RDMA/cxgb3: Fix severe limit on userspace memory registration size
[linux-2.6.git] / drivers / infiniband / hw / cxgb3 / iwch_provider.c
1 /*
2  * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/device.h>
35 #include <linux/netdevice.h>
36 #include <linux/etherdevice.h>
37 #include <linux/delay.h>
38 #include <linux/errno.h>
39 #include <linux/list.h>
40 #include <linux/spinlock.h>
41 #include <linux/ethtool.h>
42 #include <linux/rtnetlink.h>
43
44 #include <asm/io.h>
45 #include <asm/irq.h>
46 #include <asm/byteorder.h>
47
48 #include <rdma/iw_cm.h>
49 #include <rdma/ib_verbs.h>
50 #include <rdma/ib_smi.h>
51 #include <rdma/ib_umem.h>
52 #include <rdma/ib_user_verbs.h>
53
54 #include "cxio_hal.h"
55 #include "iwch.h"
56 #include "iwch_provider.h"
57 #include "iwch_cm.h"
58 #include "iwch_user.h"
59
60 static int iwch_modify_port(struct ib_device *ibdev,
61                             u8 port, int port_modify_mask,
62                             struct ib_port_modify *props)
63 {
64         return -ENOSYS;
65 }
66
67 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
68                                     struct ib_ah_attr *ah_attr)
69 {
70         return ERR_PTR(-ENOSYS);
71 }
72
73 static int iwch_ah_destroy(struct ib_ah *ah)
74 {
75         return -ENOSYS;
76 }
77
78 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
79 {
80         return -ENOSYS;
81 }
82
83 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
84 {
85         return -ENOSYS;
86 }
87
88 static int iwch_process_mad(struct ib_device *ibdev,
89                             int mad_flags,
90                             u8 port_num,
91                             struct ib_wc *in_wc,
92                             struct ib_grh *in_grh,
93                             struct ib_mad *in_mad, struct ib_mad *out_mad)
94 {
95         return -ENOSYS;
96 }
97
98 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
99 {
100         struct iwch_dev *rhp = to_iwch_dev(context->device);
101         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
102         struct iwch_mm_entry *mm, *tmp;
103
104         PDBG("%s context %p\n", __func__, context);
105         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
106                 kfree(mm);
107         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
108         kfree(ucontext);
109         return 0;
110 }
111
112 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
113                                         struct ib_udata *udata)
114 {
115         struct iwch_ucontext *context;
116         struct iwch_dev *rhp = to_iwch_dev(ibdev);
117
118         PDBG("%s ibdev %p\n", __func__, ibdev);
119         context = kzalloc(sizeof(*context), GFP_KERNEL);
120         if (!context)
121                 return ERR_PTR(-ENOMEM);
122         cxio_init_ucontext(&rhp->rdev, &context->uctx);
123         INIT_LIST_HEAD(&context->mmaps);
124         spin_lock_init(&context->mmap_lock);
125         return &context->ibucontext;
126 }
127
128 static int iwch_destroy_cq(struct ib_cq *ib_cq)
129 {
130         struct iwch_cq *chp;
131
132         PDBG("%s ib_cq %p\n", __func__, ib_cq);
133         chp = to_iwch_cq(ib_cq);
134
135         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
136         atomic_dec(&chp->refcnt);
137         wait_event(chp->wait, !atomic_read(&chp->refcnt));
138
139         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
140         kfree(chp);
141         return 0;
142 }
143
144 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
145                              struct ib_ucontext *ib_context,
146                              struct ib_udata *udata)
147 {
148         struct iwch_dev *rhp;
149         struct iwch_cq *chp;
150         struct iwch_create_cq_resp uresp;
151         struct iwch_create_cq_req ureq;
152         struct iwch_ucontext *ucontext = NULL;
153
154         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
155         rhp = to_iwch_dev(ibdev);
156         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
157         if (!chp)
158                 return ERR_PTR(-ENOMEM);
159
160         if (ib_context) {
161                 ucontext = to_iwch_ucontext(ib_context);
162                 if (!t3a_device(rhp)) {
163                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
164                                 kfree(chp);
165                                 return ERR_PTR(-EFAULT);
166                         }
167                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
168                 }
169         }
170
171         if (t3a_device(rhp)) {
172
173                 /*
174                  * T3A: Add some fluff to handle extra CQEs inserted
175                  * for various errors.
176                  * Additional CQE possibilities:
177                  *      TERMINATE,
178                  *      incoming RDMA WRITE Failures
179                  *      incoming RDMA READ REQUEST FAILUREs
180                  * NOTE: We cannot ensure the CQ won't overflow.
