2020-01-13 01:10:17

by Wanpeng Li

Subject: [PATCH RESEND v2] sched/nohz: Optimize get_nohz_timer_target()

From: Wanpeng Li <[email protected]>

On a machine, cpu 0 is used for housekeeping, the other 39 cpus in the
same socket are in nohz_full mode. We can observe huge time burn in the
loop for seaching nearest busy housekeeper cpu by ftrace.

2) | get_nohz_timer_target() {
2) 0.240 us | housekeeping_test_cpu();
2) 0.458 us | housekeeping_test_cpu();

...

2) 0.292 us | housekeeping_test_cpu();
2) 0.240 us | housekeeping_test_cpu();
2) 0.227 us | housekeeping_any_cpu();
2) + 43.460 us | }

This patch optimizes the searching logic by finding a nearest housekeeper
cpu in the housekeeping cpumask, it can minimize the worst searching time
from ~44us to < 10us in my testing. In addition, the last iterated busy
housekeeper can become a random candidate while current CPU is a better
fallback if it is a housekeeper.

Reviewed-by: Frederic Weisbecker <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Signed-off-by: Wanpeng Li <[email protected]>
---
v1 -> v2:
* current CPU is a better fallback if it is a housekeeper

kernel/sched/core.c | 19 ++++++++++++-------
1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 102dfcf..04a0f6a 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -539,27 +539,32 @@ void resched_cpu(int cpu)
*/
int get_nohz_timer_target(void)
{
- int i, cpu = smp_processor_id();
+ int i, cpu = smp_processor_id(), default_cpu = -1;
struct sched_domain *sd;

- if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
- return cpu;
+ if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) {
+ if (!idle_cpu(cpu))
+ return cpu;
+ default_cpu = cpu;
+ }

rcu_read_lock();
for_each_domain(cpu, sd) {
- for_each_cpu(i, sched_domain_span(sd)) {
+ for_each_cpu_and(i, sched_domain_span(sd),
+ housekeeping_cpumask(HK_FLAG_TIMER)) {
if (cpu == i)
continue;

- if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
+ if (!idle_cpu(i)) {
cpu = i;
goto unlock;
}
}
}

- if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
- cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+ if (default_cpu == -1)
+ default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+ cpu = default_cpu;
unlock:
rcu_read_unlock();
return cpu;
--
1.8.3.1

2020-01-20 06:17:33

by Wanpeng Li

Subject: Re: [PATCH RESEND v2] sched/nohz: Optimize get_nohz_timer_target()

ping Thomas,
On Mon, 13 Jan 2020 at 08:50, Wanpeng Li <[email protected]> wrote:
>
> From: Wanpeng Li <[email protected]>
>
> On a machine, cpu 0 is used for housekeeping, the other 39 cpus in the
> same socket are in nohz_full mode. We can observe huge time burn in the
> loop for seaching nearest busy housekeeper cpu by ftrace.
>
> 2) | get_nohz_timer_target() {
> 2) 0.240 us | housekeeping_test_cpu();
> 2) 0.458 us | housekeeping_test_cpu();
>
> ...
>
> 2) 0.292 us | housekeeping_test_cpu();
> 2) 0.240 us | housekeeping_test_cpu();
> 2) 0.227 us | housekeeping_any_cpu();
> 2) + 43.460 us | }
>
> This patch optimizes the searching logic by finding a nearest housekeeper
> cpu in the housekeeping cpumask, it can minimize the worst searching time
> from ~44us to < 10us in my testing. In addition, the last iterated busy
> housekeeper can become a random candidate while current CPU is a better
> fallback if it is a housekeeper.
>
> Reviewed-by: Frederic Weisbecker <[email protected]>
> Cc: Thomas Gleixner <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Peter Zijlstra <[email protected]>
> Cc: Frederic Weisbecker <[email protected]>
> Signed-off-by: Wanpeng Li <[email protected]>
> ---
> v1 -> v2:
> * current CPU is a better fallback if it is a housekeeper
>
> kernel/sched/core.c | 19 ++++++++++++-------
> 1 file changed, 12 insertions(+), 7 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 102dfcf..04a0f6a 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -539,27 +539,32 @@ void resched_cpu(int cpu)
> */
> int get_nohz_timer_target(void)
> {
> - int i, cpu = smp_processor_id();
> + int i, cpu = smp_processor_id(), default_cpu = -1;
> struct sched_domain *sd;
>
> - if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
> - return cpu;
> + if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) {
> + if (!idle_cpu(cpu))
> + return cpu;
> + default_cpu = cpu;
> + }
>
> rcu_read_lock();
> for_each_domain(cpu, sd) {
> - for_each_cpu(i, sched_domain_span(sd)) {
> + for_each_cpu_and(i, sched_domain_span(sd),
> + housekeeping_cpumask(HK_FLAG_TIMER)) {
> if (cpu == i)
> continue;
>
> - if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
> + if (!idle_cpu(i)) {
> cpu = i;
> goto unlock;
> }
> }
> }
>
> - if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
> - cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
> + if (default_cpu == -1)
> + default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
> + cpu = default_cpu;
> unlock:
> rcu_read_unlock();
> return cpu;
> --
> 1.8.3.1
>

