Manual merge of for-linus to upstream (fix conflicts in drivers/infiniband/core/ucm.c)
[linux-2.6] / drivers / infiniband / hw / mthca / mthca_provider.c
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 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 1397 2004-12-28 05:09:00Z 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 int mthca_query_device(struct ib_device *ibdev,
49                               struct ib_device_attr *props)
50 {
51         struct ib_smp *in_mad  = NULL;
52         struct ib_smp *out_mad = NULL;
53         int err = -ENOMEM;
54         struct mthca_dev* mdev = to_mdev(ibdev);
55
56         u8 status;
57
58         in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
59         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
60         if (!in_mad || !out_mad)
61                 goto out;
62
63         memset(props, 0, sizeof *props);
64
65         props->fw_ver              = mdev->fw_ver;
66
67         memset(in_mad, 0, sizeof *in_mad);
68         in_mad->base_version       = 1;
69         in_mad->mgmt_class         = IB_MGMT_CLASS_SUBN_LID_ROUTED;
70         in_mad->class_version      = 1;
71         in_mad->method             = IB_MGMT_METHOD_GET;
72         in_mad->attr_id            = IB_SMP_ATTR_NODE_INFO;
73
74         err = mthca_MAD_IFC(mdev, 1, 1,
75                             1, NULL, NULL, in_mad, out_mad,
76                             &status);
77         if (err)
78                 goto out;
79         if (status) {
80                 err = -EINVAL;
81                 goto out;
82         }
83
84         props->device_cap_flags    = mdev->device_cap_flags;
85         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
86                 0xffffff;
87         props->vendor_part_id      = be16_to_cpup((__be16 *) (out_mad->data + 30));
88         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
89         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
90         memcpy(&props->node_guid,      out_mad->data + 12, 8);
91
92         props->max_mr_size         = ~0ull;
93         props->max_qp              = mdev->limits.num_qps - mdev->limits.reserved_qps;
94         props->max_qp_wr           = mdev->limits.max_wqes;
95         props->max_sge             = mdev->limits.max_sg;
96         props->max_cq              = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
97         props->max_cqe             = mdev->limits.max_cqes;
98         props->max_mr              = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
99         props->max_pd              = mdev->limits.num_pds - mdev->limits.reserved_pds;
100         props->max_qp_rd_atom      = 1 << mdev->qp_table.rdb_shift;
101         props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
102         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
103         props->max_srq             = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
104         props->max_srq_wr          = mdev->limits.max_srq_wqes;
105         props->max_srq_sge         = mdev->limits.max_sg;
106         props->local_ca_ack_delay  = mdev->limits.local_ca_ack_delay;
107         props->atomic_cap          = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ? 
108                                         IB_ATOMIC_HCA : IB_ATOMIC_NONE;
109         props->max_pkeys           = mdev->limits.pkey_table_len;
110         props->max_mcast_grp       = mdev->limits.num_mgms + mdev->limits.num_amgms;
111         props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
112         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach * 
113                                            props->max_mcast_grp;
114
115         err = 0;
116  out:
117         kfree(in_mad);
118         kfree(out_mad);
119         return err;
120 }
121
122 static int mthca_query_port(struct ib_device *ibdev,
123                             u8 port, struct ib_port_attr *props)
124 {
125         struct ib_smp *in_mad  = NULL;
126         struct ib_smp *out_mad = NULL;
127         int err = -ENOMEM;
128         u8 status;
129
130         in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
131         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
132         if (!in_mad || !out_mad)
133                 goto out;
134
135         memset(props, 0, sizeof *props);
136
137         memset(in_mad, 0, sizeof *in_mad);
138         in_mad->base_version       = 1;
139         in_mad->mgmt_class         = IB_MGMT_CLASS_SUBN_LID_ROUTED;
140         in_mad->class_version      = 1;
141         in_mad->method             = IB_MGMT_METHOD_GET;
142         in_mad->attr_id            = IB_SMP_ATTR_PORT_INFO;
143         in_mad->attr_mod           = cpu_to_be32(port);
144
145         err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
146                             port, NULL, NULL, in_mad, out_mad,
147                             &status);
148         if (err)
149                 goto out;
150         if (status) {
151                 err = -EINVAL;
152                 goto out;
153         }
154
155         props->lid               = be16_to_cpup((__be16 *) (out_mad->data + 16));
156         props->lmc               = out_mad->data[34] & 0x7;
157         props->sm_lid            = be16_to_cpup((__be16 *) (out_mad->data + 18));
158         props->sm_sl             = out_mad->data[36] & 0xf;
159         props->state             = out_mad->data[32] & 0xf;
160         props->phys_state        = out_mad->data[33] >> 4;
161         props->port_cap_flags    = be32_to_cpup((__be32 *) (out_mad->data + 20));
162         props->gid_tbl_len       = to_mdev(ibdev)->limits.gid_table_len;
