Re: [PATCH v4.1] KVM, SEV: Add KVM_EXIT_SHUTDOWN metadata for SEV-ES

[Marc Zyngier]

On Fri, 08 Apr 2022 17:56:42 +0100,
Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
> 
> Queued, thanks.  But documentation was missing:
> 
> diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
> index e7a0dfdc0178..72183ae628f7 100644
> --- a/Documentation/virt/kvm/api.rst
> +++ b/Documentation/virt/kvm/api.rst
> @@ -6088,8 +6088,12 @@ should put the acknowledged interrupt vector into the 'epr' field.
>    #define KVM_SYSTEM_EVENT_SHUTDOWN       1
>    #define KVM_SYSTEM_EVENT_RESET          2
>    #define KVM_SYSTEM_EVENT_CRASH          3
> +  #define KVM_SYSTEM_EVENT_SEV_TERM       4
> +  #define KVM_SYSTEM_EVENT_NDATA_VALID    (1u << 31)
>  			__u32 type;
> +                        __u32 ndata;
>  			__u64 flags;
> +                        __u64 data[16];
>  		} system_event;
> 
>  If exit_reason is KVM_EXIT_SYSTEM_EVENT then the vcpu has triggered
> @@ -6099,7 +6103,7 @@ HVC instruction based PSCI call from the vcpu. The 'type' field describes
>  the system-level event type. The 'flags' field describes architecture
>  specific flags for the system-level event.
> 
> -Valid values for 'type' are:
> +Valid values for bits 30:0 of 'type' are:
> 
>   - KVM_SYSTEM_EVENT_SHUTDOWN -- the guest has requested a shutdown of the
>     VM. Userspace is not obliged to honour this, and if it does honour
> @@ -6112,12 +6116,18 @@ Valid values for 'type' are:
>     has requested a crash condition maintenance. Userspace can choose
>     to ignore the request, or to gather VM memory core dump and/or
>     reset/shutdown of the VM.
> + - KVM_SYSTEM_EVENT_SEV_TERM -- an AMD SEV guest requested termination.
> +   The guest physical address of the guest's GHCB is stored in `data[0]`.
> 
>  Valid flags are:
> 
>   - KVM_SYSTEM_EVENT_RESET_FLAG_PSCI_RESET2 (arm64 only) -- the guest issued
>     a SYSTEM_RESET2 call according to v1.1 of the PSCI specification.
> 
> +Extra data for this event is stored in the `data[]` array, up to index
> +`ndata-1` included, if bit 31 is set in `type`.  The data depends on the
> +`type` field.  There is no extra data if bit 31 is clear or `ndata` is zero.
> +

This has the potential to break userspace as it expects a strict match
on the whole of 'type', and does not expect to treat it as a bitfield.

Case in point, QEMU:

accel/kvm/kvm-all.c::kvm_cpu_exec()

        case KVM_EXIT_SYSTEM_EVENT:
            switch (run->system_event.type) {

CrosVM and kvmtool have similar constructs, and will break as soon as
KVM_SYSTEM_EVENT_NDATA_VALID is or'ed into 'type'.

	M.

-- 
Without deviation from the norm, progress is not possible.