2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
3 * Copyright (c) 2006 Open Grid Computing, Inc. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/module.h>
34 #include <linux/moduleparam.h>
35 #include <linux/device.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/delay.h>
39 #include <linux/errno.h>
40 #include <linux/list.h>
41 #include <linux/spinlock.h>
42 #include <linux/ethtool.h>
46 #include <asm/byteorder.h>
48 #include <rdma/iw_cm.h>
49 #include <rdma/ib_verbs.h>
50 #include <rdma/ib_smi.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,
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 notify)
298 struct iwch_dev *rhp;
300 enum t3_cq_opcode cq_op;
305 chp = to_iwch_cq(ibcq);
307 if (notify == 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,
327 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
329 int len = vma->vm_end - vma->vm_start;
330 u32 key = vma->vm_pgoff << PAGE_SHIFT;
331 struct cxio_rdev *rdev_p;
333 struct iwch_mm_entry *mm;
334 struct iwch_ucontext *ucontext;
336 PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __FUNCTION__, vma->vm_pgoff,
339 if (vma->vm_start & (PAGE_SIZE-1)) {
343 rdev_p = &(to_iwch_dev(context->device)->rdev);
344 ucontext = to_iwch_ucontext(context);
346 mm = remove_mmap(ucontext, key, len);
351 if ((mm->addr >= rdev_p->rnic_info.udbell_physbase) &&
352 (mm->addr < (rdev_p->rnic_info.udbell_physbase +
353 rdev_p->rnic_info.udbell_len))) {
356 * Map T3 DB register.
358 if (vma->vm_flags & VM_READ) {
362 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
363 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
364 vma->vm_flags &= ~VM_MAYREAD;
365 ret = io_remap_pfn_range(vma, vma->vm_start,
366 mm->addr >> PAGE_SHIFT,
367 len, vma->vm_page_prot);
371 * Map WQ or CQ contig dma memory...
373 ret = remap_pfn_range(vma, vma->vm_start,
374 mm->addr >> PAGE_SHIFT,
375 len, vma->vm_page_prot);
381 static int iwch_deallocate_pd(struct ib_pd *pd)
383 struct iwch_dev *rhp;
386 php = to_iwch_pd(pd);
388 PDBG("%s ibpd %p pdid 0x%x\n", __FUNCTION__, pd, php->pdid);
389 cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
394 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
395 struct ib_ucontext *context,
396 struct ib_udata *udata)
400 struct iwch_dev *rhp;
402 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
403 rhp = (struct iwch_dev *) ibdev;
404 pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
406 return ERR_PTR(-EINVAL);
407 php = kzalloc(sizeof(*php), GFP_KERNEL);
409 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
410 return ERR_PTR(-ENOMEM);
415 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
416 iwch_deallocate_pd(&php->ibpd);
417 return ERR_PTR(-EFAULT);
420 PDBG("%s pdid 0x%0x ptr 0x%p\n", __FUNCTION__, pdid, php);
424 static int iwch_dereg_mr(struct ib_mr *ib_mr)
426 struct iwch_dev *rhp;
430 PDBG("%s ib_mr %p\n", __FUNCTION__, ib_mr);
431 /* There can be no memory windows */
432 if (atomic_read(&ib_mr->usecnt))
435 mhp = to_iwch_mr(ib_mr);
437 mmid = mhp->attr.stag >> 8;
438 cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
440 remove_handle(rhp, &rhp->mmidr, mmid);
442 kfree((void *) (unsigned long) mhp->kva);
443 PDBG("%s mmid 0x%x ptr %p\n", __FUNCTION__, mmid, mhp);
448 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
449 struct ib_phys_buf *buffer_list,
458 struct iwch_dev *rhp;
463 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
464 php = to_iwch_pd(pd);
467 acc = iwch_convert_access(acc);
470 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
472 return ERR_PTR(-ENOMEM);
474 /* First check that we have enough alignment */
