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>
37 #include "rxe_queue.h"
38 #include "rxe_hw_counters.h"
40 static int rxe_query_device(struct ib_device *dev,
41 struct ib_device_attr *attr,
44 struct rxe_dev *rxe = to_rdev(dev);
46 if (uhw->inlen || uhw->outlen)
53 static void rxe_eth_speed_to_ib_speed(int speed, u8 *active_speed,
57 *active_width = IB_WIDTH_1X;
58 *active_speed = IB_SPEED_SDR;
59 } else if (speed <= 10000) {
60 *active_width = IB_WIDTH_1X;
61 *active_speed = IB_SPEED_FDR10;
62 } else if (speed <= 20000) {
63 *active_width = IB_WIDTH_4X;
64 *active_speed = IB_SPEED_DDR;
65 } else if (speed <= 30000) {
66 *active_width = IB_WIDTH_4X;
67 *active_speed = IB_SPEED_QDR;
68 } else if (speed <= 40000) {
69 *active_width = IB_WIDTH_4X;
70 *active_speed = IB_SPEED_FDR10;
72 *active_width = IB_WIDTH_4X;
73 *active_speed = IB_SPEED_EDR;
77 static int rxe_query_port(struct ib_device *dev,
78 u8 port_num, struct ib_port_attr *attr)
80 struct rxe_dev *rxe = to_rdev(dev);
81 struct rxe_port *port;
84 if (unlikely(port_num != 1)) {
85 pr_warn("invalid port_number %d\n", port_num);
91 /* *attr being zeroed by the caller, avoid zeroing it here */
94 mutex_lock(&rxe->usdev_lock);
95 if (rxe->ndev->ethtool_ops->get_link_ksettings) {
96 struct ethtool_link_ksettings ks;
98 rxe->ndev->ethtool_ops->get_link_ksettings(rxe->ndev, &ks);
99 speed = ks.base.speed;
100 } else if (rxe->ndev->ethtool_ops->get_settings) {
101 struct ethtool_cmd cmd;
103 rxe->ndev->ethtool_ops->get_settings(rxe->ndev, &cmd);
106 pr_warn("%s speed is unknown, defaulting to 1000\n",
110 rxe_eth_speed_to_ib_speed(speed, &attr->active_speed,
111 &attr->active_width);
112 mutex_unlock(&rxe->usdev_lock);
120 static int rxe_query_gid(struct ib_device *device,
121 u8 port_num, int index, union ib_gid *gid)
125 if (index > RXE_PORT_GID_TBL_LEN)
128 ret = ib_get_cached_gid(device, port_num, index, gid, NULL);
129 if (ret == -EAGAIN) {
130 memcpy(gid, &zgid, sizeof(*gid));
137 static int rxe_add_gid(struct ib_device *device, u8 port_num, unsigned int
138 index, const union ib_gid *gid,
139 const struct ib_gid_attr *attr, void **context)
141 if (index >= RXE_PORT_GID_TBL_LEN)
146 static int rxe_del_gid(struct ib_device *device, u8 port_num, unsigned int
147 index, void **context)
149 if (index >= RXE_PORT_GID_TBL_LEN)
154 static struct net_device *rxe_get_netdev(struct ib_device *device,
157 struct rxe_dev *rxe = to_rdev(device);
167 static int rxe_query_pkey(struct ib_device *device,
168 u8 port_num, u16 index, u16 *pkey)
170 struct rxe_dev *rxe = to_rdev(device);
171 struct rxe_port *port;
173 if (unlikely(port_num != 1)) {
174 dev_warn(device->dev.parent, "invalid port_num = %d\n",
181 if (unlikely(index >= port->attr.pkey_tbl_len)) {
182 dev_warn(device->dev.parent, "invalid index = %d\n",
187 *pkey = port->pkey_tbl[index];
194 static int rxe_modify_device(struct ib_device *dev,
195 int mask, struct ib_device_modify *attr)
197 struct rxe_dev *rxe = to_rdev(dev);
199 if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
