[PATCH v3 0/2] sched/core: Avoid obvious double update_rq_clock warning

[PATCH v3 0/2] sched/core: Avoid obvious double update_rq_clock warning

[Hao Jia]

These two patches are about the kernel scheduler:

patch 1: fixed the issue that kernel may trigger WARN_DOUBLE_CLOCK warning.
patch 2: removed some no longer needed comments in the deadline scheduler
and cleaned up the code.

Thanks for suggestion from Dietmar Eggemann & Peter Zijlstra.

v2->v3:
   - Modify some descriptions and errors in the commit information.
   - Fixed a bug that might trigger the WARN_DOUBLE_CLOCK warning in
     migrate_task_rq_dl().
   - Add Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> for
     patch 2.

v1->v2:
   - Added double_rq_clock_clear_update inline helper to clear RQCF_UPDATED
     of rq->clock_update_flags.
   - split into two separate patches.

[v1] https://lore.kernel.org/lkml/20220418090929.54005-1-jiahao.os@bytedance.com/
[v2] https://lore.kernel.org/lkml/20220422090944.52618-1-jiahao.os@bytedance.com/

Hao Jia (2):
  sched/core: Avoid obvious double update_rq_clock warning
  sched/deadline: Remove superfluous rq clock update in push_dl_task()

 kernel/sched/core.c     |  6 +++---
 kernel/sched/deadline.c | 13 ++++---------
 kernel/sched/rt.c       |  5 +++--
 kernel/sched/sched.h    | 31 +++++++++++++++++++++++++++----
 4 files changed, 37 insertions(+), 18 deletions(-)

-- 
2.32.0

[PATCH v3 1/2] sched/core: Avoid obvious double update_rq_clock warning

[Hao Jia]

When we use raw_spin_rq_lock() to acquire the rq lock and have to
update the rq clock while holding the lock, the kernel may issue
a WARN_DOUBLE_CLOCK warning.

Since we directly use raw_spin_rq_lock() to acquire rq lock instead of
rq_lock(), there is no corresponding change to rq->clock_update_flags.
In particular, we have obtained the rq lock of other cores,
the core rq->clock_update_flags may be RQCF_UPDATED at this time, and
then calling update_rq_clock() will trigger the WARN_DOUBLE_CLOCK warning.

So we need to clear RQCF_UPDATED of rq->clock_update_flags synchronously
to avoid the WARN_DOUBLE_CLOCK warning.

For the sched_rt_period_timer() and migrate_task_rq_dl() cases
we simply replace raw_spin_rq_lock()/raw_spin_rq_unlock() with
rq_lock()/rq_unlock().

For the {pull,push}_{rt,dl}_task() cases, we add the
double_rq_clock_clear_update() function to clear RQCF_UPDATED of
rq->clock_update_flags, and call double_rq_clock_clear_update()
before double_lock_balance()/double_rq_lock() returns to avoid the
WARN_DOUBLE_CLOCK warning.

Some call trace reports:
Call Trace 1:
 <IRQ>
 sched_rt_period_timer+0x10f/0x3a0
 ? enqueue_top_rt_rq+0x110/0x110
 __hrtimer_run_queues+0x1a9/0x490
 hrtimer_interrupt+0x10b/0x240
 __sysvec_apic_timer_interrupt+0x8a/0x250
 sysvec_apic_timer_interrupt+0x9a/0xd0
 </IRQ>
 <TASK>
 asm_sysvec_apic_timer_interrupt+0x12/0x20

Call Trace 2:
 <TASK>
 activate_task+0x8b/0x110
 push_rt_task.part.108+0x241/0x2c0
 push_rt_tasks+0x15/0x30
 finish_task_switch+0xaa/0x2e0
 ? __switch_to+0x134/0x420
 __schedule+0x343/0x8e0
 ? hrtimer_start_range_ns+0x101/0x340
 schedule+0x4e/0xb0
 do_nanosleep+0x8e/0x160
 hrtimer_nanosleep+0x89/0x120
 ? hrtimer_init_sleeper+0x90/0x90
 __x64_sys_nanosleep+0x96/0xd0
 do_syscall_64+0x34/0x90
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Call Trace 3:
 <TASK>
 deactivate_task+0x93/0xe0
 pull_rt_task+0x33e/0x400
 balance_rt+0x7e/0x90
 __schedule+0x62f/0x8e0
 do_task_dead+0x3f/0x50
 do_exit+0x7b8/0xbb0
 do_group_exit+0x2d/0x90
 get_signal+0x9df/0x9e0
 ? preempt_count_add+0x56/0xa0
 ? __remove_hrtimer+0x35/0x70
 arch_do_signal_or_restart+0x36/0x720
 ? nanosleep_copyout+0x39/0x50
 ? do_nanosleep+0x131/0x160
 ? audit_filter_inodes+0xf5/0x120
 exit_to_user_mode_prepare+0x10f/0x1e0
 syscall_exit_to_user_mode+0x17/0x30
 do_syscall_64+0x40/0x90
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Call Trace 4:
 update_rq_clock+0x128/0x1a0
 migrate_task_rq_dl+0xec/0x310
 set_task_cpu+0x84/0x1e4
 try_to_wake_up+0x1d8/0x5c0
 wake_up_process+0x1c/0x30
 hrtimer_wakeup+0x24/0x3c
 __hrtimer_run_queues+0x114/0x270
 hrtimer_interrupt+0xe8/0x244
 arch_timer_handler_phys+0x30/0x50
 handle_percpu_devid_irq+0x88/0x140
 generic_handle_domain_irq+0x40/0x60
 gic_handle_irq+0x48/0xe0
 call_on_irq_stack+0x2c/0x60
 do_interrupt_handler+0x80/0x84

