From patchwork Tue Sep 26 05:36:39 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yuanhan Liu X-Patchwork-Id: 818465 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (mailfrom) smtp.mailfrom=openvswitch.org (client-ip=140.211.169.12; helo=mail.linuxfoundation.org; envelope-from=ovs-dev-bounces@openvswitch.org; receiver=) Authentication-Results: ozlabs.org; dkim=fail reason="signature verification failed" (2048-bit key; unprotected) header.d=fridaylinux-org.20150623.gappssmtp.com header.i=@fridaylinux-org.20150623.gappssmtp.com header.b="yRKUStA7"; dkim-atps=neutral Received: from mail.linuxfoundation.org (mail.linuxfoundation.org [140.211.169.12]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 3y1VJv74N9z9t1t for ; Tue, 26 Sep 2017 15:44:47 +1000 (AEST) Received: from mail.linux-foundation.org (localhost [127.0.0.1]) by mail.linuxfoundation.org (Postfix) with ESMTP id A23B7C09; Tue, 26 Sep 2017 05:40:55 +0000 (UTC) X-Original-To: dev@openvswitch.org Delivered-To: ovs-dev@mail.linuxfoundation.org Received: from smtp1.linuxfoundation.org (smtp1.linux-foundation.org [172.17.192.35]) by mail.linuxfoundation.org (Postfix) with ESMTPS id 519C9BEA for ; Tue, 26 Sep 2017 05:40:54 +0000 (UTC) X-Greylist: whitelisted by SQLgrey-1.7.6 Received: from mail-pg0-f43.google.com (mail-pg0-f43.google.com [74.125.83.43]) by smtp1.linuxfoundation.org (Postfix) with ESMTPS id 85C95367 for ; Tue, 26 Sep 2017 05:40:53 +0000 (UTC) Received: by mail-pg0-f43.google.com with SMTP id 7so5351569pgd.13 for ; Mon, 25 Sep 2017 22:40:53 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=fridaylinux-org.20150623.gappssmtp.com; s=20150623; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=St7axmU79AXmtWRugAfa/hkjJQFBa8Jlx0AsVRSFMWw=; b=yRKUStA7UYwYjAL6NlaeQkHBsr8ruquHPM3LVpo6X1SuCWRHcpeHc4JgtniJg1yA3L NycDYTSm/h9PkCW/XAXWILjTv4CC930tcCxVKcMyRkzOH5kxB8aUNgFFWyMkWDwamrwF gbSLHaJFeFGe23v58kGNkxJp/zLnYJKWiszGqf8MguPaDBhadg180ZQbFINJf6Km9xLg mS6ZmcS5yQW52XJk3D/M0BlXgftC2rMpaeYYa4j9XyHWEdmZGUKL9TUppjVOw0HTmUhh LeJflNWyW6XSXaF5CnZpZs4i5J3XBDrB3RLLYsYiDvnDWoI8d2RBB1hGC00U3tOd6N1H EFIA== 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=St7axmU79AXmtWRugAfa/hkjJQFBa8Jlx0AsVRSFMWw=; b=Bz8q7DB5QkNhLlhO4hjM2W/vekDh/FImbmoI6aDT7eb27dDWwitH+NPXDnVIKWOROx uiEZp81CO+kqO4XPouCiWaUqSBxX4mHWCqAYEMwvTfKxIvi+C2Sqrv6W8N2Tegql/4XQ eU/wuoYX3/ML64guBokPKnpt3d7irT1uCqsdWiGZuoa/rdC+vwpnLviRV8qWRVVrmPWh Xqrnt7NGS5vf1NLFSlQ1qcr2ayMUBZXTH4eEHN8foUl0z/ACgRGImB9RvZVse4cNEkgn M9/lEWaPIOrPIcLMDFtXJENIorGhXU7Ltu2VXO10zvmMiyg80AbNlovZ1A/8fxWWzlLM 3r7A== X-Gm-Message-State: AHPjjUgGvdgQbpdcSePWZXzm+/oRBUijQJefpgE3tOvFceug7JFYeI93 QF9LQ8nkSmGEAOll4vPQNYK+FpOkL7DQOw== X-Google-Smtp-Source: AOwi7QCoMsqTpaBM+3BE1WfFRd+qDoQfIryAPLJS5NTlBgtFhf8RZYDGSph0l6HUVrONL2cPIA5iCw== X-Received: by 10.99.114.19 with SMTP id n19mr9516237pgc.300.1506404452750; Mon, 25 Sep 2017 22:40:52 -0700 (PDT) Received: from localhost.localdomain ([101.228.205.132]) by smtp.gmail.com with ESMTPSA id o79sm13180077pfi.108.2017.09.25.22.40.42 (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Mon, 25 Sep 2017 22:40:51 -0700 (PDT) From: Yuanhan Liu To: dev@openvswitch.org Date: Tue, 26 Sep 2017 13:36:39 +0800 Message-Id: <1506404199-23579-10-git-send-email-yliu@fridaylinux.org> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1506404199-23579-1-git-send-email-yliu@fridaylinux.org> References: <1506404199-23579-1-git-send-email-yliu@fridaylinux.org> X-Spam-Status: No, score=0.0 required=5.0 tests=DKIM_SIGNED,DKIM_VALID, RCVD_IN_DNSWL_NONE