From patchwork Fri Sep 4 14:27:38 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mark Gray X-Patchwork-Id: 1357542 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=openvswitch.org (client-ip=140.211.166.136; helo=silver.osuosl.org; envelope-from=ovs-dev-bounces@openvswitch.org; receiver=) Authentication-Results: ozlabs.org; dmarc=fail (p=none dis=none) header.from=redhat.com Authentication-Results: ozlabs.org; dkim=fail reason="signature verification failed" (1024-bit key; unprotected) header.d=redhat.com header.i=@redhat.com header.a=rsa-sha256 header.s=mimecast20190719 header.b=Uhtn5hg3; dkim-atps=neutral Received: from silver.osuosl.org (smtp3.osuosl.org [140.211.166.136]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 4Bjg5y6mlcz9sVM for ; Sat, 5 Sep 2020 00:28:54 +1000 (AEST) Received: from localhost (localhost [127.0.0.1]) by silver.osuosl.org (Postfix) with ESMTP id 64854203D4; Fri, 4 Sep 2020 14:28:52 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from silver.osuosl.org ([127.0.0.1]) by localhost (.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id qWLCf1DUXoZO; Fri, 4 Sep 2020 14:28:19 +0000 (UTC) Received: from lists.linuxfoundation.org (lf-lists.osuosl.org [140.211.9.56]) by silver.osuosl.org (Postfix) with ESMTP id 86C332270C; Fri, 4 Sep 2020 14:28:19 +0000 (UTC) Received: from lf-lists.osuosl.org (localhost [127.0.0.1]) by lists.linuxfoundation.org (Postfix) with ESMTP id 6BD4BC0891; Fri, 4 Sep 2020 14:28:19 +0000 (UTC) X-Original-To: ovs-dev@openvswitch.org Delivered-To: ovs-dev@lists.linuxfoundation.org Received: from whitealder.osuosl.org (smtp1.osuosl.org [140.211.166.138]) by lists.linuxfoundation.org (Postfix) with ESMTP id 01D60C07FF for ; Fri, 4 Sep 2020 14:28:18 +0000 (UTC) Received: from localhost (localhost [127.0.0.1]) by whitealder.osuosl.org (Postfix) with ESMTP id E4D2480A03 for ; Fri, 4 Sep 2020 14:28:17 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from whitealder.osuosl.org ([127.0.0.1]) by localhost (.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id n+3ZF8DRKu0J for ; Fri, 4 Sep 2020 14:28:15 +0000 (UTC) X-Greylist: domain auto-whitelisted by SQLgrey-1.7.6 Received: from us-smtp-delivery-1.mimecast.com (us-smtp-2.mimecast.com [207.211.31.81]) by whitealder.osuosl.org (Postfix) with ESMTPS id 8769C86E0D for ; Fri, 4 Sep 2020 14:28:15 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1599229694; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=ibaauaPojqzvaQLAWiwIgOvSO+fVgNgTPVPUz7lR/k4=; b=Uhtn5hg3Fyu5BxUCGrrY81Zd31PniDi5YICu4WDVt4HNUhorY0N1qKxU562zwgP/pTIvpL +I8ajvMcVU4iaAZaVRdi6lrbzldk94oVXXmQhpNYh7Ptv3zDao25KqxsxEpbDKxt84b+w/ idL9oSX5prsazFUhAgAMocEQTTd3eoE= Received: from mail-ed1-f70.google.com (mail-ed1-f70.google.com [209.85.208.70]) (Using TLS) by relay.mimecast.com with ESMTP id us-mta-65-FnK5IOnOMAqQqy7lg4ZYKA-1; Fri, 04 Sep 2020 10:27:59 -0400 X-MC-Unique: FnK5IOnOMAqQqy7lg4ZYKA-1 Received: by mail-ed1-f70.google.com with SMTP id y15so2780372ede.14 for ; Fri, 04 Sep 2020 07:27:59 -0700 (PDT) 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:mime-version:content-transfer-encoding; bh=ibaauaPojqzvaQLAWiwIgOvSO+fVgNgTPVPUz7lR/k4=; b=JCHhlwVvpTjehi2rLEMO9yrBHWzYj9gXFlFB2npnGwr+5Fz3ilywE2tgkXtIpMBkTn