xref: /linux/drivers/infiniband/hw/mthca/mthca_provider.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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_user_verbs.h>
41 #include <linux/mm.h>
42 
43 #include "mthca_dev.h"
44 #include "mthca_cmd.h"
45 #include "mthca_user.h"
46 #include "mthca_memfree.h"
47 
48 static void init_query_mad(struct ib_smp *mad)
49 {
50 	mad->base_version  = 1;
51 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
52 	mad->class_version = 1;
53 	mad->method    	   = IB_MGMT_METHOD_GET;
54 }
55 
56 static int mthca_query_device(struct ib_device *ibdev,
57 			      struct ib_device_attr *props)
58 {
59 	struct ib_smp *in_mad  = NULL;
60 	struct ib_smp *out_mad = NULL;
61 	int err = -ENOMEM;
62 	struct mthca_dev* mdev = to_mdev(ibdev);
63 
64 	u8 status;
65 
66 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
67 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
68 	if (!in_mad || !out_mad)
69 		goto out;
70 
71 	memset(props, 0, sizeof *props);
72 
73 	props->fw_ver              = mdev->fw_ver;
74 
75 	init_query_mad(in_mad);
76 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
77 
78 	err = mthca_MAD_IFC(mdev, 1, 1,
79 			    1, NULL, NULL, in_mad, out_mad,
80 			    &status);
81 	if (err)
82 		goto out;
83 	if (status) {
84 		err = -EINVAL;
85 		goto out;
86 	}
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_sge             = mdev->limits.max_sg;
100 	props->max_cq              = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
101 	props->max_cqe             = mdev->limits.max_cqes;
102 	props->max_mr              = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
103 	props->max_pd              = mdev->limits.num_pds - mdev->limits.reserved_pds;
104 	props->max_qp_rd_atom      = 1 << mdev->qp_table.rdb_shift;
105 	props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
106 	props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
107 	props->max_srq             = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
108 	props->max_srq_wr          = mdev->limits.max_srq_wqes;
109 	props->max_srq_sge         = mdev->limits.max_srq_sge;
110 	props->local_ca_ack_delay  = mdev->limits.local_ca_ack_delay;
111 	props->atomic_cap          = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
112 					IB_ATOMIC_HCA : IB_ATOMIC_NONE;
113 	props->max_pkeys           = mdev->limits.pkey_table_len;
114 	props->max_mcast_grp       = mdev->limits.num_mgms + mdev->limits.num_amgms;
115 	props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
116 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
117 					   props->max_mcast_grp;
118 	/*
119 	 * If Sinai memory key optimization is being used, then only
120 	 * the 8-bit key portion will change.  For other HCAs, the
121 	 * unused index bits will also be used for FMR remapping.
122 	 */
123 	if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
124 		props->max_map_per_fmr = 255;
125 	else
126 		props->max_map_per_fmr =
127 			(1 << (32 - ilog2(mdev->limits.num_mpts))) - 1;
128 
129 	err = 0;
130  out:
131 	kfree(in_mad);
132 	kfree(out_mad);
133 	return err;
134 }
135 
136 static int mthca_query_port(struct ib_device *ibdev,
137 			    u8 port, struct ib_port_attr *props)
138 {
139 	struct ib_smp *in_mad  = NULL;
140 	struct ib_smp *out_mad = NULL;
141 	int err = -ENOMEM;
142 	u8 status;
143 
144 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
145 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
146 	if (!in_mad || !out_mad)
147 		goto out;
148 
149 	memset(props, 0, sizeof *props);
150 
151 	init_query_mad(in_mad);
152 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
153 	in_mad->attr_mod = cpu_to_be32(port);
154 
155 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
156 			    port, NULL, NULL, in_mad, out_mad,
157 			    &status);
158 	if (err)
159 		goto out;
160 	if (status) {
161 		err = -EINVAL;
162 		goto out;
163 	}
164 
165 	props->lid               = be16_to_cpup((__be16 *) (out_mad->data + 16));
166 	props->lmc               = out_mad->data[34] & 0x7;
167 	props->sm_lid            = be16_to_cpup((__be16 *) (out_mad->data + 18));
168 	props->sm_sl             = out_mad->data[36] & 0xf;
169 	props->state             = out_mad->data[32] & 0xf;
170 	props->phys_state        = out_mad->data[33] >> 4;
171 	props->port_cap_flags    = be32_to_cpup((__be32 *) (out_mad->data + 20));
172 	props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
173 	props->max_msg_sz        = 0x80000000;
174 	props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
175 	props->bad_pkey_cntr     = be16_to_cpup((__be16 *) (out_mad->data + 46));
