On Fri, Jul 04, 2014 at 12:21:00PM +0200, Arnd Bergmann wrote: > However, what do we do about PCI hosts that can be used with different > kinds of systems? Do we assume that they all do PCI resource allocation? > Can we decide this on a per host driver basis, or do we need to introduce > an extension to the PCI DT binding to make that decision? If the firmware sets everything up, and standard ECAM/CAM config space is provided, then Will's simple generic driver should be selected. The kernel shouldn't even be using code to manipulate the host bridge. This is the x86 model. If it is more embedded focused and firmware doesn't do much, then a different compatible string and different driver can be used that does have any special setup code.. The HW needs to be designed to support this, so it actually has to imeplement configuration access properly, it can't split the config space for the bridge with config space for the downstream, for instance. It must implement something sensible for root port bridge windows, and a few other common sense things. Things are going to need to work like this anyhow on any HW that expects to suport ACPI... It is OK for the kernel to reconfigure the BARs and other things as it likes, so long as the configuration access mechanism is working properly according to the spec. IIRC it does a bit of a hybrid on x86 where it tries to leave things alone that are OK, and fix up things that are not OK. Jason -- 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