475 if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
480 if (num_phys_buf > 1 &&
481 ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
486 ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
487 &total_size, &npages, &shift, &page_list);
492 mhp->attr.pdid = php->pdid;
495 /* NOTE: TPT perms are backwards from BIND WR perms! */
496 mhp->attr.perms = (acc & 0x1) << 3;
497 mhp->attr.perms |= (acc & 0x2) << 1;
498 mhp->attr.perms |= (acc & 0x4) >> 1;
499 mhp->attr.perms |= (acc & 0x8) >> 3;
501 mhp->attr.va_fbo = *iova_start;
502 mhp->attr.page_size = shift - 12;
504 mhp->attr.len = (u32) total_size;
505 mhp->attr.pbl_size = npages;
506 ret = iwch_register_mem(rhp, php, mhp, shift, page_list);
518 static int iwch_reregister_phys_mem(struct ib_mr *mr,
521 struct ib_phys_buf *buffer_list,
523 int acc, u64 * iova_start)
526 struct iwch_mr mh, *mhp;
528 struct iwch_dev *rhp;
530 __be64 *page_list = NULL;
536 PDBG("%s ib_mr %p ib_pd %p\n", __FUNCTION__, mr, pd);
538 /* There can be no memory windows */
539 if (atomic_read(&mr->usecnt))
542 mhp = to_iwch_mr(mr);
544 php = to_iwch_pd(mr->pd);
546 /* make sure we are on the same adapter */
550 new_acc = mhp->attr.perms;
552 memcpy(&mh, mhp, sizeof *mhp);
554 if (mr_rereg_mask & IB_MR_REREG_PD)
555 php = to_iwch_pd(pd);
556 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
557 mh.attr.perms = iwch_convert_access(acc);
558 if (mr_rereg_mask & IB_MR_REREG_TRANS)
559 ret = build_phys_page_list(buffer_list, num_phys_buf,
561 &total_size, &npages,
564 ret = iwch_reregister_mem(rhp, php, &mh, shift, page_list, npages);
569 if (mr_rereg_mask & IB_MR_REREG_PD)
570 mhp->attr.pdid = php->pdid;
571 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
572 mhp->attr.perms = acc;
573 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
575 mhp->attr.va_fbo = *iova_start;
576 mhp->attr.page_size = shift - 12;
577 mhp->attr.len = (u32) total_size;
578 mhp->attr.pbl_size = npages;
585 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, struct ib_umem *region,
586 int acc, struct ib_udata *udata)
592 struct ib_umem_chunk *chunk;
593 struct iwch_dev *rhp;
596 struct iwch_reg_user_mr_resp uresp;
598 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
599 shift = ffs(region->page_size) - 1;
601 php = to_iwch_pd(pd);
603 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
605 return ERR_PTR(-ENOMEM);
608 list_for_each_entry(chunk, ®ion->chunk_list, list)
611 pages = kmalloc(n * sizeof(u64), GFP_KERNEL);
617 acc = iwch_convert_access(acc);
621 list_for_each_entry(chunk, ®ion->chunk_list, list)
622 for (j = 0; j < chunk->nmap; ++j) {
623 len = sg_dma_len(&chunk->page_list[j]) >> shift;
624 for (k = 0; k < len; ++k) {
625 pages[i++] = cpu_to_be64(sg_dma_address(
626 &chunk->page_list[j]) +
627 region->page_size * k);
632 mhp->attr.pdid = php->pdid;
634 mhp->attr.perms = (acc & 0x1) << 3;
635 mhp->attr.perms |= (acc & 0x2) << 1;
636 mhp->attr.perms |= (acc & 0x4) >> 1;
637 mhp->attr.perms |= (acc & 0x8) >> 3;
638 mhp->attr.va_fbo = region->virt_base;
639 mhp->attr.page_size = shift - 12;
640 mhp->attr.len = (u32) region->length;
641 mhp->attr.pbl_size = i;
642 err = iwch_register_mem(rhp, php, mhp, shift, pages);
647 if (udata && t3b_device(rhp)) {
648 uresp.pbl_addr = (mhp->attr.pbl_addr -
649 rhp->rdev.rnic_info.pbl_base) >> 3;
650 PDBG("%s user resp pbl_addr 0x%x\n", __FUNCTION__,
653 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
654 iwch_dereg_mr(&mhp->ibmr);
667 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
669 struct ib_phys_buf bl;
673 PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
676 * T3 only supports 32 bits of size.
