From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id A936AA0C45; Fri, 15 Oct 2021 18:03:58 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id AA38E41230; Fri, 15 Oct 2021 18:03:53 +0200 (CEST) Received: from AZHDRRW-EX02.NVIDIA.COM (azhdrrw-ex02.nvidia.com [20.64.145.131]) by mails.dpdk.org (Postfix) with ESMTP id 592EB4067C for ; Fri, 15 Oct 2021 18:03:50 +0200 (CEST) Received: from NAM12-MW2-obe.outbound.protection.outlook.com (104.47.66.49) by mxs.oss.nvidia.com (10.13.234.37) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.858.15; Fri, 15 Oct 2021 09:03:49 -0700 ARC-Seal: i=1; a=rsa-sha256; s=arcselector9901; d=microsoft.com; cv=none; b=BZ0w8zfsIaxWw+ljrrpXoNL6AEVsvhODkGa7SmT+CoTp4SRJkJuDQ2sKD8dCt10qIDWtXItImj1hRr3DKeP/153+/kntrHB1THzUGkrR+9VTHADRfgMzaSRe0XtHyZyrQOvqhtGtu+Eg1ypVxqCyqnqsjEvZb2EoZZ7sAt3uSbZTjnLYy2f8+hVdB+nyK8IAkc8m/gc5Kr0v7vsmswz/DNhg8bGKjYIXjXX3NwmtDgTPessNTWFo0dcxPE39FNWObDJ1bYU8MOjeOY+c56x1ityq5DUPvMySLeuuf+sBZWrsgEilpHPhpmbtTKmy6lgEOVoG7sBNASk+/Tg1hIRx4A== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=microsoft.com; s=arcselector9901; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-AntiSpam-MessageData-ChunkCount:X-MS-Exchange-AntiSpam-MessageData-0:X-MS-Exchange-AntiSpam-MessageData-1; bh=0ZVl9WA6rVz9NCADmC3Uxqzh/ysKiH5eWJ3/klTNwNc=; b=g0geiiZi/kIW+bLWJrVRHOPCVbijSdGPm0TKF77mp56n3s8qW0Ciu4ivpuhUMuhIlkOUwlg3ctu1awHd2o5MiOVVQf2gomeSEoFgjuS+ln1OOUTtULI6cnb6Z7uS9rpqmlsat7z5+mnRkwwWBFiDFhDKYu8/jpOkqf7/qcLsNc++4uJ76iGPUbZhKXPvGMPG2CI0d8CxwOTa0wQL2E2fwLSrxOS4I3lZRlc2WWazKFKim2HXmCRZx/2AV+NSitIthbhVQCJmY0eHN8MPF6NN6egwEwf8nomlKne2cZBgtMheKJT8wtickFcWTRRkbmZWuoI/1BMN9pMP7srepYbYPA== ARC-Authentication-Results: i=1; mx.microsoft.com 1; spf=pass (sender ip is 216.228.112.34) smtp.rcpttodomain=intel.com smtp.mailfrom=nvidia.com; dmarc=pass (p=quarantine sp=quarantine pct=100) action=none header.from=nvidia.com; dkim=none (message not signed); arc=none DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=Nvidia.com; s=selector2; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-SenderADCheck; bh=0ZVl9WA6rVz9NCADmC3Uxqzh/ysKiH5eWJ3/klTNwNc=; b=F8iXAL5BFv6+soh8zIrxY3IDFPLk8EZwFd6Yv+2vrm5kGv23zHVXHkFhL77BwOEdQFVSEobCj4PadNYID26jN1N9xlqXCGnxwt71liNz9QkNpFww75Fqt8Ak0jpkE4bfQKPFv3EUXpbjtT7KRrCwbbPsBMFebkE5DftC1K99Tdi3oYBj18tWLMoLsGiB1SUU4dPVrTpYfdXj8qsuJ2zsWT8jZ9w0DYcYdutqxGW1NvGHDuaPsHQUAn45MmrHIV6L7EMWEticuctd6SJGGKwva906TMyFosvJjwEvGO8KONEIM1x7VUebU+wCTQS9RtebqtF1ZJVmCezjjBZRxHNVig== Received: from DM5PR10CA0003.namprd10.prod.outlook.com (2603:10b6:4:2::13) by DM6PR12MB4105.namprd12.prod.outlook.com (2603:10b6:5:217::18) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.20.4608.16; Fri, 15 Oct 2021 16:03:48 +0000 Received: from DM6NAM11FT033.eop-nam11.prod.protection.outlook.com (2603:10b6:4:2:cafe::d9) by DM5PR10CA0003.outlook.office365.com (2603:10b6:4:2::13) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.20.4608.14 via Frontend Transport; Fri, 15 Oct 2021 16:03:48 +0000 X-MS-Exchange-Authentication-Results: spf=pass (sender IP is 216.228.112.34) smtp.mailfrom=nvidia.com; intel.com; dkim=none (message not signed) header.d=none;intel.com; dmarc=pass action=none header.from=nvidia.com; Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates 216.228.112.34 as permitted sender) receiver=protection.outlook.com; client-ip=216.228.112.34; helo=mail.nvidia.com; Received: from mail.nvidia.com (216.228.112.34) by DM6NAM11FT033.mail.protection.outlook.com (10.13.172.221) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA384) id 15.20.4608.15 via Frontend Transport; Fri, 15 Oct 2021 16:03:47 +0000 Received: from nvidia.com (172.20.187.6) by HQMAIL107.nvidia.com (172.20.187.13) with Microsoft SMTP Server (TLS) id 15.0.1497.18; Fri, 15 Oct 2021 16:03:05 +0000 From: Dmitry Kozlyuk To: CC: Matan Azrad , Viacheslav Ovsiienko , Ray Kinsella , Anatoly Burakov Date: Fri, 15 Oct 2021 19:02:43 +0300 Message-ID: <20211015160244.3097758-4-dkozlyuk@nvidia.