181                  */
182                 entries += 16;
183         }
184         entries = roundup_pow_of_two(entries);
185         chp->cq.size_log2 = ilog2(entries);
186
187         if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
188                 kfree(chp);
189                 return ERR_PTR(-ENOMEM);
190         }
191         chp->rhp = rhp;
192         chp->ibcq.cqe = 1 << chp->cq.size_log2;
193         spin_lock_init(&chp->lock);
194         atomic_set(&chp->refcnt, 1);
195         init_waitqueue_head(&chp->wait);
196         insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
197
198         if (ucontext) {
199                 struct iwch_mm_entry *mm;
200
201                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
202                 if (!mm) {
203                         iwch_destroy_cq(&chp->ibcq);
204                         return ERR_PTR(-ENOMEM);
205                 }
206                 uresp.cqid = chp->cq.cqid;
207                 uresp.size_log2 = chp->cq.size_log2;
208                 spin_lock(&ucontext->mmap_lock);
209                 uresp.key = ucontext->key;
210                 ucontext->key += PAGE_SIZE;
211                 spin_unlock(&ucontext->mmap_lock);
212                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
213                         kfree(mm);
214                         iwch_destroy_cq(&chp->ibcq);
215                         return ERR_PTR(-EFAULT);
216                 }
217                 mm->key = uresp.key;
218                 mm->addr = virt_to_phys(chp->cq.queue);
219                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
220                                              sizeof (struct t3_cqe));
221                 insert_mmap(ucontext, mm);
222         }
223         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
224              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
225              (unsigned long long) chp->cq.dma_addr);
226         return &chp->ibcq;
227 }
228
229 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
230 {
231 #ifdef notyet
232         struct iwch_cq *chp = to_iwch_cq(cq);
233         struct t3_cq oldcq, newcq;
234         int ret;
235
236         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
237
238         /* We don't downsize... */
239         if (cqe <= cq->cqe)
240                 return 0;
241
242         /* create new t3_cq with new size */
243         cqe = roundup_pow_of_two(cqe+1);
244         newcq.size_log2 = ilog2(cqe);
245
246         /* Dont allow resize to less than the current wce count */
247         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
248                 return -ENOMEM;
249         }
250
251         /* Quiesce all QPs using this CQ */
252         ret = iwch_quiesce_qps(chp);
253         if (ret) {
254                 return ret;
255         }
256
257         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
258         if (ret) {
259                 return ret;
260         }
261
262         /* copy CQEs */
263         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
264                                         sizeof(struct t3_cqe));
265
266         /* old iwch_qp gets new t3_cq but keeps old cqid */
267         oldcq = chp->cq;
268         chp->cq = newcq;
269         chp->cq.cqid = oldcq.cqid;
270
271         /* resize new t3_cq to update the HW context */
272         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
273         if (ret) {
274                 chp->cq = oldcq;
275                 return ret;
276         }
277         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
278
279         /* destroy old t3_cq */
280         oldcq.cqid = newcq.cqid;
281         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
282         if (ret) {
283                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
284                         __func__, ret);
285         }
286
287         /* add user hooks here */
288
289         /* resume qps */
290         ret = iwch_resume_qps(chp);
291         return ret;
292 #else
293         return -ENOSYS;
294 #endif
295 }
296
297 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
298 {
299         struct iwch_dev *rhp;
300         struct iwch_cq *chp;
301         enum t3_cq_opcode cq_op;
302         int err;
303         unsigned long flag;
304         u32 rptr;
305
306         chp = to_iwch_cq(ibcq);
307         rhp = chp->rhp;
308         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
309                 cq_op = CQ_ARM_SE;
310         else
311                 cq_op = CQ_ARM_AN;
312         if (chp->user_rptr_addr) {
313                 if (get_user(rptr, chp->user_rptr_addr))
314                         return -EFAULT;
315                 spin_lock_irqsave(&chp->lock, flag);
316                 chp->cq.rptr = rptr;
317         } else
318                 spin_lock_irqsave(&chp->lock, flag);
319         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
320         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
321         spin_unlock_irqrestore(&chp->lock, flag);
322         if (err < 0)
323                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
324                        chp->cq.cqid);
325         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
326                 err = 0;
327         return err;
328 }
329
330 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
331 {
332         int len = vma->vm_end - vma->vm_start;
333         u32 key = vma->vm_pgoff << PAGE_SHIFT;
334         struct cxio_rdev *rdev_p;
335         int ret = 0;
336         struct iwch_mm_entry *mm;
337         struct iwch_ucontext *ucontext;
338         u64 addr;
339
340         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
341              key, len);
342
343         if (vma->vm_start & (PAGE_SIZE-1)) {
344                 return -EINVAL;
345         }
346
347         rdev_p = &(to_iwch_dev(context->device)->rdev);
348         ucontext = to_iwch_ucontext(context);
349
350         mm = remove_mmap(ucontext, key, len);
351         if (!mm)
352                 return -EINVAL;
353         addr = mm->addr;
354         kfree(mm);
355
356         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
357             (addr < (rdev_p->rnic_info.udbell_physbase +
358                        rdev_p->rnic_info.udbell_len))) {
359
360                 /*
361                  * Map T3 DB register.
