xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 3d3337de)
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 {
137 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
138 	struct ib_smp *in_mad  = NULL;
139 	struct ib_smp *out_mad = NULL;
140 	int err = -ENOMEM;
141 	int have_ib_ports;
142 
143 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
144 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
145 	if (!in_mad || !out_mad)
146 		goto out;
147 
148 	init_query_mad(in_mad);
149 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
150 
151 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
152 			   1, NULL, NULL, in_mad, out_mad);
153 	if (err)
154 		goto out;
155 
156 	memset(props, 0, sizeof *props);
157 
158 	have_ib_ports = num_ib_ports(dev->dev);
159 
160 	props->fw_ver = dev->dev->caps.fw_ver;
161 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
162 		IB_DEVICE_PORT_ACTIVE_EVENT		|
163 		IB_DEVICE_SYS_IMAGE_GUID		|
164 		IB_DEVICE_RC_RNR_NAK_GEN		|
165 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
166 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
167 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
168 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
169 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
170 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
171 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
172 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
173 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
174 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
175 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
176 	if (dev->dev->caps.max_gso_sz &&
177 	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
178 	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
179 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
180 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
181 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
182 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
183 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
184 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
185 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
186 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
187 		props->device_cap_flags |= IB_DEVICE_XRC;
188 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
189 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
190 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
191 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
192 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
193 		else
194 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
195 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
196 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
197 	}
198 
199 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
200 		0xffffff;
201 	props->vendor_part_id	   = dev->dev->persist->pdev->device;
202 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
203 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
204 
205 	props->max_mr_size	   = ~0ull;
206 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
207 	props->max_qp		   = dev->dev->quotas.qp;
208 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
209 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
210 					 dev->dev->caps.max_rq_sg);
211 	props->max_cq		   = dev->dev->quotas.cq;
212 	props->max_cqe		   = dev->dev->caps.max_cqes;
213 	props->max_mr		   = dev->dev->quotas.mpt;
214 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
215 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
216 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
217 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
218 	props->max_srq		   = dev->dev->quotas.srq;
219 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
220 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
221 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
222 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
223 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
224 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
225 	props->masked_atomic_cap   = props->atomic_cap;
226 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
227 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
228 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
229 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
230 					   props->max_mcast_grp;
231 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
232 
233 out:
234 	kfree(in_mad);
235 	kfree(out_mad);
236 
237 	return err;
238 }
239 
240 static enum rdma_link_layer
241 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
242 {
243 	struct mlx4_dev *dev = to_mdev(device)->dev;
244 
245 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
246 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
247 }
248 
249 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
250 			      struct ib_port_attr *props, int netw_view)
251 {
252 	struct ib_smp *in_mad  = NULL;
253 	struct ib_smp *out_mad = NULL;
254 	int ext_active_speed;
255 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
256 	int err = -ENOMEM;
257 
258 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
259 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
260 	if (!in_mad || !out_mad)
261 		goto out;
262 
263 	init_query_mad(in_mad);
264 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
265 	in_mad->attr_mod = cpu_to_be32(port);
266 
267 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
268 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
269 
270 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
271 				in_mad, out_mad);
272 	if (err)
273 		goto out;
274 
275 
276 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
277 	props->lmc		= out_mad->data[34] & 0x7;
278 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
279 	props->sm_sl		= out_mad->data[36] & 0xf;
280 	props->state		= out_mad->data[32] & 0xf;
281 	props->phys_state	= out_mad->data[33] >> 4;
282 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
283 	if (netw_view)
284 		props->gid_tbl_len = out_mad->data[50];
285 	else
286 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
287 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
288 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
289 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
290 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
291 	props->active_width	= out_mad->data[31] & 0xf;
292 	props->active_speed	= out_mad->data[35] >> 4;
293 	props->max_mtu		= out_mad->data[41] & 0xf;
294 	props->active_mtu	= out_mad->data[36] >> 4;
295 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
296 	props->max_vl_num	= out_mad->data[37] >> 4;
297 	props->init_type_reply	= out_mad->data[41] >> 4;
298 
299 	/* Check if extended speeds (EDR/FDR/...) are supported */
300 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
301 		ext_active_speed = out_mad->data[62] >> 4;
302 
303 		switch (ext_active_speed) {
304 		case 1:
305 			props->active_speed = IB_SPEED_FDR;
306 			break;
307 		case 2:
308 			props->active_speed = IB_SPEED_EDR;
309 			break;
310 		}
311 	}
312 
313 	/* If reported active speed is QDR, check if is FDR-10 */
314 	if (props->active_speed == IB_SPEED_QDR) {
315 		init_query_mad(in_mad);
316 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
317 		in_mad->attr_mod = cpu_to_be32(port);
318 
319 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
320 				   NULL, NULL, in_mad, out_mad);
321 		if (err)
322 			goto out;
323 
324 		/* Checking LinkSpeedActive for FDR-10 */
325 		if (out_mad->data[15] & 0x1)
326 			props->active_speed = IB_SPEED_FDR10;
327 	}
328 
329 	/* Avoid wrong speed value returned by FW if the IB link is down. */
330 	if (props->state == IB_PORT_DOWN)
331 		 props->active_speed = IB_SPEED_SDR;
332 
333 out:
334 	kfree(in_mad);
335 	kfree(out_mad);
336 	return err;
337 }
338 
339 static u8 state_to_phys_state(enum ib_port_state state)
340 {
341 	return state == IB_PORT_ACTIVE ? 5 : 3;
342 }
343 
344 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
345 			       struct ib_port_attr *props, int netw_view)
346 {
347 
348 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
349 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
350 	struct net_device *ndev;
351 	enum ib_mtu tmp;
352 	struct mlx4_cmd_mailbox *mailbox;
353 	int err = 0;
354 	int is_bonded = mlx4_is_bonded(mdev->dev);
355 
356 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
357 	if (IS_ERR(mailbox))
358 		return PTR_ERR(mailbox);
359 
360 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
361 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
362 			   MLX4_CMD_WRAPPED);
363 	if (err)
364 		goto out;
365 
366 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
367 						IB_WIDTH_4X : IB_WIDTH_1X;
368 	props->active_speed	= IB_SPEED_QDR;
369 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
370 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
371 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
372 	props->pkey_tbl_len	= 1;
373 	props->max_mtu		= IB_MTU_4096;
374 	props->max_vl_num	= 2;
375 	props->state		= IB_PORT_DOWN;
376 	props->phys_state	= state_to_phys_state(props->state);
377 	props->active_mtu	= IB_MTU_256;
378 	if (is_bonded)
379 		rtnl_lock(); /* required to get upper dev */
380 	spin_lock_bh(&iboe->lock);
381 	ndev = iboe->netdevs[port - 1];
382 	if (ndev && is_bonded)
383 		ndev = netdev_master_upper_dev_get(ndev);
384 	if (!ndev)
385 		goto out_unlock;
386 
387 	tmp = iboe_get_mtu(ndev->mtu);
388 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
389 
390 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
391 					IB_PORT_ACTIVE : IB_PORT_DOWN;
392 	props->phys_state	= state_to_phys_state(props->state);
393 out_unlock:
394 	spin_unlock_bh(&iboe->lock);
395 	if (is_bonded)
396 		rtnl_unlock();
397 out:
398 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
399 	return err;
400 }
401 
402 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
403 			 struct ib_port_attr *props, int netw_view)
404 {
405 	int err;
406 
407 	memset(props, 0, sizeof *props);
408 
409 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
410 		ib_link_query_port(ibdev, port, props, netw_view) :
411 				eth_link_query_port(ibdev, port, props, netw_view);
412 
413 	return err;
414 }
415 
416 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
417 			      struct ib_port_attr *props)
418 {
419 	/* returns host view */
420 	return __mlx4_ib_query_port(ibdev, port, props, 0);
421 }
422 
423 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
424 			union ib_gid *gid, int netw_view)
425 {
426 	struct ib_smp *in_mad  = NULL;
427 	struct ib_smp *out_mad = NULL;
428 	int err = -ENOMEM;
429 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
430 	int clear = 0;
431 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
432 
433 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
434 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
435 	if (!in_mad || !out_mad)
436 		goto out;
437 
438 	init_query_mad(in_mad);
439 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
440 	in_mad->attr_mod = cpu_to_be32(port);
441 
442 	if (mlx4_is_mfunc(dev->dev) && netw_view)
443 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
444 
445 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
446 	if (err)
447 		goto out;
448 
449 	memcpy(gid->raw, out_mad->data + 8, 8);
450 
451 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
452 		if (index) {
453 			/* For any index > 0, return the null guid */
454 			err = 0;
455 			clear = 1;
456 			goto out;
457 		}
458 	}
459 
460 	init_query_mad(in_mad);
461 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
462 	in_mad->attr_mod = cpu_to_be32(index / 8);
463 
464 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
465 			   NULL, NULL, in_mad, out_mad);
466 	if (err)
467 		goto out;
468 
469 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
470 
471 out:
472 	if (clear)
473 		memset(gid->raw + 8, 0, 8);
474 	kfree(in_mad);
475 	kfree(out_mad);
476 	return err;
477 }
478 
479 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
480 			  union ib_gid *gid)
481 {
482 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
483 
484 	*gid = dev->iboe.gid_table[port - 1][index];
485 
486 	return 0;
487 }
488 
489 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
490 			     union ib_gid *gid)
491 {
492 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
493 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
494 	else
495 		return iboe_query_gid(ibdev, port, index, gid);
496 }
497 
498 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
499 			 u16 *pkey, int netw_view)
500 {
501 	struct ib_smp *in_mad  = NULL;
502 	struct ib_smp *out_mad = NULL;
503 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
504 	int err = -ENOMEM;
505 
506 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
507 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
508 	if (!in_mad || !out_mad)
509 		goto out;
510 
511 	init_query_mad(in_mad);
512 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
513 	in_mad->attr_mod = cpu_to_be32(index / 32);
514 
515 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
516 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
517 
518 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
519 			   in_mad, out_mad);
520 	if (err)
521 		goto out;
522 
523 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
524 
525 out:
526 	kfree(in_mad);
527 	kfree(out_mad);
528 	return err;
529 }
530 
531 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
532 {
533 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
534 }
535 
536 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
537 				 struct ib_device_modify *props)
538 {
539 	struct mlx4_cmd_mailbox *mailbox;
540 	unsigned long flags;
541 
542 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
543 		return -EOPNOTSUPP;
544 
545 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
546 		return 0;
547 
548 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
549 		return -EOPNOTSUPP;
550 
551 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
552 	memcpy(ibdev->node_desc, props->node_desc, 64);
553 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
554 
555 	/*
556 	 * If possible, pass node desc to FW, so it can generate
557 	 * a 144 trap.  If cmd fails, just ignore.
