xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision cf2f33a4e54096f90652cca3511fd6a456ea5abe)
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 = 0;
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: 0)");
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 				struct ib_udata *uhw)
137 {
138 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
139 	struct ib_smp *in_mad  = NULL;
140 	struct ib_smp *out_mad = NULL;
141 	int err = -ENOMEM;
142 	int have_ib_ports;
143 	struct mlx4_uverbs_ex_query_device cmd;
144 	struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
145 	struct mlx4_clock_params clock_params;
146 
147 	if (uhw->inlen) {
148 		if (uhw->inlen < sizeof(cmd))
149 			return -EINVAL;
150 
151 		err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
152 		if (err)
153 			return err;
154 
155 		if (cmd.comp_mask)
156 			return -EINVAL;
157 
158 		if (cmd.reserved)
159 			return -EINVAL;
160 	}
161 
162 	resp.response_length = offsetof(typeof(resp), response_length) +
163 		sizeof(resp.response_length);
164 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
165 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
166 	if (!in_mad || !out_mad)
167 		goto out;
168 
169 	init_query_mad(in_mad);
170 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
171 
172 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
173 			   1, NULL, NULL, in_mad, out_mad);
174 	if (err)
175 		goto out;
176 
177 	memset(props, 0, sizeof *props);
178 
179 	have_ib_ports = num_ib_ports(dev->dev);
180 
181 	props->fw_ver = dev->dev->caps.fw_ver;
182 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
183 		IB_DEVICE_PORT_ACTIVE_EVENT		|
184 		IB_DEVICE_SYS_IMAGE_GUID		|
185 		IB_DEVICE_RC_RNR_NAK_GEN		|
186 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
187 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
188 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
189 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
190 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
191 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
192 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
193 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
194 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
195 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
196 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
197 	if (dev->dev->caps.max_gso_sz &&
198 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
199 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
200 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
201 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
202 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
203 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
204 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
205 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
206 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
207 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
208 		props->device_cap_flags |= IB_DEVICE_XRC;
209 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
210 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
211 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
212 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
213 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
214 		else
215 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
216 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
217 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
218 	}
219 
220 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
221 		0xffffff;
222 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
223 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
224 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
225 
226 	props->max_mr_size	   = ~0ull;
227 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
228 	props->max_qp		   = dev->dev->quotas.qp;
229 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
230 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
231 					 dev->dev->caps.max_rq_sg);
232 	props->max_cq		   = dev->dev->quotas.cq;
233 	props->max_cqe		   = dev->dev->caps.max_cqes;
234 	props->max_mr		   = dev->dev->quotas.mpt;
235 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
236 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
237 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
238 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
239 	props->max_srq		   = dev->dev->quotas.srq;
240 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
241 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
242 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
243 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
244 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
245 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
246 	props->masked_atomic_cap   = props->atomic_cap;
247 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
248 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
249 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
250 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
251 					   props->max_mcast_grp;
252 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
253 	props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
254 	props->timestamp_mask = 0xFFFFFFFFFFFFULL;
255 
256 	if (!mlx4_is_slave(dev->dev))
257 		err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
258 
259 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
260 		resp.response_length += sizeof(resp.hca_core_clock_offset);
261 		if (!err && !mlx4_is_slave(dev->dev)) {
262 			resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
263 			resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
264 		}
265 	}
266 
267 	if (uhw->outlen) {
268 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
269 		if (err)
270 			goto out;
271 	}
272 out:
273 	kfree(in_mad);
274 	kfree(out_mad);
275 
276 	return err;
277 }
278 
279 static enum rdma_link_layer
280 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
281 {
282 	struct mlx4_dev *dev = to_mdev(device)->dev;
283 
284 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
285 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
286 }
287 
288 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
289 			      struct ib_port_attr *props, int netw_view)
290 {
291 	struct ib_smp *in_mad  = NULL;
292 	struct ib_smp *out_mad = NULL;
293 	int ext_active_speed;
294 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
295 	int err = -ENOMEM;
296 
297 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
298 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
299 	if (!in_mad || !out_mad)
300 		goto out;
301 
302 	init_query_mad(in_mad);
303 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
304 	in_mad->attr_mod = cpu_to_be32(port);
305 
306 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
307 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
308 
309 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
310 				in_mad, out_mad);
311 	if (err)
312 		goto out;
313 
314 
315 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
316 	props->lmc		= out_mad->data[34] & 0x7;
317 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
318 	props->sm_sl		= out_mad->data[36] & 0xf;
319 	props->state		= out_mad->data[32] & 0xf;
320 	props->phys_state	= out_mad->data[33] >> 4;
321 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
322 	if (netw_view)
323 		props->gid_tbl_len = out_mad->data[50];
324 	else
325 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
326 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
327 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
328 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
329 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
330 	props->active_width	= out_mad->data[31] & 0xf;
331 	props->active_speed	= out_mad->data[35] >> 4;
332 	props->max_mtu		= out_mad->data[41] & 0xf;
333 	props->active_mtu	= out_mad->data[36] >> 4;
334 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
335 	props->max_vl_num	= out_mad->data[37] >> 4;
336 	props->init_type_reply	= out_mad->data[41] >> 4;
337 
338 	/* Check if extended speeds (EDR/FDR/...) are supported */
339 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
340 		ext_active_speed = out_mad->data[62] >> 4;
341 
342 		switch (ext_active_speed) {
343 		case 1:
344 			props->active_speed = IB_SPEED_FDR;
345 			break;
346 		case 2:
347 			props->active_speed = IB_SPEED_EDR;
348 			break;
349 		}
350 	}
351 
352 	/* If reported active speed is QDR, check if is FDR-10 */
353 	if (props->active_speed == IB_SPEED_QDR) {
354 		init_query_mad(in_mad);
355 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
356 		in_mad->attr_mod = cpu_to_be32(port);
357 
358 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
359 				   NULL, NULL, in_mad, out_mad);
360 		if (err)
361 			goto out;
362 
363 		/* Checking LinkSpeedActive for FDR-10 */
364 		if (out_mad->data[15] & 0x1)
365 			props->active_speed = IB_SPEED_FDR10;
366 	}
367 
368 	/* Avoid wrong speed value returned by FW if the IB link is down. */
369 	if (props->state == IB_PORT_DOWN)
370 		 props->active_speed = IB_SPEED_SDR;
371 
372 out:
373 	kfree(in_mad);
374 	kfree(out_mad);
375 	return err;
376 }
377 
378 static u8 state_to_phys_state(enum ib_port_state state)
379 {
380 	return state == IB_PORT_ACTIVE ? 5 : 3;
381 }
382 
383 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
384 			       struct ib_port_attr *props, int netw_view)
385 {
386 
387 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
388 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
389 	struct net_device *ndev;
390 	enum ib_mtu tmp;
391 	struct mlx4_cmd_mailbox *mailbox;
392 	int err = 0;
393 	int is_bonded = mlx4_is_bonded(mdev->dev);
394 
395 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
396 	if (IS_ERR(mailbox))
397 		return PTR_ERR(mailbox);
398 
399 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
400 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
401 			   MLX4_CMD_WRAPPED);
402 	if (err)
403 		goto out;
404 
405 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
406 						IB_WIDTH_4X : IB_WIDTH_1X;
407 	props->active_speed	= IB_SPEED_QDR;
408 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
409 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
410 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
411 	props->pkey_tbl_len	= 1;
412 	props->max_mtu		= IB_MTU_4096;
413 	props->max_vl_num	= 2;
414 	props->state		= IB_PORT_DOWN;
415 	props->phys_state	= state_to_phys_state(props->state);
416 	props->active_mtu	= IB_MTU_256;
417 	if (is_bonded)
418 		rtnl_lock(); /* required to get upper dev */
419 	spin_lock_bh(&iboe->lock);
420 	ndev = iboe->netdevs[port - 1];
421 	if (ndev && is_bonded)
422 		ndev = netdev_master_upper_dev_get(ndev);
423 	if (!ndev)
424 		goto out_unlock;
425 
426 	tmp = iboe_get_mtu(ndev->mtu);
427 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
428 
429 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
430 					IB_PORT_ACTIVE : IB_PORT_DOWN;
431 	props->phys_state	= state_to_phys_state(props->state);
432 out_unlock:
433 	spin_unlock_bh(&iboe->lock);
434 	if (is_bonded)
435 		rtnl_unlock();
436 out:
437 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
438 	return err;
439 }
440 
441 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
442 			 struct ib_port_attr *props, int netw_view)
443 {
444 	int err;
445 
446 	memset(props, 0, sizeof *props);
447 
448 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
449 		ib_link_query_port(ibdev, port, props, netw_view) :
450 				eth_link_query_port(ibdev, port, props, netw_view);
451 
452 	return err;
453 }
454 
455 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
456 			      struct ib_port_attr *props)
457 {
458 	/* returns host view */
459 	return __mlx4_ib_query_port(ibdev, port, props, 0);
460 }
461 
462 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
463 			union ib_gid *gid, int netw_view)
464 {
465 	struct ib_smp *in_mad  = NULL;
466 	struct ib_smp *out_mad = NULL;
467 	int err = -ENOMEM;
468 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
469 	int clear = 0;
470 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
471 
472 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
473 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
474 	if (!in_mad || !out_mad)
475 		goto out;
476 
477 	init_query_mad(in_mad);
478 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
479 	in_mad->attr_mod = cpu_to_be32(port);
480 
481 	if (mlx4_is_mfunc(dev->dev) && netw_view)
482 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
483 
484 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
485 	if (err)
486 		goto out;
487 
488 	memcpy(gid->raw, out_mad->data + 8, 8);
489 
490 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
491 		if (index) {
492 			/* For any index > 0, return the null guid */
493 			err = 0;
494 			clear = 1;
495 			goto out;
496 		}
497 	}
498 
499 	init_query_mad(in_mad);
500 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
501 	in_mad->attr_mod = cpu_to_be32(index / 8);
502 
503 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
504 			   NULL, NULL, in_mad, out_mad);
505 	if (err)
506 		goto out;
507 
508 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
509 
510 out:
511 	if (clear)
512 		memset(gid->raw + 8, 0, 8);
513 	kfree(in_mad);
514 	kfree(out_mad);
515 	return err;
516 }
517 
518 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
519 			  union ib_gid *gid)
520 {
521 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
522 
523 	*gid = dev->iboe.gid_table[port - 1][index];
524 
525 	return 0;
526 }
527 
528 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
529 			     union ib_gid *gid)
530 {
531 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
532 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
533 	else
534 		return iboe_query_gid(ibdev, port, index, gid);
535 }
536 
537 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
538 			 u16 *pkey, int netw_view)
539 {
540 	struct ib_smp *in_mad  = NULL;
541 	struct ib_smp *out_mad = NULL;
542 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
543 	int err = -ENOMEM;
544 
545 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
546 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
547 	if (!in_mad || !out_mad)
548 		goto out;
549 
550 	init_query_mad(in_mad);
551 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
552 	in_mad->attr_mod = cpu_to_be32(index / 32);
553 
554 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
555 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
556 
557 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
558 			   in_mad, out_mad);
559 	if (err)
560 		goto out;
561 
562 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
563 
564 out:
565 	kfree(in_mad);
566 	kfree(out_mad);
567 	return err;
568 }
569 
570 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
571 {
572 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
573 }
574 
575 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
576 				 struct ib_device_modify *props)
577 {
578 	struct mlx4_cmd_mailbox *mailbox;
579 	unsigned long flags;
580 
581 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
582 		return -EOPNOTSUPP;
583 
584 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
585 		return 0;
586 
587 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
588 		return -EOPNOTSUPP;
589 
590 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
591 	memcpy(ibdev->node_desc, props->node_desc, 64);
592 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
593 
594 	/*
595 	 * If possible, pass node desc to FW, so it can generate
596 	 * a 144 trap.  If cmd fails, just ignore.
