On Tue, 14 Jul 2009 14:04:31 -0600
Alex Chiang <achiang@xxxxxx> wrote:
> * Jesse Barnes <jbarnes@xxxxxxxxxxxxxxxx>:
> > Haven't heard from Len yet, but I'm a bit nervous about adding this
> > to for-linus. We already have a fix upstream for the
> > acpi_get_pci_dev issue right? So there's no hurry on this set of
> > (nice) cleanups is there?
>
> The problem is that I introduced an assumption in
> acpiphp_configure_bridge, which calls acpi_get_pci_dev. That call may
> fail on systems with non-materialized PCI root bridges.
>
> Fixing that assumption was addressed by this hunk in patch 2/3:
>
> @@ -1387,16 +1363,7 @@ static void acpiphp_sanitize_bus(struct
> pci_bus *bus) /* Program resources in newly inserted bridge */
> static int acpiphp_configure_bridge (acpi_handle handle)
> {
> - struct pci_dev *dev;
> - struct pci_bus *bus;
> -
> - dev = acpi_get_pci_dev(handle);
> - if (!dev) {
> - err("cannot get PCI domain and bus number for
> bridge\n");
> - return -EINVAL;
> - }
> -
> - bus = dev->bus;
> + struct pci_bus *bus = pci_bus_from_handle(handle);
>
> pci_bus_size_bridges(bus);
> pci_bus_assign_resources(bus);
>
> Now, my systems cannot do PCI root bridge hotplug, so that line
> where we call acpi_get_pci_dev() won't cause us to fail, but I'm
> concerned about machines in the wild that:
>
> a) can do PCI root bridge hotplug
> b) have non-materialized PCI root bridges
>
> Maybe that is a small set of machines...
>
> I agree that this series is on the large side for -rc4 and
> beyond, but I'd prefer to fix it now.
>
> The thing is, a "minimal" fix wouldn't save us that much, since
> we'd still have to export acpi_pci_root stuff. We could maybe
> drop patch 3/3 which makes the final diffstat look like this:
>
> drivers/acpi/pci_root.c | 17 +-----
> include/acpi/acpi_bus.h | 14 +++++
> drivers/pci/hotplug/acpiphp_glue.c | 108
> ++++++++++++------------------------ 3 files changed, 53
> insertions(+), 86 deletions(-)
>
> Hm, actually, I can actually make that even smaller by separating
> out the interface change.
>
> Would you like me to do that and resubmit? I'd still need an ACK
> from Len re: acpi_pci_root...
How about I actually send a reply this time...
No you don't need to re-submit; assuming Len comes back quickly we can
put this into my for-linus branch. Len, we just need your ack to apply
this series.
Thanks,
--
Jesse Barnes, Intel Open Source Technology Center
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
