Message-ID: <CAJZ5v0g72sDwj19me6=9R3wOYyDsATN3KBuYK+hWc-S=UcyRKA@mail.gmail.com>
Date:	Wed, 20 Jan 2016 14:51:41 +0100
From:	"Rafael J. Wysocki" <rafael@...nel.org>
To:	"Rafael J. Wysocki" <rafael@...nel.org>
Cc:	Dmitry Torokhov <dmitry.torokhov@...il.com>,
	"Rafael J. Wysocki" <rjw@...ysocki.net>,
	Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
	Rob Herring <robh@...nel.org>,
	Grant Likely <grant.likely@...aro.org>,
	Linus Walleij <linus.walleij@...aro.org>,
	Thierry Reding <treding@...dia.com>,
	Uwe Kleine-König 
	<u.kleine-koenig@...gutronix.de>,
	lkml <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] driver-core: platform: automatically mark wakeup devices

On Wed, Jan 20, 2016 at 3:40 AM, Rafael J. Wysocki <rafael@...nel.org> wrote:
> On Wed, Jan 20, 2016 at 1:45 AM, Dmitry Torokhov
> <dmitry.torokhov@...il.com> wrote:

[cut]

>>> > I am not aware of any drivers denoting itself as being wakeup sources
>>> > and not enabling wakeup. Do you have examples?
>>>
>>> Yes, I do.
>>>
>>> i8042 is one of them, exactly to avoid enabling those devices to do
>>> system wakeup on systems where they were not enabled to do that in the
>>> past.
>>
>> Ah, yes, you added it to i8042. Well, maybe that is an argument for not
>> automatically enabling wakeup on ACPI systems because that was the
>> default behavior for them. OF/board platforms have different default
>> behavior. I guess if we do not do anything besides what the patch is
>> doing, then ACPI will not have the property I propose, so they will be
>> capable but not enabled, and OF/boards will have the property and will
>> be capable and enabled.
>
> The defaults to use should generally depend on two things IMO: On what
> the firmware tells us to use (given the argument that the firmware
> people probably have a good reason to tell us what they are telling
> us) and on what we were doing for this class of devices in the past.
> Ignoring the last bit generally leads to regressions in the cases when
> the firmware people are wrong after all, so I'm always cautious about
> this.  And this rule applies to DT as well as to ACPI, although the DT
> people usually won't admit it. :-)

OK, I realize that the example I gave was not really a good one,
because the wakeup part was added to i8042 to support wakeup from
suspend-to-idle which is really special and the implementation is
based on what the driver was doing previously.

But this also makes me think that the problem is really more
complicated than it may seem to be, so what about starting over and
looking at the wakeup problem in general?

To that end, there are two categories of wakeup support in devices.
The first one is support for something called "remote wakeup" in USB
and which means that the device is able to generate wakeup signals
when it (the device) is suspended and the rest of the system can
generally be considered as working.  I'll use the "remote wakeup" term
in general for the lack of a better one.

Remote wakeup support is used in runtime PM and suspend-to-idle (which
essentially uses the same kind of hardware mechanics, but in a
different way).

Note, though, that "device is suspended" need not map 1:1 onto a
particular hardware state.  What it really means is that either it has
been suspended using the runtime PM framework, or all devices have
gone through the device suspend sequence during suspend-to-idle.  The
hardware state the device ends up in depends on the driver/bus type/PM
domain combination and is generally expected to be low-power.

It is easy in PCI or USB and on ACPI systems where low-power states of
devices are well defined and the connection between them and the
ability to generate wakeup signals is known.  It is not so easy in the
other cases, though.  My personal opinion is that if wakeup support is
required, the device should be put into the lowest-power state from
which it can generate wakeup signals.  That very well may be the
full-power state of it if that's the only suitable one.

If that point of view is adopted, then any device that (a) can take
input and (b) uses interrupts can do remote wakeup.  We don't really
have a good way to express that in the driver core.

The second category of wakeup support is for platform-assisted
low-power states of the whole system where there's a big switch (or a
bunch of them) allowing multiple things to be powered off in one go
from the OS perspective.

I'll write about that in the next message, as I need to take care of
some urgent stuff now.

Thanks,
Rafael

Powered by blists - more mailing lists