From: Ram Chandrasekar <[email protected]>
Let userspace be another voter for cooling device state instead of the
overriding authority. It is possible that the thermal governor may find
a higher cooling state desirable than what is recommended by the
userspace. Separate out the current cooling device state from the
userspace request and aggregate the userspace request with the
governors' recommendation.
Signed-off-by: Lina Iyer <[email protected]>
---
drivers/thermal/thermal_core.c | 1 +
drivers/thermal/thermal_helpers.c | 2 +-
drivers/thermal/thermal_sysfs.c | 8 +++++++-
include/linux/thermal.h | 1 +
4 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
index 589925ac0994..c2e13e122934 100644
--- a/drivers/thermal/thermal_core.c
+++ b/drivers/thermal/thermal_core.c
@@ -970,6 +970,7 @@ __thermal_cooling_device_register(struct device_node *np,
cdev->updated = false;
cdev->device.class = &thermal_class;
cdev->devdata = devdata;
+ cdev->sysfs_req = 0;
thermal_cooling_device_setup_sysfs(cdev);
dev_set_name(&cdev->device, "cooling_device%d", cdev->id);
result = device_register(&cdev->device);
diff --git a/drivers/thermal/thermal_helpers.c b/drivers/thermal/thermal_helpers.c
index 2ba756af76b7..f550fdee0f9b 100644
--- a/drivers/thermal/thermal_helpers.c
+++ b/drivers/thermal/thermal_helpers.c
@@ -166,7 +166,7 @@ EXPORT_SYMBOL_GPL(thermal_zone_set_trips);
void thermal_cdev_update(struct thermal_cooling_device *cdev)
{
struct thermal_instance *instance;
- unsigned long target = 0;
+ unsigned long target = cdev->sysfs_req;
mutex_lock(&cdev->lock);
/* cooling device is updated*/
diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c
index 275ffee292bf..eddada715ad2 100644
--- a/drivers/thermal/thermal_sysfs.c
+++ b/drivers/thermal/thermal_sysfs.c
@@ -719,7 +719,13 @@ thermal_cooling_device_cur_state_store(struct device *dev,
result = cdev->ops->set_cur_state(cdev, state);
if (result)
return result;
- thermal_cooling_device_stats_update(cdev, state);
+
+ cdev->sysfs_req = state;
+ mutex_lock(&cdev->lock);
+ cdev->updated = false;
+ mutex_unlock(&cdev->lock);
+ thermal_cdev_update(cdev);
+
return count;
}
diff --git a/include/linux/thermal.h b/include/linux/thermal.h
index 5f4705f46c2f..7a133bd6ff86 100644
--- a/include/linux/thermal.h
+++ b/include/linux/thermal.h
@@ -139,6 +139,7 @@ struct thermal_cooling_device {
struct mutex lock; /* protect thermal_instances list */
struct list_head thermal_instances;
struct list_head node;
+ unsigned long sysfs_req;
};
struct thermal_attr {
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
On 一, 2018-05-07 at 11:55 -0600, Lina Iyer wrote:
> From: Ram Chandrasekar <[email protected]>
>
> Let userspace be another voter for cooling device state instead of
> the
> overriding authority. It is possible that the thermal governor may
> find
> a higher cooling state desirable than what is recommended by the
> userspace. Separate out the current cooling device state from the
> userspace request and aggregate the userspace request with the
> governors' recommendation.
>
hmmm, I don't understand this.
