Patchwork [V2,3/3] ARM: tegra114: cpuidle: add powered-down state

login
register
mail settings
Submitter Joseph Lo
Date June 4, 2013, 10:48 a.m.
Message ID <1370342880-422-4-git-send-email-josephl@nvidia.com>
Download mbox | patch
Permalink /patch/248531/
State Superseded, archived
Headers show

Comments

Joseph Lo - June 4, 2013, 10:48 a.m.
This supports CPU core power down on each CPU when CPU idle. When CPU go
into this state, it saves it's context and needs a proper configuration
in flow controller to power gate the CPU when CPU runs into WFI
instruction. And the CPU also needs to set the IRQ as CPU power down idle
wake up event in flow controller.

Signed-off-by: Joseph Lo <josephl@nvidia.com>
---
V2:
* remove some redundant code of memory barrier
* remove the function declaration by rearranging the coding sequence 
---
 arch/arm/mach-tegra/cpuidle-tegra114.c | 51 +++++++++++++++++++++++++++++++++-
 1 file changed, 50 insertions(+), 1 deletion(-)
Daniel Lezcano - June 4, 2013, 9:40 p.m.
On 06/04/2013 12:48 PM, Joseph Lo wrote:
> This supports CPU core power down on each CPU when CPU idle. When CPU go
> into this state, it saves it's context and needs a proper configuration
> in flow controller to power gate the CPU when CPU runs into WFI
> instruction. And the CPU also needs to set the IRQ as CPU power down idle
> wake up event in flow controller.
> 
> Signed-off-by: Joseph Lo <josephl@nvidia.com>

I would like to understand why there is a WARN with the
CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
is not the tree hiding the forest.

Thanks
  -- Daniel

> ---
> V2:
> * remove some redundant code of memory barrier
> * remove the function declaration by rearranging the coding sequence 
> ---
>  arch/arm/mach-tegra/cpuidle-tegra114.c | 51 +++++++++++++++++++++++++++++++++-
>  1 file changed, 50 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm/mach-tegra/cpuidle-tegra114.c b/arch/arm/mach-tegra/cpuidle-tegra114.c
> index 1d1c602..e0b8730 100644
> --- a/arch/arm/mach-tegra/cpuidle-tegra114.c
> +++ b/arch/arm/mach-tegra/cpuidle-tegra114.c
> @@ -17,15 +17,64 @@
>  #include <linux/kernel.h>
>  #include <linux/module.h>
>  #include <linux/cpuidle.h>
> +#include <linux/cpu_pm.h>
> +#include <linux/clockchips.h>
>  
>  #include <asm/cpuidle.h>
> +#include <asm/suspend.h>
> +#include <asm/smp_plat.h>
> +
> +#include "pm.h"
> +#include "sleep.h"
> +
> +#ifdef CONFIG_PM_SLEEP
> +#define TEGRA114_MAX_STATES 2
> +#else
> +#define TEGRA114_MAX_STATES 1
> +#endif
> +
> +#ifdef CONFIG_PM_SLEEP
> +static int tegra114_idle_power_down(struct cpuidle_device *dev,
> +				    struct cpuidle_driver *drv,
> +				    int index)
> +{
> +	local_fiq_disable();
> +
> +	tegra_set_cpu_in_lp2();
> +	cpu_pm_enter();
> +
> +	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu);
> +
> +	cpu_suspend(0, tegra30_sleep_cpu_secondary_finish);
> +
> +	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
> +
> +	cpu_pm_exit();
> +	tegra_clear_cpu_in_lp2();
> +
> +	local_fiq_enable();
> +
> +	return index;
> +}
> +#endif
>  
>  static struct cpuidle_driver tegra_idle_driver = {
>  	.name = "tegra_idle",
>  	.owner = THIS_MODULE,
> -	.state_count = 1,
> +	.state_count = TEGRA114_MAX_STATES,
>  	.states = {
>  		[0] = ARM_CPUIDLE_WFI_STATE_PWR(600),
> +#ifdef CONFIG_PM_SLEEP
> +		[1] = {
> +			.enter			= tegra114_idle_power_down,
> +			.exit_latency		= 500,
> +			.target_residency	= 1000,
> +			.power_usage		= 0,
> +			.flags			= CPUIDLE_FLAG_TIME_VALID,
> +			.name			= "powered-down",
> +			.desc			= "CPU power gated",
> +		},
> +#endif
>  	},
>  };
>  
>
Stephen Warren - June 12, 2013, 5:41 p.m.
On 06/04/2013 03:40 PM, Daniel Lezcano wrote:
> On 06/04/2013 12:48 PM, Joseph Lo wrote:
>> This supports CPU core power down on each CPU when CPU idle. When CPU go
>> into this state, it saves it's context and needs a proper configuration
>> in flow controller to power gate the CPU when CPU runs into WFI
>> instruction. And the CPU also needs to set the IRQ as CPU power down idle
>> wake up event in flow controller.
>>
>> Signed-off-by: Joseph Lo <josephl@nvidia.com>
> 
> I would like to understand why there is a WARN with the
> CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
> is not the tree hiding the forest.

