xref: /linux/drivers/infiniband/hw/mlx4/main.c (revision 005438a8eef063495ac059d128eea71b58de50e5)
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 	err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
257 	if (err)
258 		goto out;
259 
260 	if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
261 		resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
262 		resp.response_length += sizeof(resp.hca_core_clock_offset);
263 		resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
264 	}
265 
266 	if (uhw->outlen) {
267 		err = ib_copy_to_udata(uhw, &resp, resp.response_length);
268 		if (err)
269 			goto out;
270 	}
271 out:
272 	kfree(in_mad);
273 	kfree(out_mad);
274 
275 	return err;
276 }
277 
278 static enum rdma_link_layer
279 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
280 {
281 	struct mlx4_dev *dev = to_mdev(device)->dev;
282 
283 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
284 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
285 }
286 
287 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
288 			      struct ib_port_attr *props, int netw_view)
289 {
290 	struct ib_smp *in_mad  = NULL;
291 	struct ib_smp *out_mad = NULL;
292 	int ext_active_speed;
293 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
294 	int err = -ENOMEM;
295 
296 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
297 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
298 	if (!in_mad || !out_mad)
299 		goto out;
300 
301 	init_query_mad(in_mad);
302 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
303 	in_mad->attr_mod = cpu_to_be32(port);
304 
305 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
306 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
307 
308 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
309 				in_mad, out_mad);
310 	if (err)
311 		goto out;
312 
313 
314 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
315 	props->lmc		= out_mad->data[34] & 0x7;
316 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
317 	props->sm_sl		= out_mad->data[36] & 0xf;
318 	props->state		= out_mad->data[32] & 0xf;
319 	props->phys_state	= out_mad->data[33] >> 4;
320 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
321 	if (netw_view)
322 		props->gid_tbl_len = out_mad->data[50];
323 	else
324 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
325 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
326 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
327 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
328 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
329 	props->active_width	= out_mad->data[31] & 0xf;
330 	props->active_speed	= out_mad->data[35] >> 4;
331 	props->max_mtu		= out_mad->data[41] & 0xf;
332 	props->active_mtu	= out_mad->data[36] >> 4;
333 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
334 	props->max_vl_num	= out_mad->data[37] >> 4;
335 	props->init_type_reply	= out_mad->data[41] >> 4;
336 
337 	/* Check if extended speeds (EDR/FDR/...) are supported */
338 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
339 		ext_active_speed = out_mad->data[62] >> 4;
340 
341 		switch (ext_active_speed) {
342 		case 1:
343 			props->active_speed = IB_SPEED_FDR;
344 			break;
345 		case 2:
346 			props->active_speed = IB_SPEED_EDR;
347 			break;
348 		}
349 	}
350 
351 	/* If reported active speed is QDR, check if is FDR-10 */
352 	if (props->active_speed == IB_SPEED_QDR) {
353 		init_query_mad(in_mad);
354 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
355 		in_mad->attr_mod = cpu_to_be32(port);
356 
357 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
358 				   NULL, NULL, in_mad, out_mad);
359 		if (err)
360 			goto out;
361 
362 		/* Checking LinkSpeedActive for FDR-10 */
363 		if (out_mad->data[15] & 0x1)
364 			props->active_speed = IB_SPEED_FDR10;
365 	}
366 
367 	/* Avoid wrong speed value returned by FW if the IB link is down. */
368 	if (props->state == IB_PORT_DOWN)
369 		 props->active_speed = IB_SPEED_SDR;
370 
371 out:
372 	kfree(in_mad);
373 	kfree(out_mad);
374 	return err;
375 }
376 
377 static u8 state_to_phys_state(enum ib_port_state state)
378 {
379 	return state == IB_PORT_ACTIVE ? 5 : 3;
380 }
381 
382 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
383 			       struct ib_port_attr *props, int netw_view)
384 {
385 
386 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
387 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
388 	struct net_device *ndev;
389 	enum ib_mtu tmp;
390 	struct mlx4_cmd_mailbox *mailbox;
391 	int err = 0;
392 	int is_bonded = mlx4_is_bonded(mdev->dev);
393 
394 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
395 	if (IS_ERR(mailbox))
396 		return PTR_ERR(mailbox);
397 
398 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
399 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
400 			   MLX4_CMD_WRAPPED);
401 	if (err)
402 		goto out;
403 
404 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
405 						IB_WIDTH_4X : IB_WIDTH_1X;
406 	props->active_speed	= IB_SPEED_QDR;
407 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
408 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
409 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
410 	props->pkey_tbl_len	= 1;
411 	props->max_mtu		= IB_MTU_4096;
412 	props->max_vl_num	= 2;
413 	props->state		= IB_PORT_DOWN;
414 	props->phys_state	= state_to_phys_state(props->state);
415 	props->active_mtu	= IB_MTU_256;
416 	if (is_bonded)
417 		rtnl_lock(); /* required to get upper dev */
418 	spin_lock_bh(&iboe->lock);
419 	ndev = iboe->netdevs[port - 1];
420 	if (ndev && is_bonded)
421 		ndev = netdev_master_upper_dev_get(ndev);
422 	if (!ndev)
423 		goto out_unlock;
424 
425 	tmp = iboe_get_mtu(ndev->mtu);
426 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
427 
428 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
429 					IB_PORT_ACTIVE : IB_PORT_DOWN;
430 	props->phys_state	= state_to_phys_state(props->state);
431 out_unlock:
432 	spin_unlock_bh(&iboe->lock);
433 	if (is_bonded)
434 		rtnl_unlock();
435 out:
436 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
437 	return err;
438 }
439 
440 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
441 			 struct ib_port_attr *props, int netw_view)
442 {
443 	int err;
444 
445 	memset(props, 0, sizeof *props);
446 
447 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
448 		ib_link_query_port(ibdev, port, props, netw_view) :
449 				eth_link_query_port(ibdev, port, props, netw_view);
450 
451 	return err;
452 }
453 
454 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
455 			      struct ib_port_attr *props)
456 {
457 	/* returns host view */
458 	return __mlx4_ib_query_port(ibdev, port, props, 0);
459 }
460 
461 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
462 			union ib_gid *gid, int netw_view)
463 {
464 	struct ib_smp *in_mad  = NULL;
465 	struct ib_smp *out_mad = NULL;
466 	int err = -ENOMEM;
467 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
468 	int clear = 0;
469 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
470 
471 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
472 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
473 	if (!in_mad || !out_mad)
474 		goto out;
475 
476 	init_query_mad(in_mad);
477 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
478 	in_mad->attr_mod = cpu_to_be32(port);
479 
480 	if (mlx4_is_mfunc(dev->dev) && netw_view)
481 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
482 
483 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
484 	if (err)
485 		goto out;
486 
487 	memcpy(gid->raw, out_mad->data + 8, 8);
488 
489 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
490 		if (index) {
491 			/* For any index > 0, return the null guid */
492 			err = 0;
493 			clear = 1;
494 			goto out;
495 		}
496 	}
497 
498 	init_query_mad(in_mad);
499 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
500 	in_mad->attr_mod = cpu_to_be32(index / 8);
501 
502 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
503 			   NULL, NULL, in_mad, out_mad);
504 	if (err)
505 		goto out;
506 
507 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
508 
509 out:
510 	if (clear)
511 		memset(gid->raw + 8, 0, 8);
512 	kfree(in_mad);
513 	kfree(out_mad);
514 	return err;
515 }
516 
517 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
518 			  union ib_gid *gid)
519 {
520 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
521 
522 	*gid = dev->iboe.gid_table[port - 1][index];
523 
524 	return 0;
525 }
526 
527 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
528 			     union ib_gid *gid)
529 {
530 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
531 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
532 	else
533 		return iboe_query_gid(ibdev, port, index, gid);
534 }
535 
536 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
537 			 u16 *pkey, int netw_view)
538 {
539 	struct ib_smp *in_mad  = NULL;
540 	struct ib_smp *out_mad = NULL;
541 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
542 	int err = -ENOMEM;
543 
544 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
545 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
546 	if (!in_mad || !out_mad)
547 		goto out;
548 
549 	init_query_mad(in_mad);
550 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
551 	in_mad->attr_mod = cpu_to_be32(index / 32);
552 
553 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
554 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
555 
556 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
557 			   in_mad, out_mad);
558 	if (err)
559 		goto out;
560 
561 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
562 
563 out:
564 	kfree(in_mad);
565 	kfree(out_mad);
566 	return err;
567 }
568 
569 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
570 {
571 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
572 }
573 
574 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
575 				 struct ib_device_modify *props)
576 {
577 	struct mlx4_cmd_mailbox *mailbox;
578 	unsigned long flags;
579 
580 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
581 		return -EOPNOTSUPP;
582 
583 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
584 		return 0;
585 
586 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
587 		return -EOPNOTSUPP;
588 
589 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
590 	memcpy(ibdev->node_desc, props->node_desc, 64);
591 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
592 
593 	/*
594 	 * If possible, pass node desc to FW, so it can generate
595 	 * a 144 trap.  If cmd fails, just ignore.
