Re: current linux-2.6.git: cpusets completely broken

[Ingo Molnar <mingo@elte.hu>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> The thing is, we should fix the top level code to never even 
> _consider_ an invalid CPU as a target, and that in turn should mean 
> that all the other code should be able to just totally ignore CPU 
> hotplug events.

agreed. We thought we could get away by hiding the "is the CPU dead" 
information in existing data structures (by shaping sched domains to 
never lead to a dead CPU) - but this method has proven itself fragile 
via a series of bugs.

It was one micro-optimization one too many. We should just accept the 
fact that the current model is not maintainable and add your extra (and 
trivial) cpu_active_map layer that protects against migrating to CPUs 
that are going down. [we'll basically introduce a "going down" state 
inbetween 'online' and 'offline']

And this will get rid of some other fragile trickery - because from that 
point on we dont have to be super-careful about the whole sequence of 
manipulating sched domains anymore. Cpusets can do whatever it wants, it 
wont be able to break hotplug+scheduling - and that's important for any 
functionality that is not used by default.

[ I dont think Dmitry will disagree with this notion all that much, it's
  just that his personal limit for calling an algorithm unmaintainable
  is probably a lot higher than normal :-) ]

	Ingo