RDMA/cxgb3: Don't use mm after it's freed in iwch_mmap()
[linux-2.6/btrfs-unstable.git] / drivers / infiniband / hw / cxgb3 / iwch_provider.c
blobb357c11d148f6f830e3676ffaa672a8ec404ba86
1 /*
2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
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.
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.
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>
43 #include <asm/io.h>
44 #include <asm/irq.h>
45 #include <asm/byteorder.h>
47 #include <rdma/iw_cm.h>
48 #include <rdma/ib_verbs.h>
49 #include <rdma/ib_smi.h>
50 #include <rdma/ib_user_verbs.h>
52 #include "cxio_hal.h"
53 #include "iwch.h"
54 #include "iwch_provider.h"
55 #include "iwch_cm.h"
56 #include "iwch_user.h"
58 static int iwch_modify_port(struct ib_device *ibdev,
59 u8 port, int port_modify_mask,
60 struct ib_port_modify *props)
62 return -ENOSYS;
65 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
66 struct ib_ah_attr *ah_attr)
68 return ERR_PTR(-ENOSYS);
71 static int iwch_ah_destroy(struct ib_ah *ah)
73 return -ENOSYS;
76 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
78 return -ENOSYS;
81 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
83 return -ENOSYS;
86 static int iwch_process_mad(struct ib_device *ibdev,
87 int mad_flags,
88 u8 port_num,
89 struct ib_wc *in_wc,
90 struct ib_grh *in_grh,
91 struct ib_mad *in_mad, struct ib_mad *out_mad)
93 return -ENOSYS;
96 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
98 struct iwch_dev *rhp = to_iwch_dev(context->device);
99 struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
100 struct iwch_mm_entry *mm, *tmp;
102 PDBG("%s context %p\n", __FUNCTION__, context);
103 list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
104 kfree(mm);
105 cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
106 kfree(ucontext);
107 return 0;
110 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
111 struct ib_udata *udata)
113 struct iwch_ucontext *context;
114 struct iwch_dev *rhp = to_iwch_dev(ibdev);
116 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
117 context = kzalloc(sizeof(*context), GFP_KERNEL);
118 if (!context)
119 return ERR_PTR(-ENOMEM);
120 cxio_init_ucontext(&rhp->rdev, &context->uctx);
121 INIT_LIST_HEAD(&context->mmaps);
122 spin_lock_init(&context->mmap_lock);
123 return &context->ibucontext;
126 static int iwch_destroy_cq(struct ib_cq *ib_cq)
128 struct iwch_cq *chp;
130 PDBG("%s ib_cq %p\n", __FUNCTION__, ib_cq);
131 chp = to_iwch_cq(ib_cq);
133 remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
134 atomic_dec(&chp->refcnt);
135 wait_event(chp->wait, !atomic_read(&chp->refcnt));
137 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
138 kfree(chp);
139 return 0;
142 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries,
143 struct ib_ucontext *ib_context,
144 struct ib_udata *udata)
146 struct iwch_dev *rhp;
147 struct iwch_cq *chp;
148 struct iwch_create_cq_resp uresp;
149 struct iwch_create_cq_req ureq;
150 struct iwch_ucontext *ucontext = NULL;
152 PDBG("%s ib_dev %p entries %d\n", __FUNCTION__, ibdev, entries);
153 rhp = to_iwch_dev(ibdev);
154 chp = kzalloc(sizeof(*chp), GFP_KERNEL);
155 if (!chp)
156 return ERR_PTR(-ENOMEM);
158 if (ib_context) {
159 ucontext = to_iwch_ucontext(ib_context);
160 if (!t3a_device(rhp)) {
161 if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
162 kfree(chp);
163 return ERR_PTR(-EFAULT);
165 chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
169 if (t3a_device(rhp)) {
172 * T3A: Add some fluff to handle extra CQEs inserted
173 * for various errors.
174 * Additional CQE possibilities:
175 * TERMINATE,
176 * incoming RDMA WRITE Failures
177 * incoming RDMA READ REQUEST FAILUREs
178 * NOTE: We cannot ensure the CQ won't overflow.
