1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
3 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
7 #include <linux/dma-mapping.h>
8 #include <net/addrconf.h>
9 #include <rdma/uverbs_ioctl.h>
12 #include "rxe_queue.h"
13 #include "rxe_hw_counters.h"
15 static int rxe_query_device(struct ib_device *dev,
16 struct ib_device_attr *attr,
19 struct rxe_dev *rxe = to_rdev(dev);
21 if (uhw->inlen || uhw->outlen)
28 static int rxe_query_port(struct ib_device *dev,
29 u8 port_num, struct ib_port_attr *attr)
31 struct rxe_dev *rxe = to_rdev(dev);
32 struct rxe_port *port;
37 /* *attr being zeroed by the caller, avoid zeroing it here */
40 mutex_lock(&rxe->usdev_lock);
41 rc = ib_get_eth_speed(dev, port_num, &attr->active_speed,
44 if (attr->state == IB_PORT_ACTIVE)
45 attr->phys_state = IB_PORT_PHYS_STATE_LINK_UP;
46 else if (dev_get_flags(rxe->ndev) & IFF_UP)
47 attr->phys_state = IB_PORT_PHYS_STATE_POLLING;
49 attr->phys_state = IB_PORT_PHYS_STATE_DISABLED;
51 mutex_unlock(&rxe->usdev_lock);
56 static int rxe_query_pkey(struct ib_device *device,
57 u8 port_num, u16 index, u16 *pkey)
62 *pkey = IB_DEFAULT_PKEY_FULL;
66 static int rxe_modify_device(struct ib_device *dev,
67 int mask, struct ib_device_modify *attr)
69 struct rxe_dev *rxe = to_rdev(dev);
71 if (mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
72 IB_DEVICE_MODIFY_NODE_DESC))
75 if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
76 rxe->attr.sys_image_guid = cpu_to_be64(attr->sys_image_guid);
78 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
79 memcpy(rxe->ib_dev.node_desc,
80 attr->node_desc, sizeof(rxe->ib_dev.node_desc));
86 static int rxe_modify_port(struct ib_device *dev,
87 u8 port_num, int mask, struct ib_port_modify *attr)
89 struct rxe_dev *rxe = to_rdev(dev);
90 struct rxe_port *port;
94 port->attr.port_cap_flags |= attr->set_port_cap_mask;
95 port->attr.port_cap_flags &= ~attr->clr_port_cap_mask;
97 if (mask & IB_PORT_RESET_QKEY_CNTR)
98 port->attr.qkey_viol_cntr = 0;
103 static enum rdma_link_layer rxe_get_link_layer(struct ib_device *dev,
106 return IB_LINK_LAYER_ETHERNET;
109 static int rxe_alloc_ucontext(struct ib_ucontext *uctx, struct ib_udata *udata)
111 struct rxe_dev *rxe = to_rdev(uctx->device);
112 struct rxe_ucontext *uc = to_ruc(uctx);
114 return rxe_add_to_pool(&rxe->uc_pool, &uc->pelem);
117 static void rxe_dealloc_ucontext(struct ib_ucontext *ibuc)
119 struct rxe_ucontext *uc = to_ruc(ibuc);
124 static int rxe_port_immutable(struct ib_device *dev, u8 port_num,
125 struct ib_port_immutable *immutable)
128 struct ib_port_attr attr;
130 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
132 err = ib_query_port(dev, port_num, &attr);
136 immutable->pkey_tbl_len = attr.pkey_tbl_len;
137 immutable->gid_tbl_len = attr.gid_tbl_len;
138 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
143 static int rxe_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
145 struct rxe_dev *rxe = to_rdev(ibpd->device);
146 struct rxe_pd *pd = to_rpd(ibpd);
148 return rxe_add_to_pool(&rxe->pd_pool, &pd->pelem);