597 	 */
598 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
599 	if (IS_ERR(mailbox))
600 		return 0;
601 
602 	memcpy(mailbox->buf, props->node_desc, 64);
603 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
604 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
605 
606 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
607 
608 	return 0;
609 }
610 
611 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
612 			    u32 cap_mask)
613 {
614 	struct mlx4_cmd_mailbox *mailbox;
615 	int err;
616 
617 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
618 	if (IS_ERR(mailbox))
619 		return PTR_ERR(mailbox);
620 
621 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
622 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
623 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
624 	} else {
625 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
626 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
627 	}
628 
629 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
630 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
631 		       MLX4_CMD_WRAPPED);
632 
633 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
634 	return err;
635 }
636 
637 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
638 			       struct ib_port_modify *props)
639 {
640 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
641 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
642 	struct ib_port_attr attr;
643 	u32 cap_mask;
644 	int err;
645 
646 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
647 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
648 	 * violations and port capabilities are not meaningful.
649 	 */
650 	if (is_eth)
651 		return 0;
652 
653 	mutex_lock(&mdev->cap_mask_mutex);
654 
655 	err = mlx4_ib_query_port(ibdev, port, &attr);
656 	if (err)
657 		goto out;
658 
659 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
660 		~props->clr_port_cap_mask;
661 
662 	err = mlx4_ib_SET_PORT(mdev, port,
663 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
664 			       cap_mask);
665 
666 out:
667 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
668 	return err;
669 }
670 
671 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
672 						  struct ib_udata *udata)
673 {
674 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
675 	struct mlx4_ib_ucontext *context;
676 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
677 	struct mlx4_ib_alloc_ucontext_resp resp;
678 	int err;
679 
680 	if (!dev->ib_active)
681 		return ERR_PTR(-EAGAIN);
682 
683 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
684 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
685 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
686 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
687 	} else {
688 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
689 		resp.qp_tab_size      = dev->dev->caps.num_qps;
690 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
691 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
692 		resp.cqe_size	      = dev->dev->caps.cqe_size;
693 	}
694 
695 	context = kmalloc(sizeof *context, GFP_KERNEL);
696 	if (!context)
697 		return ERR_PTR(-ENOMEM);
698 
699 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
700 	if (err) {
701 		kfree(context);
702 		return ERR_PTR(err);
703 	}
704 
705 	INIT_LIST_HEAD(&context->db_page_list);
706 	mutex_init(&context->db_page_mutex);
707 
708 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
709 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
710 	else
711 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
712 
713 	if (err) {
714 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
715 		kfree(context);
716 		return ERR_PTR(-EFAULT);
717 	}
718 
719 	return &context->ibucontext;
720 }
721 
722 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
723 {
724 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
725 
726 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
727 	kfree(context);
728 
729 	return 0;
730 }
731 
732 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
733 {
734 	struct mlx4_ib_dev *dev = to_mdev(context->device);
735 
736 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
737 		return -EINVAL;
738 
739 	if (vma->vm_pgoff == 0) {
740 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
741 
742 		if (io_remap_pfn_range(vma, vma->vm_start,
743 				       to_mucontext(context)->uar.pfn,
744 				       PAGE_SIZE, vma->vm_page_prot))
745 			return -EAGAIN;
746 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
747 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
748 
749 		if (io_remap_pfn_range(vma, vma->vm_start,
750 				       to_mucontext(context)->uar.pfn +
751 				       dev->dev->caps.num_uars,
752 				       PAGE_SIZE, vma->vm_page_prot))
753 			return -EAGAIN;
754 	} else if (vma->vm_pgoff == 3) {
755 		struct mlx4_clock_params params;
756 		int ret = mlx4_get_internal_clock_params(dev->dev, &params);
757 
758 		if (ret)
759 			return ret;
760 
761 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
762 		if (io_remap_pfn_range(vma, vma->vm_start,
763 				       (pci_resource_start(dev->dev->persist->pdev,
764 							   params.bar) +
765 					params.offset)
766 				       >> PAGE_SHIFT,
767 				       PAGE_SIZE, vma->vm_page_prot))
768 			return -EAGAIN;
769 	} else {
770 		return -EINVAL;
771 	}
772 
773 	return 0;
774 }
775 
776 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
777 				      struct ib_ucontext *context,
778 				      struct ib_udata *udata)
779 {
780 	struct mlx4_ib_pd *pd;
781 	int err;
782 
783 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
784 	if (!pd)
785 		return ERR_PTR(-ENOMEM);
786 
787 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
788 	if (err) {
789 		kfree(pd);
790 		return ERR_PTR(err);
791 	}
792 
793 	if (context)
794 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
795 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
796 			kfree(pd);
797 			return ERR_PTR(-EFAULT);
798 		}
799 
800 	return &pd->ibpd;
801 }
802 
803 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
804 {
805 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
806 	kfree(pd);
807 
808 	return 0;
809 }
810 
811 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
812 					  struct ib_ucontext *context,
813 					  struct ib_udata *udata)
814 {
815 	struct mlx4_ib_xrcd *xrcd;
816 	struct ib_cq_init_attr cq_attr = {};
817 	int err;
818 
819 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
820 		return ERR_PTR(-ENOSYS);
821 
822 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
823 	if (!xrcd)
824 		return ERR_PTR(-ENOMEM);
825 
826 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
827 	if (err)
828 		goto err1;
829 
830 	xrcd->pd = ib_alloc_pd(ibdev);
831 	if (IS_ERR(xrcd->pd)) {
832 		err = PTR_ERR(xrcd->pd);
833 		goto err2;
834 	}
835 
836 	cq_attr.cqe = 1;
837 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
838 	if (IS_ERR(xrcd->cq)) {
839 		err = PTR_ERR(xrcd->cq);
840 		goto err3;
841 	}
842 
843 	return &xrcd->ibxrcd;
844 
845 err3:
846 	ib_dealloc_pd(xrcd->pd);
847 err2:
848 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
849 err1:
850 	kfree(xrcd);
851 	return ERR_PTR(err);
852 }
853 
854 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
855 {
856 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
857 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
858 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
859 	kfree(xrcd);
860 
861 	return 0;
862 }
863 
864 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
865 {
866 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
867 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
868 	struct mlx4_ib_gid_entry *ge;
869 
870 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
871 	if (!ge)
872 		return -ENOMEM;
873 
874 	ge->gid = *gid;
875 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
876 		ge->port = mqp->port;
877 		ge->added = 1;
878 	}
879 
880 	mutex_lock(&mqp->mutex);
881 	list_add_tail(&ge->list, &mqp->gid_list);
882 	mutex_unlock(&mqp->mutex);
883 
884 	return 0;
885 }
886 
887 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
888 		   union ib_gid *gid)
889 {
890 	struct net_device *ndev;
891 	int ret = 0;
892 
893 	if (!mqp->port)
894 		return 0;
895 
896 	spin_lock_bh(&mdev->iboe.lock);
897 	ndev = mdev->iboe.netdevs[mqp->port - 1];
898 	if (ndev)
899 		dev_hold(ndev);
900 	spin_unlock_bh(&mdev->iboe.lock);
901 
902 	if (ndev) {
903 		ret = 1;
904 		dev_put(ndev);
905 	}
906 
907 	return ret;
908 }
909 
910 struct mlx4_ib_steering {
911 	struct list_head list;
912 	struct mlx4_flow_reg_id reg_id;
913 	union ib_gid gid;
914 };
915 
916 static int parse_flow_attr(struct mlx4_dev *dev,
917 			   u32 qp_num,
918 			   union ib_flow_spec *ib_spec,
919 			   struct _rule_hw *mlx4_spec)
920 {
921 	enum mlx4_net_trans_rule_id type;
922 
923 	switch (ib_spec->type) {
924 	case IB_FLOW_SPEC_ETH:
925 		type = MLX4_NET_TRANS_RULE_ID_ETH;
926 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
927 		       ETH_ALEN);
928 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
929 		       ETH_ALEN);
930 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
931 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
932 		break;
933 	case IB_FLOW_SPEC_IB:
934 		type = MLX4_NET_TRANS_RULE_ID_IB;
935 		mlx4_spec->ib.l3_qpn =
936 			cpu_to_be32(qp_num);
937 		mlx4_spec->ib.qpn_mask =
938 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
939 		break;
940 
941 
942 	case IB_FLOW_SPEC_IPV4:
943 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
944 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
945 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
946 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
947 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
948 		break;
949 
950 	case IB_FLOW_SPEC_TCP:
951 	case IB_FLOW_SPEC_UDP:
952 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
953 					MLX4_NET_TRANS_RULE_ID_TCP :
954 					MLX4_NET_TRANS_RULE_ID_UDP;
955 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
956 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
957 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
958 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
959 		break;
960 
961 	default:
962 		return -EINVAL;
