From patchwork Fri Apr 21 18:57:07 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Devesh Sharma X-Patchwork-Id: 9693535 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 43FAD6038D for ; Fri, 21 Apr 2017 18:57:40 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 371612863D for ; Fri, 21 Apr 2017 18:57:40 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 2C06B28660; Fri, 21 Apr 2017 18:57:40 +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=-6.3 required=2.0 tests=BAYES_00,DKIM_SIGNED, RCVD_IN_DNSWL_HI, RCVD_IN_SORBS_SPAM, 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 ABA842865C for ; Fri, 21 Apr 2017 18:57:38 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1424299AbdDUS5h (ORCPT ); Fri, 21 Apr 2017 14:57:37 -0400 Received: from mail-wm0-f52.google.com ([74.125.82.52]:37314 "EHLO mail-wm0-f52.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1424205AbdDUS5f (ORCPT ); Fri, 21 Apr 2017 14:57:35 -0400 Received: by mail-wm0-f52.google.com with SMTP id m123so24150349wma.0 for ; Fri, 21 Apr 2017 11:57:34 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=broadcom.com; s=google; h=from:to:subject:date:message-id:in-reply-to:references; bh=2tAfaX+VEnce9g8rjSx5oQ4nsLUAluS5D+r04mX4VH0=; b=UX3PRvG6sjEaZOZ7+1flY+rgKX20FB1OXFLJVh9AI9r5uUvmTpnk/nJAvRpyp/U5zb DzaAikF+I/VZzlMcYw0G08ngcmrQZ0vd+049kePSmQTDA0RdMJQoTHv7VuNJ4mR/mYlo jwXkducRPAHowi6mDYigDADL5LMgB3Uk3DQds= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references; bh=2tAfaX+VEnce9g8rjSx5oQ4nsLUAluS5D+r04mX4VH0=; b=OxF3xgEgcMdVP934fphlsU/Acy+Q61nLoXEz0cRa9Nw9jtkrAy1t8kUk6s0OEQlyeJ YWTvwrHBPKrQ64iedeOyHUlJTwb7IgNXbBpJXfG8Mfq6K8hdFgrCXIXwwC4XP6MPyepR hme8Qcv217EfpXjQWJGMD3Do6U/6bASu3gq2url2XiZ62qVPvSBKcLhXoyE6gcrmim2d a9QlKcIQOvHPzSHHf6+xG4jLZvyp91OGPyXcegumr8GAdNTtnRGaJcVezJBnXRN92gMR z4m2A0oiiU/O++5CEy9feJix5Ynz4NwhTM6Wr3YbbyyJBUbEgq7TxDOR7OUaPouLjHBr 9+ZA== X-Gm-Message-State: AN3rC/6PvUlBkCn5lK+TFnKUKm164v1ekcfdr2t8BDL7QrQ/4k01pmtO v1FMaZv323xq79YnPzM= X-Received: by 10.28.210.9 with SMTP id j9mr96805wmg.120.1492801052575; Fri, 21 Apr 2017 11:57:32 -0700 (PDT) Received: from neo00-el73.iig.avagotech.net ([192.19.239.250]) by smtp.gmail.com with ESMTPSA id e12sm12622588wrc.43.2017.04.21.11.57.30 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Fri, 21 Apr 2017 11:57:31 -0700 (PDT) From: Devesh Sharma To: linux-rdma@vger.kernel.org Subject: [RESEND rdma-core v4 3/8] libbnxt_re: Add support for CQ and QP management Date: Fri, 21 Apr 2017 14:57:07 -0400 Message-Id: <1492801032-17587-4-git-send-email-devesh.sharma@broadcom.com> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1492801032-17587-1-git-send-email-devesh.sharma@broadcom.com> References: <1492801032-17587-1-git-send-email-devesh.sharma@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 This patch adds support for completion queue creation and destruction following are the changes: - Added User/Kernel ABI to communicate CQ specific parameters. - Added a function in a new file to allocate Page-Aligned address space. - Added a function to free page-aligned address space. - Added function to create and destroy completion queue. - Add ABI to for QP creation and WQE/RQE format. - Add functions to allocate SQ, RQ and Search PSN address space. - Add functions to store/clean qp-handles in the form of a linear table. There is table maintained in every instance of ucontext. - CQ and QP contexts now hold a pointer to the DPI mapped during PD allocation. - Removed hard-coding of page size during mapping DB page. - Renamed a variable in PD code. - Add support for create-qp. - Add support for destroy-qp. - Add support for modify-qp. - Add support for query-qp. v3->v4 -- change ABI declarations to use __leXX typedef v1->v2 -- Sorted file list -- Removed inline functions from ahi.h Signed-off-by: Sriharsha Basavapatna Signed-off-by: Somnath Kotur Signed-off-by: Selvin Xavier Signed-off-by: Devesh Sharma --- providers/bnxt_re/CMakeLists.txt | 1 + providers/bnxt_re/bnxt_re-abi.h | 106 ++++++++++++++++++ providers/bnxt_re/main.c | 5 + providers/bnxt_re/main.h | 73 ++++++++++-- providers/bnxt_re/memory.c | 76 +++++++++++++ providers/bnxt_re/memory.h | 76 +++++++++++++ providers/bnxt_re/verbs.c | 236 ++++++++++++++++++++++++++++++++++++++- 7 files changed, 555 insertions(+), 18 deletions(-) create mode 100644 providers/bnxt_re/memory.c create mode 100644 providers/bnxt_re/memory.h diff --git a/providers/bnxt_re/CMakeLists.txt b/providers/bnxt_re/CMakeLists.txt index 45e609d..7ea5ed8 100644 --- a/providers/bnxt_re/CMakeLists.txt +++ b/providers/bnxt_re/CMakeLists.txt @@ -1,4 +1,5 @@ rdma_provider(bnxt_re main.c + memory.c verbs.c ) diff --git a/providers/bnxt_re/bnxt_re-abi.h b/providers/bnxt_re/bnxt_re-abi.h index 53645d5..60ef63e 100644 --- a/providers/bnxt_re/bnxt_re-abi.h +++ b/providers/bnxt_re/bnxt_re-abi.h @@ -47,6 +47,10 @@ struct bnxt_re_cntx_resp { struct ibv_get_context_resp resp; __u32 dev_id; __u32 max_qp; /* To allocate qp-table */ + __u32 pg_size; + __u32 cqe_size; + __u32 max_cqd; + __u32 rsvd; }; struct bnxt_re_pd_resp { @@ -60,4 +64,106 @@ struct bnxt_re_mr_resp { struct ibv_reg_mr_resp resp; }; +struct bnxt_re_cq_req { + struct ibv_create_cq cmd; + __u64 cq_va; + __u64 cq_handle; +}; + +struct bnxt_re_cq_resp { + struct ibv_create_cq_resp resp; + __u32 cqid; + __u32 tail; + __u32 phase; + __u32 rsvd; +}; + +struct bnxt_re_qp_req { + struct ibv_create_qp cmd; + __u64 qpsva; + __u64 qprva; + __u64 qp_handle; +}; + +struct bnxt_re_qp_resp { + struct ibv_create_qp_resp resp; + __u32 qpid; + __u32 rsvd; +}; + +struct bnxt_re_bsqe { + __le32 rsv_ws_fl_wt; + __le32 key_immd; +}; + +struct bnxt_re_psns { + __le32 opc_spsn; + __le32 flg_npsn; +}; + +struct bnxt_re_sge { + __le64 pa; + __le32 lkey; + __le32 length; +}; + +/* Cu+ max inline data */ +#define BNXT_RE_MAX_INLINE_SIZE 0x60 + +struct bnxt_re_send { + __le32 