Re: [PATCH] implement pm_ops.valid for everybody

[David Brownell <david-b@pacbell.net>]

On Thursday 22 March 2007 6:14 pm, Matthew Locke wrote:
> >
> > So the fundamental definition needs to be in relative terms,
> > because platform-specific differences otherwise make trouble.
> 
> The problem is that a 1:1 mapping between system low power state and  
> a processor low power state is trying to be forced on every  
> platform.

Sort of.  There are only two labels available for "system" states,
and the platform-specific code entering those states will probably
kick in a processor low power state.  But there'd be no point in
preventing that code from kicking in deeper power save states.

Consider the PXA 255, which has an "idle" mode that's natural for
the idle loop, a "33 MHz idle" mode that saves more power but
means most peripherals can't be clocked, and a "sleep" mode that
turns the CPU off.  A "standby" might normally use "idle", but
might likewise use "33 MHz idle" if all the peripheral clocks
happen to be gated off after the drivers suspend().  That wouldn't
be a one-to-one mapping ... and smarter hardware might do very
similar stuff automatically, too.


> As Dave pointed out, embedded SoC's provide multiple low   
> power states that qualify for the suspend-to-ram definition.  The  
> only reasonable platform independent definition is that in STR memory  
> is powered and contents preserved.  The rest is platform specific.

That definition also applies to "standby" of course ... there may
be a LOT of states where the "standby" label can usefully apply.


> I think the right answer is that a mechanism for mapping platform  
> specific states to the system states is needed. 

That could be done, but in practice not all platform states will
necessarily be implemented by the platform code.


> Platforms define   
> their low power states and define the default for each system  
> state .  On x86 platforms, the default just works and is probably  
> never changed. 

That's the way it works today on all Linux platforms.  The x86
ones actually don't "just work" in my observation ... when either
"standby" or "STR" actually work, I've been quite surprised; the
basic issue seems to be that ACPI resume paths often misbehave.


> On embedded platforms, a policy manager can change   
> the other low power states according to its latency and operational  
> requirements.

That's not yet a proposal to change things; no details.  :)

If I were to want to change things in that area, I'd likely want
to let platforms define their own states and names (*), but in
fact I think it's probably a lot more imporant for embedded cases
to support top notch *runtime* PM rather than /sys/power/state
kinds of transitions.

- Dave

(*) For example, predefine more suspend_state_t values -- maybe
    PM_SUSPEND_PLATFORM_{0-9}, to start with -- and then add a
    label-that-state callback to "struct pm_ops".  Then make
    /sys/power/state do the obvious stuff to read and write
    additional states ... voila, platforms can now add new low
    power states with their own names.  Their clk_must_disable()
    could let drivers see some of the state differences; there
    might need to be similar mechanisms for other PM resources.