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  * $Id: mthca_provider.c 4859 2006-01-09 21:55:10Z roland $
37  */
38 
39 #include <rdma/ib_smi.h>
40 #include <rdma/ib_umem.h>
41 #include <rdma/ib_user_verbs.h>
42 #include <linux/mm.h>
43 
44 #include "mthca_dev.h"
45 #include "mthca_cmd.h"
46 #include "mthca_user.h"
47 #include "mthca_memfree.h"
48 
49 static void init_query_mad(struct ib_smp *mad)
50 {
51 	mad->base_version  = 1;
52 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
53 	mad->class_version = 1;
54 	mad->method    	   = IB_MGMT_METHOD_GET;
55 }
56 
57 static int mthca_query_device(struct ib_device *ibdev,
58 			      struct ib_device_attr *props)
59 {
60 	struct ib_smp *in_mad  = NULL;
61 	struct ib_smp *out_mad = NULL;
62 	int err = -ENOMEM;
63 	struct mthca_dev *mdev = to_mdev(ibdev);
64 
65 	u8 status;
66 
67 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
68 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
69 	if (!in_mad || !out_mad)
70 		goto out;
71 
72 	memset(props, 0, sizeof *props);
73 
74 	props->fw_ver              = mdev->fw_ver;
75 
76 	init_query_mad(in_mad);
77 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
78 
79 	err = mthca_MAD_IFC(mdev, 1, 1,
80 			    1, NULL, NULL, in_mad, out_mad,
81 			    &status);
82 	if (err)
83 		goto out;
84 	if (status) {
85 		err = -EINVAL;
86 		goto out;
87 	}
88 
89 	props->device_cap_flags    = mdev->device_cap_flags;
90 	props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
91 		0xffffff;
92 	props->vendor_part_id      = be16_to_cpup((__be16 *) (out_mad->data + 30));
93 	props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
94 	memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
95 
96 	props->max_mr_size         = ~0ull;
97 	props->page_size_cap       = mdev->limits.page_size_cap;
98 	props->max_qp              = mdev->limits.num_qps - mdev->limits.reserved_qps;
99 	props->max_qp_wr           = mdev->limits.max_wqes;
100 	props->max_sge             = mdev->limits.max_sg;
101 	props->max_cq              = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
102 	props->max_cqe             = mdev->limits.max_cqes;
103 	props->max_mr              = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
104 	props->max_pd              = mdev->limits.num_pds - mdev->limits.reserved_pds;
105 	props->max_qp_rd_atom      = 1 << mdev->qp_table.rdb_shift;
106 	props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
107 	props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
108 	props->max_srq             = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
109 	props->max_srq_wr          = mdev->limits.max_srq_wqes;
110 	props->max_srq_sge         = mdev->limits.max_srq_sge;
111 	props->local_ca_ack_delay  = mdev->limits.local_ca_ack_delay;
112 	props->atomic_cap          = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
113 					IB_ATOMIC_HCA : IB_ATOMIC_NONE;
114 	props->max_pkeys           = mdev->limits.pkey_table_len;
115 	props->max_mcast_grp       = mdev->limits.num_mgms + mdev->limits.num_amgms;
116 	props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
117 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
118 					   props->max_mcast_grp;
119 	/*
120 	 * If Sinai memory key optimization is being used, then only
121 	 * the 8-bit key portion will change.  For other HCAs, the
122 	 * unused index bits will also be used for FMR remapping.
123 	 */
124 	if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
125 		props->max_map_per_fmr = 255;
126 	else
127 		props->max_map_per_fmr =
128 			(1 << (32 - ilog2(mdev->limits.num_mpts))) - 1;
129 
130 	err = 0;
131  out:
132 	kfree(in_mad);
133 	kfree(out_mad);
134 	return err;
135 }
136 
137 static int mthca_query_port(struct ib_device *ibdev,
138 			    u8 port, struct ib_port_attr *props)
139 {
140 	struct ib_smp *in_mad  = NULL;
141 	struct ib_smp *out_mad = NULL;
142 	int err = -ENOMEM;
143 	u8 status;
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 	memset(props, 0, sizeof *props);
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 			    &status);
159 	if (err)
160 		goto out;
161 	if (status) {
162 		err = -EINVAL;
163 		goto out;
164 	}
165 
166 	props->lid               = be16_to_cpup((__be16 *) (out_mad->data + 16));
167 	props->lmc               = out_mad->data[34] & 0x7;
168 	props->sm_lid            = be16_to_cpup((__be16 *) (out_mad->data + 18));
169 	props->sm_sl             = out_mad->data[36] & 0xf;
170 	props->state             = out_mad->data[32] & 0xf;
171 	props->phys_state        = out_mad->data[33] >> 4;
172 	props->port_cap_flags    = be32_to_cpup((__be32 *) (out_mad->data + 20));
173 	props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
