Re: [PATCH v3 2/4] prtcl: introduce PR_MTE_STORE_ONLY

[Peter Collingbourne]

On Fri, May 2, 2025 at 10:37 AM Catalin Marinas <catalin.marinas@xxxxxxx> wrote:
>
> On Thu, Apr 24, 2025 at 10:34:57PM +0200, David Hildenbrand wrote:
> > On 10.04.25 10:07, Yeoreum Yun wrote:
> > > PR_MTE_STORE_ONLY is used to restrict the MTE tag check for store
> > > opeartion only.
> > >
> > > Signed-off-by: Yeoreum Yun <yeoreum.yun@xxxxxxx>
> > > ---
> > >   include/uapi/linux/prctl.h | 2 ++
> > >   1 file changed, 2 insertions(+)
> > >
> > > diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> > > index 15c18ef4eb11..83ac566251d8 100644
> > > --- a/include/uapi/linux/prctl.h
> > > +++ b/include/uapi/linux/prctl.h
> > > @@ -244,6 +244,8 @@ struct prctl_mm_map {
> > >   # define PR_MTE_TAG_MASK          (0xffffUL << PR_MTE_TAG_SHIFT)
> > >   /* Unused; kept only for source compatibility */
> > >   # define PR_MTE_TCF_SHIFT         1
> > > +/* MTE tag check store only */
> > > +# define PR_MTE_STORE_ONLY         (1UL << 19)
> >
> > That is the next available bit after PR_MTE_TAG_MASK, correct?
> >
> > Would we want to leave some space to grow PR_MTE_TAG_MASK in the future
> > (could that happen?)?
>
> The current mask covers 16 tags (bits 59:56 of a pointer) and given the
> reluctance to have a tag storage of 4 bits per 16 bytes (3% of RAM), I
> doubt we'd ever grow this.
>
> However, you have a good point, we could indeed leave 32 bits for the
> tag mask, just in case MTE gets so much traction that someone wants 8
> bits per tag (and likely a bigger granule than 16 bytes). It doesn't
> cost us anything to add additional bits from (PR_MTE_TAG_SHIFT + 32).

If it's 8 bits per tag wouldn't the exclusion mask need to be 256
bits? I probably wouldn't try to anticipate this case since it would
likely require a different API anyway.

Peter