Re: [BUG next-20260105] khugepaged: kernel BUG at ./include/linux/xarray.h:1441

[Matthew Wilcox <willy@infradead.org>]

On Mon, Jan 05, 2026 at 10:10:38AM -0800, Calvin Owens wrote:
> On Monday 01/05 at 09:51 -0800, Andrew Morton wrote:
> > That's
> > 
> > static inline struct xa_state *XAS_INVALID(struct xa_state *xas)
> > {
> > 	XA_NODE_BUG_ON(xas->xa_node, xas_valid(xas));
> > 	return xas;
> > }
> > 
> > which was added by Willy's 43b00759f21b ("XArray: Add extra debugging
> > check to xas_lock and friends").  Maybe it found a bug.
> > 
> > I'm not actually sure how 43b00759f21b found its way into linux-next. 
> > It isn't in mainline.

It's in the xarray tree.  git://git.infradead.org/users/willy/xarray.git#main

> > I'm seeing no other changes to xarray.  mm.git's khugepaged does have a
> > bunch of changes which Shivank added, so I'll cc Shivank and hopefully
> > he can spot something?
> > 
> > Can you suggest how others could reproduce this?
> 
> It triggers immediately after boot for me on any QEMU VM, one of them is
> an empty husk that runs nothing but sshd.
> 
> FWIW (not much I guess), I disabled DEBUG_XARRAY, and the first VM has
> been happily running next-20260105 that way for about an hour. The
> workload is seeding distro install ISOs with transmission.

It's checking for something which _could_ be a problem.  So it'll catch
any sloppy uses, and doesn't necessarily mean it'll be a problem.
It may be as simple as moving a call to xas_reset() or adding one
(see the other fixes in 43b00759f21b particularly the ones which affect
the test suite).