From nobody Tue Feb 15 08:03:52 2022 X-Original-To: dev-commits-src-main@mlmmj.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mlmmj.nyi.freebsd.org (Postfix) with ESMTP id 7842819BA196; Tue, 15 Feb 2022 08:03:52 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from mxrelay.nyi.freebsd.org (mxrelay.nyi.freebsd.org [IPv6:2610:1c1:1:606c::19:3]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256 client-signature RSA-PSS (4096 bits) client-digest SHA256) (Client CN "mxrelay.nyi.freebsd.org", Issuer "R3" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 4JyYWX2xQ1z3jvr; Tue, 15 Feb 2022 08:03:52 +0000 (UTC) (envelope-from git@FreeBSD.org) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1644912232; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=LPC6KVz+mWBHSiXhTx2PsJ5l7IDSCWqpBB3Y6zZgcIA=; b=poolv/5p7bAp8MAk4SCFmy61QLjf36s4YdZ55UJyi7spB2vehyWF0XdUziNbku1jN2HwNb ChsyRO9QFZ55xdkCFxFmR3aQGm/Fcd6QErf5D2T3MOiHHjKrRV/Veg9V3G+gJnaRc0/BAj +xZvUx4qMeLpEJvDSqx/KVcnIVy9S1wYcJjU0Hf96MaYVJwRvwXRRWjsn5F/rQDWFV+lz6 2l3JnPLykVbt+5A7WKoz6yv6dVr+eAeme/gxYLGNkUJakqwvRI/U28T5Kd2P9qcDYQWUeQ M38OZWmGExRzak6IcO8ruHDOKd8OrYroPAqb7XCFw2/QQjO3KpMjuvET33/+Vw== Received: from gitrepo.freebsd.org (gitrepo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:5]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (Client did not present a certificate) by mxrelay.nyi.freebsd.org (Postfix) with ESMTPS id 448901EAC9; Tue, 15 Feb 2022 08:03:52 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from gitrepo.freebsd.org ([127.0.1.44]) by gitrepo.freebsd.org (8.16.1/8.16.1) with ESMTP id 21F83qi3029498; Tue, 15 Feb 2022 08:03:52 GMT (envelope-from git@gitrepo.freebsd.org) Received: (from git@localhost) by gitrepo.freebsd.org (8.16.1/8.16.1/Submit) id 21F83qmk029497; Tue, 15 Feb 2022 08:03:52 GMT (envelope-from git) Date: Tue, 15 Feb 2022 08:03:52 GMT Message-Id: <202202150803.21F83qmk029497@gitrepo.freebsd.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org From: Kristof Provost Subject: git: 24f0bfbad57b - main - if_epair: implement fanout List-Id: Commit messages for the main branch of the src repository List-Archive: https://lists.freebsd.org/archives/dev-commits-src-main List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-dev-commits-src-main@freebsd.org X-BeenThere: dev-commits-src-main@freebsd.org MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit X-Git-Committer: kp X-Git-Repository: src X-Git-Refname: refs/heads/main X-Git-Reftype: branch X-Git-Commit: 24f0bfbad57b9c3cb9b543a60b2ba00e4812c286 Auto-Submitted: auto-generated ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1644912232; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=LPC6KVz+mWBHSiXhTx2PsJ5l7IDSCWqpBB3Y6zZgcIA=; b=VQN7oshydDHLZwQEbhqoqFdn/GJwAkcsTAV8Gw6FhqeaWRjlszJqg5LVs+2jxrwKGl0yuO cy0XQj4izLhiGBvyiMjvCcbgSmnblYAGvZmANunSQ3nOaIAmBoixUkdZPeSIODEd60hxs2 IkQbv+ZRWynXamjAAnFCIeNikTEmEH+MgCZ5je3QzWwfNYdqL35Tp7+Y/xThfz00OwVQaY TVIXuYAhR/eyOf0HzzkIDx7GANVotug4dk8DZrVZPh4glRO9si254yLR1eff/CjmAToE0Q nwe0Nl8FGBMQMEHzvGFQJJqDEyp37NX0OixMs0IiuALKFLzqhfRQPGH6HzF1Kw== ARC-Seal: i=1; s=dkim; d=freebsd.org; t=1644912232; a=rsa-sha256; cv=none; b=fVhTMIe9UqA7K/YZMOHhmd6K7b9QiGpEBBU0coz3cmkqXJPhM0X4y47MJzSMlrsAn/h3UI 0Bj5/WUzymSmZQUxFWqe4jHMOHfpXbCyXYM6Kw8XnzjPtFAmuinAHCnFyV+Q3tSoRaA/4G