Re: [PATCH 0/1] aoe: ensure we initialise the request_queue correctly

["Bruno Prémont" <bonbons@linux-vserver.org>]

On Mon, 24 August 2009 Ed Cashin <ecashin@coraid.com> wrote:
> On Fri Aug 21 12:41:59 EDT 2009, apw@canonical.com wrote:
> > We have been seeing oopses in very recent kernels when using the
> > AOE driver. When attempting to mount remote devices we get a
> > warning from the kobject layer:
> > 
> >   [ 2645.959090] kobject '<NULL>' (ffff880059ca22c0): tried to add
> > 	an uninitialized object, something is seriously wrong.
> > 
> > Looking at the driver it seems to have always had an embedded
> > request_queue and it is this that is throwing the error.  It
> > appears the intent is tha these would be allocated and released
> > using helpers, and the lack of these leaves the object
> > uninitialised and throws the error.  It is unclear how this could
> > ever have worked.
> 
> This aoe driver does not handle I/O requests but provides its own
> make_request function to blk_queue_make_request and handles bios
> instead.
> 
> The reason I was interested in a git bisect is that I suspect that
> before nobody was interested in the request_queue in the aoe driver
> until recent changes to kobject code.  A couple people have indicated
> that they'd be doing such a git bisect, so if anybody has done that,
> please Cc me.
> 
I finished bisecting the NULL object and ended up at this commit:
  cd43e26f071524647e660706b784ebcbefbd2e44

  block: Expose stacked device queues in sysfs

  Currently stacking devices do not have a queue directory in sysfs.
  However, many of the I/O characteristics like sector size, maximum
  request size, etc. are queue properties.

  This patch enables the queue directory for MD/DM devices.  The elevator
  code has been modified to deal with queues that do not have an I/O
  scheduler.

  Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
  Signed-off-by: Jens Axboe <jens.axboe@oracle.com>


This seems to generate /sys/block/$device/queue and its contents for
everyone who is using queues, not just for those queues that have a
non-NULL queue->request_fn.

Bruno