558 	 */
559 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
560 	if (IS_ERR(mailbox))
561 		return 0;
562 
563 	memcpy(mailbox->buf, props->node_desc, 64);
564 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
565 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
566 
567 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
568 
569 	return 0;
570 }
571 
572 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
573 			    u32 cap_mask)
574 {
575 	struct mlx4_cmd_mailbox *mailbox;
576 	int err;
577 
578 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
579 	if (IS_ERR(mailbox))
580 		return PTR_ERR(mailbox);
581 
582 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
583 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
584 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
585 	} else {
586 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
587 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
588 	}
589 
590 	err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
591 		       MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
592 		       MLX4_CMD_WRAPPED);
593 
594 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
595 	return err;
596 }
597 
598 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
599 			       struct ib_port_modify *props)
600 {
601 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
602 	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
603 	struct ib_port_attr attr;
604 	u32 cap_mask;
605 	int err;
606 
607 	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
608 	 * of whether port link layer is ETH or IB. For ETH ports, qkey
609 	 * violations and port capabilities are not meaningful.
610 	 */
611 	if (is_eth)
612 		return 0;
613 
614 	mutex_lock(&mdev->cap_mask_mutex);
615 
616 	err = mlx4_ib_query_port(ibdev, port, &attr);
617 	if (err)
618 		goto out;
619 
620 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
621 		~props->clr_port_cap_mask;
622 
623 	err = mlx4_ib_SET_PORT(mdev, port,
624 			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
625 			       cap_mask);
626 
627 out:
628 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
629 	return err;
630 }
631 
632 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
633 						  struct ib_udata *udata)
634 {
635 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
636 	struct mlx4_ib_ucontext *context;
637 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
638 	struct mlx4_ib_alloc_ucontext_resp resp;
639 	int err;
640 
641 	if (!dev->ib_active)
642 		return ERR_PTR(-EAGAIN);
643 
644 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
645 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
646 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
647 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
648 	} else {
649 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
650 		resp.qp_tab_size      = dev->dev->caps.num_qps;
651 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
652 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
653 		resp.cqe_size	      = dev->dev->caps.cqe_size;
654 	}
655 
656 	context = kmalloc(sizeof *context, GFP_KERNEL);
657 	if (!context)
658 		return ERR_PTR(-ENOMEM);
659 
660 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
661 	if (err) {
662 		kfree(context);
663 		return ERR_PTR(err);
664 	}
665 
666 	INIT_LIST_HEAD(&context->db_page_list);
667 	mutex_init(&context->db_page_mutex);
668 
669 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
670 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
671 	else
672 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
673 
674 	if (err) {
675 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
676 		kfree(context);
677 		return ERR_PTR(-EFAULT);
678 	}
679 
680 	return &context->ibucontext;
681 }
682 
683 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
684 {
685 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
686 
687 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
688 	kfree(context);
689 
690 	return 0;
691 }
692 
693 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
694 {
695 	struct mlx4_ib_dev *dev = to_mdev(context->device);
696 
697 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
698 		return -EINVAL;
699 
700 	if (vma->vm_pgoff == 0) {
701 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
702 
703 		if (io_remap_pfn_range(vma, vma->vm_start,
704 				       to_mucontext(context)->uar.pfn,
705 				       PAGE_SIZE, vma->vm_page_prot))
706 			return -EAGAIN;
707 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
708 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
709 
710 		if (io_remap_pfn_range(vma, vma->vm_start,
711 				       to_mucontext(context)->uar.pfn +
712 				       dev->dev->caps.num_uars,
713 				       PAGE_SIZE, vma->vm_page_prot))
714 			return -EAGAIN;
715 	} else
716 		return -EINVAL;
717 
718 	return 0;
719 }
720 
721 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
722 				      struct ib_ucontext *context,
723 				      struct ib_udata *udata)
724 {
725 	struct mlx4_ib_pd *pd;
726 	int err;
727 
728 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
729 	if (!pd)
730 		return ERR_PTR(-ENOMEM);
731 
732 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
733 	if (err) {
734 		kfree(pd);
735 		return ERR_PTR(err);
736 	}
737 
738 	if (context)
739 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
740 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
741 			kfree(pd);
742 			return ERR_PTR(-EFAULT);
743 		}
744 
745 	return &pd->ibpd;
746 }
747 
748 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
749 {
750 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
751 	kfree(pd);
752 
753 	return 0;
754 }
755 
756 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
757 					  struct ib_ucontext *context,
758 					  struct ib_udata *udata)
759 {
760 	struct mlx4_ib_xrcd *xrcd;
761 	int err;
762 
763 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
764 		return ERR_PTR(-ENOSYS);
765 
766 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
767 	if (!xrcd)
768 		return ERR_PTR(-ENOMEM);
769 
770 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
771 	if (err)
772 		goto err1;
773 
774 	xrcd->pd = ib_alloc_pd(ibdev);
775 	if (IS_ERR(xrcd->pd)) {
776 		err = PTR_ERR(xrcd->pd);
777 		goto err2;
778 	}
779 
780 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
781 	if (IS_ERR(xrcd->cq)) {
782 		err = PTR_ERR(xrcd->cq);
783 		goto err3;
784 	}
785 
786 	return &xrcd->ibxrcd;
787 
788 err3:
789 	ib_dealloc_pd(xrcd->pd);
790 err2:
791 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
792 err1:
793 	kfree(xrcd);
794 	return ERR_PTR(err);
795 }
796 
797 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
798 {
799 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
800 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
801 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
802 	kfree(xrcd);
803 
804 	return 0;
805 }
806 
807 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
808 {
809 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
810 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
811 	struct mlx4_ib_gid_entry *ge;
812 
813 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
814 	if (!ge)
815 		return -ENOMEM;
816 
817 	ge->gid = *gid;
818 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
819 		ge->port = mqp->port;
820 		ge->added = 1;
821 	}
822 
823 	mutex_lock(&mqp->mutex);
824 	list_add_tail(&ge->list, &mqp->gid_list);
825 	mutex_unlock(&mqp->mutex);
826 
827 	return 0;
828 }
829 
830 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
831 		   union ib_gid *gid)
832 {
833 	struct net_device *ndev;
834 	int ret = 0;
835 
836 	if (!mqp->port)
837 		return 0;
838 
839 	spin_lock_bh(&mdev->iboe.lock);
840 	ndev = mdev->iboe.netdevs[mqp->port - 1];
841 	if (ndev)
842 		dev_hold(ndev);
843 	spin_unlock_bh(&mdev->iboe.lock);
844 
845 	if (ndev) {
846 		ret = 1;
847 		dev_put(ndev);
848 	}
849 
850 	return ret;
851 }
852 
853 struct mlx4_ib_steering {
854 	struct list_head list;
855 	struct mlx4_flow_reg_id reg_id;
856 	union ib_gid gid;
857 };
858 
859 static int parse_flow_attr(struct mlx4_dev *dev,
860 			   u32 qp_num,
861 			   union ib_flow_spec *ib_spec,
862 			   struct _rule_hw *mlx4_spec)
863 {
864 	enum mlx4_net_trans_rule_id type;
865 
866 	switch (ib_spec->type) {
867 	case IB_FLOW_SPEC_ETH:
868 		type = MLX4_NET_TRANS_RULE_ID_ETH;
869 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
870 		       ETH_ALEN);
871 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
872 		       ETH_ALEN);
873 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
874 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
875 		break;
876 	case IB_FLOW_SPEC_IB:
877 		type = MLX4_NET_TRANS_RULE_ID_IB;
878 		mlx4_spec->ib.l3_qpn =
879 			cpu_to_be32(qp_num);
880 		mlx4_spec->ib.qpn_mask =
881 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
882 		break;
883 
884 
885 	case IB_FLOW_SPEC_IPV4:
886 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
887 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
888 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
889 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
890 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
891 		break;
892 
893 	case IB_FLOW_SPEC_TCP:
894 	case IB_FLOW_SPEC_UDP:
895 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
896 					MLX4_NET_TRANS_RULE_ID_TCP :
897 					MLX4_NET_TRANS_RULE_ID_UDP;
898 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
899 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
900 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
901 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
902 		break;
903 
904 	default:
905 		return -EINVAL;
906 	}
907 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
908 	    mlx4_hw_rule_sz(dev, type) < 0)
909 		return -EINVAL;
910 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
911 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
912 	return mlx4_hw_rule_sz(dev, type);
913 }
914 
915 struct default_rules {
916 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
917 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
