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=-12.8 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,URIBL_BLOCKED autolearn=ham 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 DEAFDC433E0 for ; Wed, 13 Jan 2021 18:16:28 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id 4719023120 for ; Wed, 13 Jan 2021 18:16:28 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 4719023120 Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=gmail.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=owner-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix) id 97B028D0086; Wed, 13 Jan 2021 13:16:27 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 92C5D8D006A; Wed, 13 Jan 2021 13:16:27 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 819DD8D0086; Wed, 13 Jan 2021 13:16:27 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0229.hostedemail.com [216.40.44.229]) by kanga.kvack.org (Postfix) with ESMTP id 67C038D006A for ; Wed, 13 Jan 2021 13:16:27 -0500 (EST) Received: from smtpin18.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay03.hostedemail.com (Postfix) with ESMTP id 1A979824556B for ; Wed, 13 Jan 2021 18:16:27 +0000 (UTC) X-FDA: 77701556814.18.frog32_48154bc27520 Received: from filter.hostedemail.com (10.5.16.251.rfc1918.com [10.5.16.251]) by smtpin18.hostedemail.com (Postfix) with ESMTP id D58D4100ED3A5 for ; Wed, 13 Jan 2021 18:16:26 +0000 (UTC) X-HE-Tag: frog32_48154bc27520 X-Filterd-Recvd-Size: 11939 Received: from mail-ej1-f47.google.com (mail-ej1-f47.google.com [209.85.218.47]) by imf48.hostedemail.com (Postfix) with ESMTP for ; Wed, 13 Jan 2021 18:16:26 +0000 (UTC) Received: by mail-ej1-f47.google.com with SMTP id 6so4439156ejz.5 for ; Wed, 13 Jan 2021 10:16:26 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc; bh=EvTiuc+2xyLYeRcrb0/52TJi7BbIHLk9GXdghe1/ck8=; b=pZPUepV9NQObFtttTju09UoUXtYfPvhPFMxJJB/J23vy/w7suSvVg4SitD9bojoBvX tZHoOYT/vsEeRa3TCCYHp3xxQfoUllu44wefJBFlV51UiI84FmW0Ccx+rFgH/qXDfVyx XGoS6McIM2OwNfboCHh9qUKM7FSTeWpqFwKMxOspG/Gi6huHl69CcfKpyCkSgs3SSNg8 1fuRawRQDqcKdpLD0a3C7MCb+1Dju9aDuQ2u+SrVUAP7GnxXh1T8UtEMLJ1dKhUn5eCx jcoW1Sj3lQAJWvYsy30ygSj+FmotUiEM7c945ROEuwgk4FwuWJQtyrXZQSVnDiKJQ9pp YtQA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=EvTiuc+2xyLYeRcrb0/52TJi7BbIHLk9GXdghe1/ck8=; b=cQOAZdtKg+P5nmtYKWIFtF6XbCy0550CMT9ikRobitFHUGuJzTaDAzTe/y5T1SBY1z NjrIQ7SBvc1IzTyfrcOIujMxyhB9be6x48Y3S3WVYcaA7Yyt2ckPjd7mocQ+i3P9JrMg r0ufmHBzVDMaFrafef2afxSfYpZm1XjP5Z+tssCCdyh9mDk5DM2QKgcWloJGpCkaNr5O hyJpxj2jkHZr4datRNdJBuauCOQRavcC3mB/5WcL/t+KrRVC3xhIyZgFUo4/FKmQVmhT +8hav4+vzUNWTd/BzzZ3vbTLsCgIeftzMKpltk9Ivo5Kdcmw/4CMIFiaBvRCW5MXRnk9 A5SQ== X-Gm-Message-State: AOAM532970GEJ0JCbPiV0n/oAcFxqb6twpLnTbzRbtVAVUQOt3K4LQG0 4lmtH3kg8a5DgkzZEFdtx7BWivQlGic1iy+LAJE= X-Google-Smtp-Source: ABdhPJxKf+o87lFKMsfYidQ89N1P8F5iDs2JRQ1uKOS/Fdx6rj7HQIU4g9yoy/JxQWjP2BUmUsOFyCecnWuNpcRGlGM= X-Received: by 2002:a17:907:546:: with SMTP id wk6mr2432952ejb.238.1610561784993; Wed, 13 Jan 2021 10:16:24 -0800 (PST) MIME-Version: 1.0 References: <20210105225817.1036378-1-shy828301@gmail.com> <20210105225817.1036378-4-shy828301@gmail.com> <56d26993-1577-3747-2d89-1275d92f7a15@virtuozzo.com> <35543012-882c-2e1e-f23b-d25a6fa41e67@virtuozzo.com> In-Reply-To: From: Yang Shi Date: Wed, 13 Jan 2021 10:16:13 -0800 Message-ID: Subject: Re: [v3 PATCH 03/11] mm: vmscan: use shrinker_rwsem to protect shrinker_maps allocation To: Kirill Tkhai Cc: Roman Gushchin , Shakeel Butt , Dave Chinner , Johannes Weiner , Michal Hocko , Andrew Morton , Linux MM , Linux FS-devel Mailing List , Linux Kernel Mailing List Content-Type: text/plain; charset="UTF-8" X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: On Tue, Jan 12, 2021 at 1:23 PM Yang Shi wrote: > > On Mon, Jan 11, 2021 at 1:34 PM Kirill Tkhai wrote: > > > > On 11.01.2021 21:57, Yang Shi wrote: > > > On Mon, Jan 11, 2021 at 9:34 AM Kirill Tkhai wrote: > > >> > > >> On 11.01.2021 20:08, Yang Shi wrote: > > >>> On Wed, Jan 6, 2021 at 1:55 AM Kirill Tkhai wrote: > > >>>> > > >>>> On 06.01.2021 01:58, Yang Shi wrote: > > >>>>> Since memcg_shrinker_map_size just can be changd under holding shrinker_rwsem > > >>>>> exclusively, the read side can be protected by holding read lock, so it sounds > > >>>>> superfluous to have a dedicated mutex. This should not exacerbate the contention > > >>>>> to shrinker_rwsem since just one read side critical section is added. > > >>>>> > > >>>>> Signed-off-by: Yang Shi > > >>>>> --- > > >>>>> mm/vmscan.c | 16 ++++++---------- > > >>>>> 1 file changed, 6 insertions(+), 10 deletions(-) > > >>>>> > > >>>>> diff --git a/mm/vmscan.c b/mm/vmscan.c > > >>>>> index 9db7b4d6d0ae..ddb9f972f856 100644 > > >>>>> --- a/mm/vmscan.c > > >>>>> +++ b/mm/vmscan.c > > >>>>> @@ -187,7 +187,6 @@ static DECLARE_RWSEM(shrinker_rwsem); > > >>>>> #ifdef CONFIG_MEMCG > > >>>>> > > >>>>> static int memcg_shrinker_map_size; > > >>>>> -static DEFINE_MUTEX(memcg_shrinker_map_mutex); > > >>>>> > > >>>>> static void memcg_free_shrinker_map_rcu(struct rcu_head *head) > > >>>>> { > > >>>>> @@ -200,8 +199,6 @@ static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, > > >>>>> struct memcg_shrinker_map *new, *old; > > >>>>> int nid; > > >>>>> > > >>>>> - lockdep_assert_held(&memcg_shrinker_map_mutex); > > >>>>> - > > >>>>> for_each_node(nid) { > > >>>>> old = rcu_dereference_protected( > > >>>>> mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); > > >>>>> @@ -250,7 +247,7 @@ int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) > > >>>>> if (mem_cgroup_is_root(memcg)) > > >>>>> return 0; > > >>>>> > > >>>>> - mutex_lock(&memcg_shrinker_map_mutex); > > >>>>> + down_read(&shrinker_rwsem); > > >>>>> size = memcg_shrinker_map_size; > > >>>>> for_each_node(nid) { > > >>>>> map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); > > >>>>> @@ -261,7 +258,7 @@ int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) > > >>>>> } > > >>>>> rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); > > >>>> > > >>>> Here we do STORE operation, and since we want the assignment is visible > > >>>> for shrink_slab_memcg() under down_read(), we have to use down_write() > > >>>> in memcg_alloc_shrinker_maps(). > > >>> > > >>> I apologize for the late reply, these emails went to my SPAM again. > > >> > > >> This is the second time the problem appeared. Just add my email address to allow list, > > >> and there won't be this problem again. > > > > > > Yes, I thought clicking "not spam" would add your email address to the > > > allow list automatically. But it turns out not true. > > > > > >> > > >>> Before this patch it was not serialized by any lock either, right? Do > > >>> we have to serialize it? As Johannes mentioned if shrinker_maps has > > >>> not been initialized yet, it means the memcg is a newborn, there > > >>> should not be significant amount of reclaimable slab caches, so it is > > >>> fine to skip it. The point makes some sense to me. > > >>> > > >>> So, the read lock seems good enough. > > >> > > >> No, this is not so. > > >> > > >> Patch "[v3 PATCH 07/11] mm: vmscan: add per memcg shrinker nr_deferred" adds > > >> new assignments: > > >> > > >> + info->map = (unsigned long *)((unsigned long)info + sizeof(*info)); > > >> + info->nr_deferred = (atomic_long_t *)((unsigned long)info + > > >> + sizeof(*info) + m_size); > > >> > > >> info->map and info->nr_deferred are not visible under READ lock in shrink_slab_memcg(), > > >> unless you use WRITE lock in memcg_alloc_shrinker_maps(). > > > > > > However map and nr_deferred are assigned before > > > rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_info, new). The > > > shrink_slab_memcg() checks