xref: /freebsd/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c (revision 448897d366c218f9fd6208427eef1e6dd51f78d0)
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 #define	LINUXKPI_PARAM_PREFIX mlx4_
35 
36 #include <linux/module.h>
37 #include <linux/slab.h>
38 #include <linux/errno.h>
39 #include <linux/netdevice.h>
40 #include <linux/inetdevice.h>
41 #include <linux/if_vlan.h>
42 #include <linux/fs.h>
43 #include <net/ipv6.h>
44 
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_user_verbs_exp.h>
48 #include <rdma/ib_addr.h>
49 
50 #include <dev/mlx4/driver.h>
51 #include <dev/mlx4/cmd.h>
52 #include <linux/sched.h>
53 #include <linux/page.h>
54 #include <linux/printk.h>
55 #include "mlx4_ib.h"
56 #include "mlx4_exp.h"
57 #include "user.h"
58 #include "wc.h"
59 
60 #define DRV_NAME	MLX4_IB_DRV_NAME
61 #define DRV_VERSION	"1.0"
62 #define DRV_RELDATE	__DATE__
63 
64 #define MLX4_IB_DRIVER_PROC_DIR_NAME "driver/mlx4_ib"
65 #define MLX4_IB_MRS_PROC_DIR_NAME "mrs"
66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
68 
69 MODULE_AUTHOR("Roland Dreier");
70 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
71 MODULE_LICENSE("Dual BSD/GPL");
72 #ifdef __linux__
73 MODULE_VERSION(DRV_VERSION);
74 #endif
75 
76 int mlx4_ib_sm_guid_assign = 1;
77 
78 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
79 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
80 
81 enum {
82 	MAX_NUM_STR_BITMAP = 1 << 15,
83 	DEFAULT_TBL_VAL = -1
84 };
85 
86 static struct mlx4_dbdf2val_lst dev_assign_str = {
87 	.name		= "dev_assign_str param",
88 	.num_vals	= 1,
89 	.def_val	= {DEFAULT_TBL_VAL},
90 	.range		= {0, MAX_NUM_STR_BITMAP - 1}
91 };
92 module_param_string(dev_assign_str, dev_assign_str.str,
93 		    sizeof(dev_assign_str.str), 0444);
94 MODULE_PARM_DESC(dev_assign_str,
95 		 "Map device function numbers to IB device numbers (e.g. '0000:04:00.0-0,002b:1c:0b.a-1,...').\n"
96 		 "\t\tHexadecimal digits for the device function (e.g. 002b:1c:0b.a) and decimal for IB device numbers (e.g. 1).\n"
97 		 "\t\tMax supported devices - 32");
98 
99 
100 static unsigned long *dev_num_str_bitmap;
101 static spinlock_t dev_num_str_lock;
102 
103 static const char mlx4_ib_version[] =
104 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
105 	DRV_VERSION " (" DRV_RELDATE ")\n";
106 
107 struct update_gid_work {
108 	struct work_struct	work;
109 	union ib_gid		gids[128];
110 	struct mlx4_ib_dev     *dev;
111 	int			port;
112 };
113 
114 struct dev_rec {
115 	int	bus;
116 	int	dev;
117 	int	func;
118 	int	nr;
119 };
120 
121 static int dr_active;
122 
123 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
124 
125 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev, struct net_device*,
126 				 unsigned long);
127 
128 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
129                                         struct mlx4_ib_dev *ibdev);
130 
131 static struct workqueue_struct *wq;
132 
133 static void init_query_mad(struct ib_smp *mad)
134 {
135 	mad->base_version  = 1;
136 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
137 	mad->class_version = 1;
138 	mad->method	   = IB_MGMT_METHOD_GET;
139 }
140 
141 static union ib_gid zgid;
142 
143 static int check_flow_steering_support(struct mlx4_dev *dev)
144 {
145 	int eth_num_ports = 0;
146 	int ib_num_ports = 0;
147 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
148 
149 	if (dmfs) {
150 		int i;
151 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
152 			eth_num_ports++;
153 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
154 			ib_num_ports++;
155 		dmfs &= (!ib_num_ports ||
156 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
157 			(!eth_num_ports ||
158 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
159 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
160 			dmfs = 0;
161 		}
162 	}
163 	return dmfs;
164 }
165 
166 int mlx4_ib_query_device(struct ib_device *ibdev,
167 				struct ib_device_attr *props)
168 {
169 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
170 	struct ib_smp *in_mad  = NULL;
171 	struct ib_smp *out_mad = NULL;
172 	int err = -ENOMEM;
173 
174 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
175 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
176 	if (!in_mad || !out_mad)
177 		goto out;
178 
179 	init_query_mad(in_mad);
180 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
181 
182 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
183 			   1, NULL, NULL, in_mad, out_mad);
184 	if (err)
185 		goto out;
186 
187 	memset(props, 0, sizeof *props);
188 
189 	props->fw_ver = dev->dev->caps.fw_ver;
190 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
191 		IB_DEVICE_PORT_ACTIVE_EVENT		|
192 		IB_DEVICE_SYS_IMAGE_GUID		|
193 		IB_DEVICE_RC_RNR_NAK_GEN		|
194 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	|
195 		IB_DEVICE_SHARED_MR;
196 
197 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
198 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
199 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
200 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
201 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
202 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
203 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
204 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
205 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
206 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
207 	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
208 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
209 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
210 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
211 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
212 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
213 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
214 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
215 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
216 		props->device_cap_flags |= IB_DEVICE_XRC;
217 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_CROSS_CHANNEL)
218 		props->device_cap_flags |= IB_DEVICE_CROSS_CHANNEL;
219 
220 	if (check_flow_steering_support(dev->dev))
221 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
222 
223 
224 	props->device_cap_flags |= IB_DEVICE_QPG;
225 	if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
226 		props->device_cap_flags |= IB_DEVICE_UD_RSS;
227 		props->max_rss_tbl_sz = dev->dev->caps.max_rss_tbl_sz;
228 	}
229 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
230 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
231 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
232 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
233 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
234 		else
235 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
236 	}
237 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
238 		0xffffff;
239 	props->vendor_part_id	   = dev->dev->pdev->device;
240 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
241 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
242 
243 	props->max_mr_size	   = ~0ull;
244 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
245 	props->max_qp		   = dev->dev->quotas.qp;
246 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
247 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
248 					 dev->dev->caps.max_rq_sg);
249 	props->max_cq		   = dev->dev->quotas.cq;
250 	props->max_cqe		   = dev->dev->caps.max_cqes;
251 	props->max_mr		   = dev->dev->quotas.mpt;
252 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
253 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
254 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
255 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
256 	props->max_srq		   = dev->dev->quotas.srq;
257 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
258 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
259 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
260 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
261 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
262 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
263 	props->masked_atomic_cap   = props->atomic_cap;
264 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
265 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
266 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
267 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
268 					   props->max_mcast_grp;
269 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
270 	props->hca_core_clock = dev->dev->caps.hca_core_clock;
271 	if (dev->dev->caps.hca_core_clock > 0)
272 		props->comp_mask |= IB_DEVICE_ATTR_WITH_HCA_CORE_CLOCK;
273 	if (dev->dev->caps.cq_timestamp) {
274 		props->timestamp_mask = 0xFFFFFFFFFFFF;
275 		props->comp_mask |= IB_DEVICE_ATTR_WITH_TIMESTAMP_MASK;
276 	}
277 
278 out:
279 	kfree(in_mad);
280 	kfree(out_mad);
281 
282 	return err;
283 }
284 
285 static enum rdma_link_layer
286 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
287 {
288 	struct mlx4_dev *dev = to_mdev(device)->dev;
289 
290 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
291 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
292 }
293 
294 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
295 			      struct ib_port_attr *props, int netw_view)
296 {
297 	struct ib_smp *in_mad  = NULL;
298 	struct ib_smp *out_mad = NULL;
299 	int ext_active_speed;
300 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
301 	int err = -ENOMEM;
302 
303 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
304 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
305 	if (!in_mad || !out_mad)
306 		goto out;
307 
308 	init_query_mad(in_mad);
309 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
310 	in_mad->attr_mod = cpu_to_be32(port);
311 
312 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
313 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
314 
315 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
316 				in_mad, out_mad);
317 	if (err)
318 		goto out;
319 
320 
321 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
322 	props->lmc		= out_mad->data[34] & 0x7;
323 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
324 	props->sm_sl		= out_mad->data[36] & 0xf;
325 	props->state		= out_mad->data[32] & 0xf;
326 	props->phys_state	= out_mad->data[33] >> 4;
327 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
328 	if (netw_view)
329 		props->gid_tbl_len = out_mad->data[50];
330 	else
331 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
332 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
333 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
334 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
335 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
336 	props->active_width	= out_mad->data[31] & 0xf;
337 	props->active_speed	= out_mad->data[35] >> 4;
338 	props->max_mtu		= out_mad->data[41] & 0xf;
339 	props->active_mtu	= out_mad->data[36] >> 4;
340 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
341 	props->max_vl_num	= out_mad->data[37] >> 4;
342 	props->init_type_reply	= out_mad->data[41] >> 4;
343 
344 	/* Check if extended speeds (EDR/FDR/...) are supported */
345 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
346 		ext_active_speed = out_mad->data[62] >> 4;
347 
348 		switch (ext_active_speed) {
349 		case 1:
350 			props->active_speed = IB_SPEED_FDR;
351 			break;
352 		case 2:
353 			props->active_speed = IB_SPEED_EDR;
354 			break;
355 		}
356 	}
357 
358 	/* If reported active speed is QDR, check if is FDR-10 */
359 	if (props->active_speed == IB_SPEED_QDR) {
360 		init_query_mad(in_mad);
361 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
362 		in_mad->attr_mod = cpu_to_be32(port);
363 
364 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
365 				   NULL, NULL, in_mad, out_mad);
366 		if (err)
367 			goto out;
368 
369 		/* Checking LinkSpeedActive for FDR-10 */
370 		if (out_mad->data[15] & 0x1)
371 			props->active_speed = IB_SPEED_FDR10;