200 rxe->attr.sys_image_guid = cpu_to_be64(attr->sys_image_guid);
202 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
203 memcpy(rxe->ib_dev.node_desc,
204 attr->node_desc, sizeof(rxe->ib_dev.node_desc));
210 static int rxe_modify_port(struct ib_device *dev,
211 u8 port_num, int mask, struct ib_port_modify *attr)
213 struct rxe_dev *rxe = to_rdev(dev);
214 struct rxe_port *port;
216 if (unlikely(port_num != 1)) {
217 pr_warn("invalid port_num = %d\n", port_num);
223 port->attr.port_cap_flags |= attr->set_port_cap_mask;
224 port->attr.port_cap_flags &= ~attr->clr_port_cap_mask;
226 if (mask & IB_PORT_RESET_QKEY_CNTR)
227 port->attr.qkey_viol_cntr = 0;
235 static enum rdma_link_layer rxe_get_link_layer(struct ib_device *dev,
238 struct rxe_dev *rxe = to_rdev(dev);
240 return rxe_link_layer(rxe, port_num);
243 static struct ib_ucontext *rxe_alloc_ucontext(struct ib_device *dev,
244 struct ib_udata *udata)
246 struct rxe_dev *rxe = to_rdev(dev);
247 struct rxe_ucontext *uc;
249 uc = rxe_alloc(&rxe->uc_pool);
250 return uc ? &uc->ibuc : ERR_PTR(-ENOMEM);
253 static int rxe_dealloc_ucontext(struct ib_ucontext *ibuc)
255 struct rxe_ucontext *uc = to_ruc(ibuc);
261 static int rxe_port_immutable(struct ib_device *dev, u8 port_num,
262 struct ib_port_immutable *immutable)
265 struct ib_port_attr attr;
267 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
269 err = ib_query_port(dev, port_num, &attr);
273 immutable->pkey_tbl_len = attr.pkey_tbl_len;
274 immutable->gid_tbl_len = attr.gid_tbl_len;
275 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
280 static struct ib_pd *rxe_alloc_pd(struct ib_device *dev,
281 struct ib_ucontext *context,
282 struct ib_udata *udata)
284 struct rxe_dev *rxe = to_rdev(dev);
287 pd = rxe_alloc(&rxe->pd_pool);
288 return pd ? &pd->ibpd : ERR_PTR(-ENOMEM);
291 static int rxe_dealloc_pd(struct ib_pd *ibpd)
293 struct rxe_pd *pd = to_rpd(ibpd);
299 static int rxe_init_av(struct rxe_dev *rxe, struct ib_ah_attr *attr,
304 struct ib_gid_attr sgid_attr;
306 err = ib_get_cached_gid(&rxe->ib_dev, attr->port_num,
307 attr->grh.sgid_index, &sgid,
310 pr_err("Failed to query sgid. err = %d\n", err);
314 err = rxe_av_from_attr(rxe, attr->port_num, av, attr);
316 err = rxe_av_fill_ip_info(rxe, av, attr, &sgid_attr, &sgid);
319 dev_put(sgid_attr.ndev);
323 static struct ib_ah *rxe_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr,
324 struct ib_udata *udata)
328 struct rxe_dev *rxe = to_rdev(ibpd->device);
329 struct rxe_pd *pd = to_rpd(ibpd);
332 err = rxe_av_chk_attr(rxe, attr);
336 ah = rxe_alloc(&rxe->ah_pool);
345 err = rxe_init_av(rxe, attr, &ah->av);
358 static int rxe_modify_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
361 struct rxe_dev *rxe = to_rdev(ibah->device);
362 struct rxe_ah *ah = to_rah(ibah);
364 err = rxe_av_chk_attr(rxe, attr);
368 err = rxe_init_av(rxe, attr, &ah->av);
375 static int rxe_query_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
377 struct rxe_dev *rxe = to_rdev(ibah->device);
378 struct rxe_ah *ah = to_rah(ibah);
380 rxe_av_to_attr(rxe, &ah->av, attr);
