From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S966355AbcBDTpt (ORCPT ); Thu, 4 Feb 2016 14:45:49 -0500 Received: from mx1.redhat.com ([209.132.183.28]:38570 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S966331AbcBDTpr (ORCPT ); Thu, 4 Feb 2016 14:45:47 -0500 From: Denys Vlasenko To: Ingo Molnar Cc: Denys Vlasenko , Thomas Graf , Peter Zijlstra , David Rientjes , Andrew Morton , linux-kernel@vger.kernel.org Subject: [PATCH] force inlining of some byteswap operations Date: Thu, 4 Feb 2016 20:45:35 +0100 Message-Id: <1454615136-32308-2-git-send-email-dvlasenk@redhat.com> In-Reply-To: <1454615136-32308-1-git-send-email-dvlasenk@redhat.com> References: <1454615136-32308-1-git-send-email-dvlasenk@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Sometimes gcc mysteriously doesn't inline very small functions we expect to be inlined. See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122 With this .config: http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os, the following functions get deinlined many times. Examples of disassembly: (12 copies, 51 calls): 66 8b 07 mov (%rdi),%ax 55 push %rbp 48 89 e5 mov %rsp,%rbp 86 e0 xchg %ah,%al 5d pop %rbp c3 retq (12 copies, 135 calls): 8b 07 mov (%rdi),%eax 55 push %rbp 48 89 e5 mov %rsp,%rbp 0f c8 bswap %eax 5d pop %rbp c3 retq (2 copies, 20 calls): 48 8b 07 mov (%rdi),%rax 55 push %rbp 48 89 e5 mov %rsp,%rbp 48 0f c8 bswap %rax 5d pop %rbp c3 retq <__swab16p> (16 copies, 146 calls): 55 push %rbp 89 f8 mov %edi,%eax 86 e0 xchg %ah,%al 48 89 e5 mov %rsp,%rbp 5d pop %rbp c3 retq <__swab32p> (43 copies, ~560 calls): 55 push %rbp 89 f8 mov %edi,%eax 0f c8 bswap %eax 48 89 e5 mov %rsp,%rbp 5d pop %rbp c3 retq <__swab64p> (21 copies, 119 calls): 55 push %rbp 48 89 f8 mov %rdi,%rax 48 0f c8 bswap %rax 48 89 e5 mov %rsp,%rbp 5d pop %rbp c3 retq <__swab32s> (6 copies, 47 calls): 8b 07 mov (%rdi),%eax 55 push %rbp 48 89 e5 mov %rsp,%rbp 0f c8 bswap %eax 89 07 mov %eax,(%rdi) 5d pop %rbp c3 retq This patch fixes this via s/inline/__always_inline/. Code size decrease after the patch is ~4.5k: text data bss dec hex filename 92202377 20826112 36417536 149446025 8e85d89 vmlinux 92197848 20826112 36417536 149441496 8e84bd8 vmlinux5_swap_after Signed-off-by: Denys Vlasenko Cc: Ingo Molnar Cc: Thomas Graf Cc: Peter Zijlstra Cc: David Rientjes Cc: Andrew Morton Cc: linux-kernel@vger.kernel.org --- include/uapi/linux/byteorder/big_endian.h | 24 ++++++++++++------------ include/uapi/linux/byteorder/little_endian.h | 24 ++++++++++++------------ include/uapi/linux/swab.h | 10 +++++----- 3 files changed, 29 insertions(+), 29 deletions(-) diff --git a/include/uapi/linux/byteorder/big_endian.h b/include/uapi/linux/byteorder/big_endian.h index 6723744..cdab17a 100644 --- a/include/uapi/linux/byteorder/big_endian.h +++ b/include/uapi/linux/byteorder/big_endian.h @@ -40,51 +40,51 @@ #define __cpu_to_be16(x) ((__force __be16)(__u16)(x)) #define __be16_to_cpu(x) ((__force __u16)(__be16)(x)) -static inline __le64 __cpu_to_le64p(const __u64 *p) +static __always_inline __le64 __cpu_to_le64p(const __u64 *p) { return (__force __le64)__swab64p(p); } -static inline __u64 __le64_to_cpup(const __le64 *p) +static __always_inline __u64 __le64_to_cpup(const __le64 *p) { return __swab64p((__u64 *)p); } -static inline __le32 __cpu_to_le32p(const __u32 *p) +static __always_inline __le32 __cpu_to_le32p(const __u32 *p) { return (__force __le32)__swab32p(p); } -static inline __u32 __le32_to_cpup(const __le32 *p) +static __always_inline __u32 __le32_to_cpup(const __le32 *p) { return __swab32p((__u32 *)p); } -static inline __le16 __cpu_to_le16p(const __u16 *p) +static __always_inline __le16 __cpu_to_le16p(const __u16 *p) { return (__force __le16)__swab16p(p); } -static inline __u16 __le16_to_cpup(const __le16 *p) +static __always_inline __u16 __le16_to_cpup(const __le16 *p) { return __swab16p((__u16 *)p); } -static inline __be64 __cpu_to_be64p(const __u64 *p) +static __always_inline __be64 __cpu_to_be64p(const __u64 *p) { return (__force __be64)*p; } -static inline __u64 __be64_to_cpup(const __be64 *p) +static __always_inline __u64 __be64_to_cpup(const __be64 *p) { return (__force __u64)*p; } -static inline __be32 __cpu_to_be32p(const __u32 *p) +static __always_inline __be32 __cpu_to_be32p(const __u32 *p) { return (__force __be32)*p; } -static inline __u32 __be32_to_cpup(const __be32 *p) +static __always_inline __u32 __be32_to_cpup(const __be32 *p) { return (__force __u32)*p; } -static inline __be16 __cpu_to_be16p(const __u16 *p) +static __always_inline __be16 __cpu_to_be16p(const __u16 *p) { return (__force __be16)*p; } -static inline __u16 __be16_to_cpup(const __be16 *p) +static __always_inline __u16 __be16_to_cpup(const __be16 *p) { return (__force __u16)*p; } diff --git a/include/uapi/linux/byteorder/little_endian.h b/include/uapi/linux/byteorder/little_endian.h index d876736..4b93f2b 100644 --- a/include/uapi/linux/byteorder/little_endian.h +++ b/include/uapi/linux/byteorder/little_endian.h @@ -40,51 +40,51 @@ #define __cpu_to_be16(x) ((__force __be16)__swab16((x))) #define __be16_to_cpu(x) __swab16((__force __u16)(__be16)(x)) -static inline __le64 __cpu_to_le64p(const __u64 *p) +static __always_inline __le64 __cpu_to_le64p(const __u64 *p) { return (__force __le64)*p; } -static inline __u64 __le64_to_cpup(const __le64 *p) +static __always_inline __u64 __le64_to_cpup(const __le64 *p) { return (__force __u64)*p; } -static inline __le32 __cpu_to_le32p(const __u32 *p) +static __always_inline __le32 __cpu_to_le32p(const __u32 *p) { return (__force __le32)*p; } -static inline __u32 __le32_to_cpup(const __le32 *p) +static __always_inline __u32 __le32_to_cpup(const __le32 *p) { return (__force __u32)*p; } -static inline __le16 __cpu_to_le16p(const __u16 *p) +static __always_inline __le16 __cpu_to_le16p(const __u16 *p) { return (__force __le16)*p; } -static inline __u16 __le16_to_cpup(const __le16 *p) +static __always_inline __u16 __le16_to_cpup(const __le16 *p) { return (__force __u16)*p; } -static inline __be64 __cpu_to_be64p(const __u64 *p) +static __always_inline __be64 __cpu_to_be64p(const __u64 *p) { return (__force __be64)__swab64p(p); } -static inline __u64 __be64_to_cpup(const __be64 *p) +static __always_inline __u64 __be64_to_cpup(const __be64 *p) { return __swab64p((__u64 *)p); } -static inline __be32 __cpu_to_be32p(const __u32 *p) +static __always_inline __be32 __cpu_to_be32p(const __u32 *p) { return (__force __be32)__swab32p(p); } -static inline __u32 __be32_to_cpup(const __be32 *p) +static __always_inline __u32 __be32_to_cpup(const __be32 *p) { return __swab32p((__u32 *)p); } -static inline __be16 __cpu_to_be16p(const __u16 *p) +static __always_inline __be16 __cpu_to_be16p(const __u16 *p) { return (__force __be16)__swab16p(p); } -static inline __u16 __be16_to_cpup(const __be16 *p) +static __always_inline __u16 __be16_to_cpup(const __be16 *p) { return __swab16p((__u16 *)p); } diff --git a/include/uapi/linux/swab.h b/include/uapi/linux/swab.h index 0e011eb..3f10e53 100644 --- a/include/uapi/linux/swab.h +++ b/include/uapi/linux/swab.h @@ -151,7 +151,7 @@ static inline __attribute_const__ __u32 __fswahb32(__u32 val) * __swab16p - return a byteswapped 16-bit value from a pointer * @p: pointer to a naturally-aligned 16-bit value */ -static inline __u16 __swab16p(const __u16 *p) +static __always_inline __u16 __swab16p(const __u16 *p) { #ifdef __arch_swab16p return __arch_swab16p(p); @@ -164,7 +164,7 @@ static inline __u16 __swab16p(const __u16 *p) * __swab32p - return a byteswapped 32-bit value from a pointer * @p: pointer to a naturally-aligned 32-bit value */ -static inline __u32 __swab32p(const __u32 *p) +static __always_inline __u32 __swab32p(const __u32 *p) { #ifdef __arch_swab32p return __arch_swab32p(p); @@ -177,7 +177,7 @@ static inline __u32 __swab32p(const __u32 *p) * __swab64p - return a byteswapped 64-bit value from a pointer * @p: pointer to a naturally-aligned 64-bit value */ -static inline __u64 __swab64p(const __u64 *p) +static __always_inline __u64 __swab64p(const __u64 *p) { #ifdef __arch_swab64p return __arch_swab64p(p); @@ -232,7 +232,7 @@ static inline void __swab16s(__u16 *p) * __swab32s - byteswap a 32-bit value in-place * @p: pointer to a naturally-aligned 32-bit value */ -static inline void __swab32s(__u32 *p) +static __always_inline void __swab32s(__u32 *p) { #ifdef __arch_swab32s __arch_swab32s(p); @@ -245,7 +245,7 @@ static inline void __swab32s(__u32 *p) * __swab64s - byteswap a 64-bit value in-place * @p: pointer to a naturally-aligned 64-bit value */ -static inline void __swab64s(__u64 *p) +static __always_inline void __swab64s(__u64 *p) { #ifdef __arch_swab64s __arch_swab64s(p); -- 1.8.1.4