On Thu, Sep 27, 2012 at 6:23 PM, Bjorn Helgaas [off-list ref] wrote:

On Thu, Sep 27, 2012 at 6:15 PM, Taku Izumi [off-list ref] wrote:

quoted

On Thu, 27 Sep 2012 10:48:09 -0600
Bjorn Helgaas [off-list ref] wrote:

quoted

On Fri, Sep 21, 2012 at 2:09 PM, Bjorn Helgaas [off-list ref] wrote:

quoted

On Tue, Sep 18, 2012 at 12:25 AM, Taku Izumi [off-list ref] wrote:

quoted

quoted

quoted

quoted

@@ -661,6 +663,11 @@ static int acpi_pci_root_remove(struct a
        device_set_run_wake(root->bus->bridge, false);
        pci_acpi_remove_bus_pm_notifier(device);

+       acpi_pci_irq_del_prt(root->bus);

acpi_pci_irq_del_prt() does not actually have a dependency on the
struct pci_bus, so I think its interface should be changed so it takes
a segment number and a bus number instead of the "struct pci_bus *".
The same applies to acpi_pci_irq_add_prt().

This basically boils down to reverting 859a3f86ca8 and d9efae3688a.  I
acked those changes at the time, but I think they were a mistake.  The
reason is that passing in the struct pci_bus * ties them into the host
bridge add/remove flow in a way that's not necessary.

If we get rid of the struct pci_bus * dependency, then we can easily
add the _PRT before doing PCI enumeration behind the bridge, and we
can remove the _PRT after removing the PCI devices.  I think this is
one small step toward getting rid of the add/start and stop/remove
split.

I'm going to work on doing this if nobody else is interested.

  I'll do that. Maybee that is separeted from this patchset.

Great, thanks!  A separate patchset is definitely fine.

Whether this is a waste of time or not, I don't know.  But I did a
little bit of work on this before you volunteered, Taku.  Sorry, I
should have pointed you at the work-in-progress patches; maybe they
could have saved you a bit of time.  Anyway, here they are:

http://git.kernel.org/?p=linux/kernel/git/helgaas/pci.git;a=shortlog;h=refs/heads/pci/bjorn-misc