Re: [PATCH v6 4/4] stackvalidate: Add ignore macros

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Tue, Jul 07, 2015 at 04:00:54PM -0700, Andy Lutomirski wrote:
> On Tue, Jul 7, 2015 at 3:59 PM, Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> > On Tue, Jul 07, 2015 at 03:00:38PM -0700, Andy Lutomirski wrote:
> >> On Tue, Jul 7, 2015 at 7:54 AM, Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> >> > Add new stackvalidate ignore macros: STACKVALIDATE_IGNORE_INSN and
> >> > STACKVALIDATE_IGNORE_FUNC.  These can be used to tell stackvalidate to
> >> > skip validation of an instruction or a function, respectively.
> >> >
> >> > Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
> >> > ---
> >> >  include/linux/stackvalidate.h | 38 ++++++++++++++++++++++++++++++++++++++
> >> >  1 file changed, 38 insertions(+)
> >> >  create mode 100644 include/linux/stackvalidate.h
> >> >
> >> > diff --git a/include/linux/stackvalidate.h b/include/linux/stackvalidate.h
> >> > new file mode 100644
> >> > index 0000000..30d4a60
> >> > --- /dev/null
> >> > +++ b/include/linux/stackvalidate.h
> >> > @@ -0,0 +1,38 @@
> >> > +#ifndef _LINUX_STACKVALIDATE_H
> >> > +#define _LINUX_STACKVALIDATE_H
> >> > +
> >> > +#ifndef __ASSEMBLY__
> >> > +
> >> > +/*
> >> > + * This C macro tells the stack validation script to ignore the function.  It
> >> > + * should only be used in special cases where you're 100% sure it won't affect
> >> > + * the reliability of frame pointers and kernel stack traces.
> >> > + *
> >> > + * For more information, see Documentation/stack-validation.txt.
> >> > + */
> >> > +#define STACKVALIDATE_IGNORE_FUNC(_func) \
> >> > +       void __attribute__((section("__stackvalidate_ignore_func,\"ae\"#"))) \
> >> > +               *__stackvalidate_ignore_func_##_func = _func
> >> > +
> >>
> >> static?  Otherwise there's some risk that ignoring a static function
> >> will cause a duplicate symbol.  Alternatively you could generate a
> >> more unique name.
> >>
> >> Also, should the linker script be updated to discard this section?
> >
> > It validates per individual object file, so there shouldn't be a risk of
> > duplicate symbols.  The 'e' flag in ,\"ae\"#" tells the linker to
> > discard the section.
> >
> > (The ,\"ae\"# stuff is a horrible hack, but it's the only way I could
> > figure out how to set the section flags from C code.  The '#' is used to
> > comment out gcc's default arguments to the .section directive in favor
> > of the "ae" flags.)
> 
> Oh, egads.  FWIW, doing it in the linker script would probably be less hackish.

Yeah.  I'll try that.

> Does the linker discard before noticing duplicate symbols in that section?

I think it does, but I'll verify.  I deliberately made it global so the
compiler doesn't discard the symbol for being unused.  But I can use the
unused attribute to avoid that.

-- 
Josh