384 static int rxe_destroy_ah(struct ib_ah *ibah)
386 struct rxe_ah *ah = to_rah(ibah);
388 rxe_drop_ref(ah->pd);
393 static int post_one_recv(struct rxe_rq *rq, struct ib_recv_wr *ibwr)
398 struct rxe_recv_wqe *recv_wqe;
399 int num_sge = ibwr->num_sge;
401 if (unlikely(queue_full(rq->queue))) {
406 if (unlikely(num_sge > rq->max_sge)) {
412 for (i = 0; i < num_sge; i++)
413 length += ibwr->sg_list[i].length;
415 recv_wqe = producer_addr(rq->queue);
416 recv_wqe->wr_id = ibwr->wr_id;
417 recv_wqe->num_sge = num_sge;
419 memcpy(recv_wqe->dma.sge, ibwr->sg_list,
420 num_sge * sizeof(struct ib_sge));
422 recv_wqe->dma.length = length;
423 recv_wqe->dma.resid = length;
424 recv_wqe->dma.num_sge = num_sge;
425 recv_wqe->dma.cur_sge = 0;
426 recv_wqe->dma.sge_offset = 0;
428 /* make sure all changes to the work queue are written before we
429 * update the producer pointer
433 advance_producer(rq->queue);
440 static struct ib_srq *rxe_create_srq(struct ib_pd *ibpd,
441 struct ib_srq_init_attr *init,
442 struct ib_udata *udata)
445 struct rxe_dev *rxe = to_rdev(ibpd->device);
446 struct rxe_pd *pd = to_rpd(ibpd);
448 struct ib_ucontext *context = udata ? ibpd->uobject->context : NULL;
450 err = rxe_srq_chk_attr(rxe, NULL, &init->attr, IB_SRQ_INIT_MASK);
454 srq = rxe_alloc(&rxe->srq_pool);
464 err = rxe_srq_from_init(rxe, srq, init, context, udata);
478 static int rxe_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
479 enum ib_srq_attr_mask mask,
480 struct ib_udata *udata)
483 struct rxe_srq *srq = to_rsrq(ibsrq);
484 struct rxe_dev *rxe = to_rdev(ibsrq->device);
486 err = rxe_srq_chk_attr(rxe, srq, attr, mask);
490 err = rxe_srq_from_attr(rxe, srq, attr, mask, udata);
500 static int rxe_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
502 struct rxe_srq *srq = to_rsrq(ibsrq);
507 attr->max_wr = srq->rq.queue->buf->index_mask;
508 attr->max_sge = srq->rq.max_sge;
509 attr->srq_limit = srq->limit;
513 static int rxe_destroy_srq(struct ib_srq *ibsrq)
515 struct rxe_srq *srq = to_rsrq(ibsrq);
518 rxe_queue_cleanup(srq->rq.queue);
520 rxe_drop_ref(srq->pd);
527 static int rxe_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
528 struct ib_recv_wr **bad_wr)
532 struct rxe_srq *srq = to_rsrq(ibsrq);
534 spin_lock_irqsave(&srq->rq.producer_lock, flags);
537 err = post_one_recv(&srq->rq, wr);
543 spin_unlock_irqrestore(&srq->rq.producer_lock, flags);
551 static struct ib_qp *rxe_create_qp(struct ib_pd *ibpd,
552 struct ib_qp_init_attr *init,
553 struct ib_udata *udata)
556 struct rxe_dev *rxe = to_rdev(ibpd->device);
557 struct rxe_pd *pd = to_rpd(ibpd);
560 err = rxe_qp_chk_init(rxe, init);
564 qp = rxe_alloc(&rxe->qp_pool);
580 err = rxe_qp_from_init(rxe, qp, pd, init, udata, ibpd);
594 static int rxe_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
595 int mask, struct ib_udata *udata)
598 struct rxe_dev *rxe = to_rdev(ibqp->device);
599 struct rxe_qp *qp = to_rqp(ibqp);
601 err = rxe_qp_chk_attr(rxe, qp, attr, mask);
605 err = rxe_qp_from_attr(qp, attr, mask, udata);
615 static int rxe_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
