From: Mike Larkin <mlarkin@nested.page>
Subject: Re: SEV-ES guest: In vctrap() KASSERT on interrupt enable flag being clear
To: tech@openbsd.org
Date: Wed, 2 Jul 2025 11:43:08 -0700

On Tue, Jul 01, 2025 at 10:32:37AM +0200, Hans-Jörg Höxer wrote:
> Hi,
>
> as discussed earlier:
>
> o KASSERT on PSL_I.  Interrupts must be off when entering vctrap().
> o explain the general workings of #VC trap handling.
>
> Take care,
> HJ.
>

ok mlarkin

> ---------
> commit d10b95bc7f4150140c4de2dcd2f0edd657fda033
> Author: Hans-Joerg Hoexer <hshoexer@genua.de>
> Date:   Mon Jun 30 13:14:35 2025 +0200
>
>     SEV-ES guest: In vctrap() KASSERT on interrupt enable flag being clear
>
>     While there, add a comment about the general workings of #VC handling.
>     Prompted and provided by mlarkin@.
>
> diff --git a/sys/arch/amd64/amd64/trap.c b/sys/arch/amd64/amd64/trap.c
> index 61a3e92f297..f68cb2c90d5 100644
> --- a/sys/arch/amd64/amd64/trap.c
> +++ b/sys/arch/amd64/amd64/trap.c
> @@ -310,7 +310,7 @@ vctrap(struct trapframe *frame)
>  	struct ghcb_sync syncout, syncin;
>  	struct ghcb_sa	*ghcb;
>
> -	intr_disable();
> +	KASSERT((read_rflags() & PSL_I) == 0);
>
>  	memset(&syncout, 0, sizeof(syncout));
>  	memset(&syncin, 0, sizeof(syncin));
> @@ -319,6 +319,16 @@ vctrap(struct trapframe *frame)
>  	sw_exitinfo1 = 0;
>  	sw_exitinfo2 = 0;
>
> +	/*
> +	 * The #VC trap occurs when the guest (us) performs an
> +	 * operation which requires sharing data with the host. In
> +	 * order to ascertain which instruction caused the #VC,
> +	 * examine the instruction by reading %rip, Then, sync the
> +	 * appropriate values out (to the host), perform VMGEXIT
> +	 * to request that the host handle the operation which
> +	 * caused the #VC, then sync the returned values back in
> +	 * (from the host).
> +	 */
>  	switch (sw_exitcode) {
>  	case SVM_VMEXIT_CPUID:
>  		ghcb_sync_val(GHCB_RAX, GHCB_SZ32, &syncout);