From patchwork Tue Sep 26 13:00:26 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Junxian Huang X-Patchwork-Id: 144878 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a59:cae8:0:b0:403:3b70:6f57 with SMTP id r8csp1896086vqu; Tue, 26 Sep 2023 06:07:00 -0700 (PDT) X-Google-Smtp-Source: AGHT+IHM9cM0jDmXbKJOKFmWdRsNBLweakLGPAyn0/xyZe4FCDoxUyCACJXQbOyS4UNcngHLWeAY X-Received: by 2002:a17:90a:e617:b0:263:829:2de with SMTP id j23-20020a17090ae61700b00263082902demr7129427pjy.2.1695733620449; Tue, 26 Sep 2023 06:07:00 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1695733620; cv=none; d=google.com; s=arc-20160816; b=QDD6cMQhXpEgXl4rNe8K96n+o0ML/Wh+xu+oygDYQ5EdwFbJyZXfFERJEYNsDbkZJ1 kwfp+IZ6c2kc+Qq678o2IJW42tgijqdLjLzTR/b2Ef6Vd3mYvNMDQdfQwmu6oAkB2iYJ O+VtKUOdOgPSWttGER/nbBaQsO4EMZMZJj0+6eSwVv7K9Ampbv7LCH6CkQKGO2MAf66Q /rwUSYCAlNa3M/R950vITk/n6TBh73PZ/ChdEUtiR56OC2TQq1J6TDT1gXJomRYAS45E O60MF9uTVrXcOhEceGQnTGOM0SKAyecYs/H57e83cpYBfCcFdP9eAJ2QfLqXPHx5RQq1 IGSA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=list-id:precedence:content-transfer-encoding:mime-version :message-id:date:subject:cc:to:from; bh=9o4k3/DKb2HKeTjx/e5tT5ZkYy73OS4tmocGUAQ4Gb0=; fh=KWvyQxL3Ff+3WPSMjlYu+P4255AmcMULAsFol6M1vNI=; b=zNnMVP33YjKVp2lxdSZCFe218Dj3Yn6rzZ4lpIu8vIbkCUfG6svBcl4PRm2Qtcc7Gv Gpo6ajzvlCoAJw+6GxQmHDNijfdN1yVXFQXWDsMAj1MEW+FmEiPGxGDGkN/UM+/ZxyQO WFOFv3oO+7NdsSUOjw0H8iNrKJLNCKJ2NkgPlfnYnrwlzyB47jlImU9ivuKvwsdVK3V4 5nc/D+9qt1nZQ1AibaR7bzNPvi9MACMz7liDMnJralu7FTh8kIdma3J2cDbqtJDOi8YT 8rZ08oRMuF8QlYweboyNlMnj+3kEMCzLiRBk+X8JBjHLzd/B6n+8Fgf1/AZV7TFTyM/B AO+Q== ARC-Authentication-Results: i=1; mx.google.com; spf=pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.35 as permitted sender) smtp.mailfrom=linux-kernel-owner@vger.kernel.org; dmarc=fail (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=hisilicon.com Received: from groat.vger.email (groat.vger.email. [23.128.96.35]) by mx.google.com with ESMTPS id md13-20020a17090b23cd00b0026d60f1627asi740329pjb.173.2023.09.26.06.07.00 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 26 Sep 2023 06:07:00 -0700 (PDT) Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.35 as permitted sender) client-ip=23.128.96.35; Authentication-Results: mx.google.com; spf=pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.35 as permitted sender) smtp.mailfrom=linux-kernel-owner@vger.kernel.org; dmarc=fail (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=hisilicon.com Received: from out1.vger.email (depot.vger.email [IPv6:2620:137:e000::3:0]) by groat.vger.email (Postfix) with ESMTP id 0CBC3830F542; Tue, 26 Sep 2023 06:03:55 -0700 (PDT) X-Virus-Status: Clean X-Virus-Scanned: clamav-milter 0.103.10 at groat.vger.email Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S234717AbjIZNDl (ORCPT + 28 others); Tue, 26 Sep 2023 09:03:41 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:52656 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232437AbjIZNDk (ORCPT ); Tue, 26 Sep 2023 09:03:40 -0400 Received: from szxga02-in.huawei.com (szxga02-in.huawei.com [45.249.212.188]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 254E9F3; Tue, 26 Sep 2023 06:03:33 -0700 (PDT) Received: from kwepemi500006.china.huawei.com (unknown [172.30.72.54]) by szxga02-in.huawei.com (SkyGuard) with ESMTP id 4Rw0GH5x8bzVkgK; Tue, 26 Sep 2023 21:00:23 +0800 (CST) Received: from localhost.localdomain (10.67.165.2) by kwepemi500006.china.huawei.com (7.221.188.68) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.31; Tue, 26 Sep 2023 21:03:30 +0800 From: Junxian Huang To: , CC: , , , Subject: [PATCH v2 for-next] RDMA/hns: Support SRQ record doorbell Date: Tue, 26 Sep 2023 21:00:26 +0800 Message-ID: <20230926130026.583088-1-huangjunxian6@hisilicon.com> X-Mailer: git-send-email 2.30.0 MIME-Version: 1.0 X-Originating-IP: [10.67.165.2] X-ClientProxiedBy: dggems701-chm.china.huawei.com (10.3.19.178) To kwepemi500006.china.huawei.com (7.221.188.68) X-CFilter-Loop: Reflected X-Spam-Status: No, score=-0.8 required=5.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS autolearn=unavailable autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on groat.vger.email Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org X-Greylist: Sender passed SPF test, not delayed by milter-greylist-4.6.4 (groat.vger.email [0.0.0.0]); Tue, 26 Sep 2023 06:03:55 -0700 (PDT) X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1777526128251158952 X-GMAIL-MSGID: 1778105576995968913 From: Yangyang Li Compared with normal doorbell, using record doorbell can shorten the process of ringing the doorbell and reduce the latency. Add a flag HNS_ROCE_CAP_FLAG_SRQ_RECORD_DB to allow FW to enable/disable SRQ record doorbell. If the flag above is set, allocate the dma buffer for SRQ record doorbell and write the buffer address into SRQC during SRQ creation. For userspace SRQ, add a flag HNS_ROCE_RSP_SRQ_CAP_RECORD_DB to notify userspace whether the SRQ record doorbell is enabled. Signed-off-by: Yangyang Li Signed-off-by: Junxian Huang --- drivers/infiniband/hw/hns/hns_roce_device.h | 3 + drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 31 +++++--- drivers/infiniband/hw/hns/hns_roce_srq.c | 85 ++++++++++++++++++++- include/uapi/rdma/hns-abi.h | 12 ++- 4 files changed, 119 insertions(+), 12 deletions(-) diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h index 2059d90f8f78..1627f3b0ef28 100644 --- a/drivers/infiniband/hw/hns/hns_roce_device.h +++ b/drivers/infiniband/hw/hns/hns_roce_device.h @@ -146,6 +146,7 @@ enum { HNS_ROCE_CAP_FLAG_SDI_MODE = BIT(14), HNS_ROCE_CAP_FLAG_STASH = BIT(17), HNS_ROCE_CAP_FLAG_CQE_INLINE = BIT(19), + HNS_ROCE_CAP_FLAG_SRQ_RECORD_DB = BIT(22), }; #define HNS_ROCE_DB_TYPE_COUNT 2 @@ -453,6 +454,8 @@ struct hns_roce_srq { spinlock_t lock; struct mutex mutex; void (*event)(struct hns_roce_srq *srq, enum hns_roce_event event); + struct hns_roce_db rdb; + u32 cap_flags; }; struct hns_roce_uar_table { diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c index b48f2c7b2fb0..a17bbf29b64b 100644 --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c @@ -941,20 +941,23 @@ static void fill_wqe_idx(struct