On Mon 13-09-21 21:32:25, Shakeel Butt wrote:

On Mon, Sep 13, 2021 at 1:37 AM Michal Hocko [off-list ref] wrote:

quoted

[...]

quoted

quoted

However Shakeel Butt pointed that there are much more popular objects
with the same life time and similar memory consumption, the accounting
of which was decided to be rejected for performance reasons.

Is there any measurable performance impact in this particular case?

I don't think there was any regression report or any performance
evaluation. Linus raised the concern on the potential performance
impact. I suggested to backoff for this allocation for now and revisit
again once we have improved the memcg accounting for kernel memory.

I am fine with the change, I am just not satisfied with the
justification. It is not really clear what the intention is except that
Linus wanted it. I have already asked Vasily to provide more
explanation. E.g. this part really begs for clarification
"
This object can consume up to 2 pages, syscall is sleeping one,
size and duration can be controlled by user, and this allocation
can be repeated by many thread at the same time.
"

It sounds like a problem, except it is not because? A worst case
scenario evaluation would be beneficial for example

Thanks!

-- 
Michal Hocko
SUSE Labs