174 	props->max_msg_sz        = 0x80000000;
175 	props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
176 	props->bad_pkey_cntr     = be16_to_cpup((__be16 *) (out_mad->data + 46));
177 	props->qkey_viol_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 48));
178 	props->active_width      = out_mad->data[31] & 0xf;
179 	props->active_speed      = out_mad->data[35] >> 4;
180 	props->max_mtu           = out_mad->data[41] & 0xf;
181 	props->active_mtu        = out_mad->data[36] >> 4;
182 	props->subnet_timeout    = out_mad->data[51] & 0x1f;
183 	props->max_vl_num        = out_mad->data[37] >> 4;
184 	props->init_type_reply   = out_mad->data[41] >> 4;
185 
186  out:
187 	kfree(in_mad);
188 	kfree(out_mad);
189 	return err;
190 }
191 
192 static int mthca_modify_device(struct ib_device *ibdev,
193 			       int mask,
194 			       struct ib_device_modify *props)
195 {
196 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
197 		return -EOPNOTSUPP;
198 
199 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
200 		if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
201 			return -ERESTARTSYS;
202 		memcpy(ibdev->node_desc, props->node_desc, 64);
203 		mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
204 	}
205 
206 	return 0;
207 }
208 
209 static int mthca_modify_port(struct ib_device *ibdev,
210 			     u8 port, int port_modify_mask,
211 			     struct ib_port_modify *props)
212 {
213 	struct mthca_set_ib_param set_ib;
214 	struct ib_port_attr attr;
215 	int err;
216 	u8 status;
217 
218 	if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
219 		return -ERESTARTSYS;
220 
221 	err = mthca_query_port(ibdev, port, &attr);
222 	if (err)
223 		goto out;
224 
225 	set_ib.set_si_guid     = 0;
226 	set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
227 
228 	set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
229 		~props->clr_port_cap_mask;
230 
231 	err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port, &status);
232 	if (err)
233 		goto out;
234 	if (status) {
235 		err = -EINVAL;
236 		goto out;
237 	}
238 
239 out:
240 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
241 	return err;
242 }
243 
244 static int mthca_query_pkey(struct ib_device *ibdev,
245 			    u8 port, u16 index, u16 *pkey)
246 {
247 	struct ib_smp *in_mad  = NULL;
248 	struct ib_smp *out_mad = NULL;
249 	int err = -ENOMEM;
250 	u8 status;
251 
252 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
253 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
254 	if (!in_mad || !out_mad)
255 		goto out;
256 
257 	init_query_mad(in_mad);
258 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
259 	in_mad->attr_mod = cpu_to_be32(index / 32);
260 
261 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
262 			    port, NULL, NULL, in_mad, out_mad,
263 			    &status);
264 	if (err)
265 		goto out;
266 	if (status) {
267 		err = -EINVAL;
268 		goto out;
269 	}
270 
271 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
272 
273  out:
274 	kfree(in_mad);
275 	kfree(out_mad);
276 	return err;
277 }
278 
279 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
280 			   int index, union ib_gid *gid)
281 {
282 	struct ib_smp *in_mad  = NULL;
283 	struct ib_smp *out_mad = NULL;
284 	int err = -ENOMEM;
285 	u8 status;
286 
287 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
288 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
289 	if (!in_mad || !out_mad)
290 		goto out;
291 
292 	init_query_mad(in_mad);
293 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
294 	in_mad->attr_mod = cpu_to_be32(port);
295 
296 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
297 			    port, NULL, NULL, in_mad, out_mad,
298 			    &status);
299 	if (err)
300 		goto out;
301 	if (status) {
302 		err = -EINVAL;
303 		goto out;
304 	}
305 
306 	memcpy(gid->raw, out_mad->data + 8, 8);
307 
308 	init_query_mad(in_mad);
309 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
310 	in_mad->attr_mod = cpu_to_be32(index / 8);
311 
312 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
313 			    port, NULL, NULL, in_mad, out_mad,
314 			    &status);
315 	if (err)
316 		goto out;
317 	if (status) {
318 		err = -EINVAL;
319 		goto out;
320 	}
321 
322 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
323 
324  out:
325 	kfree(in_mad);
326 	kfree(out_mad);
327 	return err;
328 }
329 
330 static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
331 						struct ib_udata *udata)
332 {
333 	struct mthca_alloc_ucontext_resp uresp;
334 	struct mthca_ucontext           *context;
335 	int                              err;
336 
337 	memset(&uresp, 0, sizeof uresp);
338 
339 	uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
