Created attachment 14682 [details]
acpidump output

does 2.6.24 work?

I did not test 2.6.24. 2.6.24-rc8 was OK and since there are no big changes between 2.6.24 and 2.6.24-rc8, I can assume that 2.6.24 works as well. But if you want me to check it to be sure, I can do that but it takes some time..

2.6.24 is OK, at least it does not happen every time.

This entry is being used for tracking a regression from 2.6.24.  Please don't
close it until the problem is fixed in the mainline.

Regressions list annotation:
Handled-By : Len Brown <lenb@kernel.org>

I did not see this oops in 2.6.25-rc1, but I did not huge testing. I reboot often due to various other problems so I do not highly utilize undocking before suspend as before.

well, in 2.6.25-rc2-git2, it happened again. It seems that it does not happen always, just sometimes.

It looks like this trace precedes the undocking oops. I.e., I can see this trace and after it, undocking produces the oops mentioned above.

[86836.756886] ACPI: \_SB_.GDCK - docking
[83791.360019] ata4.00: configured for UDMA/33
[86837.177682] acpi IBM0079:01: Suspicious device_add during suspend
[86837.177689] Pid: 71, comm: kacpi_notify Not tainted 2.6.25-rc2-git2 #6
[86837.177693]
[86837.177694] Call Trace:
[86837.177705]  [<ffffffff80397070>] pm_sleep_lock+0x10/0x20
[86837.177717]  [<ffffffff803905ac>] device_add+0x5c/0x590
[86837.177723]  [<ffffffff80359b83>] acpi_ut_release_mutex+0x5f/0x63
[86837.177730]  [<ffffffff8035b9c4>] acpi_bus_data_handler+0x0/0x1
[86837.177737]  [<ffffffff8035c8d8>] acpi_add_single_object+0xafa/0xcc1
[86837.177745]  [<ffffffff8028ef64>] kmem_cache_free+0x14/0xb0
[86837.177754]  [<ffffffff80342cb2>] acpi_os_release_object+0x9/0xd
[86837.177761]  [<ffffffff80358fea>] acpi_ut_update_ref_count+0x50/0x9d
[86837.177768]  [<ffffffff80359112>] acpi_ut_update_object_reference+0xdb/0x136
[86837.177776]  [<ffffffff8035ccb1>] acpi_bus_add+0x1c/0x32
[86837.177783]  [<ffffffff8036025c>] hotplug_dock_devices+0xec/0x117
[86837.177788]  [<ffffffff8035b9c4>] acpi_bus_data_handler+0x0/0x1
[86837.177794]  [<ffffffff80342eb6>] acpi_os_execute_deferred+0x0/0x2c
[86837.177801]  [<ffffffff803609fc>] dock_notify+0x7e/0xcb
[86837.177807]  [<ffffffff80348819>] acpi_ev_notify_dispatch+0x57/0x60
[86837.177813]  [<ffffffff80342ed9>] acpi_os_execute_deferred+0x23/0x2c
[86837.177820]  [<ffffffff8024630f>] run_workqueue+0xbf/0x160
[86837.177826]  [<ffffffff80246d50>] worker_thread+0x0/0x100
[86837.177831]  [<ffffffff80246d50>] worker_thread+0x0/0x100
[86837.177837]  [<ffffffff80246def>] worker_thread+0x9f/0x100
[86837.177844]  [<ffffffff80249f50>] autoremove_wake_function+0x0/0x30
[86837.177850]  [<ffffffff80246d50>] worker_thread+0x0/0x100
[86837.177856]  [<ffffffff80246d50>] worker_thread+0x0/0x100
[86837.177860]  [<ffffffff80249b5b>] kthread+0x4b/0x80
[86837.177868]  [<ffffffff8020d128>] child_rip+0xa/0x12
[86837.177874]  [<ffffffff80249b10>] kthread+0x0/0x80
[86837.177879]  [<ffffffff8020d11e>] child_rip+0x0/0x12
[86837.177883]
[86837.177885] ACPI: Error adding device IBM0079:01<7>PM: Writing back config space on device 0000:03:00.0 at offset 1 (was 100106, writing 100102)

Does it happen during a suspend?

The stack dump would be easier to interpret if the kernel was build with CONFIG_FRAME_POINTER.  Apparently acpi_add_single_object() is the culprit, and it was called from within a workqueue.  It may have been a coincidence that the workqueue was running during a suspend.

Created attachment 14961 [details]
Patch removing the locking of devices from the suspend core

Lukas, can you please test with the attached patch applied?

To any ACPI experts reading this bug report:  Is it truly necessary to register a new device (acpi_add_single_object) while the system is suspending or resuming?  

Would it be okay to block the device_add() call until after the resume is finished, instead of failing it?

Created attachment 14964 [details]
Patch removing the locking of devices from the suspend core (updated)

The previously attached patch was incomplete.  Please test this one instead.

(In reply to comment #9)
> Does it happen during a suspend?
> 

comment #8 happens during resume. (I suspend at home and resume at work in a dock station)

(why the hell I did not receive any mails as this bug was updated?)

Created attachment 14965 [details]
Patch removing the locking of devices from the suspend core (updated 2x)

Both previously attached patches were incomplete.

(In reply to comment #10)
> It may have been a coincidence that the
> workqueue was running during a suspend.

I think it is because it does not happen every time.

(In reply to comment #14)
> (In reply to comment #9)
> > Does it happen during a suspend?
> > 
> comment #8 happens during resume. (I suspend at home and resume at work in a
> dock station)

Ah, thanks, that's important.

> (why the hell I did not receive any mails as this bug was updated?)

Bugzilla problem, I guess.

Regressions list annotation:
Patch : http://marc.info/?l=linux-acpi&m=120389632114090&w=2 

Fixed by: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=7a8d37a37380e2b1500592d40b7ec384dbebe7a0