Re: [RFC][PATCH 4/4] Rename lock_cpu_hotplug/unlock_cpu_hotplug

[Gautham R Shenoy <ego@in.ibm.com>]

On Thu, Aug 24, 2006 at 05:00:26PM +0200, Ingo Molnar wrote:
> 
> * Nick Piggin <nickpiggin@yahoo.com.au> wrote:
> 
> > It really is just like a reentrant rw semaphore... I don't see the 
> > point of the name change, but I guess we don't like reentrant locks so 
> > calling it something else might go down better with Linus ;)
> 
> what would fit best is a per-cpu scalable (on the read-side) 
> self-reentrant rw mutex. We are doing cpu hotplug locking in things like 
> fork or the slab code, while most boxes will do a CPU hotplug event only 
> once in the kernel's lifetime (during bootup), so a classic global 
> read-write lock is unjustified.

I agree. However, I was not sure if anything else other than for cpu_hotplug,
needs a self-reentrent rwmutex in the kernel. 
Which is why I did not expose the locking(at least the write side of it)
outside. We don't want too many lazy programmers anyway! 

However, even in case of cpu_hotplug, if we want to prevent
a hotplug event in some critical region where we are not going to sleep, 
we may as well use preempt_disable[/enable]. Because __stop_machine_run waits 
for all the tasks in the fast-path to complete before it changes
the online_cpus map, if I am not mistaken.

Only when you want your local online cpu_map to remain intact when 
you wake up from sleep, should you use cpu_hotplug *lock*.

Ingo?

-- 
Gautham R Shenoy
Linux Technology Center
IBM India.
"Freedom comes with a price tag of responsibility, which is still a bargain,
because Freedom is priceless!"