From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-15.8 required=3.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FORGED_FROMDOMAIN,FREEMAIL_FROM, HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_CR_TRAILER,INCLUDES_PATCH, MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS,USER_AGENT_GIT autolearn=unavailable autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 92ACAC47080 for ; Tue, 1 Jun 2021 06:23:37 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 76DA361159 for ; Tue, 1 Jun 2021 06:23:37 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S233114AbhFAGZQ (ORCPT ); Tue, 1 Jun 2021 02:25:16 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:56500 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S233026AbhFAGZF (ORCPT ); Tue, 1 Jun 2021 02:25:05 -0400 Received: from mail-pl1-x62b.google.com (mail-pl1-x62b.google.com [IPv6:2607:f8b0:4864:20::62b]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 114B2C06175F; Mon, 31 May 2021 23:23:24 -0700 (PDT) Received: by mail-pl1-x62b.google.com with SMTP id e7so4241435plj.7; Mon, 31 May 2021 23:23:24 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=7SY6QzuWUSlWj33T26/GRZSVwi+Ld4IHiibBoCb52p4=; b=ePlorr20RYdU83BpwKX/s0RfyvSJGLZuiSfUt1Q5Tf+Uj6s0wJkGx8MFmTlzTq06Ly /QdSin1Qqs2l0J37C68YVuBoCY22xVUnKkbWT5c/25xPZUE4hwRl5CveuKHLOW9KwoLz WMUnhDGfB+gMrKJPdIvTLtMDs5EgLCysNJyB17ago3WfOtWSwQRzdwE9GUfx+u24xoXb YvUh+XKO7EACTncuCcfNhm9MUjn8I6F0IEFKp7B02dAPD0YE4xz1HbQBw17+5uziSWyH 6Hbt03PuVPQDsas/U9JfW6GikTKlAqSmRjtpSQ956IQanhxIOkALHMsFIlrry/0xO4L8 Jh1g== 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=7SY6QzuWUSlWj33T26/GRZSVwi+Ld4IHiibBoCb52p4=; b=FszMTsbjZf9sRRvnOKTdBb3sMsCMM1OUoRCJBgaY+wRshGR6tplPA1+2GpsFuBR+x0 bLHaMYNFHtTpNs0FH2I+RDDRqQq6OohlAfJGAIs2vm7f/mKUKil7Ih77SvSuAaXAGvae CynEOScc2XprzwkbThw1uWJikyXVvnGvMhBwtnVwcI3lF1e7ToPU6uF3ysPxLQaFQZmy 2hIQDG2BNI+UtlTe+mUyumGnH5e/EQo5Z6pfh+RlNnWj6GcAAZbsjgb6p+A2Xckp99Dq 218KsYthVpNY3qN8fltYpxj6nrU5FeaB5e/wvMfoVuF9y/vjkAUhrgZlCNFOJpiM907b Ju8A== X-Gm-Message-State: AOAM532Swguts18l4s/0CkOM/ttU2iMI1lkEau1dgl2m+jJgG8D98fxJ gDMRmgYWECgeNsPSivxyi3Q= X-Google-Smtp-Source: ABdhPJw09TNZqBmKx3WQDrsKTZlhlpstv4mz2E1lJZnWicTzFCQBo5ZLvZlMnKhALwZzNq6+juRuaA== X-Received: by 2002:a17:90a:5106:: with SMTP id t6mr3315886pjh.216.1622528603683; Mon, 31 May 2021 23:23:23 -0700 (PDT) Received: from bobo.ibm.com (60-241-69-122.static.tpgi.com.au. [60.241.69.122]) by smtp.gmail.com with ESMTPSA id h1sm12519100pfh.72.2021.05.31.23.23.19 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 31 May 2021 23:23:23 -0700 (PDT) From: Nicholas Piggin To: Andrew Morton Cc: Nicholas Piggin , Randy Dunlap , linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, linuxppc-dev@lists.ozlabs.org, linux-mm@kvack.org, Anton Blanchard , Andy Lutomirski Subject: [PATCH v3 3/4] lazy tlb: shoot lazies, a non-refcounting lazy tlb option Date: Tue, 1 Jun 2021 16:23:02 +1000 Message-Id: <20210601062303.3932513-4-npiggin@gmail.com> X-Mailer: git-send-email 2.23.0 In-Reply-To: <20210601062303.3932513-1-npiggin@gmail.com> References: <20210601062303.3932513-1-npiggin@gmail.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On big systems, the mm refcount can become highly contented when doing a lot of context switching with threaded applications (particularly switching between the idle thread and an application thread). Abandoning lazy tlb slows switching down quite a bit in the important user->idle->user cases, so instead implement a non-refcounted scheme that causes __mmdrop() to IPI all CPUs in the mm_cpumask and shoot down any remaining lazy ones. Shootdown IPIs are some concern, but they have not been observed to be a big problem with this scheme (the powerpc implementation generated 314 additional interrupts on a 144 CPU system during a kernel compile). There are a number of strategies that could be employed to reduce IPIs if they turn out to be a problem for some workload. Signed-off-by: Nicholas Piggin --- arch/Kconfig | 14 +++++++++++++- kernel/fork.