* [RFC PATCH v3 1/2] power/cpuidle: enhance the precision of state select
@ 2016-06-27 7:42 Zhaoyang Huang
2016-06-27 7:42 ` [RFC PATCH v3 2/2] power/idle: enhance the precision of sleep_length Zhaoyang Huang
0 siblings, 1 reply; 2+ messages in thread
From: Zhaoyang Huang @ 2016-06-27 7:42 UTC (permalink / raw)
To: linux-kernel, linux-pm, mingo, peterz, zhaoyang.huang,
private-pmwg, tglx
In previous version, cpu_pm_enter is invoked after the governor
select the state, which cause the executing time of cpu_pm_enter
is included in the idle time. Moving it before the state selection.
Please refer to bellowing chart for detailed information
current approach:
static void cpu_idle_loop(void)
{
while (1) {
tick_nohz_idle_enter();
--->__tick_nohz_idle_enter
--->tick_nohz_stop_sched_tick
{
...
ts->sleep_length = ktime_sub(dev->next_event, now); /*1*/
...
}
while (!need_resched()) {
...
local_irq_disable();
arch_cpu_idle_enter();
--->idle_notifier_call_chain(IDLE_START); /*2*/
cpuidle_idle_call();
--->next_state = cpuidle_select(drv, dev); /*3*/
--->static int arm_enter_idle_state(...)
{
...
ret = cpu_pm_enter(); /*4*/
}
}
---------|----------------------|----------------------------|------------------------------------|--------------------------------------------|--------------------
1.now 2.IDLE_START 3.select idle state 4.CPU_PM_ENTER next_event
(sleep_length)
modified approach:
static void cpu_idle_loop(void)
{
while (1) {
tick_nohz_idle_enter();
--->__tick_nohz_idle_enter
--->tick_nohz_stop_sched_tick
{
...
...
}
while (!need_resched()) {
...
local_irq_disable();
arch_cpu_idle_enter();
--->idle_notifier_call_chain(IDLE_START); /*1*/
cpuidle_idle_call();
--->ret = cpu_pm_enter(); /*2*/
--->tick_nohz_get_sleep_length
{
ts->sleep_length = ktime_sub(dev->next_event, now); /*3*/
}
--->next_state = cpuidle_select(drv, dev); /*4*/
--->static int arm_enter_idle_state(...)
}
----------------|---------------------------------------|---------------------|------------------|--------------------------------------------|--------------------
1.IDLE_START 2.CPU_PM_ENTER 3.now 4.select idle state next_event
(sleep_length)
Signed-off-by: Zhaoyang Huang <zhaoyang.huang@spreadtrum.com>
---
kernel/sched/idle.c | 18 ++++++++++++------
1 file changed, 12 insertions(+), 6 deletions(-)
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index bd12c6c..929da2e 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -5,6 +5,7 @@
#include <linux/cpu.h>
#include <linux/cpuidle.h>
#include <linux/cpuhotplug.h>
+#include <linux/cpu_pm.h>
#include <linux/tick.h>
#include <linux/mm.h>
#include <linux/stackprotector.h>
@@ -130,6 +131,7 @@ static void cpuidle_idle_call(void)
struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
int next_state, entered_state;
+ int ret;
/*
* Check if the idle task must be rescheduled. If it is the
@@ -174,12 +176,16 @@ static void cpuidle_idle_call(void)
/*
* Ask the cpuidle framework to choose a convenient idle state.
