Re: BUG: scheduling while atomic: chcpu/6197/0x00000002

[Borislav Petkov <bp@alien8.de>]

On Thu, Jun 28, 2012 at 11:24:56AM +0200, Thomas Sattler wrote:
> > Any chance you could try a non-rt kernel,
> 
> Earlier in this thread I wrote, that 3.4.[23] are
> in fact unaffected. Only 3.4.[234]-rt1[0123] show
> this BUG in dmesg.
> 
> > to rule out an -rt issue?
> 
> Steven Rostedt wrote (w/o CC:LKML) that he believes
> CPU hotplug is still an issue with -rt. So, are you
> still interested to follow the "amd on intel" thing?

I don't think it is an "amd on intel" issue - it is probably the
stack trace dumping code wrongly finding the init_amd symbol due to
CPU_SUP_AMD bring compiled in or so...

Rather, it looks like the issue is what Steve said.

> > Also, looks like your cpu can do 64-bit. Can you
> > try a 64-bit kernel?
> 
> Yes, it's a 32-bit userland on 64-bit hardware. So
> far I didn't compile 64-bit kernels on 32-bit user-
> land but in case you're still interested in the
> result, I'll give it a try.

Well, if you have a 64-bit cpu, you could consider running a 64-bit
kernel for many reasons. The most important one being that 64-bit code
is enjoying orders of magnitude more testing because the world has moved
to 64-bit already. -rt included.

-- 
Regards/Gruss,
Boris.