963 	}
964 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
965 	    mlx4_hw_rule_sz(dev, type) < 0)
966 		return -EINVAL;
967 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
968 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
969 	return mlx4_hw_rule_sz(dev, type);
970 }
971 
972 struct default_rules {
973 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
974 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
975 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
976 	__u8  link_layer;
977 };
978 static const struct default_rules default_table[] = {
979 	{
980 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
981 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
982 		.rules_create_list = {IB_FLOW_SPEC_IB},
983 		.link_layer = IB_LINK_LAYER_INFINIBAND
984 	}
985 };
986 
987 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
988 					 struct ib_flow_attr *flow_attr)
989 {
990 	int i, j, k;
991 	void *ib_flow;
992 	const struct default_rules *pdefault_rules = default_table;
993 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
994 
995 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
996 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
997 		memset(&field_types, 0, sizeof(field_types));
998 
999 		if (link_layer != pdefault_rules->link_layer)
1000 			continue;
1001 
1002 		ib_flow = flow_attr + 1;
1003 		/* we assume the specs are sorted */
1004 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1005 		     j < flow_attr->num_of_specs; k++) {
1006 			union ib_flow_spec *current_flow =
1007 				(union ib_flow_spec *)ib_flow;
1008 
1009 			/* same layer but different type */
1010 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1011 			     (pdefault_rules->mandatory_fields[k] &
1012 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1013 			    (current_flow->type !=
1014 			     pdefault_rules->mandatory_fields[k]))
1015 				goto out;
1016 
1017 			/* same layer, try match next one */
1018 			if (current_flow->type ==
1019 			    pdefault_rules->mandatory_fields[k]) {
1020 				j++;
1021 				ib_flow +=
1022 					((union ib_flow_spec *)ib_flow)->size;
1023 			}
1024 		}
1025 
1026 		ib_flow = flow_attr + 1;
1027 		for (j = 0; j < flow_attr->num_of_specs;
1028 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1029 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1030 				/* same layer and same type */
1031 				if (((union ib_flow_spec *)ib_flow)->type ==
1032 				    pdefault_rules->mandatory_not_fields[k])
1033 					goto out;
1034 
1035 		return i;
1036 	}
1037 out:
1038 	return -1;
1039 }
1040 
1041 static int __mlx4_ib_create_default_rules(
1042 		struct mlx4_ib_dev *mdev,
1043 		struct ib_qp *qp,
1044 		const struct default_rules *pdefault_rules,
1045 		struct _rule_hw *mlx4_spec) {
1046 	int size = 0;
1047 	int i;
1048 
1049 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1050 		int ret;
1051 		union ib_flow_spec ib_spec;
1052 		switch (pdefault_rules->rules_create_list[i]) {
1053 		case 0:
1054 			/* no rule */
1055 			continue;
1056 		case IB_FLOW_SPEC_IB:
1057 			ib_spec.type = IB_FLOW_SPEC_IB;
1058 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1059 
1060 			break;
1061 		default:
1062 			/* invalid rule */
1063 			return -EINVAL;
1064 		}
1065 		/* We must put empty rule, qpn is being ignored */
1066 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1067 				      mlx4_spec);
1068 		if (ret < 0) {
1069 			pr_info("invalid parsing\n");
1070 			return -EINVAL;
1071 		}
1072 
1073 		mlx4_spec = (void *)mlx4_spec + ret;
1074 		size += ret;
1075 	}
1076 	return size;
1077 }
1078 
1079 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1080 			  int domain,
1081 			  enum mlx4_net_trans_promisc_mode flow_type,
1082 			  u64 *reg_id)
1083 {
1084 	int ret, i;
1085 	int size = 0;
1086 	void *ib_flow;
1087 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1088 	struct mlx4_cmd_mailbox *mailbox;
1089 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1090 	int default_flow;
1091 
1092 	static const u16 __mlx4_domain[] = {
1093 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1094 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1095 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1096 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1097 	};
1098 
1099 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1100 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1101 		return -EINVAL;
1102 	}
1103 
1104 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1105 		pr_err("Invalid domain value %d\n", domain);
1106 		return -EINVAL;
1107 	}
1108 
1109 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1110 		return -EINVAL;
1111 
1112 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1113 	if (IS_ERR(mailbox))
1114 		return PTR_ERR(mailbox);
1115 	ctrl = mailbox->buf;
1116 
1117 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1118 				 flow_attr->priority);
1119 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1120 	ctrl->port = flow_attr->port;
1121 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1122 
1123 	ib_flow = flow_attr + 1;
1124 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1125 	/* Add default flows */
1126 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1127 	if (default_flow >= 0) {
1128 		ret = __mlx4_ib_create_default_rules(
1129 				mdev, qp, default_table + default_flow,
1130 				mailbox->buf + size);
1131 		if (ret < 0) {
1132 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1133 			return -EINVAL;
1134 		}
1135 		size += ret;
1136 	}
1137 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1138 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1139 				      mailbox->buf + size);
1140 		if (ret < 0) {
1141 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1142 			return -EINVAL;
1143 		}
1144 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1145 		size += ret;
1146 	}
1147 
1148 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1149 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1150 			   MLX4_CMD_WRAPPED);
1151 	if (ret == -ENOMEM)
1152 		pr_err("mcg table is full. Fail to register network rule.\n");
1153 	else if (ret == -ENXIO)
1154 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1155 	else if (ret)
1156 		pr_err("Invalid argumant. Fail to register network rule.\n");
1157 
1158 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1159 	return ret;
1160 }
1161 
1162 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1163 {
1164 	int err;
1165 	err = mlx4_cmd(dev, reg_id, 0, 0,
1166 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1167 		       MLX4_CMD_WRAPPED);
1168 	if (err)
1169 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1170 		       reg_id);
1171 	return err;
1172 }
1173 
1174 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1175 				    u64 *reg_id)
1176 {
1177 	void *ib_flow;
1178 	union ib_flow_spec *ib_spec;
1179 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1180 	int err = 0;
1181 
1182 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1183 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1184 		return 0; /* do nothing */
1185 
1186 	ib_flow = flow_attr + 1;
1187 	ib_spec = (union ib_flow_spec *)ib_flow;
1188 
1189 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1190 		return 0; /* do nothing */
1191 
1192 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1193 				    flow_attr->port, qp->qp_num,
1194 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1195 				    reg_id);
1196 	return err;
1197 }
1198 
1199 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1200 				    struct ib_flow_attr *flow_attr,
1201 				    int domain)
1202 {
1203 	int err = 0, i = 0, j = 0;
1204 	struct mlx4_ib_flow *mflow;
1205 	enum mlx4_net_trans_promisc_mode type[2];
1206 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1207 	int is_bonded = mlx4_is_bonded(dev);
1208 
1209 	memset(type, 0, sizeof(type));
1210 
1211 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1212 	if (!mflow) {
1213 		err = -ENOMEM;
1214 		goto err_free;
1215 	}
1216 
1217 	switch (flow_attr->type) {
1218 	case IB_FLOW_ATTR_NORMAL:
1219 		type[0] = MLX4_FS_REGULAR;
1220 		break;
1221 
1222 	case IB_FLOW_ATTR_ALL_DEFAULT:
1223 		type[0] = MLX4_FS_ALL_DEFAULT;
1224 		break;
1225 
1226 	case IB_FLOW_ATTR_MC_DEFAULT:
1227 		type[0] = MLX4_FS_MC_DEFAULT;
1228 		break;
1229 
1230 	case IB_FLOW_ATTR_SNIFFER:
1231 		type[0] = MLX4_FS_UC_SNIFFER;
1232 		type[1] = MLX4_FS_MC_SNIFFER;
1233 		break;
1234 
1235 	default:
1236 		err = -EINVAL;
1237 		goto err_free;
1238 	}
1239 
1240 	while (i < ARRAY_SIZE(type) && type[i]) {
1241 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1242 					    &mflow->reg_id[i].id);
1243 		if (err)
1244 			goto err_create_flow;
1245 		if (is_bonded) {
1246 			/* Application always sees one port so the mirror rule
1247 			 * must be on port #2
1248 			 */
1249 			flow_attr->port = 2;
1250 			err = __mlx4_ib_create_flow(qp, flow_attr,
1251 						    domain, type[j],
1252 						    &mflow->reg_id[j].mirror);
1253 			flow_attr->port = 1;
1254 			if (err)
1255 				goto err_create_flow;
1256 			j++;
1257 		}
1258 
1259 		i++;
1260 	}
1261 
1262 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1263 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1264 					       &mflow->reg_id[i].id);
1265 		if (err)
1266 			goto err_create_flow;
1267 
1268 		if (is_bonded) {
1269 			flow_attr->port = 2;
1270 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1271 						       &mflow->reg_id[j].mirror);
1272 			flow_attr->port = 1;
1273 			if (err)
1274 				goto err_create_flow;
1275 			j++;
1276 		}
1277 		/* function to create mirror rule */
1278 		i++;
1279 	}
1280 
1281 	return &mflow->ibflow;
1282 
1283 err_create_flow:
1284 	while (i) {
1285 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1286 					     mflow->reg_id[i].id);
1287 		i--;
1288 	}
1289 
1290 	while (j) {
1291 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1292 					     mflow->reg_id[j].mirror);
1293 		j--;
1294 	}
1295 err_free:
1296 	kfree(mflow);
1297 	return ERR_PTR(err);
1298 }
1299 
1300 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1301 {
1302 	int err, ret = 0;
1303 	int i = 0;
1304 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1305 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1306 
