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=-7.0 required=3.0 tests=BAYES_00,DKIM_ADSP_CUSTOM_MED, DKIM_INVALID,DKIM_SIGNED,FREEMAIL_FORGED_FROMDOMAIN,FREEMAIL_FROM, HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_PATCH,MAILING_LIST_MULTI,SPF_HELO_NONE, SPF_PASS,USER_AGENT_SANE_1 autolearn=no 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 2836BC433C1 for ; Wed, 24 Mar 2021 18:41:26 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id B6A0A61A17 for ; Wed, 24 Mar 2021 18:41:25 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org B6A0A61A17 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 4B1436B02E5; Wed, 24 Mar 2021 14:41:25 -0400 (EDT) Received: by kanga.kvack.org (Postfix, from userid 40) id 460276B02E7; Wed, 24 Mar 2021 14:41:25 -0400 (EDT) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 28C506B02E8; Wed, 24 Mar 2021 14:41:25 -0400 (EDT) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0175.hostedemail.com [216.40.44.175]) by kanga.kvack.org (Postfix) with ESMTP id 0A3026B02E5 for ; Wed, 24 Mar 2021 14:41:25 -0400 (EDT) Received: from smtpin32.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay05.hostedemail.com (Postfix) with ESMTP id A118A1836F64D for ; Wed, 24 Mar 2021 18:41:24 +0000 (UTC) X-FDA: 77955635688.32.39816D8 Received: from mail-lj1-f174.google.com (mail-lj1-f174.google.com [209.85.208.174]) by imf30.hostedemail.com (Postfix) with ESMTP id 00591E0011CE for ; Wed, 24 Mar 2021 18:41:20 +0000 (UTC) Received: by mail-lj1-f174.google.com with SMTP id s17so31513015ljc.5 for ; Wed, 24 Mar 2021 11:41:22 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:date:to:cc:subject:message-id:references:mime-version :content-disposition:content-transfer-encoding:in-reply-to :user-agent; bh=Lmgab50l2WbnE1Y1ChETHsaXAH3JIp5+u9sDf2Xz+2k=; b=K6n0Khh72xDM7iiKzQUHkPfe2SthSYZUSmx30mcQXYAGnlu+SnV1IzETMD4EF32r+W GwCjzIgG5PsbyQIwI1as+Ds+NcqRw3gsCQkQf1eb/RExDivWq3EbQk8yAGPNqqvK2f93 2j0jq3v0xNskXTc8JRMFsnId3Cxaqapwny/SSAREuDN0qniuRTbYFRkn3n6DwMTTpt9+ SCwu+ACFQz9Qjod9AwsN2HPDdQYmWw3hazUPYhPvE/Vtfw0X78Ctlh0UXlsXFqYUUOj2 Cfjd/+4eNB1YlQS/duiAzfXVlLYJyMwghCKhqk8FaKi/TlBhlJjFh1HIiLSCSYY61yBX Lgqw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:date:to:cc:subject:message-id:references :mime-version:content-disposition:content-transfer-encoding :in-reply-to:user-agent; bh=Lmgab50l2WbnE1Y1ChETHsaXAH3JIp5+u9sDf2Xz+2k=; b=OK1W0I1ae/ZiyEvn01opjpnoKgUKl7xq5NB8rCTm8gGYD+Imrh7xy+bzIhVulNxe5c PtsG+luIYIY4zkI1u4F7VhdmOu4SsXeJAAtPcruSB/eufNpVqavTp5Z8Z7tWycSLLrt8 auF0BbOZupl0yTE6thXpGykSz2Kz9vsQ2tzW7cWuZKN8pnxtZwyMi0SQHt5Rn6olGVWu ccNcfzFbnnjD+/lT9oKocWrIEOE8ke7QdRRpbZoFpTtvq/G5GOSWsiFDmHETGPWI+Im6 hAThjqgqeEd+6Ip7WIS6ORQTLdd8tef35Eq+5fVmN82W12LqdxgSyrVFhqVugc1OFPfM YsGA== X-Gm-Message-State: AOAM530+Q+Ahi9+mQx0Y8YVrTF3b9bqtD314aTXJtgWyaqCxpD8DPjip bvFP8yo10lMuuIQ2o0O8VF0= X-Google-Smtp-Source: ABdhPJzqqSFYVfsF7DpZqSX2LgvZ6x327mJHK44eKP+/VvVqlrwCa4uIBPAzL0DyUHzyoTcMsSRupw== X-Received: by 2002:a2e:1649:: with SMTP id 9mr3112697ljw.74.1616611281496; Wed, 24 Mar 2021 11:41:21 -0700 (PDT) Received: from pc638.lan (h5ef52e3d.seluork.dyn.perspektivbredband.net. [94.245.46.61]) by smtp.gmail.com with ESMTPSA id g14sm409662ljj.3.2021.03.24.11.41.20 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Wed, 24 Mar 2021 11:41:21 -0700 (PDT) From: Uladzislau Rezki X-Google-Original-From: Uladzislau Rezki Date: Wed, 24 Mar 2021 19:41:19 +0100 To: Matthew Wilcox Cc: Andrew Morton , Christoph Lameter , Pekka Enberg , David Rientjes , Joonsoo Kim , Vlastimil Babka , linux-mm@kvack.org, linux-kernel@vger.kernel.org, Nicholas Piggin , urezki@gmail.com Subject: Re: [PATCH 2/2] mm/vmalloc: Use kvmalloc to allocate the table of pages Message-ID: <20210324184119.GA25852@pc638.lan> References: <20210322193820.2140045-1-willy@infradead.org> <20210322193820.2140045-2-willy@infradead.org> <20210322223619.GA56503@pc638.lan> <20210322230311.GY1719932@casper.infradead.org> <20210323120436.GA1949@pc638.lan> <20210323123913.GD1719932@casper.infradead.org> <20210323203924.GA52881@pc638.lan> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20210323203924.GA52881@pc638.lan> User-Agent: Mutt/1.10.1 (2018-07-13) X-Stat-Signature: cemxc1aamd3irw9wcraajpeuaxdspns7 X-Rspamd-Server: rspam01 X-Rspamd-Queue-Id: 00591E0011CE Received-SPF: none (gmail.com>: No applicable sender policy available) receiver=imf30; identity=mailfrom; envelope-from=""; helo=mail-lj1-f174.google.com; client-ip=209.85.208.174 X-HE-DKIM-Result: pass/pass X-HE-Tag: 1616611280-143700 Content-Transfer-Encoding: quoted-printable 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, Mar 23, 2021 at 09:39:24PM +0100, Uladzislau Rezki wrote: > > On Tue, Mar 23, 2021 at 01:04:36PM +0100, Uladzislau Rezki wrote: > > > On Mon, Mar 22, 2021 at 11:03:11PM +0000, Matthew Wilcox wrote: > > > > I suspect the vast majority of the time is spent calling alloc_pa= ges_node() > > > > 1024 times. Have you looked at Mel's patch to do ... well, exact= ly what > > > > vmalloc() wants? > > > >=20 > > > > > > - __vmalloc_node_range > > > - 45.25% __alloc_pages_nodemask > > > - 37.59% get_page_from_freelist > > [...] > > > - 44.61% 0xffffffffc047348d > > > - __vunmap > > > - 35.56% free_unref_page > >=20 > > Hmm! I hadn't been thinking about the free side of things. > > Does this make a difference? > >=20 > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > index 4f5f8c907897..61d5b769fea0 100644 > > --- a/mm/vmalloc.c > > +++ b/mm/vmalloc.c > > @@ -2277,16 +2277,8 @@ static void __vunmap(const void *addr, int dea= llocate_pages) > > vm_remove_mappings(area, deallocate_pages); > > =20 > > if (deallocate_pages) { > > - int i; > > - > > - for (i =3D 0; i < area->nr_pages; i++) { > > - struct page *page =3D area->pages[i]; > > - > > - BUG_ON(!page); > > - __free_pages(page, 0); > > - } > > + release_pages(area->pages, area->nr_pages); > > atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); > > - > > kvfree(area->pages); > > } > > > Same test. 4MB allocation on a single CPU: >=20 > default: loops: 1000000 avg: 93601889 usec > patch: loops: 1000000 avg: 98217904 usec >=20 > > - __vunmap > - 41.17% free_unref_page > - 28.42% free_pcppages_bulk > - 6.38% __mod_zone_page_state > 4.79% check_preemption_disabled > 2.63% __list_del_entry_valid > 2.63% __list_add_valid > - 7.50% free_unref_page_commit > 2.15% check_preemption_disabled > 2.01% __list_add_valid > 2.31% free_unref_page_prepare.part.86 > 0.70% free_pcp_prepare > >=20 > > - __vunmap > - 45.36% release_pages > - 37.70% free_unref_page_list > - 24.70% free_pcppages_bulk > - 5.42% __mod_zone_page_state > 4.23% check_preemption_disabled > 2.31% __list_add_valid > 2.07% __list_del_entry_valid > - 7.58% free_unref_page_commit > 2.47% check_preemption_disabled > 1.75% __list_add_valid > 3.43% free_unref_page_prepare.part.86 > - 2.39% mem_cgroup_uncharge_list > uncharge_page > >=20 > It is obvious that the default version is slightly better. It requires > less things to be done comparing with release_pages() variant. >=20 > >=20 > > release_pages does a bunch of checks that are