372 	}
373 
374 	/* Avoid wrong speed value returned by FW if the IB link is down. */
375 	if (props->state == IB_PORT_DOWN)
376 		 props->active_speed = IB_SPEED_SDR;
377 
378 out:
379 	kfree(in_mad);
380 	kfree(out_mad);
381 	return err;
382 }
383 
384 static u8 state_to_phys_state(enum ib_port_state state)
385 {
386 	return state == IB_PORT_ACTIVE ? 5 : 3;
387 }
388 
389 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
390 			       struct ib_port_attr *props, int netw_view)
391 {
392 
393 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
394 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
395 	struct net_device *ndev;
396 	enum ib_mtu tmp;
397 	struct mlx4_cmd_mailbox *mailbox;
398 	unsigned long flags;
399 	int err = 0;
400 
401 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
402 	if (IS_ERR(mailbox))
403 		return PTR_ERR(mailbox);
404 
405 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
406 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
407 			   MLX4_CMD_WRAPPED);
408 	if (err)
409 		goto out;
410 
411 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
412 						IB_WIDTH_4X : IB_WIDTH_1X;
413 	props->active_speed	= IB_SPEED_QDR;
414 	props->port_cap_flags	= IB_PORT_CM_SUP;
415 	if (netw_view)
416 		props->gid_tbl_len = MLX4_ROCE_MAX_GIDS;
417 	else
418 		props->gid_tbl_len   = mdev->dev->caps.gid_table_len[port];
419 
420 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
421 	props->pkey_tbl_len	= 1;
422 	props->max_mtu		= IB_MTU_4096;
423 	props->max_vl_num	= 2;
424 	props->state		= IB_PORT_DOWN;
425 	props->phys_state	= state_to_phys_state(props->state);
426 	props->active_mtu	= IB_MTU_256;
427 	spin_lock_irqsave(&iboe->lock, flags);
428 	ndev = iboe->netdevs[port - 1];
429 	if (!ndev)
430 		goto out_unlock;
431 
432 	tmp = iboe_get_mtu(ndev->if_mtu);
433 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
434 
435 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
436 					IB_PORT_ACTIVE : IB_PORT_DOWN;
437 	props->phys_state	= state_to_phys_state(props->state);
438 out_unlock:
439 	spin_unlock_irqrestore(&iboe->lock, flags);
440 out:
441 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
442 	return err;
443 }
444 
445 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
446 			 struct ib_port_attr *props, int netw_view)
447 {
448 	int err;
449 
450 	memset(props, 0, sizeof *props);
451 
452 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
453 		ib_link_query_port(ibdev, port, props, netw_view) :
454 				eth_link_query_port(ibdev, port, props, netw_view);
455 
456 	return err;
457 }
458 
459 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
460 			      struct ib_port_attr *props)
461 {
462 	/* returns host view */
463 	return __mlx4_ib_query_port(ibdev, port, props, 0);
464 }
465 
466 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
467 			union ib_gid *gid, int netw_view)
468 {
469 	struct ib_smp *in_mad  = NULL;
470 	struct ib_smp *out_mad = NULL;
471 	int err = -ENOMEM;
472 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
473 	int clear = 0;
474 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
475 
476 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
477 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
478 	if (!in_mad || !out_mad)
479 		goto out;
480 
481 	init_query_mad(in_mad);
482 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
483 	in_mad->attr_mod = cpu_to_be32(port);
484 
485 	if (mlx4_is_mfunc(dev->dev) && netw_view)
486 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
487 
488 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
489 	if (err)
490 		goto out;
491 
492 	memcpy(gid->raw, out_mad->data + 8, 8);
493 
494 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
495 		if (index) {
496 			/* For any index > 0, return the null guid */
497 			err = 0;
498 			clear = 1;
499 			goto out;
500 		}
501 	}
502 
503 	init_query_mad(in_mad);
504 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
505 	in_mad->attr_mod = cpu_to_be32(index / 8);
506 
507 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
508 			   NULL, NULL, in_mad, out_mad);
509 	if (err)
510 		goto out;
511 
512 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
513 
514 out:
515 	if (clear)
516 		memset(gid->raw + 8, 0, 8);
517 	kfree(in_mad);
518 	kfree(out_mad);
519 	return err;
520 }
521 
522 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
523 			  union ib_gid *gid)
524 {
525 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
526 
527 	*gid = dev->iboe.gid_table[port - 1][index];
528 
529 	return 0;
530 }
531 
532 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
533 			     union ib_gid *gid)
534 {
535 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
536 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
537 	else
538 		return iboe_query_gid(ibdev, port, index, gid);
539 }
540 
541 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
542 			 u16 *pkey, int netw_view)
543 {
544 	struct ib_smp *in_mad  = NULL;
545 	struct ib_smp *out_mad = NULL;
546 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
547 	int err = -ENOMEM;
548 
549 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
550 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
551 	if (!in_mad || !out_mad)
552 		goto out;
553 
554 	init_query_mad(in_mad);
555 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
556 	in_mad->attr_mod = cpu_to_be32(index / 32);
557 
558 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
559 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
560 
561 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
562 			   in_mad, out_mad);
563 	if (err)
564 		goto out;
565 
566 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
567 
568 out:
569 	kfree(in_mad);
570 	kfree(out_mad);
571 	return err;
572 }
573 
574 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
575 {
576 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
577 }
578 
579 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
580 				 struct ib_device_modify *props)
581 {
582 	struct mlx4_cmd_mailbox *mailbox;
583 	unsigned long flags;
584 
585 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
586 		return -EOPNOTSUPP;
587 
588 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
589 		return 0;
590 
591 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
592 		return -EOPNOTSUPP;
593 
594 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
595 	memcpy(ibdev->node_desc, props->node_desc, 64);
596 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
597 
598 	/*
599 	 * If possible, pass node desc to FW, so it can generate
600 	 * a 144 trap.  If cmd fails, just ignore.
601 	 */
602 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
603 	if (IS_ERR(mailbox))
604 		return 0;
605 
606 	memset(mailbox->buf, 0, 256);
607 	memcpy(mailbox->buf, props->node_desc, 64);
608 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
609 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
610 
611 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
612 
613 	return 0;
614 }
615 
616 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
617 			 u32 cap_mask)
618 {
619 	struct mlx4_cmd_mailbox *mailbox;
620 	int err;
621 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
622 
623 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
624 	if (IS_ERR(mailbox))
625 		return PTR_ERR(mailbox);
626 
627 	memset(mailbox->buf, 0, 256);
628 
629 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
630 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
631 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
632 	} else {
633 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
634 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
635 	}
636 
637 	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
638 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
639 
640 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
641 	return err;
642 }
643 
644 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
645 			       struct ib_port_modify *props)
646 {
647 	struct ib_port_attr attr;
648 	u32 cap_mask;
649 	int err;
650 
651 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
652 
653 	err = mlx4_ib_query_port(ibdev, port, &attr);
654 	if (err)
655 		goto out;
656 
657 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
658 		~props->clr_port_cap_mask;
659 
660 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
661 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
662 			    cap_mask);
663 
664 out:
665 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
666 	return err;
667 }
668 
669 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
670 						  struct ib_udata *udata)
671 {
672 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
673 	struct mlx4_ib_ucontext *context;
674 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
675 	struct mlx4_ib_alloc_ucontext_resp resp;
676 	int err;
677 
678 	if (!dev->ib_active)
679 		return ERR_PTR(-EAGAIN);
680 
681 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
682 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
683 		if (mlx4_wc_enabled()) {
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_v3.bf_reg_size      = 0;
688 			resp_v3.bf_regs_per_page = 0;
689 		}
690 	} else {
691 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
692 		resp.qp_tab_size      = dev->dev->caps.num_qps;
693 		if (mlx4_wc_enabled()) {
694 			resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
695 			resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
696 		} else {
697 			resp.bf_reg_size      = 0;
698 			resp.bf_regs_per_page = 0;
699 		}
700 		resp.cqe_size	      = dev->dev->caps.cqe_size;
701 	}
702 
703 	context = kmalloc(sizeof *context, GFP_KERNEL);
704 	if (!context)
705 		return ERR_PTR(-ENOMEM);
706 
707 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
708 	if (err) {
709 		kfree(context);
710 		return ERR_PTR(err);
711 	}
712 
713 	INIT_LIST_HEAD(&context->db_page_list);
714 	mutex_init(&context->db_page_mutex);
715 
716 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
717 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
718 	else
719 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
720 
721 	if (err) {
722 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
723 		kfree(context);
724 		return ERR_PTR(-EFAULT);
725 	}
726 
727 	return &context->ibucontext;
728 }
729 
730 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
731 {
732 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
733 
734 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
735 	kfree(context);
736 
737 	return 0;
738 }
739 
740 /* XXX FBSD has no support for get_unmapped_area function */
741 #if 0
742 static unsigned long mlx4_ib_get_unmapped_area(struct file *file,
743 			unsigned long addr,
744 			unsigned long len, unsigned long pgoff,
745 			unsigned long flags)
746 {
747 	struct mm_struct *mm;
748 	struct vm_area_struct *vma;
749 	unsigned long start_addr;
750 	unsigned long page_size_order;
751 	unsigned long  command;
752 
753 	mm = current->mm;
754 	if (addr)
755 		return current->mm->get_unmapped_area(file, addr, len,
756 						pgoff, flags);
757 
758 	/* Last 8 bits hold the  command others are data per that command */
759 	command = pgoff & MLX4_IB_MMAP_CMD_MASK;
760 	if (command != MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES)
761 		return current->mm->get_unmapped_area(file, addr, len,
762 						pgoff, flags);
763 
764 	page_size_order = pgoff >> MLX4_IB_MMAP_CMD_BITS;
765 	/* code is based on the huge-pages get_unmapped_area code */
766 	start_addr = mm->free_area_cache;
767 
768 	if (len <= mm->cached_hole_size)
769 		start_addr = TASK_UNMAPPED_BASE;
770 
771 
772 full_search:
773 	addr = ALIGN(start_addr, 1 << page_size_order);
774 
775 	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
776 		/* At this point:  (!vma || addr < vma->vm_end). */
777 		if (TASK_SIZE - len < addr) {
778 			/*
779 			 * Start a new search - just in case we missed
780 			 * some holes.