596 	 */
597 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
598 	if (IS_ERR(mailbox))
599 		return 0;
600 
601 	memcpy(mailbox->buf, props->node_desc, 64);
602 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
603 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
604 
605 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
606 
607 	return 0;
608 }
609 
610 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
611 			    u32 cap_mask)
612 {
613 	struct mlx4_cmd_mailbox *mailbox;
614 	int err;
615 
616 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
617 	if (IS_ERR(mailbox))
618 		return PTR_ERR(mailbox);
619 
620 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
621 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
622 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
623 	} else {
624 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
625 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
626 	}
627 
628 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
629 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
630 		       MLX4_CMD_WRAPPED);
631 
632 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
633 	return err;
634 }
635 
636 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
637 			       struct ib_port_modify *props)
638 {
639 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
640 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
641 	struct ib_port_attr attr;
642 	u32 cap_mask;
643 	int err;
644 
645 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
646 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
647 	 * violations and port capabilities are not meaningful.
648 	 */
649 	if (is_eth)
650 		return 0;
651 
652 	mutex_lock(&mdev->cap_mask_mutex);
653 
654 	err = mlx4_ib_query_port(ibdev, port, &attr);
655 	if (err)
656 		goto out;
657 
658 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
659 		~props->clr_port_cap_mask;
660 
661 	err = mlx4_ib_SET_PORT(mdev, port,
662 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
663 			       cap_mask);
664 
665 out:
666 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
667 	return err;
668 }
669 
670 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
671 						  struct ib_udata *udata)
672 {
673 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
674 	struct mlx4_ib_ucontext *context;
675 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
676 	struct mlx4_ib_alloc_ucontext_resp resp;
677 	int err;
678 
679 	if (!dev->ib_active)
680 		return ERR_PTR(-EAGAIN);
681 
682 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
683 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
684 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
685 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
686 	} else {
687 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
688 		resp.qp_tab_size      = dev->dev->caps.num_qps;
689 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
690 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
691 		resp.cqe_size	      = dev->dev->caps.cqe_size;
692 	}
693 
694 	context = kmalloc(sizeof *context, GFP_KERNEL);
695 	if (!context)
696 		return ERR_PTR(-ENOMEM);
697 
698 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
699 	if (err) {
700 		kfree(context);
701 		return ERR_PTR(err);
702 	}
703 
704 	INIT_LIST_HEAD(&context->db_page_list);
705 	mutex_init(&context->db_page_mutex);
706 
707 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
708 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
709 	else
710 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
711 
712 	if (err) {
713 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
714 		kfree(context);
715 		return ERR_PTR(-EFAULT);
716 	}
717 
718 	return &context->ibucontext;
719 }
720 
721 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
722 {
723 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
724 
725 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
726 	kfree(context);
727 
728 	return 0;
729 }
730 
731 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
732 {
733 	struct mlx4_ib_dev *dev = to_mdev(context->device);
734 
735 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
736 		return -EINVAL;
737 
738 	if (vma->vm_pgoff == 0) {
739 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
740 
741 		if (io_remap_pfn_range(vma, vma->vm_start,
742 				       to_mucontext(context)->uar.pfn,
743 				       PAGE_SIZE, vma->vm_page_prot))
744 			return -EAGAIN;
745 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
746 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
747 
748 		if (io_remap_pfn_range(vma, vma->vm_start,
749 				       to_mucontext(context)->uar.pfn +
750 				       dev->dev->caps.num_uars,
751 				       PAGE_SIZE, vma->vm_page_prot))
752 			return -EAGAIN;
753 	} else if (vma->vm_pgoff == 3) {
754 		struct mlx4_clock_params params;
755 		int ret = mlx4_get_internal_clock_params(dev->dev, &params);
756 
757 		if (ret)
758 			return ret;
759 
760 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
761 		if (io_remap_pfn_range(vma, vma->vm_start,
762 				       (pci_resource_start(dev->dev->persist->pdev,
763 							   params.bar) +
764 					params.offset)
765 				       >> PAGE_SHIFT,
766 				       PAGE_SIZE, vma->vm_page_prot))
767 			return -EAGAIN;
768 	} else {
769 		return -EINVAL;
770 	}
771 
772 	return 0;
773 }
774 
775 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
776 				      struct ib_ucontext *context,
777 				      struct ib_udata *udata)
778 {
779 	struct mlx4_ib_pd *pd;
780 	int err;
781 
782 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
783 	if (!pd)
784 		return ERR_PTR(-ENOMEM);
785 
786 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
787 	if (err) {
788 		kfree(pd);
789 		return ERR_PTR(err);
790 	}
791 
792 	if (context)
793 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
794 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
795 			kfree(pd);
796 			return ERR_PTR(-EFAULT);
797 		}
798 
799 	return &pd->ibpd;
800 }
801 
802 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
803 {
804 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
805 	kfree(pd);
806 
807 	return 0;
808 }
809 
810 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
811 					  struct ib_ucontext *context,
812 					  struct ib_udata *udata)
813 {
814 	struct mlx4_ib_xrcd *xrcd;
815 	struct ib_cq_init_attr cq_attr = {};
816 	int err;
817 
818 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
819 		return ERR_PTR(-ENOSYS);
820 
821 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
822 	if (!xrcd)
823 		return ERR_PTR(-ENOMEM);
824 
825 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
826 	if (err)
827 		goto err1;
828 
829 	xrcd->pd = ib_alloc_pd(ibdev);
830 	if (IS_ERR(xrcd->pd)) {
831 		err = PTR_ERR(xrcd->pd);
832 		goto err2;
833 	}
834 
835 	cq_attr.cqe = 1;
836 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
837 	if (IS_ERR(xrcd->cq)) {
838 		err = PTR_ERR(xrcd->cq);
839 		goto err3;
840 	}
841 
842 	return &xrcd->ibxrcd;
843 
844 err3:
845 	ib_dealloc_pd(xrcd->pd);
846 err2:
847 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
848 err1:
849 	kfree(xrcd);
850 	return ERR_PTR(err);
851 }
852 
853 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
854 {
855 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
856 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
857 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
858 	kfree(xrcd);
859 
860 	return 0;
861 }
862 
863 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
864 {
865 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
866 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
867 	struct mlx4_ib_gid_entry *ge;
868 
869 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
870 	if (!ge)
871 		return -ENOMEM;
872 
873 	ge->gid = *gid;
874 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
875 		ge->port = mqp->port;
876 		ge->added = 1;
877 	}
878 
879 	mutex_lock(&mqp->mutex);
880 	list_add_tail(&ge->list, &mqp->gid_list);
881 	mutex_unlock(&mqp->mutex);
882 
883 	return 0;
884 }
885 
886 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
887 		   union ib_gid *gid)
888 {
889 	struct net_device *ndev;
890 	int ret = 0;
891 
892 	if (!mqp->port)
893 		return 0;
894 
895 	spin_lock_bh(&mdev->iboe.lock);
896 	ndev = mdev->iboe.netdevs[mqp->port - 1];
897 	if (ndev)
898 		dev_hold(ndev);
899 	spin_unlock_bh(&mdev->iboe.lock);
900 
901 	if (ndev) {
902 		ret = 1;
903 		dev_put(ndev);
904 	}
905 
906 	return ret;
907 }
908 
909 struct mlx4_ib_steering {
910 	struct list_head list;
911 	struct mlx4_flow_reg_id reg_id;
912 	union ib_gid gid;
913 };
914 
915 static int parse_flow_attr(struct mlx4_dev *dev,
916 			   u32 qp_num,
917 			   union ib_flow_spec *ib_spec,
918 			   struct _rule_hw *mlx4_spec)
919 {
920 	enum mlx4_net_trans_rule_id type;
921 
922 	switch (ib_spec->type) {
923 	case IB_FLOW_SPEC_ETH:
924 		type = MLX4_NET_TRANS_RULE_ID_ETH;
925 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
926 		       ETH_ALEN);
927 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
928 		       ETH_ALEN);
929 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
930 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
931 		break;
932 	case IB_FLOW_SPEC_IB:
933 		type = MLX4_NET_TRANS_RULE_ID_IB;
934 		mlx4_spec->ib.l3_qpn =
935 			cpu_to_be32(qp_num);
936 		mlx4_spec->ib.qpn_mask =
937 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
938 		break;
939 
940 
941 	case IB_FLOW_SPEC_IPV4:
942 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
943 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
944 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
945 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
946 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
947 		break;
948 
949 	case IB_FLOW_SPEC_TCP:
950 	case IB_FLOW_SPEC_UDP:
951 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
952 					MLX4_NET_TRANS_RULE_ID_TCP :
953 					MLX4_NET_TRANS_RULE_ID_UDP;
954 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
955 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
956 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
957 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
958 		break;
959 
960 	default:
961 		return -EINVAL;
