xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision c21b37f6)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/errno.h>
36 
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_user_verbs.h>
39 
40 #include <linux/mlx4/driver.h>
41 #include <linux/mlx4/cmd.h>
42 
43 #include "mlx4_ib.h"
44 #include "user.h"
45 
46 #define DRV_NAME	"mlx4_ib"
47 #define DRV_VERSION	"0.01"
48 #define DRV_RELDATE	"May 1, 2006"
49 
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
52 MODULE_LICENSE("Dual BSD/GPL");
53 MODULE_VERSION(DRV_VERSION);
54 
55 static const char mlx4_ib_version[] __devinitdata =
56 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
57 	DRV_VERSION " (" DRV_RELDATE ")\n";
58 
59 static void init_query_mad(struct ib_smp *mad)
60 {
61 	mad->base_version  = 1;
62 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
63 	mad->class_version = 1;
64 	mad->method	   = IB_MGMT_METHOD_GET;
65 }
66 
67 static int mlx4_ib_query_device(struct ib_device *ibdev,
68 				struct ib_device_attr *props)
69 {
70 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
71 	struct ib_smp *in_mad  = NULL;
72 	struct ib_smp *out_mad = NULL;
73 	int err = -ENOMEM;
74 
75 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
76 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
77 	if (!in_mad || !out_mad)
78 		goto out;
79 
80 	init_query_mad(in_mad);
81 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
82 
83 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
84 	if (err)
85 		goto out;
86 
87 	memset(props, 0, sizeof *props);
88 
89 	props->fw_ver = dev->dev->caps.fw_ver;
90 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
91 		IB_DEVICE_PORT_ACTIVE_EVENT		|
92 		IB_DEVICE_SYS_IMAGE_GUID		|
93 		IB_DEVICE_RC_RNR_NAK_GEN;
94 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
95 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
96 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
97 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
98 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
99 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
100 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
101 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
102 
103 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
104 		0xffffff;
105 	props->vendor_part_id	   = be16_to_cpup((__be16 *) (out_mad->data + 30));
106 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
107 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
108 
109 	props->max_mr_size	   = ~0ull;
110 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
111 	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
112 	props->max_qp_wr	   = dev->dev->caps.max_wqes;
113 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
114 					 dev->dev->caps.max_rq_sg);
115 	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
116 	props->max_cqe		   = dev->dev->caps.max_cqes;
117 	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
118 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
119 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
120 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
121 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
122 	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
123 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
124 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
125 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
126 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
127 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
128 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
129 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
130 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
131 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
132 					   props->max_mcast_grp;
133 	props->max_map_per_fmr = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
134 
135 out:
136 	kfree(in_mad);
137 	kfree(out_mad);
138 
139 	return err;
140 }
141 
142 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
143 			      struct ib_port_attr *props)
144 {
145 	struct ib_smp *in_mad  = NULL;
146 	struct ib_smp *out_mad = NULL;
147 	int err = -ENOMEM;
148 
149 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
150 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
151 	if (!in_mad || !out_mad)
152 		goto out;
153 
154 	memset(props, 0, sizeof *props);
155 
156 	init_query_mad(in_mad);
157 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
158 	in_mad->attr_mod = cpu_to_be32(port);
159 
160 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
161 	if (err)
162 		goto out;
163 
164 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
165 	props->lmc		= out_mad->data[34] & 0x7;
166 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
167 	props->sm_sl		= out_mad->data[36] & 0xf;
168 	props->state		= out_mad->data[32] & 0xf;
169 	props->phys_state	= out_mad->data[33] >> 4;
170 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
171 	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