Joseph, are you planning to post an updated series or respond to resolve
Daniel's question?
--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Joseph Lo - June 13, 2013, 2:13 a.m.
On Thu, 2013-06-13 at 01:41 +0800, Stephen Warren wrote:
> On 06/04/2013 03:40 PM, Daniel Lezcano wrote:
> > On 06/04/2013 12:48 PM, Joseph Lo wrote:
> >> This supports CPU core power down on each CPU when CPU idle. When CPU go
> >> into this state, it saves it's context and needs a proper configuration
> >> in flow controller to power gate the CPU when CPU runs into WFI
> >> instruction. And the CPU also needs to set the IRQ as CPU power down idle
> >> wake up event in flow controller.
> >>
> >> Signed-off-by: Joseph Lo <josephl@nvidia.com>
> > 
> > I would like to understand why there is a WARN with the
> > CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
> > is not the tree hiding the forest.
> 
> Joseph, are you planning to post an updated series or respond to resolve
> Daniel's question?

I need more time to investigate the detail about what caused the WARN
when apply the flag. And what's the difference if we didn't enable
CONFIG_CPU_IDLE_MULTIPLE_DRIVERS, then it only applies
CLOCK_EVT_NOTIFY_BROADCAST_ON on CPU0. Why if I am moving the
clockevent_notify of CLOCK_EVT_NOTIFY_BROADCAST_EXIT before
"local_irq_enable" in the
"cpuidle_enter_state" (drivers/cpuidle/cpuidle.c), then the warning
message gone?

Daniel,

Because I need more time to figure it out, are you OK if I can post
another patch to apply CPUIDLE_FLAG_TIMER_STOP flag for all Tegra CPU
Idle drivers later? (Once we know what's the problem behind this)

Stephen,

I don't have an update of this series, if Daniel is OK if I can post
another patch to apply TIMER_STOP flag for all Tegra idle drivers.

Thanks,
Joseph

--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Daniel Lezcano - June 13, 2013, 9:30 a.m.
On 06/13/2013 04:13 AM, Joseph Lo wrote:
> On Thu, 2013-06-13 at 01:41 +0800, Stephen Warren wrote:
>> On 06/04/2013 03:40 PM, Daniel Lezcano wrote:
>>> On 06/04/2013 12:48 PM, Joseph Lo wrote:
>>>> This supports CPU core power down on each CPU when CPU idle. When CPU go
>>>> into this state, it saves it's context and needs a proper configuration
>>>> in flow controller to power gate the CPU when CPU runs into WFI
>>>> instruction. And the CPU also needs to set the IRQ as CPU power down idle
>>>> wake up event in flow controller.
>>>>
>>>> Signed-off-by: Joseph Lo <josephl@nvidia.com>
>>>
>>> I would like to understand why there is a WARN with the
>>> CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
>>> is not the tree hiding the forest.
>>
>> Joseph, are you planning to post an updated series or respond to resolve
>> Daniel's question?
> 
> I need more time to investigate the detail about what caused the WARN
> when apply the flag. And what's the difference if we didn't enable
> CONFIG_CPU_IDLE_MULTIPLE_DRIVERS, then it only applies
> CLOCK_EVT_NOTIFY_BROADCAST_ON on CPU0.

I think I got it for this one. It is a bug in the cpuidle driver's code.

cpuidle_register_driver does get_cpu => CPU0

Then __cpuidle_register_driver(drv, <CPU0>)

 => __cpuidle_driver_init(drv, <CPU0>)

Hence, the timer broadcast is initialized only for cpu0.