ZplgYhi7ICZW5o0Zrjc5hIuFLObq6PoWiHF1cbd6eoRSQXqv/eEbuUU0j47OuwEm25bM YHb5ItRp9garQoRnx8O6AhgIwbwrab7g9ytDs9U3FdVuUXV6vjTGwxeYycDZZ2bkaEHX OH9CdVDuR5+g83H+C+86nlNUVFA4drxb8fqWKg1MMv1m4MSfNasP3T2mX/hrAZXukpw2 EZZw3wFg4GFeqeqC3PrWtADubAMl8NLUR95HAZyexwrCZxIaoQ0kLl9LBAyQC4GhvoFx +3/A== X-Gm-Message-State: AOAM532YyPHUUGZ/3grHsqDmIBk7V01NsVgRT3exn5L3QoHpKCm77GmI RxsTr3cvBGJ1vHFYZjZRAhs2iZQ7hsCocEtLnCXbyAWnCjbK1WdzV9xJJ/PNz7iGoMy/IW5TChW vY43CBoiJN6sWKRb7pg== X-Received: by 2002:a17:906:d9d9:: with SMTP id qk25mr7516139ejb.51.1599229677752; Fri, 04 Sep 2020 07:27:57 -0700 (PDT) X-Google-Smtp-Source: ABdhPJywrIyzqotMThDGpdPLP+KyEUxEkjIrPw2NCBZdSHyTCtuh4zpAHdTb5B0EEECBv+4Y1wF9Ng== X-Received: by 2002:a17:906:d9d9:: with SMTP id qk25mr7516096ejb.51.1599229677368; Fri, 04 Sep 2020 07:27:57 -0700 (PDT) Received: from magray.users.ipa.redhat.com ([64.43.139.134]) by smtp.gmail.com with ESMTPSA id a8sm6464538ejx.73.2020.09.04.07.27.56 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 04 Sep 2020 07:27:56 -0700 (PDT) From: Mark Gray To: ovs-dev@openvswitch.org Date: Fri, 4 Sep 2020 15:27:38 +0100 Message-Id: <20200904142739.26353-2-mark.d.gray@redhat.com> X-Mailer: git-send-email 2.26.2 In-Reply-To: <20200904142739.26353-1-mark.d.gray@redhat.com> References: <20200904142739.26353-1-mark.d.gray@redhat.com> MIME-Version: 1.0 Authentication-Results: relay.mimecast.com; auth=pass smtp.auth=CUSA124A263 smtp.mailfrom=mark.d.gray@redhat.com X-Mimecast-Spam-Score: 0.001 X-Mimecast-Originator: redhat.com Cc: Flavio Leitner , Matteo Croce , David Ahern Subject: [ovs-dev] [PATCH v5 2/3] dpif-netlink: distribute polling to discreet handlers X-BeenThere: ovs-dev@openvswitch.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: ovs-dev-bounces@openvswitch.org Sender: "dev" From: Aaron Conole Currently, the channel handlers are polled globally. On some systems, this causes a thundering herd issue where multiple handler threads become active, only to do no work and immediately sleep. The approach here is to push the netlink socket channels to discreet handler threads to process, rather than polling on every thread. This will eliminate the need to wake multiple threads. To check: ip netns add left ip netns add right ip link add center-left type veth peer name left0 netns left ip link add center-right type veth peer name right0 netns right ip link set center-left up ip link set center-right up ip -n left ip link set left0 up ip -n left ip addr add 172.31.110.10/24 dev left0 ip -n right ip link set right0 up ip -n right ip addr add 172.31.110.11/24 dev right0 ovs-vsctl add-br br0 ovs-vsctl add-port br0 center-right ovs-vsctl add-port br0 center-left # in one terminal perf record -e sched:sched_wakeup,irq:softirq_entry -ag # in a separate terminal ip netns exec left arping -I left0 -c 1 172.31.110.11 # in the perf terminal after exiting perf script Look for the number of 'handler' threads which were made active. Suggested-by: Ben Pfaff Co-authored-by: Mark Gray Reported-by: David Ahern Reported-at: https://mail.openvswitch.org/pipermail/ovs-dev/2019-December/365857.html Cc: Matteo Croce Cc: Flavio Leitner Fixes: 69c51582f ("dpif-netlink: don't allocate per thread netlink sockets") Signed-off-by: Aaron