xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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->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 
355 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
356 	if (IS_ERR(mailbox))
357 		return PTR_ERR(mailbox);
358 
359 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
360 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
361 			   MLX4_CMD_WRAPPED);
362 	if (err)
363 		goto out;
364 
365 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
366 						IB_WIDTH_4X : IB_WIDTH_1X;
367 	props->active_speed	= IB_SPEED_QDR;
368 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
369 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
370 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
371 	props->pkey_tbl_len	= 1;
372 	props->max_mtu		= IB_MTU_4096;
373 	props->max_vl_num	= 2;
374 	props->state		= IB_PORT_DOWN;
375 	props->phys_state	= state_to_phys_state(props->state);
376 	props->active_mtu	= IB_MTU_256;
377 	spin_lock_bh(&iboe->lock);
378 	ndev = iboe->netdevs[port - 1];
379 	if (!ndev)
380 		goto out_unlock;
381 
382 	tmp = iboe_get_mtu(ndev->mtu);
383 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
384 
385 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
386 					IB_PORT_ACTIVE : IB_PORT_DOWN;
387 	props->phys_state	= state_to_phys_state(props->state);
388 out_unlock:
389 	spin_unlock_bh(&iboe->lock);
390 out:
391 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
392 	return err;
393 }
394 
395 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
396 			 struct ib_port_attr *props, int netw_view)
397 {
398 	int err;
399 
400 	memset(props, 0, sizeof *props);
401 
402 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
403 		ib_link_query_port(ibdev, port, props, netw_view) :
404 				eth_link_query_port(ibdev, port, props, netw_view);
405 
406 	return err;
407 }
408 
409 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
410 			      struct ib_port_attr *props)
411 {
412 	/* returns host view */
413 	return __mlx4_ib_query_port(ibdev, port, props, 0);
414 }
415 
416 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
417 			union ib_gid *gid, int netw_view)
418 {
419 	struct ib_smp *in_mad  = NULL;
420 	struct ib_smp *out_mad = NULL;
421 	int err = -ENOMEM;
422 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
423 	int clear = 0;
424 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
425 
426 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
427 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
428 	if (!in_mad || !out_mad)
429 		goto out;
430 
431 	init_query_mad(in_mad);
432 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
433 	in_mad->attr_mod = cpu_to_be32(port);
434 
435 	if (mlx4_is_mfunc(dev->dev) && netw_view)
436 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
437 
438 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
439 	if (err)
440 		goto out;
441 
442 	memcpy(gid->raw, out_mad->data + 8, 8);
443 
444 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
445 		if (index) {
446 			/* For any index > 0, return the null guid */
447 			err = 0;
448 			clear = 1;
449 			goto out;
450 		}
451 	}
452 
453 	init_query_mad(in_mad);
454 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
455 	in_mad->attr_mod = cpu_to_be32(index / 8);
456 
457 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
458 			   NULL, NULL, in_mad, out_mad);
459 	if (err)
460 		goto out;
461 
462 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
463 
464 out:
465 	if (clear)
466 		memset(gid->raw + 8, 0, 8);
467 	kfree(in_mad);
468 	kfree(out_mad);
469 	return err;
470 }
471 
472 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
473 			  union ib_gid *gid)
474 {
475 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
476 
477 	*gid = dev->iboe.gid_table[port - 1][index];
478 
479 	return 0;
480 }
481 
482 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
483 			     union ib_gid *gid)
484 {
485 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
486 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
487 	else
488 		return iboe_query_gid(ibdev, port, index, gid);
489 }
490 
491 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
492 			 u16 *pkey, int netw_view)
493 {
494 	struct ib_smp *in_mad  = NULL;
495 	struct ib_smp *out_mad = NULL;
496 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
497 	int err = -ENOMEM;
498 
499 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
500 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
501 	if (!in_mad || !out_mad)
502 		goto out;
503 
504 	init_query_mad(in_mad);
505 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
506 	in_mad->attr_mod = cpu_to_be32(index / 32);
507 
508 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
509 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
510 
511 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
512 			   in_mad, out_mad);
513 	if (err)
514 		goto out;
515 
516 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
517 
518 out:
519 	kfree(in_mad);
520 	kfree(out_mad);
521 	return err;
522 }
523 
524 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
525 {
526 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
527 }
528 
529 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
530 				 struct ib_device_modify *props)
531 {
532 	struct mlx4_cmd_mailbox *mailbox;
533 	unsigned long flags;
534 
535 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
536 		return -EOPNOTSUPP;
537 
538 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
539 		return 0;
540 
541 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
542 		return -EOPNOTSUPP;
543 
544 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
545 	memcpy(ibdev->node_desc, props->node_desc, 64);
546 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
547 
548 	/*
549 	 * If possible, pass node desc to FW, so it can generate
550 	 * a 144 trap.  If cmd fails, just ignore.
551 	 */
552 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
553 	if (IS_ERR(mailbox))
554 		return 0;
555 
556 	memcpy(mailbox->buf, props->node_desc, 64);
557 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
558 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
559 
560 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
561 
562 	return 0;
563 }
564 
565 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
566 			    u32 cap_mask)
567 {
568 	struct mlx4_cmd_mailbox *mailbox;
569 	int err;
570 
571 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
572 	if (IS_ERR(mailbox))
573 		return PTR_ERR(mailbox);
574 
575 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
576 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
577 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
578 	} else {
579 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
580 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
581 	}
582 
583 	err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
584 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
585 
586 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
587 	return err;
588 }
589 
590 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
591 			       struct ib_port_modify *props)
592 {
593 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
594 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
595 	struct ib_port_attr attr;
596 	u32 cap_mask;
597 	int err;
598 
599 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
600 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
601 	 * violations and port capabilities are not meaningful.
