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
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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
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>
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_umem.h>
51 #include <rdma/ib_user_verbs.h>
55 #include "iwch_provider.h"
57 #include "iwch_user.h"
59 static int iwch_modify_port(struct ib_device
*ibdev
,
60 u8 port
, int port_modify_mask
,
61 struct ib_port_modify
*props
)
66 static struct ib_ah
*iwch_ah_create(struct ib_pd
*pd
,
67 struct ib_ah_attr
*ah_attr
)
69 return ERR_PTR(-ENOSYS
);
72 static int iwch_ah_destroy(struct ib_ah
*ah
)
77 static int iwch_multicast_attach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
82 static int iwch_multicast_detach(struct ib_qp
*ibqp
, union ib_gid
*gid
, u16 lid
)
87 static int iwch_process_mad(struct ib_device
*ibdev
,
91 struct ib_grh
*in_grh
,
92 struct ib_mad
*in_mad
, struct ib_mad
*out_mad
)
97 static int iwch_dealloc_ucontext(struct ib_ucontext
*context
)
99 struct iwch_dev
*rhp
= to_iwch_dev(context
->device
);
100 struct iwch_ucontext
*ucontext
= to_iwch_ucontext(context
);
101 struct iwch_mm_entry
*mm
, *tmp
;
103 PDBG("%s context %p\n", __FUNCTION__
, context
);
104 list_for_each_entry_safe(mm
, tmp
, &ucontext
->mmaps
, entry
)
106 cxio_release_ucontext(&rhp
->rdev
, &ucontext
->uctx
);
111 static struct ib_ucontext
*iwch_alloc_ucontext(struct ib_device
*ibdev
,
112 struct ib_udata
*udata
)
114 struct iwch_ucontext
*context
;
115 struct iwch_dev
*rhp
= to_iwch_dev(ibdev
);
117 PDBG("%s ibdev %p\n", __FUNCTION__
, ibdev
);
118 context
= kzalloc(sizeof(*context
), GFP_KERNEL
);
120 return ERR_PTR(-ENOMEM
);
121 cxio_init_ucontext(&rhp
->rdev
, &context
->uctx
);
122 INIT_LIST_HEAD(&context
->mmaps
);
123 spin_lock_init(&context
->mmap_lock
);
124 return &context
->ibucontext
;
127 static int iwch_destroy_cq(struct ib_cq
*ib_cq
)
131 PDBG("%s ib_cq %p\n", __FUNCTION__
, ib_cq
);
132 chp
= to_iwch_cq(ib_cq
);
134 remove_handle(chp
->rhp
, &chp
->rhp
->cqidr
, chp
->cq
.cqid
);
135 atomic_dec(&chp
->refcnt
);
136 wait_event(chp
->wait
, !atomic_read(&chp
->refcnt
));
138 cxio_destroy_cq(&chp
->rhp
->rdev
, &chp
->cq
);
143 static struct ib_cq
*iwch_create_cq(struct ib_device
*ibdev
, int entries
, int vector
,
144 struct ib_ucontext
*ib_context
,
145 struct ib_udata
*udata
)
147 struct iwch_dev
*rhp
;
149 struct iwch_create_cq_resp uresp
;
150 struct iwch_create_cq_req ureq
;
151 struct iwch_ucontext
*ucontext
= NULL
;
153 PDBG("%s ib_dev %p entries %d\n", __FUNCTION__
, ibdev
, entries
);
154 rhp
= to_iwch_dev(ibdev
);
155 chp
= kzalloc(sizeof(*chp
), GFP_KERNEL
);
157 return ERR_PTR(-ENOMEM
);
160 ucontext
= to_iwch_ucontext(ib_context
);
161 if (!t3a_device(rhp
)) {
162 if (ib_copy_from_udata(&ureq
, udata
, sizeof (ureq
))) {
164 return ERR_PTR(-EFAULT
);
166 chp
->user_rptr_addr
= (u32 __user
*)(unsigned long)ureq
.user_rptr_addr
;
170 if (t3a_device(rhp
)) {
173 * T3A: Add some fluff to handle extra CQEs inserted
174 * for various errors.
175 * Additional CQE possibilities:
177 * incoming RDMA WRITE Failures
178 * incoming RDMA READ REQUEST FAILUREs
179 * NOTE: We cannot ensure the CQ won't overflow.