151 static int rxe_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
153 struct rxe_pd *pd = to_rpd(ibpd);
159 static int rxe_create_ah(struct ib_ah *ibah,
160 struct rdma_ah_init_attr *init_attr,
161 struct ib_udata *udata)
165 struct rxe_dev *rxe = to_rdev(ibah->device);
166 struct rxe_ah *ah = to_rah(ibah);
168 err = rxe_av_chk_attr(rxe, init_attr->ah_attr);
172 err = rxe_add_to_pool(&rxe->ah_pool, &ah->pelem);
176 rxe_init_av(init_attr->ah_attr, &ah->av);
180 static int rxe_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *attr)
183 struct rxe_dev *rxe = to_rdev(ibah->device);
184 struct rxe_ah *ah = to_rah(ibah);
186 err = rxe_av_chk_attr(rxe, attr);
190 rxe_init_av(attr, &ah->av);
194 static int rxe_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *attr)
196 struct rxe_ah *ah = to_rah(ibah);
198 memset(attr, 0, sizeof(*attr));
199 attr->type = ibah->type;
200 rxe_av_to_attr(&ah->av, attr);
204 static int rxe_destroy_ah(struct ib_ah *ibah, u32 flags)
206 struct rxe_ah *ah = to_rah(ibah);
212 static int post_one_recv(struct rxe_rq *rq, const struct ib_recv_wr *ibwr)
217 struct rxe_recv_wqe *recv_wqe;
218 int num_sge = ibwr->num_sge;
220 if (unlikely(queue_full(rq->queue))) {
225 if (unlikely(num_sge > rq->max_sge)) {
231 for (i = 0; i < num_sge; i++)
232 length += ibwr->sg_list[i].length;
234 recv_wqe = producer_addr(rq->queue);
235 recv_wqe->wr_id = ibwr->wr_id;
236 recv_wqe->num_sge = num_sge;
238 memcpy(recv_wqe->dma.sge, ibwr->sg_list,
239 num_sge * sizeof(struct ib_sge));
241 recv_wqe->dma.length = length;
242 recv_wqe->dma.resid = length;
243 recv_wqe->dma.num_sge = num_sge;
244 recv_wqe->dma.cur_sge = 0;
245 recv_wqe->dma.sge_offset = 0;
247 /* make sure all changes to the work queue are written before we
248 * update the producer pointer
252 advance_producer(rq->queue);
259 static int rxe_create_srq(struct ib_srq *ibsrq, struct ib_srq_init_attr *init,
260 struct ib_udata *udata)
263 struct rxe_dev *rxe = to_rdev(ibsrq->device);
264 struct rxe_pd *pd = to_rpd(ibsrq->pd);
265 struct rxe_srq *srq = to_rsrq(ibsrq);
266 struct rxe_create_srq_resp __user *uresp = NULL;
268 if (init->srq_type != IB_SRQT_BASIC)
272 if (udata->outlen < sizeof(*uresp))
274 uresp = udata->outbuf;
277 err = rxe_srq_chk_attr(rxe, NULL, &init->attr, IB_SRQ_INIT_MASK);
281 err = rxe_add_to_pool(&rxe->srq_pool, &srq->pelem);
288 err = rxe_srq_from_init(rxe, srq, init, udata, uresp);
301 static int rxe_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
302 enum ib_srq_attr_mask mask,
303 struct ib_udata *udata)
306 struct rxe_srq *srq = to_rsrq(ibsrq);
307 struct rxe_dev *rxe = to_rdev(ibsrq->device);
308 struct rxe_modify_srq_cmd ucmd = {};
311 if (udata->inlen < sizeof(ucmd))
314 err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
319 err = rxe_srq_chk_attr(rxe, srq, attr, mask);
323 err = rxe_srq_from_attr(rxe, srq, attr, mask, &ucmd, udata);
333 static int rxe_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
335 struct rxe_srq *srq = to_rsrq(ibsrq);
340 attr->max_wr = srq->rq.queue->buf->index_mask;
341 attr->max_sge = srq->rq.max_sge;