1307 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1308 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1309 		if (err)
1310 			ret = err;
1311 		if (mflow->reg_id[i].mirror) {
1312 			err = __mlx4_ib_destroy_flow(mdev->dev,
1313 						     mflow->reg_id[i].mirror);
1314 			if (err)
1315 				ret = err;
1316 		}
1317 		i++;
1318 	}
1319 
1320 	kfree(mflow);
1321 	return ret;
1322 }
1323 
1324 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1325 {
1326 	int err;
1327 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1328 	struct mlx4_dev	*dev = mdev->dev;
1329 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1330 	struct mlx4_ib_steering *ib_steering = NULL;
1331 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1332 	struct mlx4_flow_reg_id	reg_id;
1333 
1334 	if (mdev->dev->caps.steering_mode ==
1335 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1336 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1337 		if (!ib_steering)
1338 			return -ENOMEM;
1339 	}
1340 
1341 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1342 				    !!(mqp->flags &
1343 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1344 				    prot, &reg_id.id);
1345 	if (err) {
1346 		pr_err("multicast attach op failed, err %d\n", err);
1347 		goto err_malloc;
1348 	}
1349 
1350 	reg_id.mirror = 0;
1351 	if (mlx4_is_bonded(dev)) {
1352 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1353 					    (mqp->port == 1) ? 2 : 1,
1354 					    !!(mqp->flags &
1355 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1356 					    prot, &reg_id.mirror);
1357 		if (err)
1358 			goto err_add;
1359 	}
1360 
1361 	err = add_gid_entry(ibqp, gid);
1362 	if (err)
1363 		goto err_add;
1364 
1365 	if (ib_steering) {
1366 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1367 		ib_steering->reg_id = reg_id;
1368 		mutex_lock(&mqp->mutex);
1369 		list_add(&ib_steering->list, &mqp->steering_rules);
1370 		mutex_unlock(&mqp->mutex);
1371 	}
1372 	return 0;
1373 
1374 err_add:
1375 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1376 			      prot, reg_id.id);
1377 	if (reg_id.mirror)
1378 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1379 				      prot, reg_id.mirror);
1380 err_malloc:
1381 	kfree(ib_steering);
1382 
1383 	return err;
1384 }
1385 
1386 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1387 {
1388 	struct mlx4_ib_gid_entry *ge;
1389 	struct mlx4_ib_gid_entry *tmp;
1390 	struct mlx4_ib_gid_entry *ret = NULL;
1391 
1392 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1393 		if (!memcmp(raw, ge->gid.raw, 16)) {
1394 			ret = ge;
1395 			break;
1396 		}
1397 	}
1398 
1399 	return ret;
1400 }
1401 
1402 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1403 {
1404 	int err;
1405 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1406 	struct mlx4_dev *dev = mdev->dev;
1407 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1408 	struct net_device *ndev;
1409 	struct mlx4_ib_gid_entry *ge;
1410 	struct mlx4_flow_reg_id reg_id = {0, 0};
1411 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1412 
1413 	if (mdev->dev->caps.steering_mode ==
1414 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1415 		struct mlx4_ib_steering *ib_steering;
1416 
1417 		mutex_lock(&mqp->mutex);
1418 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1419 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1420 				list_del(&ib_steering->list);
1421 				break;
1422 			}
1423 		}
1424 		mutex_unlock(&mqp->mutex);
1425 		if (&ib_steering->list == &mqp->steering_rules) {
1426 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1427 			return -EINVAL;
1428 		}
1429 		reg_id = ib_steering->reg_id;
1430 		kfree(ib_steering);
1431 	}
1432 
1433 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1434 				    prot, reg_id.id);
1435 	if (err)
1436 		return err;
1437 
1438 	if (mlx4_is_bonded(dev)) {
1439 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1440 					    prot, reg_id.mirror);
1441 		if (err)
1442 			return err;
1443 	}
1444 
1445 	mutex_lock(&mqp->mutex);
1446 	ge = find_gid_entry(mqp, gid->raw);
1447 	if (ge) {
1448 		spin_lock_bh(&mdev->iboe.lock);
1449 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1450 		if (ndev)
1451 			dev_hold(ndev);
1452 		spin_unlock_bh(&mdev->iboe.lock);
1453 		if (ndev)
1454 			dev_put(ndev);
1455 		list_del(&ge->list);
1456 		kfree(ge);
1457 	} else
1458 		pr_warn("could not find mgid entry\n");
1459 
1460 	mutex_unlock(&mqp->mutex);
1461 
1462 	return 0;
1463 }
1464 
1465 static int init_node_data(struct mlx4_ib_dev *dev)
1466 {
1467 	struct ib_smp *in_mad  = NULL;
1468 	struct ib_smp *out_mad = NULL;
1469 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1470 	int err = -ENOMEM;
1471 
1472 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1473 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1474 	if (!in_mad || !out_mad)
1475 		goto out;
1476 
1477 	init_query_mad(in_mad);
1478 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1479 	if (mlx4_is_master(dev->dev))
1480 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1481 
1482 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1483 	if (err)
1484 		goto out;
1485 
1486 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1487 
1488 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1489 
1490 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1491 	if (err)
1492 		goto out;
1493 
1494 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1495 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1496 
1497 out:
1498 	kfree(in_mad);
1499 	kfree(out_mad);
1500 	return err;
1501 }
1502 
1503 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1504 			char *buf)
1505 {
1506 	struct mlx4_ib_dev *dev =
1507 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1508 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1509 }
1510 
1511 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1512 			   char *buf)
1513 {
1514 	struct mlx4_ib_dev *dev =
1515 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1516 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1517 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1518 		       (int) dev->dev->caps.fw_ver & 0xffff);
1519 }
1520 
1521 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1522 			char *buf)
1523 {
1524 	struct mlx4_ib_dev *dev =
1525 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1526 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1527 }
1528 
1529 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1530 			  char *buf)
1531 {
1532 	struct mlx4_ib_dev *dev =
1533 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1534 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1535 		       dev->dev->board_id);
1536 }
1537 
1538 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1539 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1540 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1541 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1542 
1543 static struct device_attribute *mlx4_class_attributes[] = {
1544 	&dev_attr_hw_rev,
1545 	&dev_attr_fw_ver,
1546 	&dev_attr_hca_type,
1547 	&dev_attr_board_id
1548 };
1549 
1550 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1551 				     struct net_device *dev)
1552 {
1553 	memcpy(eui, dev->dev_addr, 3);
1554 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1555 	if (vlan_id < 0x1000) {
1556 		eui[3] = vlan_id >> 8;
1557 		eui[4] = vlan_id & 0xff;
1558 	} else {
1559 		eui[3] = 0xff;
1560 		eui[4] = 0xfe;
1561 	}
1562 	eui[0] ^= 2;
1563 }
1564 
1565 static void update_gids_task(struct work_struct *work)
1566 {
1567 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1568 	struct mlx4_cmd_mailbox *mailbox;
1569 	union ib_gid *gids;
1570 	int err;
1571 	struct mlx4_dev	*dev = gw->dev->dev;
1572 	int is_bonded = mlx4_is_bonded(dev);
1573 
1574 	if (!gw->dev->ib_active)
1575 		return;
1576 
1577 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1578 	if (IS_ERR(mailbox)) {
1579 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1580 		return;
1581 	}
1582 
1583 	gids = mailbox->buf;
1584 	memcpy(gids, gw->gids, sizeof gw->gids);
1585 
1586 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1587 		       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1588 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1589 	if (err)
1590 		pr_warn("set port command failed\n");
1591 	else
1592 		if ((gw->port == 1) || !is_bonded)
1593 			mlx4_ib_dispatch_event(gw->dev,
1594 					       is_bonded ? 1 : gw->port,
1595 					       IB_EVENT_GID_CHANGE);
1596 
1597 	mlx4_free_cmd_mailbox(dev, mailbox);
1598 	kfree(gw);
1599 }
1600 
1601 static void reset_gids_task(struct work_struct *work)
1602 {
1603 	struct update_gid_work *gw =
1604 			container_of(work, struct update_gid_work, work);
1605 	struct mlx4_cmd_mailbox *mailbox;
1606 	union ib_gid *gids;
1607 	int err;
1608 	struct mlx4_dev	*dev = gw->dev->dev;
1609 
1610 	if (!gw->dev->ib_active)
1611 		return;
1612 
1613 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1614 	if (IS_ERR(mailbox)) {
1615 		pr_warn("reset gid table failed\n");
1616 		goto free;
1617 	}
1618 
1619 	gids = mailbox->buf;
1620 	memcpy(gids, gw->gids, sizeof(gw->gids));
1621 
1622 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1623 				    IB_LINK_LAYER_ETHERNET) {
1624 		err = mlx4_cmd(dev, mailbox->dma,
1625 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1626 			       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1627 			       MLX4_CMD_TIME_CLASS_B,
1628 			       MLX4_CMD_WRAPPED);
1629 		if (err)
1630 			pr_warn("set port %d command failed\n", gw->port);
1631 	}
1632 
1633 	mlx4_free_cmd_mailbox(dev, mailbox);
1634 free:
1635 	kfree(gw);
1636 }
1637 
1638 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1639 			    union ib_gid *gid, int clear,
1640 			    int default_gid)
1641 {
1642 	struct update_gid_work *work;
1643 	int i;
1644 	int need_update = 0;
1645 	int free = -1;
1646 	int found = -1;
1647 	int max_gids;
1648 
1649 	if (default_gid) {
1650 		free = 0;
1651 	} else {
1652 		max_gids = dev->dev->caps.gid_table_len[port];
1653 		for (i = 1; i < max_gids; ++i) {
1654 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1655 				    sizeof(*gid)))
1656 				found = i;
1657 
1658 			if (clear) {
