From patchwork Tue Jul 17 07:38:17 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Potnuri Bharat Teja X-Patchwork-Id: 10528235 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork.web.codeaurora.org (Postfix) with ESMTP id 64CF0600D0 for ; Tue, 17 Jul 2018 07:38:41 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 486AA28614 for ; Tue, 17 Jul 2018 07:38:41 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 3D16E28623; Tue, 17 Jul 2018 07:38:41 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.9 required=2.0 tests=BAYES_00, MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 6757D28614 for ; Tue, 17 Jul 2018 07:38:40 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1728515AbeGQIJy (ORCPT ); Tue, 17 Jul 2018 04:09:54 -0400 Received: from stargate.chelsio.com ([12.32.117.8]:25884 "EHLO stargate.chelsio.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1727052AbeGQIJy (ORCPT ); Tue, 17 Jul 2018 04:09:54 -0400 Received: from localhost (chittorgarh.blr.asicdesigners.com [10.193.184.65]) by stargate.chelsio.com (8.13.8/8.13.8) with ESMTP id w6H7cY6q011952; Tue, 17 Jul 2018 00:38:35 -0700 From: Potnuri Bharat Teja To: jgg@ziepe.ca, dledford@redhat.com Cc: swise@opengridcomputing.com, rajur@chelsio.com, bharat@chelsio.com, linux-rdma@vger.kernel.org Subject: [PATCH for-next v2 1/2] iw_cxgb4: RDMA write with immediate support Date: Tue, 17 Jul 2018 13:08:17 +0530 Message-Id: <20180717073818.13442-2-bharat@chelsio.com> X-Mailer: git-send-email 2.9.3 In-Reply-To: <20180717073818.13442-1-bharat@chelsio.com> References: <20180717073818.13442-1-bharat@chelsio.com> Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP Adds iw_cxgb4 functionality to support RDMA_WRITE_WITH_IMMEDATE opcode. Signed-off-by: Potnuri Bharat Teja Signed-off-by: Steve Wise --- drivers/infiniband/hw/cxgb4/cq.c | 23 ++++++++++++++++--- drivers/infiniband/hw/cxgb4/qp.c | 37 ++++++++++++++++++++++++------- drivers/infiniband/hw/cxgb4/t4.h | 16 ++++++++++++- drivers/infiniband/hw/cxgb4/t4fw_ri_api.h | 18 ++++++++++++--- include/uapi/rdma/cxgb4-abi.h | 3 ++- 5 files changed, 81 insertions(+), 16 deletions(-) diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c index 64f70800945b..5792675150e0 100644 --- a/drivers/infiniband/hw/cxgb4/cq.c +++ b/drivers/infiniband/hw/cxgb4/cq.c @@ -816,15 +816,32 @@ static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc) wc->byte_len = CQE_LEN(&cqe); else wc->byte_len = 0; - wc->opcode = IB_WC_RECV; - if (CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_INV || - CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_SE_INV) { + + switch (CQE_OPCODE(&cqe)) { + case FW_RI_SEND: + wc->opcode = IB_WC_RECV; + break; + case FW_RI_SEND_WITH_INV: + case FW_RI_SEND_WITH_SE_INV: + wc->opcode = IB_WC_RECV; wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe); wc->wc_flags |= IB_WC_WITH_INVALIDATE; c4iw_invalidate_mr(qhp->rhp, wc->ex.invalidate_rkey); + break; + case FW_RI_WRITE_IMMEDIATE: + wc->opcode = IB_WC_RECV_RDMA_WITH_IMM; + wc->ex.imm_data = CQE_IMM_DATA(&cqe); + wc->wc_flags |= IB_WC_WITH_IMM; + break; + default: + pr_err("Unexpected opcode %d in the CQE received for QPID=0x%0x\n", + CQE_OPCODE(&cqe), CQE_QPID(&cqe)); + ret = -EINVAL; + goto out; } } else { switch (CQE_OPCODE(&cqe)) { + case