342 attr->srq_limit = srq->limit;
346 static int rxe_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata)
348 struct rxe_srq *srq = to_rsrq(ibsrq);
351 rxe_queue_cleanup(srq->rq.queue);
353 rxe_drop_ref(srq->pd);
358 static int rxe_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
359 const struct ib_recv_wr **bad_wr)
363 struct rxe_srq *srq = to_rsrq(ibsrq);
365 spin_lock_irqsave(&srq->rq.producer_lock, flags);
368 err = post_one_recv(&srq->rq, wr);
374 spin_unlock_irqrestore(&srq->rq.producer_lock, flags);
382 static struct ib_qp *rxe_create_qp(struct ib_pd *ibpd,
383 struct ib_qp_init_attr *init,
384 struct ib_udata *udata)
387 struct rxe_dev *rxe = to_rdev(ibpd->device);
388 struct rxe_pd *pd = to_rpd(ibpd);
390 struct rxe_create_qp_resp __user *uresp = NULL;
393 if (udata->outlen < sizeof(*uresp))
394 return ERR_PTR(-EINVAL);
395 uresp = udata->outbuf;
398 err = rxe_qp_chk_init(rxe, init);
402 qp = rxe_alloc(&rxe->qp_pool);
418 err = rxe_qp_from_init(rxe, qp, pd, init, uresp, ibpd, udata);
432 static int rxe_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
433 int mask, struct ib_udata *udata)
436 struct rxe_dev *rxe = to_rdev(ibqp->device);
437 struct rxe_qp *qp = to_rqp(ibqp);
439 if (mask & ~IB_QP_ATTR_STANDARD_BITS)
442 err = rxe_qp_chk_attr(rxe, qp, attr, mask);
446 err = rxe_qp_from_attr(qp, attr, mask, udata);
456 static int rxe_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
457 int mask, struct ib_qp_init_attr *init)
459 struct rxe_qp *qp = to_rqp(ibqp);
461 rxe_qp_to_init(qp, init);
462 rxe_qp_to_attr(qp, attr, mask);
467 static int rxe_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
469 struct rxe_qp *qp = to_rqp(ibqp);
477 static int validate_send_wr(struct rxe_qp *qp, const struct ib_send_wr *ibwr,
478 unsigned int mask, unsigned int length)
480 int num_sge = ibwr->num_sge;
481 struct rxe_sq *sq = &qp->sq;
483 if (unlikely(num_sge > sq->max_sge))
486 if (unlikely(mask & WR_ATOMIC_MASK)) {
490 if (atomic_wr(ibwr)->remote_addr & 0x7)
494 if (unlikely((ibwr->send_flags & IB_SEND_INLINE) &&
495 (length > sq->max_inline)))
504 static void init_send_wr(struct rxe_qp *qp, struct rxe_send_wr *wr,
505 const struct ib_send_wr *ibwr)
507 wr->wr_id = ibwr->wr_id;
508 wr->num_sge = ibwr->num_sge;
509 wr->opcode = ibwr->opcode;
510 wr->send_flags = ibwr->send_flags;
512 if (qp_type(qp) == IB_QPT_UD ||
513 qp_type(qp) == IB_QPT_SMI ||
514 qp_type(qp) == IB_QPT_GSI) {
515 wr->wr.ud.remote_qpn = ud_wr(ibwr)->remote_qpn;
516 wr->wr.ud.remote_qkey = ud_wr(ibwr)->remote_qkey;
517 if (qp_type(qp) == IB_QPT_GSI)
518 wr->wr.ud.pkey_index = ud_wr(ibwr)->pkey_index;
519 if (wr->opcode == IB_WR_SEND_WITH_IMM)
520 wr->ex.imm_data = ibwr->ex.imm_data;
522 switch (wr->opcode) {
523 case IB_WR_RDMA_WRITE_WITH_IMM:
524 wr->ex.imm_data = ibwr->ex.imm_data;
526 case IB_WR_RDMA_READ:
527 case IB_WR_RDMA_WRITE:
528 wr->wr.rdma.remote_addr = rdma_wr(ibwr)->remote_addr;
529 wr->wr.rdma.rkey = rdma_wr(ibwr)->rkey;
531 case IB_WR_SEND_WITH_IMM:
532 wr->ex.imm_data = ibwr->ex.imm_data;
534 case IB_WR_SEND_WITH_INV:
535 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
537 case IB_WR_ATOMIC_CMP_AND_SWP:
538 case IB_WR_ATOMIC_FETCH_AND_ADD:
539 wr->wr.atomic.remote_addr =
540 atomic_wr(ibwr)->remote_addr;
541 wr->wr.atomic.compare_add =
542 atomic_wr(ibwr)->compare_add;
543 wr->wr.atomic.swap = atomic_wr(ibwr)->swap;
544 wr->wr.atomic.rkey = atomic_wr(ibwr)->rkey;
546 case IB_WR_LOCAL_INV:
547 wr->ex.invalidate_rkey = ibwr->ex.invalidate_rkey;
550 wr->wr.reg.mr = reg_wr(ibwr)->mr;
551 wr->wr.reg.key = reg_wr(ibwr)->key;
552 wr->wr.reg.access = reg_wr(ibwr)->access;
560 static int init_send_wqe(struct rxe_qp *qp, const struct ib_send_wr *ibwr,
561 unsigned int mask, unsigned int length,
562 struct rxe_send_wqe *wqe)
564 int num_sge = ibwr->num_sge;
569 init_send_wr(qp, &wqe->wr, ibwr);
571 if (qp_type(qp) == IB_QPT_UD ||
572 qp_type(qp) == IB_QPT_SMI ||
573 qp_type(qp) == IB_QPT_GSI)
574 memcpy(&wqe->av, &to_rah(ud_wr(ibwr)->ah)->av, sizeof(wqe->av));
576 if (unlikely(ibwr->send_flags & IB_SEND_INLINE)) {
577 p = wqe->dma.inline_data;
580 for (i = 0; i < num_sge; i++, sge++) {
581 memcpy(p, (void *)(uintptr_t)sge->addr,
586 } else if (mask & WR_REG_MASK) {
588 wqe->state = wqe_state_posted;
591 memcpy(wqe->dma.sge, ibwr->sg_list,
592 num_sge * sizeof(struct ib_sge));
594 wqe->iova = mask & WR_ATOMIC_MASK ? atomic_wr(ibwr)->remote_addr :
595 mask & WR_READ_OR_WRITE_MASK ? rdma_wr(ibwr)->remote_addr : 0;
597 wqe->dma.length = length;
598 wqe->dma.resid = length;
599 wqe->dma.num_sge = num_sge;
600 wqe->dma.cur_sge = 0;
601 wqe->dma.sge_offset = 0;
602 wqe->state = wqe_state_posted;
603 wqe->ssn = atomic_add_return(1, &qp->ssn);
608 static int post_one_send(struct rxe_qp *qp, const struct ib_send_wr *ibwr,
609 unsigned int mask, u32 length)
612 struct rxe_sq *sq = &qp->sq;
613 struct rxe_send_wqe *send_wqe;
616 err = validate_send_wr(qp, ibwr, mask, length);
620 spin_lock_irqsave(&qp->sq.sq_lock, flags);
622 if (unlikely(queue_full(sq->queue))) {
627 send_wqe = producer_addr(sq->queue);
629 err = init_send_wqe(qp, ibwr, mask, length, send_wqe);
634 * make sure all changes to the work queue are
635 * written before we update the producer pointer
639 advance_producer(sq->queue);
640 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
645 spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
649 static int rxe_post_send_kernel(struct rxe_qp *qp, const struct ib_send_wr *wr,
650 const struct ib_send_wr **bad_wr)
654 unsigned int length = 0;
656 struct ib_send_wr *next;
659 mask = wr_opcode_mask(wr->opcode, qp);
660 if (unlikely(!mask)) {
666 if (unlikely((wr->send_flags & IB_SEND_INLINE) &&
667 !(mask & WR_INLINE_MASK))) {
676 for (i = 0; i < wr->num_sge; i++)
677 length += wr->sg_list[i].length;
679 err = post_one_send(qp, wr, mask, length);
688 rxe_run_task(&qp->req.task, 1);
689 if (unlikely(qp->req.state == QP_STATE_ERROR))
690 rxe_run_task(&qp->comp.task, 1);
695 static int rxe_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
696 const struct ib_send_wr **bad_wr)
698 struct rxe_qp *qp = to_rqp(ibqp);
700 if (unlikely(!qp->valid)) {
705 if (unlikely(qp->req.state < QP_STATE_READY)) {