340 	if (mthca_is_memfree(to_mdev(ibdev)))
341 		uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
342 	else
343 		uresp.uarc_size = 0;
344 
345 	context = kmalloc(sizeof *context, GFP_KERNEL);
346 	if (!context)
347 		return ERR_PTR(-ENOMEM);
348 
349 	err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
350 	if (err) {
351 		kfree(context);
352 		return ERR_PTR(err);
353 	}
354 
355 	context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
356 	if (IS_ERR(context->db_tab)) {
357 		err = PTR_ERR(context->db_tab);
358 		mthca_uar_free(to_mdev(ibdev), &context->uar);
359 		kfree(context);
360 		return ERR_PTR(err);
361 	}
362 
363 	if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) {
364 		mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
365 		mthca_uar_free(to_mdev(ibdev), &context->uar);
366 		kfree(context);
367 		return ERR_PTR(-EFAULT);
368 	}
369 
370 	return &context->ibucontext;
371 }
372 
373 static int mthca_dealloc_ucontext(struct ib_ucontext *context)
374 {
375 	mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
376 				  to_mucontext(context)->db_tab);
377 	mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
378 	kfree(to_mucontext(context));
379 
380 	return 0;
381 }
382 
383 static int mthca_mmap_uar(struct ib_ucontext *context,
384 			  struct vm_area_struct *vma)
385 {
386 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
387 		return -EINVAL;
388 
389 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
390 
391 	if (io_remap_pfn_range(vma, vma->vm_start,
392 			       to_mucontext(context)->uar.pfn,
393 			       PAGE_SIZE, vma->vm_page_prot))
394 		return -EAGAIN;
395 
396 	return 0;
397 }
398 
399 static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev,
400 				    struct ib_ucontext *context,
401 				    struct ib_udata *udata)
402 {
403 	struct mthca_pd *pd;
404 	int err;
405 
406 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
407 	if (!pd)
408 		return ERR_PTR(-ENOMEM);
409 
410 	err = mthca_pd_alloc(to_mdev(ibdev), !context, pd);
411 	if (err) {
412 		kfree(pd);
413 		return ERR_PTR(err);
414 	}
415 
416 	if (context) {
417 		if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
418 			mthca_pd_free(to_mdev(ibdev), pd);
419 			kfree(pd);
420 			return ERR_PTR(-EFAULT);
421 		}
422 	}
423 
424 	return &pd->ibpd;
425 }
426 
427 static int mthca_dealloc_pd(struct ib_pd *pd)
428 {
429 	mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
430 	kfree(pd);
431 
432 	return 0;
433 }
434 
435 static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
436 				     struct ib_ah_attr *ah_attr)
437 {
438 	int err;
439 	struct mthca_ah *ah;
440 
441 	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
442 	if (!ah)
443 		return ERR_PTR(-ENOMEM);
444 
445 	err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
446 	if (err) {
447 		kfree(ah);
448 		return ERR_PTR(err);
449 	}
450 
451 	return &ah->ibah;
452 }
453 
454 static int mthca_ah_destroy(struct ib_ah *ah)
455 {
456 	mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
457 	kfree(ah);
458 
459 	return 0;
460 }
461 
462 static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
463 				       struct ib_srq_init_attr *init_attr,
464 				       struct ib_udata *udata)
465 {
466 	struct mthca_create_srq ucmd;
467 	struct mthca_ucontext *context = NULL;
468 	struct mthca_srq *srq;
469 	int err;
470 
471 	srq = kmalloc(sizeof *srq, GFP_KERNEL);
472 	if (!srq)
473 		return ERR_PTR(-ENOMEM);
474 
475 	if (pd->uobject) {
476 		context = to_mucontext(pd->uobject->context);
477 
478 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
479 			err = -EFAULT;
480 			goto err_free;
481 		}
482 
483 		err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
484 					context->db_tab, ucmd.db_index,
485 					ucmd.db_page);
486 
487 		if (err)
488 			goto err_free;
489 
490 		srq->mr.ibmr.lkey = ucmd.lkey;
491 		srq->db_index     = ucmd.db_index;
492 	}
493 
494 	err = mthca_alloc_srq(to_mdev(pd->device), to_mpd(pd),
495 			      &init_attr->attr, srq);
496 
497 	if (err && pd->uobject)
498 		mthca_unmap_user_db(to_mdev(pd->device), &context->uar,
499 				    context->db_tab, ucmd.db_index);
500 
501 	if (err)
502 		goto err_free;
503 
504 	if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof (__u32))) {
505 		mthca_free_srq(to_mdev(pd->device), srq);
506 		err = -EFAULT;
507 		goto err_free;
508 	}
509 
510 	return &srq->ibsrq;
511 
512 err_free:
513 	kfree(srq);
514 
515 	return ERR_PTR(err);
516 }
517 
518 static int mthca_destroy_srq(struct ib_srq *srq)
519 {
520 	struct mthca_ucontext *context;
521 
522 	if (srq->uobject) {
523 		context = to_mucontext(srq->uobject->context);
524 
525 		mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
