Re: 3.5.0: BUG: Bad page state in process Compositor pfn:16834f

[Michal Hocko]

On Sat 16-04-16 18:37:53, Frederik Himpe wrote:
> With Linux 3.5.0 I suddenly had Firefox which suddenly seemed
> to get stuck, and then I tried to kill it. I also noticed that
> the ps aux command was hanging. Then I found this in my kernel
> logs: 
> 
> 
> Apr 16 20:21:53 piranha kernel: [27926.414148] BUG: Bad page state in process Compositor  pfn:16834f
> Apr 16 20:21:53 piranha kernel: [27926.414153] page:ffffea0005a0d3c0 count:0 mapcount:0 mapping:          (null) index:0x0
> Apr 16 20:21:53 piranha kernel: [27926.414155] flags: 0x1ffff8000000000()

Flags look weird

> Apr 16 20:21:53 piranha kernel: [27926.414156] page dumped because: page still charged to cgroup
> Apr 16 20:21:53 piranha kernel: [27926.414157] page->mem_cgroup:4000000000000000

And we think that the page is still charged because page->mem_cgroup !=
NULL. The value is not a pointer though. It seems like somebody is
corrupting the struct page. Hard to tell who that might be though from
the available information. This will be nasty to track down I am afraid.
How reproducible is it?

-- 
Michal Hocko
SUSE Labs