Re: [PATCH v12 6/7] x86/arch_prctl: Add ARCH_[GET|SET]_CPUID

[Andy Lutomirski]

On Nov 21, 2016 12:27 AM, "Ingo Molnar" <mingo@xxxxxxxxxx> wrote:
>
>
> * Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>
> > On Fri, 18 Nov 2016, Ingo Molnar wrote:
> > > * Kyle Huey <me@xxxxxxxxxxxx> wrote:
> > > > + if (test_tsk_thread_flag(prev_p, TIF_NOCPUID) ^
> > > > +     test_tsk_thread_flag(next_p, TIF_NOCPUID)) {
> > > > +         set_cpuid_faulting(test_tsk_thread_flag(next_p, TIF_NOCPUID));
> > > > + }
> > > > +
> > >
> > > Why not cache the required MSR value in the task struct instead?
> > >
> > > That would allow something much more obvious and much faster, like:
> > >
> > >     if (prev_p->thread.misc_features_val != next_p->thread.misc_features_val)
> > >             wrmsrl(MSR_MISC_FEATURES_ENABLES, next_p->thread.misc_features_val);
> > >
> > > (The TIF flag maintenance is still required to get into __switch_to_xtra().)
> > >
> > > It would also be easy to extend without extra overhead, should any other feature
> > > bit be added to the MSR in the future.
> >
> > I doubt that. There are feature enable bits coming up which are not related to
> > tasks.
>
> Any inefficiencies resulting from such features should IMHO be carried by those
> features, not by per task features - but:
>
> > [...] So if we have switches enabling/disabling global features, then we would
> > be forced to chase all threads in order to update all misc_features thread
> > variables. Surely not what we want to do.
>
> What switches would those be? We generally don't twiddle global CPU features post
> bootup - we pick a model on bootup and go with that.

I don't see what problem we're trying to solve here.  If we end up
with a mix of global (and changeable!) features and per-task features,
we can just do:

wrmsrl(MSR_MISC_FEATURES_ENABLES, global_misc_features_val |
next_p->thread.misc_features_val);

This is *still* way faster than rdmsr.