That should have been fixed with:

commit 82467a5a885ddd9f80309682159da8db510e7832
Author: Daniel Lezcano <daniel.lezcano@linaro.org>
Date:   Fri Jun 7 21:53:09 2013 +0000

    cpuidle: simplify multiple driver support

    Commit bf4d1b5 (cpuidle: support multiple drivers) introduced support
    for using multiple cpuidle drivers at the same time.  It added a
    couple of new APIs to register the driver per CPU, but that led to
    some unnecessary code complexity related to the kernel config options
    deciding whether or not the multiple driver support is enabled.  The
    code has to work as it did before when the multiple driver support is
    not enabled and the multiple driver support has to be compatible with
    the previously existing API.

    Remove the new API, not used by any driver in the tree yet (but
    needed for the HMP cpuidle drivers that will be submitted soon), and
    add a new cpumask pointer to the cpuidle driver structure that will
    point to the mask of CPUs handled by the given driver.  That will
    allow the cpuidle_[un]register_driver() API to be used for the
    multiple driver support along with the cpuidle_[un]register()
    functions added recently.

    [rjw: Changelog]
    Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
    Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>


> Why if I am moving the
> clockevent_notify of CLOCK_EVT_NOTIFY_BROADCAST_EXIT before
> "local_irq_enable" in the
> "cpuidle_enter_state" (drivers/cpuidle/cpuidle.c), then the warning
> message gone?

Is the warning coming immediately, at the first time
CLOCK_EVT_NOTIFY_BROADCAST_EXIT is invoked or can it occur at different
moment ?

> Daniel,
> 
> Because I need more time to figure it out, are you OK if I can post
> another patch to apply CPUIDLE_FLAG_TIMER_STOP flag for all Tegra CPU
> Idle drivers later? (Once we know what's the problem behind this)
> 
> Stephen,
> 
> I don't have an update of this series, if Daniel is OK if I can post
> another patch to apply TIMER_STOP flag for all Tegra idle drivers.
> 
> Thanks,
> Joseph
>
Joseph Lo - June 13, 2013, 2:32 p.m.
On Thu, 2013-06-13 at 17:30 +0800, Daniel Lezcano wrote:
> On 06/13/2013 04:13 AM, Joseph Lo wrote:
> > On Thu, 2013-06-13 at 01:41 +0800, Stephen Warren wrote:
> >> On 06/04/2013 03:40 PM, Daniel Lezcano wrote:
> >>> On 06/04/2013 12:48 PM, Joseph Lo wrote:
> >>>> This supports CPU core power down on each CPU when CPU idle. When CPU go
> >>>> into this state, it saves it's context and needs a proper configuration
> >>>> in flow controller to power gate the CPU when CPU runs into WFI
> >>>> instruction. And the CPU also needs to set the IRQ as CPU power down idle
> >>>> wake up event in flow controller.
> >>>>
> >>>> Signed-off-by: Joseph Lo <josephl@nvidia.com>
> >>>
> >>> I would like to understand why there is a WARN with the
> >>> CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
> >>> is not the tree hiding the forest.
> >>
> >> Joseph, are you planning to post an updated series or respond to resolve
> >> Daniel's question?
> > 
> > I need more time to investigate the detail about what caused the WARN
> > when apply the flag. And what's the difference if we didn't enable
> > CONFIG_CPU_IDLE_MULTIPLE_DRIVERS, then it only applies
> > CLOCK_EVT_NOTIFY_BROADCAST_ON on CPU0.
> 
> I think I got it for this one. It is a bug in the cpuidle driver's code.
> 
> cpuidle_register_driver does get_cpu => CPU0
> 
> Then __cpuidle_register_driver(drv, <CPU0>)
> 
>  => __cpuidle_driver_init(drv, <CPU0>)
> 
> Hence, the timer broadcast is initialized only for cpu0.
> 
> That should have been fixed with:
> 
OK, thanks.

