On Thu, Sep 9, 2021 at 4:10 PM Rob Herring <robh@xxxxxxxxxx> wrote: > On Wed, Sep 8, 2021 at 11:39 AM Arnd Bergmann <arnd@xxxxxxxx> wrote: > > > > > It might be time for default implementations here that can be shared > > > with arm64. The functions look the same or similar to the arm64 > > > version in many cases and why they are different isn't that clear to me > > > not being all that familar with the ACPI code. > > > > I think it can be simplified quite a bit if we restructure the acpi pci > > code to behave like a normal pci host bridge driver. > > That is exactly what I want to see happen! I'm not that familiar with > the ACPI device probing piece of it or I probably would have done that > by now. I gather there's not a normal acpi_device (or platform_device > with ACPI matching?) so we'd have to create the device(s) based on the > MCFG table. I have a patch that I did as part of a longer series to modernize some of the more unusual pci host bridges, it's only a small step but should help follow up for the rest: https://git.kernel.org/pub/scm/linux/kernel/git/arnd/playground.git/commit/?h=pci-probe-rework-20210320&id=7346fbf1938e547833726ebdf25dfe0ef185cbff What I noticed is that there are a couple of data structures that each exist for every acpi host bridge: struct acpi_pci_root struct acpi_pci_root_ops struct acpi_pci_root_info struct acpi_pci_generic_root_info struct pci_sysdata (arch specific, but includes data used by acpi) struct pci_host_bridge I think we can pretty much move all the struct members from those into the generic pci_host_bridge, removing the duplicates and adding #ifdef CONFIG_ACPI for some of them. Similarly, for the global functions from arch/arm64/kernel/pci.c etc, I think they should mostly get turned into callback handlers that get set by the probe function, replacing the __weak defaults. Arnd