183 entries
= roundup_pow_of_two(entries
);
184 chp
->cq
.size_log2
= ilog2(entries
);
186 if (cxio_create_cq(&rhp
->rdev
, &chp
->cq
)) {
188 return ERR_PTR(-ENOMEM
);
191 chp
->ibcq
.cqe
= (1 << chp
->cq
.size_log2
) - 1;
192 spin_lock_init(&chp
->lock
);
193 atomic_set(&chp
->refcnt
, 1);
194 init_waitqueue_head(&chp
->wait
);
195 insert_handle(rhp
, &rhp
->cqidr
, chp
, chp
->cq
.cqid
);
198 struct iwch_mm_entry
*mm
;
200 mm
= kmalloc(sizeof *mm
, GFP_KERNEL
);
202 iwch_destroy_cq(&chp
->ibcq
);
203 return ERR_PTR(-ENOMEM
);
205 uresp
.cqid
= chp
->cq
.cqid
;
206 uresp
.size_log2
= chp
->cq
.size_log2
;
207 spin_lock(&ucontext
->mmap_lock
);
208 uresp
.key
= ucontext
->key
;
209 ucontext
->key
+= PAGE_SIZE
;
210 spin_unlock(&ucontext
->mmap_lock
);
211 if (ib_copy_to_udata(udata
, &uresp
, sizeof (uresp
))) {
213 iwch_destroy_cq(&chp
->ibcq
);
214 return ERR_PTR(-EFAULT
);
217 mm
->addr
= virt_to_phys(chp
->cq
.queue
);
218 mm
->len
= PAGE_ALIGN((1UL << uresp
.size_log2
) *
219 sizeof (struct t3_cqe
));
220 insert_mmap(ucontext
, mm
);
222 PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
223 chp
->cq
.cqid
, chp
, (1 << chp
->cq
.size_log2
),
224 (unsigned long long) chp
->cq
.dma_addr
);
228 static int iwch_resize_cq(struct ib_cq
*cq
, int cqe
, struct ib_udata
*udata
)
231 struct iwch_cq
*chp
= to_iwch_cq(cq
);
232 struct t3_cq oldcq
, newcq
;
235 PDBG("%s ib_cq %p cqe %d\n", __FUNCTION__
, cq
, cqe
);
237 /* We don't downsize... */
241 /* create new t3_cq with new size */
242 cqe
= roundup_pow_of_two(cqe
+1);
243 newcq
.size_log2
= ilog2(cqe
);
245 /* Dont allow resize to less than the current wce count */
246 if (cqe
< Q_COUNT(chp
->cq
.rptr
, chp
->cq
.wptr
)) {
250 /* Quiesce all QPs using this CQ */
251 ret
= iwch_quiesce_qps(chp
);
256 ret
= cxio_create_cq(&chp
->rhp
->rdev
, &newcq
);
262 memcpy(newcq
.queue
, chp
->cq
.queue
, (1 << chp
->cq
.size_log2
) *
263 sizeof(struct t3_cqe
));
265 /* old iwch_qp gets new t3_cq but keeps old cqid */
268 chp
->cq
.cqid
= oldcq
.cqid
;
270 /* resize new t3_cq to update the HW context */
271 ret
= cxio_resize_cq(&chp
->rhp
->rdev
, &chp
->cq
);
276 chp
->ibcq
.cqe
= (1<<chp
->cq
.size_log2
) - 1;
278 /* destroy old t3_cq */
279 oldcq
.cqid
= newcq
.cqid
;
280 ret
= cxio_destroy_cq(&chp
->rhp
->rdev
, &oldcq
);
282 printk(KERN_ERR MOD
"%s - cxio_destroy_cq failed %d\n",
286 /* add user hooks here */
289 ret
= iwch_resume_qps(chp
);
296 static int iwch_arm_cq(struct ib_cq
*ibcq
, enum ib_cq_notify_flags flags
)
298 struct iwch_dev
*rhp
;
300 enum t3_cq_opcode cq_op
;
305 chp
= to_iwch_cq(ibcq
);
307 if ((flags
& IB_CQ_SOLICITED_MASK
) == IB_CQ_SOLICITED
)
311 if (chp
->user_rptr_addr
) {
312 if (get_user(rptr
, chp
->user_rptr_addr
))
314 spin_lock_irqsave(&chp
->lock
, flag
);
317 spin_lock_irqsave(&chp
->lock
, flag
);
318 PDBG("%s rptr 0x%x\n", __FUNCTION__
, chp
->cq
.rptr
);
319 err
= cxio_hal_cq_op(&rhp
->rdev
, &chp
->cq
, cq_op
, 0);
320 spin_unlock_irqrestore(&chp
->lock
, flag
);
322 printk(KERN_ERR MOD
"Error %d rearming CQID 0x%x\n", err
,
324 if (err
> 0 && !(flags
& IB_CQ_REPORT_MISSED_EVENTS
))
329 static int iwch_mmap(struct ib_ucontext
*context
, struct vm_area_struct
*vma
)
331 int len
= vma
->vm_end
- vma
->vm_start
;
332 u32 key
= vma
->vm_pgoff
<< PAGE_SHIFT
;
333 struct cxio_rdev
*rdev_p
;
335 struct iwch_mm_entry
*mm
;
336 struct iwch_ucontext
*ucontext
;
339 PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __FUNCTION__
, vma
->vm_pgoff
,
342 if (vma
->vm_start
& (PAGE_SIZE
-1)) {
346 rdev_p
= &(to_iwch_dev(context
->device
)->rdev
);
347 ucontext
= to_iwch_ucontext(context
);
349 mm
= remove_mmap(ucontext
, key
, len
);
355 if ((addr
>= rdev_p
->rnic_info
.udbell_physbase
) &&
356 (addr
< (rdev_p
->rnic_info
.udbell_physbase
+
357 rdev_p
->rnic_info
.udbell_len
))) {
360 * Map T3 DB register.
362 if (vma
->vm_flags
& VM_READ
) {
366 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
367 vma
->vm_flags
|= VM_DONTCOPY
| VM_DONTEXPAND
;
368 vma
->vm_flags
&= ~VM_MAYREAD
;
369 ret
= io_remap_pfn_range(vma
, vma
->vm_start
,
371 len
, vma
->vm_page_prot
);
375 * Map WQ or CQ contig dma memory...
