Re: [PATCH] x86/bugs: Allow STIBP with IBRS

[Josh Poimboeuf <jpoimboe@kernel.org>]

On Mon, Feb 20, 2023 at 11:04:56AM -0800, KP Singh wrote:
> On Mon, Feb 20, 2023 at 11:00 AM Josh Poimboeuf <jpoimboe@kernel.org> wrote:
> >
> > On Mon, Feb 20, 2023 at 10:33:56AM -0800, KP Singh wrote:
> > >  static char *stibp_state(void)
> > >  {
> > > -       if (spectre_v2_in_ibrs_mode(spectre_v2_enabled))
> > > +       if (!spectre_v2_user_needs_stibp(spectre_v2_enabled))
> > >                 return "";
> > >
> > >         switch (spectre_v2_user_stibp) {
> > >
> > > Also Josh, is it okay for us to have a discussion and have me write
> > > the patch as a v2? Your current patch does not even credit me at all.
> > > Seems a bit unfair, but I don't really care. I was going to rev up the
> > > patch with your suggestions.
> >
> > Well, frankly the patch needed a complete rewrite.  The patch
> > description was unclear about what the problem is and what's being
> 
> Josh, this is a complex issue, we are figuring it out together on the
> list. It's complex, that's why folks got it wrong in the first place.
> Calling the patch obtuse and unclear is unfair!
> 
> > fixed.  The code was obtuse and the comments didn't help.  I could tell
> > by the other replies that I wasn't the only one confused.
> 
> The patch you sent is not clear either, it implicitly ties in STIBP
> with eIBRS. There is no explanation anywhere that IBRS just means
> KERNEL_IBRS.

Ok, so something like this on top?

diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
index b97c0d28e573..fb3079445700 100644
--- a/arch/x86/kernel/cpu/bugs.c
+++ b/arch/x86/kernel/cpu/bugs.c
@@ -1201,6 +1201,10 @@ spectre_v2_user_select_mitigation(void)
 	/*
 	 * If no STIBP, enhanced IBRS is enabled, or SMT impossible,
 	 * STIBP is not required.
+	 *
+	 * For legacy IBRS, STIBP may still be needed because IBRS is only
+	 * enabled in kernel space, so user space isn't protected from indirect
+	 * branch prediction attacks from a sibling CPU thread.
 	 */
 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
 	    !smt_possible ||