The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
A new idle_exit function is called when the prev task is the idle function so the elapsed time will be accounted as idle time in the rq's load.
Changes since V5: - Rename idle_enter/exit function to idle_enter/exit_fair
Changes since V4: - Rebase on v3.9-rc6 instead of Steven Rostedt's patches - Create the post_schedule_idle function that was previously created by Steven's patches
Changes since V3: - Remove dependancy with CONFIG_FAIR_GROUP_SCHED - Add a new idle_enter function and create a post_schedule callback for idle class - Remove the update_runnable_avg from idle_balance
Changes since V2: - remove useless definition for UP platform - rebased on top of Steven Rostedt's patches : https://lkml.org/lkml/2013/2/12/558
Changes since V1: - move code out of schedule function and create a pre_schedule callback for idle class instead.
Signed-off-by: Vincent Guittot vincent.guittot@linaro.org --- kernel/sched/fair.c | 23 +++++++++++++++++++++-- kernel/sched/idle_task.c | 16 ++++++++++++++++ kernel/sched/sched.h | 12 ++++++++++++ 3 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 7a33e59..1de3df0 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1562,6 +1562,27 @@ static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq, se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter); } /* migrations, e.g. sleep=0 leave decay_count == 0 */ } + +/* + * Update the rq's load with the elapsed running time before entering + * idle. if the last scheduled task is not a CFS task, idle_enter will + * be the only way to update the runnable statistic. + */ +void idle_enter_fair(struct rq *this_rq) +{ + update_rq_runnable_avg(this_rq, 1); +} + +/* + * Update the rq's load with the elapsed idle time before a task is + * scheduled. if the newly scheduled task is not a CFS task, idle_exit will + * be the only way to update the runnable statistic. + */ +void idle_exit_fair(struct rq *this_rq) +{ + update_rq_runnable_avg(this_rq, 0); +} + #else static inline void update_entity_load_avg(struct sched_entity *se, int update_cfs_rq) {} @@ -5219,8 +5240,6 @@ void idle_balance(int this_cpu, struct rq *this_rq) if (this_rq->avg_idle < sysctl_sched_migration_cost) return;
- update_rq_runnable_avg(this_rq, 1); - /* * Drop the rq->lock, but keep IRQ/preempt disabled. */ diff --git a/kernel/sched/idle_task.c b/kernel/sched/idle_task.c index b6baf37..b8ce773 100644 --- a/kernel/sched/idle_task.c +++ b/kernel/sched/idle_task.c @@ -13,6 +13,16 @@ select_task_rq_idle(struct task_struct *p, int sd_flag, int flags) { return task_cpu(p); /* IDLE tasks as never migrated */ } + +static void pre_schedule_idle(struct rq *rq, struct task_struct *prev) +{ + idle_exit_fair(rq); +} + +static void post_schedule_idle(struct rq *rq) +{ + idle_enter_fair(rq); +} #endif /* CONFIG_SMP */ /* * Idle tasks are unconditionally rescheduled: @@ -25,6 +35,10 @@ static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int fl static struct task_struct *pick_next_task_idle(struct rq *rq) { schedstat_inc(rq, sched_goidle); +#ifdef CONFIG_SMP + /* Trigger the post schedule to do an idle_enter for CFS */ + rq->post_schedule = 1; +#endif return rq->idle; }
@@ -86,6 +100,8 @@ const struct sched_class idle_sched_class = {
#ifdef CONFIG_SMP .select_task_rq = select_task_rq_idle, + .pre_schedule = pre_schedule_idle, + .post_schedule = post_schedule_idle, #endif
.set_curr_task = set_curr_task_idle, diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index cc03cfd..8f1d80e 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -880,6 +880,18 @@ extern const struct sched_class idle_sched_class; extern void trigger_load_balance(struct rq *rq, int cpu); extern void idle_balance(int this_cpu, struct rq *this_rq);
+/* + * Only depends on SMP, FAIR_GROUP_SCHED may be removed when runnable_avg + * becomes useful in lb + */ +#if defined(CONFIG_FAIR_GROUP_SCHED) +extern void idle_enter_fair(struct rq *this_rq); +extern void idle_exit_fair(struct rq *this_rq); +#else +static inline void idle_enter_fair(struct rq *this_rq) {} +static inline void idle_exit_fair(struct rq *this_rq) {} +#endif + #else /* CONFIG_SMP */
static inline void idle_balance(int cpu, struct rq *rq)
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the throttle specifically to not waste cycles doing that on every microscopic idle.
