ARM: exynos: factor out the idle states

What we have now is (1) cpu0 going always to WFI when cpu1 is up,
(2) cpu0 going to all states when cpu1 is down.

In other words, cpuidle is disabled when cpu1 is up and enabled when cpu1 is down.

This patch use the cpu hotplug notifier to enable/disable cpuidle, when the cpu1 is plugged or unplugged. That clarifies the code and make it simpler.

Signed-off: Daniel Lezcano <[email protected]>

---
 arch/arm/mach-exynos/cpuidle.c |  117 +++++++++++++++++++++++-----------------
 1 file changed, 69 insertions(+), 48 deletions(-)

arch/arm/mach-exynos/cpuidle.c +69 -49

index e6f006b..d8f6f33 100644
@@ -12,6 +12,8 @@
 #include <linux/init.h>
 #include <linux/cpuidle.h>
 #include <linux/cpu_pm.h>
+#include <linux/notifier.h>
+#include <linux/cpu.h>
 #include <linux/io.h>
 #include <linux/export.h>
 #include <linux/time.h>

@@ -36,31 +38,8 @@
 
 #define S5P_CHECK_AFTR         0xFCBA0D10
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index);
-
 static struct cpuidle_device exynos4_cpuidle_device;
 
-static struct cpuidle_driver exynos4_idle_driver = {
-       .name                   = "exynos4_idle",
-       .owner                  = THIS_MODULE,
-       .en_core_tk_irqen       = 1,
-       .states = {
-               [0] = ARM_CPUIDLE_WFI_STATE,
-               [1] = {
-                       .enter            = exynos4_enter_lowpower,
-                       .exit_latency     = 300,
-                       .target_residency = 100000,
-                       .flags            = CPUIDLE_FLAG_TIME_VALID,
-                       .name             = "C1",
-                       .desc             = "ARM power down",
-               },
-       },
-       .state_count = 2,
-       .safe_state_index = 0,
-};
-
 /* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
 static void exynos4_set_wakeupmask(void)
 {

@@ -93,9 +72,9 @@ static int idle_finisher(unsigned long flags)
        return 1;
 }
 
-static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
+static int exynos4_enter_lowpower(struct cpuidle_device *dev,
+                                 struct cpuidle_driver *drv,
+                                 int index)
 {
        unsigned long tmp;
 

@@ -143,22 +122,6 @@ static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
        return index;
 }
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
-{
-       int new_index = index;
-
-       /* This mode only can be entered when other core's are offline */
-       if (num_online_cpus() > 1)
-               new_index = drv->safe_state_index;
-
-       if (new_index == 0)
-               return arm_cpuidle_simple_enter(dev, drv, new_index);
-       else
-               return exynos4_enter_core0_aftr(dev, drv, new_index);
-}
-
 static void __init exynos5_core_down_clk(void)
 {
        unsigned int tmp;

@@ -191,6 +154,62 @@ static void __init exynos5_core_down_clk(void)
        __raw_writel(tmp, EXYNOS5_PWR_CTRL2);
 }
 
+static int cpu_hotplug_notify(struct notifier_block *n,
+                             unsigned long action, void *hcpu)
+{
+       int ret;
+
+       /*
+        * cpu0 can't be shutdown on Origen, so this routine is
+        * called only for cpu1.
+        */
+
+       switch (action & 0xf) {
+
+       case CPU_ONLINE:
+               cpuidle_pause_and_lock();
+               cpuidle_disable_device(&exynos4_cpuidle_device);
+               cpuidle_resume_and_unlock();
+               break;
+
+       case CPU_DEAD:
+               if (!exynos4_cpuidle_device.registered) {
+                       ret = cpuidle_register_device(&exynos4_cpuidle_device);
+                       WARN_ONCE(ret, "Failed to register cpuidle device");
+               } else {
+                       cpuidle_pause_and_lock();
+                       cpuidle_enable_device(&exynos4_cpuidle_device);
+                       cpuidle_resume_and_unlock();
+               }
+               break;
+       }
+
+       return NOTIFY_OK;
+}
+
+static struct notifier_block cpu_hotplug_notifier = {
+       .notifier_call = cpu_hotplug_notify,
+};
+
+static struct cpuidle_driver exynos4_idle_driver = {
+       .name                   = "exynos4_idle",
+       .owner                  = THIS_MODULE,
+       .en_core_tk_irqen       = 1,
+       .states = {
+               [0] = ARM_CPUIDLE_WFI_STATE,
+               [1] = {
+                       .enter            = exynos4_enter_lowpower,
+                       .exit_latency     = 300,
+                       .target_residency = 100000,
+                       .flags            = CPUIDLE_FLAG_TIME_VALID,
+                       .name             = "C1",
+                       .desc             = "ARM power down",
+               },
+       },
+       .state_count = 2,
+       .safe_state_index = 0,
+};
+
 static int __init exynos4_init_cpuidle(void)
 {
        int ret;