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20211015160244.3097758-1-dkozlyuk@nvidia.com> References: <20211013110131.2909604-1-dkozlyuk@nvidia.com> <20211015160244.3097758-1-dkozlyuk@nvidia.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Content-Type: text/plain X-Originating-IP: [172.20.187.6] X-ClientProxiedBy: HQMAIL105.nvidia.com (172.20.187.12) To HQMAIL107.nvidia.com (172.20.187.13) X-EOPAttributedMessage: 0 X-MS-PublicTrafficType: Email X-MS-Office365-Filtering-Correlation-Id: 30c98b6a-174f-4a8f-a449-08d98ff55f3a X-MS-TrafficTypeDiagnostic: DM6PR12MB4105: X-Microsoft-Antispam-PRVS: X-MS-Exchange-Transport-Forked: True X-MS-Oob-TLC-OOBClassifiers: OLM:34; X-MS-Exchange-SenderADCheck: 1 X-MS-Exchange-AntiSpam-Relay: 0 X-Microsoft-Antispam: BCL:0; X-Microsoft-Antispam-Message-Info: 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 X-Forefront-Antispam-Report: CIP:216.228.112.34; CTRY:US; LANG:en; SCL:1; SRV:; IPV:NLI; SFV:NSPM; H:mail.nvidia.com; PTR:schybrid03.nvidia.com; CAT:NONE; SFS:(4636009)(46966006)(36840700001)(508600001)(356005)(30864003)(70586007)(83380400001)(26005)(7636003)(7696005)(5660300002)(2906002)(70206006)(1076003)(186003)(6286002)(4326008)(8936002)(36756003)(316002)(55016002)(16526019)(426003)(107886003)(6666004)(54906003)(2616005)(8676002)(36860700001)(336012)(86362001)(6916009)(47076005)(82310400003); DIR:OUT; SFP:1101; X-MS-Exchange-CrossTenant-OriginalArrivalTime: 15 Oct 2021 16:03:47.5884 (UTC) X-MS-Exchange-CrossTenant-Network-Message-Id: 30c98b6a-174f-4a8f-a449-08d98ff55f3a X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a; Ip=[216.228.112.34]; Helo=[mail.nvidia.com] X-MS-Exchange-CrossTenant-AuthSource: DM6NAM11FT033.eop-nam11.prod.protection.outlook.com X-MS-Exchange-CrossTenant-AuthAs: Anonymous X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem X-MS-Exchange-Transport-CrossTenantHeadersStamped: DM6PR12MB4105 Subject: [dpdk-dev] [PATCH v5 3/4] common/mlx5: add mempool registration facilities X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add internal API to register mempools, that is, to create memory regions (MR) for their memory and store them in a separate database. Implementation deals with multi-process, so that class drivers don't need to. Each protection domain has its own database. Memory regions can be shared within a database if they represent a single hugepage covering one or more mempools entirely. Add internal API to lookup an MR key for an address that belongs to a known mempool. It is a responsibility of a class driver to extract the mempool from an mbuf. Signed-off-by: Dmitry Kozlyuk Acked-by: Matan Azrad --- drivers/common/mlx5/mlx5_common_mp.c | 50 +++ drivers/common/mlx5/mlx5_common_mp.h | 14 + drivers/common/mlx5/mlx5_common_mr.c | 580 +++++++++++++++++++++++++++ drivers/common/mlx5/mlx5_common_mr.h | 17 + drivers/common/mlx5/version.map | 5 + 5 files changed, 666 insertions(+) diff --git a/drivers/common/mlx5/mlx5_common_mp.c b/drivers/common/mlx5/mlx5_common_mp.c index 673a7c31de..6dfc5535e0 100644 --- a/drivers/common/mlx5/mlx5_common_mp.c +++ b/drivers/common/mlx5/mlx5_common_mp.c @@ -54,6 +54,56 @@ mlx5_mp_req_mr_create(struct mlx5_mp_id *mp_id, uintptr_t addr) return ret; } +/** + * @param mp_id + * ID of the MP process. + * @param share_cache + * Shared MR