362                  */
363                 if (vma->vm_flags & VM_READ) {
364                         return -EPERM;
365                 }
366
367                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
368                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
369                 vma->vm_flags &= ~VM_MAYREAD;
370                 ret = io_remap_pfn_range(vma, vma->vm_start,
371                                          addr >> PAGE_SHIFT,
372                                          len, vma->vm_page_prot);
373         } else {
374
375                 /*
376                  * Map WQ or CQ contig dma memory...
377                  */
378                 ret = remap_pfn_range(vma, vma->vm_start,
379                                       addr >> PAGE_SHIFT,
380                                       len, vma->vm_page_prot);
381         }
382
383         return ret;
384 }
385
386 static int iwch_deallocate_pd(struct ib_pd *pd)
387 {
388         struct iwch_dev *rhp;
389         struct iwch_pd *php;
390
391         php = to_iwch_pd(pd);
392         rhp = php->rhp;
393         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
394         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
395         kfree(php);
396         return 0;
397 }
398
399 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
400                                struct ib_ucontext *context,
401                                struct ib_udata *udata)
402 {
403         struct iwch_pd *php;
404         u32 pdid;
405         struct iwch_dev *rhp;
406
407         PDBG("%s ibdev %p\n", __func__, ibdev);
408         rhp = (struct iwch_dev *) ibdev;
409         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
410         if (!pdid)
411                 return ERR_PTR(-EINVAL);
412         php = kzalloc(sizeof(*php), GFP_KERNEL);
413         if (!php) {
414                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
415                 return ERR_PTR(-ENOMEM);
416         }
417         php->pdid = pdid;
418         php->rhp = rhp;
419         if (context) {
420                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
421                         iwch_deallocate_pd(&php->ibpd);
422                         return ERR_PTR(-EFAULT);
423                 }
424         }
425         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
426         return &php->ibpd;
427 }
428
429 static int iwch_dereg_mr(struct ib_mr *ib_mr)
430 {
431         struct iwch_dev *rhp;
432         struct iwch_mr *mhp;
433         u32 mmid;
434
435         PDBG("%s ib_mr %p\n", __func__, ib_mr);
436         /* There can be no memory windows */
437         if (atomic_read(&ib_mr->usecnt))
438                 return -EINVAL;
439
440         mhp = to_iwch_mr(ib_mr);
441         rhp = mhp->rhp;
442         mmid = mhp->attr.stag >> 8;
443         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
444                        mhp->attr.pbl_addr);
445         iwch_free_pbl(mhp);
446         remove_handle(rhp, &rhp->mmidr, mmid);
447         if (mhp->kva)
448                 kfree((void *) (unsigned long) mhp->kva);
449         if (mhp->umem)
450                 ib_umem_release(mhp->umem);
451         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
452         kfree(mhp);
453         return 0;
454 }
455
456 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
457                                         struct ib_phys_buf *buffer_list,
458                                         int num_phys_buf,
459                                         int acc,
460                                         u64 *iova_start)
461 {
462         __be64 *page_list;
463         int shift;
464         u64 total_size;
465         int npages;
466         struct iwch_dev *rhp;
467         struct iwch_pd *php;
468         struct iwch_mr *mhp;
469         int ret;
470
471         PDBG("%s ib_pd %p\n", __func__, pd);
472         php = to_iwch_pd(pd);
473         rhp = php->rhp;
474
475         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
476         if (!mhp)
477                 return ERR_PTR(-ENOMEM);
478
479         mhp->rhp = rhp;
480
481         /* First check that we have enough alignment */
482         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
483                 ret = -EINVAL;
484                 goto err;
485         }
486
487         if (num_phys_buf > 1 &&
488             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
489                 ret = -EINVAL;
490                 goto err;
491         }
492
493         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
494                                    &total_size, &npages, &shift, &page_list);
495         if (ret)
496                 goto err;
497
498         ret = iwch_alloc_pbl(mhp, npages);
499         if (ret) {
500                 kfree(page_list);
501                 goto err_pbl;
502         }
503
504         ret = iwch_write_pbl(mhp, page_list, npages, 0);
505         kfree(page_list);
506         if (ret)
507                 goto err_pbl;
508
509         mhp->attr.pdid = php->pdid;
510         mhp->attr.zbva = 0;
511
512         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
513         mhp->attr.va_fbo = *iova_start;
514         mhp->attr.page_size = shift - 12;
515
516         mhp->attr.len = (u32) total_size;
517         mhp->attr.pbl_size = npages;
518         ret = iwch_register_mem(rhp, php, mhp, shift);
519         if (ret)
520                 goto err_pbl;
521
