xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision ce932d0c5589e9766e089c22c66890dfc48fbd94)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
46 
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
49 
50 #include "mlx4_ib.h"
51 #include "user.h"
52 
53 #define DRV_NAME	"mlx4_ib"
54 #define DRV_VERSION	"1.0"
55 #define DRV_RELDATE	"April 4, 2008"
56 
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61 
62 static const char mlx4_ib_version[] =
63 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
64 	DRV_VERSION " (" DRV_RELDATE ")\n";
65 
66 struct update_gid_work {
67 	struct work_struct	work;
68 	union ib_gid		gids[128];
69 	struct mlx4_ib_dev     *dev;
70 	int			port;
71 };
72 
73 static struct workqueue_struct *wq;
74 
75 static void init_query_mad(struct ib_smp *mad)
76 {
77 	mad->base_version  = 1;
78 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
79 	mad->class_version = 1;
80 	mad->method	   = IB_MGMT_METHOD_GET;
81 }
82 
83 static union ib_gid zgid;
84 
85 static int mlx4_ib_query_device(struct ib_device *ibdev,
86 				struct ib_device_attr *props)
87 {
88 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
89 	struct ib_smp *in_mad  = NULL;
90 	struct ib_smp *out_mad = NULL;
91 	int err = -ENOMEM;
92 
93 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
94 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
95 	if (!in_mad || !out_mad)
96 		goto out;
97 
98 	init_query_mad(in_mad);
99 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
100 
101 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
102 	if (err)
103 		goto out;
104 
105 	memset(props, 0, sizeof *props);
106 
107 	props->fw_ver = dev->dev->caps.fw_ver;
108 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
109 		IB_DEVICE_PORT_ACTIVE_EVENT		|
110 		IB_DEVICE_SYS_IMAGE_GUID		|
111 		IB_DEVICE_RC_RNR_NAK_GEN		|
112 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
113 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
114 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
115 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
116 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
117 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
118 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
119 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
120 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
121 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
122 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
123 	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
124 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
125 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
126 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
127 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
128 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
131 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
132 		props->device_cap_flags |= IB_DEVICE_XRC;
133 
134 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
135 		0xffffff;
136 	props->vendor_part_id	   = be16_to_cpup((__be16 *) (out_mad->data + 30));
137 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
138 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
139 
140 	props->max_mr_size	   = ~0ull;
141 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
142 	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
143 	props->max_qp_wr	   = dev->dev->caps.max_wqes;
144 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
145 					 dev->dev->caps.max_rq_sg);
146 	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
147 	props->max_cqe		   = dev->dev->caps.max_cqes;
148 	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
149 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
150 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
151 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
152 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
153 	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
154 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
155 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
156 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
157 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
158 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
159 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
160 	props->masked_atomic_cap   = IB_ATOMIC_HCA;
161 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
162 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
163 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
164 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
165 					   props->max_mcast_grp;
166 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
167 
168 out:
169 	kfree(in_mad);
170 	kfree(out_mad);
171 
172 	return err;
173 }
174 
175 static enum rdma_link_layer
176 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
177 {
178 	struct mlx4_dev *dev = to_mdev(device)->dev;
179 
180 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
181 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
