From: Hailong Liu <liu.hailo...@zte.com.cn> The rq->rd->span of a cpu in a system with isolated cpus splited into two different parts: one is for isolated cpus, another for non-isolated cpus.
When CONFIG_RT_GROUP_SCHED enabled, the handler of sched_rt_period_timer updates rt_time and rt_runtime for every cpus in rq(this_cpu)->rd->span. It means that other parts cpus out of this_cpu's rd->span will be missed by sched_rt_period_timer handler, when CONFIG_RT_GROUP_SCHED enabled and isolated cpus presents in system. E.g problem will be triggered as follows on my 8 cores machine: 1 enable CONFIG_RT_GROUP_SCHED=y, and boot kernel with command-line "isolcpus=4-7" 2 create a child group and init it: mount -t cgroup -o cpu cpu /sys/fs/cgruop mkdir /sys/fs/cgroup/child0 echo 950000 > /sys/fs/cgroup/child0/cpu.rt_runtime_us 3 run two rt-loop tasks, assume their pids are $pid1 and $pid2 4 affinity a rt task to the isolated cpu-sets taskset -p 0xf0 $pid2 5 add tasks created above into child cpu-group echo $pid1 > /sys/fs/cgroup/child0/tasks echo $pid2 > /sys/fs/cgroup/child0/tasks 6 check wat happened: "top": one of the task will fail to has cpu usage, but its stat is "R" "kill": the task on the problem rt_rq can't be killed This patch will fix this problem. Signed-off-by: Hailong Liu <liu.hailo...@zte.com.cn> --- kernel/sched/rt.c | 15 +++------------ 1 file changed, 3 insertions(+), 12 deletions(-) diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 49ec096a8..c5c39695c 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -855,19 +855,10 @@ static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun) int i, idle = 1, throttled = 0; const struct cpumask *span; - span = sched_rt_period_mask(); #ifdef CONFIG_RT_GROUP_SCHED - /* - * FIXME: isolated CPUs should really leave the root task group, - * whether they are isolcpus or were isolated via cpusets, lest - * the timer run on a CPU which does not service all runqueues, - * potentially leaving other CPUs indefinitely throttled. If - * isolation is really required, the user will turn the throttle - * off to kill the perturbations it causes anyway. Meanwhile, - * this maintains functionality for boot and/or troubleshooting. - */ - if (rt_b == &root_task_group.rt_bandwidth) - span = cpu_online_mask; + span = cpu_online_mask; +#else + span = sched_rt_period_mask(); #endif for_each_cpu(i, span) { int enqueue = 0; -- 2.17.1