172 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
173 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
174 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
175 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
176 	props->active_width	= out_mad->data[31] & 0xf;
177 	props->active_speed	= out_mad->data[35] >> 4;
178 	props->max_mtu		= out_mad->data[41] & 0xf;
179 	props->active_mtu	= out_mad->data[36] >> 4;
180 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
181 	props->max_vl_num	= out_mad->data[37] >> 4;
182 	props->init_type_reply	= out_mad->data[41] >> 4;
183 
184 out:
185 	kfree(in_mad);
186 	kfree(out_mad);
187 
188 	return err;
189 }
190 
191 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
192 			     union ib_gid *gid)
193 {
194 	struct ib_smp *in_mad  = NULL;
195 	struct ib_smp *out_mad = NULL;
196 	int err = -ENOMEM;
197 
198 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
199 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
200 	if (!in_mad || !out_mad)
201 		goto out;
202 
203 	init_query_mad(in_mad);
204 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
205 	in_mad->attr_mod = cpu_to_be32(port);
206 
207 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
208 	if (err)
209 		goto out;
210 
211 	memcpy(gid->raw, out_mad->data + 8, 8);
212 
213 	init_query_mad(in_mad);
214 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
215 	in_mad->attr_mod = cpu_to_be32(index / 8);
216 
217 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
218 	if (err)
219 		goto out;
220 
221 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
222 
223 out:
224 	kfree(in_mad);
225 	kfree(out_mad);
226 	return err;
227 }
228 
229 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
230 			      u16 *pkey)
231 {
232 	struct ib_smp *in_mad  = NULL;
233 	struct ib_smp *out_mad = NULL;
234 	int err = -ENOMEM;
235 
236 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
237 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
238 	if (!in_mad || !out_mad)
239 		goto out;
240 
241 	init_query_mad(in_mad);
242 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
243 	in_mad->attr_mod = cpu_to_be32(index / 32);
244 
245 	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
246 	if (err)
247 		goto out;
248 
249 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
250 
251 out:
252 	kfree(in_mad);
253 	kfree(out_mad);
254 	return err;
255 }
256 
257 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
258 				 struct ib_device_modify *props)
259 {
260 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
261 		return -EOPNOTSUPP;
262 
263 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
264 		spin_lock(&to_mdev(ibdev)->sm_lock);
265 		memcpy(ibdev->node_desc, props->node_desc, 64);
266 		spin_unlock(&to_mdev(ibdev)->sm_lock);
267 	}
268 
269 	return 0;
270 }
271 
272 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
273 			 u32 cap_mask)
274 {
275 	struct mlx4_cmd_mailbox *mailbox;
276 	int err;
277 
278 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
279 	if (IS_ERR(mailbox))
280 		return PTR_ERR(mailbox);
281 
282 	memset(mailbox->buf, 0, 256);
283 
284 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
285 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
286 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
287 	} else {
288 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
289 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
290 	}
291 
292 	err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
293 		       MLX4_CMD_TIME_CLASS_B);
294 
295 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
296 	return err;
297 }
298 
299 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
300 			       struct ib_port_modify *props)
301 {
302 	struct ib_port_attr attr;
303 	u32 cap_mask;
304 	int err;
305 
306 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
307 
308 	err = mlx4_ib_query_port(ibdev, port, &attr);
309 	if (err)
310 		goto out;
311 
312 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
313 		~props->clr_port_cap_mask;
314 
315 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
316 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
317 			    cap_mask);
318 
319 out:
320 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
321 	return err;
322 }
323 
324 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
325 						  struct ib_udata *udata)
326 {
327 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
328 	struct mlx4_ib_ucontext *context;
329 	struct mlx4_ib_alloc_ucontext_resp resp;
330 	int err;
331 
332 	resp.qp_tab_size      = dev->dev->caps.num_qps;
333 	resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
334 	resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
335 
336 	context = kmalloc(sizeof *context, GFP_KERNEL);
337 	if (!context)
338 		return ERR_PTR(-ENOMEM);
339 
340 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
341 	if (err) {
342 		kfree(context);
343 		return ERR_PTR(err);
344 	}