377 ret
= remap_pfn_range(vma
, vma
->vm_start
,
379 len
, vma
->vm_page_prot
);
385 static int iwch_deallocate_pd(struct ib_pd
*pd
)
387 struct iwch_dev
*rhp
;
390 php
= to_iwch_pd(pd
);
392 PDBG("%s ibpd %p pdid 0x%x\n", __FUNCTION__
, pd
, php
->pdid
);
393 cxio_hal_put_pdid(rhp
->rdev
.rscp
, php
->pdid
);
398 static struct ib_pd
*iwch_allocate_pd(struct ib_device
*ibdev
,
399 struct ib_ucontext
*context
,
400 struct ib_udata
*udata
)
404 struct iwch_dev
*rhp
;
406 PDBG("%s ibdev %p\n", __FUNCTION__
, ibdev
);
407 rhp
= (struct iwch_dev
*) ibdev
;
408 pdid
= cxio_hal_get_pdid(rhp
->rdev
.rscp
);
410 return ERR_PTR(-EINVAL
);
411 php
= kzalloc(sizeof(*php
), GFP_KERNEL
);
413 cxio_hal_put_pdid(rhp
->rdev
.rscp
, pdid
);
414 return ERR_PTR(-ENOMEM
);
419 if (ib_copy_to_udata(udata
, &php
->pdid
, sizeof (__u32
))) {
420 iwch_deallocate_pd(&php
->ibpd
);
421 return ERR_PTR(-EFAULT
);
424 PDBG("%s pdid 0x%0x ptr 0x%p\n", __FUNCTION__
, pdid
, php
);
428 static int iwch_dereg_mr(struct ib_mr
*ib_mr
)
430 struct iwch_dev
*rhp
;
434 PDBG("%s ib_mr %p\n", __FUNCTION__
, ib_mr
);
435 /* There can be no memory windows */
436 if (atomic_read(&ib_mr
->usecnt
))
439 mhp
= to_iwch_mr(ib_mr
);
441 mmid
= mhp
->attr
.stag
>> 8;
442 cxio_dereg_mem(&rhp
->rdev
, mhp
->attr
.stag
, mhp
->attr
.pbl_size
,
444 remove_handle(rhp
, &rhp
->mmidr
, mmid
);
446 kfree((void *) (unsigned long) mhp
->kva
);
448 ib_umem_release(mhp
->umem
);
449 PDBG("%s mmid 0x%x ptr %p\n", __FUNCTION__
, mmid
, mhp
);
454 static struct ib_mr
*iwch_register_phys_mem(struct ib_pd
*pd
,
455 struct ib_phys_buf
*buffer_list
,
464 struct iwch_dev
*rhp
;
469 PDBG("%s ib_pd %p\n", __FUNCTION__
, pd
);
470 php
= to_iwch_pd(pd
);
473 mhp
= kzalloc(sizeof(*mhp
), GFP_KERNEL
);
475 return ERR_PTR(-ENOMEM
);
477 /* First check that we have enough alignment */
478 if ((*iova_start
& ~PAGE_MASK
) != (buffer_list
[0].addr
& ~PAGE_MASK
)) {
483 if (num_phys_buf
> 1 &&
484 ((buffer_list
[0].addr
+ buffer_list
[0].size
) & ~PAGE_MASK
)) {
489 ret
= build_phys_page_list(buffer_list
, num_phys_buf
, iova_start
,
490 &total_size
, &npages
, &shift
, &page_list
);
495 mhp
->attr
.pdid
= php
->pdid
;
498 mhp
->attr
.perms
= iwch_ib_to_tpt_access(acc
);
499 mhp
->attr
.va_fbo
= *iova_start
;
500 mhp
->attr
.page_size
= shift
- 12;
502 mhp
->attr
.len
= (u32
) total_size
;
503 mhp
->attr
.pbl_size
= npages
;
504 ret
= iwch_register_mem(rhp
, php
, mhp
, shift
, page_list
);
516 static int iwch_reregister_phys_mem(struct ib_mr
*mr
,
519 struct ib_phys_buf
*buffer_list
,
521 int acc
, u64
* iova_start
)
524 struct iwch_mr mh
, *mhp
;
526 struct iwch_dev
*rhp
;
527 __be64
*page_list
= NULL
;
533 PDBG("%s ib_mr %p ib_pd %p\n", __FUNCTION__
, mr
, pd
);
535 /* There can be no memory windows */
536 if (atomic_read(&mr
->usecnt
))
539 mhp
= to_iwch_mr(mr
);
541 php
= to_iwch_pd(mr
->pd
);
543 /* make sure we are on the same adapter */
547 memcpy(&mh
, mhp
, sizeof *mhp
);
549 if (mr_rereg_mask
& IB_MR_REREG_PD
)
550 php
= to_iwch_pd(pd
);
551 if (mr_rereg_mask
& IB_MR_REREG_ACCESS
)
552 mh
.attr
.perms
= iwch_ib_to_tpt_access(acc
);
553 if (mr_rereg_mask
& IB_MR_REREG_TRANS
) {
554 ret
= build_phys_page_list(buffer_list
, num_phys_buf
,
556 &total_size
, &npages
