On Tue 03-09-24 14:34:05, Yafang Shao wrote: > On Mon, Sep 2, 2024 at 5:09 PM Michal Hocko <mhocko@xxxxxxxx> wrote: > > > > On Mon 02-09-24 17:01:12, Yafang Shao wrote: > > > > I really do not see why GFP_NOFAIL should be any special in this > > > > specific case. > > > > > > I believe there's no way to stop it from looping, even if you > > > implement a sophisticated user space OOM killer. ;) > > > > User space OOM killer should be helping to replenish a free memory and > > we have some heuristics to help NOFAIL users out with some portion of > > memory reserves already IIRC. So we do already give them some special > > treatment in the page allocator path. Not so much in the reclaim path. > > When setting GFP_NOFAIL, it's important to not only enable direct > reclaim but also the OOM killer. In scenarios where swap is off and > there is minimal page cache, setting GFP_NOFAIL without __GFP_FS can > result in an infinite loop. In other words, GFP_NOFAIL should not be > used with GFP_NOFS. Unfortunately, many call sites do combine them. This is the case with GFP_NOFS on its own already. NOFAIL is no different and both will be looping for ever. We heavily rely on kswapd or other GFP_KERNEL's direct reclaim to allow for forward progress. Unfortunatelly we haven't really found a better way to deal with NOFS only/predominant workloads. -- Michal Hocko SUSE Labs
