On Tue, 2022-08-02 at 10:18 +0100, David Woodhouse wrote: > On Tue, 2022-08-02 at 09:46 +0200, Ard Biesheuvel wrote: > > > If we want this, I would propose (happy to provide the implementation > > > but let's discuss the design first) something along the line of a > > > generic mechanism to "register" such a system device, which would add > > > it to a list. That list would be scanned on PCI device discovery for > > > BAR address matches, and the pci_dev/BAR# added to the entry (that or > > > put a pointer to the entry into pci_dev for speed/efficiency). > > > > This means that bus numbers cannot be reassigned, which I don't think > > we rely on today. To positively identify a PCI device, you'll need > > some kind of path notation to avoid relying on the bus numbers > > assigned by the firmware (this could happen for hot-pluggable root > > ports where no bus range has been reserved by the firmware) > > That kind of path notation already exists for the Intel IOMMU, and > probably others. See dmar_match_pci_path(), dmar_pci_bus_add_dev() etc. > > It would be good to lift that out and make it generic, rather than > reinventing another version. I think this is a completely orthogonal issue to what I'm trying to solve. I don't think we actually have a problem with bus numbers changing (see my other response). Yes, bus-number-agnostic PCI paths have been a thing for a long time in device-tree land :) Cheers, Ben.