163         props->max_msg_sz        = 0x80000000;
164         props->pkey_tbl_len      = to_mdev(ibdev)->limits.pkey_table_len;
165         props->bad_pkey_cntr     = be16_to_cpup((__be16 *) (out_mad->data + 46));
166         props->qkey_viol_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 48));
167         props->active_width      = out_mad->data[31] & 0xf;
168         props->active_speed      = out_mad->data[35] >> 4;
169         props->max_mtu           = out_mad->data[41] & 0xf;
170         props->active_mtu        = out_mad->data[36] >> 4;
171         props->subnet_timeout    = out_mad->data[51] & 0x1f;
172
173  out:
174         kfree(in_mad);
175         kfree(out_mad);
176         return err;
177 }
178
179 static int mthca_modify_port(struct ib_device *ibdev,
180                              u8 port, int port_modify_mask,
181                              struct ib_port_modify *props)
182 {
183         struct mthca_set_ib_param set_ib;
184         struct ib_port_attr attr;
185         int err;
186         u8 status;
187
188         if (down_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
189                 return -ERESTARTSYS;
190
191         err = mthca_query_port(ibdev, port, &attr);
192         if (err)
193                 goto out;
194
195         set_ib.set_si_guid     = 0;
196         set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
197
198         set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
199                 ~props->clr_port_cap_mask;
200
201         err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port, &status);
202         if (err)
203                 goto out;
204         if (status) {
205                 err = -EINVAL;
206                 goto out;
207         }
208
209 out:
210         up(&to_mdev(ibdev)->cap_mask_mutex);
211         return err;
212 }
213
214 static int mthca_query_pkey(struct ib_device *ibdev,
215                             u8 port, u16 index, u16 *pkey)
216 {
217         struct ib_smp *in_mad  = NULL;
218         struct ib_smp *out_mad = NULL;
219         int err = -ENOMEM;
220         u8 status;
221
222         in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
223         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
224         if (!in_mad || !out_mad)
225                 goto out;
226
227         memset(in_mad, 0, sizeof *in_mad);
228         in_mad->base_version       = 1;
229         in_mad->mgmt_class         = IB_MGMT_CLASS_SUBN_LID_ROUTED;
230         in_mad->class_version      = 1;
231         in_mad->method             = IB_MGMT_METHOD_GET;
232         in_mad->attr_id            = IB_SMP_ATTR_PKEY_TABLE;
233         in_mad->attr_mod           = cpu_to_be32(index / 32);
234
235         err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
236                             port, NULL, NULL, in_mad, out_mad,
237                             &status);
238         if (err)
239                 goto out;
240         if (status) {
241                 err = -EINVAL;
242                 goto out;
243         }
244
245         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
246
247  out:
248         kfree(in_mad);
249         kfree(out_mad);
250         return err;
251 }
252
253 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
254                            int index, union ib_gid *gid)
255 {
256         struct ib_smp *in_mad  = NULL;
257         struct ib_smp *out_mad = NULL;
258         int err = -ENOMEM;
259         u8 status;
260
261         in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
262         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
263         if (!in_mad || !out_mad)
264                 goto out;
265
266         memset(in_mad, 0, sizeof *in_mad);
267         in_mad->base_version       = 1;
268         in_mad->mgmt_class         = IB_MGMT_CLASS_SUBN_LID_ROUTED;
269         in_mad->class_version      = 1;
270         in_mad->method             = IB_MGMT_METHOD_GET;
271         in_mad->attr_id            = IB_SMP_ATTR_PORT_INFO;
272         in_mad->attr_mod           = cpu_to_be32(port);
273
274         err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
275                             port, NULL, NULL, in_mad, out_mad,
276                             &status);
277         if (err)
278                 goto out;
279         if (status) {
280                 err = -EINVAL;
281                 goto out;
282         }
283
284         memcpy(gid->raw, out_mad->data + 8, 8);
285
286         memset(in_mad, 0, sizeof *in_mad);
287         in_mad->base_version       = 1;
288         in_mad->mgmt_class         = IB_MGMT_CLASS_SUBN_LID_ROUTED;
289         in_mad->class_version      = 1;
290         in_mad->method             = IB_MGMT_METHOD_GET;
291         in_mad->attr_id            = IB_SMP_ATTR_GUID_INFO;
292         in_mad->attr_mod           = cpu_to_be32(index / 8);
293
294         err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
295                             port, NULL, NULL, in_mad, out_mad,
296                             &status);
297         if (err)
298                 goto out;
299         if (status) {
300                 err = -EINVAL;
301                 goto out;
302         }
303
304         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 16, 8);
305
306  out:
307         kfree(in_mad);
308         kfree(out_mad);
309         return err;
310 }
311
312 static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
313                                                 struct ib_udata *udata)
314 {
315         struct mthca_alloc_ucontext_resp uresp;
