Re: [PATCH] x86/entry: Remove UNTRAIN_RET from native_irq_return_ldt

[Peter Zijlstra]

On Wed, Jul 13, 2022 at 09:58:08PM +0200, Alexandre Chartre wrote:
> UNTRAIN_RET is not needed in native_irq_return_ldt because ret untrain
> has already be done at this point. In addition, when the RETBleed
> mitigation is IBPB, UNTRAIN_RET clobbers several registers (AX, CX, DX)
> so here it trashes user values which are in these registers.
> 
> Signed-off-by: Alexandre Chartre <alexandre.chartre@xxxxxxxxxx>

Acked-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>

> ---
>  arch/x86/entry/entry_64.S | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
> index 285e043a3e40..9953d966d124 100644
> --- a/arch/x86/entry/entry_64.S
> +++ b/arch/x86/entry/entry_64.S
> @@ -727,7 +727,6 @@ native_irq_return_ldt:
>  	pushq	%rdi				/* Stash user RDI */
>  	swapgs					/* to kernel GS */
>  	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi	/* to kernel CR3 */
> -	UNTRAIN_RET
>  
>  	movq	PER_CPU_VAR(espfix_waddr), %rdi
>  	movq	%rax, (0*8)(%rdi)		/* user RAX */
> -- 
> 2.31.1
>