2020-01-29 11:35:36

by tip-bot2 for Alexey Makhalov

Subject: [tip: sched/core] sched/nohz: Optimize get_nohz_timer_target()

The following commit has been merged into the sched/core branch of tip:

Commit-ID: e938b9c94164e4d981039f1cf6007d7453883e5a
Gitweb: https://git.kernel.org/tip/e938b9c94164e4d981039f1cf6007d7453883e5a
Author: Wanpeng Li <[email protected]>
AuthorDate: Mon, 13 Jan 2020 08:50:27 +08:00
Committer: Ingo Molnar <[email protected]>
CommitterDate: Tue, 28 Jan 2020 21:36:57 +01:00

sched/nohz: Optimize get_nohz_timer_target()

On a machine, CPU 0 is used for housekeeping, the other 39 CPUs in the
same socket are in nohz_full mode. We can observe huge time burn in the
loop for seaching nearest busy housekeeper cpu by ftrace.

2) | get_nohz_timer_target() {
2) 0.240 us | housekeeping_test_cpu();
2) 0.458 us | housekeeping_test_cpu();

...

2) 0.292 us | housekeeping_test_cpu();
2) 0.240 us | housekeeping_test_cpu();
2) 0.227 us | housekeeping_any_cpu();
2) + 43.460 us | }

This patch optimizes the searching logic by finding a nearest housekeeper
CPU in the housekeeping cpumask, it can minimize the worst searching time
from ~44us to < 10us in my testing. In addition, the last iterated busy
housekeeper can become a random candidate while current CPU is a better
fallback if it is a housekeeper.

Signed-off-by: Wanpeng Li <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Reviewed-by: Frederic Weisbecker <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
---
kernel/sched/core.c | 19 ++++++++++++-------
1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 55b9a9c..a8a5d5b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -552,27 +552,32 @@ void resched_cpu(int cpu)
*/
int get_nohz_timer_target(void)
{
- int i, cpu = smp_processor_id();
+ int i, cpu = smp_processor_id(), default_cpu = -1;
struct sched_domain *sd;

- if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
- return cpu;
+ if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) {
+ if (!idle_cpu(cpu))
+ return cpu;
+ default_cpu = cpu;
+ }

rcu_read_lock();
for_each_domain(cpu, sd) {
- for_each_cpu(i, sched_domain_span(sd)) {
+ for_each_cpu_and(i, sched_domain_span(sd),
+ housekeeping_cpumask(HK_FLAG_TIMER)) {
if (cpu == i)
continue;

- if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
+ if (!idle_cpu(i)) {
cpu = i;
goto unlock;
}
}
}

- if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
- cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+ if (default_cpu == -1)
+ default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+ cpu = default_cpu;
unlock:
rcu_read_unlock();
return cpu;