2 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2015 System Fabric Works, 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
34 #include <linux/dma-mapping.h>
35 #include <net/addrconf.h>
38 #include "rxe_queue.h"
39 #include "rxe_hw_counters.h"
41 static int rxe_query_device(struct ib_device *dev,
42 struct ib_device_attr *attr,
45 struct rxe_dev *rxe = to_rdev(dev);
47 if (uhw->inlen || uhw->outlen)
54 static int rxe_query_port(struct ib_device *dev,
55 u8 port_num, struct ib_port_attr *attr)
57 struct rxe_dev *rxe = to_rdev(dev);
58 struct rxe_port *port;
61 if (unlikely(port_num != 1)) {
62 pr_warn("invalid port_number %d\n", port_num);
68 /* *attr being zeroed by the caller, avoid zeroing it here */
71 mutex_lock(&rxe->usdev_lock);
72 rc = ib_get_eth_speed(dev, port_num, &attr->active_speed,
74 mutex_unlock(&rxe->usdev_lock);
80 static struct net_device *rxe_get_netdev(struct ib_device *device,
83 struct rxe_dev *rxe = to_rdev(device);
93 static int rxe_query_pkey(struct ib_device *device,
94 u8 port_num, u16 index, u16 *pkey)
96 struct rxe_dev *rxe = to_rdev(device);
97 struct rxe_port *port;
99 if (unlikely(port_num != 1)) {
100 dev_warn(device->dev.parent, "invalid port_num = %d\n",
107 if (unlikely(index >= port->attr.pkey_tbl_len)) {
108 dev_warn(device->dev.parent, "invalid index = %d\n",
113 *pkey = port->pkey_tbl[index];
120 static int rxe_modify_device(struct ib_device *dev,
121 int mask, struct ib_device_modify *attr)
123 struct rxe_dev *rxe = to_rdev(dev);
125 if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
126 rxe->attr.sys_image_guid = cpu_to_be64(attr->sys_image_guid);
128 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
129 memcpy(rxe->ib_dev.node_desc,
130 attr->node_desc, sizeof(rxe->ib_dev.node_desc));
136 static int rxe_modify_port(struct ib_device *dev,
137 u8 port_num, int mask, struct ib_port_modify *attr)
139 struct rxe_dev *rxe = to_rdev(dev);
140 struct rxe_port *port;
142 if (unlikely(port_num != 1)) {
143 pr_warn("invalid port_num = %d\n", port_num);
149 port->attr.port_cap_flags |= attr->set_port_cap_mask;
150 port->attr.port_cap_flags &= ~attr->clr_port_cap_mask;
152 if (mask & IB_PORT_RESET_QKEY_CNTR)
153 port->attr.qkey_viol_cntr = 0;
161 static enum rdma_link_layer rxe_get_link_layer(struct ib_device *dev,
164 struct rxe_dev *rxe = to_rdev(dev);
166 return rxe_link_layer(rxe, port_num);
169 static struct ib_ucontext *rxe_alloc_ucontext(struct ib_device *dev,
170 struct ib_udata *udata)
172 struct rxe_dev *rxe = to_rdev(dev);
173 struct rxe_ucontext *uc;
175 uc = rxe_alloc(&rxe->uc_pool);
176 return uc ? &uc->ibuc : ERR_PTR(-ENOMEM);
179 static int rxe_dealloc_ucontext(struct ib_ucontext *ibuc)
181 struct rxe_ucontext *uc = to_ruc(ibuc);
187 static int rxe_port_immutable(struct ib_device *dev, u8 port_num,
188 struct ib_port_immutable *immutable)
191 struct ib_port_attr attr;
193 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
