Hi Carlo & Jan,
On 26/03/2024 17:04, Jan Beulich wrote:
No, we need to get to the root of the issue. Since osstest has hit it quite
easily as it seems, I'm somewhat surprised you didn't hit it in your testing.
In any event, as per my earlier reply, my present guess is that your change
has merely uncovered a previously latent issue elsewhere.
Ok, what about removing PGC_extra in free_heap_pages() before the
mark_page_free() call?
Question is: How would you justify such a change? IOW I'm not convinced
(yet) this wants doing there.
Looking at the code, the flag is originally set in
alloc_domheap_pages(). So I guess it would make sense to do it in
free_domheap_pages().
For PGC_static, I don't think we can reach free_domheap_pages() with the
flag set (the pages should live in a separate pool). So I believe there
is nothing to do for them.
Cheers,
--
Julien Grall
