On Sat, Aug 18, 2012 at 02:16:28PM +0200, Wolfram Sang wrote:

On Sat, Aug 18, 2012 at 01:17:45PM +0200, Thierry Reding wrote:

quoted

On Sat, Aug 18, 2012 at 11:48:46AM +0200, Wolfram Sang wrote:

quoted

On Wed, Aug 08, 2012 at 09:06:47AM +0200, Thierry Reding wrote:

quoted

This commit adds support for probing slave devices parsed from the
device tree.

Signed-off-by: Thierry Reding <redacted>

How are the devicetree nodes of the slaves connected to this PCI device
which should not have a node itself?

The matching is done by pci_set_of_node() and pci_set_bus_of_node().
The latter will eventually call pcibios_get_phb_of_node(). What makes
you say that PCI devices shouldn't have nodes themselves?

I thought to recall that busses which can be probed at runtime should
not have nodes but should be probed. Maybe that was wrong. Can you point
me to a dts-file with an example?

That approach doesn't work for I2C controllers, though because you
couldn't attach an slaves to them. The same goes for SPI controllers or
any other dumb busses that can't be probed.

arch/x86/platform/ce4100/falconfalls.dts is an example for an X86
platform where PCI devices have associated device tree nodes.

Thierry