195 err = ib_query_port(dev, port_num, &attr);
199 immutable->pkey_tbl_len = attr.pkey_tbl_len;
200 immutable->gid_tbl_len = attr.gid_tbl_len;
201 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
206 static struct ib_pd *rxe_alloc_pd(struct ib_device *dev,
207 struct ib_ucontext *context,
208 struct ib_udata *udata)
210 struct rxe_dev *rxe = to_rdev(dev);
213 pd = rxe_alloc(&rxe->pd_pool);
214 return pd ? &pd->ibpd : ERR_PTR(-ENOMEM);
217 static int rxe_dealloc_pd(struct ib_pd *ibpd)
219 struct rxe_pd *pd = to_rpd(ibpd);
225 static int rxe_init_av(struct rxe_dev *rxe, struct rdma_ah_attr *attr,
230 struct ib_gid_attr sgid_attr;
232 err = ib_get_cached_gid(&rxe->ib_dev, rdma_ah_get_port_num(attr),
233 rdma_ah_read_grh(attr)->sgid_index, &sgid,
236 pr_err("Failed to query sgid. err = %d\n", err);
240 rxe_av_from_attr(rdma_ah_get_port_num(attr), av, attr);
241 rxe_av_fill_ip_info(av, attr, &sgid_attr, &sgid);
242 dev_put(sgid_attr.ndev);
246 static struct ib_ah *rxe_create_ah(struct ib_pd *ibpd,
247 struct rdma_ah_attr *attr,
248 struct ib_udata *udata)
252 struct rxe_dev *rxe = to_rdev(ibpd->device);
253 struct rxe_pd *pd = to_rpd(ibpd);
256 err = rxe_av_chk_attr(rxe, attr);
260 ah = rxe_alloc(&rxe->ah_pool);
269 err = rxe_init_av(rxe, attr, &ah->av);
282 static int rxe_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *attr)
285 struct rxe_dev *rxe = to_rdev(ibah->device);
286 struct rxe_ah *ah = to_rah(ibah);
288 err = rxe_av_chk_attr(rxe, attr);
292 err = rxe_init_av(rxe, attr, &ah->av);
299 static int rxe_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *attr)
301 struct rxe_ah *ah = to_rah(ibah);
303 memset(attr, 0, sizeof(*attr));
304 attr->type = ibah->type;
305 rxe_av_to_attr(&ah->av, attr);
309 static int rxe_destroy_ah(struct ib_ah *ibah)
311 struct rxe_ah *ah = to_rah(ibah);
313 rxe_drop_ref(ah->pd);
318 static int post_one_recv(struct rxe_rq *rq, struct ib_recv_wr *ibwr)
323 struct rxe_recv_wqe *recv_wqe;
324 int num_sge = ibwr->num_sge;
326 if (unlikely(queue_full(rq->queue))) {
331 if (unlikely(num_sge > rq->max_sge)) {
337 for (i = 0; i < num_sge; i++)
338 length += ibwr->sg_list[i].length;
340 recv_wqe = producer_addr(rq->queue);
341 recv_wqe->wr_id = ibwr->wr_id;
342 recv_wqe->num_sge = num_sge;
344 memcpy(recv_wqe->dma.sge, ibwr->sg_list,
345 num_sge * sizeof(struct ib_sge));
347 recv_wqe->dma.length = length;
348 recv_wqe->dma.resid = length;
349 recv_wqe->dma.num_sge = num_sge;
350 recv_wqe->dma.cur_sge = 0;
351 recv_wqe->dma.sge_offset = 0;
353 /* make sure all changes to the work queue are written before we
354 * update the producer pointer
358 advance_producer(rq->queue);
365 static struct ib_srq *rxe_create_srq(struct ib_pd *ibpd,
366 struct ib_srq_init_attr *init,
367 struct ib_udata *udata)
370 struct rxe_dev *rxe = to_rdev(ibpd->device);
371 struct rxe_pd *pd = to_rpd(ibpd);
373 struct ib_ucontext *context = udata ? ibpd->uobject->context : NULL;