616 int mask, struct ib_qp_init_attr *init)
618 struct rxe_qp *qp = to_rqp(ibqp);
620 rxe_qp_to_init(qp, init);
621 rxe_qp_to_attr(qp, attr, mask);
626 static int rxe_destroy_qp(struct ib_qp *ibqp)
628 struct rxe_qp *qp = to_rqp(ibqp);
636 static int validate_send_wr(struct rxe_qp *qp, struct ib_send_wr *ibwr,
637 unsigned int mask, unsigned int length)
639 int num_sge = ibwr->num_sge;
640 struct rxe_sq *sq = &qp->sq;
642 if (unlikely(num_sge > sq->max_sge))
645 if (unlikely(mask & WR_ATOMIC_MASK)) {
649 if (atomic_wr(ibwr)->remote_addr & 0x7)
653 if (unlikely((ibwr->send_flags & IB_SEND_INLINE) &&
654 (length > sq->max_inline)))
663 static void init_send_wr(struct rxe_qp *qp, struct rxe_send_wr *wr,
664 struct ib_send_wr *ibwr)
666 wr->wr_id = ibwr->wr_id;
667 wr->num_sge = ibwr->num_sge;
668 wr->opcode = ibwr->opcode;
669 wr->send_flags = ibwr->send_flags;
671 if (qp_type(qp) == IB_QPT_UD ||
672 qp_type(qp) == IB_QPT_SMI ||
673 qp_type(qp) == IB_QPT_GSI) {
674 wr->wr.ud.remote_qpn = ud_wr(ibwr)->remote_qpn;
675 wr->wr.ud.remote_qkey = ud_wr(ibwr)->remote_qkey;
676 if (qp_type(qp) == IB_QPT_GSI)
677 wr->wr.ud.pkey_index = ud_wr(ibwr)->pkey_index;
678 if (wr->opcode == IB_WR_SEND_WITH_IMM)
679 wr->ex.imm_data = ibwr->ex.imm_data;
681 switch (wr->opcode) {
682 case IB_WR_RDMA_WRITE_WITH_IMM:
683 wr->ex.imm_data = ibwr->ex.imm_data;
684 case IB_WR_RDMA_READ:
685 case IB_WR_RDMA_WRITE:
686 wr->wr.rdma.remote_addr = rdma_wr(ibwr)->remote_addr;
687 wr->wr.rdma.rkey = rdma_wr(ibwr)->rkey;
689 case IB_WR_SEND_WITH_IMM:
690 wr->ex.imm_data = ibwr->ex.imm_data;
692 case IB_WR_SEND_WITH_INV:
693 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
695 case IB_WR_ATOMIC_CMP_AND_SWP:
696 case IB_WR_ATOMIC_FETCH_AND_ADD:
697 wr->wr.atomic.remote_addr =
698 atomic_wr(ibwr)->remote_addr;
699 wr->wr.atomic.compare_add =
700 atomic_wr(ibwr)->compare_add;
701 wr->wr.atomic.swap = atomic_wr(ibwr)->swap;
702 wr->wr.atomic.rkey = atomic_wr(ibwr)->rkey;
704 case IB_WR_LOCAL_INV:
705 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
708 wr->wr.reg.mr = reg_wr(ibwr)->mr;
709 wr->wr.reg.key = reg_wr(ibwr)->key;
710 wr->wr.reg.access = reg_wr(ibwr)->access;
718 static int init_send_wqe(struct rxe_qp *qp, struct ib_send_wr *ibwr,
719 unsigned int mask, unsigned int length,
720 struct rxe_send_wqe *wqe)
722 int num_sge = ibwr->num_sge;
727 init_send_wr(qp, &wqe->wr, ibwr);
729 if (qp_type(qp) == IB_QPT_UD ||
730 qp_type(qp) == IB_QPT_SMI ||
731 qp_type(qp) == IB_QPT_GSI)
732 memcpy(&wqe->av, &to_rah(ud_wr(ibwr)->ah)->av, sizeof(wqe->av));
734 if (unlikely(ibwr->send_flags & IB_SEND_INLINE)) {
735 p = wqe->dma.inline_data;
738 for (i = 0; i < num_sge; i++, sge++) {
739 if (qp->is_user && copy_from_user(p, (__user void *)
740 (uintptr_t)sge->addr, sge->length))
743 else if (!qp->is_user)
744 memcpy(p, (void *)(uintptr_t)sge->addr,
749 } else if (mask & WR_REG_MASK) {
751 wqe->state = wqe_state_posted;
754 memcpy(wqe->dma.sge, ibwr->sg_list,
755 num_sge * sizeof(struct ib_sge));
757 wqe->iova = (mask & WR_ATOMIC_MASK) ?