316         struct mthca_ucontext           *context;
317         int                              err;
318
319         memset(&uresp, 0, sizeof uresp);
320
321         uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
322         if (mthca_is_memfree(to_mdev(ibdev)))
323                 uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
324         else
325                 uresp.uarc_size = 0;
326
327         context = kmalloc(sizeof *context, GFP_KERNEL);
328         if (!context)
329                 return ERR_PTR(-ENOMEM);
330
331         err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
332         if (err) {
333                 kfree(context);
334                 return ERR_PTR(err);
335         }
336
337         context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
338         if (IS_ERR(context->db_tab)) {
339                 err = PTR_ERR(context->db_tab);
340                 mthca_uar_free(to_mdev(ibdev), &context->uar);
341                 kfree(context);
342                 return ERR_PTR(err);
343         }
344
345         if (ib_copy_to_udata(udata, &uresp, sizeof uresp)) {
346                 mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
347                 mthca_uar_free(to_mdev(ibdev), &context->uar);
348                 kfree(context);
349                 return ERR_PTR(-EFAULT);
350         }
351
352         return &context->ibucontext;
353 }
354
355 static int mthca_dealloc_ucontext(struct ib_ucontext *context)
356 {
357         mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
358                                   to_mucontext(context)->db_tab);
359         mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
360         kfree(to_mucontext(context));
361
362         return 0;
363 }
364
365 static int mthca_mmap_uar(struct ib_ucontext *context,
366                           struct vm_area_struct *vma)
367 {
368         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
369                 return -EINVAL;
370
371         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
372
373         if (io_remap_pfn_range(vma, vma->vm_start,
374                                to_mucontext(context)->uar.pfn,
375                                PAGE_SIZE, vma->vm_page_prot))
376                 return -EAGAIN;
377
378         return 0;
379 }
380
381 static struct ib_pd *mthca_alloc_pd(struct ib_device *ibdev,
382                                     struct ib_ucontext *context,
383                                     struct ib_udata *udata)
384 {
385         struct mthca_pd *pd;
386         int err;
387
388         pd = kmalloc(sizeof *pd, GFP_KERNEL);
389         if (!pd)
390                 return ERR_PTR(-ENOMEM);
391
392         err = mthca_pd_alloc(to_mdev(ibdev), !context, pd);
393         if (err) {
394                 kfree(pd);
395                 return ERR_PTR(err);
396         }
397
398         if (context) {
399                 if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
400                         mthca_pd_free(to_mdev(ibdev), pd);
401                         kfree(pd);
402                         return ERR_PTR(-EFAULT);
403                 }
404         }
405
406         return &pd->ibpd;
407 }
408
409 static int mthca_dealloc_pd(struct ib_pd *pd)
410 {
411         mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
412         kfree(pd);
413
414         return 0;
415 }
416
417 static struct ib_ah *mthca_ah_create(struct ib_pd *pd,
418                                      struct ib_ah_attr *ah_attr)
419 {
420         int err;
421         struct mthca_ah *ah;
422
423         ah = kmalloc(sizeof *ah, GFP_ATOMIC);
424         if (!ah)
425                 return ERR_PTR(-ENOMEM);
426
427         err = mthca_create_ah(to_mdev(pd->device), to_mpd(pd), ah_attr, ah);
428         if (err) {
429                 kfree(ah);
430                 return ERR_PTR(err);
431         }
432
433         return &ah->ibah;
434 }
435
436 static int mthca_ah_destroy(struct ib_ah *ah)
437 {
438         mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
439         kfree(ah);
440
441         return 0;
442 }
443
444 static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
445                                        struct ib_srq_init_attr *init_attr,
446                                        struct ib_udata *udata)
447 {
448         struct mthca_create_srq ucmd;
449         struct mthca_ucontext *context = NULL;
450         struct mthca_srq *srq;
451         int err;
452
453         srq = kmalloc(sizeof *srq, GFP_KERNEL);
454         if (!srq)
455                 return ERR_PTR(-ENOMEM);
456
457         if (pd->uobject) {
458                 context = to_mucontext(pd->uobject->context);
459
460                 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
461                         return ERR_PTR(-EFAULT);
462
463                 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
464                                         context->db_tab, ucmd.db_index,
465                                         ucmd.db_page);
466
467                 if (err)
468                         goto err_free;
469
470                 srq->mr.ibmr.lkey = ucmd.lkey;
471                 srq->db_index     = ucmd.db_index;
472         }
473
474         err = mthca_alloc_srq(to_mdev(pd->device), to_mpd(pd),
475                               &init_attr->attr, srq);
476
477         if (err && pd->uobject)
478                 mthca_unmap_user_db(to_mdev(pd->device), &context->uar,
479                                     context->db_tab, ucmd.db_index);
480
481         if (err)
482                 goto err_free;
