Re: [PATCH 2/3] objtool: Ignore retpoline alternatives

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Thu, Jan 11, 2018 at 04:39:38PM +0000, David Woodhouse wrote:
> On Thu, 2018-01-11 at 10:33 -0600, Josh Poimboeuf wrote:
> > On Thu, Jan 11, 2018 at 04:27:38PM +0000, David Woodhouse wrote:
> > > 
> > > On Wed, 2018-01-10 at 19:48 -0600, Josh Poimboeuf wrote:
> > > > 
> > > > 
> > > > +#define ANNOTATE_NOSPEC_ALTERNATIVE                            \
> > > > +       "999:\n\t"                                              \
> > > > +       ".pushsection .discard.nospec\n\t"                      \
> > > > +       ".long 999b - .\n\t"                                    \
> > > > +       ".popsection\n\t"
> > > > +
> > > 
> > >   Ick, numbers. Use .Lfoo_%= instead.
> > > 
> > I seem to recall that not working with inline asm, maybe old versions of
> > GCC don't like it or something?  I can try it and see if 0-day bot
> > complains.
> 
> You just need %= (for inline asm) instead of \@ (for .macro).
> 
> I already fixed it up in
> http://git.infradead.org/users/dwmw2/linux-retpoline.git/ and will get
> nagmail from 0day shortly if it doesn't work :)
> 
> (I love you Fengguang)

I found a description from an old commit of mine:

  3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")

> A workaround for this issue is to ensure that each instance of the
> inline asm statement uses a different label, so that GCC sees the
> statements are unique and leaves them alone.  The inline asm ‘%=’ token
> could be used for that, but unfortunately older versions of GCC don't
> support it.  So I implemented a poor man's version of it with the
> __LINE__ macro.

The above macro is protected by '#ifdef RETPOLINE', and I seriously
doubt 0-day is testing with an unreleased version of GCC.  So you
shouldn't see a 0-day warning.

I think I heard that retpolines won't be ported to anything older than
GCC 4.9, so maybe it's safe to use '%='.  I don't remember when it was
introduced into GCC though.

-- 
Josh