Re: [PATCH] x86/lib: Use REP MOVSB for large memcpy() on ERMS CPUs without FSRM

[Kiryl Shutsemau <kas@kernel.org>]

On Sat, Oct 17, 2026 at 12:27:25PM -0700, Usama Arif wrote:
> diff --git a/arch/x86/lib/memcpy_64.S b/arch/x86/lib/memcpy_64.S
> index 12a23fa7c44c5..6b3e203b6087c 100644
> --- a/arch/x86/lib/memcpy_64.S
> +++ b/arch/x86/lib/memcpy_64.S
> @@ -31,9 +31,10 @@
>   * which the compiler could/should do much better anyway.
>   */
>  SYM_TYPED_FUNC_START(__memcpy)
> -	ALTERNATIVE "jmp memcpy_orig", "", X86_FEATURE_FSRM
> +	/* The FSRM destination move shares the space used by the ERMS jump. */
> +	ALTERNATIVE_2 "jmp memcpy_orig", "jmp memcpy_erms", X86_FEATURE_ERMS, \
> +		      "movq %rdi, %rax", X86_FEATURE_FSRM
>  
> -	movq %rdi, %rax

Moving MOVQ inside ALTERNATIVE_2 is ugly...

And it only turns a 5-byte NOP into a 2-byte one on FSRM.
What is the point?

>  	movq %rdx, %rcx
>  	rep movsb
>  	RET
> @@ -171,3 +172,11 @@ SYM_FUNC_START_LOCAL(memcpy_orig)
>  	RET
>  SYM_FUNC_END(memcpy_orig)
>  
> +SYM_FUNC_START_LOCAL(memcpy_erms)
> +	cmpq $2048, %rdx
> +	jb memcpy_orig
> +	movq %rdi, %rax
> +	movq %rdx, %rcx
> +	rep movsb
> +	RET
> +SYM_FUNC_END(memcpy_erms)

Copies below 2048 now take two jumps to reach memcpy_orig() instead of
one, and David's point is that nearly all kernel copies are that small.

Can the check go in memcpy_orig() after the existing 32-byte compare?

-- 
  Kiryl Shutsemau / Kirill A. Shutemov