On Mon, 29 Apr 2002, Jeff Garzik wrote: > My suggestion has always been to give Linux what it wants. For memory, > this means providing the E820 table, and for PCI $PIR, this means > providing the PCI IRQ routing table... This is the cost of having your > own data table, etc. -- you have to modify all the software you are > working with just to read the data table.
I provide $PIR to linux now. But some hunting around in the pci source seemed to have #ifdef CONFIG_PCIBIOS around the linux code that hunts for $PIR. Did I misread? I normally disable CONFIG_PCIBIOS because linuxbios doesn't have one, so having linux call int 0x1a is a bad thing. > Why not just provide the tables existing software is already tested and > proven to work with? we have that. I'm worried about what combination of CONFIG options will make linux look for it. I'll go back and look and try to verify that I am not missing something. ron
