From patchwork Sat Jan 15 21:39:50 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Luc Pelletier X-Patchwork-Id: 105873 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 55866A034D; Sat, 15 Jan 2022 22:40:48 +0100 (CET) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id DDF8E406A2; Sat, 15 Jan 2022 22:40:47 +0100 (CET) Received: from mail-qk1-f175.google.com (mail-qk1-f175.google.com [209.85.222.175]) by mails.dpdk.org (Postfix) with ESMTP id 638CD4013F; Sat, 15 Jan 2022 22:40:46 +0100 (CET) Received: by mail-qk1-f175.google.com with SMTP id u3so4729138qku.1; Sat, 15 Jan 2022 13:40:46 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=DKN5b91/LsO2GhDKf2EywNX+avjQv5jzNpTWqA99K2E=; b=fgv9wFesyzntZVYbGlJsQRiV2Buo4QYecPXWIkENGPktwy34i6HaB5CpnGmWXldryN CV7hJuh4Lyhq9P6plpCbsZAEtv7FsXYVt7/PSsl6l247HWcqxfRw1aGOkV4LQFJpFXXY m7oSh6Qnx3miUchnB3gR+MVLg7wHkL6J5jBgW9vLnBGLd426sAT3KCkdoVyr/6q7Nmqx czMEd2T3hUfFRIZG+GZxj6HwUr6kS8gY8l+QFn7OxQKoqExAy5cd0BfZ6n36/w/1fZYI L3S5JineAmiFTkb9GKN4BDJzDwnEwCsxOck7cxsAh5Kk1EOYpsZfaMq8dO4DSQ8IQzTM hz/A== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=DKN5b91/LsO2GhDKf2EywNX+avjQv5jzNpTWqA99K2E=; b=wz2BkKkqeUoYTQ7N7iFZTossHURggo31N8c4hWLAqFJiPGcua+aac9LiZXxyw3FGdq 9Zp0QWEKaFVuI1XdqXxZLppBdnPO9ZETSEkIEbzL184czR9QOltH8D94igrj9gjRcKyE I7dEUjpUTbCc469pgXJv3FFqgyvuqKeIE3CKELKb3tmIFDpyFNQSW2vjkh8wjzx8ePqi qPzxtzaaJPNPIh0hx9IeREEkVrBJtArTs1zR3PJYKrLARTBgrk7DC5NGf3iS2HtXvX7/ XqOgQ/NrtYJfJ7lxo4jnuKWXO/4AN73G9erFV7psXk6FiNuFGBnorimhPlEkJRqYfOwR 7pAg== X-Gm-Message-State: AOAM530owFmMOlbmi3lSD2Aky/IamPEvDwBFJrvE57NFiHwW/aQOX5kH +2sBR5659BG6zSdUbT14WRE= X-Google-Smtp-Source: ABdhPJw4OXXPgRxvm6s46rz3M6/U8X16qptKWuGaA5GvAhEyeveT6R/fO8mBeg9FNtF9fjigJNlG+A== X-Received: by 2002:a37:6947:: with SMTP id e68mr4290307qkc.26.1642282845726; Sat, 15 Jan 2022 13:40:45 -0800 (PST) Received: from localhost.localdomain (bras-base-hullpq2034w-grc-18-74-15-213-135.dsl.bell.ca. [74.15.213.135]) by smtp.gmail.com with ESMTPSA id j22sm6403775qko.46.2022.01.15.13.40.44 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Sat, 15 Jan 2022 13:40:45 -0800 (PST) From: Luc Pelletier To: bruce.richardson@intel.com, konstantin.ananyev@intel.com Cc: dev@dpdk.org, Luc Pelletier , Xiaoyun Li , stable@dpdk.org Subject: [PATCH v2] eal: fix unaligned loads/stores in rte_memcpy_generic Date: Sat, 15 Jan 2022 16:39:50 -0500 Message-Id: <20220115213949.449313-1-lucp.at.work@gmail.com> In-Reply-To: <20220115194102.444140-1-lucp.at.work@gmail.com> References: <20220115194102.444140-1-lucp.at.work@gmail.com> MIME-Version: 1.0 X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Calls to rte_memcpy_generic could result in unaligned loads/stores for 1 < n < 16. This is undefined behavior according to the C standard, and it gets flagged by the clang undefined behavior sanitizer. rte_memcpy_generic is called with unaligned src and dst addresses. When 1 < n < 16, the code would cast both src and dst to a qword, dword or word pointer, without verifying the alignment of src/dst. The code was changed to use a for loop to copy the bytes one by one. Experimentation on compiler explorer indicates that gcc 7+ (released in 2017) and clang 7+ (released in 2018) both optimize out the for loop with the least number of memory loads and stores, if n is known at compile-time. If n is only known at compile-time, gcc and clang have different behaviour but they both seem to recognize that a memcpy is being done. More recent versions of both gcc/clang seem to also produce even more optimized results. Fixes: d35cc1fe6a7a ("eal/x86: revert select optimized memcpy at run-time") Cc: Xiaoyun Li Cc: stable@dpdk.org Signed-off-by: Luc Pelletier --- I forgot that code under x86 also needs to compile for 32-bit (obviously). So, I did some more experimentation and replaced the assembly code with a regular for loop. Explanations are in the updated commit message. Experimentation was done on compiler explorer here: https://godbolt.org/z/zK54rzPEn