758 atomic_wr(ibwr)->remote_addr :
759 rdma_wr(ibwr)->remote_addr;
761 wqe->dma.length = length;
762 wqe->dma.resid = length;
763 wqe->dma.num_sge = num_sge;
764 wqe->dma.cur_sge = 0;
765 wqe->dma.sge_offset = 0;
766 wqe->state = wqe_state_posted;
767 wqe->ssn = atomic_add_return(1, &qp->ssn);
772 static int post_one_send(struct rxe_qp *qp, struct ib_send_wr *ibwr,
773 unsigned int mask, u32 length)
776 struct rxe_sq *sq = &qp->sq;
777 struct rxe_send_wqe *send_wqe;
780 err = validate_send_wr(qp, ibwr, mask, length);
784 spin_lock_irqsave(&qp->sq.sq_lock, flags);
786 if (unlikely(queue_full(sq->queue))) {
791 send_wqe = producer_addr(sq->queue);
793 err = init_send_wqe(qp, ibwr, mask, length, send_wqe);
798 * make sure all changes to the work queue are
799 * written before we update the producer pointer
803 advance_producer(sq->queue);
804 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
809 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
813 static int rxe_post_send_kernel(struct rxe_qp *qp, struct ib_send_wr *wr,
814 struct ib_send_wr **bad_wr)
818 unsigned int length = 0;
823 mask = wr_opcode_mask(wr->opcode, qp);
824 if (unlikely(!mask)) {
830 if (unlikely((wr->send_flags & IB_SEND_INLINE) &&
831 !(mask & WR_INLINE_MASK))) {
838 for (i = 0; i < wr->num_sge; i++)
839 length += wr->sg_list[i].length;
841 err = post_one_send(qp, wr, mask, length);
851 * Must sched in case of GSI QP because ib_send_mad() hold irq lock,
852 * and the requester call ip_local_out_sk() that takes spin_lock_bh.
854 must_sched = (qp_type(qp) == IB_QPT_GSI) ||
855 (queue_count(qp->sq.queue) > 1);
857 rxe_run_task(&qp->req.task, must_sched);
862 static int rxe_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
863 struct ib_send_wr **bad_wr)
865 struct rxe_qp *qp = to_rqp(ibqp);
867 if (unlikely(!qp->valid)) {
872 if (unlikely(qp->req.state < QP_STATE_READY)) {
878 /* Utilize process context to do protocol processing */
879 rxe_run_task(&qp->req.task, 0);
882 return rxe_post_send_kernel(qp, wr, bad_wr);
885 static int rxe_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
886 struct ib_recv_wr **bad_wr)
889 struct rxe_qp *qp = to_rqp(ibqp);
890 struct rxe_rq *rq = &qp->rq;
893 if (unlikely((qp_state(qp) < IB_QPS_INIT) || !qp->valid)) {
899 if (unlikely(qp->srq)) {
905 spin_lock_irqsave(&rq->producer_lock, flags);
908 err = post_one_recv(rq, wr);
916 spin_unlock_irqrestore(&rq->producer_lock, flags);
922 static struct ib_cq *rxe_create_cq(struct ib_device *dev,
923 const struct ib_cq_init_attr *attr,
924 struct ib_ucontext *context,
925 struct ib_udata *udata)
928 struct rxe_dev *rxe = to_rdev(dev);
932 return ERR_PTR(-EINVAL);
934 err = rxe_cq_chk_attr(rxe, NULL, attr->cqe, attr->comp_vector, udata);
938 cq = rxe_alloc(&rxe->cq_pool);
944 err = rxe_cq_from_init(rxe, cq, attr->cqe, attr->comp_vector,
957 static int rxe_destroy_cq(struct ib_cq *ibcq)
