W dniu 26.11.2025 o 23:58, John Johansen pisze:
On 11/26/25 02:04, Zygmunt Krynicki wrote:
Hello
As a part of investigation into issue affecting mmap with MAP_HUGETLB
[1], I was thinking if huge pages should have a dedicated mediation
class and be handled similarly to how mqueue was recently added.
yes we want to split it off, it could potentially go into its own class
or potentially share a class with a few other memory operations that we
want to handle better.
What other operations would go there?
The trade-off being whether we want to handle them in a separate feature
abi or not. If we put them in the same class then they need to move
together
Right
In the kernel, `aa_file_perm` function could special case hugetlbfs so
that mmap would not end up using odd (possibly disconnected) paths for
accessing it.
I'd be happy to pick up this work if there is consensus on the general
direction. I would need some help with reviews and guidance along the
way.
I am not opposed to free work, it should likely use the newer task based
pattern/operations as a starting point. I need to get those posted.
Though its probably not going to happen this week.
I'm looking forward to those then.
Best regards
ZK