918 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
919 	__u8  link_layer;
920 };
921 static const struct default_rules default_table[] = {
922 	{
923 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
924 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
925 		.rules_create_list = {IB_FLOW_SPEC_IB},
926 		.link_layer = IB_LINK_LAYER_INFINIBAND
927 	}
928 };
929 
930 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
931 					 struct ib_flow_attr *flow_attr)
932 {
933 	int i, j, k;
934 	void *ib_flow;
935 	const struct default_rules *pdefault_rules = default_table;
936 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
937 
938 	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
939 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
940 		memset(&field_types, 0, sizeof(field_types));
941 
942 		if (link_layer != pdefault_rules->link_layer)
943 			continue;
944 
945 		ib_flow = flow_attr + 1;
946 		/* we assume the specs are sorted */
947 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
948 		     j < flow_attr->num_of_specs; k++) {
949 			union ib_flow_spec *current_flow =
950 				(union ib_flow_spec *)ib_flow;
951 
952 			/* same layer but different type */
953 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
954 			     (pdefault_rules->mandatory_fields[k] &
955 			      IB_FLOW_SPEC_LAYER_MASK)) &&
956 			    (current_flow->type !=
957 			     pdefault_rules->mandatory_fields[k]))
958 				goto out;
959 
960 			/* same layer, try match next one */
961 			if (current_flow->type ==
962 			    pdefault_rules->mandatory_fields[k]) {
963 				j++;
964 				ib_flow +=
965 					((union ib_flow_spec *)ib_flow)->size;
966 			}
967 		}
968 
969 		ib_flow = flow_attr + 1;
970 		for (j = 0; j < flow_attr->num_of_specs;
971 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
972 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
973 				/* same layer and same type */
974 				if (((union ib_flow_spec *)ib_flow)->type ==
975 				    pdefault_rules->mandatory_not_fields[k])
976 					goto out;
977 
978 		return i;
979 	}
980 out:
981 	return -1;
982 }
983 
984 static int __mlx4_ib_create_default_rules(
985 		struct mlx4_ib_dev *mdev,
986 		struct ib_qp *qp,
987 		const struct default_rules *pdefault_rules,
988 		struct _rule_hw *mlx4_spec) {
989 	int size = 0;
990 	int i;
991 
992 	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
993 		int ret;
994 		union ib_flow_spec ib_spec;
995 		switch (pdefault_rules->rules_create_list[i]) {
996 		case 0:
997 			/* no rule */
998 			continue;
999 		case IB_FLOW_SPEC_IB:
1000 			ib_spec.type = IB_FLOW_SPEC_IB;
1001 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
1002 
1003 			break;
1004 		default:
1005 			/* invalid rule */
1006 			return -EINVAL;
1007 		}
1008 		/* We must put empty rule, qpn is being ignored */
1009 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1010 				      mlx4_spec);
1011 		if (ret < 0) {
1012 			pr_info("invalid parsing\n");
1013 			return -EINVAL;
1014 		}
1015 
1016 		mlx4_spec = (void *)mlx4_spec + ret;
1017 		size += ret;
1018 	}
1019 	return size;
1020 }
1021 
1022 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1023 			  int domain,
1024 			  enum mlx4_net_trans_promisc_mode flow_type,
1025 			  u64 *reg_id)
1026 {
1027 	int ret, i;
1028 	int size = 0;
1029 	void *ib_flow;
1030 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1031 	struct mlx4_cmd_mailbox *mailbox;
1032 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1033 	int default_flow;
1034 
1035 	static const u16 __mlx4_domain[] = {
1036 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1037 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1038 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1039 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1040 	};
1041 
1042 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1043 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1044 		return -EINVAL;
1045 	}
1046 
1047 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1048 		pr_err("Invalid domain value %d\n", domain);
1049 		return -EINVAL;
1050 	}
1051 
1052 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1053 		return -EINVAL;
1054 
1055 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1056 	if (IS_ERR(mailbox))
1057 		return PTR_ERR(mailbox);
1058 	ctrl = mailbox->buf;
1059 
1060 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1061 				 flow_attr->priority);
1062 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1063 	ctrl->port = flow_attr->port;
1064 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1065 
1066 	ib_flow = flow_attr + 1;
1067 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1068 	/* Add default flows */
1069 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1070 	if (default_flow >= 0) {
1071 		ret = __mlx4_ib_create_default_rules(
1072 				mdev, qp, default_table + default_flow,
1073 				mailbox->buf + size);
1074 		if (ret < 0) {
1075 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1076 			return -EINVAL;
1077 		}
1078 		size += ret;
1079 	}
1080 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1081 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1082 				      mailbox->buf + size);
1083 		if (ret < 0) {
1084 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1085 			return -EINVAL;
1086 		}
1087 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1088 		size += ret;
1089 	}
1090 
1091 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1092 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1093 			   MLX4_CMD_NATIVE);
1094 	if (ret == -ENOMEM)
1095 		pr_err("mcg table is full. Fail to register network rule.\n");
1096 	else if (ret == -ENXIO)
1097 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1098 	else if (ret)
1099 		pr_err("Invalid argumant. Fail to register network rule.\n");
1100 
1101 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1102 	return ret;
1103 }
1104 
1105 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1106 {
1107 	int err;
1108 	err = mlx4_cmd(dev, reg_id, 0, 0,
1109 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1110 		       MLX4_CMD_NATIVE);
1111 	if (err)
1112 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1113 		       reg_id);
1114 	return err;
1115 }
1116 
1117 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1118 				    u64 *reg_id)
1119 {
1120 	void *ib_flow;
1121 	union ib_flow_spec *ib_spec;
1122 	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
1123 	int err = 0;
1124 
1125 	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1126 	    dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1127 		return 0; /* do nothing */
1128 
1129 	ib_flow = flow_attr + 1;
1130 	ib_spec = (union ib_flow_spec *)ib_flow;
1131 
1132 	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1133 		return 0; /* do nothing */
1134 
1135 	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1136 				    flow_attr->port, qp->qp_num,
1137 				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1138 				    reg_id);
1139 	return err;
1140 }
1141 
1142 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1143 				    struct ib_flow_attr *flow_attr,
1144 				    int domain)
1145 {
1146 	int err = 0, i = 0, j = 0;
1147 	struct mlx4_ib_flow *mflow;
1148 	enum mlx4_net_trans_promisc_mode type[2];
1149 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1150 	int is_bonded = mlx4_is_bonded(dev);
1151 
1152 	memset(type, 0, sizeof(type));
1153 
1154 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1155 	if (!mflow) {
1156 		err = -ENOMEM;
1157 		goto err_free;
1158 	}
1159 
1160 	switch (flow_attr->type) {
1161 	case IB_FLOW_ATTR_NORMAL:
1162 		type[0] = MLX4_FS_REGULAR;
1163 		break;
1164 
1165 	case IB_FLOW_ATTR_ALL_DEFAULT:
1166 		type[0] = MLX4_FS_ALL_DEFAULT;
1167 		break;
1168 
1169 	case IB_FLOW_ATTR_MC_DEFAULT:
1170 		type[0] = MLX4_FS_MC_DEFAULT;
1171 		break;
1172 
1173 	case IB_FLOW_ATTR_SNIFFER:
1174 		type[0] = MLX4_FS_UC_SNIFFER;
1175 		type[1] = MLX4_FS_MC_SNIFFER;
1176 		break;
1177 
1178 	default:
1179 		err = -EINVAL;
1180 		goto err_free;
1181 	}
1182 
1183 	while (i < ARRAY_SIZE(type) && type[i]) {
1184 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1185 					    &mflow->reg_id[i].id);
1186 		if (err)
1187 			goto err_create_flow;
1188 		i++;
1189 		if (is_bonded) {
1190 			/* Application always sees one port so the mirror rule
1191 			 * must be on port #2
1192 			 */
1193 			flow_attr->port = 2;
1194 			err = __mlx4_ib_create_flow(qp, flow_attr,
1195 						    domain, type[j],
1196 						    &mflow->reg_id[j].mirror);
1197 			flow_attr->port = 1;
1198 			if (err)
1199 				goto err_create_flow;
1200 			j++;
1201 		}
1202 
1203 	}
1204 
1205 	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1206 		err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1207 					       &mflow->reg_id[i].id);
1208 		if (err)
1209 			goto err_create_flow;
1210 		i++;
1211 		if (is_bonded) {
1212 			flow_attr->port = 2;
1213 			err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1214 						       &mflow->reg_id[j].mirror);
1215 			flow_attr->port = 1;
1216 			if (err)
1217 				goto err_create_flow;
1218 			j++;
1219 		}
1220 		/* function to create mirror rule */
1221 	}
1222 
1223 	return &mflow->ibflow;
1224 
1225 err_create_flow:
1226 	while (i) {
1227 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1228 					     mflow->reg_id[i].id);
1229 		i--;
1230 	}
1231 
1232 	while (j) {
1233 		(void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1234 					     mflow->reg_id[j].mirror);
1235 		j--;
1236 	}
1237 err_free:
1238 	kfree(mflow);
1239 	return ERR_PTR(err);
1240 }
1241 
1242 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1243 {
1244 	int err, ret = 0;
1245 	int i = 0;
1246 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1247 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1248 
1249 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1250 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1251 		if (err)
1252 			ret = err;
1253 		if (mflow->reg_id[i].mirror) {
1254 			err = __mlx4_ib_destroy_flow(mdev->dev,
1255 						     mflow->reg_id[i].mirror);
