From mboxrd@z Thu Jan 1 00:00:00 1970 Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.remote-ip=209.85.221.41 ARC-Seal: i=1; d=subspace.kernel.org; s=arc20261015; a=rsa-sha256; cv=none; t=1795212231; b=4QWTqUTEC2Z7DlcolIrdbo/4scLcu9kTpNNzyOtm9u61KaYzViMvVrWcDluQYBdmEFAs u/+2fDxL1EMQbKxKezkzuC9d4eauJVdyvoKw2YzbMdFFsEc8kjLL6W9PDYVYKY9TWu+qr 5lZk6f8tGo6eLEvQZfNkn0cKfy6Nv92nN4de0PHI/uR3sMMS+GCdAmPNCkaz7SS8AG2fn vPvjCUbK5Hk6r+RKPyz5lgAbZGsUYe95TY9T1gijySqWFMafJwdq8eY0PSKBxj/QitC72 MH/iR9IbO9eUO4kpUBuWCQjlkyANqvhtpxIIIwNbY9ToaRPrYjpPkpF4RRlyBxaFT/A== ARC-Message-Signature: i=1; d=subspace.kernel.org; s=arc20261015; a=rsa-sha256; c=relaxed/relaxed; t=1795212231; h=DMARC-Filter:Received:DKIM-Signature:X-Google-DKIM-Signature: X-Forwarded-Encrypted:X-Gm-Message-State:X-Gm-Gg:X-Received:Received: From:To:Cc:Subject:Date:Message-ID:X-Mailer:In-Reply-To:References: MIME-Version:Content-Transfer-Encoding; bh=FYgilbWDAFhzqBX4pr5QCnT909akeptX2wNGaYFkpnY=; b=U6wsFw9NO2qoKzscQZlYuN0jeTpxlrLojMFqDkWosmeOi5lfyyOrewPMiBhhUynvdWOd fHIBfCO/nqq6UPI34GVYGY+wXiFH995v9s5yIOXCZEi00cZOdySdGiqa1rjUZ5JRZtaSR ATMHb8Lp3oqFkX+cM/DWQR18/xtvuLxkZVphWQpjIHuJLc910NH9HdqS7J8Nn8YaNmXE3 TrPYpGRDoKGC4mBAXuP1RCGiB4BBw3jNQd6KNBDVfdjTmLAKMN8Mod/M49NJilzNJMKWt /x/URg5vtVoDSczk9qD8seVVn5TsahoyX2fO+YRfjPa1YeHNAm8iXLw3oHMsYnc+6xw== ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass header.from=gmail.com; spf=pass smtp.mailfrom=gmail.com; dkim=pass header.d=gmail.com header.i=@gmail.com header.a=rsa-sha256 header.s=20260707 header.b=ycsH7AuE; arc=none smtp.remote-ip=209.85.221.41 Received: from mail-wr1-f41.google.com (mail-wr1-f41.google.com [209.85.221.41]) (using TLSv1.3 with cipher TLS_AES_128_GCM_SHA256 (128/128 bits) key-exchange X25519 server-signature ECDSA (prime256v1) server-digest SHA256) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id E0086606A69D for ; Fri, 20 Nov 2026 22:03:41 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.4.2 smtp.subspace.kernel.org E0086606A69D Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=gmail.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.a=rsa-sha256 header.s=20260707 header.b=ycsH7AuE Received: by mail-wr1-f41.google.com with SMTP id ffacd0b85a97d-4ad6219f7d8so681300f8f.0 for ; Fri, 20 Nov 2026 14:03:41 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20260707; t=1795212220; x=1795817020; darn=vger.kernel.org; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:from:to:cc:subject:date :message-id:reply-to:content-type; bh=FYgilbWDAFhzqBX4pr5QCnT909akeptX2wNGaYFkpnY=; b=ycsH7AuEwi/PG5y/j3/+nKuyPibue+HdCEuLWy1JtDiD1DR8F35ujuFKyeuDodYdU1 