522         return &mhp->ibmr;
523
524 err_pbl:
525         iwch_free_pbl(mhp);
526
527 err:
528         kfree(mhp);
529         return ERR_PTR(ret);
530
531 }
532
533 static int iwch_reregister_phys_mem(struct ib_mr *mr,
534                                      int mr_rereg_mask,
535                                      struct ib_pd *pd,
536                                      struct ib_phys_buf *buffer_list,
537                                      int num_phys_buf,
538                                      int acc, u64 * iova_start)
539 {
540
541         struct iwch_mr mh, *mhp;
542         struct iwch_pd *php;
543         struct iwch_dev *rhp;
544         __be64 *page_list = NULL;
545         int shift = 0;
546         u64 total_size;
547         int npages;
548         int ret;
549
550         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
551
552         /* There can be no memory windows */
553         if (atomic_read(&mr->usecnt))
554                 return -EINVAL;
555
556         mhp = to_iwch_mr(mr);
557         rhp = mhp->rhp;
558         php = to_iwch_pd(mr->pd);
559
560         /* make sure we are on the same adapter */
561         if (rhp != php->rhp)
562                 return -EINVAL;
563
564         memcpy(&mh, mhp, sizeof *mhp);
565
566         if (mr_rereg_mask & IB_MR_REREG_PD)
567                 php = to_iwch_pd(pd);
568         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
569                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
570         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
571                 ret = build_phys_page_list(buffer_list, num_phys_buf,
572                                            iova_start,
573                                            &total_size, &npages,
574                                            &shift, &page_list);
575                 if (ret)
576                         return ret;
577         }
578
579         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
580         kfree(page_list);
581         if (ret) {
582                 return ret;
583         }
584         if (mr_rereg_mask & IB_MR_REREG_PD)
585                 mhp->attr.pdid = php->pdid;
586         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
587                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
588         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
589                 mhp->attr.zbva = 0;
590                 mhp->attr.va_fbo = *iova_start;
591                 mhp->attr.page_size = shift - 12;
592                 mhp->attr.len = (u32) total_size;
593                 mhp->attr.pbl_size = npages;
594         }
595
596         return 0;
597 }
598
599
600 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
601                                       u64 virt, int acc, struct ib_udata *udata)
602 {
603         __be64 *pages;
604         int shift, n, len;
605         int i, j, k;
606         int err = 0;
607         struct ib_umem_chunk *chunk;
608         struct iwch_dev *rhp;
609         struct iwch_pd *php;
610         struct iwch_mr *mhp;
611         struct iwch_reg_user_mr_resp uresp;
612
613         PDBG("%s ib_pd %p\n", __func__, pd);
614
615         php = to_iwch_pd(pd);
616         rhp = php->rhp;
617         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
618         if (!mhp)
619                 return ERR_PTR(-ENOMEM);
620
621         mhp->rhp = rhp;
622
623         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
624         if (IS_ERR(mhp->umem)) {
625                 err = PTR_ERR(mhp->umem);
626                 kfree(mhp);
627                 return ERR_PTR(err);
628         }
629
630         shift = ffs(mhp->umem->page_size) - 1;
631
632         n = 0;
633         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
634                 n += chunk->nents;
635
636         err = iwch_alloc_pbl(mhp, n);
637         if (err)
638                 goto err;
639
640         pages = (__be64 *) __get_free_page(GFP_KERNEL);
641         if (!pages) {
642                 err = -ENOMEM;
643                 goto err_pbl;
644         }
645
646         i = n = 0;
647
648         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
649                 for (j = 0; j < chunk->nmap; ++j) {
650                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
651                         for (k = 0; k < len; ++k) {
652                                 pages[i++] = cpu_to_be64(sg_dma_address(
653                                         &chunk->page_list[j]) +
654                                         mhp->umem->page_size * k);
655                                 if (i == PAGE_SIZE / sizeof *pages) {
656                                         err = iwch_write_pbl(mhp, pages, i, n);
657                                         if (err)
658                                                 goto pbl_done;
659                                         n += i;
660                                         i = 0;
661                                 }
662                         }
663                 }
664
665         if (i)
666                 err = iwch_write_pbl(mhp, pages, i, n);
667
668 pbl_done:
669         free_page((unsigned long) pages);
670         if (err)
671                 goto err_pbl;
672
673         mhp->attr.pdid = php->pdid;
674         mhp->attr.zbva = 0;
675         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
676         mhp->attr.va_fbo = virt;
677         mhp->attr.page_size = shift - 12;
678         mhp->attr.len = (u32) length;