1256 			if (err)
1257 				ret = err;
1258 		}
1259 		i++;
1260 	}
1261 
1262 	kfree(mflow);
1263 	return ret;
1264 }
1265 
1266 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1267 {
1268 	int err;
1269 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1270 	struct mlx4_dev	*dev = mdev->dev;
1271 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1272 	struct mlx4_ib_steering *ib_steering = NULL;
1273 	enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1274 	struct mlx4_flow_reg_id	reg_id;
1275 
1276 	if (mdev->dev->caps.steering_mode ==
1277 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1278 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1279 		if (!ib_steering)
1280 			return -ENOMEM;
1281 	}
1282 
1283 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1284 				    !!(mqp->flags &
1285 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1286 				    prot, &reg_id.id);
1287 	if (err) {
1288 		pr_err("multicast attach op failed, err %d\n", err);
1289 		goto err_malloc;
1290 	}
1291 
1292 	reg_id.mirror = 0;
1293 	if (mlx4_is_bonded(dev)) {
1294 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1295 					    (mqp->port == 1) ? 2 : 1,
1296 					    !!(mqp->flags &
1297 					    MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1298 					    prot, &reg_id.mirror);
1299 		if (err)
1300 			goto err_add;
1301 	}
1302 
1303 	err = add_gid_entry(ibqp, gid);
1304 	if (err)
1305 		goto err_add;
1306 
1307 	if (ib_steering) {
1308 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1309 		ib_steering->reg_id = reg_id;
1310 		mutex_lock(&mqp->mutex);
1311 		list_add(&ib_steering->list, &mqp->steering_rules);
1312 		mutex_unlock(&mqp->mutex);
1313 	}
1314 	return 0;
1315 
1316 err_add:
1317 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1318 			      prot, reg_id.id);
1319 	if (reg_id.mirror)
1320 		mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1321 				      prot, reg_id.mirror);
1322 err_malloc:
1323 	kfree(ib_steering);
1324 
1325 	return err;
1326 }
1327 
1328 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1329 {
1330 	struct mlx4_ib_gid_entry *ge;
1331 	struct mlx4_ib_gid_entry *tmp;
1332 	struct mlx4_ib_gid_entry *ret = NULL;
1333 
1334 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1335 		if (!memcmp(raw, ge->gid.raw, 16)) {
1336 			ret = ge;
1337 			break;
1338 		}
1339 	}
1340 
1341 	return ret;
1342 }
1343 
1344 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1345 {
1346 	int err;
1347 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1348 	struct mlx4_dev *dev = mdev->dev;
1349 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1350 	struct net_device *ndev;
1351 	struct mlx4_ib_gid_entry *ge;
1352 	struct mlx4_flow_reg_id reg_id = {0, 0};
1353 	enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1354 
1355 	if (mdev->dev->caps.steering_mode ==
1356 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1357 		struct mlx4_ib_steering *ib_steering;
1358 
1359 		mutex_lock(&mqp->mutex);
1360 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1361 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1362 				list_del(&ib_steering->list);
1363 				break;
1364 			}
1365 		}
1366 		mutex_unlock(&mqp->mutex);
1367 		if (&ib_steering->list == &mqp->steering_rules) {
1368 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1369 			return -EINVAL;
1370 		}
1371 		reg_id = ib_steering->reg_id;
1372 		kfree(ib_steering);
1373 	}
1374 
1375 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1376 				    prot, reg_id.id);
1377 	if (err)
1378 		return err;
1379 
1380 	if (mlx4_is_bonded(dev)) {
1381 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1382 					    prot, reg_id.mirror);
1383 		if (err)
1384 			return err;
1385 	}
1386 
1387 	mutex_lock(&mqp->mutex);
1388 	ge = find_gid_entry(mqp, gid->raw);
1389 	if (ge) {
1390 		spin_lock_bh(&mdev->iboe.lock);
1391 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1392 		if (ndev)
1393 			dev_hold(ndev);
1394 		spin_unlock_bh(&mdev->iboe.lock);
1395 		if (ndev)
1396 			dev_put(ndev);
1397 		list_del(&ge->list);
1398 		kfree(ge);
1399 	} else
1400 		pr_warn("could not find mgid entry\n");
1401 
1402 	mutex_unlock(&mqp->mutex);
1403 
1404 	return 0;
1405 }
1406 
1407 static int init_node_data(struct mlx4_ib_dev *dev)
1408 {
1409 	struct ib_smp *in_mad  = NULL;
1410 	struct ib_smp *out_mad = NULL;
1411 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1412 	int err = -ENOMEM;
1413 
1414 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1415 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1416 	if (!in_mad || !out_mad)
1417 		goto out;
1418 
1419 	init_query_mad(in_mad);
1420 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1421 	if (mlx4_is_master(dev->dev))
1422 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1423 
1424 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1425 	if (err)
1426 		goto out;
1427 
1428 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1429 
1430 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1431 
1432 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1433 	if (err)
1434 		goto out;
1435 
1436 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1437 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1438 
1439 out:
1440 	kfree(in_mad);
1441 	kfree(out_mad);
1442 	return err;
1443 }
1444 
1445 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1446 			char *buf)
1447 {
1448 	struct mlx4_ib_dev *dev =
1449 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1450 	return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1451 }
1452 
1453 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1454 			   char *buf)
1455 {
1456 	struct mlx4_ib_dev *dev =
1457 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1458 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1459 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1460 		       (int) dev->dev->caps.fw_ver & 0xffff);
1461 }
1462 
1463 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1464 			char *buf)
1465 {
1466 	struct mlx4_ib_dev *dev =
1467 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1468 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1469 }
1470 
1471 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1472 			  char *buf)
1473 {
1474 	struct mlx4_ib_dev *dev =
1475 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1476 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1477 		       dev->dev->board_id);
1478 }
1479 
1480 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1481 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1482 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1483 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1484 
1485 static struct device_attribute *mlx4_class_attributes[] = {
1486 	&dev_attr_hw_rev,
1487 	&dev_attr_fw_ver,
1488 	&dev_attr_hca_type,
1489 	&dev_attr_board_id
1490 };
1491 
1492 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1493 				     struct net_device *dev)
1494 {
1495 	memcpy(eui, dev->dev_addr, 3);
1496 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1497 	if (vlan_id < 0x1000) {
1498 		eui[3] = vlan_id >> 8;
1499 		eui[4] = vlan_id & 0xff;
1500 	} else {
1501 		eui[3] = 0xff;
1502 		eui[4] = 0xfe;
1503 	}
1504 	eui[0] ^= 2;
1505 }
1506 
1507 static void update_gids_task(struct work_struct *work)
1508 {
1509 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1510 	struct mlx4_cmd_mailbox *mailbox;
1511 	union ib_gid *gids;
1512 	int err;
1513 	struct mlx4_dev	*dev = gw->dev->dev;
1514 	int is_bonded = mlx4_is_bonded(dev);
1515 
1516 	if (!gw->dev->ib_active)
1517 		return;
1518 
1519 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1520 	if (IS_ERR(mailbox)) {
1521 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1522 		return;
1523 	}
1524 
1525 	gids = mailbox->buf;
1526 	memcpy(gids, gw->gids, sizeof gw->gids);
1527 
1528 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1529 		       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1530 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1531 	if (err)
1532 		pr_warn("set port command failed\n");
1533 	else
1534 		if ((gw->port == 1) || !is_bonded)
1535 			mlx4_ib_dispatch_event(gw->dev,
1536 					       is_bonded ? 1 : gw->port,
1537 					       IB_EVENT_GID_CHANGE);
1538 
1539 	mlx4_free_cmd_mailbox(dev, mailbox);
1540 	kfree(gw);
1541 }
1542 
1543 static void reset_gids_task(struct work_struct *work)
1544 {
1545 	struct update_gid_work *gw =
1546 			container_of(work, struct update_gid_work, work);
1547 	struct mlx4_cmd_mailbox *mailbox;
1548 	union ib_gid *gids;
1549 	int err;
1550 	struct mlx4_dev	*dev = gw->dev->dev;
1551 
1552 	if (!gw->dev->ib_active)
1553 		return;
1554 
1555 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1556 	if (IS_ERR(mailbox)) {
1557 		pr_warn("reset gid table failed\n");
1558 		goto free;
1559 	}
1560 
1561 	gids = mailbox->buf;
1562 	memcpy(gids, gw->gids, sizeof(gw->gids));
1563 
1564 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1565 				    IB_LINK_LAYER_ETHERNET) {
1566 		err = mlx4_cmd(dev, mailbox->dma,
1567 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1568 			       MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1569 			       MLX4_CMD_TIME_CLASS_B,
1570 			       MLX4_CMD_WRAPPED);
1571 		if (err)
1572 			pr_warn(KERN_WARNING
1573 				"set port %d command failed\n", gw->port);
1574 	}
1575 
1576 	mlx4_free_cmd_mailbox(dev, mailbox);
1577 free:
1578 	kfree(gw);
1579 }
1580 
1581 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1582 			    union ib_gid *gid, int clear,
1583 			    int default_gid)
1584 {
1585 	struct update_gid_work *work;
1586 	int i;
1587 	int need_update = 0;
1588 	int free = -1;
1589 	int found = -1;
1590 	int max_gids;
1591 
1592 	if (default_gid) {
1593 		free = 0;
1594 	} else {
1595 		max_gids = dev->dev->caps.gid_table_len[port];
1596 		for (i = 1; i < max_gids; ++i) {
1597 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1598 				    sizeof(*gid)))
1599 				found = i;
1600 
1601 			if (clear) {
1602 				if (found >= 0) {
1603 					need_update = 1;
1604 					dev->iboe.gid_table[port - 1][found] =
1605 						zgid;
1606 					break;