526 				    context->db_tab, to_msrq(srq)->db_index);
527 	}
528 
529 	mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
530 	kfree(srq);
531 
532 	return 0;
533 }
534 
535 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
536 				     struct ib_qp_init_attr *init_attr,
537 				     struct ib_udata *udata)
538 {
539 	struct mthca_create_qp ucmd;
540 	struct mthca_qp *qp;
541 	int err;
542 
543 	if (init_attr->create_flags)
544 		return ERR_PTR(-EINVAL);
545 
546 	switch (init_attr->qp_type) {
547 	case IB_QPT_RC:
548 	case IB_QPT_UC:
549 	case IB_QPT_UD:
550 	{
551 		struct mthca_ucontext *context;
552 
553 		qp = kmalloc(sizeof *qp, GFP_KERNEL);
554 		if (!qp)
555 			return ERR_PTR(-ENOMEM);
556 
557 		if (pd->uobject) {
558 			context = to_mucontext(pd->uobject->context);
559 
560 			if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
561 				kfree(qp);
562 				return ERR_PTR(-EFAULT);
563 			}
564 
565 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
566 						context->db_tab,
567 						ucmd.sq_db_index, ucmd.sq_db_page);
568 			if (err) {
569 				kfree(qp);
570 				return ERR_PTR(err);
571 			}
572 
573 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
574 						context->db_tab,
575 						ucmd.rq_db_index, ucmd.rq_db_page);
576 			if (err) {
577 				mthca_unmap_user_db(to_mdev(pd->device),
578 						    &context->uar,
579 						    context->db_tab,
580 						    ucmd.sq_db_index);
581 				kfree(qp);
582 				return ERR_PTR(err);
583 			}
584 
585 			qp->mr.ibmr.lkey = ucmd.lkey;
586 			qp->sq.db_index  = ucmd.sq_db_index;
587 			qp->rq.db_index  = ucmd.rq_db_index;
588 		}
589 
590 		err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
591 				     to_mcq(init_attr->send_cq),
592 				     to_mcq(init_attr->recv_cq),
593 				     init_attr->qp_type, init_attr->sq_sig_type,
594 				     &init_attr->cap, qp);
595 
596 		if (err && pd->uobject) {
597 			context = to_mucontext(pd->uobject->context);
598 
599 			mthca_unmap_user_db(to_mdev(pd->device),
600 					    &context->uar,
601 					    context->db_tab,
602 					    ucmd.sq_db_index);
603 			mthca_unmap_user_db(to_mdev(pd->device),
604 					    &context->uar,
605 					    context->db_tab,
606 					    ucmd.rq_db_index);
607 		}
608 
609 		qp->ibqp.qp_num = qp->qpn;
610 		break;
611 	}
612 	case IB_QPT_SMI:
613 	case IB_QPT_GSI:
614 	{
615 		/* Don't allow userspace to create special QPs */
616 		if (pd->uobject)
617 			return ERR_PTR(-EINVAL);
618 
619 		qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
620 		if (!qp)
621 			return ERR_PTR(-ENOMEM);
622 
623 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
624 
625 		err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
626 				      to_mcq(init_attr->send_cq),
627 				      to_mcq(init_attr->recv_cq),
628 				      init_attr->sq_sig_type, &init_attr->cap,
629 				      qp->ibqp.qp_num, init_attr->port_num,
630 				      to_msqp(qp));
631 		break;
632 	}
633 	default:
634 		/* Don't support raw QPs */
635 		return ERR_PTR(-ENOSYS);
636 	}
637 
638 	if (err) {
639 		kfree(qp);
640 		return ERR_PTR(err);
641 	}
642 
643 	init_attr->cap.max_send_wr     = qp->sq.max;
644 	init_attr->cap.max_recv_wr     = qp->rq.max;
645 	init_attr->cap.max_send_sge    = qp->sq.max_gs;
646 	init_attr->cap.max_recv_sge    = qp->rq.max_gs;
647 	init_attr->cap.max_inline_data = qp->max_inline_data;
648 
649 	return &qp->ibqp;
650 }
651 
652 static int mthca_destroy_qp(struct ib_qp *qp)
653 {
654 	if (qp->uobject) {
655 		mthca_unmap_user_db(to_mdev(qp->device),
656 				    &to_mucontext(qp->uobject->context)->uar,
657 				    to_mucontext(qp->uobject->context)->db_tab,
658 				    to_mqp(qp)->sq.db_index);
659 		mthca_unmap_user_db(to_mdev(qp->device),
660 				    &to_mucontext(qp->uobject->context)->uar,
661 				    to_mucontext(qp->uobject->context)->db_tab,
662 				    to_mqp(qp)->rq.db_index);
663 	}
664 	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
665 	kfree(qp);
666 	return 0;
667 }
668 
669 static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries,
670 				     int comp_vector,
671 				     struct ib_ucontext *context,
672 				     struct ib_udata *udata)
673 {
674 	struct mthca_create_cq ucmd;
675 	struct mthca_cq *cq;
676 	int nent;
677 	int err;
678 
679 	if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
680 		return ERR_PTR(-EINVAL);
681 
682 	if (context) {
683 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
684 			return ERR_PTR(-EFAULT);
685 
686 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
687 					to_mucontext(context)->db_tab,
688 					ucmd.set_db_index, ucmd.set_db_page);
689 		if (err)
690 			return ERR_PTR(err);
691 
