Re: [PATCH v2 10/11] KVM: arm64: nVHE: Check hypercall handlers against the declared ABI

On Mon, 03 Aug 2026 19:34:52 +0100,
Fuad Tabba <[email protected]> wrote:
> 
> On Mon, 3 Aug 2026 at 19:03, Marc Zyngier <[email protected]> wrote:
> ...
> >
> > I've also dropped this with the following hacks. The set_cpu_reg()
> > stuff isn't brilliant, and requires a single word type. But I find the
> > overall scheme less invasive.
> >
> > WDYT?
> 
> Works, and hyp-main.nvhe.o comes out byte identical. I did look for a
> way to dispatch on void without the type table, but _Generic doesn't
> take an incomplete type and __builtin_choose_expr type-checks both
> arms on gcc and clang, so I don't think there's a better option.
> 
> One nit: the declaration in kvm_hcall.h still says "unsigned long"
> while the definition now says "ulong". Could we make both ulong?

Yeah, that's an oversight. I don't like the ulong stuff at all, but I
don't see how to do this without it.

> 
> I'm happy to respin with both of these, or would you like to fold them in?

Oliver is in charge of 7.3, so I'll let him reply on this.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.