slave_init(ibdev, j, 1);
2403 		}
2404 	}
2405 	return ibdev;
2406 
2407 err_notif:
2408 	if (ibdev->iboe.nb.notifier_call) {
2409 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2410 			pr_warn("failure unregistering notifier\n");
2411 		ibdev->iboe.nb.notifier_call = NULL;
2412 	}
2413 	if (ibdev->iboe.nb_inet.notifier_call) {
2414 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2415 			pr_warn("failure unregistering notifier\n");
2416 		ibdev->iboe.nb_inet.notifier_call = NULL;
2417 	}
2418 #if IS_ENABLED(CONFIG_IPV6)
2419 	if (ibdev->iboe.nb_inet6.notifier_call) {
2420 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2421 			pr_warn("failure unregistering notifier\n");
2422 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2423 	}
2424 #endif
2425 	flush_workqueue(wq);
2426 
2427 	mlx4_ib_close_sriov(ibdev);
2428 
2429 err_mad:
2430 	mlx4_ib_mad_cleanup(ibdev);
2431 
2432 err_reg:
2433 	ib_unregister_device(&ibdev->ib_dev);
2434 
2435 err_steer_free_bitmap:
2436 	kfree(ibdev->ib_uc_qpns_bitmap);
2437 
2438 err_steer_qp_release:
2439 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2440 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2441 				      ibdev->steer_qpn_count);
2442 err_counter:
2443 	for (; i; --i)
2444 		if (ibdev->counters[i - 1] != -1)
2445 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2446 
2447 err_map:
2448 	iounmap(ibdev->uar_map);
2449 
2450 err_uar:
2451 	mlx4_uar_free(dev, &ibdev->priv_uar);
2452 
2453 err_pd:
2454 	mlx4_pd_free(dev, ibdev->priv_pdn);
2455 
2456 err_dealloc:
2457 	ib_dealloc_device(&ibdev->ib_dev);
2458 
2459 	return NULL;
2460 }
2461 
2462 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2463 {
2464 	int offset;
2465 
2466 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2467 
2468 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2469 					 dev->steer_qpn_count,
2470 					 get_count_order(count));
2471 	if (offset < 0)
2472 		return offset;
2473 
2474 	*qpn = dev->steer_qpn_base + offset;
2475 	return 0;
2476 }
2477 
2478 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2479 {
2480 	if (!qpn ||
2481 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2482 		return;
2483 
2484 	BUG_ON(qpn < dev->steer_qpn_base);
2485 
2486 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2487 			      qpn - dev->steer_qpn_base,
2488 			      get_count_order(count));
2489 }
2490 
2491 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2492 			 int is_attach)
2493 {
2494 	int err;
2495 	size_t flow_size;
2496 	struct ib_flow_attr *flow = NULL;
2497 	struct ib_flow_spec_ib *ib_spec;
2498 
2499 	if (is_attach) {
2500 		flow_size = sizeof(struct ib_flow_attr) +
2501 			    sizeof(struct ib_flow_spec_ib);
2502 		flow = kzalloc(flow_size, GFP_KERNEL);
2503 		if (!flow)
2504 			return -ENOMEM;
2505 		flow->port = mqp->port;
2506 		flow->num_of_specs = 1;
2507 		flow->size = flow_size;
2508 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2509 		ib_spec->type = IB_FLOW_SPEC_IB;
2510 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2511 		/* Add an empty rule for IB L2 */
2512 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2513 
2514 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2515 					    IB_FLOW_DOMAIN_NIC,
2516 					    MLX4_FS_REGULAR,
2517 					    &mqp->reg_id);
2518 	} else {
2519 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2520 	}
2521 	kfree(flow);
2522 	return err;
2523 }
2524 
2525 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2526 {
2527 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2528 	int p;
2529 
2530 	ibdev->ib_active = false;
2531 	flush_workqueue(wq);
2532 
2533 	mlx4_ib_close_sriov(ibdev);
2534 	mlx4_ib_mad_cleanup(ibdev);
2535 	ib_unregister_device(&ibdev->ib_dev);
2536 	if (ibdev->iboe.nb.notifier_call) {
2537 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2538 			pr_warn("failure unregistering notifier\n");
2539 		ibdev->iboe.nb.notifier_call = NULL;
2540 	}
2541 
2542 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2543 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2544 				      ibdev->steer_qpn_count);
2545 		kfree(ibdev->ib_uc_qpns_bitmap);
2546 	}
2547 
2548 	if (ibdev->iboe.nb_inet.notifier_call) {
2549 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2550 			pr_warn("failure unregistering notifier\n");
2551 		ibdev->iboe.nb_inet.notifier_call = NULL;
2552 	}
2553 #if IS_ENABLED(CONFIG_IPV6)
2554 	if (ibdev->iboe.nb_inet6.notifier_call) {
2555 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2556 			pr_warn("failure unregistering notifier\n");
2557 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2558 	}
2559 #endif
2560 
2561 	iounmap(ibdev->uar_map);
2562 	for (p = 0; p < ibdev->num_ports; ++p)
2563 		if (ibdev->counters[p] != -1)