efDxqM7PZfq0ObpZuX+GRe7pH/8QDXd746FnF3tBu5Qqnu74a0ywuz6JOp/IKqcpnyfs BZEecralbSxUvc25sdjAmrWaCNv8tEtoiOyqPoow7HLYdDhOOBkrc8CNq34klDkTjt5d 7Gl6eAStcFMl5h5BDH2qzNke+cN/1HGyVWwc/0nWy8GQgxcS6JYpIUTUOXwwlF9IYXan W+h4utjdM6e+a18hfqQqoqsHPE4oXOU1U8rvvOusitKoe5FRxzkLmTAEvQiiWS5M/H59 dLyg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20260707; t=1795212220; x=1795817020; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-gg:x-gm-message-state:from :to:cc:subject:date:message-id:reply-to:content-type; bh=FYgilbWDAFhzqBX4pr5QCnT909akeptX2wNGaYFkpnY=; b=DrCrk0MXQrT8+4RpiqYZgwhR1nMKphug+dGJnnvHZptaWKR/UTytYH6z0QKLAVQLKW AnfF0yNC+MQn8FpY6XjEIE80bYt1D/EckufgwN7pFda9lTi1/UddYLKHsU4aqu3TvU2b ufPnenLWQEOuATBk+t7put9OguGTVy5dB/WECEcN/9HTNlmgB65qGF7Fhq1JuilZYF41 7WB7pT37lEFKleAVa2Q04KbEjfBr/XcK+9EKN/F6lmzk1JhxmlOVa9mXkEHqBlXfGRLw 5446pev/sradztg0VM9FbcA+TnuwEst2CARFdeH4qoTJ/Iad6QYuM6R2L6A8F0mhhgde +6eQ== X-Forwarded-Encrypted: i=1; AL+hErFgc2XYwl4gcK7lIp8jmA4XaJ8QDdQKgTj//O/yAAly8osAdqCDq8ObydkB5sPe3pkupiMXnvs=@vger.kernel.org X-Gm-Message-State: APuU4Z9kchcE7KSBAjFTwmZ3xcR7cILqBdFyfpTsfrfm65jFi4sGUbp/ Q7dnhPBMv2j/rHwavmyvQI7IYQsWoibjcqv0xFE1ZXQJSFvV9TtR7gnj X-Gm-Gg: AYBFou2rwRu7a/k+MgYYakIQWxn3Ukx5IWbHmGdaTgt7XbfKrX51U4prdjeFR12SwYB 8HcVcU8ijVIMMgb8O+TUrB91edY4vCChbaSW4xy7W6laJCDPZVRmaKdtWT074tYDSJEqZ3FQ7Ow FYY2neJsFFD3SrxgMW321pQjTlfYyCaA1vV9XJ3lwCZHY2v2UlInv24yzEOMM56skgUTagYKf3F sw/+F0ugBJojplEcGRxF4lCvFP4Fyg5WnB4D+ROx5mvjlj+172+dtdfIhrwB0d4xl9VgzROj8Zf yD85j5U8+kMnzIUS40E6fm7jbvqPpmQU3UkgVklwmOrFEXTMpgFfE9zm+N7eUcOWBnl2ntgKAGa fjdNbvOXlTDyeY2yZcyP4tcUrPKnvhVXER2eSd2lV+u/DlK9eXUnJgP2I6YXX3GskrvcC3fo41E 2oNXsZYhXvIiRoCKrqnmC0aaMmiBLeMjHjWo7lB2CS77C750uws76SGhlEPOFEoX//6kdlbdy2B pCqrMsDQ5gzncQVS7KK/ym3oKxudhjHFBiInLesY6CjLVm3eO6zugEwl8Ksggvx1bVyIGElsI2v Gm4UVNsNZ3C17122UxsDqL5QsKYU+Rt1QfVMdRygsWDDgS5BrQZBoL2K6p306CjrMIihYwsEf0B w8dhGvYKn X-Received: by 2002:a05:600c:64c9:b0:4bb:679a:86ee with SMTP id 5b1f17b1804b1-4bb88938dbemr105794155e9.29.1795212220053; Fri, 20 Nov 2026 14:03:40 -0800 (PST) Received: from demian.play.pl (83-175-177-170.dynamic.play.pl. [83.175.177.170]) by smtp.gmail.com with ESMTPSA id ffacd0b85a97d-4ab6b9abf15sm7396031f8f.43.2026.11.20.14.03.37 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 20 Nov 2026 14:03:38 -0800 (PST) From: Demian Shulhan To: Catalin Marinas , Will Deacon Cc: Robin Murphy , David Laight , Nathan Chancellor , Jakub Kicinski , Eric Dumazet , Andrew Morton , David Gow , Brendan Higgins , linux-arm-kernel@lists.infradead.org, netdev@vger.kernel.org, kunit-dev@googlegroups.com, linux-kselftest@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH v3 1/2] arm64: csum: Add