If the governor does not work well, we should either improve the
governor or use userspace governor instead.
do you have any examples that the kernel governor chooses improper
cooling state?
thanks,
rui
> Signed-off-by: Lina Iyer <[email protected]>
> ---
> drivers/thermal/thermal_core.c | 1 +
> drivers/thermal/thermal_helpers.c | 2 +-
> drivers/thermal/thermal_sysfs.c | 8 +++++++-
> include/linux/thermal.h | 1 +
> 4 files changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/thermal/thermal_core.c
> b/drivers/thermal/thermal_core.c
> index 589925ac0994..c2e13e122934 100644
> --- a/drivers/thermal/thermal_core.c
> +++ b/drivers/thermal/thermal_core.c
> @@ -970,6 +970,7 @@ __thermal_cooling_device_register(struct
> device_node *np,
> cdev->updated = false;
> cdev->device.class = &thermal_class;
> cdev->devdata = devdata;
> + cdev->sysfs_req = 0;
> thermal_cooling_device_setup_sysfs(cdev);
> dev_set_name(&cdev->device, "cooling_device%d", cdev->id);
> result = device_register(&cdev->device);
> diff --git a/drivers/thermal/thermal_helpers.c
> b/drivers/thermal/thermal_helpers.c
> index 2ba756af76b7..f550fdee0f9b 100644
> --- a/drivers/thermal/thermal_helpers.c
> +++ b/drivers/thermal/thermal_helpers.c
> @@ -166,7 +166,7 @@ EXPORT_SYMBOL_GPL(thermal_zone_set_trips);
> void thermal_cdev_update(struct thermal_cooling_device *cdev)
> {
> struct thermal_instance *instance;
> - unsigned long target = 0;
> + unsigned long target = cdev->sysfs_req;
>
> mutex_lock(&cdev->lock);
> /* cooling device is updated*/
> diff --git a/drivers/thermal/thermal_sysfs.c
> b/drivers/thermal/thermal_sysfs.c
> index 275ffee292bf..eddada715ad2 100644
> --- a/drivers/thermal/thermal_sysfs.c
> +++ b/drivers/thermal/thermal_sysfs.c
> @@ -719,7 +719,13 @@ thermal_cooling_device_cur_state_store(struct
> device *dev,
> result = cdev->ops->set_cur_state(cdev, state);
> if (result)
> return result;
> - thermal_cooling_device_stats_update(cdev, state);
> +
> + cdev->sysfs_req = state;
> + mutex_lock(&cdev->lock);
> + cdev->updated = false;
> + mutex_unlock(&cdev->lock);
> + thermal_cdev_update(cdev);
> +
> return count;
> }
>
> diff --git a/include/linux/thermal.h b/include/linux/thermal.h
> index 5f4705f46c2f..7a133bd6ff86 100644
> --- a/include/linux/thermal.h
> +++ b/include/linux/thermal.h
> @@ -139,6 +139,7 @@ struct thermal_cooling_device {
> struct mutex lock; /* protect thermal_instances list */
> struct list_head thermal_instances;
> struct list_head node;
> + unsigned long sysfs_req;
> };
>
> struct thermal_attr {
Adding Ram, so he can respond.
-- Lina
On Thu, Jul 26 2018 at 02:55 -0600, Zhang Rui wrote:
>On 一, 2018-05-07 at 11:55 -0600, Lina Iyer wrote:
>> From: Ram Chandrasekar <[email protected]>
>>
>> Let userspace be another voter for cooling device state instead of
>> the
>> overriding authority. It is possible that the thermal governor may
>> find
>> a higher cooling state desirable than what is recommended by the
>> userspace. Separate out the current cooling device state from the
>> userspace request and aggregate the userspace request with the
>> governors' recommendation.
>>
>
>hmmm, I don't understand this.
>If the governor does not work well, we should either improve the
>governor or use userspace governor instead.
>do you have any examples that the kernel governor chooses improper
>cooling state?
