Re: [tip:x86/setup] x86, setup: Use string copy operation to optimze copy in kernel compression

[Yinghai Lu <yinghai@kernel.org>]

On 10/07/2010 10:40 PM, tip-bot for Zhao Yakui wrote:
> Commit-ID:  68f4d5a00adaab33b136fce2c72d5c377b39b0b0
> Gitweb:     http://git.kernel.org/tip/68f4d5a00adaab33b136fce2c72d5c377b39b0b0
> Author:     Zhao Yakui <yakui.zhao@intel.com>
> AuthorDate: Fri, 8 Oct 2010 09:47:33 +0800
> Committer:  H. Peter Anvin <hpa@zytor.com>
> CommitDate: Thu, 7 Oct 2010 21:23:09 -0700
> 
> x86, setup: Use string copy operation to optimze copy in kernel compression
> 
> The kernel decompression code parses the ELF header and then copies
> the segment to the corresponding destination.  Currently it uses slow
> byte-copy code.  This patch makes it use the string copy operations
> instead.
> 
> In the test the copy performance can be improved very significantly after using
> the string copy operation mechanism.
>         1. The copy time can be reduced from 150ms to 20ms on one Atom machine
> 	2. The copy time can be reduced about 80% on another machine
> 		The time is reduced from 7ms to 1.5ms when using 32-bit kernel.
> 		The time is reduced from 10ms to 2ms when using 64-bit kernel.
> 
> Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
> LKML-Reference: <1286502453-7043-1-git-send-email-yakui.zhao@intel.com>
> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
> ---
>  arch/x86/boot/compressed/misc.c |   29 +++++++++++++++++++++++------
>  1 files changed, 23 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/x86/boot/compressed/misc.c b/arch/x86/boot/compressed/misc.c
> index 8f7bef8..23f315c 100644
> --- a/arch/x86/boot/compressed/misc.c
> +++ b/arch/x86/boot/compressed/misc.c
> @@ -229,18 +229,35 @@ void *memset(void *s, int c, size_t n)
>  		ss[i] = c;
>  	return s;
>  }
> -
> +#ifdef CONFIG_X86_32
>  void *memcpy(void *dest, const void *src, size_t n)
>  {
> -	int i;
> -	const char *s = src;
> -	char *d = dest;
> +	int d0, d1, d2;
> +	asm volatile(
> +		"rep ; movsl\n\t"
> +		"movl %4,%%ecx\n\t"
> +		"rep ; movsb\n\t"
> +		: "=&c" (d0), "=&D" (d1), "=&S" (d2)
> +		: "0" (n >> 2), "g" (n & 3), "1" (dest), "2" (src)
> +		: "memory");
>  
> -	for (i = 0; i < n; i++)
> -		d[i] = s[i];
>  	return dest;
>  }
> +#else
> +void *memcpy(void *dest, const void *src, size_t n)
> +{
> +	long d0, d1, d2;
> +	asm volatile(
> +		"rep ; movsq\n\t"
> +		"movq %4,%%rcx\n\t"
> +		"rep ; movsb\n\t"
> +		: "=&c" (d0), "=&D" (d1), "=&S" (d2)
> +		: "0" (n >> 3), "g" (n & 7), "1" (dest), "2" (src)
> +		: "memory");
>  
> +	return dest;
> +}
> +#endif
>  
>  static void error(char *x)
>  {

wonder if it would have problem with some old AMD K8 systems.

in amd.c

        /* On C+ stepping K8 rep microcode works well for copy/memset */
        if (c->x86 == 0xf) {
                u32 level;

                level = cpuid_eax(1);
                if ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
                        set_cpu_cap(c, X86_FEATURE_REP_GOOD);
...

        }
        if (c->x86 >= 0x10)
                set_cpu_cap(c, X86_FEATURE_REP_GOOD);

Yinghai