962 	}
963 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
964 	    mlx4_hw_rule_sz(dev, type) < 0)
965 		return -EINVAL;
966 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
967 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
968 	return mlx4_hw_rule_sz(dev, type);
969 }
970 
971 struct default_rules {
972 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
973 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
974 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
975 	__u8  link_layer;
976 };
977 static const struct default_rules default_table[] = {
978 	{
979 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
980 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
981 		.rules_create_list = {IB_FLOW_SPEC_IB},
982 		.link_layer = IB_LINK_LAYER_INFINIBAND
983 	}
984 };
985 
986 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
987 					 struct ib_flow_attr *flow_attr)
988 {
989 	int i, j, k;
990 	void *ib_flow;
991 	const struct default_rules *pdefault_rules = default_table;
992 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
993 
994 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
995 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
996 		memset(&field_types, 0, sizeof(field_types));
997 
998 		if (link_layer != pdefault_rules->link_layer)
999 			continue;
1000 
1001 		ib_flow = flow_attr + 1;
1002 		/* we assume the specs are sorted */
1003 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1004 		     j < flow_attr->num_of_specs; k++) {
1005 			union ib_flow_spec *current_flow =
1006 				(union ib_flow_spec *)ib_flow;
1007 
1008 			/* same layer but different type */
1009 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1010 			     (pdefault_rules->mandatory_fields[k] &
1011 			      IB_FLOW_SPEC_LAYER_MASK)) &&
1012 			    (current_flow->type !=
1013 			     pdefault_rules->mandatory_fields[k]))
1014 				goto out;
1015 
1016 			/* same layer, try match next one */
1017 			if (current_flow->type ==
1018 			    pdefault_rules->mandatory_fields[k]) {
1019 				j++;
1020 				ib_flow +=
1021 					((union ib_flow_spec *)ib_flow)->size;
1022 			}
1023 		}
1024 
1025 		ib_flow = flow_attr + 1;
1026 		for (j = 0; j < flow_attr->num_of_specs;
1027 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1028 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1029 				/* same layer and same type */
1030 				if (((union ib_flow_spec *)ib_flow)->type ==
1031 				    pdefault_rules->mandatory_not_fields[k])
1032 					goto out;
1033 
1034 		return i;
1035 	}
1036 out:
1037 	return -1;
1038 }
1039 
1040 static int __mlx4_ib_create_default_rules(
1041 		struct mlx4_ib_dev *mdev,
1042 		struct ib_qp *qp,
1043 		const struct default_rules *pdefault_rules,
1044 		struct _rule_hw *mlx4_spec) {
1045 	int size = 0;
1046 	int i;
1047 
1048 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1049 		int ret;
1050 		union ib_flow_spec ib_spec;
1051 		switch (pdefault_rules->rules_create_list[i]) {
1052 		case 0:
1053 			/* no rule */
1054 			continue;
1055 		case IB_FLOW_SPEC_IB:
1056 			ib_spec.type = IB_FLOW_SPEC_IB;
1057 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1058 
1059 			break;
1060 		default:
1061 			/* invalid rule */
1062 			return -EINVAL;
1063 		}
1064 		/* We must put empty rule, qpn is being ignored */
1065 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1066 				      mlx4_spec);
1067 		if (ret < 0) {
1068 			pr_info("invalid parsing\n");
1069 			return -EINVAL;
1070 		}
1071 
1072 		mlx4_spec = (void *)mlx4_spec + ret;
1073 		size += ret;
1074 	}
1075 	return size;
1076 }
1077 
1078 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1079 			  int domain,
1080 			  enum mlx4_net_trans_promisc_mode flow_type,
1081 			  u64 *reg_id)
1082 {
1083 	int ret, i;
1084 	int size = 0;
1085 	void *ib_flow;
1086 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1087 	struct mlx4_cmd_mailbox *mailbox;
1088 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1089 	int default_flow;
1090 
1091 	static const u16 __mlx4_domain[] = {
1092 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1093 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1094 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1095 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1096 	};
1097 
1098 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1099 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1100 		return -EINVAL;
1101 	}
1102 
1103 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1104 		pr_err("Invalid domain value %d\n", domain);
1105 		return -EINVAL;
1106 	}
1107 
1108 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1109 		return -EINVAL;
1110 
1111 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1112 	if (IS_ERR(mailbox))
1113 		return PTR_ERR(mailbox);
1114 	ctrl = mailbox->buf;
1115 
1116 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1117 				 flow_attr->priority);
1118 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1119 	ctrl->port = flow_attr->port;
1120 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1121 
1122 	ib_flow = flow_attr + 1;
1123 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1124 	/* Add default flows */
1125 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1126 	if (default_flow >= 0) {
1127 		ret = __mlx4_ib_create_default_rules(
1128 				mdev, qp, default_table + default_flow,
1129 				mailbox->buf + size);
1130 		if (ret < 0) {
1131 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1132 			return -EINVAL;
1133 		}
1134 		size += ret;
1135 	}
1136 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1137 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1138 				      mailbox->buf + size);
1139 		if (ret < 0) {
1140 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1141 			return -EINVAL;
1142 		}
1143 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1144 		size += ret;
1145 	}
1146 
1147 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1148 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1149 			   MLX4_CMD_WRAPPED);
1150 	if (ret == -ENOMEM)
1151 		pr_err("mcg table is full. Fail to register network rule.\n");
1152 	else if (ret == -ENXIO)
1153 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1154 	else if (ret)
1155 		pr_err("Invalid argumant. Fail to register network rule.\n");
1156 
1157 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1158 	return ret;
1159 }
1160 
1161 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1162 {
1163 	int err;
1164 	err = mlx4_cmd(dev, reg_id, 0, 0,
1165 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1166 		       MLX4_CMD_WRAPPED);
1167 	if (err)
1168 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1169 		       reg_id);
1170 	return err;
1171 }
1172 
1173 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1174 				    u64 *reg_id)
1175 {
1176 	void *ib_flow;
1177 	union ib_flow_spec *ib_spec;
1178 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1179 	int err = 0;
1180 
1181 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1182 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1183 		return 0; /* do nothing */
1184 
1185 	ib_flow = flow_attr + 1;
1186 	ib_spec = (union ib_flow_spec *)ib_flow;
1187 
1188 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1189 		return 0; /* do nothing */
1190 
1191 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1192 				    flow_attr->port, qp->qp_num,
1193 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1194 				    reg_id);
1195 	return err;
1196 }
1197 
1198 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1199 				    struct ib_flow_attr *flow_attr,
1200 				    int domain)
1201 {
1202 	int err = 0, i = 0, j = 0;
1203 	struct mlx4_ib_flow *mflow;
1204 	enum mlx4_net_trans_promisc_mode type[2];
1205 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1206 	int is_bonded = mlx4_is_bonded(dev);
1207 
1208 	memset(type, 0, sizeof(type));
1209 
1210 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1211 	if (!mflow) {
1212 		err = -ENOMEM;
1213 		goto err_free;
1214 	}
1215 
1216 	switch (flow_attr->type) {
1217 	case IB_FLOW_ATTR_NORMAL:
1218 		type[0] = MLX4_FS_REGULAR;
1219 		break;
1220 
1221 	case IB_FLOW_ATTR_ALL_DEFAULT:
1222 		type[0] = MLX4_FS_ALL_DEFAULT;
1223 		break;
1224 
1225 	case IB_FLOW_ATTR_MC_DEFAULT:
1226 		type[0] = MLX4_FS_MC_DEFAULT;
1227 		break;
1228 
1229 	case IB_FLOW_ATTR_SNIFFER:
1230 		type[0] = MLX4_FS_UC_SNIFFER;
1231 		type[1] = MLX4_FS_MC_SNIFFER;
1232 		break;
1233 
1234 	default:
1235 		err = -EINVAL;
1236 		goto err_free;
1237 	}
1238 
1239 	while (i < ARRAY_SIZE(type) && type[i]) {
1240 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1241 					    &mflow->reg_id[i].id);
1242 		if (err)
1243 			goto err_create_flow;
1244 		if (is_bonded) {
1245 			/* Application always sees one port so the mirror rule
1246 			 * must be on port #2
1247 			 */
1248 			flow_attr->port = 2;
1249 			err = __mlx4_ib_create_flow(qp, flow_attr,
1250 						    domain, type[j],
1251 						    &mflow->reg_id[j].mirror);
1252 			flow_attr->port = 1;
1253 			if (err)
1254 				goto err_create_flow;
1255 			j++;
1256 		}
1257 
1258 		i++;
1259 	}
1260 
1261 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1262 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1263 					       &mflow->reg_id[i].id);
1264 		if (err)
1265 			goto err_create_flow;
1266 
1267 		if (is_bonded) {
1268 			flow_attr->port = 2;
1269 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1270 						       &mflow->reg_id[j].mirror);
1271 			flow_attr->port = 1;
1272 			if (err)
1273 				goto err_create_flow;
1274 			j++;
1275 		}
1276 		/* function to create mirror rule */
1277 		i++;
1278 	}
1279 
1280 	return &mflow->ibflow;
1281 
1282 err_create_flow:
1283 	while (i) {
1284 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1285 					     mflow->reg_id[i].id);
1286 		i--;
1287 	}
1288 
1289 	while (j) {
1290 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1291 					     mflow->reg_id[j].mirror);
1292 		j--;
1293 	}
1294 err_free:
1295 	kfree(mflow);
1296 	return ERR_PTR(err);
1297 }
1298 
1299 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1300 {
1301 	int err, ret = 0;
1302 	int i = 0;
1303 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1304 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1305 