781 			 */
782 			if (start_addr != TASK_UNMAPPED_BASE) {
783 				start_addr = TASK_UNMAPPED_BASE;
784 				goto full_search;
785 			}
786 			return -ENOMEM;
787 		}
788 
789 		if (!vma || addr + len <= vma->vm_start)
790 			return addr;
791 		addr = ALIGN(vma->vm_end, 1 << page_size_order);
792 	}
793 }
794 #endif
795 
796 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
797 {
798 	struct mlx4_ib_dev *dev = to_mdev(context->device);
799 
800 	/* Last 8 bits hold the  command others are data per that command */
801 	unsigned long  command = vma->vm_pgoff & MLX4_IB_MMAP_CMD_MASK;
802 
803 	if (command < MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) {
804 		/* compatibility handling for commands 0 & 1*/
805 		if (vma->vm_end - vma->vm_start != PAGE_SIZE)
806 			return -EINVAL;
807 	}
808 	if (command == MLX4_IB_MMAP_UAR_PAGE) {
809 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
810 
811 		if (io_remap_pfn_range(vma, vma->vm_start,
812 				       to_mucontext(context)->uar.pfn,
813 				       PAGE_SIZE, vma->vm_page_prot))
814 			return -EAGAIN;
815 	} else if (command == MLX4_IB_MMAP_BLUE_FLAME_PAGE &&
816 			dev->dev->caps.bf_reg_size != 0) {
817 		vma->vm_page_prot = pgprot_wc(vma->vm_page_prot);
818 
819 		if (io_remap_pfn_range(vma, vma->vm_start,
820 				       to_mucontext(context)->uar.pfn +
821 				       dev->dev->caps.num_uars,
822 				       PAGE_SIZE, vma->vm_page_prot))
823 			return -EAGAIN;
824 	} else if (command == MLX4_IB_MMAP_GET_HW_CLOCK) {
825 		struct mlx4_clock_params params;
826 		int ret;
827 
828 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
829 		if (ret)
830 			return ret;
831 
832 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
833 
834 		if (io_remap_pfn_range(vma, vma->vm_start,
835 				       (pci_resource_start(dev->dev->pdev,
836 				       params.bar) + params.offset)
837 				       >> PAGE_SHIFT,
838 				       PAGE_SIZE, vma->vm_page_prot))
839 			return -EAGAIN;
840 	} else
841 		return -EINVAL;
842 
843 	return 0;
844 }
845 
846 static int mlx4_ib_ioctl(struct ib_ucontext *context, unsigned int cmd,
847 			 unsigned long arg)
848 {
849 	struct mlx4_ib_dev *dev = to_mdev(context->device);
850 	int ret;
851         int offset;
852 
853 	switch (cmd) {
854 	case MLX4_IOCHWCLOCKOFFSET: {
855 		struct mlx4_clock_params params;
856 		int ret;
857 		ret = mlx4_get_internal_clock_params(dev->dev, &params);
858 		if (!ret) {
859                         offset = params.offset % PAGE_SIZE;
860 			ret = put_user(offset,
861 					 (int *)arg);
862 			return sizeof(int);
863 		} else {
864 			return ret;
865 		}
866 	}
867 	default: {
868 		pr_err("mlx4_ib: invalid ioctl %u command with arg %lX\n",
869 		       cmd, arg);
870 		return -ENOTTY;
871 	}
872 	}
873 
874 	return ret;
875 }
876 
877 static int mlx4_ib_query_values(struct ib_device *device, int q_values,
878 				struct ib_device_values *values)
879 {
880 	struct mlx4_ib_dev *dev = to_mdev(device);
881 	cycle_t cycles;
882 
883 	values->values_mask = 0;
884 	if (q_values & IBV_VALUES_HW_CLOCK) {
885 		cycles = mlx4_read_clock(dev->dev);
886 		if (cycles < 0) {
887 			values->hwclock = cycles & CORE_CLOCK_MASK;
888 			values->values_mask |= IBV_VALUES_HW_CLOCK;
889 		}
890 		q_values &= ~IBV_VALUES_HW_CLOCK;
891 	}
892 
893 	if (q_values)
894 		return -ENOTTY;
895 
896 	return 0;
897 }
898 
899 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
900 				      struct ib_ucontext *context,
901 				      struct ib_udata *udata)
902 {
903 	struct mlx4_ib_pd *pd;
904 	int err;
905 
906 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
907 	if (!pd)
908 		return ERR_PTR(-ENOMEM);
909 
910 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
911 	if (err) {
912 		kfree(pd);
913 		return ERR_PTR(err);
914 	}
915 
916 	if (context)
917 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
918 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
919 			kfree(pd);
920 			return ERR_PTR(-EFAULT);
921 		}
922 
923 	return &pd->ibpd;
924 }
925 
926 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
927 {
928 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
929 	kfree(pd);
930 
931 	return 0;
932 }
933 
934 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
935 					  struct ib_ucontext *context,
936 					  struct ib_udata *udata)
937 {
938 	struct mlx4_ib_xrcd *xrcd;
939 	int err;
940 
941 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
942 		return ERR_PTR(-ENOSYS);
943 
944 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
945 	if (!xrcd)
946 		return ERR_PTR(-ENOMEM);
947 
948 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
949 	if (err)
950 		goto err1;
951 
952 	xrcd->pd = ib_alloc_pd(ibdev);
953 	if (IS_ERR(xrcd->pd)) {
954 		err = PTR_ERR(xrcd->pd);
955 		goto err2;
956 	}
957 
958 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
959 	if (IS_ERR(xrcd->cq)) {
960 		err = PTR_ERR(xrcd->cq);
961 		goto err3;
962 	}
963 
964 	return &xrcd->ibxrcd;
965 
966 err3:
967 	ib_dealloc_pd(xrcd->pd);
968 err2:
969 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
970 err1:
971 	kfree(xrcd);
972 	return ERR_PTR(err);
973 }
974 
975 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
976 {
977 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
978 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
979 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
980 	kfree(xrcd);
981 
982 	return 0;
983 }
984 
985 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
986 {
987 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
988 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
989 	struct mlx4_ib_gid_entry *ge;
990 
991 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
992 	if (!ge)
993 		return -ENOMEM;
994 
995 	ge->gid = *gid;
996 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
997 		ge->port = mqp->port;
998 		ge->added = 1;
999 	}
1000 
1001 	mutex_lock(&mqp->mutex);
1002 	list_add_tail(&ge->list, &mqp->gid_list);
1003 	mutex_unlock(&mqp->mutex);
1004 
1005 	return 0;
1006 }
1007 
1008 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1009 		   union ib_gid *gid)
1010 {
1011 	u8 mac[6];
1012 	struct net_device *ndev;
1013 	int ret = 0;
1014 
1015 	if (!mqp->port)
1016 		return 0;
1017 
1018 	spin_lock(&mdev->iboe.lock);
1019 	ndev = mdev->iboe.netdevs[mqp->port - 1];
1020 	if (ndev)
1021 		dev_hold(ndev);
1022 	spin_unlock(&mdev->iboe.lock);
1023 
1024 	if (ndev) {
1025 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1026 		rtnl_lock();
1027 		dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac, 6, 0);
1028 		ret = 1;
1029 		rtnl_unlock();
1030 		dev_put(ndev);
1031 	}
1032 
1033 	return ret;
1034 }
1035 
1036 struct mlx4_ib_steering {
1037 	struct list_head list;
1038 	u64 reg_id;
1039 	union ib_gid gid;
1040 };
1041 
1042 static int parse_flow_attr(struct mlx4_dev *dev,
1043 			   union ib_flow_spec *ib_spec,
1044 			   struct _rule_hw *mlx4_spec)
1045 {
1046 	enum mlx4_net_trans_rule_id type;
1047 
1048 	switch (ib_spec->type) {
1049 	case IB_FLOW_SPEC_ETH:
1050 		type = MLX4_NET_TRANS_RULE_ID_ETH;
1051 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1052 		       ETH_ALEN);
1053 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1054 		       ETH_ALEN);
1055 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1056 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1057 		break;
1058 
1059 	case IB_FLOW_SPEC_IB:
1060 		type = MLX4_NET_TRANS_RULE_ID_IB;
1061 		mlx4_spec->ib.l3_qpn = ib_spec->ib.val.l3_type_qpn;
1062 		mlx4_spec->ib.qpn_mask = ib_spec->ib.mask.l3_type_qpn;
1063 		memcpy(&mlx4_spec->ib.dst_gid, ib_spec->ib.val.dst_gid, 16);
1064 		memcpy(&mlx4_spec->ib.dst_gid_msk,
1065 		       ib_spec->ib.mask.dst_gid, 16);
1066 		break;
1067 
1068 	case IB_FLOW_SPEC_IPV4:
1069 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
1070 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1071 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1072 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1073 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1074 		break;
1075 
1076 	case IB_FLOW_SPEC_TCP:
1077 	case IB_FLOW_SPEC_UDP:
1078 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1079 					MLX4_NET_TRANS_RULE_ID_TCP :
1080 					MLX4_NET_TRANS_RULE_ID_UDP;
1081 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1082 		mlx4_spec->tcp_udp.dst_port_msk =
1083 			ib_spec->tcp_udp.mask.dst_port;
1084 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1085 		mlx4_spec->tcp_udp.src_port_msk =
1086 			ib_spec->tcp_udp.mask.src_port;
1087 		break;
1088 
1089 	default:
1090 		return -EINVAL;
1091 	}
1092 	if (map_sw_to_hw_steering_id(dev, type) < 0 ||
1093 	    hw_rule_sz(dev, type) < 0)
1094 		return -EINVAL;
1095 	mlx4_spec->id = cpu_to_be16(map_sw_to_hw_steering_id(dev, type));
1096 	mlx4_spec->size = hw_rule_sz(dev, type) >> 2;
1097 	return hw_rule_sz(dev, type);
1098 }
1099 
1100 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1101 			  int domain,
1102 			  enum mlx4_net_trans_promisc_mode flow_type,
1103 			  u64 *reg_id)
1104 {
1105 	int ret, i;
1106 	int size = 0;
1107 	void *ib_flow;
1108 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1109 	struct mlx4_cmd_mailbox *mailbox;
1110 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1111 	size_t rule_size = sizeof(struct mlx4_net_trans_rule_hw_ctrl) +
1112 			   (sizeof(struct _rule_hw) * flow_attr->num_of_specs);
1113 
1114 	static const u16 __mlx4_domain[] = {
1115 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1116 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1117 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1118 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1119 	};
1120 
1121 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1122 		pr_err("Invalid priority value.\n");
1123 		return -EINVAL;
1124 	}
1125 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1126 		pr_err("Invalid domain value.\n");
1127 		return -EINVAL;
1128 	}
1129 	if (map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1130 		return -EINVAL;
1131 
1132 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1133 	if (IS_ERR(mailbox))
1134 		return PTR_ERR(mailbox);
1135 	memset(mailbox->buf, 0, rule_size);
1136 	ctrl = mailbox->buf;
1137 
1138 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1139 				 flow_attr->priority);
1140 	ctrl->type = map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1141 	ctrl->port = flow_attr->port;
1142 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1143 
1144 	if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_ALLOW_LOOP_BACK)
1145 		ctrl->flags = (1 << 3);
1146 
1147 	ib_flow = flow_attr + 1;
1148 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1149 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1150 		ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
1151 		if (ret < 0) {
1152 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1153 			return -EINVAL;