> commit 82467a5a885ddd9f80309682159da8db510e7832
> Author: Daniel Lezcano <daniel.lezcano@linaro.org>
> Date:   Fri Jun 7 21:53:09 2013 +0000
> 
>     cpuidle: simplify multiple driver support
> 
>     Commit bf4d1b5 (cpuidle: support multiple drivers) introduced support
>     for using multiple cpuidle drivers at the same time.  It added a
>     couple of new APIs to register the driver per CPU, but that led to
>     some unnecessary code complexity related to the kernel config options
>     deciding whether or not the multiple driver support is enabled.  The
>     code has to work as it did before when the multiple driver support is
>     not enabled and the multiple driver support has to be compatible with
>     the previously existing API.
> 
>     Remove the new API, not used by any driver in the tree yet (but
>     needed for the HMP cpuidle drivers that will be submitted soon), and
>     add a new cpumask pointer to the cpuidle driver structure that will
>     point to the mask of CPUs handled by the given driver.  That will
>     allow the cpuidle_[un]register_driver() API to be used for the
>     multiple driver support along with the cpuidle_[un]register()
>     functions added recently.
> 
>     [rjw: Changelog]
>     Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
>     Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> 
> > Why if I am moving the
> > clockevent_notify of CLOCK_EVT_NOTIFY_BROADCAST_EXIT before
> > "local_irq_enable" in the
> > "cpuidle_enter_state" (drivers/cpuidle/cpuidle.c), then the warning
> > message gone?
> 
> Is the warning coming immediately, at the first time
> CLOCK_EVT_NOTIFY_BROADCAST_EXIT is invoked or can it occur at different
> moment ?
> 
Not exactly at the 1st time, it may happen in the 3rd to 10th time. But
it only happens once. I mean I did check the
tick_broadcast_pending_mask, it only happens once the CPU didn't clear
the tick_broadcast_pending_mask after CLOCK_EVT_NOTIFY_BROADCAST_EXIT.
And cause the warning message. Then I didn't see it happen anymore.


--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Daniel Lezcano - June 13, 2013, 3:02 p.m.
On 06/13/2013 04:32 PM, Joseph Lo wrote:
> On Thu, 2013-06-13 at 17:30 +0800, Daniel Lezcano wrote:
>> On 06/13/2013 04:13 AM, Joseph Lo wrote:
>>> On Thu, 2013-06-13 at 01:41 +0800, Stephen Warren wrote:
>>>> On 06/04/2013 03:40 PM, Daniel Lezcano wrote:
>>>>> On 06/04/2013 12:48 PM, Joseph Lo wrote:
>>>>>> This supports CPU core power down on each CPU when CPU idle. When CPU go
>>>>>> into this state, it saves it's context and needs a proper configuration
>>>>>> in flow controller to power gate the CPU when CPU runs into WFI
>>>>>> instruction. And the CPU also needs to set the IRQ as CPU power down idle
>>>>>> wake up event in flow controller.
>>>>>>
>>>>>> Signed-off-by: Joseph Lo <josephl@nvidia.com>
>>>>>
>>>>> I would like to understand why there is a WARN with the
>>>>> CPUIDLE_FLAG_TIMER_STOP flag set before queuing this patch and ensure it
>>>>> is not the tree hiding the forest.
>>>>
>>>> Joseph, are you planning to post an updated series or respond to resolve
>>>> Daniel's question?
>>>
>>> I need more time to investigate the detail about what caused the WARN
>>> when apply the flag. And what's the difference if we didn't enable
>>> CONFIG_CPU_IDLE_MULTIPLE_DRIVERS, then it only applies
>>> CLOCK_EVT_NOTIFY_BROADCAST_ON on CPU0.
>>
>> I think I got it for this one. It is a bug in the cpuidle driver's code.
>>
>> cpuidle_register_driver does get_cpu => CPU0
>>
>> Then __cpuidle_register_driver(drv, <CPU0>)
>>
>>  => __cpuidle_driver_init(drv, <CPU0>)
>>
>> Hence, the timer broadcast is initialized only for cpu0.
>>
>> That should have been fixed with:
>>
> OK, thanks.
> 
>> commit 82467a5a885ddd9f80309682159da8db510e7832
>> Author: Daniel Lezcano <daniel.lezcano@linaro.org>
>> Date:   Fri Jun 7 21:53:09 2013 +0000
>>
>>     cpuidle: simplify multiple driver support
>>
>>     Commit bf4d1b5 (cpuidle: support multiple drivers) introduced support
>>     for using multiple cpuidle drivers at the same time.  It added a
>>     couple of new APIs to register the driver per CPU, but that led to
>>     some unnecessary code complexity related to the kernel config options
>>     deciding whether or not the multiple driver support is enabled.  The
>>     code has to work as it did before when the multiple driver support is
>>     not enabled and the multiple driver support has to be compatible with
>>     the previously existing API.
>>
>>     Remove the new API, not used by any driver in the tree yet (but
>>     needed for the HMP cpuidle drivers that will be submitted soon), and
>>     add a new cpumask pointer to the cpuidle driver structure that will
>>     point to the mask of CPUs handled by the given driver.  That will
>>     allow the cpuidle_[un]register_driver() API to be used for the
>>     multiple driver support along with the cpuidle_[un]register()
>>     functions added recently.
>>
>>     [rjw: Changelog]
>>     Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
>>     Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>>
>>
>>> Why if I am moving the
>>> clockevent_notify of CLOCK_EVT_NOTIFY_BROADCAST_EXIT before
>>> "local_irq_enable" in the
>>> "cpuidle_enter_state" (drivers/cpuidle/cpuidle.c), then the warning
>>> message gone?
>>
>> Is the warning coming immediately, at the first time
>> CLOCK_EVT_NOTIFY_BROADCAST_EXIT is invoked or can it occur at different
>> moment ?
>>
> Not exactly at the 1st time, it may happen in the 3rd to 10th time. But
> it only happens once. I mean I did check the
> tick_broadcast_pending_mask, it only happens once the CPU didn't clear
> the tick_broadcast_pending_mask after CLOCK_EVT_NOTIFY_BROADCAST_EXIT.
> And cause the warning message. Then I didn't see it happen anymore.

