Re: Fw: Re: [mm PATCH 4/6] RCU: (now) CPU hotplug

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Tue, Jan 30, 2007 at 08:33:40AM +0100, Ingo Molnar wrote:
> 
> * Paul E. McKenney <paulmck@linux.vnet.ibm.com> wrote:
> 
> > > in fact (new) kprobes uses the freezer, and it's far more 
> > > performance sensitive than the handling of CPU hotplug events.
> > 
> > Outside of realtime workloads, I agree that performance should not be 
> > a problem.  And I don't know of any reason why realtime systems need 
> > to be able to do hotplug CPU.  Yet, anyway.  ;-)
> 
> even for -rt it's not really an issue: the hotplug locks are so 
> all-encompassing and so unbound at the moment that there's no realistic 
> expectation for them to ever become deterministic. So we might as well 
> make them encompass "everything" - without any noticeable effect.
> 
> > So the thought is to make _cpu_down() and _cpu_up() do something like 
> > the following (untested, probably does not even compile), perhaps with 
> > suitable adjustments elsewhere as well?
> > 
> > 	static int _cpu_down(unsigned int cpu)
> > 	{
> > 		int err;
> > 		struct task_struct *p;
> > 		cpumask_t old_allowed, tmp;
> > 
> > 		if (num_online_cpus() == 1)
> > 			return -EBUSY;
> > 
> > 		if (!cpu_online(cpu))
> > 			return -EINVAL;
> > 
> > 		if (freeze_processes()) {
> > 			err = -EBUSY;
> > 			goto out_freeze_notify_failed;
> > 		}
> > 		err = raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE,
> > 							(void *)(long)cpu);
> 
> yeah. This all looks so nice that i almost cannot believe it's true :-) 

Well, it turns out that maybe it is in fact untrue.  :-/

I need to look at all uses of PF_NOFREEZE -- as I understand the
code, processes marked PF_NOFREEZE will continue running, potentially
interfering with the hotplug operation.  :-(

I will pass my findings on to this list.

> This would allow us to rip out all the cpu-hotplug locking: wow! If only
> someone would volunteer to try to pull this off and then to touch so
> many subsystems ;-)

Hey, just ending the debates on how to do CPU-hotplug locking would be
worth something!  ;-)

> i fully agree that the opposite notifications should be traversed in 
> inverse order [but this is an orthogonal improvement]. Too bad the 
> notifier list is a single linked list.

:-(

But there can't be -that- many elements in that list...  But agreed,
separate issue.

							Thanx, Paul