>
>thanks,
>rui
>
>> Signed-off-by: Lina Iyer <[email protected]>
>> ---
>> drivers/thermal/thermal_core.c | 1 +
>> drivers/thermal/thermal_helpers.c | 2 +-
>> drivers/thermal/thermal_sysfs.c | 8 +++++++-
>> include/linux/thermal.h | 1 +
>> 4 files changed, 10 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/thermal/thermal_core.c
>> b/drivers/thermal/thermal_core.c
>> index 589925ac0994..c2e13e122934 100644
>> --- a/drivers/thermal/thermal_core.c
>> +++ b/drivers/thermal/thermal_core.c
>> @@ -970,6 +970,7 @@ __thermal_cooling_device_register(struct
>> device_node *np,
>> cdev->updated = false;
>> cdev->device.class = &thermal_class;
>> cdev->devdata = devdata;
>> + cdev->sysfs_req = 0;
>> thermal_cooling_device_setup_sysfs(cdev);
>> dev_set_name(&cdev->device, "cooling_device%d", cdev->id);
>> result = device_register(&cdev->device);
>> diff --git a/drivers/thermal/thermal_helpers.c
>> b/drivers/thermal/thermal_helpers.c
>> index 2ba756af76b7..f550fdee0f9b 100644
>> --- a/drivers/thermal/thermal_helpers.c
>> +++ b/drivers/thermal/thermal_helpers.c
>> @@ -166,7 +166,7 @@ EXPORT_SYMBOL_GPL(thermal_zone_set_trips);
>> void thermal_cdev_update(struct thermal_cooling_device *cdev)
>> {
>> struct thermal_instance *instance;
>> - unsigned long target = 0;
>> + unsigned long target = cdev->sysfs_req;
>>
>> mutex_lock(&cdev->lock);
>> /* cooling device is updated*/
>> diff --git a/drivers/thermal/thermal_sysfs.c
>> b/drivers/thermal/thermal_sysfs.c
>> index 275ffee292bf..eddada715ad2 100644
>> --- a/drivers/thermal/thermal_sysfs.c
>> +++ b/drivers/thermal/thermal_sysfs.c
>> @@ -719,7 +719,13 @@ thermal_cooling_device_cur_state_store(struct
>> device *dev,
>> result = cdev->ops->set_cur_state(cdev, state);
>> if (result)
>> return result;
>> - thermal_cooling_device_stats_update(cdev, state);
>> +
>> + cdev->sysfs_req = state;
>> + mutex_lock(&cdev->lock);
>> + cdev->updated = false;
>> + mutex_unlock(&cdev->lock);
>> + thermal_cdev_update(cdev);
>> +
>> return count;
>> }
>>
>> diff --git a/include/linux/thermal.h b/include/linux/thermal.h
>> index 5f4705f46c2f..7a133bd6ff86 100644
>> --- a/include/linux/thermal.h
>> +++ b/include/linux/thermal.h
>> @@ -139,6 +139,7 @@ struct thermal_cooling_device {
>> struct mutex lock; /* protect thermal_instances list */
>> struct list_head thermal_instances;
>> struct list_head node;
>> + unsigned long sysfs_req;
>> };
>>
>> struct thermal_attr {
On 8/14/2018 12:04 PM, Lina Iyer wrote:
> Adding Ram, so he can respond.
> -- Lina
>
> On Thu, Jul 26 2018 at 02:55 -0600, Zhang Rui wrote:
>> On 一, 2018-05-07 at 11:55 -0600, Lina Iyer wrote:
>>> From: Ram Chandrasekar <[email protected]>
>>>
>>> Let userspace be another voter for cooling device state instead of
>>> the
>>> overriding authority. It is possible that the thermal governor may
>>> find
>>> a higher cooling state desirable than what is recommended by the
>>> userspace. Separate out the current cooling device state from the
>>> userspace request and aggregate the userspace request with the
>>> governors' recommendation.
>>>
>>
>> hmmm, I don't understand this.
>> If the governor does not work well, we should either improve the
>> governor or use userspace governor instead.
>> do you have any examples that the kernel governor chooses improper
>> cooling state?
>>
Userspace governor and kernel governor could monitor different thermal
zones for different thermal conditions and mitigate the same cooling
device. In this case, the cooling device state request from userspace
governor shouldn’t overwrite the request from kernel governor, rather it
needs to be aggregated. This change is to aggregate the cooling device
state requests between the userspace and kernel governors.
Thanks,
Ram