Oops report for the week upto Feb 10th 2008

Oops report for the week upto Feb 10th 2008

[Arjan van de Ven]

The http://www.kerneloops.org website collects kernel oops and
warning reports from various mailing lists and bugzillas as well as
with a client users can install to auto-submit oopses.
Below is a top 10 list of the oopses/backtraces collected in the last 7 days.
(Reports prior to 2.6.23 have been omitted in collecting the top 10)

This week, a total of 323 oopses and warnings have been reported,
compared to 110 reports in the previous week.

(This sharp increase is due to Fedora 9 alpha shipping the oops data
collection client in the default install, giving us much wider coverage
in the issues that actual users hit; many thanks to the Fedora project
for this)

With the 2.6.25-rc1 release out, this will be the last report that includes
2.6.23; future reports will only include issues from 2.6.24 and later.


Rank 1: set_dentry_child_flags
	WARN_ON at fs/inotify.c:172 set_dentry_child_flags
	Reported 93 times (116 total reports)
	This is a user triggered WARN_ON in inotify. Sadly inotify seems to be unmaintained.
	More info: http://www.kerneloops.org/search.php?search=set_dentry_child_flags

Rank 2: __ieee80211_rx_handle_packet
	WARN_ON at net/mac80211/rx.c:1704 __ieee80211_rx_handle_packet
	Reported 35 times (82 total reports)
	This is the well known packet alignment issue; fix in progress
	More info: http://www.kerneloops.org/search.php?search=__ieee80211_rx_handle_packet

Rank 3: remove_proc_entry
	WARN_ON at fs/proc/generic.c:736
	Reported 20 times (38 total reports)
	This WARN_ON is there if code tries to remove a non-empty /proc directory.
	Most reports are tainted, pointing at a bug in a binary module.
	Without the WARN_ON improvements from 2.6.25-rc1 it'll be hard to figure out which module is guilty.
	More info: http://www.kerneloops.org/search.php?search=remove_proc_entry

Rank 4: bad_io_access
	WARN_ON at lib/iomap.c:44
	Reported 15 times (21 total reports)
	This is a bug in the ata_bmdma code
	More info: http://www.kerneloops.org/search.php?search=bad_io_access

Rank 5: sysctl_print_path
	Reported 28 times (41 total reports)
	This is a bug in the binary-only madwifi module.
	More info: http://www.kerneloops.org/search.php?search=sysctl_print_path

Rank 6: dma_free_coherent
	WARN_ON at arch/x86/kernel/pci-dma_32.c:66
	Reported 9 times (11 total reports)
	This bug happened with the acx, arcmsr, sym2 and generic ata drivers
	More info: http://www.kerneloops.org/search.php?search=dma_free_coherent

Rank 7: _spin_unlock_irqrestore
	Soft lockup
	Reported 7 times (11 total reports)
	In all cases it happened via scsi_dispatch_cmd
	More info: http://www.kerneloops.org/search.php?search=_spin_unlock_irqrestore

Rank 8: uart_flush_buffer
	The well known bluetooth tty flush bug
	A fix for this bug has been merged into 2.6.25-rc1
	Reported 6 times (67 total reports)
	More info: http://www.kerneloops.org/search.php?search=uart_flush_buffer

Rank 9: mark_buffer_dirty
	WARN_ON at fs/buffer.c:1169
	This indicates that a non-uptodate buffer is marked dirty.
	This can lead to data corruption!
	Reported 5 times (12 total reports) - Only seen since 2.6.24-rc6
	Usually happens during umount()
	More info: http://www.kerneloops.org/search.php?search=mark_buffer_dirty

Rank 10: sysfs_add_one
	WARN_ON at fs/sysfs/dir.c:424 (duplicate filename)
	Reported 5 times (57 total reports)
	Seems to happen mostly with ALSA
	More info: http://www.kerneloops.org/search.php?search=sysfs_add_one

Re: Oops report for the week upto Feb 10th 2008

[Nick Piggin]

On Monday 11 February 2008 11:35, Arjan van de Ven wrote:
> The http://www.kerneloops.org website collects kernel oops and
> warning reports from various mailing lists and bugzillas as well as
> with a client users can install to auto-submit oopses.
> Below is a top 10 list of the oopses/backtraces collected in the last 7
> days. (Reports prior to 2.6.23 have been omitted in collecting the top 10)
>
> This week, a total of 323 oopses and warnings have been reported,
> compared to 110 reports in the previous week.
>
> (This sharp increase is due to Fedora 9 alpha shipping the oops data
> collection client in the default install, giving us much wider coverage
> in the issues that actual users hit; many thanks to the Fedora project
> for this)
>
> With the 2.6.25-rc1 release out, this will be the last report that includes
> 2.6.23; future reports will only include issues from 2.6.24 and later.
>
>
> Rank 1: set_dentry_child_flags
> 	WARN_ON at fs/inotify.c:172 set_dentry_child_flags
> 	Reported 93 times (116 total reports)
> 	This is a user triggered WARN_ON in inotify. Sadly inotify seems to be
> unmaintained. More info:
> http://www.kerneloops.org/search.php?search=set_dentry_child_flags

I was never able to trigger this or get anyone to reliably trigger it with
a debug patch in. Which is why it has taken so long to fix. It looks like
kde4 is triggering this big rash of new reports.

Anyway, I have fixed a race or two and removed that warning code (which was
also a little racy). So I think that should be OK.


> Rank 9: mark_buffer_dirty
> 	WARN_ON at fs/buffer.c:1169
> 	This indicates that a non-uptodate buffer is marked dirty.
> 	This can lead to data corruption!
> 	Reported 5 times (12 total reports) - Only seen since 2.6.24-rc6
> 	Usually happens during umount()
> 	More info: http://www.kerneloops.org/search.php?search=mark_buffer_dirty

That's interesting.

Re: Oops report for the week upto Feb 10th 2008

[Alexey Dobriyan]

On Sun, Feb 10, 2008 at 04:35:51PM -0800, Arjan van de Ven wrote:
> Rank 3: remove_proc_entry
> 	WARN_ON at fs/proc/generic.c:736
> 	Reported 20 times (38 total reports)
> 	This WARN_ON is there if code tries to remove a non-empty /proc 
> 	directory.
> 	Most reports are tainted, pointing at a bug in a binary module.
> 	Without the WARN_ON improvements from 2.6.25-rc1 it'll be hard to 
> 	figure out which module is guilty.

Easy, I'll submit patch RSN.

Re: Oops report for the week upto Feb 10th 2008

[Arjan van de Ven]

Alexey Dobriyan wrote:
> On Sun, Feb 10, 2008 at 04:35:51PM -0800, Arjan van de Ven wrote:
>> Rank 3: remove_proc_entry
>> 	WARN_ON at fs/proc/generic.c:736
>> 	Reported 20 times (38 total reports)
>> 	This WARN_ON is there if code tries to remove a non-empty /proc 
>> 	directory.
>> 	Most reports are tainted, pointing at a bug in a binary module.
>> 	Without the WARN_ON improvements from 2.6.25-rc1 it'll be hard to 
>> 	figure out which module is guilty.
> 
> Easy, I'll submit patch RSN.

to fix it or to print modules?
(the later is already done in 2.6.25-rc1, just not in 2.6.24 yet)