Hi Darrick, Sorry for the late answer. On Fri, 10 Oct 2008 10:54:15 -0700, Darrick J. Wong wrote: > On Thu, Oct 09, 2008 at 01:43:53PM +0200, Jean Delvare wrote: > > > sure why the AMB device would be any different. How comes that the PCI > > devices know the AMB memory address if they aren't related? If the PCI > > devices do not own this memory mapping, who does? > > On _most_ of my systems it's listed in the ACPI DSDT as one of the > "system board resources" (SBD1). Unfortunately, this SBD1 object lists > a bunch of memory regions and I/O ports, without any indication of what > each of those regions is for, other than "not for general OS use". It's > a little confusing here--the PCI config register has ultimate control > over where the region lives, but ACPI claims it as part of its own > system board "do not use" map. ACPI being weird and inconsistent? Wow, that's totally news to me! ;) > I suppose it wouldn't be hard to do this with a pci_device_id and > MODULE_DEVICE_TABLE instead of a direct MODULE_ALIAS. But since I only > need the PCI device for a brief moment I'm trying to avoid bloating the > driver with all the pci_driver support code. I'm not sure why a pci driver would be more code than a platform driver looping over pci_get_device(). In fact I suspect it would be less code. And if we are going to auto-load the driver based on PCI devices, then it makes a lot of sense for i5k_amb to be a pci driver. Back to the initial object of the patch: please do not hard-code the PCI device aliases. No other driver does this. Instead, please define the IDs in a proper pci_device_id null-terminated array and call MODULE_DEVICE_TABLE(pci, x) on it. This is easier to read and not subject to break if the format of the pci aliases ever changes. Thanks, -- Jean Delvare