1306 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1307 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1308 		if (err)
1309 			ret = err;
1310 		if (mflow->reg_id[i].mirror) {
1311 			err = __mlx4_ib_destroy_flow(mdev->dev,
1312 						     mflow->reg_id[i].mirror);
1313 			if (err)
1314 				ret = err;
1315 		}
1316 		i++;
1317 	}
1318 
1319 	kfree(mflow);
1320 	return ret;
1321 }
1322 
1323 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1324 {
1325 	int err;
1326 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1327 	struct mlx4_dev	*dev = mdev->dev;
1328 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1329 	struct mlx4_ib_steering *ib_steering = NULL;
1330 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1331 	struct mlx4_flow_reg_id	reg_id;
1332 
1333 	if (mdev->dev->caps.steering_mode ==
1334 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1335 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1336 		if (!ib_steering)
1337 			return -ENOMEM;
1338 	}
1339 
1340 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1341 				    !!(mqp->flags &
1342 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1343 				    prot, &reg_id.id);
1344 	if (err) {
1345 		pr_err("multicast attach op failed, err %d\n", err);
1346 		goto err_malloc;
1347 	}
1348 
1349 	reg_id.mirror = 0;
1350 	if (mlx4_is_bonded(dev)) {
1351 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1352 					    (mqp->port == 1) ? 2 : 1,
1353 					    !!(mqp->flags &
1354 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1355 					    prot, &reg_id.mirror);
1356 		if (err)
1357 			goto err_add;
1358 	}
1359 
1360 	err = add_gid_entry(ibqp, gid);
1361 	if (err)
1362 		goto err_add;
1363 
1364 	if (ib_steering) {
1365 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1366 		ib_steering->reg_id = reg_id;
1367 		mutex_lock(&mqp->mutex);
1368 		list_add(&ib_steering->list, &mqp->steering_rules);
1369 		mutex_unlock(&mqp->mutex);
1370 	}
1371 	return 0;
1372 
1373 err_add:
1374 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1375 			      prot, reg_id.id);
1376 	if (reg_id.mirror)
1377 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1378 				      prot, reg_id.mirror);
1379 err_malloc:
1380 	kfree(ib_steering);
1381 
1382 	return err;
1383 }
1384 
1385 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1386 {
1387 	struct mlx4_ib_gid_entry *ge;
1388 	struct mlx4_ib_gid_entry *tmp;
1389 	struct mlx4_ib_gid_entry *ret = NULL;
1390 
1391 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1392 		if (!memcmp(raw, ge->gid.raw, 16)) {
1393 			ret = ge;
1394 			break;
1395 		}
1396 	}
1397 
1398 	return ret;
1399 }
1400 
1401 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1402 {
1403 	int err;
1404 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1405 	struct mlx4_dev *dev = mdev->dev;
1406 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1407 	struct net_device *ndev;
1408 	struct mlx4_ib_gid_entry *ge;
1409 	struct mlx4_flow_reg_id reg_id = {0, 0};
1410 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1411 
1412 	if (mdev->dev->caps.steering_mode ==
1413 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1414 		struct mlx4_ib_steering *ib_steering;
1415 
1416 		mutex_lock(&mqp->mutex);
1417 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1418 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1419 				list_del(&ib_steering->list);
1420 				break;
1421 			}
1422 		}
1423 		mutex_unlock(&mqp->mutex);
1424 		if (&ib_steering->list == &mqp->steering_rules) {
1425 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1426 			return -EINVAL;
1427 		}
1428 		reg_id = ib_steering->reg_id;
1429 		kfree(ib_steering);
1430 	}
1431 
1432 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1433 				    prot, reg_id.id);
1434 	if (err)
1435 		return err;
1436 
1437 	if (mlx4_is_bonded(dev)) {
1438 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1439 					    prot, reg_id.mirror);
1440 		if (err)
1441 			return err;
1442 	}
1443 
1444 	mutex_lock(&mqp->mutex);
1445 	ge = find_gid_entry(mqp, gid->raw);
1446 	if (ge) {
1447 		spin_lock_bh(&mdev->iboe.lock);
1448 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1449 		if (ndev)
1450 			dev_hold(ndev);
1451 		spin_unlock_bh(&mdev->iboe.lock);
1452 		if (ndev)
1453 			dev_put(ndev);
1454 		list_del(&ge->list);
1455 		kfree(ge);
1456 	} else
1457 		pr_warn("could not find mgid entry\n");
1458 
1459 	mutex_unlock(&mqp->mutex);
1460 
1461 	return 0;
1462 }
1463 
1464 static int init_node_data(struct mlx4_ib_dev *dev)
1465 {
1466 	struct ib_smp *in_mad  = NULL;
1467 	struct ib_smp *out_mad = NULL;
1468 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1469 	int err = -ENOMEM;
1470 
1471 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1472 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1473 	if (!in_mad || !out_mad)
1474 		goto out;
1475 
1476 	init_query_mad(in_mad);
1477 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1478 	if (mlx4_is_master(dev->dev))
1479 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1480 
1481 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1482 	if (err)
1483 		goto out;
1484 
1485 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1486 
1487 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1488 
1489 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1490 	if (err)
1491 		goto out;
1492 
1493 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1494 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1495 
1496 out:
1497 	kfree(in_mad);
1498 	kfree(out_mad);
1499 	return err;
1500 }
1501 
1502 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1503 			char *buf)
1504 {
1505 	struct mlx4_ib_dev *dev =
1506 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1507 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1508 }
1509 
1510 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1511 			   char *buf)
1512 {
1513 	struct mlx4_ib_dev *dev =
1514 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1515 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1516 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1517 		       (int) dev->dev->caps.fw_ver & 0xffff);
1518 }
1519 
1520 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1521 			char *buf)
1522 {
1523 	struct mlx4_ib_dev *dev =
1524 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1525 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1526 }
1527 
1528 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1529 			  char *buf)
1530 {
1531 	struct mlx4_ib_dev *dev =
1532 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1533 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1534 		       dev->dev->board_id);
1535 }
1536 
1537 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1538 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1539 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1540 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1541 
1542 static struct device_attribute *mlx4_class_attributes[] = {
1543 	&dev_attr_hw_rev,
1544 	&dev_attr_fw_ver,
1545 	&dev_attr_hca_type,
1546 	&dev_attr_board_id
1547 };
1548 
1549 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1550 				     struct net_device *dev)
1551 {
1552 	memcpy(eui, dev->dev_addr, 3);
1553 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1554 	if (vlan_id < 0x1000) {
1555 		eui[3] = vlan_id >> 8;
1556 		eui[4] = vlan_id & 0xff;
1557 	} else {
1558 		eui[3] = 0xff;
1559 		eui[4] = 0xfe;
1560 	}
1561 	eui[0] ^= 2;
1562 }
1563 
1564 static void update_gids_task(struct work_struct *work)
1565 {
1566 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1567 	struct mlx4_cmd_mailbox *mailbox;
1568 	union ib_gid *gids;
1569 	int err;
1570 	struct mlx4_dev	*dev = gw->dev->dev;
1571 	int is_bonded = mlx4_is_bonded(dev);
1572 
1573 	if (!gw->dev->ib_active)
1574 		return;
1575 
1576 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1577 	if (IS_ERR(mailbox)) {
1578 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1579 		return;
1580 	}
1581 
1582 	gids = mailbox->buf;
1583 	memcpy(gids, gw->gids, sizeof gw->gids);
1584 
1585 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1586 		       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1587 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1588 	if (err)
1589 		pr_warn("set port command failed\n");
1590 	else
1591 		if ((gw->port == 1) || !is_bonded)
1592 			mlx4_ib_dispatch_event(gw->dev,
1593 					       is_bonded ? 1 : gw->port,
1594 					       IB_EVENT_GID_CHANGE);
1595 
1596 	mlx4_free_cmd_mailbox(dev, mailbox);
1597 	kfree(gw);
1598 }
1599 
1600 static void reset_gids_task(struct work_struct *work)
1601 {
1602 	struct update_gid_work *gw =
1603 			container_of(work, struct update_gid_work, work);
1604 	struct mlx4_cmd_mailbox *mailbox;
1605 	union ib_gid *gids;
1606 	int err;
1607 	struct mlx4_dev	*dev = gw->dev->dev;
1608 
1609 	if (!gw->dev->ib_active)
1610 		return;
1611 
1612 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1613 	if (IS_ERR(mailbox)) {
1614 		pr_warn("reset gid table failed\n");
1615 		goto free;
1616 	}
1617 
1618 	gids = mailbox->buf;
1619 	memcpy(gids, gw->gids, sizeof(gw->gids));
1620 
1621 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1622 				    IB_LINK_LAYER_ETHERNET) {
1623 		err = mlx4_cmd(dev, mailbox->dma,
1624 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1625 			       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1626 			       MLX4_CMD_TIME_CLASS_B,
1627 			       MLX4_CMD_WRAPPED);
1628 		if (err)
1629 			pr_warn("set port %d command failed\n", gw->port);
1630 	}
1631 
1632 	mlx4_free_cmd_mailbox(dev, mailbox);
1633 free:
1634 	kfree(gw);
1635 }
1636 
1637 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1638 			    union ib_gid *gid, int clear,
1639 			    int default_gid)
1640 {
1641 	struct update_gid_work *work;
1642 	int i;
1643 	int need_update = 0;
1644 	int free = -1;
1645 	int found = -1;
1646 	int max_gids;
1647 
1648 	if (default_gid) {
1649 		free = 0;
1650 	} else {
1651 		max_gids = dev->dev->caps.gid_table_len[port];
1652 		for (i = 1; i < max_gids; ++i) {
1653 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1654 				    sizeof(*gid)))
1655 				found = i;
1656 
1657 			if (clear) {
