On Wed, Sep 02, 2026 at 10:18:00AM -0400, Dave Voutila wrote:
> Both VMX and SVM currently encode an injected #DB as a software
> exception.
>
> Using the software-exception type also gives the hardware the wrong
> instruction-completion semantics while delivering the event. Move #DB
> next to #UD in both injection paths, where it is encoded as a hardware
> exception without an error code.
>
> ok?
>

hm this is another one I thought we (me? hshoexer?) already fixed but if it
never landed please go ahead. I know we also discussed this one before,
a few months ago.

ok mlarkin

> ---
> diff e7450dffd6c02a765d93e405c0a5751cf3f2b9f3 c597b118780d9e1062c05ddde4ba76f673c3fede
> commit - e7450dffd6c02a765d93e405c0a5751cf3f2b9f3
> commit + c597b118780d9e1062c05ddde4ba76f673c3fede
> blob - 48b8ac55514e79d373b80601e6ca04f236613c93
> blob + 82b2ec36d9d1149814d736fdf6cc9b5bbb3380a5
> --- sys/arch/amd64/amd64/vmm_machdep.c
> +++ sys/arch/amd64/amd64/vmm_machdep.c
> @@ -3879,6 +3879,7 @@ vcpu_run_vmx(struct vcpu *vcpu, struct vm_run_params *
>  				/* Software Exceptions */
>  				eii |= (4ULL << 8);
>  				break;
> +			case VMM_EX_DB:
>  			case VMM_EX_UD:
>  				/* Hardware exception, no error code. */
>  				eii |= (3ULL << 8);
> @@ -6781,13 +6782,13 @@ vcpu_run_svm(struct vcpu *vcpu, struct vm_run_params *
>  			switch (vcpu->vc_inject.vie_vector) {
>  			case VMM_EX_BP:
>  			case VMM_EX_OF:
> -			case VMM_EX_DB:
>  				/*
>  				 * Software exception.
>  				 * XXX check nRIP support.
>  				 */
>  				vmcb->v_eventinj |= (4ULL << 8);
>  				break;
> +			case VMM_EX_DB:
>  			case VMM_EX_UD:
>  				/* Hardware exception, no error code. */
>  				vmcb->v_eventinj |= (3ULL << 8);