length; + __le32 qkey; + __le32 dst_qp; + __le32 avid; + __le64 rsvd; +}; + +struct bnxt_re_raw { + __le32 length; + __le32 rsvd1; + __le32 cfa_meta; + __le32 rsvd2; + __le64 rsvd3; +}; + +struct bnxt_re_rdma { + __le32 length; + __le32 rsvd1; + __le64 rva; + __le32 rkey; + __le32 rsvd2; +}; + +struct bnxt_re_atomic { + __le64 rva; + __le64 swp_dt; + __le64 cmp_dt; +}; + +struct bnxt_re_inval { + __le64 rsvd[3]; +}; + +struct bnxt_re_bind { + __le32 plkey; + __le32 lkey; + __le64 va; + __le64 len; /* only 40 bits are valid */ +}; + +struct bnxt_re_brqe { + __le32 rsv_ws_fl_wt; + __le32 rsvd; +}; + +struct bnxt_re_rqe { + __le64 rsvd[3]; +}; + +struct bnxt_re_srqe { + __le32 srq_tag; /* 20 bits are valid */ + __le32 rsvd1; + __le64 rsvd[2]; +}; + #endif diff --git a/providers/bnxt_re/main.c b/providers/bnxt_re/main.c index 4e12e87..34f75fd 100644 --- a/providers/bnxt_re/main.c +++ b/providers/bnxt_re/main.c @@ -110,8 +110,10 @@ static int bnxt_re_init_context(struct verbs_device *vdev, { struct ibv_get_context cmd; struct bnxt_re_cntx_resp resp; + struct bnxt_re_dev *dev; struct bnxt_re_context *cntx; + dev = to_bnxt_re_dev(&vdev->device); cntx = to_bnxt_re_context(ibvctx); memset(&resp, 0, sizeof(resp)); @@ -122,6 +124,9 @@ static int bnxt_re_init_context(struct verbs_device *vdev, cntx->dev_id = resp.dev_id; cntx->max_qp = resp.max_qp; + dev->pg_size = resp.pg_size; + dev->cqe_size = resp.cqe_size; + dev->max_cq_depth = resp.max_cqd; ibvctx->ops = bnxt_re_cntx_ops; return 0; diff --git a/providers/bnxt_re/main.h b/providers/bnxt_re/main.h index 2113446..1e3de75 100644 --- a/providers/bnxt_re/main.h +++ b/providers/bnxt_re/main.h @@ -48,6 +48,15 @@ #include #include "bnxt_re-abi.h" +#include "memory.h" + +#define DEV "bnxt_re : " + +struct bnxt_re_dpi { + __u32 dpindx; + __u64 *dbpage; + pthread_spinlock_t db_lock; +}; struct bnxt_re_pd { struct ibv_pd ibvpd; @@ -56,31 +65,48 @@ struct bnxt_re_pd { struct bnxt_re_cq { struct ibv_cq ibvcq; -}; - -struct bnxt_re_qp { - struct ibv_qp ibvqp; + uint32_t cqid; + struct bnxt_re_queue cqq; + struct bnxt_re_dpi *udpi; + uint32_t cqe_size; + uint8_t phase; }; struct bnxt_re_srq { struct ibv_srq ibvsrq; }; -struct bnxt_re_mr { - struct ibv_mr ibvmr; +struct bnxt_re_qp { + struct ibv_qp ibvqp; + struct bnxt_re_queue *sqq; + struct bnxt_re_psns *psns; /* start ptr. */ + struct bnxt_re_queue *rqq; + struct bnxt_re_srq *srq; + struct bnxt_re_cq *scq; + struct bnxt_re_cq *rcq; + struct bnxt_re_dpi *udpi; + uint64_t *swrid; + uint64_t *rwrid; + uint32_t qpid; + uint32_t tbl_indx; + uint16_t mtu; + uint16_t qpst; + uint8_t qptyp; + /* wrid? */ + /* irdord? */ }; -#define DEV "bnxt_re : " - -struct bnxt_re_dpi { - __u32 dpindx; - __u64 *dbpage; - pthread_spinlock_t db_lock; +struct bnxt_re_mr { + struct ibv_mr ibvmr; }; struct bnxt_re_dev { struct verbs_device vdev; uint8_t abi_version; + uint32_t pg_size; + + uint32_t cqe_size; + uint32_t