@@ -204,12 +223,14 @@ static int __init exynos4_init_cpuidle(void)
                return ret;
        }
 
-       ret = cpuidle_register_device(&exynos4_cpuidle_device);
-       if (ret) {
-               printk(KERN_ERR "CPUidle register device failed\n");
-               return ret;
-       }
+       ret = register_cpu_notifier(&cpu_hotplug_notifier);
+       if (ret)
+               goto out_unregister_driver;
+out:
+       return ret;
 
-       return 0;
+out_unregister_driver:
+       cpuidle_unregister_driver(&exynos4_idle_driver);
+       goto out;
 }
 device_initcall(exynos4_init_cpuidle);
--

1.7.9.5

_______________________________________________
linaro-dev mailing list
linaro-dev at lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-dev

Hi Daniel,

This hotplug noifiers looks fine. I suppose it should add extra state C1 in cpu0. If it is done like below than for normal cases(when all cpu's are online) there wont be any statistics for C0 state

also which
is required. Other patches look good.

Thanks,
Amit

Quoted 184 lines · Daniel Lezcano

What we have now is (1) cpu0 going always to WFI when cpu1 is up,
(2) cpu0 going to all states when cpu1 is down.

In other words, cpuidle is disabled when cpu1 is up and enabled when cpu1 is down.

This patch use the cpu hotplug notifier to enable/disable cpuidle, when the cpu1 is plugged or unplugged. That clarifies the code and make it simpler.

Signed-off: Daniel Lezcano <[email protected]>

---
 arch/arm/mach-exynos/cpuidle.c |  117 +++++++++++++++++++++++-----------------
 1 file changed, 69 insertions(+), 48 deletions(-)

arch/arm/mach-exynos/cpuidle.c +69 -49

index e6f006b..d8f6f33 100644
@@ -12,6 +12,8 @@
 #include <linux/init.h>
 #include <linux/cpuidle.h>
 #include <linux/cpu_pm.h>
+#include <linux/notifier.h>
+#include <linux/cpu.h>
 #include <linux/io.h>
 #include <linux/export.h>
 #include <linux/time.h>

@@ -36,31 +38,8 @@
 
 #define S5P_CHECK_AFTR         0xFCBA0D10
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index);
-
 static struct cpuidle_device exynos4_cpuidle_device;
 
-static struct cpuidle_driver exynos4_idle_driver = {
-       .name                   = "exynos4_idle",
-       .owner                  = THIS_MODULE,
-       .en_core_tk_irqen       = 1,
-       .states = {
-               [0] = ARM_CPUIDLE_WFI_STATE,
-               [1] = {
-                       .enter            = exynos4_enter_lowpower,
-                       .exit_latency     = 300,
-                       .target_residency = 100000,
-                       .flags            = CPUIDLE_FLAG_TIME_VALID,
-                       .name             = "C1",
-                       .desc             = "ARM power down",
-               },
-       },
-       .state_count = 2,
-       .safe_state_index = 0,
-};
-
 /* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
 static void exynos4_set_wakeupmask(void)
 {

@@ -93,9 +72,9 @@ static int idle_finisher(unsigned long flags)
        return 1;
 }
 
-static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
+static int exynos4_enter_lowpower(struct cpuidle_device *dev,
+                                 struct cpuidle_driver *drv,
+                                 int index)
 {
        unsigned long tmp;
 

@@ -143,22 +122,6 @@ static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
        return index;
 }
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
-{
-       int new_index = index;
-
-       /* This mode only can be entered when other core's are offline */
-       if (num_online_cpus() > 1)
-               new_index = drv->safe_state_index;
-
-       if (new_index == 0)
-               return arm_cpuidle_simple_enter(dev, drv, new_index);
-       else
-               return exynos4_enter_core0_aftr(dev, drv, new_index);
-}
-
 static void __init exynos5_core_down_clk(void)
 {
        unsigned int tmp;

@@ -191,6 +154,62 @@ static void __init exynos5_core_down_clk(void)
        __raw_writel(tmp, EXYNOS5_PWR_CTRL2);
 }
 
