1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
5  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
6  * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_umem.h>
39 #include <rdma/ib_user_verbs.h>
40 
41 #include <linux/sched.h>
42 #include <linux/slab.h>
43 #include <linux/stat.h>
44 #include <linux/mm.h>
45 #include <linux/export.h>
46 
47 #include "mthca_dev.h"
48 #include "mthca_cmd.h"
49 #include <rdma/mthca-abi.h>
50 #include "mthca_memfree.h"
51 
52 static void init_query_mad(struct ib_smp *mad)
53 {
54 	mad->base_version  = 1;
55 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
56 	mad->class_version = 1;
57 	mad->method    	   = IB_MGMT_METHOD_GET;
58 }
59 
60 static int mthca_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
61 			      struct ib_udata *uhw)
62 {
63 	struct ib_smp *in_mad  = NULL;
64 	struct ib_smp *out_mad = NULL;
65 	int err = -ENOMEM;
66 	struct mthca_dev *mdev = to_mdev(ibdev);
67 
68 	if (uhw->inlen || uhw->outlen)
69 		return -EINVAL;
70 
71 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
72 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
73 	if (!in_mad || !out_mad)
74 		goto out;
75 
76 	memset(props, 0, sizeof *props);
77 
78 	props->fw_ver              = mdev->fw_ver;
79 
80 	init_query_mad(in_mad);
81 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
82 
83 	err = mthca_MAD_IFC(mdev, 1, 1,
84 			    1, NULL, NULL, in_mad, out_mad);
85 	if (err)
86 		goto out;
87 
88 	props->device_cap_flags    = mdev->device_cap_flags;
89 	props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
90 		0xffffff;
91 	props->vendor_part_id      = be16_to_cpup((__be16 *) (out_mad->data + 30));
92 	props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
93 	memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
94 
95 	props->max_mr_size         = ~0ull;
96 	props->page_size_cap       = mdev->limits.page_size_cap;
97 	props->max_qp              = mdev->limits.num_qps - mdev->limits.reserved_qps;
98 	props->max_qp_wr           = mdev->limits.max_wqes;
99 	props->max_send_sge        = mdev->limits.max_sg;
100 	props->max_recv_sge        = mdev->limits.max_sg;
101 	props->max_sge_rd          = mdev->limits.max_sg;
102 	props->max_cq              = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
103 	props->max_cqe             = mdev->limits.max_cqes;
104 	props->max_mr              = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
105 	props->max_pd              = mdev->limits.num_pds - mdev->limits.reserved_pds;
106 	props->max_qp_rd_atom      = 1 << mdev->qp_table.rdb_shift;
107 	props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
108 	props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
109 	props->max_srq             = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
110 	props->max_srq_wr          = mdev->limits.max_srq_wqes;
111 	props->max_srq_sge         = mdev->limits.max_srq_sge;
112 	props->local_ca_ack_delay  = mdev->limits.local_ca_ack_delay;
113 	props->atomic_cap          = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
114 					IB_ATOMIC_HCA : IB_ATOMIC_NONE;
115 	props->max_pkeys           = mdev->limits.pkey_table_len;
116 	props->max_mcast_grp       = mdev->limits.num_mgms + mdev->limits.num_amgms;
117 	props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
118 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
119 					   props->max_mcast_grp;
120 	/*
121 	 * If Sinai memory key optimization is being used, then only
122 	 * the 8-bit key portion will change.  For other HCAs, the
123 	 * unused index bits will also be used for FMR remapping.
124 	 */
125 	if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
126 		props->max_map_per_fmr = 255;
127 	else
128 		props->max_map_per_fmr =
129 			(1 << (32 - ilog2(mdev->limits.num_mpts))) - 1;
130 
131 	err = 0;
132  out:
133 	kfree(in_mad);
134 	kfree(out_mad);
135 	return err;
136 }
137 
138 static int mthca_query_port(struct ib_device *ibdev,
139 			    u8 port, struct ib_port_attr *props)
140 {
141 	struct ib_smp *in_mad  = NULL;
142 	struct ib_smp *out_mad = NULL;
143 	int err = -ENOMEM;
144 
145 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
146 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
147 	if (!in_mad || !out_mad)
148 		goto out;
149 
150 	/* props being zeroed by the caller, avoid zeroing it here */
151 
152 	init_query_mad(in_mad);
153 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
154 	in_mad->attr_mod = cpu_to_be32(port);
155 
156 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
157 			    port, NULL, NULL, in_mad, out_mad);
158 	if (err)
159 		goto out;
160 
161 	props->lid               = be16_to_cpup((__be16 *) (out_mad->data + 16));
162 	props->lmc               = out_mad->data[34] & 0x7;