1607 				}
1608 			} else {
1609 				if (found >= 0)
1610 					break;
1611 
1612 				if (free < 0 &&
1613 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1614 					    &zgid, sizeof(*gid)))
1615 					free = i;
1616 			}
1617 		}
1618 	}
1619 
1620 	if (found == -1 && !clear && free >= 0) {
1621 		dev->iboe.gid_table[port - 1][free] = *gid;
1622 		need_update = 1;
1623 	}
1624 
1625 	if (!need_update)
1626 		return 0;
1627 
1628 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1629 	if (!work)
1630 		return -ENOMEM;
1631 
1632 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1633 	INIT_WORK(&work->work, update_gids_task);
1634 	work->port = port;
1635 	work->dev = dev;
1636 	queue_work(wq, &work->work);
1637 
1638 	return 0;
1639 }
1640 
1641 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1642 {
1643 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1644 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1645 }
1646 
1647 
1648 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1649 {
1650 	struct update_gid_work *work;
1651 
1652 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1653 	if (!work)
1654 		return -ENOMEM;
1655 
1656 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1657 	memset(work->gids, 0, sizeof(work->gids));
1658 	INIT_WORK(&work->work, reset_gids_task);
1659 	work->dev = dev;
1660 	work->port = port;
1661 	queue_work(wq, &work->work);
1662 	return 0;
1663 }
1664 
1665 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1666 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1667 {
1668 	struct mlx4_ib_iboe *iboe;
1669 	int port = 0;
1670 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1671 				rdma_vlan_dev_real_dev(event_netdev) :
1672 				event_netdev;
1673 	union ib_gid default_gid;
1674 
1675 	mlx4_make_default_gid(real_dev, &default_gid);
1676 
1677 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1678 		return 0;
1679 
1680 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1681 		return 0;
1682 
1683 	if ((real_dev != event_netdev) &&
1684 	    (event == NETDEV_DOWN) &&
1685 	    rdma_link_local_addr((struct in6_addr *)gid))
1686 		return 0;
1687 
1688 	iboe = &ibdev->iboe;
1689 	spin_lock_bh(&iboe->lock);
1690 
1691 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1692 		if ((netif_is_bond_master(real_dev) &&
1693 		     (real_dev == iboe->masters[port - 1])) ||
1694 		     (!netif_is_bond_master(real_dev) &&
1695 		     (real_dev == iboe->netdevs[port - 1])))
1696 			update_gid_table(ibdev, port, gid,
1697 					 event == NETDEV_DOWN, 0);
1698 
1699 	spin_unlock_bh(&iboe->lock);
1700 	return 0;
1701 
1702 }
1703 
1704 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1705 			       struct mlx4_ib_dev *ibdev)
1706 {
1707 	u8 port = 0;
1708 	struct mlx4_ib_iboe *iboe;
1709 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1710 				rdma_vlan_dev_real_dev(dev) : dev;
1711 
1712 	iboe = &ibdev->iboe;
1713 
1714 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1715 		if ((netif_is_bond_master(real_dev) &&
1716 		     (real_dev == iboe->masters[port - 1])) ||
1717 		     (!netif_is_bond_master(real_dev) &&
1718 		     (real_dev == iboe->netdevs[port - 1])))
1719 			break;
1720 
1721 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1722 		return 0;
1723 	else
1724 		return port;
1725 }
1726 
1727 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1728 				void *ptr)
1729 {
1730 	struct mlx4_ib_dev *ibdev;
1731 	struct in_ifaddr *ifa = ptr;
1732 	union ib_gid gid;
1733 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1734 
1735 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1736 
1737 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1738 
1739 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1740 	return NOTIFY_DONE;
1741 }
1742 
1743 #if IS_ENABLED(CONFIG_IPV6)
1744 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1745 				void *ptr)
1746 {
1747 	struct mlx4_ib_dev *ibdev;
1748 	struct inet6_ifaddr *ifa = ptr;
1749 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1750 	struct net_device *event_netdev = ifa->idev->dev;
1751 
1752 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1753 
1754 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1755 	return NOTIFY_DONE;
1756 }
1757 #endif
1758 
1759 #define MLX4_IB_INVALID_MAC	((u64)-1)
1760 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1761 			       struct net_device *dev,
1762 			       int port)
1763 {
1764 	u64 new_smac = 0;
1765 	u64 release_mac = MLX4_IB_INVALID_MAC;
1766 	struct mlx4_ib_qp *qp;
1767 
1768 	read_lock(&dev_base_lock);
1769 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1770 	read_unlock(&dev_base_lock);
1771 
1772 	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1773 
1774 	/* no need for update QP1 and mac registration in non-SRIOV */
1775 	if (!mlx4_is_mfunc(ibdev->dev))
1776 		return;
1777 
1778 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1779 	qp = ibdev->qp1_proxy[port - 1];
1780 	if (qp) {
1781 		int new_smac_index;
1782 		u64 old_smac;
1783 		struct mlx4_update_qp_params update_params;
1784 
1785 		mutex_lock(&qp->mutex);
1786 		old_smac = qp->pri.smac;
1787 		if (new_smac == old_smac)
1788 			goto unlock;
1789 
1790 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1791 
1792 		if (new_smac_index < 0)
1793 			goto unlock;
1794 
1795 		update_params.smac_index = new_smac_index;
1796 		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1797 				   &update_params)) {
1798 			release_mac = new_smac;
1799 			goto unlock;
1800 		}
1801 		/* if old port was zero, no mac was yet registered for this QP */
1802 		if (qp->pri.smac_port)
1803 			release_mac = old_smac;
1804 		qp->pri.smac = new_smac;
1805 		qp->pri.smac_port = port;
1806 		qp->pri.smac_index = new_smac_index;
1807 	}
1808 
1809 unlock:
1810 	if (release_mac != MLX4_IB_INVALID_MAC)
1811 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1812 	if (qp)
1813 		mutex_unlock(&qp->mutex);
1814 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1815 }
1816 
1817 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1818 				 struct mlx4_ib_dev *ibdev, u8 port)
1819 {
1820 	struct in_device *in_dev;
1821 #if IS_ENABLED(CONFIG_IPV6)
1822 	struct inet6_dev *in6_dev;
1823 	union ib_gid  *pgid;
1824 	struct inet6_ifaddr *ifp;
1825 	union ib_gid default_gid;
1826 #endif
1827 	union ib_gid gid;
1828 
1829 
1830 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1831 		return;
1832 
1833 	/* IPv4 gids */
1834 	in_dev = in_dev_get(dev);
1835 	if (in_dev) {
1836 		for_ifa(in_dev) {
1837 			/*ifa->ifa_address;*/
1838 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1839 					       (struct in6_addr *)&gid);
1840 			update_gid_table(ibdev, port, &gid, 0, 0);
1841 		}
1842 		endfor_ifa(in_dev);
1843 		in_dev_put(in_dev);
1844 	}
1845 #if IS_ENABLED(CONFIG_IPV6)
1846 	mlx4_make_default_gid(dev, &default_gid);
1847 	/* IPv6 gids */
1848 	in6_dev = in6_dev_get(dev);
1849 	if (in6_dev) {
1850 		read_lock_bh(&in6_dev->lock);
1851 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1852 			pgid = (union ib_gid *)&ifp->addr;
1853 			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1854 				continue;
1855 			update_gid_table(ibdev, port, pgid, 0, 0);
1856 		}
1857 		read_unlock_bh(&in6_dev->lock);
1858 		in6_dev_put(in6_dev);
1859 	}
1860 #endif
1861 }
1862 
1863 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1864 				 struct  net_device *dev, u8 port)
1865 {
1866 	union ib_gid gid;
1867 	mlx4_make_default_gid(dev, &gid);
1868 	update_gid_table(ibdev, port, &gid, 0, 1);
1869 }
1870 
1871 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1872 {
1873 	struct	net_device *dev;
1874 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1875 	int i;
1876 	int err = 0;
1877 
1878 	for (i = 1; i <= ibdev->num_ports; ++i) {
1879 		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1880 		    IB_LINK_LAYER_ETHERNET) {
1881 			err = reset_gid_table(ibdev, i);
1882 			if (err)
1883 				goto out;
1884 		}
1885 	}
1886 
1887 	read_lock(&dev_base_lock);
1888 	spin_lock_bh(&iboe->lock);
1889 
1890 	for_each_netdev(&init_net, dev) {
1891 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1892 		/* port will be non-zero only for ETH ports */
1893 		if (port) {
1894 			mlx4_ib_set_default_gid(ibdev, dev, port);
1895 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1896 		}
1897 	}
1898 
1899 	spin_unlock_bh(&iboe->lock);
1900 	read_unlock(&dev_base_lock);
1901 out:
1902 	return err;
1903 }
1904 
1905 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1906 				 struct net_device *dev,
1907 				 unsigned long event)
1908 
1909 {
1910 	struct mlx4_ib_iboe *iboe;
1911 	int update_qps_port = -1;
1912 	int port;
1913 
1914 	iboe = &ibdev->iboe;
1915 
1916 	spin_lock_bh(&iboe->lock);
1917 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1918 		enum ib_port_state	port_state = IB_PORT_NOP;
1919 		struct net_device *old_master = iboe->masters[port - 1];
1920 		struct net_device *curr_netdev;
1921 		struct net_device *curr_master;
1922 
1923 		iboe->netdevs[port - 1] =
1924 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1925 		if (iboe->netdevs[port - 1])
1926 			mlx4_ib_set_default_gid(ibdev,
1927 						iboe->netdevs[port - 1], port);
1928 		curr_netdev = iboe->netdevs[port - 1];
1929 
1930 		if (iboe->netdevs[port - 1] &&
1931 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1932 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1933 				iboe->netdevs[port - 1]);
1934 		} else {
1935 			iboe->masters[port - 1] = NULL;
1936 		}
1937 		curr_master = iboe->masters[port - 1];
1938 
1939 		if (dev == iboe->netdevs[port - 1] &&
1940 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1941 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1942 			update_qps_port = port;
1943 
1944 		if (curr_netdev) {
1945 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1946 						IB_PORT_ACTIVE : IB_PORT_DOWN;
1947 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1948 			if (curr_master) {
1949 				/* if using bonding/team and a slave port is down, we
1950 				 * don't want the bond IP based gids in the table since
1951 				 * flows that select port by gid may get the down port.