374 struct rxe_create_srq_resp __user *uresp = NULL;
377 if (udata->outlen < sizeof(*uresp))
378 return ERR_PTR(-EINVAL);
379 uresp = udata->outbuf;
382 err = rxe_srq_chk_attr(rxe, NULL, &init->attr, IB_SRQ_INIT_MASK);
386 srq = rxe_alloc(&rxe->srq_pool);
396 err = rxe_srq_from_init(rxe, srq, init, context, uresp);
410 static int rxe_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
411 enum ib_srq_attr_mask mask,
412 struct ib_udata *udata)
415 struct rxe_srq *srq = to_rsrq(ibsrq);
416 struct rxe_dev *rxe = to_rdev(ibsrq->device);
417 struct rxe_modify_srq_cmd ucmd = {};
420 if (udata->inlen < sizeof(ucmd))
423 err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
428 err = rxe_srq_chk_attr(rxe, srq, attr, mask);
432 err = rxe_srq_from_attr(rxe, srq, attr, mask, &ucmd);
442 static int rxe_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
444 struct rxe_srq *srq = to_rsrq(ibsrq);
449 attr->max_wr = srq->rq.queue->buf->index_mask;
450 attr->max_sge = srq->rq.max_sge;
451 attr->srq_limit = srq->limit;
455 static int rxe_destroy_srq(struct ib_srq *ibsrq)
457 struct rxe_srq *srq = to_rsrq(ibsrq);
460 rxe_queue_cleanup(srq->rq.queue);
462 rxe_drop_ref(srq->pd);
469 static int rxe_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
470 struct ib_recv_wr **bad_wr)
474 struct rxe_srq *srq = to_rsrq(ibsrq);
476 spin_lock_irqsave(&srq->rq.producer_lock, flags);
479 err = post_one_recv(&srq->rq, wr);
485 spin_unlock_irqrestore(&srq->rq.producer_lock, flags);
493 static struct ib_qp *rxe_create_qp(struct ib_pd *ibpd,
494 struct ib_qp_init_attr *init,
495 struct ib_udata *udata)
498 struct rxe_dev *rxe = to_rdev(ibpd->device);
499 struct rxe_pd *pd = to_rpd(ibpd);
501 struct rxe_create_qp_resp __user *uresp = NULL;
504 if (udata->outlen < sizeof(*uresp))
505 return ERR_PTR(-EINVAL);
506 uresp = udata->outbuf;
509 err = rxe_qp_chk_init(rxe, init);
513 qp = rxe_alloc(&rxe->qp_pool);
529 err = rxe_qp_from_init(rxe, qp, pd, init, uresp, ibpd);
543 static int rxe_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
544 int mask, struct ib_udata *udata)
547 struct rxe_dev *rxe = to_rdev(ibqp->device);
548 struct rxe_qp *qp = to_rqp(ibqp);
550 err = rxe_qp_chk_attr(rxe, qp, attr, mask);
554 err = rxe_qp_from_attr(qp, attr, mask, udata);
564 static int rxe_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
565 int mask, struct ib_qp_init_attr *init)
567 struct rxe_qp *qp = to_rqp(ibqp);
569 rxe_qp_to_init(qp, init);
570 rxe_qp_to_attr(qp, attr, mask);
575 static int rxe_destroy_qp(struct ib_qp *ibqp)
577 struct rxe_qp *qp = to_rqp(ibqp);
585 static int validate_send_wr(struct rxe_qp *qp, struct ib_send_wr *ibwr,
586 unsigned int mask, unsigned int length)
588 int num_sge = ibwr->num_sge;
589 struct rxe_sq *sq = &qp->sq;
591 if (unlikely(num_sge > sq->max_sge))
594 if (unlikely(mask & WR_ATOMIC_MASK)) {
598 if (atomic_wr(ibwr)->remote_addr & 0x7)