163 	props->sm_lid            = be16_to_cpup((__be16 *) (out_mad->data + 18));
164 	props->sm_sl             = out_mad->data[36] & 0xf;
165 	props->state             = out_mad->data[32] & 0xf;
166 	props->phys_state        = out_mad->data[33] >> 4;
167 	props->port_cap_flags    = be32_to_cpup((__be32 *) (out_mad->data + 20));
168 	props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
169 	props->max_msg_sz        = 0x80000000;
170 	props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
171 	props->bad_pkey_cntr     = be16_to_cpup((__be16 *) (out_mad->data + 46));
172 	props->qkey_viol_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 48));
173 	props->active_width      = out_mad->data[31] & 0xf;
174 	props->active_speed      = out_mad->data[35] >> 4;
175 	props->max_mtu           = out_mad->data[41] & 0xf;
176 	props->active_mtu        = out_mad->data[36] >> 4;
177 	props->subnet_timeout    = out_mad->data[51] & 0x1f;
178 	props->max_vl_num        = out_mad->data[37] >> 4;
179 	props->init_type_reply   = out_mad->data[41] >> 4;
180 
181  out:
182 	kfree(in_mad);
183 	kfree(out_mad);
184 	return err;
185 }
186 
187 static int mthca_modify_device(struct ib_device *ibdev,
188 			       int mask,
189 			       struct ib_device_modify *props)
190 {
191 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
192 		return -EOPNOTSUPP;
193 
194 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
195 		if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
196 			return -ERESTARTSYS;
197 		memcpy(ibdev->node_desc, props->node_desc,
198 		       IB_DEVICE_NODE_DESC_MAX);
199 		mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
200 	}
201 
202 	return 0;
203 }
204 
205 static int mthca_modify_port(struct ib_device *ibdev,
206 			     u8 port, int port_modify_mask,
207 			     struct ib_port_modify *props)
208 {
209 	struct mthca_set_ib_param set_ib;
210 	struct ib_port_attr attr;
211 	int err;
212 
213 	if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
214 		return -ERESTARTSYS;
215 
216 	err = ib_query_port(ibdev, port, &attr);
217 	if (err)
218 		goto out;
219 
220 	set_ib.set_si_guid     = 0;
221 	set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
222 
223 	set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
224 		~props->clr_port_cap_mask;
225 
226 	err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port);
227 	if (err)
228 		goto out;
229 out:
230 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
231 	return err;
232 }
233 
234 static int mthca_query_pkey(struct ib_device *ibdev,
235 			    u8 port, u16 index, u16 *pkey)
236 {
237 	struct ib_smp *in_mad  = NULL;
238 	struct ib_smp *out_mad = NULL;
239 	int err = -ENOMEM;
240 
241 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
242 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
243 	if (!in_mad || !out_mad)
244 		goto out;
245 
246 	init_query_mad(in_mad);
247 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
248 	in_mad->attr_mod = cpu_to_be32(index / 32);
249 
250 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
251 			    port, NULL, NULL, in_mad, out_mad);
252 	if (err)
253 		goto out;
254 
255 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
256 
257  out:
258 	kfree(in_mad);
259 	kfree(out_mad);
260 	return err;
261 }
262 
263 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
264 			   int index, union ib_gid *gid)
265 {
266 	struct ib_smp *in_mad  = NULL;
267 	struct ib_smp *out_mad = NULL;
268 	int err = -ENOMEM;
269 
270 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
271 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
272 	if (!in_mad || !out_mad)
273 		goto out;
274 
275 	init_query_mad(in_mad);
276 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
277 	in_mad->attr_mod = cpu_to_be32(port);
278 
279 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
280 			    port, NULL, NULL, in_mad, out_mad);
281 	if (err)
282 		goto out;
283 
284 	memcpy(gid->raw, out_mad->data + 8, 8);
285 
286 	init_query_mad(in_mad);
287 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
288 	in_mad->attr_mod = cpu_to_be32(index / 8);
289 
290 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
291 			    port, NULL, NULL, in_mad, out_mad);
292 	if (err)
293 		goto out;
294 
295 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
296 
297  out:
298 	kfree(in_mad);
299 	kfree(out_mad);
300 	return err;
301 }
302 
303 static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
304 						struct ib_udata *udata)
305 {
306 	struct mthca_alloc_ucontext_resp uresp;
307 	struct mthca_ucontext           *context;
308 	int                              err;
309 
310 	if (!(to_mdev(ibdev)->active))
311 		return ERR_PTR(-EAGAIN);
312 
313 	memset(&uresp, 0, sizeof uresp);
314 
315 	uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