483
484         if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof (__u32))) {
485                 mthca_free_srq(to_mdev(pd->device), srq);
486                 err = -EFAULT;
487                 goto err_free;
488         }
489
490         return &srq->ibsrq;
491
492 err_free:
493         kfree(srq);
494
495         return ERR_PTR(err);
496 }
497
498 static int mthca_destroy_srq(struct ib_srq *srq)
499 {
500         struct mthca_ucontext *context;
501
502         if (srq->uobject) {
503                 context = to_mucontext(srq->uobject->context);
504
505                 mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
506                                     context->db_tab, to_msrq(srq)->db_index);
507         }
508
509         mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
510         kfree(srq);
511
512         return 0;
513 }
514
515 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
516                                      struct ib_qp_init_attr *init_attr,
517                                      struct ib_udata *udata)
518 {
519         struct mthca_create_qp ucmd;
520         struct mthca_qp *qp;
521         int err;
522
523         switch (init_attr->qp_type) {
524         case IB_QPT_RC:
525         case IB_QPT_UC:
526         case IB_QPT_UD:
527         {
528                 struct mthca_ucontext *context;
529
530                 qp = kmalloc(sizeof *qp, GFP_KERNEL);
531                 if (!qp)
532                         return ERR_PTR(-ENOMEM);
533
534                 if (pd->uobject) {
535                         context = to_mucontext(pd->uobject->context);
536
537                         if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
538                                 return ERR_PTR(-EFAULT);
539
540                         err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
541                                                 context->db_tab,
542                                                 ucmd.sq_db_index, ucmd.sq_db_page);
543                         if (err) {
544                                 kfree(qp);
545                                 return ERR_PTR(err);
546                         }
547
548                         err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
549                                                 context->db_tab,
550                                                 ucmd.rq_db_index, ucmd.rq_db_page);
551                         if (err) {
552                                 mthca_unmap_user_db(to_mdev(pd->device),
553                                                     &context->uar,
554                                                     context->db_tab,
555                                                     ucmd.sq_db_index);
556                                 kfree(qp);
557                                 return ERR_PTR(err);
558                         }
559
560                         qp->mr.ibmr.lkey = ucmd.lkey;
561                         qp->sq.db_index  = ucmd.sq_db_index;
562                         qp->rq.db_index  = ucmd.rq_db_index;
563                 }
564
565                 err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
566                                      to_mcq(init_attr->send_cq),
567                                      to_mcq(init_attr->recv_cq),
568                                      init_attr->qp_type, init_attr->sq_sig_type,
569                                      &init_attr->cap, qp);
570
571                 if (err && pd->uobject) {
572                         context = to_mucontext(pd->uobject->context);
573
574                         mthca_unmap_user_db(to_mdev(pd->device),
575                                             &context->uar,
576                                             context->db_tab,
577                                             ucmd.sq_db_index);
578                         mthca_unmap_user_db(to_mdev(pd->device),
579                                             &context->uar,
580                                             context->db_tab,
581                                             ucmd.rq_db_index);
582                 }
583
584                 qp->ibqp.qp_num = qp->qpn;
585                 break;
586         }
587         case IB_QPT_SMI:
588         case IB_QPT_GSI:
589         {
590                 /* Don't allow userspace to create special QPs */
591                 if (pd->uobject)
592                         return ERR_PTR(-EINVAL);
593
594                 qp = kmalloc(sizeof (struct mthca_sqp), GFP_KERNEL);
595                 if (!qp)
596                         return ERR_PTR(-ENOMEM);
597
598                 qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
599
600                 err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
601                                       to_mcq(init_attr->send_cq),
602                                       to_mcq(init_attr->recv_cq),
603                                       init_attr->sq_sig_type, &init_attr->cap,
604                                       qp->ibqp.qp_num, init_attr->port_num,
605                                       to_msqp(qp));
606                 break;
607         }
608         default:
609                 /* Don't support raw QPs */
610                 return ERR_PTR(-ENOSYS);
611         }
612
613         if (err) {
614                 kfree(qp);
615                 return ERR_PTR(err);
616         }
617
618         init_attr->cap.max_inline_data = 0;
619         init_attr->cap.max_send_wr     = qp->sq.max;
620         init_attr->cap.max_recv_wr     = qp->rq.max;