678 bl.size = 0xffffffff;
681 ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
685 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
687 struct iwch_dev *rhp;
694 php = to_iwch_pd(pd);
696 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
698 return ERR_PTR(-ENOMEM);
699 ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
705 mhp->attr.pdid = php->pdid;
706 mhp->attr.type = TPT_MW;
707 mhp->attr.stag = stag;
709 insert_handle(rhp, &rhp->mmidr, mhp, mmid);
710 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __FUNCTION__, mmid, mhp, stag);
714 static int iwch_dealloc_mw(struct ib_mw *mw)
716 struct iwch_dev *rhp;
720 mhp = to_iwch_mw(mw);
722 mmid = (mw->rkey) >> 8;
723 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
724 remove_handle(rhp, &rhp->mmidr, mmid);
726 PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __FUNCTION__, mw, mmid, mhp);
730 static int iwch_destroy_qp(struct ib_qp *ib_qp)
732 struct iwch_dev *rhp;
734 struct iwch_qp_attributes attrs;
735 struct iwch_ucontext *ucontext;
737 qhp = to_iwch_qp(ib_qp);
740 if (qhp->attr.state == IWCH_QP_STATE_RTS) {
741 attrs.next_state = IWCH_QP_STATE_ERROR;
742 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);
798 * NOTE: The SQ and total WQ sizes don't need to be
799 * a power of two. However, all the code assumes
800 * they are. EG: Q_FREECNT() and friends.
802 sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
803 wqsize = roundup_pow_of_two(rqsize + sqsize);
804 PDBG("%s wqsize %d sqsize %d rqsize %d\n", __FUNCTION__,
805 wqsize, sqsize, rqsize);
806 qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
808 return ERR_PTR(-ENOMEM);
809 qhp->wq.size_log2 = ilog2(wqsize);
810 qhp->wq.rq_size_log2 = ilog2(rqsize);
811 qhp->wq.sq_size_log2 = ilog2(sqsize);
812 ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
813 if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
814 ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
816 return ERR_PTR(-ENOMEM);
818 attrs->cap.max_recv_wr = rqsize - 1;
819 attrs->cap.max_send_wr = sqsize;
821 qhp->attr.pd = php->pdid;
822 qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
823 qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
824 qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
825 qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
826 qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
827 qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
828 qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
829 qhp->attr.state = IWCH_QP_STATE_IDLE;
830 qhp->attr.next_state = IWCH_QP_STATE_IDLE;
833 * XXX - These don't get passed in from the openib user
834 * at create time. The CM sets them via a QP modify.
835 * Need to fix... I think the CM should
837 qhp->attr.enable_rdma_read = 1;
838 qhp->attr.enable_rdma_write = 1;
839 qhp->attr.enable_bind = 1;
840 qhp->attr.max_ord = 1;
841 qhp->attr.max_ird = 1;
843 spin_lock_init(&qhp->lock);
844 init_waitqueue_head(&qhp->wait);
845 atomic_set(&qhp->refcnt, 1);
846 insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
850 struct iwch_mm_entry *mm1, *mm2;
852 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
854 iwch_destroy_qp(&qhp->ibqp);
855 return ERR_PTR(-ENOMEM);
858 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
861 iwch_destroy_qp(&qhp->ibqp);
862 return ERR_PTR(-ENOMEM);
865 uresp.qpid = qhp->wq.qpid;
866 uresp.size_log2 = qhp->wq.size_log2;
867 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
868 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
869 spin_lock(&ucontext->mmap_lock);
870 uresp.key = ucontext->key;
871 ucontext->key += PAGE_SIZE;
872 uresp.db_key = ucontext->key;
873 ucontext->key += PAGE_SIZE;
874 spin_unlock(&ucontext->mmap_lock);
875 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
878 iwch_destroy_qp(&qhp->ibqp);
879 return ERR_PTR(-EFAULT);
881 mm1->key = uresp.key;
882 mm1->addr = virt_to_phys(qhp->wq.queue);
883 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
884 insert_mmap(ucontext, mm1);
885 mm2->key = uresp.db_key;
886 mm2->addr = qhp->wq.udb & PAGE_MASK;
887 mm2->len = PAGE_SIZE;
888 insert_mmap(ucontext, mm2);
890 qhp->ibqp.qp_num = qhp->wq.qpid;
891 init_timer(&(qhp->timer));
892 PDBG("%s sq_num_entries %d, rq_num_entries %d "
893 "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
894 __FUNCTION__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
895 qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
896 1 << qhp->wq.size_log2);
900 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
901 int attr_mask, struct ib_udata *udata)
903 struct iwch_dev *rhp;
905 enum iwch_qp_attr_mask mask = 0;
906 struct iwch_qp_attributes attrs;
908 PDBG("%s ib_qp %p\n", __FUNCTION__, ibqp);
910 /* iwarp does not support the RTR state */