316 	if (mthca_is_memfree(to_mdev(ibdev)))
317 		uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
318 	else
319 		uresp.uarc_size = 0;
320 
321 	context = kmalloc(sizeof *context, GFP_KERNEL);
322 	if (!context)
323 		return ERR_PTR(-ENOMEM);
324 
325 	err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
326 	if (err) {
327 		kfree(context);
328 		return ERR_PTR(err);
329 	}
330 
331 	context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
332 	if (IS_ERR(context->db_tab)) {
333 		err = PTR_ERR(context->db_tab);
334 		mthca_uar_free(to_mdev(ibdev), &context->uar);
335 		kfree(context);
336 		return ERR_PTR(err);
337 	}
338 
339 	if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) {
340 		mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
341 		mthca_uar_free(to_mdev(ibdev), &context->uar);
342 		kfree(context);
343 		return ERR_PTR(-EFAULT);
344 	}
345 
346 	context->reg_mr_warned = 0;
347 
348 	return &context->ibucontext;
349 }
350 
351 static int mthca_dealloc_ucontext(struct ib_ucontext *context)
352 {
353 	mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
354 				  to_mucontext(context)->db_tab);
355 	mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
356 	kfree(to_mucontext(context));
357 
358 	return 0;
359 }
360 
361 static int mthca_mmap_uar(struct ib_ucontext *context,
362 			  struct vm_area_struct *vma)
363 {
364 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
365 		return -EINVAL;
366 
367 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
368 
369 	if (io_remap_pfn_range(vma, vma->vm_start,
370 			       to_mucontext(context)->uar.pfn,
371 			       PAGE_SIZE, vma->vm_page_prot))
372 		return -EAGAIN;
373 
374 	return 0;
375 }
376 
377 static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev,
378 				    struct ib_ucontext *context,
379 				    struct ib_udata *udata)
380 {
381 	struct mthca_pd *pd;
382 	int err;
383 
384 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
385 	if (!pd)
386 		return ERR_PTR(-ENOMEM);
387 
388 	err = mthca_pd_alloc(to_mdev(ibdev), !context, pd);
389 	if (err) {
390 		kfree(pd);
391 		return ERR_PTR(err);
392 	}
393 
394 	if (context) {
395 		if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
396 			mthca_pd_free(to_mdev(ibdev), pd);
397 			kfree(pd);
398 			return ERR_PTR(-EFAULT);
399 		}
400 	}
401 
402 	return &pd->ibpd;
403 }
404 
405 static int mthca_dealloc_pd(struct ib_pd *pd)
406 {
407 	mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
408 	kfree(pd);
409 
410 	return 0;
411 }
412 
413 static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
414 				     struct rdma_ah_attr *ah_attr,
415 				     struct ib_udata *udata)
416 
417 {
418 	int err;
419 	struct mthca_ah *ah;
420 
421 	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
422 	if (!ah)
423 		return ERR_PTR(-ENOMEM);
424 
425 	err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
426 	if (err) {
427 		kfree(ah);
428 		return ERR_PTR(err);
429 	}
430 
431 	return &ah->ibah;
432 }
433 
434 static int mthca_ah_destroy(struct ib_ah *ah)
435 {
436 	mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
437 	kfree(ah);
438 
439 	return 0;
440 }
441 
442 static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
443 				       struct ib_srq_init_attr *init_attr,
444 				       struct ib_udata *udata)
445 {
446 	struct mthca_create_srq ucmd;
447 	struct mthca_ucontext *context = NULL;
448 	struct mthca_srq *srq;
449 	int err;
450 
451 	if (init_attr->srq_type != IB_SRQT_BASIC)
452 		return ERR_PTR(-EOPNOTSUPP);
453 
454 	srq = kmalloc(sizeof *srq, GFP_KERNEL);
455 	if (!srq)
456 		return ERR_PTR(-ENOMEM);
457 
458 	if (pd->uobject) {
459 		context = to_mucontext(pd->uobject->context);
460 
461 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
462 			err = -EFAULT;
463 			goto err_free;
464 		}
465 
466 		err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
467 					context->db_tab, ucmd.db_index,
468 					ucmd.db_page);
469 
470 		if (err)
471 			goto err_free;
472 
473 		srq->mr.ibmr.lkey = ucmd.lkey;
474 		srq->db_index     = ucmd.db_index;
475 	}
476 
477 	err = mthca_alloc_srq(to_mdev(pd->device), to_mpd(pd),
478 			      &init_attr->attr, srq);
479 
480 	if (err && pd->uobject)
481 		mthca_unmap_user_db(to_mdev(pd->device), &context->uar,
482 				    context->db_tab, ucmd.db_index);
483 
484 	if (err)
485 		goto err_free;
486 
487 	if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof (__u32))) {
488 		mthca_free_srq(to_mdev(pd->device), srq);
489 		err = -EFAULT;
490 		goto err_free;
491 	}
492 
493 	return &srq->ibsrq;
494 
495 err_free:
496 	kfree(srq);
497 
498 	return ERR_PTR(err);
499 }
500 
501 static int mthca_destroy_srq(struct ib_srq *srq)
502 {
503 	struct mthca_ucontext *context;
504 
