Re: kmemcheck: Fatal error; system fails to boot when kmemcheck enabled

[Sami Liedes <sami.liedes@iki.fi>]

[-- Attachment #1: Type: text/plain, Size: 1736 bytes --]

On Sat, Jul 14, 2012 at 04:09:02AM +0300, Sami Liedes wrote:
> > Another thing you can try is try to figure out when things went wrong.
> > If you test an old kernel like 3.0 or so -- and find that kmemcheck
> > works, that's great, it means we have a way to narrow it down further.
> > I may also try your config here to see if I can reproduce it.
> 
> I'll probably test 3.0 and could even try to git bisect within a few
> days.

I tested with 3.0.36. It does not boot either.

Also, I tested with both 3.0.36 and 3.4.4 with kmemcheck=2. With that
parameter, both boot fine. Curiously, when booting with kmemcheck=2, I
can see absolutely no messages about kmemcheck, just as with
kmemcheck=0, and the machine boots fine. The crashes happen only with
kmemcheck=1. The speed suggests to me that with kmemcheck=2, kmemcheck
is actually possibly not enabled at all for some reason.

One curious result with 3.0.36: With nosmp, it panics so early that
I'm unable to capture the dmesg (I may experiment with getting some
other video mode than 80x25; not sure if it's possible...). *Without*
nosmp, it boots much further, and I can actually get the dmesg via
netconsole. I tried this a couple of times, with always the same
result.

Dmesg from 3.0.36 without nosmp:

   http://www.niksula.hut.fi/~sliedes/kmemcheck/nc.log.3.0.36

A photo of the (final portion of the) backtrace for 3.0.36 with nosmp:

   http://www.niksula.hut.fi/~sliedes/3.0.36.kmemcheck.nosmp.bt.jpg

Dmesg from 3.4.4 with kmemcheck=2:

   http://www.niksula.hut.fi/~sliedes/kmemcheck/nc.log.3.4.4.kmemcheck-2

Dmesg from 3.0.36 with kmemcheck=2:

   http://www.niksula.hut.fi/~sliedes/kmemcheck/nc.log.3.0.36.kmemcheck-2

	Sami

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]