602 	 */
603 	if (is_eth)
604 		return 0;
605 
606 	mutex_lock(&mdev->cap_mask_mutex);
607 
608 	err = mlx4_ib_query_port(ibdev, port, &attr);
609 	if (err)
610 		goto out;
611 
612 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
613 		~props->clr_port_cap_mask;
614 
615 	err = mlx4_ib_SET_PORT(mdev, port,
616 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
617 			       cap_mask);
618 
619 out:
620 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
621 	return err;
622 }
623 
624 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
625 						  struct ib_udata *udata)
626 {
627 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
628 	struct mlx4_ib_ucontext *context;
629 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
630 	struct mlx4_ib_alloc_ucontext_resp resp;
631 	int err;
632 
633 	if (!dev->ib_active)
634 		return ERR_PTR(-EAGAIN);
635 
636 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
637 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
638 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
639 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
640 	} else {
641 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
642 		resp.qp_tab_size      = dev->dev->caps.num_qps;
643 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
644 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
645 		resp.cqe_size	      = dev->dev->caps.cqe_size;
646 	}
647 
648 	context = kmalloc(sizeof *context, GFP_KERNEL);
649 	if (!context)
650 		return ERR_PTR(-ENOMEM);
651 
652 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
653 	if (err) {
654 		kfree(context);
655 		return ERR_PTR(err);
656 	}
657 
658 	INIT_LIST_HEAD(&context->db_page_list);
659 	mutex_init(&context->db_page_mutex);
660 
661 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
662 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
663 	else
664 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
665 
666 	if (err) {
667 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
668 		kfree(context);
669 		return ERR_PTR(-EFAULT);
670 	}
671 
672 	return &context->ibucontext;
673 }
674 
675 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
676 {
677 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
678 
679 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
680 	kfree(context);
681 
682 	return 0;
683 }
684 
685 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
686 {
687 	struct mlx4_ib_dev *dev = to_mdev(context->device);
688 
689 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
690 		return -EINVAL;
691 
692 	if (vma->vm_pgoff == 0) {
693 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
694 
695 		if (io_remap_pfn_range(vma, vma->vm_start,
696 				       to_mucontext(context)->uar.pfn,
697 				       PAGE_SIZE, vma->vm_page_prot))
698 			return -EAGAIN;
699 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
700 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
701 
702 		if (io_remap_pfn_range(vma, vma->vm_start,
703 				       to_mucontext(context)->uar.pfn +
704 				       dev->dev->caps.num_uars,
705 				       PAGE_SIZE, vma->vm_page_prot))
706 			return -EAGAIN;
707 	} else
708 		return -EINVAL;
709 
710 	return 0;
711 }
712 
713 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
714 				      struct ib_ucontext *context,
715 				      struct ib_udata *udata)
716 {
717 	struct mlx4_ib_pd *pd;
718 	int err;
719 
720 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
721 	if (!pd)
722 		return ERR_PTR(-ENOMEM);
723 
724 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
725 	if (err) {
726 		kfree(pd);
727 		return ERR_PTR(err);
728 	}
729 
730 	if (context)
731 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
732 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
733 			kfree(pd);
734 			return ERR_PTR(-EFAULT);
735 		}
736 
737 	return &pd->ibpd;
738 }
739 
740 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
741 {
742 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
743 	kfree(pd);
744 
745 	return 0;
746 }
747 
748 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
749 					  struct ib_ucontext *context,
750 					  struct ib_udata *udata)
751 {
752 	struct mlx4_ib_xrcd *xrcd;
753 	int err;
754 
755 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
756 		return ERR_PTR(-ENOSYS);
757 
758 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
759 	if (!xrcd)
760 		return ERR_PTR(-ENOMEM);
761 
762 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
763 	if (err)
764 		goto err1;
765 
766 	xrcd->pd = ib_alloc_pd(ibdev);
767 	if (IS_ERR(xrcd->pd)) {
768 		err = PTR_ERR(xrcd->pd);
769 		goto err2;
770 	}
771 
772 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
773 	if (IS_ERR(xrcd->cq)) {
774 		err = PTR_ERR(xrcd->cq);
775 		goto err3;
776 	}
777 
778 	return &xrcd->ibxrcd;
779 
780 err3:
781 	ib_dealloc_pd(xrcd->pd);
782 err2:
783 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
784 err1:
785 	kfree(xrcd);
786 	return ERR_PTR(err);
787 }
788 
789 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
790 {
791 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
792 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
793 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
794 	kfree(xrcd);
795 
796 	return 0;
797 }
798 
799 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
800 {
801 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
802 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
803 	struct mlx4_ib_gid_entry *ge;
804 
805 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
806 	if (!ge)
807 		return -ENOMEM;
808 
809 	ge->gid = *gid;
810 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
811 		ge->port = mqp->port;
812 		ge->added = 1;
813 	}
814 
815 	mutex_lock(&mqp->mutex);
816 	list_add_tail(&ge->list, &mqp->gid_list);
817 	mutex_unlock(&mqp->mutex);
818 
819 	return 0;
820 }
821 
822 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
823 		   union ib_gid *gid)
824 {
825 	struct net_device *ndev;
826 	int ret = 0;
827 
828 	if (!mqp->port)
829 		return 0;
830 
831 	spin_lock_bh(&mdev->iboe.lock);
832 	ndev = mdev->iboe.netdevs[mqp->port - 1];
833 	if (ndev)
834 		dev_hold(ndev);
835 	spin_unlock_bh(&mdev->iboe.lock);
836 
837 	if (ndev) {
838 		ret = 1;
839 		dev_put(ndev);
840 	}
841 
842 	return ret;
843 }
844 
845 struct mlx4_ib_steering {
846 	struct list_head list;
847 	u64 reg_id;
848 	union ib_gid gid;
849 };
850 
851 static int parse_flow_attr(struct mlx4_dev *dev,
852 			   u32 qp_num,
853 			   union ib_flow_spec *ib_spec,
854 			   struct _rule_hw *mlx4_spec)
855 {
856 	enum mlx4_net_trans_rule_id type;
857 
858 	switch (ib_spec->type) {
859 	case IB_FLOW_SPEC_ETH:
860 		type = MLX4_NET_TRANS_RULE_ID_ETH;
861 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
862 		       ETH_ALEN);
863 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
864 		       ETH_ALEN);
865 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
866 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
867 		break;
868 	case IB_FLOW_SPEC_IB:
869 		type = MLX4_NET_TRANS_RULE_ID_IB;
870 		mlx4_spec->ib.l3_qpn =
871 			cpu_to_be32(qp_num);
872 		mlx4_spec->ib.qpn_mask =
873 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
874 		break;
875 
876 
877 	case IB_FLOW_SPEC_IPV4:
878 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
879 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
880 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
881 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
882 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
883 		break;
884 
885 	case IB_FLOW_SPEC_TCP:
886 	case IB_FLOW_SPEC_UDP:
887 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
888 					MLX4_NET_TRANS_RULE_ID_TCP :
889 					MLX4_NET_TRANS_RULE_ID_UDP;
890 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
891 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
892 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
893 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
894 		break;
895 
896 	default:
897 		return -EINVAL;
898 	}
899 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
900 	    mlx4_hw_rule_sz(dev, type) < 0)
901 		return -EINVAL;
902 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
903 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
904 	return mlx4_hw_rule_sz(dev, type);
905 }
906 
907 struct default_rules {
908 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
909 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
910 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
911 	__u8  link_layer;
912 };
913 static const struct default_rules default_table[] = {
914 	{
915 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
916 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
917 		.rules_create_list = {IB_FLOW_SPEC_IB},
918 		.link_layer = IB_LINK_LAYER_INFINIBAND
919 	}
920 };
921 
922 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
923 					 struct ib_flow_attr *flow_attr)
924 {
925 	int i, j, k;
926 	void *ib_flow;
927 	const struct default_rules *pdefault_rules = default_table;
928 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
929 