505 	if (srq->uobject) {
506 		context = to_mucontext(srq->uobject->context);
507 
508 		mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
509 				    context->db_tab, to_msrq(srq)->db_index);
510 	}
511 
512 	mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
513 	kfree(srq);
514 
515 	return 0;
516 }
517 
518 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
519 				     struct ib_qp_init_attr *init_attr,
520 				     struct ib_udata *udata)
521 {
522 	struct mthca_create_qp ucmd;
523 	struct mthca_qp *qp;
524 	int err;
525 
526 	if (init_attr->create_flags)
527 		return ERR_PTR(-EINVAL);
528 
529 	switch (init_attr->qp_type) {
530 	case IB_QPT_RC:
531 	case IB_QPT_UC:
532 	case IB_QPT_UD:
533 	{
534 		struct mthca_ucontext *context;
535 
536 		qp = kmalloc(sizeof *qp, GFP_KERNEL);
537 		if (!qp)
538 			return ERR_PTR(-ENOMEM);
539 
540 		if (pd->uobject) {
541 			context = to_mucontext(pd->uobject->context);
542 
543 			if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
544 				kfree(qp);
545 				return ERR_PTR(-EFAULT);
546 			}
547 
548 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
549 						context->db_tab,
550 						ucmd.sq_db_index, ucmd.sq_db_page);
551 			if (err) {
552 				kfree(qp);
553 				return ERR_PTR(err);
554 			}
555 
556 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
557 						context->db_tab,
558 						ucmd.rq_db_index, ucmd.rq_db_page);
559 			if (err) {
560 				mthca_unmap_user_db(to_mdev(pd->device),
561 						    &context->uar,
562 						    context->db_tab,
563 						    ucmd.sq_db_index);
564 				kfree(qp);
565 				return ERR_PTR(err);
566 			}
567 
568 			qp->mr.ibmr.lkey = ucmd.lkey;
569 			qp->sq.db_index  = ucmd.sq_db_index;
570 			qp->rq.db_index  = ucmd.rq_db_index;
571 		}
572 
573 		err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
574 				     to_mcq(init_attr->send_cq),
575 				     to_mcq(init_attr->recv_cq),
576 				     init_attr->qp_type, init_attr->sq_sig_type,
577 				     &init_attr->cap, qp);
578 
579 		if (err && pd->uobject) {
580 			context = to_mucontext(pd->uobject->context);
581 
582 			mthca_unmap_user_db(to_mdev(pd->device),
583 					    &context->uar,
584 					    context->db_tab,
585 					    ucmd.sq_db_index);
586 			mthca_unmap_user_db(to_mdev(pd->device),
587 					    &context->uar,
588 					    context->db_tab,
589 					    ucmd.rq_db_index);
590 		}
591 
592 		qp->ibqp.qp_num = qp->qpn;
593 		break;
594 	}
595 	case IB_QPT_SMI:
596 	case IB_QPT_GSI:
597 	{
598 		/* Don't allow userspace to create special QPs */
599 		if (pd->uobject)
600 			return ERR_PTR(-EINVAL);
601 
602 		qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
603 		if (!qp)
604 			return ERR_PTR(-ENOMEM);
605 
606 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
607 
608 		err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
609 				      to_mcq(init_attr->send_cq),
610 				      to_mcq(init_attr->recv_cq),
611 				      init_attr->sq_sig_type, &init_attr->cap,
612 				      qp->ibqp.qp_num, init_attr->port_num,
613 				      to_msqp(qp));
614 		break;
615 	}
616 	default:
617 		/* Don't support raw QPs */
618 		return ERR_PTR(-ENOSYS);
619 	}
620 
621 	if (err) {
622 		kfree(qp);
623 		return ERR_PTR(err);
624 	}
625 
626 	init_attr->cap.max_send_wr     = qp->sq.max;
627 	init_attr->cap.max_recv_wr     = qp->rq.max;
628 	init_attr->cap.max_send_sge    = qp->sq.max_gs;
629 	init_attr->cap.max_recv_sge    = qp->rq.max_gs;
630 	init_attr->cap.max_inline_data = qp->max_inline_data;
631 
632 	return &qp->ibqp;
633 }
634 
635 static int mthca_destroy_qp(struct ib_qp *qp)
636 {
637 	if (qp->uobject) {
638 		mthca_unmap_user_db(to_mdev(qp->device),
639 				    &to_mucontext(qp->uobject->context)->uar,
640 				    to_mucontext(qp->uobject->context)->db_tab,
641 				    to_mqp(qp)->sq.db_index);
642 		mthca_unmap_user_db(to_mdev(qp->device),
643 				    &to_mucontext(qp->uobject->context)->uar,
644 				    to_mucontext(qp->uobject->context)->db_tab,
645 				    to_mqp(qp)->rq.db_index);
646 	}
647 	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
648 	kfree(qp);
649 	return 0;
650 }
651 
652 static struct ib_cq *mthca_create_cq(struct ib_device *ibdev,
653 				     const struct ib_cq_init_attr *attr,
654 				     struct ib_ucontext *context,
655 				     struct ib_udata *udata)
656 {
657 	int entries = attr->cqe;
658 	struct mthca_create_cq ucmd;
659 	struct mthca_cq *cq;
660 	int nent;
661 	int err;
662 
663 	if (attr->flags)
664 		return ERR_PTR(-EINVAL);
665 
666 	if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
667 		return ERR_PTR(-EINVAL);
668 
669 	if (context) {
670 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
671 			return ERR_PTR(-EFAULT);
672 
673 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
674 					to_mucontext(context)->db_tab,
