Re: [PATCH 2/2] x86: Shorten RESET_CALL_DEPTH

[Andrew . Cooper3]

On 15/05/2023 10:28 am, Peter Zijlstra wrote:
> RESET_CALL_DEPTH is a pretty fat monster and blows up UNTRAIN_RET to
> 20 bytes:
>
>   19:       48 c7 c0 80 00 00 00    mov    $0x80,%rax
>   20:       48 c1 e0 38             shl    $0x38,%rax
>   24:       65 48 89 04 25 00 00 00 00      mov    %rax,%gs:0x0     29: R_X86_64_32S        pcpu_hot+0x10
>
> Shrink it by 4 bytes:
>
>   0:   31 c0                   xor    %eax,%eax
>   2:   48 0f ba e8 3f          bts    $0x3f,%rax
>   7:   65 48 89 04 25 00 00 00 00      mov    %rax,%gs:0x0
>
> Shrink RESET_CALL_DEPTH_FROM_CALL by 5 bytes by only setting al, the
> other bits are shifted out (the same could be done for
> RESET_CALL_DEPTH, but the xor+bts sequence has less depencies due to
> the zeroing).
>
> Suggested-by: Andrew.Cooper3@xxxxxxxxxx

Andrew Cooper <andrew.cooper3@xxxxxxxxxx>

> Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> ---
>  arch/x86/include/asm/nospec-branch.h |    6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> --- a/arch/x86/include/asm/nospec-branch.h
> +++ b/arch/x86/include/asm/nospec-branch.h
> @@ -84,12 +84,12 @@
>  	movq	$-1, PER_CPU_VAR(pcpu_hot + X86_call_depth);
>  
>  #define RESET_CALL_DEPTH					\
> -	mov	$0x80, %rax;					\
> -	shl	$56, %rax;					\
> +	xor	%eax, %eax;					\
> +	bts	$59, %rax;					\

$63 ?

The disassembly looks correct.

~Andrew