From patchwork Wed Dec 21 11:42:03 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Selvin Xavier X-Patchwork-Id: 9482955 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 8E3A760772 for ; Wed, 21 Dec 2016 11:43:59 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 6D70C283FC for ; Wed, 21 Dec 2016 11:43:59 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 61CE428402; Wed, 21 Dec 2016 11:43:59 +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=-6.8 required=2.0 tests=BAYES_00,DKIM_SIGNED, RCVD_IN_DNSWL_HI,T_DKIM_INVALID 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 BC55C283FC for ; Wed, 21 Dec 2016 11:43:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758185AbcLULnx (ORCPT ); Wed, 21 Dec 2016 06:43:53 -0500 Received: from mail-wm0-f54.google.com ([74.125.82.54]:35212 "EHLO mail-wm0-f54.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932591AbcLULnq (ORCPT ); Wed, 21 Dec 2016 06:43:46 -0500 Received: by mail-wm0-f54.google.com with SMTP id a197so145846572wmd.0 for ; Wed, 21 Dec 2016 03:43:45 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=broadcom.com; s=google; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=cYzu/7ZKxAF9GrXYZQNIjWm/fYZhsiRYjjHYtqU4wbY=; b=Ot14TrV+iN+GXf36LI3vXFqRwAOjtMISqB8NmN92Zqfa/DELOm8LiYfnRv4Y6Y/jD+ sYN1uUA5NjWF99D+NHzoC3oo3YCmCLYmEjvxrrpXgVmxV0NQql9JtA+1dC0CLZCRHn0B /MxQQ6/oAc3wnV1h9YXE+CMAt1GIXgexMC8KM= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=cYzu/7ZKxAF9GrXYZQNIjWm/fYZhsiRYjjHYtqU4wbY=; b=NL/kYJ/35L6zqxMJCEiDRAGo+lMoH+EhOt6wPDCoQOk29iUgOcGWM3JPoUfU41Of2M 6oCjs7KokQ+IFuJPlHl7PLNS4FY2xWAbavx39uhlkIy6bcr4J0iQdPgHWOztdYXoOqj3 asq6G0W6CwWvucYh+k8eyXGBFVVJWajKxVEnMcbM8i6a+vXNT6aFaySBicNPp8M3JBOx F+newZGTka9eK7WqICeoRZfLYLrAlg1fo6gzymasLy4xwAhM53jdVsT/ONvyqce3HMkf h8HCjcrSrI7tP5C5+K/J8O2PEIRH3MKylKvzPIsBkIWs+tuskuaEdIqs7TpwuUe0dTKM 3swQ== X-Gm-Message-State: AIkVDXLhn+rfKIb7Jscjy5WCx9sMewX9uvAmTZW0aSAhcIsllpBUMdkwoL9uKtGWXrbwA/52 X-Received: by 10.28.140.136 with SMTP id o130mr6716014wmd.76.1482320624124; Wed, 21 Dec 2016 03:43:44 -0800 (PST) Received: from dhcp-10-192-206-197.iig.avagotech.net ([192.19.239.250]) by smtp.gmail.com with ESMTPSA id t194sm2997257wmd.1.2016.12.21.03.43.39 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Wed, 21 Dec 2016 03:43:43 -0800 (PST) From: Selvin Xavier To: dledford@redhat.com, linux-rdma@vger.kernel.org Cc: netdev@vger.kernel.org, michael.chan@broadcom.com, Selvin Xavier , Eddie Wai , Devesh Sharma , Somnath Kotur , Sriharsha Basavapatna Subject: [PATCH for bnxt_re V4 14/21] RDMA/bnxt_re: Support post_send verb Date: Wed, 21 Dec 2016 03:42:03 -0800 Message-Id: <1482320530-5344-15-git-send-email-selvin.xavier@broadcom.com> X-Mailer: git-send-email 2.5.5 In-Reply-To: <1482320530-5344-1-git-send-email-selvin.xavier@broadcom.com> References: <1482320530-5344-1-git-send-email-selvin.xavier@broadcom.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 Enables the ib_post_send fastpath verb for posting Send work requests on QPs. v2: Fixed some sparse warnings v3: Fixes endianness related warnings reported by sparse. Changes some of the macros to inline functions. Signed-off-by: Eddie Wai Signed-off-by: Devesh Sharma Signed-off-by: Somnath Kotur Signed-off-by: Sriharsha Basavapatna --- drivers/infiniband/hw/bnxt_re/bnxt_re.h | 4 + drivers/infiniband/hw/bnxt_re/ib_verbs.c | 544 ++++++++++++++++++++++++++++++- drivers/infiniband/hw/bnxt_re/ib_verbs.h | 2 + drivers/infiniband/hw/bnxt_re/main.c | 1 + drivers/infiniband/hw/bnxt_re/qplib_fp.c | 268 +++++++++++++++ drivers/infiniband/hw/bnxt_re/qplib_fp.h | 8 + 6 files changed, 824 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h index f53ebd9..6281e23 100644 --- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h +++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h @@ -131,6 +131,10 @@ struct bnxt_re_dev { #define