*/
- next_state = cpuidle_select(drv, dev);
- entered_state = call_cpuidle(drv, dev, next_state);
- /*
- * Give the governor an opportunity to reflect on the outcome
- */
- cpuidle_reflect(dev, entered_state);
+ ret = cpu_pm_enter();
+ if (!ret) {
+ next_state = cpuidle_select(drv, dev);
+ entered_state = call_cpuidle(drv, dev, next_state);
+ cpu_pm_exit();
+ /*
+ * Give the governor an opportunity to reflect on the outcome
+ */
+ cpuidle_reflect(dev, entered_state);
+ }
}
exit_idle:
--
1.9.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* [RFC PATCH v3 2/2] power/idle: enhance the precision of sleep_length
2016-06-27 7:42 [RFC PATCH v3 1/2] power/cpuidle: enhance the precision of state select Zhaoyang Huang
@ 2016-06-27 7:42 ` Zhaoyang Huang
0 siblings, 0 replies; 2+ messages in thread
From: Zhaoyang Huang @ 2016-06-27 7:42 UTC (permalink / raw)
To: linux-kernel, linux-pm, mingo, peterz, zhaoyang.huang,
private-pmwg, tglx
As shown in bellowing chart, there should be a gap between
tick_nohz_idle_enter(step '1') and tick_nohz_get_sleep_length
(step '3') when idle, which comprise of the execution time of
the code behind step '1' and a series of notify call(step2,4).
Yes, for most of the scenarios, it would NOT be a problem.
However,all time consumed by the steps above will cause the
sleep_length is not very precised in the scenario where the
idle state is shallow(imagin an idle state last less than 500us
under a 200Mhz's P-state). Change it in this series of patch.
Please refer to bellowing chart for detailed information
current approach:
static void cpu_idle_loop(void)
{
while (1) {
tick_nohz_idle_enter();
--->__tick_nohz_idle_enter
--->tick_nohz_stop_sched_tick
{
...
ts->sleep_length = ktime_sub(dev->next_event, now); /*1*/
...
}
while (!need_resched()) {
...
local_irq_disable();
arch_cpu_idle_enter();
--->idle_notifier_call_chain(IDLE_START); /*2*/
cpuidle_idle_call();
--->next_state = cpuidle_select(drv, dev); /*3*/
--->static int arm_enter_idle_state(...)
{
...
ret = cpu_pm_enter(); /*4*/
}
}
---------|----------------------|----------------------------|------------------------------------|--------------------------------------------|--------------------
1.now 2.IDLE_START 3.select idle state 4.CPU_PM_ENTER next_event
(sleep_length)
modified approach:
static void cpu_idle_loop(void)
{
while (1) {
tick_nohz_idle_enter();
--->__tick_nohz_idle_enter
--->tick_nohz_stop_sched_tick
{
...
...
}
while (!need_resched()) {
...
local_irq_disable();
arch_cpu_idle_enter();
--->idle_notifier_call_chain(IDLE_START); /*1*/
cpuidle_idle_call();
--->ret = cpu_pm_enter(); /*2*/
--->tick_nohz_get_sleep_length
{
ts->sleep_length = ktime_sub(dev->next_event, now); /*3*/
}
--->next_state = cpuidle_select(drv, dev); /*4*/
--->static int arm_enter_idle_state(...)
}
----------------|---------------------------------------|---------------------|------------------|--------------------------------------------|--------------------
1.IDLE_START 2.CPU_PM_ENTER 3.now 4.select idle state next_event
(sleep_length)
Signed-off-by: Zhaoyang Huang <zhaoyang.huang@spreadtrum.com>
---
kernel/time/tick-sched.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 536ada8..ee3be3d 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -975,6 +975,11 @@ void tick_nohz_irq_exit(void)
ktime_t tick_nohz_get_sleep_length(void)
{
struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
+ struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
+ ktime_t now;
+
+ now = ktime_get();
+ ts->sleep_length = ktime_sub(dev->next_event, now);
return ts->sleep_length;
}
--
1.9.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
end of thread, other threads:[~2016-06-27 7:42 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-06-27 7:42 [RFC PATCH v3 1/2] power/cpuidle: enhance the precision of state select Zhaoyang Huang
2016-06-27 7:42 ` [RFC PATCH v3 2/2] power/idle: enhance the precision of sleep_length Zhaoyang Huang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).