675 					ucmd.set_db_index, ucmd.set_db_page);
676 		if (err)
677 			return ERR_PTR(err);
678 
679 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
680 					to_mucontext(context)->db_tab,
681 					ucmd.arm_db_index, ucmd.arm_db_page);
682 		if (err)
683 			goto err_unmap_set;
684 	}
685 
686 	cq = kmalloc(sizeof *cq, GFP_KERNEL);
687 	if (!cq) {
688 		err = -ENOMEM;
689 		goto err_unmap_arm;
690 	}
691 
692 	if (context) {
693 		cq->buf.mr.ibmr.lkey = ucmd.lkey;
694 		cq->set_ci_db_index  = ucmd.set_db_index;
695 		cq->arm_db_index     = ucmd.arm_db_index;
696 	}
697 
698 	for (nent = 1; nent <= entries; nent <<= 1)
699 		; /* nothing */
700 
701 	err = mthca_init_cq(to_mdev(ibdev), nent,
702 			    context ? to_mucontext(context) : NULL,
703 			    context ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
704 			    cq);
705 	if (err)
706 		goto err_free;
707 
708 	if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
709 		mthca_free_cq(to_mdev(ibdev), cq);
710 		err = -EFAULT;
711 		goto err_free;
712 	}
713 
714 	cq->resize_buf = NULL;
715 
716 	return &cq->ibcq;
717 
718 err_free:
719 	kfree(cq);
720 
721 err_unmap_arm:
722 	if (context)
723 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
724 				    to_mucontext(context)->db_tab, ucmd.arm_db_index);
725 
726 err_unmap_set:
727 	if (context)
728 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
729 				    to_mucontext(context)->db_tab, ucmd.set_db_index);
730 
731 	return ERR_PTR(err);
732 }
733 
734 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
735 				  int entries)
736 {
737 	int ret;
738 
739 	spin_lock_irq(&cq->lock);
740 	if (cq->resize_buf) {
741 		ret = -EBUSY;
742 		goto unlock;
743 	}
744 
745 	cq->resize_buf = kmalloc(sizeof *cq->resize_buf, GFP_ATOMIC);
746 	if (!cq->resize_buf) {
747 		ret = -ENOMEM;
748 		goto unlock;
749 	}
750 
751 	cq->resize_buf->state = CQ_RESIZE_ALLOC;
752 
753 	ret = 0;
754 
755 unlock:
756 	spin_unlock_irq(&cq->lock);
757 
758 	if (ret)
759 		return ret;
760 
761 	ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
762 	if (ret) {
763 		spin_lock_irq(&cq->lock);
764 		kfree(cq->resize_buf);
765 		cq->resize_buf = NULL;
766 		spin_unlock_irq(&cq->lock);
767 		return ret;
768 	}
769 
770 	cq->resize_buf->cqe = entries - 1;
771 
772 	spin_lock_irq(&cq->lock);
773 	cq->resize_buf->state = CQ_RESIZE_READY;
774 	spin_unlock_irq(&cq->lock);
775 
776 	return 0;
777 }
778 
779 static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
780 {
781 	struct mthca_dev *dev = to_mdev(ibcq->device);
782 	struct mthca_cq *cq = to_mcq(ibcq);
783 	struct mthca_resize_cq ucmd;
784 	u32 lkey;
785 	int ret;
786 
787 	if (entries < 1 || entries > dev->limits.max_cqes)
788 		return -EINVAL;
789 
790 	mutex_lock(&cq->mutex);
791 
792 	entries = roundup_pow_of_two(entries + 1);
793 	if (entries == ibcq->cqe + 1) {
794 		ret = 0;
795 		goto out;
796 	}
797 
798 	if (cq->is_kernel) {
799 		ret = mthca_alloc_resize_buf(dev, cq, entries);
800 		if (ret)
801 			goto out;
802 		lkey = cq->resize_buf->buf.mr.ibmr.lkey;
803 	} else {
804 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
805 			ret = -EFAULT;
806 			goto out;
807 		}
808 		lkey = ucmd.lkey;
809 	}
810 
811 	ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries));
812 
813 	if (ret) {
814 		if (cq->resize_buf) {
815 			mthca_free_cq_buf(dev, &cq->resize_buf->buf,
816 					  cq->resize_buf->cqe);
817 			kfree(cq->resize_buf);
818 			spin_lock_irq(&cq->lock);
819 			cq->resize_buf = NULL;
820 			spin_unlock_irq(&cq->lock);
821 		}
822 		goto out;
823 	}
824 
825 	if (cq->is_kernel) {
826 		struct mthca_cq_buf tbuf;
827 		int tcqe;
828 
829 		spin_lock_irq(&cq->lock);
830 		if (cq->resize_buf->state == CQ_RESIZE_READY) {
831 			mthca_cq_resize_copy_cqes(cq);
832 			tbuf         = cq->buf;
833 			tcqe         = cq->ibcq.cqe;
834 			cq->buf      = cq->resize_buf->buf;
835 			cq->ibcq.cqe = cq->resize_buf->cqe;
836 		} else {
837 			tbuf = cq->resize_buf->buf;
838 			tcqe = cq->resize_buf->cqe;
839 		}
840 
841 		kfree(cq->resize_buf);
842 		cq->resize_buf = NULL;
843 		spin_unlock_irq(&cq->lock);
844 
845 		mthca_free_cq_buf(dev, &tbuf, tcqe);
846 	} else
847 		ibcq->cqe = entries - 1;
848 
849 out:
850 	mutex_unlock(&cq->mutex);
851 
852 	return ret;
853 }
854 
855 static int mthca_destroy_cq(struct ib_cq *cq)
856 {
857 	if (cq->uobject) {
858 		mthca_unmap_user_db(to_mdev(cq->device),
859 				    &to_mucontext(cq->uobject->context)->uar,
860 				    to_mucontext(cq->uobject->context)->db_tab,
861 				    to_mcq(cq)->arm_db_index);
862 		mthca_unmap_user_db(to_mdev(cq->device),
863 				    &to_mucontext(cq->uobject->context)->uar,
864 				    to_mucontext(cq->uobject->context)->db_tab,
865 				    to_mcq(cq)->set_ci_db_index);
866 	}