602 if (unlikely((ibwr->send_flags & IB_SEND_INLINE) &&
603 (length > sq->max_inline)))
612 static void init_send_wr(struct rxe_qp *qp, struct rxe_send_wr *wr,
613 struct ib_send_wr *ibwr)
615 wr->wr_id = ibwr->wr_id;
616 wr->num_sge = ibwr->num_sge;
617 wr->opcode = ibwr->opcode;
618 wr->send_flags = ibwr->send_flags;
620 if (qp_type(qp) == IB_QPT_UD ||
621 qp_type(qp) == IB_QPT_SMI ||
622 qp_type(qp) == IB_QPT_GSI) {
623 wr->wr.ud.remote_qpn = ud_wr(ibwr)->remote_qpn;
624 wr->wr.ud.remote_qkey = ud_wr(ibwr)->remote_qkey;
625 if (qp_type(qp) == IB_QPT_GSI)
626 wr->wr.ud.pkey_index = ud_wr(ibwr)->pkey_index;
627 if (wr->opcode == IB_WR_SEND_WITH_IMM)
628 wr->ex.imm_data = ibwr->ex.imm_data;
630 switch (wr->opcode) {
631 case IB_WR_RDMA_WRITE_WITH_IMM:
632 wr->ex.imm_data = ibwr->ex.imm_data;
634 case IB_WR_RDMA_READ:
635 case IB_WR_RDMA_WRITE:
636 wr->wr.rdma.remote_addr = rdma_wr(ibwr)->remote_addr;
637 wr->wr.rdma.rkey = rdma_wr(ibwr)->rkey;
639 case IB_WR_SEND_WITH_IMM:
640 wr->ex.imm_data = ibwr->ex.imm_data;
642 case IB_WR_SEND_WITH_INV:
643 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
645 case IB_WR_ATOMIC_CMP_AND_SWP:
646 case IB_WR_ATOMIC_FETCH_AND_ADD:
647 wr->wr.atomic.remote_addr =
648 atomic_wr(ibwr)->remote_addr;
649 wr->wr.atomic.compare_add =
650 atomic_wr(ibwr)->compare_add;
651 wr->wr.atomic.swap = atomic_wr(ibwr)->swap;
652 wr->wr.atomic.rkey = atomic_wr(ibwr)->rkey;
654 case IB_WR_LOCAL_INV:
655 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
658 wr->wr.reg.mr = reg_wr(ibwr)->mr;
659 wr->wr.reg.key = reg_wr(ibwr)->key;
660 wr->wr.reg.access = reg_wr(ibwr)->access;
668 static int init_send_wqe(struct rxe_qp *qp, struct ib_send_wr *ibwr,
669 unsigned int mask, unsigned int length,
670 struct rxe_send_wqe *wqe)
672 int num_sge = ibwr->num_sge;
677 init_send_wr(qp, &wqe->wr, ibwr);
679 if (qp_type(qp) == IB_QPT_UD ||
680 qp_type(qp) == IB_QPT_SMI ||
681 qp_type(qp) == IB_QPT_GSI)
682 memcpy(&wqe->av, &to_rah(ud_wr(ibwr)->ah)->av, sizeof(wqe->av));
684 if (unlikely(ibwr->send_flags & IB_SEND_INLINE)) {
685 p = wqe->dma.inline_data;
688 for (i = 0; i < num_sge; i++, sge++) {
689 memcpy(p, (void *)(uintptr_t)sge->addr,
694 } else if (mask & WR_REG_MASK) {
696 wqe->state = wqe_state_posted;
699 memcpy(wqe->dma.sge, ibwr->sg_list,
700 num_sge * sizeof(struct ib_sge));
702 wqe->iova = mask & WR_ATOMIC_MASK ? atomic_wr(ibwr)->remote_addr :
703 mask & WR_READ_OR_WRITE_MASK ? rdma_wr(ibwr)->remote_addr : 0;
705 wqe->dma.length = length;
706 wqe->dma.resid = length;
707 wqe->dma.num_sge = num_sge;
708 wqe->dma.cur_sge = 0;
709 wqe->dma.sge_offset = 0;
710 wqe->state = wqe_state_posted;
711 wqe->ssn = atomic_add_return(1, &qp->ssn);
716 static int post_one_send(struct rxe_qp *qp, struct ib_send_wr *ibwr,
717 unsigned int mask, u32 length)
720 struct rxe_sq *sq = &qp->sq;
721 struct rxe_send_wqe *send_wqe;
724 err = validate_send_wr(qp, ibwr, mask, length);