Conole Signed-off-by: Mark Gray --- v2: Oops - forgot to commit my whitespace cleanups. v3: one port latency results as per Matteo's comments Stock: min/avg/max/mdev = 21.5/36.5/96.5/1.0 us With Patch: min/avg/max/mdev = 5.3/9.7/98.4/0.5 us v4: Oops - first email did not send v5: min/avg/max/mdev = 9.3/10.4/33.6/2.2 us lib/dpif-netlink.c | 319 +++++++++++++++++++++++++-------------------- 1 file changed, 177 insertions(+), 142 deletions(-) diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c index 7da4fb54d..0e38f697c 100644 --- a/lib/dpif-netlink.c +++ b/lib/dpif-netlink.c @@ -37,6 +37,7 @@ #include "dpif-provider.h" #include "fat-rwlock.h" #include "flow.h" +#include "id-pool.h" #include "netdev-linux.h" #include "netdev-offload.h" #include "netdev-provider.h" @@ -160,6 +161,9 @@ static void dpif_netlink_flow_to_dpif_flow(struct dpif_flow *, /* One of the dpif channels between the kernel and userspace. */ struct dpif_channel { + struct hmap_node hmap_node; + struct dpif_handler *handler; + uint32_t port_idx; struct nl_sock *sock; /* Netlink socket. */ long long int last_poll; /* Last time this channel was polled. */ }; @@ -183,6 +187,8 @@ struct dpif_handler { int n_events; /* Num events returned by epoll_wait(). */ int event_offset; /* Offset into 'epoll_events'. */ + struct hmap channels; /* map of channels for each port. */ + #ifdef _WIN32 /* Pool of sockets. */ struct dpif_windows_vport_sock *vport_sock_pool; @@ -201,9 +207,8 @@ struct dpif_netlink { struct fat_rwlock upcall_lock; struct dpif_handler *handlers; uint32_t n_handlers; /* Num of upcall handlers. */ - struct dpif_channel *channels; /* Array of channels for each port. */ - int uc_array_size; /* Size of 'handler->channels' and */ - /* 'handler->epoll_events'. */ + + struct id_pool *port_ids; /* Set of added port ids */ /* Change notification. */ struct nl_sock *port_notifier; /* vport multicast group subscriber. */ @@ -287,6 +292,27 @@ close_nl_sock(struct nl_sock *sock) #endif } +static size_t +dpif_handler_channels_count(const struct dpif_handler *handler) +{ + return hmap_count(&handler->channels); +} + +static struct dpif_channel * +dpif_handler_channel_find(struct dpif_handler *handler, uint32_t idx) +{ + struct dpif_channel *channel; + + HMAP_FOR_EACH_WITH_HASH (channel, hmap_node, hash_int(idx, 0), + &handler->channels) { + if (channel->port_idx == idx) { + return channel; + } + } + + return NULL; +} + static struct dpif_netlink * dpif_netlink_cast(const struct dpif *dpif) { @@ -385,6 +411,8 @@ open_dpif(const struct dpif_netlink_dp *dp, struct dpif **dpifp) dpif->dp_ifindex = dp->dp_ifindex; dpif->user_features = dp->user_features; + dpif->port_ids = id_pool_create(0, MAX_PORTS); + *dpifp = &dpif->dpif; return 0; @@ -452,161 +480,139 @@ static bool vport_get_pid(struct dpif_netlink *dpif, uint32_t port_idx, uint32_t *upcall_pid) { - /* Since the nl_sock can only be assigned in either all - * or none "dpif" channels, the following check - * would suffice. */ - if (!dpif->channels[port_idx].sock) { + size_t i; + ovs_assert(!WINDOWS || dpif->n_handlers <= 1); + + if (!id_pool_has_id(dpif->port_ids, port_idx)) { return false; } - ovs_assert(!WINDOWS || dpif->n_handlers <= 1); - *upcall_pid = nl_sock_pid(dpif->channels[port_idx].sock); + /* The 'port_idx' should only be valid for a single handler. */ + for (i = 0; i < dpif->n_handlers; i++) { + struct dpif_channel *channel = + dpif_handler_channel_find(&dpif->handlers[i], port_idx); - return true; + if (channel) { + *upcall_pid = nl_sock_pid(channel->sock); + return true; + } + } + + return false; } static int vport_add_channel(struct dpif_netlink *dpif, odp_port_t port_no, struct nl_sock *sock) { - struct epoll_event event; uint32_t port_idx = odp_to_u32(port_no); + struct dpif_channel *channel; + struct dpif_handler *handler; + struct epoll_event event; size_t i; - int error; - if (dpif->handlers == NULL) { + if (dpif->handlers == NULL || + id_pool_has_id(dpif->port_ids, port_idx)) { close_nl_sock(sock); - return 0; + return EINVAL; } - /* We assume that the datapath densely chooses port numbers, which can - * therefore be used as an index into 'channels' and 'epoll_events' of - * 'dpif'. */ - if (port_idx >= dpif->uc_array_size) { - uint32_t new_size = port_idx + 1; - - if (new_size > MAX_PORTS) { - VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big", - dpif_name(&dpif->dpif), port_no); - return EFBIG; - } + if (port_idx >= MAX_PORTS) { + VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big", + dpif_name(&dpif->dpif), port_no); + return EFBIG; + } - dpif->channels = xrealloc(dpif->channels, - new_size * sizeof *dpif->channels); + handler = &dpif->handlers[0]; - for (i = dpif->uc_array_size; i < new_size; i++) { - dpif->channels[i].sock = NULL; + for (i = 0; i < dpif->n_handlers; i++) { + if (dpif_handler_channels_count(&dpif->handlers[i]) < + dpif_handler_channels_count(handler)) { + handler = &dpif->handlers[i]; } + } - for (i = 0; i < dpif->n_handlers; i++) { - struct dpif_handler *handler = &dpif->handlers[i]; - - handler->epoll_events = xrealloc(handler->epoll_events, - new_size * sizeof *handler->epoll_events); + channel = xmalloc(sizeof *channel); + channel->port_idx = port_idx; + channel->sock = sock; + channel->last_poll = LLONG_MIN; + channel->handler = handler; + hmap_insert(&handler->channels, &channel->hmap_node, hash_int(port_idx, 0)); - } - dpif->uc_array_size = new_size; - } + handler->epoll_events = xrealloc(handler->epoll_events, + dpif_handler_channels_count(handler) * + sizeof *handler->epoll_events); memset(&event, 0, sizeof event); event.events = EPOLLIN | EPOLLEXCLUSIVE; event.data.u32 = port_idx; - for (i = 0; i < dpif->n_handlers; i++) { - struct dpif_handler *handler = &dpif->handlers[i]; - -#ifndef _WIN32 - if (epoll_ctl(handler->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), - &event) < 0) { - error = errno; - goto error; - } -#endif - } - dpif->channels[port_idx].sock = sock; - dpif->channels[port_idx].last_poll = LLONG_MIN; - - return 0; - -error: #ifndef _WIN32 - while (i--) { - epoll_ctl(dpif->handlers[i].epoll_fd, EPOLL_CTL_DEL, - nl_sock_fd(sock), NULL); + if (epoll_ctl(handler->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock), + &event) < 0) { + hmap_remove(&handler->channels, &channel->hmap_node); + free(channel); + id_pool_free_id(dpif->port_ids, port_idx); + return errno; } #endif - dpif->channels[port_idx].sock = NULL; - return error; + id_pool_add(dpif->port_ids, port_idx); + return 