959 struct rxe_cq *cq = to_rcq(ibcq);
965 static int rxe_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
968 struct rxe_cq *cq = to_rcq(ibcq);
969 struct rxe_dev *rxe = to_rdev(ibcq->device);
971 err = rxe_cq_chk_attr(rxe, cq, cqe, 0, udata);
975 err = rxe_cq_resize_queue(cq, cqe, udata);
985 static int rxe_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
988 struct rxe_cq *cq = to_rcq(ibcq);
992 spin_lock_irqsave(&cq->cq_lock, flags);
993 for (i = 0; i < num_entries; i++) {
994 cqe = queue_head(cq->queue);
998 memcpy(wc++, &cqe->ibwc, sizeof(*wc));
999 advance_consumer(cq->queue);
1001 spin_unlock_irqrestore(&cq->cq_lock, flags);
1006 static int rxe_peek_cq(struct ib_cq *ibcq, int wc_cnt)
1008 struct rxe_cq *cq = to_rcq(ibcq);
1009 int count = queue_count(cq->queue);
1011 return (count > wc_cnt) ? wc_cnt : count;
1014 static int rxe_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
1016 struct rxe_cq *cq = to_rcq(ibcq);
1017 unsigned long irq_flags;
1020 spin_lock_irqsave(&cq->cq_lock, irq_flags);
1021 if (cq->notify != IB_CQ_NEXT_COMP)
1022 cq->notify = flags & IB_CQ_SOLICITED_MASK;
1024 if ((flags & IB_CQ_REPORT_MISSED_EVENTS) && !queue_empty(cq->queue))
1027 spin_unlock_irqrestore(&cq->cq_lock, irq_flags);
1032 static struct ib_mr *rxe_get_dma_mr(struct ib_pd *ibpd, int access)
1034 struct rxe_dev *rxe = to_rdev(ibpd->device);
1035 struct rxe_pd *pd = to_rpd(ibpd);
1039 mr = rxe_alloc(&rxe->mr_pool);
1049 err = rxe_mem_init_dma(rxe, pd, access, mr);
1060 return ERR_PTR(err);
1063 static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
1067 int access, struct ib_udata *udata)
1070 struct rxe_dev *rxe = to_rdev(ibpd->device);
1071 struct rxe_pd *pd = to_rpd(ibpd);
1074 mr = rxe_alloc(&rxe->mr_pool);
1084 err = rxe_mem_init_user(rxe, pd, start, length, iova,
1096 return ERR_PTR(err);
1099 static int rxe_dereg_mr(struct ib_mr *ibmr)
1101 struct rxe_mem *mr = to_rmr(ibmr);
1103 mr->state = RXE_MEM_STATE_ZOMBIE;
1104 rxe_drop_ref(mr->pd);
1110 static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd,
1111 enum ib_mr_type mr_type,
1114 struct rxe_dev *rxe = to_rdev(ibpd->device);
1115 struct rxe_pd *pd = to_rpd(ibpd);
1119 if (mr_type != IB_MR_TYPE_MEM_REG)
1120 return ERR_PTR(-EINVAL);
1122 mr = rxe_alloc(&rxe->mr_pool);
1132 err = rxe_mem_init_fast(rxe, pd, max_num_sg, mr);
1143 return ERR_PTR(err);
1146 static int rxe_set_page(struct ib_mr *ibmr, u64 addr)
1148 struct rxe_mem *mr = to_rmr(ibmr);
1149 struct rxe_map *map;
1150 struct rxe_phys_buf *buf;
1152 if (unlikely(mr->nbuf == mr->num_buf))
1155 map = mr->map[mr->nbuf / RXE_BUF_PER_MAP];
1156 buf = &map->buf[mr->nbuf % RXE_BUF_PER_MAP];
1159 buf->size = ibmr->page_size;
1165 static int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
1166 int sg_nents, unsigned int *sg_offset)
1168 struct rxe_mem *mr = to_rmr(ibmr);
1173 n = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rxe_set_page);