to_bnxt_re_dev(ptr, member) \ container_of((ptr), struct bnxt_re_dev, member) +#define BNXT_RE_ROCE_V1_PACKET 0 +#define BNXT_RE_ROCEV2_IPV4_PACKET 2 +#define BNXT_RE_ROCEV2_IPV6_PACKET 3 + static inline struct device *rdev_to_dev(struct bnxt_re_dev *rdev) { if (rdev) diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 2189cad..e659490 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -60,6 +60,20 @@ #include "ib_verbs.h" #include +static int bnxt_re_build_sgl(struct ib_sge *ib_sg_list, + struct bnxt_qplib_sge *sg_list, int num) +{ + int i, total = 0; + + for (i = 0; i < num; i++) { + sg_list[i].addr = ib_sg_list[i].addr; + sg_list[i].lkey = ib_sg_list[i].lkey; + sg_list[i].size = ib_sg_list[i].length; + total += sg_list[i].size; + } + return total; +} + /* Device */ struct net_device *bnxt_re_get_netdev(struct ib_device *ibdev, u8 port_num) { @@ -698,8 +712,6 @@ static u8 __from_ib_qp_type(enum ib_qp_type type) return CMDQ_CREATE_QP_TYPE_RC; case IB_QPT_UD: return CMDQ_CREATE_QP_TYPE_UD; - case IB_QPT_RAW_ETHERTYPE: - return CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE; default: return IB_QPT_MAX; } @@ -873,7 +885,6 @@ struct ib_qp *bnxt_re_create_qp(struct ib_pd *ib_pd, struct bnxt_re_dev *rdev = pd->rdev; struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr; struct bnxt_re_qp *qp; - struct bnxt_re_srq *srq; struct bnxt_re_cq *cq; int rc, entries; @@ -1445,6 +1456,533 @@ int bnxt_re_query_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr, return 0; } +/* Routine for sending QP1 packets for RoCE V1 an V2 + */ +static int bnxt_re_build_qp1_send_v2(struct bnxt_re_qp *qp, + struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe, + int payload_size) +{ + struct ib_device *ibdev = &qp->rdev->ibdev; + struct bnxt_re_ah *ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah, + ib_ah); + struct bnxt_qplib_ah *qplib_ah = &ah->qplib_ah; + struct bnxt_qplib_sge sge; + union ib_gid sgid; + u8 nw_type; + u16 ether_type; + struct ib_gid_attr sgid_attr; + union ib_gid dgid; + bool is_eth = false; + bool is_vlan = false; + bool is_grh = false; + bool is_udp = false; + u8 ip_version = 0; + u16 vlan_id = 0xFFFF; + void *buf; + int i, rc = 0, size; + + memset(&qp->qp1_hdr, 0, sizeof(qp->qp1_hdr)); + + rc = ib_get_cached_gid(ibdev, 1, + qplib_ah->host_sgid_index, &sgid, + &sgid_attr); + if (rc) { + dev_err(rdev_to_dev(qp->rdev), + "Failed to query gid at index %d", + qplib_ah->host_sgid_index); + return rc; + } + if (sgid_attr.ndev) { + if (is_vlan_dev(sgid_attr.ndev)) + vlan_id = vlan_dev_vlan_id(sgid_attr.ndev); + dev_put(sgid_attr.ndev); + } + /* Get network header type for this GID */ + nw_type = ib_gid_to_network_type(sgid_attr.gid_type, &sgid); + switch (nw_type) { + case RDMA_NETWORK_IPV4: + nw_type = BNXT_RE_ROCEV2_IPV4_PACKET; + break; + case RDMA_NETWORK_IPV6: + nw_type = BNXT_RE_ROCEV2_IPV6_PACKET; + break; + default: + nw_type = BNXT_RE_ROCE_V1_PACKET; + break; + } + memcpy(&dgid.raw, &qplib_ah->dgid, 16); + is_udp = sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP; + if (is_udp) { + if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) { + ip_version = 4; + ether_type = ETH_P_IP; + } else { + ip_version = 6; + ether_type = ETH_P_IPV6; + } + is_grh = false; + } else { + ether_type = BNXT_QPLIB_ETHTYPE_ROCEV1; + is_grh = true; + } + + is_eth = true; + is_vlan = (vlan_id && (vlan_id < 0x1000)) ? true : false; + + ib_ud_header_init(payload_size, !is_eth, is_eth, is_vlan, is_grh, + ip_version, is_udp, 0, &qp->qp1_hdr); + + /* ETH */ + ether_addr_copy(qp->qp1_hdr.eth.dmac_h, ah->qplib_ah.dmac); + ether_addr_copy(qp->qp1_hdr.eth.smac_h, qp->qplib_qp.smac); + + /* For vlan, check the sgid for vlan existence */ + + if (!is_vlan) { + qp->qp1_hdr.eth.type = cpu_to_be16(ether_type); + } else { + qp->qp1_hdr.vlan.type = cpu_to_be16(ether_type); + qp->qp1_hdr.vlan.tag = cpu_to_be16(vlan_id); + } + + if (is_grh || (ip_version == 6)) { + memcpy(qp->qp1_hdr.grh.source_gid.raw, sgid.raw, sizeof(sgid)); + memcpy(qp->qp1_hdr.grh.destination_gid.raw, qplib_ah->dgid.data, + sizeof(sgid)); + qp->qp1_hdr.grh.hop_limit = qplib_ah->hop_limit; + } + + if (ip_version == 4) { + qp->qp1_hdr.ip4.tos = 0; + qp->qp1_hdr.ip4.id = 0; + qp->qp1_hdr.ip4.frag_off = htons(IP_DF); + qp->qp1_hdr.ip4.ttl = qplib_ah->hop_limit; + + memcpy(&qp->qp1_hdr.ip4.saddr, sgid.raw + 12, 4); + memcpy(&qp->qp1_hdr.ip4.daddr, qplib_ah->dgid.data + 12, 4); + qp->qp1_hdr.ip4.check = ib_ud_ip4_csum(&qp->qp1_hdr); + } + + if (is_udp) { + qp->qp1_hdr.udp.dport = htons(ROCE_V2_UDP_DPORT); + qp->qp1_hdr.udp.sport = htons(0x8CD1); + qp->qp1_hdr.udp.csum = 0; + } + + /* BTH */ + if (wr->opcode == IB_WR_SEND_WITH_IMM) { + qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE; + qp->qp1_hdr.immediate_present = 1; + } else { + qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY; + } + if (wr->send_flags & IB_SEND_SOLICITED) + qp->qp1_hdr.bth.solicited_event = 1; + /* pad_count */ + qp->qp1_hdr.bth.pad_count = (4 - payload_size) & 3; + + /* P_key for QP1 is for all members */ + qp->qp1_hdr.bth.pkey = cpu_to_be16(0xFFFF); + qp->qp1_hdr.bth.destination_qpn = IB_QP1; + qp->qp1_hdr.bth.ack_req = 0; + qp->send_psn++; + qp->send_psn &= BTH_PSN_MASK; + qp->qp1_hdr.bth.psn = cpu_to_be32(qp->send_psn); + /* DETH */ + /* Use the priviledged Q_Key for QP1 */ + qp->qp1_hdr.deth.qkey = cpu_to_be32(IB_QP1_QKEY); + qp->qp1_hdr.deth.source_qpn = IB_QP1; + + /* Pack the QP1 to the transmit buffer */ + buf = bnxt_qplib_get_qp1_sq_buf(&qp->qplib_qp, &sge); + if (buf) { + size = ib_ud_header_pack(&qp->qp1_hdr, buf); + for (i = wqe->num_sge; i; i--) { + wqe->sg_list[i].addr = wqe->sg_list[i - 1].addr; + wqe->sg_list[i].lkey = wqe->sg_list[i - 1].lkey; + wqe->sg_list[i].size = wqe->sg_list[i - 1].size; + } + + /* + * Max Header buf size for IPV6 RoCE V2 is 86, + * which is same as the QP1 SQ header buffer. + * Header buf size for IPV4 RoCE V2 can be 66. + * ETH(14) + VLAN(4)+ IP(20) + UDP (8) + BTH(20). + * Subtract 20 bytes from QP1 SQ header buf size + */ + if (is_udp && ip_version == 4) + sge.size -= 20; + /* + * Max Header buf size for RoCE V1 is 78. + * ETH(14) + VLAN(4) + GRH(40) + BTH(20). + * Subtract 8 bytes from QP1 SQ header buf size + */ + if (!is_udp) + sge.size -= 8; + + /* Subtract 4 bytes for non vlan packets */ + if (!is_vlan) + sge.size -= 4; + + wqe->sg_list[0].addr = sge.addr; + wqe->sg_list[0].lkey = sge.lkey; + wqe->sg_list[0].size = sge.size; + wqe->num_sge++; + + } else { + dev_err(rdev_to_dev(qp->rdev), "QP1 buffer is empty!"); + rc = -ENOMEM; + } + return rc; +} + +static int is_ud_qp(struct bnxt_re_qp *qp) +{ + return qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD; +} + +static int bnxt_re_build_send_wqe(struct bnxt_re_qp *qp, + struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + struct bnxt_re_ah *ah = NULL; + + if (is_ud_qp(qp)) { + ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah, ib_ah); + wqe->send.q_key = ud_wr(wr)->remote_qkey; + wqe->send.dst_qp = ud_wr(wr)->remote_qpn; + wqe->send.avid = ah->qplib_ah.id; + } + switch (wr->opcode) { + case IB_WR_SEND: + wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND; + break; + case IB_WR_SEND_WITH_IMM: + wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM; + wqe->send.imm_data = wr->ex.imm_data; + break; + case IB_WR_SEND_WITH_INV: + wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV; + wqe->send.inv_key = wr->ex.invalidate_rkey; + break; + default: + return -EINVAL; + } + if (wr->send_flags & IB_SEND_SIGNALED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; + if (wr->send_flags & IB_SEND_FENCE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->send_flags & IB_SEND_SOLICITED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT; + if (wr->send_flags & IB_SEND_INLINE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE; + + return 0; +} + +static int bnxt_re_build_rdma_wqe(struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + switch (wr->opcode) { + case IB_WR_RDMA_WRITE: + wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE; + break; + case IB_WR_RDMA_WRITE_WITH_IMM: + wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM; + wqe->rdma.imm_data = wr->ex.imm_data; + break; + case IB_WR_RDMA_READ: + wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_READ; + wqe->rdma.inv_key = wr->ex.invalidate_rkey; + break; + default: + return -EINVAL; + } + wqe->rdma.remote_va = rdma_wr(wr)->remote_addr; + wqe->rdma.r_key = rdma_wr(wr)->rkey; + if (wr->send_flags & IB_SEND_SIGNALED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; + if (wr->send_flags & IB_SEND_FENCE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->send_flags & IB_SEND_SOLICITED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT; + if (wr->send_flags & IB_SEND_INLINE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE; + + return 0; +} + +static int bnxt_re_build_atomic_wqe(struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + switch (wr->opcode) { + case IB_WR_ATOMIC_CMP_AND_SWP: + wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP; + wqe->atomic.swap_data = atomic_wr(wr)->swap; + break; + case IB_WR_ATOMIC_FETCH_AND_ADD: + wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD; + wqe->atomic.cmp_data = atomic_wr(wr)->compare_add; + break; + default: + return -EINVAL; + } + wqe->atomic.remote_va = atomic_wr(wr)->remote_addr; + wqe->atomic.r_key = atomic_wr(wr)->rkey; + if (wr->send_flags & IB_SEND_SIGNALED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; + if (wr->send_flags & IB_SEND_FENCE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->send_flags & IB_SEND_SOLICITED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT; + return 0; +} + +static int bnxt_re_build_inv_wqe(struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + wqe->type = BNXT_QPLIB_SWQE_TYPE_LOCAL_INV; + wqe->local_inv.inv_l_key = wr->ex.invalidate_rkey; + + if (wr->send_flags & IB_SEND_SIGNALED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; + if (wr->send_flags & IB_SEND_FENCE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->send_flags & IB_SEND_SOLICITED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT; + + return 0; +} + +static int bnxt_re_build_reg_wqe(struct ib_reg_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + struct bnxt_re_mr *mr = container_of(wr->mr, struct bnxt_re_mr, ib_mr); + struct bnxt_qplib_frpl *qplib_frpl = &mr->qplib_frpl; + int access = wr->access; + + wqe->frmr.pbl_ptr = (__le64 *)qplib_frpl->hwq.pbl_ptr[0]; + wqe->frmr.pbl_dma_ptr = qplib_frpl->hwq.pbl_dma_ptr[0]; + wqe->frmr.page_list = mr->pages; + wqe->frmr.page_list_len = mr->npages; + wqe->frmr.levels = qplib_frpl->hwq.level + 1; + wqe->type = BNXT_QPLIB_SWQE_TYPE_REG_MR; + + if (wr->wr.send_flags & IB_SEND_FENCE) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE; + if (wr->wr.send_flags & IB_SEND_SIGNALED) + wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP; + + if (access & IB_ACCESS_LOCAL_WRITE) + wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE; + if (access & IB_ACCESS_REMOTE_READ) + wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_READ; + if (access & IB_ACCESS_REMOTE_WRITE) + wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_WRITE; + if (access & IB_ACCESS_REMOTE_ATOMIC) + wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_ATOMIC; + if (access & IB_ACCESS_MW_BIND) + wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_WINDOW_BIND; + + wqe->frmr.l_key = wr->key; + wqe->frmr.length = wr->mr->length; + wqe->frmr.pbl_pg_sz_log = (wr->mr->page_size >> PAGE_SHIFT_4K) - 