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 65 insertions(+), 1 deletion(-) diff --git a/arch/Kconfig b/arch/Kconfig index 276e1c1c0219..91e1882e3284 100644 --- a/arch/Kconfig +++ b/arch/Kconfig @@ -439,11 +439,23 @@ config NO_MMU_LAZY_TLB def_bool n # Use normal mm refcounting for MMU_LAZY_TLB kernel thread references. -# For now, this must be enabled if MMU_LAZY_TLB is enabled. config MMU_LAZY_TLB_REFCOUNT def_bool y depends on MMU_LAZY_TLB +# Instead of refcounting the lazy mm struct for kernel thread references +# (which can cause contention with multi-threaded apps on large multiprocessor +# systems), this option causes __mmdrop to IPI all CPUs in the mm_cpumask and +# switch to init_mm if they were using the to-be-freed mm as the lazy tlb. To +# implement this, architectures must use _lazy_tlb variants of mm refcounting +# when releasing kernel thread mm references, and mm_cpumask must include at +# least all possible CPUs in which the mm might be lazy, at the time of the +# final mmdrop. mmgrab/mmdrop in arch/ code must be switched to _lazy_tlb +# postfix as necessary. +config MMU_LAZY_TLB_SHOOTDOWN + bool + depends on MMU_LAZY_TLB + config ARCH_HAVE_NMI_SAFE_CMPXCHG bool diff --git a/kernel/fork.c b/kernel/fork.c index dc06afd725cb..d485c24426a0 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -674,6 +674,53 @@ static void check_mm(struct mm_struct *mm) #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL)) #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) +static void do_shoot_lazy_tlb(void *arg) +{ + struct mm_struct *mm = arg; + + if (current->active_mm == mm) { + WARN_ON_ONCE(current->mm); + current->active_mm = &init_mm; + switch_mm(mm, &init_mm, current); + } +} + +static void do_check_lazy_tlb(void *arg) +{ + struct mm_struct *mm = arg; + + WARN_ON_ONCE(current->active_mm == mm); +} + +static void shoot_lazy_tlbs(struct mm_struct *mm) +{ + if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) { + /* + * IPI overheads have not found to be expensive, but they could + * be reduced in a number of possible ways, for example (in + * roughly increasing order of complexity): + * - A batch of mms requiring IPIs could be gathered and freed + * at once. + * - CPUs could store their active mm somewhere that can be + * remotely checked without a lock, to filter out + * false-positives in the cpumask. + * - After mm_users or mm_count reaches zero, switching away + * from the mm could clear mm_cpumask to reduce some IPIs + * (some batching or delaying would help). + * - A delayed freeing and RCU-like quiescing sequence based on + * mm switching to avoid IPIs completely. + */ + on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1); + if (IS_ENABLED(CONFIG_DEBUG_VM)) + on_each_cpu(do_check_lazy_tlb, (void *)mm, 1); + } else { + /* + * In this case, lazy tlb mms are refounted and would not reach + * __mmdrop until all CPUs have switched away and mmdrop()ed. + */ + } +} + /* * Called when the last reference to the mm * is dropped: either by a lazy thread or by @@ -683,7 +730,12 @@ void __mmdrop(struct mm_struct *mm) { BUG_ON(mm == &init_mm); WARN_ON_ONCE(mm == current->mm); + + /* Ensure no CPUs are using this as their lazy tlb mm */ + shoot_lazy_tlbs(mm); + WARN_ON_ONCE(mm == current->active_mm); + mm_free_pgd(mm); destroy_context(mm); mmu_notifier_subscriptions_destroy(mm); -- 2.23.0