1175 mr->va = ibmr->iova;
1176 mr->iova = ibmr->iova;
1177 mr->length = ibmr->length;
1178 mr->page_shift = ilog2(ibmr->page_size);
1179 mr->page_mask = ibmr->page_size - 1;
1180 mr->offset = mr->iova & mr->page_mask;
1185 static int rxe_attach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1188 struct rxe_dev *rxe = to_rdev(ibqp->device);
1189 struct rxe_qp *qp = to_rqp(ibqp);
1190 struct rxe_mc_grp *grp;
1192 /* takes a ref on grp if successful */
1193 err = rxe_mcast_get_grp(rxe, mgid, &grp);
1197 err = rxe_mcast_add_grp_elem(rxe, qp, grp);
1203 static int rxe_detach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1205 struct rxe_dev *rxe = to_rdev(ibqp->device);
1206 struct rxe_qp *qp = to_rqp(ibqp);
1208 return rxe_mcast_drop_grp_elem(rxe, qp, mgid);
1211 static ssize_t rxe_show_parent(struct device *device,
1212 struct device_attribute *attr, char *buf)
1214 struct rxe_dev *rxe = container_of(device, struct rxe_dev,
1217 return snprintf(buf, 16, "%s\n", rxe_parent_name(rxe, 1));
1220 static DEVICE_ATTR(parent, S_IRUGO, rxe_show_parent, NULL);
1222 static struct device_attribute *rxe_dev_attributes[] = {
1226 int rxe_register_device(struct rxe_dev *rxe)
1230 struct ib_device *dev = &rxe->ib_dev;
1232 strlcpy(dev->name, "rxe%d", IB_DEVICE_NAME_MAX);
1233 strlcpy(dev->node_desc, "rxe", sizeof(dev->node_desc));
1235 dev->owner = THIS_MODULE;
1236 dev->node_type = RDMA_NODE_IB_CA;
1237 dev->phys_port_cnt = 1;
1238 dev->num_comp_vectors = RXE_NUM_COMP_VECTORS;
1239 dev->dev.parent = rxe_dma_device(rxe);
1240 dev->local_dma_lkey = 0;
1241 dev->node_guid = rxe_node_guid(rxe);
1242 dev->dev.dma_ops = &dma_virt_ops;
1244 dev->uverbs_abi_ver = RXE_UVERBS_ABI_VERSION;
1245 dev->uverbs_cmd_mask = BIT_ULL(IB_USER_VERBS_CMD_GET_CONTEXT)
1246 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)
1247 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_DEVICE)
1248 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_PORT)
1249 | BIT_ULL(IB_USER_VERBS_CMD_ALLOC_PD)
1250 | BIT_ULL(IB_USER_VERBS_CMD_DEALLOC_PD)
1251 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_SRQ)
1252 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_SRQ)
1253 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_SRQ)
1254 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_SRQ)
1255 | BIT_ULL(IB_USER_VERBS_CMD_POST_SRQ_RECV)
1256 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_QP)
1257 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_QP)
1258 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_QP)
1259 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_QP)
1260 | BIT_ULL(IB_USER_VERBS_CMD_POST_SEND)
1261 | BIT_ULL(IB_USER_VERBS_CMD_POST_RECV)
1262 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_CQ)
1263 | BIT_ULL(IB_USER_VERBS_CMD_RESIZE_CQ)
1264 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_CQ)
1265 | BIT_ULL(IB_USER_VERBS_CMD_POLL_CQ)