182 }
183 
184 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
185 			      struct ib_port_attr *props)
186 {
187 	struct ib_smp *in_mad  = NULL;
188 	struct ib_smp *out_mad = NULL;
189 	int ext_active_speed;
190 	int err = -ENOMEM;
191 
192 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
193 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
194 	if (!in_mad || !out_mad)
195 		goto out;
196 
197 	init_query_mad(in_mad);
198 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
199 	in_mad->attr_mod = cpu_to_be32(port);
200 
201 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL,
202 				in_mad, out_mad);
203 	if (err)
204 		goto out;
205 
206 
207 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
208 	props->lmc		= out_mad->data[34] & 0x7;
209 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
210 	props->sm_sl		= out_mad->data[36] & 0xf;
211 	props->state		= out_mad->data[32] & 0xf;
212 	props->phys_state	= out_mad->data[33] >> 4;
213 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
214 	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
215 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
216 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
217 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
218 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
219 	props->active_width	= out_mad->data[31] & 0xf;
220 	props->active_speed	= out_mad->data[35] >> 4;
221 	props->max_mtu		= out_mad->data[41] & 0xf;
222 	props->active_mtu	= out_mad->data[36] >> 4;
223 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
224 	props->max_vl_num	= out_mad->data[37] >> 4;
225 	props->init_type_reply	= out_mad->data[41] >> 4;
226 
227 	/* Check if extended speeds (EDR/FDR/...) are supported */
228 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
229 		ext_active_speed = out_mad->data[62] >> 4;
230 
231 		switch (ext_active_speed) {
232 		case 1:
233 			props->active_speed = IB_SPEED_FDR;
234 			break;
235 		case 2:
236 			props->active_speed = IB_SPEED_EDR;
237 			break;
238 		}
239 	}
240 
241 	/* If reported active speed is QDR, check if is FDR-10 */
242 	if (props->active_speed == IB_SPEED_QDR) {
243 		init_query_mad(in_mad);
244 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
245 		in_mad->attr_mod = cpu_to_be32(port);
246 
247 		err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
248 				   NULL, NULL, in_mad, out_mad);
249 		if (err)
250 			goto out;
251 
252 		/* Checking LinkSpeedActive for FDR-10 */
253 		if (out_mad->data[15] & 0x1)
254 			props->active_speed = IB_SPEED_FDR10;
255 	}
256 
257 	/* Avoid wrong speed value returned by FW if the IB link is down. */
258 	if (props->state == IB_PORT_DOWN)
259 		 props->active_speed = IB_SPEED_SDR;
260 
261 out:
262 	kfree(in_mad);
263 	kfree(out_mad);
264 	return err;
265 }
266 
267 static u8 state_to_phys_state(enum ib_port_state state)
268 {
269 	return state == IB_PORT_ACTIVE ? 5 : 3;
270 }
271 
272 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
273 			       struct ib_port_attr *props)
274 {
275 
276 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
277 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
278 	struct net_device *ndev;
279 	enum ib_mtu tmp;
280 	struct mlx4_cmd_mailbox *mailbox;
281 	int err = 0;
282 
283 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
284 	if (IS_ERR(mailbox))
285 		return PTR_ERR(mailbox);
286 
287 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
288 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
289 			   MLX4_CMD_WRAPPED);
290 	if (err)
291 		goto out;
292 
293 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
294 						IB_WIDTH_4X : IB_WIDTH_1X;
295 	props->active_speed	= IB_SPEED_QDR;
296 	props->port_cap_flags	= IB_PORT_CM_SUP;
297 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
298 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
299 	props->pkey_tbl_len	= 1;
300 	props->max_mtu		= IB_MTU_4096;
301 	props->max_vl_num	= 2;
302 	props->state		= IB_PORT_DOWN;
303 	props->phys_state	= state_to_phys_state(props->state);
304 	props->active_mtu	= IB_MTU_256;
305 	spin_lock(&iboe->lock);
306 	ndev = iboe->netdevs[port - 1];
307 	if (!ndev)
308 		goto out_unlock;
309 
310 	tmp = iboe_get_mtu(ndev->mtu);
311 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
312 
313 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
314 					IB_PORT_ACTIVE : IB_PORT_DOWN;
315 	props->phys_state	= state_to_phys_state(props->state);
316 out_unlock:
317 	spin_unlock(&iboe->lock);
318 out:
319 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
320 	return err;
321 }
322 
323 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
324 			      struct ib_port_attr *props)
325 {
326 	int err;
327 
328 	memset(props, 0, sizeof *props);
329 
330 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
331 		ib_link_query_port(ibdev, port, props) :
332 				eth_link_query_port(ibdev, port, props);
333 
334 	return err;
335 }
336 
337 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