176 	props->qkey_viol_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 48));
177 	props->active_width      = out_mad->data[31] & 0xf;
178 	props->active_speed      = out_mad->data[35] >> 4;
179 	props->max_mtu           = out_mad->data[41] & 0xf;
180 	props->active_mtu        = out_mad->data[36] >> 4;
181 	props->subnet_timeout    = out_mad->data[51] & 0x1f;
182 	props->max_vl_num        = out_mad->data[37] >> 4;
183 	props->init_type_reply   = out_mad->data[41] >> 4;
184 
185  out:
186 	kfree(in_mad);
187 	kfree(out_mad);
188 	return err;
189 }
190 
191 static int mthca_modify_device(struct ib_device *ibdev,
192 			       int mask,
193 			       struct ib_device_modify *props)
194 {
195 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
196 		return -EOPNOTSUPP;
197 
198 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
199 		if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
200 			return -ERESTARTSYS;
201 		memcpy(ibdev->node_desc, props->node_desc, 64);
202 		mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
203 	}
204 
205 	return 0;
206 }
207 
208 static int mthca_modify_port(struct ib_device *ibdev,
209 			     u8 port, int port_modify_mask,
210 			     struct ib_port_modify *props)
211 {
212 	struct mthca_set_ib_param set_ib;
213 	struct ib_port_attr attr;
214 	int err;
215 	u8 status;
216 
217 	if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
218 		return -ERESTARTSYS;
219 
220 	err = mthca_query_port(ibdev, port, &attr);
221 	if (err)
222 		goto out;
223 
224 	set_ib.set_si_guid     = 0;
225 	set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
226 
227 	set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
228 		~props->clr_port_cap_mask;
229 
230 	err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port, &status);
231 	if (err)
232 		goto out;
233 	if (status) {
234 		err = -EINVAL;
235 		goto out;
236 	}
237 
238 out:
239 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
240 	return err;
241 }
242 
243 static int mthca_query_pkey(struct ib_device *ibdev,
244 			    u8 port, u16 index, u16 *pkey)
245 {
246 	struct ib_smp *in_mad  = NULL;
247 	struct ib_smp *out_mad = NULL;
248 	int err = -ENOMEM;
249 	u8 status;
250 
251 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
252 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
253 	if (!in_mad || !out_mad)
254 		goto out;
255 
256 	init_query_mad(in_mad);
257 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
258 	in_mad->attr_mod = cpu_to_be32(index / 32);
259 
260 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
261 			    port, NULL, NULL, in_mad, out_mad,
262 			    &status);
263 	if (err)
264 		goto out;
265 	if (status) {
266 		err = -EINVAL;
267 		goto out;
268 	}
269 
270 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
271 
272  out:
273 	kfree(in_mad);
274 	kfree(out_mad);
275 	return err;
276 }
277 
278 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
279 			   int index, union ib_gid *gid)
280 {
281 	struct ib_smp *in_mad  = NULL;
282 	struct ib_smp *out_mad = NULL;
283 	int err = -ENOMEM;
284 	u8 status;
285 
286 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
287 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
288 	if (!in_mad || !out_mad)
289 		goto out;
290 
291 	init_query_mad(in_mad);
292 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
293 	in_mad->attr_mod = cpu_to_be32(port);
294 
295 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
296 			    port, NULL, NULL, in_mad, out_mad,
297 			    &status);
298 	if (err)
299 		goto out;
300 	if (status) {
301 		err = -EINVAL;
302 		goto out;
303 	}
304 
305 	memcpy(gid->raw, out_mad->data + 8, 8);
306 
307 	init_query_mad(in_mad);
308 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
309 	in_mad->attr_mod = cpu_to_be32(index / 8);
310 
311 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
312 			    port, NULL, NULL, in_mad, out_mad,
313 			    &status);
314 	if (err)
315 		goto out;
316 	if (status) {
317 		err = -EINVAL;
318 		goto out;
319 	}
320 
321 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
322 
323  out:
324 	kfree(in_mad);
325 	kfree(out_mad);
326 	return err;
327 }
328 
329 static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
330 						struct ib_udata *udata)
331 {
332 	struct mthca_alloc_ucontext_resp uresp;
333 	struct mthca_ucontext           *context;
334 	int                              err;
335 
336 	memset(&uresp, 0, sizeof uresp);
337 
338 	uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
339 	if (mthca_is_memfree(to_mdev(ibdev)))
340 		uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
341 	else
342 		uresp.uarc_size = 0;
343 