728 spin_lock_irqsave(&qp->sq.sq_lock, flags);
730 if (unlikely(queue_full(sq->queue))) {
735 send_wqe = producer_addr(sq->queue);
737 err = init_send_wqe(qp, ibwr, mask, length, send_wqe);
742 * make sure all changes to the work queue are
743 * written before we update the producer pointer
747 advance_producer(sq->queue);
748 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
753 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
757 static int rxe_post_send_kernel(struct rxe_qp *qp, struct ib_send_wr *wr,
758 struct ib_send_wr **bad_wr)
762 unsigned int length = 0;
767 mask = wr_opcode_mask(wr->opcode, qp);
768 if (unlikely(!mask)) {
774 if (unlikely((wr->send_flags & IB_SEND_INLINE) &&
775 !(mask & WR_INLINE_MASK))) {
782 for (i = 0; i < wr->num_sge; i++)
783 length += wr->sg_list[i].length;
785 err = post_one_send(qp, wr, mask, length);
795 * Must sched in case of GSI QP because ib_send_mad() hold irq lock,
796 * and the requester call ip_local_out_sk() that takes spin_lock_bh.
798 must_sched = (qp_type(qp) == IB_QPT_GSI) ||
799 (queue_count(qp->sq.queue) > 1);
801 rxe_run_task(&qp->req.task, must_sched);
802 if (unlikely(qp->req.state == QP_STATE_ERROR))
803 rxe_run_task(&qp->comp.task, 1);
808 static int rxe_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
809 struct ib_send_wr **bad_wr)
811 struct rxe_qp *qp = to_rqp(ibqp);
813 if (unlikely(!qp->valid)) {
818 if (unlikely(qp->req.state < QP_STATE_READY)) {
824 /* Utilize process context to do protocol processing */
825 rxe_run_task(&qp->req.task, 0);
828 return rxe_post_send_kernel(qp, wr, bad_wr);
831 static int rxe_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
832 struct ib_recv_wr **bad_wr)
835 struct rxe_qp *qp = to_rqp(ibqp);
836 struct rxe_rq *rq = &qp->rq;
839 if (unlikely((qp_state(qp) < IB_QPS_INIT) || !qp->valid)) {
845 if (unlikely(qp->srq)) {
851 spin_lock_irqsave(&rq->producer_lock, flags);
854 err = post_one_recv(rq, wr);
862 spin_unlock_irqrestore(&rq->producer_lock, flags);
864 if (qp->resp.state == QP_STATE_ERROR)
865 rxe_run_task(&qp->resp.task, 1);
871 static struct ib_cq *rxe_create_cq(struct ib_device *dev,
872 const struct ib_cq_init_attr *attr,
873 struct ib_ucontext *context,
874 struct ib_udata *udata)
877 struct rxe_dev *rxe = to_rdev(dev);
879 struct rxe_create_cq_resp __user *uresp = NULL;
882 if (udata->outlen < sizeof(*uresp))
883 return ERR_PTR(-EINVAL);
884 uresp = udata->outbuf;
888 return ERR_PTR(-EINVAL);
890 err = rxe_cq_chk_attr(rxe, NULL, attr->cqe, attr->comp_vector);
894 cq = rxe_alloc(&rxe->cq_pool);
900 err = rxe_cq_from_init(rxe, cq, attr->cqe, attr->comp_vector,
913 static int rxe_destroy_cq(struct ib_cq *ibcq)
915 struct rxe_cq *cq = to_rcq(ibcq);
923 static int rxe_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
926 struct rxe_cq *cq = to_rcq(ibcq);
927 struct rxe_dev *rxe = to_rdev(ibcq->device);
928 struct rxe_resize_cq_resp __user *uresp = NULL;
931 if (udata->outlen < sizeof(*uresp))