930 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
931 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
932 		memset(&field_types, 0, sizeof(field_types));
933 
934 		if (link_layer != pdefault_rules->link_layer)
935 			continue;
936 
937 		ib_flow = flow_attr + 1;
938 		/* we assume the specs are sorted */
939 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
940 		     j < flow_attr->num_of_specs; k++) {
941 			union ib_flow_spec *current_flow =
942 				(union ib_flow_spec *)ib_flow;
943 
944 			/* same layer but different type */
945 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
946 			     (pdefault_rules->mandatory_fields[k] &
947 			      IB_FLOW_SPEC_LAYER_MASK)) &&
948 			    (current_flow->type !=
949 			     pdefault_rules->mandatory_fields[k]))
950 				goto out;
951 
952 			/* same layer, try match next one */
953 			if (current_flow->type ==
954 			    pdefault_rules->mandatory_fields[k]) {
955 				j++;
956 				ib_flow +=
957 					((union ib_flow_spec *)ib_flow)->size;
958 			}
959 		}
960 
961 		ib_flow = flow_attr + 1;
962 		for (j = 0; j < flow_attr->num_of_specs;
963 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
964 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
965 				/* same layer and same type */
966 				if (((union ib_flow_spec *)ib_flow)->type ==
967 				    pdefault_rules->mandatory_not_fields[k])
968 					goto out;
969 
970 		return i;
971 	}
972 out:
973 	return -1;
974 }
975 
976 static int __mlx4_ib_create_default_rules(
977 		struct mlx4_ib_dev *mdev,
978 		struct ib_qp *qp,
979 		const struct default_rules *pdefault_rules,
980 		struct _rule_hw *mlx4_spec) {
981 	int size = 0;
982 	int i;
983 
984 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
985 		int ret;
986 		union ib_flow_spec ib_spec;
987 		switch (pdefault_rules->rules_create_list[i]) {
988 		case 0:
989 			/* no rule */
990 			continue;
991 		case IB_FLOW_SPEC_IB:
992 			ib_spec.type = IB_FLOW_SPEC_IB;
993 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
994 
995 			break;
996 		default:
997 			/* invalid rule */
998 			return -EINVAL;
999 		}
1000 		/* We must put empty rule, qpn is being ignored */
1001 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1002 				      mlx4_spec);
1003 		if (ret < 0) {
1004 			pr_info("invalid parsing\n");
1005 			return -EINVAL;
1006 		}
1007 
1008 		mlx4_spec = (void *)mlx4_spec + ret;
1009 		size += ret;
1010 	}
1011 	return size;
1012 }
1013 
1014 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1015 			  int domain,
1016 			  enum mlx4_net_trans_promisc_mode flow_type,
1017 			  u64 *reg_id)
1018 {
1019 	int ret, i;
1020 	int size = 0;
1021 	void *ib_flow;
1022 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1023 	struct mlx4_cmd_mailbox *mailbox;
1024 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1025 	int default_flow;
1026 
1027 	static const u16 __mlx4_domain[] = {
1028 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1029 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1030 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1031 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1032 	};
1033 
1034 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1035 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1036 		return -EINVAL;
1037 	}
1038 
1039 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1040 		pr_err("Invalid domain value %d\n", domain);
1041 		return -EINVAL;
1042 	}
1043 
1044 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1045 		return -EINVAL;
1046 
1047 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1048 	if (IS_ERR(mailbox))
1049 		return PTR_ERR(mailbox);
1050 	ctrl = mailbox->buf;
1051 
1052 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1053 				 flow_attr->priority);
1054 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1055 	ctrl->port = flow_attr->port;
1056 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1057 
1058 	ib_flow = flow_attr + 1;
1059 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1060 	/* Add default flows */
1061 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1062 	if (default_flow >= 0) {
1063 		ret = __mlx4_ib_create_default_rules(
1064 				mdev, qp, default_table + default_flow,
1065 				mailbox->buf + size);
1066 		if (ret < 0) {
1067 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1068 			return -EINVAL;
1069 		}
1070 		size += ret;
1071 	}
1072 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1073 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1074 				      mailbox->buf + size);
1075 		if (ret < 0) {
1076 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1077 			return -EINVAL;
1078 		}
1079 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1080 		size += ret;
1081 	}
1082 
1083 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1084 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1085 			   MLX4_CMD_NATIVE);
1086 	if (ret == -ENOMEM)
1087 		pr_err("mcg table is full. Fail to register network rule.\n");
1088 	else if (ret == -ENXIO)
1089 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1090 	else if (ret)
1091 		pr_err("Invalid argumant. Fail to register network rule.\n");
1092 
1093 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1094 	return ret;
1095 }
1096 
1097 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1098 {
1099 	int err;
1100 	err = mlx4_cmd(dev, reg_id, 0, 0,
1101 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1102 		       MLX4_CMD_NATIVE);
1103 	if (err)
1104 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1105 		       reg_id);
1106 	return err;
1107 }
1108 
1109 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1110 				    u64 *reg_id)
1111 {
1112 	void *ib_flow;
1113 	union ib_flow_spec *ib_spec;
1114 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1115 	int err = 0;
1116 
1117 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1118 		return 0; /* do nothing */
1119 
1120 	ib_flow = flow_attr + 1;
1121 	ib_spec = (union ib_flow_spec *)ib_flow;
1122 
1123 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1124 		return 0; /* do nothing */
1125 
1126 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1127 				    flow_attr->port, qp->qp_num,
1128 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1129 				    reg_id);
1130 	return err;
1131 }
1132 
1133 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1134 				    struct ib_flow_attr *flow_attr,
1135 				    int domain)
1136 {
1137 	int err = 0, i = 0;
1138 	struct mlx4_ib_flow *mflow;
1139 	enum mlx4_net_trans_promisc_mode type[2];
1140 
1141 	memset(type, 0, sizeof(type));
1142 
1143 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1144 	if (!mflow) {
1145 		err = -ENOMEM;
1146 		goto err_free;
1147 	}
1148 
1149 	switch (flow_attr->type) {
1150 	case IB_FLOW_ATTR_NORMAL:
1151 		type[0] = MLX4_FS_REGULAR;
1152 		break;
1153 
1154 	case IB_FLOW_ATTR_ALL_DEFAULT:
1155 		type[0] = MLX4_FS_ALL_DEFAULT;
1156 		break;
1157 
1158 	case IB_FLOW_ATTR_MC_DEFAULT:
1159 		type[0] = MLX4_FS_MC_DEFAULT;
1160 		break;
1161 
1162 	case IB_FLOW_ATTR_SNIFFER:
1163 		type[0] = MLX4_FS_UC_SNIFFER;
1164 		type[1] = MLX4_FS_MC_SNIFFER;
1165 		break;
1166 
1167 	default:
1168 		err = -EINVAL;
1169 		goto err_free;
1170 	}
1171 
1172 	while (i < ARRAY_SIZE(type) && type[i]) {
1173 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1174 					    &mflow->reg_id[i]);
1175 		if (err)
1176 			goto err_free;
1177 		i++;
1178 	}
1179 
1180 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1181 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr, &mflow->reg_id[i]);
1182 		if (err)
1183 			goto err_free;
1184 	}
1185 
1186 	return &mflow->ibflow;
1187 
1188 err_free:
1189 	kfree(mflow);
1190 	return ERR_PTR(err);
1191 }
1192 
1193 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1194 {
1195 	int err, ret = 0;
1196 	int i = 0;
1197 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1198 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1199 
1200 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1201 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1202 		if (err)
1203 			ret = err;
1204 		i++;
1205 	}
1206 
1207 	kfree(mflow);
1208 	return ret;
1209 }
1210 
1211 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1212 {
1213 	int err;
1214 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1215 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1216 	u64 reg_id;
1217 	struct mlx4_ib_steering *ib_steering = NULL;
1218 	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1219 		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1220 
1221 	if (mdev->dev->caps.steering_mode ==
1222 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1223 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1224 		if (!ib_steering)
1225 			return -ENOMEM;
1226 	}
1227 
1228 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1229 				    !!(mqp->flags &
1230 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1231 				    prot, &reg_id);
1232 	if (err)
1233 		goto err_malloc;
1234 
1235 	err = add_gid_entry(ibqp, gid);
1236 	if (err)
1237 		goto err_add;
1238 
1239 	if (ib_steering) {