Steps to reproduce:
1. Enable CONFIG_SCHED_DEBUG when compiling the kernel
2. echo 1 > /sys/kernel/debug/clear_warn_once
   echo "WARN_DOUBLE_CLOCK" > /sys/kernel/debug/sched/features
   echo "NO_RT_PUSH_IPI" > /sys/kernel/debug/sched/features
3. Run some rt/dl tasks that periodically work and sleep, e.g.
Create 2*n rt or dl (90% running) tasks via rt-app (on a system
with n CPUs), and Dietmar Eggemann reports Call Trace 4 when running
on PREEMPT_RT kernel.

Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
---
 kernel/sched/core.c     |  6 +++---
 kernel/sched/deadline.c |  5 +++--
 kernel/sched/rt.c       |  5 +++--
 kernel/sched/sched.h    | 31 +++++++++++++++++++++++++++----
 4 files changed, 36 insertions(+), 11 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 51efaabac3e4..84538271b4eb 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -610,10 +610,10 @@ void double_rq_lock(struct rq *rq1, struct rq *rq2)
 		swap(rq1, rq2);
 
 	raw_spin_rq_lock(rq1);
-	if (__rq_lockp(rq1) == __rq_lockp(rq2))
-		return;
+	if (__rq_lockp(rq1) != __rq_lockp(rq2))
+		raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
 
-	raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
+	double_rq_clock_clear_update(rq1, rq2);
 }
 #endif
 
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index fb4255ae0b2c..a562bb271031 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -1833,6 +1833,7 @@ select_task_rq_dl(struct task_struct *p, int cpu, int flags)
 static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused)
 {
 	struct rq *rq;
+	struct rq_flags rf;
 
 	if (READ_ONCE(p->__state) != TASK_WAKING)
 		return;
@@ -1843,7 +1844,7 @@ static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused
 	 * from try_to_wake_up(). Hence, p->pi_lock is locked, but
 	 * rq->lock is not... So, lock it
 	 */
-	raw_spin_rq_lock(rq);
+	rq_lock(rq, &rf);
 	if (p->dl.dl_non_contending) {
 		update_rq_clock(rq);
 		sub_running_bw(&p->dl, &rq->dl);
@@ -1859,7 +1860,7 @@ static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused
 			put_task_struct(p);
 	}
 	sub_rq_bw(&p->dl, &rq->dl);
-	raw_spin_rq_unlock(rq);
+	rq_unlock(rq, &rf);
 }
 
 static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
index a32c46889af8..7891c0f0e1ff 100644
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -871,6 +871,7 @@ static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
 		int enqueue = 0;
 		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
 		struct rq *rq = rq_of_rt_rq(rt_rq);
+		struct rq_flags rf;
 		int skip;
 
 		/*
@@ -885,7 +886,7 @@ static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
 		if (skip)
 			continue;
 
-		raw_spin_rq_lock(rq);
+		rq_lock(rq, &rf);
 		update_rq_clock(rq);
 
 		if (rt_rq->rt_time) {
@@ -923,7 +924,7 @@ static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
 
 		if (enqueue)
 			sched_rt_rq_enqueue(rt_rq);
-		raw_spin_rq_unlock(rq);
+		rq_unlock(rq, &rf);
 	}
 
 	if (!throttled && (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF))
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 8dccb34eb190..975e7d990a0b 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2478,6 +2478,27 @@ unsigned long arch_scale_freq_capacity(int cpu)
 }
 #endif
 
