Re: [PATCH] Hotplug CPU prep V: x86 non-linear CPU numbers

[Rusty Russell]

In message <[e-mail]> you write:
> > @@ -1585,7 +1581,7 @@
> >  
> >  	p = buf;
> >  
> > -	if ((smp_num_cpus == 1) &&
> > +	if ((num_online_cpus() == 1) &&
> >  	    !(error = apm_get_power_status(&bx, &cx, &dx))) {
> >  		ac_line_status = (bx >> 8) & 0xff;
> >  		battery_status = bx & 0xff;
> 
> Are you sure? What if you add another CPU just after the test?

APM is special: what it *really* wants to know is if only CPU 0 is
online (this test will almost certainly fail miserably if the single
online CPU is CPU 1, which previously wasn't possible).

I'll check with the APM maintainer what guarantees he needs: we may
actually need to grab to cpu control lock here.  Anyway apparently I
broke something else in APM.

(BTW, in general, we use magic to avoid locks around num_online_cpus()
et al: we make changes then wait for everyone to voluntarily schedule
before exposing it).

Cheers,
Rusty.
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.