The -rt scheduler currently evaluates all online CPUs whenever one or more
enters an rt_overload condition. This suffers from scalability limitations
as the # of online CPUs increases since each rq->lock needs to be acquired by
each cpu. So we introduce a cpumask to track exactly which CPUs need RT
balancing.
Signed-off-by: Gregory Haskins <[EMAIL PROTECTED]>
---
kernel/sched.c | 12 +++++++++---
1 files changed, 9 insertions(+), 3 deletions(-)
diff --git a/kernel/sched.c b/kernel/sched.c
index 93fd6de..aaacec2 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -631,6 +631,7 @@ static inline struct rq *this_rq_lock(void)
#if defined(CONFIG_PREEMPT_RT) && defined(CONFIG_SMP)
static __cacheline_aligned_in_smp atomic_t rt_overload;
+static cpumask_t rto_cpus;
#endif
static inline void inc_rt_tasks(struct task_struct *p, struct rq *rq)
@@ -639,8 +640,11 @@ static inline void inc_rt_tasks(struct task_struct *p,
struct rq *rq)
if (rt_task(p)) {
rq->rt_nr_running++;
# ifdef CONFIG_SMP
- if (rq->rt_nr_running == 2)
+ if (rq->rt_nr_running == 2) {
+ cpu_set(rq->cpu, rto_cpus);
+ smp_wmb();
atomic_inc(&rt_overload);
+ }
# endif
}
#endif
@@ -653,8 +657,10 @@ static inline void dec_rt_tasks(struct task_struct *p,
struct rq *rq)
WARN_ON(!rq->rt_nr_running);
rq->rt_nr_running--;
# ifdef CONFIG_SMP
- if (rq->rt_nr_running == 1)
+ if (rq->rt_nr_running == 1) {
atomic_dec(&rt_overload);
+ cpu_clear(rq->cpu, rto_cpus);
+ }
# endif
}
#endif
@@ -1503,7 +1509,7 @@ static void balance_rt_tasks(struct rq *this_rq, int
this_cpu)
*/
next = pick_next_task(this_rq, this_rq->curr);
- for_each_online_cpu(cpu) {
+ for_each_cpu_mask(cpu, rto_cpus) {
if (cpu == this_cpu)
continue;
src_rq = cpu_rq(cpu);
-
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