621         init_attr->cap.max_send_sge    = qp->sq.max_gs;
622         init_attr->cap.max_recv_sge    = qp->rq.max_gs;
623
624         return &qp->ibqp;
625 }
626
627 static int mthca_destroy_qp(struct ib_qp *qp)
628 {
629         if (qp->uobject) {
630                 mthca_unmap_user_db(to_mdev(qp->device),
631                                     &to_mucontext(qp->uobject->context)->uar,
632                                     to_mucontext(qp->uobject->context)->db_tab,
633                                     to_mqp(qp)->sq.db_index);
634                 mthca_unmap_user_db(to_mdev(qp->device),
635                                     &to_mucontext(qp->uobject->context)->uar,
636                                     to_mucontext(qp->uobject->context)->db_tab,
637                                     to_mqp(qp)->rq.db_index);
638         }
639         mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
640         kfree(qp);
641         return 0;
642 }
643
644 static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries,
645                                      struct ib_ucontext *context,
646                                      struct ib_udata *udata)
647 {
648         struct mthca_create_cq ucmd;
649         struct mthca_cq *cq;
650         int nent;
651         int err;
652
653         if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
654                 return ERR_PTR(-EINVAL);
655
656         if (context) {
657                 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
658                         return ERR_PTR(-EFAULT);
659
660                 err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
661                                         to_mucontext(context)->db_tab,
662                                         ucmd.set_db_index, ucmd.set_db_page);
663                 if (err)
664                         return ERR_PTR(err);
665
666                 err = mthca_map_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
667                                         to_mucontext(context)->db_tab,
668                                         ucmd.arm_db_index, ucmd.arm_db_page);
669                 if (err)
670                         goto err_unmap_set;
671         }
672
673         cq = kmalloc(sizeof *cq, GFP_KERNEL);
674         if (!cq) {
675                 err = -ENOMEM;
676                 goto err_unmap_arm;
677         }
678
679         if (context) {
680                 cq->mr.ibmr.lkey    = ucmd.lkey;
681                 cq->set_ci_db_index = ucmd.set_db_index;
682                 cq->arm_db_index    = ucmd.arm_db_index;
683         }
684
685         for (nent = 1; nent <= entries; nent <<= 1)
686                 ; /* nothing */
687
688         err = mthca_init_cq(to_mdev(ibdev), nent,
689                             context ? to_mucontext(context) : NULL,
690                             context ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
691                             cq);
692         if (err)
693                 goto err_free;
694
695         if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
696                 mthca_free_cq(to_mdev(ibdev), cq);
697                 goto err_free;
698         }
699
700         return &cq->ibcq;
701
702 err_free:
703         kfree(cq);
704
705 err_unmap_arm:
706         if (context)
707                 mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
708                                     to_mucontext(context)->db_tab, ucmd.arm_db_index);
709
710 err_unmap_set:
711         if (context)
712                 mthca_unmap_user_db(to_mdev(ibdev), &to_mucontext(context)->uar,
713                                     to_mucontext(context)->db_tab, ucmd.set_db_index);
714
715         return ERR_PTR(err);
716 }
717
718 static int mthca_destroy_cq(struct ib_cq *cq)
719 {
720         if (cq->uobject) {
721                 mthca_unmap_user_db(to_mdev(cq->device),
722                                     &to_mucontext(cq->uobject->context)->uar,
723                                     to_mucontext(cq->uobject->context)->db_tab,
724                                     to_mcq(cq)->arm_db_index);
725                 mthca_unmap_user_db(to_mdev(cq->device),
726                                     &to_mucontext(cq->uobject->context)->uar,
727                                     to_mucontext(cq->uobject->context)->db_tab,
728                                     to_mcq(cq)->set_ci_db_index);
729         }
730         mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
731         kfree(cq);
732
733         return 0;
734 }
735
736 static inline u32 convert_access(int acc)
737 {
738         return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC       : 0) |
739                (acc & IB_ACCESS_REMOTE_WRITE  ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
740                (acc & IB_ACCESS_REMOTE_READ   ? MTHCA_MPT_FLAG_REMOTE_READ  : 0) |
741                (acc & IB_ACCESS_LOCAL_WRITE   ? MTHCA_MPT_FLAG_LOCAL_WRITE  : 0) |
742                MTHCA_MPT_FLAG_LOCAL_READ;
743 }
744
745 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
746 {
747         struct mthca_mr *mr;
748         int err;
749
750         mr = kmalloc(sizeof *mr, GFP_KERNEL);
751         if (!mr)
752                 return ERR_PTR(-ENOMEM);
753
754         err = mthca_mr_alloc_notrans(to_mdev(pd->device),
755                                      to_mpd(pd)->pd_num,
756                                      convert_access(acc), mr);
757
758         if (err) {
759                 kfree(mr);
760                 return ERR_PTR(err);
761         }
762
763         return &mr->ibmr;
764 }
765