692 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
693 					to_mucontext(context)->db_tab,
694 					ucmd.arm_db_index, ucmd.arm_db_page);
695 		if (err)
696 			goto err_unmap_set;
697 	}
698 
699 	cq = kmalloc(sizeof *cq, GFP_KERNEL);
700 	if (!cq) {
701 		err = -ENOMEM;
702 		goto err_unmap_arm;
703 	}
704 
705 	if (context) {
706 		cq->buf.mr.ibmr.lkey = ucmd.lkey;
707 		cq->set_ci_db_index  = ucmd.set_db_index;
708 		cq->arm_db_index     = ucmd.arm_db_index;
709 	}
710 
711 	for (nent = 1; nent <= entries; nent <<= 1)
712 		; /* nothing */
713 
714 	err = mthca_init_cq(to_mdev(ibdev), nent,
715 			    context ? to_mucontext(context) : NULL,
716 			    context ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
717 			    cq);
718 	if (err)
719 		goto err_free;
720 
721 	if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
722 		mthca_free_cq(to_mdev(ibdev), cq);
723 		goto err_free;
724 	}
725 
726 	cq->resize_buf = NULL;
727 
728 	return &cq->ibcq;
729 
730 err_free:
731 	kfree(cq);
732 
733 err_unmap_arm:
734 	if (context)
735 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
736 				    to_mucontext(context)->db_tab, ucmd.arm_db_index);
737 
738 err_unmap_set:
739 	if (context)
740 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
741 				    to_mucontext(context)->db_tab, ucmd.set_db_index);
742 
743 	return ERR_PTR(err);
744 }
745 
746 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
747 				  int entries)
748 {
749 	int ret;
750 
751 	spin_lock_irq(&cq->lock);
752 	if (cq->resize_buf) {
753 		ret = -EBUSY;
754 		goto unlock;
755 	}
756 
757 	cq->resize_buf = kmalloc(sizeof *cq->resize_buf, GFP_ATOMIC);
758 	if (!cq->resize_buf) {
759 		ret = -ENOMEM;
760 		goto unlock;
761 	}
762 
763 	cq->resize_buf->state = CQ_RESIZE_ALLOC;
764 
765 	ret = 0;
766 
767 unlock:
768 	spin_unlock_irq(&cq->lock);
769 
770 	if (ret)
771 		return ret;
772 
773 	ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
774 	if (ret) {
775 		spin_lock_irq(&cq->lock);
776 		kfree(cq->resize_buf);
777 		cq->resize_buf = NULL;
778 		spin_unlock_irq(&cq->lock);
779 		return ret;
780 	}
781 
782 	cq->resize_buf->cqe = entries - 1;
783 
784 	spin_lock_irq(&cq->lock);
785 	cq->resize_buf->state = CQ_RESIZE_READY;
786 	spin_unlock_irq(&cq->lock);
787 
788 	return 0;
789 }
790 
791 static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
792 {
793 	struct mthca_dev *dev = to_mdev(ibcq->device);
794 	struct mthca_cq *cq = to_mcq(ibcq);
795 	struct mthca_resize_cq ucmd;
796 	u32 lkey;
797 	u8 status;
798 	int ret;
799 
800 	if (entries < 1 || entries > dev->limits.max_cqes)
801 		return -EINVAL;
802 
803 	mutex_lock(&cq->mutex);
804 
805 	entries = roundup_pow_of_two(entries + 1);
806 	if (entries == ibcq->cqe + 1) {
807 		ret = 0;
808 		goto out;
809 	}
810 
811 	if (cq->is_kernel) {
812 		ret = mthca_alloc_resize_buf(dev, cq, entries);
813 		if (ret)
814 			goto out;
815 		lkey = cq->resize_buf->buf.mr.ibmr.lkey;
816 	} else {
817 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
818 			ret = -EFAULT;
819 			goto out;
820 		}
821 		lkey = ucmd.lkey;
822 	}
823 
824 	ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries), &status);
825 	if (status)
826 		ret = -EINVAL;
827 
828 	if (ret) {
829 		if (cq->resize_buf) {
830 			mthca_free_cq_buf(dev, &cq->resize_buf->buf,
831 					  cq->resize_buf->cqe);
832 			kfree(cq->resize_buf);
833 			spin_lock_irq(&cq->lock);
834 			cq->resize_buf = NULL;
835 			spin_unlock_irq(&cq->lock);
836 		}
837 		goto out;
838 	}
839 
840 	if (cq->is_kernel) {
841 		struct mthca_cq_buf tbuf;
842 		int tcqe;
843 
844 		spin_lock_irq(&cq->lock);
845 		if (cq->resize_buf->state == CQ_RESIZE_READY) {
846 			mthca_cq_resize_copy_cqes(cq);
847 			tbuf         = cq->buf;
848 			tcqe         = cq->ibcq.cqe;
849 			cq->buf      = cq->resize_buf->buf;
850 			cq->ibcq.cqe = cq->resize_buf->cqe;
851 		} else {
852 			tbuf = cq->resize_buf->buf;
853 			tcqe = cq->resize_buf->cqe;
854 		}
855 
856 		kfree(cq->resize_buf);
857 		cq->resize_buf = NULL;
858 		spin_unlock_irq(&cq->lock);
859 
860 		mthca_free_cq_buf(dev, &tbuf, tcqe);
861 	} else
862 		ibcq->cqe = entries - 1;
863 
864 out:
865 	mutex_unlock(&cq->mutex);
866 
867 	return ret;
868 }
869 
870 static int mthca_destroy_cq(struct ib_cq *cq)
871 {
872 	if (cq->uobject) {
873 		mthca_unmap_user_db(to_mdev(cq->device),
874 				    &to_mucontext(cq->uobject->context)->uar,
875 				    to_mucontext(cq->uobject->context)->db_tab,
876 				    to_mcq(cq)->arm_db_index);
877 		mthca_unmap_user_db(to_mdev(cq->device),
878 				    &to_mucontext(cq->uobject->context)->uar,
879 				    to_mucontext(cq->uobject->context)->db_tab,
880 				    to_mcq(cq)->set_ci_db_index);
881 	}