1658 				if (found >= 0) {
1659 					need_update = 1;
1660 					dev->iboe.gid_table[port - 1][found] =
1661 						zgid;
1662 					break;
1663 				}
1664 			} else {
1665 				if (found >= 0)
1666 					break;
1667 
1668 				if (free < 0 &&
1669 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1670 					    &zgid, sizeof(*gid)))
1671 					free = i;
1672 			}
1673 		}
1674 	}
1675 
1676 	if (found == -1 && !clear && free >= 0) {
1677 		dev->iboe.gid_table[port - 1][free] = *gid;
1678 		need_update = 1;
1679 	}
1680 
1681 	if (!need_update)
1682 		return 0;
1683 
1684 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1685 	if (!work)
1686 		return -ENOMEM;
1687 
1688 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1689 	INIT_WORK(&work->work, update_gids_task);
1690 	work->port = port;
1691 	work->dev = dev;
1692 	queue_work(wq, &work->work);
1693 
1694 	return 0;
1695 }
1696 
1697 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1698 {
1699 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1700 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1701 }
1702 
1703 
1704 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1705 {
1706 	struct update_gid_work *work;
1707 
1708 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1709 	if (!work)
1710 		return -ENOMEM;
1711 
1712 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1713 	memset(work->gids, 0, sizeof(work->gids));
1714 	INIT_WORK(&work->work, reset_gids_task);
1715 	work->dev = dev;
1716 	work->port = port;
1717 	queue_work(wq, &work->work);
1718 	return 0;
1719 }
1720 
1721 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1722 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1723 {
1724 	struct mlx4_ib_iboe *iboe;
1725 	int port = 0;
1726 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1727 				rdma_vlan_dev_real_dev(event_netdev) :
1728 				event_netdev;
1729 	union ib_gid default_gid;
1730 
1731 	mlx4_make_default_gid(real_dev, &default_gid);
1732 
1733 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1734 		return 0;
1735 
1736 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1737 		return 0;
1738 
1739 	if ((real_dev != event_netdev) &&
1740 	    (event == NETDEV_DOWN) &&
1741 	    rdma_link_local_addr((struct in6_addr *)gid))
1742 		return 0;
1743 
1744 	iboe = &ibdev->iboe;
1745 	spin_lock_bh(&iboe->lock);
1746 
1747 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1748 		if ((netif_is_bond_master(real_dev) &&
1749 		     (real_dev == iboe->masters[port - 1])) ||
1750 		     (!netif_is_bond_master(real_dev) &&
1751 		     (real_dev == iboe->netdevs[port - 1])))
1752 			update_gid_table(ibdev, port, gid,
1753 					 event == NETDEV_DOWN, 0);
1754 
1755 	spin_unlock_bh(&iboe->lock);
1756 	return 0;
1757 
1758 }
1759 
1760 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1761 			       struct mlx4_ib_dev *ibdev)
1762 {
1763 	u8 port = 0;
1764 	struct mlx4_ib_iboe *iboe;
1765 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1766 				rdma_vlan_dev_real_dev(dev) : dev;
1767 
1768 	iboe = &ibdev->iboe;
1769 
1770 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1771 		if ((netif_is_bond_master(real_dev) &&
1772 		     (real_dev == iboe->masters[port - 1])) ||
1773 		     (!netif_is_bond_master(real_dev) &&
1774 		     (real_dev == iboe->netdevs[port - 1])))
1775 			break;
1776 
1777 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1778 		return 0;
1779 	else
1780 		return port;
1781 }
1782 
1783 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1784 				void *ptr)
1785 {
1786 	struct mlx4_ib_dev *ibdev;
1787 	struct in_ifaddr *ifa = ptr;
1788 	union ib_gid gid;
1789 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1790 
1791 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1792 
1793 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1794 
1795 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1796 	return NOTIFY_DONE;
1797 }
1798 
1799 #if IS_ENABLED(CONFIG_IPV6)
1800 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1801 				void *ptr)
1802 {
1803 	struct mlx4_ib_dev *ibdev;
1804 	struct inet6_ifaddr *ifa = ptr;
1805 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1806 	struct net_device *event_netdev = ifa->idev->dev;
1807 
1808 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1809 
1810 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1811 	return NOTIFY_DONE;
1812 }
1813 #endif
1814 
1815 #define MLX4_IB_INVALID_MAC	((u64)-1)
1816 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1817 			       struct net_device *dev,
1818 			       int port)
1819 {
1820 	u64 new_smac = 0;
1821 	u64 release_mac = MLX4_IB_INVALID_MAC;
1822 	struct mlx4_ib_qp *qp;
1823 
1824 	read_lock(&dev_base_lock);
1825 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1826 	read_unlock(&dev_base_lock);
1827 
1828 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1829 
1830 	/* no need for update QP1 and mac registration in non-SRIOV */
1831 	if (!mlx4_is_mfunc(ibdev->dev))
1832 		return;
1833 
1834 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1835 	qp = ibdev->qp1_proxy[port - 1];
1836 	if (qp) {
1837 		int new_smac_index;
1838 		u64 old_smac;
1839 		struct mlx4_update_qp_params update_params;
1840 
1841 		mutex_lock(&qp->mutex);
1842 		old_smac = qp->pri.smac;
1843 		if (new_smac == old_smac)
1844 			goto unlock;
1845 
1846 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1847 
1848 		if (new_smac_index < 0)
1849 			goto unlock;
1850 
1851 		update_params.smac_index = new_smac_index;
1852 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1853 				   &update_params)) {
1854 			release_mac = new_smac;
1855 			goto unlock;
1856 		}
1857 		/* if old port was zero, no mac was yet registered for this QP */
1858 		if (qp->pri.smac_port)
1859 			release_mac = old_smac;
1860 		qp->pri.smac = new_smac;
1861 		qp->pri.smac_port = port;
1862 		qp->pri.smac_index = new_smac_index;
1863 	}
1864 
1865 unlock:
1866 	if (release_mac != MLX4_IB_INVALID_MAC)
1867 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1868 	if (qp)
1869 		mutex_unlock(&qp->mutex);
1870 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1871 }
1872 
1873 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1874 				 struct mlx4_ib_dev *ibdev, u8 port)
1875 {
1876 	struct in_device *in_dev;
1877 #if IS_ENABLED(CONFIG_IPV6)
1878 	struct inet6_dev *in6_dev;
1879 	union ib_gid  *pgid;
1880 	struct inet6_ifaddr *ifp;
1881 	union ib_gid default_gid;
1882 #endif
1883 	union ib_gid gid;
1884 
1885 
1886 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1887 		return;
1888 
1889 	/* IPv4 gids */
1890 	in_dev = in_dev_get(dev);
1891 	if (in_dev) {
1892 		for_ifa(in_dev) {
1893 			/*ifa->ifa_address;*/
1894 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1895 					       (struct in6_addr *)&gid);
1896 			update_gid_table(ibdev, port, &gid, 0, 0);
1897 		}
1898 		endfor_ifa(in_dev);
1899 		in_dev_put(in_dev);
1900 	}
1901 #if IS_ENABLED(CONFIG_IPV6)
1902 	mlx4_make_default_gid(dev, &default_gid);
1903 	/* IPv6 gids */
1904 	in6_dev = in6_dev_get(dev);
1905 	if (in6_dev) {
1906 		read_lock_bh(&in6_dev->lock);
1907 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1908 			pgid = (union ib_gid *)&ifp->addr;
1909 			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1910 				continue;
1911 			update_gid_table(ibdev, port, pgid, 0, 0);
1912 		}
1913 		read_unlock_bh(&in6_dev->lock);
1914 		in6_dev_put(in6_dev);
1915 	}
1916 #endif
1917 }
1918 
1919 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1920 				 struct  net_device *dev, u8 port)
1921 {
1922 	union ib_gid gid;
1923 	mlx4_make_default_gid(dev, &gid);
1924 	update_gid_table(ibdev, port, &gid, 0, 1);
1925 }
1926 
1927 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1928 {
1929 	struct	net_device *dev;
1930 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1931 	int i;
1932 	int err = 0;
1933 
1934 	for (i = 1; i <= ibdev->num_ports; ++i) {
1935 		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1936 		    IB_LINK_LAYER_ETHERNET) {
1937 			err = reset_gid_table(ibdev, i);
1938 			if (err)
1939 				goto out;
1940 		}
1941 	}
1942 
1943 	read_lock(&dev_base_lock);
1944 	spin_lock_bh(&iboe->lock);
1945 
1946 	for_each_netdev(&init_net, dev) {
1947 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1948 		/* port will be non-zero only for ETH ports */
1949 		if (port) {
1950 			mlx4_ib_set_default_gid(ibdev, dev, port);
1951 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1952 		}
1953 	}
1954 
1955 	spin_unlock_bh(&iboe->lock);
1956 	read_unlock(&dev_base_lock);
1957 out:
1958 	return err;
1959 }
1960 
1961 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1962 				 struct net_device *dev,
1963 				 unsigned long event)
1964 
1965 {
1966 	struct mlx4_ib_iboe *iboe;
1967 	int update_qps_port = -1;
1968 	int port;
1969 
1970 	iboe = &ibdev->iboe;
1971 
1972 	spin_lock_bh(&iboe->lock);
1973 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1974 		enum ib_port_state	port_state = IB_PORT_NOP;
1975 		struct net_device *old_master = iboe->masters[port - 1];
1976 		struct net_device *curr_netdev;
1977 		struct net_device *curr_master;
1978 
1979 		iboe->netdevs[port - 1] =
1980 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1981 		if (iboe->netdevs[port - 1])
1982 			mlx4_ib_set_default_gid(ibdev,
1983 						iboe->netdevs[port - 1], port);
1984 		curr_netdev = iboe->netdevs[port - 1];
1985 
1986 		if (iboe->netdevs[port - 1] &&
1987 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1988 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1989 				iboe->netdevs[port - 1]);
1990 		} else {
1991 			iboe->masters[port - 1] = NULL;
1992 		}
1993 		curr_master = iboe->masters[port - 1];
1994 
1995 		if (dev == iboe->netdevs[port - 1] &&
1996 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1997 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1998 			update_qps_port = port;
1999 
2000 		if (curr_netdev) {
2001 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
2002 						IB_PORT_ACTIVE : IB_PORT_DOWN;
2003 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
2004 			if (curr_master) {
2005 				/* if using bonding/team and a slave port is down, we
2006 				 * don't want the bond IP based gids in the table since
2007 				 * flows that select port by gid may get the down port.
