On Mon, Jun 28, 2010 at 2:49 PM, Kevin Hilman
[off-list ref] wrote:

Grant Likely [off-list ref] writes:

quoted

On Fri, Jun 25, 2010 at 4:46 PM, Kevin Hilman
[off-list ref] wrote:

quoted

Grant Likely [off-list ref] writes:

quoted

Another way to look at the problem is that these runtime
customizations are kind of a property of the parent device (the bus,
not the bus_type). ?Would it make sense for parent devices to have
runtime ops to perform for each child that is suspended/resumed? ?That
would make it simple to register another device that implements the
bus behaviour and attach it at runtime instead of compile time.

Maybe I didn't fully understand your idea, but the problem here is
devices do not have dev_pm_ops. ?Only busses, classes, and types have
dev_pm_ops.

Sorry, I mistyped. ?What I meant was for the parent device's
device_driver to be able to have a set of child dev_pm_ops; but I'm
wading into territory (power management) I'm not particularly familiar
with, and that might be making things far too complex.

I see your point now, but I think this indeed might complicate things
too much.

Also, I'm not crazy about how this would delay the per-device PM hooks
to be essentially batched until all devices under the parent are
"ready."

No, delaying until all children are ready wasn't my intent.  I was
thinking about a set of childs_dev_pm_ops() that would be called once
for each child as each child suspends/resumes...  but probably a moot
point since we both agree that it adds a lot of complexity.  :-)

But anyways, it just needs some more research on my part.
Unfortunately, I'll be away from this work on vacation for most of July,
so this won't get any attention from me until August.

quoted

quoted

Though I'm horribly unfamiliar with the intended usage of 'struct
device_type', an interesing discovery is that dev->type also has
dev_pm_ops, and it takes precedence over the bus type in the
suspend/resume. ?IOW, when suspending, when deciding which dev_pm_ops to
use, it checks class, type, then bus in that order.

So I guess my suggestion above boils down to somehow inserting
"parent" between type and bus in that list.

quoted

I need to explore this 'type' feature a little more, but using that or
the 'class' might be another way to have custom PM ops for certain
platform_devices.

Should maybe start cc'ing Greg and linux-kernel/linux-pm in this discussion.

They were involved in the early discussions about overriding the
platform_bus dev_pm_ops methods, which led us to the current
implementation.

But as I explore the custom bus approach a bit more (in August) I'll
broaden the audience.

Thanks again for all your helpful suggestions,

np.  Talk you you later.  Have a great vacation.

g.

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.