autolearn=disabled version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on smtp1.linux-foundation.org Cc: Simon Horman Subject: [ovs-dev] [PATCH v3 9/9] dpif-netdev: do hw flow offload in another thread X-BeenThere: ovs-dev@openvswitch.org X-Mailman-Version: 2.1.12 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , MIME-Version: 1.0 Sender: ovs-dev-bounces@openvswitch.org Errors-To: ovs-dev-bounces@openvswitch.org Currently, the major trigger for hw flow offload is at upcall handling, which is actually in the datapath. Moreover, the hw offload installation and modification is not that lightweight. Meaning, if there are so many flows being added or modified frequently, it could stall the datapath, which could result to packet loss. To diminish that, all those flow operations will be recorded and appended to a list. A thread is then introduced to process this list (to do the real flow offloading put/del operations). This could leave the datapath as lightweight as possible. Signed-off-by: Yuanhan Liu --- lib/dpif-netdev.c | 301 ++++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 249 insertions(+), 52 deletions(-) diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index 13fd012..ef5c2e9 100644 --- a/lib/dpif-netdev.c +++ b/lib/dpif-netdev.c @@ -1949,6 +1949,223 @@ dp_netdev_pmd_find_flow_by_mark(const uint32_t mark) return NULL; } +struct dp_flow_offload_item { + struct dp_netdev_pmd_thread *pmd; + odp_port_t in_port; + struct dp_netdev_flow *flow; + ovs_u128 ufid; + int op; + struct match match; + struct nlattr *actions; + size_t actions_len; + int rxq; + uint32_t flow_mark; + + struct ovs_list node; +}; + +struct dp_flow_offload { + struct ovs_mutex mutex; + struct ovs_list list; + pthread_cond_t cond; +}; + +static struct dp_flow_offload dp_flow_offload = { + .mutex = OVS_MUTEX_INITIALIZER, + .list = OVS_LIST_INITIALIZER(&dp_flow_offload.list), +}; + +static struct ovsthread_once offload_thread_once + = OVSTHREAD_ONCE_INITIALIZER; + +enum { + DP_NETDEV_FLOW_OFFLOAD_OP_ADD, + DP_NETDEV_FLOW_OFFLOAD_OP_MOD, + DP_NETDEV_FLOW_OFFLOAD_OP_DEL, +}; + +static struct dp_flow_offload_item * +dp_netdev_alloc_flow_offload(struct dp_netdev_pmd_thread *pmd, + struct dp_netdev_flow *flow, + odp_port_t in_port, const ovs_u128 *ufid, + int op) +{ + struct dp_flow_offload_item *offload; + + offload = xzalloc(sizeof(*offload)); + offload->pmd = pmd; + offload->flow = flow; + offload->in_port = in_port; + offload->ufid = *ufid; + offload->op = op; + + ovs_refcount_ref(&pmd->ref_cnt); + ovs_refcount_ref(&flow->ref_cnt); + + return offload; +} + +static void +dp_netdev_free_flow_offload(struct dp_flow_offload_item *offload) +{ + ovs_refcount_unref(&offload->pmd->ref_cnt); + ovs_refcount_unref(&offload->flow->ref_cnt); + + free(offload->actions); + free(offload); +} + +static void +dp_netdev_append_flow_offload(struct dp_flow_offload_item *offload) +{ + ovs_mutex_lock(&dp_flow_offload.mutex); + ovs_list_push_back(&dp_flow_offload.list, &offload->node); + ovs_mutex_unlock(&dp_flow_offload.mutex); + + pthread_cond_signal(&dp_flow_offload.cond); +} + +static int +dp_netdev_flow_offload_del(struct dp_flow_offload_item *offload) +{ + struct dp_netdev_flow *flow = offload->flow; + struct dp_netdev_port *port; + int ret = -1; + + ovs_mutex_lock(&offload->pmd->flow_mutex); + port = dp_netdev_lookup_port(offload->pmd->dp, offload->in_port); + if (flow->has_mark && port) { + ret = netdev_flow_del(port->netdev, &offload->ufid, NULL); + } + + dp_netdev_remove_flow_mark_map(flow->mark); + ovsrcu_quiesce_start(); + flow->has_mark = false; + ovs_mutex_unlock(&offload->pmd->flow_mutex); + + return ret; +} + +static int +dp_netdev_flow_offload_put(struct dp_flow_offload_item *offload) +{ + struct dp_netdev_port *port; + struct dp_netdev_flow *flow = offload->flow; + bool create = offload->op == DP_NETDEV_FLOW_OFFLOAD_OP_ADD; + struct offload_info info; + int ret; + + port = dp_netdev_lookup_port(offload->pmd->dp, offload->in_port); + if (!port) { + return -1; + } + + if (create) { + if (!dp_netdev_alloc_flow_mark(&info.flow_mark)) { + VLOG_ERR("failed to allocate flow mark!\n"); + return -1; + } + offload->flow_mark = info.flow_mark; + } else { + info.flow_mark = offload->flow_mark; + } + info.rxq = offload->rxq; + + ret = netdev_flow_put(port->netdev, &offload->match, + offload->actions, offload->actions_len, + &offload->ufid, &info, NULL); + if (ret) { + if (create) { + dp_netdev_remove_flow_mark_map(info.flow_mark); + } + return ret; + } + + ovs_mutex_lock(&offload->pmd->flow_mutex); + if (create) { + flow->has_mark = true; + flow->mark = info.flow_mark; + if (!flow->dead) { + /* + * A flow could have been dead after we regain the lock, + * while the flow has offloaded to the netdev. When that + * happens, there should be an offload item in the offload + * list for the flow removal. To make sure the flow will + * be deleted successfully later, above 2 fields (has_mark + * and mark) have to be set properly. + */ + dp_netdev_install_flow_mark_map(flow->mark, flow); + ovsrcu_quiesce_start(); + } + } + ovs_mutex_unlock(&offload->pmd->flow_mutex); + + return 0; +} + +static void * +dp_netdev_flow_offload_main(void *data OVS_UNUSED) +{ + struct dp_flow_offload_item *offload; + struct ovs_list *list; + const char *op; + int ret; + + for (;;) { + ovs_mutex_lock(&dp_flow_offload.mutex); + if (ovs_list_is_empty(&dp_flow_offload.list)) { + pthread_cond_wait(&dp_flow_offload.cond, + &dp_flow_offload.mutex.lock); + } + list = ovs_list_pop_front(&dp_flow_offload.list); + offload = CONTAINER_OF(list, struct dp_flow_offload_item, node); + ovs_mutex_unlock(&dp_flow_offload.mutex); + + switch (offload->op) { + case DP_NETDEV_FLOW_OFFLOAD_OP_ADD: + op = "install"; + ret = dp_netdev_flow_offload_put(offload); + break; + case DP_NETDEV_FLOW_OFFLOAD_OP_MOD: + op = "modify"; + ret = dp_netdev_flow_offload_put(offload); + break; + case DP_NETDEV_FLOW_OFFLOAD_OP_DEL: + op = "delete"; + ret = dp_netdev_flow_offload_del(offload); + break; + default: + OVS_NOT_REACHED(); + } + + VLOG_INFO("%s to %s netdev flow with mark %u\n", + ret == 0 ? "succeed" : "failed", + op, offload->flow_mark); + dp_netdev_free_flow_offload(offload); + } + + return NULL; +} + +static void +queue_netdev_flow_del(struct dp_netdev_pmd_thread *pmd, + struct dp_netdev_flow *flow, + odp_port_t in_port, const ovs_u128 *ufid, + int op) +{ + struct dp_flow_offload_item *offload; + + if (ovsthread_once_start(&offload_thread_once)) { + pthread_cond_init(&dp_flow_offload.cond, NULL); + ovs_thread_create("dp_netdev_flow_offload", + dp_netdev_flow_offload_main, NULL); + ovsthread_once_done(&offload_thread_once); + } + + offload = dp_netdev_alloc_flow_offload(pmd, flow, in_port, ufid, op); + offload->flow_mark = flow->mark; + dp_netdev_append_flow_offload(offload); +} static void dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, @@ -1963,16 +2180,8 @@ dp_netdev_pmd_remove_flow(struct dp_netdev_pmd_thread *pmd, ovs_assert(cls != NULL); dpcls_remove(cls, &flow->cr); cmap_remove(&pmd->flow_table, node, dp_netdev_flow_hash(&flow->ufid)); - if (flow->has_mark) { - struct dp_netdev_port *port; - - port = dp_netdev_lookup_port(pmd->dp, in_port); - if (port) { - netdev_flow_del(port->netdev, &flow->ufid, NULL); - } - dp_netdev_remove_flow_mark_map(flow->mark); - flow->has_mark = false; - } + queue_netdev_flow_del(pmd, flow, in_port, &flow->ufid, + DP_NETDEV_FLOW_OFFLOAD_OP_DEL); flow->dead = true; dp_netdev_flow_unref(flow); @@ -2553,53 +2762,41 @@ out: } static void -try_netdev_flow_put(struct dp_netdev_pmd_thread *pmd, odp_port_t in_port, - struct dp_netdev_flow *flow, struct match *match, - const ovs_u128 *ufid, const struct nlattr *actions, - size_t actions_len, int rxq) +queue_netdev_flow_put(struct dp_netdev_pmd_thread *pmd, odp_port_t in_port, + struct dp_netdev_flow *flow, struct match *match, + const ovs_u128 *ufid, const struct nlattr *actions, + size_t actions_len, int rxq) { - struct offload_info info; - struct dp_netdev_port *port; - bool modification = flow->has_mark; - const char *op = modification ? "modify" : "install"; - int ret; + struct dp_flow_offload_item *offload; + int op; - port = dp_netdev_lookup_port(pmd->dp, in_port); - if (!port) { + if (!netdev_is_flow_api_enabled()) { return; } - info.rxq = rxq; - - if (modification) { - info.flow_mark = flow->mark; - } else { - if (!netdev_is_flow_api_enabled()) { - return; - } - - if (!dp_netdev_alloc_flow_mark(&info.flow_mark)) { - VLOG_ERR("failed to allocate flow mark!\n"); - return; - } - } - ret = netdev_flow_put(port->netdev, match, - CONST_CAST(struct nlattr *, actions), - actions_len, ufid, &info, NULL); - if (ret) { - VLOG_ERR("failed to %s netdev flow with mark %u\n", - op, info.flow_mark); - return; + if (ovsthread_once_start(&offload_thread_once)) { + pthread_cond_init(&dp_flow_offload.cond, NULL); + ovs_thread_create("dp_netdev_flow_offload", + dp_netdev_flow_offload_main, NULL); + ovsthread_once_done(&offload_thread_once); } - if (!modification) { - flow->has_mark = true; - flow->mark = info.flow_mark; - dp_netdev_install_flow_mark_map(info.flow_mark, flow); + if (flow->has_mark) { + op = DP_NETDEV_FLOW_OFFLOAD_OP_MOD; + } else { + op = DP_NETDEV_FLOW_OFFLOAD_OP_ADD; + } + offload = dp_netdev_alloc_flow_offload(pmd, flow, in_port, ufid, op); + offload->match = *match; + offload->rxq = rxq; + offload->actions = xmalloc(actions_len); + memcpy(offload->actions, actions, actions_len); + offload->actions_len = actions_len; + if (flow->has_mark) { + offload->flow_mark = flow->mark; } - VLOG_INFO("succeed to %s netdev flow with mark %u\n", - op, flow->mark); + dp_netdev_append_flow_offload(offload); } static struct dp_netdev_flow * @@ -2654,8 +2851,8 @@ dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd, cmap_insert(&pmd->flow_table, CONST_CAST(struct cmap_node *, &flow->node), dp_netdev_flow_hash(&flow->ufid)); - try_netdev_flow_put(pmd, in_port, flow, match, ufid, - actions, actions_len, rxq); + queue_netdev_flow_put(pmd, in_port, flow, match, ufid, + actions, actions_len, rxq); if (OVS_UNLIKELY(!VLOG_DROP_DBG((&upcall_rl)))) { struct ds ds = DS_EMPTY_INITIALIZER; @@ -2745,8 +2942,8 @@ flow_put_on_pmd(struct dp_netdev_pmd_thread *pmd, old_actions = dp_netdev_flow_get_actions(netdev_flow); ovsrcu_set(&netdev_flow->actions, new_actions); - try_netdev_flow_put(pmd, in_port, netdev_flow, match, ufid, - put->actions, put->actions_len, -1); + queue_netdev_flow_put(pmd, in_port, netdev_flow, match, ufid, + put->actions, put->actions_len, -1); if (stats) { get_dpif_flow_stats(netdev_flow, stats);