Re: [PATCH] x86/tsx: Use MSR_TSX_CTRL to clear CPUID bits

[Pawan Gupta]

On 15.02.2022 11:24, Borislav Petkov wrote:
> On Mon, Feb 14, 2022 at 04:20:14PM -0800, Pawan Gupta wrote:
> > ... we are calling tsx_clear_cpuid() unconditionally.
>
> I know, that's why I asked...
>
> > > If those CPUs which support only disabling TSX through MSR_IA32_TSX_CTRL
> > > but don't have MSR_TSX_FORCE_ABORT - if those CPUs set
> > > X86_FEATURE_RTM_ALWAYS_ABORT too, then this should work.
>
> ... this^^.
>
> IOW, what are you fixing here exactly?
>
> Let's look at the two callsites of tsx_clear_cpuid():
>
> 1. tsx_init: that will do something on X86_FEATURE_RTM_ALWAYS_ABORT CPUs.
>
> 2. init_intel: that will get called when
>
> 	tsx_ctrl_state == TSX_CTRL_RTM_ALWAYS_ABORT
>
> But TSX_CTRL_RTM_ALWAYS_ABORT gets set only when
> X86_FEATURE_RTM_ALWAYS_ABORT is set. I.e., the first case, in
> tsx_init().
>
> So, IIUC, you wanna fix the case where CPUs which set
> X86_FEATURE_RTM_ALWAYS_ABORT but *don't* have MSR_TSX_FORCE_ABORT, those
> CPUs should still disable TSX through MSR_IA32_TSX_CTRL.
>
> Correct?

That is exactly what this patch is fixing. Please let me know if you
have any questions.

Thanks,
Pawan