911 if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
912 attr_mask &= ~IB_QP_STATE;
914 /* Make sure we still have something left to do */
918 memset(&attrs, 0, sizeof attrs);
919 qhp = to_iwch_qp(ibqp);
922 attrs.next_state = iwch_convert_state(attr->qp_state);
923 attrs.enable_rdma_read = (attr->qp_access_flags &
924 IB_ACCESS_REMOTE_READ) ? 1 : 0;
925 attrs.enable_rdma_write = (attr->qp_access_flags &
926 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
927 attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
930 mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
931 mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
932 (IWCH_QP_ATTR_ENABLE_RDMA_READ |
933 IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
934 IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
936 return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
939 void iwch_qp_add_ref(struct ib_qp *qp)
941 PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
942 atomic_inc(&(to_iwch_qp(qp)->refcnt));
945 void iwch_qp_rem_ref(struct ib_qp *qp)
947 PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
948 if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
949 wake_up(&(to_iwch_qp(qp)->wait));
952 struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
954 PDBG("%s ib_dev %p qpn 0x%x\n", __FUNCTION__, dev, qpn);
955 return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
959 static int iwch_query_pkey(struct ib_device *ibdev,
960 u8 port, u16 index, u16 * pkey)
962 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
967 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
968 int index, union ib_gid *gid)
970 struct iwch_dev *dev;
972 PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
973 __FUNCTION__, ibdev, port, index, gid);
974 dev = to_iwch_dev(ibdev);
975 BUG_ON(port == 0 || port > 2);
976 memset(&(gid->raw[0]), 0, sizeof(gid->raw));
977 memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
981 static int iwch_query_device(struct ib_device *ibdev,
982 struct ib_device_attr *props)
985 struct iwch_dev *dev;
986 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
988 dev = to_iwch_dev(ibdev);
989 memset(props, 0, sizeof *props);
990 memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
991 props->device_cap_flags = dev->device_cap_flags;
992 props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
993 props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
994 props->max_mr_size = ~0ull;
995 props->max_qp = dev->attr.max_qps;
996 props->max_qp_wr = dev->attr.max_wrs;
997 props->max_sge = dev->attr.max_sge_per_wr;
998 props->max_sge_rd = 1;
999 props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1000 props->max_cq = dev->attr.max_cqs;
1001 props->max_cqe = dev->attr.max_cqes_per_cq;
1002 props->max_mr = dev->attr.max_mem_regs;
1003 props->max_pd = dev->attr.max_pds;
1004 props->local_ca_ack_delay = 0;
1009 static int iwch_query_port(struct ib_device *ibdev,
1010 u8 port, struct ib_port_attr *props)
1012 PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
1013 props->max_mtu = IB_MTU_4096;
1018 props->state = IB_PORT_ACTIVE;
1019 props->phys_state = 0;
1020 props->port_cap_flags =
1022 IB_PORT_SNMP_TUNNEL_SUP |
1023 IB_PORT_REINIT_SUP |
1024 IB_PORT_DEVICE_MGMT_SUP |
1025 IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1026 props->gid_tbl_len = 1;
1027 props->pkey_tbl_len = 1;
1028 props->qkey_viol_cntr = 0;
1029 props->active_width = 2;
1030 props->active_speed = 2;
1031 props->max_msg_sz = -1;
1036 static ssize_t show_rev(struct class_device *cdev, char *buf)
1038 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1040 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1041 return sprintf(buf, "%d\n", dev->rdev.t3cdev_p->type);
1044 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
1046 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1048 struct ethtool_drvinfo info;
1049 struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
1051 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1052 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1053 return sprintf(buf, "%s\n", info.fw_version);
1056 static ssize_t show_hca(struct class_device *cdev, char *buf)
1058 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1060 struct ethtool_drvinfo info;
1061 struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
1063 PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
1064 lldev->ethtool_ops->get_drvinfo(lldev, &info);
1065 return sprintf(buf, "%s\n", info.driver);
1068 static ssize_t show_board(struct class_device *cdev, char *buf)
1070 struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
1072 PDBG("%s class dev 0x%p\n", __FUNCTION__, dev);
1073 return sprintf(buf, "%x.%x\n", dev->rdev.rnic_info.pdev->vendor,
1074 dev->rdev.rnic_info.pdev->device);