1; + wqe->frmr.va = wr->mr->iova; + return 0; +} + +static int bnxt_re_copy_inline_data(struct bnxt_re_dev *rdev, + struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + /* Copy the inline data to the data field */ + u8 *in_data; + u32 i, sge_len; + void *sge_addr; + + in_data = wqe->inline_data; + for (i = 0; i < wr->num_sge; i++) { + sge_addr = (void *)(unsigned long) + wr->sg_list[i].addr; + sge_len = wr->sg_list[i].length; + + if ((sge_len + wqe->inline_len) > + BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) { + dev_err(rdev_to_dev(rdev), + "Inline data size requested > supported value"); + return -EINVAL; + } + sge_len = wr->sg_list[i].length; + + memcpy(in_data, sge_addr, sge_len); + in_data += wr->sg_list[i].length; + wqe->inline_len += wr->sg_list[i].length; + } + return wqe->inline_len; +} + +static int bnxt_re_copy_wr_payload(struct bnxt_re_dev *rdev, + struct ib_send_wr *wr, + struct bnxt_qplib_swqe *wqe) +{ + int payload_sz = 0; + + if (wr->send_flags & IB_SEND_INLINE) + payload_sz = bnxt_re_copy_inline_data(rdev, wr, wqe); + else + payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe->sg_list, + wqe->num_sge); + + return payload_sz; +} + +static int bnxt_re_post_send_shadow_qp(struct bnxt_re_dev *rdev, + struct bnxt_re_qp *qp, + struct ib_send_wr *wr) +{ + struct bnxt_qplib_swqe wqe; + int rc = 0, payload_sz = 0; + unsigned long flags; + + spin_lock_irqsave(&qp->sq_lock, flags); + memset(&wqe, 0, sizeof(wqe)); + while (wr) { + /* House keeping */ + memset(&wqe, 0, sizeof(wqe)); + + /* Common */ + wqe.num_sge = wr->num_sge; + if (wr->num_sge > qp->qplib_qp.sq.max_sge) { + dev_err(rdev_to_dev(rdev), + "Limit exceeded for Send SGEs"); + rc = -EINVAL; + goto bad; + } + + payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe); + if (payload_sz < 0) { + rc = -EINVAL; + goto bad; + } + wqe.wr_id = wr->wr_id; + + wqe.type = BNXT_QPLIB_SWQE_TYPE_SEND; + + rc = bnxt_re_build_send_wqe(qp, wr, &wqe); + if (!rc) + rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe); +bad: + if (rc) { + dev_err(rdev_to_dev(rdev), + "Post send failed opcode = %#x rc = %d", + wr->opcode, rc); + break; + } + wr = wr->next; + } + bnxt_qplib_post_send_db(&qp->qplib_qp); + spin_unlock_irqrestore(&qp->sq_lock, flags); + return rc; +} + +int bnxt_re_post_send(struct ib_qp *ib_qp, struct ib_send_wr *wr, + struct ib_send_wr **bad_wr) +{ + struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp); + struct bnxt_qplib_swqe wqe; + int rc = 0, payload_sz = 0; + unsigned long flags; + + spin_lock_irqsave(&qp->sq_lock, flags); + while (wr) { + /* House keeping */ + memset(&wqe, 0, sizeof(wqe)); + + /* Common */ + wqe.num_sge = wr->num_sge; + if (wr->num_sge > qp->qplib_qp.sq.max_sge) { + dev_err(rdev_to_dev(qp->rdev), + "Limit exceeded for Send SGEs"); + rc = -EINVAL; + goto bad; + } + + payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe); + if (payload_sz < 0) { + rc = -EINVAL; + goto bad; + } + wqe.wr_id = wr->wr_id; + + switch (wr->opcode) { + case IB_WR_SEND: + case IB_WR_SEND_WITH_IMM: + if (ib_qp->qp_type == IB_QPT_GSI) { + rc = bnxt_re_build_qp1_send_v2(qp, wr, &wqe, + payload_sz); + if (rc) + goto bad; + wqe.rawqp1.lflags |= + SQ_SEND_RAWETH_QP1_LFLAGS_ROCE_CRC; + } + switch (wr->send_flags) { + case IB_SEND_IP_CSUM: + wqe.rawqp1.lflags |= + SQ_SEND_RAWETH_QP1_LFLAGS_IP_CHKSUM; + break; + default: + break; + } + /* Fall thru to build the wqe */ + case IB_WR_SEND_WITH_INV: + rc = bnxt_re_build_send_wqe(qp, wr, &wqe); + break; + case IB_WR_RDMA_WRITE: + case IB_WR_RDMA_WRITE_WITH_IMM: + case IB_WR_RDMA_READ: + rc = bnxt_re_build_rdma_wqe(wr, &wqe); + break; + case IB_WR_ATOMIC_CMP_AND_SWP: + case IB_WR_ATOMIC_FETCH_AND_ADD: + rc = bnxt_re_build_atomic_wqe(wr, &wqe); + break; + case IB_WR_RDMA_READ_WITH_INV: + dev_err(rdev_to_dev(qp->rdev), + "RDMA Read with Invalidate is not supported"); + rc = -EINVAL; + goto bad; + case IB_WR_LOCAL_INV: + rc = bnxt_re_build_inv_wqe(wr, &wqe); + break; + case IB_WR_REG_MR: + rc = bnxt_re_build_reg_wqe(reg_wr(wr), &wqe); + break; + default: + /* Unsupported WRs */ + dev_err(rdev_to_dev(qp->rdev), + "WR (%#x) is not supported", wr->opcode); + rc = -EINVAL; + goto bad; + } + if (!rc) + rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe); +bad: + if (rc) { + dev_err(rdev_to_dev(qp->rdev), + "post_send failed op:%#x qps = %#x rc = %d\n", + wr->opcode, qp->qplib_qp.state, rc); + *bad_wr = wr; + break; + } + wr = wr->next; + } + bnxt_qplib_post_send_db(&qp->qplib_qp); + spin_unlock_irqrestore(&qp->sq_lock, flags); + + return rc; +} + /* Completion Queues */ int bnxt_re_destroy_cq(struct ib_cq *ib_cq) { diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h index 75ee88a..becdcdc 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h @@ -162,6 +162,8 @@ int bnxt_re_modify_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr, int bnxt_re_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr, int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); int bnxt_re_destroy_qp(struct ib_qp *qp); +int bnxt_re_post_send(struct ib_qp *qp, struct ib_send_wr *send_wr, + struct ib_send_wr **bad_send_wr); struct ib_cq *bnxt_re_create_cq(struct ib_device *ibdev, const struct ib_cq_init_attr *attr, struct ib_ucontext *context, diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c index d68f08d..dbf402e 100644 --- a/drivers/infiniband/hw/bnxt_re/main.c +++ b/drivers/infiniband/hw/bnxt_re/main.c @@ -462,6 +462,7 @@ static int bnxt_re_register_ib(struct bnxt_re_dev *rdev) ibdev->query_qp = bnxt_re_query_qp; ibdev->destroy_qp = bnxt_re_destroy_qp; + ibdev->post_send = bnxt_re_post_send; ibdev->create_cq = bnxt_re_create_cq; ibdev->destroy_cq = bnxt_re_destroy_cq; ibdev->req_notify_cq = bnxt_re_req_notify_cq; diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c index f1f9095..edf52b9 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c @@ -1078,6 +1078,274 @@ int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res, return 0; } +void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp, + struct bnxt_qplib_sge *sge) +{ + struct bnxt_qplib_q *sq = &qp->sq; + u32 sw_prod; + + memset(sge, 0, sizeof(*sge)); + + if (qp->sq_hdr_buf) { + sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq); + sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map + + sw_prod * qp->sq_hdr_buf_size); + sge->lkey = 0xFFFFFFFF; + sge->size = qp->sq_hdr_buf_size; + return qp->sq_hdr_buf + sw_prod * sge->size; + } + return NULL; +} + +void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp) +{ + struct bnxt_qplib_q *sq = &qp->sq; + struct dbr_dbr db_msg = { 0 }; + u32 sw_prod; + + sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq); + + db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) & + DBR_DBR_INDEX_MASK); + db_msg.type_xid = + cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) | + DBR_DBR_TYPE_SQ); + /* Flush all the WQE writes to HW */ + wmb(); + __iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64)); +} + +int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp, + struct bnxt_qplib_swqe *wqe) +{ + struct bnxt_qplib_q *sq = &qp->sq; + struct bnxt_qplib_swq *swq; + struct sq_send *hw_sq_send_hdr, **hw_sq_send_ptr; + struct sq_sge *hw_sge; + u32 sw_prod; + u8 wqe_size16; + int i, rc = 0, data_len = 0, pkt_num = 0; + __le32 temp32; + + if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS) { + rc = -EINVAL; + goto done; + } + if (HWQ_CMP((sq->hwq.prod + 1), &sq->hwq) == + HWQ_CMP(sq->hwq.cons, &sq->hwq)) { + rc = -ENOMEM; + goto done; + } + sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq); + swq = &sq->swq[sw_prod]; + swq->wr_id = wqe->wr_id; + swq->type = wqe->type; + swq->flags = wqe->flags; + if (qp->sig_type) + swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP; + swq->start_psn = sq->psn & BTH_PSN_MASK; + + hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr; + hw_sq_send_hdr = &hw_sq_send_ptr[get_sqe_pg(sw_prod)] + [get_sqe_idx(sw_prod)]; + + memset(hw_sq_send_hdr, 0, BNXT_QPLIB_MAX_SQE_ENTRY_SIZE); + + if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) { + /* Copy the inline data */ + if (wqe->inline_len > BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) { + dev_warn(&sq->hwq.pdev->dev, + "QPLIB: Inline data length > 96 detected"); + data_len = BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH; + } else { + data_len = wqe->inline_len; + } + memcpy(hw_sq_send_hdr->data, wqe->inline_data, data_len); + wqe_size16 = (data_len + 15) >> 4; + } else { + for (i = 0, hw_sge = (struct sq_sge *)hw_sq_send_hdr->data; + i < wqe->num_sge; i++, hw_sge++) { + hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr); + hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey); + hw_sge->size = cpu_to_le32(wqe->sg_list[i].size); + data_len += wqe->sg_list[i].size; + } + /* Each SGE entry = 1 WQE size16 */ + wqe_size16 = wqe->num_sge; + } + + /* Specifics */ + switch (wqe->type) { + case BNXT_QPLIB_SWQE_TYPE_SEND: + if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) { + /* Assemble info for Raw Ethertype QPs */ + struct sq_send_raweth_qp1 *sqe = + (struct sq_send_raweth_qp1 *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->wqe_size = wqe_size16 + + ((offsetof(typeof(*sqe), data) + 15) >> 4); + sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action); + sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags); + sqe->length = cpu_to_le32(data_len); + sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta & + SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) << + SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT); + + break; + } + /* else, just fall thru */ + case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM: + case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV: + { + struct sq_send *sqe = (struct sq_send *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->wqe_size = wqe_size16 + + ((offsetof(typeof(*sqe), data) + 15) >> 4); + sqe->inv_key_or_imm_data = cpu_to_le32( + wqe->send.inv_key); + if (qp->type == CMDQ_CREATE_QP_TYPE_UD) { + sqe->q_key = cpu_to_le32(wqe->send.q_key); + sqe->dst_qp = cpu_to_le32( + wqe->send.dst_qp & SQ_SEND_DST_QP_MASK); + sqe->length = cpu_to_le32(data_len); + sqe->avid = cpu_to_le32(wqe->send.avid & + SQ_SEND_AVID_MASK); + sq->psn = (sq->psn + 1) & BTH_PSN_MASK; + } else { + sqe->length = cpu_to_le32(data_len); + sqe->dst_qp = 0; + sqe->avid = 0; + if (qp->mtu) + pkt_num = (data_len + qp->mtu - 1) / qp->mtu; + if (!pkt_num) + pkt_num = 1; + sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK; + } + break; + } + case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE: + case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM: + case BNXT_QPLIB_SWQE_TYPE_RDMA_READ: + { + struct sq_rdma *sqe = (struct sq_rdma *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->wqe_size = wqe_size16 + + ((offsetof(typeof(*sqe), data) + 15) >> 4); + sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key); + sqe->length = cpu_to_le32((u32)data_len); + sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va); + sqe->remote_key = cpu_to_le32(wqe->rdma.r_key); + if (qp->mtu) + pkt_num = (data_len + qp->mtu - 1) / qp->mtu; + if (!pkt_num) + pkt_num = 1; + sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK; + break; + } + case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP: + case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD: + { + struct sq_atomic *sqe = (struct sq_atomic *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->remote_key = cpu_to_le32(wqe->atomic.r_key); + sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va); + sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data); + sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data); + if (qp->mtu) + pkt_num = (data_len + qp->mtu - 1) / qp->mtu; + if (!pkt_num) + pkt_num = 1; + sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK; + break; + } + case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV: + { + struct sq_localinvalidate *sqe = + (struct sq_localinvalidate *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key); + + break; + } + case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR: + { + struct sq_fr_pmr *sqe = (struct sq_fr_pmr *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->access_cntl = wqe->frmr.access_cntl | + SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE; + sqe->zero_based_page_size_log = + (wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) << + SQ_FR_PMR_PAGE_SIZE_LOG_SFT | + (wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0); + sqe->l_key = cpu_to_le32(wqe->frmr.l_key); + temp32 = cpu_to_le32(wqe->frmr.length); + memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length)); + sqe->numlevels_pbl_page_size_log = + ((wqe->frmr.pbl_pg_sz_log << + SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) & + SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) | + ((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) & + SQ_FR_PMR_NUMLEVELS_MASK); + + for (i = 0; i < wqe->frmr.page_list_len; i++) + wqe->frmr.pbl_ptr[i] = cpu_to_le64( + wqe->frmr.page_list[i] | + PTU_PTE_VALID); + sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr); + sqe->va = cpu_to_le64(wqe->frmr.va); + + break; + } + case BNXT_QPLIB_SWQE_TYPE_BIND_MW: + { + struct sq_bind *sqe = (struct sq_bind *)hw_sq_send_hdr; + + sqe->wqe_type = wqe->type; + sqe->flags = wqe->flags; + sqe->access_cntl = wqe->bind.access_cntl; + sqe->mw_type_zero_based = wqe->bind.mw_type | + (wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0); + sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key); + sqe->l_key = cpu_to_le32(wqe->bind.r_key); + sqe->va = cpu_to_le64(wqe->bind.va); + temp32 = cpu_to_le32(wqe->bind.length); + memcpy(&sqe->length, &temp32, sizeof(wqe->bind.length)); + break; + } + default: + /* Bad wqe, return error */ + rc = -EINVAL; + goto done; + } + swq->next_psn = sq->psn & BTH_PSN_MASK; + if (swq->psn_search) { + swq->psn_search->opcode_start_psn = cpu_to_le32( + ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) & + SQ_PSN_SEARCH_START_PSN_MASK) | + ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) & + SQ_PSN_SEARCH_OPCODE_MASK)); + swq->psn_search->flags_next_psn = cpu_to_le32( + ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) & + SQ_PSN_SEARCH_NEXT_PSN_MASK)); + } + + sq->hwq.prod++; +done: + return rc; +} + /* CQ */ /* Spinlock must be held */ diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.h b/drivers/infiniband/hw/bnxt_re/qplib_fp.h index e62c750..0a87920 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_fp.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.h @@ -38,6 +38,9 @@ #ifndef __BNXT_QPLIB_FP_H__ #define __BNXT_QPLIB_FP_H__ + +#define BNXT_QPLIB_ETHTYPE_ROCEV1 0x8915 + struct bnxt_qplib_sge { u64 addr; u32 lkey; @@ -415,6 +418,11 @@ int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp); int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp); int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp); int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp); +void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp, + struct bnxt_qplib_sge *sge); +void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp); +int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp, + struct bnxt_qplib_swqe *wqe); int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq); int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);