Re: [PATCH v2 14/23] mm: allow specifying custom oom constraint for BPF triggers

[Roman Gushchin]

Michal Hocko <mhocko@xxxxxxxx> writes:

> On Mon 27-10-25 16:21:57, Roman Gushchin wrote:
>> Currently there is a hard-coded list of possible oom constraints:
>> NONE, CPUSET, MEMORY_POLICY & MEMCG. Add a new one: CONSTRAINT_BPF.
>> Also, add an ability to specify a custom constraint name
>> when calling bpf_out_of_memory(). If an empty string is passed
>> as an argument, CONSTRAINT_BPF is displayed.
>
> Constrain is meant to define the scope of the oom handler but to me it
> seems like you want to specify the oom handler and (ab)using scope for
> that. In other words it still makes sense to distinguesh memcg, global,
> mempolicy wide OOMs with global vs. bpf handler, right?

I use the word "constraint" as the "reason" why an OOM was declared (in
other words which constraint was violated). And memcg vs global define
the scope. Right now the only way to trigger a memcg oom is to exceed
the memory.max limit. But with bpf oom there will others, e.g. exceed a
certain PSI threshold. So you can have different constraints violated
within the same scope.