Created attachment 99331 [details] Full dmesg from 3.9-rc5 with patch from bug 56501 On my Sony laptop, if I dock, undock and dock again, acpiphp_glue fails to register interrupt notify handler and then the kernel dereferences the NULL pointer in pci_bus_read_config_dword. The full dmesg (captured in part locally, and in part over netconsole) is attached. The lines starting with /sys/bus/pci/devices were written to /dev/kmsg for debugging another bug, you can ignore them now. Note: there are also failures to assign resources to PCI devices in the dock. They might share the root cause, but are tracked in a different bug #56531. Bugs that were relevant to the same laptop and may contain useful attachments such as dmesgs and acpidump output: * bug #55611 * bug #56501 * bug #56531
Created attachment 99341 [details] Full dmesg from 3.9-rc5 with patch from bug 56501 Oops, Bugzilla could not copy the attachment from another bug. Reattaching.
Sorry, due to a physical mishap, the laptop is now out of order. I will not be able to test any patches, thus closing the bug.
Sorry to hear this. I just managed to re-produce this with a fake DSDT table and some hack code to pci layer on v3.9 and found the problematic code, then I did a git pull and found 3.10 has some big changes to pci hotplug that made my investigation useless since those code are gone. But 3.10 has other problems regarding dock and pci hotplug, I'll probably raise the question to mailing list. Thanks again for all the tests you have done.
I have contacted the authorized Sony service center. They told me that they will fix the laptop once the needed replacement part arrives, and this can take up to three weeks. So maybe not everything is lost - stay tuned. If/when they repair the laptop, I will reopen this bug.
The laptop has been repaired. However, given the refactoring, it is pointless to try to fix the bug in 3.9, and in 3.10 a separate (yet to be reported) bug exists that leads to non-rescanning of the PCI bus after docking, thus possibly hiding this oops.