I suspect when the CLOCK_EVENT_NOTIFY_EXIT is after the
local_irq_enable, we have the timer handler executed before where it
sets the tick_broadcast_pending_mask.

Hence in the tick_broadcast_oneshot_control function when we fall into
the condition:

 if (dev->next_event.tv64 == KTIME_MAX)
	goto out;

The tick_broadcast_pending_mask is not cleared and when re-entering with
CLOCK_EVENT_NOTIFY_ENTER, this mask is set and that raises the warning.

So, if I am not wrong, the condition to raise this warning would be:

 1. use the flag (of course), because the
CLOCK_EVT_NOTIFY_BROADCAST_EXIT is called after the local_irq_enabled

 2. there is no timer planned after the expired one.

 3. a new timer is created and the cpu enters the low power state again

The first step would be to create a simple program to reproduce easily
the warning.

First boot the kernel with init=/bin/bash (or whatever, the init process
is replaced by a shell), thus reducing considerably the number of timers.

Then try to raise the warning with cpu2:

	taskset 4 /bin/sleep 0.5

... several times until the warning appears.

As I don't have the hardware, this is based on assumptions, so may be I
am wrong.

Patch

diff --git a/arch/arm/mach-tegra/cpuidle-tegra114.c b/arch/arm/mach-tegra/cpuidle-tegra114.c
index 1d1c602..e0b8730 100644
--- a/arch/arm/mach-tegra/cpuidle-tegra114.c
+++ b/arch/arm/mach-tegra/cpuidle-tegra114.c
@@ -17,15 +17,64 @@ 
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/cpuidle.h>
+#include <linux/cpu_pm.h>
+#include <linux/clockchips.h>
 
 #include <asm/cpuidle.h>
+#include <asm/suspend.h>
+#include <asm/smp_plat.h>
+
+#include "pm.h"
+#include "sleep.h"
+
+#ifdef CONFIG_PM_SLEEP
+#define TEGRA114_MAX_STATES 2
+#else
+#define TEGRA114_MAX_STATES 1
+#endif
+
+#ifdef CONFIG_PM_SLEEP
+static int tegra114_idle_power_down(struct cpuidle_device *dev,
+				    struct cpuidle_driver *drv,
+				    int index)
+{
+	local_fiq_disable();
+
+	tegra_set_cpu_in_lp2();
+	cpu_pm_enter();
+
+	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu);
+
+	cpu_suspend(0, tegra30_sleep_cpu_secondary_finish);
+
+	clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu);
+
+	cpu_pm_exit();
+	tegra_clear_cpu_in_lp2();
+
+	local_fiq_enable();
+
+	return index;
+}
+#endif
 
 static struct cpuidle_driver tegra_idle_driver = {
 	.name = "tegra_idle",
 	.owner = THIS_MODULE,
-	.state_count = 1,
+	.state_count = TEGRA114_MAX_STATES,
 	.states = {
 		[0] = ARM_CPUIDLE_WFI_STATE_PWR(600),
+#ifdef CONFIG_PM_SLEEP
+		[1] = {
+			.enter			= tegra114_idle_power_down,
+			.exit_latency		= 500,
+			.target_residency	= 1000,
+			.power_usage		= 0,
+			.flags			= CPUIDLE_FLAG_TIME_VALID,
+			.name			= "powered-down",
+			.desc			= "CPU power gated",
+		},
+#endif
 	},
 };