Re: [PATCH v2 15/45] rcu: Use get/put_online_cpus_atomic() to prevent CPU offline

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 06/26/2013 07:59 PM, David Laight wrote:
>> Once stop_machine() is gone from the CPU offline path, we won't be able
>> to depend on disabling preemption to prevent CPUs from going offline
>> from under us.
> 
> Could you use an rcu-like sequence so that disabling pre-emption
> would be enough?
> 
> Something like rebuilding the cpu list, then forcing yourself
> to run on all the cpu.
> 

Certainly, and we had debated schemes similar to that (including
schemes invoking synchronize_sched() itself) in earlier discussions.
(But IIRC even those schemes required converting call-sites from
preempt_disable() to get/put_online_cpus_atomic(), to properly
synchronize).

> That would be far less intrusive.
> 

But that would increase the latency of hotplug operations like Paul
pointed out, and that in turn is not good for use-cases such as
suspend/resume, where we take all non-boot CPUs offline in a loop.
(That would mean executing the above logic num_online_cpus() times!).

So we started hunting for ideas that can make the hotplug writer
side not only less intrusive, but also remain fast. That's how
the later designs evolved.
 
Regards,
Srivatsa S. Bhat