,
562 ret
= iwch_reregister_mem(rhp
, php
, &mh
, shift
, page_list
, npages
);
567 if (mr_rereg_mask
& IB_MR_REREG_PD
)
568 mhp
->attr
.pdid
= php
->pdid
;
569 if (mr_rereg_mask
& IB_MR_REREG_ACCESS
)
570 mhp
->attr
.perms
= iwch_ib_to_tpt_access(acc
);
571 if (mr_rereg_mask
& IB_MR_REREG_TRANS
) {
573 mhp
->attr
.va_fbo
= *iova_start
;
574 mhp
->attr
.page_size
= shift
- 12;
575 mhp
->attr
.len
= (u32
) total_size
;
576 mhp
->attr
.pbl_size
= npages
;
583 static struct ib_mr
*iwch_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
584 u64 virt
, int acc
, struct ib_udata
*udata
)
590 struct ib_umem_chunk
*chunk
;
591 struct iwch_dev
*rhp
;
594 struct iwch_reg_user_mr_resp uresp
;
596 PDBG("%s ib_pd %p\n", __FUNCTION__
, pd
);
598 php
= to_iwch_pd(pd
);
600 mhp
= kzalloc(sizeof(*mhp
), GFP_KERNEL
);
602 return ERR_PTR(-ENOMEM
);
604 mhp
->umem
= ib_umem_get(pd
->uobject
->context
, start
, length
, acc
);
605 if (IS_ERR(mhp
->umem
)) {
606 err
= PTR_ERR(mhp
->umem
);
611 shift
= ffs(mhp
->umem
->page_size
) - 1;
614 list_for_each_entry(chunk
, &mhp
->umem
->chunk_list
, list
)
617 pages
= kmalloc(n
* sizeof(u64
), GFP_KERNEL
);
625 list_for_each_entry(chunk
, &mhp
->umem
->chunk_list
, list
)
626 for (j
= 0; j
< chunk
->nmap
; ++j
) {
627 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
628 for (k
= 0; k
< len
; ++k
) {
629 pages
[i
++] = cpu_to_be64(sg_dma_address(
630 &chunk
->page_list
[j
]) +
631 mhp
->umem
->page_size
* k
);
636 mhp
->attr
.pdid
= php
->pdid
;
638 mhp
->attr
.perms
= iwch_ib_to_tpt_access(acc
);
639 mhp
->attr
.va_fbo
= virt
;
640 mhp
->attr
.page_size
= shift
- 12;
641 mhp
->attr
.len
= (u32
) length
;
642 mhp
->attr
.pbl_size
= i
;
643 err
= iwch_register_mem(rhp
, php
, mhp
, shift
, pages
);
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__
,
654 if (ib_copy_to_udata(udata
, &uresp
, sizeof (uresp
))) {
655 iwch_dereg_mr(&mhp
->ibmr
);
664 ib_umem_release(mhp
->umem
);
669 static struct ib_mr
*iwch_get_dma_mr(struct ib_pd
*pd
, int acc
)
671 struct ib_phys_buf bl
;
675 PDBG("%s ib_pd %p\n", __FUNCTION__
, pd
);
678 * T3 only supports 32 bits of size.
680 bl
.size
= 0xffffffff;
683 ibmr
= iwch_register_phys_mem(pd
, &bl
, 1, acc
, &kva
);
687 static struct ib_mw
*iwch_alloc_mw(struct ib_pd
*pd
)
689 struct iwch_dev
*rhp
;
696 php
= to_iwch_pd(pd
);
698 mhp
= kzalloc(sizeof(*mhp
), GFP_KERNEL
);
700 return ERR_PTR(-ENOMEM
);
701 ret
= cxio_allocate_window(&rhp
->rdev
, &stag
, php
->pdid
);
707 mhp
->attr
.pdid
= php
->pdid
;
708 mhp
->attr
.type
= TPT_MW
;
709 mhp
->attr
.stag
= stag
;
711 insert_handle(rhp
, &rhp
->mmidr
, mhp
, mmid
);
712 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __FUNCTION__
, mmid
, mhp
, stag
);
716 static int iwch_dealloc_mw(struct ib_mw
*mw
)
718 struct iwch_dev
*rhp
;
722 mhp
= to_iwch_mw(mw
);
724 mmid
= (mw
->rkey
) >> 8;
725 cxio_deallocate_window(&rhp
->rdev
, mhp
->attr
.stag
);
726 remove_handle(rhp
, &rhp
->mmidr
, mmid
);
728 PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __FUNCTION__
, mw
, mmid
, mhp
);
732 static int iwch_destroy_qp(struct ib_qp
*ib_qp
)
734 struct iwch_dev
*rhp
;
736 struct iwch_qp_attributes attrs
;
737 struct iwch_ucontext
*ucontext
;
739 qhp
= to_iwch_qp(ib_qp
);