1659 				if (found >= 0) {
1660 					need_update = 1;
1661 					dev->iboe.gid_table[port - 1][found] =
1662 						zgid;
1663 					break;
1664 				}
1665 			} else {
1666 				if (found >= 0)
1667 					break;
1668 
1669 				if (free < 0 &&
1670 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1671 					    &zgid, sizeof(*gid)))
1672 					free = i;
1673 			}
1674 		}
1675 	}
1676 
1677 	if (found == -1 && !clear && free >= 0) {
1678 		dev->iboe.gid_table[port - 1][free] = *gid;
1679 		need_update = 1;
1680 	}
1681 
1682 	if (!need_update)
1683 		return 0;
1684 
1685 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1686 	if (!work)
1687 		return -ENOMEM;
1688 
1689 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1690 	INIT_WORK(&work->work, update_gids_task);
1691 	work->port = port;
1692 	work->dev = dev;
1693 	queue_work(wq, &work->work);
1694 
1695 	return 0;
1696 }
1697 
1698 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1699 {
1700 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1701 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1702 }
1703 
1704 
1705 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1706 {
1707 	struct update_gid_work *work;
1708 
1709 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1710 	if (!work)
1711 		return -ENOMEM;
1712 
1713 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1714 	memset(work->gids, 0, sizeof(work->gids));
1715 	INIT_WORK(&work->work, reset_gids_task);
1716 	work->dev = dev;
1717 	work->port = port;
1718 	queue_work(wq, &work->work);
1719 	return 0;
1720 }
1721 
1722 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1723 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1724 {
1725 	struct mlx4_ib_iboe *iboe;
1726 	int port = 0;
1727 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1728 				rdma_vlan_dev_real_dev(event_netdev) :
1729 				event_netdev;
1730 	union ib_gid default_gid;
1731 
1732 	mlx4_make_default_gid(real_dev, &default_gid);
1733 
1734 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1735 		return 0;
1736 
1737 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1738 		return 0;
1739 
1740 	if ((real_dev != event_netdev) &&
1741 	    (event == NETDEV_DOWN) &&
1742 	    rdma_link_local_addr((struct in6_addr *)gid))
1743 		return 0;
1744 
1745 	iboe = &ibdev->iboe;
1746 	spin_lock_bh(&iboe->lock);
1747 
1748 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1749 		if ((netif_is_bond_master(real_dev) &&
1750 		     (real_dev == iboe->masters[port - 1])) ||
1751 		     (!netif_is_bond_master(real_dev) &&
1752 		     (real_dev == iboe->netdevs[port - 1])))
1753 			update_gid_table(ibdev, port, gid,
1754 					 event == NETDEV_DOWN, 0);
1755 
1756 	spin_unlock_bh(&iboe->lock);
1757 	return 0;
1758 
1759 }
1760 
1761 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1762 			       struct mlx4_ib_dev *ibdev)
1763 {
1764 	u8 port = 0;
1765 	struct mlx4_ib_iboe *iboe;
1766 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1767 				rdma_vlan_dev_real_dev(dev) : dev;
1768 
1769 	iboe = &ibdev->iboe;
1770 
1771 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1772 		if ((netif_is_bond_master(real_dev) &&
1773 		     (real_dev == iboe->masters[port - 1])) ||
1774 		     (!netif_is_bond_master(real_dev) &&
1775 		     (real_dev == iboe->netdevs[port - 1])))
1776 			break;
1777 
1778 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1779 		return 0;
1780 	else
1781 		return port;
1782 }
1783 
1784 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1785 				void *ptr)
1786 {
1787 	struct mlx4_ib_dev *ibdev;
1788 	struct in_ifaddr *ifa = ptr;
1789 	union ib_gid gid;
1790 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1791 
1792 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1793 
1794 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1795 
1796 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1797 	return NOTIFY_DONE;
1798 }
1799 
1800 #if IS_ENABLED(CONFIG_IPV6)
1801 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1802 				void *ptr)
1803 {
1804 	struct mlx4_ib_dev *ibdev;
1805 	struct inet6_ifaddr *ifa = ptr;
1806 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1807 	struct net_device *event_netdev = ifa->idev->dev;
1808 
1809 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1810 
1811 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1812 	return NOTIFY_DONE;
1813 }
1814 #endif
1815 
1816 #define MLX4_IB_INVALID_MAC	((u64)-1)
1817 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1818 			       struct net_device *dev,
1819 			       int port)
1820 {
1821 	u64 new_smac = 0;
1822 	u64 release_mac = MLX4_IB_INVALID_MAC;
1823 	struct mlx4_ib_qp *qp;
1824 
1825 	read_lock(&dev_base_lock);
1826 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1827 	read_unlock(&dev_base_lock);
1828 
1829 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1830 
1831 	/* no need for update QP1 and mac registration in non-SRIOV */
1832 	if (!mlx4_is_mfunc(ibdev->dev))
1833 		return;
1834 
1835 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1836 	qp = ibdev->qp1_proxy[port - 1];
1837 	if (qp) {
1838 		int new_smac_index;
1839 		u64 old_smac;
1840 		struct mlx4_update_qp_params update_params;
1841 
1842 		mutex_lock(&qp->mutex);
1843 		old_smac = qp->pri.smac;
1844 		if (new_smac == old_smac)
1845 			goto unlock;
1846 
1847 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1848 
1849 		if (new_smac_index < 0)
1850 			goto unlock;
1851 
1852 		update_params.smac_index = new_smac_index;
1853 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1854 				   &update_params)) {
1855 			release_mac = new_smac;
1856 			goto unlock;
1857 		}
1858 		/* if old port was zero, no mac was yet registered for this QP */
1859 		if (qp->pri.smac_port)
1860 			release_mac = old_smac;
1861 		qp->pri.smac = new_smac;
1862 		qp->pri.smac_port = port;
1863 		qp->pri.smac_index = new_smac_index;
1864 	}
1865 
1866 unlock:
1867 	if (release_mac != MLX4_IB_INVALID_MAC)
1868 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1869 	if (qp)
1870 		mutex_unlock(&qp->mutex);
1871 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1872 }
1873 
1874 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1875 				 struct mlx4_ib_dev *ibdev, u8 port)
1876 {
1877 	struct in_device *in_dev;
1878 #if IS_ENABLED(CONFIG_IPV6)
1879 	struct inet6_dev *in6_dev;
1880 	union ib_gid  *pgid;
1881 	struct inet6_ifaddr *ifp;
1882 	union ib_gid default_gid;
1883 #endif
1884 	union ib_gid gid;
1885 
1886 
1887 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1888 		return;
1889 
1890 	/* IPv4 gids */
1891 	in_dev = in_dev_get(dev);
1892 	if (in_dev) {
1893 		for_ifa(in_dev) {
1894 			/*ifa->ifa_address;*/
1895 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1896 					       (struct in6_addr *)&gid);
1897 			update_gid_table(ibdev, port, &gid, 0, 0);
1898 		}
1899 		endfor_ifa(in_dev);
1900 		in_dev_put(in_dev);
1901 	}
1902 #if IS_ENABLED(CONFIG_IPV6)
1903 	mlx4_make_default_gid(dev, &default_gid);
1904 	/* IPv6 gids */
1905 	in6_dev = in6_dev_get(dev);
1906 	if (in6_dev) {
1907 		read_lock_bh(&in6_dev->lock);
1908 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1909 			pgid = (union ib_gid *)&ifp->addr;
1910 			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1911 				continue;
1912 			update_gid_table(ibdev, port, pgid, 0, 0);
1913 		}
1914 		read_unlock_bh(&in6_dev->lock);
1915 		in6_dev_put(in6_dev);
1916 	}
1917 #endif
1918 }
1919 
1920 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1921 				 struct  net_device *dev, u8 port)
1922 {
1923 	union ib_gid gid;
1924 	mlx4_make_default_gid(dev, &gid);
1925 	update_gid_table(ibdev, port, &gid, 0, 1);
1926 }
1927 
1928 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1929 {
1930 	struct	net_device *dev;
1931 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1932 	int i;
1933 	int err = 0;
1934 
1935 	for (i = 1; i <= ibdev->num_ports; ++i) {
1936 		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1937 		    IB_LINK_LAYER_ETHERNET) {
1938 			err = reset_gid_table(ibdev, i);
1939 			if (err)
1940 				goto out;
1941 		}
1942 	}
1943 
1944 	read_lock(&dev_base_lock);
1945 	spin_lock_bh(&iboe->lock);
1946 
1947 	for_each_netdev(&init_net, dev) {
1948 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1949 		/* port will be non-zero only for ETH ports */
1950 		if (port) {
1951 			mlx4_ib_set_default_gid(ibdev, dev, port);
1952 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1953 		}
1954 	}
1955 
1956 	spin_unlock_bh(&iboe->lock);
1957 	read_unlock(&dev_base_lock);
1958 out:
1959 	return err;
1960 }
1961 
1962 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1963 				 struct net_device *dev,
1964 				 unsigned long event)
1965 
1966 {
1967 	struct mlx4_ib_iboe *iboe;
1968 	int update_qps_port = -1;
1969 	int port;
1970 
1971 	iboe = &ibdev->iboe;
1972 
1973 	spin_lock_bh(&iboe->lock);
1974 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1975 		enum ib_port_state	port_state = IB_PORT_NOP;
1976 		struct net_device *old_master = iboe->masters[port - 1];
1977 		struct net_device *curr_netdev;
1978 		struct net_device *curr_master;
1979 
1980 		iboe->netdevs[port - 1] =
1981 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1982 		if (iboe->netdevs[port - 1])
1983 			mlx4_ib_set_default_gid(ibdev,
1984 						iboe->netdevs[port - 1], port);
1985 		curr_netdev = iboe->netdevs[port - 1];
1986 
1987 		if (iboe->netdevs[port - 1] &&
1988 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1989 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1990 				iboe->netdevs[port - 1]);
1991 		} else {
1992 			iboe->masters[port - 1] = NULL;
1993 		}
1994 		curr_master = iboe->masters[port - 1];
1995 
1996 		if (dev == iboe->netdevs[port - 1] &&
1997 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1998 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1999 			update_qps_port = port;
2000 
2001 		if (curr_netdev) {
2002 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
2003 						IB_PORT_ACTIVE : IB_PORT_DOWN;
2004 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
2005 			if (curr_master) {
2006 				/* if using bonding/team and a slave port is down, we
2007 				 * don't want the bond IP based gids in the table since
2008 				 * flows that select port by gid may get the down port.