1154 		}
1155 		ib_flow += ((union ib_flow_spec *)ib_flow)->size;
1156 		size += ret;
1157 	}
1158 
1159 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1160 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1161 			   MLX4_CMD_NATIVE);
1162 	if (ret == -ENOMEM)
1163 		pr_err("mcg table is full. Fail to register network rule.\n");
1164 	else if (ret == -ENXIO)
1165 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1166 	else if (ret)
1167 		pr_err("Invalid argumant. Fail to register network rule.\n");
1168 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1169 	return ret;
1170 }
1171 
1172 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1173 {
1174 	int err;
1175 	err = mlx4_cmd(dev, reg_id, 0, 0,
1176 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1177 		       MLX4_CMD_NATIVE);
1178 	if (err)
1179 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1180 		       (unsigned long long)reg_id);
1181 	return err;
1182 }
1183 
1184 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1185 				    struct ib_flow_attr *flow_attr,
1186 				    int domain)
1187 {
1188 	int err = 0, i = 0;
1189 	struct mlx4_ib_flow *mflow;
1190 	enum mlx4_net_trans_promisc_mode type[2];
1191 
1192 	memset(type, 0, sizeof(type));
1193 
1194 	mflow = kzalloc(sizeof(struct mlx4_ib_flow), GFP_KERNEL);
1195 	if (!mflow) {
1196 		err = -ENOMEM;
1197 		goto err_free;
1198 	}
1199 
1200 	switch (flow_attr->type) {
1201 	case IB_FLOW_ATTR_NORMAL:
1202 		type[0] = MLX4_FS_REGULAR;
1203 		break;
1204 
1205 	case IB_FLOW_ATTR_ALL_DEFAULT:
1206 		type[0] = MLX4_FS_ALL_DEFAULT;
1207 		break;
1208 
1209 	case IB_FLOW_ATTR_MC_DEFAULT:
1210 		type[0] = MLX4_FS_MC_DEFAULT;
1211 		break;
1212 
1213 	case IB_FLOW_ATTR_SNIFFER:
1214 		type[0] = MLX4_FS_UC_SNIFFER;
1215 		type[1] = MLX4_FS_MC_SNIFFER;
1216 		break;
1217 
1218 	default:
1219 		err = -EINVAL;
1220 		goto err_free;
1221 	}
1222 
1223 	while (i < ARRAY_SIZE(type) && type[i]) {
1224 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1225 					    &mflow->reg_id[i]);
1226 		if (err)
1227 			goto err_free;
1228 		i++;
1229 	}
1230 
1231 	return &mflow->ibflow;
1232 
1233 err_free:
1234 	kfree(mflow);
1235 	return ERR_PTR(err);
1236 }
1237 
1238 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1239 {
1240 	int err, ret = 0;
1241 	int i = 0;
1242 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1243 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1244 
1245 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1246 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1247 		if (err)
1248 			ret = err;
1249 		i++;
1250 	}
1251 
1252 	kfree(mflow);
1253 	return ret;
1254 }
1255 
1256 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1257 {
1258 	struct mlx4_ib_gid_entry *ge;
1259 	struct mlx4_ib_gid_entry *tmp;
1260 	struct mlx4_ib_gid_entry *ret = NULL;
1261 
1262 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1263 		if (!memcmp(raw, ge->gid.raw, 16)) {
1264 			ret = ge;
1265 			break;
1266 		}
1267 	}
1268 
1269 	return ret;
1270 }
1271 
1272 
1273 static int del_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1274 {
1275 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1276 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1277 	struct mlx4_ib_gid_entry *ge;
1278 	struct net_device *ndev;
1279 	u8 mac[6];
1280 
1281 	mutex_lock(&mqp->mutex);
1282 	ge = find_gid_entry(mqp, gid->raw);
1283 	if (ge) {
1284 		spin_lock(&mdev->iboe.lock);
1285 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1286 		if (ndev)
1287 			dev_hold(ndev);
1288 		spin_unlock(&mdev->iboe.lock);
1289 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1290 		if (ndev) {
1291 			rtnl_lock();
1292 			dev_mc_delete(mdev->iboe.netdevs[ge->port - 1], mac, 6, 0);
1293 			rtnl_unlock();
1294 			dev_put(ndev);
1295 		}
1296 		list_del(&ge->list);
1297 		kfree(ge);
1298 	} else
1299 		pr_warn("could not find mgid entry\n");
1300 
1301 	mutex_unlock(&mqp->mutex);
1302 	return ge != NULL ? 0 : -EINVAL;
1303 }
1304 
1305 static int _mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid,
1306 			       int count)
1307 {
1308 	int err;
1309 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1310 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1311 	u64 reg_id = 0;
1312 	int record_err = 0;
1313 
1314 	if (mdev->dev->caps.steering_mode ==
1315 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1316 		struct mlx4_ib_steering *ib_steering;
1317 		struct mlx4_ib_steering *tmp;
1318 		LIST_HEAD(temp);
1319 
1320 		mutex_lock(&mqp->mutex);
1321 		list_for_each_entry_safe(ib_steering, tmp, &mqp->steering_rules,
1322 					 list) {
1323 			if (memcmp(ib_steering->gid.raw, gid->raw, 16))
1324 				continue;
1325 
1326 			if (--count < 0)
1327 				break;
1328 
1329 			list_del(&ib_steering->list);
1330 			list_add(&ib_steering->list, &temp);
1331 		}
1332 		mutex_unlock(&mqp->mutex);
1333 		list_for_each_entry_safe(ib_steering, tmp, &temp,
1334 					 list) {
1335 			reg_id = ib_steering->reg_id;
1336 
1337 			err = mlx4_multicast_detach(mdev->dev, &mqp->mqp,
1338 					gid->raw,
1339 					(ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1340 					MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1341 					reg_id);
1342 			if (err) {
1343 				record_err = record_err ?: err;
1344 				continue;
1345 			}
1346 
1347 			err = del_gid_entry(ibqp, gid);
1348 			if (err) {
1349 				record_err = record_err ?: err;
1350 				continue;
1351 			}
1352 
1353 			list_del(&ib_steering->list);
1354 			kfree(ib_steering);
1355 		}
1356 		mutex_lock(&mqp->mutex);
1357 		list_for_each_entry(ib_steering, &temp, list) {
1358 			list_add(&ib_steering->list, &mqp->steering_rules);
1359 		}
1360 		mutex_unlock(&mqp->mutex);
1361 		if (count) {
1362 			pr_warn("Couldn't release all reg_ids for mgid. Steering rule is left attached\n");
1363 			return -EINVAL;
1364 		}
1365 
1366 	} else {
1367 		if (mdev->dev->caps.steering_mode == MLX4_STEERING_MODE_B0 &&
1368 		    ibqp->qp_type == IB_QPT_RAW_PACKET)
1369 			gid->raw[5] = mqp->port;
1370 
1371 		err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1372 				(ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1373 				MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1374 				reg_id);
1375 		if (err)
1376 			return err;
1377 
1378 		err = del_gid_entry(ibqp, gid);
1379 
1380 		if (err)
1381 			return err;
1382 	}
1383 
1384 	return record_err;
1385 }
1386 
1387 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1388 {
1389 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1390 	int count = (mdev->dev->caps.steering_mode ==
1391 		     MLX4_STEERING_MODE_DEVICE_MANAGED) ?
1392 		    mdev->dev->caps.num_ports : 1;
1393 
1394 	return _mlx4_ib_mcg_detach(ibqp, gid, lid, count);
1395 }
1396 
1397 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1398 {
1399 	int err = -ENODEV;
1400 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1401 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1402 	DECLARE_BITMAP(ports, MLX4_MAX_PORTS);
1403 	int i = 0;
1404 
1405 	if (mdev->dev->caps.steering_mode == MLX4_STEERING_MODE_B0 &&
1406 	    ibqp->qp_type == IB_QPT_RAW_PACKET)
1407 		gid->raw[5] = mqp->port;
1408 
1409 	if (mdev->dev->caps.steering_mode ==
1410 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1411 		bitmap_fill(ports, mdev->dev->caps.num_ports);
1412 	} else {
1413 		if (mqp->port <= mdev->dev->caps.num_ports) {
1414 			bitmap_zero(ports, mdev->dev->caps.num_ports);
1415 			set_bit(0, ports);
1416 		} else {
1417 			return -EINVAL;
1418 		}
1419 	}
1420 
1421 	for (; i < mdev->dev->caps.num_ports; i++) {
1422 		u64 reg_id;
1423 		struct mlx4_ib_steering *ib_steering = NULL;
1424 		if (!test_bit(i, ports))
1425 			continue;
1426 		if (mdev->dev->caps.steering_mode ==
1427 		    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1428 			ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1429 			if (!ib_steering)
1430 				goto err_add;
1431 		}
1432 
1433 		err = mlx4_multicast_attach(mdev->dev, &mqp->mqp,
1434 			gid->raw, i + 1,
1435 			!!(mqp->flags &
1436 				MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1437 			(ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1438 			MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1439 			&reg_id);
1440 		if (err) {
1441 			kfree(ib_steering);
1442 			goto err_add;
1443 		}
1444 
1445 		err = add_gid_entry(ibqp, gid);
1446 		if (err) {
1447 			mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1448 					      MLX4_PROT_IB_IPV6, reg_id);
1449 			kfree(ib_steering);
1450 			goto err_add;
1451 		}
1452 
1453 		if (ib_steering) {
1454 			memcpy(ib_steering->gid.raw, gid->raw, 16);
1455 			mutex_lock(&mqp->mutex);
1456 			list_add(&ib_steering->list, &mqp->steering_rules);
1457 			mutex_unlock(&mqp->mutex);
1458 			ib_steering->reg_id = reg_id;
1459 		}
1460 	}
1461 
1462 
1463 	return 0;
1464 
1465 err_add:
1466 	if (i > 0)
1467 		_mlx4_ib_mcg_detach(ibqp, gid, lid, i);
1468 
1469 	return err;
1470 }
1471 
1472 static int init_node_data(struct mlx4_ib_dev *dev)
1473 {
1474 	struct ib_smp *in_mad  = NULL;
1475 	struct ib_smp *out_mad = NULL;
1476 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1477 	int err = -ENOMEM;
1478 
1479 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1480 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1481 	if (!in_mad || !out_mad)
1482 		goto out;
1483 
1484 	init_query_mad(in_mad);
1485 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1486 	if (mlx4_is_master(dev->dev))
1487 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
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 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1494 
1495 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1496 
1497 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1498 	if (err)
1499 		goto out;
1500 
1501 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1502 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1503 
1504 out:
1505 	kfree(in_mad);
1506 	kfree(out_mad);
1507 	return err;
1508 }
1509 
1510 static ssize_t show_hca(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, "MT%d\n", dev->dev->pdev->device);
1516 }
1517 
1518 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1519 			   char *buf)
1520 {
1521 	struct mlx4_ib_dev *dev =
1522 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1523 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1524 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1525 		       (int) dev->dev->caps.fw_ver & 0xffff);
1526 }
1527 
1528 static ssize_t show_rev(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, "%x\n", dev->dev->rev_id);
1534 }
1535 
1536 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1537 			  char *buf)
1538 {
1539 	struct mlx4_ib_dev *dev =
1540 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1541 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1542 		       dev->dev->board_id);
1543 }
1544 