+#ifdef CONFIG_SCHED_DEBUG
+/*
+ * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to acquire
+ * rq lock instead of rq_lock(). So at the end of these two functions we need to
+ * call double_rq_clock_clear_update() synchronously to clear RQCF_UPDATED of
+ * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
+ */
+static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
+{
+	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
+	/*
+	 * If CONFIG_SMP is not defined, rq1 and rq2 are the same,
+	 * so we just clear RQCF_UPDATED one of them.
+	 */
+#ifdef CONFIG_SMP
+	rq2->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
+#endif
+}
+#else
+static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2) {}
+#endif
 
 #ifdef CONFIG_SMP
 
@@ -2543,14 +2564,15 @@ static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
 	__acquires(busiest->lock)
 	__acquires(this_rq->lock)
 {
-	if (__rq_lockp(this_rq) == __rq_lockp(busiest))
-		return 0;
-
-	if (likely(raw_spin_rq_trylock(busiest)))
+	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
+		likely(raw_spin_rq_trylock(busiest))) {
+		double_rq_clock_clear_update(this_rq, busiest);
 		return 0;
+	}
 
 	if (rq_order_less(this_rq, busiest)) {
 		raw_spin_rq_lock_nested(busiest, SINGLE_DEPTH_NESTING);
+		double_rq_clock_clear_update(this_rq, busiest);
 		return 0;
 	}
 
@@ -2644,6 +2666,7 @@ static inline void double_rq_lock(struct rq *rq1, struct rq *rq2)
 	BUG_ON(rq1 != rq2);
 	raw_spin_rq_lock(rq1);
 	__acquire(rq2->lock);	/* Fake it out ;) */
+	double_rq_clock_clear_update(rq1, rq2);
 }
 
 /*
-- 
2.32.0

[PATCH v3 2/2] sched/deadline: Remove superfluous rq clock update in push_dl_task()

[Hao Jia]

The change to call update_rq_clock() before activate_task()
commit 840d719604b0 ("sched/deadline: Update rq_clock of later_rq
when pushing a task") is no longer needed since commit f4904815f97a
("sched/deadline: Fix double accounting of rq/running bw in push & pull")
removed the add_running_bw() before the activate_task().

So we remove some comments that are no longer needed and update
rq clock in activate_task().

Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
---
 kernel/sched/deadline.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index a562bb271031..83636a7fcd32 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -2320,13 +2320,7 @@ static int push_dl_task(struct rq *rq)
 
 	deactivate_task(rq, next_task, 0);
 	set_task_cpu(next_task, later_rq->cpu);
-
-	/*
-	 * Update the later_rq clock here, because the clock is used
-	 * by the cpufreq_update_util() inside __add_running_bw().
-	 */
-	update_rq_clock(later_rq);
-	activate_task(later_rq, next_task, ENQUEUE_NOCLOCK);
+	activate_task(later_rq, next_task, 0);
 	ret = 1;
 
 	resched_curr(later_rq);
-- 
2.32.0

Re: [PATCH v3 1/2] sched/core: Avoid obvious double update_rq_clock warning

[Dietmar Eggemann]

On 27/04/2022 10:00, Hao Jia wrote:

[...]

LGTM, comments are only nit-picks.

> Since we directly use raw_spin_rq_lock() to acquire rq lock instead of
> rq_lock(), there is no corresponding change to rq->clock_update_flags.
> In particular, we have obtained the rq lock of other cores,

s/cores/CPUs, with SMT, a core can have multiple (logical) CPUs.

[...]

> So we need to clear RQCF_UPDATED of rq->clock_update_flags synchronously
> to avoid the WARN_DOUBLE_CLOCK warning.

Why you mention `synchronously` here? Isn't this obvious? (1)

[...]

> @@ -1833,6 +1833,7 @@ select_task_rq_dl(struct task_struct *p, int cpu, int flags)
>  static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused)
>  {
>  	struct rq *rq;
> +	struct rq_flags rf;

-       struct rq *rq;
        struct rq_flags rf;
+       struct rq *rq;

Use reverse fir tree order for variable declarations.
(./Documentation/process/maintainer-tip.rst)

[...]

> +#ifdef CONFIG_SCHED_DEBUG
> +/*
> + * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to acquire
> + * rq lock instead of rq_lock(). So at the end of these two functions we need to
> + * call double_rq_clock_clear_update() synchronously to clear RQCF_UPDATED of
                                          ^^^^^^^^^^^^^
                                          (1)

> + * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
> + */
> +static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
> +{
> +	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
> +	/*
> +	 * If CONFIG_SMP is not defined, rq1 and rq2 are the same,
> +	 * so we just clear RQCF_UPDATED one of them.
> +	 */

Maybe shorter:

      /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */

[...]