766 static struct ib_mr *mthca_reg_phys_mr(struct ib_pd       *pd,
767                                        struct ib_phys_buf *buffer_list,
768                                        int                 num_phys_buf,
769                                        int                 acc,
770                                        u64                *iova_start)
771 {
772         struct mthca_mr *mr;
773         u64 *page_list;
774         u64 total_size;
775         u64 mask;
776         int shift;
777         int npages;
778         int err;
779         int i, j, n;
780
781         /* First check that we have enough alignment */
782         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK))
783                 return ERR_PTR(-EINVAL);
784
785         if (num_phys_buf > 1 &&
786             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK))
787                 return ERR_PTR(-EINVAL);
788
789         mask = 0;
790         total_size = 0;
791         for (i = 0; i < num_phys_buf; ++i) {
792                 if (i != 0 && buffer_list[i].addr & ~PAGE_MASK)
793                         return ERR_PTR(-EINVAL);
794                 if (i != 0 && i != num_phys_buf - 1 &&
795                     (buffer_list[i].size & ~PAGE_MASK))
796                         return ERR_PTR(-EINVAL);
797
798                 total_size += buffer_list[i].size;
799                 if (i > 0)
800                         mask |= buffer_list[i].addr;
801         }
802
803         /* Find largest page shift we can use to cover buffers */
804         for (shift = PAGE_SHIFT; shift < 31; ++shift)
805                 if (num_phys_buf > 1) {
806                         if ((1ULL << shift) & mask)
807                                 break;
808                 } else {
809                         if (1ULL << shift >=
810                             buffer_list[0].size +
811                             (buffer_list[0].addr & ((1ULL << shift) - 1)))
812                                 break;
813                 }
814
815         buffer_list[0].size += buffer_list[0].addr & ((1ULL << shift) - 1);
816         buffer_list[0].addr &= ~0ull << shift;
817
818         mr = kmalloc(sizeof *mr, GFP_KERNEL);
819         if (!mr)
820                 return ERR_PTR(-ENOMEM);
821
822         npages = 0;
823         for (i = 0; i < num_phys_buf; ++i)
824                 npages += (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
825
826         if (!npages)
827                 return &mr->ibmr;
828
829         page_list = kmalloc(npages * sizeof *page_list, GFP_KERNEL);
830         if (!page_list) {
831                 kfree(mr);
832                 return ERR_PTR(-ENOMEM);
833         }
834
835         n = 0;
836         for (i = 0; i < num_phys_buf; ++i)
837                 for (j = 0;
838                      j < (buffer_list[i].size + (1ULL << shift) - 1) >> shift;
839                      ++j)
840                         page_list[n++] = buffer_list[i].addr + ((u64) j << shift);
841
842         mthca_dbg(to_mdev(pd->device), "Registering memory at %llx (iova %llx) "
843                   "in PD %x; shift %d, npages %d.\n",
844                   (unsigned long long) buffer_list[0].addr,
845                   (unsigned long long) *iova_start,
846                   to_mpd(pd)->pd_num,
847                   shift, npages);
848
849         err = mthca_mr_alloc_phys(to_mdev(pd->device),
850                                   to_mpd(pd)->pd_num,
851                                   page_list, shift, npages,
852                                   *iova_start, total_size,
853                                   convert_access(acc), mr);
854
855         if (err) {
856                 kfree(page_list);
857                 kfree(mr);
858                 return ERR_PTR(err);
859         }
860
861         kfree(page_list);
862         return &mr->ibmr;
863 }
864
865 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
866                                        int acc, struct ib_udata *udata)
867 {
868         struct mthca_dev *dev = to_mdev(pd->device);
869         struct ib_umem_chunk *chunk;
870         struct mthca_mr *mr;
871         u64 *pages;
872         int shift, n, len;
873         int i, j, k;
874         int err = 0;
875
876         shift = ffs(region->page_size) - 1;
877
878         mr = kmalloc(sizeof *mr, GFP_KERNEL);
879         if (!mr)
880                 return ERR_PTR(-ENOMEM);
881
882         n = 0;
883         list_for_each_entry(chunk, &region->chunk_list, list)
884                 n += chunk->nents;
885
886         mr->mtt = mthca_alloc_mtt(dev, n);
887         if (IS_ERR(mr->mtt)) {
888                 err = PTR_ERR(mr->mtt);
889                 goto err;
890         }
891
892         pages = (u64 *) __get_free_page(GFP_KERNEL);
893         if (!pages) {
894                 err = -ENOMEM;
895                 goto err_mtt;
896         }
897
898         i = n = 0;
899
900         list_for_each_entry(chunk, &region->chunk_list, list)
901                 for (j = 0; j < chunk->nmap; ++j) {
902                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
903                         for (k = 0; k < len; ++k) {
904                                 pages[i++] = sg_dma_address(&chunk->page_list[j]) +
905                                         region->page_size * k;
906                                 /*
907                                  * Be friendly to WRITE_MTT command
908                                  * and leave two empty slots for the
909                                  * index and reserved fields of the
910                                  * mailbox.
