On 12/06/2012 02:27 AM, Tejun Heo wrote:

Hello,

On Thu, Dec 06, 2012 at 02:01:35AM +0530, Srivatsa S. Bhat wrote:

quoted

Yes, that _sounds_ sufficient, but IMHO it won't be, in practice. The
*number* of call-sites that you need to convert from preempt_disable/enable
to get/put_online_cpus_atomic() won't be too many, however the *frequency*
of usage of those call-sites can potentially be very high.

I don't think that will be the case and, even if it is, doing it this
way would make it difficult to tell.  The right thing to do is
replacing stop_machine with finer grained percpu locking first.
Refining it further should happen iff that isn't enough and there
isn't an simpler solution.  So, let's please do the simple conversion
first.

Hmm, OK, that sounds like a good plan. So I'll drop the "light" and
"full" variants for now and work on providing a straight-forward
get/put_online_cpus_atomic() APIs.

Thank you!
 
Regards,
Srivatsa S. Bhat