Re: [PATCH v3 2/4] tests/acceptance: boot_linux_console: Add boot Linux with kvm tests

[Andrew Jones]

On Thu, Jan 23, 2020 at 07:47:19PM -0200, Wainer dos Santos Moschetta wrote:
> 
> On 1/22/20 7:02 AM, Andrew Jones wrote:
> > On Tue, Jan 21, 2020 at 10:27:51PM -0300, Wainer dos Santos Moschetta wrote:
> > > +    def test_aarch64_virt_kvm(self):
> > > +        """
> > > +        :avocado: tags=arch:aarch64
> > > +        :avocado: tags=machine:virt
> > > +        :avocado: tags=accel:kvm
> > > +        """
> > > +        self.do_test_aarch64_virt()
> > > +
> > > +    def test_aarch64_virt_tcg(self):
> > > +        """
> > > +        :avocado: tags=arch:aarch64
> > > +        :avocado: tags=machine:virt
> > > +        :avocado: tags=accel:tcg
> > > +        """
> > > +        self.do_test_aarch64_virt()
> > > +
> > Does do_test_aarch64_virt() add more machine parameters? Also, which cpu
> > type does it choose? The reason I ask is because aarch64 virt will fail to
> > run with KVM unless the appropriate gic version is specified (the
> > gic-version machine parameter). Also the cpu type must be 'host' or 'max'.
> > 'max' is the better choice as it also works for tcg. gic-version also
> > takes 'max' allowing it to auto-select the appropriate version. So if it's
> > not already there somewhere, then please ensure aarch64 has these
> > additional parameters:
> > 
> >   machine:gic-version=max
> >   cpu:max
> 
> 
> The test was passing '-cpu cortex-a56', I replaced with '-cpu max'. Also,
> now, it passes the gic version as you pointed out. I will send those changes
> on a v4.
> 
> Other than that, I tried '-cpu max -machine virt' (without gic-version) and
> QEMU crashed:
> 
> [root@virtlab-arm03 build]# ./aarch64-softmmu/qemu-system-aarch64 -accel kvm
> -cpu max -machine virt -display none -vga none
> qemu-system-aarch64: PMU: KVM_SET_DEVICE_ATTR: Invalid argument
> qemu-system-aarch64: failed to set irq for PMU
> Aborted (core dumped)
> 
> ----
> 
> Should I expect it to crash or rather nicely fail?

crash, unfortunately. I recall we once had a plan to send patches to
fail nicer, and to output a hint on how to resolve the issue, but I
guess that never happened...

Thanks,
drew