Re: [PATCH] vmd: handle the reset control register at 0xcf9

On Sun, Aug 23, 2026 at 12:28:28AM +0000, ssnf wrote:
> Handle guest writes to the x86 reset control register and terminate
> the VM on a reset request.
>
> Tested on a 9front guest with fshalt -r.
> ---
>  usr.sbin/vmd/x86_vm.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/usr.sbin/vmd/x86_vm.c b/usr.sbin/vmd/x86_vm.c
> index 1b5ade90f9b..a0197c34b4c 100644
> --- a/usr.sbin/vmd/x86_vm.c
> +++ b/usr.sbin/vmd/x86_vm.c
> @@ -507,6 +507,7 @@ vcpu_exit_inout(struct vm_run_params *vrp)
>  int
>  vcpu_exit(struct vm_run_params *vrp)
>  {
> +	struct vm_exit *vei = vrp->vrp_exit;
>  	int ret;
>
>  	switch (vrp->vrp_exit_reason) {
> @@ -533,6 +534,10 @@ vcpu_exit(struct vm_run_params *vrp)
>  		break;
>  	case VMX_EXIT_IO:
>  	case SVM_VMEXIT_IOIO:
> +		if (vei->vei.vei_dir == VEI_DIR_OUT &&
> +		    vei->vei.vei_port == 0xcf9 &&
> +		    (vei->vei.vei_data & 0x06) == 0x06)
> +			return (EAGAIN);
>  		vcpu_exit_inout(vrp);
>  		break;
>  	case VMX_EXIT_HLT:
> --
> 2.51.0
>

I am not opposed to the general idea of supporting this register, but I would
ask for the following changes:

1. move the checks above into vcpu_exit_inout ; that function is where in/out
   instruction emulation is done, not directly in the vcpu_exit handler. make
   vcpu_exit_inout return a value, and you can return EAGAIN there and pass
   it up the chain like you are doing above.

2. use #defines for 0xcf9 and whatever 0x06 means

Happy to take a look again if those changes are made. Thx!