diff options
Diffstat (limited to 'drivers/infiniband/hw/qedr/verbs.c')
| -rw-r--r-- | drivers/infiniband/hw/qedr/verbs.c | 1020 |
1 files changed, 1020 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c index e70e808e2b9b..e4b4d47c6a2d 100644 --- a/drivers/infiniband/hw/qedr/verbs.c +++ b/drivers/infiniband/hw/qedr/verbs.c @@ -2447,3 +2447,1023 @@ err1: kfree(mr); return ERR_PTR(rc); } + +static inline int qedr_wq_is_full(struct qedr_qp_hwq_info *wq) +{ + return (((wq->prod + 1) % wq->max_wr) == wq->cons); +} + +static int sge_data_len(struct ib_sge *sg_list, int num_sge) +{ + int i, len = 0; + + for (i = 0; i < num_sge; i++) + len += sg_list[i].length; + + return len; +} + +static void swap_wqe_data64(u64 *p) +{ + int i; + + for (i = 0; i < QEDR_SQE_ELEMENT_SIZE / sizeof(u64); i++, p++) + *p = cpu_to_be64(cpu_to_le64(*p)); +} + +static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev, + struct qedr_qp *qp, u8 *wqe_size, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr, u8 *bits, + u8 bit) +{ + u32 data_size = sge_data_len(wr->sg_list, wr->num_sge); + char *seg_prt, *wqe; + int i, seg_siz; + + if (data_size > ROCE_REQ_MAX_INLINE_DATA_SIZE) { + DP_ERR(dev, "Too much inline data in WR: %d\n", data_size); + *bad_wr = wr; + return 0; + } + + if (!data_size) + return data_size; + + *bits |= bit; + + seg_prt = NULL; + wqe = NULL; + seg_siz = 0; + + /* Copy data inline */ + for (i = 0; i < wr->num_sge; i++) { + u32 len = wr->sg_list[i].length; + void *src = (void *)(uintptr_t)wr->sg_list[i].addr; + + while (len > 0) { + u32 cur; + + /* New segment required */ + if (!seg_siz) { + wqe = (char *)qed_chain_produce(&qp->sq.pbl); + seg_prt = wqe; + seg_siz = sizeof(struct rdma_sq_common_wqe); + (*wqe_size)++; + } + + /* Calculate currently allowed length */ + cur = min_t(u32, len, seg_siz); + memcpy(seg_prt, src, cur); + + /* Update segment variables */ + seg_prt += cur; + seg_siz -= cur; + + /* Update sge variables */ + src += cur; + len -= cur; + + /* Swap fully-completed segments */ + if (!seg_siz) + swap_wqe_data64((u64 *)wqe); + } + } + + /* swap last not completed segment */ + if (seg_siz) + swap_wqe_data64((u64 *)wqe); + + return data_size; +} + +#define RQ_SGE_SET(sge, vaddr, vlength, vflags) \ + do { \ + DMA_REGPAIR_LE(sge->addr, vaddr); \ + (sge)->length = cpu_to_le32(vlength); \ + (sge)->flags = cpu_to_le32(vflags); \ + } while (0) + +#define SRQ_HDR_SET(hdr, vwr_id, num_sge) \ + do { \ + DMA_REGPAIR_LE(hdr->wr_id, vwr_id); \ + (hdr)->num_sges = num_sge; \ + } while (0) + +#define SRQ_SGE_SET(sge, vaddr, vlength, vlkey) \ + do { \ + DMA_REGPAIR_LE(sge->addr, vaddr); \ + (sge)->length = cpu_to_le32(vlength); \ + (sge)->l_key = cpu_to_le32(vlkey); \ + } while (0) + +static u32 qedr_prepare_sq_sges(struct qedr_qp *qp, u8 *wqe_size, + struct ib_send_wr *wr) +{ + u32 data_size = 0; + int i; + + for (i = 0; i < wr->num_sge; i++) { + struct rdma_sq_sge *sge = qed_chain_produce(&qp->sq.pbl); + + DMA_REGPAIR_LE(sge->addr, wr->sg_list[i].addr); + sge->l_key = cpu_to_le32(wr->sg_list[i].lkey); + sge->length = cpu_to_le32(wr->sg_list[i].length); + data_size += wr->sg_list[i].length; + } + + if (wqe_size) + *wqe_size += wr->num_sge; + + return data_size; +} + +static u32 qedr_prepare_sq_rdma_data(struct qedr_dev *dev, + struct qedr_qp *qp, + struct rdma_sq_rdma_wqe_1st *rwqe, + struct rdma_sq_rdma_wqe_2nd *rwqe2, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey); + DMA_REGPAIR_LE(rwqe2->remote_va, rdma_wr(wr)->remote_addr); + + if (wr->send_flags & IB_SEND_INLINE) { + u8 flags = 