[Bug 215807] Bad page state in process systemd-udevd

[Matthew Wilcox]

On Tue, Apr 12, 2022 at 02:08:04PM -0700, Andrew Morton wrote:
> hm, that's my third `bad page' report in three emails.  But I'm not
> seeing a pattern - this one might be a DRM thing.

I noticed it was an order-9 page and was set to take responsibility
for it, but it's clearly not a filesystem page, but a DRM page.
Let me help decode this for the benefit of the DRM folks.

> > [    8.453363] amdgpu: Topology: Add CPU node
> > [    8.467169] BUG: Bad page state in process systemd-udevd  pfn:11b403
> > [    8.467180] fbcon: Taking over console
> > [    8.467188] page:000000002cc06944 refcount:0 mapcount:0
> > mapping:0000000000000000 index:0x3 pfn:0x11b403
> > [    8.467195] head:00000000aa25e58e order:9 compound_mapcount:0
> > compound_pincount:0
> > [    8.467198] flags: 0x17ffffc0010000(head|node=0|zone=2|lastcpupid=0x1fffff)
> > [    8.467205] raw: 0017ffffc0000000 fffff2da846d0001 fffff2da846d00c8
> > 0000000000000000
> > [    8.467208] raw: 0000000000000000 0000000000000000 00000000ffffffff
> > 0000000000000000
> > [    8.467210] head: 0017ffffc0010000 0000000000000000 dead000000000122
> > 0000000000000000
> > [    8.467212] head: 0000000000000000 0000000000000000 00000000ffffffff
> > 0000000000000000
> > [    8.467214] page dumped because: corrupted mapping in tail page

Tail pages are all supposed to have page->mapping == TAIL_MAPPING
(0x400 + POISON_POINTER_DELTA).  This one has page->mapping == NULL.
I say "all", but tail pages 1 and 2 aren't checked because those
get other useful things stored in them instead.  This is tail page
number 3, so it's the first one checked.

So DRM has probably been overly enthusiastic about cleaning up
something.  Hope that's helpful!