Re: [libvirt] Re: [PATCH 2/6] host ("node") device enumeration - internal API additions

On Tue, 2008-10-28 at 14:13 +0100, Daniel Veillard wrote: > On Tue, Oct 21, 2008 at 01:54:13PM -0400, David Lively wrote: > > This patch contains the internal API additions. > > Looks fine to me, just a couple of questions: > > > + // TODO: Implement PCI bus devs > > + // TODO: Implement USB bus devs > > Any reason it's not in ? I would expect HAL to report them, no ? I posted a new patch set yesterday that (among other things) removed the PCI_BUS and USB_BUS devices. Currently HAL doesn't report them as anything special, e.g.: udi = '/org/freedesktop/Hal/devices/pci_8086_27cc' info.linux.driver = 'ehci_hcd' (string) info.parent = '/org/freedesktop/Hal/devices/computer' (string) info.product = '82801G (ICH7 Family) USB2 EHCI Controller' (string) info.subsystem = 'pci' (string) info.udi = '/org/freedesktop/Hal/devices/pci_8086_27cc' (string) info.vendor = 'Intel Corporation' (string) linux.hotplug_type = 2 (0x2) (int) linux.subsystem = 'pci' (string) linux.sysfs_path = '/sys/devices/pci0000:00/0000:00:1d.7' (string) pci.device_class = 12 (0xc) (int) pci.device_protocol = 32 (0x20) (int) pci.device_subclass = 3 (0x3) (int) pci.linux.sysfs_path = '/sys/devices/pci0000:00/0000:00:1d.7' (string) pci.product = '82801G (ICH7 Family) USB2 EHCI Controller' (string) pci.product_id = 10188 (0x27cc) (int) pci.subsys_product_id = 22594 (0x5842) (int) pci.subsys_vendor = 'Intel Corporation' (string) pci.subsys_vendor_id = 32902 (0x8086) (int) pci.vendor = 'Intel Corporation' (string) pci.vendor_id = 32902 (0x8086) (int) udi = '/org/freedesktop/Hal/devices/pci_8086_244e' info.parent = '/org/freedesktop/Hal/devices/computer' (string) info.product = '82801 PCI Bridge' (string) info.subsystem = 'pci' (string) info.udi = '/org/freedesktop/Hal/devices/pci_8086_244e' (string) info.vendor = 'Intel Corporation' (string) linux.hotplug_type = 2 (0x2) (int) linux.subsystem = 'pci' (string) linux.sysfs_path = '/sys/devices/pci0000:00/0000:00:1e.0' (string) pci.device_class = 6 (0x6) (int) pci.device_protocol = 1 (0x1) (int) pci.device_subclass = 4 (0x4) (int) pci.linux.sysfs_path = '/sys/devices/pci0000:00/0000:00:1e.0' (string) pci.product = '82801 PCI Bridge' (string) pci.product_id = 9294 (0x244e) (int) pci.subsys_product_id = 0 (0x0) (int) pci.subsys_vendor_id = 0 (0x0) (int) pci.vendor = 'Intel Corporation' (string) pci.vendor_id = 32902 (0x8086) (int) So at this point, I just report them as PCI devices, though of course the PCI bridge will be the parent of PCI devices on that bus, and the USB controller will be the parent of USB devices controlled by it. Any client code that needs to recognize them specifically as PCI or USB bus controllers can use the PCI device attributes for that. Hmmm ... currently I expose the product & vendor info for PCI devices, along with their bus addressing info. But perhaps I should also expose their device_class, protocol, and subclass?? > > + > > +// TODO: virNodeDeviceDefParseString/File/Node for virNodeDeviceCreate Currently I have no need to parse a NodeDevice definition. But virNodeDeviceCreate will need to do that, once it's implemented. (I don't have any NPIV NICs or other such toys to play with right now. Daniel B says he'll implement Create/Destroy since he does.) I also thought it would be better to postpone the parser work until needed since the node definition seems likely to change as we work out the details here.