0; } static void -vport_del_channels(struct dpif_netlink *dpif, odp_port_t port_no) +vport_del_channel(struct dpif_netlink *dpif, + struct dpif_channel *channel) { - uint32_t port_idx = odp_to_u32(port_no); - size_t i; - - if (!dpif->handlers || port_idx >= dpif->uc_array_size - || !dpif->channels[port_idx].sock) { - return; - } - - for (i = 0; i < dpif->n_handlers; i++) { - struct dpif_handler *handler = &dpif->handlers[i]; + if (channel) { #ifndef _WIN32 - epoll_ctl(handler->epoll_fd, EPOLL_CTL_DEL, - nl_sock_fd(dpif->channels[port_idx].sock), NULL); + epoll_ctl(channel->handler->epoll_fd, EPOLL_CTL_DEL, + nl_sock_fd(channel->sock), NULL); + nl_sock_destroy(channel->sock); #endif - handler->event_offset = handler->n_events = 0; + hmap_remove(&channel->handler->channels, &channel->hmap_node); + id_pool_free_id(dpif->port_ids, channel->port_idx); + channel->handler->event_offset = channel->handler->n_events = 0; + free(channel); } -#ifndef _WIN32 - nl_sock_destroy(dpif->channels[port_idx].sock); -#endif - dpif->channels[port_idx].sock = NULL; } static void destroy_all_channels(struct dpif_netlink *dpif) OVS_REQ_WRLOCK(dpif->upcall_lock) { - unsigned int i; + size_t i; if (!dpif->handlers) { return; } - for (i = 0; i < dpif->uc_array_size; i++ ) { - struct dpif_netlink_vport vport_request; - uint32_t upcall_pids = 0; - - if (!dpif->channels[i].sock) { - continue; - } - - /* Turn off upcalls. */ - dpif_netlink_vport_init(&vport_request); - vport_request.cmd = OVS_VPORT_CMD_SET; - vport_request.dp_ifindex = dpif->dp_ifindex; - vport_request.port_no = u32_to_odp(i); - vport_request.n_upcall_pids = 1; - vport_request.upcall_pids = &upcall_pids; - dpif_netlink_vport_transact(&vport_request, NULL, NULL); - - vport_del_channels(dpif, u32_to_odp(i)); - } - for (i = 0; i < dpif->n_handlers; i++) { struct dpif_handler *handler = &dpif->handlers[i]; + struct dpif_channel *channel, *next; + + HMAP_FOR_EACH_SAFE (channel, next, hmap_node, &handler->channels) { + struct dpif_netlink_vport vport_request; + uint32_t upcall_pids = 0; + /* Turn off upcalls. */ + dpif_netlink_vport_init(&vport_request); + vport_request.cmd = OVS_VPORT_CMD_SET; + vport_request.dp_ifindex = dpif->dp_ifindex; + vport_request.port_no = u32_to_odp(channel->port_idx); + vport_request.n_upcall_pids = 1; + vport_request.upcall_pids = &upcall_pids; + dpif_netlink_vport_transact(&vport_request, NULL, NULL); + + vport_del_channel(dpif, channel); + } dpif_netlink_handler_uninit(handler); free(handler->epoll_events); } - free(dpif->channels); - free(dpif->handlers); + dpif->handlers = NULL; - dpif->channels = NULL; dpif->n_handlers = 0; - dpif->uc_array_size = 0; } static void @@ -618,9 +624,11 @@ dpif_netlink_close(struct dpif *dpif_) fat_rwlock_wrlock(&dpif->upcall_lock); destroy_all_channels(dpif); + id_pool_destroy(dpif->port_ids); fat_rwlock_unlock(&dpif->upcall_lock); fat_rwlock_destroy(&dpif->upcall_lock); + free(dpif); } @@ -981,8 +989,10 @@ dpif_netlink_port_del__(struct dpif_netlink *dpif, odp_port_t port_no) OVS_REQ_WRLOCK(dpif->upcall_lock) { struct dpif_netlink_vport vport; + struct dpif_channel *channel; struct dpif_port dpif_port; int error; + size_t i; error = dpif_netlink_port_query__(dpif, port_no, NULL, &dpif_port); if (error) { @@ -1003,7 +1013,18 @@ dpif_netlink_port_del__(struct dpif_netlink *dpif, odp_port_t port_no) #endif error = dpif_netlink_vport_transact(&vport, NULL, NULL); - vport_del_channels(dpif, port_no); + for (i = 0; i < dpif->n_handlers; i++) { + channel = dpif_handler_channel_find(&dpif->handlers[i], + odp_to_u32(port_no)); + if (channel) { + vport_del_channel(dpif, channel); + break; + } + } + + if (!channel) { + return EINVAL; + } if (!error && !ovs_tunnels_out_of_tree) { error = dpif_netlink_rtnl_port_destroy(dpif_port.name, dpif_port.type); @@ -1084,23 +1105,39 @@ dpif_netlink_port_get_pid__(const struct dpif_netlink *dpif, odp_port_t port_no) OVS_REQ_RDLOCK(dpif->upcall_lock) { + struct dpif_channel *min_channel = NULL, *found_channel = NULL; uint32_t port_idx = odp_to_u32(port_no); uint32_t pid = 0; + size_t i; + + /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s + * channel, since it is not heavily loaded. */ + uint32_t idx = port_no == ODPP_NONE ? 0 : port_idx; - if (dpif->handlers && dpif->uc_array_size > 0) { - /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s - * channel, since it is not heavily loaded. */ - uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx; - - /* Needs to check in case the socket pointer is changed in between - * the holding of upcall_lock. A known case happens when the main - * thread deletes the vport while the handler thread is handling - * the upcall from that port. */ - if (dpif->channels[idx].sock) { - pid = nl_sock_pid(dpif->channels[idx].sock); + if (!id_pool_has_id(dpif->port_ids, port_idx)) { + return 0; + } + + if (dpif->handlers) { + for (i = 0; i < dpif->n_handlers; i++) { + if (dpif_handler_channel_find(&dpif->handlers[i], idx)) { + found_channel = dpif_handler_channel_find(&dpif->handlers[i], + idx); + break; + } + + if (dpif_handler_channel_find(&dpif->handlers[i], 0)) { + min_channel = dpif_handler_channel_find(&dpif->handlers[i], 0); + } } } + if (found_channel) { + pid = nl_sock_pid(found_channel->sock); + } else if (min_channel) { + pid = nl_sock_pid(min_channel->sock); + } + return pid; } @@ -2342,12 +2379,14 @@ dpif_netlink_operate(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops, static void dpif_netlink_handler_uninit(struct dpif_handler *handler) { + hmap_destroy(&handler->channels); vport_delete_sock_pool(handler); } static int dpif_netlink_handler_init(struct dpif_handler *handler) { + hmap_init(&handler->channels); return vport_create_sock_pool(handler); } #else @@ -2355,6 +2394,7 @@ dpif_netlink_handler_init(struct dpif_handler *handler) static int dpif_netlink_handler_init(struct dpif_handler *handler) { + hmap_init(&handler->channels); handler->epoll_fd = epoll_create(10); return handler->epoll_fd < 0 ? errno : 0; } @@ -2362,6 +2402,7 @@ dpif_netlink_handler_init(struct dpif_handler *handler) static void dpif_netlink_handler_uninit(struct dpif_handler *handler) { + hmap_destroy(&handler->channels); close(handler->epoll_fd); } #endif @@ -2374,12 +2415,11 @@ static int dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) OVS_REQ_WRLOCK(dpif->upcall_lock) { - unsigned long int *keep_channels; struct dpif_netlink_vport vport; - size_t keep_channels_nbits; struct nl_dump