2564 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2565 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2566 		mlx4_CLOSE_PORT(dev, p);
2567 
2568 	mlx4_ib_free_eqs(dev, ibdev);
2569 
2570 	mlx4_uar_free(dev, &ibdev->priv_uar);
2571 	mlx4_pd_free(dev, ibdev->priv_pdn);
2572 	ib_dealloc_device(&ibdev->ib_dev);
2573 }
2574 
2575 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2576 {
2577 	struct mlx4_ib_demux_work **dm = NULL;
2578 	struct mlx4_dev *dev = ibdev->dev;
2579 	int i;
2580 	unsigned long flags;
2581 	struct mlx4_active_ports actv_ports;
2582 	unsigned int ports;
2583 	unsigned int first_port;
2584 
2585 	if (!mlx4_is_master(dev))
2586 		return;
2587 
2588 	actv_ports = mlx4_get_active_ports(dev, slave);
2589 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2590 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2591 
2592 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2593 	if (!dm) {
2594 		pr_err("failed to allocate memory for tunneling qp update\n");
2595 		goto out;
2596 	}
2597 
2598 	for (i = 0; i < ports; i++) {
2599 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2600 		if (!dm[i]) {
2601 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2602 			for (i = 0; i < dev->caps.num_ports; i++) {
2603 				if (dm[i])
2604 					kfree(dm[i]);
2605 			}
2606 			goto out;
2607 		}
2608 	}
2609 	/* initialize or tear down tunnel QPs for the slave */
2610 	for (i = 0; i < ports; i++) {
2611 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2612 		dm[i]->port = first_port + i + 1;
2613 		dm[i]->slave = slave;
2614 		dm[i]->do_init = do_init;
2615 		dm[i]->dev = ibdev;
2616 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2617 		if (!ibdev->sriov.is_going_down)
2618 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2619 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2620 	}
2621 out:
2622 	kfree(dm);
2623 	return;
2624 }
2625 
2626 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2627 {
2628 	struct mlx4_ib_qp *mqp;
2629 	unsigned long flags_qp;
2630 	unsigned long flags_cq;
2631 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2632 	struct list_head    cq_notify_list;
2633 	struct mlx4_cq *mcq;
2634 	unsigned long flags;
2635 
2636 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2637 	INIT_LIST_HEAD(&cq_notify_list);
2638 
2639 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2640 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2641 
2642 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2643 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2644 		if (mqp->sq.tail != mqp->sq.head) {
2645 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2646 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2647 			if (send_mcq->mcq.comp &&
2648 			    mqp->ibqp.send_cq->comp_handler) {
2649 				if (!send_mcq->mcq.reset_notify_added) {
2650 					send_mcq->mcq.reset_notify_added = 1;
2651 					list_add_tail(&send_mcq->mcq.reset_notify,
2652 						      &cq_notify_list);
2653 				}
2654 			}
2655 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2656 		}
2657 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2658 		/* Now, handle the QP's receive queue */
2659 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2660 		/* no handling is needed for SRQ */
2661 		if (!mqp->ibqp.srq) {
2662 			if (mqp->rq.tail != mqp->rq.head) {
2663 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2664 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2665 				if (recv_mcq->mcq.comp &&
2666 				    mqp->ibqp.recv_cq->comp_handler) {
2667 					if (!recv_mcq->mcq.reset_notify_added) {
2668 						recv_mcq->mcq.reset_notify_added = 1;
2669 						list_add_tail(&recv_mcq->mcq.reset_notify,
2670 							      &cq_notify_list);
2671 					}
2672 				}
2673 				spin_unlock_irqrestore(&recv_mcq->lock,
2674 						       flags_cq);
2675 			}
2676 		}
2677 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2678 	}
2679 
2680 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2681 		mcq->comp(mcq);
2682 	}
2683 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2684 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2685 }
2686 
2687 static void handle_bonded_port_state_event(struct work_struct *work)
2688 {
2689 	struct ib_event_work *ew =
2690 		container_of(work, struct ib_event_work, work);
2691 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2692 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2693 	int i;
2694 	struct ib_event ibev;
2695 
2696 	kfree(ew);
2697 	spin_lock_bh(&ibdev->iboe.lock);
2698 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2699 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2700 		enum ib_port_state curr_port_state;
2701 
2702 		if (!curr_netdev)
2703 			continue;
2704 
2705 		curr_port_state =
2706 			(netif_running(curr_netdev) &&
2707 			 netif_carrier_ok(curr_netdev)) ?