cache. + * @param pd + * Protection domain. + * @param mempool + * Mempool to register or unregister. + * @param reg + * True to register the mempool, False to unregister. + */ +int +mlx5_mp_req_mempool_reg(struct mlx5_mp_id *mp_id, + struct mlx5_mr_share_cache *share_cache, void *pd, + struct rte_mempool *mempool, bool reg) +{ + struct rte_mp_msg mp_req; + struct rte_mp_msg *mp_res; + struct rte_mp_reply mp_rep; + struct mlx5_mp_param *req = (struct mlx5_mp_param *)mp_req.param; + struct mlx5_mp_arg_mempool_reg *arg = &req->args.mempool_reg; + struct mlx5_mp_param *res; + struct timespec ts = {.tv_sec = MLX5_MP_REQ_TIMEOUT_SEC, .tv_nsec = 0}; + enum mlx5_mp_req_type type; + int ret; + + MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_SECONDARY); + type = reg ? MLX5_MP_REQ_MEMPOOL_REGISTER : + MLX5_MP_REQ_MEMPOOL_UNREGISTER; + mp_init_msg(mp_id, &mp_req, type); + arg->share_cache = share_cache; + arg->pd = pd; + arg->mempool = mempool; + ret = rte_mp_request_sync(&mp_req, &mp_rep, &ts); + if (ret) { + DRV_LOG(ERR, "port %u request to primary process failed", + mp_id->port_id); + return -rte_errno; + } + MLX5_ASSERT(mp_rep.nb_received == 1); + mp_res = &mp_rep.msgs[0]; + res = (struct mlx5_mp_param *)mp_res->param; + ret = res->result; + if (ret) + rte_errno = -ret; + mlx5_free(mp_rep.msgs); + return ret; +} + /** * Request Verbs queue state modification to the primary process. * diff --git a/drivers/common/mlx5/mlx5_common_mp.h b/drivers/common/mlx5/mlx5_common_mp.h index 6829141fc7..527bf3cad8 100644 --- a/drivers/common/mlx5/mlx5_common_mp.h +++ b/drivers/common/mlx5/mlx5_common_mp.h @@ -14,6 +14,8 @@ enum mlx5_mp_req_type { MLX5_MP_REQ_VERBS_CMD_FD = 1, MLX5_MP_REQ_CREATE_MR, + MLX5_MP_REQ_MEMPOOL_REGISTER, + MLX5_MP_REQ_MEMPOOL_UNREGISTER, MLX5_MP_REQ_START_RXTX, MLX5_MP_REQ_STOP_RXTX, MLX5_MP_REQ_QUEUE_STATE_MODIFY, @@ -33,6 +35,12 @@ struct mlx5_mp_arg_queue_id { uint16_t queue_id; /* DPDK queue ID. */ }; +struct mlx5_mp_arg_mempool_reg { + struct mlx5_mr_share_cache *share_cache; + void *pd; /* NULL for MLX5_MP_REQ_MEMPOOL_UNREGISTER */ + struct rte_mempool *mempool; +}; + /* Pameters for IPC. */ struct mlx5_mp_param { enum mlx5_mp_req_type type; @@ -41,6 +49,8 @@ struct mlx5_mp_param { RTE_STD_C11 union { uintptr_t addr; /* MLX5_MP_REQ_CREATE_MR */ + struct mlx5_mp_arg_mempool_reg mempool_reg; + /* MLX5_MP_REQ_MEMPOOL_(UN)REGISTER */ struct mlx5_mp_arg_queue_state_modify state_modify; /* MLX5_MP_REQ_QUEUE_STATE_MODIFY */ struct mlx5_mp_arg_queue_id queue_id; @@ -91,6 +101,10 @@ void mlx5_mp_uninit_secondary(const char *name); __rte_internal int mlx5_mp_req_mr_create(struct mlx5_mp_id *mp_id, uintptr_t addr); __rte_internal +int mlx5_mp_req_mempool_reg(struct mlx5_mp_id *mp_id, + struct mlx5_mr_share_cache *share_cache, void *pd, + struct rte_mempool *mempool, bool reg); +__rte_internal int mlx5_mp_req_queue_state_modify(struct mlx5_mp_id *mp_id, struct mlx5_mp_arg_queue_state_modify *sm); __rte_internal diff --git a/drivers/common/mlx5/mlx5_common_mr.c b/drivers/common/mlx5/mlx5_common_mr.c index 98fe8698e2..2e039a4e70 100644 --- a/drivers/common/mlx5/mlx5_common_mr.c +++ b/drivers/common/mlx5/mlx5_common_mr.c @@ -2,7 +2,10 @@ * Copyright 2016 6WIND S.A. * Copyright 2020 Mellanox Technologies, Ltd */ +#include + #include +#include #include #include #include @@ -21,6 +24,29 @@ struct mr_find_contig_memsegs_data { const struct rte_memseg_list *msl; }; +/* Virtual memory range. */ +struct mlx5_range { + uintptr_t start; + uintptr_t end; +}; + +/** Memory region for a mempool. */ +struct mlx5_mempool_mr { + struct mlx5_pmd_mr pmd_mr; + uint32_t refcnt; /**< Number of mempools sharing this MR. */ +}; + +/* Mempool registration. */ +struct mlx5_mempool_reg { + LIST_ENTRY(mlx5_mempool_reg) next; + /** Registered mempool, used to designate registrations. */ + struct rte_mempool *mp; + /** Memory regions for the address ranges of the mempool. */ + struct mlx5_mempool_mr *mrs; + /** Number of memory regions. */ + unsigned int mrs_n; +}; + /** * Expand B-tree table to a given size. Can't be called with holding * memory_hotplug_lock or share_cache.rwlock due to rte_realloc(). @@ -1191,3 +1217,557 @@ mlx5_mr_dump_cache(struct mlx5_mr_share_cache *share_cache __rte_unused) rte_rwlock_read_unlock(&share_cache->rwlock); #endif } + +static int +mlx5_range_compare_start(const void *lhs, const void *rhs) +{ + const struct mlx5_range *r1 = lhs, *r2 = rhs; + + if (r1->start > r2->start) + return 1; + else if (r1->start < r2->start) + return -1; + return 0; +} + +static void +mlx5_range_from_mempool_chunk(struct rte_mempool *mp, void *opaque, + struct rte_mempool_memhdr *memhdr, + unsigned int idx) +{ + struct mlx5_range *ranges = opaque, *range = &ranges[idx]; + uint64_t page_size = rte_mem_page_size(); + + RTE_SET_USED(mp); + range->start = RTE_ALIGN_FLOOR((uintptr_t)memhdr->addr, page_size); + range->end = RTE_ALIGN_CEIL(range->start + memhdr->len, page_size); +} + +/** + * Get VA-contiguous ranges of the mempool memory. + * Each range start and end is aligned to the system page size. + * + * @param[in] mp + * Analyzed mempool. + * @param[out] out + * Receives the ranges, caller must release it with free(). + * @param[out] ount_n + * Receives the number of @p out elements. + * + * @return + * 0 on success, (-1) on failure. + */ +static int +mlx5_get_mempool_ranges(struct rte_mempool *mp, struct mlx5_range **out, + unsigned int *out_n) +{ + struct mlx5_range *chunks; + unsigned int chunks_n = mp->nb_mem_chunks, contig_n, i; + + /* Collect page-aligned memory ranges of the mempool. */ + chunks = calloc(sizeof(chunks[0]), chunks_n); + if (chunks == NULL) + return -1; + rte_mempool_mem_iter(mp, mlx5_range_from_mempool_chunk, chunks); + /* Merge adjacent chunks and place them at the beginning. */ + qsort(chunks, chunks_n, sizeof(chunks[0]), mlx5_range_compare_start); + contig_n = 1; + for (i = 1; i < chunks_n; i++) + if (chunks[i - 1].end != chunks[i].start) { + chunks[contig_n - 1].end = chunks[i - 1].end; + chunks[contig_n] = chunks[i]; + contig_n++; + } + /* Extend the last contiguous chunk to the end of the mempool. */ + chunks[contig_n - 1].end = chunks[i - 1].end; + *out = chunks; + *out_n = contig_n; + return 0; +} + +/** + * Analyze mempool memory to select memory ranges to register. + * + * @param[in] mp + * Mempool to analyze. + * @param[out] out + * Receives memory ranges to register, aligned to the system page size. + * The caller must release them with free(). + * @param[out] out_n + * Receives the number of @p out items. + * @param[out] share_hugepage + * Receives True if the entire pool resides within a single hugepage. + * + * @return + * 0 on success, (-1) on failure. + */ +static int +mlx5_mempool_reg_analyze(struct rte_mempool *mp, struct mlx5_range **out, + unsigned int *out_n, bool *share_hugepage) +{ + struct mlx5_range *ranges = NULL; + unsigned int i, ranges_n = 0; + struct rte_memseg_list *msl; + + if (mlx5_get_mempool_ranges(mp, &ranges, &ranges_n) < 0) { + DRV_LOG(ERR, "Cannot get address ranges for mempool %s", + mp->name); + return -1; + } + /* Check if the hugepage of the pool can be shared. */ + *share_hugepage = false; + msl = rte_mem_virt2memseg_list((void *)ranges[0].start); + if (msl != NULL) { + uint64_t hugepage_sz = 0; + + /* Check that all ranges are on pages of the same size. */ + for (i = 0; i < ranges_n; i++) { + if (hugepage_sz != 0 && hugepage_sz != msl->page_sz) + break; + hugepage_sz = msl->page_sz; + } + if (i == ranges_n) { + /* + * If the entire pool is within one hugepage, + * combine all ranges into one of the hugepage size. + */ + uintptr_t reg_start = ranges[0].start; + uintptr_t reg_end = ranges[ranges_n - 1].end; + uintptr_t hugepage_start = + RTE_ALIGN_FLOOR(reg_start, hugepage_sz); + uintptr_t hugepage_end = hugepage_start + hugepage_sz; + if (reg_end < hugepage_end) { + ranges[0].start = hugepage_start; + ranges[0].end = hugepage_end; + ranges_n = 1; + *share_hugepage = true; + } + } + } + *out = ranges; + *out_n = ranges_n; + return 0; +} + +/** Create a registration object for the mempool. */ +static struct mlx5_mempool_reg * +mlx5_mempool_reg_create(struct rte_mempool *mp, unsigned int mrs_n) +{ + struct mlx5_mempool_reg *mpr = NULL; + + mpr = mlx5_malloc(MLX5_MEM_RTE | MLX5_MEM_ZERO, + sizeof(*mpr) + mrs_n * sizeof(mpr->mrs[0]), + RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY); + if (mpr == NULL) { + DRV_LOG(ERR, "Cannot allocate mempool %s registration object", + mp->name); + return NULL; + } + mpr->mp = mp; + mpr->mrs = (struct mlx5_mempool_mr *)(mpr + 1); + mpr->mrs_n = mrs_n; + return mpr; +} + +/** + * Destroy a mempool registration object. + * + * @param standalone + * Whether @p mpr owns its MRs excludively, i.e. they are not shared. + */ +static void +mlx5_mempool_reg_destroy(struct mlx5_mr_share_cache *share_cache, + struct mlx5_mempool_reg *mpr, bool standalone) +{ + if (standalone) { + unsigned int i; + + for (i = 0; i < mpr->mrs_n; i++) + share_cache->dereg_mr_cb(&mpr->mrs[i].pmd_mr); + } + mlx5_free(mpr); +} + +/** Find registration object of a mempool. */ +static struct mlx5_mempool_reg * +mlx5_mempool_reg_lookup(struct mlx5_mr_share_cache *share_cache, + struct rte_mempool *mp) +{ + struct mlx5_mempool_reg *mpr; + + LIST_FOREACH(mpr, &share_cache->mempool_reg_list, next) + if (mpr->mp == mp) + break; + return mpr; +} + +/** Increment reference counters of MRs used in the registration. */ +static void +mlx5_mempool_reg_attach(struct mlx5_mempool_reg *mpr) +{ + unsigned int i; + + for (i = 0; i < mpr->mrs_n; i++) + __atomic_add_fetch(&mpr->mrs[i].refcnt, 1, __ATOMIC_RELAXED); +} + +/** + * Decrement reference counters of MRs used in the registration. + * + * @return True if no more references to @p mpr MRs exist, False otherwise. + */ +static bool +mlx5_mempool_reg_detach(struct mlx5_mempool_reg *mpr) +{ + unsigned int i; + bool ret = false; + + for (i = 0; i < mpr->mrs_n; i++) + ret |= __atomic_sub_fetch(&mpr->mrs[i].refcnt, 1, + __ATOMIC_RELAXED) == 0; + return ret; +} + +static int +mlx5_mr_mempool_register_primary(struct mlx5_mr_share_cache *share_cache, + void *pd, struct rte_mempool *mp) +{ + struct mlx5_range *ranges = NULL; + struct mlx5_mempool_reg *mpr, *new_mpr; + unsigned int i, ranges_n; + bool share_hugepage; + int ret = -1; + + /* Early check to avoid unnecessary creation of MRs. */ + rte_rwlock_read_lock(&share_cache->rwlock); + mpr = mlx5_mempool_reg_lookup(share_cache, mp); + rte_rwlock_read_unlock(&share_cache->rwlock); + if (mpr != NULL) { + DRV_LOG(DEBUG, "Mempool %s is already registered for PD %p", + mp->name, pd); + rte_errno = EEXIST; + goto exit; + } + if (mlx5_mempool_reg_analyze(mp, &ranges, &ranges_n, + &share_hugepage) < 0) { + DRV_LOG(ERR, "Cannot get mempool %s memory ranges", mp->name); + rte_errno = ENOMEM; + goto exit; + } + new_mpr = mlx5_mempool_reg_create(mp, ranges_n); + if (new_mpr == NULL) { + DRV_LOG(ERR, + "Cannot create a registration object for mempool %s in PD %p", + mp->name, pd); + rte_errno = ENOMEM; + goto exit; + } + /* + * If the entire mempool fits in a single hugepage, the MR for this + * hugepage can be shared across mempools that also fit in it. + */ + if (share_hugepage) { + rte_rwlock_write_lock(&share_cache->rwlock); + LIST_FOREACH(mpr, &share_cache->mempool_reg_list, next) { + if (mpr->mrs[0].pmd_mr.addr == (void *)ranges[0].start) + break; + } + if (mpr != NULL) { + new_mpr->mrs = mpr->mrs; + mlx5_mempool_reg_attach(new_mpr); + LIST_INSERT_HEAD(&share_cache->mempool_reg_list, + new_mpr, next); + } + rte_rwlock_write_unlock(&share_cache->rwlock); + if (mpr != NULL) { + DRV_LOG(DEBUG, "Shared MR %#x in PD %p for mempool %s with mempool %s", + mpr->mrs[0].pmd_mr.lkey, pd, mp->name, + mpr->mp->name); + ret = 0; + goto exit; + } + } + for (i = 0; i < ranges_n; i++) { + struct mlx5_mempool_mr *mr = &new_mpr->mrs[i]; + const struct mlx5_range *range = &ranges[i]; + size_t len = range->end - range->start; + + if (share_cache->reg_mr_cb(pd, (void *)range->start, len, + &mr->pmd_mr) < 0) { + DRV_LOG(ERR, + "Failed to create an MR in PD %p for address range " + "[0x%" PRIxPTR ", 0x%" PRIxPTR "] (%zu bytes) for mempool %s", + pd, range->start, range->end, len, mp->name); + break; + } + DRV_LOG(DEBUG, + "Created a new MR %#x in PD %p for address range " + "[0x%" PRIxPTR ", 0x%" PRIxPTR "] (%zu bytes) for mempool %s", + mr->pmd_mr.lkey, pd, range->start, range->end, len, + mp->name); + } + if (i != ranges_n) { + mlx5_mempool_reg_destroy(share_cache, new_mpr, true); + rte_errno = EINVAL; + goto exit; + } + /* Concurrent registration is not supposed to happen. */ + rte_rwlock_write_lock(&share_cache->rwlock); + mpr = mlx5_mempool_reg_lookup(share_cache, mp); + if (mpr == NULL) { + mlx5_mempool_reg_attach(new_mpr); + LIST_INSERT_HEAD(&share_cache->mempool_reg_list, + new_mpr, next); + ret = 0; + } + rte_rwlock_write_unlock(&share_cache->rwlock); + if (mpr != NULL) { + DRV_LOG(DEBUG, "Mempool %s is already registered for PD %p", + mp->name, pd); + mlx5_mempool_reg_destroy(share_cache, new_mpr, true); + rte_errno = EEXIST; + goto exit; + } +exit: + free(ranges); + return ret; +} + +static int +mlx5_mr_mempool_register_secondary(struct mlx5_mr_share_cache *share_cache, + void *pd, struct rte_mempool *mp, + struct mlx5_mp_id *mp_id) +{ + if (mp_id == NULL) { + rte_errno = EINVAL; + return -1; + } + return mlx5_mp_req_mempool_reg(mp_id, share_cache, pd, mp, true); +} + +/** + * Register the memory of a mempool in the protection domain. + * + * @param share_cache + * Shared MR cache of the protection domain. + * @param pd + * Protection domain object. + * @param mp + * Mempool to register. + * @param mp_id + * Multi-process identifier, may be NULL for the primary process. + * + * @return + * 0 on success, (-1) on failure and rte_errno is set. + */ +int +mlx5_mr_mempool_register(struct mlx5_mr_share_cache *share_cache, void *pd, + struct rte_mempool *mp, struct mlx5_mp_id *mp_id) +{ + if (mp->flags & MEMPOOL_F_NON_IO) + return 0; + switch (rte_eal_process_type()) { + case RTE_PROC_PRIMARY: + return mlx5_mr_mempool_register_primary(share_cache, pd, mp); + case RTE_PROC_SECONDARY: + return mlx5_mr_mempool_register_secondary(share_cache, pd, mp, + mp_id); + default: + return -1; + } +} + +static int +mlx5_mr_mempool_unregister_primary(struct mlx5_mr_share_cache *share_cache, + struct rte_mempool *mp) +{ + struct mlx5_mempool_reg *mpr; + bool standalone = false; + + rte_rwlock_write_lock(&share_cache->rwlock); + LIST_FOREACH(mpr, &share_cache->mempool_reg_list, next) + if (mpr->mp == mp) { + LIST_REMOVE(mpr, next); + standalone = mlx5_mempool_reg_detach(mpr); + if (standalone) + /* + * The unlock operation below provides a memory + * barrier due to its store-release semantics. + */ + ++share_cache->dev_gen; + break; + } + rte_rwlock_write_unlock(&share_cache->rwlock); + if (mpr == NULL) { + rte_errno = ENOENT; + return -1; + } + mlx5_mempool_reg_destroy(share_cache, mpr, standalone); + return 0; +} + +static int +mlx5_mr_mempool_unregister_secondary(struct mlx5_mr_share_cache *share_cache, + struct rte_mempool *mp, + struct mlx5_mp_id *mp_id) +{ + if (mp_id == NULL) { + rte_errno = EINVAL; + return -1; + } + return mlx5_mp_req_mempool_reg(mp_id, share_cache, NULL, mp, false); +} + +/** + * Unregister the memory of a mempool from the protection domain. + * + * @param share_cache + * Shared MR cache of the protection domain. + * @param mp + * Mempool to unregister. + * @param mp_id + * Multi-process identifier, may be NULL for the primary process. + * + * @return + * 0 on success, (-1) on failure and rte_errno is set. + */ +int +mlx5_mr_mempool_unregister(struct mlx5_mr_share_cache *share_cache, + struct rte_mempool *mp, struct mlx5_mp_id *mp_id) +{ + if (mp->flags & MEMPOOL_F_NON_IO) + return 0; + switch (rte_eal_process_type()) { + case RTE_PROC_PRIMARY: + return mlx5_mr_mempool_unregister_primary(share_cache, mp); + case RTE_PROC_SECONDARY: + return mlx5_mr_mempool_unregister_secondary(share_cache, mp, + mp_id); + default: + return -1; + } +} + +/** + * Lookup a MR key by and address in a registered mempool. + * + * @param mpr + * Mempool registration object. + * @param addr + * Address within the mempool. + * @param entry + * Bottom-half cache entry to fill. + * + * @return + * MR key or UINT32_MAX on failure, which can only happen + * if the address is not from within the mempool. + */ +static uint32_t +mlx5_mempool_reg_addr2mr(struct mlx5_mempool_reg *mpr, uintptr_t addr, + struct mr_cache_entry *entry) +{ + uint32_t lkey = UINT32_MAX; + unsigned int i; + + for (i = 0; i < mpr->mrs_n; i++) { + const struct mlx5_pmd_mr *mr = &mpr->mrs[i].pmd_mr; + uintptr_t mr_addr = (uintptr_t)mr->addr; + + if (mr_addr <= addr) { + lkey = rte_cpu_to_be_32(mr->lkey); + entry->start = mr_addr; + entry->end = mr_addr + mr->len; + entry->lkey = lkey; + break; + } + } + return lkey; +} + +/** + * Update bottom-half cache from the list of mempool registrations. + * + * @param share_cache + * Pointer to a global shared MR cache. + * @param mr_ctrl + * Per-queue MR control handle. + * @param entry + * Pointer to an entry in the bottom-half cache to update + * with the MR lkey looked up. + * @param mp + * Mempool containing the address. + * @param addr + * Address to lookup. + * @return + * MR lkey on success, UINT32_MAX on failure. + */ +static uint32_t +mlx5_lookup_mempool_regs(struct mlx5_mr_share_cache *share_cache, + struct mlx5_mr_ctrl *mr_ctrl, + struct mr_cache_entry *entry, + struct rte_mempool *mp, uintptr_t addr) +{ + struct mlx5_mr_btree *bt = &mr_ctrl->cache_bh; + struct mlx5_mempool_reg *mpr; + uint32_t lkey = UINT32_MAX; + + /* If local cache table is full, try to double it. */ + if (unlikely(bt->len == bt->size)) + mr_btree_expand(bt, bt->size << 1); + /* Look up in mempool registrations. */ + rte_rwlock_read_lock(&share_cache->rwlock); + mpr = mlx5_mempool_reg_lookup(share_cache, mp); + if (mpr != NULL) + lkey = mlx5_mempool_reg_addr2mr(mpr, addr, entry); + rte_rwlock_read_unlock(&share_cache->rwlock); + /* + * Update local cache. Even if it fails, return the found entry + * to update top-half cache. Next time, this entry will be found + * in