Re: [PATCH] target/arm: Fix Rt/Rt2 in ESR_ELx for copro traps from AArch32 to 64

[Peter Maydell]

On Tue, 4 Aug 2020 at 21:38, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> On 8/3/20 9:54 AM, Peter Maydell wrote:
> > +    case 14:
> > +        switch (mode) {
> > +        case ARM_CPU_MODE_USR:
> > +        case ARM_CPU_MODE_SYS:
> > +            return 14;
> > +        case ARM_CPU_MODE_HYP:
> > +            return 16;
>
> Hyp uses LR_usr...
>
> > +        case ARM_CPU_MODE_IRQ:
> > +            return 18;
> > +        case ARM_CPU_MODE_SVC:
> > +            return 20;
> > +        case ARM_CPU_MODE_ABT:
> > +            return 22;
> > +        case ARM_CPU_MODE_UND:
> > +            return 24;
>
> ... making all of these off-by-2.

Yeah, this is the bug fixed in v2.

> > +        case ARM_CPU_MODE_FIQ:
> > +            return 30;
> > +        default:
> > +            g_assert_not_reached();
> > +        }
> > +    case 15:
> > +        return 31;
>
> I don't see that R15 is mapped at all.  Is this really reachable?

It is -- you can get Rt=0b1111 for MRC to APSR_nzcv (as well
as for various UNPREDICTABLE MCR etc cases that will get trapped
here that we have to do something sensible for). Rt=0b1111
is mapped to 0b11111 in the pseudocode (see
AArch64.AArch32SystemAccessTrapSyndrome()); it's just not
documented in the text (yet) that this is the required behaviour.

thanks
-- PMM