1545 static ssize_t show_vsd(struct device *device, struct device_attribute *attr,
1546 			  char *buf)
1547 {
1548 	struct mlx4_ib_dev *dev =
1549 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1550 	ssize_t len = MLX4_VSD_LEN;
1551 
1552 	if (dev->dev->vsd_vendor_id == PCI_VENDOR_ID_MELLANOX)
1553 		len = sprintf(buf, "%.*s\n", MLX4_VSD_LEN, dev->dev->vsd);
1554 	else
1555 		memcpy(buf, dev->dev->vsd, MLX4_VSD_LEN);
1556 
1557 	return len;
1558 }
1559 
1560 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1561 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1562 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1563 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1564 static DEVICE_ATTR(vsd,      S_IRUGO, show_vsd,    NULL);
1565 
1566 static struct device_attribute *mlx4_class_attributes[] = {
1567 	&dev_attr_hw_rev,
1568 	&dev_attr_fw_ver,
1569 	&dev_attr_hca_type,
1570 	&dev_attr_board_id,
1571 	&dev_attr_vsd
1572 };
1573 
1574 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev, u8 port)
1575 {
1576         memcpy(eui, IF_LLADDR(dev), 3);
1577         memcpy(eui + 5, IF_LLADDR(dev) + 3, 3);
1578 	if (vlan_id < 0x1000) {
1579 		eui[3] = vlan_id >> 8;
1580 		eui[4] = vlan_id & 0xff;
1581 	} else {
1582 		eui[3] = 0xff;
1583 		eui[4] = 0xfe;
1584 	}
1585 	eui[0] ^= 2;
1586 }
1587 
1588 static void update_gids_task(struct work_struct *work)
1589 {
1590 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1591 	struct mlx4_cmd_mailbox *mailbox;
1592 	union ib_gid *gids;
1593 	int err;
1594 	struct mlx4_dev	*dev = gw->dev->dev;
1595 
1596 
1597 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1598 	if (IS_ERR(mailbox)) {
1599 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1600 		goto free;
1601 	}
1602 
1603 	gids = mailbox->buf;
1604 	memcpy(gids, gw->gids, sizeof gw->gids);
1605 
1606 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1607 					IB_LINK_LAYER_ETHERNET) {
1608 		err = mlx4_cmd(dev, mailbox->dma,
1609 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1610 			       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1611 			       MLX4_CMD_WRAPPED);
1612 
1613 		if (err)
1614 			pr_warn("set port command failed\n");
1615 		else
1616 			mlx4_ib_dispatch_event(gw->dev, gw->port,
1617 					       IB_EVENT_GID_CHANGE);
1618 	}
1619 
1620 	mlx4_free_cmd_mailbox(dev, mailbox);
1621 free:
1622 	kfree(gw);
1623 }
1624 
1625 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
1626 {
1627 	struct mlx4_ib_dev *ibdev = to_mdev(device);
1628 	return mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
1629 }
1630 
1631 static void reset_gids_task(struct work_struct *work)
1632 {
1633 	struct update_gid_work *gw =
1634 			container_of(work, struct update_gid_work, work);
1635 	struct mlx4_cmd_mailbox *mailbox;
1636 	union ib_gid *gids;
1637 	int err;
1638 	struct mlx4_dev	*dev = gw->dev->dev;
1639 
1640 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1641 	if (IS_ERR(mailbox)) {
1642 		pr_warn("reset gid table failed\n");
1643 		goto free;
1644 	}
1645 
1646 	gids = mailbox->buf;
1647 	memcpy(gids, gw->gids, sizeof(gw->gids));
1648 
1649 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, 1) ==
1650 					IB_LINK_LAYER_ETHERNET &&
1651 					dev->caps.num_ports > 0) {
1652 		err = mlx4_cmd(dev, mailbox->dma,
1653 			       MLX4_SET_PORT_GID_TABLE << 8 | 1,
1654 			       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1655 			       MLX4_CMD_WRAPPED);
1656 		if (err)
1657 			pr_warn("set port 1 command failed\n");
1658 	}
1659 
1660 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, 2) ==
1661 					IB_LINK_LAYER_ETHERNET &&
1662 					dev->caps.num_ports > 1) {
1663 		err = mlx4_cmd(dev, mailbox->dma,
1664 			       MLX4_SET_PORT_GID_TABLE << 8 | 2,
1665 			       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1666 			       MLX4_CMD_WRAPPED);
1667 		if (err)
1668 			pr_warn("set port 2 command failed\n");
1669 	}
1670 
1671 	mlx4_free_cmd_mailbox(dev, mailbox);
1672 free:
1673 	kfree(gw);
1674 }
1675 
1676 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1677 		union ib_gid *gid, int clear, int default_gid)
1678 {
1679 	struct update_gid_work *work;
1680 	int i;
1681 	int need_update = 0;
1682 	int free = -1;
1683 	int found = -1;
1684 	int max_gids;
1685 	int start_index = !default_gid;
1686 
1687 	max_gids = dev->dev->caps.gid_table_len[port];
1688 	for (i = start_index; i < max_gids; ++i) {
1689 		if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1690 		    sizeof(*gid)))
1691 			found = i;
1692 
1693 		if (clear) {
1694 			if (found >= 0) {
1695 				need_update = 1;
1696 				dev->iboe.gid_table[port - 1][found] = zgid;
1697 				break;
1698 			}
1699 		} else {
1700 			if (found >= 0)
1701 				break;
1702 
1703 			if (free < 0 &&
1704 			    !memcmp(&dev->iboe.gid_table[port - 1][i],
1705 				    &zgid, sizeof(*gid)))
1706 				free = i;
1707 		}
1708 	}
1709 
1710 	if (found == -1 && !clear && free < 0) {
1711 		pr_err("GID table of port %d is full. Can't add "GID_PRINT_FMT"\n",
1712 		       port, GID_PRINT_ARGS(gid));
1713 		return -ENOMEM;
1714 	}
1715 	if (found == -1 && clear) {
1716 		pr_err(GID_PRINT_FMT" is not in GID table of port %d\n", GID_PRINT_ARGS(gid), port);
1717 		return -EINVAL;
1718 	}
1719 	if (found == -1 && !clear && free >= 0) {
1720 		dev->iboe.gid_table[port - 1][free] = *gid;
1721 		need_update = 1;
1722 	}
1723 
1724 	if (!need_update)
1725 		return 0;
1726 
1727 	work = kzalloc(sizeof *work, GFP_ATOMIC);
1728 	if (!work)
1729 		return -ENOMEM;
1730 
1731 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1732 	INIT_WORK(&work->work, update_gids_task);
1733 	work->port = port;
1734 	work->dev = dev;
1735 	queue_work(wq, &work->work);
1736 
1737 	return 0;
1738 }
1739 
1740 static int reset_gid_table(struct mlx4_ib_dev *dev)
1741 {
1742 	struct update_gid_work *work;
1743 
1744 
1745 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1746 	if (!work)
1747 		return -ENOMEM;
1748 
1749 	memset(dev->iboe.gid_table, 0, sizeof(dev->iboe.gid_table));
1750 	memset(work->gids, 0, sizeof(work->gids));
1751 	INIT_WORK(&work->work, reset_gids_task);
1752 	work->dev = dev;
1753 	queue_work(wq, &work->work);
1754 	return 0;
1755 }
1756 
1757 /* XXX BOND Related - stub (no support for these flags in FBSD)*/
1758 static inline int netif_is_bond_master(struct net_device *dev)
1759 {
1760 #if 0
1761 	return (dev->flags & IFF_MASTER) && (dev->priv_flags & IFF_BONDING);
1762 #endif
1763         return 0;
1764 }
1765 
1766 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid, u8 port)
1767 {
1768 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1769 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev, port);
1770 }
1771 
1772 static u8 mlx4_ib_get_dev_port(struct net_device *dev, struct mlx4_ib_dev *ibdev)
1773 {
1774 	u8 port = 0;
1775 	struct mlx4_ib_iboe *iboe;
1776 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1777 				rdma_vlan_dev_real_dev(dev) : dev;
1778 
1779 	iboe = &ibdev->iboe;
1780 
1781 	for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1782 		if ((netif_is_bond_master(real_dev) && (real_dev == iboe->masters[port - 1])) ||
1783 		    (!netif_is_bond_master(real_dev) && (real_dev == iboe->netdevs[port - 1])))
1784 			break;
1785 
1786 	return port > MLX4_MAX_PORTS ? 0 : port;
1787 }
1788 
1789 static void mlx4_ib_get_dev_addr(struct net_device *dev, struct mlx4_ib_dev *ibdev, u8 port)
1790 {
1791         struct ifaddr *ifa;
1792 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1793 	struct inet6_dev *in6_dev;
1794 	union ib_gid  *pgid;
1795 	struct inet6_ifaddr *ifp;
1796 #endif
1797 	union ib_gid gid;
1798 
1799 
1800 	if ((port == 0) || (port > MLX4_MAX_PORTS))
1801 		return;
1802 
1803 	/* IPv4 gids */
1804         TAILQ_FOREACH(ifa, &dev->if_addrhead, ifa_link) {
1805                 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET){
1806                         ipv6_addr_set_v4mapped(
1807 				((struct sockaddr_in *) ifa->ifa_addr)->sin_addr.s_addr,
1808 				(struct in6_addr *)&gid);
1809                         update_gid_table(ibdev, port, &gid, 0, 0);
1810                 }
1811 
1812         }
1813 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1814 	/* IPv6 gids */
1815 	in6_dev = in6_dev_get(dev);
1816 	if (in6_dev) {
1817 		read_lock_bh(&in6_dev->lock);
1818 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1819 			pgid = (union ib_gid *)&ifp->addr;
1820 				update_gid_table(ibdev, port, pgid, 0, 0);
1821 			}
1822 		read_unlock_bh(&in6_dev->lock);
1823 		in6_dev_put(in6_dev);
1824 		}
1825 #endif
1826 }
1827 
1828 static void mlx4_set_default_gid(struct mlx4_ib_dev *ibdev,
1829 				 struct  net_device *dev, u8 port)
1830 {
1831 	union ib_gid gid;
1832 	mlx4_make_default_gid(dev, &gid, port);
1833 	update_gid_table(ibdev, port, &gid, 0, 1);
1834 }
1835 
1836 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1837 {
1838 	struct	net_device *dev;
1839 
1840 	if (reset_gid_table(ibdev))
1841 		return -1;
1842 
1843         IFNET_RLOCK_NOSLEEP();
1844         TAILQ_FOREACH(dev, &V_ifnet, if_link) {
1845 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1846 		if (port) {
1847 			if (!rdma_vlan_dev_real_dev(dev) &&
1848 			    !netif_is_bond_master(dev))
1849 				mlx4_set_default_gid(ibdev, dev, port);
1850 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1851 		}
1852 	}
1853 
1854         IFNET_RUNLOCK_NOSLEEP();
1855 
1856 	return 0;
1857 }
1858 
1859 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1860 				 struct net_device *dev, unsigned long event)
1861 {
1862 	struct mlx4_ib_iboe *iboe;
1863 	int port;
1864 	int init = 0;
1865 	unsigned long flags;
1866 
1867 	iboe = &ibdev->iboe;
1868 
1869 	spin_lock_irqsave(&iboe->lock, flags);
1870 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1871 		struct net_device *old_netdev = iboe->netdevs[port - 1];
1872 /* XXX BOND related */
1873 #if 0
1874 		struct net_device *old_master = iboe->masters[port - 1];
1875 #endif
1876 		iboe->masters[port - 1] = NULL;
1877 		iboe->netdevs[port - 1] =
1878 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1879 
1880 
1881 		if (old_netdev != iboe->netdevs[port - 1])
1882 			init = 1;
1883 		if (dev == iboe->netdevs[port - 1] &&
1884 		    event == NETDEV_CHANGEADDR)
1885 			init = 1;
1886 /* XXX BOND related */
1887 #if 0
1888                 if (iboe->netdevs[port - 1] && netif_is_bond_slave(iboe->netdevs[port - 1]))
1889 			iboe->masters[port - 1] = iboe->netdevs[port - 1]->master;
1890 
1891 		/* if bonding is used it is possible that we add it to masters only after
1892 		   IP address is assigned to the net bonding interface */
1893 		if (old_master != iboe->masters[port - 1])
1894 			init = 1;
1895 #endif
1896 	}
1897 
1898 	spin_unlock_irqrestore(&iboe->lock, flags);
1899 
1900 	if (init)
1901 		if (mlx4_ib_init_gid_table(ibdev))
1902 			pr_warn("Fail to reset gid table\n");
1903 }
1904 
1905 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1906 				void *ptr)