344 	context = kmalloc(sizeof *context, GFP_KERNEL);
345 	if (!context)
346 		return ERR_PTR(-ENOMEM);
347 
348 	err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
349 	if (err) {
350 		kfree(context);
351 		return ERR_PTR(err);
352 	}
353 
354 	context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
355 	if (IS_ERR(context->db_tab)) {
356 		err = PTR_ERR(context->db_tab);
357 		mthca_uar_free(to_mdev(ibdev), &context->uar);
358 		kfree(context);
359 		return ERR_PTR(err);
360 	}
361 
362 	if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) {
363 		mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
364 		mthca_uar_free(to_mdev(ibdev), &context->uar);
365 		kfree(context);
366 		return ERR_PTR(-EFAULT);
367 	}
368 
369 	return &context->ibucontext;
370 }
371 
372 static int mthca_dealloc_ucontext(struct ib_ucontext *context)
373 {
374 	mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
375 				  to_mucontext(context)->db_tab);
376 	mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
377 	kfree(to_mucontext(context));
378 
379 	return 0;
380 }
381 
382 static int mthca_mmap_uar(struct ib_ucontext *context,
383 			  struct vm_area_struct *vma)
384 {
385 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
386 		return -EINVAL;
387 
388 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
389 
390 	if (io_remap_pfn_range(vma, vma->vm_start,
391 			       to_mucontext(context)->uar.pfn,
392 			       PAGE_SIZE, vma->vm_page_prot))
393 		return -EAGAIN;
394 
395 	return 0;
396 }
397 
398 static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev,
399 				    struct ib_ucontext *context,
400 				    struct ib_udata *udata)
401 {
402 	struct mthca_pd *pd;
403 	int err;
404 
405 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
406 	if (!pd)
407 		return ERR_PTR(-ENOMEM);
408 
409 	err = mthca_pd_alloc(to_mdev(ibdev), !context, pd);
410 	if (err) {
411 		kfree(pd);
412 		return ERR_PTR(err);
413 	}
414 
415 	if (context) {
416 		if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
417 			mthca_pd_free(to_mdev(ibdev), pd);
418 			kfree(pd);
419 			return ERR_PTR(-EFAULT);
420 		}
421 	}
422 
423 	return &pd->ibpd;
424 }
425 
426 static int mthca_dealloc_pd(struct ib_pd *pd)
427 {
428 	mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
429 	kfree(pd);
430 
431 	return 0;
432 }
433 
434 static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
435 				     struct ib_ah_attr *ah_attr)
436 {
437 	int err;
438 	struct mthca_ah *ah;
439 
440 	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
441 	if (!ah)
442 		return ERR_PTR(-ENOMEM);
443 
444 	err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
445 	if (err) {
446 		kfree(ah);
447 		return ERR_PTR(err);
448 	}
449 
450 	return &ah->ibah;
451 }
452 
453 static int mthca_ah_destroy(struct ib_ah *ah)
454 {
455 	mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
456 	kfree(ah);
457 
458 	return 0;
459 }
460 
461 static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
462 				       struct ib_srq_init_attr *init_attr,
463 				       struct ib_udata *udata)
464 {
465 	struct mthca_create_srq ucmd;
466 	struct mthca_ucontext *context = NULL;
467 	struct mthca_srq *srq;
468 	int err;
469 
470 	srq = kmalloc(sizeof *srq, GFP_KERNEL);
471 	if (!srq)
472 		return ERR_PTR(-ENOMEM);
473 
474 	if (pd->uobject) {
475 		context = to_mucontext(pd->uobject->context);
476 
477 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
478 			err = -EFAULT;
479 			goto err_free;
480 		}
481 
482 		err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
483 					context->db_tab, ucmd.db_index,
484 					ucmd.db_page);
485 
486 		if (err)
487 			goto err_free;
488 
489 		srq->mr.ibmr.lkey = ucmd.lkey;
490 		srq->db_index     = ucmd.db_index;
491 	}
492 
493 	err = mthca_alloc_srq(to_mdev(pd->device), to_mpd(pd),
494 			      &init_attr->attr, srq);
495 
496 	if (err && pd->uobject)
497 		mthca_unmap_user_db(to_mdev(pd->device), &context->uar,
498 				    context->db_tab, ucmd.db_index);
499 
500 	if (err)
501 		goto err_free;
502 
503 	if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof (__u32))) {
504 		mthca_free_srq(to_mdev(pd->device), srq);
505 		err = -EFAULT;
506 		goto err_free;
507 	}
508 
509 	return &srq->ibsrq;
510 
511 err_free:
512 	kfree(srq);
513 
514 	return ERR_PTR(err);
515 }
516 
517 static int mthca_destroy_srq(struct ib_srq *srq)
518 {
519 	struct mthca_ucontext *context;
520 
521 	if (srq->uobject) {
522 		context = to_mucontext(srq->uobject->context);
523 
524 		mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
525 				    context->db_tab, to_msrq(srq)->db_index);
526 	}
527 