180 entries += 16;
182 entries = roundup_pow_of_two(entries);
183 chp->cq.size_log2 = ilog2(entries);
185 if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
186 kfree(chp);
187 return ERR_PTR(-ENOMEM);
189 chp->rhp = rhp;
190 chp->ibcq.cqe = (1 << chp->cq.size_log2) - 1;
191 spin_lock_init(&chp->lock);
192 atomic_set(&chp->refcnt, 1);
193 init_waitqueue_head(&chp->wait);
194 insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
196 if (ucontext) {
197 struct iwch_mm_entry *mm;
199 mm = kmalloc(sizeof *mm, GFP_KERNEL);
200 if (!mm) {
201 iwch_destroy_cq(&chp->ibcq);
202 return ERR_PTR(-ENOMEM);
204 uresp.cqid = chp->cq.cqid;
205 uresp.size_log2 = chp->cq.size_log2;
206 spin_lock(&ucontext->mmap_lock);
207 uresp.key = ucontext->key;
208 ucontext->key += PAGE_SIZE;
209 spin_unlock(&ucontext->mmap_lock);
210 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
211 kfree(mm);
212 iwch_destroy_cq(&chp->ibcq);
213 return ERR_PTR(-EFAULT);
215 mm->key = uresp.key;
216 mm->addr = virt_to_phys(chp->cq.queue);
217 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
218 sizeof (struct t3_cqe));
219 insert_mmap(ucontext, mm);
221 PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
222 chp->cq.cqid, chp, (1 << chp->cq.size_log2),
223 (unsigned long long) chp->cq.dma_addr);
224 return &chp->ibcq;
227 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
229 #ifdef notyet
230 struct iwch_cq *chp = to_iwch_cq(cq);
231 struct t3_cq oldcq, newcq;
232 int ret;
234 PDBG("%s ib_cq %p cqe %d\n", __FUNCTION__, cq, cqe);
236 /* We don't downsize... */
237 if (cqe <= cq->cqe)
238 return 0;
240 /* create new t3_cq with new size */
241 cqe = roundup_pow_of_two(cqe+1);
242 newcq.size_log2 = ilog2(cqe);
244 /* Dont allow resize to less than the current wce count */
245 if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
246 return -ENOMEM;
249 /* Quiesce all QPs using this CQ */
250 ret = iwch_quiesce_qps(chp);
251 if (ret) {
252 return ret;
255 ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
256 if (ret) {
257 return ret;
260 /* copy CQEs */
261 memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
262 sizeof(struct t3_cqe));
264 /* old iwch_qp gets new t3_cq but keeps old cqid */
265 oldcq = chp->cq;
266 chp->cq = newcq;
267 chp->cq.cqid = oldcq.cqid;
269 /* resize new t3_cq to update the HW context */
270 ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
271 if (ret) {
272 chp->cq = oldcq;
273 return ret;
275 chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
277 /* destroy old t3_cq */
278 oldcq.cqid = newcq.cqid;
279 ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
280 if (ret) {
281 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
282 __FUNCTION__, ret);
285 /* add user hooks here */
287 /* resume qps */
288 ret = iwch_resume_qps(chp);
289 return ret;
290 #else
291 return -ENOSYS;
292 #endif
295 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify)
297 struct iwch_dev *rhp;
298 struct iwch_cq *chp;
299 enum t3_cq_opcode cq_op;
300 int err;
301 unsigned long flag;
302 u32 rptr;
304 chp = to_iwch_cq(ibcq);
305 rhp = chp->rhp;
306 if (notify == IB_CQ_SOLICITED)
307 cq_op = CQ_ARM_SE;
308 else
309 cq_op = CQ_ARM_AN;
310 if (chp->user_rptr_addr) {
311 if (get_user(rptr, chp->user_rptr_addr))
312 return -EFAULT;
313 spin_lock_irqsave(&chp->lock, flag);
314 chp->cq.rptr = rptr;
315 } else
316 spin_lock_irqsave(&chp->lock, flag);
317 PDBG("%s rptr 0x%x\n", __FUNCTION__, chp->cq.rptr);
318 err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
319 spin_unlock_irqrestore(&chp->lock, flag);
320 if (err)
321 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
322 chp->cq.cqid);
323 return err;
326 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
328 int len = vma->vm_end - vma->vm_start;
329 u32 key = vma->vm_pgoff << PAGE_SHIFT;
330 struct cxio_rdev *rdev_p;
331 int ret = 0;
332 struct iwch_mm_entry *mm;
333 struct iwch_ucontext *ucontext;
334 u64 addr;
336 PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __FUNCTION__, vma->vm_pgoff,
337 key, len);
339 if (vma->vm_start & (PAGE_SIZE-1)) {
340 return -EINVAL;
343 rdev_p = &(to_iwch_dev(context->device)->rdev);
344 ucontext = to_iwch_ucontext(context);
346 mm = remove_mmap(ucontext, key, len);
347 if (!mm)
348 return -EINVAL;
349 addr = mm->addr;
350 kfree(mm);
352 if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
353 (addr < (rdev_p->rnic_info.udbell_physbase +
354 rdev_p->rnic_info.udbell_len))) {
357 * Map T3 DB register.
