Re: [Qemu-devel] [PATCH 1/8] target-arm: Properly support EL2 and EL3 in arm_el_is_aa64()

["Edgar E. Iglesias" <edgar.iglesias@gmail.com>]

On Fri, Jan 15, 2016 at 03:47:17PM +0000, Peter Maydell wrote:
> On 15 January 2016 at 15:37, Edgar E. Iglesias <edgar.iglesias@gmail.com> wrote:
> > On Fri, Jan 15, 2016 at 02:50:24PM +0000, Peter Maydell wrote:
> >> Do you have much locally extra that you needed for enabling
> >> EL3 in the Cortex-A53? I have an ARM Trusted Firmware + OP-TEE
> >> setup now that I'm going to use to work through the missing bits,
> >> but if you've already gone through that effort there's no need
> >> my duplicating work...
> >
> > I don't have anything immediate for EL3 beyond enabling it and some
> > boot thing for a15/aarch32 to allow me to run my tests.
> 
> Cool. I'm not sure at what point to add the patch that turns
> on the EL3 feature bit, but I guess in the not too distant future :-)

IMO, we should have enabled it by now :-)
I guess the EL2 needs the 2 stage MMU error handling and maybe the GIC virt features
to be reasonably useful, I don't know.

> 
> > I haven't
> > really looked at the boot in detail for aa32 so I haven't bothered
> > submitting it. This is it:
> >
> > commit b30c7102624241a67ebb2d3df70e88a4148f68a4
> > Author: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
> > Date:   Sun Sep 13 09:52:01 2015 +0200
> >
> >     target-arm: Start EL3 capable ARMv7 cores in MON mode
> >
> >     Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
> >
> > diff --git a/target-arm/cpu.c b/target-arm/cpu.c
> > index f6f5539..485965f 100644
> > --- a/target-arm/cpu.c
> > +++ b/target-arm/cpu.c
> > @@ -164,6 +164,9 @@ static void arm_cpu_reset(CPUState *s)
> >  #else
> >      /* SVC mode with interrupts disabled.  */
> >      env->uncached_cpsr = ARM_CPU_MODE_SVC;
> > +    if (arm_feature(env, ARM_FEATURE_EL3)) {
> > +        env->uncached_cpsr = ARM_CPU_MODE_MON;
> > +    }
> >      env->daif = PSTATE_D | PSTATE_A | PSTATE_I | PSTATE_F;
> >      /* On ARMv7-M the CPSR_I is the value of the PRIMASK register, and is
> >       * clear at reset. Initial SP and PC are loaded from ROM.
> 
> This doesn't look right. A 32-bit CPU with TrustZone boots into
> Secure-SVC, not Mon. This works because in the v7 security model
> Secure-SVC and Mon are at the same privilege level, unlike AArch64
> where EL3 is higher privilege than EL1. If guest code needs to
> get into Mon mode it can do so from S-SVC (eg set MVBAR, make
> sure there's sensible code at that vector entrypoint, execute
> an SMC).

Thanks. I figured my zero research would prove me wrong :-)
Thinking about it, I have never ran my aa32 TZ testsuite on real
HW so there are tons of stuff that could be wrong!

Best regards,
Edgar