Re: i386 9.99.17 build fails for NET4501 kernel

[Maxime Villard <max%m00nbsd.net@localhost>]

Le 17/10/2019 à 14:07, Joerg Sonnenberger a écrit :
> On Thu, Oct 17, 2019 at 12:43:48PM +0200, Rhialto wrote:
> > On Thu 17 Oct 2019 at 11:07:44 +0200, Maxime Villard wrote:
> > > I guess we whould disable the messages on this particular file for now as
> > > the others have suggested, and file a PR in LLVM/GCC, because apart from
> > > a compiler bug I don't see what it could be.
> >
> > The original error has no place in a freestanding build anyway, since
> > vfork (and longjmp probably too) are concepts of hosted implementations
> > only.
>
> Yes and no. GCC has a hard-coded list of functions that always get the
> returns_twice attribute. (v)fork is on that list, but also savectx. So
> the fix is not to disable the warning, but rename the function.
>
> Joerg

It appears that the savectx in this function is unused.

It copies %rsp and %rbp in pcb1, but nothing can be done with that. pcb1 gets
copied into pcb2, but pcb2->pcb_{rsp,rbp} get overwritten at the end of the
function.

So the savectx block can be dropped.