359 if (vma->vm_flags & VM_READ) {
360 return -EPERM;
363 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
364 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
365 vma->vm_flags &= ~VM_MAYREAD;
366 ret = io_remap_pfn_range(vma, vma->vm_start,
367 addr >> PAGE_SHIFT,
368 len, vma->vm_page_prot);
369 } else {
372 * Map WQ or CQ contig dma memory...
374 ret = remap_pfn_range(vma, vma->vm_start,
375 addr >> PAGE_SHIFT,
376 len, vma->vm_page_prot);
379 return ret;
382 static int iwch_deallocate_pd(struct ib_pd *pd)
384 struct iwch_dev *rhp;
385 struct iwch_pd *php;
387 php = to_iwch_pd(pd);
388 rhp = php->rhp;
389 PDBG("%s ibpd %p pdid 0x%x\n", __FUNCTION__, pd, php->pdid);
390 cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
391 kfree(php);
392 return 0;
395 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
396 struct ib_ucontext *context,
397 struct ib_udata *udata)
399 struct iwch_pd *php;
400 u32 pdid;
401 struct iwch_dev *rhp;
403 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
404 rhp = (struct iwch_dev *) ibdev;
405 pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
406 if (!pdid)
407 return ERR_PTR(-EINVAL);
408 php = kzalloc(sizeof(*php), GFP_KERNEL);
409 if (!php) {
410 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
411 return ERR_PTR(-ENOMEM);
413 php->pdid = pdid;
414 php->rhp = rhp;
415 if (context) {
416 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
417 iwch_deallocate_pd(&php->ibpd);
418 return ERR_PTR(-EFAULT);
421 PDBG("%s pdid 0x%0x ptr 0x%p\n", __FUNCTION__, pdid, php);
422 return &php->ibpd;
425 static int iwch_dereg_mr(struct ib_mr *ib_mr)
427 struct iwch_dev *rhp;
428 struct iwch_mr *mhp;
429 u32 mmid;
431 PDBG("%s ib_mr %p\n", __FUNCTION__, ib_mr);
432 /* There can be no memory windows */
433 if (atomic_read(&ib_mr->usecnt))
434 return -EINVAL;
436 mhp = to_iwch_mr(ib_mr);
437 rhp = mhp->rhp;
438 mmid = mhp->attr.stag >> 8;
439 cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
440 mhp->attr.pbl_addr);
441 remove_handle(rhp, &rhp->mmidr, mmid);
442 if (mhp->kva)
443 kfree((void *) (unsigned long) mhp->kva);
444 PDBG("%s mmid 0x%x ptr %p\n", __FUNCTION__, mmid, mhp);
445 kfree(mhp);
446 return 0;
449 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
450 struct ib_phys_buf *buffer_list,
451 int num_phys_buf,
452 int acc,
453 u64 *iova_start)
455 __be64 *page_list;
456 int shift;
457 u64 total_size;
458 int npages;
459 struct iwch_dev *rhp;
460 struct iwch_pd *php;
461 struct iwch_mr *mhp;
462 int ret;
464 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
465 php = to_iwch_pd(pd);
466 rhp = php->rhp;
468 acc = iwch_convert_access(acc);
471 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
472 if (!mhp)
473 return ERR_PTR(-ENOMEM);
475 /* First check that we have enough alignment */
476 if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
477 ret = -EINVAL;
478 goto err;
481 if (num_phys_buf > 1 &&
482 ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
483 ret = -EINVAL;
484 goto err;
487 ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
488 &total_size, &npages, &shift, &page_list);
489 if (ret)
490 goto err;
492 mhp->rhp = rhp;
493 mhp->attr.pdid = php->pdid;
494 mhp->attr.zbva = 0;
496 /* NOTE: TPT perms are backwards from BIND WR perms! */
497 mhp->attr.perms = (acc & 0x1) << 3;
498 mhp->attr.perms |= (acc & 0x2) << 1;
499 mhp->attr.perms |= (acc & 0x4) >> 1;
500 mhp->attr.perms |= (acc & 0x8) >> 3;
502 mhp->attr.va_fbo = *iova_start;
503 mhp->attr.page_size = shift - 12;
505 mhp->attr.len = (u32) total_size;
506 mhp->attr.pbl_size = npages;
507 ret = iwch_register_mem(rhp, php, mhp, shift, page_list);
508 kfree(page_list);
509 if (ret) {
510 goto err;
512 return &mhp->ibmr;
513 err:
514 kfree(mhp);
515 return ERR_PTR(ret);
