Re: RCU question

[George Anzinger <george@mvista.com>]

Lee Revell wrote:
> On Sat, 2004-12-11 at 02:10 +0530, Dipankar Sarma wrote:
> 
>>On Fri, Dec 10, 2004 at 11:42:55AM -0800, George Anzinger wrote:
>>
>>>Dipankar Sarma wrote:
>>>
>>>>And yes, RCU processing in softirq context can re-raise the softirq.
>>>>AFAICS, it is perfectly normal.
>>>
>>>My assumption was that, this being the idle task, RCU would be more than 
>>>happy to finish all its pending tasks.
>>
>>We try to avoid really long running softirqs (RCU tasklet in this case)
>>for better scheduling latency. A long running rcu tasklet during
>>an idle cpu may delay running of an RT process that becomes runnable
>>during the rcu tasklet.
>>
> 
> 
> Well, softirqs should really be preemptible if you care about RT task
> latency.  Ingo's patches have had this for months.  Works great.  Maybe
> it's time to push it upstream.

Yes, I understand, and soft_irq() does turn on interrupts...
I was thinking of something like:

	while(softirq_pending()) {
		local_irq_enable();
		do_softirq();
		local_irq_disable();
	}
	
	<proceed to idle hlt...>
-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/