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 Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by smtp.lore.kernel.org (Postfix) with ESMTP id 5A424C433EF for ; Wed, 2 Mar 2022 10:05:43 +0000 (UTC) Received: by kanga.kvack.org (Postfix) id E4F148D0003; Wed, 2 Mar 2022 05:05:42 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id DFE308D0001; Wed, 2 Mar 2022 05:05:42 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id CC70A8D0003; Wed, 2 Mar 2022 05:05:42 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from relay.hostedemail.com (relay.hostedemail.com [64.99.140.28]) by kanga.kvack.org (Postfix) with ESMTP id BE2158D0001 for ; Wed, 2 Mar 2022 05:05:42 -0500 (EST) Received: from smtpin09.hostedemail.com (a10.router.float.18 [10.200.18.1]) by unirelay09.hostedemail.com (Postfix) with ESMTP id 8C36724B6B for ; Wed, 2 Mar 2022 10:05:42 +0000 (UTC) X-FDA: 79199014524.09.7EEEBF9 Received: from mail-vs1-f43.google.com (mail-vs1-f43.google.com [209.85.217.43]) by imf10.hostedemail.com (Postfix) with ESMTP id EADFFC000F for ; Wed, 2 Mar 2022 10:05:41 +0000 (UTC) Received: by mail-vs1-f43.google.com with SMTP id w4so1272916vsq.1 for ; Wed, 02 Mar 2022 02:05:41 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc:content-transfer-encoding; bh=EYfrVL/3Fk55o7PeesHa/rcefVwDuQvh3Zqyw9HxNck=; b=Yi+z0/rW44vXrDSIHTaPtBDo80l5jRBxzsz4MDKEte1zksBv1y442Rh60qqep9usRd rYX9XOACpKrHi8UAprCcx7kuIHTzSj2VMWWOIr/vxKaTqokEeIC+C34uBdgJC5hnxrHv qV7wHTtYvNjwrM3TZt4T27EDR+DrCD0ce8oNi0XlTCAV+YC1tGoPrI+Voioz/i6bfahx GIIdoeCjJ7b1ZaWzxsoAGuH2FqZGm7Khn2xhgzvefvxNBffu6ql6nAz/5P4JQg71fqGJ a0hYnE/c7M8w/XtBZkIkaQGIRMHJr8vkhofMS5m+tpGi+3/gKh6Al12prPZsElHnC/cO 0Bqg== X-Gm-Message-State: AOAM533SKf6T4DDDHzqmmMNMpfST2FaJcvXaC76nEdnHmX9Py8lT4cXg bMdrHFfItur7Bdf4cepIiAGX2eUn37w0YA== X-Google-Smtp-Source: ABdhPJzWzs0g7WJaIcB0hHKAPcm/X9QsDcY+XPFaFygh348uYp260vAGde4I9FKH8+1odX2WZqj/Gw== X-Received: by 2002:a05:6102:2434:b0:31e:2234:851 with SMTP id l20-20020a056102243400b0031e22340851mr11387163vsi.79.1646215540955; Wed, 02 Mar 2022 02:05:40 -0800 (PST) Received: from mail-vs1-f43.google.com (mail-vs1-f43.google.com. [209.85.217.43]) by smtp.gmail.com with ESMTPSA id g8-20020a67ea08000000b0031b71a95501sm2195954vso.29.2022.03.02.02.05.39 for (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Wed, 02 Mar 2022 02:05:40 -0800 (PST) Received: by mail-vs1-f43.google.com with SMTP id d11so1259279vsm.5 for ; Wed, 02 Mar 2022 02:05:39 -0800 (PST) X-Received: by 2002:a67:c499:0:b0:320:2cd8:9e1a with SMTP id d25-20020a67c499000000b003202cd89e1amr266158vsk.38.1646215539723; Wed, 02 Mar 2022 02:05:39 -0800 (PST) MIME-Version: 1.0 References: <1646045273-9343-1-git-send-email-anshuman.khandual@arm.com> <1646045273-9343-10-git-send-email-anshuman.khandual@arm.com> <542fa048-131e-240b-cc3a-fd4fff7ce4ba@arm.com> <52866c88-59f9-2d1c-6f5a-5afcaf23f2bb@csgroup.eu> <9caa90f5-c10d-75dd-b403-1388b7a3d296@arm.com> In-Reply-To: <9caa90f5-c10d-75dd-b403-1388b7a3d296@arm.com> From: Geert Uytterhoeven Date: Wed, 2 Mar 2022 11:05:28 +0100 X-Gmail-Original-Message-ID: Message-ID: Subject: Re: [PATCH V3 09/30] arm/mm: Enable ARCH_HAS_VM_GET_PAGE_PROT To: Anshuman Khandual Cc: Christophe Leroy , "Russell King (Oracle)" , "linux-ia64@vger.kernel.org" , "linux-sh@vger.kernel.org" , "linux-mips@vger.kernel.org" , "linux-mm@kvack.org" , "sparclinux@vger.kernel.org" , "linux-riscv@lists.infradead.org" , "linux-arch@vger.kernel.org" , "linux-s390@vger.kernel.org" , "linux-hexagon@vger.kernel.org" , "linux-csky@vger.kernel.org" , Christoph