FW_RI_WRITE_IMMEDIATE: case FW_RI_RDMA_WRITE: wc->opcode = IB_WC_RDMA_WRITE; break; diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index 08dc555942af..88bd67d7dc77 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -555,7 +555,15 @@ static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe, if (wr->num_sge > T4_MAX_SEND_SGE) return -EINVAL; - wqe->write.r2 = 0; + + /* + * iWARP protocol supports 64 bit immediate data but rdma api + * limits it to 32bit. + */ + if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) + wqe->write.iw_imm_data.ib_imm_data.imm_data32 = wr->ex.imm_data; + else + wqe->write.iw_imm_data.ib_imm_data.imm_data32 = 0; wqe->write.stag_sink = cpu_to_be32(rdma_wr(wr)->rkey); wqe->write.to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr); if (wr->num_sge) { @@ -846,6 +854,9 @@ static int ib_to_fw_opcode(int ib_opcode) case IB_WR_RDMA_WRITE: opcode = FW_RI_RDMA_WRITE; break; + case IB_WR_RDMA_WRITE_WITH_IMM: + opcode = FW_RI_WRITE_IMMEDIATE; + break; case IB_WR_RDMA_READ: case IB_WR_RDMA_READ_WITH_INV: opcode = FW_RI_READ_REQ; @@ -964,6 +975,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, enum fw_wr_opcodes fw_opcode = 0; enum fw_ri_wr_flags fw_flags; struct c4iw_qp *qhp; + struct c4iw_dev *rhp; union t4_wr *wqe = NULL; u32 num_wrs; struct t4_swsqe *swsqe; @@ -971,6 +983,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, u16 idx = 0; qhp = to_c4iw_qp(ibqp); + rhp = qhp->rhp; spin_lock_irqsave(&qhp->lock, flag); /* @@ -1015,6 +1028,13 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, swsqe->opcode = FW_RI_SEND_WITH_INV; err = build_rdma_send(&qhp->wq.sq, wqe, wr, &len16); break; + case IB_WR_RDMA_WRITE_WITH_IMM: + if (unlikely(!rhp->rdev.lldi.write_w_imm_support)) { + err = -EINVAL; + break; + } + fw_flags |= FW_RI_RDMA_WRITE_WITH_IMMEDIATE; + /*FALLTHROUGH*/ case IB_WR_RDMA_WRITE: fw_opcode = FW_RI_RDMA_WRITE_WR; swsqe->opcode = FW_RI_RDMA_WRITE; @@ -1025,8 +1045,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, fw_opcode = FW_RI_RDMA_READ_WR; swsqe->opcode = FW_RI_READ_REQ; if (wr->opcode == IB_WR_RDMA_READ_WITH_INV) { - c4iw_invalidate_mr(qhp->rhp, - wr->sg_list[0].lkey); + c4iw_invalidate_mr(rhp, wr->sg_list[0].lkey); fw_flags = FW_RI_RDMA_READ_INVALIDATE; } else { fw_flags = 0; @@ -1042,7 +1061,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, struct c4iw_mr *mhp = to_c4iw_mr(reg_wr(wr)->mr); swsqe->opcode = FW_RI_FAST_REGISTER; - if (qhp->rhp->rdev.lldi.fr_nsmr_tpte_wr_support && + if (rhp->rdev.lldi.fr_nsmr_tpte_wr_support && !mhp->attr.state && mhp->mpl_len <= 2) { fw_opcode = FW_RI_FR_NSMR_TPTE_WR; build_tpte_memreg(&wqe->fr_tpte, reg_wr(wr), @@ -1051,7 +1070,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, fw_opcode = FW_RI_FR_NSMR_WR; err = build_memreg(&qhp->wq.sq, wqe, reg_wr(wr), mhp, &len16, - qhp->rhp->rdev.lldi.ulptx_memwrite_dsgl); + rhp->rdev.lldi.ulptx_memwrite_dsgl); if (err) break; } @@ -1064,7 +1083,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, fw_opcode = FW_RI_INV_LSTAG_WR; swsqe->opcode = FW_RI_LOCAL_INV; err = build_inv_stag(wqe, wr, &len16); - c4iw_invalidate_mr(qhp->rhp, wr->ex.invalidate_rkey); + c4iw_invalidate_mr(rhp, wr->ex.invalidate_rkey); break; default: pr_warn("%s post of type=%d TBD!