On Thu, 7 Jan 2021 at 19:10, Richard Henderson
<richard.hender...@linaro.org> wrote:
>
> On 1/7/21 7:54 AM, Peter Maydell wrote:
> >> - && (sctlr & SCTLR_TCF0)
> >> + && (sctlr & SCTLR_TCF)
> >> && allocation_tag_access_enabled(env, 0, sctlr)) {
> >> flags = FIELD_DP32(flags, TBFLAG_A64, MTE0_ACTIVE, 1);
> >> }
> >
> >
> > I don't understand this change, could you explain a bit more?
> > In commit 50244cc76abcac we change to looking at the TCF
> > field corresponding to the actual current EL instead of the
> > EL for the memory-access.
>
> Correct.
>
> > But if we're doing that then why
> > should we be looking at exclusively SCTLR_TCF0 in this
> > for-unpriv-access code rather than doing the same thing we do
> > for normal accesses and checking
> > (sctlr & (el == 0 ? SCTLR_TCF0 : SCTLR_TCF))
>
> Because this is for the UNPRIV instructions which are UNDEF at el == 0.
Ah, right. (It didn't help that I'd read the diff backwards:
the new code looks at SCTLR_TCF, not SCTLR_TCF0.)
Further, the SCTLR_*.ATA/ATA0 checks *are* based on the
privilege of the access, which is why calling
allocation_tag_access_enabled(env, 0, sctlr)
is still correct.
Applied to target-arm.next, thanks.
-- PMM