1907 {
1908 	struct net_device *dev = ptr;
1909 	struct mlx4_ib_dev *ibdev;
1910 
1911 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1912 
1913 	mlx4_ib_scan_netdevs(ibdev, dev, event);
1914 
1915 	return NOTIFY_DONE;
1916 }
1917 
1918 /* This function initializes the gid table only if the event_netdev real device is an iboe
1919  * device, will be invoked by the inet/inet6 events */
1920 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1921                                 void *ptr)
1922 {
1923         struct net_device *event_netdev = ptr;
1924         struct mlx4_ib_dev *ibdev;
1925         struct mlx4_ib_iboe *ibdev_iboe;
1926         int port = 0;
1927 
1928         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1929 
1930         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1931                         rdma_vlan_dev_real_dev(event_netdev) :
1932                         event_netdev;
1933 
1934         ibdev_iboe = &ibdev->iboe;
1935 
1936         port = mlx4_ib_get_dev_port(real_dev, ibdev);
1937 
1938         /* Perform init_gid_table if the event real_dev is the net_device which represents this port,
1939          * otherwise this event is not related and would be ignored.*/
1940         if(port && (real_dev == ibdev_iboe->netdevs[port - 1]))
1941                 if (mlx4_ib_init_gid_table(ibdev))
1942                         pr_warn("Fail to reset gid table\n");
1943 
1944         return NOTIFY_DONE;
1945 }
1946 
1947 
1948 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1949 {
1950 	int port;
1951 	int slave;
1952 	int i;
1953 
1954 	if (mlx4_is_master(ibdev->dev)) {
1955 		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1956 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1957 				for (i = 0;
1958 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1959 				     ++i) {
1960 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1961 					/* master has the identity virt2phys pkey mapping */
1962 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1963 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1964 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1965 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1966 				}
1967 			}
1968 		}
1969 		/* initialize pkey cache */
1970 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1971 			for (i = 0;
1972 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1973 			     ++i)
1974 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
1975 					(i) ? 0 : 0xFFFF;
1976 		}
1977 	}
1978 }
1979 
1980 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1981 {
1982 	char name[32];
1983 	int eq_per_port = 0;
1984 	int added_eqs = 0;
1985 	int total_eqs = 0;
1986 	int i, j, eq;
1987 
1988 	/* Legacy mode or comp_pool is not large enough */
1989 	if (dev->caps.comp_pool == 0 ||
1990 	    dev->caps.num_ports > dev->caps.comp_pool)
1991 		return;
1992 
1993 	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1994 					dev->caps.num_ports);
1995 
1996 	/* Init eq table */
1997 	added_eqs = 0;
1998 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1999 		added_eqs += eq_per_port;
2000 
2001 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
2002 
2003 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2004 	if (!ibdev->eq_table)
2005 		return;
2006 
2007 	ibdev->eq_added = added_eqs;
2008 
2009 	eq = 0;
2010 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2011 		for (j = 0; j < eq_per_port; j++) {
2012 			sprintf(name, "mlx4-ib-%d-%d@%d:%d:%d:%d", i, j,
2013 						pci_get_domain(dev->pdev->dev.bsddev),
2014 						pci_get_bus(dev->pdev->dev.bsddev),
2015 						PCI_SLOT(dev->pdev->devfn),
2016 						PCI_FUNC(dev->pdev->devfn));
2017 
2018 			/* Set IRQ for specific name (per ring) */
2019 			if (mlx4_assign_eq(dev, name,
2020 					   &ibdev->eq_table[eq])) {
2021 				/* Use legacy (same as mlx4_en driver) */
2022 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2023 				ibdev->eq_table[eq] =
2024 					(eq % dev->caps.num_comp_vectors);
2025 			}
2026 			eq++;
2027 		}
2028 	}
2029 
2030 	/* Fill the reset of the vector with legacy EQ */
2031 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2032 		ibdev->eq_table[eq++] = i;
2033 
2034 	/* Advertise the new number of EQs to clients */
2035 	ibdev->ib_dev.num_comp_vectors = total_eqs;
2036 }
2037 
2038 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2039 {
2040 	int i;
2041 
2042 	/* no additional eqs were added */
2043 	if (!ibdev->eq_table)
2044 		return;
2045 
2046 	/* Reset the advertised EQ number */
2047 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2048 
2049 	/* Free only the added eqs */
2050 	for (i = 0; i < ibdev->eq_added; i++) {
2051 		/* Don't free legacy eqs if used */
2052 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2053 			continue;
2054 		mlx4_release_eq(dev, ibdev->eq_table[i]);
2055 	}
2056 
2057 	kfree(ibdev->eq_table);
2058 }
2059 
2060 /*
2061  * create show function and a device_attribute struct pointing to
2062  * the function for _name
2063  */
2064 #define DEVICE_DIAG_RPRT_ATTR(_name, _offset, _op_mod)		\
2065 static ssize_t show_rprt_##_name(struct device *dev,		\
2066 				 struct device_attribute *attr,	\
2067 				 char *buf){			\
2068 	return show_diag_rprt(dev, buf, _offset, _op_mod);	\
2069 }								\
2070 static DEVICE_ATTR(_name, S_IRUGO, show_rprt_##_name, NULL);
2071 
2072 #define MLX4_DIAG_RPRT_CLEAR_DIAGS 3
2073 
2074 static size_t show_diag_rprt(struct device *device, char *buf,
2075 			     u32 offset, u8 op_modifier)
2076 {
2077 	size_t ret;
2078 	u32 counter_offset = offset;
2079 	u32 diag_counter = 0;
2080 	struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
2081 					       ib_dev.dev);
2082 
2083 	ret = mlx4_query_diag_counters(dev->dev, 1, op_modifier,
2084 				       &counter_offset, &diag_counter);
2085 	if (ret)
2086 		return ret;
2087 
2088 	return sprintf(buf, "%d\n", diag_counter);
2089 }
2090 
2091 static ssize_t clear_diag_counters(struct device *device,
2092 				   struct device_attribute *attr,
2093 				   const char *buf, size_t length)
2094 {
2095 	size_t ret;
2096 	struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
2097 					       ib_dev.dev);
2098 
2099 	ret = mlx4_query_diag_counters(dev->dev, 0, MLX4_DIAG_RPRT_CLEAR_DIAGS,
2100 				       NULL, NULL);
2101 	if (ret)
2102 		return ret;
2103 
2104 	return length;
2105 }
2106 
2107 DEVICE_DIAG_RPRT_ATTR(rq_num_lle	, 0x00, 2);
2108 DEVICE_DIAG_RPRT_ATTR(sq_num_lle	, 0x04, 2);
2109 DEVICE_DIAG_RPRT_ATTR(rq_num_lqpoe	, 0x08, 2);
2110 DEVICE_DIAG_RPRT_ATTR(sq_num_lqpoe 	, 0x0C, 2);
2111 DEVICE_DIAG_RPRT_ATTR(rq_num_lpe	, 0x18, 2);
2112 DEVICE_DIAG_RPRT_ATTR(sq_num_lpe	, 0x1C, 2);
2113 DEVICE_DIAG_RPRT_ATTR(rq_num_wrfe	, 0x20, 2);
2114 DEVICE_DIAG_RPRT_ATTR(sq_num_wrfe	, 0x24, 2);
2115 DEVICE_DIAG_RPRT_ATTR(sq_num_mwbe	, 0x2C, 2);
2116 DEVICE_DIAG_RPRT_ATTR(sq_num_bre	, 0x34, 2);
2117 DEVICE_DIAG_RPRT_ATTR(rq_num_lae	, 0x38, 2);
2118 DEVICE_DIAG_RPRT_ATTR(sq_num_rire	, 0x44, 2);
2119 DEVICE_DIAG_RPRT_ATTR(rq_num_rire	, 0x48, 2);
2120 DEVICE_DIAG_RPRT_ATTR(sq_num_rae	, 0x4C, 2);
2121 DEVICE_DIAG_RPRT_ATTR(rq_num_rae	, 0x50, 2);
2122 DEVICE_DIAG_RPRT_ATTR(sq_num_roe	, 0x54, 2);
2123 DEVICE_DIAG_RPRT_ATTR(sq_num_tree	, 0x5C, 2);
2124 DEVICE_DIAG_RPRT_ATTR(sq_num_rree	, 0x64, 2);
2125 DEVICE_DIAG_RPRT_ATTR(rq_num_rnr	, 0x68, 2);
2126 DEVICE_DIAG_RPRT_ATTR(sq_num_rnr	, 0x6C, 2);
2127 DEVICE_DIAG_RPRT_ATTR(rq_num_oos	, 0x100, 2);
2128 DEVICE_DIAG_RPRT_ATTR(sq_num_oos	, 0x104, 2);
2129 DEVICE_DIAG_RPRT_ATTR(rq_num_mce	, 0x108, 2);
2130 DEVICE_DIAG_RPRT_ATTR(rq_num_udsdprd	, 0x118, 2);
2131 DEVICE_DIAG_RPRT_ATTR(rq_num_ucsdprd	, 0x120, 2);
2132 DEVICE_DIAG_RPRT_ATTR(num_cqovf		, 0x1A0, 2);
2133 DEVICE_DIAG_RPRT_ATTR(num_eqovf		, 0x1A4, 2);
2134 DEVICE_DIAG_RPRT_ATTR(num_baddb		, 0x1A8, 2);
2135 
2136 static DEVICE_ATTR(clear_diag, S_IWUSR, NULL, clear_diag_counters);
2137 
2138 static struct attribute *diag_rprt_attrs[] = {
2139 	&dev_attr_rq_num_lle.attr,
2140 	&dev_attr_sq_num_lle.attr,
2141 	&dev_attr_rq_num_lqpoe.attr,
2142 	&dev_attr_sq_num_lqpoe.attr,
2143 	&dev_attr_rq_num_lpe.attr,
2144 	&dev_attr_sq_num_lpe.attr,
2145 	&dev_attr_rq_num_wrfe.attr,
2146 	&dev_attr_sq_num_wrfe.attr,
2147 	&dev_attr_sq_num_mwbe.attr,
2148 	&dev_attr_sq_num_bre.attr,
2149 	&dev_attr_rq_num_lae.attr,
2150 	&dev_attr_sq_num_rire.attr,
2151 	&dev_attr_rq_num_rire.attr,
2152 	&dev_attr_sq_num_rae.attr,
2153 	&dev_attr_rq_num_rae.attr,
2154 	&dev_attr_sq_num_roe.attr,
2155 	&dev_attr_sq_num_tree.attr,
2156 	&dev_attr_sq_num_rree.attr,
2157 	&dev_attr_rq_num_rnr.attr,
2158 	&dev_attr_sq_num_rnr.attr,
2159 	&dev_attr_rq_num_oos.attr,
2160 	&dev_attr_sq_num_oos.attr,
2161 	&dev_attr_rq_num_mce.attr,
2162 	&dev_attr_rq_num_udsdprd.attr,
2163 	&dev_attr_rq_num_ucsdprd.attr,
2164 	&dev_attr_num_cqovf.attr,
2165 	&dev_attr_num_eqovf.attr,
2166 	&dev_attr_num_baddb.attr,
2167 	&dev_attr_clear_diag.attr,
2168 	NULL
2169 };
2170 
2171 static struct attribute_group diag_counters_group = {
2172 	.name  = "diag_counters",
2173 	.attrs  = diag_rprt_attrs
2174 };
2175 
2176 static void init_dev_assign(void)
2177 {
2178 	int i = 1;
2179 
2180 	spin_lock_init(&dev_num_str_lock);
2181 	if (mlx4_fill_dbdf2val_tbl(&dev_assign_str))
2182 		return;
2183 	dev_num_str_bitmap =
2184 		kmalloc(BITS_TO_LONGS(MAX_NUM_STR_BITMAP) * sizeof(long),
2185 			GFP_KERNEL);
2186 	if (!dev_num_str_bitmap) {
2187 		pr_warn("bitmap alloc failed -- cannot apply dev_assign_str parameter\n");
2188 		return;
2189 	}
2190 	bitmap_zero(dev_num_str_bitmap, MAX_NUM_STR_BITMAP);
2191 	while ((i < MLX4_DEVS_TBL_SIZE) && (dev_assign_str.tbl[i].dbdf !=
2192 	       MLX4_ENDOF_TBL)) {
2193 		if (bitmap_allocate_region(dev_num_str_bitmap,
2194 					   dev_assign_str.tbl[i].val[0], 0))
2195 			goto err;
2196 		i++;
2197 	}
2198 	dr_active = 1;
2199 	return;
2200 
2201 err:
2202 	kfree(dev_num_str_bitmap);
2203 	dev_num_str_bitmap = NULL;
2204 	pr_warn("mlx4_ib: The value of 'dev_assign_str' parameter "
2205 			    "is incorrect. The parameter value is discarded!");
2206 }
2207 
2208 static int mlx4_ib_dev_idx(struct mlx4_dev *dev)
2209 {
2210 	int i, val;
2211 
2212 	if (!dr_active)
2213 		return -1;
2214 	if (!dev)
2215 		return -1;
2216 	if (mlx4_get_val(dev_assign_str.tbl, dev->pdev, 0, &val))
2217 		return -1;
2218 
2219 	if (val != DEFAULT_TBL_VAL) {
2220 		dev->flags |= MLX4_FLAG_DEV_NUM_STR;
2221 		return val;
2222 	}
2223 
2224 	spin_lock(&dev_num_str_lock);
2225 	i = bitmap_find_free_region(dev_num_str_bitmap, MAX_NUM_STR_BITMAP, 0);
2226 	spin_unlock(&dev_num_str_lock);
2227 	if (i >= 0)
2228 		return i;
2229 
2230 	return -1;
2231 }
2232 
2233 static void *mlx4_ib_add(struct mlx4_dev *dev)
2234 {
2235 	struct mlx4_ib_dev *ibdev;
2236 	int num_ports = 0;
2237 	int i, j;
2238 	int err;
2239 	struct mlx4_ib_iboe *iboe;
2240 	int dev_idx;
2241 
2242         pr_info_once("%s", mlx4_ib_version);
2243 
2244 	mlx4_foreach_ib_transport_port(i, dev)
