From patchwork Sun May 22 20:48:14 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Maciej W. Rozycki" X-Patchwork-Id: 12858349 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id DFED4C433F5 for ; Sun, 22 May 2022 20:48:20 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1344311AbiEVUsT (ORCPT ); Sun, 22 May 2022 16:48:19 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:36042 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1343506AbiEVUsT (ORCPT ); Sun, 22 May 2022 16:48:19 -0400 Received: from angie.orcam.me.uk (angie.orcam.me.uk [78.133.224.34]) by lindbergh.monkeyblade.net (Postfix) with ESMTP id E63F22A268; Sun, 22 May 2022 13:48:17 -0700 (PDT) Received: by angie.orcam.me.uk (Postfix, from userid 500) id ACFBB92009C; Sun, 22 May 2022 22:48:14 +0200 (CEST) Received: from localhost (localhost [127.0.0.1]) by angie.orcam.me.uk (Postfix) with ESMTP id A5EEE92009B; Sun, 22 May 2022 21:48:14 +0100 (BST) Date: Sun, 22 May 2022 21:48:14 +0100 (BST) From: "Maciej W. Rozycki" To: Thomas Bogendoerfer cc: Paul Cercueil , Nathan Chancellor , Tiezhu Yang , linux-mips@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH] MIPS: Rewrite `csum_tcpudp_nofold' in plain C Message-ID: User-Agent: Alpine 2.21 (DEB 202 2017-01-01) MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-mips@vger.kernel.org Recent commit 198688edbf77 ("MIPS: Fix inline asm input/output type mismatch in checksum.h used with Clang") introduced a code size and performance regression with 64-bit code emitted for `csum_tcpudp_nofold' by GCC, caused by a redundant truncation operation produced due to a data type change made to the variable associated with the inline assembly's output operand. The intent previously expressed here with operands and constraints for optimal code was to have the output operand share a register with one inputs, both of a different integer type each. This is perfectly valid with the MIPS psABI where a register can hold integer data of different types and the assembly code used here makes data stored in the output register match the data type used with the output operand, however it has turned out impossible to express this arrangement in source code such as to satisfy LLVM, apparently due to the compiler's internal limitations. There is nothing peculiar about the inline assembly `csum_tcpudp_nofold' includes however, though it does choose assembly instructions carefully. Rewrite this piece of assembly in plain C then, using corresponding C language operations, making GCC produce the same assembly instructions, possibly shuffled, in the general case and sometimes actually fewer of them where an input is constant, because the compiler does not have to reload it to a register (operand constraints could be adjusted for that, but the plain C approach is cleaner anyway). Example code size changes are as follows, for a 32-bit configuration: text data bss total filename 5920480 1347236 126592 7394308 vmlinux-old 5920480 1347236 126592 7394308 vmlinux-now 5919728 1347236 126592 7393556 vmlinux-c and for a 64-bit configuration: text data bss total filename 6024112 1790828 225728 8040668 vmlinux-old 6024128 1790828 225728 8040684 vmlinux-now 6023760 1790828 225728 8040316 vmlinux-c respectively, where "old" is with the commit referred reverted, "now" is with no change, and "c" is with this change applied. Signed-off-by: Maciej W. Rozycki --- Hi, I have visually inspected code produced and verified this change to boot with TCP networking performing just fine, both with a 32-bit and a 64-bit configuration. Sadly with the little endianness only, because in the course of this verification I have discovered the core card of my Malta board bit the dust a few days ago, apparently in a permanent manner, and I have no other big-endian MIPS system available here to try. The only difference between the two endiannesses is the left-shift operation on (proto + len) however, which doesn't happen for big-endian configurations, so the little endianness should in principle provide enough coverage. Also I'm leaving it to LLVM folks to verify, however this is plain C, so it is expected to just work. Please apply. Maciej --- arch/mips/include/asm/checksum.h | 71 ++++++++++++++++++--------------------- 1 file changed, 34 insertions(+), 37 deletions(-) linux-mips-csum-tcpudp-nofold-proper-fix.diff Index: linux-macro/arch/mips/include/asm/checksum.h =================================================================== --- linux-macro.orig/arch/mips/include/asm/checksum.h +++ linux-macro/arch/mips/include/asm/checksum.h @@ -128,48 +128,45 @@ static inline __sum16 ip_fast_csum(const static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len, __u8 proto, - __wsum sum) + __wsum isum) { - unsigned long tmp = (__force unsigned long)sum; + const unsigned int sh32 = IS_ENABLED(CONFIG_64BIT) ? 32 : 0; + unsigned long sum = (__force unsigned long)daddr; + unsigned long tmp; + __u32 osum; - __asm__( - " .set push # csum_tcpudp_nofold\n" - " .set noat \n" -#ifdef CONFIG_32BIT - " addu %0, %2 \n" - " sltu $1, %0, %2 \n" - " addu %0, $1 \n" + tmp = (__force unsigned long)saddr; + sum += tmp; - " addu %0, %3 \n" - " sltu $1, %0, %3 \n" - " addu %0, $1 \n" + if (IS_ENABLED(CONFIG_32BIT)) + sum += sum < tmp; - " addu %0, %4 \n" - " sltu $1, %0, %4 \n" - " addu %0, $1 \n" -#endif -#ifdef CONFIG_64BIT - " daddu %0, %2 \n" - " daddu %0, %3 \n" - " daddu %0, %4 \n" - " dsll32 $1, %0, 0 \n" - " daddu %0, $1 \n" - " sltu $1, %0, $1 \n" - " dsra32 %0, %0, 0 \n" - " addu %0, $1 \n" -#endif - " .set pop" - : "=r" (tmp) - : "0" ((__force unsigned long)daddr), - "r" ((__force unsigned long)saddr), -#ifdef __MIPSEL__ - "r" ((proto + len) << 8), -#else - "r" (proto + len), -#endif - "r" ((__force unsigned long)sum)); + /* + * We know PROTO + LEN has the sign bit clear, so cast to a signed + * type to avoid an extraneous zero-extension where TMP is 64-bit. + */ + tmp = (__s32)(proto + len); + tmp <<= IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? 8 : 0; + sum += tmp; + if (IS_ENABLED(CONFIG_32BIT)) + sum += sum < tmp; - return (__force __wsum)tmp; + tmp = (__force unsigned long)isum; + sum += tmp; + + if (IS_ENABLED(CONFIG_32BIT)) { + sum += sum < tmp; + osum = sum; + } else if (IS_ENABLED(CONFIG_64BIT)) { + tmp = sum << sh32; + sum += tmp; + osum = sum < tmp; + osum += sum >> sh32; + } else { + BUILD_BUG(); + } + + return (__force __wsum)osum; } #define csum_tcpudp_nofold csum_tcpudp_nofold