Seems to me when scheduling an rt task, you want to do as little other than switching to/from the rt task as possible. I don't let rt tasks do idle balancing either, their job isn't to balance fair class on the way out the door, it's to get off/onto the cpu ASAP, and do rt work.
-Mike
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
Seems to me when scheduling an rt task, you want to do as little other than switching to/from the rt task as possible. I don't let rt tasks do idle balancing either, their job isn't to balance fair class on the way out the door, it's to get off/onto the cpu ASAP, and do rt work.
I agree but the patch is not about balancing fair task but keep coherent runnable value
Vincent
-Mike
On Fri, 2013-04-19 at 09:49 +0200, Vincent Guittot wrote:
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
If it only scares fair class tasks away from the periodic rt load, that seems like a benefit to me, not a liability. If we really really need perfect load numbers, fine, we have to eat some cycles, but when I look at it, it looks like one of those "Perfect is the enemy of good" things.
-Mike
On 19 April 2013 10:14, Mike Galbraith efault@gmx.de wrote:
On Fri, 2013-04-19 at 09:49 +0200, Vincent Guittot wrote:
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
If it only scares fair class tasks away from the periodic rt load, that seems like a benefit to me, not a liability. If we really really need
I'm not sure that such behavior that is only based on erroneous value, is good one.
perfect load numbers, fine, we have to eat some cycles, but when I look at it, it looks like one of those "Perfect is the enemy of good" things.
The target is not perfect number but good enough to be usable. The systctl_migration_cost threshold is good for idle balancing but can generates wrong load value
Vincent
-Mike
On Fri, 2013-04-19 at 10:50 +0200, Vincent Guittot wrote:
On 19 April 2013 10:14, Mike Galbraith efault@gmx.de wrote:
On Fri, 2013-04-19 at 09:49 +0200, Vincent Guittot wrote:
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
If it only scares fair class tasks away from the periodic rt load, that seems like a benefit to me, not a liability. If we really really need
I'm not sure that such behavior that is only based on erroneous value, is good one.
perfect load numbers, fine, we have to eat some cycles, but when I look at it, it looks like one of those "Perfect is the enemy of good" things.
The target is not perfect number but good enough to be usable. The systctl_migration_cost threshold is good for idle balancing but can generates wrong load value
But again, why do we care? To be able to mix rt and fair loads and still make pretty mixed load utilization numbers? Paying a general case fast path price to make strange (to me) load utilization numbers pretty is not very attractive. If you muck about with rt classes, you need to have a good reason for doing that. If you do have a good reason, you also allocated all resources, including CPU, so don't need the kernel to balance the load for you. Paying any fast path price to make the kernel balance a mixed rt/fair load just seems fundamentally wrong to me.
-Mike
On Fri, 2013-04-19 at 11:21 +0200, Mike Galbraith wrote:
On Fri, 2013-04-19 at 10:50 +0200, Vincent Guittot wrote:
On 19 April 2013 10:14, Mike Galbraith efault@gmx.de wrote:
On Fri, 2013-04-19 at 09:49 +0200, Vincent Guittot wrote:
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
If it only scares fair class tasks away from the periodic rt load, that seems like a benefit to me, not a liability. If we really really need
I'm not sure that such behavior that is only based on erroneous value, is good one.
perfect load numbers, fine, we have to eat some cycles, but when I look at it, it looks like one of those "Perfect is the enemy of good" things.
The target is not perfect number but good enough to be usable. The systctl_migration_cost threshold is good for idle balancing but can generates wrong load value
But again, why do we care? To be able to mix rt and fair loads and still make pretty mixed load utilization numbers? Paying a general case fast path price to make strange (to me) load utilization numbers pretty is not very attractive.