dump; uint64_t reply_stub[NL_DUMP_BUFSIZE / 8]; struct ofpbuf buf; + struct dpif_channel *channel; int retval = 0; size_t i; @@ -2412,13 +2452,9 @@ dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) for (i = 0; i < n_handlers; i++) { struct dpif_handler *handler = &dpif->handlers[i]; - handler->event_offset = handler->n_events = 0; } - keep_channels_nbits = dpif->uc_array_size; - keep_channels = bitmap_allocate(keep_channels_nbits); - ofpbuf_use_stub(&buf, reply_stub, sizeof reply_stub); dpif_netlink_port_dump_start__(dpif, &dump); while (!dpif_netlink_port_dump_next__(dpif, &dump, &vport, &buf)) { @@ -2426,8 +2462,7 @@ dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) uint32_t upcall_pid; int error; - if (port_no >= dpif->uc_array_size - || !vport_get_pid(dpif, port_no, &upcall_pid)) { + if (!vport_get_pid(dpif, port_no, &upcall_pid)) { struct nl_sock *sock; error = create_nl_sock(dpif, &sock); @@ -2475,25 +2510,17 @@ dpif_netlink_refresh_channels(struct dpif_netlink *dpif, uint32_t n_handlers) } } - if (port_no < keep_channels_nbits) { - bitmap_set1(keep_channels, port_no); - } continue; error: - vport_del_channels(dpif, vport.port_no); + for (i = 0; i < n_handlers; i++) { + channel = dpif_handler_channel_find(&dpif->handlers[i], port_no); + vport_del_channel(dpif, channel); + } } nl_dump_done(&dump); ofpbuf_uninit(&buf); - /* Discard any saved channels that we didn't reuse. */ - for (i = 0; i < keep_channels_nbits; i++) { - if (!bitmap_is_set(keep_channels, i)) { - vport_del_channels(dpif, u32_to_odp(i)); - } - } - free(keep_channels); - return retval; } @@ -2714,6 +2741,8 @@ dpif_netlink_recv__(struct dpif_netlink *dpif, uint32_t handler_id, OVS_REQ_RDLOCK(dpif->upcall_lock) { struct dpif_handler *handler; + size_t n_channels; + int read_tries = 0; if (!dpif->handlers || handler_id >= dpif->n_handlers) { @@ -2721,14 +2750,15 @@ dpif_netlink_recv__(struct dpif_netlink *dpif, uint32_t handler_id, } handler = &dpif->handlers[handler_id]; - if (handler->event_offset >= handler->n_events) { + n_channels = dpif_handler_channels_count(handler); + if (handler->event_offset >= handler->n_events && n_channels) { int retval; handler->event_offset = handler->n_events = 0; do { retval = epoll_wait(handler->epoll_fd, handler->epoll_events, - dpif->uc_array_size, 0); + n_channels, 0); } while (retval < 0 && errno == EINTR); if (retval < 0) { @@ -2741,7 +2771,10 @@ dpif_netlink_recv__(struct dpif_netlink *dpif, uint32_t handler_id, while (handler->event_offset < handler->n_events) { int idx = handler->epoll_events[handler->event_offset].data.u32; - struct dpif_channel *ch = &dpif->channels[idx]; + struct dpif_channel *ch = dpif_handler_channel_find(handler, idx); + if (!ch) { + return EAGAIN; + } handler->event_offset++; @@ -2845,12 +2878,14 @@ dpif_netlink_recv_purge__(struct dpif_netlink *dpif) if (dpif->handlers) { size_t i; - if (!dpif->channels[0].sock) { - return; - } - for (i = 0; i < dpif->uc_array_size; i++ ) { + for (i = 0; i < dpif->n_handlers; i++) { + struct dpif_handler *handler = &dpif->handlers[i]; + struct dpif_channel *channel; + + HMAP_FOR_EACH (channel, hmap_node, &handler->channels) { + nl_sock_drain(channel->sock); + } - nl_sock_drain(dpif->channels[i].sock); } } }