338 			       union ib_gid *gid)
339 {
340 	struct ib_smp *in_mad  = NULL;
341 	struct ib_smp *out_mad = NULL;
342 	int err = -ENOMEM;
343 
344 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
345 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
346 	if (!in_mad || !out_mad)
347 		goto out;
348 
349 	init_query_mad(in_mad);
350 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
351 	in_mad->attr_mod = cpu_to_be32(port);
352 
353 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
354 	if (err)
355 		goto out;
356 
357 	memcpy(gid->raw, out_mad->data + 8, 8);
358 
359 	init_query_mad(in_mad);
360 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
361 	in_mad->attr_mod = cpu_to_be32(index / 8);
362 
363 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
364 	if (err)
365 		goto out;
366 
367 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
368 
369 out:
370 	kfree(in_mad);
371 	kfree(out_mad);
372 	return err;
373 }
374 
375 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
376 			  union ib_gid *gid)
377 {
378 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
379 
380 	*gid = dev->iboe.gid_table[port - 1][index];
381 
382 	return 0;
383 }
384 
385 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
386 			     union ib_gid *gid)
387 {
388 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
389 		return __mlx4_ib_query_gid(ibdev, port, index, gid);
390 	else
391 		return iboe_query_gid(ibdev, port, index, gid);
392 }
393 
394 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
395 			      u16 *pkey)
396 {
397 	struct ib_smp *in_mad  = NULL;
398 	struct ib_smp *out_mad = NULL;
399 	int err = -ENOMEM;
400 
401 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
402 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
403 	if (!in_mad || !out_mad)
404 		goto out;
405 
406 	init_query_mad(in_mad);
407 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
408 	in_mad->attr_mod = cpu_to_be32(index / 32);
409 
410 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
411 	if (err)
412 		goto out;
413 
414 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
415 
416 out:
417 	kfree(in_mad);
418 	kfree(out_mad);
419 	return err;
420 }
421 
422 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
423 				 struct ib_device_modify *props)
424 {
425 	struct mlx4_cmd_mailbox *mailbox;
426 
427 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
428 		return -EOPNOTSUPP;
429 
430 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
431 		return 0;
432 
433 	spin_lock(&to_mdev(ibdev)->sm_lock);
434 	memcpy(ibdev->node_desc, props->node_desc, 64);
435 	spin_unlock(&to_mdev(ibdev)->sm_lock);
436 
437 	/*
438 	 * If possible, pass node desc to FW, so it can generate
439 	 * a 144 trap.  If cmd fails, just ignore.
440 	 */
441 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
442 	if (IS_ERR(mailbox))
443 		return 0;
444 
445 	memset(mailbox->buf, 0, 256);
446 	memcpy(mailbox->buf, props->node_desc, 64);
447 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
448 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
449 
450 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
451 
452 	return 0;
453 }
454 
455 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
456 			 u32 cap_mask)
457 {
458 	struct mlx4_cmd_mailbox *mailbox;
459 	int err;
460 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
461 
462 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
463 	if (IS_ERR(mailbox))
464 		return PTR_ERR(mailbox);
465 
466 	memset(mailbox->buf, 0, 256);
467 
468 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
469 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
470 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
471 	} else {
472 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
473 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
474 	}
475 
476 	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
477 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
478 
479 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
480 	return err;
481 }
482 
483 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
484 			       struct ib_port_modify *props)
485 {
486 	struct ib_port_attr attr;
487 	u32 cap_mask;
488 	int err;
489 
490 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
491 
492 	err = mlx4_ib_query_port(ibdev, port, &attr);
493 	if (err)
494 		goto out;
495 
496 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
497 		~props->clr_port_cap_mask;
498 
499 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
500 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
501 			    cap_mask);
502 
503 out:
504 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
505 	return err;
506 }
507 
508 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
509 						  struct ib_udata *udata)
510 {
511 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
512 	struct mlx4_ib_ucontext *context;
513 	struct mlx4_ib_alloc_ucontext_resp resp;
514 	int err;
515 
516 	if (!dev->ib_active)