0; + + SET_FIELD2(flags, RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG, 1); + return qedr_prepare_sq_inline_data(dev, qp, &rwqe->wqe_size, wr, + bad_wr, &rwqe->flags, flags); + } + + return qedr_prepare_sq_sges(qp, &rwqe->wqe_size, wr); +} + +static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev, + struct qedr_qp *qp, + struct rdma_sq_send_wqe_1st *swqe, + struct rdma_sq_send_wqe_2st *swqe2, + struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + memset(swqe2, 0, sizeof(*swqe2)); + if (wr->send_flags & IB_SEND_INLINE) { + u8 flags = 0; + + SET_FIELD2(flags, RDMA_SQ_SEND_WQE_INLINE_FLG, 1); + return qedr_prepare_sq_inline_data(dev, qp, &swqe->wqe_size, wr, + bad_wr, &swqe->flags, flags); + } + + return qedr_prepare_sq_sges(qp, &swqe->wqe_size, wr); +} + +static int qedr_prepare_reg(struct qedr_qp *qp, + struct rdma_sq_fmr_wqe_1st *fwqe1, + struct ib_reg_wr *wr) +{ + struct qedr_mr *mr = get_qedr_mr(wr->mr); + struct rdma_sq_fmr_wqe_2nd *fwqe2; + + fwqe2 = (struct rdma_sq_fmr_wqe_2nd *)qed_chain_produce(&qp->sq.pbl); + fwqe1->addr.hi = upper_32_bits(mr->ibmr.iova); + fwqe1->addr.lo = lower_32_bits(mr->ibmr.iova); + fwqe1->l_key = wr->key; + + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_READ, + !!(wr->access & IB_ACCESS_REMOTE_READ)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE, + !!(wr->access & IB_ACCESS_REMOTE_WRITE)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC, + !!(wr->access & IB_ACCESS_REMOTE_ATOMIC)); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_READ, 1); + SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE, + !!(wr->access & IB_ACCESS_LOCAL_WRITE)); + fwqe2->fmr_ctrl = 0; + + SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG, + ilog2(mr->ibmr.page_size) - 12); + + fwqe2->length_hi = 0; + fwqe2->length_lo = mr->ibmr.length; + fwqe2->pbl_addr.hi = upper_32_bits(mr->info.pbl_table->pa); + fwqe2->pbl_addr.lo = lower_32_bits(mr->info.pbl_table->pa); + + qp->wqe_wr_id[qp->sq.prod].mr = mr; + + return 0; +} + +enum ib_wc_opcode qedr_ib_to_wc_opcode(enum ib_wr_opcode opcode) +{ + switch (opcode) { + case IB_WR_RDMA_WRITE: + case IB_WR_RDMA_WRITE_WITH_IMM: + return IB_WC_RDMA_WRITE; + case IB_WR_SEND_WITH_IMM: + case IB_WR_SEND: + case IB_WR_SEND_WITH_INV: + return IB_WC_SEND; + case IB_WR_RDMA_READ: + return IB_WC_RDMA_READ; + case IB_WR_ATOMIC_CMP_AND_SWP: + return IB_WC_COMP_SWAP; + case IB_WR_ATOMIC_FETCH_AND_ADD: + return IB_WC_FETCH_ADD; + case IB_WR_REG_MR: + return IB_WC_REG_MR; + case IB_WR_LOCAL_INV: + return IB_WC_LOCAL_INV; + default: + return IB_WC_SEND; + } +} + +inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr) +{ + int wq_is_full, err_wr, pbl_is_full; + struct qedr_dev *dev = qp->dev; + + /* prevent SQ overflow and/or processing of a bad WR */ + err_wr = wr->num_sge > qp->sq.max_sges; + wq_is_full = qedr_wq_is_full(&qp->sq); + pbl_is_full = qed_chain_get_elem_left_u32(&qp->sq.pbl) < + QEDR_MAX_SQE_ELEMENTS_PER_SQE; + if (wq_is_full || err_wr || pbl_is_full) { + if (wq_is_full && !(qp->err_bitmap & QEDR_QP_ERR_SQ_FULL)) { + DP_ERR(dev, + "error: WQ is full. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_SQ_FULL; + } + + if (err_wr && !(qp->err_bitmap & QEDR_QP_ERR_BAD_SR)) { + DP_ERR(dev, + "error: WR is bad. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_BAD_SR; + } + + if (pbl_is_full && + !