1240 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1241 		ib_steering->reg_id = reg_id;
1242 		mutex_lock(&mqp->mutex);
1243 		list_add(&ib_steering->list, &mqp->steering_rules);
1244 		mutex_unlock(&mqp->mutex);
1245 	}
1246 	return 0;
1247 
1248 err_add:
1249 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1250 			      prot, reg_id);
1251 err_malloc:
1252 	kfree(ib_steering);
1253 
1254 	return err;
1255 }
1256 
1257 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1258 {
1259 	struct mlx4_ib_gid_entry *ge;
1260 	struct mlx4_ib_gid_entry *tmp;
1261 	struct mlx4_ib_gid_entry *ret = NULL;
1262 
1263 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1264 		if (!memcmp(raw, ge->gid.raw, 16)) {
1265 			ret = ge;
1266 			break;
1267 		}
1268 	}
1269 
1270 	return ret;
1271 }
1272 
1273 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1274 {
1275 	int err;
1276 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1277 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1278 	struct net_device *ndev;
1279 	struct mlx4_ib_gid_entry *ge;
1280 	u64 reg_id = 0;
1281 	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1282 		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1283 
1284 	if (mdev->dev->caps.steering_mode ==
1285 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1286 		struct mlx4_ib_steering *ib_steering;
1287 
1288 		mutex_lock(&mqp->mutex);
1289 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1290 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1291 				list_del(&ib_steering->list);
1292 				break;
1293 			}
1294 		}
1295 		mutex_unlock(&mqp->mutex);
1296 		if (&ib_steering->list == &mqp->steering_rules) {
1297 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1298 			return -EINVAL;
1299 		}
1300 		reg_id = ib_steering->reg_id;
1301 		kfree(ib_steering);
1302 	}
1303 
1304 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1305 				    prot, reg_id);
1306 	if (err)
1307 		return err;
1308 
1309 	mutex_lock(&mqp->mutex);
1310 	ge = find_gid_entry(mqp, gid->raw);
1311 	if (ge) {
1312 		spin_lock_bh(&mdev->iboe.lock);
1313 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1314 		if (ndev)
1315 			dev_hold(ndev);
1316 		spin_unlock_bh(&mdev->iboe.lock);
1317 		if (ndev)
1318 			dev_put(ndev);
1319 		list_del(&ge->list);
1320 		kfree(ge);
1321 	} else
1322 		pr_warn("could not find mgid entry\n");
1323 
1324 	mutex_unlock(&mqp->mutex);
1325 
1326 	return 0;
1327 }
1328 
1329 static int init_node_data(struct mlx4_ib_dev *dev)
1330 {
1331 	struct ib_smp *in_mad  = NULL;
1332 	struct ib_smp *out_mad = NULL;
1333 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1334 	int err = -ENOMEM;
1335 
1336 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1337 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1338 	if (!in_mad || !out_mad)
1339 		goto out;
1340 
1341 	init_query_mad(in_mad);
1342 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1343 	if (mlx4_is_master(dev->dev))
1344 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1345 
1346 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1347 	if (err)
1348 		goto out;
1349 
1350 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1351 
1352 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1353 
1354 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1355 	if (err)
1356 		goto out;
1357 
1358 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1359 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1360 
1361 out:
1362 	kfree(in_mad);
1363 	kfree(out_mad);
1364 	return err;
1365 }
1366 
1367 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1368 			char *buf)
1369 {
1370 	struct mlx4_ib_dev *dev =
1371 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1372 	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1373 }
1374 
1375 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1376 			   char *buf)
1377 {
1378 	struct mlx4_ib_dev *dev =
1379 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1380 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1381 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1382 		       (int) dev->dev->caps.fw_ver & 0xffff);
1383 }
1384 
1385 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1386 			char *buf)
1387 {
1388 	struct mlx4_ib_dev *dev =
1389 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1390 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1391 }
1392 
1393 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1394 			  char *buf)
1395 {
1396 	struct mlx4_ib_dev *dev =
1397 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1398 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1399 		       dev->dev->board_id);
1400 }
1401 
1402 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1403 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1404 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1405 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1406 
1407 static struct device_attribute *mlx4_class_attributes[] = {
1408 	&dev_attr_hw_rev,
1409 	&dev_attr_fw_ver,
1410 	&dev_attr_hca_type,
1411 	&dev_attr_board_id
1412 };
1413 
1414 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1415 				     struct net_device *dev)
1416 {
1417 	memcpy(eui, dev->dev_addr, 3);
1418 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1419 	if (vlan_id < 0x1000) {
1420 		eui[3] = vlan_id >> 8;
1421 		eui[4] = vlan_id & 0xff;
1422 	} else {
1423 		eui[3] = 0xff;
1424 		eui[4] = 0xfe;
1425 	}
1426 	eui[0] ^= 2;
1427 }
1428 
1429 static void update_gids_task(struct work_struct *work)
1430 {
1431 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1432 	struct mlx4_cmd_mailbox *mailbox;
1433 	union ib_gid *gids;
1434 	int err;
1435 	struct mlx4_dev	*dev = gw->dev->dev;
1436 
1437 	if (!gw->dev->ib_active)
1438 		return;
1439 
1440 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1441 	if (IS_ERR(mailbox)) {
1442 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1443 		return;
1444 	}
1445 
1446 	gids = mailbox->buf;
1447 	memcpy(gids, gw->gids, sizeof gw->gids);
1448 
1449 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1450 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1451 		       MLX4_CMD_WRAPPED);
1452 	if (err)
1453 		pr_warn("set port command failed\n");
1454 	else
1455 		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1456 
1457 	mlx4_free_cmd_mailbox(dev, mailbox);
1458 	kfree(gw);
1459 }
1460 
1461 static void reset_gids_task(struct work_struct *work)
1462 {
1463 	struct update_gid_work *gw =
1464 			container_of(work, struct update_gid_work, work);
1465 	struct mlx4_cmd_mailbox *mailbox;
1466 	union ib_gid *gids;
1467 	int err;
1468 	struct mlx4_dev	*dev = gw->dev->dev;
1469 
1470 	if (!gw->dev->ib_active)
1471 		return;
1472 
1473 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1474 	if (IS_ERR(mailbox)) {
1475 		pr_warn("reset gid table failed\n");
1476 		goto free;
1477 	}
1478 
1479 	gids = mailbox->buf;
1480 	memcpy(gids, gw->gids, sizeof(gw->gids));
1481 
1482 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1483 				    IB_LINK_LAYER_ETHERNET) {
1484 		err = mlx4_cmd(dev, mailbox->dma,
1485 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1486 			       1, MLX4_CMD_SET_PORT,
1487 			       MLX4_CMD_TIME_CLASS_B,
1488 			       MLX4_CMD_WRAPPED);
1489 		if (err)
1490 			pr_warn(KERN_WARNING
1491 				"set port %d command failed\n", gw->port);
1492 	}
1493 
1494 	mlx4_free_cmd_mailbox(dev, mailbox);
1495 free:
1496 	kfree(gw);
1497 }
1498 
1499 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1500 			    union ib_gid *gid, int clear,
1501 			    int default_gid)
1502 {
1503 	struct update_gid_work *work;
1504 	int i;
1505 	int need_update = 0;
1506 	int free = -1;
1507 	int found = -1;
1508 	int max_gids;
1509 
1510 	if (default_gid) {
1511 		free = 0;
1512 	} else {
1513 		max_gids = dev->dev->caps.gid_table_len[port];
1514 		for (i = 1; i < max_gids; ++i) {
1515 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1516 				    sizeof(*gid)))
1517 				found = i;
1518 
1519 			if (clear) {
1520 				if (found >= 0) {
1521 					need_update = 1;
1522 					dev->iboe.gid_table[port - 1][found] =
1523 						zgid;
1524 					break;
1525 				}
1526 			} else {
1527 				if (found >= 0)
1528 					break;
1529 
1530 				if (free < 0 &&
1531 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1532 					    &zgid, sizeof(*gid)))
1533 					free = i;
1534 			}
1535 		}
1536 	}
1537 
1538 	if (found == -1 && !clear && free >= 0) {
1539 		dev->iboe.gid_table[port - 1][free] = *gid;
1540 		need_update = 1;
1541 	}
1542 
1543 	if (!need_update)
1544 		return 0;
1545 
1546 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1547 	if (!work)
1548 		return -ENOMEM;
1549 
1550 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1551 	INIT_WORK(&work->work, update_gids_task);
1552 	work->port = port;
1553 	work->dev = dev;
1554 	queue_work(wq, &work->work);
1555 
1556 	return 0;
1557 }
1558 
1559 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1560 {
1561 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1562 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1563 }
1564 
1565 
1566 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1567 {