517 		return ERR_PTR(-EAGAIN);
518 
519 	resp.qp_tab_size      = dev->dev->caps.num_qps;
520 	resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
521 	resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
522 
523 	context = kmalloc(sizeof *context, GFP_KERNEL);
524 	if (!context)
525 		return ERR_PTR(-ENOMEM);
526 
527 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
528 	if (err) {
529 		kfree(context);
530 		return ERR_PTR(err);
531 	}
532 
533 	INIT_LIST_HEAD(&context->db_page_list);
534 	mutex_init(&context->db_page_mutex);
535 
536 	err = ib_copy_to_udata(udata, &resp, sizeof resp);
537 	if (err) {
538 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
539 		kfree(context);
540 		return ERR_PTR(-EFAULT);
541 	}
542 
543 	return &context->ibucontext;
544 }
545 
546 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
547 {
548 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
549 
550 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
551 	kfree(context);
552 
553 	return 0;
554 }
555 
556 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
557 {
558 	struct mlx4_ib_dev *dev = to_mdev(context->device);
559 
560 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
561 		return -EINVAL;
562 
563 	if (vma->vm_pgoff == 0) {
564 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
565 
566 		if (io_remap_pfn_range(vma, vma->vm_start,
567 				       to_mucontext(context)->uar.pfn,
568 				       PAGE_SIZE, vma->vm_page_prot))
569 			return -EAGAIN;
570 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
571 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
572 
573 		if (io_remap_pfn_range(vma, vma->vm_start,
574 				       to_mucontext(context)->uar.pfn +
575 				       dev->dev->caps.num_uars,
576 				       PAGE_SIZE, vma->vm_page_prot))
577 			return -EAGAIN;
578 	} else
579 		return -EINVAL;
580 
581 	return 0;
582 }
583 
584 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
585 				      struct ib_ucontext *context,
586 				      struct ib_udata *udata)
587 {
588 	struct mlx4_ib_pd *pd;
589 	int err;
590 
591 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
592 	if (!pd)
593 		return ERR_PTR(-ENOMEM);
594 
595 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
596 	if (err) {
597 		kfree(pd);
598 		return ERR_PTR(err);
599 	}
600 
601 	if (context)
602 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
603 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
604 			kfree(pd);
605 			return ERR_PTR(-EFAULT);
606 		}
607 
608 	return &pd->ibpd;
609 }
610 
611 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
612 {
613 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
614 	kfree(pd);
615 
616 	return 0;
617 }
618 
619 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
620 					  struct ib_ucontext *context,
621 					  struct ib_udata *udata)
622 {
623 	struct mlx4_ib_xrcd *xrcd;
624 	int err;
625 
626 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
627 		return ERR_PTR(-ENOSYS);
628 
629 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
630 	if (!xrcd)
631 		return ERR_PTR(-ENOMEM);
632 
633 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
634 	if (err)
635 		goto err1;
636 
637 	xrcd->pd = ib_alloc_pd(ibdev);
638 	if (IS_ERR(xrcd->pd)) {
639 		err = PTR_ERR(xrcd->pd);
640 		goto err2;
641 	}
642 
643 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
644 	if (IS_ERR(xrcd->cq)) {
645 		err = PTR_ERR(xrcd->cq);
646 		goto err3;
647 	}
648 
649 	return &xrcd->ibxrcd;
650 
651 err3:
652 	ib_dealloc_pd(xrcd->pd);
653 err2:
654 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
655 err1:
656 	kfree(xrcd);
657 	return ERR_PTR(err);
658 }
659 
660 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
661 {
662 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
663 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
664 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
665 	kfree(xrcd);
666 
667 	return 0;
668 }
669 
670 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
671 {
672 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
673 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
674 	struct mlx4_ib_gid_entry *ge;
675 
676 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
677 	if (!ge)
678 		return -ENOMEM;
679 
680 	ge->gid = *gid;
681 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
682 		ge->port = mqp->port;
683 		ge->added = 1;
684 	}
685 
686 	mutex_lock(&mqp->mutex);
687 	list_add_tail(&ge->list, &mqp->gid_list);
688 	mutex_unlock(&mqp->mutex);
689 
690 	return 0;
691 }
692 
693 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
694 		   union ib_gid *gid)
695 {
696 	u8 mac[6];
697 	struct net_device *ndev;
698 	int ret = 0;
699 
700 	if (!mqp->port)
701 		return 0;
702 
703 	spin_lock(&mdev->iboe.lock);
704 	ndev = mdev->iboe.netdevs[mqp->port - 1];
705 	if (ndev)
706 		dev_hold(ndev);
707 	spin_unlock(&mdev->iboe.lock);
708 
709 	if (ndev) {
710 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
711 		rtnl_lock();
