Re: post 2.6.26 requires pciehp_slot_with_bus

[Alex Chiang <achiang@hp.com>]

* Jesse Barnes <jbarnes@virtuousgeek.org>:
> On Thursday, July 24, 2008 2:51 pm Pierre Ossman wrote:
> > On Thu, 24 Jul 2008 14:07:18 -0700
> >
> > Jesse Barnes <jbarnes@virtuousgeek.org> wrote:
> > > On Thursday, July 24, 2008 1:39 pm Pierre Ossman wrote:
> > > > This is a laptop with a single expressport that worked
> > > > fine up until this merge window. What changed and why is
> > > > it no longer possible to support this hardware without a
> > > > kernel parameter?
> > >
> > > Yeah we're being a bit more careful about registering
> > > hotplug slots these days.  The fact that you got a conflict
> > > message indicates that more than one driver is trying to
> > > bind to that PCIe port and handle hotplug for it.  I guess
> > > acpiphp must already be loaded?
> >
> > Nope. Only pciehp is builtin, and acpiphp is not loaded.
> > Anything non-hotplug related that can block stuff?
> 
> Theoretically only one of the other hotplug drivers could claim
> the slot (e.g.  acpi, cpci, cpqhp, fakephp etc.).  You could
> put a dump_stack() or something into
> drivers/pci/hotplug/pci_hotplug_core.c:pci_hp_register to see
> if we're getting multiple callers on the same slot for some
> reason...  Maybe the pciehp driver itself is trying to register
> the same slot twice?

This is a good idea.

Any chance you could get that debug info for us, Pierre?

Thanks.

/ac