boot-qemu.sh: Support using KVM on arm64 hardware

[nathanchance]

No description provided.

[nickdesaulniers]

Is there anything we can do to make the host check more consistent with the x86 check?

[nathanchance]

Is there anything we can do to make the host check more consistent with the x86 check?

What kind of consistency were you looking for? The logic for this check was copied from the cpu-checker package on Debian and Ubuntu, which contains kvm-ok. Apparently all AArch64 hardware is KVM compatible, all the script does is check if the KVM module is loaded (i.e. /dev/kvm exists).

I can probably refine the second half of the x86 check to just be "does /dev/kvm exist" but I would need to test it later.

[nickdesaulniers]

why do we have uname guards on one and not both? Is there something in cpuinfo that we should check on aarch64?

I would expect both or neither to have cpuinfo checks.

I would expect both or neither to have uname checks.

Not different approaches per target.

[nathanchance]

Not all x86 CPUs have hardware virtualization; we have to check /proc/cpuinfo for the feature flags (vmx for Intel, svm for AMD). We do not need a uname check for that as non-x86 processors will not have those flags. All AArch64 CPUs have hardware virtualization so we just need to check if we are running on an AArch64 processor to determine if we have hardware virtualization; I do not see any flags in /proc/cpuinfo that we can use to check.

[nickdesaulniers]

Will /dev/kvm exist on x86 cpus that don't support vmx|svm? If not, then it seems better to do a uname -m check for x86 hosts, rather than check cpuinfo. What if svm appears in an aarch64 host's /dev/cpuinfo trying to boot x86?

[nathanchance]

Will /dev/kvm exist on x86 cpus that don't support vmx|svm? If not, then it seems better to do a uname -m check for x86 hosts, rather than check cpuinfo. What if svm appears in an aarch64 host's /dev/cpuinfo trying to boot x86?

Ah, I see what you are saying. I do not think that /dev/kvm can exist if the CPU does not have hardware virtualization. I highly doubt that svm would appear on an AArch64 device but who knows? I still think the check is consistent with how it is in kvm-ok but I am fine with changing it (although I guess I will need to check for either x86_64 or i386 through i686 right?)