711 /* Utilize process context to do protocol processing */
712 rxe_run_task(&qp->req.task, 0);
715 return rxe_post_send_kernel(qp, wr, bad_wr);
718 static int rxe_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
719 const struct ib_recv_wr **bad_wr)
722 struct rxe_qp *qp = to_rqp(ibqp);
723 struct rxe_rq *rq = &qp->rq;
726 if (unlikely((qp_state(qp) < IB_QPS_INIT) || !qp->valid)) {
732 if (unlikely(qp->srq)) {
738 spin_lock_irqsave(&rq->producer_lock, flags);
741 err = post_one_recv(rq, wr);
749 spin_unlock_irqrestore(&rq->producer_lock, flags);
751 if (qp->resp.state == QP_STATE_ERROR)
752 rxe_run_task(&qp->resp.task, 1);
758 static int rxe_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
759 struct ib_udata *udata)
762 struct ib_device *dev = ibcq->device;
763 struct rxe_dev *rxe = to_rdev(dev);
764 struct rxe_cq *cq = to_rcq(ibcq);
765 struct rxe_create_cq_resp __user *uresp = NULL;
768 if (udata->outlen < sizeof(*uresp))
770 uresp = udata->outbuf;
776 err = rxe_cq_chk_attr(rxe, NULL, attr->cqe, attr->comp_vector);
780 err = rxe_cq_from_init(rxe, cq, attr->cqe, attr->comp_vector, udata,
785 return rxe_add_to_pool(&rxe->cq_pool, &cq->pelem);
788 static int rxe_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
790 struct rxe_cq *cq = to_rcq(ibcq);
798 static int rxe_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
801 struct rxe_cq *cq = to_rcq(ibcq);
802 struct rxe_dev *rxe = to_rdev(ibcq->device);
803 struct rxe_resize_cq_resp __user *uresp = NULL;
806 if (udata->outlen < sizeof(*uresp))
808 uresp = udata->outbuf;
811 err = rxe_cq_chk_attr(rxe, cq, cqe, 0);
815 err = rxe_cq_resize_queue(cq, cqe, uresp, udata);
825 static int rxe_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
828 struct rxe_cq *cq = to_rcq(ibcq);
832 spin_lock_irqsave(&cq->cq_lock, flags);
833 for (i = 0; i < num_entries; i++) {
834 cqe = queue_head(cq->queue);
838 memcpy(wc++, &cqe->ibwc, sizeof(*wc));
839 advance_consumer(cq->queue);
841 spin_unlock_irqrestore(&cq->cq_lock, flags);
846 static int rxe_peek_cq(struct ib_cq *ibcq, int wc_cnt)
848 struct rxe_cq *cq = to_rcq(ibcq);
849 int count = queue_count(cq->queue);
851 return (count > wc_cnt) ? wc_cnt : count;
854 static int rxe_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
856 struct rxe_cq *cq = to_rcq(ibcq);
857 unsigned long irq_flags;
860 spin_lock_irqsave(&cq->cq_lock, irq_flags);
861 if (cq->notify != IB_CQ_NEXT_COMP)
862 cq->notify = flags & IB_CQ_SOLICITED_MASK;
864 if ((flags & IB_CQ_REPORT_MISSED_EVENTS) && !queue_empty(cq->queue))
867 spin_unlock_irqrestore(&cq->cq_lock, irq_flags);
872 static struct ib_mr *rxe_get_dma_mr(struct ib_pd *ibpd, int access)
874 struct rxe_dev *rxe = to_rdev(ibpd->device);
875 struct rxe_pd *pd = to_rpd(ibpd);
878 mr = rxe_alloc(&rxe->mr_pool);
880 return ERR_PTR(-ENOMEM);
884 rxe_mem_init_dma(pd, access, mr);
889 static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
893 int access, struct ib_udata *udata)
896 struct rxe_dev *rxe = to_rdev(ibpd->device);
897 struct rxe_pd *pd = to_rpd(ibpd);