2009 				*/
2010 				if (port_state == IB_PORT_DOWN &&
2011 				    !mlx4_is_bonded(ibdev->dev)) {
2012 					reset_gid_table(ibdev, port);
2013 					mlx4_ib_set_default_gid(ibdev,
2014 								curr_netdev,
2015 								port);
2016 				} else {
2017 					/* gids from the upper dev (bond/team)
2018 					 * should appear in port's gid table
2019 					*/
2020 					mlx4_ib_get_dev_addr(curr_master,
2021 							     ibdev, port);
2022 				}
2023 			}
2024 			/* if bonding is used it is possible that we add it to
2025 			 * masters only after IP address is assigned to the
2026 			 * net bonding interface.
2027 			*/
2028 			if (curr_master && (old_master != curr_master)) {
2029 				reset_gid_table(ibdev, port);
2030 				mlx4_ib_set_default_gid(ibdev,
2031 							curr_netdev, port);
2032 				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
2033 			}
2034 
2035 			if (!curr_master && (old_master != curr_master)) {
2036 				reset_gid_table(ibdev, port);
2037 				mlx4_ib_set_default_gid(ibdev,
2038 							curr_netdev, port);
2039 				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
2040 			}
2041 		} else {
2042 			reset_gid_table(ibdev, port);
2043 		}
2044 	}
2045 
2046 	spin_unlock_bh(&iboe->lock);
2047 
2048 	if (update_qps_port > 0)
2049 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2050 }
2051 
2052 static int mlx4_ib_netdev_event(struct notifier_block *this,
2053 				unsigned long event, void *ptr)
2054 {
2055 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2056 	struct mlx4_ib_dev *ibdev;
2057 
2058 	if (!net_eq(dev_net(dev), &init_net))
2059 		return NOTIFY_DONE;
2060 
2061 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2062 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2063 
2064 	return NOTIFY_DONE;
2065 }
2066 
2067 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2068 {
2069 	int port;
2070 	int slave;
2071 	int i;
2072 
2073 	if (mlx4_is_master(ibdev->dev)) {
2074 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2075 		     ++slave) {
2076 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2077 				for (i = 0;
2078 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2079 				     ++i) {
2080 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2081 					/* master has the identity virt2phys pkey mapping */
2082 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2083 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2084 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2085 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2086 				}
2087 			}
2088 		}
2089 		/* initialize pkey cache */
2090 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2091 			for (i = 0;
2092 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2093 			     ++i)
2094 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2095 					(i) ? 0 : 0xFFFF;
2096 		}
2097 	}
2098 }
2099 
2100 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2101 {
2102 	int i, j, eq = 0, total_eqs = 0;
2103 
2104 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2105 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2106 	if (!ibdev->eq_table)
2107 		return;
2108 
2109 	for (i = 1; i <= dev->caps.num_ports; i++) {
2110 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2111 		     j++, total_eqs++) {
2112 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2113 				continue;
2114 			ibdev->eq_table[eq] = total_eqs;
2115 			if (!mlx4_assign_eq(dev, i,
2116 					    &ibdev->eq_table[eq]))
2117 				eq++;
2118 			else
2119 				ibdev->eq_table[eq] = -1;
2120 		}
2121 	}
2122 
2123 	for (i = eq; i < dev->caps.num_comp_vectors;
2124 	     ibdev->eq_table[i++] = -1)
2125 		;
2126 
2127 	/* Advertise the new number of EQs to clients */
2128 	ibdev->ib_dev.num_comp_vectors = eq;
2129 }
2130 
2131 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2132 {
2133 	int i;
2134 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2135 
2136 	/* no eqs were allocated */
2137 	if (!ibdev->eq_table)
2138 		return;
2139 
2140 	/* Reset the advertised EQ number */
2141 	ibdev->ib_dev.num_comp_vectors = 0;
2142 
2143 	for (i = 0; i < total_eqs; i++)
2144 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2145 
2146 	kfree(ibdev->eq_table);
2147 	ibdev->eq_table = NULL;
2148 }
2149 
2150 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2151 			       struct ib_port_immutable *immutable)
2152 {
2153 	struct ib_port_attr attr;
2154 	int err;
2155 
2156 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2157 	if (err)
2158 		return err;
2159 
2160 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2161 	immutable->gid_tbl_len = attr.gid_tbl_len;
2162 
2163 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2164 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2165 	else
2166 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2167 
2168 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2169 
2170 	return 0;
2171 }
2172 
2173 static void *mlx4_ib_add(struct mlx4_dev *dev)
2174 {
2175 	struct mlx4_ib_dev *ibdev;
2176 	int num_ports = 0;
2177 	int i, j;
2178 	int err;
2179 	struct mlx4_ib_iboe *iboe;
2180 	int ib_num_ports = 0;
2181 	int num_req_counters;
2182 	int allocated;
2183 	u32 counter_index;
2184 
2185 	pr_info_once("%s", mlx4_ib_version);
2186 
2187 	num_ports = 0;
2188 	mlx4_foreach_ib_transport_port(i, dev)
2189 		num_ports++;
2190 
2191 	/* No point in registering a device with no ports... */
2192 	if (num_ports == 0)
2193 		return NULL;
2194 
2195 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2196 	if (!ibdev) {
2197 		dev_err(&dev->persist->pdev->dev,
2198 			"Device struct alloc failed\n");
2199 		return NULL;
2200 	}
2201 
2202 	iboe = &ibdev->iboe;
2203 
2204 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2205 		goto err_dealloc;
2206 
2207 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2208 		goto err_pd;
2209 
2210 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2211 				 PAGE_SIZE);
2212 	if (!ibdev->uar_map)
2213 		goto err_uar;
2214 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2215 
2216 	ibdev->dev = dev;
2217 	ibdev->bond_next_port	= 0;
2218 
2219 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2220 	ibdev->ib_dev.owner		= THIS_MODULE;
2221 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2222 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2223 	ibdev->num_ports		= num_ports;
2224 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2225 						1 : ibdev->num_ports;
2226 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2227 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2228 
2229 	if (dev->caps.userspace_caps)
2230 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2231 	else
2232 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2233 
2234 	ibdev->ib_dev.uverbs_cmd_mask	=
2235 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2236 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2237 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2238 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2239 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2240 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2241 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2242 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2243 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2244 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2245 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2246 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2247 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2248 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2249 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2250 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2251 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2252 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2253 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2254 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2255 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2256 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2257 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2258 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2259 
2260 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2261 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2262 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2263 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2264 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2265 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2266 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2267 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2268 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2269 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2270 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2271 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2272 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2273 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2274 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2275 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2276 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2277 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2278 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2279 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2280 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2281 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2282 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2283 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2284 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2285 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2286 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2287 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2288 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2289 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2290 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2291 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2292 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2293 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2294 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2295 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2296 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2297 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2298 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2299 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2300 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2301 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2302 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2303 
