Re: [RFC][PATCH 3/4] PCI / ACPI PM: Platform support for PCI PME wake-up

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

On Monday 14 September 2009, Matthew Garrett wrote:
> On Mon, Sep 14, 2009 at 12:53:05AM +0200, Rafael J. Wysocki wrote:
> > On Monday 14 September 2009, Matthew Garrett wrote:
> > > On Sun, Sep 13, 2009 at 11:24:03PM +0200, Rafael J. Wysocki wrote:
> > > > +	} else if (!dev->wakeup.flags.run_wake) {
> > > > +		acpi_set_gpe_type(dev->wakeup.gpe_device,
> > > > +					dev->wakeup.gpe_number,
> > > > +					ACPI_GPE_TYPE_WAKE);
> > > 
> > > Is this going to work for cases where we have multiple devices attached 
> > > to the same GPE? The common one is EHCI, where both EHCI HCDs will be 
> > > one a single GPE. If we wake one, that'll then disable the GPE for the 
> > > other. Further wakeup events will then be lost.
> > 
> > You're right, I overlooked that.  Some kind of refcounting is needed here.
> 
> I've sent patches to implement this at the GPE level, which also change 
> the API for requesting them. I'm waiting on feedback from Bob Moore.
> 
> > > > +	if (device->wakeup.flags.valid)
> > > > +		acpi_install_notify_handler(device->handle, ACPI_SYSTEM_NOTIFY,
> > > > +						pci_acpi_device_wakeup,
> > > > +						&dev->dev);
> > > > +
> > > 
> > > I think this will fail for the root bridge if acpiphp has already 
> > > grabbed it to check for hotplug events.
> > 
> > The root bridge is not a struct pci_dev and this is called only for these.
> 
> Are you sure?

Yup.  ACPI associates the root bridge handle with the struct device created by
pci_create_bus() as the bridge of the root bus.

> When I did this I saw conflicts with acpiphp. I may have my terminology
> confused. I sent another patch that abstract that out.

I'll have a look at acpihp, then.

Thanks,
Rafael