1952 				*/
1953 				if (port_state == IB_PORT_DOWN &&
1954 				    !mlx4_is_bonded(ibdev->dev)) {
1955 					reset_gid_table(ibdev, port);
1956 					mlx4_ib_set_default_gid(ibdev,
1957 								curr_netdev,
1958 								port);
1959 				} else {
1960 					/* gids from the upper dev (bond/team)
1961 					 * should appear in port's gid table
1962 					*/
1963 					mlx4_ib_get_dev_addr(curr_master,
1964 							     ibdev, port);
1965 				}
1966 			}
1967 			/* if bonding is used it is possible that we add it to
1968 			 * masters only after IP address is assigned to the
1969 			 * net bonding interface.
1970 			*/
1971 			if (curr_master && (old_master != curr_master)) {
1972 				reset_gid_table(ibdev, port);
1973 				mlx4_ib_set_default_gid(ibdev,
1974 							curr_netdev, port);
1975 				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1976 			}
1977 
1978 			if (!curr_master && (old_master != curr_master)) {
1979 				reset_gid_table(ibdev, port);
1980 				mlx4_ib_set_default_gid(ibdev,
1981 							curr_netdev, port);
1982 				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1983 			}
1984 		} else {
1985 			reset_gid_table(ibdev, port);
1986 		}
1987 	}
1988 
1989 	spin_unlock_bh(&iboe->lock);
1990 
1991 	if (update_qps_port > 0)
1992 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1993 }
1994 
1995 static int mlx4_ib_netdev_event(struct notifier_block *this,
1996 				unsigned long event, void *ptr)
1997 {
1998 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1999 	struct mlx4_ib_dev *ibdev;
2000 
2001 	if (!net_eq(dev_net(dev), &init_net))
2002 		return NOTIFY_DONE;
2003 
2004 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2005 	mlx4_ib_scan_netdevs(ibdev, dev, event);
2006 
2007 	return NOTIFY_DONE;
2008 }
2009 
2010 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2011 {
2012 	int port;
2013 	int slave;
2014 	int i;
2015 
2016 	if (mlx4_is_master(ibdev->dev)) {
2017 		for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2018 		     ++slave) {
2019 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2020 				for (i = 0;
2021 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2022 				     ++i) {
2023 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2024 					/* master has the identity virt2phys pkey mapping */
2025 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2026 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2027 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2028 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2029 				}
2030 			}
2031 		}
2032 		/* initialize pkey cache */
2033 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2034 			for (i = 0;
2035 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2036 			     ++i)
2037 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
2038 					(i) ? 0 : 0xFFFF;
2039 		}
2040 	}
2041 }
2042 
2043 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2044 {
2045 	char name[80];
2046 	int eq_per_port = 0;
2047 	int added_eqs = 0;
2048 	int total_eqs = 0;
2049 	int i, j, eq;
2050 
2051 	/* Legacy mode or comp_pool is not large enough */
2052 	if (dev->caps.comp_pool == 0 ||
2053 	    dev->caps.num_ports > dev->caps.comp_pool)
2054 		return;
2055 
2056 	eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
2057 
2058 	/* Init eq table */
2059 	added_eqs = 0;
2060 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2061 		added_eqs += eq_per_port;
2062 
2063 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
2064 
2065 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2066 	if (!ibdev->eq_table)
2067 		return;
2068 
2069 	ibdev->eq_added = added_eqs;
2070 
2071 	eq = 0;
2072 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2073 		for (j = 0; j < eq_per_port; j++) {
2074 			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
2075 				 i, j, dev->persist->pdev->bus->name);
2076 			/* Set IRQ for specific name (per ring) */
2077 			if (mlx4_assign_eq(dev, name, NULL,
2078 					   &ibdev->eq_table[eq])) {
2079 				/* Use legacy (same as mlx4_en driver) */
2080 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2081 				ibdev->eq_table[eq] =
2082 					(eq % dev->caps.num_comp_vectors);
2083 			}
2084 			eq++;
2085 		}
2086 	}
2087 
2088 	/* Fill the reset of the vector with legacy EQ */
2089 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2090 		ibdev->eq_table[eq++] = i;
2091 
2092 	/* Advertise the new number of EQs to clients */
2093 	ibdev->ib_dev.num_comp_vectors = total_eqs;
2094 }
2095 
2096 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2097 {
2098 	int i;
2099 
2100 	/* no additional eqs were added */
2101 	if (!ibdev->eq_table)
2102 		return;
2103 
2104 	/* Reset the advertised EQ number */
2105 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2106 
2107 	/* Free only the added eqs */
2108 	for (i = 0; i < ibdev->eq_added; i++) {
2109 		/* Don't free legacy eqs if used */
2110 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2111 			continue;
2112 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2113 	}
2114 
2115 	kfree(ibdev->eq_table);
2116 }
2117 
2118 static void *mlx4_ib_add(struct mlx4_dev *dev)
2119 {
2120 	struct mlx4_ib_dev *ibdev;
2121 	int num_ports = 0;
2122 	int i, j;
2123 	int err;
2124 	struct mlx4_ib_iboe *iboe;
2125 	int ib_num_ports = 0;
2126 	int num_req_counters;
2127 
2128 	pr_info_once("%s", mlx4_ib_version);
2129 
2130 	num_ports = 0;
2131 	mlx4_foreach_ib_transport_port(i, dev)
2132 		num_ports++;
2133 
2134 	/* No point in registering a device with no ports... */
2135 	if (num_ports == 0)
2136 		return NULL;
2137 
2138 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2139 	if (!ibdev) {
2140 		dev_err(&dev->persist->pdev->dev,
2141 			"Device struct alloc failed\n");
2142 		return NULL;
2143 	}
2144 
2145 	iboe = &ibdev->iboe;
2146 
2147 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2148 		goto err_dealloc;
2149 
2150 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2151 		goto err_pd;
2152 
2153 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2154 				 PAGE_SIZE);
2155 	if (!ibdev->uar_map)
2156 		goto err_uar;
2157 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2158 
2159 	ibdev->dev = dev;
2160 	ibdev->bond_next_port	= 0;
2161 
2162 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2163 	ibdev->ib_dev.owner		= THIS_MODULE;
2164 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2165 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2166 	ibdev->num_ports		= num_ports;
2167 	ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2168 						1 : ibdev->num_ports;
2169 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2170 	ibdev->ib_dev.dma_device	= &dev->persist->pdev->dev;
2171 
2172 	if (dev->caps.userspace_caps)
2173 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2174 	else
2175 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2176 
2177 	ibdev->ib_dev.uverbs_cmd_mask	=
2178 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2179 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2180 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2181 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2182 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2183 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2184 		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
2185 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2186 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2187 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2188 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2189 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2190 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2191 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2192 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2193 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2194 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2195 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2196 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2197 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2198 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2199 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2200 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2201 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2202 
2203 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2204 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2205 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2206 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2207 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2208 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2209 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2210 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2211 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2212 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2213 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2214 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2215 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2216 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2217 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2218 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2219 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2220 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2221 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2222 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2223 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2224 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2225 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2226 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2227 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2228 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2229 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2230 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2231 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2232 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2233 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2234 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2235 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2236 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2237 	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
