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

[Borislav Petkov <bp@alien8.de>]

On Thu, Jan 25, 2018 at 09:10:24AM -0600, Josh Poimboeuf wrote:
> Huh?  GCC doesn't even look inside the inline asm.  That's why we had to
> implement ASM_CALL_CONSTRAINT.

That wasn't very correct. What I meant was: *we* need to tell gcc that
the inline asm *might* clobber registers and which those might be. What
David suggested with ASM_NO_INPUT_CLOBBER is the proper thing to do.

> And the seg fault is objtool's way of telling you you need a
> ANNOTATE_NOSPEC_ALTERNATIVE above the alternative ;-)

Except that it blew up when I did this which doesn't have ALTERNATIVE
(it's the diff I saved :-))

diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
index 390b3dc3d438..16cc2e73d17d 100644
--- a/arch/x86/kernel/cpu/bugs.c
+++ b/arch/x86/kernel/cpu/bugs.c
@@ -281,3 +281,19 @@ ssize_t cpu_show_spectre_v2(struct device *dev,
 	return sprintf(buf, "%s\n", spectre_v2_strings[spectre_v2_enabled]);
 }
 #endif
+
+#ifdef CONFIG_RETPOLINE
+void __fill_rsb_nop(void)
+{
+	cpu_relax();
+}
+
+void __fill_rsb(void)
+{
+	unsigned long loops;
+
+	asm volatile (__stringify(__FILL_RETURN_BUFFER(%0, RSB_CLEAR_LOOPS, %1))
+		      : "=r" (loops), ASM_CALL_CONSTRAINT
+		      : : "memory" );
+}
+#endif
--

> (I know that's not the best answer, will fix it soon.)

Thx.

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.