519 static int iwch_reregister_phys_mem(struct ib_mr *mr,
520 int mr_rereg_mask,
521 struct ib_pd *pd,
522 struct ib_phys_buf *buffer_list,
523 int num_phys_buf,
524 int acc, u64 * iova_start)
527 struct iwch_mr mh, *mhp;
528 struct iwch_pd *php;
529 struct iwch_dev *rhp;
530 int new_acc;
531 __be64 *page_list = NULL;
532 int shift = 0;
533 u64 total_size;
534 int npages;
535 int ret;
537 PDBG("%s ib_mr %p ib_pd %p\n", __FUNCTION__, mr, pd);
539 /* There can be no memory windows */
540 if (atomic_read(&mr->usecnt))
541 return -EINVAL;
543 mhp = to_iwch_mr(mr);
544 rhp = mhp->rhp;
545 php = to_iwch_pd(mr->pd);
547 /* make sure we are on the same adapter */
548 if (rhp != php->rhp)
549 return -EINVAL;
551 new_acc = mhp->attr.perms;
553 memcpy(&mh, mhp, sizeof *mhp);
555 if (mr_rereg_mask & IB_MR_REREG_PD)
556 php = to_iwch_pd(pd);
557 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
558 mh.attr.perms = iwch_convert_access(acc);
559 if (mr_rereg_mask & IB_MR_REREG_TRANS)
560 ret = build_phys_page_list(buffer_list, num_phys_buf,
561 iova_start,
562 &total_size, &npages,
563 &shift, &page_list);
565 ret = iwch_reregister_mem(rhp, php, &mh, shift, page_list, npages);
566 kfree(page_list);
567 if (ret) {
568 return ret;
570 if (mr_rereg_mask & IB_MR_REREG_PD)
571 mhp->attr.pdid = php->pdid;
572 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
573 mhp->attr.perms = acc;
574 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
575 mhp->attr.zbva = 0;
576 mhp->attr.va_fbo = *iova_start;
577 mhp->attr.page_size = shift - 12;
578 mhp->attr.len = (u32) total_size;
579 mhp->attr.pbl_size = npages;
582 return 0;
586 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
587 int acc, struct ib_udata *udata)
589 __be64 *pages;
590 int shift, n, len;
591 int i, j, k;
592 int err = 0;
593 struct ib_umem_chunk *chunk;
594 struct iwch_dev *rhp;
595 struct iwch_pd *php;
596 struct iwch_mr *mhp;
597 struct iwch_reg_user_mr_resp uresp;
599 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
600 shift = ffs(region->page_size) - 1;
602 php = to_iwch_pd(pd);
603 rhp = php->rhp;
604 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
605 if (!mhp)
606 return ERR_PTR(-ENOMEM);
608 n = 0;
609 list_for_each_entry(chunk, &region->chunk_list, list)
610 n += chunk->nents;
612 pages = kmalloc(n * sizeof(u64), GFP_KERNEL);
613 if (!pages) {
614 err = -ENOMEM;
615 goto err;
618 acc = iwch_convert_access(acc);
620 i = n = 0;
622 list_for_each_entry(chunk, &region->chunk_list, list)
623 for (j = 0; j < chunk->nmap; ++j) {
624 len = sg_dma_len(&chunk->page_list[j]) >> shift;
625 for (k = 0; k < len; ++k) {
626 pages[i++] = cpu_to_be64(sg_dma_address(
627 &chunk->page_list[j]) +
628 region->page_size * k);
632 mhp->rhp = rhp;
633 mhp->attr.pdid = php->pdid;
634 mhp->attr.zbva = 0;
635 mhp->attr.perms = (acc & 0x1) << 3;
636 mhp->attr.perms |= (acc & 0x2) << 1;
637 mhp->attr.perms |= (acc & 0x4) >> 1;
638 mhp->attr.perms |= (acc & 0x8) >> 3;
639 mhp->attr.va_fbo = region->virt_base;
640 mhp->attr.page_size = shift - 12;
641 mhp->attr.len = (u32) region->length;
642 mhp->attr.pbl_size = i;
643 err = iwch_register_mem(rhp, php, mhp, shift, pages);
644 kfree(pages);
645 if (err)
646 goto err;
648 if (udata && t3b_device(rhp)) {
649 uresp.pbl_addr = (mhp->attr.pbl_addr -
650 rhp->rdev.rnic_info.pbl_base) >> 3;
651 PDBG("%s user resp pbl_addr 0x%x\n", __FUNCTION__,
652 uresp.pbl_addr);
654 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
655 iwch_dereg_mr(&mhp->ibmr);
656 err = -EFAULT;
657 goto err;
661 return &mhp->ibmr;
663 err:
664 kfree(mhp);
665 return ERR_PTR(err);
668 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
670 struct ib_phys_buf bl;
671 u64 kva;
672 struct ib_mr *ibmr;
674 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
677 * T3 only supports 32 bits of size.
