• Michal Hocko's avatar
    mm: exclude memoryless nodes from zone_reclaim · 70ef57e6
    Michal Hocko authored
    We had a report about strange OOM killer strikes on a PPC machine
    although there was a lot of swap free and a tons of anonymous memory
    which could be swapped out.  In the end it turned out that the OOM was a
    side effect of zone reclaim which wasn't unmapping and swapping out and
    so the system was pushed to the OOM.  Although this sounds like a bug
    somewhere in the kswapd vs.  zone reclaim vs.  direct reclaim
    interaction numactl on the said hardware suggests that the zone reclaim
    should not have been set in the first place:
    
      node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
      node 0 size: 0 MB
      node 0 free: 0 MB
      node 2 cpus:
      node 2 size: 7168 MB
      node 2 free: 6019 MB
      node distances:
      node   0   2
      0:  10  40
      2:  40  10
    
    So all the CPUs are associated with Node0 which doesn't have any memory
    while Node2 contains all the available memory.  Node distances cause an
    automatic zone_reclaim_mode enabling.
    
    Zone reclaim is intended to keep the allocations local but this doesn't
    make any sense on the memoryless nodes.  So let's exclude such nodes for
    init_zone_allows_reclaim which evaluates zone reclaim behavior and
    suitable reclaim_nodes.
    Signed-off-by: default avatarMichal Hocko <mhocko@suse.cz>
    Acked-by: default avatarDavid Rientjes <rientjes@google.com>
    Acked-by: default avatarNishanth Aravamudan <nacc@linux.vnet.ibm.com>
    Tested-by: default avatarNishanth Aravamudan <nacc@linux.vnet.ibm.com>
    Acked-by: default avatarMel Gorman <mgorman@suse.de>
    Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
    70ef57e6
page_alloc.c 182 KB