[4.22 BLOCKER] Re: [PATCH v4 5/6] xen/mem_access: wrap memory access when VM_EVENT=n

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 15/01/2026 9:28 am, Penny Zheng wrote:
> diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
> index 07e54890d9..b34cd29629 100644
> --- a/xen/arch/x86/hvm/hvm.c
> +++ b/xen/arch/x86/hvm/hvm.c
> @@ -4804,6 +4810,12 @@ static int do_altp2m_op(
>          break;
>  
>      case HVMOP_altp2m_set_mem_access:
> +        if ( !vm_event_is_enabled(current) )
> +        {
> +            rc = -EOPNOTSUPP;
> +            break;
> +        }
> +
>          if ( a.u.mem_access.pad )
>              rc = -EINVAL;
>          else
> @@ -4813,6 +4825,12 @@ static int do_altp2m_op(
>          break;
>  
>      case HVMOP_altp2m_set_mem_access_multi:
> +        if ( !vm_event_is_enabled(current) )
> +        {
> +            rc = -EOPNOTSUPP;
> +            break;
> +        }
> +
>          if ( a.u.set_mem_access_multi.pad ||
>               a.u.set_mem_access_multi.opaque > a.u.set_mem_access_multi.nr )
>          {
> @@ -4844,6 +4862,12 @@ static int do_altp2m_op(
>          break;
>  
>      case HVMOP_altp2m_get_mem_access:
> +        if ( !vm_event_is_enabled(current) )
> +        {
> +            rc = -EOPNOTSUPP;
> +            break;
> +        }
> +
>          if ( a.u.mem_access.pad )
>              rc = -EINVAL;
>          else

(Forwarding a report from Hady Azzam, CC'd).

This breaks introspection generally.  It is (mostly) erroneous to check
current like this; it's checking dom0's vm_event-ness, not the target of
the altp2m op.

P2M changes are also inherently domain-scope, not vcpu scope, further
calling into question the use of a vcpu-oriented check.

As far as I can tell, the misuse of vm_event_is_enabled() is just to
wrap the IS_ENABLED(CONFIG_VM_EVENT) condition.

~Andrew