528 	mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
529 	kfree(srq);
530 
531 	return 0;
532 }
533 
534 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
535 				     struct ib_qp_init_attr *init_attr,
536 				     struct ib_udata *udata)
537 {
538 	struct mthca_create_qp ucmd;
539 	struct mthca_qp *qp;
540 	int err;
541 
542 	switch (init_attr->qp_type) {
543 	case IB_QPT_RC:
544 	case IB_QPT_UC:
545 	case IB_QPT_UD:
546 	{
547 		struct mthca_ucontext *context;
548 
549 		qp = kmalloc(sizeof *qp, GFP_KERNEL);
550 		if (!qp)
551 			return ERR_PTR(-ENOMEM);
552 
553 		if (pd->uobject) {
554 			context = to_mucontext(pd->uobject->context);
555 
556 			if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
557 				kfree(qp);
558 				return ERR_PTR(-EFAULT);
559 			}
560 
561 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
562 						context->db_tab,
563 						ucmd.sq_db_index, ucmd.sq_db_page);
564 			if (err) {
565 				kfree(qp);
566 				return ERR_PTR(err);
567 			}
568 
569 			err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
570 						context->db_tab,
571 						ucmd.rq_db_index, ucmd.rq_db_page);
572 			if (err) {
573 				mthca_unmap_user_db(to_mdev(pd->device),
574 						    &context->uar,
575 						    context->db_tab,
576 						    ucmd.sq_db_index);
577 				kfree(qp);
578 				return ERR_PTR(err);
579 			}
580 
581 			qp->mr.ibmr.lkey = ucmd.lkey;
582 			qp->sq.db_index  = ucmd.sq_db_index;
583 			qp->rq.db_index  = ucmd.rq_db_index;
584 		}
585 
586 		err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
587 				     to_mcq(init_attr->send_cq),
588 				     to_mcq(init_attr->recv_cq),
589 				     init_attr->qp_type, init_attr->sq_sig_type,
590 				     &init_attr->cap, qp);
591 
592 		if (err && pd->uobject) {
593 			context = to_mucontext(pd->uobject->context);
594 
595 			mthca_unmap_user_db(to_mdev(pd->device),
596 					    &context->uar,
597 					    context->db_tab,
598 					    ucmd.sq_db_index);
599 			mthca_unmap_user_db(to_mdev(pd->device),
600 					    &context->uar,
601 					    context->db_tab,
602 					    ucmd.rq_db_index);
603 		}
604 
605 		qp->ibqp.qp_num = qp->qpn;
606 		break;
607 	}
608 	case IB_QPT_SMI:
609 	case IB_QPT_GSI:
610 	{
611 		/* Don't allow userspace to create special QPs */
612 		if (pd->uobject)
613 			return ERR_PTR(-EINVAL);
614 
615 		qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
616 		if (!qp)
617 			return ERR_PTR(-ENOMEM);
618 
619 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
620 
621 		err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
622 				      to_mcq(init_attr->send_cq),
623 				      to_mcq(init_attr->recv_cq),
624 				      init_attr->sq_sig_type, &init_attr->cap,
625 				      qp->ibqp.qp_num, init_attr->port_num,
626 				      to_msqp(qp));
627 		break;
628 	}
629 	default:
630 		/* Don't support raw QPs */
631 		return ERR_PTR(-ENOSYS);
632 	}
633 
634 	if (err) {
635 		kfree(qp);
636 		return ERR_PTR(err);
637 	}
638 
639 	init_attr->cap.max_send_wr     = qp->sq.max;
640 	init_attr->cap.max_recv_wr     = qp->rq.max;
641 	init_attr->cap.max_send_sge    = qp->sq.max_gs;
642 	init_attr->cap.max_recv_sge    = qp->rq.max_gs;
643 	init_attr->cap.max_inline_data = qp->max_inline_data;
644 
645 	return &qp->ibqp;
646 }
647 
648 static int mthca_destroy_qp(struct ib_qp *qp)
649 {
650 	if (qp->uobject) {
651 		mthca_unmap_user_db(to_mdev(qp->device),
652 				    &to_mucontext(qp->uobject->context)->uar,
653 				    to_mucontext(qp->uobject->context)->db_tab,
654 				    to_mqp(qp)->sq.db_index);
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)->rq.db_index);
659 	}
660 	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
661 	kfree(qp);
662 	return 0;
663 }
664 
665 static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries,
666 				     struct ib_ucontext *context,
667 				     struct ib_udata *udata)
668 {
669 	struct mthca_create_cq ucmd;
670 	struct mthca_cq *cq;
671 	int nent;
672 	int err;
673 
674 	if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
675 		return ERR_PTR(-EINVAL);
676 
677 	if (context) {
678 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
679 			return ERR_PTR(-EFAULT);
680 
681 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
682 					to_mucontext(context)->db_tab,
683 					ucmd.set_db_index, ucmd.set_db_page);
684 		if (err)
685 			return ERR_PTR(err);
686 
687 		err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
688 					to_mucontext(context)->db_tab,
689 					ucmd.arm_db_index, ucmd.arm_db_page);
690 		if (err)
691 			goto err_unmap_set;
692 	}
693 
694 	cq = kmalloc(sizeof *cq, GFP_KERNEL);
695 	if (!cq) {
696 		err = -ENOMEM;
697 		goto err_unmap_arm;
