2009-03-27 16:20:31

by Dhaval Giani

Subject: [PATCH] sched: rt group scheduling, compare child's limit against parent's

Hi Ingo, Peter,

I've not been able to follow the reasoning behind checking the sum of
the children's runtime not exceeding the parent's runtime as opposed to
the child's runtime not exceeding that of the parent.

>From what I understand, the limits only guarantee that the group cannot
exceed a certain limit and not guarantee a group gets to run. If my
understanding is corrct, maybe the following patch is needed?

Thanks,

--
sched: rt group scheduling, compare child's limit against parent's

The current logic checks that the sum of the children's runtime in rt
group scheduling does not exceed the parent's. This logic is wrong,
all we claim is that the group cannot exceed the amount of time allowed.

Change the code to reflect this.

Signed-off-by: Dhaval Giani <[email protected]>
---
kernel/sched.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -9601,7 +9601,7 @@ static int tg_schedulable(struct task_gr
{
struct rt_schedulable_data *d = data;
struct task_group *child;
- unsigned long total, sum = 0;
+ unsigned long total;
u64 period, runtime;

period = ktime_to_ns(tg->rt_bandwidth.rt_period);
@@ -9640,9 +9640,10 @@ static int tg_schedulable(struct task_gr
return -EINVAL;

/*
- * The sum of our children's runtime should not exceed our own.
+ * The child's runtime should not exceed our own.
*/
list_for_each_entry_rcu(child, &tg->children, siblings) {
+ unsigned long ratio;
period = ktime_to_ns(child->rt_bandwidth.rt_period);
runtime = child->rt_bandwidth.rt_runtime;

@@ -9651,11 +9652,11 @@ static int tg_schedulable(struct task_gr
runtime = d->rt_runtime;
}

- sum += to_ratio(period, runtime);
- }
+ ratio = to_ratio(period, runtime);

- if (sum > total)
- return -EINVAL;
+ if (ratio > total)
+ return -EINVAL;
+ }

return 0;
}
--
regards,
Dhaval

2009-03-27 16:24:12

by Peter Zijlstra

Subject: Re: [PATCH] sched: rt group scheduling, compare child's limit against parent's

On Fri, 2009-03-27 at 21:49 +0530, Dhaval Giani wrote:
> Hi Ingo, Peter,
>
> I've not been able to follow the reasoning behind checking the sum of
> the children's runtime not exceeding the parent's runtime as opposed to
> the child's runtime not exceeding that of the parent.
>
> From what I understand, the limits only guarantee that the group cannot
> exceed a certain limit and not guarantee a group gets to run. If my
> understanding is corrct, maybe the following patch is needed?

NAK, rt needs that as it guarantees a min too.