+static int cpu_hotplug_notify(struct notifier_block *n,
+                             unsigned long action, void *hcpu)
+{
+       int ret;
+
+       /*
+        * cpu0 can't be shutdown on Origen, so this routine is
+        * called only for cpu1.
+        */
+
+       switch (action & 0xf) {
+
+       case CPU_ONLINE:
+               cpuidle_pause_and_lock();
+               cpuidle_disable_device(&exynos4_cpuidle_device);
+               cpuidle_resume_and_unlock();
+               break;
+
+       case CPU_DEAD:
+               if (!exynos4_cpuidle_device.registered) {
+                       ret = cpuidle_register_device(&exynos4_cpuidle_device);
+                       WARN_ONCE(ret, "Failed to register cpuidle device");
+               } else {
+                       cpuidle_pause_and_lock();
+                       cpuidle_enable_device(&exynos4_cpuidle_device);
+                       cpuidle_resume_and_unlock();
+               }
+               break;
+       }
+
+       return NOTIFY_OK;
+}
+
+static struct notifier_block cpu_hotplug_notifier = {
+       .notifier_call = cpu_hotplug_notify,
+};
+
+static struct cpuidle_driver exynos4_idle_driver = {
+       .name                   = "exynos4_idle",
+       .owner                  = THIS_MODULE,
+       .en_core_tk_irqen       = 1,
+       .states = {
+               [0] = ARM_CPUIDLE_WFI_STATE,
+               [1] = {
+                       .enter            = exynos4_enter_lowpower,
+                       .exit_latency     = 300,
+                       .target_residency = 100000,
+                       .flags            = CPUIDLE_FLAG_TIME_VALID,
+                       .name             = "C1",
+                       .desc             = "ARM power down",
+               },
+       },
+       .state_count = 2,
+       .safe_state_index = 0,
+};
+
 static int __init exynos4_init_cpuidle(void)
 {
        int ret;

@@ -204,12 +223,14 @@ static int __init exynos4_init_cpuidle(void)
                return ret;
        }
 
-       ret = cpuidle_register_device(&exynos4_cpuidle_device);
-       if (ret) {
-               printk(KERN_ERR "CPUidle register device failed\n");
-               return ret;
-       }
+       ret = register_cpu_notifier(&cpu_hotplug_notifier);
+       if (ret)
+               goto out_unregister_driver;
+out:
+       return ret;
 
-       return 0;
+out_unregister_driver:
+       cpuidle_unregister_driver(&exynos4_idle_driver);
+       goto out;
 }
 device_initcall(exynos4_init_cpuidle);
--

1.7.9.5

Quoted 4 lines

_______________________________________________
linaro-dev mailing list
linaro-dev at lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-dev

_______________________________________________
linaro-dev mailing list
linaro-dev at lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-dev

Quoted 1 line · amit daniel kachhap

Hi Daniel,

Hi Amit Daniel,

Quoted 3 lines

This hotplug noifiers looks fine. I suppose it should add extra state C1 in cpu0. If it is done like below than for normal cases(when all cpu's are online) there wont be any statistics for C0 state

I guess you meant state 0 which is WFI, right ? C0 state is the intel semantic for cpu fully turned on.

Quoted 3 lines

also which
is required. Other patches look good.

Ok, that makes sense to have statistics even if they are only doing WFI.

Then the patch 4/5 is not ok, no ?

Thanks for review
-- Daniel

Quoted 190 lines

Thanks,
Amit

Quoted 184 lines · Daniel Lezcano

What we have now is (1) cpu0 going always to WFI when cpu1 is up,
(2) cpu0 going to all states when cpu1 is down.

In other words, cpuidle is disabled when cpu1 is up and enabled when cpu1 is down.

This patch use the cpu hotplug notifier to enable/disable cpuidle, when the cpu1 is plugged or unplugged. That clarifies the code and make it simpler.

