Re: CONFIG_DM_MQ_DEFAULT makes my kernel unhappy..

[Mike Snitzer <snitzer@redhat.com>]

On Sun, Apr 26 2015 at 10:16am -0400,
Christoph Hellwig <hch@infradead.org> wrote:

> On Sat, Apr 25, 2015 at 08:37:32AM -0700, Christoph Hellwig wrote:
> > I'll give it a spin.
> > 
> > > What were you actually testing when this happened?  Were you failing
> > > paths under IO load or something?
> > 
> > This happened on the initial udev-direct scan of the devices.
> > 
> > > Also, which upstream commit was your tree based on?  Did you have any
> > > non-upstream patches applied that I should be aware of?
> > 
> > Latest Linus tree as of this morning.
> 
> Seems like the first warning goes away without CONFIG_DEBUG_OBJECTS.
> Note that in either way I/O over dm-mpath actually works fine.

I'm not seeing any warnings with use_blk_mq=Y but I don't have
CONFIG_DEBUG_OBJECTS set.  I expected to still see the later errors you
saw though.  Unfortunately (or fortunately? ;) I'm not... with 4.1-rc1
on a pretty fast bare-metal system.

Since you can easily reproduce you might need to chase down the problem.

BTW, I'm liking LIO and these lio-utils for quick test setups.. much
more useful than scsi_debug!

For now I'll move on to reviewing/testing your v2 patch.