shrinker_info pointer. > > > But that order might be not guaranteed, so it seems a memory barrier > > > before rcu_assign_pointer should be good enough, right? > > > > Yes, and here are some more: > > > > 1)There is rcu_dereference_protected() dereferrencing in rcu_dereference_protected(), > > but in case of we use READ lock in memcg_alloc_shrinker_maps(), the dereferrencing > > is not actually protected. > > > > 2)READ lock makes memcg_alloc_shrinker_info() racy against memory allocation fail. > > memcg_alloc_shrinker_info()->memcg_free_shrinker_info() may free memory right > > after shrink_slab_memcg() dereferenced it. You may say shrink_slab_memcg()->mem_cgroup_online() > > protects us from it?! Yes, sure, but this is not the thing we want to remember > > in the future, since this spreads modularity. > > > > Why don't we use WRITE lock? It prohibits shrinking of SLAB during memcg_alloc_shrinker_info()->kvzalloc()? > > Yes, it is the main concern. > > > Yes, but it is not a problem, since page cache is still shrinkable, and we are able to > > allocate memory. WRITE lock means better modularity, and it gives us a possibility > > not to think about corner cases. > > I do agree using write lock makes life easier. I'm just not sure how > bad the impact would be, particularly with vfs metadata heavy workload > (the most memory is consumed by slab cache rather than page cache). > But I think I can design a simple test case, which generates global > memory pressure with slab cache (i.e. negative dentry cache), then > create significant amount of memcgs (i.e. 10k), then check if the > memcgs creation time is lengthened or not. Did a test on a VM with two nodes (80 cpus) + 16GB memory. The test does the below firstly: * Generate negative dentry cache from all cpus to fill up the memory * Run kernel build with 80 processes The memory would be filled up and there should be multiple parallel reclaimers running simultaneously (at least 2 kswapd processes, at most 80 reclaimers), then create 10K memcgs (memcgs creation need allocate shrinker_info with acquiring shrinker_rwsem). The result is: Read lock real 7m17.891s user 0m28.061s sys 2m33.170s Write lock real 7m5.431s user 0m20.400s sys 2m53.162s The one with write lock has longer sys time, it should not be caused by the lock contention since the lock is rwsem, it might spend more time in reclaiming pages. But it had a little bit shorter wall time spent. And OOMs didn't happen either. So, it seems using write lock didn't have a noticeable impact. > > > > > >> > > >> Nowhere in your patchset you convert READ lock to WRITE lock in memcg_alloc_shrinker_maps(). > > >> > > >> So, just use the true lock in this patch from the first time. > > >> > > >>>> > > >>>>> } > > >>>>> - mutex_unlock(&memcg_shrinker_map_mutex); > > >>>>> + up_read(&shrinker_rwsem); > > >>>>> > > >>>>> return ret; > > >>>>> } > > >>>>> @@ -276,9 +273,8 @@ static int memcg_expand_shrinker_maps(int new_id) > > >>>>> if (size <= old_size) > > >>>>> return 0; > > >>>>> > > >>>>> - mutex_lock(&memcg_shrinker_map_mutex); > > >>>>> if (!root_mem_cgroup) > > >>>>> - goto unlock; > > >>>>> + goto out; > > >>>>> > > >>>>> memcg = mem_cgroup_iter(NULL, NULL, NULL); > > >>>>> do { > > >>>>> @@ -287,13 +283,13 @@ static int memcg_expand_shrinker_maps(int new_id) > > >>>>> ret = memcg_expand_one_shrinker_map(memcg, size, old_size); > > >>>>> if (ret) { > > >>>>> mem_cgroup_iter_break(NULL, memcg); > > >>>>> - goto unlock; > > >>>>> + goto out; > > >>>>> } > > >>>>> } while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL); > > >>>>> -unlock: > > >>>>> +out: > > >>>>> if (!ret) > > >>>>> memcg_shrinker_map_size = size; > > >>>>> - mutex_unlock(&memcg_shrinker_map_mutex); > > >>>>> + > > >>>>> return ret; > > >>>>> } > > >>>>> > > >>>>> > > >>>> > > >>>> > > >> > > >> > > > >