911                                  */
912                                 if (i == PAGE_SIZE / sizeof (u64) - 2) {
913                                         err = mthca_write_mtt(dev, mr->mtt,
914                                                               n, pages, i);
915                                         if (err)
916                                                 goto mtt_done;
917                                         n += i;
918                                         i = 0;
919                                 }
920                         }
921                 }
922
923         if (i)
924                 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
925 mtt_done:
926         free_page((unsigned long) pages);
927         if (err)
928                 goto err_mtt;
929
930         err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, region->virt_base,
931                              region->length, convert_access(acc), mr);
932
933         if (err)
934                 goto err_mtt;
935
936         return &mr->ibmr;
937
938 err_mtt:
939         mthca_free_mtt(dev, mr->mtt);
940
941 err:
942         kfree(mr);
943         return ERR_PTR(err);
944 }
945
946 static int mthca_dereg_mr(struct ib_mr *mr)
947 {
948         struct mthca_mr *mmr = to_mmr(mr);
949         mthca_free_mr(to_mdev(mr->device), mmr);
950         kfree(mmr);
951         return 0;
952 }
953
954 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
955                                       struct ib_fmr_attr *fmr_attr)
956 {
957         struct mthca_fmr *fmr;
958         int err;
959
960         fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
961         if (!fmr)
962                 return ERR_PTR(-ENOMEM);
963
964         memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
965         err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
966                              convert_access(mr_access_flags), fmr);
967
968         if (err) {
969                 kfree(fmr);
970                 return ERR_PTR(err);
971         }
972
973         return &fmr->ibmr;
974 }
975
976 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
977 {
978         struct mthca_fmr *mfmr = to_mfmr(fmr);
979         int err;
980
981         err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
982         if (err)
983                 return err;
984
985         kfree(mfmr);
986         return 0;
987 }
988
989 static int mthca_unmap_fmr(struct list_head *fmr_list)
990 {
991         struct ib_fmr *fmr;
992         int err;
993         u8 status;
994         struct mthca_dev *mdev = NULL;
995
996         list_for_each_entry(fmr, fmr_list, list) {
997                 if (mdev && to_mdev(fmr->device) != mdev)
998                         return -EINVAL;
999                 mdev = to_mdev(fmr->device);
1000         }
1001
1002         if (!mdev)
1003                 return 0;
1004
1005         if (mthca_is_memfree(mdev)) {
1006                 list_for_each_entry(fmr, fmr_list, list)
1007                         mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1008
1009                 wmb();
1010         } else
1011                 list_for_each_entry(fmr, fmr_list, list)
1012                         mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1013
1014         err = mthca_SYNC_TPT(mdev, &status);
1015         if (err)
1016                 return err;
1017         if (status)
1018                 return -EINVAL;
1019         return 0;
1020 }
1021
1022 static ssize_t show_rev(struct class_device *cdev, char *buf)
1023 {
1024         struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1025         return sprintf(buf, "%x\n", dev->rev_id);
1026 }
1027
1028 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
1029 {
1030         struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1031         return sprintf(buf, "%x.%x.%x\n", (int) (dev->fw_ver >> 32),
1032                        (int) (dev->fw_ver >> 16) & 0xffff,
1033                        (int) dev->fw_ver & 0xffff);
1034 }
1035
1036 static ssize_t show_hca(struct class_device *cdev, char *buf)
1037 {
1038         struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1039         switch (dev->pdev->device) {
1040         case PCI_DEVICE_ID_MELLANOX_TAVOR:
1041                 return sprintf(buf, "MT23108\n");
1042         case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1043                 return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1044         case PCI_DEVICE_ID_MELLANOX_ARBEL:
1045                 return sprintf(buf, "MT25208\n");
1046         case PCI_DEVICE_ID_MELLANOX_SINAI:
1047         case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1048                 return sprintf(buf, "MT25204\n");
1049         default:
1050                 return sprintf(buf, "unknown\n");
1051         }
1052 }
1053
1054 static ssize_t show_board(struct class_device *cdev, char *buf)
1055 {
1056         struct mthca_dev *dev = container_of(cdev, struct mthca_dev, ib_dev.class_dev);
1057         return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1058 }
1059
1060 static CLASS_DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1061 static CLASS_DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1062 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1063 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1064
1065 static struct class_device_attribute *mthca_class_attributes[] = {
1066         &class_device_attr_hw_rev,
1067         &class_device_attr_fw_ver,
1068         &class_device_attr_hca_type,
1069         &class_device_attr_board_id
1070 };
1071
1072 int mthca_register_device(struct mthca_dev *dev)
1073 {
1074         int ret;
1075         int i;
1076
1077         strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