900 mr = rxe_alloc(&rxe->mr_pool);
910 err = rxe_mem_init_user(pd, start, length, iova,
925 static int rxe_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata)
927 struct rxe_mem *mr = to_rmr(ibmr);
929 mr->state = RXE_MEM_STATE_ZOMBIE;
930 rxe_drop_ref(mr_pd(mr));
936 static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd, enum ib_mr_type mr_type,
939 struct rxe_dev *rxe = to_rdev(ibpd->device);
940 struct rxe_pd *pd = to_rpd(ibpd);
944 if (mr_type != IB_MR_TYPE_MEM_REG)
945 return ERR_PTR(-EINVAL);
947 mr = rxe_alloc(&rxe->mr_pool);
957 err = rxe_mem_init_fast(pd, max_num_sg, mr);
971 static int rxe_set_page(struct ib_mr *ibmr, u64 addr)
973 struct rxe_mem *mr = to_rmr(ibmr);
975 struct rxe_phys_buf *buf;
977 if (unlikely(mr->nbuf == mr->num_buf))
980 map = mr->map[mr->nbuf / RXE_BUF_PER_MAP];
981 buf = &map->buf[mr->nbuf % RXE_BUF_PER_MAP];
984 buf->size = ibmr->page_size;
990 static int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
991 int sg_nents, unsigned int *sg_offset)
993 struct rxe_mem *mr = to_rmr(ibmr);
998 n = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rxe_set_page);
1000 mr->va = ibmr->iova;
1001 mr->iova = ibmr->iova;
1002 mr->length = ibmr->length;
1003 mr->page_shift = ilog2(ibmr->page_size);
1004 mr->page_mask = ibmr->page_size - 1;
1005 mr->offset = mr->iova & mr->page_mask;
1010 static int rxe_attach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1013 struct rxe_dev *rxe = to_rdev(ibqp->device);
1014 struct rxe_qp *qp = to_rqp(ibqp);
1015 struct rxe_mc_grp *grp;
1017 /* takes a ref on grp if successful */
1018 err = rxe_mcast_get_grp(rxe, mgid, &grp);
1022 err = rxe_mcast_add_grp_elem(rxe, qp, grp);
1028 static int rxe_detach_mcast(struct ib_qp *ibqp, union ib_gid *mgid, u16 mlid)
1030 struct rxe_dev *rxe = to_rdev(ibqp->device);
1031 struct rxe_qp *qp = to_rqp(ibqp);
1033 return rxe_mcast_drop_grp_elem(rxe, qp, mgid);
1036 static ssize_t parent_show(struct device *device,
1037 struct device_attribute *attr, char *buf)
1039 struct rxe_dev *rxe =
1040 rdma_device_to_drv_device(device, struct rxe_dev, ib_dev);
1042 return scnprintf(buf, PAGE_SIZE, "%s\n", rxe_parent_name(rxe, 1));
1045 static DEVICE_ATTR_RO(parent);
1047 static struct attribute *rxe_dev_attributes[] = {
1048 &dev_attr_parent.attr,
1052 static const struct attribute_group rxe_attr_group = {
1053 .attrs = rxe_dev_attributes,
1056 static int rxe_enable_driver(struct ib_device *ib_dev)
1058 struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev);
1060 rxe_set_port_state(rxe);
1061 dev_info(&rxe->ib_dev.dev, "added %s\n", netdev_name(rxe->ndev));
1065 static const struct ib_device_ops rxe_dev_ops = {
1066 .owner = THIS_MODULE,
1067 .driver_id = RDMA_DRIVER_RXE,
1068 .uverbs_abi_ver = RXE_UVERBS_ABI_VERSION,
1070 .alloc_hw_stats = rxe_ib_alloc_hw_stats,
1071 .alloc_mr = rxe_alloc_mr,
1072 .alloc_pd = rxe_alloc_pd,
1073 .alloc_ucontext = rxe_alloc_ucontext,
1074 .attach_mcast = rxe_attach_mcast,
1075 .create_ah = rxe_create_ah,
1076 .create_cq = rxe_create_cq,
1077 .create_qp = rxe_create_qp,
1078 .create_srq = rxe_create_srq,
