Re: [PATCH 7/8] Clean up workqueue.c with respect to the freezer based cpu-hotplug

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

On Thu, Apr 12, 2007 at 07:52:20AM +0530, Srivatsa Vaddagiri wrote:
> I wonder if there is some value in "enforcing" an order in which
> processes get frozen i.e freeze A first before B. That may solve the
> deadlocks we have been discussing wrt kthread_stop and flush_workqueue
> as well.
>
> The idea is similar to how deadlock wrt multiple locks are solved -
> where a ordering is enforced. Take Lock A first before Lock B. 
> 
> If process A waits on B (like in kthread_stop or flush_workqueue), then if we:
> 
> 	1. Insert A and B in a list (freeze_me_first_list)

In that case, A should insert the dependency into the
freeze_me_first_list as B is unaware of the dependency yet.

What if by the time A has inserted the dependency B is already frozen?

Can very well happen right?

A:			B:			freezer
-------------------------------------------------------------------------------
						*Check the list. don't find  A/B
						*Mark A freezeable.
						*Mark B freezeable.
			*try_to_freeze();
*insert A, B		.		
 into the		.	
 list.			.
			.		
		  	.
* wait_for_		.
  completion(done);	.			/* Freezer fails at this
  			.			 *  point 
			.			 */
		  	.
		  	.
		  	*complete(done);

*try_to_freeze();

Example: A = a thread doing flush_workqueue.
	 B = worker thread.


Of course, we can always use the freezer_skip around this
wait_for_completion as well as long as the thread A is not marked
PF_NOFREEZE. But with multiple freeze events, it won't be as simple as
that.

> -- 
> Regards,
> vatsa

Thanks and Regards
gautham.
-- 
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!"
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/