1568 	struct update_gid_work *work;
1569 
1570 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1571 	if (!work)
1572 		return -ENOMEM;
1573 
1574 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1575 	memset(work->gids, 0, sizeof(work->gids));
1576 	INIT_WORK(&work->work, reset_gids_task);
1577 	work->dev = dev;
1578 	work->port = port;
1579 	queue_work(wq, &work->work);
1580 	return 0;
1581 }
1582 
1583 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1584 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1585 {
1586 	struct mlx4_ib_iboe *iboe;
1587 	int port = 0;
1588 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1589 				rdma_vlan_dev_real_dev(event_netdev) :
1590 				event_netdev;
1591 	union ib_gid default_gid;
1592 
1593 	mlx4_make_default_gid(real_dev, &default_gid);
1594 
1595 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1596 		return 0;
1597 
1598 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1599 		return 0;
1600 
1601 	if ((real_dev != event_netdev) &&
1602 	    (event == NETDEV_DOWN) &&
1603 	    rdma_link_local_addr((struct in6_addr *)gid))
1604 		return 0;
1605 
1606 	iboe = &ibdev->iboe;
1607 	spin_lock_bh(&iboe->lock);
1608 
1609 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1610 		if ((netif_is_bond_master(real_dev) &&
1611 		     (real_dev == iboe->masters[port - 1])) ||
1612 		     (!netif_is_bond_master(real_dev) &&
1613 		     (real_dev == iboe->netdevs[port - 1])))
1614 			update_gid_table(ibdev, port, gid,
1615 					 event == NETDEV_DOWN, 0);
1616 
1617 	spin_unlock_bh(&iboe->lock);
1618 	return 0;
1619 
1620 }
1621 
1622 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1623 			       struct mlx4_ib_dev *ibdev)
1624 {
1625 	u8 port = 0;
1626 	struct mlx4_ib_iboe *iboe;
1627 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1628 				rdma_vlan_dev_real_dev(dev) : dev;
1629 
1630 	iboe = &ibdev->iboe;
1631 
1632 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1633 		if ((netif_is_bond_master(real_dev) &&
1634 		     (real_dev == iboe->masters[port - 1])) ||
1635 		     (!netif_is_bond_master(real_dev) &&
1636 		     (real_dev == iboe->netdevs[port - 1])))
1637 			break;
1638 
1639 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1640 		return 0;
1641 	else
1642 		return port;
1643 }
1644 
1645 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1646 				void *ptr)
1647 {
1648 	struct mlx4_ib_dev *ibdev;
1649 	struct in_ifaddr *ifa = ptr;
1650 	union ib_gid gid;
1651 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1652 
1653 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1654 
1655 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1656 
1657 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1658 	return NOTIFY_DONE;
1659 }
1660 
1661 #if IS_ENABLED(CONFIG_IPV6)
1662 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1663 				void *ptr)
1664 {
1665 	struct mlx4_ib_dev *ibdev;
1666 	struct inet6_ifaddr *ifa = ptr;
1667 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1668 	struct net_device *event_netdev = ifa->idev->dev;
1669 
1670 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1671 
1672 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1673 	return NOTIFY_DONE;
1674 }
1675 #endif
1676 
1677 #define MLX4_IB_INVALID_MAC	((u64)-1)
1678 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1679 			       struct net_device *dev,
1680 			       int port)
1681 {
1682 	u64 new_smac = 0;
1683 	u64 release_mac = MLX4_IB_INVALID_MAC;
1684 	struct mlx4_ib_qp *qp;
1685 
1686 	read_lock(&dev_base_lock);
1687 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1688 	read_unlock(&dev_base_lock);
1689 
1690 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1691 
1692 	/* no need for update QP1 and mac registration in non-SRIOV */
1693 	if (!mlx4_is_mfunc(ibdev->dev))
1694 		return;
1695 
1696 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1697 	qp = ibdev->qp1_proxy[port - 1];
1698 	if (qp) {
1699 		int new_smac_index;
1700 		u64 old_smac;
1701 		struct mlx4_update_qp_params update_params;
1702 
1703 		mutex_lock(&qp->mutex);
1704 		old_smac = qp->pri.smac;
1705 		if (new_smac == old_smac)
1706 			goto unlock;
1707 
1708 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1709 
1710 		if (new_smac_index < 0)
1711 			goto unlock;
1712 
1713 		update_params.smac_index = new_smac_index;
1714 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1715 				   &update_params)) {
1716 			release_mac = new_smac;
1717 			goto unlock;
1718 		}
1719 		/* if old port was zero, no mac was yet registered for this QP */
1720 		if (qp->pri.smac_port)
1721 			release_mac = old_smac;
1722 		qp->pri.smac = new_smac;
1723 		qp->pri.smac_port = port;
1724 		qp->pri.smac_index = new_smac_index;
1725 	}
1726 
1727 unlock:
1728 	if (release_mac != MLX4_IB_INVALID_MAC)
1729 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1730 	if (qp)
1731 		mutex_unlock(&qp->mutex);
1732 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1733 }
1734 
1735 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1736 				 struct mlx4_ib_dev *ibdev, u8 port)
1737 {
1738 	struct in_device *in_dev;
1739 #if IS_ENABLED(CONFIG_IPV6)
1740 	struct inet6_dev *in6_dev;
1741 	union ib_gid  *pgid;
1742 	struct inet6_ifaddr *ifp;
1743 	union ib_gid default_gid;
1744 #endif
1745 	union ib_gid gid;
1746 
1747 
1748 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1749 		return;
1750 
1751 	/* IPv4 gids */
1752 	in_dev = in_dev_get(dev);
1753 	if (in_dev) {
1754 		for_ifa(in_dev) {
1755 			/*ifa->ifa_address;*/
1756 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1757 					       (struct in6_addr *)&gid);
1758 			update_gid_table(ibdev, port, &gid, 0, 0);
1759 		}
1760 		endfor_ifa(in_dev);
1761 		in_dev_put(in_dev);
1762 	}
1763 #if IS_ENABLED(CONFIG_IPV6)
1764 	mlx4_make_default_gid(dev, &default_gid);
1765 	/* IPv6 gids */
1766 	in6_dev = in6_dev_get(dev);
1767 	if (in6_dev) {
1768 		read_lock_bh(&in6_dev->lock);
1769 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1770 			pgid = (union ib_gid *)&ifp->addr;
1771 			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1772 				continue;
1773 			update_gid_table(ibdev, port, pgid, 0, 0);
1774 		}
1775 		read_unlock_bh(&in6_dev->lock);
1776 		in6_dev_put(in6_dev);
1777 	}
1778 #endif
1779 }
1780 
1781 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1782 				 struct  net_device *dev, u8 port)
1783 {
1784 	union ib_gid gid;
1785 	mlx4_make_default_gid(dev, &gid);
1786 	update_gid_table(ibdev, port, &gid, 0, 1);
1787 }
1788 
1789 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1790 {
1791 	struct	net_device *dev;
1792 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1793 	int i;
1794 	int err = 0;
1795 
1796 	for (i = 1; i <= ibdev->num_ports; ++i) {
1797 		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1798 		    IB_LINK_LAYER_ETHERNET) {
1799 			err = reset_gid_table(ibdev, i);
1800 			if (err)
1801 				goto out;
1802 		}
1803 	}
1804 
1805 	read_lock(&dev_base_lock);
1806 	spin_lock_bh(&iboe->lock);
1807 
1808 	for_each_netdev(&init_net, dev) {
1809 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1810 		/* port will be non-zero only for ETH ports */
1811 		if (port) {
1812 			mlx4_ib_set_default_gid(ibdev, dev, port);
1813 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1814 		}
1815 	}
1816 
1817 	spin_unlock_bh(&iboe->lock);
1818 	read_unlock(&dev_base_lock);
1819 out:
1820 	return err;
1821 }
1822 
1823 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1824 				 struct net_device *dev,
1825 				 unsigned long event)
1826 
1827 {
1828 	struct mlx4_ib_iboe *iboe;
1829 	int update_qps_port = -1;
1830 	int port;
1831 
1832 	iboe = &ibdev->iboe;
1833 
1834 	spin_lock_bh(&iboe->lock);
1835 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1836 		enum ib_port_state	port_state = IB_PORT_NOP;
1837 		struct net_device *old_master = iboe->masters[port - 1];
1838 		struct net_device *curr_netdev;
1839 		struct net_device *curr_master;
1840 
1841 		iboe->netdevs[port - 1] =
1842 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1843 		if (iboe->netdevs[port - 1])
1844 			mlx4_ib_set_default_gid(ibdev,
1845 						iboe->netdevs[port - 1], port);
1846 		curr_netdev = iboe->netdevs[port - 1];
1847 
1848 		if (iboe->netdevs[port - 1] &&
1849 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1850 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1851 				iboe->netdevs[port - 1]);
1852 		} else {
1853 			iboe->masters[port - 1] = NULL;
1854 		}
1855 		curr_master = iboe->masters[port - 1];
1856 
1857 		if (dev == iboe->netdevs[port - 1] &&
1858 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1859 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1860 			update_qps_port = port;
1861 
1862 		if (curr_netdev) {
1863 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1864 						IB_PORT_ACTIVE : IB_PORT_DOWN;
1865 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1866 			if (curr_master) {
1867 				/* if using bonding/team and a slave port is down, we
1868 				 * don't want the bond IP based gids in the table since
1869 				 * flows that select port by gid may get the down port.