679
680         err = iwch_register_mem(rhp, php, mhp, shift);
681         if (err)
682                 goto err_pbl;
683
684         if (udata && !t3a_device(rhp)) {
685                 uresp.pbl_addr = (mhp->attr.pbl_addr -
686                                  rhp->rdev.rnic_info.pbl_base) >> 3;
687                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
688                      uresp.pbl_addr);
689
690                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
691                         iwch_dereg_mr(&mhp->ibmr);
692                         err = -EFAULT;
693                         goto err;
694                 }
695         }
696
697         return &mhp->ibmr;
698
699 err_pbl:
700         iwch_free_pbl(mhp);
701
702 err:
703         ib_umem_release(mhp->umem);
704         kfree(mhp);
705         return ERR_PTR(err);
706 }
707
708 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
709 {
710         struct ib_phys_buf bl;
711         u64 kva;
712         struct ib_mr *ibmr;
713
714         PDBG("%s ib_pd %p\n", __func__, pd);
715
716         /*
717          * T3 only supports 32 bits of size.
718          */
719         bl.size = 0xffffffff;
720         bl.addr = 0;
721         kva = 0;
722         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
723         return ibmr;
724 }
725
726 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
727 {
728         struct iwch_dev *rhp;
729         struct iwch_pd *php;
730         struct iwch_mw *mhp;
731         u32 mmid;
732         u32 stag = 0;
733         int ret;
734
735         php = to_iwch_pd(pd);
736         rhp = php->rhp;
737         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
738         if (!mhp)
739                 return ERR_PTR(-ENOMEM);
740         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
741         if (ret) {
742                 kfree(mhp);
743                 return ERR_PTR(ret);
744         }
745         mhp->rhp = rhp;
746         mhp->attr.pdid = php->pdid;
747         mhp->attr.type = TPT_MW;
748         mhp->attr.stag = stag;
749         mmid = (stag) >> 8;
750         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
751         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
752         return &(mhp->ibmw);
753 }
754
755 static int iwch_dealloc_mw(struct ib_mw *mw)
756 {
757         struct iwch_dev *rhp;
758         struct iwch_mw *mhp;
759         u32 mmid;
760
761         mhp = to_iwch_mw(mw);
762         rhp = mhp->rhp;
763         mmid = (mw->rkey) >> 8;
764         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
765         remove_handle(rhp, &rhp->mmidr, mmid);
766         kfree(mhp);
767         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
768         return 0;
769 }
770
771 static int iwch_destroy_qp(struct ib_qp *ib_qp)
772 {
773         struct iwch_dev *rhp;
774         struct iwch_qp *qhp;
775         struct iwch_qp_attributes attrs;
776         struct iwch_ucontext *ucontext;
777
778         qhp = to_iwch_qp(ib_qp);
779         rhp = qhp->rhp;
780
781         attrs.next_state = IWCH_QP_STATE_ERROR;
782         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
783         wait_event(qhp->wait, !qhp->ep);
784
785         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
786
787         atomic_dec(&qhp->refcnt);
788         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
789
790         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
791                                   : NULL;
792         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
793                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
794
795         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
796              ib_qp, qhp->wq.qpid, qhp);
797         kfree(qhp);
798         return 0;
799 }
800
801 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
802                              struct ib_qp_init_attr *attrs,
803                              struct ib_udata *udata)
804 {
805         struct iwch_dev *rhp;
806         struct iwch_qp *qhp;
807         struct iwch_pd *php;
808         struct iwch_cq *schp;
809         struct iwch_cq *rchp;
810         struct iwch_create_qp_resp uresp;
811         int wqsize, sqsize, rqsize;
812         struct iwch_ucontext *ucontext;
813
814         PDBG("%s ib_pd %p\n", __func__, pd);
815         if (attrs->qp_type != IB_QPT_RC)
816                 return ERR_PTR(-EINVAL);
817         php = to_iwch_pd(pd);
818         rhp = php->rhp;
819         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
820         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
821         if (!schp || !rchp)
822                 return ERR_PTR(-EINVAL);
823
824         /* The RQT size must be # of entries + 1 rounded up to a power of two */
825         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
826         if (rqsize == attrs->cap.max_recv_wr)
827                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
828
829         /* T3 doesn't support RQT depth < 16 */
830         if (rqsize < 16)
831                 rqsize = 16;
832
833         if (rqsize > T3_MAX_RQ_SIZE)
834                 return ERR_PTR(-EINVAL);
835
836         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
837                 return ERR_PTR(-EINVAL);
838
839         /*
840          * NOTE: The SQ and total WQ sizes don't need to be
841          * a power of two.  However, all the code assumes
842          * they are. EG: Q_FREECNT() and friends.
