On Fri, Apr 26, 2013 at 3:41 PM, Yinghai Lu <yinghai@xxxxxxxxxx> wrote:
> On Fri, Apr 26, 2013 at 2:25 PM, Bjorn Helgaas <bhelgaas@xxxxxxxxxx> wrote:
>> I'm trying to figure out where we call device_unregister() when
>> removing a PCI host bridge. I see the device_register() in the
>> following path:
>>
>> acpi_pci_root_add
>> pci_acpi_scan_root
>> pci_create_root_bus
>> bridge = pci_alloc_host_bridge
>> bridge->dev.release = pci_release_bus_bridge_dev
>> device_register(&bridge->dev)
>> pci_set_host_bridge_release(..., release_pci_root_info, ...)
>>
>> But I don't see any device_unregister() in the remove path:
>>
>> acpi_pci_root_remove
>> pci_remove_root_bus
>> put_device(&host_bridge->dev)
>>
>> pci_release_bus_bridge_dev
>> bridge->release_fn # e.g., release_pci_root_info
>> release_pci_root_info
>> __release_pci_root_info
>> release_resource
>> free_pci_root_info_res
>> teardown_mcfg_map
>>
>> What am I missing?
>
> pci_stop_root_bus
> /* stop the host bridge */
> device_del(&host_bridge->dev);
Hmm... OK, thanks for the pointer. So we don't actually call
device_unregister() itself, but since device_unregister() consists of
nothing more than "device_del(dev); put_device(dev);" I guess we're
doing the equivalent in two pieces:
acpi_pci_root_remove
pci_stop_root_bus
device_del(&host_bridge->dev)
pci_remove_root_bus
put_device(&host_bridge->dev)
This seems unnecessarily difficult to analyze. It would be better if
we had symmetric device_register()/device_unregister() paths, or if
that can't be done, symmetric device_add()/device_del() paths.
Don't bother to generate a quick patch to just replace
device_register() with explicit calls to device_initialize() and
device_add(), though. Maybe someday we can clean up that and more.
Anyway, thanks for fixing my confusion.
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