698 	}
699 
700 	if (context) {
701 		cq->buf.mr.ibmr.lkey = ucmd.lkey;
702 		cq->set_ci_db_index  = ucmd.set_db_index;
703 		cq->arm_db_index     = ucmd.arm_db_index;
704 	}
705 
706 	for (nent = 1; nent <= entries; nent <<= 1)
707 		; /* nothing */
708 
709 	err = mthca_init_cq(to_mdev(ibdev), nent,
710 			    context ? to_mucontext(context) : NULL,
711 			    context ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
712 			    cq);
713 	if (err)
714 		goto err_free;
715 
716 	if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
717 		mthca_free_cq(to_mdev(ibdev), cq);
718 		goto err_free;
719 	}
720 
721 	cq->resize_buf = NULL;
722 
723 	return &cq->ibcq;
724 
725 err_free:
726 	kfree(cq);
727 
728 err_unmap_arm:
729 	if (context)
730 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
731 				    to_mucontext(context)->db_tab, ucmd.arm_db_index);
732 
733 err_unmap_set:
734 	if (context)
735 		mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
736 				    to_mucontext(context)->db_tab, ucmd.set_db_index);
737 
738 	return ERR_PTR(err);
739 }
740 
741 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
742 				  int entries)
743 {
744 	int ret;
745 
746 	spin_lock_irq(&cq->lock);
747 	if (cq->resize_buf) {
748 		ret = -EBUSY;
749 		goto unlock;
750 	}
751 
752 	cq->resize_buf = kmalloc(sizeof *cq->resize_buf, GFP_ATOMIC);
753 	if (!cq->resize_buf) {
754 		ret = -ENOMEM;
755 		goto unlock;
756 	}
757 
758 	cq->resize_buf->state = CQ_RESIZE_ALLOC;
759 
760 	ret = 0;
761 
762 unlock:
763 	spin_unlock_irq(&cq->lock);
764 
765 	if (ret)
766 		return ret;
767 
768 	ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
769 	if (ret) {
770 		spin_lock_irq(&cq->lock);
771 		kfree(cq->resize_buf);
772 		cq->resize_buf = NULL;
773 		spin_unlock_irq(&cq->lock);
774 		return ret;
775 	}
776 
777 	cq->resize_buf->cqe = entries - 1;
778 
779 	spin_lock_irq(&cq->lock);
780 	cq->resize_buf->state = CQ_RESIZE_READY;
781 	spin_unlock_irq(&cq->lock);
782 
783 	return 0;
784 }
785 
786 static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
787 {
788 	struct mthca_dev *dev = to_mdev(ibcq->device);
789 	struct mthca_cq *cq = to_mcq(ibcq);
790 	struct mthca_resize_cq ucmd;
791 	u32 lkey;
792 	u8 status;
793 	int ret;
794 
795 	if (entries < 1 || entries > dev->limits.max_cqes)
796 		return -EINVAL;
797 
798 	mutex_lock(&cq->mutex);
799 
800 	entries = roundup_pow_of_two(entries + 1);
801 	if (entries == ibcq->cqe + 1) {
802 		ret = 0;
803 		goto out;
804 	}
805 
806 	if (cq->is_kernel) {
807 		ret = mthca_alloc_resize_buf(dev, cq, entries);
808 		if (ret)
809 			goto out;
810 		lkey = cq->resize_buf->buf.mr.ibmr.lkey;
811 	} else {
812 		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
813 			ret = -EFAULT;
814 			goto out;
815 		}
816 		lkey = ucmd.lkey;
817 	}
818 
819 	ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries), &status);
820 	if (status)
821 		ret = -EINVAL;
822 
823 	if (ret) {
824 		if (cq->resize_buf) {
825 			mthca_free_cq_buf(dev, &cq->resize_buf->buf,
826 					  cq->resize_buf->cqe);
827 			kfree(cq->resize_buf);
828 			spin_lock_irq(&cq->lock);
829 			cq->resize_buf = NULL;
830 			spin_unlock_irq(&cq->lock);
831 		}
832 		goto out;
833 	}
834 
835 	if (cq->is_kernel) {
836 		struct mthca_cq_buf tbuf;
837 		int tcqe;
838 
839 		spin_lock_irq(&cq->lock);
840 		if (cq->resize_buf->state == CQ_RESIZE_READY) {
841 			mthca_cq_resize_copy_cqes(cq);
842 			tbuf         = cq->buf;
843 			tcqe         = cq->ibcq.cqe;
844 			cq->buf      = cq->resize_buf->buf;
845 			cq->ibcq.cqe = cq->resize_buf->cqe;
846 		} else {
847 			tbuf = cq->resize_buf->buf;
848 			tcqe = cq->resize_buf->cqe;
849 		}
850 
851 		kfree(cq->resize_buf);
852 		cq->resize_buf = NULL;
853 		spin_unlock_irq(&cq->lock);
854 
855 		mthca_free_cq_buf(dev, &tbuf, tcqe);
856 	} else
857 		ibcq->cqe = entries - 1;
858 
859 out:
860 	mutex_unlock(&cq->mutex);
861 
862 	return ret;
863 }
864 
865 static int mthca_destroy_cq(struct ib_cq *cq)
866 {
867 	if (cq->uobject) {
868 		mthca_unmap_user_db(to_mdev(cq->device),
869 				    &to_mucontext(cq->uobject->context)->uar,
870 				    to_mucontext(cq->uobject->context)->db_tab,
871 				    to_mcq(cq)->arm_db_index);
872 		mthca_unmap_user_db(to_mdev(cq->device),
873 				    &to_mucontext(cq->uobject->context)->uar,
874 				    to_mucontext(cq->uobject->context)->db_tab,
875 				    to_mcq(cq)->set_ci_db_index);
876 	}
877 	mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
878 	kfree(cq);
879 
880 	return 0;
881 }
882 
883 static inline u32 convert_access(int acc)
884 {
885 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
