Re: [kvm-unit-tests PATCH v2] arm64: Add basic MTE test

[Alexandru Elisei <alexandru.elisei@arm.com>]

Hi Drew,

On Mon, Dec 30, 2024 at 05:52:54PM +0100, Andrew Jones wrote:
> On Mon, Dec 30, 2024 at 04:28:30PM +0000, Alexandru Elisei wrote:
> > Hi Drew,
> > 
> > On Mon, Dec 30, 2024 at 04:45:01PM +0100, Andrew Jones wrote:
> > > On Mon, Dec 30, 2024 at 03:19:48PM +0000, Alexandru Elisei wrote:
> > > ...
> > > > Interestingly though, when I try to run a test (either with ./run_tests.sh
> > > > mte-sync, or by trying out your example), I get this error:
> > > > 
> > > > qemu-system-aarch64: MTE requested, but not supported by the guest CPU
> > > > 
> > > > My guess is that's caused by kvm-unit-tests defaulting to cortex-a57 as the
> > > > model CPU. Changing the cpu to 'max' makes the test run. How are you getting
> > > > around this on your end?
> > > >
> > > 
> > > We can (and I think should) change to 'max' for the default model.
> > 
> > I think so too, something like this maybe?
> > 
> > diff --git a/configure b/configure
> > index 86cf1da36467..4babbd5f5e74 100755
> > --- a/configure
> > +++ b/configure
> > @@ -286,7 +286,7 @@ fi
> >  [ -z "$processor" ] && processor="$arch"
> > 
> >  if [ "$processor" = "arm64" ]; then
> > -    processor="cortex-a57"
> > +    processor="max"
> >  elif [ "$processor" = "arm" ]; then
> >      processor="cortex-a15"
> >  fi
> > 
> > Can't do the same for arm, because Makefile.arm passes the processor to
> > -mcpu and that causes a build error.
> 
> I wonder if we can improve the mcpu thing by just selecting some generic
> cpu type at this point? Or just dropping it and counting on the compiler
> to use generic cpu type? It'd be nice to use max for both arm and arm64,

I tried setting the default to 'max' for both arm and arm64. If I pass
'max' as the cpu type to arm, I get the build error above. If I don't pass
-mcpu to the arm compiler when $PROCESSOR=max, I get different errors
because of unsupported instructions:

/tmp/ccOHIrg7.s: Assembler messages:
/tmp/ccOHIrg7.s:46: Error: selected processor does not support `yield' in ARM mode
/tmp/ccGCB0zP.s: Assembler messages:
/tmp/ccGCB0zP.s:55: Error: selected processor does not support `ldrex r2,[r4]' in ARM mode
/tmp/ccGCB0zP.s:59: Error: selected processor does not support `strex r3,r2,[r4]' in ARM mode
/tmp/ccGCB0zP.s:71: Error: selected processor does not support `dmb ish' in ARM mode
/tmp/ccGCB0zP.s:99: Error: selected processor does not support `dmb ish' in ARM mode
/tmp/ccGCB0zP.s:144: Error: selected processor does not support `dmb ish' in ARM mode
[..]

and the errors go on. This means that the default cannot be 'max' for arm.

I also realized that the configure --processor option for arm64:

   --processor=PROCESSOR  processor to compile for (aarch64)

does not work as advertised, because the $PROCESSOR variable is not passed
down to the compiler.

So what I'm proposing is this:

- Patch #1: pass $PROCESSOR to the arm64 compiler, to reconcile the help
  text for --processor and to achieve parity with arm.

- Patch #2: have arm64 ignore $PROCESSOR when compiling if it's 'max.  By
  ignore I mean don't add a -mcpu argument to the compiler invocation.
  Change configure to raise an error if --processor=max is used for arm
  (users might think that if it works for arm64, it also works for arm).

- Patch #3: change the default for arm64 to 'max', but don't touch the
  default for arm.

What do you think?

Thanks,
Alex

> but if nobody has time to do the arm work / testing to be sure we can,
> then I'm fine with just a patch changing arm64.
> 
> Thanks,
> drew