fused copy and Internet checksum Date: Fri, 20 Nov 2026 23:03:23 +0100 Message-ID: <20261120220330.1067165-2-demyansh@gmail.com> X-Mailer: git-send-email 2.53.0 In-Reply-To: <20261120220330.1067165-1-demyansh@gmail.com> References: <20261120220330.1067165-1-demyansh@gmail.com> X-Mailing-List: netdev@vger.kernel.org List-Id: List-Post: List-Help: List-Subscribe: List-Unsubscribe: List-Owner: Precedence: list MIME-Version: 1.0 Content-Transfer-Encoding: 8bit arm64 falls back to the generic csum_partial_copy_nocheck(), which does two passes (memcpy() + csum_partial()) and hits software GSO paths (e.g. TCP over WireGuard) and UDP/ICMP replies. Implement a single-pass csum_partial_copy_nocheck() in arch/arm64/lib/csum.c using a 64-byte carry-tree primitive unrolled 4x to 256 bytes (~21 instructions per 64B, pure leaf function with zero spills on GCC 15 and Clang 21) and 16-byte @src alignment for len >= 272. Each summed word is read once into GPRs before storing and summing, so the checksum always matches @dst under concurrent @src modification. In-kernel KUnit benchmark on ARM Neoverse-V2 (3.0 GHz, CONFIG_CHECKSUM_KUNIT_BENCHMARK=y, GCC 15.2 / Clang 21.1, median of 3): case (src/dst) memcpy+csum_partial fused (GCC / Clang) speedup ---------------- ------------------- ------------------- ----------- 16B (+0 / +0) 4.0 / 3.7 GB/s 6.9 / 5.9 GB/s 1.73x/1.61x 64B (+0 / +0) 9.4 / 10.5 GB/s 19.1 / 22.0 GB/s 2.04x/2.08x 256B (+0 / +0) 16.3 / 18.4 GB/s 33.2 / 35.2 GB/s 2.04x/1.92x 512B (+0 / +0) 19.4 / 21.0 GB/s 36.5 / 39.3 GB/s 1.88x/1.87x 1368B (+0 / +0) 21.7 / 22.2 GB/s 39.5 / 41.2 GB/s 1.82x/1.85x 1448B (+0 / +0) 21.8 / 22.2 GB/s 39.2 / 41.1 GB/s 1.79x/1.86x 4096B (+0 / +0) 23.4 / 23.5 GB/s 42.5 / 43.9 GB/s 1.82x/1.87x 16384B (+0 / +0) 23.8 / 24.0 GB/s 43.3 / 45.1 GB/s 1.82x/1.87x 16B (+1 / +3) 1.5 / 2.6 GB/s 6.9 / 5.9 GB/s 4.44x/2.25x 128B (+1 / +3) 6.4 / 10.5 GB/s 24.3 / 27.2 GB/s 3.78x/2.60x 1448B (+1 / +3) 17.8 / 18.0 GB/s 38.0 / 39.0 GB/s 2.14x/2.16x 1500B (+1 / +3) 17.9 / 18.2 GB/s 36.5 / 39.6 GB/s 2.04x/2.18x 4096B (+1 / +3) 19.1 / 19.5 GB/s 40.6 / 42.0 GB/s 2.12x/2.15x 16384B (+1 / +3) 19.4 / 19.6 GB/s 42.3 / 43.8 GB/s 2.18x/2.24x Suggested-by: David Laight Signed-off-by: Demian Shulhan --- arch/arm64/include/asm/checksum.h | 3 + arch/arm64/lib/csum.c | 189 ++++++++++++++++++++++++++++++ 2 files changed, 192 insertions(+) diff --git a/arch/arm64/include/asm/checksum.h b/arch/arm64/include/asm/checksum.h index dc52b733675d..9f766db01152 100644 --- a/arch/arm64/include/asm/checksum.h +++ b/arch/arm64/include/asm/checksum.h @@ -44,6 +44,9 @@ static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl) extern