742 attrs
.next_state
= IWCH_QP_STATE_ERROR
;
743 iwch_modify_qp(rhp
, qhp
, IWCH_QP_ATTR_NEXT_STATE
, &attrs
, 0);
744 wait_event(qhp
->wait
, !qhp
->ep
);
746 remove_handle(rhp
, &rhp
->qpidr
, qhp
->wq
.qpid
);
748 atomic_dec(&qhp
->refcnt
);
749 wait_event(qhp
->wait
, !atomic_read(&qhp
->refcnt
));
751 ucontext
= ib_qp
->uobject
? to_iwch_ucontext(ib_qp
->uobject
->context
)
753 cxio_destroy_qp(&rhp
->rdev
, &qhp
->wq
,
754 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
756 PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __FUNCTION__
,
757 ib_qp
, qhp
->wq
.qpid
, qhp
);
762 static struct ib_qp
*iwch_create_qp(struct ib_pd
*pd
,
763 struct ib_qp_init_attr
*attrs
,
764 struct ib_udata
*udata
)
766 struct iwch_dev
*rhp
;
769 struct iwch_cq
*schp
;
770 struct iwch_cq
*rchp
;
771 struct iwch_create_qp_resp uresp
;
772 int wqsize
, sqsize
, rqsize
;
773 struct iwch_ucontext
*ucontext
;
775 PDBG("%s ib_pd %p\n", __FUNCTION__
, pd
);
776 if (attrs
->qp_type
!= IB_QPT_RC
)
777 return ERR_PTR(-EINVAL
);
778 php
= to_iwch_pd(pd
);
780 schp
= get_chp(rhp
, ((struct iwch_cq
*) attrs
->send_cq
)->cq
.cqid
);
781 rchp
= get_chp(rhp
, ((struct iwch_cq
*) attrs
->recv_cq
)->cq
.cqid
);
783 return ERR_PTR(-EINVAL
);
785 /* The RQT size must be # of entries + 1 rounded up to a power of two */
786 rqsize
= roundup_pow_of_two(attrs
->cap
.max_recv_wr
);
787 if (rqsize
== attrs
->cap
.max_recv_wr
)
788 rqsize
= roundup_pow_of_two(attrs
->cap
.max_recv_wr
+1);
790 /* T3 doesn't support RQT depth < 16 */
794 if (rqsize
> T3_MAX_RQ_SIZE
)
795 return ERR_PTR(-EINVAL
);
797 if (attrs
->cap
.max_inline_data
> T3_MAX_INLINE
)
798 return ERR_PTR(-EINVAL
);
801 * NOTE: The SQ and total WQ sizes don't need to be
802 * a power of two. However, all the code assumes
803 * they are. EG: Q_FREECNT() and friends.
805 sqsize
= roundup_pow_of_two(attrs
->cap
.max_send_wr
);
806 wqsize
= roundup_pow_of_two(rqsize
+ sqsize
);
807 PDBG("%s wqsize %d sqsize %d rqsize %d\n", __FUNCTION__
,
808 wqsize
, sqsize
, rqsize
);
809 qhp
= kzalloc(sizeof(*qhp
), GFP_KERNEL
);
811 return ERR_PTR(-ENOMEM
);
812 qhp
->wq
.size_log2
= ilog2(wqsize
);
813 qhp
->wq
.rq_size_log2
= ilog2(rqsize
);
814 qhp
->wq
.sq_size_log2
= ilog2(sqsize
);
815 ucontext
= pd
->uobject
? to_iwch_ucontext(pd
->uobject
->context
) : NULL
;
816 if (cxio_create_qp(&rhp
->rdev
, !udata
, &qhp
->wq
,
817 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
)) {
819 return ERR_PTR(-ENOMEM
);
821 attrs
->cap
.max_recv_wr
= rqsize
- 1;
822 attrs
->cap
.max_send_wr
= sqsize
;
824 qhp
->attr
.pd
= php
->pdid
;
825 qhp
->attr
.scq
= ((struct iwch_cq
*) attrs
->send_cq
)->cq
.cqid
;
826 qhp
->attr
.rcq
= ((struct iwch_cq
*) attrs
->recv_cq
)->cq
.cqid
;
827 qhp
->attr
.sq_num_entries
= attrs
->cap
.max_send_wr
;
828 qhp
->attr
.rq_num_entries
= attrs
->cap
.max_recv_wr
;
829 qhp
->attr
.sq_max_sges
= attrs
->cap
.max_send_sge
;
830 qhp
->attr
.sq_max_sges_rdma_write
= attrs
->cap
.max_send_sge
;
831 qhp
->attr
.rq_max_sges
= attrs
->cap
.max_recv_sge
;
832 qhp
->attr
.state
= IWCH_QP_STATE_IDLE
;
833 qhp
->attr
.next_state
= IWCH_QP_STATE_IDLE
;
836 * XXX - These don't get passed in from the openib user
837 * at create time. The CM sets them via a QP modify.
