On Mon, Jun 18, 2012 at 5:40 PM, Yinghai Lu [off-list ref] wrote:

On Mon, Jun 18, 2012 at 3:30 PM, Ulrich Drepper [off-list ref] wrote:

quoted

On Sat, Jun 16, 2012 at 5:56 PM, Bjorn Helgaas [off-list ref] wrote:

quoted

If ACPI provides a perfectly usable generic way to describe this
topology and the vendor BIOS doesn't bother to use it, I'm not very
interested in trying to compensate for that BIOS deficiency by adding
a bunch of non-portable CPU-specific gunk to Linux.

The problem is that all machines get this wrong.  I've tested varies
models from Dell and HP and none of them have the _PXM entry and the
local_cpus fields are wrong.  If there is a reasonably sane way to
compensate for broken BIOSes it should be considered.  We all know how
good BIOS authors are...

please check attached one. and  you may append "pci=busnum_node=00:00,80:01"
to change node for root bus.
assume you have two root bus: 00, 80

I'm not opposed to something like this, if people think it's useful.

This patch sets the node quite early, before we even look at the _PXM
information in pci_acpi_scan_root().  That means if the BIOS does
supply a _PXM method and the user gives this argument, the
user-supplied info is silently overwritten.  To me it would make more
sense to handle an option like this *after* we look for _PXM info.
That way it could be used to compensate for both missing and incorrect
_PXM info.