867 	mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
868 	kfree(cq);
869 
870 	return 0;
871 }
872 
873 static inline u32 convert_access(int acc)
874 {
875 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
876 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
877 	       (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
878 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
879 	       MTHCA_MPT_FLAG_LOCAL_READ;
880 }
881 
882 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
883 {
884 	struct mthca_mr *mr;
885 	int err;
886 
887 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
888 	if (!mr)
889 		return ERR_PTR(-ENOMEM);
890 
891 	err = mthca_mr_alloc_notrans(to_mdev(pd->device),
892 				     to_mpd(pd)->pd_num,
893 				     convert_access(acc), mr);
894 
895 	if (err) {
896 		kfree(mr);
897 		return ERR_PTR(err);
898 	}
899 
900 	mr->umem = NULL;
901 
902 	return &mr->ibmr;
903 }
904 
905 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
906 				       u64 virt, int acc, struct ib_udata *udata)
907 {
908 	struct mthca_dev *dev = to_mdev(pd->device);
909 	struct scatterlist *sg;
910 	struct mthca_mr *mr;
911 	struct mthca_reg_mr ucmd;
912 	u64 *pages;
913 	int shift, n, len;
914 	int i, k, entry;
915 	int err = 0;
916 	int write_mtt_size;
917 
918 	if (udata->inlen < sizeof ucmd) {
919 		if (!to_mucontext(pd->uobject->context)->reg_mr_warned) {
920 			mthca_warn(dev, "Process '%s' did not pass in MR attrs.\n",
921 				   current->comm);
922 			mthca_warn(dev, "  Update libmthca to fix this.\n");
923 		}
924 		++to_mucontext(pd->uobject->context)->reg_mr_warned;
925 		ucmd.mr_attrs = 0;
926 	} else if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
927 		return ERR_PTR(-EFAULT);
928 
929 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
930 	if (!mr)
931 		return ERR_PTR(-ENOMEM);
932 
933 	mr->umem = ib_umem_get(pd->uobject->context, start, length, acc,
934 			       ucmd.mr_attrs & MTHCA_MR_DMASYNC);
935 
936 	if (IS_ERR(mr->umem)) {
937 		err = PTR_ERR(mr->umem);
938 		goto err;
939 	}
940 
941 	shift = mr->umem->page_shift;
942 	n = mr->umem->nmap;
943 
944 	mr->mtt = mthca_alloc_mtt(dev, n);
945 	if (IS_ERR(mr->mtt)) {
946 		err = PTR_ERR(mr->mtt);
947 		goto err_umem;
948 	}
949 
950 	pages = (u64 *) __get_free_page(GFP_KERNEL);
951 	if (!pages) {
952 		err = -ENOMEM;
953 		goto err_mtt;
954 	}
955 
956 	i = n = 0;
957 
958 	write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
959 
960 	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
961 		len = sg_dma_len(sg) >> shift;
962 		for (k = 0; k < len; ++k) {
963 			pages[i++] = sg_dma_address(sg) + (k << shift);
964 			/*
965 			 * Be friendly to write_mtt and pass it chunks
966 			 * of appropriate size.
967 			 */
968 			if (i == write_mtt_size) {
969 				err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
970 				if (err)
971 					goto mtt_done;
972 				n += i;
973 				i = 0;
974 			}
975 		}
976 	}
977 
978 	if (i)
979 		err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
980 mtt_done:
981 	free_page((unsigned long) pages);
982 	if (err)
983 		goto err_mtt;
984 
985 	err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, virt, length,
986 			     convert_access(acc), mr);
987 
988 	if (err)
989 		goto err_mtt;
990 
991 	return &mr->ibmr;
992 
993 err_mtt:
994 	mthca_free_mtt(dev, mr->mtt);
995 
996 err_umem:
997 	ib_umem_release(mr->umem);
998 
999 err:
1000 	kfree(mr);
1001 	return ERR_PTR(err);
1002 }
1003 
1004 static int mthca_dereg_mr(struct ib_mr *mr)
1005 {
1006 	struct mthca_mr *mmr = to_mmr(mr);
1007 
1008 	mthca_free_mr(to_mdev(mr->device), mmr);
1009 	if (mmr->umem)
1010 		ib_umem_release(mmr->umem);
1011 	kfree(mmr);
1012 
1013 	return 0;
1014 }
1015 
1016 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1017 				      struct ib_fmr_attr *fmr_attr)
1018 {
1019 	struct mthca_fmr *fmr;
1020 	int err;
1021 
1022 	fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
1023 	if (!fmr)
1024 		return ERR_PTR(-ENOMEM);
1025 
1026 	memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
1027 	err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
1028 			     convert_access(mr_access_flags), fmr);
1029 
1030 	if (err) {
1031 		kfree(fmr);
1032 		return ERR_PTR(err);
1033 	}
1034 
1035 	return &fmr->ibmr;
1036 }
1037 
1038 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
1039 {
1040 	struct mthca_fmr *mfmr = to_mfmr(fmr);
1041 	int err;
1042 
1043 	err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
1044 	if (err)
1045 		return err;
1046 
1047 	kfree(mfmr);
1048 	return 0;
1049 }
1050 
1051 static int mthca_unmap_fmr(struct list_head *fmr_list)
1052 {
1053 	struct ib_fmr *fmr;
1054 	int err;
1055 	struct mthca_dev *mdev = NULL;
1056 
1057 	list_for_each_entry(fmr, fmr_list, list) {
