Re: [PATCH] x86/retpoline: Fix NOSPEC_JMP for tip

From: Andi Kleen
Date: Tue Jan 09 2018 - 20:30:16 EST


On Tue, Jan 09, 2018 at 06:45:34PM -0600, Tom Lendacky wrote:
> On 1/9/2018 6:40 PM, Thomas Gleixner wrote:
> > On Tue, 9 Jan 2018, Linus Torvalds wrote:
> >
> >> On Tue, Jan 9, 2018 at 4:31 PM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
> >>>
> >>> The following patch fixes it for me. Something doesn't
> >>> seem to work with ALTERNATIVE_2. It adds only a few bytes
> >>> more code, so seems acceptable.
> >>
> >> Ugh. It's kind of stupid, though.
> >>
> >> Why is the code sequence not simply:
> >>
> >> ALTERNATIVE "", "lfence", X86_FEATURE_RETPOLINE_AMD
> >> ALTERNATIVE __stringify(jmp *\reg), __stringify(RETPOLINE_JMP \reg),
> >> X86_FEATURE_RETPOLINE
> >>
> >> ie make that X86_FEATURE_RETPOLINE_AMD _only_ emit the "lfence", and
> >> simply fall through to what will be the "jmp *\reg" of the
> >> non-RETPOLINE version.
> >>
> >> Then just make sure X86_FEATURE_RETPOLINE_AMD disables X86_FEATURE_RETPOLINE.
>
> I think there are areas that rely on X86_FEATURE_RETPOLINE being set
> even if X86_FEATURE_RETPOLINE_AMD is set. For example, line 261 in
> arch/x86/entry/entry_32.S is only checking for X86_FEATURE_RETPOLINE.

I audited the difference places. They all seem ok.

I assume you don't need FILL_RETURN_BUFFER on AMD. If not let me know
and we can add a X86_FEATURE_RETPOLINE_COMMON

-Andi