Signed-off: Daniel Lezcano <[email protected]>

---
 arch/arm/mach-exynos/cpuidle.c |  117 +++++++++++++++++++++++-----------------
 1 file changed, 69 insertions(+), 48 deletions(-)

arch/arm/mach-exynos/cpuidle.c +69 -49

index e6f006b..d8f6f33 100644
@@ -12,6 +12,8 @@
 #include <linux/init.h>
 #include <linux/cpuidle.h>
 #include <linux/cpu_pm.h>
+#include <linux/notifier.h>
+#include <linux/cpu.h>
 #include <linux/io.h>
 #include <linux/export.h>
 #include <linux/time.h>

@@ -36,31 +38,8 @@
 
 #define S5P_CHECK_AFTR         0xFCBA0D10
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index);
-
 static struct cpuidle_device exynos4_cpuidle_device;
 
-static struct cpuidle_driver exynos4_idle_driver = {
-       .name                   = "exynos4_idle",
-       .owner                  = THIS_MODULE,
-       .en_core_tk_irqen       = 1,
-       .states = {
-               [0] = ARM_CPUIDLE_WFI_STATE,
-               [1] = {
-                       .enter            = exynos4_enter_lowpower,
-                       .exit_latency     = 300,
-                       .target_residency = 100000,
-                       .flags            = CPUIDLE_FLAG_TIME_VALID,
-                       .name             = "C1",
-                       .desc             = "ARM power down",
-               },
-       },
-       .state_count = 2,
-       .safe_state_index = 0,
-};
-
 /* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
 static void exynos4_set_wakeupmask(void)
 {

@@ -93,9 +72,9 @@ static int idle_finisher(unsigned long flags)
        return 1;
 }
 
-static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
+static int exynos4_enter_lowpower(struct cpuidle_device *dev,
+                                 struct cpuidle_driver *drv,
+                                 int index)
 {
        unsigned long tmp;
 

@@ -143,22 +122,6 @@ static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
        return index;
 }
 
-static int exynos4_enter_lowpower(struct cpuidle_device *dev,
-                               struct cpuidle_driver *drv,
-                               int index)
-{
-       int new_index = index;
-
-       /* This mode only can be entered when other core's are offline */
-       if (num_online_cpus() > 1)
-               new_index = drv->safe_state_index;
-
-       if (new_index == 0)
-               return arm_cpuidle_simple_enter(dev, drv, new_index);
-       else
-               return exynos4_enter_core0_aftr(dev, drv, new_index);
-}
-
 static void __init exynos5_core_down_clk(void)
 {
        unsigned int tmp;

@@ -191,6 +154,62 @@ static void __init exynos5_core_down_clk(void)
        __raw_writel(tmp, EXYNOS5_PWR_CTRL2);
 }
 
+static int cpu_hotplug_notify(struct notifier_block *n,
+                             unsigned long action, void *hcpu)
+{
+       int ret;
+
+       /*
+        * cpu0 can't be shutdown on Origen, so this routine is
+        * called only for cpu1.
+        */
+
+       switch (action & 0xf) {
+
+       case CPU_ONLINE:
+               cpuidle_pause_and_lock();
+               cpuidle_disable_device(&exynos4_cpuidle_device);
+               cpuidle_resume_and_unlock();
+               break;
+
+       case CPU_DEAD:
+               if (!exynos4_cpuidle_device.registered) {
+                       ret = cpuidle_register_device(&exynos4_cpuidle_device);
+                       WARN_ONCE(ret, "Failed to register cpuidle device");
+               } else {
+                       cpuidle_pause_and_lock();
+                       cpuidle_enable_device(&exynos4_cpuidle_device);
+                       cpuidle_resume_and_unlock();
+               }
+               break;
+       }
+
+       return NOTIFY_OK;
+}
+
+static struct notifier_block cpu_hotplug_notifier = {
+       .notifier_call = cpu_hotplug_notify,
+};
+
+static struct cpuidle_driver exynos4_idle_driver = {
+       .name                   = "exynos4_idle",
+       .owner                  = THIS_MODULE,
+       .en_core_tk_irqen       = 1,
+       .states = {
+               [0] = ARM_CPUIDLE_WFI_STATE,
+               [1] = {
+                       .enter            = exynos4_enter_lowpower,
+                       .exit_latency     = 300,
+                       .target_residency = 100000,
+                       .flags            = CPUIDLE_FLAG_TIME_VALID,
+                       .name             = "C1",
+                       .desc             = "ARM power down",
+               },
+       },
+       .state_count = 2,
+       .safe_state_index = 0,
+};
+
 static int __init exynos4_init_cpuidle(void)
 {
        int ret;

@@ -204,12 +223,14 @@ static int __init exynos4_init_cpuidle(void)
                return ret;
        }
 
-       ret = cpuidle_register_device(&exynos4_cpuidle_device);
-       if (ret) {
-               printk(KERN_ERR "CPUidle register device failed\n");
-               return ret;
-       }
+       ret = register_cpu_notifier(&cpu_hotplug_notifier);
+       if (ret)
+               goto out_unregister_driver;
+out:
+       return ret;
 
