Re: [PATCH] sched: fix priority leakage in pick_next_highest_task_rt()

[Yong Zhang <yong.zhang0@gmail.com>]

On Wed, May 18, 2011 at 9:14 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> On Tue, May 17, 2011 at 10:53:22PM +0800, Hillf Danton wrote:
>> On Tue, May 17, 2011 at 10:28 AM, Yong Zhang <yong.zhang0@gmail.com> wrote:
>> > On Mon, May 16, 2011 at 8:55 PM, Hillf Danton <dhillf@gmail.com> wrote:
>> >> When picking the second highest RT task for a given runqueue, if no
>> >> task found after scanning the queue of priority == idx, the next idx
>> >> should also be checked even in case that next is already existing, or
>> >> the window of priority leakage could be opened.
>> >
>> > I don't see what kind of problem you patch will fix.
>> > And mind explaining how priority leakage could happen?
>> >
>> Hi Yong
>>
>> If no task is found after scanning the list at array->queue + idx,
>> what should we operate on next?
>> And why is the list scanned?
>>
>
> The patch looks correct.
>
> The code looks like so:
>
>        for_each_leaf_rt_rq(rt_rq, rq) {
>                array = &rt_rq->active;
>                idx = sched_find_first_bit(array->bitmap);
> next_idx:
>                if (idx >= MAX_RT_PRIO)
>                        continue;
>                if (next && next->prio < idx)
>                        continue;
>                list_for_each_entry(rt_se, array->queue + idx, run_list) {
>                        struct task_struct *p;
>
>                        if (!rt_entity_is_task(rt_se))
>                                continue;
>
>                        p = rt_task_of(rt_se);
>                        if (pick_rt_task(rq, p, cpu)) {
>                                next = p;
>                                break;
>                        }
>                }
>                if (!next) {
>                        idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
>                        goto next_idx;
>                }
>        }
>
> What we are doing is looking for the next highest prio task that we can
> migrate. When we find the next highest priority task that can migrate,
> we pick it. But the issue comes with the cgroups. If we are looping
> through the cgroups, and we pick a task in one cgroup, but when we check
> the next cgroup, if it has a higher priority task, but that task can't
> migrate, but the next one, also of higher priority, can, that "if (!next)"
> wont catch it.

Yup, I misread the patch at the first time.

Now I think Hillf's patch is correct.

Thanks for your explanation Steven.

Thanks,
Yong

>
> Although, I don't know the cgroup code very well, and I wonder what it
> means to pull a task from a run queue onto another run queue that has
> dropped in priority.
>
> But, anyway, for the patch:
>
> Acked-by: Steven Rostedt <rostedt@goodmis.org>
>
> -- Steve
>
>



-- 
Only stand for myself