max_cq_depth; }; struct bnxt_re_context { @@ -107,4 +133,27 @@ static inline struct bnxt_re_pd *to_bnxt_re_pd(struct ibv_pd *ibvpd) return container_of(ibvpd, struct bnxt_re_pd, ibvpd); } +static inline struct bnxt_re_cq *to_bnxt_re_cq(struct ibv_cq *ibvcq) +{ + return container_of(ibvcq, struct bnxt_re_cq, ibvcq); +} + +static inline struct bnxt_re_qp *to_bnxt_re_qp(struct ibv_qp *ibvqp) +{ + return container_of(ibvqp, struct bnxt_re_qp, ibvqp); +} + +static inline uint32_t bnxt_re_get_sqe_sz(void) +{ + return sizeof(struct bnxt_re_bsqe) + + sizeof(struct bnxt_re_send) + + BNXT_RE_MAX_INLINE_SIZE; +} + +static inline uint32_t bnxt_re_get_rqe_sz(void) +{ + return sizeof(struct bnxt_re_brqe) + + sizeof(struct bnxt_re_rqe) + + BNXT_RE_MAX_INLINE_SIZE; +} #endif diff --git a/providers/bnxt_re/memory.c b/providers/bnxt_re/memory.c new file mode 100644 index 0000000..67125e9 --- /dev/null +++ b/providers/bnxt_re/memory.c @@ -0,0 +1,76 @@ +/* + * Broadcom NetXtreme-E User Space RoCE driver + * + * Copyright (c) 2015-2017, Broadcom. All rights reserved. The term + * Broadcom refers to Broadcom Limited and/or its subsidiaries. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * BSD license below: + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' + * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, + * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR + * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, + * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE + * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN + * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * Description: Implements method to allocate page-aligned memory + * buffers. + */ + +#include +#include + +#include "main.h" + +int bnxt_re_alloc_aligned(struct bnxt_re_queue *que, uint32_t pg_size) +{ + int ret, bytes; + + bytes = (que->depth * que->stride); + que->bytes = get_aligned(bytes, pg_size); + que->va = mmap(NULL, que->bytes, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if (que->va == MAP_FAILED) { + que->bytes = 0; + return errno; + } + /* Touch pages before proceeding. */ + memset(que->va, 0, que->bytes); + + ret = ibv_dontfork_range(que->va, que->bytes); + if (ret) { + munmap(que->va, que->bytes); + que->bytes = 0; + } + + return ret; +} + +void bnxt_re_free_aligned(struct bnxt_re_queue *que) +{ + if (que->bytes) { + ibv_dofork_range(que->va, que->bytes); + munmap(que->va, que->bytes); + que->bytes = 0; + } +} diff --git a/providers/bnxt_re/memory.h b/providers/bnxt_re/memory.h new file mode 100644 index 0000000..ea29a24 --- /dev/null +++ b/providers/bnxt_re/memory.h @@ -0,0 +1,76 @@ +/* + * Broadcom NetXtreme-E User Space RoCE driver + * + * Copyright (c) 2015-2017, Broadcom. All rights reserved. The term + * Broadcom refers to Broadcom Limited and/or its subsidiaries. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * BSD license below: + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' + * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, + * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR + * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, + * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE + * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN + * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * Description: Implements data-struture to allocate page-aligned + * memory buffer. + */ + +#ifndef __MEMORY_H__ +#define __MEMORY_H__ + +#include + +struct bnxt_re_queue { + void *va; + uint32_t bytes; /* for munmap */ + uint32_t depth; /* no. of entries */ + uint32_t head; + uint32_t tail; + uint32_t stride; + pthread_spinlock_t qlock; +}; + +static inline unsigned long get_aligned(uint32_t size, uint32_t al_size) +{ + return (unsigned long)(size + al_size - 1) & ~(al_size - 1); +} + +static inline unsigned long roundup_pow_of_two(unsigned long val) +{ + unsigned long roundup = 1; + + if (val == 1) + return (roundup << 1); + + while (roundup < val) + roundup <<= 1; + + return roundup; +} + +int bnxt_re_alloc_aligned(struct bnxt_re_queue *que, uint32_t pg_size); +void bnxt_re_free_aligned(struct bnxt_re_queue *que); + +#endif diff --git a/providers/bnxt_re/verbs.c b/providers/bnxt_re/verbs.c index f67e37d..d5b081d 100644 --- a/providers/bnxt_re/verbs.c +++ b/providers/bnxt_re/verbs.c @@ -79,6 +79,7 @@ struct ibv_pd *bnxt_re_alloc_pd(struct ibv_context *ibvctx) struct ibv_alloc_pd cmd; struct bnxt_re_pd_resp resp; struct bnxt_re_context *cntx = to_bnxt_re_context(ibvctx); + struct bnxt_re_dev *dev = to_bnxt_re_dev(ibvctx->device); struct bnxt_re_pd *pd; uint64_t dbr; @@ -96,7 +97,7 @@ struct ibv_pd *bnxt_re_alloc_pd(struct ibv_context *ibvctx) /* Map DB page now. */ cntx->udpi.dpindx = resp.dpi; - cntx->udpi.dbpage = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED, + cntx->udpi.dbpage = mmap(NULL, dev->pg_size, PROT_WRITE, MAP_SHARED, ibvctx->cmd_fd, dbr); if (cntx->udpi.dbpage == MAP_FAILED) { (void)ibv_cmd_dealloc_pd(&pd->ibvpd); @@ -114,6 +115,7 @@ int bnxt_re_free_pd(struct ibv_pd *ibvpd) { struct bnxt_re_pd *pd = to_bnxt_re_pd(ibvpd); struct bnxt_re_context *cntx = to_bnxt_re_context(ibvpd->context); + struct bnxt_re_dev *dev = to_bnxt_re_dev(cntx->ibvctx.device); int status; status = ibv_cmd_dealloc_pd(ibvpd); @@ -122,7 +124,8 @@ int bnxt_re_free_pd(struct ibv_pd *ibvpd) pthread_spin_destroy(&cntx->udpi.db_lock); if (cntx->udpi.dbpage && (cntx->udpi.dbpage != MAP_FAILED)) - munmap(cntx->udpi.dbpage, 4096); + munmap(cntx->udpi.dbpage, dev->pg_size); + free(pd); return 0; @@ -164,6 +167,48 @@ int bnxt_re_dereg_mr(struct ibv_mr *ibvmr) struct ibv_cq *bnxt_re_create_cq(struct ibv_context *ibvctx, int ncqe, struct ibv_comp_channel *channel, int vec) { + struct bnxt_re_cq *cq; + struct bnxt_re_cq_req cmd; + struct bnxt_re_cq_resp resp; + + struct bnxt_re_context *cntx = to_bnxt_re_context(ibvctx); + struct bnxt_re_dev *dev = to_bnxt_re_dev(ibvctx->device); + + if (ncqe > dev->max_cq_depth) + return NULL; + + cq = calloc(1, sizeof(*cq)); + if (!cq) + return NULL; + + cq->cqq.depth = roundup_pow_of_two(ncqe + 1); + if (cq->cqq.depth > dev->max_cq_depth + 1) + cq->cqq.depth = dev->max_cq_depth + 1; + cq->cqq.stride = dev->cqe_size; + if (bnxt_re_alloc_aligned(&cq->cqq, dev->pg_size)) + goto fail; + + pthread_spin_init(&cq->cqq.qlock, PTHREAD_PROCESS_PRIVATE); + + cmd.cq_va = (uintptr_t)cq->cqq.va; + cmd.cq_handle = (uintptr_t)cq; + + memset(&resp, 0, sizeof(resp)); + if (ibv_cmd_create_cq(ibvctx, ncqe, channel, vec, + &cq->ibvcq, &cmd.cmd, sizeof(cmd), + &resp.resp, sizeof(resp))) + goto cmdfail; + + cq->cqid = resp.cqid; + cq->phase = resp.phase; + cq->cqq.tail = resp.tail; + cq->udpi = &cntx->udpi; + + return &cq->ibvcq; +cmdfail: + bnxt_re_free_aligned(&cq->cqq); +fail: + free(cq); return NULL; } @@ -174,7 +219,17 @@ int bnxt_re_resize_cq(struct ibv_cq *ibvcq, int ncqe) int bnxt_re_destroy_cq(struct ibv_cq *ibvcq) { - return -ENOSYS; + int status; + struct bnxt_re_cq *cq = to_bnxt_re_cq(ibvcq); + + status = ibv_cmd_destroy_cq(ibvcq); + if (status) + return status; + + bnxt_re_free_aligned(&cq->cqq); + free(cq); + + return 0; } int bnxt_re_poll_cq(struct ibv_cq *ibvcq, int nwc, struct ibv_wc *wc) @@ -192,27 +247,196 @@ int bnxt_re_arm_cq(struct ibv_cq *ibvcq, int flags) return -ENOSYS; } +static int bnxt_re_check_qp_limits(struct ibv_qp_init_attr *attr) +{ + return 0; +} + +static void bnxt_re_free_queue_ptr(struct bnxt_re_qp *qp) +{ + if (qp->rqq) + free(qp->rqq); + if (qp->sqq) + free(qp->sqq); +} + +static int bnxt_re_alloc_queue_ptr(struct bnxt_re_qp *qp, + struct ibv_qp_init_attr *attr) +{ + qp->sqq = calloc(1, sizeof(struct bnxt_re_queue)); + if (!qp->sqq) + return -ENOMEM; + if (attr->srq) + qp->srq = NULL;/*TODO: to_bnxt_re_srq(attr->srq);*/ + else { + qp->rqq = calloc(1, sizeof(struct bnxt_re_queue)); + if (!qp->rqq) { + free(qp->sqq); + return -ENOMEM; + } + } + + return 0; +} + +static void bnxt_re_free_queues(struct bnxt_re_qp *qp) +{ + if (qp->rwrid) + free(qp->rwrid); + pthread_spin_destroy(&qp->rqq->qlock); + bnxt_re_free_aligned(qp->rqq); + + if (qp->swrid) + free(qp->swrid); + pthread_spin_destroy(&qp->sqq->qlock); + bnxt_re_free_aligned(qp->sqq); +} + +static int bnxt_re_alloc_queues(struct bnxt_re_qp *qp, + struct ibv_qp_init_attr *attr, + uint32_t pg_size) { + struct bnxt_re_queue *que; + uint32_t psn_depth; + int ret; + + if (attr->cap.max_send_wr) { + que = qp->sqq; + que->stride = bnxt_re_get_sqe_sz(); + que->depth = roundup_pow_of_two(attr->cap.max_send_wr); + /* psn_depth extra entries of size que->stride */ + psn_depth = (que->depth * sizeof(struct bnxt_re_psns)) / + que->stride; + que->depth += psn_depth; + ret = bnxt_re_alloc_aligned(qp->sqq, pg_size); + if (ret) + return ret; + /* exclude psns depth*/ + que->depth -= psn_depth; + /* start of spsn space sizeof(struct