886 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
887 	       (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
888 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
889 	       MTHCA_MPT_FLAG_LOCAL_READ;
890 }
891 
892 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
893 {
894 	struct mthca_mr *mr;
895 	int err;
896 
897 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
898 	if (!mr)
899 		return ERR_PTR(-ENOMEM);
900 
901 	err = mthca_mr_alloc_notrans(to_mdev(pd->device),
902 				     to_mpd(pd)->pd_num,
903 				     convert_access(acc), mr);
904 
905 	if (err) {
906 		kfree(mr);
907 		return ERR_PTR(err);
908 	}
909 
910 	return &mr->ibmr;
911 }
912 
913 static struct ib_mr *mthca_reg_phys_mr(struct ib_pd       *pd,
914 				       struct ib_phys_buf *buffer_list,
915 				       int                 num_phys_buf,
916 				       int                 acc,
917 				       u64                *iova_start)
918 {
919 	struct mthca_mr *mr;
920 	u64 *page_list;
921 	u64 total_size;
922 	u64 mask;
923 	int shift;
924 	int npages;
925 	int err;
926 	int i, j, n;
927 
928 	/* First check that we have enough alignment */
929 	if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK))
930 		return ERR_PTR(-EINVAL);
931 
932 	mask = 0;
933 	total_size = 0;
934 	for (i = 0; i < num_phys_buf; ++i) {
935 		if (i != 0)
936 			mask |= buffer_list[i].addr;
937 		if (i != num_phys_buf - 1)
938 			mask |= buffer_list[i].addr + buffer_list[i].size;
939 
940 		total_size += buffer_list[i].size;
941 	}
942 
943 	if (mask & ~PAGE_MASK)
944 		return ERR_PTR(-EINVAL);
945 
946 	/* Find largest page shift we can use to cover buffers */
947 	for (shift = PAGE_SHIFT; shift < 31; ++shift)
948 		if (num_phys_buf > 1) {
949 			if ((1ULL << shift) & mask)
950 				break;
951 		} else {
952 			if (1ULL << shift >=
953 			    buffer_list[0].size +
954 			    (buffer_list[0].addr & ((1ULL << shift) - 1)))
955 				break;
956 		}
957 
958 	buffer_list[0].size += buffer_list[0].addr & ((1ULL << shift) - 1);
959 	buffer_list[0].addr &= ~0ull << shift;
960 
961 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
962 	if (!mr)
963 		return ERR_PTR(-ENOMEM);
964 
965 	npages = 0;
966 	for (i = 0; i < num_phys_buf; ++i)
967 		npages += (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
968 
969 	if (!npages)
970 		return &mr->ibmr;
971 
972 	page_list = kmalloc(npages * sizeof *page_list, GFP_KERNEL);
973 	if (!page_list) {
974 		kfree(mr);
975 		return ERR_PTR(-ENOMEM);
976 	}
977 
978 	n = 0;
979 	for (i = 0; i < num_phys_buf; ++i)
980 		for (j = 0;
981 		     j < (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
982 		     ++j)
983 			page_list[n++] = buffer_list[i].addr + ((u64) j << shift);
984 
985 	mthca_dbg(to_mdev(pd->device), "Registering memory at %llx (iova %llx) "
986 		  "in PD %x; shift %d, npages %d.\n",
987 		  (unsigned long long) buffer_list[0].addr,
988 		  (unsigned long long) *iova_start,
989 		  to_mpd(pd)->pd_num,
990 		  shift, npages);
991 
992 	err = mthca_mr_alloc_phys(to_mdev(pd->device),
993 				  to_mpd(pd)->pd_num,
994 				  page_list, shift, npages,
995 				  *iova_start, total_size,
996 				  convert_access(acc), mr);
997 
998 	if (err) {
999 		kfree(page_list);
1000 		kfree(mr);
1001 		return ERR_PTR(err);
1002 	}
1003 
1004 	kfree(page_list);
1005 	return &mr->ibmr;
1006 }
1007 
1008 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
1009 				       int acc, struct ib_udata *udata)
1010 {
1011 	struct mthca_dev *dev = to_mdev(pd->device);
1012 	struct ib_umem_chunk *chunk;
1013 	struct mthca_mr *mr;
1014 	u64 *pages;
1015 	int shift, n, len;
1016 	int i, j, k;
1017 	int err = 0;
1018 	int write_mtt_size;
1019 
1020 	shift = ffs(region->page_size) - 1;
1021 
1022 	mr = kmalloc(sizeof *mr, GFP_KERNEL);
1023 	if (!mr)
1024 		return ERR_PTR(-ENOMEM);
1025 
1026 	n = 0;
1027 	list_for_each_entry(chunk, &region->chunk_list, list)
1028 		n += chunk->nents;
1029 
1030 	mr->mtt = mthca_alloc_mtt(dev, n);
1031 	if (IS_ERR(mr->mtt)) {
1032 		err = PTR_ERR(mr->mtt);
1033 		goto err;
1034 	}
1035 
1036 	pages = (u64 *) __get_free_page(GFP_KERNEL);
1037 	if (!pages) {
1038 		err = -ENOMEM;
1039 		goto err_mtt;
1040 	}
1041 
1042 	i = n = 0;
1043 
1044 	write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
1045 
1046 	list_for_each_entry(chunk, &region->chunk_list, list)
1047 		for (j = 0; j < chunk->nmap; ++j) {
1048 			len = sg_dma_len(&chunk->page_list[j]) >> shift;
1049 			for (k = 0; k < len; ++k) {
1050 				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
1051 					region->page_size * k;
1052 				/*
1053 				 * Be friendly to write_mtt and pass it chunks
1054 				 * of appropriate size.