unnecessary ... we coul= d > > probably just do: > >=20 > > LIST_HEAD(pages_to_free); > >=20 > > for (i =3D 0; i < area->nr_pages; i++) { > > struct page *page =3D area->pages[i]; > > if (put_page_testzero(page)) > > list_add(&page->lru, &pages_to_free); > > } > > free_unref_page_list(&pages_to_free); > >=20 > > but let's see if the provided interface gets us the performance we wa= nt. > > =20 > I will test it tomorrow. From the first glance it looks like a more lig= ht version :) >=20 Here we go: diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 4f5f8c907897..349024768ba6 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -2254,6 +2254,7 @@ static void vm_remove_mappings(struct vm_struct *ar= ea, int deallocate_pages) static void __vunmap(const void *addr, int deallocate_pages) { struct vm_struct *area; + LIST_HEAD(pages_to_free); =20 if (!addr) return; @@ -2282,11 +2283,12 @@ static void __vunmap(const void *addr, int deallo= cate_pages) for (i =3D 0; i < area->nr_pages; i++) { struct page *page =3D area->pages[i]; =20 - BUG_ON(!page); - __free_pages(page, 0); + if (put_page_testzero(page)) + list_add(&page->lru, &pages_to_free); } - atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); =20 + free_unref_page_list(&pages_to_free); + atomic_long_sub(area->nr_pages, &nr_vmalloc_pages); kvfree(area->pages); } # patch 4MB allocation, single cpu, loops: 1000000 avg: 89065758 usec 4MB allocation, single cpu, loops: 1000000 avg: 90258523 usec 4MB allocation, single cpu, loops: 1000000 avg: 89363057 usec 4MB allocation, single cpu, loops: 1000000 avg: 89271685 usec 4MB allocation, single cpu, loops: 1000000 avg: 89247375 usec # default 4MB allocation, single cpu, loops: 1000000 avg: 89258814 usec 4MB allocation, single cpu, loops: 1000000 avg: 89364194 usec 4MB allocation, single cpu, loops: 1000000 avg: 89226816 usec 4MB allocation, single cpu, loops: 1000000 avg: 89247360 usec 4MB allocation, single cpu, loops: 1000000 avg: 89330116 usec Do not see any difference. See below some profiling regarding cache misses: - __vunmap - 32.15% free_unref_page_list = = =20 - 23.54% free_pcppages_bulk - 6.33% __mod_zone_page_state 4.65% check_preemption_disabled free_unref_page_list(): =E2=94=82 free_unref_page_list(): =E2=94=82ffffffff8125152a: mov 0x8(%rbp),%rax 31.81 =E2=94=82ffffffff8125152e: lea 0x8(%rbp),%r12 = = =20 =E2=94=82ffffffff81251532: mov %rbp,%r14 14.40 =E2=94=82ffffffff81251535: lea -0x8(%rax),%rbp =20 (gdb) l *0xffffffff8125152e 0xffffffff8125152e is in free_unref_page_list (mm/page_alloc.c:3271). 3266 struct page *page, *next; 3267 unsigned long flags, pfn; 3268 int batch_count =3D 0; 3269 3270 /* Prepare pages for freeing */ 3271 list_for_each_entry_safe(page, next, list, lru) { 3272 pfn =3D page_to_pfn(page); 3273 if (!free_unref_page_prepare(page, pfn)) 3274 list_del(&page->lru); 3275 set_page_private(page, pfn); (gdb) free_pcppages_bulk(): =E2=94=82 PageBuddy(): 0.59 =E2=94=82ffffffff8124f523: mov 0x30(%rax),%edi 13.59 =E2=94=82ffffffff8124f526: and $0xf0000080,%edi =20 (gdb) l *0xffffffff8124f526 0xffffffff8124f526 is in free_pcppages_bulk (./include/linux/page-flags.h= :742). 737 738 /* 739 * PageBuddy() indicates that the page is free and in the buddy s= ystem 740 * (see mm/page_alloc.c). 741 */ 742 PAGE_TYPE_OPS(Buddy, buddy) 743 744 /* 745 * PageOffline() indicates that the page is logically offline alt= hough the 746 * containing section is online. (e.g. inflated in a balloon driv= er or (gdb) Looks like it would be good to have a free_pages_bulk_array() :) -- Vlad Rezki