On 04/20/2018 04:53 PM, Christopher Lameter wrote: > On Thu, 19 Apr 2018, Michal Hocko wrote: > >> Overriding __GFP_NORETRY is just a bad idea. It will make the semantic >> of the flag just more confusing. Note there are users who use >> __GFP_NORETRY as a way to suppress heavy memory pressure and/or the OOM >> killer. You do not want to change the semantic for them. > > Redoing the allocation after failing a large order alloc is a retry. I > would say its confusing right now because a retry occurs despite > specifying GFP_NORETRY, > >> Besides that the changelog is less than optimal. What is the actual >> problem? Why somebody doesn't want a fallback? Is there a configuration >> that could prevent the same? > > The problem is that SLUB does not honor GFP_NORETRY. The semantics of > GFP_NORETRY are not followed. The caller might want SLUB to try hard to get that high-order page that will minimize memory waste (e.g. 2MB page for 3 640k objects), and __GFP_NORETRY will kill the effort on allocating that high-order page. Thus, using __GPF_NORETRY for "please give me a space-optimized object, or nothing (because I have a fallback that's better than wasting memory, e.g. by using 1MB page for 640kb object)" is not ideal. Maybe __GFP_RETRY_MAYFAIL is a better fit? Or perhaps indicate this situation to SLUB with e.g. __GFP_COMP, although that's rather ugly?
