Re: [PATCH v4 29/39] arm64: mte: Switch GCR_EL1 in kernel entry and exit

[Catalin Marinas]

On Fri, Oct 09, 2020 at 10:56:02AM +0100, Vincenzo Frascino wrote:
> On 10/9/20 9:11 AM, Catalin Marinas wrote:
> > On Thu, Oct 08, 2020 at 07:24:12PM +0100, Vincenzo Frascino wrote:
> >> On 10/2/20 3:06 PM, Catalin Marinas wrote:
> >>> On Fri, Oct 02, 2020 at 01:10:30AM +0200, Andrey Konovalov wrote:
> >>>> diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
> >>>> index 7c67ac6f08df..d1847f29f59b 100644
> >>>> --- a/arch/arm64/kernel/mte.c
> >>>> +++ b/arch/arm64/kernel/mte.c
> >>>> @@ -23,6 +23,8 @@
> >>>>  #include <asm/ptrace.h>
> >>>>  #include <asm/sysreg.h>
> >>>>  
> >>>> +u64 gcr_kernel_excl __ro_after_init;
> >>>> +
> >>>>  static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap)
> >>>>  {
> >>>>  	pte_t old_pte = READ_ONCE(*ptep);
> >>>> @@ -120,6 +122,13 @@ void *mte_set_mem_tag_range(void *addr, size_t size, u8 tag)
> >>>>  	return ptr;
> >>>>  }
> >>>>  
> >>>> +void mte_init_tags(u64 max_tag)
> >>>> +{
> >>>> +	u64 incl = GENMASK(max_tag & MTE_TAG_MAX, 0);
> >>>
> >>> Nitpick: it's not obvious that MTE_TAG_MAX is a mask, so better write
> >>> this as GENMASK(min(max_tag, MTE_TAG_MAX), 0).
> >>
> >> The two things do not seem equivalent because the format of the tags in KASAN is
> >> 0xFF and in MTE is 0xF, hence if extract the minimum whatever is the tag passed
> >> by KASAN it will always be MTE_TAG_MAX.
> >>
> >> To make it cleaner I propose: GENMASK(FIELD_GET(MTE_TAG_MAX, max_tag), 0);
> > 
> > I don't think that's any clearer since FIELD_GET still assumes that
> > MTE_TAG_MAX is a mask. I think it's better to add a comment on why this
> > is needed, as you explained above that the KASAN tags go to 0xff.
> > 
> > If you want to get rid of MTE_TAG_MAX altogether, just do a
> > 
> > 	max_tag &= (1 << MAX_TAG_SIZE) - 1;
> > 
> > before setting incl (a comment is still useful).
> > 
> 
> Agree, but still think we should use FIELD_GET here since it is common language
> in the kernel.
> 
> How about we get rid of MTE_TAG_MAX and we do something like:
> 
> GENMASK(FIELD_GET(MTE_TAG_MASK >> MTE_TAG_SHIFT, max_tag), 0);

It works for me and you can drop the MTE_TAG_MAX definition (I think
it's only used here).

-- 
Catalin