wait_task_inactive() spend too much time on system startup
v2 cover wait_task_inactive() spend too much time on system startup
From: Liang Chen <[email protected]>
Changelog:
v1: wait_task_inactive() frequently call schedule_hrtimeout() and spend a lot of time, i am trying to optimize it on rockchip platform.
v2: Use atomic_flags(PFA) instead of TIF flag, and add some comments.
Liang Chen (1):
sched/fair: do not preempt current task if it is going to call
schedule()
include/linux/sched.h | 5 +++++ kernel/kthread.c | 4 ++++ kernel/sched/fair.c | 13 +++++++++++++ 3 files changed, 22 insertions(+)
Signature · 1 line
2.17.1
PATCH v2 1/1 sched/fair: do not preempt current task if it is going to call schedule()
From: Liang Chen <[email protected]>
when we create a kthread with ktrhead_create_on_cpu(),the child thread entry is ktread.c:ktrhead() which will be preempted by the parent after call complete(done) while schedule() is not called yet,then the parent will call wait_task_inactive(child) but the child is still on the runqueue, so the parent will schedule_hrtimeout() for 1 jiffy,it will waste a lot of time,especially on startup.
parent childktrhead_create_on_cpu()
wait_fo_completion(&done) -----> ktread.c:ktrhead()
|----- complete(done);--wakeup and preempted by parent
kthread_bind() <------------| |-> schedule();--dequeue here wait_task_inactive(child) | schedule_hrtimeout(1 jiffy) -|
So we hope the child just wakeup parent but not preempted by parent, and the child is going to call schedule() soon,then the parent will not call schedule_hrtimeout(1 jiffy) as the child is already dequeue.
The same issue for ktrhead_park()&&kthread_parkme(). This patch can save 120ms on rk312x startup with CONFIG_HZ=300.
--- include/linux/sched.h | 5 +++++ kernel/kthread.c | 4 ++++ kernel/sched/fair.c | 13 +++++++++++++ 3 files changed, 22 insertions(+)
include/linux/sched.h +5 -0
@@ -1533,6 +1533,7 @@ static inline bool is_percpu_thread(void)#define PFA_SPEC_IB_DISABLE 5/* Indirect branch speculation restricted */ #define PFA_SPEC_IB_FORCE_DISABLE 6/* Indirect branch speculation permanently restricted */ #define PFA_SPEC_SSB_NOEXEC 7/* Speculative Store Bypass clear on execve() */ +#define PFA_GOING_TO_SCHED 8/* task is going to call schedule() */ #define TASK_PFA_TEST (name, func) \ static inline bool task_##func(struct task_struct *p) \
@@ -1575,6 +1576,10 @@ TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable) TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable) TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable) +TASK_PFA_TEST(GOING_TO_SCHED, going_to_sched) +TASK_PFA_SET(GOING_TO_SCHED, going_to_sched) +TASK_PFA_CLEAR(GOING_TO_SCHED, going_to_sched) +static inline void current_restore_flags (unsigned long orig_flags,unsigned long flags) {
kernel/kthread.c +4 -0
@@ -199,8 +199,10 @@ static void __kthread_parkme(struct kthread *self)if (!test_bit (KTHREAD_SHOULD_PARK , &self->flags))break ; +task_set_going_to_sched (current);complete (&self->parked);schedule (); +task_clear_going_to_sched (current); }__set_current_state (TASK_RUNNING ); }
@@ -245,8 +247,10 @@ static int kthread(void *_create)/* OK, tell user we're spawned, wait for stop or wakeup */ __set_current_state (TASK_UNINTERRUPTIBLE ); create->result = current; +task_set_going_to_sched (current);complete (done);schedule (); +task_clear_going_to_sched (current); ret = -EINTR ;if (!test_bit (KTHREAD_SHOULD_STOP , &self->flags)) {
kernel/sched/fair.c +13 -0
@@ -4330,6 +4330,12 @@ entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) return;#endif +/* +* current task is going to call schedule(), do not preempt it or +* it will casue more useless contex_switch(). +*/ +if (task_going_to_sched (rq_of (cfs_rq)->curr)) +return ;if (cfs_rq->nr_running >1 )check_preempt_tick (cfs_rq, curr);
@@ -6634,6 +6640,13 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_if (test_tsk_need_resched (curr))return ; +/* +* current task is going to call schedule(), do not preempt it or +* it will casue more useless contex_switch(). +*/ +if (task_going_to_sched (curr)) +return ; +/* Idle tasks are by definition preempted by non-idle tasks. */ if (unlikely (task_has_idle_policy (curr)) &&likely (!task_has_idle_policy (p)))
