Re: [PATCH 1/2] UNREACHABLE() macro

[Jakub Jelinek <jakub@redhat.com>]

On Fri, Sep 11, 2009 at 10:52:28PM -0700, Andrew Morton wrote:
> On Fri, 11 Sep 2009 21:43:42 -0700 "H. Peter Anvin" <hpa@kernel.org> wrote:
> 
> > On 09/11/2009 08:49 PM, Arjan van de Ven wrote:
> > > On Fri, 11 Sep 2009 14:55:25 -0700 (PDT)
> > > Roland McGrath <roland@redhat.com> wrote:
> > > 
> > >>> That's a bit of a mouthful.  Did you consider a runtime probe with
> > >>> scripts/Kbuild.include's try-run, cc-option, etc?  
> > >>
> > >> I did not see any precedent in the sources for using those to test for
> > >> features by compiling particular test sources (i.e. in autoconf
> > > 
> > > look at the stackprotector flags.. they work this way already.
> > > It gets done once per kernel build...
> > > 
> > 
> > That works for flags, but not for the presence of builtin functions.
> > You can't even just try compiling something, since it will turn into an
> > ordinary function if not present... not obvious until link.
> > 
> 
> Use -Wall -Werror and if the compiler doesn't know about
> __builtin_unreachable() it will error out.

Yeah, e.g. checking exit status of:
echo '__attribute__((noreturn)) void foo (void) { asm volatile ("" : : : "memory"); __builtin_unreachable (); }' | gcc -S -O2 -W -Wall -Werror -xc - -o /dev/null > /dev/null 2>/dev/null
works.

	Jakub