Re: [RESEND patch 2.6.25] ACPI uses device_may_wakeup() policy inputs

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

On Monday 21 April 2008, Zhang Rui wrote:
> 
> On Fri, 2008-04-18 at 21:18 -0700, David Brownell wrote:
> > This imports the driver model device.power.may_wakeup flags to ACPI,
> > using it to *REPLACE* the /proc/acpi/wakeup flags for some devices.
> > It depends on the previous patch making device.power.can_wakeup
> > behave. It does that by:
> > 
> >  - Implementing platform_enable_wakeup(), which is currently invoked
> >    only by pci_enable_wake().  When that's called -- probably in the
> >    driver suspend() call -- it updates acpi_device.wakeup.state.enabled
> >    flag in the same way writing to /proc/acpi/wakeup updates it.
> >    
> >  - Updating the usage of the corresponding ACPI flags when turning on
> >    wakeup power domains and GPEs.
> > 
> > THIS PATCH NEEDS MORE ATTENTION because of the way the ACPI method
> > invocations have been changing, e.g. the 1.0 vs 2.0 sequencing.
> > 
> > Right now it's not clear to me whether the GPEs are always enabled at
> > the right time, and for that matter whether the rules haven't changed
> > so that drivers can no longer effectively control those settings from
> > suspend() unless acpi_new_pts_ordering is in effect.
>
> Sorry. It's such a long sentence which is hard for me to understand. :(

Apologies.

On the bright side ... didn't all the new_pts_ordering stuff
get removed?  If that stays gone, it removes the main concern
I had.  That comment was written when I observed what looked
to be troublesome semantic changes from that "new" ordering.


> > it's not clear to me whether the GPEs are always enabled at
> > the right time
>  
> this patch doesn't change the time when GPEs are enabled.

No it doesn't.  Maybe I'm just more paranoid about it than
someone who knows ACPI (and its version-specific issues) a
lot better than me.


> > NOT YET SIGNED-OFF ... primarily because of the confusion about
> > the order in which ACPI methods get called during entry to suspend
> > states.
>
> I think it's safe to apply this patch.

I did this work before the "new_pts_ordering" stuff happened.  Then
after "new_pts_ordering", it looked a bit problematic ... originally,
I would have agreed with you.  Maybe now I can agree again.

- Dave



> thanks,
> rui
> 
> > Presumably one of the "new style" PM methods calls will
> > now always work for drivers wanting to enable wakeup methods...
>