Hellwig , "linux-snps-arc@lists.infradead.org" , "linux-xtensa@linux-xtensa.org" , Arnd Bergmann , "linux-um@lists.infradead.org" , "linux-m68k@lists.linux-m68k.org" , "openrisc@lists.librecores.org" , "linux-arm-kernel@lists.infradead.org" , "linux-parisc@vger.kernel.org" , "linux-kernel@vger.kernel.org" , "linux-alpha@vger.kernel.org" , "akpm@linux-foundation.org" , "linuxppc-dev@lists.ozlabs.org" Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Rspam-User: X-Rspamd-Server: rspam02 X-Rspamd-Queue-Id: EADFFC000F X-Stat-Signature: mm1p8pqb74pqoqjda13wuiynqmk6inum Authentication-Results: imf10.hostedemail.com; dkim=none; dmarc=none; spf=pass (imf10.hostedemail.com: domain of geert.uytterhoeven@gmail.com designates 209.85.217.43 as permitted sender) smtp.mailfrom=geert.uytterhoeven@gmail.com X-HE-Tag: 1646215541-129765 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: Hi Anshuman, On Wed, Mar 2, 2022 at 10:51 AM Anshuman Khandual wrote: > On 3/2/22 12:35 PM, Christophe Leroy wrote: > > Le 02/03/2022 =C3=A0 04:22, Anshuman Khandual a =C3=A9crit : > >> On 3/1/22 1:46 PM, Christophe Leroy wrote: > >>> Le 01/03/2022 =C3=A0 01:31, Russell King (Oracle) a =C3=A9crit : > >>>> On Tue, Mar 01, 2022 at 05:30:41AM +0530, Anshuman Khandual wrote: > >>>>> On 2/28/22 4:27 PM, Russell King (Oracle) wrote: > >>>>>> On Mon, Feb 28, 2022 at 04:17:32PM +0530, Anshuman Khandual wrote: > >>>>>>> This defines and exports a platform specific custom vm_get_page_p= rot() via > >>>>>>> subscribing ARCH_HAS_VM_GET_PAGE_PROT. Subsequently all __SXXX an= d __PXXX > >>>>>>> macros can be dropped which are no longer needed. > >>>>>> > >>>>>> What I would really like to know is why having to run _code_ to wo= rk out > >>>>>> what the page protections need to be is better than looking it up = in a > >>>>>> table. > >>>>>> > >>>>>> Not only is this more expensive in terms of CPU cycles, it also br= ings > >>>>>> additional code size with it. > >>>>>> > >>>>>> I'm struggling to see what the benefit is. > >>>>> > >>>>> Currently vm_get_page_prot() is also being _run_ to fetch required = page > >>>>> protection values. Although that is being run in the core MM and fr= om a > >>>>> platform perspective __SXXX, __PXXX are just being exported for a t= able. > >>>>> Looking it up in a table (and applying more constructs there after)= is > >>>>> not much different than a clean switch case statement in terms of C= PU > >>>>> usage. So this is not more expensive in terms of CPU cycles. > >>>> > >>>> I disagree. > >>> > >>> So do I. > >>> > >>>> > >>>> However, let's base this disagreement on some evidence. Here is the > >>>> present 32-bit ARM implementation: > >>>> > >>>> 00000048 : > >>>> 48: e200000f and r0, r0, #15 > >>>> 4c: e3003000 movw r3, #0 > >>>> 4c: R_ARM_MOVW_ABS_NC .LANCHOR1 > >>>> 50: e3403000 movt r3, #0 > >>>> 50: R_ARM_MOVT_ABS .LANCHOR1 > >>>> 54: e7930100 ldr r0, [r3, r0, lsl #2] > >>>> 58: e12fff1e bx lr > >>>> > >>>> That is five instructions long. > >>> > >>> On ppc32 I get: > >>> > >>> 00000094 : > >>> 94: 3d 20 00 00 lis r9,0 > >>> 96: R_PPC_ADDR16_HA .data..ro_after_init > >>> 98: 54 84 16 ba rlwinm r4,r4,2,26,29 > >>> 9c: 39 29 00 00 addi r9,r9,0 > >>> 9e: R_PPC_ADDR16_LO .data..ro_after_init > >>> a0: 7d 29 20 2e