Re: [PATCH -v2] x86/retpoline: Ensure default return thunk isn't used at runtime

[Josh Poimboeuf <jpoimboe@kernel.org>]

On Thu, Jan 04, 2024 at 02:26:23PM +0100, Borislav Petkov wrote:
> On Thu, Jan 04, 2024 at 02:24:46PM +0100, Borislav Petkov wrote:
> > +void __warn_thunk(void)
> > +{
> > +	pr_warn_once("\n");
> > +	pr_warn_once("**********************************************************\n");
> > +	pr_warn_once("**   NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE   **\n");
> > +	pr_warn_once("**                                                      **\n");
> > +	pr_warn_once("**   Unpatched return thunk in use. This should not     **\n");
> > +	pr_warn_once("**   happen on a production kernel. Please report this  **\n");
> > +	pr_warn_once("**   to x86@kernel.org.                                 **\n");
> 
> I'm not yet sure here whether this should say "upstream kernels" because
> otherwise we'll get a bunch of distro or whatnot downstream kernels
> reports where we can't really do anything about...
> 
> Hmmm.

At the very least, the dump_stack() should be a WARN_ON_ONCE().
Otherwise this is actually *more* likely to be ignored since automated
tools don't have a way to catch it: no taint, no "WARNING" string, no
panic_on_warn, etc.

But also, I'm not a fan of the banner.  A warning is enough IMO.

Many/most warnings can be "security" issues.  A production server which
ignores warnings/taints/etc would be a much bigger problem.

And as you say, there are many frankenkernels out there and upstream
doesn't want to be in the business of debugging them.

-- 
Josh