838 * Need to fix... I think the CM should
840 qhp
->attr
.enable_rdma_read
= 1;
841 qhp
->attr
.enable_rdma_write
= 1;
842 qhp
->attr
.enable_bind
= 1;
843 qhp
->attr
.max_ord
= 1;
844 qhp
->attr
.max_ird
= 1;
846 spin_lock_init(&qhp
->lock
);
847 init_waitqueue_head(&qhp
->wait
);
848 atomic_set(&qhp
->refcnt
, 1);
849 insert_handle(rhp
, &rhp
->qpidr
, qhp
, qhp
->wq
.qpid
);
853 struct iwch_mm_entry
*mm1
, *mm2
;
855 mm1
= kmalloc(sizeof *mm1
, GFP_KERNEL
);
857 iwch_destroy_qp(&qhp
->ibqp
);
858 return ERR_PTR(-ENOMEM
);
861 mm2
= kmalloc(sizeof *mm2
, GFP_KERNEL
);
864 iwch_destroy_qp(&qhp
->ibqp
);
865 return ERR_PTR(-ENOMEM
);
868 uresp
.qpid
= qhp
->wq
.qpid
;
869 uresp
.size_log2
= qhp
->wq
.size_log2
;
870 uresp
.sq_size_log2
= qhp
->wq
.sq_size_log2
;
871 uresp
.rq_size_log2
= qhp
->wq
.rq_size_log2
;
872 spin_lock(&ucontext
->mmap_lock
);
873 uresp
.key
= ucontext
->key
;
874 ucontext
->key
+= PAGE_SIZE
;
875 uresp
.db_key
= ucontext
->key
;
876 ucontext
->key
+= PAGE_SIZE
;
877 spin_unlock(&ucontext
->mmap_lock
);
878 if (ib_copy_to_udata(udata
, &uresp
, sizeof (uresp
))) {
881 iwch_destroy_qp(&qhp
->ibqp
);
882 return ERR_PTR(-EFAULT
);
884 mm1
->key
= uresp
.key
;
885 mm1
->addr
= virt_to_phys(qhp
->wq
.queue
);
886 mm1
->len
= PAGE_ALIGN(wqsize
* sizeof (union t3_wr
));
887 insert_mmap(ucontext
, mm1
);
888 mm2
->key
= uresp
.db_key
;
889 mm2
->addr
= qhp
->wq
.udb
& PAGE_MASK
;
890 mm2
->len
= PAGE_SIZE
;
891 insert_mmap(ucontext
, mm2
);
893 qhp
->ibqp
.qp_num
= qhp
->wq
.qpid
;
894 init_timer(&(qhp
->timer
));
895 PDBG("%s sq_num_entries %d, rq_num_entries %d "
896 "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
897 __FUNCTION__
, qhp
->attr
.sq_num_entries
, qhp
->attr
.rq_num_entries
,
898 qhp
->wq
.qpid
, qhp
, (unsigned long long) qhp
->wq
.dma_addr
,
899 1 << qhp
->wq
.size_log2
);
903 static int iwch_ib_modify_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*attr
,
904 int attr_mask
, struct ib_udata
*udata
)
906 struct iwch_dev
*rhp
;
908 enum iwch_qp_attr_mask mask
= 0;
909 struct iwch_qp_attributes attrs
;
911 PDBG("%s ib_qp %p\n", __FUNCTION__
, ibqp
);
913 /* iwarp does not support the RTR state */
914 if ((attr_mask
& IB_QP_STATE
) && (attr
->qp_state
== IB_QPS_RTR
))
915 attr_mask
&= ~IB_QP_STATE
;
917 /* Make sure we still have something left to do */
921 memset(&attrs
, 0, sizeof attrs
);
922 qhp
= to_iwch_qp(ibqp
);
925 attrs
.next_state
= iwch_convert_state(attr
->qp_state
);
926 attrs
.enable_rdma_read
= (attr
->qp_access_flags
&
927 IB_ACCESS_REMOTE_READ
) ? 1 : 0;
928 attrs
.enable_rdma_write
= (attr
->qp_access_flags
&
929 IB_ACCESS_REMOTE_WRITE
) ? 1 : 0;
930 attrs
.enable_bind
= (attr
->qp_access_flags
& IB_ACCESS_MW_BIND
) ? 1 : 0;
933 mask
|= (attr_mask
& IB_QP_STATE
) ? IWCH_QP_ATTR_NEXT_STATE
: 0;
934 mask
|= (attr_mask
& IB_QP_ACCESS_FLAGS
) ?
935 (IWCH_QP_ATTR_ENABLE_RDMA_READ
|
936 IWCH_QP_ATTR_ENABLE_RDMA_WRITE
|
937 IWCH_QP_ATTR_ENABLE_RDMA_BIND
) : 0;
939 return iwch_modify_qp(rhp
, qhp
, mask
, &attrs
, 0);
942 void iwch_qp_add_ref(struct ib_qp
*qp
)
944 PDBG("%s ib_qp %p\n", __FUNCTION__
, qp
);
945 atomic_inc(&(to_iwch_qp(qp
)->refcnt
));
948 void iwch_qp_rem_ref(struct ib_qp
*qp
)
950 PDBG("%s ib_qp %p\n", __FUNCTION__
, qp
);
951 if (atomic_dec_and_test(&(to_iwch_qp(qp
)->refcnt
)))
952 wake_up(&(to_iwch_qp(qp
)->wait
));
955 static struct ib_qp
*iwch_get_qp(struct ib_device
*dev
, int qpn
)
957 PDBG("%s ib_dev %p qpn 0x%x\n", __FUNCTION__
, dev
, qpn
);
958 return (struct ib_qp
*)get_qhp(to_iwch_dev(dev
), qpn
);
962 static int iwch_query_pkey(struct ib_device
*ibdev
,
963 u8 port
, u16 index
, u16
* pkey
)
965 PDBG("%s ibdev %p\n", __FUNCTION__
, ibdev
);
970 static int iwch_query_gid(struct ib_device
*ibdev
, u8 port
,
971 int index
, union ib_gid
*gid
)
973 struct iwch_dev
*dev
;
975 PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
976 __FUNCTION__
, ibdev
, port
, index
, gid
);
977 dev
= to_iwch_dev(ibdev
);
978 BUG_ON(port
== 0 || port
> 2);
979 memset(&(gid
->raw
[0]), 0, sizeof(gid
->raw