2008 				*/
2009 				if (port_state == IB_PORT_DOWN &&
2010 				    !mlx4_is_bonded(ibdev->dev)) {
2011 					reset_gid_table(ibdev, port);
2012 					mlx4_ib_set_default_gid(ibdev,
2013 								curr_netdev,
2014 								port);
2015 				} else {
2016 					/* gids from the upper dev (bond/team)
2017 					 * should appear in port's gid table
2018 					*/
2019 					mlx4_ib_get_dev_addr(curr_master,
2020 							     ibdev, port);
2021 				}
2022 			}
2023 			/* if bonding is used it is possible that we add it to
2024 			 * masters only after IP address is assigned to the
2025 			 * net bonding interface.
2026 			*/
2027 			if (curr_master && (old_master != curr_master)) {
2028 				reset_gid_table(ibdev, port);
2029 				mlx4_ib_set_default_gid(ibdev,
2030 							curr_netdev, port);
2031 				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
2032 			}
2033 
2034 			if (!curr_master && (old_master != curr_master)) {
2035 				reset_gid_table(ibdev, port);
2036 				mlx4_ib_set_default_gid(ibdev,
2037 							curr_netdev, port);
2038 				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
2039 			}
2040 		} else {
2041 			reset_gid_table(ibdev, port);
2042 		}
2043 	}
2044 
2045 	spin_unlock_bh(&iboe->lock);
2046 
2047 	if (update_qps_port > 0)
2048 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2049 }
2050 
2051 static int mlx4_ib_netdev_event(struct notifier_block *this,
2052 				unsigned long event, void *ptr)
2053 {
2054 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2055 	struct mlx4_ib_dev *ibdev;
2056 
2057 	if (!net_eq(dev_net(dev), &init_net))
2058 		return NOTIFY_DONE;
2059 
2060 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2061 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2062 
2063 	return NOTIFY_DONE;
2064 }
2065 
2066 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2067 {
2068 	int port;
2069 	int slave;
2070 	int i;
2071 
2072 	if (mlx4_is_master(ibdev->dev)) {
2073 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2074 		     ++slave) {
2075 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2076 				for (i = 0;
2077 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2078 				     ++i) {
2079 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2080 					/* master has the identity virt2phys pkey mapping */
2081 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2082 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2083 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2084 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2085 				}
2086 			}
2087 		}
2088 		/* initialize pkey cache */
2089 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2090 			for (i = 0;
2091 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2092 			     ++i)
2093 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2094 					(i) ? 0 : 0xFFFF;
2095 		}
2096 	}
2097 }
2098 
2099 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2100 {
2101 	int i, j, eq = 0, total_eqs = 0;
2102 
2103 	ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2104 				  sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2105 	if (!ibdev->eq_table)
2106 		return;
2107 
2108 	for (i = 1; i <= dev->caps.num_ports; i++) {
2109 		for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2110 		     j++, total_eqs++) {
2111 			if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2112 				continue;
2113 			ibdev->eq_table[eq] = total_eqs;
2114 			if (!mlx4_assign_eq(dev, i,
2115 					    &ibdev->eq_table[eq]))
2116 				eq++;
2117 			else
2118 				ibdev->eq_table[eq] = -1;
2119 		}
2120 	}
2121 
2122 	for (i = eq; i < dev->caps.num_comp_vectors;
2123 	     ibdev->eq_table[i++] = -1)
2124 		;
2125 
2126 	/* Advertise the new number of EQs to clients */
2127 	ibdev->ib_dev.num_comp_vectors = eq;
2128 }
2129 
2130 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2131 {
2132 	int i;
2133 	int total_eqs = ibdev->ib_dev.num_comp_vectors;
2134 
2135 	/* no eqs were allocated */
2136 	if (!ibdev->eq_table)
2137 		return;
2138 
2139 	/* Reset the advertised EQ number */
2140 	ibdev->ib_dev.num_comp_vectors = 0;
2141 
2142 	for (i = 0; i < total_eqs; i++)
2143 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2144 
2145 	kfree(ibdev->eq_table);
2146 	ibdev->eq_table = NULL;
2147 }
2148 
2149 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2150 			       struct ib_port_immutable *immutable)
2151 {
2152 	struct ib_port_attr attr;
2153 	int err;
2154 
2155 	err = mlx4_ib_query_port(ibdev, port_num, &attr);
2156 	if (err)
2157 		return err;
2158 
2159 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
2160 	immutable->gid_tbl_len = attr.gid_tbl_len;
2161 
2162 	if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2163 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2164 	else
2165 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2166 
2167 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2168 
2169 	return 0;
2170 }
2171 
2172 static void *mlx4_ib_add(struct mlx4_dev *dev)
2173 {
2174 	struct mlx4_ib_dev *ibdev;
2175 	int num_ports = 0;
2176 	int i, j;
2177 	int err;
2178 	struct mlx4_ib_iboe *iboe;
2179 	int ib_num_ports = 0;
2180 	int num_req_counters;
2181 	int allocated;
2182 	u32 counter_index;
2183 
2184 	pr_info_once("%s", mlx4_ib_version);
2185 
2186 	num_ports = 0;
2187 	mlx4_foreach_ib_transport_port(i, dev)
2188 		num_ports++;
2189 
2190 	/* No point in registering a device with no ports... */
2191 	if (num_ports == 0)
2192 		return NULL;
2193 
2194 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2195 	if (!ibdev) {
2196 		dev_err(&dev->persist->pdev->dev,
2197 			"Device struct alloc failed\n");
2198 		return NULL;
2199 	}
2200 
2201 	iboe = &ibdev->iboe;
2202 
2203 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2204 		goto err_dealloc;
2205 
2206 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2207 		goto err_pd;
2208 
2209 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2210 				 PAGE_SIZE);
2211 	if (!ibdev->uar_map)
2212 		goto err_uar;
2213 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2214 
2215 	ibdev->dev = dev;
2216 	ibdev->bond_next_port	= 0;
2217 
2218 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2219 	ibdev->ib_dev.owner		= THIS_MODULE;
2220 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2221 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2222 	ibdev->num_ports		= num_ports;
2223 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2224 						1 : ibdev->num_ports;
2225 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2226 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2227 
2228 	if (dev->caps.userspace_caps)
2229 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2230 	else
2231 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2232 
2233 	ibdev->ib_dev.uverbs_cmd_mask	=
2234 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2235 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2236 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2237 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2238 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2239 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2240 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2241 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2242 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2243 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2244 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2245 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2246 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2247 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2248 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2249 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2250 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2251 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2252 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2253 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2254 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2255 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2256 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2257 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2258 
2259 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2260 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2261 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2262 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2263 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2264 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2265 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2266 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2267 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2268 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2269 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2270 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2271 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2272 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2273 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2274 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2275 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2276 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2277 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2278 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2279 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2280 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2281 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2282 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2283 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2284 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2285 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2286 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2287 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2288 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2289 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2290 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2291 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2292 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2293 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2294 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2295 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2296 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2297 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2298 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2299 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2300 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2301 	ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2302 