933 uresp = udata->outbuf;
936 err = rxe_cq_chk_attr(rxe, cq, cqe, 0);
940 err = rxe_cq_resize_queue(cq, cqe, uresp);
950 static int rxe_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
953 struct rxe_cq *cq = to_rcq(ibcq);
957 spin_lock_irqsave(&cq->cq_lock, flags);
958 for (i = 0; i < num_entries; i++) {
959 cqe = queue_head(cq->queue);
963 memcpy(wc++, &cqe->ibwc, sizeof(*wc));
964 advance_consumer(cq->queue);
966 spin_unlock_irqrestore(&cq->cq_lock, flags);
971 static int rxe_peek_cq(struct ib_cq *ibcq, int wc_cnt)
973 struct rxe_cq *cq = to_rcq(ibcq);
974 int count = queue_count(cq->queue);
976 return (count > wc_cnt) ? wc_cnt : count;
979 static int rxe_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
981 struct rxe_cq *cq = to_rcq(ibcq);
982 unsigned long irq_flags;
985 spin_lock_irqsave(&cq->cq_lock, irq_flags);
986 if (cq->notify != IB_CQ_NEXT_COMP)
987 cq->notify = flags & IB_CQ_SOLICITED_MASK;
989 if ((flags & IB_CQ_REPORT_MISSED_EVENTS) && !queue_empty(cq->queue))
992 spin_unlock_irqrestore(&cq->cq_lock, irq_flags);
997 static struct ib_mr *rxe_get_dma_mr(struct ib_pd *ibpd, int access)
999 struct rxe_dev *rxe = to_rdev(ibpd->device);
1000 struct rxe_pd *pd = to_rpd(ibpd);
1004 mr = rxe_alloc(&rxe->mr_pool);
1014 err = rxe_mem_init_dma(rxe, pd, access, mr);
1025 return ERR_PTR(err);
1028 static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
1032 int access, struct ib_udata *udata)
1035 struct rxe_dev *rxe = to_rdev(ibpd->device);
1036 struct rxe_pd *pd = to_rpd(ibpd);
1039 mr = rxe_alloc(&rxe->mr_pool);
1049 err = rxe_mem_init_user(rxe, pd, start, length, iova,
1061 return ERR_PTR(err);
1064 static int rxe_dereg_mr(struct ib_mr *ibmr)
1066 struct rxe_mem *mr = to_rmr(ibmr);
1068 mr->state = RXE_MEM_STATE_ZOMBIE;
1069 rxe_drop_ref(mr->pd);
1075 static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd,
1076 enum ib_mr_type mr_type,
1079 struct rxe_dev *rxe = to_rdev(ibpd->device);
1080 struct rxe_pd *pd = to_rpd(ibpd);
1084 if (mr_type != IB_MR_TYPE_MEM_REG)
1085 return ERR_PTR(-EINVAL);
1087 mr = rxe_alloc(&rxe->mr_pool);
1097 err = rxe_mem_init_fast(rxe, pd, max_num_sg, mr);
1108 return ERR_PTR(err);
1111 static int rxe_set_page(struct ib_mr *ibmr, u64 addr)
1113 struct rxe_mem *mr = to_rmr(ibmr);
1114 struct rxe_map *map;
1115 struct rxe_phys_buf *buf;
1117 if (unlikely(mr->nbuf == mr->num_buf))
1120 map = mr->map[mr->nbuf / RXE_BUF_PER_MAP];
1121 buf = &map->buf[mr->nbuf % RXE_BUF_PER_MAP];
1124 buf->size = ibmr->page_size;
1130 static int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
1131 int sg_nents, unsigned int *sg_offset)
1133 struct rxe_mem *mr = to_rmr(ibmr);
1138 n = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rxe_set_page);
1140 mr->va = ibmr->iova;
1141 mr->iova = ibmr->iova;
1142 mr->length = ibmr->length;
1143 mr->page_shift = ilog2(ibmr->page_size);
1144 mr->page_mask = ibmr->page_size - 1;
