On Thu 27-07-23 08:57:03, Alexei Starovoitov wrote: > On Thu, Jul 27, 2023 at 4:45 AM Alan Maguire <alan.maguire@xxxxxxxxxx> wrote: > > > > On 27/07/2023 08:36, Chuyi Zhou wrote: > > > This patchset tries to add a new bpf prog type and use it to select > > > a victim memcg when global OOM is invoked. The mainly motivation is > > > the need to customizable OOM victim selection functionality so that > > > we can protect more important app from OOM killer. > > > > > > > It's a nice use case, but at a high level, the approach pursued here > > is, as I understand it, discouraged for new BPF program development. > > Specifically, adding a new BPF program type with semantics like this > > is not preferred. Instead, can you look at using something like > > > > - using "fmod_ret" instead of a new program type > > - use BPF kfuncs instead of helpers. > > - add selftests in tools/testing/selftests/bpf not samples. > > +1 to what Alan said above and below. > > Also as Michal said there needs to be a design doc with pros and cons. > We see far too often that people attempt to use BPF in places where it > shouldn't be. > If programmability is not strictly necessary then BPF is not a good fit. To be completely honest I am not sure whether BPF is the right fit myself. It is definitely a path to be explored but maybe we will learn not the proper one in the end. The primary reason for considering it though is that there is endless (+1) different policies how to select a victim to kill on OOM. So having an interface to hook into and make that decision sounds very promissing. This might be something very static like EXPORT_SYMBOL that a kernel module can hook into or a more broader BPF callback that could be more generic and therefore allow for easier to define policy AFAIU. -- Michal Hocko SUSE Labs