1266 | BIT_ULL(IB_USER_VERBS_CMD_PEEK_CQ)
1267 | BIT_ULL(IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)
1268 | BIT_ULL(IB_USER_VERBS_CMD_REG_MR)
1269 | BIT_ULL(IB_USER_VERBS_CMD_DEREG_MR)
1270 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_AH)
1271 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_AH)
1272 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_AH)
1273 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_AH)
1274 | BIT_ULL(IB_USER_VERBS_CMD_ATTACH_MCAST)
1275 | BIT_ULL(IB_USER_VERBS_CMD_DETACH_MCAST)
1278 dev->query_device = rxe_query_device;
1279 dev->modify_device = rxe_modify_device;
1280 dev->query_port = rxe_query_port;
1281 dev->modify_port = rxe_modify_port;
1282 dev->get_link_layer = rxe_get_link_layer;
1283 dev->query_gid = rxe_query_gid;
1284 dev->get_netdev = rxe_get_netdev;
1285 dev->add_gid = rxe_add_gid;
1286 dev->del_gid = rxe_del_gid;
1287 dev->query_pkey = rxe_query_pkey;
1288 dev->alloc_ucontext = rxe_alloc_ucontext;
1289 dev->dealloc_ucontext = rxe_dealloc_ucontext;
1290 dev->mmap = rxe_mmap;
1291 dev->get_port_immutable = rxe_port_immutable;
1292 dev->alloc_pd = rxe_alloc_pd;
1293 dev->dealloc_pd = rxe_dealloc_pd;
1294 dev->create_ah = rxe_create_ah;
1295 dev->modify_ah = rxe_modify_ah;
1296 dev->query_ah = rxe_query_ah;
1297 dev->destroy_ah = rxe_destroy_ah;
1298 dev->create_srq = rxe_create_srq;
1299 dev->modify_srq = rxe_modify_srq;
1300 dev->query_srq = rxe_query_srq;
1301 dev->destroy_srq = rxe_destroy_srq;
1302 dev->post_srq_recv = rxe_post_srq_recv;
1303 dev->create_qp = rxe_create_qp;
1304 dev->modify_qp = rxe_modify_qp;
1305 dev->query_qp = rxe_query_qp;
1306 dev->destroy_qp = rxe_destroy_qp;
1307 dev->post_send = rxe_post_send;
1308 dev->post_recv = rxe_post_recv;
1309 dev->create_cq = rxe_create_cq;
1310 dev->destroy_cq = rxe_destroy_cq;
1311 dev->resize_cq = rxe_resize_cq;
1312 dev->poll_cq = rxe_poll_cq;
1313 dev->peek_cq = rxe_peek_cq;
1314 dev->req_notify_cq = rxe_req_notify_cq;
1315 dev->get_dma_mr = rxe_get_dma_mr;
1316 dev->reg_user_mr = rxe_reg_user_mr;
1317 dev->dereg_mr = rxe_dereg_mr;
1318 dev->alloc_mr = rxe_alloc_mr;
1319 dev->map_mr_sg = rxe_map_mr_sg;
1320 dev->attach_mcast = rxe_attach_mcast;
1321 dev->detach_mcast = rxe_detach_mcast;
1322 dev->get_hw_stats = rxe_ib_get_hw_stats;
1323 dev->alloc_hw_stats = rxe_ib_alloc_hw_stats;
1325 err = ib_register_device(dev, NULL);
1327 pr_warn("rxe_register_device failed, err = %d\n", err);
1331 for (i = 0; i < ARRAY_SIZE(rxe_dev_attributes); ++i) {
1332 err = device_create_file(&dev->dev, rxe_dev_attributes[i]);
1334 pr_warn("device_create_file failed, i = %d, err = %d\n",
1343 ib_unregister_device(dev);
1348 int rxe_unregister_device(struct rxe_dev *rxe)
1351 struct ib_device *dev = &rxe->ib_dev;
1353 for (i = 0; i < ARRAY_SIZE(rxe_dev_attributes); ++i)
1354 device_remove_file(&dev->dev, rxe_dev_attributes[i]);
1356 ib_unregister_device(dev);