2303 	if (!mlx4_is_slave(ibdev->dev)) {
2304 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2305 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2306 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2307 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2308 	}
2309 
2310 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2311 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2312 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2313 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2314 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2315 
2316 		ibdev->ib_dev.uverbs_cmd_mask |=
2317 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2318 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2319 	}
2320 
2321 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2322 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2323 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2324 		ibdev->ib_dev.uverbs_cmd_mask |=
2325 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2326 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2327 	}
2328 
2329 	if (check_flow_steering_support(dev)) {
2330 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2331 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2332 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2333 
2334 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2335 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2336 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2337 	}
2338 
2339 	ibdev->ib_dev.uverbs_ex_cmd_mask |=
2340 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2341 		(1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ);
2342 
2343 	mlx4_ib_alloc_eqs(dev, ibdev);
2344 
2345 	spin_lock_init(&iboe->lock);
2346 
2347 	if (init_node_data(ibdev))
2348 		goto err_map;
2349 
2350 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2351 	for (i = 0; i < num_req_counters; ++i) {
2352 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2353 		allocated = 0;
2354 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2355 						IB_LINK_LAYER_ETHERNET) {
2356 			err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2357 			/* if failed to allocate a new counter, use default */
2358 			if (err)
2359 				counter_index =
2360 					mlx4_get_default_counter_index(dev,
2361 								       i + 1);
2362 			else
2363 				allocated = 1;
2364 		} else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2365 			counter_index = mlx4_get_default_counter_index(dev,
2366 								       i + 1);
2367 		}
2368 		ibdev->counters[i].index = counter_index;
2369 		ibdev->counters[i].allocated = allocated;
2370 		pr_info("counter index %d for port %d allocated %d\n",
2371 			counter_index, i + 1, allocated);
2372 	}
2373 	if (mlx4_is_bonded(dev))
2374 		for (i = 1; i < ibdev->num_ports ; ++i) {
2375 			ibdev->counters[i].index = ibdev->counters[0].index;
2376 			ibdev->counters[i].allocated = 0;
2377 		}
2378 
2379 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2380 		ib_num_ports++;
2381 
2382 	spin_lock_init(&ibdev->sm_lock);
2383 	mutex_init(&ibdev->cap_mask_mutex);
2384 	INIT_LIST_HEAD(&ibdev->qp_list);
2385 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2386 
2387 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2388 	    ib_num_ports) {
2389 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2390 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2391 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2392 					    &ibdev->steer_qpn_base, 0);
2393 		if (err)
2394 			goto err_counter;
2395 
2396 		ibdev->ib_uc_qpns_bitmap =
2397 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2398 				sizeof(long),
2399 				GFP_KERNEL);
2400 		if (!ibdev->ib_uc_qpns_bitmap) {
2401 			dev_err(&dev->persist->pdev->dev,
2402 				"bit map alloc failed\n");
2403 			goto err_steer_qp_release;
2404 		}
2405 
2406 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2407 
2408 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2409 				dev, ibdev->steer_qpn_base,
2410 				ibdev->steer_qpn_base +
2411 				ibdev->steer_qpn_count - 1);
2412 		if (err)
2413 			goto err_steer_free_bitmap;
2414 	}
2415 
2416 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2417 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2418 
2419 	if (ib_register_device(&ibdev->ib_dev, NULL))
2420 		goto err_steer_free_bitmap;
2421 
2422 	if (mlx4_ib_mad_init(ibdev))
2423 		goto err_reg;
2424 
2425 	if (mlx4_ib_init_sriov(ibdev))
2426 		goto err_mad;
2427 
2428 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2429 		if (!iboe->nb.notifier_call) {
2430 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2431 			err = register_netdevice_notifier(&iboe->nb);
2432 			if (err) {
2433 				iboe->nb.notifier_call = NULL;
2434 				goto err_notif;
2435 			}
2436 		}
2437 		if (!iboe->nb_inet.notifier_call) {
2438 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2439 			err = register_inetaddr_notifier(&iboe->nb_inet);
2440 			if (err) {
2441 				iboe->nb_inet.notifier_call = NULL;
2442 				goto err_notif;
2443 			}
2444 		}
2445 #if IS_ENABLED(CONFIG_IPV6)
2446 		if (!iboe->nb_inet6.notifier_call) {
2447 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2448 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2449 			if (err) {
2450 				iboe->nb_inet6.notifier_call = NULL;
2451 				goto err_notif;
2452 			}
2453 		}
2454 #endif
2455 		if (mlx4_ib_init_gid_table(ibdev))
2456 			goto err_notif;
2457 	}
2458 
2459 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2460 		if (device_create_file(&ibdev->ib_dev.dev,
2461 				       mlx4_class_attributes[j]))
2462 			goto err_notif;
2463 	}
2464 
2465 	ibdev->ib_active = true;
2466 
2467 	if (mlx4_is_mfunc(ibdev->dev))
2468 		init_pkeys(ibdev);
2469 
2470 	/* create paravirt contexts for any VFs which are active */
2471 	if (mlx4_is_master(ibdev->dev)) {
2472 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2473 			if (j == mlx4_master_func_num(ibdev->dev))
2474 				continue;
2475 			if (mlx4_is_slave_active(ibdev->dev, j))
2476 				do_slave_init(ibdev, j, 1);
2477 		}
2478 	}
2479 	return ibdev;
2480 
2481 err_notif:
2482 	if (ibdev->iboe.nb.notifier_call) {
2483 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2484 			pr_warn("failure unregistering notifier\n");
2485 		ibdev->iboe.nb.notifier_call = NULL;
2486 	}
2487 	if (ibdev->iboe.nb_inet.notifier_call) {
2488 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2489 			pr_warn("failure unregistering notifier\n");
2490 		ibdev->iboe.nb_inet.notifier_call = NULL;
2491 	}
2492 #if IS_ENABLED(CONFIG_IPV6)
2493 	if (ibdev->iboe.nb_inet6.notifier_call) {
2494 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2495 			pr_warn("failure unregistering notifier\n");
2496 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2497 	}
2498 #endif
2499 	flush_workqueue(wq);
2500 
2501 	mlx4_ib_close_sriov(ibdev);
2502 
2503 err_mad:
2504 	mlx4_ib_mad_cleanup(ibdev);
2505 
2506 err_reg:
2507 	ib_unregister_device(&ibdev->ib_dev);
2508 
2509 err_steer_free_bitmap:
2510 	kfree(ibdev->ib_uc_qpns_bitmap);
2511 
2512 err_steer_qp_release:
2513 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2514 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2515 				      ibdev->steer_qpn_count);
2516 err_counter:
2517 	for (i = 0; i < ibdev->num_ports; ++i) {
2518 		if (ibdev->counters[i].index != -1 &&
2519 		    ibdev->counters[i].allocated)
2520 			mlx4_counter_free(ibdev->dev,
2521 					  ibdev->counters[i].index);
2522 	}
2523 err_map:
2524 	iounmap(ibdev->uar_map);
2525 
2526 err_uar:
2527 	mlx4_uar_free(dev, &ibdev->priv_uar);
2528 
2529 err_pd:
2530 	mlx4_pd_free(dev, ibdev->priv_pdn);
2531 
2532 err_dealloc:
2533 	ib_dealloc_device(&ibdev->ib_dev);
2534 
2535 	return NULL;
2536 }
2537 
2538 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2539 {
2540 	int offset;
2541 
2542 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2543 
2544 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2545 					 dev->steer_qpn_count,
2546 					 get_count_order(count));
2547 	if (offset < 0)
2548 		return offset;
2549 
2550 	*qpn = dev->steer_qpn_base + offset;
2551 	return 0;
2552 }
2553 
2554 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2555 {
2556 	if (!qpn ||
2557 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2558 		return;
2559 
2560 	BUG_ON(qpn < dev->steer_qpn_base);
2561 
2562 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2563 			      qpn - dev->steer_qpn_base,
2564 			      get_count_order(count));
2565 }
2566 
2567 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2568 			 int is_attach)
2569 {
2570 	int err;
2571 	size_t flow_size;
2572 	struct ib_flow_attr *flow = NULL;
2573 	struct ib_flow_spec_ib *ib_spec;
2574 
2575 	if (is_attach) {
2576 		flow_size = sizeof(struct ib_flow_attr) +
2577 			    sizeof(struct ib_flow_spec_ib);
2578 		flow = kzalloc(flow_size, GFP_KERNEL);
2579 		if (!flow)
2580 			return -ENOMEM;
2581 		flow->port = mqp->port;
2582 		flow->num_of_specs = 1;
2583 		flow->size = flow_size;
2584 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2585 		ib_spec->type = IB_FLOW_SPEC_IB;
2586 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2587 		/* Add an empty rule for IB L2 */
2588 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2589 
2590 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2591 					    IB_FLOW_DOMAIN_NIC,
2592 					    MLX4_FS_REGULAR,
2593 					    &mqp->reg_id);
2594 	} else {
2595 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2596 	}
2597 	kfree(flow);
2598 	return err;
2599 }
2600 
2601 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2602 {
2603 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2604 	int p;
2605 
2606 	ibdev->ib_active = false;
2607 	flush_workqueue(wq);
2608 
2609 	mlx4_ib_close_sriov(ibdev);
2610 	mlx4_ib_mad_cleanup(ibdev);
2611 	ib_unregister_device(&ibdev->ib_dev);
2612 	if (ibdev->iboe.nb.notifier_call) {
2613 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2614 			pr_warn("failure unregistering notifier\n");
2615 		ibdev->iboe.nb.notifier_call = NULL;