2708 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2709 
2710 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2711 			curr_port_state : IB_PORT_ACTIVE;
2712 	}
2713 	spin_unlock_bh(&ibdev->iboe.lock);
2714 
2715 	ibev.device = &ibdev->ib_dev;
2716 	ibev.element.port_num = 1;
2717 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2718 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2719 
2720 	ib_dispatch_event(&ibev);
2721 }
2722 
2723 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2724 			  enum mlx4_dev_event event, unsigned long param)
2725 {
2726 	struct ib_event ibev;
2727 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2728 	struct mlx4_eqe *eqe = NULL;
2729 	struct ib_event_work *ew;
2730 	int p = 0;
2731 
2732 	if (mlx4_is_bonded(dev) &&
2733 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2734 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2735 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2736 		if (!ew)
2737 			return;
2738 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2739 		ew->ib_dev = ibdev;
2740 		queue_work(wq, &ew->work);
2741 		return;
2742 	}
2743 
2744 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2745 		eqe = (struct mlx4_eqe *)param;
2746 	else
2747 		p = (int) param;
2748 
2749 	switch (event) {
2750 	case MLX4_DEV_EVENT_PORT_UP:
2751 		if (p > ibdev->num_ports)
2752 			return;
2753 		if (mlx4_is_master(dev) &&
2754 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2755 			IB_LINK_LAYER_INFINIBAND) {
2756 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2757 		}
2758 		ibev.event = IB_EVENT_PORT_ACTIVE;
2759 		break;
2760 
2761 	case MLX4_DEV_EVENT_PORT_DOWN:
2762 		if (p > ibdev->num_ports)
2763 			return;
2764 		ibev.event = IB_EVENT_PORT_ERR;
2765 		break;
2766 
2767 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2768 		ibdev->ib_active = false;
2769 		ibev.event = IB_EVENT_DEVICE_FATAL;
2770 		mlx4_ib_handle_catas_error(ibdev);
2771 		break;
2772 
2773 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2774 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2775 		if (!ew) {
2776 			pr_err("failed to allocate memory for events work\n");
2777 			break;
2778 		}
2779 
2780 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2781 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2782 		ew->ib_dev = ibdev;
2783 		/* need to queue only for port owner, which uses GEN_EQE */
2784 		if (mlx4_is_master(dev))
2785 			queue_work(wq, &ew->work);
2786 		else
2787 			handle_port_mgmt_change_event(&ew->work);
2788 		return;
2789 
2790 	case MLX4_DEV_EVENT_SLAVE_INIT:
2791 		/* here, p is the slave id */
2792 		do_slave_init(ibdev, p, 1);
2793 		return;
2794 
2795 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2796 		/* here, p is the slave id */
2797 		do_slave_init(ibdev, p, 0);
2798 		return;
2799 
2800 	default:
2801 		return;
2802 	}
2803 
2804 	ibev.device	      = ibdev_ptr;
2805 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2806 
2807 	ib_dispatch_event(&ibev);
2808 }
2809 
2810 static struct mlx4_interface mlx4_ib_interface = {
2811 	.add		= mlx4_ib_add,
2812 	.remove		= mlx4_ib_remove,
2813 	.event		= mlx4_ib_event,
2814 	.protocol	= MLX4_PROT_IB_IPV6,
2815 	.flags		= MLX4_INTFF_BONDING
2816 };
2817 
2818 static int __init mlx4_ib_init(void)
2819 {
2820 	int err;
2821 
2822 	wq = create_singlethread_workqueue("mlx4_ib");
2823 	if (!wq)
2824 		return -ENOMEM;
2825 
2826 	err = mlx4_ib_mcg_init();
2827 	if (err)
2828 		goto clean_wq;
2829 
2830 	err = mlx4_register_interface(&mlx4_ib_interface);
2831 	if (err)
2832 		goto clean_mcg;
2833 
2834 	return 0;
2835 
2836 clean_mcg:
2837 	mlx4_ib_mcg_destroy();
2838 
2839 clean_wq:
2840 	destroy_workqueue(wq);
2841 	return err;
2842 }
2843 
2844 static void __exit mlx4_ib_cleanup(void)
2845 {
2846 	mlx4_unregister_interface(&mlx4_ib_interface);
2847 	mlx4_ib_mcg_destroy();
2848 	destroy_workqueue(wq);
2849 }
2850 
2851 module_init(mlx4_ib_init);
2852 module_exit(mlx4_ib_cleanup);
2853