2304 	if (!mlx4_is_slave(ibdev->dev)) {
2305 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2306 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2307 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2308 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2309 	}
2310 
2311 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2312 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2313 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2314 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2315 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2316 
2317 		ibdev->ib_dev.uverbs_cmd_mask |=
2318 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2319 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2320 	}
2321 
2322 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2323 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2324 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2325 		ibdev->ib_dev.uverbs_cmd_mask |=
2326 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2327 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2328 	}
2329 
2330 	if (check_flow_steering_support(dev)) {
2331 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2332 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2333 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2334 
2335 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2336 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2337 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2338 	}
2339 
2340 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2341 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2342 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ);
2343 
2344 	mlx4_ib_alloc_eqs(dev, ibdev);
2345 
2346 	spin_lock_init(&iboe->lock);
2347 
2348 	if (init_node_data(ibdev))
2349 		goto err_map;
2350 
2351 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2352 	for (i = 0; i < num_req_counters; ++i) {
2353 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2354 		allocated = 0;
2355 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2356 						IB_LINK_LAYER_ETHERNET) {
2357 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2358 			/* if failed to allocate a new counter, use default */
2359 			if (err)
2360 				counter_index =
2361 					mlx4_get_default_counter_index(dev,
2362 								       i + 1);
2363 			else
2364 				allocated = 1;
2365 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2366 			counter_index = mlx4_get_default_counter_index(dev,
2367 								       i + 1);
2368 		}
2369 		ibdev->counters[i].index = counter_index;
2370 		ibdev->counters[i].allocated = allocated;
2371 		pr_info("counter index %d for port %d allocated %d\n",
2372 			counter_index, i + 1, allocated);
2373 	}
2374 	if (mlx4_is_bonded(dev))
2375 		for (i = 1; i < ibdev->num_ports ; ++i) {
2376 			ibdev->counters[i].index = ibdev->counters[0].index;
2377 			ibdev->counters[i].allocated = 0;
2378 		}
2379 
2380 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2381 		ib_num_ports++;
2382 
2383 	spin_lock_init(&ibdev->sm_lock);
2384 	mutex_init(&ibdev->cap_mask_mutex);
2385 	INIT_LIST_HEAD(&ibdev->qp_list);
2386 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2387 
2388 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2389 	    ib_num_ports) {
2390 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2391 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2392 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2393 					    &ibdev->steer_qpn_base, 0);
2394 		if (err)
2395 			goto err_counter;
2396 
2397 		ibdev->ib_uc_qpns_bitmap =
2398 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2399 				sizeof(long),
2400 				GFP_KERNEL);
2401 		if (!ibdev->ib_uc_qpns_bitmap) {
2402 			dev_err(&dev->persist->pdev->dev,
2403 				"bit map alloc failed\n");
2404 			goto err_steer_qp_release;
2405 		}
2406 
2407 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2408 
2409 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2410 				dev, ibdev->steer_qpn_base,
2411 				ibdev->steer_qpn_base +
2412 				ibdev->steer_qpn_count - 1);
2413 		if (err)
2414 			goto err_steer_free_bitmap;
2415 	}
2416 
2417 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2418 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2419 
2420 	if (ib_register_device(&ibdev->ib_dev, NULL))
2421 		goto err_steer_free_bitmap;
2422 
2423 	if (mlx4_ib_mad_init(ibdev))
2424 		goto err_reg;
2425 
2426 	if (mlx4_ib_init_sriov(ibdev))
2427 		goto err_mad;
2428 
2429 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2430 		if (!iboe->nb.notifier_call) {
2431 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2432 			err = register_netdevice_notifier(&iboe->nb);
2433 			if (err) {
2434 				iboe->nb.notifier_call = NULL;
2435 				goto err_notif;
2436 			}
2437 		}
2438 		if (!iboe->nb_inet.notifier_call) {
2439 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2440 			err = register_inetaddr_notifier(&iboe->nb_inet);
2441 			if (err) {
2442 				iboe->nb_inet.notifier_call = NULL;
2443 				goto err_notif;
2444 			}
2445 		}
2446 #if IS_ENABLED(CONFIG_IPV6)
2447 		if (!iboe->nb_inet6.notifier_call) {
2448 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2449 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2450 			if (err) {
2451 				iboe->nb_inet6.notifier_call = NULL;
2452 				goto err_notif;
2453 			}
2454 		}
2455 #endif
2456 		if (mlx4_ib_init_gid_table(ibdev))
2457 			goto err_notif;
2458 	}
2459 
2460 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2461 		if (device_create_file(&ibdev->ib_dev.dev,
2462 				       mlx4_class_attributes[j]))
2463 			goto err_notif;
2464 	}
2465 
2466 	ibdev->ib_active = true;
2467 
2468 	if (mlx4_is_mfunc(ibdev->dev))
2469 		init_pkeys(ibdev);
2470 
2471 	/* create paravirt contexts for any VFs which are active */
2472 	if (mlx4_is_master(ibdev->dev)) {
2473 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2474 			if (j == mlx4_master_func_num(ibdev->dev))
2475 				continue;
2476 			if (mlx4_is_slave_active(ibdev->dev, j))
2477 				do_slave_init(ibdev, j, 1);
2478 		}
2479 	}
2480 	return ibdev;
2481 
2482 err_notif:
2483 	if (ibdev->iboe.nb.notifier_call) {
2484 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2485 			pr_warn("failure unregistering notifier\n");
2486 		ibdev->iboe.nb.notifier_call = NULL;
2487 	}
2488 	if (ibdev->iboe.nb_inet.notifier_call) {
2489 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2490 			pr_warn("failure unregistering notifier\n");
2491 		ibdev->iboe.nb_inet.notifier_call = NULL;
2492 	}
2493 #if IS_ENABLED(CONFIG_IPV6)
2494 	if (ibdev->iboe.nb_inet6.notifier_call) {
2495 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2496 			pr_warn("failure unregistering notifier\n");
2497 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2498 	}
2499 #endif
2500 	flush_workqueue(wq);
2501 
2502 	mlx4_ib_close_sriov(ibdev);
2503 
2504 err_mad:
2505 	mlx4_ib_mad_cleanup(ibdev);
2506 
2507 err_reg:
2508 	ib_unregister_device(&ibdev->ib_dev);
2509 
2510 err_steer_free_bitmap:
2511 	kfree(ibdev->ib_uc_qpns_bitmap);
2512 
2513 err_steer_qp_release:
2514 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2515 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2516 				      ibdev->steer_qpn_count);
2517 err_counter:
2518 	for (i = 0; i < ibdev->num_ports; ++i) {
2519 		if (ibdev->counters[i].index != -1 &&
2520 		    ibdev->counters[i].allocated)
2521 			mlx4_counter_free(ibdev->dev,
2522 					  ibdev->counters[i].index);
2523 	}
2524 err_map:
2525 	iounmap(ibdev->uar_map);
2526 
2527 err_uar:
2528 	mlx4_uar_free(dev, &ibdev->priv_uar);
2529 
2530 err_pd:
2531 	mlx4_pd_free(dev, ibdev->priv_pdn);
2532 
2533 err_dealloc:
2534 	ib_dealloc_device(&ibdev->ib_dev);
2535 
2536 	return NULL;
2537 }
2538 
2539 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2540 {
2541 	int offset;
2542 
2543 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2544 
2545 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2546 					 dev->steer_qpn_count,
2547 					 get_count_order(count));
2548 	if (offset < 0)
2549 		return offset;
2550 
2551 	*qpn = dev->steer_qpn_base + offset;
2552 	return 0;
2553 }
2554 
2555 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2556 {
2557 	if (!qpn ||
2558 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2559 		return;
2560 
2561 	BUG_ON(qpn < dev->steer_qpn_base);
2562 
2563 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2564 			      qpn - dev->steer_qpn_base,
2565 			      get_count_order(count));
2566 }
2567 
2568 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2569 			 int is_attach)
2570 {
2571 	int err;
2572 	size_t flow_size;
2573 	struct ib_flow_attr *flow = NULL;
2574 	struct ib_flow_spec_ib *ib_spec;
2575 
2576 	if (is_attach) {
2577 		flow_size = sizeof(struct ib_flow_attr) +
2578 			    sizeof(struct ib_flow_spec_ib);
2579 		flow = kzalloc(flow_size, GFP_KERNEL);
2580 		if (!flow)
2581 			return -ENOMEM;
2582 		flow->port = mqp->port;
2583 		flow->num_of_specs = 1;
2584 		flow->size = flow_size;
2585 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2586 		ib_spec->type = IB_FLOW_SPEC_IB;
2587 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2588 		/* Add an empty rule for IB L2 */
2589 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2590 
2591 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2592 					    IB_FLOW_DOMAIN_NIC,
2593 					    MLX4_FS_REGULAR,
2594 					    &mqp->reg_id);
2595 	} else {
2596 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2597 	}
2598 	kfree(flow);
2599 	return err;
2600 }
2601 
2602 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2603 {
2604 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2605 	int p;
2606 
2607 	ibdev->ib_active = false;
2608 	flush_workqueue(wq);
2609 
2610 	mlx4_ib_close_sriov(ibdev);
2611 	mlx4_ib_mad_cleanup(ibdev);
2612 	ib_unregister_device(&ibdev->ib_dev);
2613 	if (ibdev->iboe.nb.notifier_call) {
2614 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2615 			pr_warn("failure unregistering notifier\n");
2616 		ibdev->iboe.nb.notifier_call = NULL;
2617 	}
2618 
