From: Hans-Jörg Höxer <hshoexer@genua.de>
Subject: Re: [EXT] Re: SEV-ES guest: early GHCB allocation and early #VC trap handling 1/3
To: <tech@openbsd.org>
Date: Tue, 1 Jul 2025 10:27:11 +0200

Hi Mike,

thanks again for your review and feedback!  Answers inline below.

> ...
> > @@ -297,6 +301,52 @@ kpageflttrap(struct trapframe *frame, uint64_t cr2)
> >  	return 1;
> >  }
> >
> > +int
> > +vctrap(struct trapframe *frame)
> > +{
> > +	uint64_t	 sw_exitcode, sw_exitinfo1, sw_exitinfo2;
> > +	struct ghcb_sync syncout, syncin;
> > +	struct ghcb_sa	*ghcb;
> > +
> > +	intr_disable();
> > +
> > +	memset(&syncout, 0, sizeof(syncout));
> > +	memset(&syncin, 0, sizeof(syncin));
> > +
> > +	sw_exitcode = frame->tf_err;
> > +	sw_exitinfo1 = 0;
> > +	sw_exitinfo2 = 0;
> > +
> > +	switch (sw_exitcode) {
> > +	default:
> > +		panic("invalid exit code 0x%llx", sw_exitcode);
> > +	}
> 
> I was scratching my head on this until tb@ pointed out that this is filled out
> in a later diff. What are your thoughts around combining the diffs together
> so that if we need to revert diff #3 later for some reason, we aren't left with
> a halfway working vctrap()? Eg, "all or nothing"? I'll defer to you and bluhm@.

good point, sorry for the confusion.

> ...
> > diff --git a/sys/arch/amd64/include/cpufunc.h b/sys/arch/amd64/include/cpufunc.h
> > index e4c8d6924d2..9f01b3cb989 100644
> > --- a/sys/arch/amd64/include/cpufunc.h
> > +++ b/sys/arch/amd64/include/cpufunc.h
> > @@ -439,6 +439,13 @@ breakpoint(void)
> >  	__asm volatile("int $3");
> >  }
> >
> > +/* VMGEXIT */
> > +static __inline void
> > +vmgexit(void)
> > +{
> > +	__asm volatile("rep; vmmcall");
> 
> 
> A comment here about the conversation we had the other day would be useful for
> posterity:
> 
> 	/* rep; vmmcall encodes the vmgexit instruction */

ack! bluhm@ already added that comment, thanks!

> ...