712 		dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
713 		ret = 1;
714 		rtnl_unlock();
715 		dev_put(ndev);
716 	}
717 
718 	return ret;
719 }
720 
721 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
722 {
723 	int err;
724 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
725 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
726 
727 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
728 				    !!(mqp->flags & MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
729 				    MLX4_PROT_IB_IPV6);
730 	if (err)
731 		return err;
732 
733 	err = add_gid_entry(ibqp, gid);
734 	if (err)
735 		goto err_add;
736 
737 	return 0;
738 
739 err_add:
740 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
741 	return err;
742 }
743 
744 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
745 {
746 	struct mlx4_ib_gid_entry *ge;
747 	struct mlx4_ib_gid_entry *tmp;
748 	struct mlx4_ib_gid_entry *ret = NULL;
749 
750 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
751 		if (!memcmp(raw, ge->gid.raw, 16)) {
752 			ret = ge;
753 			break;
754 		}
755 	}
756 
757 	return ret;
758 }
759 
760 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
761 {
762 	int err;
763 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
764 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
765 	u8 mac[6];
766 	struct net_device *ndev;
767 	struct mlx4_ib_gid_entry *ge;
768 
769 	err = mlx4_multicast_detach(mdev->dev,
770 				    &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
771 	if (err)
772 		return err;
773 
774 	mutex_lock(&mqp->mutex);
775 	ge = find_gid_entry(mqp, gid->raw);
776 	if (ge) {
777 		spin_lock(&mdev->iboe.lock);
778 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
779 		if (ndev)
780 			dev_hold(ndev);
781 		spin_unlock(&mdev->iboe.lock);
782 		rdma_get_mcast_mac((struct in6_addr *)gid, mac);
783 		if (ndev) {
784 			rtnl_lock();
785 			dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
786 			rtnl_unlock();
787 			dev_put(ndev);
788 		}
789 		list_del(&ge->list);
790 		kfree(ge);
791 	} else
792 		printk(KERN_WARNING "could not find mgid entry\n");
793 
794 	mutex_unlock(&mqp->mutex);
795 
796 	return 0;
797 }
798 
799 static int init_node_data(struct mlx4_ib_dev *dev)
800 {
801 	struct ib_smp *in_mad  = NULL;
802 	struct ib_smp *out_mad = NULL;
803 	int err = -ENOMEM;
804 
805 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
806 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
807 	if (!in_mad || !out_mad)
808 		goto out;
809 
810 	init_query_mad(in_mad);
811 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
812 
813 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
814 	if (err)
815 		goto out;
816 
817 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
818 
819 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
820 
821 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
822 	if (err)
823 		goto out;
824 
825 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
826 
827 out:
828 	kfree(in_mad);
829 	kfree(out_mad);
830 	return err;
831 }
832 
833 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
834 			char *buf)
835 {
836 	struct mlx4_ib_dev *dev =
837 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
838 	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
839 }
840 
841 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
842 			   char *buf)
843 {
844 	struct mlx4_ib_dev *dev =
845 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
846 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
847 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
848 		       (int) dev->dev->caps.fw_ver & 0xffff);
849 }
850 
851 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
852 			char *buf)
853 {
854 	struct mlx4_ib_dev *dev =
855 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
856 	return sprintf(buf, "%x\n", dev->dev->rev_id);
857 }
858 
859 static ssize_t show_board(struct device *device, struct device_attribute *attr,
860 			  char *buf)
861 {
862 	struct mlx4_ib_dev *dev =
863 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
864 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
865 		       dev->dev->board_id);
866 }
867 
868 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
869 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
870 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
871 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
872 
873 static struct device_attribute *mlx4_class_attributes[] = {
874 	&dev_attr_hw_rev,
875 	&dev_attr_fw_ver,
876 	&dev_attr_hca_type,
877 	&dev_attr_board_id
878 };
879 
880 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
881 {
882 	memcpy(eui, dev->dev_addr, 3);
883 	memcpy(eui + 5, dev->dev_addr + 3, 3);
884 	if (vlan_id < 0x1000) {
885 		eui[3] = vlan_id >> 8;
886 		eui[4] = vlan_id & 0xff;
887 	} else {
888 		eui[3] = 0xff;
889 		eui[4] = 0xfe;
890 	}
891 	eui[0] ^= 2;
892 }
893 
894 static void update_gids_task(struct work_struct *work)
895 {
896 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
