W dniu 14 września 2009 00:17 użytkownik Rafał Miłecki
<zajec5@gmail.com> napisał:
> W dniu 11 września 2009 20:41 użytkownik Rafał Miłecki
> <zajec5@gmail.com> napisał:
>> This patchset implements features needed to implement real power management.
>>
>> Last one (0005) even shows that it's possible to implement
>> downclocking on DPMS_OFF with vious patches. It's just prove of
>> working infrastructure, nothing acceptable to commit. However I hope
>> it makes you decide commiting 0001-0004.
>>
>> Please review/test/commit. If you decide to apply last patch, just use:
>> xset dpms force off
>
> I was thinking over weekend about possible bad decisions made here
> that will hit us later.
>
> It's quite possible we won't use only limited amount of power modes
> but quite big amount of dynamically generated modes between "minimum"
> and "maximum" (with some step for clocks and voltage). Anyway even
> then we will need to keep "minimum" and "maximum" (and "default"?).
> And keeping that in 1 struct properties will be pain:
> radeon_pm->minimum_engine_clock
> radeon_pm->minimum_memory_clock
> radeon_pm->minimum_voltage
> *3
Probably not *3, but *4. I think we will also need sth like "user mode".
> So I believe introducing radeom_pm_state is still a good idea.
> Eventually we will modify state setting function, but probably nothing
> more.
Maybe one bad decision is using last_set introduced in 4th patch.
For very dynamic power management we may want have "dynamic_mode"
radeon_pm_state. We will just modify it's properties (ex. clock+=100)
and then call setting function again. In that case last_set will point
to "dynamic_mode" and we will use "dynamic_mode" again. That would
cause not changing clocks, even when something in "dynamic_mode"
changed.
So I'm still for applying 0001-0003.
--
Rafał
|