679 bl.size = 0xffffffff;
680 bl.addr = 0;
681 kva = 0;
682 ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
683 return ibmr;
686 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
688 struct iwch_dev *rhp;
689 struct iwch_pd *php;
690 struct iwch_mw *mhp;
691 u32 mmid;
692 u32 stag = 0;
693 int ret;
695 php = to_iwch_pd(pd);
696 rhp = php->rhp;
697 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
698 if (!mhp)
699 return ERR_PTR(-ENOMEM);
700 ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
701 if (ret) {
702 kfree(mhp);
703 return ERR_PTR(ret);
705 mhp->rhp = rhp;
706 mhp->attr.pdid = php->pdid;
707 mhp->attr.type = TPT_MW;
708 mhp->attr.stag = stag;
709 mmid = (stag) >> 8;
710 insert_handle(rhp, &rhp->mmidr, mhp, mmid);
711 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __FUNCTION__, mmid, mhp, stag);
712 return &(mhp->ibmw);
715 static int iwch_dealloc_mw(struct ib_mw *mw)
717 struct iwch_dev *rhp;
718 struct iwch_mw *mhp;
719 u32 mmid;
721 mhp = to_iwch_mw(mw);
722 rhp = mhp->rhp;
723 mmid = (mw->rkey) >> 8;
724 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
725 remove_handle(rhp, &rhp->mmidr, mmid);
726 kfree(mhp);
727 PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __FUNCTION__, mw, mmid, mhp);
728 return 0;
731 static int iwch_destroy_qp(struct ib_qp *ib_qp)
733 struct iwch_dev *rhp;
734 struct iwch_qp *qhp;
735 struct iwch_qp_attributes attrs;
736 struct iwch_ucontext *ucontext;
738 qhp = to_iwch_qp(ib_qp);
739 rhp = qhp->rhp;
741 if (qhp->attr.state == IWCH_QP_STATE_RTS) {
742 attrs.next_state = IWCH_QP_STATE_ERROR;
743 iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
745 wait_event(qhp->wait, !qhp->ep);
747 remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
749 atomic_dec(&qhp->refcnt);
750 wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
752 ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
753 : NULL;
754 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
755 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
757 PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __FUNCTION__,
758 ib_qp, qhp->wq.qpid, qhp);
759 kfree(qhp);
760 return 0;
763 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
764 struct ib_qp_init_attr *attrs,
765 struct ib_udata *udata)
767 struct iwch_dev *rhp;
768 struct iwch_qp *qhp;
769 struct iwch_pd *php;
770 struct iwch_cq *schp;
771 struct iwch_cq *rchp;
772 struct iwch_create_qp_resp uresp;
773 int wqsize, sqsize, rqsize;
774 struct iwch_ucontext *ucontext;
776 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
777 if (attrs->qp_type != IB_QPT_RC)
778 return ERR_PTR(-EINVAL);
779 php = to_iwch_pd(pd);
780 rhp = php->rhp;
781 schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
782 rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
783 if (!schp || !rchp)
784 return ERR_PTR(-EINVAL);
786 /* The RQT size must be # of entries + 1 rounded up to a power of two */
787 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
788 if (rqsize == attrs->cap.max_recv_wr)
789 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
791 /* T3 doesn't support RQT depth < 16 */
792 if (rqsize < 16)
793 rqsize = 16;
795 if (rqsize > T3_MAX_RQ_SIZE)
796 return ERR_PTR(-EINVAL);
799 * NOTE: The SQ and total WQ sizes don't need to be
800 * a power of two. However, all the code assumes
801 * they are. EG: Q_FREECNT() and friends.