-       return 0;
+out_unregister_driver:
+       cpuidle_unregister_driver(&exynos4_idle_driver);
+       goto out;
 }
 device_initcall(exynos4_init_cpuidle);
--

1.7.9.5

Quoted 9 lines

Quoted 4 lines

_______________________________________________
linaro-dev mailing list
linaro-dev at lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-dev

_______________________________________________
linaro-dev mailing list
linaro-dev at lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-dev

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Quoted 11 lines · Daniel Lezcano

Quoted 1 line · amit daniel kachhap

Hi Daniel,

Hi Amit Daniel,

Quoted 3 lines

This hotplug noifiers looks fine. I suppose it should add extra state C1 in cpu0. If it is done like below than for normal cases(when all cpu's are online) there wont be any statistics for C0 state

I guess you meant state 0 which is WFI, right ? C0 state is the intel semantic for cpu fully turned on.

Yes I meant C0 as wfi

Quoted 7 lines

Quoted 3 lines

also which
is required. Other patches look good.

Ok, that makes sense to have statistics even if they are only doing WFI.

Then the patch 4/5 is not ok, no ?

yes I suppose patch 4 and patch 5 are related and depends how you frame patch 5. I think it is better to create C0/C1 sysfs and other things in the beginning because it is a filesystem call and may increase the cpu hotplug time which is not worth. May be if cpuidle framework exposes some API to enable/disable states then it is better.

For patch 1,2 and 3,

Acked-by Amit Daniel Kachhap <[email protected]>

Quoted 1 line · amit daniel kachhap

On Thu, Jan 10, 2013 at 1:32 PM, Daniel Lezcano <[email protected]>

Quoted 25 lines · wrote:

Quoted 11 lines · amit daniel kachhap

Quoted 1 line

Hi Daniel,

Hi Amit Daniel,

Quoted 3 lines

This hotplug noifiers looks fine. I suppose it should add extra state C1 in cpu0. If it is done like below than for normal cases(when all cpu's are online) there wont be any statistics for C0 state

I guess you meant state 0 which is WFI, right ? C0 state is the intel semantic for cpu fully turned on.

Yes I meant C0 as wfi

Quoted 7 lines

Quoted 3 lines

also which
is required. Other patches look good.

Ok, that makes sense to have statistics even if they are only doing WFI.

Then the patch 4/5 is not ok, no ?

yes I suppose patch 4 and patch 5 are related and depends how you frame patch 5. I think it is better to create C0/C1 sysfs and other things in the beginning because it is a filesystem call and may increase the cpu hotplug time which is not worth. May be if cpuidle framework exposes some API to enable/disable states then it is better.

Quoted 2 lines

For patch 1,2 and 3,

Acked-by Amit Daniel Kachhap <[email protected]>

Hi Kukjin,

is it possible to take these patches [1-3/5] ?

The patches [3-4/5] could be ignored.

Quoted 24 lines · Daniel Lezcano

Quoted 1 line · amit daniel kachhap

On Thu, Jan 10, 2013 at 1:32 PM, Daniel Lezcano <[email protected]>

Quoted 21 lines · wrote:

Quoted 9 lines · amit daniel kachhap

Quoted 1 line

Hi Daniel,

Hi Amit Daniel,

Quoted 3 lines

This hotplug noifiers looks fine. I suppose it should add extra state C1 in cpu0. If it is done like below than for normal cases(when all cpu's are online) there wont be any statistics for C0 state

I guess you meant state 0 which is WFI, right ? C0 state is the intel semantic for cpu fully turned on.

Yes I meant C0 as wfi

Quoted 5 lines

Quoted 2 lines

also which
is required. Other patches look good.

Ok, that makes sense to have statistics even if they are only doing WFI.

Then the patch 4/5 is not ok, no ?

yes I suppose patch 4 and patch 5 are related and depends how you frame patch 5. I think it is better to create C0/C1 sysfs and other things in the beginning because it is a filesystem call and may increase the cpu hotplug time which is not worth. May be if cpuidle framework exposes some API to enable/disable states then it is better.

Quoted 6 lines

Quoted 2 lines

For patch 1,2 and 3,

Acked-by Amit Daniel Kachhap <[email protected]>

Hi Kukjin,

is it possible to take these patches [1-3/5] ?

Looks OK to me, I will apply with Amit's ack.

Quoted 2 lines

The patches [3-4/5] could be ignored.

Probably, you mean [4-5/5] :-)