Re: [PATCH 2/3] x86/mm: Fix LAM inconsistency during context switch

[Kirill A. Shutemov]

On Tue, Mar 12, 2024 at 03:59:50AM +0000, Yosry Ahmed wrote:
> diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
> index 76e91fc68c5f3..748d2b3bdb985 100644
> --- a/arch/x86/kernel/process_64.c
> +++ b/arch/x86/kernel/process_64.c
> @@ -753,10 +753,12 @@ static long prctl_map_vdso(const struct vdso_image *image, unsigned long addr)
>  static void enable_lam_func(void *__mm)
>  {
>  	struct mm_struct *mm = __mm;
> +	unsigned long lam = mm_lam_cr3_mask(mm);
> +	u64 untag_mask = mm_untag_mask(mm);
>  

Maybe push these mm dereferences inside the if block below?
I am not sure if compiler can re-order operations past this_cpu_read().

>  	if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm) {
> -		write_cr3(__read_cr3() | mm->context.lam_cr3_mask);
> -		set_tlbstate_lam_mode(mm);
> +		write_cr3(__read_cr3() | lam);
> +		cpu_tlbstate_update_lam(lam, untag_mask);
>  	}
>  }
>  

-- 
  Kiryl Shutsemau / Kirill A. Shutemov