On 02/03/23 14:34, Valentin Schneider wrote:
> On 02/03/23 13:56, Michal Hocko wrote:
>> What I am trying to evaluate here is whether it makes sense to support
>> and maintain a non-trivial code for something that might be working
>> sub-optimally or even not working properly in some corner cases. The
>> price for the maintenance is certainly not free.
>
> That's also what I'm trying to make sense of. Either way this will be
> frankenkernel territory, the cgroupv1 ship has already sailed for upstream
> IMO.
So, if someone ends up in a similar situation and considers kludging those
notifications back in:
Don't.
Have a look at:
memcg_check_events()
`\
mem_cgroup_threshold()
`\
__mem_cgroup_threshold()
`\
eventfd_signal()
Having IRQs off, percpu reads, and the the eventfd signal is a nice recipe
for disaster.
The OOM notification sits outside of that, but any other memcg notification
will cause pain.