345 
346 	INIT_LIST_HEAD(&context->db_page_list);
347 	mutex_init(&context->db_page_mutex);
348 
349 	err = ib_copy_to_udata(udata, &resp, sizeof resp);
350 	if (err) {
351 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
352 		kfree(context);
353 		return ERR_PTR(-EFAULT);
354 	}
355 
356 	return &context->ibucontext;
357 }
358 
359 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
360 {
361 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
362 
363 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
364 	kfree(context);
365 
366 	return 0;
367 }
368 
369 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
370 {
371 	struct mlx4_ib_dev *dev = to_mdev(context->device);
372 
373 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
374 		return -EINVAL;
375 
376 	if (vma->vm_pgoff == 0) {
377 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
378 
379 		if (io_remap_pfn_range(vma, vma->vm_start,
380 				       to_mucontext(context)->uar.pfn,
381 				       PAGE_SIZE, vma->vm_page_prot))
382 			return -EAGAIN;
383 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
384 		/* FIXME want pgprot_writecombine() for BlueFlame pages */
385 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
386 
387 		if (io_remap_pfn_range(vma, vma->vm_start,
388 				       to_mucontext(context)->uar.pfn +
389 				       dev->dev->caps.num_uars,
390 				       PAGE_SIZE, vma->vm_page_prot))
391 			return -EAGAIN;
392 	} else
393 		return -EINVAL;
394 
395 	return 0;
396 }
397 
398 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
399 				      struct ib_ucontext *context,
400 				      struct ib_udata *udata)
401 {
402 	struct mlx4_ib_pd *pd;
403 	int err;
404 
405 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
406 	if (!pd)
407 		return ERR_PTR(-ENOMEM);
408 
409 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
410 	if (err) {
411 		kfree(pd);
412 		return ERR_PTR(err);
413 	}
414 
415 	if (context)
416 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
417 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
418 			kfree(pd);
419 			return ERR_PTR(-EFAULT);
420 		}
421 
422 	return &pd->ibpd;
423 }
424 
425 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
426 {
427 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
428 	kfree(pd);
429 
430 	return 0;
431 }
432 
433 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
434 {
435 	return mlx4_multicast_attach(to_mdev(ibqp->device)->dev,
436 				     &to_mqp(ibqp)->mqp, gid->raw);
437 }
438 
439 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
440 {
441 	return mlx4_multicast_detach(to_mdev(ibqp->device)->dev,
442 				     &to_mqp(ibqp)->mqp, gid->raw);
443 }
444 
445 static int init_node_data(struct mlx4_ib_dev *dev)
446 {
447 	struct ib_smp *in_mad  = NULL;
448 	struct ib_smp *out_mad = NULL;
449 	int err = -ENOMEM;
450 
451 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
452 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
453 	if (!in_mad || !out_mad)
454 		goto out;
455 
456 	init_query_mad(in_mad);
457 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
458 
459 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
460 	if (err)
461 		goto out;
462 
463 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
464 
465 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
466 
467 	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
468 	if (err)
469 		goto out;
470 
471 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
472 
473 out:
474 	kfree(in_mad);
475 	kfree(out_mad);
476 	return err;
477 }
478 
479 static void *mlx4_ib_add(struct mlx4_dev *dev)
480 {
481 	struct mlx4_ib_dev *ibdev;
482 
483 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
484 	if (!ibdev) {
485 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
486 		return NULL;
487 	}
488 
489 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
490 		goto err_dealloc;
491 
492 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
493 		goto err_pd;
494 
495 	ibdev->uar_map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
496 	if (!ibdev->uar_map)
497 		goto err_uar;
498 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
499 
500 	INIT_LIST_HEAD(&ibdev->pgdir_list);
501 	mutex_init(&ibdev->pgdir_mutex);
502 
503 	ibdev->dev = dev;
504 
505 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
506 	ibdev->ib_dev.owner		= THIS_MODULE;
507 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
508 	ibdev->ib_dev.phys_port_cnt	= dev->caps.num_ports;
509 	ibdev->ib_dev.num_comp_vectors	= 1;
510 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
511 
512 	ibdev->ib_dev.uverbs_abi_ver	= MLX4_IB_UVERBS_ABI_VERSION;
513 	ibdev->ib_dev.uverbs_cmd_mask	=
514 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
515 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
516 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
517 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
518 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
519 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
