Re: [PATCH 1/1] PM / Runtime: let rpm_resume fail if rpm disabled and device suspended.

From: "Rafael J. Wysocki" <rjw@rjwysocki.net>
To: Kevin Hilman <khilman@linaro.org>
Cc: Alan Stern <stern@rowland.harvard.edu>,
	Allen Yu <alleny@nvidia.com>, Pavel Machek <pavel@ucw.cz>,
	Len Brown <len.brown@intel.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	"linux-pm@vger.kernel.org" <linux-pm@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 1/1] PM / Runtime: let rpm_resume fail if rpm disabled and device suspended.
Date: Sun, 22 Jun 2014 15:40:17 +0200	[thread overview]
Message-ID: <1645345.yNjrKp8T0k@vostro.rjw.lan> (raw)
In-Reply-To: <7h4mzftgdl.fsf@paris.lan>

On Friday, June 20, 2014 02:34:14 PM Kevin Hilman wrote:
> Alan Stern <stern@rowland.harvard.edu> writes:
> 
> > On Fri, 20 Jun 2014, Rafael J. Wysocki wrote:
> >
> >> > For a general device, the fact that dev->power.is_suspended is set
> >> > means the device _has_ been powered down.  Even though the
> >> > runtime_status may not have changed, the PM core has to assume the
> >> > device is not available for use.
> >> 
> >> This seems to go a bit too far.  What power.is_suspended actually means is
> >> that __device_suspend() has run for the device successfully.  What the
> >> implications of that are depends on the bus type (or subsystem in general)
> >> and device driver.
> >> 
> >> > While your I2C devices may be useable even after the ->suspend callback
> >> > returns, for most devices this isn't true.  So we shouldn't allow
> >> > rpm_resume() to return imediately when is_suspended is set.
> >> 
> >> I can agree with that.
> >
> > We really do need to decide more precisely how runtime PM and system PM 
> > will interact.  
> 
> Yes!
> 
> > Should ->runtime_resume callbacks be allowed after ->suspend has
> > returned?
> 
> Abolutely.
> 
> > Kevin has stated that some devices do need this ability.  But most
> > don't.  
> 
> Does it matter if most don't?  As long a some do, we need to support
> this.  It may not be "most" devices, but on the (mostly embedded) SoCs I
> work on, the devices that do need this tend to be rather crucial core
> devices that are used during the PM of other devices (e.g. I2C, SPI,
> GPIOs, etc. etc.)
> 
> > The PM core needs to handle these conflicting requirements
> > somehow.
> 
> I agree.  We've gone back and forth a few times on the various
> interactions between system PM and runtime PM over the years but it
> seems there are still things to clarify.

Well, we only considered a specific use case every time without looking at
the big picture, mostly because we didn't really know what the big picture
was.  I guess today we have enough experience to try to address all of these
problems together.

I guess we need to start with making a list of different types of bus type/driver
behavior existing today and how the core is supposed to interact with them.

Rafael