hns_roce_srq *srq, unsigned int wqe_idx) idx_que->head++; } -static void update_srq_db(struct hns_roce_v2_db *db, struct hns_roce_srq *srq) +static void update_srq_db(struct hns_roce_srq *srq) { - hr_reg_write(db, DB_TAG, srq->srqn); - hr_reg_write(db, DB_CMD, HNS_ROCE_V2_SRQ_DB); - hr_reg_write(db, DB_PI, srq->idx_que.head); + struct hns_roce_dev *hr_dev = to_hr_dev(srq->ibsrq.device); + struct hns_roce_v2_db db; + + hr_reg_write(&db, DB_TAG, srq->srqn); + hr_reg_write(&db, DB_CMD, HNS_ROCE_V2_SRQ_DB); + hr_reg_write(&db, DB_PI, srq->idx_que.head); + + hns_roce_write64(hr_dev, (__le32 *)&db, srq->db_reg); } static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr, const struct ib_recv_wr **bad_wr) { - struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device); struct hns_roce_srq *srq = to_hr_srq(ibsrq); - struct hns_roce_v2_db srq_db; unsigned long flags; int ret = 0; u32 max_sge; @@ -985,9 +988,11 @@ static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq, } if (likely(nreq)) { - update_srq_db(&srq_db, srq); - - hns_roce_write64(hr_dev, (__le32 *)&srq_db, srq->db_reg); + if (srq->cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB) + *srq->rdb.db_record = srq->idx_que.head & + V2_DB_PRODUCER_IDX_M; + else + update_srq_db(srq); } spin_unlock_irqrestore(&srq->lock, flags); @@ -5630,6 +5635,14 @@ static int hns_roce_v2_write_srqc(struct hns_roce_srq *srq, void *mb_buf) hr_reg_write(ctx, SRQC_WQE_BUF_PG_SZ, to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.buf_pg_shift)); + if (srq->cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB) { + hr_reg_enable(ctx, SRQC_DB_RECORD_EN); + hr_reg_write(ctx, SRQC_DB_RECORD_ADDR_L, + lower_32_bits(srq->rdb.dma) >> 1); + hr_reg_write(ctx, SRQC_DB_RECORD_ADDR_H, + upper_32_bits(srq->rdb.dma)); + } + return hns_roce_v2_write_srqc_index_queue(srq, ctx); } diff --git a/drivers/infiniband/hw/hns/hns_roce_srq.c b/drivers/infiniband/hw/hns/hns_roce_srq.c index 8dae98f827eb..4e2d1c8e164a 100644 --- a/drivers/infiniband/hw/hns/hns_roce_srq.c +++ b/drivers/infiniband/hw/hns/hns_roce_srq.c @@ -5,6 +5,7 @@ #include #include +#include #include "hns_roce_device.h" #include "hns_roce_cmd.h" #include "hns_roce_hem.h" @@ -387,6 +388,79 @@ static void free_srq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq) free_srq_idx(hr_dev, srq); } +static int get_srq_ucmd(struct hns_roce_srq *srq, struct ib_udata *udata, + struct hns_roce_ib_create_srq *ucmd) +{ + struct ib_device *ibdev = srq->ibsrq.device; + int ret; + + ret = ib_copy_from_udata(ucmd, udata, min(udata->inlen, sizeof(*ucmd))); + if (ret) { + ibdev_err(ibdev, "failed to copy SRQ udata, ret = %d.\n", ret); + return ret; + } + + return 0; +} + +static void free_srq_db(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq, + struct ib_udata *udata) +{ + struct hns_roce_ucontext *uctx; + + if (!