1055 				 */
1056 				if (i == write_mtt_size) {
1057 					err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
1058 					if (err)
1059 						goto mtt_done;
1060 					n += i;
1061 					i = 0;
1062 				}
1063 			}
1064 		}
1065 
1066 	if (i)
1067 		err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
1068 mtt_done:
1069 	free_page((unsigned long) pages);
1070 	if (err)
1071 		goto err_mtt;
1072 
1073 	err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, region->virt_base,
1074 			     region->length, convert_access(acc), mr);
1075 
1076 	if (err)
1077 		goto err_mtt;
1078 
1079 	return &mr->ibmr;
1080 
1081 err_mtt:
1082 	mthca_free_mtt(dev, mr->mtt);
1083 
1084 err:
1085 	kfree(mr);
1086 	return ERR_PTR(err);
1087 }
1088 
1089 static int mthca_dereg_mr(struct ib_mr *mr)
1090 {
1091 	struct mthca_mr *mmr = to_mmr(mr);
1092 	mthca_free_mr(to_mdev(mr->device), mmr);
1093 	kfree(mmr);
1094 	return 0;
1095 }
1096 
1097 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1098 				      struct ib_fmr_attr *fmr_attr)
1099 {
1100 	struct mthca_fmr *fmr;
1101 	int err;
1102 
1103 	fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
1104 	if (!fmr)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
1108 	err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
1109 			     convert_access(mr_access_flags), fmr);
1110 
1111 	if (err) {
1112 		kfree(fmr);
1113 		return ERR_PTR(err);
1114 	}
1115 
1116 	return &fmr->ibmr;
1117 }
1118 
1119 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
1120 {
1121 	struct mthca_fmr *mfmr = to_mfmr(fmr);
1122 	int err;
1123 
1124 	err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
1125 	if (err)
1126 		return err;
1127 
1128 	kfree(mfmr);
1129 	return 0;
1130 }
1131 
1132 static int mthca_unmap_fmr(struct list_head *fmr_list)
1133 {
1134 	struct ib_fmr *fmr;
1135 	int err;
1136 	u8 status;
1137 	struct mthca_dev *mdev = NULL;
1138 
1139 	list_for_each_entry(fmr, fmr_list, list) {
1140 		if (mdev && to_mdev(fmr->device) != mdev)
1141 			return -EINVAL;
1142 		mdev = to_mdev(fmr->device);
1143 	}
1144 
1145 	if (!mdev)
1146 		return 0;
1147 
1148 	if (mthca_is_memfree(mdev)) {
1149 		list_for_each_entry(fmr, fmr_list, list)
1150 			mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1151 
1152 		wmb();
1153 	} else
1154 		list_for_each_entry(fmr, fmr_list, list)
1155 			mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1156 
1157 	err = mthca_SYNC_TPT(mdev, &status);
1158 	if (err)
1159 		return err;
1160 	if (status)
1161 		return -EINVAL;
1162 	return 0;
1163 }
1164 
1165 static ssize_t show_rev(struct class_device *cdev, char *buf)
1166 {
1167 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1168 	return sprintf(buf, "%x\n", dev->rev_id);
1169 }
1170 
1171 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
1172 {
1173 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1174 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->fw_ver >> 32),
1175 		       (int) (dev->fw_ver >> 16) & 0xffff,
1176 		       (int) dev->fw_ver & 0xffff);
1177 }
1178 
1179 static ssize_t show_hca(struct class_device *cdev, char *buf)
1180 {
1181 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1182 	switch (dev->pdev->device) {
1183 	case PCI_DEVICE_ID_MELLANOX_TAVOR:
1184 		return sprintf(buf, "MT23108\n");
1185 	case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1186 		return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1187 	case PCI_DEVICE_ID_MELLANOX_ARBEL:
1188 		return sprintf(buf, "MT25208\n");
1189 	case PCI_DEVICE_ID_MELLANOX_SINAI:
1190 	case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1191 		return sprintf(buf, "MT25204\n");
1192 	default:
1193 		return sprintf(buf, "unknown\n");
1194 	}
1195 }
1196 
1197 static ssize_t show_board(struct class_device *cdev, char *buf)
1198 {
1199 	struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1200 	return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1201 }
1202 
1203 static CLASS_DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1204 static CLASS_DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1205 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1206 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1207 
1208 static struct class_device_attribute *mthca_class_attributes[] = {
1209 	&class_device_attr_hw_rev,
1210 	&class_device_attr_fw_ver,
1211 	&class_device_attr_hca_type,
1212 	&class_device_attr_board_id
1213 };
1214 
1215 static int mthca_init_node_data(struct mthca_dev *dev)
1216 {
1217 	struct ib_smp *in_mad  = NULL;
1218 	struct ib_smp *out_mad = NULL;
1219 	int err = -ENOMEM;
1220 	u8 status;
1221 
1222 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1223 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1224 	if (!in_mad || !out_mad)
1225 		goto out;
1226 
1227 	init_query_mad(in_mad);
1228 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1229 
1230 	err = mthca_MAD_IFC(dev, 1, 1,
1231 			    1, NULL, NULL, in_mad, out_mad,
1232 			    &status);
1233 	if (err)
1234 		goto out;
1235 	if (status) {
1236 		err = -EINVAL;
1237 		goto out;
1238 	}
1239 
1240 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1241 
1242 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1243 
1244 	err = mthca_MAD_IFC(dev, 1, 1,
1245 			    1, NULL, NULL, in_mad, out_mad,
1246 			    &status);
1247 	if (err)
1248 		goto out;
1249 	if (status) {
1250 		err = -EINVAL;
1251 		goto out;
1252 	}
1253 
1254 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1255 
1256 out:
1257 	kfree(in_mad);
1258 	kfree(out_mad);
1259 	return err;
1260 }
1261 
