On Tue 27-08-24 17:29:34, Uladzislau Rezki wrote: > On Tue, Aug 27, 2024 at 03:37:38PM +0200, Michal Hocko wrote: > > On Tue 27-08-24 14:47:30, Uladzislau Rezki wrote: > > > On Tue, Aug 26, 2024 at 08:49:35AM +0200, Michal Hocko wrote: > > [...] > > > > > 2. High-order allocations. Do you think we should not care much about > > > > > it when __GFP_NOFAIL is set? Same here, there is a fallback for order-0 > > > > > if "high" fails, it is more likely NO_FAIL succeed for order-0. Thus > > > > > keeping NOFAIL for high-order sounds like not a good approach to me. > > > > > > > > We should avoid high order allocations with GFP_NOFAIL at all cost. > > > > > > > What do you propose here? Fail such request? > > > > We shouldn't have any hard requirements for higher order allocations in the vmalloc > > right? In other words we can always fallback to base pages. > > > We always drop NOFAIL for high-order, if it fails we fall-back to > order-0. I got the feeling that you wanted just bail-out fully if > high-order and NOFAIL. Nope. We should always fall back to order 0 for both NOFAIL and regular vmalloc allocations. -- Michal Hocko SUSE Labs