1077 static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1078 static CLASS_DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1079 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1080 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1082 static struct class_device_attribute *iwch_class_attributes[] = {
1083 &class_device_attr_hw_rev,
1084 &class_device_attr_fw_ver,
1085 &class_device_attr_hca_type,
1086 &class_device_attr_board_id
1089 int iwch_register_device(struct iwch_dev *dev)
1094 PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
1095 strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1096 memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1097 memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1098 dev->ibdev.owner = THIS_MODULE;
1099 dev->device_cap_flags =
1100 (IB_DEVICE_ZERO_STAG |
1101 IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW);
1103 dev->ibdev.uverbs_cmd_mask =
1104 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1105 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1106 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1107 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1108 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1109 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1110 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1111 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1112 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1113 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1114 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1115 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1116 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1117 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1118 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1119 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1120 (1ull << IB_USER_VERBS_CMD_POST_RECV);
1121 dev->ibdev.node_type = RDMA_NODE_RNIC;
1122 memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1123 dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1124 dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1125 dev->ibdev.class_dev.dev = &(dev->rdev.rnic_info.pdev->dev);
1126 dev->ibdev.query_device = iwch_query_device;
1127 dev->ibdev.query_port = iwch_query_port;
1128 dev->ibdev.modify_port = iwch_modify_port;
1129 dev->ibdev.query_pkey = iwch_query_pkey;
1130 dev->ibdev.query_gid = iwch_query_gid;
1131 dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1132 dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1133 dev->ibdev.mmap = iwch_mmap;
1134 dev->ibdev.alloc_pd = iwch_allocate_pd;
1135 dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1136 dev->ibdev.create_ah = iwch_ah_create;
1137 dev->ibdev.destroy_ah = iwch_ah_destroy;
1138 dev->ibdev.create_qp = iwch_create_qp;
1139 dev->ibdev.modify_qp = iwch_ib_modify_qp;
1140 dev->ibdev.destroy_qp = iwch_destroy_qp;
1141 dev->ibdev.create_cq = iwch_create_cq;
1142 dev->ibdev.destroy_cq = iwch_destroy_cq;
1143 dev->ibdev.resize_cq = iwch_resize_cq;
1144 dev->ibdev.poll_cq = iwch_poll_cq;
1145 dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1146 dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1147 dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1148 dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1149 dev->ibdev.dereg_mr = iwch_dereg_mr;
1150 dev->ibdev.alloc_mw = iwch_alloc_mw;
1151 dev->ibdev.bind_mw = iwch_bind_mw;
1152 dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1154 dev->ibdev.attach_mcast = iwch_multicast_attach;
1155 dev->ibdev.detach_mcast = iwch_multicast_detach;
1156 dev->ibdev.process_mad = iwch_process_mad;
1158 dev->ibdev.req_notify_cq = iwch_arm_cq;
1159 dev->ibdev.post_send = iwch_post_send;
1160 dev->ibdev.post_recv = iwch_post_receive;
1164 (struct iw_cm_verbs *) kmalloc(sizeof(struct iw_cm_verbs),
1166 dev->ibdev.iwcm->connect = iwch_connect;
1167 dev->ibdev.iwcm->accept = iwch_accept_cr;
1168 dev->ibdev.iwcm->reject = iwch_reject_cr;
1169 dev->ibdev.iwcm->create_listen = iwch_create_listen;
1170 dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1171 dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1172 dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1173 dev->ibdev.iwcm->get_qp = iwch_get_qp;
1175 ret = ib_register_device(&dev->ibdev);
1179 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1180 ret = class_device_create_file(&dev->ibdev.class_dev,
1181 iwch_class_attributes[i]);
1188 ib_unregister_device(&dev->ibdev);
1193 void iwch_unregister_device(struct iwch_dev *dev)
1197 PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
1198 for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1199 class_device_remove_file(&dev->ibdev.class_dev,
1200 iwch_class_attributes[i]);
1201 ib_unregister_device(&dev->ibdev);