YgepJhW1UWk+B9sBmR7qgmDmiFB7KOA9FLlz4fwdbJUasP6e10LM4iNa7RK3c9e1tdKElv pUYVnKq0FpF3+BXVYlU0DbDqsyFFEz/oNVvhGe9gopwvb+4/sfeF6oVp2yJug/6wtjsLoG 436X5MBoBrkwWPyBEUz9WMo5wWIdHPnk7FgViQwWJjPvgr7eMRJcQHmMaKPZyg== ARC-Authentication-Results: i=1; mx1.freebsd.org; none X-ThisMailContainsUnwantedMimeParts: N The branch main has been updated by kp: URL: https://cgit.FreeBSD.org/src/commit/?id=24f0bfbad57b9c3cb9b543a60b2ba00e4812c286 commit 24f0bfbad57b9c3cb9b543a60b2ba00e4812c286 Author: Kristof Provost AuthorDate: 2021-12-09 13:24:13 +0000 Commit: Kristof Provost CommitDate: 2022-02-15 08:03:24 +0000 if_epair: implement fanout Allow multiple cores to be used to process if_epair traffic. We do this (if RSS is enabled) based on the RSS hash of the incoming packet. This allows us to distribute the load over multiple cores, rather than sending everything to the same one. We also switch from swi_sched() to taskqueues, which also contributes to better throughput. Benchmark results: With net.isr.maxthreads=-1 Setup A: (cc0 - bridge0 - epair0a) (epair0b - bridge1 - cc1) Before 627 Kpps After (no RSS) 1.198 Mpps After (RSS) 3.148 Mpps Setup B: (cc0 - bridge0 - epaira0) (epair0b - vnet jail - epair1a) (epair1b - bridge1 - cc1) Before 7.705 Kpps After (no RSS) 1.017 Mpps After (RSS) 2.083 Mpps MFC after: 3 weeks Sponsored by: Orange Business Services Differential Revision: https://reviews.freebsd.org/D33731 --- sys/net/if_epair.c | 309 ++++++++++++++++++++++++++++++++--------------------- 1 file changed, 186 insertions(+), 123 deletions(-) diff --git a/sys/net/if_epair.c b/sys/net/if_epair.c index ff20107307ff..6cc0a1a954ef 100644 --- a/sys/net/if_epair.c +++ b/sys/net/if_epair.c @@ -40,6 +40,8 @@ #include __FBSDID("$FreeBSD$"); +#include "opt_rss.h" + #include #include #include @@ -50,10 +52,11 @@ __FBSDID("$FreeBSD$"); #include #include #include +#include #include #include #include -#include +#include #include #include #include @@ -68,6 +71,11 @@ __FBSDID("$FreeBSD$"); #include #include #include +#ifdef RSS +#include +#include +#include +#endif #include static int epair_clone_match(struct if_clone *, const char *); @@ -90,21 +98,32 @@ static unsigned int next_index = 0; #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx) #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx) -static void *swi_cookie[MAXCPU]; /* swi(9). */ -static STAILQ_HEAD(, epair_softc) swi_sc[MAXCPU]; +struct epair_softc; +struct epair_queue { + int id; + struct buf_ring *rxring[2]; + volatile int ridx; /* 0 || 1 */ + struct task tx_task; + struct epair_softc *sc; +}; static struct mtx epair_n_index_mtx; struct epair_softc { - struct ifnet *ifp; /* This ifp. */ - struct ifnet *oifp; /* other ifp of pair. */ - void *swi_cookie; /* swi(9). */ - struct buf_ring *rxring[2]; - volatile int ridx; /* 0 || 1 */ - struct ifmedia media; /* Media config (fake). */ - uint32_t cpuidx; + struct ifnet *ifp; /* This ifp. */ + struct ifnet *oifp; /* other ifp of pair. */ + int num_queues; + struct epair_queue *queues; + struct ifmedia media; /* Media config (fake). */ STAILQ_ENTRY(epair_softc) entry; }; +struct epair_tasks_t { + int tasks; + struct taskqueue *tq[MAXCPU]; +}; + +static struct epair_tasks_t epair_tasks; + static void epair_clear_mbuf(struct mbuf *m) { @@ -119,59 +138,43 @@ epair_clear_mbuf(struct mbuf *m) } static