));
980 memcpy(&(gid
->raw
[0]), dev
->rdev
.port_info
.lldevs
[port
-1]->dev_addr
, 6);
984 static int iwch_query_device(struct ib_device
*ibdev
,
985 struct ib_device_attr
*props
)
988 struct iwch_dev
*dev
;
989 PDBG("%s ibdev %p\n", __FUNCTION__
, ibdev
);
991 dev
= to_iwch_dev(ibdev
);
992 memset(props
, 0, sizeof *props
);
993 memcpy(&props
->sys_image_guid
, dev
->rdev
.t3cdev_p
->lldev
->dev_addr
, 6);
994 props
->device_cap_flags
= dev
->device_cap_flags
;
995 props
->vendor_id
= (u32
)dev
->rdev
.rnic_info
.pdev
->vendor
;
996 props
->vendor_part_id
= (u32
)dev
->rdev
.rnic_info
.pdev
->device
;
997 props
->max_mr_size
= ~0ull;
998 props
->max_qp
= dev
->attr
.max_qps
;
999 props
->max_qp_wr
= dev
->attr
.max_wrs
;
1000 props
->max_sge
= dev
->attr
.max_sge_per_wr
;
1001 props
->max_sge_rd
= 1;
1002 props
->max_qp_rd_atom
= dev
->attr
.max_rdma_reads_per_qp
;
1003 props
->max_cq
= dev
->attr
.max_cqs
;
1004 props
->max_cqe
= dev
->attr
.max_cqes_per_cq
;
1005 props
->max_mr
= dev
->attr
.max_mem_regs
;
1006 props
->max_pd
= dev
->attr
.max_pds
;
1007 props
->local_ca_ack_delay
= 0;
1012 static int iwch_query_port(struct ib_device
*ibdev
,
1013 u8 port
, struct ib_port_attr
*props
)
1015 PDBG("%s ibdev %p\n", __FUNCTION__
, ibdev
);
1016 props
->max_mtu
= IB_MTU_4096
;
1021 props
->state
= IB_PORT_ACTIVE
;
1022 props
->phys_state
= 0;
1023 props
->port_cap_flags
=
1025 IB_PORT_SNMP_TUNNEL_SUP
|
1026 IB_PORT_REINIT_SUP
|
1027 IB_PORT_DEVICE_MGMT_SUP
|
1028 IB_PORT_VENDOR_CLASS_SUP
| IB_PORT_BOOT_MGMT_SUP
;
1029 props
->gid_tbl_len
= 1;
1030 props
->pkey_tbl_len
= 1;
1031 props
->qkey_viol_cntr
= 0;
1032 props
->active_width
= 2;
1033 props
->active_speed
= 2;
1034 props
->max_msg_sz
= -1;
1039 static ssize_t
show_rev(struct class_device
*cdev
, char *buf
)
1041 struct iwch_dev
*dev
= container_of(cdev
, struct iwch_dev
,
1043 PDBG("%s class dev 0x%p\n", __FUNCTION__
, cdev
);
1044 return sprintf(buf
, "%d\n", dev
->rdev
.t3cdev_p
->type
);
1047 static ssize_t
show_fw_ver(struct class_device
*cdev
, char *buf
)
1049 struct iwch_dev
*dev
= container_of(cdev
, struct iwch_dev
,
1051 struct ethtool_drvinfo info
;
1052 struct net_device
*lldev
= dev
->rdev
.t3cdev_p
->lldev
;
1054 PDBG("%s class dev 0x%p\n", __FUNCTION__
, cdev
);
1055 lldev
->ethtool_ops
->get_drvinfo(lldev
, &info
);
1056 return sprintf(buf
, "%s\n", info
.fw_version
);
1059 static ssize_t
show_hca(struct class_device
*cdev
, char *buf
)
1061 struct iwch_dev
*dev
= container_of(cdev
, struct iwch_dev
,
1063 struct ethtool_drvinfo info
;
1064 struct net_device
*lldev
= dev
->rdev
.t3cdev_p
->lldev
;
1066 PDBG("%s class dev 0x%p\n", __FUNCTION__
, cdev
);
1067 lldev
->ethtool_ops
->get_drvinfo(lldev
, &info
);
1068 return sprintf(buf
, "%s\n", info
.driver
);
1071 static ssize_t
show_board(struct class_device
*cdev
, char *buf
)
1073 struct iwch_dev
*dev
= container_of(cdev
, struct iwch_dev
,
1075 PDBG("%s class dev 0x%p\n", __FUNCTION__
, dev
);
1076 return sprintf(buf
, "%x.%x\n", dev
->rdev
.rnic_info
.pdev
->vendor
,
1077 dev
->rdev
.rnic_info
.pdev
->device
);
1080 static CLASS_DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1081 static CLASS_DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1082 static CLASS_DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1083 static CLASS_DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1085 static struct class_device_attribute
*iwch_class_attributes
[] = {
1086 &class_device_attr_hw_rev
,
1087 &class_device_attr_fw_ver
,
1088 &class_device_attr_hca_type
,
1089 &class_device_attr_board_id
1092 int iwch_register_device(struct iwch_dev
*dev
)
1097 PDBG("%s iwch_dev %p\n", __FUNCTION__
, dev
);
1098 strlcpy(dev
->ibdev
.name
, "cxgb3_%d", IB_DEVICE_NAME_MAX
);
1099 memset(&dev
->ibdev
.node_guid
, 0, sizeof(dev
->ibdev
.node_guid
));
1100 memcpy(&dev
->ibdev
.node_guid
, dev
->rdev
.t3cdev_p
->lldev
->dev_addr
, 6);
1101 dev
->ibdev
.owner
= THIS_MODULE
;
1102 dev
->device_cap_flags
=
1103 (IB_DEVICE_ZERO_STAG
|
1104 IB_DEVICE_SEND_W_INV