So I'm not convinced this is a good thing to do, but it's not my call, that's Peter and Ingos job, so having expressed my opinion, I'll shut up and let them do their thing ;-)
-Mike
On 19 April 2013 11:21, Mike Galbraith efault@gmx.de wrote:
On Fri, 2013-04-19 at 10:50 +0200, Vincent Guittot wrote:
On 19 April 2013 10:14, Mike Galbraith efault@gmx.de wrote:
On Fri, 2013-04-19 at 09:49 +0200, Vincent Guittot wrote:
On 19 April 2013 06:30, Mike Galbraith efault@gmx.de wrote:
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
Why do we care what rq's load says, if the only thing running is a periodic RT task? I _think_ I recall that stuff being put under the
cfs scheduler will use a wrong rq load the next time it wants to schedule a task
throttle specifically to not waste cycles doing that on every microscopic idle.
yes but this lead to the wrong computation of runnable_avg_sum. To be more precise, we only need to call __update_entity_runnable_avg, __update_tg_runnable_avg is not mandatory in this step.
If it only scares fair class tasks away from the periodic rt load, that seems like a benefit to me, not a liability. If we really really need
I'm not sure that such behavior that is only based on erroneous value, is good one.
perfect load numbers, fine, we have to eat some cycles, but when I look at it, it looks like one of those "Perfect is the enemy of good" things.
The target is not perfect number but good enough to be usable. The systctl_migration_cost threshold is good for idle balancing but can generates wrong load value
But again, why do we care? To be able to mix rt and fair loads and still make pretty mixed load utilization numbers? Paying a general case
If runnable_avg_sum can be wrong, it becomes unusable and all the stuff around becomes useless.
fast path price to make strange (to me) load utilization numbers pretty is not very attractive. If you muck about with rt classes, you need to have a good reason for doing that. If you do have a good reason, you also allocated all resources, including CPU, so don't need the kernel to
Some tasks have responsiveness constraints so they use rt class but they also live with cfs tasks.
Vincent
balance the load for you. Paying any fast path price to make the kernel balance a mixed rt/fair load just seems fundamentally wrong to me.
-Mike
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
A new idle_exit function is called when the prev task is the idle function so the elapsed time will be accounted as idle time in the rq's load.
Acked-by: Peter Zijlstra a.p.zijlstra@chello.nl
Thanks Vince!
On Thu, 2013-04-18 at 18:34 +0200, Vincent Guittot wrote:
The current update of the rq's load can be erroneous when RT tasks are involved
The update of the load of a rq that becomes idle, is done only if the avg_idle is less than sysctl_sched_migration_cost. If RT tasks and short idle duration alternate, the runnable_avg will not be updated correctly and the time will be accounted as idle time when a CFS task wakes up.
A new idle_enter function is called when the next task is the idle function so the elapsed time will be accounted as run time in the load of the rq, whatever the average idle time is. The function update_rq_runnable_avg is removed from idle_balance.
When a RT task is scheduled on an idle CPU, the update of the rq's load is not done when the rq exit idle state because CFS's functions are not called. Then, the idle_balance, which is called just before entering the idle function, updates the rq's load and makes the assumption that the elapsed time since the last update, was only running time.
As a consequence, the rq's load of a CPU that only runs a periodic RT task, is close to LOAD_AVG_MAX whatever the running duration of the RT task is.
A new idle_exit function is called when the prev task is the idle function so the elapsed time will be accounted as idle time in the rq's load.
Changes since V5:
- Rename idle_enter/exit function to idle_enter/exit_fair
Changes since V4:
- Rebase on v3.9-rc6 instead of Steven Rostedt's patches
Acked-by: Steven Rostedt rostedt@goodmis.org
-- Steve
- Create the post_schedule_idle function that was previously created by Steven's patches
Changes since V3:
- Remove dependancy with CONFIG_FAIR_GROUP_SCHED
- Add a new idle_enter function and create a post_schedule callback for
idle class
- Remove the update_runnable_avg from idle_balance
Changes since V2:
- remove useless definition for UP platform
- rebased on top of Steven Rostedt's patches :
https://lkml.org/lkml/2013/2/12/558
Changes since V1:
- move code out of schedule function and create a pre_schedule callback for idle class instead.
Signed-off-by: Vincent Guittot vincent.guittot@linaro.org
linaro-kernel@lists.linaro.org
-
Mike Galbraith -
Peter Zijlstra -
Steven Rostedt -
Vincent Guittot