843          */
844         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
845         wqsize = roundup_pow_of_two(rqsize + sqsize);
846         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
847              wqsize, sqsize, rqsize);
848         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
849         if (!qhp)
850                 return ERR_PTR(-ENOMEM);
851         qhp->wq.size_log2 = ilog2(wqsize);
852         qhp->wq.rq_size_log2 = ilog2(rqsize);
853         qhp->wq.sq_size_log2 = ilog2(sqsize);
854         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
855         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
856                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
857                 kfree(qhp);
858                 return ERR_PTR(-ENOMEM);
859         }
860
861         attrs->cap.max_recv_wr = rqsize - 1;
862         attrs->cap.max_send_wr = sqsize;
863         attrs->cap.max_inline_data = T3_MAX_INLINE;
864
865         qhp->rhp = rhp;
866         qhp->attr.pd = php->pdid;
867         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
868         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
869         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
870         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
871         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
872         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
873         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
874         qhp->attr.state = IWCH_QP_STATE_IDLE;
875         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
876
877         /*
878          * XXX - These don't get passed in from the openib user
879          * at create time.  The CM sets them via a QP modify.
880          * Need to fix...  I think the CM should
881          */
882         qhp->attr.enable_rdma_read = 1;
883         qhp->attr.enable_rdma_write = 1;
884         qhp->attr.enable_bind = 1;
885         qhp->attr.max_ord = 1;
886         qhp->attr.max_ird = 1;
887
888         spin_lock_init(&qhp->lock);
889         init_waitqueue_head(&qhp->wait);
890         atomic_set(&qhp->refcnt, 1);
891         insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
892
893         if (udata) {
894
895                 struct iwch_mm_entry *mm1, *mm2;
896
897                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
898                 if (!mm1) {
899                         iwch_destroy_qp(&qhp->ibqp);
900                         return ERR_PTR(-ENOMEM);
901                 }
902
903                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
904                 if (!mm2) {
905                         kfree(mm1);
906                         iwch_destroy_qp(&qhp->ibqp);
907                         return ERR_PTR(-ENOMEM);
908                 }
909
910                 uresp.qpid = qhp->wq.qpid;
911                 uresp.size_log2 = qhp->wq.size_log2;
912                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
913                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
914                 spin_lock(&ucontext->mmap_lock);
915                 uresp.key = ucontext->key;
916                 ucontext->key += PAGE_SIZE;
917                 uresp.db_key = ucontext->key;
918                 ucontext->key += PAGE_SIZE;
919                 spin_unlock(&ucontext->mmap_lock);
920                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
921                         kfree(mm1);
922                         kfree(mm2);
923                         iwch_destroy_qp(&qhp->ibqp);
924                         return ERR_PTR(-EFAULT);
925                 }
926                 mm1->key = uresp.key;
927                 mm1->addr = virt_to_phys(qhp->wq.queue);
928                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
929                 insert_mmap(ucontext, mm1);
930                 mm2->key = uresp.db_key;
931                 mm2->addr = qhp->wq.udb & PAGE_MASK;
932                 mm2->len = PAGE_SIZE;
933                 insert_mmap(ucontext, mm2);
934         }
935         qhp->ibqp.qp_num = qhp->wq.qpid;
936         init_timer(&(qhp->timer));
937         PDBG("%s sq_num_entries %d, rq_num_entries %d "
938              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
939              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
940              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
941              1 << qhp->wq.size_log2);
942         return &qhp->ibqp;
943 }
944
945 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
946                       int attr_mask, struct ib_udata *udata)
947 {
948         struct iwch_dev *rhp;
949         struct iwch_qp *qhp;
950         enum iwch_qp_attr_mask mask = 0;
951         struct iwch_qp_attributes attrs;
952
953         PDBG("%s ib_qp %p\n", __func__, ibqp);
954
955         /* iwarp does not support the RTR state */
956         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
957                 attr_mask &= ~IB_QP_STATE;
958
959         /* Make sure we still have something left to do */
960         if (!attr_mask)
961                 return 0;
962
963         memset(&attrs, 0, sizeof attrs);
964         qhp = to_iwch_qp(ibqp);
965         rhp = qhp->rhp;
966
967         attrs.next_state = iwch_convert_state(attr->qp_state);
968         attrs.enable_rdma_read = (attr->qp_access_flags &
969                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
970         attrs.enable_rdma_write = (attr->qp_access_flags &
971                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
972         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
973
974
975         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
976         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
