Re: [Alternative 2][PATCH] ACPI / PCI: Set root bridge ACPI handle in advance
From: Rafael J. Wysocki
Date: Wed Jan 02 2013 - 19:35:06 EST
On Wednesday, January 02, 2013 04:07:32 PM Bjorn Helgaas wrote:
> On Thu, Dec 27, 2012 at 10:32:13PM +0100, Rafael J. Wysocki wrote:
> > To that end, split pci_create_root_bus() into two functions,
> > pci_alloc_root() and pci_add_root(), that will allocate memory for
> > the new PCI bus and bridge representations and register them with
> > the driver core, respectively, and that may be called directly by
> > the architectures that need to set the root bridge's ACPI handle
> > before registering it.
> I'm trying to *reduce* the interfaces for creating and scanning PCI
> host bridges, and this is a step in the opposite direction.
Yes it is.
The alternative is to make the root bridge initialization code more complex.
> > Next, Make both x86 and ia64 (the only architectures using ACPI at
> > the moment) call pci_alloc_root(), set the root bridge's ACPI handle
> > and then call pci_add_root() in their pci_acpi_scan_root() routines
> > instead of calling pci_create_root_bus(). For the other code paths
> > adding PCI root bridges define a new pci_create_root_bus() as a
> > simple combination of pci_alloc_root() and pci_add_root().
> pci_create_root_bus() takes a "struct device *parent" argument. That
> seems like a logical place to tell the PCI core about the host bridge
> device, but x86 and ia64 currently pass NULL there.
And there's a reason for that. Namely, on these architectures PCI host
bridges have no physical parents (well, at least in current practice).
> The patch below shows what I'm thinking. It does have the side-effect
> of changing the sysfs topology from this:
> to this:
> because it puts the PCI root bus (pci0000:00) under the PNP0A08 device
> rather than at the top level.
Which is wrong.
PNP0A08 is not a parent of the host bridge, but its ACPI "companion" (ie. ACPI
namespace node representing the host bridge itself).
> That seems like an improvement to me, but it *is* different.
Well, then we should make every ACPI device node corresponding to a PCI device
be a parent of that device's struct pci_dev and so on for other bus types. It
doesn't sound like an attractive idea. :-) Moreover, it is impossible, because
those things generally already have parents (struct pci_dev objects have them
That said the idea to pass something meaningful in the parent argument
of pci_create_root_bus() can be implemented if we create a "physical" device
object corresponding to "device:00" (which is an ACPI namespace node) in your
>From what I can tell, "device:00" always corresponds to the ACPI _SB scope
(which is mandatory), so in principle we can create an abstract "physical"
device object for it and call it something like "system_root". Then, if we
use it as the parent of pci0000:00 (the host bridge), then we'll have
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/