Re: [PATCH v5 02/12] x86/retpoline: Add initial retpoline support

[Tom Lendacky]

On 1/6/2018 3:21 PM, Woodhouse, David wrote:
> On Sat, 2018-01-06 at 21:16 +0000, Andrew Cooper wrote:
>> On 06/01/18 11:49, David Woodhouse wrote:
>>> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
>>> index 372ba3f..40e6e54 100644
>>> --- a/arch/x86/kernel/cpu/common.c
>>> +++ b/arch/x86/kernel/cpu/common.c
>>> @@ -904,6 +904,11 @@ static void __init early_identify_cpu(struct cpuinfo_x86 *c)
>>> ÂÂ
>>> ÂÂÂÂÂÂÂsetup_force_cpu_bug(X86_BUG_SPECTRE_V1);
>>> ÂÂÂÂÂÂÂsetup_force_cpu_bug(X86_BUG_SPECTRE_V2);
>>> +#ifdef CONFIG_RETPOLINE
>>> +ÂÂÂÂÂsetup_force_cpu_cap(X86_FEATURE_RETPOLINE);
>>> +ÂÂÂÂÂif (c->x86_vendor == X86_VENDOR_AMD)
>>> +ÂÂÂÂÂÂÂÂÂÂÂÂÂsetup_force_cpu_cap(X86_FEATURE_RETPOLINE_AMD);
>>
>> This isn't safe. It needs to be dependant on finding that LFENCEs are
>> actually dispatch serialising.
>>
>> In particular, when virtualised, you'll most likely be saddled with the
>> hypervisors choice of setting, in which case you need to use retpoline
>> as a fallback.
> 
> Thanks. I was about to rebase on top of tip/x86/pti which has Tom's
> patches to make lfence serialising â which seem to say that if the MSR
> isn't available, it *will* be serialising.
> 
> I think I'll just refrain from setting X86_FEATURE_RETPOLINE_AMD for
> now, and let Tom turn that on in his own time.

I can do that.  I'll move it to arch/x86/kernel/cpu/amd.c to just after
the line that sets the MSR bit making lfence serializing.  I'll submit
that once your patches are pulled in (or at least the feature bits).

Thanks,
Tom

>