On Fri, Nov 06, 2009 at 02:47:54PM -0600, Bjorn Helgaas wrote: > On Friday 06 November 2009 12:57:17 pm Ira W. Snyder wrote: > > On Thu, Nov 05, 2009 at 06:34:56PM -0600, Bjorn Helgaas wrote: > > > > [ big snip ] > > > > > > > > Of course, this doesn't help if you actually power off the slot or > > > press your reset button. In that case, we have to assign things from > > > scratch, and we really need to know what the host bridge apertures are. > > > It'd be best if the BIOS gave us an ACPI description, so we could just > > > use "pci=use_crs". Without that, I think we'll need a host bridge > > > driver to read them out of the hardware, and it sounds like we don't > > > have the specs required to write such a driver. > > > > > > > What would actually be needed to write this driver? I ran an the command > > "lspci -MH1 -vv -xxx", which I found in a very old email. It manages to > > dump out all 0x100 bytes of configuration space on the bridges. Using > > this and lspci, it is clear what memory lives behind the bridge, I > > think. > > > > Here is a dump from my Force computer's Broadcom host bridges. Look > > closely at the 4 16-bit registers starting at 0xC0. If you correlate > > them with the BAR dump, shown below, it shows: > > > > Bridge 00:00.0 > > non-prefetchable memory from 0xf9000000 - 0xf93fffff > > prefectchable memory from 0xf9400000 - 0xf96fffff > > > > Bridge 00:00.1 > > non-prefetchable memory from 0xf98c0000 - 0xfaffffff > > prefectchable memory from NONE PRESENT > > Hey, this is cool! Nice detective work :-) > > There should also be a bus number range: it looks like the 00.0 > host bridge has bus 00 below it, and 00.1 has at least buses 01-02 > below it. > > And there should be an I/O port range. Your Freescale device doesn't > use any I/O ports, but presumably you could plug in a card that did, > and the BIOS should allocate space for it. > > That's basically all the info we should need to write a driver. There's > not really a good model for one yet because we don't have any kind of > generic PCI host bridge structure. We do have drivers/acpi/pci_root.c > and arch/x86/pci/amd_bus.c, which are two different approaches to this. > amd_bus.c is closer to what we need here, but it's not well integrated > into PCI, so I don't really want to add more stuff like it. > Ok, I've started to write a driver for this, but I can't get anything to change when I run the hotplug. I've tried: 1) add a quirk on the host bridge, like intel_bus.c from the quirk, change the dev->bus->resource[] start, end, flags This totally crashes and burns, it probably isn't what I want to do. 2) call update_res() from amd_bus.c (also used by intel_bus.c) this doesn't make any difference. Reading through the update_res() code, I don't see that it changes anything to do with the PCI subsystem data structures at all... I know you said you didn't want more code similar to amd_bus.c or intel_bus.c, but for the time being I'd like to try this method. Ira -- 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