897 	struct mlx4_cmd_mailbox *mailbox;
898 	union ib_gid *gids;
899 	int err;
900 	struct mlx4_dev	*dev = gw->dev->dev;
901 	struct ib_event event;
902 
903 	mailbox = mlx4_alloc_cmd_mailbox(dev);
904 	if (IS_ERR(mailbox)) {
905 		printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
906 		return;
907 	}
908 
909 	gids = mailbox->buf;
910 	memcpy(gids, gw->gids, sizeof gw->gids);
911 
912 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
913 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
914 		       MLX4_CMD_NATIVE);
915 	if (err)
916 		printk(KERN_WARNING "set port command failed\n");
917 	else {
918 		memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
919 		event.device = &gw->dev->ib_dev;
920 		event.element.port_num = gw->port;
921 		event.event    = IB_EVENT_GID_CHANGE;
922 		ib_dispatch_event(&event);
923 	}
924 
925 	mlx4_free_cmd_mailbox(dev, mailbox);
926 	kfree(gw);
927 }
928 
929 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
930 {
931 	struct net_device *ndev = dev->iboe.netdevs[port - 1];
932 	struct update_gid_work *work;
933 	struct net_device *tmp;
934 	int i;
935 	u8 *hits;
936 	int ret;
937 	union ib_gid gid;
938 	int free;
939 	int found;
940 	int need_update = 0;
941 	u16 vid;
942 
943 	work = kzalloc(sizeof *work, GFP_ATOMIC);
944 	if (!work)
945 		return -ENOMEM;
946 
947 	hits = kzalloc(128, GFP_ATOMIC);
948 	if (!hits) {
949 		ret = -ENOMEM;
950 		goto out;
951 	}
952 
953 	rcu_read_lock();
954 	for_each_netdev_rcu(&init_net, tmp) {
955 		if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
956 			gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
957 			vid = rdma_vlan_dev_vlan_id(tmp);
958 			mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
959 			found = 0;
960 			free = -1;
961 			for (i = 0; i < 128; ++i) {
962 				if (free < 0 &&
963 				    !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
964 					free = i;
965 				if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
966 					hits[i] = 1;
967 					found = 1;
968 					break;
969 				}
970 			}
971 
972 			if (!found) {
973 				if (tmp == ndev &&
974 				    (memcmp(&dev->iboe.gid_table[port - 1][0],
975 					    &gid, sizeof gid) ||
976 				     !memcmp(&dev->iboe.gid_table[port - 1][0],
977 					     &zgid, sizeof gid))) {
978 					dev->iboe.gid_table[port - 1][0] = gid;
979 					++need_update;
980 					hits[0] = 1;
981 				} else if (free >= 0) {
982 					dev->iboe.gid_table[port - 1][free] = gid;
983 					hits[free] = 1;
984 					++need_update;
985 				}
986 			}
987 		}
988 	}
989 	rcu_read_unlock();
990 
991 	for (i = 0; i < 128; ++i)
992 		if (!hits[i]) {
993 			if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
994 				++need_update;
995 			dev->iboe.gid_table[port - 1][i] = zgid;
996 		}
997 
998 	if (need_update) {
999 		memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1000 		INIT_WORK(&work->work, update_gids_task);
1001 		work->port = port;
1002 		work->dev = dev;
1003 		queue_work(wq, &work->work);
1004 	} else
1005 		kfree(work);
1006 
1007 	kfree(hits);
1008 	return 0;
1009 
1010 out:
1011 	kfree(work);
1012 	return ret;
1013 }
1014 
1015 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1016 {
1017 	switch (event) {
1018 	case NETDEV_UP:
1019 	case NETDEV_CHANGEADDR:
1020 		update_ipv6_gids(dev, port, 0);
1021 		break;
1022 
1023 	case NETDEV_DOWN:
1024 		update_ipv6_gids(dev, port, 1);
1025 		dev->iboe.netdevs[port - 1] = NULL;
1026 	}
1027 }
1028 
1029 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1030 {
1031 	update_ipv6_gids(dev, port, 0);
1032 }
1033 
1034 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1035 {
1036 	update_ipv6_gids(dev, port, 1);
1037 }
1038 
1039 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1040 				void *ptr)
1041 {
1042 	struct net_device *dev = ptr;
1043 	struct mlx4_ib_dev *ibdev;
1044 	struct net_device *oldnd;
1045 	struct mlx4_ib_iboe *iboe;
1046 	int port;
1047 
1048 	if (!net_eq(dev_net(dev), &init_net))
1049 		return NOTIFY_DONE;
1050 
1051 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1052 	iboe = &ibdev->iboe;
1053 
1054 	spin_lock(&iboe->lock);
1055 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1056 		oldnd = iboe->netdevs[port - 1];
1057 		iboe->netdevs[port - 1] =
1058 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1059 		if (oldnd != iboe->netdevs[port - 1]) {
1060 			if (iboe->netdevs[port - 1])
1061 				netdev_added(ibdev, port);
1062 			else
1063 				netdev_removed(ibdev, port);
1064 		}
1065 	}
1066 
1067 	if (dev == iboe->netdevs[0] ||
1068 	    (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1069 		handle_en_event(ibdev, 1, event);
1070 	else if (dev == iboe->netdevs[1]
1071 		 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1072 		handle_en_event(ibdev, 2, event);
1073 
1074 	spin_unlock(&iboe->lock);
1075 
1076 	return NOTIFY_DONE;
1077 }
1078 
1079 static void *mlx4_ib_add(struct mlx4_dev *dev)
1080 {
1081 	struct mlx4_ib_dev *ibdev;
1082 	int num_ports = 0;
1083 	int i;