2616 	}
2617 
2618 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2619 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2620 				      ibdev->steer_qpn_count);
2621 		kfree(ibdev->ib_uc_qpns_bitmap);
2622 	}
2623 
2624 	if (ibdev->iboe.nb_inet.notifier_call) {
2625 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2626 			pr_warn("failure unregistering notifier\n");
2627 		ibdev->iboe.nb_inet.notifier_call = NULL;
2628 	}
2629 #if IS_ENABLED(CONFIG_IPV6)
2630 	if (ibdev->iboe.nb_inet6.notifier_call) {
2631 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2632 			pr_warn("failure unregistering notifier\n");
2633 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2634 	}
2635 #endif
2636 
2637 	iounmap(ibdev->uar_map);
2638 	for (p = 0; p < ibdev->num_ports; ++p)
2639 		if (ibdev->counters[p].index != -1 &&
2640 		    ibdev->counters[p].allocated)
2641 			mlx4_counter_free(ibdev->dev, ibdev->counters[p].index);
2642 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2643 		mlx4_CLOSE_PORT(dev, p);
2644 
2645 	mlx4_ib_free_eqs(dev, ibdev);
2646 
2647 	mlx4_uar_free(dev, &ibdev->priv_uar);
2648 	mlx4_pd_free(dev, ibdev->priv_pdn);
2649 	ib_dealloc_device(&ibdev->ib_dev);
2650 }
2651 
2652 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2653 {
2654 	struct mlx4_ib_demux_work **dm = NULL;
2655 	struct mlx4_dev *dev = ibdev->dev;
2656 	int i;
2657 	unsigned long flags;
2658 	struct mlx4_active_ports actv_ports;
2659 	unsigned int ports;
2660 	unsigned int first_port;
2661 
2662 	if (!mlx4_is_master(dev))
2663 		return;
2664 
2665 	actv_ports = mlx4_get_active_ports(dev, slave);
2666 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2667 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2668 
2669 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2670 	if (!dm) {
2671 		pr_err("failed to allocate memory for tunneling qp update\n");
2672 		goto out;
2673 	}
2674 
2675 	for (i = 0; i < ports; i++) {
2676 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2677 		if (!dm[i]) {
2678 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2679 			for (i = 0; i < dev->caps.num_ports; i++) {
2680 				if (dm[i])
2681 					kfree(dm[i]);
2682 			}
2683 			goto out;
2684 		}
2685 	}
2686 	/* initialize or tear down tunnel QPs for the slave */
2687 	for (i = 0; i < ports; i++) {
2688 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2689 		dm[i]->port = first_port + i + 1;
2690 		dm[i]->slave = slave;
2691 		dm[i]->do_init = do_init;
2692 		dm[i]->dev = ibdev;
2693 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2694 		if (!ibdev->sriov.is_going_down)
2695 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2696 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2697 	}
2698 out:
2699 	kfree(dm);
2700 	return;
2701 }
2702 
2703 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2704 {
2705 	struct mlx4_ib_qp *mqp;
2706 	unsigned long flags_qp;
2707 	unsigned long flags_cq;
2708 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2709 	struct list_head    cq_notify_list;
2710 	struct mlx4_cq *mcq;
2711 	unsigned long flags;
2712 
2713 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2714 	INIT_LIST_HEAD(&cq_notify_list);
2715 
2716 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2717 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2718 
2719 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2720 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2721 		if (mqp->sq.tail != mqp->sq.head) {
2722 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2723 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2724 			if (send_mcq->mcq.comp &&
2725 			    mqp->ibqp.send_cq->comp_handler) {
2726 				if (!send_mcq->mcq.reset_notify_added) {
2727 					send_mcq->mcq.reset_notify_added = 1;
2728 					list_add_tail(&send_mcq->mcq.reset_notify,
2729 						      &cq_notify_list);
2730 				}
2731 			}
2732 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2733 		}
2734 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2735 		/* Now, handle the QP's receive queue */
2736 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2737 		/* no handling is needed for SRQ */
2738 		if (!mqp->ibqp.srq) {
2739 			if (mqp->rq.tail != mqp->rq.head) {
2740 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2741 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2742 				if (recv_mcq->mcq.comp &&
2743 				    mqp->ibqp.recv_cq->comp_handler) {
2744 					if (!recv_mcq->mcq.reset_notify_added) {
2745 						recv_mcq->mcq.reset_notify_added = 1;
2746 						list_add_tail(&recv_mcq->mcq.reset_notify,
2747 							      &cq_notify_list);
2748 					}
2749 				}
2750 				spin_unlock_irqrestore(&recv_mcq->lock,
2751 						       flags_cq);
2752 			}
2753 		}
2754 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2755 	}
2756 
2757 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2758 		mcq->comp(mcq);
2759 	}
2760 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2761 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2762 }
2763 
2764 static void handle_bonded_port_state_event(struct work_struct *work)
2765 {
2766 	struct ib_event_work *ew =
2767 		container_of(work, struct ib_event_work, work);
2768 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2769 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2770 	int i;
2771 	struct ib_event ibev;
2772 
2773 	kfree(ew);
2774 	spin_lock_bh(&ibdev->iboe.lock);
2775 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2776 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2777 		enum ib_port_state curr_port_state;
2778 
2779 		if (!curr_netdev)
2780 			continue;
2781 
2782 		curr_port_state =
2783 			(netif_running(curr_netdev) &&
2784 			 netif_carrier_ok(curr_netdev)) ?
2785 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2786 
2787 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2788 			curr_port_state : IB_PORT_ACTIVE;
2789 	}
2790 	spin_unlock_bh(&ibdev->iboe.lock);
2791 
2792 	ibev.device = &ibdev->ib_dev;
2793 	ibev.element.port_num = 1;
2794 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2795 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2796 
2797 	ib_dispatch_event(&ibev);
2798 }
2799 
2800 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2801 			  enum mlx4_dev_event event, unsigned long param)
2802 {
2803 	struct ib_event ibev;
2804 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2805 	struct mlx4_eqe *eqe = NULL;
2806 	struct ib_event_work *ew;
2807 	int p = 0;
2808 
2809 	if (mlx4_is_bonded(dev) &&
2810 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2811 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2812 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2813 		if (!ew)
2814 			return;
2815 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2816 		ew->ib_dev = ibdev;
2817 		queue_work(wq, &ew->work);
2818 		return;
2819 	}
2820 
2821 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2822 		eqe = (struct mlx4_eqe *)param;
2823 	else
2824 		p = (int) param;
2825 
2826 	switch (event) {
2827 	case MLX4_DEV_EVENT_PORT_UP:
2828 		if (p > ibdev->num_ports)
2829 			return;
2830 		if (mlx4_is_master(dev) &&
2831 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2832 			IB_LINK_LAYER_INFINIBAND) {
2833 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2834 		}
2835 		ibev.event = IB_EVENT_PORT_ACTIVE;
2836 		break;
2837 
2838 	case MLX4_DEV_EVENT_PORT_DOWN:
2839 		if (p > ibdev->num_ports)
2840 			return;
2841 		ibev.event = IB_EVENT_PORT_ERR;
2842 		break;
2843 
2844 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2845 		ibdev->ib_active = false;
2846 		ibev.event = IB_EVENT_DEVICE_FATAL;
2847 		mlx4_ib_handle_catas_error(ibdev);
2848 		break;
2849 
2850 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2851 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2852 		if (!ew) {
2853 			pr_err("failed to allocate memory for events work\n");
2854 			break;
2855 		}
2856 
2857 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2858 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2859 		ew->ib_dev = ibdev;
2860 		/* need to queue only for port owner, which uses GEN_EQE */
2861 		if (mlx4_is_master(dev))
2862 			queue_work(wq, &ew->work);
2863 		else
2864 			handle_port_mgmt_change_event(&ew->work);
2865 		return;
2866 
2867 	case MLX4_DEV_EVENT_SLAVE_INIT:
2868 		/* here, p is the slave id */
2869 		do_slave_init(ibdev, p, 1);
2870 		if (mlx4_is_master(dev)) {
2871 			int i;
2872 
2873 			for (i = 1; i <= ibdev->num_ports; i++) {
2874 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2875 					== IB_LINK_LAYER_INFINIBAND)
2876 					mlx4_ib_slave_alias_guid_event(ibdev,
2877 								       p, i,
2878 								       1);
2879 			}
2880 		}
2881 		return;
2882 
2883 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2884 		if (mlx4_is_master(dev)) {
2885 			int i;
2886 
2887 			for (i = 1; i <= ibdev->num_ports; i++) {
2888 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2889 					== IB_LINK_LAYER_INFINIBAND)
2890 					mlx4_ib_slave_alias_guid_event(ibdev,
2891 								       p, i,
2892 								       0);
2893 			}
2894 		}
2895 		/* here, p is the slave id */
2896 		do_slave_init(ibdev, p, 0);
2897 		return;
2898 
2899 	default:
2900 		return;
2901 	}
2902 
2903 	ibev.device	      = ibdev_ptr;
2904 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2905 
2906 	ib_dispatch_event(&ibev);
2907 }
2908 
2909 static struct mlx4_interface mlx4_ib_interface = {
2910 	.add		= mlx4_ib_add,
2911 	.remove		= mlx4_ib_remove,
2912 	.event		= mlx4_ib_event,
2913 	.protocol	= MLX4_PROT_IB_IPV6,
2914 	.flags		= MLX4_INTFF_BONDING
2915 };
2916 
2917 static int __init mlx4_ib_init(void)
2918 {
2919 	int err;
2920 
2921 	wq = create_singlethread_workqueue("mlx4_ib");
2922 	if (!wq)
2923 		return -ENOMEM;
2924 
2925 	err = mlx4_ib_mcg_init();
2926 	if (err)
2927 		goto clean_wq;
2928 
2929 	err = mlx4_register_interface(&mlx4_ib_interface);
2930 	if (err)
2931 		goto clean_mcg;
2932 
2933 	return 0;
2934 
2935 clean_mcg:
2936 	mlx4_ib_mcg_destroy();
2937 
2938 clean_wq:
2939 	destroy_workqueue(wq);
2940 	return err;
2941 }
2942 
2943 static void __exit mlx4_ib_cleanup(void)
2944 {
2945 	mlx4_unregister_interface(&mlx4_ib_interface);
2946 	mlx4_ib_mcg_destroy();
2947 	destroy_workqueue(wq);
2948 }
2949 
2950 module_init(mlx4_ib_init);
2951 module_exit(mlx4_ib_cleanup);
2952