Re: [PATCH v2] thermal: re-calculate k_po/k_pu when update sustainable power

[Punit Agrawal <punit.agrawal@arm.com>]

Daniel Thompson <daniel.thompson@linaro.org> writes:

> On 05/01/16 16:33, Javi Merino wrote:
>> On Mon, Jan 04, 2016 at 11:22:26AM +0000, Daniel Thompson wrote:
>>> On 01/01/16 00:03, Leo Yan wrote:
>>>> Hi Eduardo,
>>>>
>>>> Thanks for review.
>>>>
>>>> On Thu, Dec 31, 2015 at 09:21:26AM -0800, Eduardo Valentin wrote:
>>>>> On Wed, Dec 30, 2015 at 06:38:40PM +0800, Leo Yan wrote:
>>>>>> k_po/k_pu are in essence ratio values compared with sustainable power.
>>>>>> So when update sustainable power, we can recalculate k_po/k_pu simply
>>>>>> with below formula:
>>>>>>
>>>>>>                 sustainable_power(new)
>>>>>>      k_p(new) = ---------------------- * k_p(old)
>>>>>>                 sustainable_power(old)
>>>>>>
>>>>>> Signed-off-by: Leo Yan <leo.yan@linaro.org>
>>>>>> ---
>>>>>>   drivers/thermal/thermal_core.c | 6 +++++-
>>>>>>   1 file changed, 5 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/thermal/thermal_core.c b/drivers/thermal/thermal_core.c
>>>>>> index d9e525c..223f8df 100644
>>>>>> --- a/drivers/thermal/thermal_core.c
>>>>>> +++ b/drivers/thermal/thermal_core.c
>>>>>> @@ -908,7 +908,7 @@ sustainable_power_store(struct device *dev, struct device_attribute *devattr,
>>>>>>   			const char *buf, size_t count)
>>>>>>   {
>>>>>>   	struct thermal_zone_device *tz = to_thermal_zone(dev);
>>>>>> -	u32 sustainable_power;
>>>>>> +	u32 sustainable_power, old_val;
>>>>>>
>>>>>>   	if (!tz->tzp)
>>>>>>   		return -EIO;
>>>>>> @@ -916,8 +916,12 @@ sustainable_power_store(struct device *dev, struct device_attribute *devattr,
>>>>>>   	if (kstrtou32(buf, 10, &sustainable_power))
>>>>>>   		return -EINVAL;
>>>>>>
>>>>>> +	old_val = tz->tzp->sustainable_power;
>>>>>> +
>>>>>>   	tz->tzp->sustainable_power = sustainable_power;
>>>>>>
>>>>>> +	tz->tzp->k_po = (tz->tzp->k_po * sustainable_power) / old_val;
>>>>>> +	tz->tzp->k_pu = (tz->tzp->k_pu * sustainable_power) / old_val;
>>>>>
>>>>> I believe this has to be done by the governor. These properties are
>>>>> power_allocator specific. thermal_core should not really care about
>>>>> them.
>>>>
>>>> Okay, I will try to update these properties in power_allocator.c.
>>>> Javi, do you think this is fine for you?
>>>
>>> If sustainable power were to change frequently (perhaps on
>>> entry/exit of a fan inhibitor mode?) then we would accumulate
>>> rounding errors here...
>>
>> Can you elaborate on the use case?  What is this fan inhibitor mode?
>
> It's made up.
>
> I was trying to think of use cases which might result in the userspace
> wishing to make frequent changes the maximum sustainable power. For
> designs where the system remains thermally overcommited even with the
> fan running then having a mode where the fan must not spin up would
> require such a change.

If there isn't an immediate use case, we should hold off making any
changes until they're really needed.

[...]