(srq->cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB)) + return; + + srq->cap_flags &= ~HNS_ROCE_SRQ_CAP_RECORD_DB; + if (udata) { + uctx = rdma_udata_to_drv_context(udata, + struct hns_roce_ucontext, + ibucontext); + hns_roce_db_unmap_user(uctx, &srq->rdb); + } else { + hns_roce_free_db(hr_dev, &srq->rdb); + } +} + +static int alloc_srq_db(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq, + struct ib_udata *udata, + struct hns_roce_ib_create_srq_resp *resp) +{ + struct hns_roce_ib_create_srq ucmd = {}; + struct hns_roce_ucontext *uctx; + int ret; + + if (udata) { + ret = get_srq_ucmd(srq, udata, &ucmd); + if (ret) + return ret; + + if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ_RECORD_DB) && + (ucmd.req_cap_flags & HNS_ROCE_SRQ_CAP_RECORD_DB)) { + uctx = rdma_udata_to_drv_context(udata, + struct hns_roce_ucontext, ibucontext); + ret = hns_roce_db_map_user(uctx, ucmd.db_addr, + &srq->rdb); + if (ret) + return ret; + + srq->cap_flags |= HNS_ROCE_RSP_SRQ_CAP_RECORD_DB; + } + } else { + if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ_RECORD_DB) { + ret = hns_roce_alloc_db(hr_dev, &srq->rdb, 1); + if (ret) + return ret; + + *srq->rdb.db_record = 0; + srq->cap_flags |= HNS_ROCE_RSP_SRQ_CAP_RECORD_DB; + } + srq->db_reg = hr_dev->reg_base + SRQ_DB_REG; + } + + return 0; +} + int hns_roce_create_srq(struct ib_srq *ib_srq, struct ib_srq_init_attr *init_attr, struct ib_udata *udata) @@ -407,15 +481,20 @@ int hns_roce_create_srq(struct ib_srq *ib_srq, if (ret) return ret; - ret = alloc_srqn(hr_dev, srq); + ret = alloc_srq_db(hr_dev, srq, udata, &resp); if (ret) goto err_srq_buf; + ret = alloc_srqn(hr_dev, srq); + if (ret) + goto err_srq_db; + ret = alloc_srqc(hr_dev, srq); if (ret) goto err_srqn; if (udata) { + resp.cap_flags = srq->cap_flags; resp.srqn = srq->srqn; if (ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)))) { @@ -424,7 +503,6 @@ int hns_roce_create_srq(struct ib_srq *ib_srq, } } - srq->db_reg = hr_dev->reg_base + SRQ_DB_REG; srq->event = hns_roce_ib_srq_event; refcount_set(&srq->refcount, 1); init_completion(&srq->free); @@ -435,6 +513,8 @@ int hns_roce_create_srq(struct ib_srq *ib_srq, free_srqc(hr_dev, srq); err_srqn: free_srqn(hr_dev, srq); +err_srq_db: + free_srq_db(hr_dev, srq, udata); err_srq_buf: free_srq_buf(hr_dev, srq); @@ -448,6 +528,7 @@ int hns_roce_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata) free_srqc(hr_dev, srq); free_srqn(hr_dev, srq); + free_srq_db(hr_dev, srq, udata); free_srq_buf(hr_dev, srq); return 0; } diff --git a/include/uapi/rdma/hns-abi.h b/include/uapi/rdma/hns-abi.h index 2e68a8b0c92c..ce0f37f83416 100644 --- a/include/uapi/rdma/hns-abi.h +++ b/include/uapi/rdma/hns-abi.h @@ -52,15 +52,25 @@ struct hns_roce_ib_create_cq_resp { __aligned_u64 cap_flags; }; +enum hns_roce_srq_cap_flags { + HNS_ROCE_SRQ_CAP_RECORD_DB = 1 << 0, +}; + +enum hns_roce_srq_cap_flags_resp { + HNS_ROCE_RSP_SRQ_CAP_RECORD_DB = 1 << 0, +}; + struct hns_roce_ib_create_srq { __aligned_u64 buf_addr; __aligned_u64 db_addr; __aligned_u64 que_addr; + __u32 req_cap_flags; /* Use enum hns_roce_srq_cap_flags */ + __u32 reserved; }; struct hns_roce_ib_create_srq_resp { __u32 srqn; - __u32 reserved; + __u32 cap_flags; /* Use enum hns_roce_srq_cap_flags */ }; struct hns_roce_ib_create_qp {