Re: [PATCH v4 00/12] target/arm mte fixes

[Peter Maydell]

On Thu, 8 Apr 2021 at 15:25, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> On 4/8/21 5:47 AM, Peter Maydell wrote:
> >> Richard Henderson (12):
> >>    accel/tcg: Preserve PAGE_ANON when changing page permissions
> >>    target/arm: Check PAGE_WRITE_ORG for MTE writeability
> >>    target/arm: Fix mte_checkN
> >>    target/arm: Split out mte_probe_int
> >>    target/arm: Fix unaligned checks for mte_check1, mte_probe1
> >>    test/tcg/aarch64: Add mte-5
> >>    target/arm: Replace MTEDESC ESIZE+TSIZE with SIZEM1
> >>    target/arm: Merge mte_check1, mte_checkN
> >>    target/arm: Rename mte_probe1 to mte_probe
> >>    target/arm: Simplify sve mte checking
> >>    target/arm: Remove log2_esize parameter to gen_mte_checkN
> >>    exec: Fix overlap of PAGE_ANON and PAGE_TARGET_1
> >
> > So, what should we do with this series for 6.0 ? We'll be at rc3
> > next week, and this big a change seems risky at this point
> > in the cycle. Is there anything in here that's worth extracting
> > as a for-6.0 change? (maybe patches 1, 2, 12?)
>
> Definitely 12, since that broke BTI.
>
> Patches 1 and 2 are certainly simple enough.
>
> Otherwise... the rest would be nice to have.  It's quite isolated to mte=on.
> If you defer, I guess that's fine too -- the bug report did come in quite 
> late.

OK, I'm going to take 1, 2 and 12 into target-arm.next for 6.0, and we'll
defer the rest. AIUI the bug fixed by the remaining patches is basically
"we didn't do some tag checks we should have done in a corner case of
misaligned accesses that span two differently tagged regions", and I think
that's not a terrible bug to leave in 6.0, compared to the risk of breaking
MTE more seriously while fixing it.

thanks
-- PMM