For those of keen eye, my last patch contained a (harmless)
bug: a ',' in place of '&&'.  

The attached patch replaces yesterday's patch and contains
a number of additional needed changes.

-  "struct resource" was mistyped "struct ressource" in one place.
-  build_devlist() contained a memory leak: struct pci_devs were
   kmalloc()ed but not freed if left unused.
-  Made all printk()s consistent: start with "PnPBIOS:"
-  Made some code formatting and other trivial changes
-  Incorrect nodenum was recorded in dev->devfn
-  But most importantly, the driver now distinguishes between
   resource data that has not been set and resources that are
   reported by the PnP BIOS as extant but disabled.  Whereas
   previously a struct resource .start value of 0 was ambiguous
   between  "resource does not exist" and "resource is disabled",
   now the .flags field (specifically, the IORESOURCE_UNSET bit) is
   used to indicate an unused struct, while a .start value of -1
   is used to indicate a disabled irq or dma.

This patch supersedes the previous one I submitted yesterday
in a file called "pnpbios-patch-20010924-1" attached to a message
entitled "Small change to pnp_bios.c".

--
Thomas