On Tuesday, December 18, 2012 10:59:46 AM Yinghai Lu wrote:

On Tue, Dec 18, 2012 at 8:10 AM, Toshi Kani [off-list ref] wrote:

quoted

On Tue, 2012-12-18 at 02:48 +0100, Rafael J. Wysocki wrote:

So, I would propose the following changes.

 - Move the acpi_hot_add_bind() call back to the original place after
the device_attach() call.
 - Rename the name of acpi_hot_add_bind() to something like
acpi_bind_adr_device() since it is no longer hot-add only (and is
specific to _ADR devices).
 - Create its pair function, acpi_unbind_adr_device(), which is called
from acpi_bus_remove().  When a constructor interface is introduced, its
destructor should be introduced as well.
 - Remove the binding procedure from acpi_pci_root_add().  This should
be done in patch [2/6].

i think we should put jiang four patches before Rafael's patches.

http://git.kernel.org/?p=linux/kernel/git/yinghai/linux-yinghai.git;a=shortlog;h=refs/heads/for-pci-jiang-hotplug

Actually, I have something more radical than that in mind. :-)

Namely, we don't need to call the wakeup-related stuff from acpi_pci_bind() and
acpi_pci_unbind().  It's only there, because I did't find a better place for it
when I added it.

If we can set the ACPI handles of PCI devices in pci_scan_device(), which isn't
too difficult to do (I actually have a patch for that and it's not too invasive),
we can easily move the wakeup enabling stuff to pci_pm_init() and wakeup
disabling somewhere near pci_release_capabilities().

Then we are only left with the _PRT setup in there, but that could be done
as soon as in pci_scan_device() too, if I'm not mistaken.

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.