Re: Couple of kernel tracebacks

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Mon, 2017-08-28 at 08:54 -0400, Bruce Ashfield wrote:
> On 08/26/2017 06:53 PM, Richard Purdie wrote:
> > 
> > Hi Bruce,
> > 
> > We are seeing a few teething issues which seem kernel related on
> > the
> > autobuilder. The x86 lsb build saw this traceback in the logs:
> I'll start running some stress tests and see if I can get anything
> to happen.

Thanks!

> > This has happened on multiple builders and on multiple images
> > (sato,
> > sato-sdk and I think minimal). Could be the new kernel, could be
> > qemu
> > :/. If has occurred on lsb and non-lsb ppc which makes it less
> > kernel
> > version specific I guess. For some reason I keep wanting to blame
> > the
> > IDE drivers but it is using virtio. We never get any backtrace for
> > this, the log just stop dead and then we hit timeouts, it never
> > boots
> > fully in these cases. It stops after:
> It could be the virtio back end interacting in ways that we've
> never hit before.
> 
> I'll take another look at that IDE mess in 4.12 and see if the
> driver is fixable.
> 
> Is there anyway that we could do a few runs with only virtio on
> the 4.12 kernel and confirm that the hang goes away with the
> lsb configuration ? That would definitely point the finger at some
> sort of virtio interaction and force us into that IDE driver for
> a fix.

I did try booting the system with a "CONFIG_IDE is not set" from a
config fragment and confirmed I can turn IDE off at least for ppc and
it still works. I could put that in master-next and test that a bit,
see if things keep working and if any of the hangs occur? Its a pure
guess whether its related to IDE or not at this point...

> FYI: that IDE issue is already logged in kernel.org bugzilla (by
> someone else) and was reported to the mailing list. Neither the
> bug or the post got any attention at all. I also tried to fix the
> code and it is really detailed stuff that is going to take a few
> days of study to actually understand and fix.

Understandable. I can't help wonder if we shouldn't concentrate on on
dropping the IDE bits where we can?

Cheers,

Richard