> @@ -2543,14 +2564,15 @@ static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
>  	__acquires(busiest->lock)
>  	__acquires(this_rq->lock)
>  {
> -	if (__rq_lockp(this_rq) == __rq_lockp(busiest))
> -		return 0;
> -
> -	if (likely(raw_spin_rq_trylock(busiest)))
> +	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
> +		likely(raw_spin_rq_trylock(busiest))) {

indention:

        if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
-               likely(raw_spin_rq_trylock(busiest))) {
+           likely(raw_spin_rq_trylock(busiest))) {

[...]

Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>

Tested on Arm64 Juno with mainline & preempt-rt (linux-5.17.y-rt).

Re: [PATCH v3 2/2] sched/deadline: Remove superfluous rq clock update in push_dl_task()

[Daniel Bristot de Oliveira]

On 4/27/22 10:00, Hao Jia wrote:
> The change to call update_rq_clock() before activate_task()
> commit 840d719604b0 ("sched/deadline: Update rq_clock of later_rq
> when pushing a task") is no longer needed since commit f4904815f97a
> ("sched/deadline: Fix double accounting of rq/running bw in push & pull")
> removed the add_running_bw() before the activate_task().
> 
> So we remove some comments that are no longer needed and update
> rq clock in activate_task().
> 
> Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>

Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>

-- Daniel

Re: [External] Re: [PATCH v3 1/2] sched/core: Avoid obvious double update_rq_clock warning

[Hao Jia]

On 2022/4/29 Dietmar Eggemann wrote:
> On 27/04/2022 10:00, Hao Jia wrote:
> 
> [...]
> 
> LGTM, comments are only nit-picks.
> 
>> Since we directly use raw_spin_rq_lock() to acquire rq lock instead of
>> rq_lock(), there is no corresponding change to rq->clock_update_flags.
>> In particular, we have obtained the rq lock of other cores,
> 
> s/cores/CPUs, with SMT, a core can have multiple (logical) CPUs.

Thanks for your review comments.
I will do it in patch v4.
Thanks.

> 
> [...]
> 
>> So we need to clear RQCF_UPDATED of rq->clock_update_flags synchronously
>> to avoid the WARN_DOUBLE_CLOCK warning.
> 
> Why you mention `synchronously` here? Isn't this obvious? (1)

I will do it in patch v4.
Thanks.

> 
> [...]
> 
>> @@ -1833,6 +1833,7 @@ select_task_rq_dl(struct task_struct *p, int cpu, int flags)
>>   static void migrate_task_rq_dl(struct task_struct *p, int new_cpu __maybe_unused)
>>   {
>>   	struct rq *rq;
>> +	struct rq_flags rf;
> 
> -       struct rq *rq;
>          struct rq_flags rf;
> +       struct rq *rq;
> 
> Use reverse fir tree order for variable declarations.
> (./Documentation/process/maintainer-tip.rst)

I will do it in patch v4.
Thanks.

> 
> [...]
> 
>> +#ifdef CONFIG_SCHED_DEBUG
>> +/*
>> + * In double_lock_balance()/double_rq_lock(), we use raw_spin_rq_lock() to acquire
>> + * rq lock instead of rq_lock(). So at the end of these two functions we need to
>> + * call double_rq_clock_clear_update() synchronously to clear RQCF_UPDATED of
>                                            ^^^^^^^^^^^^^
>                                            (1)
> 
>> + * rq->clock_update_flags to avoid the WARN_DOUBLE_CLOCK warning.
>> + */
>> +static inline void double_rq_clock_clear_update(struct rq *rq1, struct rq *rq2)
>> +{
>> +	rq1->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
>> +	/*
>> +	 * If CONFIG_SMP is not defined, rq1 and rq2 are the same,
>> +	 * so we just clear RQCF_UPDATED one of them.
>> +	 */
> 
> Maybe shorter:
> 
>        /* rq1 == rq2 for !CONFIG_SMP, so just clear RQCF_UPDATED once. */

I will do it in patch v4.
Thanks.
> 
> [...]
> 
>> @@ -2543,14 +2564,15 @@ static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
>>   	__acquires(busiest->lock)
>>   	__acquires(this_rq->lock)
>>   {
>> -	if (__rq_lockp(this_rq) == __rq_lockp(busiest))
>> -		return 0;
>> -
>> -	if (likely(raw_spin_rq_trylock(busiest)))
>> +	if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
>> +		likely(raw_spin_rq_trylock(busiest))) {
> 
> indention:
> 
>          if (__rq_lockp(this_rq) == __rq_lockp(busiest) ||
> -               likely(raw_spin_rq_trylock(busiest))) {
> +           likely(raw_spin_rq_trylock(busiest))) {

Thanks again for your review and suggestion.
I will do it in patch v4.
Thanks.

> 
> [...]
> 
> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
> 
> Tested on Arm64 Juno with mainline & preempt-rt (linux-5.17.y-rt).