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

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Wed, Jun 10, 2015 at 03:08:14PM +0200, Andi Kleen wrote:
> 
> > 2. Each callable function must never leave its own bounds (i.e. with a
> >    jump to outside the function) except when returning.
> 
> That prevents a lot of optimizations with out of line code.

In most cases there are ways to keep the optimizations.  For example:

- grow the function bounds to keep the jump internal
- duplicate the destination code inside the function
- convert the jump to a call

Also note that these rules only affect _callable_ functions, so the
entry code and other non-function asm code can still be a pile of
spaghetti (though I think Andy is working on improving that).

> In fact even gcc with the right options can generate code that violates
> this. Standard Linux constructions, such as exception handling,
> also violate this.
>
> If your tool needs that your tool is broken.

This tool only validates asm code, so I don't see how whatever gcc does
is relevant.

> BTW any other frame pointer requirement should be also optional,
> as it slows down a number of CPUs, such as Atoms.

Yes.  This patch set is a first step towards being able to disable frame
pointers.  Once all callable functions are reasonably structured, we can
start generating and validating DWARF CFI data.  Then we can make a
reliable DWARF unwinder and get rid of frame pointers.

-- 
Josh