Commit adfcf423 authored by Linus Torvalds's avatar Linus Torvalds
Browse files

x86: don't use REP_GOOD or ERMS for user memory copies 



The modern target to use is FSRM (Fast Short REP MOVS), and the other
cases should only be used for bigger areas (ie mainly things like page
clearing).

Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
parent 20f3337d

arch/x86/include/asm/uaccess_64.h

+5 −10
Original line number Diff line number Diff line
@@ -18,9 +18,7 @@ 


/* Handles exceptions in both to and from, but doesn't do access_ok */

__must_check unsigned long

copy_user_enhanced_fast_string(void *to, const void *from, unsigned len);

__must_check unsigned long

copy_user_generic_string(void *to, const void *from, unsigned len);

copy_user_fast_string(void *to, const void *from, unsigned len);

__must_check unsigned long

copy_user_generic_unrolled(void *to, const void *from, unsigned len);
@@ -30,15 +28,12 @@ copy_user_generic(void *to, const void *from, unsigned len) 
	unsigned ret;



	/*

	 * If CPU has ERMS feature, use copy_user_enhanced_fast_string.

	 * Otherwise, if CPU has rep_good feature, use copy_user_generic_string.

	 * If CPU has FSRM feature, use 'rep movs'.

	 * Otherwise, use copy_user_generic_unrolled.

	 */

	alternative_call_2(copy_user_generic_unrolled,

			 copy_user_generic_string,

			 X86_FEATURE_REP_GOOD,

			 copy_user_enhanced_fast_string,

			 X86_FEATURE_ERMS,

	alternative_call(copy_user_generic_unrolled,

			 copy_user_fast_string,

			 X86_FEATURE_FSRM,

			 ASM_OUTPUT2("=a" (ret), "=D" (to), "=S" (from),

				     "=d" (len)),

			 "1" (to), "2" (from), "3" (len)

arch/x86/lib/copy_user_64.S

+7 −44
Original line number Diff line number Diff line
@@ -104,8 +104,8 @@ SYM_FUNC_START(copy_user_generic_unrolled) 
SYM_FUNC_END(copy_user_generic_unrolled)

EXPORT_SYMBOL(copy_user_generic_unrolled)



/* Some CPUs run faster using the string copy instructions.

 * This is also a lot simpler. Use them when possible.

/*

 * Some CPUs support FSRM for Fast Short REP MOVS.

 *

 * Only 4GB of copy is supported. This shouldn't be a problem

 * because the kernel normally only writes from/to page sized chunks
@@ -122,58 +122,21 @@ EXPORT_SYMBOL(copy_user_generic_unrolled) 
 * Output:

 * eax uncopied bytes or 0 if successful.

 */

SYM_FUNC_START(copy_user_generic_string)

SYM_FUNC_START(copy_user_fast_string)

	ASM_STAC

	cmpl $8,%edx

	jb 2f		/* less than 8 bytes, go to byte copy loop */

	ALIGN_DESTINATION

	movl %edx,%ecx

	shrl $3,%ecx

	andl $7,%edx

1:	rep movsq

2:	movl %edx,%ecx

3:	rep movsb

1:	rep movsb

	xorl %eax,%eax

	ASM_CLAC

	RET



11:	leal (%rdx,%rcx,8),%ecx

12:	movl %ecx,%edx		/* ecx is zerorest also */

	jmp .Lcopy_user_handle_tail



	_ASM_EXTABLE_CPY(1b, 11b)

	_ASM_EXTABLE_CPY(3b, 12b)

SYM_FUNC_END(copy_user_generic_string)

EXPORT_SYMBOL(copy_user_generic_string)



/*

 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.

 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.

 *

 * Input:

 * rdi destination

 * rsi source

 * rdx count

 *

 * Output:

 * eax uncopied bytes or 0 if successful.

 */

SYM_FUNC_START(copy_user_enhanced_fast_string)

	ASM_STAC

	/* CPUs without FSRM should avoid rep movsb for short copies */

	ALTERNATIVE "cmpl $64, %edx; jb copy_user_short_string", "", X86_FEATURE_FSRM

	movl %edx,%ecx

1:	rep movsb

	xorl %eax,%eax

12:	movl %ecx,%eax		/* ecx is zerorest also */

	ASM_CLAC

	RET



12:	movl %ecx,%edx		/* ecx is zerorest also */

	jmp .Lcopy_user_handle_tail



	_ASM_EXTABLE_CPY(1b, 12b)

SYM_FUNC_END(copy_user_enhanced_fast_string)

EXPORT_SYMBOL(copy_user_enhanced_fast_string)

SYM_FUNC_END(copy_user_fast_string)

EXPORT_SYMBOL(copy_user_fast_string)



/*

 * Try to copy last bytes and clear the rest if needed.