On Wed 30-06-21 11:04:55, Mel Gorman wrote: [...] > I didn't do a full audit to determine what fallout, if any, there is > to ultimately passing __GFP_NOFAIL to kvmalloc although kvmalloc_node > explicitly notes that __GFP_NOFAIL is not supported. Adding Michal Hocko > to the cc to see if he remembers why __GFP_NOFAIL was problematic. This is because there are allocations in the vmalloc path which do not get the full gfp context. E.g. page table allocations. This could be likely handled somewhere at the vmalloc layer and retry but I do not remember anybody would be really requesting the support. -- Michal Hocko SUSE Labs