1870 				*/
1871 				if (port_state == IB_PORT_DOWN) {
1872 					reset_gid_table(ibdev, port);
1873 					mlx4_ib_set_default_gid(ibdev,
1874 								curr_netdev,
1875 								port);
1876 				} else {
1877 					/* gids from the upper dev (bond/team)
1878 					 * should appear in port's gid table
1879 					*/
1880 					mlx4_ib_get_dev_addr(curr_master,
1881 							     ibdev, port);
1882 				}
1883 			}
1884 			/* if bonding is used it is possible that we add it to
1885 			 * masters only after IP address is assigned to the
1886 			 * net bonding interface.
1887 			*/
1888 			if (curr_master && (old_master != curr_master)) {
1889 				reset_gid_table(ibdev, port);
1890 				mlx4_ib_set_default_gid(ibdev,
1891 							curr_netdev, port);
1892 				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1893 			}
1894 
1895 			if (!curr_master && (old_master != curr_master)) {
1896 				reset_gid_table(ibdev, port);
1897 				mlx4_ib_set_default_gid(ibdev,
1898 							curr_netdev, port);
1899 				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1900 			}
1901 		} else {
1902 			reset_gid_table(ibdev, port);
1903 		}
1904 	}
1905 
1906 	spin_unlock_bh(&iboe->lock);
1907 
1908 	if (update_qps_port > 0)
1909 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1910 }
1911 
1912 static int mlx4_ib_netdev_event(struct notifier_block *this,
1913 				unsigned long event, void *ptr)
1914 {
1915 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1916 	struct mlx4_ib_dev *ibdev;
1917 
1918 	if (!net_eq(dev_net(dev), &init_net))
1919 		return NOTIFY_DONE;
1920 
1921 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1922 	mlx4_ib_scan_netdevs(ibdev, dev, event);
1923 
1924 	return NOTIFY_DONE;
1925 }
1926 
1927 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1928 {
1929 	int port;
1930 	int slave;
1931 	int i;
1932 
1933 	if (mlx4_is_master(ibdev->dev)) {
1934 		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1935 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1936 				for (i = 0;
1937 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1938 				     ++i) {
1939 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1940 					/* master has the identity virt2phys pkey mapping */
1941 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1942 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1943 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1944 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1945 				}
1946 			}
1947 		}
1948 		/* initialize pkey cache */
1949 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1950 			for (i = 0;
1951 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1952 			     ++i)
1953 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
1954 					(i) ? 0 : 0xFFFF;
1955 		}
1956 	}
1957 }
1958 
1959 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1960 {
1961 	char name[80];
1962 	int eq_per_port = 0;
1963 	int added_eqs = 0;
1964 	int total_eqs = 0;
1965 	int i, j, eq;
1966 
1967 	/* Legacy mode or comp_pool is not large enough */
1968 	if (dev->caps.comp_pool == 0 ||
1969 	    dev->caps.num_ports > dev->caps.comp_pool)
1970 		return;
1971 
1972 	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1973 					dev->caps.num_ports);
1974 
1975 	/* Init eq table */
1976 	added_eqs = 0;
1977 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1978 		added_eqs += eq_per_port;
1979 
1980 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
1981 
1982 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1983 	if (!ibdev->eq_table)
1984 		return;
1985 
1986 	ibdev->eq_added = added_eqs;
1987 
1988 	eq = 0;
1989 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1990 		for (j = 0; j < eq_per_port; j++) {
1991 			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1992 				 i, j, dev->pdev->bus->name);
1993 			/* Set IRQ for specific name (per ring) */
1994 			if (mlx4_assign_eq(dev, name, NULL,
1995 					   &ibdev->eq_table[eq])) {
1996 				/* Use legacy (same as mlx4_en driver) */
1997 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1998 				ibdev->eq_table[eq] =
1999 					(eq % dev->caps.num_comp_vectors);
2000 			}
2001 			eq++;
2002 		}
2003 	}
2004 
2005 	/* Fill the reset of the vector with legacy EQ */
2006 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2007 		ibdev->eq_table[eq++] = i;
2008 
2009 	/* Advertise the new number of EQs to clients */
2010 	ibdev->ib_dev.num_comp_vectors = total_eqs;
2011 }
2012 
2013 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2014 {
2015 	int i;
2016 
2017 	/* no additional eqs were added */
2018 	if (!ibdev->eq_table)
2019 		return;
2020 
2021 	/* Reset the advertised EQ number */
2022 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2023 
2024 	/* Free only the added eqs */
2025 	for (i = 0; i < ibdev->eq_added; i++) {
2026 		/* Don't free legacy eqs if used */
2027 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2028 			continue;
2029 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2030 	}
2031 
2032 	kfree(ibdev->eq_table);
2033 }
2034 
2035 static void *mlx4_ib_add(struct mlx4_dev *dev)
2036 {
2037 	struct mlx4_ib_dev *ibdev;
2038 	int num_ports = 0;
2039 	int i, j;
2040 	int err;
2041 	struct mlx4_ib_iboe *iboe;
2042 	int ib_num_ports = 0;
2043 
2044 	pr_info_once("%s", mlx4_ib_version);
2045 
2046 	num_ports = 0;
2047 	mlx4_foreach_ib_transport_port(i, dev)
2048 		num_ports++;
2049 
2050 	/* No point in registering a device with no ports... */
2051 	if (num_ports == 0)
2052 		return NULL;
2053 
2054 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2055 	if (!ibdev) {
2056 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
2057 		return NULL;
2058 	}
2059 
2060 	iboe = &ibdev->iboe;
2061 
2062 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2063 		goto err_dealloc;
2064 
2065 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2066 		goto err_pd;
2067 
2068 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2069 				 PAGE_SIZE);
2070 	if (!ibdev->uar_map)
2071 		goto err_uar;
2072 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2073 
2074 	ibdev->dev = dev;
2075 
2076 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2077 	ibdev->ib_dev.owner		= THIS_MODULE;
2078 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2079 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2080 	ibdev->num_ports		= num_ports;
2081 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
2082 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2083 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
2084 
2085 	if (dev->caps.userspace_caps)
2086 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2087 	else
2088 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2089 
2090 	ibdev->ib_dev.uverbs_cmd_mask	=
2091 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2092 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2093 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2094 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2095 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2096 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2097 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2098 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2099 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2100 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2101 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2102 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2103 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2104 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2105 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2106 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2107 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2108 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2109 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2110 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2111 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2112 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2113 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2114 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2115 