1058 		if (mdev && to_mdev(fmr->device) != mdev)
1059 			return -EINVAL;
1060 		mdev = to_mdev(fmr->device);
1061 	}
1062 
1063 	if (!mdev)
1064 		return 0;
1065 
1066 	if (mthca_is_memfree(mdev)) {
1067 		list_for_each_entry(fmr, fmr_list, list)
1068 			mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1069 
1070 		wmb();
1071 	} else
1072 		list_for_each_entry(fmr, fmr_list, list)
1073 			mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1074 
1075 	err = mthca_SYNC_TPT(mdev);
1076 	return err;
1077 }
1078 
1079 static ssize_t hw_rev_show(struct device *device,
1080 			   struct device_attribute *attr, char *buf)
1081 {
1082 	struct mthca_dev *dev =
1083 		container_of(device, struct mthca_dev, ib_dev.dev);
1084 	return sprintf(buf, "%x\n", dev->rev_id);
1085 }
1086 static DEVICE_ATTR_RO(hw_rev);
1087 
1088 static ssize_t hca_type_show(struct device *device,
1089 			     struct device_attribute *attr, char *buf)
1090 {
1091 	struct mthca_dev *dev =
1092 		container_of(device, struct mthca_dev, ib_dev.dev);
1093 	switch (dev->pdev->device) {
1094 	case PCI_DEVICE_ID_MELLANOX_TAVOR:
1095 		return sprintf(buf, "MT23108\n");
1096 	case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1097 		return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1098 	case PCI_DEVICE_ID_MELLANOX_ARBEL:
1099 		return sprintf(buf, "MT25208\n");
1100 	case PCI_DEVICE_ID_MELLANOX_SINAI:
1101 	case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1102 		return sprintf(buf, "MT25204\n");
1103 	default:
1104 		return sprintf(buf, "unknown\n");
1105 	}
1106 }
1107 static DEVICE_ATTR_RO(hca_type);
1108 
1109 static ssize_t board_id_show(struct device *device,
1110 			     struct device_attribute *attr, char *buf)
1111 {
1112 	struct mthca_dev *dev =
1113 		container_of(device, struct mthca_dev, ib_dev.dev);
1114 	return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1115 }
1116 static DEVICE_ATTR_RO(board_id);
1117 
1118 static struct attribute *mthca_dev_attributes[] = {
1119 	&dev_attr_hw_rev.attr,
1120 	&dev_attr_hca_type.attr,
1121 	&dev_attr_board_id.attr,
1122 	NULL
1123 };
1124 
1125 static const struct attribute_group mthca_attr_group = {
1126 	.attrs = mthca_dev_attributes,
1127 };
1128 
1129 static int mthca_init_node_data(struct mthca_dev *dev)
1130 {
1131 	struct ib_smp *in_mad  = NULL;
1132 	struct ib_smp *out_mad = NULL;
1133 	int err = -ENOMEM;
1134 
1135 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1136 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1137 	if (!in_mad || !out_mad)
1138 		goto out;
1139 
1140 	init_query_mad(in_mad);
1141 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1142 
1143 	err = mthca_MAD_IFC(dev, 1, 1,
1144 			    1, NULL, NULL, in_mad, out_mad);
1145 	if (err)
1146 		goto out;
1147 
1148 	memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1149 
1150 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1151 
1152 	err = mthca_MAD_IFC(dev, 1, 1,
1153 			    1, NULL, NULL, in_mad, out_mad);
1154 	if (err)
1155 		goto out;
1156 
1157 	if (mthca_is_memfree(dev))
1158 		dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1159 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1160 
1161 out:
1162 	kfree(in_mad);
1163 	kfree(out_mad);
1164 	return err;
1165 }
1166 
1167 static int mthca_port_immutable(struct ib_device *ibdev, u8 port_num,
1168 			        struct ib_port_immutable *immutable)
1169 {
1170 	struct ib_port_attr attr;
1171 	int err;
1172 
1173 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
1174 
1175 	err = ib_query_port(ibdev, port_num, &attr);
1176 	if (err)
1177 		return err;
1178 
1179 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
1180 	immutable->gid_tbl_len = attr.gid_tbl_len;
1181 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
1182 
1183 	return 0;
1184 }
1185 
1186 static void get_dev_fw_str(struct ib_device *device, char *str)
1187 {
1188 	struct mthca_dev *dev =
1189 		container_of(device, struct mthca_dev, ib_dev);
1190 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
1191 		 (int) (dev->fw_ver >> 32),
1192 		 (int) (dev->fw_ver >> 16) & 0xffff,
1193 		 (int) dev->fw_ver & 0xffff);
1194 }
1195 
1196 int mthca_register_device(struct mthca_dev *dev)
1197 {
1198 	int ret;
1199 
1200 	ret = mthca_init_node_data(dev);
1201 	if (ret)
1202 		return ret;
1203 
1204 	dev->ib_dev.owner                = THIS_MODULE;
1205 
1206 	dev->ib_dev.uverbs_abi_ver	 = MTHCA_UVERBS_ABI_VERSION;
1207 	dev->ib_dev.uverbs_cmd_mask	 =
