Re: [PATCH v5 00/10] x86/asm: Compile-time asm code validation

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Wed, Jun 10, 2015 at 11:24:05AM -0700, Andy Lutomirski wrote:
> Slightly off-topic, but this reminds me: when writing inline asm that
> needs to push to the stack (for whatever reason), it's incredibly
> messy to get the annotations right -- they're different depending on
> whether the previous frame base (is that what "CFA" is?) is currently
> sp + constant, in which case we need an annotation adjusting the
> constant or whether it's independent of sp (bp + constant), in which
> case we shouldn't adjust the offset.  (If it's some other function of
> sp, we're screwed.)
> 
> Regardless of whether these types of annotations end up being done by
> hand or by script, should we consider asking the binutils people to
> give us some nice .cfi_adjust_for_push and .cfi_adjust_for_pop or
> similar directives?

Hm, that's a tough one.  Might be worth asking...

Another alternative would be to ask gcc to make a change to always setup
the frame pointer for any function which has inline assembly, so that
you know (hopefully) that CFA is based on bp.

Or, maybe there's already a way to force gcc to do that with the asm
directive somehow?

> 
> See here for Jan Beulich's solution, which is incomprehensible to me:
> 
> http://thread.gmane.org/gmane.linux.kernel/1820765

<brain explodes>

-- 
Josh