Re: [tip:x86/pti] x86/retpoline: Fill return stack buffer on vmexit

From: Jiri Kosina
Date: Thu Jan 25 2018 - 13:41:11 EST


On Thu, 25 Jan 2018, Andy Lutomirski wrote:

> Distros that use retpolines need their driver vendors to recompile no
> matter what.

Absolutely. Tainting a kernel, issuing a warning, or even voluntarily
deciding to not load modules loaded without retpolines, that all sounds
like reasonable aproaches.

Artificially introducing kernel ABI breakage which is not there (as
retpolines are fully compatible when it comes to ABI between modules and
kernel ... the fact that it potentially brings non-retpolined indirect
jump into the kernel is a security concent, but not ABI issue) sounds like
a bad idea.

Those two things (ABI and security concerns) are independent.

> Distros that use IBRS and refuse to use retpolines should get put on a
> list of "didn't actually adequately mitigate spectre".

Oh absolutely, especially on archs where there is no IBRS. But how is this
relevant to ABI?

Thanks,

--
Jiri Kosina
SUSE Labs