[RFC PATCH v2 1/2] cpuidle: Add common init interface and idle functionality

[Turquette, Mike]

On Wed, Jan 4, 2012 at 3:15 PM, Rob Lee <rob.lee at linaro.org> wrote:
> On 22 December 2011 16:57, Rob Herring <robherring2 at gmail.com> wrote:
>> On 12/14/2011 01:02 AM, Robert Lee wrote:
>>> +static struct cpuidle_driver exynos4_idle_driver = {
>>> +     .name           = "exynos4_idle",
>>> +     .owner          = THIS_MODULE,
>>> +     .states[0] = {
>>> +             .enter                  = cpuidle_def_idle,
>>>               .exit_latency           = 1,
>>>               .target_residency       = 100000,
>>>               .flags                  = CPUIDLE_FLAG_TIME_VALID,
>>>               .name                   = "IDLE",
>>>               .desc                   = "ARM clock gating(WFI)",
>>>       },
>>
>> As this is just plain wfi and shouldn't really be different per
>> platform, it would be nice to get rid of all of this state info. Perhaps
>> a macro with all the data since each driver needs to init each state.
>> The target residency value looks kind of suspect. You should only go to
>> wfi if you expect to be idle for 100ms?
>>
>
> Ok, I'll look at add a ARM specific macro for a generic WFI state in v3.
>
> For the target_residency value, I don't understand why the values
> being used are there other than some of the original ARM platform
> implementations were based on the Intel implementations per their
> comments, and the Intel value was used.  From the comments in
> drivers/cpuidle/governors/menu.c: "state entry and exit have an energy
> cost, and a certain amount of time in the  C state is required to
> actually break even on this cost. CPUIDLE provides us this duration in
> the target_residency field".  The governor code uses it accordingly.
> I'm not aware that a pure WFI only state uses additional energy to
> enter and exit compared to spinning in a loop, so I think the
> "target_residency" of a pure WFI state value should 0 or 1.  If a
> platform skips a pure WFI idle state and also implements additional
> platform specific power savings in their "WFI" idle state, then they
> may need to adjust exit_latency and target_residency accordingly.

Extra data point: on OMAP4 we chose .target_residency = 5 for simple WFI state:
http://git.omapzoom.org/?p=kernel/omap.git;a=blob;f=arch/arm/mach-omap2/cpuidle44xx.c;h=383944076085f808b8764baf11df0589229010e5;hb=p-android-omap-3.0#l113

Regards,
Mike