void -epair_if_input(struct epair_softc *sc, int ridx) +epair_if_input(struct epair_softc *sc, struct epair_queue *q, int ridx) { - struct epoch_tracker et; struct ifnet *ifp; struct mbuf *m; ifp = sc->ifp; - NET_EPOCH_ENTER(et); - do { - m = buf_ring_dequeue_sc(sc->rxring[ridx]); + CURVNET_SET(ifp->if_vnet); + while (! buf_ring_empty(q->rxring[ridx])) { + m = buf_ring_dequeue_mc(q->rxring[ridx]); if (m == NULL) - break; + continue; MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); (*ifp->if_input)(ifp, m); - } while (1); - NET_EPOCH_EXIT(et); + } + CURVNET_RESTORE(); } static void -epair_sintr(struct epair_softc *sc) +epair_tx_start_deferred(void *arg, int pending) { + struct epair_queue *q = (struct epair_queue *)arg; + struct epair_softc *sc = q->sc; int ridx, nidx; if_ref(sc->ifp); + ridx = atomic_load_int(&q->ridx); do { - ridx = sc->ridx; nidx = (ridx == 0) ? 1 : 0; - } while (!atomic_cmpset_int(&sc->ridx, ridx, nidx)); - epair_if_input(sc, ridx); + } while (!atomic_fcmpset_int(&q->ridx, &ridx, nidx)); + epair_if_input(sc, q, ridx); - if_rele(sc->ifp); -} + if (! buf_ring_empty(q->rxring[nidx])) + taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task); -static void -epair_intr(void *arg) -{ - struct epair_softc *sc; - uint32_t cpuidx; - - cpuidx = (uintptr_t)arg; - /* If this is a problem, this is a read-mostly situation. */ - EPAIR_LOCK(); - STAILQ_FOREACH(sc, &swi_sc[cpuidx], entry) { - /* Do this lockless. */ - if (buf_ring_empty(sc->rxring[sc->ridx])) - continue; - epair_sintr(sc); - } - EPAIR_UNLOCK(); - - return; + if_rele(sc->ifp); } static int @@ -181,7 +184,12 @@ epair_menq(struct mbuf *m, struct epair_softc *osc) int len, ret; int ridx; short mflags; + struct epair_queue *q = NULL; + uint32_t bucket; bool was_empty; +#ifdef RSS + struct ether_header *eh; +#endif /* * I know this looks weird. We pass the "other sc" as we need that one @@ -202,13 +210,38 @@ epair_menq(struct mbuf *m, struct epair_softc *osc) MPASS(m->m_nextpkt == NULL); MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); - ridx = atomic_load_int(&osc->ridx); - was_empty = buf_ring_empty(osc->rxring[ridx]); - ret = buf_ring_enqueue(osc->rxring[ridx], m); +#ifdef RSS + ret = rss_m2bucket(m, &bucket); + if (ret) { + /* Actually hash the packet. */ + eh = mtod(m, struct ether_header *); + + switch (ntohs(eh->ether_type)) { + case ETHERTYPE_IP: + rss_soft_m2cpuid_v4(m, 0, &bucket); + break; + case ETHERTYPE_IPV6: + rss_soft_m2cpuid_v6(m, 0, &bucket); + break; + default: + bucket = 0; + break; + } + } + bucket %= osc->num_queues; +#else + bucket = 0; +#endif + q = &osc->queues[bucket]; + + ridx = atomic_load_int(&q->ridx); + was_empty = buf_ring_empty(q->rxring[ridx]); + ret = buf_ring_enqueue(q->rxring[ridx], m); if (ret != 0) { /* Ring is full. */ + if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); m_freem(m); - return (0); + goto done; } if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); @@ -223,9 +256,9 @@ epair_menq(struct mbuf *m, struct epair_softc *osc) /* Someone else received the packet. */ if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); - /* Kick the interrupt handler for the first packet. */ - if (was_empty && osc->swi_cookie != NULL) - swi_sched(osc->swi_cookie, 0); +done: + if (was_empty) + taskqueue_enqueue(epair_tasks.tq[bucket], &q->tx_task); return (0); } @@ -495,16 +528,27 @@ epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) /* Allocate memory for both [ab] interfaces */ sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); sca->ifp = if_alloc(IFT_ETHER); + sca->num_queues = epair_tasks.tasks; if (sca->ifp == NULL) { free(sca, M_EPAIR); ifc_free_unit(ifc, unit); return (ENOSPC); } - sca->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK,NULL); - sca->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + sca->queues = mallocarray(sca->num_queues, sizeof(struct epair_queue), + M_EPAIR, M_WAITOK); + for (int i = 0; i < sca->num_queues; i++) { + struct epair_queue *q = &sca->queues[i]; + q->id = i; + q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + q->ridx = 0; + q->sc = sca; + NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q); + } scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); scb->ifp = if_alloc(IFT_ETHER); + scb->num_queues = epair_tasks.tasks; if (scb->ifp == NULL) { free(scb, M_EPAIR); if_free(sca->ifp); @@ -512,8 +556,17 @@ epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) ifc_free_unit(ifc, unit); return (ENOSPC); } - scb->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); - scb->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + scb->queues = mallocarray(scb->num_queues, sizeof(struct epair_queue), + M_EPAIR, M_WAITOK); + for (int i = 0; i < scb->num_queues; i++) { + struct epair_queue *q = &scb->queues[i]; + q->id = i; + q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); + q->ridx = 0; + q->sc = scb; + NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q); + } /* * Cross-reference the interfaces so we will be able to free both. @@ -528,41 +581,6 @@ epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) #else hash = 0; #endif - if (swi_cookie[hash] == NULL) { - void *cookie; - - EPAIR_UNLOCK(); - error = swi_add(NULL, epairname, - epair_intr, (void *)(uintptr_t)hash, - SWI_NET, INTR_MPSAFE, &cookie); - if (error) { - buf_ring_free(scb->rxring[0], M_EPAIR); - buf_ring_free(scb->rxring[1], M_EPAIR); - if_free(scb->ifp); - free(scb, M_EPAIR); - buf_ring_free(sca->rxring[0], M_EPAIR); - buf_ring_free(sca->rxring[1], M_EPAIR); - if_free(sca->ifp); - free(sca, M_EPAIR); - ifc_free_unit(ifc, unit); - return (ENOSPC); - } - EPAIR_LOCK(); - /* Recheck under lock even though a race is very unlikely. */ - if (swi_cookie[hash] == NULL) { - swi_cookie[hash] = cookie; - } else { - EPAIR_UNLOCK(); - (void) swi_remove(cookie); - EPAIR_LOCK(); - } - } - sca->cpuidx = hash; - STAILQ_INSERT_TAIL(&swi_sc[hash], sca, entry); - sca->swi_cookie = swi_cookie[hash]; - scb->cpuidx = hash; - STAILQ_INSERT_TAIL(&swi_sc[hash], scb, entry); - scb->swi_cookie = swi_cookie[hash]; EPAIR_UNLOCK(); /* Initialise pseudo media types. */ @@ -669,12 +687,15 @@ epair_drain_rings(struct epair_softc *sc) struct mbuf *m; for (ridx = 0; ridx < 2; ridx++) { - do { - m = buf_ring_dequeue_sc(sc->rxring[ridx]); - if (m == NULL) - break; - m_freem(m); - } while (1); + for (int i = 0; i < sc->num_queues; i++) { + struct epair_queue *q = &sc->queues[i]; + do { + m = buf_ring_dequeue_sc(q->rxring[ridx]); + if (m == NULL) + break; + m_freem(m); + } while (1); + } } } @@ -707,14 +728,6 @@ epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) ether_ifdetach(ifp); ether_ifdetach(oifp); - /* Second