]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
sched: fix bandwidth validation for UID grouping
authorPeter Zijlstra <peterz@infradead.org>
Wed, 14 Jan 2009 09:56:32 +0000 (10:56 +0100)
committerIngo Molnar <mingo@elte.hu>
Thu, 15 Jan 2009 14:07:27 +0000 (15:07 +0100)
Impact: make rt-limit tunables work again

Mark Glines reported:

> I've got an issue on x86-64 where I can't configure the system to allow
> RT tasks for a non-root user.
>
> In 2.6.26.5, I was able to do the following to set things up nicely:
> echo 450000 >/sys/kernel/uids/0/cpu_rt_runtime
> echo 450000 >/sys/kernel/uids/1000/cpu_rt_runtime
>
> Seems like every value I try to echo into the /sys files returns EINVAL.

For UID grouping we initialize the root group with infinite bandwidth
which by default is actually more than the global limit, therefore the
bandwidth check always fails.

Because the root group is a phantom group (for UID grouping) we cannot
runtime adjust it, therefore we let it reflect the global bandwidth
settings.

Reported-by: Mark Glines <mark@glines.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched.c

index 3b630d8826605eb4e579a2b2ae48208b52f2de22..ed62d1cee05c0310f56976fec9d542d044d1ee59 100644 (file)
@@ -9050,6 +9050,13 @@ static int tg_schedulable(struct task_group *tg, void *data)
                runtime = d->rt_runtime;
        }
 
+#ifdef CONFIG_USER_SCHED
+       if (tg == &root_task_group) {
+               period = global_rt_period();
+               runtime = global_rt_runtime();
+       }
+#endif
+
        /*
         * Cannot have more runtime than the period.
         */