1145 mr->offset = mr->iova & mr->page_mask;
1150 static int rxe_attach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1153 struct rxe_dev *rxe = to_rdev(ibqp->device);
1154 struct rxe_qp *qp = to_rqp(ibqp);
1155 struct rxe_mc_grp *grp;
1157 /* takes a ref on grp if successful */
1158 err = rxe_mcast_get_grp(rxe, mgid, &grp);
1162 err = rxe_mcast_add_grp_elem(rxe, qp, grp);
1168 static int rxe_detach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1170 struct rxe_dev *rxe = to_rdev(ibqp->device);
1171 struct rxe_qp *qp = to_rqp(ibqp);
1173 return rxe_mcast_drop_grp_elem(rxe, qp, mgid);
1176 static ssize_t parent_show(struct device *device,
1177 struct device_attribute *attr, char *buf)
1179 struct rxe_dev *rxe = container_of(device, struct rxe_dev,
1182 return snprintf(buf, 16, "%s\n", rxe_parent_name(rxe, 1));
1185 static DEVICE_ATTR_RO(parent);
1187 static struct device_attribute *rxe_dev_attributes[] = {
1191 int rxe_register_device(struct rxe_dev *rxe)
1195 struct ib_device *dev = &rxe->ib_dev;
1196 struct crypto_shash *tfm;
1198 strlcpy(dev->name, "rxe%d", IB_DEVICE_NAME_MAX);
1199 strlcpy(dev->node_desc, "rxe", sizeof(dev->node_desc));
1201 dev->owner = THIS_MODULE;
1202 dev->node_type = RDMA_NODE_IB_CA;
1203 dev->phys_port_cnt = 1;
1204 dev->num_comp_vectors = num_possible_cpus();
1205 dev->dev.parent = rxe_dma_device(rxe);
1206 dev->local_dma_lkey = 0;
1207 addrconf_addr_eui48((unsigned char *)&dev->node_guid,
1208 rxe->ndev->dev_addr);
1209 dev->dev.dma_ops = &dma_virt_ops;
1210 dma_coerce_mask_and_coherent(&dev->dev,
1211 dma_get_required_mask(&dev->dev));
1213 dev->uverbs_abi_ver = RXE_UVERBS_ABI_VERSION;
1214 dev->uverbs_cmd_mask = BIT_ULL(IB_USER_VERBS_CMD_GET_CONTEXT)
1215 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)
1216 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_DEVICE)
1217 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_PORT)
1218 | BIT_ULL(IB_USER_VERBS_CMD_ALLOC_PD)
1219 | BIT_ULL(IB_USER_VERBS_CMD_DEALLOC_PD)
1220 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_SRQ)
1221 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_SRQ)
1222 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_SRQ)
1223 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_SRQ)
1224 | BIT_ULL(IB_USER_VERBS_CMD_POST_SRQ_RECV)
1225 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_QP)
1226 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_QP)
1227 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_QP)
1228 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_QP)
1229 | BIT_ULL(IB_USER_VERBS_CMD_POST_SEND)
1230 | BIT_ULL(IB_USER_VERBS_CMD_POST_RECV)
1231 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_CQ)
1232 | BIT_ULL(IB_USER_VERBS_CMD_RESIZE_CQ)
1233 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_CQ)
1234 | BIT_ULL(IB_USER_VERBS_CMD_POLL_CQ)
1235 | BIT_ULL(IB_USER_VERBS_CMD_PEEK_CQ)
1236 | BIT_ULL(IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)