2619 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2620 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2621 				      ibdev->steer_qpn_count);
2622 		kfree(ibdev->ib_uc_qpns_bitmap);
2623 	}
2624 
2625 	if (ibdev->iboe.nb_inet.notifier_call) {
2626 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2627 			pr_warn("failure unregistering notifier\n");
2628 		ibdev->iboe.nb_inet.notifier_call = NULL;
2629 	}
2630 #if IS_ENABLED(CONFIG_IPV6)
2631 	if (ibdev->iboe.nb_inet6.notifier_call) {
2632 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2633 			pr_warn("failure unregistering notifier\n");
2634 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2635 	}
2636 #endif
2637 
2638 	iounmap(ibdev->uar_map);
2639 	for (p = 0; p < ibdev->num_ports; ++p)
2640 		if (ibdev->counters[p].index != -1 &&
2641 		    ibdev->counters[p].allocated)
2642 			mlx4_counter_free(ibdev->dev, ibdev->counters[p].index);
2643 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2644 		mlx4_CLOSE_PORT(dev, p);
2645 
2646 	mlx4_ib_free_eqs(dev, ibdev);
2647 
2648 	mlx4_uar_free(dev, &ibdev->priv_uar);
2649 	mlx4_pd_free(dev, ibdev->priv_pdn);
2650 	ib_dealloc_device(&ibdev->ib_dev);
2651 }
2652 
2653 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2654 {
2655 	struct mlx4_ib_demux_work **dm = NULL;
2656 	struct mlx4_dev *dev = ibdev->dev;
2657 	int i;
2658 	unsigned long flags;
2659 	struct mlx4_active_ports actv_ports;
2660 	unsigned int ports;
2661 	unsigned int first_port;
2662 
2663 	if (!mlx4_is_master(dev))
2664 		return;
2665 
2666 	actv_ports = mlx4_get_active_ports(dev, slave);
2667 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2668 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2669 
2670 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2671 	if (!dm) {
2672 		pr_err("failed to allocate memory for tunneling qp update\n");
2673 		return;
2674 	}
2675 
2676 	for (i = 0; i < ports; i++) {
2677 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2678 		if (!dm[i]) {
2679 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2680 			while (--i >= 0)
2681 				kfree(dm[i]);
2682 			goto out;
2683 		}
2684 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2685 		dm[i]->port = first_port + i + 1;
2686 		dm[i]->slave = slave;
2687 		dm[i]->do_init = do_init;
2688 		dm[i]->dev = ibdev;
2689 	}
2690 	/* initialize or tear down tunnel QPs for the slave */
2691 	spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2692 	if (!ibdev->sriov.is_going_down) {
2693 		for (i = 0; i < ports; i++)
2694 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2695 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2696 	} else {
2697 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2698 		for (i = 0; i < ports; i++)
2699 			kfree(dm[i]);
2700 	}
2701 out:
2702 	kfree(dm);
2703 	return;
2704 }
2705 
2706 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2707 {
2708 	struct mlx4_ib_qp *mqp;
2709 	unsigned long flags_qp;
2710 	unsigned long flags_cq;
2711 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2712 	struct list_head    cq_notify_list;
2713 	struct mlx4_cq *mcq;
2714 	unsigned long flags;
2715 
2716 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2717 	INIT_LIST_HEAD(&cq_notify_list);
2718 
2719 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2720 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2721 
2722 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2723 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2724 		if (mqp->sq.tail != mqp->sq.head) {
2725 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2726 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2727 			if (send_mcq->mcq.comp &&
2728 			    mqp->ibqp.send_cq->comp_handler) {
2729 				if (!send_mcq->mcq.reset_notify_added) {
2730 					send_mcq->mcq.reset_notify_added = 1;
2731 					list_add_tail(&send_mcq->mcq.reset_notify,
2732 						      &cq_notify_list);
2733 				}
2734 			}
2735 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2736 		}
2737 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2738 		/* Now, handle the QP's receive queue */
2739 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2740 		/* no handling is needed for SRQ */
2741 		if (!mqp->ibqp.srq) {
2742 			if (mqp->rq.tail != mqp->rq.head) {
2743 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2744 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2745 				if (recv_mcq->mcq.comp &&
2746 				    mqp->ibqp.recv_cq->comp_handler) {
2747 					if (!recv_mcq->mcq.reset_notify_added) {
2748 						recv_mcq->mcq.reset_notify_added = 1;
2749 						list_add_tail(&recv_mcq->mcq.reset_notify,
2750 							      &cq_notify_list);
2751 					}
2752 				}
2753 				spin_unlock_irqrestore(&recv_mcq->lock,
2754 						       flags_cq);
2755 			}
2756 		}
2757 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2758 	}
2759 
2760 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2761 		mcq->comp(mcq);
2762 	}
2763 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2764 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2765 }
2766 
2767 static void handle_bonded_port_state_event(struct work_struct *work)
2768 {
2769 	struct ib_event_work *ew =
2770 		container_of(work, struct ib_event_work, work);
2771 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2772 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2773 	int i;
2774 	struct ib_event ibev;
2775 
2776 	kfree(ew);
2777 	spin_lock_bh(&ibdev->iboe.lock);
2778 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2779 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2780 		enum ib_port_state curr_port_state;
2781 
2782 		if (!curr_netdev)
2783 			continue;
2784 
2785 		curr_port_state =
2786 			(netif_running(curr_netdev) &&
2787 			 netif_carrier_ok(curr_netdev)) ?
2788 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2789 
2790 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2791 			curr_port_state : IB_PORT_ACTIVE;
2792 	}
2793 	spin_unlock_bh(&ibdev->iboe.lock);
2794 
2795 	ibev.device = &ibdev->ib_dev;
2796 	ibev.element.port_num = 1;
2797 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2798 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2799 
2800 	ib_dispatch_event(&ibev);
2801 }
2802 
2803 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2804 			  enum mlx4_dev_event event, unsigned long param)
2805 {
2806 	struct ib_event ibev;
2807 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2808 	struct mlx4_eqe *eqe = NULL;
2809 	struct ib_event_work *ew;
2810 	int p = 0;
2811 
2812 	if (mlx4_is_bonded(dev) &&
2813 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2814 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2815 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2816 		if (!ew)
2817 			return;
2818 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2819 		ew->ib_dev = ibdev;
2820 		queue_work(wq, &ew->work);
2821 		return;
2822 	}
2823 
2824 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2825 		eqe = (struct mlx4_eqe *)param;
2826 	else
2827 		p = (int) param;
2828 
2829 	switch (event) {
2830 	case MLX4_DEV_EVENT_PORT_UP:
2831 		if (p > ibdev->num_ports)
2832 			return;
2833 		if (mlx4_is_master(dev) &&
2834 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2835 			IB_LINK_LAYER_INFINIBAND) {
2836 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2837 		}
2838 		ibev.event = IB_EVENT_PORT_ACTIVE;
2839 		break;
2840 
2841 	case MLX4_DEV_EVENT_PORT_DOWN:
2842 		if (p > ibdev->num_ports)
2843 			return;
2844 		ibev.event = IB_EVENT_PORT_ERR;
2845 		break;
2846 
2847 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2848 		ibdev->ib_active = false;
2849 		ibev.event = IB_EVENT_DEVICE_FATAL;
2850 		mlx4_ib_handle_catas_error(ibdev);
2851 		break;
2852 
2853 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2854 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2855 		if (!ew) {
2856 			pr_err("failed to allocate memory for events work\n");
2857 			break;
2858 		}
2859 
2860 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2861 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2862 		ew->ib_dev = ibdev;
2863 		/* need to queue only for port owner, which uses GEN_EQE */
2864 		if (mlx4_is_master(dev))
2865 			queue_work(wq, &ew->work);
2866 		else
2867 			handle_port_mgmt_change_event(&ew->work);
2868 		return;
2869 
2870 	case MLX4_DEV_EVENT_SLAVE_INIT:
2871 		/* here, p is the slave id */
2872 		do_slave_init(ibdev, p, 1);
2873 		if (mlx4_is_master(dev)) {
2874 			int i;
2875 
2876 			for (i = 1; i <= ibdev->num_ports; i++) {
2877 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2878 					== IB_LINK_LAYER_INFINIBAND)
2879 					mlx4_ib_slave_alias_guid_event(ibdev,
2880 								       p, i,
2881 								       1);
2882 			}
2883 		}
2884 		return;
2885 
2886 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2887 		if (mlx4_is_master(dev)) {
2888 			int i;
2889 
2890 			for (i = 1; i <= ibdev->num_ports; i++) {
2891 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2892 					== IB_LINK_LAYER_INFINIBAND)
2893 					mlx4_ib_slave_alias_guid_event(ibdev,
2894 								       p, i,
2895 								       0);
2896 			}
2897 		}
2898 		/* here, p is the slave id */
2899 		do_slave_init(ibdev, p, 0);
2900 		return;
2901 
2902 	default:
2903 		return;
2904 	}
2905 
2906 	ibev.device	      = ibdev_ptr;
2907 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2908 
2909 	ib_dispatch_event(&ibev);
2910 }
2911 
2912 static struct mlx4_interface mlx4_ib_interface = {
2913 	.add		= mlx4_ib_add,
2914 	.remove		= mlx4_ib_remove,
2915 	.event		= mlx4_ib_event,
2916 	.protocol	= MLX4_PROT_IB_IPV6,
2917 	.flags		= MLX4_INTFF_BONDING
2918 };
2919 
2920 static int __init mlx4_ib_init(void)
2921 {
2922 	int err;
2923 
2924 	wq = create_singlethread_workqueue("mlx4_ib");
2925 	if (!wq)
2926 		return -ENOMEM;
2927 
2928 	err = mlx4_ib_mcg_init();
2929 	if (err)
2930 		goto clean_wq;
2931 
2932 	err = mlx4_register_interface(&mlx4_ib_interface);
2933 	if (err)
2934 		goto clean_mcg;
2935 
2936 	return 0;
2937 
2938 clean_mcg:
2939 	mlx4_ib_mcg_destroy();
2940 
2941 clean_wq:
2942 	destroy_workqueue(wq);
2943 	return err;
2944 }
2945 
2946 static void __exit mlx4_ib_cleanup(void)
2947 {
2948 	mlx4_unregister_interface(&mlx4_ib_interface);
2949 	mlx4_ib_mcg_destroy();
2950 	destroy_workqueue(wq);
2951 }
2952 
2953 module_init(mlx4_ib_init);
2954 module_exit(mlx4_ib_cleanup);
2955