1208 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1209 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1210 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1211 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1212 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1213 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1214 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1215 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1216 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1217 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1218 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1219 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1220 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1221 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1222 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1223 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1224 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1225 	dev->ib_dev.node_type            = RDMA_NODE_IB_CA;
1226 	dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
1227 	dev->ib_dev.num_comp_vectors     = 1;
1228 	dev->ib_dev.dev.parent           = &dev->pdev->dev;
1229 	dev->ib_dev.query_device         = mthca_query_device;
1230 	dev->ib_dev.query_port           = mthca_query_port;
1231 	dev->ib_dev.modify_device        = mthca_modify_device;
1232 	dev->ib_dev.modify_port          = mthca_modify_port;
1233 	dev->ib_dev.query_pkey           = mthca_query_pkey;
1234 	dev->ib_dev.query_gid            = mthca_query_gid;
1235 	dev->ib_dev.alloc_ucontext       = mthca_alloc_ucontext;
1236 	dev->ib_dev.dealloc_ucontext     = mthca_dealloc_ucontext;
1237 	dev->ib_dev.mmap                 = mthca_mmap_uar;
1238 	dev->ib_dev.alloc_pd             = mthca_alloc_pd;
1239 	dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
1240 	dev->ib_dev.create_ah            = mthca_ah_create;
1241 	dev->ib_dev.query_ah             = mthca_ah_query;
1242 	dev->ib_dev.destroy_ah           = mthca_ah_destroy;
1243 
1244 	if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1245 		dev->ib_dev.create_srq           = mthca_create_srq;
1246 		dev->ib_dev.modify_srq           = mthca_modify_srq;
1247 		dev->ib_dev.query_srq            = mthca_query_srq;
1248 		dev->ib_dev.destroy_srq          = mthca_destroy_srq;
1249 		dev->ib_dev.uverbs_cmd_mask	|=
1250 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1251 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1252 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1253 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1254 
1255 		if (mthca_is_memfree(dev))
1256 			dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1257 		else
1258 			dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1259 	}
1260 
1261 	dev->ib_dev.create_qp            = mthca_create_qp;
1262 	dev->ib_dev.modify_qp            = mthca_modify_qp;
1263 	dev->ib_dev.query_qp             = mthca_query_qp;
1264 	dev->ib_dev.destroy_qp           = mthca_destroy_qp;
1265 	dev->ib_dev.create_cq            = mthca_create_cq;
1266 	dev->ib_dev.resize_cq            = mthca_resize_cq;
1267 	dev->ib_dev.destroy_cq           = mthca_destroy_cq;
1268 	dev->ib_dev.poll_cq              = mthca_poll_cq;
1269 	dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
1270 	dev->ib_dev.reg_user_mr          = mthca_reg_user_mr;
1271 	dev->ib_dev.dereg_mr             = mthca_dereg_mr;
1272 	dev->ib_dev.get_port_immutable   = mthca_port_immutable;
1273 	dev->ib_dev.get_dev_fw_str       = get_dev_fw_str;
1274 
1275 	if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1276 		dev->ib_dev.alloc_fmr            = mthca_alloc_fmr;
1277 		dev->ib_dev.unmap_fmr            = mthca_unmap_fmr;
1278 		dev->ib_dev.dealloc_fmr          = mthca_dealloc_fmr;
1279 		if (mthca_is_memfree(dev))
1280 			dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1281 		else
1282 			dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1283 	}
1284 
1285 	dev->ib_dev.attach_mcast         = mthca_multicast_attach;
1286 	dev->ib_dev.detach_mcast         = mthca_multicast_detach;
1287 	dev->ib_dev.process_mad          = mthca_process_mad;
1288 
1289 	if (mthca_is_memfree(dev)) {
1290 		dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1291 		dev->ib_dev.post_send     = mthca_arbel_post_send;
1292 		dev->ib_dev.post_recv     = mthca_arbel_post_receive;
1293 	} else {
1294 		dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1295 		dev->ib_dev.post_send     = mthca_tavor_post_send;
1296 		dev->ib_dev.post_recv     = mthca_tavor_post_receive;
1297 	}
1298 
1299 	mutex_init(&dev->cap_mask_mutex);
1300 
1301 	rdma_set_device_sysfs_group(&dev->ib_dev, &mthca_attr_group);
1302 	dev->ib_dev.driver_id = RDMA_DRIVER_MTHCA;
1303 	ret = ib_register_device(&dev->ib_dev, "mthca%d", NULL);
1304 	if (ret)
1305 		return ret;
1306 
1307 	mthca_start_catas_poll(dev);
1308 
1309 	return 0;
1310 }
1311 
1312 void mthca_unregister_device(struct mthca_dev *dev)
1313 {
1314 	mthca_stop_catas_poll(dev);
1315 	ib_unregister_device(&dev->ib_dev);
1316 }
1317