On 12/07/2012 11:46 PM, Tejun Heo wrote:

Hello, again.

On Fri, Dec 07, 2012 at 09:57:24AM -0800, Tejun Heo wrote:

quoted

possible.  Also, I think the right approach would be auditing each
get_online_cpus_atomic() callsites and figure out proper locking order
rather than implementing a construct this unusual especially as
hunting down the incorrect cases shouldn't be difficult given proper
lockdep annotation.

On the second look, it looks like you're implementing proper
percpu_rwlock semantics

Ah, nice! I didn't realize that I was actually doing what I intended
to avoid! ;-)

Looking at the implementation of lglocks, and going by Oleg's earlier
comment that we just need to replace spinlock_t with rwlock_t in them
to get percpu_rwlocks, I was horrified at the kinds of circular locking
dependencies that they would be prone to, unless used carefully.

So I devised this scheme to be safe, while still having relaxed rules.
But if *this* is what percpu_rwlocks should ideally look like,
then great! :-)

as readers aren't supposed to induce circular
dependency directly.

Yep, in this scheme, nobody will end up in circular dependency.

 Can you please work with Oleg to implement
proper percpu-rwlock and use that for CPU hotplug rather than
implementing it inside CPU hotplug?

Sure, I'd be more than happy to!
 
Regards,
Srivatsa S. Bhat