Re: [kvm-unit-tests PATCH] arm64: mte: Fix MTE granule mask

[Alexandru Elisei <alexandru.elisei@arm.com>]

Hi Vladimir,

On Fri, Sep 05, 2025 at 01:41:29PM +0100, Vladimir Murzin wrote:
> So it generates correct mask when used together with MTE_TAG_SHIFT
> 
> Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
> ---
>  arm/mte.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arm/mte.c b/arm/mte.c
> index 95d58aaa..a1bed8a7 100644
> --- a/arm/mte.c
> +++ b/arm/mte.c
> @@ -26,7 +26,7 @@
>  #define MTE_TCF_ASYMM	0b11
>  
>  #define MTE_GRANULE_SIZE        UL(16)
> -#define MTE_GRANULE_MASK        (~(MTE_GRANULE_SIZE - 1))
> +#define MTE_GRANULE_MASK        (MTE_GRANULE_SIZE - 1)

And this is how it's used:

#define untagged(p)									\
({											\
	unsigned long __in = (unsigned long)(p);					\
	typeof(p) __out = (typeof(p))(__in & ~(MTE_GRANULE_MASK << MTE_TAG_SHIFT));	\

I'm going to walk through how MTE_GRANULE_MASK is used, just to make sure I got
things right this time.

Without this change:

MTE_GRANULE_MASK = ~(0..01..1) = 1..10..0 -> LSB 4 bits are zero
~(MTE_GRANULE_MASK << MTE_TAG_SHIFT) = ~(1..10..0) = 0..01..1 -> bits 63:60 are
zero, and what is masked is actually the PAC field, not the logical tag.

With this patch:

MTE_GRANULE_MASK = 0..01..1 -> LSB 4 bits are 1
~(MTE_GRANULE_MASK << MTE_TAG_SHIFT) = ~(0..01..10..0) = 1..10..01..1 -> bits
59:56 are 0, and the logical tag is masked, which is what we want, so:

Reviewed-by: Alexandru Elisei <alexandru.elisei@arm.com>

Also curious, I remember running the tests and I didn't see an error, how did
you catch this?

Thanks,
Alex

>  #define MTE_TAG_SHIFT           56
>  
>  #define untagged(p)									\
> -- 
> 2.34.1
>