Re: [PATCH v2 10/17] KVM: arm64: Prevent host PC adjustments for protected vCPUs

[Fuad Tabba]

Hi Joey,

[...]

>  For a non-protected vCPU, use PKVM_HOST_STATE_DIRTY to determine
>  whether to update the host vCPU or hyp vCPU.
[...]
> Feel free to not take the rewording, but it helped me understand a bit,
> and felt clearer.

Marc had a second look at this patch and it came out reworked: the
copy selection moved into __kvm_adjust_pc() behind a get/put pair, and
the pinning moved out into its own fix. The pinning paragraph you
reworded is gone from this patch, and the PKVM_HOST_STATE_DIRTY one is
reworded along your lines.

> Not sure here why we use hyp_vcpu->vcpu.kvm over host_vcpu->kvm, in
> non-protected pKVM do we still not trust it?

It was the pointer, not the VM: EL2 wasn't dereferencing a
host-written pointer without pinning what it points at. Moot now, the
host copy is adjusted with its own struct kvm, which the fix pins.

>  /* Loaded protected vCPU: do nothing. */

The else clause went with the rework. The drop is a NULL from the get
helper, with a comment there.

> Acked-by: Joey Gouly <joey.gouly@xxxxxxx>

Thanks! I've left it off v3 since the code you acked has changed under
you, so I'd appreciate a second look there.

Cheers,
/fuad