[PATCH kvmtool] aarch64: Give up with MTE for AArch32 guest

[Alexandru Elisei]

Hi,

On Fri, May 20, 2022 at 02:48:10PM +0100, Vladimir Murzin wrote:
> Hi Alexandru,
> 
> On 5/20/22 14:31, Alexandru Elisei wrote:
> > Hi Vladimir,
> > 
> > When I run an --aarch32 guest with --debug this is the message that I'm
> > getting:
> > 
> >   Info: (arm/aarch64/kvm.c) kvm__arch_enable_mte:146: MTE capability not available
> > 
> > Would you mind elaborating on the merits of the message that you are
> > proposing:
> > 
> >   Info: (arm/aarch64/kvm.c) kvm__arch_enable_mte:124: MTE is incompatible with AArch32
> > 
> > Is it because it explains why the capability is not available?
> 
> Hmm, without this patch I'm getting
> 
>  lkvm run -k /opt/kvm/guest-0/Image -d /opt/kvm/guest-0/fs.ext2 -c 4 -m 681 --console virtio '--irqchip=gicv3' --params 'console=hvc earlycon swiotlb=1024' --aarch32 --debug
> 
> + lkvm run -k /opt/kvm/guest-0/Image -m 681 -c 4 --name guest-115
>   Info: (arm/aarch64/kvm.c) kvm__get_vm_type:76: max_ipa aa8fffff ipa_bits 32 max_ipa_bits 40
>   Info: (arm/aarch64/kvm.c) kvm__arch_enable_mte:104: MTE capability enabled
>   Info: (arm/kvm.c) kvm__arch_load_kernel_image:118: Loaded kernel to 0x80008000 (17280980 bytes)
>   Info: (arm/kvm.c) kvm__arch_load_kernel_image:134: Placing fdt at 0x8fe00000 - 0x8fffffff
>   Fatal: Unable to initialise vcpu
> 
> with patch applied
> 
> lkvm run -k /opt/kvm/guest-0/Image -d /opt/kvm/guest-0/fs.ext2 -c 4 -m 681 --console virtio '--irqchip=gicv3' --params 'console=hvc earlycon swiotlb=1024' --aarch32 --debug
> + lkvm run -k /opt/kvm/guest-0/Image -d /opt/kvm/guest-0/fs.ext2 -c 4 -m 681 --console virtio '--irqchip=gicv3' --params 'console=hvc earlycon swiotlb=1024' --aarch32 --debug
>   # lkvm run -k /opt/kvm/guest-0/Image -m 681 -c 4 --name guest-114
>   Info: (arm/aarch64/kvm.c) kvm__get_vm_type:76: max_ipa aa8fffff ipa_bits 32 max_ipa_bits 40
>   Info: (arm/aarch64/kvm.c) kvm__arch_enable_mte:92: MTE is incompatible with AArch32
>   Info: (arm/kvm.c) kvm__arch_load_kernel_image:118: Loaded kernel to 0x80008000 (17280980 bytes)
>   Info: (arm/kvm.c) kvm__arch_load_kernel_image:134: Placing fdt at 0x8fe00000 - 0x8fffffff
>   Info: (virtio/mmio.c) virtio_mmio_init:325: virtio-mmio.devices=0x200 at 0x3000000:36
>   Info: (virtio/mmio.c) virtio_mmio_init:325: virtio-mmio.devices=0x200 at 0x3000200:37
>   Info: (virtio/mmio.c) virtio_mmio_init:325: virtio-mmio.devices=0x200 at 0x3000400:38
> 
> My host and guest kernel is 5.15, maybe I'm just missing some backport?

I'm embarassed now, I didn't check that MTE was enabled on the model.
Without this patch I am indeed seeing the error message:

  Fatal: Unable to initialise vcpu

which goes away after I apply your patch.

> 
> Cheers
> Vladimir
> 
> > 
> > Thanks,
> > Alex
> > 
> > On Fri, May 20, 2022 at 01:38:44PM +0100, Vladimir Murzin wrote:
> >> KVM doesn't support combination of MTE and AArch32 guest, so do not
> >> even try.
> >>
> >> Signed-off-by: Vladimir Murzin <vladimir.murzin at arm.com>
> >> ---
> >>  arm/aarch64/kvm.c | 5 +++++
> >>  1 file changed, 5 insertions(+)
> >>
> >> diff --git a/arm/aarch64/kvm.c b/arm/aarch64/kvm.c
> >> index 1b992dd..f3fe854 100644
> >> --- a/arm/aarch64/kvm.c
> >> +++ b/arm/aarch64/kvm.c
> >> @@ -120,6 +120,11 @@ void kvm__arch_enable_mte(struct kvm *kvm)
> >>  		.cap = KVM_CAP_ARM_MTE,
> >>  	};
> >>  
> >> +	if (kvm->cfg.arch.aarch32_guest) {
> >> +		pr_debug("MTE is incompatible with AArch32");

Nitpick: I think "MTE disabled because it is incompatible with AArch32" is
slightly better, as it better matches the disabled by user message below.
It's up to you if you want to change the patch, I don't have a preference
either way.

> >> +		return;
> >> +	}
> >> +

I checked and this also matches the documentation for KVM_CAP_ARM_MTE:

Tested-by: Alexandru Elisei <alexandru.elisei at arm.com>
Reviewed-by: Alexandru Elisei <alexandru.elisei at arm.com>

Thanks,
Alex

> >>  	if (kvm->cfg.arch.mte_disabled) {
> >>  		pr_debug("MTE disabled by user");
> >>  		return;
> >> -- 
> >> 2.25.1
> >>
>