Re: 900af0d breaks some embedded suspend/resume

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Saturday 18 April 2009, Russell King wrote:
> On Sat, Apr 18, 2009 at 03:23:35PM +0200, Rafael J. Wysocki wrote:
> > The patchset in question had been discussed quite extensively before it was
> > merged and it's a pity none of the people caring for the affected platforms
> > took part in those discussions.  That would allow us to avoid the breakage.
> 
> Maybe on some list, but not everyone is subscribed to a million and one
> mailing lists.  I don't have enough time to read those which I'm currently
> subscribed to, so I definitely don't want any more.
> 
> > > or provide alternative equivalent functionality where the platform code is
> > > notified of the PM event prior to the late suspend callback being issued.
> > 
> > There is the .begin() callback that could be used, but if you need the
> > platform to be notified _just_ prior to the late suspend callback, then the
> > only thing I can think of at the moment is the appended patch.
> > 
> > It shouldn't break anything in theory, because the majority of drivers put
> > their devices into low power states in the "regular" suspend callbacks anyway.
> 
> Okay, my requirement is:
> 
> What I need to be able to do is to suspend most devices on the host side
> which may involve talking to a separate microcontroller via I2C to shut
> down power to peripherals.
> 
> Once that's complete, I then need to inform this microcontroller via I2C
> that we're going to be entering suspend mode, and wait for it to acknowledge
> that (after it's done its own suspend preparations).  At that point I can
> shutdown the I2C controller, and enter suspend mode.

Would it be an option to use a sysdev for that?

> Upon resume (which is activated by this microcontroller, including jtagging
> the boot code across to the host CPU), we need to tell this microcontroller
> that we're going back to 'run' mode again via I2C, and then resume the
> devices.
> 
> This is why we hooked the PXA I2C driver into the late suspend and
> early resume methods, so the I2C driver would be the last to suspend and
> the first to resume, thus allowing it to be used to talk to this micro-
> controller when required.  This worked out nicely because the late suspend
> used to before the platform prepare and platform finish used to happen
> after the early resume methods were called.

Unfortunately I'm not familiar with I2C, so I'm not sure whether or not this
would work, but it looks like sysdev could be used instead of platform_driver
for i2c_pxa_driver.

If using sysdev here (and analogously in i2c-s3c2410.c) is an option, I'd
prefer to do that instead of reordering suspend and resume code once again.
 
> So, it looks like your patch to kernel/power/main.c will allow me to
> continue to achieve this.  (Don't know about the disk.c bits, I've
> never used suspend-to-disk.)

This patch undoes some previous changes, but I'm not too comfortable with it,
because I think that putting devices into low power states after .prepare() may
not work on some ACPI systems.  Since devices should generally be put into low
power states during the "late" suspend (ie. when their interrupt handlers are
not invoked), it may be quite inconvenient.

Thanks,
Rafael