1078         dev->ib_dev.owner                = THIS_MODULE;
1079
1080         dev->ib_dev.uverbs_abi_ver       = MTHCA_UVERBS_ABI_VERSION;
1081         dev->ib_dev.uverbs_cmd_mask      =
1082                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1083                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1084                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1085                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1086                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1087                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1088                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1089                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1090                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1091                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1092                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1093                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1094                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1095                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1096                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1097                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1098                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1099                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1100         dev->ib_dev.node_type            = IB_NODE_CA;
1101         dev->ib_dev.phys_port_cnt        = dev->limits.num_ports;
1102         dev->ib_dev.dma_device           = &dev->pdev->dev;
1103         dev->ib_dev.class_dev.dev        = &dev->pdev->dev;
1104         dev->ib_dev.query_device         = mthca_query_device;
1105         dev->ib_dev.query_port           = mthca_query_port;
1106         dev->ib_dev.modify_port          = mthca_modify_port;
1107         dev->ib_dev.query_pkey           = mthca_query_pkey;
1108         dev->ib_dev.query_gid            = mthca_query_gid;
1109         dev->ib_dev.alloc_ucontext       = mthca_alloc_ucontext;
1110         dev->ib_dev.dealloc_ucontext     = mthca_dealloc_ucontext;
1111         dev->ib_dev.mmap                 = mthca_mmap_uar;
1112         dev->ib_dev.alloc_pd             = mthca_alloc_pd;
1113         dev->ib_dev.dealloc_pd           = mthca_dealloc_pd;
1114         dev->ib_dev.create_ah            = mthca_ah_create;
1115         dev->ib_dev.destroy_ah           = mthca_ah_destroy;
1116
1117         if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1118                 dev->ib_dev.create_srq           = mthca_create_srq;
1119                 dev->ib_dev.modify_srq           = mthca_modify_srq;
1120                 dev->ib_dev.destroy_srq          = mthca_destroy_srq;
1121
1122                 if (mthca_is_memfree(dev))
1123                         dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1124                 else
1125                         dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1126         }
1127
1128         dev->ib_dev.create_qp            = mthca_create_qp;
1129         dev->ib_dev.modify_qp            = mthca_modify_qp;
1130         dev->ib_dev.destroy_qp           = mthca_destroy_qp;
1131         dev->ib_dev.create_cq            = mthca_create_cq;
1132         dev->ib_dev.destroy_cq           = mthca_destroy_cq;
1133         dev->ib_dev.poll_cq              = mthca_poll_cq;
1134         dev->ib_dev.get_dma_mr           = mthca_get_dma_mr;
1135         dev->ib_dev.reg_phys_mr          = mthca_reg_phys_mr;
1136         dev->ib_dev.reg_user_mr          = mthca_reg_user_mr;
1137         dev->ib_dev.dereg_mr             = mthca_dereg_mr;
1138
1139         if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1140                 dev->ib_dev.alloc_fmr            = mthca_alloc_fmr;
1141                 dev->ib_dev.unmap_fmr            = mthca_unmap_fmr;
1142                 dev->ib_dev.dealloc_fmr          = mthca_dealloc_fmr;
1143                 if (mthca_is_memfree(dev))
1144                         dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1145                 else
1146                         dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1147         }
1148
1149         dev->ib_dev.attach_mcast         = mthca_multicast_attach;
1150         dev->ib_dev.detach_mcast         = mthca_multicast_detach;
1151         dev->ib_dev.process_mad          = mthca_process_mad;
1152
1153         if (mthca_is_memfree(dev)) {
1154                 dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1155                 dev->ib_dev.post_send     = mthca_arbel_post_send;
1156                 dev->ib_dev.post_recv     = mthca_arbel_post_receive;
1157         } else {
1158                 dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1159                 dev->ib_dev.post_send     = mthca_tavor_post_send;
1160                 dev->ib_dev.post_recv     = mthca_tavor_post_receive;
1161         }
1162
1163         init_MUTEX(&dev->cap_mask_mutex);
1164
1165         ret = ib_register_device(&dev->ib_dev);
1166         if (ret)
1167                 return ret;
1168
1169         for (i = 0; i < ARRAY_SIZE(mthca_class_attributes); ++i) {
1170                 ret = class_device_create_file(&dev->ib_dev.class_dev,
1171                                                mthca_class_attributes[i]);
1172                 if (ret) {
1173                         ib_unregister_device(&dev->ib_dev);
1174                         return ret;
1175                 }
1176         }
1177
1178         return 0;
1179 }
1180
1181 void mthca_unregister_device(struct mthca_dev *dev)
1182 {
1183         ib_unregister_device(&dev->ib_dev);
1184 }