stop interrupt handler. */ - EPAIR_LOCK(); - STAILQ_REMOVE(&swi_sc[sca->cpuidx], sca, epair_softc, entry); - STAILQ_REMOVE(&swi_sc[scb->cpuidx], scb, epair_softc, entry); - EPAIR_UNLOCK(); - sca->swi_cookie = NULL; - scb->swi_cookie = NULL; - /* Third free any queued packets and all the resources. */ CURVNET_SET_QUIET(oifp->if_vnet); epair_drain_rings(scb); @@ -725,16 +738,24 @@ epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) __func__, error); if_free(oifp); ifmedia_removeall(&scb->media); - buf_ring_free(scb->rxring[0], M_EPAIR); - buf_ring_free(scb->rxring[1], M_EPAIR); + for (int i = 0; i < scb->num_queues; i++) { + struct epair_queue *q = &scb->queues[i]; + buf_ring_free(q->rxring[0], M_EPAIR); + buf_ring_free(q->rxring[1], M_EPAIR); + } + free(scb->queues, M_EPAIR); free(scb, M_EPAIR); CURVNET_RESTORE(); epair_drain_rings(sca); if_free(ifp); ifmedia_removeall(&sca->media); - buf_ring_free(sca->rxring[0], M_EPAIR); - buf_ring_free(sca->rxring[1], M_EPAIR); + for (int i = 0; i < sca->num_queues; i++) { + struct epair_queue *q = &sca->queues[i]; + buf_ring_free(q->rxring[0], M_EPAIR); + buf_ring_free(q->rxring[1], M_EPAIR); + } + free(sca->queues, M_EPAIR); free(sca, M_EPAIR); /* Last free the cloner unit. */ @@ -762,34 +783,76 @@ vnet_epair_uninit(const void *unused __unused) VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_epair_uninit, NULL); +static int +epair_mod_init() +{ + char name[32]; + epair_tasks.tasks = 0; + +#ifdef RSS + struct pcpu *pcpu; + int cpu; + + CPU_FOREACH(cpu) { + cpuset_t cpu_mask; + + /* Pin to this CPU so we get appropriate NUMA allocations. */ + pcpu = pcpu_find(cpu); + thread_lock(curthread); + sched_bind(curthread, cpu); + thread_unlock(curthread); + + snprintf(name, sizeof(name), "epair_task_%d", cpu); + + epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK, + taskqueue_thread_enqueue, + &epair_tasks.tq[cpu]); + CPU_SETOF(cpu, &cpu_mask); + taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET, + &cpu_mask, "%s", name); + + epair_tasks.tasks++; + } +#else + snprintf(name, sizeof(name), "epair_task"); + + epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK, + taskqueue_thread_enqueue, + &epair_tasks.tq[0]); + taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name); + + epair_tasks.tasks = 1; +#endif + + return (0); +} + +static void +epair_mod_cleanup() +{ + + for (int i = 0; i < epair_tasks.tasks; i++) { + taskqueue_drain_all(epair_tasks.tq[i]); + taskqueue_free(epair_tasks.tq[i]); + } +} + static int epair_modevent(module_t mod, int type, void *data) { - int i; + int ret; switch (type) { case MOD_LOAD: - for (i = 0; i < MAXCPU; i++) { - swi_cookie[i] = NULL; - STAILQ_INIT(&swi_sc[i]); - } EPAIR_LOCK_INIT(); + ret = epair_mod_init(); + if (ret != 0) + return (ret); if (bootverbose) printf("%s: %s initialized.\n", __func__, epairname); break; case MOD_UNLOAD: - EPAIR_LOCK(); - for (i = 0; i < MAXCPU; i++) { - if (!STAILQ_EMPTY(&swi_sc[i])) { - printf("%s: swi_sc[%d] active\n", __func__, i); - EPAIR_UNLOCK(); - return (EBUSY); - } - } - EPAIR_UNLOCK(); - for (i = 0; i < MAXCPU; i++) - if (swi_cookie[i] != NULL) - (void) swi_remove(swi_cookie[i]); + epair_mod_cleanup(); EPAIR_LOCK_DESTROY(); if (bootverbose) printf("%s: %s unloaded.\n", __func__, epairname);