2238 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2239 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2240 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2241 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2242 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2243 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2244 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2245 
2246 	if (!mlx4_is_slave(ibdev->dev)) {
2247 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2248 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2249 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2250 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2251 	}
2252 
2253 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2254 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2255 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2256 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2257 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2258 
2259 		ibdev->ib_dev.uverbs_cmd_mask |=
2260 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2261 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2262 	}
2263 
2264 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2265 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2266 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2267 		ibdev->ib_dev.uverbs_cmd_mask |=
2268 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2269 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2270 	}
2271 
2272 	if (check_flow_steering_support(dev)) {
2273 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2274 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2275 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2276 
2277 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2278 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2279 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2280 	}
2281 
2282 	mlx4_ib_alloc_eqs(dev, ibdev);
2283 
2284 	spin_lock_init(&iboe->lock);
2285 
2286 	if (init_node_data(ibdev))
2287 		goto err_map;
2288 
2289 	num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2290 	for (i = 0; i < num_req_counters; ++i) {
2291 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2292 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2293 						IB_LINK_LAYER_ETHERNET) {
2294 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2295 			if (err)
2296 				ibdev->counters[i] = -1;
2297 		} else {
2298 			ibdev->counters[i] = -1;
2299 		}
2300 	}
2301 	if (mlx4_is_bonded(dev))
2302 		for (i = 1; i < ibdev->num_ports ; ++i)
2303 			ibdev->counters[i] = ibdev->counters[0];
2304 
2305 
2306 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2307 		ib_num_ports++;
2308 
2309 	spin_lock_init(&ibdev->sm_lock);
2310 	mutex_init(&ibdev->cap_mask_mutex);
2311 	INIT_LIST_HEAD(&ibdev->qp_list);
2312 	spin_lock_init(&ibdev->reset_flow_resource_lock);
2313 
2314 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2315 	    ib_num_ports) {
2316 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2317 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2318 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2319 					    &ibdev->steer_qpn_base, 0);
2320 		if (err)
2321 			goto err_counter;
2322 
2323 		ibdev->ib_uc_qpns_bitmap =
2324 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2325 				sizeof(long),
2326 				GFP_KERNEL);
2327 		if (!ibdev->ib_uc_qpns_bitmap) {
2328 			dev_err(&dev->persist->pdev->dev,
2329 				"bit map alloc failed\n");
2330 			goto err_steer_qp_release;
2331 		}
2332 
2333 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2334 
2335 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2336 				dev, ibdev->steer_qpn_base,
2337 				ibdev->steer_qpn_base +
2338 				ibdev->steer_qpn_count - 1);
2339 		if (err)
2340 			goto err_steer_free_bitmap;
2341 	}
2342 
2343 	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2344 		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2345 
2346 	if (ib_register_device(&ibdev->ib_dev, NULL))
2347 		goto err_steer_free_bitmap;
2348 
2349 	if (mlx4_ib_mad_init(ibdev))
2350 		goto err_reg;
2351 
2352 	if (mlx4_ib_init_sriov(ibdev))
2353 		goto err_mad;
2354 
2355 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2356 		if (!iboe->nb.notifier_call) {
2357 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2358 			err = register_netdevice_notifier(&iboe->nb);
2359 			if (err) {
2360 				iboe->nb.notifier_call = NULL;
2361 				goto err_notif;
2362 			}
2363 		}
2364 		if (!iboe->nb_inet.notifier_call) {
2365 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2366 			err = register_inetaddr_notifier(&iboe->nb_inet);
2367 			if (err) {
2368 				iboe->nb_inet.notifier_call = NULL;
2369 				goto err_notif;
2370 			}
2371 		}
2372 #if IS_ENABLED(CONFIG_IPV6)
2373 		if (!iboe->nb_inet6.notifier_call) {
2374 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2375 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2376 			if (err) {
2377 				iboe->nb_inet6.notifier_call = NULL;
2378 				goto err_notif;
2379 			}
2380 		}
2381 #endif
2382 		if (mlx4_ib_init_gid_table(ibdev))
2383 			goto err_notif;
2384 	}
2385 
2386 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2387 		if (device_create_file(&ibdev->ib_dev.dev,
2388 				       mlx4_class_attributes[j]))
2389 			goto err_notif;
2390 	}
2391 
2392 	ibdev->ib_active = true;
2393 
2394 	if (mlx4_is_mfunc(ibdev->dev))
2395 		init_pkeys(ibdev);
2396 
2397 	/* create paravirt contexts for any VFs which are active */
2398 	if (mlx4_is_master(ibdev->dev)) {
2399 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2400 			if (j == mlx4_master_func_num(ibdev->dev))
2401 				continue;
2402 			if (mlx4_is_slave_active(ibdev->dev, j))
2403 				do_slave_init(ibdev, j, 1);
2404 		}
2405 	}
2406 	return ibdev;
2407 
2408 err_notif:
2409 	if (ibdev->iboe.nb.notifier_call) {
2410 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2411 			pr_warn("failure unregistering notifier\n");
2412 		ibdev->iboe.nb.notifier_call = NULL;
2413 	}
2414 	if (ibdev->iboe.nb_inet.notifier_call) {
2415 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2416 			pr_warn("failure unregistering notifier\n");
2417 		ibdev->iboe.nb_inet.notifier_call = NULL;
2418 	}
2419 #if IS_ENABLED(CONFIG_IPV6)
2420 	if (ibdev->iboe.nb_inet6.notifier_call) {
2421 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2422 			pr_warn("failure unregistering notifier\n");
2423 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2424 	}
2425 #endif
2426 	flush_workqueue(wq);
2427 
2428 	mlx4_ib_close_sriov(ibdev);
2429 
2430 err_mad:
2431 	mlx4_ib_mad_cleanup(ibdev);
2432 
2433 err_reg:
2434 	ib_unregister_device(&ibdev->ib_dev);
2435 
2436 err_steer_free_bitmap:
2437 	kfree(ibdev->ib_uc_qpns_bitmap);
2438 
2439 err_steer_qp_release:
2440 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2441 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2442 				      ibdev->steer_qpn_count);
2443 err_counter:
2444 	for (; i; --i)
2445 		if (ibdev->counters[i - 1] != -1)
2446 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2447 
2448 err_map:
2449 	iounmap(ibdev->uar_map);
2450 
2451 err_uar:
2452 	mlx4_uar_free(dev, &ibdev->priv_uar);
2453 
2454 err_pd:
2455 	mlx4_pd_free(dev, ibdev->priv_pdn);
2456 
2457 err_dealloc:
2458 	ib_dealloc_device(&ibdev->ib_dev);
2459 
2460 	return NULL;
2461 }
2462 
2463 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2464 {
2465 	int offset;
2466 
2467 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2468 
2469 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2470 					 dev->steer_qpn_count,
2471 					 get_count_order(count));
2472 	if (offset < 0)
2473 		return offset;
2474 
2475 	*qpn = dev->steer_qpn_base + offset;
2476 	return 0;
2477 }
2478 
2479 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2480 {
2481 	if (!qpn ||
2482 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2483 		return;
2484 
2485 	BUG_ON(qpn < dev->steer_qpn_base);
2486 
2487 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2488 			      qpn - dev->steer_qpn_base,
2489 			      get_count_order(count));
2490 }
2491 
2492 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2493 			 int is_attach)
2494 {
2495 	int err;
2496 	size_t flow_size;
2497 	struct ib_flow_attr *flow = NULL;
2498 	struct ib_flow_spec_ib *ib_spec;
2499 
2500 	if (is_attach) {
2501 		flow_size = sizeof(struct ib_flow_attr) +
2502 			    sizeof(struct ib_flow_spec_ib);
2503 		flow = kzalloc(flow_size, GFP_KERNEL);
2504 		if (!flow)
2505 			return -ENOMEM;
2506 		flow->port = mqp->port;
2507 		flow->num_of_specs = 1;
2508 		flow->size = flow_size;
2509 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2510 		ib_spec->type = IB_FLOW_SPEC_IB;
2511 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2512 		/* Add an empty rule for IB L2 */
2513 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2514 
2515 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2516 					    IB_FLOW_DOMAIN_NIC,
2517 					    MLX4_FS_REGULAR,
2518 					    &mqp->reg_id);
2519 	} else {
2520 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2521 	}
2522 	kfree(flow);
2523 	return err;
2524 }
2525 
2526 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2527 {
2528 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2529 	int p;
2530 
2531 	ibdev->ib_active = false;
2532 	flush_workqueue(wq);
2533 
2534 	mlx4_ib_close_sriov(ibdev);
2535 	mlx4_ib_mad_cleanup(ibdev);
2536 	ib_unregister_device(&ibdev->ib_dev);
2537 	if (ibdev->iboe.nb.notifier_call) {
2538 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2539 			pr_warn("failure unregistering notifier\n");
2540 		ibdev->iboe.nb.notifier_call = NULL;
2541 	}
2542 
2543 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2544 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2545 				      ibdev->steer_qpn_count);
2546 		kfree(ibdev->ib_uc_qpns_bitmap);