2116 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2117 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2118 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2119 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2120 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2121 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2122 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2123 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2124 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2125 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2126 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2127 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2128 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2129 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2130 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2131 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2132 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2133 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2134 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2135 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2136 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2137 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2138 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2139 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2140 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2141 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2142 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2143 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2144 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2145 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2146 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2147 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2148 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2149 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2150 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2151 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2152 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2153 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2154 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2155 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2156 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2157 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2158 
2159 	if (!mlx4_is_slave(ibdev->dev)) {
2160 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2161 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2162 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2163 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2164 	}
2165 
2166 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2167 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2168 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2169 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2170 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2171 
2172 		ibdev->ib_dev.uverbs_cmd_mask |=
2173 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2174 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2175 	}
2176 
2177 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2178 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2179 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2180 		ibdev->ib_dev.uverbs_cmd_mask |=
2181 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2182 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2183 	}
2184 
2185 	if (check_flow_steering_support(dev)) {
2186 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2187 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2188 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2189 
2190 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2191 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2192 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2193 	}
2194 
2195 	mlx4_ib_alloc_eqs(dev, ibdev);
2196 
2197 	spin_lock_init(&iboe->lock);
2198 
2199 	if (init_node_data(ibdev))
2200 		goto err_map;
2201 
2202 	for (i = 0; i < ibdev->num_ports; ++i) {
2203 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2204 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2205 						IB_LINK_LAYER_ETHERNET) {
2206 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2207 			if (err)
2208 				ibdev->counters[i] = -1;
2209 		} else {
2210 			ibdev->counters[i] = -1;
2211 		}
2212 	}
2213 
2214 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2215 		ib_num_ports++;
2216 
2217 	spin_lock_init(&ibdev->sm_lock);
2218 	mutex_init(&ibdev->cap_mask_mutex);
2219 
2220 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2221 	    ib_num_ports) {
2222 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2223 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2224 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2225 					    &ibdev->steer_qpn_base);
2226 		if (err)
2227 			goto err_counter;
2228 
2229 		ibdev->ib_uc_qpns_bitmap =
2230 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2231 				sizeof(long),
2232 				GFP_KERNEL);
2233 		if (!ibdev->ib_uc_qpns_bitmap) {
2234 			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2235 			goto err_steer_qp_release;
2236 		}
2237 
2238 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2239 
2240 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2241 				dev, ibdev->steer_qpn_base,
2242 				ibdev->steer_qpn_base +
2243 				ibdev->steer_qpn_count - 1);
2244 		if (err)
2245 			goto err_steer_free_bitmap;
2246 	}
2247 
2248 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2249 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2250 
2251 	if (ib_register_device(&ibdev->ib_dev, NULL))
2252 		goto err_steer_free_bitmap;
2253 
2254 	if (mlx4_ib_mad_init(ibdev))
2255 		goto err_reg;
2256 
2257 	if (mlx4_ib_init_sriov(ibdev))
2258 		goto err_mad;
2259 
2260 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2261 		if (!iboe->nb.notifier_call) {
2262 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2263 			err = register_netdevice_notifier(&iboe->nb);
2264 			if (err) {
2265 				iboe->nb.notifier_call = NULL;
2266 				goto err_notif;
2267 			}
2268 		}
2269 		if (!iboe->nb_inet.notifier_call) {
2270 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2271 			err = register_inetaddr_notifier(&iboe->nb_inet);
2272 			if (err) {
2273 				iboe->nb_inet.notifier_call = NULL;
2274 				goto err_notif;
2275 			}
2276 		}
2277 #if IS_ENABLED(CONFIG_IPV6)
2278 		if (!iboe->nb_inet6.notifier_call) {
2279 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2280 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2281 			if (err) {
2282 				iboe->nb_inet6.notifier_call = NULL;
2283 				goto err_notif;
2284 			}
2285 		}
2286 #endif
2287 		if (mlx4_ib_init_gid_table(ibdev))
2288 			goto err_notif;
2289 	}
2290 
2291 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2292 		if (device_create_file(&ibdev->ib_dev.dev,
2293 				       mlx4_class_attributes[j]))
2294 			goto err_notif;
2295 	}
2296 
2297 	ibdev->ib_active = true;
2298 
2299 	if (mlx4_is_mfunc(ibdev->dev))
2300 		init_pkeys(ibdev);
2301 
2302 	/* create paravirt contexts for any VFs which are active */
2303 	if (mlx4_is_master(ibdev->dev)) {
2304 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2305 			if (j == mlx4_master_func_num(ibdev->dev))
2306 				continue;
2307 			if (mlx4_is_slave_active(ibdev->dev, j))
2308 				do_slave_init(ibdev, j, 1);
2309 		}
2310 	}
2311 	return ibdev;
2312 
2313 err_notif:
2314 	if (ibdev->iboe.nb.notifier_call) {
2315 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2316 			pr_warn("failure unregistering notifier\n");
2317 		ibdev->iboe.nb.notifier_call = NULL;
2318 	}
2319 	if (ibdev->iboe.nb_inet.notifier_call) {
2320 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2321 			pr_warn("failure unregistering notifier\n");
2322 		ibdev->iboe.nb_inet.notifier_call = NULL;
2323 	}
2324 #if IS_ENABLED(CONFIG_IPV6)
2325 	if (ibdev->iboe.nb_inet6.notifier_call) {
2326 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2327 			pr_warn("failure unregistering notifier\n");
2328 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2329 	}
2330 #endif
2331 	flush_workqueue(wq);
2332 
2333 	mlx4_ib_close_sriov(ibdev);
2334 
2335 err_mad:
2336 	mlx4_ib_mad_cleanup(ibdev);
2337 
2338 err_reg:
2339 	ib_unregister_device(&ibdev->ib_dev);
2340 
2341 err_steer_free_bitmap:
2342 	kfree(ibdev->ib_uc_qpns_bitmap);
2343 
2344 err_steer_qp_release:
2345 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2346 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2347 				      ibdev->steer_qpn_count);
2348 err_counter:
2349 	for (; i; --i)
2350 		if (ibdev->counters[i - 1] != -1)
