[PATCH 5/7] Serial: OMAP: add runtime pm support for omap-serial driver

[Kevin Hilman]

Govindraj <govindraj.ti at gmail.com> writes:

[...]

>>
>> So here's an experiment to try with autosuspend.  I suspect this will
>> work, just hack it up to prove the concept.  If it works, we can make
>> something more generic.  Here are a few alternatives to try.  I may
>> experiment with some of them tomorrow as well, but please let me know
>> what you try:
>>
>> Using autosuspend, clocks will get cut independently of the idle path.
>> Then, use the PRCM ISR detection of UART module wakeups to call the
>> UART's interrupt handler.  The interrupt handler will pm_runtime_get(),
>> enable the clocks, and then take care of the interrupt.  Done.
>>
>> Alternatively, you could test it on current code by simply removing the
>> resume_from_idle call from the idle path and calling it instead from the
>> PRCM ISR when UART module wakeups are detected.
>
> I remember doing similar experiment didn't seem to help,
>

To show it's possible, I did really hacky proof of concept, hard-coded
to UART3 console for n900/beagle, but at least it shows that this
approach can work, and the module-level wakeups are working and can be
used as the trigger for UART wakeup instead of resume_from_idle.  Of
course, it still has problems with using serial after non-UART wakeups,
but once omap-serial is using runtime PM, that will no longer be an
issue.

My hacky branch is called pm-wip/uart-wkup and is in my pm git tree[1]
(based at the pm-core branch)

To test module-level wakeups (instead of IO-ring) I again forced CORE to
stay on during suspend, and tested wakeup from suspend on 3430/n900:

  echo 3 > /debug/pm_debug/core_pwrdm/suspend
  echo mem > /sys/power/state

Kevin

[1] git://git.kernel.org/pub/scm/linux/kernel/git/khilman/linux-omap-pm.git