The HURD needs to provide interfaces to let tools such as lspci and setpci access PCI configuration space.  The current pciutils backends for the hurd support only reading a dump file and access through i386 ports (0xcf8 et al).  This has the following problems:
 - non-i386 architectures do not work in this way
 - this mechanism doesn't support access to extended pci config space (bytes 256-4095) as found on PCI-X 2.0 and PCI Express cards.
 - this mechanism doesn't support PCI domains/segments which are increasingly commonly found on workstation i386 machines as well as other architectures and large i386 servers

Feel free to ask me about the Linux implementations of accessing PCI config space; I propose you come up with an abstraction somewhat like sysfs (ie you have a directory per PCI device named something like 0001:04:02.3 and a file inside that called 'config'), but I'm interested in discussing other mechanisms that you might come up with.