803 sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
804 wqsize = roundup_pow_of_two(rqsize + sqsize);
805 PDBG("%s wqsize %d sqsize %d rqsize %d\n", __FUNCTION__,
806 wqsize, sqsize, rqsize);
807 qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
808 if (!qhp)
809 return ERR_PTR(-ENOMEM);
810 qhp->wq.size_log2 = ilog2(wqsize);
811 qhp->wq.rq_size_log2 = ilog2(rqsize);
812 qhp->wq.sq_size_log2 = ilog2(sqsize);
813 ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
814 if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
815 ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
816 kfree(qhp);
817 return ERR_PTR(-ENOMEM);
819 attrs->cap.max_recv_wr = rqsize - 1;
820 attrs->cap.max_send_wr = sqsize;
821 qhp->rhp = rhp;
822 qhp->attr.pd = php->pdid;
823 qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
824 qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
825 qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
826 qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
827 qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
828 qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
829 qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
830 qhp->attr.state = IWCH_QP_STATE_IDLE;
831 qhp->attr.next_state = IWCH_QP_STATE_IDLE;
834 * XXX - These don't get passed in from the openib user
835 * at create time. The CM sets them via a QP modify.
836 * Need to fix... I think the CM should
838 qhp->attr.enable_rdma_read = 1;
839 qhp->attr.enable_rdma_write = 1;
840 qhp->attr.enable_bind = 1;
841 qhp->attr.max_ord = 1;
842 qhp->attr.max_ird = 1;
844 spin_lock_init(&qhp->lock);
845 init_waitqueue_head(&qhp->wait);
846 atomic_set(&qhp->refcnt, 1);
847 insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
849 if (udata) {
851 struct iwch_mm_entry *mm1, *mm2;
853 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
854 if (!mm1) {
855 iwch_destroy_qp(&qhp->ibqp);
856 return ERR_PTR(-ENOMEM);
859 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
860 if (!mm2) {
861 kfree(mm1);
862 iwch_destroy_qp(&qhp->ibqp);
863 return ERR_PTR(-ENOMEM);
866 uresp.qpid = qhp->wq.qpid;
867 uresp.size_log2 = qhp->wq.size_log2;
868 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
869 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
870 spin_lock(&ucontext->mmap_lock);
871 uresp.key = ucontext->key;
872 ucontext->key += PAGE_SIZE;
873 uresp.db_key = ucontext->key;
874 ucontext->key += PAGE_SIZE;
875 spin_unlock(&ucontext->mmap_lock);
876 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
877 kfree(mm1);
878 kfree(mm2);
879 iwch_destroy_qp(&qhp->ibqp);
880 return ERR_PTR(-EFAULT);
882 mm1->key = uresp.key;
883 mm1->addr = virt_to_phys(qhp->wq.queue);
884 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
885 insert_mmap(ucontext, mm1);
886 mm2->key = uresp.db_key;
887 mm2->addr = qhp->wq.udb & PAGE_MASK;
888 mm2->len = PAGE_SIZE;
889 insert_mmap(ucontext, mm2);
891 qhp->ibqp.qp_num = qhp->wq.qpid;
892 init_timer(&(qhp->timer));
893 PDBG("%s sq_num_entries %d, rq_num_entries %d "
894 "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
895 __FUNCTION__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
896 qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
897 1 << qhp->wq.size_log2);
898 return &qhp->ibqp;
901 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
902 int attr_mask, struct ib_udata *udata)
904 struct iwch_dev *rhp;
905 struct iwch_qp *qhp;
906 enum iwch_qp_attr_mask mask = 0;
907 struct iwch_qp_attributes attrs;
909 PDBG("%s ib_qp %p\n", __FUNCTION__, ibqp);
911 /* iwarp does not support the RTR state */
912 if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
913 attr_mask &= ~IB_QP_STATE;
915 /* Make sure we still have something left to do */
916 if (!attr_mask)
917 return 0;
919 memset(&attrs, 0, sizeof attrs);
920 qhp = to_iwch_qp(ibqp);
921 rhp = qhp->rhp;
923 attrs.next_state = iwch_convert_state(attr->qp_state);
924 attrs.enable_rdma_read = (attr->qp_access_flags &
925 IB_ACCESS_REMOTE_READ) ? 1 : 0;
926 attrs.enable_rdma_write = (attr->qp_access_flags &
927 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
928 attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
931 mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
932 mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
933 (IWCH_QP_ATTR_ENABLE_RDMA_READ |
934 IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
935 IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
937 return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
940 void iwch_qp_add_ref(struct ib_qp *qp)