lwzx r9,r9,r4 > >>> a4: 91 23 00 00 stw r9,0(r3) > >>> a8: 4e 80 00 20 blr > >>> > >>> > >>>> > >>>> Please show that your new implementation is not more expensive on > >>>> 32-bit ARM. Please do so by building a 32-bit kernel, and providing > >>>> the disassembly. > >>> > >>> With your series I get: > >>> > >>> 00000000 : > >>> 0: 3d 20 00 00 lis r9,0 > >>> 2: R_PPC_ADDR16_HA .rodata > >>> 4: 39 29 00 00 addi r9,r9,0 > >>> 6: R_PPC_ADDR16_LO .rodata > >>> 8: 54 84 16 ba rlwinm r4,r4,2,26,29 > >>> c: 7d 49 20 2e lwzx r10,r9,r4 > >>> 10: 7d 4a 4a 14 add r10,r10,r9 > >>> 14: 7d 49 03 a6 mtctr r10 > >>> 18: 4e 80 04 20 bctr > >>> 1c: 39 20 03 15 li r9,789 > >>> 20: 91 23 00 00 stw r9,0(r3) > >>> 24: 4e 80 00 20 blr > >>> 28: 39 20 01 15 li r9,277 > >>> 2c: 91 23 00 00 stw r9,0(r3) > >>> 30: 4e 80 00 20 blr > >>> 34: 39 20 07 15 li r9,1813 > >>> 38: 91 23 00 00 stw r9,0(r3) > >>> 3c: 4e 80 00 20 blr > >>> 40: 39 20 05 15 li r9,1301 > >>> 44: 91 23 00 00 stw r9,0(r3) > >>> 48: 4e 80 00 20 blr > >>> 4c: 39 20 01 11 li r9,273 > >>> 50: 4b ff ff d0 b 20 > >>> > >>> > >>> That is definitely more expensive, it implements a table of branches. > >> > >> Okay, will split out the PPC32 implementation that retains existing > >> table look up method. Also planning to keep that inside same file > >> (arch/powerpc/mm/mmap.c), unless you have a difference preference. > > > > My point was not to get something specific for PPC32, but to amplify on > > Russell's objection. > > > > As this is bad for ARM and bad for PPC32, do we have any evidence that > > your change is good for any other architecture ? > > > > I checked PPC64 and there is exactly the same drawback. With the curren= t > > implementation it is a small function performing table read then a few > > adjustment. After your change it is a bigger function implementing a > > table of branches. > > I am wondering if this would not be the case for any other switch case > statement on the platform ? Is there something specific/different just > on vm_get_page_prot() implementation ? Are you suggesting that switch > case statements should just be avoided instead ? > > > > > So, as requested by Russell, could you look at the disassembly for othe= r > > architectures and show us that ARM and POWERPC are the only ones for > > which your change is not optimal ? > > But the primary purpose of this series is not to guarantee optimized > code on platform by platform basis, while migrating from a table based > look up method into a switch case statement. > > But instead, the purposes is to remove current levels of unnecessary > abstraction while converting a vm_flags access combination into page > protection. The switch case statement for platform implementation of > vm_get_page_prot() just seemed logical enough. Christoph's original > suggestion patch for x86 had the same implementation as well. > > But if the table look up is still better/preferred method on certain > platforms like arm or ppc32, will be happy to preserve that. I doubt the switch() variant would give better code on any platform. What about using tables everywhere, using designated initializers to improve readability? Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k= .org In personal conversations with technical people, I call myself a hacker. Bu= t when I'm talking to journalists I just say "programmer" or something like t= hat. -- Linus Torvalds