1084 	int err;
1085 	struct mlx4_ib_iboe *iboe;
1086 
1087 	printk_once(KERN_INFO "%s", mlx4_ib_version);
1088 
1089 	if (mlx4_is_mfunc(dev)) {
1090 		printk(KERN_WARNING "IB not yet supported in SRIOV\n");
1091 		return NULL;
1092 	}
1093 
1094 	mlx4_foreach_ib_transport_port(i, dev)
1095 		num_ports++;
1096 
1097 	/* No point in registering a device with no ports... */
1098 	if (num_ports == 0)
1099 		return NULL;
1100 
1101 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1102 	if (!ibdev) {
1103 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1104 		return NULL;
1105 	}
1106 
1107 	iboe = &ibdev->iboe;
1108 
1109 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1110 		goto err_dealloc;
1111 
1112 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1113 		goto err_pd;
1114 
1115 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1116 				 PAGE_SIZE);
1117 	if (!ibdev->uar_map)
1118 		goto err_uar;
1119 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1120 
1121 	ibdev->dev = dev;
1122 
1123 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1124 	ibdev->ib_dev.owner		= THIS_MODULE;
1125 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1126 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
1127 	ibdev->num_ports		= num_ports;
1128 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1129 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
1130 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
1131 
1132 	ibdev->ib_dev.uverbs_abi_ver	= MLX4_IB_UVERBS_ABI_VERSION;
1133 	ibdev->ib_dev.uverbs_cmd_mask	=
1134 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1135 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1136 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1137 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1138 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1139 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1140 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1141 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1142 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1143 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1144 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1145 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1146 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1147 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1148 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1149 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1150 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
1151 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1152 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1153 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1154 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
1155 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
1156 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
1157 
1158 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
1159 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
1160 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
1161 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
1162 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
1163 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
1164 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
1165 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
1166 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
1167 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
1168 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
1169 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
1170 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
1171 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
1172 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
1173 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
1174 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
1175 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
1176 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
1177 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
1178 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
1179 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
1180 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
1181 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
1182 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
1183 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
1184 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
1185 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
1186 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
1187 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
1188 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
1189 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
1190 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
1191 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
1192 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