lib/eal/x86/include/rte_memcpy.h | 82 ++++++++------------------------ 1 file changed, 20 insertions(+), 62 deletions(-) diff --git a/lib/eal/x86/include/rte_memcpy.h b/lib/eal/x86/include/rte_memcpy.h index 1b6c6e585f..e422397e49 100644 --- a/lib/eal/x86/include/rte_memcpy.h +++ b/lib/eal/x86/include/rte_memcpy.h @@ -45,6 +45,23 @@ extern "C" { static __rte_always_inline void * rte_memcpy(void *dst, const void *src, size_t n); +/** + * Copy bytes from one location to another, + * locations should not overlap. + * Use with unaligned src/dst, and n <= 15. + */ +static __rte_always_inline void * +rte_mov15_or_less_unaligned(void *dst, const void *src, size_t n) +{ + void *ret = dst; + for (; n; n--) { + *((char *)dst) = *((const char *) src); + dst = ((char *)dst) + 1; + src = ((const char *)src) + 1; + } + return ret; +} + #if defined __AVX512F__ && defined RTE_MEMCPY_AVX512 #define ALIGNMENT_MASK 0x3F @@ -171,8 +188,6 @@ rte_mov512blocks(uint8_t *dst, const uint8_t *src, size_t n) static __rte_always_inline void * rte_memcpy_generic(void *dst, const void *src, size_t n) { - uintptr_t dstu = (uintptr_t)dst; - uintptr_t srcu = (uintptr_t)src; void *ret = dst; size_t dstofss; size_t bits; @@ -181,24 +196,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) * Copy less than 16 bytes */ if (n < 16) { - if (n & 0x01) { - *(uint8_t *)dstu = *(const uint8_t *)srcu; - srcu = (uintptr_t)((const uint8_t *)srcu + 1); - dstu = (uintptr_t)((uint8_t *)dstu + 1); - } - if (n & 0x02) { - *(uint16_t *)dstu = *(const uint16_t *)srcu; - srcu = (uintptr_t)((const uint16_t *)srcu + 1); - dstu = (uintptr_t)((uint16_t *)dstu + 1); - } - if (n & 0x04) { - *(uint32_t *)dstu = *(const uint32_t *)srcu; - srcu = (uintptr_t)((const uint32_t *)srcu + 1); - dstu = (uintptr_t)((uint32_t *)dstu + 1); - } - if (n & 0x08) - *(uint64_t *)dstu = *(const uint64_t *)srcu; - return ret; + return rte_mov15_or_less_unaligned(dst, src, n); } /** @@ -379,8 +377,6 @@ rte_mov128blocks(uint8_t *dst, const uint8_t *src, size_t n) static __rte_always_inline void * rte_memcpy_generic(void *dst, const void *src, size_t n) { - uintptr_t dstu = (uintptr_t)dst; - uintptr_t srcu = (uintptr_t)src; void *ret = dst; size_t dstofss; size_t bits; @@ -389,25 +385,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) * Copy less than 16 bytes */ if (n < 16) { - if (n & 0x01) { - *(uint8_t *)dstu = *(const uint8_t *)srcu; - srcu = (uintptr_t)((const uint8_t *)srcu + 1); - dstu = (uintptr_t)((uint8_t *)dstu + 1); - } - if (n & 0x02) { - *(uint16_t *)dstu = *(const uint16_t *)srcu; - srcu = (uintptr_t)((const uint16_t *)srcu + 1); - dstu = (uintptr_t)((uint16_t *)dstu + 1); - } - if (n & 0x04) { - *(uint32_t *)dstu = *(const uint32_t *)srcu; - srcu = (uintptr_t)((const uint32_t *)srcu + 1); - dstu = (uintptr_t)((uint32_t *)dstu + 1); - } - if (n & 0x08) { - *(uint64_t *)dstu = *(const uint64_t *)srcu; - } - return ret; + return rte_mov15_or_less_unaligned(dst, src, n); } /** @@ -672,8 +650,6 @@ static __rte_always_inline void * rte_memcpy_generic(void *dst, const void *src, size_t n) { __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8; - uintptr_t dstu = (uintptr_t)dst; - uintptr_t srcu = (uintptr_t)src; void *ret = dst; size_t dstofss; size_t srcofs; @@ -682,25 +658,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) * Copy less than 16 bytes */ if (n < 16) { - if (n & 0x01) { - *(uint8_t *)dstu = *(const uint8_t *)srcu; - srcu = (uintptr_t)((const uint8_t *)srcu + 1); - dstu = (uintptr_t)((uint8_t *)dstu + 1); - } - if (n & 0x02) { - *(uint16_t *)dstu = *(const uint16_t *)srcu; - srcu = (uintptr_t)((const uint16_t *)srcu + 1); - dstu = (uintptr_t)((uint16_t *)dstu + 1); - } - if (n & 0x04) { - *(uint32_t *)dstu = *(const uint32_t *)srcu; - srcu = (uintptr_t)((const uint32_t *)srcu + 1); - dstu = (uintptr_t)((uint32_t *)dstu + 1); - } - if (n & 0x08) { - *(uint64_t *)dstu = *(const uint64_t *)srcu; - } - return ret; + return rte_mov15_or_less_unaligned(dst, src, n); } /**