| IB_DEVICE_MEM_WINDOW
);
1106 dev
->ibdev
.uverbs_cmd_mask
=
1107 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1108 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1109 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1110 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1111 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1112 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1113 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1114 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1115 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1116 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1117 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ
) |
1118 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1119 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1120 (1ull << IB_USER_VERBS_CMD_POLL_CQ
) |
1121 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1122 (1ull << IB_USER_VERBS_CMD_POST_SEND
) |
1123 (1ull << IB_USER_VERBS_CMD_POST_RECV
);
1124 dev
->ibdev
.node_type
= RDMA_NODE_RNIC
;
1125 memcpy(dev
->ibdev
.node_desc
, IWCH_NODE_DESC
, sizeof(IWCH_NODE_DESC
));
1126 dev
->ibdev
.phys_port_cnt
= dev
->rdev
.port_info
.nports
;
1127 dev
->ibdev
.num_comp_vectors
= 1;
1128 dev
->ibdev
.dma_device
= &(dev
->rdev
.rnic_info
.pdev
->dev
);
1129 dev
->ibdev
.query_device
= iwch_query_device
;
1130 dev
->ibdev
.query_port
= iwch_query_port
;
1131 dev
->ibdev
.modify_port
= iwch_modify_port
;
1132 dev
->ibdev
.query_pkey
= iwch_query_pkey
;
1133 dev
->ibdev
.query_gid
= iwch_query_gid
;
1134 dev
->ibdev
.alloc_ucontext
= iwch_alloc_ucontext
;
1135 dev
->ibdev
.dealloc_ucontext
= iwch_dealloc_ucontext
;
1136 dev
->ibdev
.mmap
= iwch_mmap
;
1137 dev
->ibdev
.alloc_pd
= iwch_allocate_pd
;
1138 dev
->ibdev
.dealloc_pd
= iwch_deallocate_pd
;
1139 dev
->ibdev
.create_ah
= iwch_ah_create
;
1140 dev
->ibdev
.destroy_ah
= iwch_ah_destroy
;
1141 dev
->ibdev
.create_qp
= iwch_create_qp
;
1142 dev
->ibdev
.modify_qp
= iwch_ib_modify_qp
;
1143 dev
->ibdev
.destroy_qp
= iwch_destroy_qp
;
1144 dev
->ibdev
.create_cq
= iwch_create_cq
;
1145 dev
->ibdev
.destroy_cq
= iwch_destroy_cq
;
1146 dev
->ibdev
.resize_cq
= iwch_resize_cq
;
1147 dev
->ibdev
.poll_cq
= iwch_poll_cq
;
1148 dev
->ibdev
.get_dma_mr
= iwch_get_dma_mr
;
1149 dev
->ibdev
.reg_phys_mr
= iwch_register_phys_mem
;
1150 dev
->ibdev
.rereg_phys_mr
= iwch_reregister_phys_mem
;
1151 dev
->ibdev
.reg_user_mr
= iwch_reg_user_mr
;
1152 dev
->ibdev
.dereg_mr
= iwch_dereg_mr
;
1153 dev
->ibdev
.alloc_mw
= iwch_alloc_mw
;
1154 dev
->ibdev
.bind_mw
= iwch_bind_mw
;
1155 dev
->ibdev
.dealloc_mw
= iwch_dealloc_mw
;
1157 dev
->ibdev
.attach_mcast
= iwch_multicast_attach
;
1158 dev
->ibdev
.detach_mcast
= iwch_multicast_detach
;
1159 dev
->ibdev
.process_mad
= iwch_process_mad
;
1161 dev
->ibdev
.req_notify_cq
= iwch_arm_cq
;
1162 dev
->ibdev
.post_send
= iwch_post_send
;
1163 dev
->ibdev
.post_recv
= iwch_post_receive
;
1166 dev
->ibdev
.iwcm
= kmalloc(sizeof(struct iw_cm_verbs
), GFP_KERNEL
);
1167 if (!dev
->ibdev
.iwcm
)
1170 dev
->ibdev
.iwcm
->connect
= iwch_connect
;
1171 dev
->ibdev
.iwcm
->accept
= iwch_accept_cr
;
1172 dev
->ibdev
.iwcm
->reject
= iwch_reject_cr
;
1173 dev
->ibdev
.iwcm
->create_listen
= iwch_create_listen
;
1174 dev
->ibdev
.iwcm
->destroy_listen
= iwch_destroy_listen
;
1175 dev
->ibdev
.iwcm
->add_ref
= iwch_qp_add_ref
;
1176 dev
->ibdev
.iwcm
->rem_ref
= iwch_qp_rem_ref
;
1177 dev
->ibdev
.iwcm
->get_qp
= iwch_get_qp
;
1179 ret
= ib_register_device(&dev
->ibdev
);
1183 for (i
= 0; i
< ARRAY_SIZE(iwch_class_attributes
); ++i
) {
1184 ret
= class_device_create_file(&dev
->ibdev
.class_dev
,
1185 iwch_class_attributes
[i
]);
1192 ib_unregister_device(&dev
->ibdev
);
1197 void iwch_unregister_device(struct iwch_dev
*dev
)
1201 PDBG("%s iwch_dev %p\n", __FUNCTION__
, dev
);
1202 for (i
= 0; i
< ARRAY_SIZE(iwch_class_attributes
); ++i
)
1203 class_device_remove_file(&dev
->ibdev
.class_dev
,
1204 iwch_class_attributes
[i
]);
1205 ib_unregister_device(&dev
->ibdev
);