977                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
978                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
979                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
980
981         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
982 }
983
984 void iwch_qp_add_ref(struct ib_qp *qp)
985 {
986         PDBG("%s ib_qp %p\n", __func__, qp);
987         atomic_inc(&(to_iwch_qp(qp)->refcnt));
988 }
989
990 void iwch_qp_rem_ref(struct ib_qp *qp)
991 {
992         PDBG("%s ib_qp %p\n", __func__, qp);
993         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
994                 wake_up(&(to_iwch_qp(qp)->wait));
995 }
996
997 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
998 {
999         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1000         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1001 }
1002
1003
1004 static int iwch_query_pkey(struct ib_device *ibdev,
1005                            u8 port, u16 index, u16 * pkey)
1006 {
1007         PDBG("%s ibdev %p\n", __func__, ibdev);
1008         *pkey = 0;
1009         return 0;
1010 }
1011
1012 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1013                           int index, union ib_gid *gid)
1014 {
1015         struct iwch_dev *dev;
1016
1017         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1018                __func__, ibdev, port, index, gid);
1019         dev = to_iwch_dev(ibdev);
1020         BUG_ON(port == 0 || port > 2);
1021         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1022         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1023         return 0;
1024 }
1025
1026 static int iwch_query_device(struct ib_device *ibdev,
1027                              struct ib_device_attr *props)
1028 {
1029
1030         struct iwch_dev *dev;
1031         PDBG("%s ibdev %p\n", __func__, ibdev);
1032
1033         dev = to_iwch_dev(ibdev);
1034         memset(props, 0, sizeof *props);
1035         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1036         props->device_cap_flags = dev->device_cap_flags;
1037         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1038         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1039         props->max_mr_size = dev->attr.max_mr_size;
1040         props->max_qp = dev->attr.max_qps;
1041         props->max_qp_wr = dev->attr.max_wrs;
1042         props->max_sge = dev->attr.max_sge_per_wr;
1043         props->max_sge_rd = 1;
1044         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1045         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1046         props->max_cq = dev->attr.max_cqs;
1047         props->max_cqe = dev->attr.max_cqes_per_cq;
1048         props->max_mr = dev->attr.max_mem_regs;
1049         props->max_pd = dev->attr.max_pds;
1050         props->local_ca_ack_delay = 0;
1051
1052         return 0;
1053 }
1054
1055 static int iwch_query_port(struct ib_device *ibdev,
1056                            u8 port, struct ib_port_attr *props)
1057 {
1058         PDBG("%s ibdev %p\n", __func__, ibdev);
1059         props->max_mtu = IB_MTU_4096;
1060         props->lid = 0;
1061         props->lmc = 0;
1062         props->sm_lid = 0;
1063         props->sm_sl = 0;
1064         props->state = IB_PORT_ACTIVE;
1065         props->phys_state = 0;
1066         props->port_cap_flags =
1067             IB_PORT_CM_SUP |
1068             IB_PORT_SNMP_TUNNEL_SUP |
1069             IB_PORT_REINIT_SUP |
1070             IB_PORT_DEVICE_MGMT_SUP |
1071             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1072         props->gid_tbl_len = 1;
1073         props->pkey_tbl_len = 1;
1074         props->qkey_viol_cntr = 0;
1075         props->active_width = 2;
1076         props->active_speed = 2;
1077         props->max_msg_sz = -1;
1078
1079         return 0;
1080 }
1081
1082 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1083                         char *buf)
1084 {
1085         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1086                                                  ibdev.dev);
1087         PDBG("%s dev 0x%p\n", __func__, dev);
1088         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1089 }
1090
1091 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1092 {
1093         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1094                                                  ibdev.dev);
1095         struct ethtool_drvinfo info;
1096         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1097
1098         PDBG("%s dev 0x%p\n", __func__, dev);
1099         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1100         return sprintf(buf, "%s\n", info.fw_version);
1101 }
1102
1103 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1104                         char *buf)
1105 {
1106         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1107                                                  ibdev.dev);
1108         struct ethtool_drvinfo info;
1109         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1110
1111         PDBG("%s dev 0x%p\n", __func__, dev);
1112         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1113         return sprintf(buf, "%s\n", info.driver);
1114 }
1115
1116 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1117                           char *buf)
1118 {
1119         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1120                                                  ibdev.dev);
1121         PDBG("%s dev 0x%p\n", __func__, dev);
1122         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1123                        iwch_dev->rdev.rnic_info.pdev->device);
1124 }
1125
1126 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1127 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1128 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1129 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1130