942 PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
943 atomic_inc(&(to_iwch_qp(qp)->refcnt));
946 void iwch_qp_rem_ref(struct ib_qp *qp)
948 PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
949 if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
950 wake_up(&(to_iwch_qp(qp)->wait));
953 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
955 PDBG("%s ib_dev %p qpn 0x%x\n", __FUNCTION__, dev, qpn);
956 return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
960 static int iwch_query_pkey(struct ib_device *ibdev,
961 u8 port, u16 index, u16 * pkey)
963 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
964 *pkey = 0;
965 return 0;
968 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
969 int index, union ib_gid *gid)
971 struct iwch_dev *dev;
973 PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
974 __FUNCTION__, ibdev, port, index, gid);
975 dev = to_iwch_dev(ibdev);
976 BUG_ON(port == 0 || port > 2);
977 memset(&(gid->raw[0]), 0, sizeof(gid->raw));
978 memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
979 return 0;
982 static int iwch_query_device(struct ib_device *ibdev,
983 struct ib_device_attr *props)
986 struct iwch_dev *dev;
987 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
989 dev = to_iwch_dev(ibdev);
990 memset(props, 0, sizeof *props);
991 memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
992 props->device_cap_flags = dev->device_cap_flags;
993 props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
994 props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
995 props->max_mr_size = ~0ull;
996 props->max_qp = dev->attr.max_qps;
997 props->max_qp_wr = dev->attr.max_wrs;
998 props->max_sge = dev->attr.max_sge_per_wr;
999 props->max_sge_rd = 1;
1000 props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1001 props->max_cq = dev->attr.max_cqs;
1002 props->max_cqe = dev->attr.max_cqes_per_cq;
1003 props->max_mr = dev->attr.max_mem_regs;
1004 props->max_pd = dev->attr.max_pds;
1005 props->local_ca_ack_delay = 0;
1007 return 0;
1010 static int iwch_query_port(struct ib_device *ibdev,
1011 u8 port, struct ib_port_attr *props)
1013 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
1014 props->max_mtu = IB_MTU_4096;
1015 props->lid = 0;
1016 props->lmc = 0;
1017 props->sm_lid = 0;
1018 props->sm_sl = 0;
1019 props->state = IB_PORT_ACTIVE;
1020 props->phys_state = 0;
1021 props->port_cap_flags =
1022 IB_PORT_CM_SUP |
1023 IB_PORT_SNMP_TUNNEL_SUP |
1024 IB_PORT_REINIT_SUP |
1025 IB_PORT_DEVICE_MGMT_SUP |
1026 IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1027 props->gid_tbl_len = 1;
1028 props->pkey_tbl_len = 1;
1029 props->qkey_viol_cntr = 0;
1030 props->active_width = 2;
1031 props->active_speed = 2;
1032 props->max_msg_sz = -1;
1034 return 0;
1037 static ssize_t show_rev(struct class_device *cdev, char *buf)
1039 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1040 ibdev.class_dev);
1041 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1042 return sprintf(buf, "%d\n", dev->rdev.t3cdev_p->type);
1045 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
1047 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1048 ibdev.class_dev);
1049 struct ethtool_drvinfo info;
1050 struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
1052 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1053 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1054 return sprintf(buf, "%s\n", info.fw_version);
1057 static ssize_t show_hca(struct class_device *cdev, char *buf)
1059 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1060 ibdev.class_dev);
1061 struct ethtool_drvinfo info;
1062 struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
1064 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1065 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1066 return sprintf(buf, "%s\n", info.driver);
1069 static ssize_t show_board(struct class_device *cdev, char *buf)
1071 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1072 ibdev.class_dev);
1073 PDBG("%s class dev 0x%p\n", __FUNCTION__, dev);
1074 return sprintf(buf, "%x.%x\n", dev->rdev.rnic_info.pdev->vendor,
1075 dev->rdev.rnic_info.pdev->device);
1078 static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1079 static CLASS_DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1080 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1081 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1083 static struct class_device_attribute *iwch_class_attributes[] = {
1084 &class_device_attr_hw_rev,
1085 &class_device_attr_fw_ver,
1086 &class_device_attr_hca_type,
1087 &class_device_attr_board_id