882 	mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
883 	kfree(cq);
884 
885 	return 0;
886 }
887 
888 static inline u32 convert_access(int acc)
889 {
890 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
891 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
892 	       (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
893 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
894 	       MTHCA_MPT_FLAG_LOCAL_READ;
895 }
896 
897 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
898 {
899 	struct mthca_mr *mr;
900 	int err;
901 
902 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
903 	if (!mr)
904 		return ERR_PTR(-ENOMEM);
905 
906 	err = mthca_mr_alloc_notrans(to_mdev(pd->device),
907 				     to_mpd(pd)->pd_num,
908 				     convert_access(acc), mr);
909 
910 	if (err) {
911 		kfree(mr);
912 		return ERR_PTR(err);
913 	}
914 
915 	mr->umem = NULL;
916 
917 	return &mr->ibmr;
918 }
919 
920 static struct ib_mr *mthca_reg_phys_mr(struct ib_pd       *pd,
921 				       struct ib_phys_buf *buffer_list,
922 				       int                 num_phys_buf,
923 				       int                 acc,
924 				       u64                *iova_start)
925 {
926 	struct mthca_mr *mr;
927 	u64 *page_list;
928 	u64 total_size;
929 	unsigned long mask;
930 	int shift;
931 	int npages;
932 	int err;
933 	int i, j, n;
934 
935 	mask = buffer_list[0].addr ^ *iova_start;
936 	total_size = 0;
937 	for (i = 0; i < num_phys_buf; ++i) {
938 		if (i != 0)
939 			mask |= buffer_list[i].addr;
940 		if (i != num_phys_buf - 1)
941 			mask |= buffer_list[i].addr + buffer_list[i].size;
942 
943 		total_size += buffer_list[i].size;
944 	}
945 
946 	if (mask & ~PAGE_MASK)
947 		return ERR_PTR(-EINVAL);
948 
949 	shift = __ffs(mask | 1 << 31);
950 
951 	buffer_list[0].size += buffer_list[0].addr & ((1ULL << shift) - 1);
952 	buffer_list[0].addr &= ~0ull << shift;
953 
954 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
955 	if (!mr)
956 		return ERR_PTR(-ENOMEM);
957 
958 	npages = 0;
959 	for (i = 0; i < num_phys_buf; ++i)
960 		npages += (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
961 
962 	if (!npages)
963 		return &mr->ibmr;
964 
965 	page_list = kmalloc(npages * sizeof *page_list, GFP_KERNEL);
966 	if (!page_list) {
967 		kfree(mr);
968 		return ERR_PTR(-ENOMEM);
969 	}
970 
971 	n = 0;
972 	for (i = 0; i < num_phys_buf; ++i)
973 		for (j = 0;
974 		     j < (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
975 		     ++j)
976 			page_list[n++] = buffer_list[i].addr + ((u64) j << shift);
977 
978 	mthca_dbg(to_mdev(pd->device), "Registering memory at %llx (iova %llx) "
979 		  "in PD %x; shift %d, npages %d.\n",
980 		  (unsigned long long) buffer_list[0].addr,
981 		  (unsigned long long) *iova_start,
982 		  to_mpd(pd)->pd_num,
983 		  shift, npages);
984 
985 	err = mthca_mr_alloc_phys(to_mdev(pd->device),
986 				  to_mpd(pd)->pd_num,
987 				  page_list, shift, npages,
988 				  *iova_start, total_size,
989 				  convert_access(acc), mr);
990 
991 	if (err) {
992 		kfree(page_list);
993 		kfree(mr);
994 		return ERR_PTR(err);
995 	}
996 
997 	kfree(page_list);
998 	mr->umem = NULL;
999 
1000 	return &mr->ibmr;
1001 }
1002 
1003 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1004 				       u64 virt, int acc, struct ib_udata *udata)
1005 {
1006 	struct mthca_dev *dev = to_mdev(pd->device);
1007 	struct ib_umem_chunk *chunk;
1008 	struct mthca_mr *mr;
1009 	u64 *pages;
1010 	int shift, n, len;
1011 	int i, j, k;
1012 	int err = 0;
1013 	int write_mtt_size;
1014 
1015 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
1016 	if (!mr)
1017 		return ERR_PTR(-ENOMEM);
1018 
1019 	mr->umem = ib_umem_get(pd->uobject->context, start, length, acc);
1020 	if (IS_ERR(mr->umem)) {
1021 		err = PTR_ERR(mr->umem);
1022 		goto err;
1023 	}
1024 
1025 	shift = ffs(mr->umem->page_size) - 1;
1026 
1027 	n = 0;
1028 	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
1029 		n += chunk->nents;
1030 
1031 	mr->mtt = mthca_alloc_mtt(dev, n);
1032 	if (IS_ERR(mr->mtt)) {
1033 		err = PTR_ERR(mr->mtt);
1034 		goto err_umem;
1035 	}
1036 
1037 	pages = (u64 *) __get_free_page(GFP_KERNEL);
1038 	if (!pages) {
1039 		err = -ENOMEM;
1040 		goto err_mtt;
1041 	}
1042 
1043 	i = n = 0;
1044 
1045 	write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
1046 
1047 	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
1048 		for (j = 0; j < chunk->nmap; ++j) {
1049 			len = sg_dma_len(&chunk->page_list[j]) >> shift;
1050 			for (k = 0; k < len; ++k) {
1051 				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
1052 					mr->umem->page_size * k;
1053 				/*
1054 				 * Be friendly to write_mtt and pass it chunks
1055 				 * of appropriate size.