unsigned int do_csum(const unsigned char *buff, int len); #define do_csum do_csum +#define _HAVE_ARCH_CSUM_AND_COPY +__wsum csum_partial_copy_nocheck(const void *src, void *dst, int len); + #include #endif /* __ASM_CHECKSUM_H */ diff --git a/arch/arm64/lib/csum.c b/arch/arm64/lib/csum.c index 75db718e5b6a..2bf2d63cfd42 100644 --- a/arch/arm64/lib/csum.c +++ b/arch/arm64/lib/csum.c @@ -1,9 +1,12 @@ // SPDX-License-Identifier: GPL-2.0-only // Copyright (C) 2019-2020 Arm Ltd. +#include #include +#include #include #include +#include #include @@ -116,6 +119,192 @@ unsigned int __no_sanitize_address do_csum(const unsigned char *buff, int len) return sum >> 16; } +static __always_inline __uint128_t get_unaligned128(const u8 *p) +{ + u64 lo = get_unaligned((const u64 *)p); + u64 hi = get_unaligned((const u64 *)(p + 8)); + + return ((__uint128_t)hi << 64) | lo; +} + +static __always_inline void put_unaligned128(__uint128_t val, u8 *p) +{ + put_unaligned((u64)val, (u64 *)p); + put_unaligned((u64)(val >> 64), (u64 *)(p + 8)); +} + +static __always_inline __wsum csum_fold_acc_cy(__uint128_t acc, u64 cy) +{ + __uint128_t t; + u64 s; + + acc += (acc >> 64) | (acc << 64); + t = (__uint128_t)(u64)(acc >> 64) + cy; + s = (u64)t + (u64)(t >> 64); + s += (s >> 32) | (s << 32); + return (__force __wsum)(u32)(s >> 32); +} + +static __always_inline void csum_tree64_ovf(const u8 *s, u8 *d, + __uint128_t *acc, s64 *cy) +{ + __uint128_t v0 = get_unaligned128(s); + __uint128_t v1 = get_unaligned128(s + 16); + __uint128_t v2, v3; + s64 o; + + put_unaligned128(v0, d); + put_unaligned128(v1, d + 16); + o = __builtin_add_overflow(v0, v1, &v0); + + v2 = get_unaligned128(s + 32); + v3 = get_unaligned128(s + 48); + put_unaligned128(v2, d + 32); + put_unaligned128(v3, d + 48); + o += __builtin_add_overflow(v2, v3, &v2); + + o += __builtin_add_overflow(v0, v2, &v0); + *cy += o + __builtin_add_overflow(*acc, v0, acc); +} + +static __always_inline void csum_align_src16(const u8 **ps, u8 **pd, + unsigned int *pulen, + __uint128_t *pacc, s64 *pc0) +{ + uintptr_t off = (uintptr_t)(*ps) & 15; + + if (unlikely(off) && *pulen >= 272U) { + unsigned int adv = 16U - (unsigned int)off; + u64 lo = get_unaligned((const u64 *)(*ps)); + u64 hi = get_unaligned((const u64 *)(*ps + 8)); + u64 odd = (u64)off & 1ULL; + u64 m = ~0ULL >> ((off & 7) * 8); + u64 h; + + put_unaligned(lo, (u64 *)(*pd)); + put_unaligned(hi, (u64 *)(*pd + 8)); + lo = (off & 8) ? (lo & m) : lo; + hi = (off & 8) ? 