520 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
521 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
522 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
523 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
524 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
525 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
526 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
527 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
528 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
529 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
530 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
531 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
532 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
533 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
534 
535 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
536 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
537 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
538 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
539 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
540 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
541 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
542 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
543 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
544 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
545 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
546 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
547 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
548 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
549 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
550 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
551 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
552 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
553 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
554 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
555 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
556 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
557 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
558 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
559 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
560 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
561 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
562 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
563 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
564 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
565 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
566 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
567 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
568 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
569 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
570 
571 	if (init_node_data(ibdev))
572 		goto err_map;
573 
574 	spin_lock_init(&ibdev->sm_lock);
575 	mutex_init(&ibdev->cap_mask_mutex);
576 
577 	if (ib_register_device(&ibdev->ib_dev))
578 		goto err_map;
579 
580 	if (mlx4_ib_mad_init(ibdev))
581 		goto err_reg;
582 
583 	return ibdev;
584 
585 err_reg:
586 	ib_unregister_device(&ibdev->ib_dev);
587 
588 err_map:
589 	iounmap(ibdev->uar_map);
590 
591 err_uar:
592 	mlx4_uar_free(dev, &ibdev->priv_uar);
593 
594 err_pd:
595 	mlx4_pd_free(dev, ibdev->priv_pdn);
596 
597 err_dealloc:
598 	ib_dealloc_device(&ibdev->ib_dev);
599 
600 	return NULL;
601 }
602 
603 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
604 {
605 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
606 	int p;
607 
608 	for (p = 1; p <= dev->caps.num_ports; ++p)
609 		mlx4_CLOSE_PORT(dev, p);
610 
611 	mlx4_ib_mad_cleanup(ibdev);
612 	ib_unregister_device(&ibdev->ib_dev);
613 	iounmap(ibdev->uar_map);
614 	mlx4_uar_free(dev, &ibdev->priv_uar);
615 	mlx4_pd_free(dev, ibdev->priv_pdn);
616 	ib_dealloc_device(&ibdev->ib_dev);
617 }
618 
619 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
620 			  enum mlx4_dev_event event, int subtype,
621 			  int port)
622 {
623 	struct ib_event ibev;
624 
625 	switch (event) {
626 	case MLX4_EVENT_TYPE_PORT_CHANGE:
627 		ibev.event = subtype == MLX4_PORT_CHANGE_SUBTYPE_ACTIVE ?
628 			IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
629 		break;
630 
631 	case MLX4_EVENT_TYPE_LOCAL_CATAS_ERROR:
632 		ibev.event = IB_EVENT_DEVICE_FATAL;
633 		break;
634 
635 	default:
636 		return;
637 	}
638 
639 	ibev.device	      = ibdev_ptr;
640 	ibev.element.port_num = port;
641 
642 	ib_dispatch_event(&ibev);
643 }
644 
645 static struct mlx4_interface mlx4_ib_interface = {
646 	.add	= mlx4_ib_add,
647 	.remove	= mlx4_ib_remove,
648 	.event	= mlx4_ib_event
649 };
650 
651 static int __init mlx4_ib_init(void)
652 {
653 	return mlx4_register_interface(&mlx4_ib_interface);
654 }
655 
656 static void __exit mlx4_ib_cleanup(void)
657 {
658 	mlx4_unregister_interface(&mlx4_ib_interface);
659 }
660 
661 module_init(mlx4_ib_init);
662 module_exit(mlx4_ib_cleanup);
663