1193 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1194 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1195 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1196 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
1197 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
1198 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
1199 
1200 	ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
1201 	ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
1202 	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
1203 	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
1204 
1205 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1206 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1207 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1208 		ibdev->ib_dev.uverbs_cmd_mask |=
1209 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1210 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1211 	}
1212 
1213 	spin_lock_init(&iboe->lock);
1214 
1215 	if (init_node_data(ibdev))
1216 		goto err_map;
1217 
1218 	for (i = 0; i < ibdev->num_ports; ++i) {
1219 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1220 						IB_LINK_LAYER_ETHERNET) {
1221 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1222 			if (err)
1223 				ibdev->counters[i] = -1;
1224 		} else
1225 				ibdev->counters[i] = -1;
1226 	}
1227 
1228 	spin_lock_init(&ibdev->sm_lock);
1229 	mutex_init(&ibdev->cap_mask_mutex);
1230 
1231 	if (ib_register_device(&ibdev->ib_dev, NULL))
1232 		goto err_counter;
1233 
1234 	if (mlx4_ib_mad_init(ibdev))
1235 		goto err_reg;
1236 
1237 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1238 		iboe->nb.notifier_call = mlx4_ib_netdev_event;
1239 		err = register_netdevice_notifier(&iboe->nb);
1240 		if (err)
1241 			goto err_reg;
1242 	}
1243 
1244 	for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1245 		if (device_create_file(&ibdev->ib_dev.dev,
1246 				       mlx4_class_attributes[i]))
1247 			goto err_notif;
1248 	}
1249 
1250 	ibdev->ib_active = true;
1251 
1252 	return ibdev;
1253 
1254 err_notif:
1255 	if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1256 		printk(KERN_WARNING "failure unregistering notifier\n");
1257 	flush_workqueue(wq);
1258 
1259 err_reg:
1260 	ib_unregister_device(&ibdev->ib_dev);
1261 
1262 err_counter:
1263 	for (; i; --i)
1264 		if (ibdev->counters[i - 1] != -1)
1265 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1266 
1267 err_map:
1268 	iounmap(ibdev->uar_map);
1269 
1270 err_uar:
1271 	mlx4_uar_free(dev, &ibdev->priv_uar);
1272 
1273 err_pd:
1274 	mlx4_pd_free(dev, ibdev->priv_pdn);
1275 
1276 err_dealloc:
1277 	ib_dealloc_device(&ibdev->ib_dev);
1278 
1279 	return NULL;
1280 }
1281 
1282 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1283 {
1284 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
1285 	int p;
1286 
1287 	mlx4_ib_mad_cleanup(ibdev);
1288 	ib_unregister_device(&ibdev->ib_dev);
1289 	if (ibdev->iboe.nb.notifier_call) {
1290 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1291 			printk(KERN_WARNING "failure unregistering notifier\n");
1292 		ibdev->iboe.nb.notifier_call = NULL;
1293 	}
1294 	iounmap(ibdev->uar_map);
1295 	for (p = 0; p < ibdev->num_ports; ++p)
1296 		if (ibdev->counters[p] != -1)
1297 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1298 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1299 		mlx4_CLOSE_PORT(dev, p);
1300 
1301 	mlx4_uar_free(dev, &ibdev->priv_uar);
1302 	mlx4_pd_free(dev, ibdev->priv_pdn);
1303 	ib_dealloc_device(&ibdev->ib_dev);
1304 }
1305 
1306 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1307 			  enum mlx4_dev_event event, int port)
1308 {
1309 	struct ib_event ibev;
1310 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1311 
1312 	if (port > ibdev->num_ports)
1313 		return;
1314 
1315 	switch (event) {
1316 	case MLX4_DEV_EVENT_PORT_UP:
1317 		ibev.event = IB_EVENT_PORT_ACTIVE;
1318 		break;
1319 
1320 	case MLX4_DEV_EVENT_PORT_DOWN:
1321 		ibev.event = IB_EVENT_PORT_ERR;
1322 		break;
1323 
1324 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1325 		ibdev->ib_active = false;
1326 		ibev.event = IB_EVENT_DEVICE_FATAL;
1327 		break;
1328 
1329 	default:
1330 		return;
1331 	}
1332 
1333 	ibev.device	      = ibdev_ptr;
1334 	ibev.element.port_num = port;
1335 
1336 	ib_dispatch_event(&ibev);
1337 }
1338 
1339 static struct mlx4_interface mlx4_ib_interface = {
1340 	.add		= mlx4_ib_add,
1341 	.remove		= mlx4_ib_remove,
1342 	.event		= mlx4_ib_event,
1343 	.protocol	= MLX4_PROT_IB_IPV6
1344 };
1345 
1346 static int __init mlx4_ib_init(void)
1347 {
1348 	int err;
1349 
1350 	wq = create_singlethread_workqueue("mlx4_ib");
1351 	if (!wq)
1352 		return -ENOMEM;
1353 
1354 	err = mlx4_register_interface(&mlx4_ib_interface);
1355 	if (err) {
1356 		destroy_workqueue(wq);
1357 		return err;
1358 	}
1359 
1360 	return 0;
1361 }
1362 
1363 static void __exit mlx4_ib_cleanup(void)
1364 {
1365 	mlx4_unregister_interface(&mlx4_ib_interface);
1366 	destroy_workqueue(wq);
1367 }
1368 
1369 module_init(mlx4_ib_init);
1370 module_exit(mlx4_ib_cleanup);
1371