Re: [PATCH 0/7] CPU controller - V1

[Srivatsa Vaddagiri <vatsa@in.ibm.com>]

On Mon, Aug 21, 2006 at 05:10:41PM +0000, Mike Galbraith wrote:
> I must be missing something.  If current and awakening tasks have
> separate runqueues, task_rq(awakening)->curr != current.  We won't look
> at current->prio, so won't resched(current).

Ok ..we have two types of runqueues here:

1. struct task_grp_rq
	per-task-group-per-cpu runqueue, which holds ready-to-run tasks 
	belonging to the group in active and expired arrays.

2. struct rq
	per-cpu runqueue, which holds ready-to-run task-groups in active and 
	expired arrays. This structure also holds some members like
	curr, nr_running etc which more or less have the same significance as 
	the current runqueue members.

task_rq(tsk) still extracts "struct rq", while
task_grp(tsk)->rq[task_cpu(tsk)] extracts "struct task_grp_rq".

Hence task_rq(awakening)->curr == current, which should be sufficient to 
resched(current), although I think there is a bug in current code 
(irrespective of these patches):

try_to_wake_up() :
	
	...

        if (!sync || cpu != this_cpu) {
                if (TASK_PREEMPTS_CURR(p, rq))
                        resched_task(rq->curr);
        }
        success = 1;

	...

TASK_PREEMPTS_CURR() is examined and resched_task() is called only if 
(cpu != this_cpu). What about the case (cpu == this_cpu) - who will
call resched_task() on current? I had expected the back-end of interrupt
handling to do that, but didnt find any code to do so.

-- 
Regards,
vatsa