[Bug 18252] spinlock lockup in __make_request <- submit_bio <- ondemand_readahead

[Stefan Richter <stefanr@s5r6.in-berlin.de>]

Full quote for lkml:

bugzilla-daemon@bugzilla.kernel.org wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=18252
> 
>            Summary: spinlock lockup in __make_request <- submit_bio <-
>                     ondemand_readahead
>            Product: IO/Storage
>            Version: 2.5
>     Kernel Version: 2.6.36-rc3
>           Platform: All
>         OS/Version: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: Block Layer
>         AssignedTo: axboe@kernel.dk
>         ReportedBy: stefanr@s5r6.in-berlin.de
>         Regression: No
> 
> 
> Created an attachment (id=29562)
>  --> (https://bugzilla.kernel.org/attachment.cgi?id=29562)
> BUG screenshot
> 
> After a week uptime of 2.6.36-rc3 (I ran 2.6.35 before that),

Almost two weeks uptime actually.

> I was greeted by a black screen of death today in the morning:
> 
> (see screenshot in attachment; partial transcript:)
> 
> sending NMI to all CPUs:
> BUG: soinlock lockup on CPU#0, ktorrent/4313, ffff8802...
> PID: 4313, comm: ktorrent Tainted: G  M D W   2.6.36-rc3 #3
> Call Trace:
>  [...] do_raw_spin_lock+0x118/0x147
>  [...] _raw_spin_lock_irq+0x44/0x49
>  [...] ? __make_request+0x5c/0x400
>  [...] __make_request+0x5c/0x400
>  [...] generic_make_request+0x23a/0x2a9
>  [...] submit_bio+0xad/b6
>  [...] mpage_bio_submit...
>  [...] do_mpage_readpage...
>  [...] ? get_parent_ip...
>  [...] ? sub_preempt_count...
>  [...] ? __lru_cache_add...
>  [...] mpage_readpages...
>  [...] ? ext4_get_block...
>  [...] ? __alloc_pages_nodemask...
>  [...] ? ext4_get_block...
>  [...] ext4_readpages...
>  [...] __do_page_cache_readahead...
>  [...] ? __do_page_cache_readahead...
>  [...] ra_submit...
>  [...] ondemand_readahead...
> 
> This is a system with Phenom II x4 and Radeon graphics.  Since kernel mode
> setting is fairly new for radeon, it is possible that the lockup happened with
> earlier kernels too but simply ended in a lockup without trace dump to the
> screen.  IOW, it is not clear to me whether this is a regression or not.
> 
> The bug happened while kaffeine wrote an MPEG 2 TS to the same filesystem from
> which ktorrent was reading.  Of course this kind of commonplace workload
> happened without problem two or three times before during the week in which I
> ran 2.6.36-rc3.
> 

(The screenshot is a bit large, hence I reported in bugzilla instead of the list.)

The kernel taint was due to prior apparently unrelated lockdep report, bug
17752 "2.6.36-rc3: inconsistent lock state (iprune_sem, shrink_icache_memory".
 And there were three machine check events ten days ago due to corrected ECC
memory errors.
-- 
Stefan Richter
-=====-==-=- =--= -=-==
http://arcgraph.de/sr/