the global cache. + */ + if (lkey != UINT32_MAX) + mr_btree_insert(bt, entry); + return lkey; +} + +/** + * Bottom-half lookup for the address from the mempool. + * + * @param share_cache + * Pointer to a global shared MR cache. + * @param mr_ctrl + * Per-queue MR control handle. + * @param mp + * Mempool containing the address. + * @param addr + * Address to lookup. + * @return + * MR lkey on success, UINT32_MAX on failure. + */ +uint32_t +mlx5_mr_mempool2mr_bh(struct mlx5_mr_share_cache *share_cache, + struct mlx5_mr_ctrl *mr_ctrl, + struct rte_mempool *mp, uintptr_t addr) +{ + struct mr_cache_entry *repl = &mr_ctrl->cache[mr_ctrl->head]; + uint32_t lkey; + uint16_t bh_idx = 0; + + /* Binary-search MR translation table. */ + lkey = mr_btree_lookup(&mr_ctrl->cache_bh, &bh_idx, addr); + /* Update top-half cache. */ + if (likely(lkey != UINT32_MAX)) { + *repl = (*mr_ctrl->cache_bh.table)[bh_idx]; + } else { + lkey = mlx5_lookup_mempool_regs(share_cache, mr_ctrl, repl, + mp, addr); + /* Can only fail if the address is not from the mempool. */ + if (unlikely(lkey == UINT32_MAX)) + return UINT32_MAX; + } + /* Update the most recently used entry. */ + mr_ctrl->mru = mr_ctrl->head; + /* Point to the next victim, the oldest. */ + mr_ctrl->head = (mr_ctrl->head + 1) % MLX5_MR_CACHE_N; + return lkey; +} diff --git a/drivers/common/mlx5/mlx5_common_mr.h b/drivers/common/mlx5/mlx5_common_mr.h index 6e465a05e9..685ac98e08 100644 --- a/drivers/common/mlx5/mlx5_common_mr.h +++ b/drivers/common/mlx5/mlx5_common_mr.h @@ -13,6 +13,7 @@ #include #include +#include #include #include "mlx5_glue.h" @@ -75,6 +76,7 @@ struct mlx5_mr_ctrl { } __rte_packed; LIST_HEAD(mlx5_mr_list, mlx5_mr); +LIST_HEAD(mlx5_mempool_reg_list, mlx5_mempool_reg); /* Global per-device MR cache. */ struct mlx5_mr_share_cache { @@ -83,6 +85,7 @@ struct mlx5_mr_share_cache { struct mlx5_mr_btree cache; /* Global MR cache table. */ struct mlx5_mr_list mr_list; /* Registered MR list. */ struct mlx5_mr_list mr_free_list; /* Freed MR list. */ + struct mlx5_mempool_reg_list mempool_reg_list; /* Mempool database. */ mlx5_reg_mr_t reg_mr_cb; /* Callback to reg_mr func */ mlx5_dereg_mr_t dereg_mr_cb; /* Callback to dereg_mr func */ } __rte_packed; @@ -136,6 +139,10 @@ uint32_t mlx5_mr_addr2mr_bh(void *pd, struct mlx5_mp_id *mp_id, struct mlx5_mr_ctrl *mr_ctrl, uintptr_t addr, unsigned int mr_ext_memseg_en); __rte_internal +uint32_t mlx5_mr_mempool2mr_bh(struct mlx5_mr_share_cache *share_cache, + struct mlx5_mr_ctrl *mr_ctrl, + struct rte_mempool *mp, uintptr_t addr); +__rte_internal void mlx5_mr_release_cache(struct mlx5_mr_share_cache *mr_cache); __rte_internal void mlx5_mr_dump_cache(struct mlx5_mr_share_cache *share_cache __rte_unused); @@ -179,4 +186,14 @@ mlx5_common_verbs_dereg_mr(struct mlx5_pmd_mr *pmd_mr); __rte_internal void mlx5_mr_free(struct mlx5_mr *mr, mlx5_dereg_mr_t dereg_mr_cb); + +__rte_internal +int +mlx5_mr_mempool_register(struct mlx5_mr_share_cache *share_cache, void *pd, + struct rte_mempool *mp, struct mlx5_mp_id *mp_id); +__rte_internal +int +mlx5_mr_mempool_unregister(struct mlx5_mr_share_cache *share_cache, + struct rte_mempool *mp, struct mlx5_mp_id *mp_id); + #endif /* RTE_PMD_MLX5_COMMON_MR_H_ */ diff --git a/drivers/common/mlx5/version.map b/drivers/common/mlx5/version.map index d3c5040aac..85100d5afb 100644 --- a/drivers/common/mlx5/version.map +++ b/drivers/common/mlx5/version.map @@ -152,4 +152,9 @@ INTERNAL { mlx5_realloc; mlx5_translate_port_name; # WINDOWS_NO_EXPORT + + mlx5_mr_mempool_register; + mlx5_mr_mempool_unregister; + mlx5_mp_req_mempool_reg; + mlx5_mr_mempool2mr_bh; }; -- 2.25.1