1262 int mthca_register_device(struct mthca_dev *dev)
1263 {
1264 	int ret;
1265 	int i;
1266 
1267 	ret = mthca_init_node_data(dev);
1268 	if (ret)
1269 		return ret;
1270 
1271 	strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
1272 	dev->ib_dev.owner                = THIS_MODULE;
1273 
1274 	dev->ib_dev.uverbs_abi_ver	 = MTHCA_UVERBS_ABI_VERSION;
1275 	dev->ib_dev.uverbs_cmd_mask	 =
1276 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1277 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1278 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1279 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1280 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1281 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1282 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1283 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1284 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1285 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1286 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1287 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1288 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1289 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1290 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1291 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1292 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1293 	dev->ib_dev.node_type            = RDMA_NODE_IB_CA;
1294 	dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
1295 	dev->ib_dev.dma_device           = &dev->pdev->dev;
1296 	dev->ib_dev.class_dev.dev        = &dev->pdev->dev;
1297 	dev->ib_dev.query_device         = mthca_query_device;
1298 	dev->ib_dev.query_port           = mthca_query_port;
1299 	dev->ib_dev.modify_device        = mthca_modify_device;
1300 	dev->ib_dev.modify_port          = mthca_modify_port;
1301 	dev->ib_dev.query_pkey           = mthca_query_pkey;
1302 	dev->ib_dev.query_gid            = mthca_query_gid;
1303 	dev->ib_dev.alloc_ucontext       = mthca_alloc_ucontext;
1304 	dev->ib_dev.dealloc_ucontext     = mthca_dealloc_ucontext;
1305 	dev->ib_dev.mmap                 = mthca_mmap_uar;
1306 	dev->ib_dev.alloc_pd             = mthca_alloc_pd;
1307 	dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
1308 	dev->ib_dev.create_ah            = mthca_ah_create;
1309 	dev->ib_dev.query_ah             = mthca_ah_query;
1310 	dev->ib_dev.destroy_ah           = mthca_ah_destroy;
1311 
1312 	if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1313 		dev->ib_dev.create_srq           = mthca_create_srq;
1314 		dev->ib_dev.modify_srq           = mthca_modify_srq;
1315 		dev->ib_dev.query_srq            = mthca_query_srq;
1316 		dev->ib_dev.destroy_srq          = mthca_destroy_srq;
1317 		dev->ib_dev.uverbs_cmd_mask	|=
1318 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1319 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1320 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1321 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1322 
1323 		if (mthca_is_memfree(dev))
1324 			dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1325 		else
1326 			dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1327 	}
1328 
1329 	dev->ib_dev.create_qp            = mthca_create_qp;
1330 	dev->ib_dev.modify_qp            = mthca_modify_qp;
1331 	dev->ib_dev.query_qp             = mthca_query_qp;
1332 	dev->ib_dev.destroy_qp           = mthca_destroy_qp;
1333 	dev->ib_dev.create_cq            = mthca_create_cq;
1334 	dev->ib_dev.resize_cq            = mthca_resize_cq;
1335 	dev->ib_dev.destroy_cq           = mthca_destroy_cq;
1336 	dev->ib_dev.poll_cq              = mthca_poll_cq;
1337 	dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
1338 	dev->ib_dev.reg_phys_mr          = mthca_reg_phys_mr;
1339 	dev->ib_dev.reg_user_mr          = mthca_reg_user_mr;
1340 	dev->ib_dev.dereg_mr             = mthca_dereg_mr;
1341 
1342 	if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1343 		dev->ib_dev.alloc_fmr            = mthca_alloc_fmr;
1344 		dev->ib_dev.unmap_fmr            = mthca_unmap_fmr;
1345 		dev->ib_dev.dealloc_fmr          = mthca_dealloc_fmr;
1346 		if (mthca_is_memfree(dev))
1347 			dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1348 		else
1349 			dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1350 	}
1351 
1352 	dev->ib_dev.attach_mcast         = mthca_multicast_attach;
1353 	dev->ib_dev.detach_mcast         = mthca_multicast_detach;
1354 	dev->ib_dev.process_mad          = mthca_process_mad;
1355 
1356 	if (mthca_is_memfree(dev)) {
1357 		dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1358 		dev->ib_dev.post_send     = mthca_arbel_post_send;
1359 		dev->ib_dev.post_recv     = mthca_arbel_post_receive;
1360 	} else {
1361 		dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1362 		dev->ib_dev.post_send     = mthca_tavor_post_send;
1363 		dev->ib_dev.post_recv     = mthca_tavor_post_receive;
1364 	}
1365 
1366 	mutex_init(&dev->cap_mask_mutex);
1367 
1368 	ret = ib_register_device(&dev->ib_dev);
1369 	if (ret)
1370 		return ret;
1371 
1372 	for (i = 0; i < ARRAY_SIZE(mthca_class_attributes); ++i) {
1373 		ret = class_device_create_file(&dev->ib_dev.class_dev,
1374 					       mthca_class_attributes[i]);
1375 		if (ret) {
1376 			ib_unregister_device(&dev->ib_dev);
1377 			return ret;
1378 		}
1379 	}
1380 
1381 	mthca_start_catas_poll(dev);
1382 
1383 	return 0;
1384 }
1385 
1386 void mthca_unregister_device(struct mthca_dev *dev)
1387 {
1388 	mthca_stop_catas_poll(dev);
1389 	ib_unregister_device(&dev->ib_dev);
1390 }
1391