1056 				 */
1057 				if (i == write_mtt_size) {
1058 					err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
1059 					if (err)
1060 						goto mtt_done;
1061 					n += i;
1062 					i = 0;
1063 				}
1064 			}
1065 		}
1066 
1067 	if (i)
1068 		err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
1069 mtt_done:
1070 	free_page((unsigned long) pages);
1071 	if (err)
1072 		goto err_mtt;
1073 
1074 	err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, virt, length,
1075 			     convert_access(acc), mr);
1076 
1077 	if (err)
1078 		goto err_mtt;
1079 
1080 	return &mr->ibmr;
1081 
1082 err_mtt:
1083 	mthca_free_mtt(dev, mr->mtt);
1084 
1085 err_umem:
1086 	ib_umem_release(mr->umem);
1087 
1088 err:
1089 	kfree(mr);
1090 	return ERR_PTR(err);
1091 }
1092 
1093 static int mthca_dereg_mr(struct ib_mr *mr)
1094 {
1095 	struct mthca_mr *mmr = to_mmr(mr);
1096 
1097 	mthca_free_mr(to_mdev(mr->device), mmr);
1098 	if (mmr->umem)
1099 		ib_umem_release(mmr->umem);
1100 	kfree(mmr);
1101 
1102 	return 0;
1103 }
1104 
1105 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1106 				      struct ib_fmr_attr *fmr_attr)
1107 {
1108 	struct mthca_fmr *fmr;
1109 	int err;
1110 
1111 	fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
1112 	if (!fmr)
1113 		return ERR_PTR(-ENOMEM);
1114 
1115 	memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
1116 	err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
1117 			     convert_access(mr_access_flags), fmr);
1118 
1119 	if (err) {
1120 		kfree(fmr);
1121 		return ERR_PTR(err);
1122 	}
1123 
1124 	return &fmr->ibmr;
1125 }
1126 
1127 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
1128 {
1129 	struct mthca_fmr *mfmr = to_mfmr(fmr);
1130 	int err;
1131 
1132 	err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
1133 	if (err)
1134 		return err;
1135 
1136 	kfree(mfmr);
1137 	return 0;
1138 }
1139 
1140 static int mthca_unmap_fmr(struct list_head *fmr_list)
1141 {
1142 	struct ib_fmr *fmr;
1143 	int err;
1144 	u8 status;
1145 	struct mthca_dev *mdev = NULL;
1146 
1147 	list_for_each_entry(fmr, fmr_list, list) {
1148 		if (mdev && to_mdev(fmr->device) != mdev)
1149 			return -EINVAL;
1150 		mdev = to_mdev(fmr->device);
1151 	}
1152 
1153 	if (!mdev)
1154 		return 0;
1155 
1156 	if (mthca_is_memfree(mdev)) {
1157 		list_for_each_entry(fmr, fmr_list, list)
1158 			mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1159 
1160 		wmb();
1161 	} else
1162 		list_for_each_entry(fmr, fmr_list, list)
1163 			mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1164 
1165 	err = mthca_SYNC_TPT(mdev, &status);
1166 	if (err)
1167 		return err;
1168 	if (status)
1169 		return -EINVAL;
1170 	return 0;
1171 }
1172 
1173 static ssize_t show_rev(struct class_device *cdev, char *buf)
1174 {
1175 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1176 	return sprintf(buf, "%x\n", dev->rev_id);
1177 }
1178 
1179 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
1180 {
1181 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1182 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->fw_ver >> 32),
1183 		       (int) (dev->fw_ver >> 16) & 0xffff,
1184 		       (int) dev->fw_ver & 0xffff);
1185 }
1186 
1187 static ssize_t show_hca(struct class_device *cdev, char *buf)
1188 {
1189 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1190 	switch (dev->pdev->device) {
1191 	case PCI_DEVICE_ID_MELLANOX_TAVOR:
1192 		return sprintf(buf, "MT23108\n");
1193 	case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1194 		return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1195 	case PCI_DEVICE_ID_MELLANOX_ARBEL:
1196 		return sprintf(buf, "MT25208\n");
1197 	case PCI_DEVICE_ID_MELLANOX_SINAI:
1198 	case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1199 		return sprintf(buf, "MT25204\n");
1200 	default:
1201 		return sprintf(buf, "unknown\n");
1202 	}
1203 }
1204 
1205 static ssize_t show_board(struct class_device *cdev, char *buf)
1206 {
1207 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1208 	return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1209 }
1210 
1211 static CLASS_DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1212 static CLASS_DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1213 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1214 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1215 
1216 static struct class_device_attribute *mthca_class_attributes[] = {
1217 	&class_device_attr_hw_rev,
1218 	&class_device_attr_fw_ver,
1219 	&class_device_attr_hca_type,
1220 	&class_device_attr_board_id
1221 };
1222 
1223 static int mthca_init_node_data(struct mthca_dev *dev)
1224 {
1225 	struct ib_smp *in_mad  = NULL;
1226 	struct ib_smp *out_mad = NULL;
1227 	int err = -ENOMEM;
1228 	u8 status;
1229 
1230 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1231 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1232 	if (!in_mad || !out_mad)
1233 		goto out;
1234 
1235 	init_query_mad(in_mad);
1236 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1237 
1238 	err = mthca_MAD_IFC(dev, 1, 1,
1239 			    1, NULL, NULL, in_mad, out_mad,
1240 			    &status);
1241 	if (err)
1242 		goto out;
1243 	if (status) {
1244 		err = -EINVAL;
1245 		goto out;
1246 	}
1247 
1248 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1249 
1250 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1251 
1252 	err = mthca_MAD_IFC(dev, 1, 1,
1253 			    1, NULL, NULL, in_mad, out_mad,
1254 			    &status);
1255 	if (err)
1256 		goto out;
1257 	if (status) {
1258 		err = -EINVAL;
1259 		goto out;
1260 	}
1261 
1262 	if (mthca_is_memfree(dev))
1263 		dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1264 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1265 
