Re: [PATCH block/for-3.3/core] block: an exiting task should be allowed to create io_context

[Tejun Heo <tj@kernel.org>]

Hello, again.

Adding Shaohua Li as he fixed a similar issue in 4a0b75c7d0 "block,
cfq: fix empty queue crash caused by request merge".  The original
thread can be read from

  http://thread.gmane.org/gmane.linux.kernel.next/20064/focus=20159

On Tue, Jan 03, 2012 at 09:35:00AM -0800, Tejun Heo wrote:
> Happy new year, guys.
> 
> On Wed, Dec 28, 2011 at 01:19:18PM -0800, Tejun Heo wrote:
> > > On Wed, Dec 28, 2011 at 9:50 AM, Hugh Dickins <hughd@google.com> wrote:
> > > > "It's the tmpfs swapping test that I've been running, with variations,
> > > > for years.  System booted with mem=700M and 1.5G swap, two repetitious
> > > > make -j20 kernel builds (of a 2.6.24 kernel: I stuck with that because
> > > > the balance of built to unbuilt source grows smaller with later kernels),
> > > > one directly in a tmpfs, the other in a 1k-block ext2 (that I drive with
> > > > ext4's CONFIG_EXT4_USE_FOR_EXT23) on /dev/loop0 on a 450MB tmpfs file."
> > > >
> > > > I doubt much of that (quoted from an older mail to someone else about
> > > > one of the many other bugs it's found) is relevant: maybe just plenty
> > > > of file I/O and swapping.
> > > 
> > > Plain -j4 build isn't triggering anything.  I'll try to replicate the condition.
> > 
> > It's not too reliable but I can reproduce it with -j 22 allmodconfig
> > build inside qemu w/ 512M of memory.  I'll try to find out what's
> > going on.
> 
> I misread the code, the problem is empty cfqq on the cfq prio tree.  I
> don't think this is caused by recent io_context changes.  It looks
> like somebody is forgetting to remove cfqq from the dispatch prio tree
> after emptying a cfqq by removing a request from it.  Jens, any ideas?

That should have been service tree.  I couldn't find more missing
removals other than the one Shaohua's patch already fixed.  Close
cooperator selection in cfq_select_queue() seems suspicious tho.  I
can't see what prevents it from returning an empty coopeator cfqq.
I'm trying to verify whether that's the case.  Will update when I know
more.

Thanks.

-- 
tejun