1131 static struct device_attribute *iwch_class_attributes[] = {
1132         &dev_attr_hw_rev,
1133         &dev_attr_fw_ver,
1134         &dev_attr_hca_type,
1135         &dev_attr_board_id
1136 };
1137
1138 int iwch_register_device(struct iwch_dev *dev)
1139 {
1140         int ret;
1141         int i;
1142
1143         PDBG("%s iwch_dev %p\n", __func__, dev);
1144         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1145         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1146         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1147         dev->ibdev.owner = THIS_MODULE;
1148         dev->device_cap_flags =
1149             (IB_DEVICE_ZERO_STAG | IB_DEVICE_MEM_WINDOW);
1150
1151         dev->ibdev.uverbs_cmd_mask =
1152             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1153             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1154             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1155             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1156             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1157             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1158             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1159             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1160             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1161             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1162             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1163             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1164             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1165             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1166             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1167             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1168             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1169         dev->ibdev.node_type = RDMA_NODE_RNIC;
1170         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1171         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1172         dev->ibdev.num_comp_vectors = 1;
1173         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1174         dev->ibdev.query_device = iwch_query_device;
1175         dev->ibdev.query_port = iwch_query_port;
1176         dev->ibdev.modify_port = iwch_modify_port;
1177         dev->ibdev.query_pkey = iwch_query_pkey;
1178         dev->ibdev.query_gid = iwch_query_gid;
1179         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1180         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1181         dev->ibdev.mmap = iwch_mmap;
1182         dev->ibdev.alloc_pd = iwch_allocate_pd;
1183         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1184         dev->ibdev.create_ah = iwch_ah_create;
1185         dev->ibdev.destroy_ah = iwch_ah_destroy;
1186         dev->ibdev.create_qp = iwch_create_qp;
1187         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1188         dev->ibdev.destroy_qp = iwch_destroy_qp;
1189         dev->ibdev.create_cq = iwch_create_cq;
1190         dev->ibdev.destroy_cq = iwch_destroy_cq;
1191         dev->ibdev.resize_cq = iwch_resize_cq;
1192         dev->ibdev.poll_cq = iwch_poll_cq;
1193         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1194         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1195         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1196         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1197         dev->ibdev.dereg_mr = iwch_dereg_mr;
1198         dev->ibdev.alloc_mw = iwch_alloc_mw;
1199         dev->ibdev.bind_mw = iwch_bind_mw;
1200         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1201
1202         dev->ibdev.attach_mcast = iwch_multicast_attach;
1203         dev->ibdev.detach_mcast = iwch_multicast_detach;
1204         dev->ibdev.process_mad = iwch_process_mad;
1205
1206         dev->ibdev.req_notify_cq = iwch_arm_cq;
1207         dev->ibdev.post_send = iwch_post_send;
1208         dev->ibdev.post_recv = iwch_post_receive;
1209
1210
1211         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1212         if (!dev->ibdev.iwcm)
1213                 return -ENOMEM;
1214
1215         dev->ibdev.iwcm->connect = iwch_connect;
1216         dev->ibdev.iwcm->accept = iwch_accept_cr;
1217         dev->ibdev.iwcm->reject = iwch_reject_cr;
1218         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1219         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1220         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1221         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1222         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1223
1224         ret = ib_register_device(&dev->ibdev);
1225         if (ret)
1226                 goto bail1;
1227
1228         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1229                 ret = device_create_file(&dev->ibdev.dev,
1230                                          iwch_class_attributes[i]);
1231                 if (ret) {
1232                         goto bail2;
1233                 }
1234         }
1235         return 0;
1236 bail2:
1237         ib_unregister_device(&dev->ibdev);
1238 bail1:
1239         return ret;
1240 }
1241
1242 void iwch_unregister_device(struct iwch_dev *dev)
1243 {
1244         int i;
1245
1246         PDBG("%s iwch_dev %p\n", __func__, dev);
1247         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1248                 device_remove_file(&dev->ibdev.dev,
1249                                    iwch_class_attributes[i]);
1250         ib_unregister_device(&dev->ibdev);
1251         return;
1252 }