2245 		num_ports++;
2246 
2247 	/* No point in registering a device with no ports... */
2248 	if (num_ports == 0)
2249 		return NULL;
2250 
2251 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2252 	if (!ibdev) {
2253 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
2254 		return NULL;
2255 	}
2256 
2257 	iboe = &ibdev->iboe;
2258 
2259 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2260 		goto err_dealloc;
2261 
2262 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2263 		goto err_pd;
2264 
2265 	ibdev->priv_uar.map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT,
2266 		PAGE_SIZE);
2267 
2268 	if (!ibdev->priv_uar.map)
2269 		goto err_uar;
2270 
2271 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2272 
2273 	ibdev->dev = dev;
2274 
2275 	dev_idx = mlx4_ib_dev_idx(dev);
2276 	if (dev_idx >= 0)
2277 		sprintf(ibdev->ib_dev.name, "mlx4_%d", dev_idx);
2278 	else
2279 		strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2280 
2281 	ibdev->ib_dev.owner		= THIS_MODULE;
2282 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
2283 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
2284 	ibdev->num_ports		= num_ports;
2285 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
2286 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
2287 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
2288 
2289 	if (dev->caps.userspace_caps)
2290 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2291 	else
2292 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2293 
2294 	ibdev->ib_dev.uverbs_cmd_mask	=
2295 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
2296 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
2297 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
2298 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2299 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2300 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2301 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2302 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2303 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2304 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2305 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2306 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2307 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2308 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2309 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2310 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2311 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2312 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2313 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2314 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2315 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2316 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2317 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2318 
2319 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2320 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2321 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2322 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2323 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2324 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2325 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2326 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2327 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2328 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2329 /* XXX FBSD has no support for get_unmapped_area function */
2330 #if 0
2331 	ibdev->ib_dev.get_unmapped_area = mlx4_ib_get_unmapped_area;
2332 #endif
2333 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2334 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2335 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2336 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2337 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2338 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2339 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2340 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2341 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2342 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2343 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2344 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2345 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2346 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2347 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2348 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2349 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2350 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2351 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2352 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2353 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2354 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2355 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2356 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2357 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2358 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2359 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2360 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2361 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2362 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2363 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2364 	ibdev->ib_dev.get_netdev	= mlx4_ib_get_netdev;
2365 	ibdev->ib_dev.ioctl		= mlx4_ib_ioctl;
2366 	ibdev->ib_dev.query_values	= mlx4_ib_query_values;
2367 
2368 	if (!mlx4_is_slave(ibdev->dev)) {
2369 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2370 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2371 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2372 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2373 	}
2374 
2375 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW) {
2376 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2377 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2378 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2379 
2380 		ibdev->ib_dev.uverbs_cmd_mask |=
2381 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2382 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2383 	}
2384 
2385 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2386 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2387 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2388 		ibdev->ib_dev.uverbs_cmd_mask |=
2389 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2390 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2391 	}
2392 
2393 	/*
2394 	 * Set experimental data
2395 	 */
2396 	ibdev->ib_dev.uverbs_exp_cmd_mask	=
2397 		(1ull << IB_USER_VERBS_EXP_CMD_CREATE_QP)	|
2398 		(1ull << IB_USER_VERBS_EXP_CMD_MODIFY_CQ)	|
2399 		(1ull << IB_USER_VERBS_EXP_CMD_QUERY_DEVICE)	|
2400 		(1ull << IB_USER_VERBS_EXP_CMD_CREATE_CQ);
2401 	ibdev->ib_dev.exp_create_qp	= mlx4_ib_exp_create_qp;
2402 	ibdev->ib_dev.exp_query_device	= mlx4_ib_exp_query_device;
2403 	if (check_flow_steering_support(dev)) {
2404 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2405 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2406 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2407 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2408 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2409 	} else {
2410 		pr_debug("Device managed flow steering is unavailable for this configuration.\n");
2411 	}
2412 	/*
2413 	 * End of experimental data
2414 	 */
2415 
2416 	mlx4_ib_alloc_eqs(dev, ibdev);
2417 
2418 	spin_lock_init(&iboe->lock);
2419 
2420 	if (init_node_data(ibdev))
2421 		goto err_map;
2422 
2423 	for (i = 0; i < ibdev->num_ports; ++i) {
2424 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2425 						IB_LINK_LAYER_ETHERNET) {
2426 			if (mlx4_is_slave(dev)) {
2427 				ibdev->counters[i].status = mlx4_counter_alloc(ibdev->dev,
2428 									       i + 1,
2429 									       &ibdev->counters[i].counter_index);
2430 			} else {/* allocating the PF IB default counter indices reserved in mlx4_init_counters_table */
2431 				ibdev->counters[i].counter_index = ((i + 1) << 1) - 1;
2432 				ibdev->counters[i].status = 0;
2433 			}
2434 
2435 			dev_info(&dev->pdev->dev,
2436 				 "%s: allocated counter index %d for port %d\n",
2437 				 __func__, ibdev->counters[i].counter_index, i+1);
2438 		} else {
2439 			ibdev->counters[i].counter_index = MLX4_SINK_COUNTER_INDEX;
2440 			ibdev->counters[i].status = -ENOSPC;
2441 		}
2442 	}
2443 
2444 	spin_lock_init(&ibdev->sm_lock);
2445 	mutex_init(&ibdev->cap_mask_mutex);
2446 
2447 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2448 	    !mlx4_is_mfunc(dev)) {
2449 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2450 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2451 					    MLX4_IB_UC_STEER_QPN_ALIGN, &ibdev->steer_qpn_base, 0);
2452 		if (err)
2453 			goto err_counter;
2454 
2455 		ibdev->ib_uc_qpns_bitmap =
2456 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2457 				sizeof(long),
2458 				GFP_KERNEL);
2459 		if (!ibdev->ib_uc_qpns_bitmap) {
2460 			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2461 			goto err_steer_qp_release;
2462 		}
2463 
2464 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2465 
2466 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(dev, ibdev->steer_qpn_base,
2467 				ibdev->steer_qpn_base + ibdev->steer_qpn_count - 1);
2468 		if (err)
2469 			goto err_steer_free_bitmap;
2470 	}
2471 
2472 	if (ib_register_device(&ibdev->ib_dev, NULL))
2473 		goto err_steer_free_bitmap;
2474 
2475 	if (mlx4_ib_mad_init(ibdev))
2476 		goto err_reg;
2477 
2478 	if (mlx4_ib_init_sriov(ibdev))
2479 		goto err_mad;
2480 
2481 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2482 		if (!iboe->nb.notifier_call) {
2483 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2484 			err = register_netdevice_notifier(&iboe->nb);
2485 			if (err) {
2486 				iboe->nb.notifier_call = NULL;
2487 				goto err_notify;
2488 			}
2489 		}
2490 		if (!iboe->nb_inet.notifier_call) {
2491 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2492 			err = register_inetaddr_notifier(&iboe->nb_inet);
2493 			if (err) {
2494 				iboe->nb_inet.notifier_call = NULL;
2495 				goto err_notify;
2496 			}
2497 		}
2498 		mlx4_ib_scan_netdevs(ibdev, NULL, 0);
2499 	}
2500 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2501 		if (device_create_file(&ibdev->ib_dev.dev,
2502 				       mlx4_class_attributes[j]))
2503 			goto err_notify;
2504 	}
2505 	if (sysfs_create_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group))
2506 		goto err_notify;
2507 
2508 	ibdev->ib_active = true;
2509 
2510 	if (mlx4_is_mfunc(ibdev->dev))
2511 		init_pkeys(ibdev);
2512 
2513 	/* create paravirt contexts for any VFs which are active */
2514 	if (mlx4_is_master(ibdev->dev)) {
2515 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2516 			if (j == mlx4_master_func_num(ibdev->dev))