1090 int iwch_register_device(struct iwch_dev *dev)
1092 int ret;
1093 int i;
1095 PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
1096 strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1097 memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1098 memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1099 dev->ibdev.owner = THIS_MODULE;
1100 dev->device_cap_flags =
1101 (IB_DEVICE_ZERO_STAG |
1102 IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW);
1104 dev->ibdev.uverbs_cmd_mask =
1105 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1106 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1107 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1108 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1109 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1110 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1111 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1112 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1113 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1114 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1115 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1116 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1117 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1118 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1119 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1120 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1121 (1ull << IB_USER_VERBS_CMD_POST_RECV);
1122 dev->ibdev.node_type = RDMA_NODE_RNIC;
1123 memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1124 dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1125 dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1126 dev->ibdev.class_dev.dev = &(dev->rdev.rnic_info.pdev->dev);
1127 dev->ibdev.query_device = iwch_query_device;
1128 dev->ibdev.query_port = iwch_query_port;
1129 dev->ibdev.modify_port = iwch_modify_port;
1130 dev->ibdev.query_pkey = iwch_query_pkey;
1131 dev->ibdev.query_gid = iwch_query_gid;
1132 dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1133 dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1134 dev->ibdev.mmap = iwch_mmap;
1135 dev->ibdev.alloc_pd = iwch_allocate_pd;
1136 dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1137 dev->ibdev.create_ah = iwch_ah_create;
1138 dev->ibdev.destroy_ah = iwch_ah_destroy;
1139 dev->ibdev.create_qp = iwch_create_qp;
1140 dev->ibdev.modify_qp = iwch_ib_modify_qp;
1141 dev->ibdev.destroy_qp = iwch_destroy_qp;
1142 dev->ibdev.create_cq = iwch_create_cq;
1143 dev->ibdev.destroy_cq = iwch_destroy_cq;
1144 dev->ibdev.resize_cq = iwch_resize_cq;
1145 dev->ibdev.poll_cq = iwch_poll_cq;
1146 dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1147 dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1148 dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1149 dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1150 dev->ibdev.dereg_mr = iwch_dereg_mr;
1151 dev->ibdev.alloc_mw = iwch_alloc_mw;
1152 dev->ibdev.bind_mw = iwch_bind_mw;
1153 dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1155 dev->ibdev.attach_mcast = iwch_multicast_attach;
1156 dev->ibdev.detach_mcast = iwch_multicast_detach;
1157 dev->ibdev.process_mad = iwch_process_mad;
1159 dev->ibdev.req_notify_cq = iwch_arm_cq;
1160 dev->ibdev.post_send = iwch_post_send;
1161 dev->ibdev.post_recv = iwch_post_receive;
1164 dev->ibdev.iwcm =
1165 (struct iw_cm_verbs *) kmalloc(sizeof(struct iw_cm_verbs),
1166 GFP_KERNEL);
1167 dev->ibdev.iwcm->connect = iwch_connect;
1168 dev->ibdev.iwcm->accept = iwch_accept_cr;
1169 dev->ibdev.iwcm->reject = iwch_reject_cr;
1170 dev->ibdev.iwcm->create_listen = iwch_create_listen;
1171 dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1172 dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1173 dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1174 dev->ibdev.iwcm->get_qp = iwch_get_qp;
1176 ret = ib_register_device(&dev->ibdev);
1177 if (ret)
1178 goto bail1;
1180 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1181 ret = class_device_create_file(&dev->ibdev.class_dev,
1182 iwch_class_attributes[i]);
1183 if (ret) {
1184 goto bail2;
1187 return 0;
1188 bail2:
1189 ib_unregister_device(&dev->ibdev);
1190 bail1:
1191 return ret;
1194 void iwch_unregister_device(struct iwch_dev *dev)
1196 int i;
1198 PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
1199 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1200 class_device_remove_file(&dev->ibdev.class_dev,
1201 iwch_class_attributes[i]);
1202 ib_unregister_device(&dev->ibdev);
1203 return;