bnxt_re_psns) each. */ + qp->psns = (que->va + que->stride * que->depth); + pthread_spin_init(&que->qlock, PTHREAD_PROCESS_PRIVATE); + qp->swrid = calloc(que->depth, sizeof(uint64_t)); + if (!qp->swrid) { + ret = -ENOMEM; + goto fail; + } + } + + if (attr->cap.max_recv_wr && qp->rqq) { + que = qp->rqq; + que->stride = bnxt_re_get_rqe_sz(); + que->depth = roundup_pow_of_two(attr->cap.max_recv_wr); + ret = bnxt_re_alloc_aligned(qp->rqq, pg_size); + if (ret) + goto fail; + pthread_spin_init(&que->qlock, PTHREAD_PROCESS_PRIVATE); + qp->rwrid = calloc(que->depth, sizeof(uint64_t)); + if (!qp->rwrid) { + ret = -ENOMEM; + goto fail; + } + } + + return 0; + +fail: + bnxt_re_free_queues(qp); + return ret; +} + struct ibv_qp *bnxt_re_create_qp(struct ibv_pd *ibvpd, struct ibv_qp_init_attr *attr) { + struct bnxt_re_qp *qp; + struct bnxt_re_qp_req req; + struct bnxt_re_qp_resp resp; + + struct bnxt_re_context *cntx = to_bnxt_re_context(ibvpd->context); + struct bnxt_re_dev *dev = to_bnxt_re_dev(cntx->ibvctx.device); + + if (bnxt_re_check_qp_limits(attr)) + return NULL; + + qp = calloc(1, sizeof(*qp)); + if (!qp) + return NULL; + /* alloc queue pointers */ + if (bnxt_re_alloc_queue_ptr(qp, attr)) + goto fail; + /* alloc queues */ + if (bnxt_re_alloc_queues(qp, attr, dev->pg_size)) + goto failq; + /* Fill ibv_cmd */ + req.qpsva = (uintptr_t)qp->sqq->va; + req.qprva = qp->rqq ? (uintptr_t)qp->rqq->va : 0; + req.qp_handle = (uintptr_t)qp; + + if (ibv_cmd_create_qp(ibvpd, &qp->ibvqp, attr, &req.cmd, sizeof(req), + &resp.resp, sizeof(resp))) { + goto failcmd; + } + + qp->qpid = resp.qpid; + qp->qptyp = attr->qp_type; + qp->qpst = IBV_QPS_RESET; + qp->scq = to_bnxt_re_cq(attr->send_cq); + qp->rcq = to_bnxt_re_cq(attr->recv_cq); + qp->udpi = &cntx->udpi; + + return &qp->ibvqp; +failcmd: + bnxt_re_free_queues(qp); +failq: + bnxt_re_free_queue_ptr(qp); +fail: + free(qp); + return NULL; } int bnxt_re_modify_qp(struct ibv_qp *ibvqp, struct ibv_qp_attr *attr, int attr_mask) { - return -ENOSYS; + struct ibv_modify_qp cmd = {}; + struct bnxt_re_qp *qp = to_bnxt_re_qp(ibvqp); + int rc; + + rc = ibv_cmd_modify_qp(ibvqp, attr, attr_mask, &cmd, sizeof(cmd)); + if (!rc) + qp->qpst = ibvqp->state; + + return rc; } int bnxt_re_query_qp(struct ibv_qp *ibvqp, struct ibv_qp_attr *attr, int attr_mask, struct ibv_qp_init_attr *init_attr) { - return -ENOSYS; + struct ibv_query_qp cmd; + struct bnxt_re_qp *qp = to_bnxt_re_qp(ibvqp); + int rc; + + rc = ibv_cmd_query_qp(ibvqp, attr, attr_mask, init_attr, + &cmd, sizeof(cmd)); + if (!rc) + qp->qpst = ibvqp->state; + + return rc; } int bnxt_re_destroy_qp(struct ibv_qp *ibvqp) { - return -ENOSYS; + struct bnxt_re_qp *qp = to_bnxt_re_qp(ibvqp); + int status; + + status = ibv_cmd_destroy_qp(ibvqp); + if (status) + return status; + + bnxt_re_free_queues(qp); + bnxt_re_free_queue_ptr(qp); + free(qp); + + return 0; } int bnxt_re_post_send(struct ibv_qp *ibvqp, struct ibv_send_wr *wr,