On Thursday 01 April 2010 12:00:06 pm Ira W. Snyder wrote: > On Wed, Mar 31, 2010 at 05:18:21PM -0600, Bjorn Helgaas wrote: > > > [ 0.340009] pci 0000:00:00.0: host bridge window [io 0x01f0-0x01f7] > > > [ 0.344008] pci 0000:00:00.0: host bridge window [io 0x03f6] > > > [ 0.348008] pci 0000:00:00.0: host bridge window [io 0x0170-0x0177] > > > [ 0.352008] pci 0000:00:00.0: host bridge window [io 0x0376] > > > > Seems surprising that these are the only ports in the 0x0-0xcfff range > > that are routed anywhere. I would naively expect them to be at least > > subtractively decoded by one of the bridges, so all the legacy builtin > > hardware (APICs, serial ports, etc) would still work, but the unused > > stuff would be available for PCI. > > I believe that the legacy IDE ports have a fixed address. See the > drivers/ide/ide-legacy.c ide_legacy_init_one() function. The driver uses > hardcoded fixed addresses. Yep, there's no question that these IDE ranges are forwarded by the bridge. I think it's likely that the bridge forwards *more* than that, but you'd have to find that in a spec or do a lot of experimenting, e.g., by plugging in a card that uses I/O ports, forcibly writing the BARs to different addresses, and seeing whether the card works. > However, if I do not include the reservation of the legacy IDE IO ports, > I get messages like this during kernel boot, and the IDE controller is > disabled: > > [ 0.500137] pci 0000:00:0f.1: no compatible bridge window for [io 0x01f0-0x01f7] > [ 0.504009] pci 0000:00:0f.1: no compatible bridge window for [io 0x03f6] > [ 0.508009] pci 0000:00:0f.1: no compatible bridge window for [io 0x0170-0x0177] > [ 0.512008] pci 0000:00:0f.1: no compatible bridge window for [io 0x0376] > [ 0.516009] pci 0000:00:0f.1: no compatible bridge window for [io 0xffa0-0xffaf] Right. I wasn't suggesting removing those bridge windows, only that there's a lot of unused space in the 0x0-0xcfff range, even after reserving the legacy ports, and it's possible that one of the host bridges would forward that unused space to the PCI bus. I don't know much about PCMCIA, but I *think* there might be PCMCIA devices that are not fully configurable, e.g., a device might only have a few options for where its registers can be mapped. If all of those options were in the 0x0-0xcfff range, there'd be no way to make that device work in this system with your current patch. But this is just idle speculation on my part; no action needed on your part :-) Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
