Re: [Patch v2 2/2] mm/khugepaged: remove definition of struct khugepaged_mm_slot

[SeongJae Park <sj@kernel.org>]

On Sat, 20 Sep 2025 12:29:58 +0000 Wei Yang <richard.weiyang@gmail.com> wrote:

> On Sat, Sep 20, 2025 at 04:52:33AM -0700, SeongJae Park wrote:
> [...]
> >
> >On latest mm-new tree, I am getting below error while building UML mode kernel
> >for kunit.  And 'git bisect' points me this patch.  I'm not familiar with this
> >code and have no time to dive deep for now, so reporting first.
> 
> Thanks for reporting.

Thank you for fast reply!

> 
> >
> >Oops: general protection fault, probably for non-canonical adI
> >[  356.456907] CPU: 34 UID: 0 PID: 309 Comm: khugepaged Not tainted 6.17.0-rc4+ #370 PREEMPT(voluntary)
> >[  356.457702] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-4.el9 04/01/2014
> >[  356.458484] RIP: 0010:collect_mm_slot (mm/khugepaged.c:1427)
> 
> If my understanding is correct, the error happens in following code flow:
> 
>   khugepaged_scan_mm_slot()
>     mm = slot->mm;                (1)
>     collect_mm_slot()
>       mm = slot->mm;              (2)
> 
> Looks the reason is slot is NULL at (2), but we have already accessed it at (1).
> 
> Not find the cause yet.
> 
> Would you mind sharing your UML config and test step? So that I can reproduce
> it.

To reproduce the issue, I do below:

    $ ./tools/testing/kunit/kunit.py run --kunitconfig mm/damon/tests

Then the OOPs happens a few seconds aftr the kunit run, or during the
kunit-purpose UML build.  The issue sometimes doesn't happen or take time more
than seconds.  Since the code is related with khugepaged, apparently there are
some timing factors.  But I was able to reproduce about 3/4 times.

It was also reproducible when I build and install kernel on my QEMU machine
using 'make install'.


Thanks,
SJ

[...]