2547 	}
2548 
2549 	if (ibdev->iboe.nb_inet.notifier_call) {
2550 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2551 			pr_warn("failure unregistering notifier\n");
2552 		ibdev->iboe.nb_inet.notifier_call = NULL;
2553 	}
2554 #if IS_ENABLED(CONFIG_IPV6)
2555 	if (ibdev->iboe.nb_inet6.notifier_call) {
2556 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2557 			pr_warn("failure unregistering notifier\n");
2558 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2559 	}
2560 #endif
2561 
2562 	iounmap(ibdev->uar_map);
2563 	for (p = 0; p < ibdev->num_ports; ++p)
2564 		if (ibdev->counters[p] != -1)
2565 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2566 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2567 		mlx4_CLOSE_PORT(dev, p);
2568 
2569 	mlx4_ib_free_eqs(dev, ibdev);
2570 
2571 	mlx4_uar_free(dev, &ibdev->priv_uar);
2572 	mlx4_pd_free(dev, ibdev->priv_pdn);
2573 	ib_dealloc_device(&ibdev->ib_dev);
2574 }
2575 
2576 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2577 {
2578 	struct mlx4_ib_demux_work **dm = NULL;
2579 	struct mlx4_dev *dev = ibdev->dev;
2580 	int i;
2581 	unsigned long flags;
2582 	struct mlx4_active_ports actv_ports;
2583 	unsigned int ports;
2584 	unsigned int first_port;
2585 
2586 	if (!mlx4_is_master(dev))
2587 		return;
2588 
2589 	actv_ports = mlx4_get_active_ports(dev, slave);
2590 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2591 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2592 
2593 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2594 	if (!dm) {
2595 		pr_err("failed to allocate memory for tunneling qp update\n");
2596 		goto out;
2597 	}
2598 
2599 	for (i = 0; i < ports; i++) {
2600 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2601 		if (!dm[i]) {
2602 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2603 			for (i = 0; i < dev->caps.num_ports; i++) {
2604 				if (dm[i])
2605 					kfree(dm[i]);
2606 			}
2607 			goto out;
2608 		}
2609 	}
2610 	/* initialize or tear down tunnel QPs for the slave */
2611 	for (i = 0; i < ports; i++) {
2612 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2613 		dm[i]->port = first_port + i + 1;
2614 		dm[i]->slave = slave;
2615 		dm[i]->do_init = do_init;
2616 		dm[i]->dev = ibdev;
2617 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2618 		if (!ibdev->sriov.is_going_down)
2619 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2620 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2621 	}
2622 out:
2623 	kfree(dm);
2624 	return;
2625 }
2626 
2627 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2628 {
2629 	struct mlx4_ib_qp *mqp;
2630 	unsigned long flags_qp;
2631 	unsigned long flags_cq;
2632 	struct mlx4_ib_cq *send_mcq, *recv_mcq;
2633 	struct list_head    cq_notify_list;
2634 	struct mlx4_cq *mcq;
2635 	unsigned long flags;
2636 
2637 	pr_warn("mlx4_ib_handle_catas_error was started\n");
2638 	INIT_LIST_HEAD(&cq_notify_list);
2639 
2640 	/* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2641 	spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2642 
2643 	list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2644 		spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2645 		if (mqp->sq.tail != mqp->sq.head) {
2646 			send_mcq = to_mcq(mqp->ibqp.send_cq);
2647 			spin_lock_irqsave(&send_mcq->lock, flags_cq);
2648 			if (send_mcq->mcq.comp &&
2649 			    mqp->ibqp.send_cq->comp_handler) {
2650 				if (!send_mcq->mcq.reset_notify_added) {
2651 					send_mcq->mcq.reset_notify_added = 1;
2652 					list_add_tail(&send_mcq->mcq.reset_notify,
2653 						      &cq_notify_list);
2654 				}
2655 			}
2656 			spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2657 		}
2658 		spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2659 		/* Now, handle the QP's receive queue */
2660 		spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2661 		/* no handling is needed for SRQ */
2662 		if (!mqp->ibqp.srq) {
2663 			if (mqp->rq.tail != mqp->rq.head) {
2664 				recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2665 				spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2666 				if (recv_mcq->mcq.comp &&
2667 				    mqp->ibqp.recv_cq->comp_handler) {
2668 					if (!recv_mcq->mcq.reset_notify_added) {
2669 						recv_mcq->mcq.reset_notify_added = 1;
2670 						list_add_tail(&recv_mcq->mcq.reset_notify,
2671 							      &cq_notify_list);
2672 					}
2673 				}
2674 				spin_unlock_irqrestore(&recv_mcq->lock,
2675 						       flags_cq);
2676 			}
2677 		}
2678 		spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2679 	}
2680 
2681 	list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2682 		mcq->comp(mcq);
2683 	}
2684 	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2685 	pr_warn("mlx4_ib_handle_catas_error ended\n");
2686 }
2687 
2688 static void handle_bonded_port_state_event(struct work_struct *work)
2689 {
2690 	struct ib_event_work *ew =
2691 		container_of(work, struct ib_event_work, work);
2692 	struct mlx4_ib_dev *ibdev = ew->ib_dev;
2693 	enum ib_port_state bonded_port_state = IB_PORT_NOP;
2694 	int i;
2695 	struct ib_event ibev;
2696 
2697 	kfree(ew);
2698 	spin_lock_bh(&ibdev->iboe.lock);
2699 	for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2700 		struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2701 		enum ib_port_state curr_port_state;
2702 
2703 		if (!curr_netdev)
2704 			continue;
2705 
2706 		curr_port_state =
2707 			(netif_running(curr_netdev) &&
2708 			 netif_carrier_ok(curr_netdev)) ?
2709 			IB_PORT_ACTIVE : IB_PORT_DOWN;
2710 
2711 		bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2712 			curr_port_state : IB_PORT_ACTIVE;
2713 	}
2714 	spin_unlock_bh(&ibdev->iboe.lock);
2715 
2716 	ibev.device = &ibdev->ib_dev;
2717 	ibev.element.port_num = 1;
2718 	ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2719 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2720 
2721 	ib_dispatch_event(&ibev);
2722 }
2723 
2724 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2725 			  enum mlx4_dev_event event, unsigned long param)
2726 {
2727 	struct ib_event ibev;
2728 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2729 	struct mlx4_eqe *eqe = NULL;
2730 	struct ib_event_work *ew;
2731 	int p = 0;
2732 
2733 	if (mlx4_is_bonded(dev) &&
2734 	    ((event == MLX4_DEV_EVENT_PORT_UP) ||
2735 	    (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2736 		ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2737 		if (!ew)
2738 			return;
2739 		INIT_WORK(&ew->work, handle_bonded_port_state_event);
2740 		ew->ib_dev = ibdev;
2741 		queue_work(wq, &ew->work);
2742 		return;
2743 	}
2744 
2745 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2746 		eqe = (struct mlx4_eqe *)param;
2747 	else
2748 		p = (int) param;
2749 
2750 	switch (event) {
2751 	case MLX4_DEV_EVENT_PORT_UP:
2752 		if (p > ibdev->num_ports)
2753 			return;
2754 		if (mlx4_is_master(dev) &&
2755 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2756 			IB_LINK_LAYER_INFINIBAND) {
2757 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2758 		}
2759 		ibev.event = IB_EVENT_PORT_ACTIVE;
2760 		break;
2761 
2762 	case MLX4_DEV_EVENT_PORT_DOWN:
2763 		if (p > ibdev->num_ports)
2764 			return;
2765 		ibev.event = IB_EVENT_PORT_ERR;
2766 		break;
2767 
2768 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2769 		ibdev->ib_active = false;
2770 		ibev.event = IB_EVENT_DEVICE_FATAL;
2771 		mlx4_ib_handle_catas_error(ibdev);
2772 		break;
2773 
2774 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2775 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2776 		if (!ew) {
2777 			pr_err("failed to allocate memory for events work\n");
2778 			break;
2779 		}
2780 
2781 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2782 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2783 		ew->ib_dev = ibdev;
2784 		/* need to queue only for port owner, which uses GEN_EQE */
2785 		if (mlx4_is_master(dev))
2786 			queue_work(wq, &ew->work);
2787 		else
2788 			handle_port_mgmt_change_event(&ew->work);
2789 		return;
2790 
2791 	case MLX4_DEV_EVENT_SLAVE_INIT:
2792 		/* here, p is the slave id */
2793 		do_slave_init(ibdev, p, 1);
2794 		if (mlx4_is_master(dev)) {
2795 			int i;
2796 
2797 			for (i = 1; i <= ibdev->num_ports; i++) {
2798 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2799 					== IB_LINK_LAYER_INFINIBAND)
2800 					mlx4_ib_slave_alias_guid_event(ibdev,
2801 								       p, i,
2802 								       1);
2803 			}
2804 		}
2805 		return;
2806 
2807 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2808 		if (mlx4_is_master(dev)) {
2809 			int i;
2810 
2811 			for (i = 1; i <= ibdev->num_ports; i++) {
2812 				if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2813 					== IB_LINK_LAYER_INFINIBAND)
2814 					mlx4_ib_slave_alias_guid_event(ibdev,
2815 								       p, i,
2816 								       0);
2817 			}
2818 		}
2819 		/* here, p is the slave id */
2820 		do_slave_init(ibdev, p, 0);
2821 		return;
2822 
2823 	default:
2824 		return;
2825 	}
2826 
2827 	ibev.device	      = ibdev_ptr;
2828 	ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2829 
2830 	ib_dispatch_event(&ibev);
2831 }
2832 
2833 static struct mlx4_interface mlx4_ib_interface = {
2834 	.add		= mlx4_ib_add,
2835 	.remove		= mlx4_ib_remove,
2836 	.event		= mlx4_ib_event,
2837 	.protocol	= MLX4_PROT_IB_IPV6,
2838 	.flags		= MLX4_INTFF_BONDING
2839 };
2840 
2841 static int __init mlx4_ib_init(void)
2842 {
2843 	int err;
2844 
2845 	wq = create_singlethread_workqueue("mlx4_ib");
2846 	if (!wq)
2847 		return -ENOMEM;
2848 
2849 	err = mlx4_ib_mcg_init();
2850 	if (err)
2851 		goto clean_wq;
2852 
2853 	err = mlx4_register_interface(&mlx4_ib_interface);
2854 	if (err)
2855 		goto clean_mcg;
2856 
2857 	return 0;
2858 
2859 clean_mcg:
2860 	mlx4_ib_mcg_destroy();
2861 
2862 clean_wq:
2863 	destroy_workqueue(wq);
2864 	return err;
2865 }
2866 
2867 static void __exit mlx4_ib_cleanup(void)
2868 {
2869 	mlx4_unregister_interface(&mlx4_ib_interface);
2870 	mlx4_ib_mcg_destroy();
2871 	destroy_workqueue(wq);
2872 }
2873 
2874 module_init(mlx4_ib_init);
2875 module_exit(mlx4_ib_cleanup);
2876