On 09.05.24 13:59, Mikhail Gavrilov wrote:
On Wed, May 8, 2024 at 10:45 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
"page dumped because: non-NULL mapping"
Is the relevant bit. We are freeing a page, but page->mapping is not
NULL. IIUC, it might happen under memory pressure when reclaiming memory.
It's weird that only you are seeing that, if it would be something
"obvious" I would expect multiple reports :/
Maybe because the problem is really difficult to reproduce (rare
combination of the kernel build option and workload). I even thought
that the problem was fixed because it did not reproduce itself for a
week.
But yesterday when it happened again with the kworker process, I
stopped doubting that it was fixed.
Now I am concerned with the question of how to be as useful as
possible when reproducing bug again?
Do you have the other stracktrace as well?
Maybe triggering memory reclaim (e.g., using "stress" or "memhog") could
trigger it, that might be reasonable to trey. Once we have a reproducer
we could at least bisect.
--
Cheers,
David / dhildenb
