[PATCH v2] arm64/lib: add accelerated do_csum for arm64
do_csum() in lib/checksum.c is too slow in ARM64, and we can use assembly and algorithm to accelerate it. V2: Fix compile warning reported by kbuild test robot Signed-off-by: huhai --- arch/arm64/include/asm/checksum.h | 3 + arch/arm64/lib/Makefile | 2 +- arch/arm64/lib/checksum.c | 145 ++ 3 files changed, 149 insertions(+), 1 deletion(-) create mode 100644 arch/arm64/lib/checksum.c diff --git a/arch/arm64/include/asm/checksum.h b/arch/arm64/include/asm/checksum.h index 0b6f5a7d4027..0d7b80fb300e 100644 --- a/arch/arm64/include/asm/checksum.h +++ b/arch/arm64/include/asm/checksum.h @@ -26,6 +26,9 @@ static inline __sum16 csum_fold(__wsum csum) } #define csum_fold csum_fold +#define do_csum do_csum +unsigned int do_csum(const unsigned char *buff, unsigned int len); + static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl) { __uint128_t tmp; diff --git a/arch/arm64/lib/Makefile b/arch/arm64/lib/Makefile index 69ff9887f724..4134730a121b 100644 --- a/arch/arm64/lib/Makefile +++ b/arch/arm64/lib/Makefile @@ -1,5 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 -lib-y := clear_user.o delay.o copy_from_user.o\ +lib-y := checksum.o clear_user.o delay.o copy_from_user.o \ copy_to_user.o copy_in_user.o copy_page.o\ clear_page.o memchr.o memcpy.o memmove.o memset.o\ memcmp.o strcmp.o strncmp.o strlen.o strnlen.o \ diff --git a/arch/arm64/lib/checksum.c b/arch/arm64/lib/checksum.c new file mode 100644 index ..5df8c6f9a8ac --- /dev/null +++ b/arch/arm64/lib/checksum.c @@ -0,0 +1,145 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Authors: huhai + * Copyright (C) 2019,Tianjin KYLIN Information Technology Co., Ltd. + * + * This file contains network checksum routines that are better done + * in an architecture-specific manner due to speed. + * + * Acknowledgements: + * This file is based on arch/x86/lib/csum-partial_64.c and + * arch/alpha/lib/checksum.c, which was written by Thomas Gleixner + * and Rick Gorton respectively. + */ + +#include +#include +#include + +static inline unsigned short from64to16(unsigned long x) +{ + /* +* Using extract instructions is a bit more efficient +* than the original shift/bitmask version. +*/ + union { + unsigned long ul; + unsigned intui[2]; + unsigned short us[4]; + } in_v, tmp_v, out_v; + + in_v.ul = x; + tmp_v.ul = (unsigned long) in_v.ui[0] + (unsigned long) in_v.ui[1]; + + /* +* Since the bits of tmp_v.sh[3] are going to always be zero, +* we don't have to bother to add that in. +*/ + out_v.ul = (unsigned long) tmp_v.us[0] + (unsigned long) tmp_v.us[1] + + (unsigned long) tmp_v.us[2]; + + /* Similarly, out_v.us[2] is always zero for the final add. */ + return out_v.us[0] + out_v.us[1]; +} + +/* + * Do a 64-bit checksum on an arbitrary memory area. + * Returns a 16bit checksum. + */ +unsigned int do_csum(const unsigned char *buff, unsigned len) +{ + unsigned odd, count; + unsigned long result = 0; + + if (unlikely(len == 0)) + return result; + odd = 1 & (unsigned long) buff; + if (odd) { + result = *buff << 8; + len--; + buff++; + } + count = len >> 1; /* nr of 16-bit words.. */ + if (count) { + if (2 & (unsigned long) buff) { + result += *(unsigned short *)buff; + count--; + len -= 2; + buff += 2; + } + count >>= 1;/* nr of 32-bit words.. */ + if (count) { + unsigned long zero = 0; + unsigned count64; + + if (4 & (unsigned long) buff) { + result += *(unsigned int *) buff; + count--; + len -= 4; + buff += 4; + } + count >>= 1;/* nr of 64-bit words.. */ + + /* main loop using 64byte blocks */ + count64 = count >> 3; + while (count64) { + __asm__ __volatile__( + "ldr x3, [%x1, #0]\n" + "adds %x0, %x0, x3\n" + "ldr x3, [%x1, #8]\n" + "adcs %x0, %x0, x3\n" + "ldr x3, [%x1, #16]\n" +