1237 | BIT_ULL(IB_USER_VERBS_CMD_REG_MR)
1238 | BIT_ULL(IB_USER_VERBS_CMD_DEREG_MR)
1239 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_AH)
1240 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_AH)
1241 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_AH)
1242 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_AH)
1243 | BIT_ULL(IB_USER_VERBS_CMD_ATTACH_MCAST)
1244 | BIT_ULL(IB_USER_VERBS_CMD_DETACH_MCAST)
1247 dev->query_device = rxe_query_device;
1248 dev->modify_device = rxe_modify_device;
1249 dev->query_port = rxe_query_port;
1250 dev->modify_port = rxe_modify_port;
1251 dev->get_link_layer = rxe_get_link_layer;
1252 dev->get_netdev = rxe_get_netdev;
1253 dev->query_pkey = rxe_query_pkey;
1254 dev->alloc_ucontext = rxe_alloc_ucontext;
1255 dev->dealloc_ucontext = rxe_dealloc_ucontext;
1256 dev->mmap = rxe_mmap;
1257 dev->get_port_immutable = rxe_port_immutable;
1258 dev->alloc_pd = rxe_alloc_pd;
1259 dev->dealloc_pd = rxe_dealloc_pd;
1260 dev->create_ah = rxe_create_ah;
1261 dev->modify_ah = rxe_modify_ah;
1262 dev->query_ah = rxe_query_ah;
1263 dev->destroy_ah = rxe_destroy_ah;
1264 dev->create_srq = rxe_create_srq;
1265 dev->modify_srq = rxe_modify_srq;
1266 dev->query_srq = rxe_query_srq;
1267 dev->destroy_srq = rxe_destroy_srq;
1268 dev->post_srq_recv = rxe_post_srq_recv;
1269 dev->create_qp = rxe_create_qp;
1270 dev->modify_qp = rxe_modify_qp;
1271 dev->query_qp = rxe_query_qp;
1272 dev->destroy_qp = rxe_destroy_qp;
1273 dev->post_send = rxe_post_send;
1274 dev->post_recv = rxe_post_recv;
1275 dev->create_cq = rxe_create_cq;
1276 dev->destroy_cq = rxe_destroy_cq;
1277 dev->resize_cq = rxe_resize_cq;
1278 dev->poll_cq = rxe_poll_cq;
1279 dev->peek_cq = rxe_peek_cq;
1280 dev->req_notify_cq = rxe_req_notify_cq;
1281 dev->get_dma_mr = rxe_get_dma_mr;
1282 dev->reg_user_mr = rxe_reg_user_mr;
1283 dev->dereg_mr = rxe_dereg_mr;
1284 dev->alloc_mr = rxe_alloc_mr;
1285 dev->map_mr_sg = rxe_map_mr_sg;
1286 dev->attach_mcast = rxe_attach_mcast;
1287 dev->detach_mcast = rxe_detach_mcast;
1288 dev->get_hw_stats = rxe_ib_get_hw_stats;
1289 dev->alloc_hw_stats = rxe_ib_alloc_hw_stats;
1291 tfm = crypto_alloc_shash("crc32", 0, 0);
1293 pr_err("failed to allocate crc algorithm err:%ld\n",
1295 return PTR_ERR(tfm);
1299 dev->driver_id = RDMA_DRIVER_RXE;
1300 err = ib_register_device(dev, NULL);
1302 pr_warn("%s failed with error %d\n", __func__, err);
1306 for (i = 0; i < ARRAY_SIZE(rxe_dev_attributes); ++i) {
1307 err = device_create_file(&dev->dev, rxe_dev_attributes[i]);
1309 pr_warn("%s failed with error %d for attr number %d\n",
1318 ib_unregister_device(dev);
1320 crypto_free_shash(rxe->tfm);
1325 int rxe_unregister_device(struct rxe_dev *rxe)
1328 struct ib_device *dev = &rxe->ib_dev;
1330 for (i = 0; i < ARRAY_SIZE(rxe_dev_attributes); ++i)
1331 device_remove_file(&dev->dev, rxe_dev_attributes[i]);
1333 ib_unregister_device(dev);