Re: Stackleak vs noinstr (Was: [GIT pull] objtool/core for v5.16-rc1)

[Borislav Petkov]

On Tue, Nov 02, 2021 at 10:50:06AM -0700, Linus Torvalds wrote:
> On Tue, Nov 2, 2021 at 3:05 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >
> > Having the plugin gate on section name seems a lot hacky, but given it's
> > already doing that, one more doesn't hurt.
> 
> Looks sane to me.
> 
> Some of the other warnings are just odd.
> 
> Why is mce_setup() 'noinst'? I'm not seeing any reason for it, but
> maybe I'm just blind. That one complains about the memcpy() call.
> 
> Of course, I suspect memcpy/memset might be better off noinstr anyway,
> exactly because they can happen for very regular C code (struct
> assignments etc). But mce_setup() doesn't really seem to have much
> reason to not be instrumented.

That is going away in my local patchset here. The aim is to have the #MC
handler be noinstr, ofc, but that thing calls a bunch of other functions
and even external ones so it needs careful massaging without destroying
the whole house of cards in the process. :-)

I should have something palatable - read: properly split patches - soon.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette