Re: [tip:sched/core] [sched/fair] 79104becf4: BUG:kernel_NULL_pointer_dereference,address

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Oct 28, 2025 at 10:30:08AM +0800, Chen, Yu C wrote:
> On 10/27/2025 10:09 PM, Peter Zijlstra wrote:
> > On Mon, Oct 27, 2025 at 03:07:18PM +0100, Peter Zijlstra wrote:
> > > On Mon, Oct 27, 2025 at 02:55:16PM +0100, Peter Zijlstra wrote:
> > > 
> > > > > May I know if you are using the kernel config 0day attached?
> > > > > I found that the config 0day attached
> > > > > (https://download.01.org/0day-ci/archive/20251021/202510211205.1e0f5223-lkp@intel.com/config-6.18.0-rc1-00001-g79104becf42b)
> > > > > has
> > > > > CONFIG_IA32_EMULATION=y
> > > > > CONFIG_IA32_EMULATION_DEFAULT_DISABLED=y
> > > 
> > > Yep, deleting that entry makes it all work.
> > 
> > 'work' might be over stating, it boots and starts trinity, which then
> > promptly (as in a handful of seconds) triggers OOM and dies. Not
> > actually reproducing the NULL deref I was looking for.
> 
> Change the following line in job-script
> export memory='16G'
> to
> export memory='64G'
> ?

Yes, that seems to help.

> I had a try and can reproduce the NULL except at first run:

Took me two runs, but yes, I can see it now.

Anyway, this is two bugs in the robot, can we please fix all this to not
happen again?

 - .config has 32bit disabled while robot provides 32bit images. Clearly
   the actual robot runs 64bit images and the reproduction should
   provide those too.

 - job description is inaccurate in the amount of memory required.

The reproduction steps must exactly match what the real robot runs, not
something else.