0ULL : (hi & m); + h = lo + hi; + h += h < lo; + *pacc = ror64(h, (unsigned int)(odd * 56ULL)); + /* + * At len <= INT_MAX, the 256B loop adds <= 16 carries per + * iteration (< 2^27 total before bit 62 is cleared, and c0 + * stays < 2^33 overall), so bit 62 of @c0 cannot overflow. + */ + *pc0 = (s64)(odd << 62); + *ps += adv; + *pd += adv; + *pulen -= adv; + } +} + +static __always_inline __wsum csum_fold_final(__uint128_t acc, u64 cy, u64 odd) +{ + __wsum res = csum_fold_acc_cy(acc, cy); + + if (unlikely(odd)) + return (__force __wsum)ror32((__force u32)res, 8); + return res; +} + +/* + * Single-pass copy and ones' complement sum. Each summed word is read once + * into a GPR before storing and summing, so the returned sum always matches + * @dst even if @src is concurrently modified. + */ +__wsum __no_sanitize_address csum_partial_copy_nocheck(const void *src, void *dst, int len) +{ + const u8 *s = src; + u8 *d = dst; + __uint128_t acc = 0; + s64 c0 = 0; + u64 odd = 0; + unsigned int ulen = (unsigned int)len; + + if (unlikely(len <= 0)) + return 0; + + kasan_check_read(src, len); + kasan_check_write(dst, len); + + if (ulen >= 64U) { + if (ulen >= 256U) { + csum_align_src16(&s, &d, &ulen, &acc, &c0); + ulen -= 256U; + do { + csum_tree64_ovf(s, d, &acc, &c0); + csum_tree64_ovf(s + 64, d + 64, &acc, &c0); + csum_tree64_ovf(s + 128, d + 128, &acc, &c0); + csum_tree64_ovf(s + 192, d + 192, &acc, &c0); + s += 256; + d += 256; + } while (!__builtin_sub_overflow(ulen, 256U, &ulen)); + odd = (u64)c0 >> 62; + c0 &= (1LL << 62) - 1; + ulen &= 255U; + if (!ulen) + goto fold; + } + if (ulen >= 64U) { + ulen -= 64U; + do { + csum_tree64_ovf(s, d, &acc, &c0); + s += 64; + d += 64; + } while (!__builtin_sub_overflow(ulen, 64U, &ulen)); + ulen &= 63U; + if (!ulen) + goto fold; + } + } + if (likely(ulen >= 16U)) { + ulen -= 16U; + do { + __uint128_t v = get_unaligned128(s); + + put_unaligned128(v, d); + s += 16; + d += 16; + c0 += __builtin_add_overflow(acc, v, &acc); + } while (!__builtin_sub_overflow(ulen, 16U, &ulen)); + } + if (likely(ulen & 8U)) { + u64 v = get_unaligned((const u64 *)s); + + put_unaligned(v, (u64 *)d); + s += 8; + d += 8; + c0 += __builtin_add_overflow(acc, v, &acc); + } + if (likely(ulen & 4U)) { + u32 v = get_unaligned((const u32 *)s); + + put_unaligned(v, (u32 *)d); + s += 4; + d += 4; + c0 += v; + } + if (likely(ulen & 2U)) { + u16 v = get_unaligned((const u16 *)s); + + put_unaligned(v, (u16 *)d); + s += 2; + d += 2; + c0 += v; + } + if (ulen & 1U) { + u8 v = *s; + + *d = v; + c0 += v; + } +fold: + return csum_fold_final(acc, (u64)c0, odd); +} +EXPORT_SYMBOL(csum_partial_copy_nocheck); + __sum16 csum_ipv6_magic(const struct in6_addr *saddr, const struct in6_addr *daddr, __u32 len, __u8 proto, __wsum csum) -- 2.53.0