2020년 3월 23일 (월) 오후 12:59, Baoquan He <bhe@xxxxxxxxxx>님이 작성: > > On 03/23/20 at 12:50pm, Joonsoo Kim wrote: > > Hello, Baoquan. > > > > 2020년 3월 20일 (금) 오후 7:30, Baoquan He <bhe@xxxxxxxxxx>님이 작성: > > > > > > > > > On 03/20/20 at 05:32pm, js1304@xxxxxxxxx wrote: > > > > From: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx> > > > > > > > > Currently, we use the zone index of preferred_zone which represents > > > > the best matching zone for allocation, as classzone_idx. It has > > > > a problem on NUMA systems when the lowmem reserve protection exists > > > > for some zones on a node that do not exist on other nodes. > > > > > > > > In NUMA system, it can be possible that each node has different populated > > > > zones. For example, node 0 could have DMA/DMA32/NORMAL/MOVABLE zone and > > > > node 1 could have only NORMAL zone. In this setup, allocation request > > > > initiated on node 0 and the one on node 1 would have different > > > > classzone_idx, 3 and 2, respectively, since their preferred_zones are > > > > different. If the allocation is local, there is no problem. However, > > > > if it is handled by the remote node due to memory shortage, the problem > > > > would happen. > > > > > > Hi Joonsoo, > > > > > > Not sure if adding one sentence into above paragraph would be make it > > > easier to understand. Assume you are only talking about the high_zoneidx > > > is MOVABLE_ZONE with calculation of gfp_zone(gfp_mask), since any other > > > case doesn't have this problem. Please correct me if I am wrong. > > > > You're right. This example is for the allocation request with > > gfp_zone(gfp_mask), > > MOVABLE_ZONE. > > > > > In NUMA system, it can be possible that each node has different populated > > > zones. For example, node 0 could have DMA/DMA32/NORMAL/MOVABLE zone and > > > node 1 could have only NORMAL zone. In this setup, if we get high_zoneidx > > > as 3 (namely MOVABLE zone), with gfp_zone(gfp_mask), allocation request > > > initiated on node 0 and the one on node 1 would have different > > > classzone_idx, 3 and 2, respectively, since their preferred_zones are > > > different. If the allocation is local, there is no problem. However, > > > if it is handled by the remote node due to memory shortage, the problem > > > would happen. > > > > I'm okay with your change but I try again to be better. Please check the > > following rewritten commit message and please let me know if it is better > > than before. > > Yeah, this new one looks very detailed, I believe anyone interested can > get what's going on. Thanks for doing this. Thanks for checking it. I will submit v4 with including this re-written commit message. Thanks.