patch submission: x86 trapsleds

[Todd Mortimer via llvm-commits]

Hi Joerg,

Thanks for taking the time to have a look.

On Thu, Jul 20, 2017 at 11:02:16AM +0200, Joerg Sonnenberger wrote:
> I don't believe turning a single padding instruction into a jump is a
> good idea for any intra-function place. You should at the very least
> demonstrate the performance (non-)regression by running LNT.

No problem. I am figuring LNT out now, and will happily post results
when I have them.

> You will also need at least a functional test case before this can be
> further considered. 

Okay. I am not 100% clear what this means. This patch doesn't change the
behaviour of programs at all, unless they (a) jump into the middle of a
padding section (which they should not do) or (b) are out of control via
exploitation or other error. What kind of functional test case would be
appropriate? Is there an existing test case that targets a similar
condition that could guide me to writing an acceptable test?

Thanks again for your time. I am hopeful that this change can be
accepted into llvm, even if it is behind an option or something.

Todd