[PATCH v2] arm64: mte: avoid clearing PSTATE.TCO on entry unless necessary

[Peter Collingbourne]

On Fri, Dec 10, 2021 at 4:06 AM Catalin Marinas <catalin.marinas at arm.com> wrote:
>
> On Wed, Nov 10, 2021 at 02:07:35PM -0800, Peter Collingbourne wrote:
> > diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
> > index 2f69ae43941d..a78ec15f5bbc 100644
> > --- a/arch/arm64/kernel/entry.S
> > +++ b/arch/arm64/kernel/entry.S
> > @@ -189,6 +189,27 @@ alternative_cb_end
> >  #endif
> >       .endm
> >
> > +     .macro mte_clear_tco, sctlr
> > +     /*
> > +      * Re-enable tag checking (TCO set on exception entry). This is only
> > +      * necessary if MTE is enabled in either the kernel or the userspace
> > +      * task in synchronous mode. With MTE disabled in the kernel and
> > +      * disabled or asynchronous in userspace, tag check faults (including in
> > +      * uaccesses) are not reported, therefore there is no need to re-enable
> > +      * checking. This is beneficial on microarchitectures where re-enabling
> > +      * TCO is expensive.
> > +      */
> > +#ifdef CONFIG_ARM64_MTE
> > +alternative_cb       kasan_hw_tags_enable
> > +     tbz     \sctlr, #SCTLR_EL1_TCF0_SHIFT, 1f
> > +alternative_cb_end
> > +alternative_if ARM64_MTE
> > +     SET_PSTATE_TCO(0)
> > +alternative_else_nop_endif
> > +1:
> > +#endif
> > +     .endm
>
> The patch looks fine to me but I recall in a private conversation with
> Mark he proposed the idea of moving this to entry-common.c (unless it
> was about something completely different). The downside is that we run
> with the TCO set for slightly longer. There shouldn't be a major
> drawback currently as we don't have stack tagging anyway.

Yes, Mark made that suggestion on the list. I tried it and found that
it led to a performance regression relative to baseline [1].

Peter

[1] https://lore.kernel.org/all/CAMn1gO51k1Dqts=THYq28nVMSvO6ZQB5sEG1wuzEVpAXBTfFjg@mail.gmail.com/