Re: [PATCH RFC] gcc-plugins: Handle GCC version mismatch for OOT modules

[Josh Poimboeuf]

On Wed, Mar 03, 2021 at 11:25:34AM -0800, Linus Torvalds wrote:
> On Wed, Mar 3, 2021 at 11:15 AM Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> >
> > Adding Linus, who indicated in another thread that we shouldn't force
> > exact GCC versions because there's no technical reason to do so.
> 
> I do not believe we should recompile everything just because the gcc
> version changes.
> 
> But gcc _plugins_ certainly should depend on the kernel version.
> 
> Very few people should be enabling the gcc plugins in the first place.
> Honestly, most of them are bad, and the people who really care about
> those things have already moved to clang which does the important
> parts natively without the need for a plugin. I'm personally waiting
> for the day when we can just say "let's remove them".

You might be sad to learn that some of the plugins are useful for
hardening of a production distro kernel, like stackleak and structleak.

> But in the meantime, making the plugins depend on the gcc version some
> way is certainly better than not doing so.

So currently, the plugins already so that.  They require the GCC version
to be exact.  If there's a mismatch, then it fails the OOT module build.

But that's not usable for a distro.  When users build OOT modules with a
slight GCC mismatch, it breaks the build, effectively requiring the
exact same GCC version for *all* OOT builds going forward.

So there have been a few proposals to better handle GCC version
mismatches:

1) disable the plugin - this works fine for most plugins except
   randstruct

2) rebuild the plugin whenever the GCC version changes

3) fail the build, like today - effectively blocks distros from using
   plugins

-- 
Josh