[PATCH] arm64/lib: add accelerated do_csum for arm64
do_csum() in lib/checksum.c is too slow in ARM64, and we can use assembly and algorithm to accelerate it. Signed-off-by: huhai --- arch/arm64/include/asm/checksum.h | 3 + arch/arm64/lib/Makefile | 2 +- arch/arm64/lib/checksum.c | 144 ++ 3 files changed, 148 insertions(+), 1 deletion(-) create mode 100644 arch/arm64/lib/checksum.c diff --git a/arch/arm64/include/asm/checksum.h b/arch/arm64/include/asm/checksum.h index 0b6f5a7d4027..0d7b80fb300e 100644 --- a/arch/arm64/include/asm/checksum.h +++ b/arch/arm64/include/asm/checksum.h @@ -26,6 +26,9 @@ static inline __sum16 csum_fold(__wsum csum) } #define csum_fold csum_fold +#define do_csum do_csum +unsigned int do_csum(const unsigned char *buff, unsigned int len); + static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl) { __uint128_t tmp; diff --git a/arch/arm64/lib/Makefile b/arch/arm64/lib/Makefile index 69ff9887f724..4134730a121b 100644 --- a/arch/arm64/lib/Makefile +++ b/arch/arm64/lib/Makefile @@ -1,5 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 -lib-y := clear_user.o delay.o copy_from_user.o\ +lib-y := checksum.o clear_user.o delay.o copy_from_user.o \ copy_to_user.o copy_in_user.o copy_page.o\ clear_page.o memchr.o memcpy.o memmove.o memset.o\ memcmp.o strcmp.o strncmp.o strlen.o strnlen.o \ diff --git a/arch/arm64/lib/checksum.c b/arch/arm64/lib/checksum.c new file mode 100644 index ..6931ef13ef87 --- /dev/null +++ b/arch/arm64/lib/checksum.c @@ -0,0 +1,144 @@ +/* + * arch/arm64/lib/checksum.c + * + * This file contains network checksum routines that are better done + * in an architecture-specific manner due to speed. + * + * Acknowledgements: + * This file is based on arch/x86/lib/csum-partial_64.c and + * arch/alpha/lib/checksum.c, which was written by Thomas Gleixner + * and Rick Gorton respectively. + */ + +#include +#include +#include + +static inline unsigned short from64to16(unsigned long x) +{ + /* Using extract instructions is a bit more efficient +* than the original shift/bitmask version. +*/ + + union { + unsigned long ul; + unsigned intui[2]; + unsigned short us[4]; + } in_v, tmp_v, out_v; + + in_v.ul = x; + tmp_v.ul = (unsigned long) in_v.ui[0] + (unsigned long) in_v.ui[1]; + + /* Since the bits of tmp_v.sh[3] are going to always be zero, +* we don't have to bother to add that in. +*/ + out_v.ul = (unsigned long) tmp_v.us[0] + (unsigned long) tmp_v.us[1] + + (unsigned long) tmp_v.us[2]; + + /* Similarly, out_v.us[2] is always zero for the final add. */ + return out_v.us[0] + out_v.us[1]; +} + +/* + * Do a 64-bit checksum on an arbitrary memory area. + * Returns a 16bit checksum. + */ +unsigned int do_csum(const unsigned char *buff, unsigned len) +{ + unsigned odd, count; + unsigned long result = 0; + + if (unlikely(len == 0)) + return result; + odd = 1 & (unsigned long) buff; + if (odd) { + result = *buff << 8; + len--; + buff++; + } + count = len >> 1; /* nr of 16-bit words.. */ + if (count) { + if (2 & (unsigned long) buff) { + result += *(unsigned short *)buff; + count--; + len -= 2; + buff += 2; + } + count >>= 1;/* nr of 32-bit words.. */ + if (count) { + unsigned long zero; + unsigned long tmp1; + unsigned count64; + + if (4 & (unsigned long) buff) { + result += *(unsigned int *) buff; + count--; + len -= 4; + buff += 4; + } + count >>= 1;/* nr of 64-bit words.. */ + + /* main loop using 64byte blocks */ + zero = 0; + count64 = count >> 3; + while (count64) { + __asm__ __volatile__( + "ldr %x3, [%x1, #0]\n" + "adds %x0, %x0, %x3\n" + "ldr %x3, [%x1, #8]\n" + "adcs %x0, %x0, %x3\n" + "ldr %x3, [%x1, #16]\n" + "adcs %x0, %x0, %x3\n" + "ldr %x3, [%x1, #24]\n&q