Signature · 1 line
2.17.1
Re: [PATCH v2 1/1] sched/fair: do not preempt current task if it is going to call schedule()
Quoted 17 lines · On Thu, Mar 05, 2020 at 05:59:48PM +0800, [email protected] wrote:
From: Liang Chen <[email protected]>
when we create a kthread with ktrhead_create_on_cpu(),the child thread entry is ktread.c:ktrhead() which will be preempted by the parent after call complete(done) while
schedule()is not called yet,then the parent will callwait_task_inactive(child)but the child is still on the runqueue, so the parent willschedule_hrtimeout()for 1 jiffy,it will waste a lot of time,especially on startup.parent child
ktrhead_create_on_cpu()
wait_fo_completion(&done)-----> ktread.c:ktrhead()
|----- complete(done);--wakeup and preempted by parentkthread_bind() <------------| |-> schedule();--dequeue here wait_task_inactive(child) | schedule_hrtimeout(1 jiffy) -|
Quoted 8 lines
So we hope the child just wakeup parent but not preempted by parent, and the child is going to call
schedule()soon,then the parent will not callschedule_hrtimeout(1 jiffy)as the child is already dequeue.The same issue for ktrhead_park()&&kthread_parkme(). This patch can save 120ms on rk312x startup with
CONFIG_HZ=300.Signed-off-by Liang Chen <[email protected]>
I'm not familiar with the subtleties of scheduler internals (e.g. is there a race between the end of "schedule();" and calling "task_clear_going_to_sched();" that effects the preemption test logic?), so I'll leave that review to the others. But speaking to the PFA change, it looks sane to me:
-Kees
Quoted 91 lines
--- include/linux/sched.h | 5 +++++ kernel/kthread.c | 4 ++++ kernel/sched/fair.c | 13 +++++++++++++ 3 files changed, 22 insertions(+)include/linux/sched.h +5
-0@@ -1533,6 +1533,7 @@ static inline bool is_percpu_thread(void)#define PFA_SPEC_IB_DISABLE 5/* Indirect branch speculation restricted */ #define PFA_SPEC_IB_FORCE_DISABLE 6/* Indirect branch speculation permanently restricted */ #define PFA_SPEC_SSB_NOEXEC 7/* Speculative Store Bypass clear on execve() */ +#define PFA_GOING_TO_SCHED 8/* task is going to call schedule() */ #define TASK_PFA_TEST (name, func) \ static inline bool task_##func(struct task_struct *p) \@@ -1575,6 +1576,10 @@ TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable) TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable) TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable) +TASK_PFA_TEST(GOING_TO_SCHED, going_to_sched) +TASK_PFA_SET(GOING_TO_SCHED, going_to_sched) +TASK_PFA_CLEAR(GOING_TO_SCHED, going_to_sched) +static inline void current_restore_flags (unsigned long orig_flags,unsigned long flags) {kernel/kthread.c +4
-0@@ -199,8 +199,10 @@ static void __kthread_parkme(struct kthread *self)if (!test_bit (KTHREAD_SHOULD_PARK , &self->flags))break ; +task_set_going_to_sched (current);complete (&self->parked);schedule (); +task_clear_going_to_sched (current); }__set_current_state (TASK_RUNNING ); }@@ -245,8 +247,10 @@ static int kthread(void *_create)/* OK, tell user we're spawned, wait for stop or wakeup */ __set_current_state (TASK_UNINTERRUPTIBLE ); create->result = current; +task_set_going_to_sched (current);complete (done);schedule (); +task_clear_going_to_sched (current); ret = -EINTR ;if (!test_bit (KTHREAD_SHOULD_STOP , &self->flags)) {kernel/sched/fair.c +13
-0@@ -4330,6 +4330,12 @@ entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued) hrtimer_active(&rq_of(cfs_rq)->hrtick_timer)) return;#endif +/* +* current task is going to call schedule(), do not preempt it or +* it will casue more useless contex_switch(). +*/ +if (task_going_to_sched (rq_of (cfs_rq)->curr)) +return ;if (cfs_rq->nr_running >1 )check_preempt_tick (cfs_rq, curr);@@ -6634,6 +6640,13 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_if (test_tsk_need_resched (curr))return ; +/* +* current task is going to call schedule(), do not preempt it or +* it will casue more useless contex_switch(). +*/ +if (task_going_to_sched (curr)) +return ; +/* Idle tasks are by definition preempted by non-idle tasks. */ if (unlikely (task_has_idle_policy (curr)) &&likely (!task_has_idle_policy (p)))Signature · 1 line
2.17.1
Signature · 1 line
Kees Cook