On Fri, Nov 21, 2025 at 03:45:06PM +0100, hshoexer wrote:
> Hi,
>
> right now, vmm(4) only injects #UD correctly, when running on
> intel/VMX.  Running on amd/SVM we hit the default case.
>
> #UD does not provide an error code, so bit 11 of EVENTINJ must be
> clear.
>
> ok?
>
> --------------------------------------------------------------
> diff --git a/sys/arch/amd64/amd64/vmm_machdep.c b/sys/arch/amd64/amd64/vmm_machdep.c
> index 9e63a0415bb..a85e1908c72 100644
> --- a/sys/arch/amd64/amd64/vmm_machdep.c
> +++ b/sys/arch/amd64/amd64/vmm_machdep.c
> @@ -6739,6 +6739,10 @@ vcpu_run_svm(struct vcpu *vcpu, struct vm_run_params *vrp)
>  				 */
>  				vmcb->v_eventinj |= (4ULL << 8);
>  				break;
> +			case VMM_EX_UD:
> +				/* Hardware exception, no error code. */
> +				vmcb->v_eventinj |= (3ULL << 8);
> +				break;
>  			case VMM_EX_AC:
>  				vcpu->vc_inject.vie_errorcode = 0;
>  				/* fallthrough */
>

I have a better diff for this that I made at the hackathon. I'll send it
tomorrow when I'm back on that machine. There are/were errors also in the
#UD handling on VMX.