\n", __func__, @@ -1083,7 +1102,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, swsqe->wr_id = wr->wr_id; if (c4iw_wr_log) { swsqe->sge_ts = cxgb4_read_sge_timestamp( - qhp->rhp->rdev.lldi.ports[0]); + rhp->rdev.lldi.ports[0]); swsqe->host_time = ktime_get(); } @@ -1097,7 +1116,7 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, t4_sq_produce(&qhp->wq, len16); idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE); } - if (!qhp->rhp->rdev.status_page->db_off) { + if (!rhp->rdev.status_page->db_off) { t4_ring_sq_db(&qhp->wq, idx, wqe); spin_unlock_irqrestore(&qhp->lock, flag); } else { @@ -2092,6 +2111,8 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, uresp.flags = C4IW_QPF_ONCHIP; } else uresp.flags = 0; + if (rhp->rdev.lldi.write_w_imm_support) + uresp.flags |= C4IW_QPF_WRITE_W_IMM; uresp.qid_mask = rhp->rdev.qpmask; uresp.sqid = qhp->wq.sq.qid; uresp.sq_size = qhp->wq.sq.size; diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h index 11d55fc2ded7..3fdcc445ff16 100644 --- a/drivers/infiniband/hw/cxgb4/t4.h +++ b/drivers/infiniband/hw/cxgb4/t4.h @@ -190,7 +190,19 @@ struct t4_cqe { __be32 abs_rqe_idx; } srcqe; struct { - __be64 imm_data; + __be32 mo; + __be32 msn; + /* + * Use union for immediate data to be consistent with + * stack's 32 bit data and iWARP spec's 64 bit data. + */ + union { + struct { + u32 reserved; + __be32 imm_data32; + } ib_imm_data; + __be64 imm_data64; + } iw_imm_data; } imm_data_rcqe; u64 drain_cookie; @@ -253,6 +265,8 @@ struct t4_cqe { #define CQE_WRID_STAG(x) (be32_to_cpu((x)->u.rcqe.stag)) #define CQE_WRID_MSN(x) (be32_to_cpu((x)->u.rcqe.msn)) #define CQE_ABS_RQE_IDX(x) (be32_to_cpu((x)->u.srcqe.abs_rqe_idx)) +#define CQE_IMM_DATA(x)( \ + (x)->u.imm_data_rcqe.iw_imm_data.ib_imm_data.imm_data32) /* used for SQ completion processing */ #define CQE_WRID_SQ_IDX(x) ((x)->u.scqe.cidx) diff --git a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h index 0f4f86b004d6..45e7211e94ac 100644 --- a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h +++ b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h @@ -50,7 +50,8 @@ enum fw_ri_wr_opcode { FW_RI_BYPASS = 0xd, FW_RI_RECEIVE = 0xe, - FW_RI_SGE_EC_CR_RETURN = 0xf + FW_RI_SGE_EC_CR_RETURN = 0xf, + FW_RI_WRITE_IMMEDIATE = FW_RI_RDMA_INIT }; enum fw_ri_wr_flags { @@ -59,7 +60,8 @@ enum fw_ri_wr_flags { FW_RI_SOLICITED_EVENT_FLAG = 0x04, FW_RI_READ_FENCE_FLAG = 0x08, FW_RI_LOCAL_FENCE_FLAG = 0x10, - FW_RI_RDMA_READ_INVALIDATE = 0x20 + FW_RI_RDMA_READ_INVALIDATE = 0x20, + FW_RI_RDMA_WRITE_WITH_IMMEDIATE = 0x40 }; enum fw_ri_mpa_attrs { @@ -546,7 +548,17 @@ struct fw_ri_rdma_write_wr { __u16 wrid; __u8 r1[3]; __u8 len16; - __be64 r2; + /* + * Use union for immediate data to be consistent with stack's 32 bit + * data and iWARP spec's 64 bit data. + */ + union { + struct { + u32 reserved; + __be32 imm_data32; + } ib_imm_data; + __be64 imm_data64; + } iw_imm_data; __be32 plen; __be32 stag_sink; __be64 to_sink; diff --git a/include/uapi/rdma/cxgb4-abi.h b/include/uapi/rdma/cxgb4-abi.h index d0b2d829471a..f85ec1a3f727 100644 --- a/include/uapi/rdma/cxgb4-abi.h +++ b/include/uapi/rdma/cxgb4-abi.h @@ -65,7 +65,8 @@ struct c4iw_create_cq_resp { }; enum { - C4IW_QPF_ONCHIP = (1 << 0) + C4IW_QPF_ONCHIP = (1 << 0), + C4IW_QPF_WRITE_W_IMM = (1 << 1) }; struct c4iw_create_qp_resp {