1079 .dealloc_driver = rxe_dealloc,
1080 .dealloc_pd = rxe_dealloc_pd,
1081 .dealloc_ucontext = rxe_dealloc_ucontext,
1082 .dereg_mr = rxe_dereg_mr,
1083 .destroy_ah = rxe_destroy_ah,
1084 .destroy_cq = rxe_destroy_cq,
1085 .destroy_qp = rxe_destroy_qp,
1086 .destroy_srq = rxe_destroy_srq,
1087 .detach_mcast = rxe_detach_mcast,
1088 .enable_driver = rxe_enable_driver,
1089 .get_dma_mr = rxe_get_dma_mr,
1090 .get_hw_stats = rxe_ib_get_hw_stats,
1091 .get_link_layer = rxe_get_link_layer,
1092 .get_port_immutable = rxe_port_immutable,
1093 .map_mr_sg = rxe_map_mr_sg,
1095 .modify_ah = rxe_modify_ah,
1096 .modify_device = rxe_modify_device,
1097 .modify_port = rxe_modify_port,
1098 .modify_qp = rxe_modify_qp,
1099 .modify_srq = rxe_modify_srq,
1100 .peek_cq = rxe_peek_cq,
1101 .poll_cq = rxe_poll_cq,
1102 .post_recv = rxe_post_recv,
1103 .post_send = rxe_post_send,
1104 .post_srq_recv = rxe_post_srq_recv,
1105 .query_ah = rxe_query_ah,
1106 .query_device = rxe_query_device,
1107 .query_pkey = rxe_query_pkey,
1108 .query_port = rxe_query_port,
1109 .query_qp = rxe_query_qp,
1110 .query_srq = rxe_query_srq,
1111 .reg_user_mr = rxe_reg_user_mr,
1112 .req_notify_cq = rxe_req_notify_cq,
1113 .resize_cq = rxe_resize_cq,
1115 INIT_RDMA_OBJ_SIZE(ib_ah, rxe_ah, ibah),
1116 INIT_RDMA_OBJ_SIZE(ib_cq, rxe_cq, ibcq),
1117 INIT_RDMA_OBJ_SIZE(ib_pd, rxe_pd, ibpd),
1118 INIT_RDMA_OBJ_SIZE(ib_srq, rxe_srq, ibsrq),
1119 INIT_RDMA_OBJ_SIZE(ib_ucontext, rxe_ucontext, ibuc),
1122 int rxe_register_device(struct rxe_dev *rxe, const char *ibdev_name)
1125 struct ib_device *dev = &rxe->ib_dev;
1126 struct crypto_shash *tfm;
1128 strlcpy(dev->node_desc, "rxe", sizeof(dev->node_desc));
1130 dev->node_type = RDMA_NODE_IB_CA;
1131 dev->phys_port_cnt = 1;
1132 dev->num_comp_vectors = num_possible_cpus();
1133 dev->dev.parent = rxe_dma_device(rxe);
1134 dev->local_dma_lkey = 0;
1135 addrconf_addr_eui48((unsigned char *)&dev->node_guid,
1136 rxe->ndev->dev_addr);
1137 dev->dev.dma_parms = &rxe->dma_parms;
1138 dma_set_max_seg_size(&dev->dev, UINT_MAX);
1139 dma_set_coherent_mask(&dev->dev, dma_get_required_mask(&dev->dev));
1141 dev->uverbs_cmd_mask |=
1142 BIT_ULL(IB_USER_VERBS_CMD_POST_SRQ_RECV)
1143 | BIT_ULL(IB_USER_VERBS_CMD_POST_SEND)
1144 | BIT_ULL(IB_USER_VERBS_CMD_POST_RECV)
1145 | BIT_ULL(IB_USER_VERBS_CMD_POLL_CQ)
1146 | BIT_ULL(IB_USER_VERBS_CMD_PEEK_CQ)
1147 | BIT_ULL(IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)
1148 | BIT_ULL(IB_USER_VERBS_CMD_CREATE_AH)
1149 | BIT_ULL(IB_USER_VERBS_CMD_MODIFY_AH)
1150 | BIT_ULL(IB_USER_VERBS_CMD_QUERY_AH)
1151 | BIT_ULL(IB_USER_VERBS_CMD_DESTROY_AH)
1154 ib_set_device_ops(dev, &rxe_dev_ops);
1155 err = ib_device_set_netdev(&rxe->ib_dev, rxe->ndev, 1);
1159 tfm = crypto_alloc_shash("crc32", 0, 0);
1161 pr_err("failed to allocate crc algorithm err:%ld\n",
1163 return PTR_ERR(tfm);
1167 rdma_set_device_sysfs_group(dev, &rxe_attr_group);
1168 err = ib_register_device(dev, ibdev_name, NULL);
1170 pr_warn("%s failed with error %d\n", __func__, err);
1173 * Note that rxe may be invalid at this point if another thread