Re: cpuidle - minimum time for sleep

[Nishanth Menon <nm@ti.com>]

On 10/09/2014 02:01 PM, Valdis.Kletnieks@vt.edu wrote:
> On Thu, 09 Oct 2014 21:28:23 +0300, Ran Shalit said:
> 
>> Does anybody know what is the minimum expected time for sleep period
>> with the cpuidle ?
> 
> Both processor dependent and sleep level dependent.  There's a certain
> amount of latency induced by the hardware waking up.
> 
> Look at /sys/devices/system/cpu/cpu*/cpuidle/state*/latency
> 
Yes, that is correct. the sleep and wakeup time are dependent on the
power state we attempt (may or maynot achieve)

I personally toggle an unused pin using padmux register write with
weak pull up/down in controlled tests (mostly using disable =1 for
states I am not measuring), then capture pinctrl toggles using [1]
into a csv for many thousands of iterations then use the conservative
values. I usually do this at the slowest frequency to capture the
worst case values that i feed into cpuidle_driver.states.exit_latency
and appropriate value for target_residency - I usually ignore
power_usage as the value is never a constant and depends on quiet a
few factors that i cannot discuss in public domain.

Here are some helpful links on OMAP specific strategies (these may be
a little old, but just search for CPUIDLE latency measure in google)

https://lwn.net/Articles/384146/

http://www.omappedia.org/wiki/Power_Management_Device_Latencies_Measurement

http://lists.linaro.org/pipermail/linaro-dev/2010-August/000568.html


[1] https://www.saleae.com/

-- 
Regards,
Nishanth Menon