2351 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2352 
2353 err_map:
2354 	iounmap(ibdev->uar_map);
2355 
2356 err_uar:
2357 	mlx4_uar_free(dev, &ibdev->priv_uar);
2358 
2359 err_pd:
2360 	mlx4_pd_free(dev, ibdev->priv_pdn);
2361 
2362 err_dealloc:
2363 	ib_dealloc_device(&ibdev->ib_dev);
2364 
2365 	return NULL;
2366 }
2367 
2368 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2369 {
2370 	int offset;
2371 
2372 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2373 
2374 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2375 					 dev->steer_qpn_count,
2376 					 get_count_order(count));
2377 	if (offset < 0)
2378 		return offset;
2379 
2380 	*qpn = dev->steer_qpn_base + offset;
2381 	return 0;
2382 }
2383 
2384 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2385 {
2386 	if (!qpn ||
2387 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2388 		return;
2389 
2390 	BUG_ON(qpn < dev->steer_qpn_base);
2391 
2392 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2393 			      qpn - dev->steer_qpn_base,
2394 			      get_count_order(count));
2395 }
2396 
2397 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2398 			 int is_attach)
2399 {
2400 	int err;
2401 	size_t flow_size;
2402 	struct ib_flow_attr *flow = NULL;
2403 	struct ib_flow_spec_ib *ib_spec;
2404 
2405 	if (is_attach) {
2406 		flow_size = sizeof(struct ib_flow_attr) +
2407 			    sizeof(struct ib_flow_spec_ib);
2408 		flow = kzalloc(flow_size, GFP_KERNEL);
2409 		if (!flow)
2410 			return -ENOMEM;
2411 		flow->port = mqp->port;
2412 		flow->num_of_specs = 1;
2413 		flow->size = flow_size;
2414 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2415 		ib_spec->type = IB_FLOW_SPEC_IB;
2416 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2417 		/* Add an empty rule for IB L2 */
2418 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2419 
2420 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2421 					    IB_FLOW_DOMAIN_NIC,
2422 					    MLX4_FS_REGULAR,
2423 					    &mqp->reg_id);
2424 	} else {
2425 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2426 	}
2427 	kfree(flow);
2428 	return err;
2429 }
2430 
2431 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2432 {
2433 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2434 	int p;
2435 
2436 	ibdev->ib_active = false;
2437 	flush_workqueue(wq);
2438 
2439 	mlx4_ib_close_sriov(ibdev);
2440 	mlx4_ib_mad_cleanup(ibdev);
2441 	ib_unregister_device(&ibdev->ib_dev);
2442 	if (ibdev->iboe.nb.notifier_call) {
2443 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2444 			pr_warn("failure unregistering notifier\n");
2445 		ibdev->iboe.nb.notifier_call = NULL;
2446 	}
2447 
2448 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2449 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2450 				      ibdev->steer_qpn_count);
2451 		kfree(ibdev->ib_uc_qpns_bitmap);
2452 	}
2453 
2454 	if (ibdev->iboe.nb_inet.notifier_call) {
2455 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2456 			pr_warn("failure unregistering notifier\n");
2457 		ibdev->iboe.nb_inet.notifier_call = NULL;
2458 	}
2459 #if IS_ENABLED(CONFIG_IPV6)
2460 	if (ibdev->iboe.nb_inet6.notifier_call) {
2461 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2462 			pr_warn("failure unregistering notifier\n");
2463 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2464 	}
2465 #endif
2466 
2467 	iounmap(ibdev->uar_map);
2468 	for (p = 0; p < ibdev->num_ports; ++p)
2469 		if (ibdev->counters[p] != -1)
2470 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2471 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2472 		mlx4_CLOSE_PORT(dev, p);
2473 
2474 	mlx4_ib_free_eqs(dev, ibdev);
2475 
2476 	mlx4_uar_free(dev, &ibdev->priv_uar);
2477 	mlx4_pd_free(dev, ibdev->priv_pdn);
2478 	ib_dealloc_device(&ibdev->ib_dev);
2479 }
2480 
2481 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2482 {
2483 	struct mlx4_ib_demux_work **dm = NULL;
2484 	struct mlx4_dev *dev = ibdev->dev;
2485 	int i;
2486 	unsigned long flags;
2487 	struct mlx4_active_ports actv_ports;
2488 	unsigned int ports;
2489 	unsigned int first_port;
2490 
2491 	if (!mlx4_is_master(dev))
2492 		return;
2493 
2494 	actv_ports = mlx4_get_active_ports(dev, slave);
2495 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2496 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2497 
2498 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2499 	if (!dm) {
2500 		pr_err("failed to allocate memory for tunneling qp update\n");
2501 		goto out;
2502 	}
2503 
2504 	for (i = 0; i < ports; i++) {
2505 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2506 		if (!dm[i]) {
2507 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2508 			for (i = 0; i < dev->caps.num_ports; i++) {
2509 				if (dm[i])
2510 					kfree(dm[i]);
2511 			}
2512 			goto out;
2513 		}
2514 	}
2515 	/* initialize or tear down tunnel QPs for the slave */
2516 	for (i = 0; i < ports; i++) {
2517 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2518 		dm[i]->port = first_port + i + 1;
2519 		dm[i]->slave = slave;
2520 		dm[i]->do_init = do_init;
2521 		dm[i]->dev = ibdev;
2522 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2523 		if (!ibdev->sriov.is_going_down)
2524 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2525 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2526 	}
2527 out:
2528 	kfree(dm);
2529 	return;
2530 }
2531 
2532 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2533 			  enum mlx4_dev_event event, unsigned long param)
2534 {
2535 	struct ib_event ibev;
2536 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2537 	struct mlx4_eqe *eqe = NULL;
2538 	struct ib_event_work *ew;
2539 	int p = 0;
2540 
2541 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2542 		eqe = (struct mlx4_eqe *)param;
2543 	else
2544 		p = (int) param;
2545 
2546 	switch (event) {
2547 	case MLX4_DEV_EVENT_PORT_UP:
2548 		if (p > ibdev->num_ports)
2549 			return;
2550 		if (mlx4_is_master(dev) &&
2551 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2552 			IB_LINK_LAYER_INFINIBAND) {
2553 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2554 		}
2555 		ibev.event = IB_EVENT_PORT_ACTIVE;
2556 		break;
2557 
2558 	case MLX4_DEV_EVENT_PORT_DOWN:
2559 		if (p > ibdev->num_ports)
2560 			return;
2561 		ibev.event = IB_EVENT_PORT_ERR;
2562 		break;
2563 
2564 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2565 		ibdev->ib_active = false;
2566 		ibev.event = IB_EVENT_DEVICE_FATAL;
2567 		break;
2568 
2569 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2570 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2571 		if (!ew) {
2572 			pr_err("failed to allocate memory for events work\n");
2573 			break;
2574 		}
2575 
2576 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2577 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2578 		ew->ib_dev = ibdev;
2579 		/* need to queue only for port owner, which uses GEN_EQE */
2580 		if (mlx4_is_master(dev))
2581 			queue_work(wq, &ew->work);
2582 		else
2583 			handle_port_mgmt_change_event(&ew->work);
2584 		return;
2585 
2586 	case MLX4_DEV_EVENT_SLAVE_INIT:
2587 		/* here, p is the slave id */
2588 		do_slave_init(ibdev, p, 1);
2589 		return;
2590 
2591 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2592 		/* here, p is the slave id */
2593 		do_slave_init(ibdev, p, 0);
2594 		return;
2595 
2596 	default:
2597 		return;
2598 	}
2599 
2600 	ibev.device	      = ibdev_ptr;
2601 	ibev.element.port_num = (u8) p;
2602 
2603 	ib_dispatch_event(&ibev);
2604 }
2605 
2606 static struct mlx4_interface mlx4_ib_interface = {
2607 	.add		= mlx4_ib_add,
2608 	.remove		= mlx4_ib_remove,
2609 	.event		= mlx4_ib_event,
2610 	.protocol	= MLX4_PROT_IB_IPV6
2611 };
2612 
2613 static int __init mlx4_ib_init(void)
2614 {
2615 	int err;
2616 
2617 	wq = create_singlethread_workqueue("mlx4_ib");
2618 	if (!wq)
2619 		return -ENOMEM;
2620 
2621 	err = mlx4_ib_mcg_init();
2622 	if (err)
2623 		goto clean_wq;
2624 
2625 	err = mlx4_register_interface(&mlx4_ib_interface);
2626 	if (err)
2627 		goto clean_mcg;
2628 
2629 	return 0;
2630 
2631 clean_mcg:
2632 	mlx4_ib_mcg_destroy();
2633 
2634 clean_wq:
2635 	destroy_workqueue(wq);
2636 	return err;
2637 }
2638 
2639 static void __exit mlx4_ib_cleanup(void)
2640 {
2641 	mlx4_unregister_interface(&mlx4_ib_interface);
2642 	mlx4_ib_mcg_destroy();
2643 	destroy_workqueue(wq);
2644 }
2645 
2646 module_init(mlx4_ib_init);
2647 module_exit(mlx4_ib_cleanup);
2648