On 27.09.22 12:10, Borislav Petkov wrote:
> On Tue, Sep 27, 2022 at 10:57:37AM +0200, Juergen Gross wrote:
>> TBH I don't see the point of having an accessor which is just setting a
>> variable to "true". But if you like it better, I can keep it.
> 
> Accessors are always better, no matter how silly. :)

Okay, then I'll keep it.

> But, in trying to grok your next patch - you really should split those
> more complex ones because they're a pain to review - I'm starting to
> wonder whether we could even remove mtrr_aps_delayed_init and make the
> delayed init the default.
> 
> Because, AFAICT, set_mtrr_aps_delayed_init() is called by default
> by native_smp_prepare_cpus(). Which is called by hyperv and
> arch/x86/xen/smp_hvm.c.
> 
> The only one that's not calling it is arch/x86/xen/smp_pv.c but that
> thing doesn't support MTRRs in the first place, right?

Correct.

> Which means, it doesn't need delayed MTRR init anyway.
> 
> Which would then mean that this would simplify this ugly logic even more.
> 
> Or am I missing an angle?

Yes: cpu hotplug.


Juergen