2517 				continue;
2518 			if (mlx4_is_slave_active(ibdev->dev, j))
2519 				do_slave_init(ibdev, j, 1);
2520 		}
2521 	}
2522 	return ibdev;
2523 
2524 err_notify:
2525 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2526                 device_remove_file(&ibdev->ib_dev.dev,
2527                         mlx4_class_attributes[j]);
2528         }
2529 
2530 	if (ibdev->iboe.nb.notifier_call) {
2531 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2532 			pr_warn("failure unregistering notifier\n");
2533 		ibdev->iboe.nb.notifier_call = NULL;
2534 	}
2535 	if (ibdev->iboe.nb_inet.notifier_call) {
2536 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2537 			pr_warn("failure unregistering notifier\n");
2538 		ibdev->iboe.nb_inet.notifier_call = NULL;
2539 	}
2540 	flush_workqueue(wq);
2541 
2542 	mlx4_ib_close_sriov(ibdev);
2543 
2544 err_mad:
2545 	mlx4_ib_mad_cleanup(ibdev);
2546 
2547 err_reg:
2548 	ib_unregister_device(&ibdev->ib_dev);
2549 
2550 err_steer_free_bitmap:
2551 	kfree(ibdev->ib_uc_qpns_bitmap);
2552 
2553 err_steer_qp_release:
2554 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED)
2555 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2556 				ibdev->steer_qpn_count);
2557 err_counter:
2558 	for (; i; --i) {
2559 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i) ==
2560 						IB_LINK_LAYER_ETHERNET) {
2561 			mlx4_counter_free(ibdev->dev,
2562 					  i,
2563 					  ibdev->counters[i - 1].counter_index);
2564 		}
2565 	}
2566 
2567 err_map:
2568 	iounmap(ibdev->priv_uar.map);
2569 	mlx4_ib_free_eqs(dev, ibdev);
2570 
2571 err_uar:
2572 	mlx4_uar_free(dev, &ibdev->priv_uar);
2573 
2574 err_pd:
2575 	mlx4_pd_free(dev, ibdev->priv_pdn);
2576 
2577 err_dealloc:
2578 	ib_dealloc_device(&ibdev->ib_dev);
2579 
2580 	return NULL;
2581 }
2582 
2583 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2584 {
2585 	int offset;
2586 
2587 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2588 
2589 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2590 					 dev->steer_qpn_count,
2591 					 get_count_order(count));
2592 	if (offset < 0)
2593 		return offset;
2594 
2595 	*qpn = dev->steer_qpn_base + offset;
2596 	return 0;
2597 }
2598 
2599 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2600 {
2601 	if (!qpn ||
2602 	    dev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
2603 		return;
2604 
2605 	BUG_ON(qpn < dev->steer_qpn_base);
2606 
2607 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2608 			qpn - dev->steer_qpn_base, get_count_order(count));
2609 }
2610 
2611 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2612 			 int is_attach)
2613 {
2614 	int err;
2615 	size_t flow_size;
2616 	struct ib_flow_attr *flow = NULL;
2617 	struct ib_flow_spec_ib *ib_spec;
2618 
2619 	if (is_attach) {
2620 		flow_size = sizeof(struct ib_flow_attr) +
2621 			    sizeof(struct ib_flow_spec_ib);
2622 		flow = kzalloc(flow_size, GFP_KERNEL);
2623 		if (!flow)
2624 			return -ENOMEM;
2625 		flow->port = mqp->port;
2626 		flow->num_of_specs = 1;
2627 		flow->size = flow_size;
2628 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2629 		ib_spec->type = IB_FLOW_SPEC_IB;
2630 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2631 		ib_spec->val.l3_type_qpn = mqp->ibqp.qp_num;
2632 		ib_spec->mask.l3_type_qpn = MLX4_IB_FLOW_QPN_MASK;
2633 
2634 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2635 					    IB_FLOW_DOMAIN_NIC,
2636 					    MLX4_FS_REGULAR,
2637 					    &mqp->reg_id);
2638 	} else {
2639 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2640 	}
2641 	kfree(flow);
2642 	return err;
2643 }
2644 
2645 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2646 {
2647 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2648 	int p, j;
2649 	int dev_idx, ret;
2650 
2651 	if (ibdev->iboe.nb_inet.notifier_call) {
2652 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2653 			pr_warn("failure unregistering notifier\n");
2654 		ibdev->iboe.nb_inet.notifier_call = NULL;
2655 	}
2656 
2657 	mlx4_ib_close_sriov(ibdev);
2658 	sysfs_remove_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group);
2659 	mlx4_ib_mad_cleanup(ibdev);
2660 
2661 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2662 		device_remove_file(&ibdev->ib_dev.dev,
2663 			mlx4_class_attributes[j]);
2664 	}
2665 
2666 
2667 	dev_idx = -1;
2668 	if (dr_active && !(ibdev->dev->flags & MLX4_FLAG_DEV_NUM_STR)) {
2669 		ret = sscanf(ibdev->ib_dev.name, "mlx4_%d", &dev_idx);
2670 		if (ret != 1)
2671 			dev_idx = -1;
2672 	}
2673 	ib_unregister_device(&ibdev->ib_dev);
2674 	if (dev_idx >= 0) {
2675 		spin_lock(&dev_num_str_lock);
2676 		bitmap_release_region(dev_num_str_bitmap, dev_idx, 0);
2677 		spin_unlock(&dev_num_str_lock);
2678 	}
2679 
2680 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2681 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2682 				ibdev->steer_qpn_count);
2683 		kfree(ibdev->ib_uc_qpns_bitmap);
2684 	}
2685 
2686 	if (ibdev->iboe.nb.notifier_call) {
2687 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2688 			pr_warn("failure unregistering notifier\n");
2689 		ibdev->iboe.nb.notifier_call = NULL;
2690 	}
2691 	iounmap(ibdev->priv_uar.map);
2692 
2693 	for (p = 0; p < ibdev->num_ports; ++p) {
2694 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, p + 1) ==
2695 						IB_LINK_LAYER_ETHERNET) {
2696 			mlx4_counter_free(ibdev->dev,
2697 					  p + 1,
2698 					  ibdev->counters[p].counter_index);
2699 		}
2700 	}
2701 
2702 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2703 		mlx4_CLOSE_PORT(dev, p);
2704 
2705 	mlx4_ib_free_eqs(dev, ibdev);
2706 
2707 	mlx4_uar_free(dev, &ibdev->priv_uar);
2708 	mlx4_pd_free(dev, ibdev->priv_pdn);
2709 	ib_dealloc_device(&ibdev->ib_dev);
2710 }
2711 
2712 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2713 {
2714 	struct mlx4_ib_demux_work **dm = NULL;
2715 	struct mlx4_dev *dev = ibdev->dev;
2716 	int i;
2717 	unsigned long flags;
2718 
2719 	if (!mlx4_is_master(dev))
2720 		return;
2721 
2722 	dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2723 	if (!dm) {
2724 		pr_err("failed to allocate memory for tunneling qp update\n");
2725 		goto out;
2726 	}
2727 
2728 	for (i = 0; i < dev->caps.num_ports; i++) {
2729 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2730 		if (!dm[i]) {
2731 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2732 			for (i = 0; i < dev->caps.num_ports; i++) {
2733 				if (dm[i])
2734 					kfree(dm[i]);
2735 			}
2736 			goto out;
2737 		}
2738 	}
2739 	/* initialize or tear down tunnel QPs for the slave */
2740 	for (i = 0; i < dev->caps.num_ports; i++) {
2741 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2742 		dm[i]->port = i + 1;
2743 		dm[i]->slave = slave;
2744 		dm[i]->do_init = do_init;
2745 		dm[i]->dev = ibdev;
2746 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2747 		if (!ibdev->sriov.is_going_down)
2748 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2749 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2750 	}
2751 out:
2752 	if (dm)
2753 		kfree(dm);
2754 	return;
2755 }
2756 
2757 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2758 			  enum mlx4_dev_event event, unsigned long param)
2759 {
2760 	struct ib_event ibev;
2761 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2762 	struct mlx4_eqe *eqe = NULL;
2763 	struct ib_event_work *ew;
2764 	int p = 0;
2765 
2766 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2767 		eqe = (struct mlx4_eqe *)param;
2768 	else
2769 		p = (int) param;
2770 
2771 	switch (event) {
2772 	case MLX4_DEV_EVENT_PORT_UP:
2773 		if (p > ibdev->num_ports)
2774 			return;
2775 		if (mlx4_is_master(dev) &&
2776 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2777 			IB_LINK_LAYER_INFINIBAND) {
2778 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2779 		}
2780 		mlx4_ib_info((struct ib_device *) ibdev_ptr,
2781 			     "Port %d logical link is up\n", p);
2782 		ibev.event = IB_EVENT_PORT_ACTIVE;
2783 		break;
2784 
2785 	case MLX4_DEV_EVENT_PORT_DOWN:
2786 		if (p > ibdev->num_ports)
2787 			return;
2788 		mlx4_ib_info((struct ib_device *) ibdev_ptr,
2789 			     "Port %d logical link is down\n", p);
2790 		ibev.event = IB_EVENT_PORT_ERR;
2791 		break;
2792 
2793 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2794 		ibdev->ib_active = false;
2795 		ibev.event = IB_EVENT_DEVICE_FATAL;
2796 		break;
2797 
2798 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2799 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2800 		if (!ew) {
2801 			pr_err("failed to allocate memory for events work\n");
2802 			break;
2803 		}
2804 
2805 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2806 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2807 		ew->ib_dev = ibdev;
2808 		/* need to queue only for port owner, which uses GEN_EQE */
2809 		if (mlx4_is_master(dev))
2810 			queue_work(wq, &ew->work);
2811 		else
2812 			handle_port_mgmt_change_event(&ew->work);
2813 		return;
2814 
2815 	case MLX4_DEV_EVENT_SLAVE_INIT:
2816 		/* here, p is the slave id */
2817 		do_slave_init(ibdev, p, 1);
2818 		return;
2819 
2820 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2821 		/* here, p is the slave id */
2822 		do_slave_init(ibdev, p, 0);
2823 		return;
2824 
2825 	default:
2826 		return;
2827 	}
2828 
2829 	ibev.device	      = ibdev_ptr;
2830 	ibev.element.port_num = (u8) p;
2831 
2832 	ib_dispatch_event(&ibev);
2833 }
2834 
2835 static struct mlx4_interface mlx4_ib_interface = {
2836 	.add		= mlx4_ib_add,
2837 	.remove		= mlx4_ib_remove,
2838 	.event		= mlx4_ib_event,
2839 	.protocol	= MLX4_PROT_IB_IPV6
2840 };
2841 
2842 static int __init mlx4_ib_init(void)
2843 {
2844 	int err;
2845 
2846 	wq = create_singlethread_workqueue("mlx4_ib");
2847 	if (!wq)
2848 		return -ENOMEM;
2849 
2850 	err = mlx4_ib_mcg_init();
2851 	if (err)
2852 		goto clean_proc;
2853 
2854 	init_dev_assign();
2855 
2856 	err = mlx4_register_interface(&mlx4_ib_interface);
2857 	if (err)
2858 		goto clean_mcg;
2859 
2860 	return 0;
2861 
2862 clean_mcg:
2863 	mlx4_ib_mcg_destroy();
2864 
2865 clean_proc:
2866 	destroy_workqueue(wq);
2867 	return err;
2868 }
2869 
2870 static void __exit mlx4_ib_cleanup(void)
2871 {
2872 	mlx4_unregister_interface(&mlx4_ib_interface);
2873 	mlx4_ib_mcg_destroy();
2874 	destroy_workqueue(wq);
2875 
2876 	kfree(dev_num_str_bitmap);
2877 }
2878 
2879 module_init_order(mlx4_ib_init, SI_ORDER_MIDDLE);
2880 module_exit(mlx4_ib_cleanup);
2881 
2882 static int
2883 mlx4ib_evhand(module_t mod, int event, void *arg)
2884 {
2885 	return (0);
2886 }
2887 
2888 static moduledata_t mlx4ib_mod = {
2889 	.name = "mlx4ib",
2890 	.evhand = mlx4ib_evhand,
2891 };
2892 
2893 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_LAST, SI_ORDER_ANY);
2894 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
2895 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);
2896 MODULE_DEPEND(mlx4ib, linuxkpi, 1, 1, 1);
2897