(qp->err_bitmap & QEDR_QP_ERR_SQ_PBL_FULL)) { + DP_ERR(dev, + "error: WQ PBL is full. Post send on QP %p failed (this error appears only once)\n", + qp); + qp->err_bitmap |= QEDR_QP_ERR_SQ_PBL_FULL; + } + return false; + } + return true; +} + +int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct qedr_dev *dev = get_qedr_dev(ibqp->device); + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct rdma_sq_atomic_wqe_1st *awqe1; + struct rdma_sq_atomic_wqe_2nd *awqe2; + struct rdma_sq_atomic_wqe_3rd *awqe3; + struct rdma_sq_send_wqe_2st *swqe2; + struct rdma_sq_local_inv_wqe *iwqe; + struct rdma_sq_rdma_wqe_2nd *rwqe2; + struct rdma_sq_send_wqe_1st *swqe; + struct rdma_sq_rdma_wqe_1st *rwqe; + struct rdma_sq_fmr_wqe_1st *fwqe1; + struct rdma_sq_common_wqe *wqe; + u32 length; + int rc = 0; + bool comp; + + if (!qedr_can_post_send(qp, wr)) { + *bad_wr = wr; + return -ENOMEM; + } + + wqe = qed_chain_produce(&qp->sq.pbl); + qp->wqe_wr_id[qp->sq.prod].signaled = + !!(wr->send_flags & IB_SEND_SIGNALED) || qp->signaled; + + wqe->flags = 0; + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_SE_FLG, + !!(wr->send_flags & IB_SEND_SOLICITED)); + comp = (!!(wr->send_flags & IB_SEND_SIGNALED)) || qp->signaled; + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_COMP_FLG, comp); + SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_RD_FENCE_FLG, + !!(wr->send_flags & IB_SEND_FENCE)); + wqe->prev_wqe_size = qp->prev_wqe_size; + + qp->wqe_wr_id[qp->sq.prod].opcode = qedr_ib_to_wc_opcode(wr->opcode); + + switch (wr->opcode) { + case IB_WR_SEND_WITH_IMM: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_IMM; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + swqe->wqe_size = 2; + swqe2 = qed_chain_produce(&qp->sq.pbl); + + swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.imm_data); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + case IB_WR_SEND: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + + swqe->wqe_size = 2; + swqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + case IB_WR_SEND_WITH_INV: + wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE; + swqe = (struct rdma_sq_send_wqe_1st *)wqe; + swqe2 = qed_chain_produce(&qp->sq.pbl); + swqe->wqe_size = 2; + swqe->inv_key_or_imm_data = cpu_to_le32(wr->ex.invalidate_rkey); + length = qedr_prepare_sq_send_data(dev, qp, swqe, swqe2, + wr, bad_wr); + swqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size; + qp->prev_wqe_size = swqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length; + break; + + case IB_WR_RDMA_WRITE_WITH_IMM: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe->imm_data = htonl(cpu_to_le32(wr->ex.imm_data)); + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + case IB_WR_RDMA_WRITE: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + case IB_WR_RDMA_READ_WITH_INV: + DP_ERR(dev, + "RDMA READ WITH INVALIDATE not supported\n"); + *bad_wr = wr; + rc = -EINVAL; + break; + + case IB_WR_RDMA_READ: + wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_RD; + rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe; + + rwqe->wqe_size = 2; + rwqe2 = qed_chain_produce(&qp->sq.pbl); + length = qedr_prepare_sq_rdma_data(dev, qp, rwqe, rwqe2, + wr, bad_wr); + rwqe->length = cpu_to_le32(length); + qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size; + qp->prev_wqe_size = rwqe->wqe_size; + qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length; + break; + + case IB_WR_ATOMIC_CMP_AND_SWP: + case IB_WR_ATOMIC_FETCH_AND_ADD: + awqe1 = (struct rdma_sq_atomic_wqe_1st *)wqe; + awqe1->wqe_size = 4; + + awqe2 = qed_chain_produce(&qp->sq.pbl); + DMA_REGPAIR_LE(awqe2->remote_va, atomic_wr(wr)->remote_addr); + awqe2->r_key = cpu_to_le32(atomic_wr(wr)->rkey); + + awqe3 = qed_chain_produce(&qp->sq.pbl); + + if (wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) { + wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_ADD; + DMA_REGPAIR_LE(awqe3->swap_data, + atomic_wr(wr)->compare_add); + } else { + wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP; + DMA_REGPAIR_LE(awqe3->swap_data, + atomic_wr(wr)->swap); + DMA_REGPAIR_LE(awqe3->cmp_data, + atomic_wr(wr)->compare_add); + } + + qedr_prepare_sq_sges(qp, NULL, wr); + + qp->wqe_wr_id[qp->sq.prod].wqe_size = awqe1->wqe_size; + qp->prev_wqe_size = awqe1->wqe_size; + break; + + case IB_WR_LOCAL_INV: + iwqe = (struct rdma_sq_local_inv_wqe *)wqe; + iwqe->wqe_size = 1; + + iwqe->req_type = RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE; + iwqe->inv_l_key = wr->ex.invalidate_rkey; + qp->wqe_wr_id[qp->sq.prod].wqe_size = iwqe->wqe_size; + qp->prev_wqe_size = iwqe->wqe_size; + break; + case IB_WR_REG_MR: + DP_DEBUG(dev, QEDR_MSG_CQ, "REG_MR\n"); + wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR; + fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe; + fwqe1->wqe_size = 2; + + rc = qedr_prepare_reg(qp, fwqe1, reg_wr(wr)); + if (rc) { + DP_ERR(dev, "IB_REG_MR failed rc=%d\n", rc); + *bad_wr = wr; + break; + } + + qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->wqe_size; + qp->prev_wqe_size = fwqe1->wqe_size; + break; + default: + DP_ERR(dev, "invalid opcode 0x%x!\n", wr->opcode); + rc = -EINVAL; + *bad_wr = wr; + break; + } + + if (*bad_wr) { + u16 value; + + /* Restore prod to its position before + * this WR was processed + */ + value = le16_to_cpu(qp->sq.db_data.data.value); + qed_chain_set_prod(&qp->sq.pbl, value, wqe); + + /* Restore prev_wqe_size */ + qp->prev_wqe_size = wqe->prev_wqe_size; + rc = -EINVAL; + DP_ERR(dev, "POST SEND FAILED\n"); + } + + return rc; +} + +int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct qedr_dev *dev = get_qedr_dev(ibqp->device); + struct qedr_qp *qp = get_qedr_qp(ibqp); + unsigned long flags; + int rc = 0; + + *bad_wr = NULL; + + spin_lock_irqsave(&qp->q_lock, flags); + + if ((qp->state == QED_ROCE_QP_STATE_RESET) || + (qp->state == QED_ROCE_QP_STATE_ERR)) { + spin_unlock_irqrestore(&qp->q_lock, flags); + *bad_wr = wr; + DP_DEBUG(dev, QEDR_MSG_CQ, + "QP in wrong state! QP icid=0x%x state %d\n", + qp->icid, qp->state); + return -EINVAL; + } + + if (!wr) { + DP_ERR(dev, "Got an empty post send.\n"); + return -EINVAL; + } + + while (wr) { + rc = __qedr_post_send(ibqp, wr, bad_wr); + if (rc) + break; + + qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id; + + qedr_inc_sw_prod(&qp->sq); + + qp->sq.db_data.data.value++; + + wr = wr->next; + } + + /* Trigger doorbell + * If there was a failure in the first WR then it will be triggered in + * vane. However this is not harmful (as long as the producer value is + * unchanged). For performance reasons we avoid checking for this + * redundant doorbell. + */ + wmb(); + writel(qp->sq.db_data.raw, qp->sq.db); + + /* Make sure write sticks */ + mmiowb(); + + spin_unlock_irqrestore(&qp->q_lock, flags); + + return rc; +} + +int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr) +{ + struct qedr_qp *qp = get_qedr_qp(ibqp); + struct qedr_dev *dev = qp->dev; + unsigned long flags; + int status = 0; + + spin_lock_irqsave(&qp->q_lock, flags); + + if ((qp->state == QED_ROCE_QP_STATE_RESET) || + (qp->state == QED_ROCE_QP_STATE_ERR)) { + spin_unlock_irqrestore(&qp->q_lock, flags); + *bad_wr = wr; + return -EINVAL; + } + + while (wr) { + int i; + + if (qed_chain_get_elem_left_u32(&qp->rq.pbl) < + QEDR_MAX_RQE_ELEMENTS_PER_RQE || + wr->num_sge > qp->rq.max_sges) { + DP_ERR(dev, "Can't post WR (%d < %d) || (%d > %d)\n", + qed_chain_get_elem_left_u32(&qp->rq.pbl), + QEDR_MAX_RQE_ELEMENTS_PER_RQE, wr->num_sge, + qp->rq.max_sges); + status = -ENOMEM; + *bad_wr = wr; + break; + } + for (i = 0; i < wr->num_sge; i++) { + u32 flags = 0; + struct rdma_rq_sge *rqe = + qed_chain_produce(&qp->rq.pbl); + + /* First one must include the number + * of SGE in the list + */ + if (!i) + SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, + wr->num_sge); + + SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, + wr->sg_list[i].lkey); + + RQ_SGE_SET(rqe, wr->sg_list[i].addr, + wr->sg_list[i].length, flags); + } + + /* Special case of no sges. FW requires between 1-4 sges... + * in this case we need to post 1 sge with length zero. this is + * because rdma write with immediate consumes an RQ. + */ + if (!wr->num_sge) { + u32 flags = 0; + struct rdma_rq_sge *rqe = + qed_chain_produce(&qp->rq.pbl); + + /* First one must include the number + * of SGE in the list + */ + SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, 0); + SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, 1); + + RQ_SGE_SET(rqe, 0, 0, flags); + i = 1; + } + + qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id; + qp->rqe_wr_id[qp->rq.prod].wqe_size = i; + + qedr_inc_sw_prod(&qp->rq); + + /* Flush all the writes before signalling doorbell */ + wmb(); + + qp->rq.db_data.data.value++; + + writel(qp->rq.db_data.raw, qp->rq.db); + + /* Make sure write sticks */ + mmiowb(); + + wr = wr->next; + } + + spin_unlock_irqrestore(&qp->q_lock, flags); + + return status; +} + +static int is_valid_cqe(struct qedr_cq *cq, union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + + return (resp_cqe->flags & RDMA_CQE_REQUESTER_TOGGLE_BIT_MASK) == + cq->pbl_toggle; +} + +static struct qedr_qp *cqe_get_qp(union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + struct qedr_qp *qp; + + qp = (struct qedr_qp *)(uintptr_t)HILO_GEN(resp_cqe->qp_handle.hi, + resp_cqe->qp_handle.lo, + u64); + return qp; +} + +static enum rdma_cqe_type cqe_get_type(union rdma_cqe *cqe) +{ + struct rdma_cqe_requester *resp_cqe = &cqe->req; + + return GET_FIELD(resp_cqe->flags, RDMA_CQE_REQUESTER_TYPE); +} + +/* Return latest CQE (needs processing) */ +static union rdma_cqe *get_cqe(struct qedr_cq *cq) +{ + return cq->latest_cqe; +} + +/* In fmr we need to increase the number of fmr completed counter for the fmr + * algorithm determining whether we can free a pbl or not. + * we need to perform this whether the work request was signaled or not. for + * this purpose we call this function from the condition that checks if a wr + * should be skipped, to make sure we don't miss it ( possibly this fmr + * operation was not signalted) + */ +static inline void qedr_chk_if_fmr(struct qedr_qp *qp) +{ + if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_REG_MR) + qp->wqe_wr_id[qp->sq.cons].mr->info.completed++; +} + +static int process_req(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, int num_entries, + struct ib_wc *wc, u16 hw_cons, enum ib_wc_status status, + int force) +{ + u16 cnt = 0; + + while (num_entries && qp->sq.wqe_cons != hw_cons) { + if (!qp->wqe_wr_id[qp->sq.cons].signaled && !force) { + qedr_chk_if_fmr(qp); + /* skip WC */ + goto next_cqe; + } + + /* fill WC */ + wc->status = status; + wc->wc_flags = 0; + wc->src_qp = qp->id; + wc->qp = &qp->ibqp; + + wc->wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id; + wc->opcode = qp->wqe_wr_id[qp->sq.cons].opcode; + + switch (wc->opcode) { + case IB_WC_RDMA_WRITE: + wc->byte_len = qp->wqe_wr_id[qp->sq.cons].bytes_len; + break; + case IB_WC_COMP_SWAP: + case IB_WC_FETCH_ADD: + wc->byte_len = 8; + break; + case IB_WC_REG_MR: + qp->wqe_wr_id[qp->sq.cons].mr->info.completed++; + break; + default: + break; + } + + num_entries--; + wc++; + cnt++; +next_cqe: + while (qp->wqe_wr_id[qp->sq.cons].wqe_size--) + qed_chain_consume(&qp->sq.pbl); + qedr_inc_sw_cons(&qp->sq); + } + + return cnt; +} + +static int qedr_poll_cq_req(struct qedr_dev *dev, + struct qedr_qp *qp, struct qedr_cq *cq, + int num_entries, struct ib_wc *wc, + struct rdma_cqe_requester *req) +{ + int cnt = 0; + + switch (req->status) { + case RDMA_CQE_REQ_STS_OK: + cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons, + IB_WC_SUCCESS, 0); + break; + case RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons, + IB_WC_WR_FLUSH_ERR, 0); + break; + default: + /* process all WQE before the cosumer */ + qp->state = QED_ROCE_QP_STATE_ERR; + cnt = process_req(dev, qp, cq, num_entries, wc, + req->sq_cons - 1, IB_WC_SUCCESS, 0); + wc += cnt; + /* if we have extra WC fill it with actual error info */ + if (cnt < num_entries) { + enum ib_wc_status wc_status; + + switch (req->status) { + case RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_BAD_RESP_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_LEN_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_QP_OP_ERR; + break; + case RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_LOC_PROT_ERR; + break; + case RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_MW_BIND_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_INV_REQ_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_ACCESS_ERR; + break; + case RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_REM_OP_ERR; + break; + case RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR: + DP_ERR(dev, + "Error: POLL CQ with RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_RNR_RETRY_EXC_ERR; + break; + case RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR: + DP_ERR(dev, + "Error: POLL CQ with ROCE_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_RETRY_EXC_ERR; + break; + default: + DP_ERR(dev, + "Error: POLL CQ with IB_WC_GENERAL_ERR. CQ icid=0x%x, QP icid=0x%x\n", + cq->icid, qp->icid); + wc_status = IB_WC_GENERAL_ERR; + } + cnt += process_req(dev, qp, cq, 1, wc, req->sq_cons, + wc_status, 1); + } + } + + return cnt; +} + +static void __process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, struct ib_wc *wc, + struct rdma_cqe_responder *resp, u64 wr_id) +{ + enum ib_wc_status wc_status = IB_WC_SUCCESS; + u8 flags; + + wc->opcode = IB_WC_RECV; + wc->wc_flags = 0; + + switch (resp->status) { + case RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR: + wc_status = IB_WC_LOC_ACCESS_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR: + wc_status = IB_WC_LOC_LEN_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR: + wc_status = IB_WC_LOC_QP_OP_ERR; + break; + case RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR: + wc_status = IB_WC_LOC_PROT_ERR; + break; + case RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR: + wc_status = IB_WC_MW_BIND_ERR; + break; + case RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR: + wc_status = IB_WC_REM_INV_RD_REQ_ERR; + break; + case RDMA_CQE_RESP_STS_OK: + wc_status = IB_WC_SUCCESS; + wc->byte_len = le32_to_cpu(resp->length); + + flags = resp->flags & QEDR_RESP_RDMA_IMM; + + if (flags == QEDR_RESP_RDMA_IMM) + wc->opcode = IB_WC_RECV_RDMA_WITH_IMM; + + if (flags == QEDR_RESP_RDMA_IMM || flags == QEDR_RESP_IMM) { + wc->ex.imm_data = + le32_to_cpu(resp->imm_data_or_inv_r_Key); + wc->wc_flags |= IB_WC_WITH_IMM; + } + break; + default: + wc->status = IB_WC_GENERAL_ERR; + DP_ERR(dev, "Invalid CQE status detected\n"); + } + + /* fill WC */ + wc->status = wc_status; + wc->src_qp = qp->id; + wc->qp = &qp->ibqp; + wc->wr_id = wr_id; +} + +static int process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, struct ib_wc *wc, + struct rdma_cqe_responder *resp) +{ + u64 wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id; + + __process_resp_one(dev, qp, cq, wc, resp, wr_id); + + while (qp->rqe_wr_id[qp->rq.cons].wqe_size--) + qed_chain_consume(&qp->rq.pbl); + qedr_inc_sw_cons(&qp->rq); + + return 1; +} + +static int process_resp_flush(struct qedr_qp *qp, struct qedr_cq *cq, + int num_entries, struct ib_wc *wc, u16 hw_cons) +{ + u16 cnt = 0; + + while (num_entries && qp->rq.wqe_cons != hw_cons) { + /* fill WC */ + wc->status = IB_WC_WR_FLUSH_ERR; + wc->wc_flags = 0; + wc->src_qp = qp->id; + wc->byte_len = 0; + wc->wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id; + wc->qp = &qp->ibqp; + num_entries--; + wc++; + cnt++; + while (qp->rqe_wr_id[qp->rq.cons].wqe_size--) + qed_chain_consume(&qp->rq.pbl); + qedr_inc_sw_cons(&qp->rq); + } + + return cnt; +} + +static void try_consume_resp_cqe(struct qedr_cq *cq, struct qedr_qp *qp, + struct rdma_cqe_responder *resp, int *update) +{ + if (le16_to_cpu(resp->rq_cons) == qp->rq.wqe_cons) { + consume_cqe(cq); + *update |= 1; + } +} + +static int qedr_poll_cq_resp(struct qedr_dev *dev, struct qedr_qp *qp, + struct qedr_cq *cq, int num_entries, + struct ib_wc *wc, struct rdma_cqe_responder *resp, + int *update) +{ + int cnt; + + if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) { + cnt = process_resp_flush(qp, cq, num_entries, wc, + resp->rq_cons); + try_consume_resp_cqe(cq, qp, resp, update); + } else { + cnt = process_resp_one(dev, qp, cq, wc, resp); + consume_cqe(cq); + *update |= 1; + } + + return cnt; +} + +static void try_consume_req_cqe(struct qedr_cq *cq, struct qedr_qp *qp, + struct rdma_cqe_requester *req, int *update) +{ + if (le16_to_cpu(req->sq_cons) == qp->sq.wqe_cons) { + consume_cqe(cq); + *update |= 1; + } +} + +int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc) +{ + struct qedr_dev *dev = get_qedr_dev(ibcq->device); + struct qedr_cq *cq = get_qedr_cq(ibcq); + union rdma_cqe *cqe = cq->latest_cqe; + u32 old_cons, new_cons; + unsigned long flags; + int update = 0; + int done = 0; + + spin_lock_irqsave(&cq->cq_lock, flags); + old_cons = qed_chain_get_cons_idx_u32(&cq->pbl); + while (num_entries && is_valid_cqe(cq, cqe)) { + struct qedr_qp *qp; + int cnt = 0; + + /* prevent speculative reads of any field of CQE */ + rmb(); + + qp = cqe_get_qp(cqe); + if (!qp) { + WARN(1, "Error: CQE QP pointer is NULL. CQE=%p\n", cqe); + break; + } + + wc->qp = &qp->ibqp; + + switch (cqe_get_type(cqe)) { + case RDMA_CQE_TYPE_REQUESTER: + cnt = qedr_poll_cq_req(dev, qp, cq, num_entries, wc, + &cqe->req); + try_consume_req_cqe(cq, qp, &cqe->req, &update); + break; + case RDMA_CQE_TYPE_RESPONDER_RQ: + cnt = qedr_poll_cq_resp(dev, qp, cq, num_entries, wc, + &cqe->resp, &update); + break; + case RDMA_CQE_TYPE_INVALID: + default: + DP_ERR(dev, "Error: invalid CQE type = %d\n", + cqe_get_type(cqe)); + } + num_entries -= cnt; + wc += cnt; + done += cnt; + + cqe = get_cqe(cq); + } + new_cons = qed_chain_get_cons_idx_u32(&cq->pbl); + + cq->cq_cons += new_cons - old_cons; + + if (update) + /* doorbell notifies abount latest VALID entry, + * but chain already point to the next INVALID one + */ + doorbell_cq(cq, cq->cq_cons - 1, cq->arm_flags); + + spin_unlock_irqrestore(&cq->cq_lock, flags); + return done; +} |