1266 out:
1267 	kfree(in_mad);
1268 	kfree(out_mad);
1269 	return err;
1270 }
1271 
1272 int mthca_register_device(struct mthca_dev *dev)
1273 {
1274 	int ret;
1275 	int i;
1276 
1277 	ret = mthca_init_node_data(dev);
1278 	if (ret)
1279 		return ret;
1280 
1281 	strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
1282 	dev->ib_dev.owner                = THIS_MODULE;
1283 
1284 	dev->ib_dev.uverbs_abi_ver	 = MTHCA_UVERBS_ABI_VERSION;
1285 	dev->ib_dev.uverbs_cmd_mask	 =
1286 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1287 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1288 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1289 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1290 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1291 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1292 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1293 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1294 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1295 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1296 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1297 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1298 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1299 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1300 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1301 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1302 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1303 	dev->ib_dev.node_type            = RDMA_NODE_IB_CA;
1304 	dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
1305 	dev->ib_dev.num_comp_vectors     = 1;
1306 	dev->ib_dev.dma_device           = &dev->pdev->dev;
1307 	dev->ib_dev.query_device         = mthca_query_device;
1308 	dev->ib_dev.query_port           = mthca_query_port;
1309 	dev->ib_dev.modify_device        = mthca_modify_device;
1310 	dev->ib_dev.modify_port          = mthca_modify_port;
1311 	dev->ib_dev.query_pkey           = mthca_query_pkey;
1312 	dev->ib_dev.query_gid            = mthca_query_gid;
1313 	dev->ib_dev.alloc_ucontext       = mthca_alloc_ucontext;
1314 	dev->ib_dev.dealloc_ucontext     = mthca_dealloc_ucontext;
1315 	dev->ib_dev.mmap                 = mthca_mmap_uar;
1316 	dev->ib_dev.alloc_pd             = mthca_alloc_pd;
1317 	dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
1318 	dev->ib_dev.create_ah            = mthca_ah_create;
1319 	dev->ib_dev.query_ah             = mthca_ah_query;
1320 	dev->ib_dev.destroy_ah           = mthca_ah_destroy;
1321 
1322 	if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1323 		dev->ib_dev.create_srq           = mthca_create_srq;
1324 		dev->ib_dev.modify_srq           = mthca_modify_srq;
1325 		dev->ib_dev.query_srq            = mthca_query_srq;
1326 		dev->ib_dev.destroy_srq          = mthca_destroy_srq;
1327 		dev->ib_dev.uverbs_cmd_mask	|=
1328 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1329 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1330 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1331 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1332 
1333 		if (mthca_is_memfree(dev))
1334 			dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1335 		else
1336 			dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1337 	}
1338 
1339 	dev->ib_dev.create_qp            = mthca_create_qp;
1340 	dev->ib_dev.modify_qp            = mthca_modify_qp;
1341 	dev->ib_dev.query_qp             = mthca_query_qp;
1342 	dev->ib_dev.destroy_qp           = mthca_destroy_qp;
1343 	dev->ib_dev.create_cq            = mthca_create_cq;
1344 	dev->ib_dev.resize_cq            = mthca_resize_cq;
1345 	dev->ib_dev.destroy_cq           = mthca_destroy_cq;
1346 	dev->ib_dev.poll_cq              = mthca_poll_cq;
1347 	dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
1348 	dev->ib_dev.reg_phys_mr          = mthca_reg_phys_mr;
1349 	dev->ib_dev.reg_user_mr          = mthca_reg_user_mr;
1350 	dev->ib_dev.dereg_mr             = mthca_dereg_mr;
1351 
1352 	if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1353 		dev->ib_dev.alloc_fmr            = mthca_alloc_fmr;
1354 		dev->ib_dev.unmap_fmr            = mthca_unmap_fmr;
1355 		dev->ib_dev.dealloc_fmr          = mthca_dealloc_fmr;
1356 		if (mthca_is_memfree(dev))
1357 			dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1358 		else
1359 			dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1360 	}
1361 
1362 	dev->ib_dev.attach_mcast         = mthca_multicast_attach;
1363 	dev->ib_dev.detach_mcast         = mthca_multicast_detach;
1364 	dev->ib_dev.process_mad          = mthca_process_mad;
1365 
1366 	if (mthca_is_memfree(dev)) {
1367 		dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1368 		dev->ib_dev.post_send     = mthca_arbel_post_send;
1369 		dev->ib_dev.post_recv     = mthca_arbel_post_receive;
1370 	} else {
1371 		dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1372 		dev->ib_dev.post_send     = mthca_tavor_post_send;
1373 		dev->ib_dev.post_recv     = mthca_tavor_post_receive;
1374 	}
1375 
1376 	mutex_init(&dev->cap_mask_mutex);
1377 
1378 	ret = ib_register_device(&dev->ib_dev);
1379 	if (ret)
1380 		return ret;
1381 
1382 	for (i = 0; i < ARRAY_SIZE(mthca_class_attributes); ++i) {
1383 		ret = class_device_create_file(&dev->ib_dev.class_dev,
1384 					       mthca_class_attributes[i]);
1385 		if (ret) {
1386 			ib_unregister_device(&dev->ib_dev);
1387 			return ret;
1388 		}
1389 	}
1390 
1391 	mthca_start_catas_poll(dev);
1392 
1393 	return 0;
1394 }
1395 
1396 void mthca_unregister_device(struct mthca_dev *dev)
1397 {
1398 	mthca_stop_catas_poll(dev);
1399 	ib_unregister_device(&dev->ib_dev);
1400 }
1401