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=-15.3 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_CR_TRAILER,INCLUDES_PATCH, MAILING_LIST_MULTI,NICE_REPLY_A,SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED, USER_AGENT_SANE_1 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 A82D5C433FE for ; Fri, 4 Dec 2020 10:28:24 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id 201E22251D for ; Fri, 4 Dec 2020 10:28:24 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 201E22251D Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=suse.cz Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=owner-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix) id 6D5C06B0036; Fri, 4 Dec 2020 05:28:23 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 6AD876B005C; Fri, 4 Dec 2020 05:28:23 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 574FC6B0068; Fri, 4 Dec 2020 05:28:23 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0072.hostedemail.com [216.40.44.72]) by kanga.kvack.org (Postfix) with ESMTP id 418C16B0036 for ; Fri, 4 Dec 2020 05:28:23 -0500 (EST) Received: from smtpin22.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay02.hostedemail.com (Postfix) with ESMTP id 06D63362D for ; Fri, 4 Dec 2020 10:28:23 +0000 (UTC) X-FDA: 77555225286.22.table18_6011bd8273c3 Received: from filter.hostedemail.com (10.5.16.251.rfc1918.com [10.5.16.251]) by smtpin22.hostedemail.com (Postfix) with ESMTP id CC42F18038E67 for ; Fri, 4 Dec 2020 10:28:22 +0000 (UTC) X-HE-Tag: table18_6011bd8273c3 X-Filterd-Recvd-Size: 5341 Received: from mx2.suse.de (mx2.suse.de [195.135.220.15]) by imf03.hostedemail.com (Postfix) with ESMTP for ; Fri, 4 Dec 2020 10:28:22 +0000 (UTC) X-Virus-Scanned: by amavisd-new at test-mx.suse.de Received: from relay2.suse.de (unknown [195.135.221.27]) by mx2.suse.de (Postfix) with ESMTP id 02476AC9A; Fri, 4 Dec 2020 10:28:21 +0000 (UTC) To: Muchun Song Cc: Andrew Morton , Linux Memory Management List , LKML References: <20201202121838.75218-1-songmuchun@bytedance.com> <320c8522-4ed5-809f-e6fc-8a185587519c@suse.cz> From: Vlastimil Babka Subject: Re: [External] Re: [PATCH] mm/page_alloc: speeding up the iteration of max_order Message-ID: <69367ce1-eb9b-d76d-0141-da871bd826ec@suse.cz> Date: Fri, 4 Dec 2020 11:28:20 +0100 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.5.0 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-US 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 12/4/20 5:03 AM, Muchun Song wrote: > On Fri, Dec 4, 2020 at 1:37 AM Vlastimil Babka wrote: >> >> On 12/2/20 1:18 PM, Muchun Song wrote: >> > When we free a page whose order is very close to MAX_ORDER and great= er >> > than pageblock_order, it wastes some CPU cycles to increase max_orde= r >> > to MAX_ORDER one by one and check the pageblock migratetype of that = page >> >> But we have to do that. It's not the same page, it's the merged page a= nd the new >> buddy is a different pageblock and we need to check if they have compa= tible >> migratetypes and can merge, or we have to bail out. So the patch is wr= ong. >> >> > repeatedly especially when MAX_ORDER is much larger than pageblock_o= rder. >> >> Do we have such architectures/configurations anyway? >> >> > Signed-off-by: Muchun Song >> > --- >> > mm/page_alloc.c | 4 +++- >> > 1 file changed, 3 insertions(+), 1 deletion(-) >> > >> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c >> > index 141f12e5142c..959541234e1d 100644 >> > --- a/mm/page_alloc.c >> > +++ b/mm/page_alloc.c >> > @@ -1041,7 +1041,7 @@ static inline void __free_one_page(struct page= *page, >> > pfn =3D combined_pfn; >> > order++; >> > } >> > - if (max_order < MAX_ORDER) { >=20 > If we free a page with order =3D=3D MAX_ORDER - 1, it has no buddy. > The following pageblock operation is also pointless. OK, I see. >> > + if (max_order < MAX_ORDER && order < MAX_ORDER - 1) { Yes, this makes sense, as in your other patch we shouldn't check the budd= y when order =3D=3D MAX_ORDER - 1 already. >> > /* If we are here, it means order is >=3D pageblock_or= der. >> > * We want to prevent merge between freepages on isola= te >> > * pageblock and normal pageblock. Without this, pageb= lock >> > @@ -1062,6 +1062,8 @@ static inline void __free_one_page(struct page= *page, >> > is_migrate_isolate(bud= dy_mt))) >> > goto done_merging; >> > } >> > + if (unlikely(order !=3D max_order - 1)) >> > + max_order =3D order + 1; >> > max_order++; OK I see now what you want to do here. the "if" may be true if we already entered the function with order > pageblock_order. I think we could just simplfy the "if" and "max_order++" above to: max_order =3D order + 2 which starts to get a bit ugly, so why not change max_order to be -1 (com= pared to now) in the whole function: max_order =3D min_t(unsigned int, MAX_ORDER - 1, pageblock_order); ... continue_merging: while (order < max_order) { ... if (order < MAX_ORDER - 1) { // it's redundant to keep checking max_order < MAX_ORDER - 1 here after y= our change, right? ... max_order =3D order + 1; // less weird than "+ 2" Off by one errors, here we go! >> Or maybe I just don't understand what this is doing. When is the new '= if' even >> true? We just bailed out of "while (order < max_order - 1)" after the = last >> "order++", which means it should hold that "order =3D=3D max_order - 1= ")? >=20 > No, I do not agree. The MAX_ORDER may be greater than 11. >=20 > # git grep "CONFIG_FORCE_MAX_ZONEORDER" > # arch/arm/configs/imx_v6_v7_defconfig:CONFIG_FORCE_MAX_ZONEORDER=3D14 > # arch/powerpc/configs/85xx/ge_imp3a_defconfig:CONFIG_FORCE_MAX_ZONEORD= ER=3D17 > # arch/powerpc/configs/fsl-emb-nonhw.config:CONFIG_FORCE_MAX_ZONEORDER=3D= 13 >=20 > Have you seen it? On some architecture, the MAX_ORDER > can be 17. When we free a page with an order 16. Without this > patch, the max_order should be increased one by one from 10 to > 17. >=20 > Thanks. >=20 >=20 >> Your description sounds like you want to increase max_order to MAX_ORD= ER in one >> step, which as I explained would be wrong. But the implementation look= s actually >> like a no-op. >> >> > max_order++; >> > goto continue_merging; >> > } >> > >> >=20 >=20 > -- > Yours, > Muchun >=20