mm: fixed nodemask args in set_mempolicy()
inside oom0[2|4], ksm0[2|4] cases, they are running under
numa system, and allocte memory in the second numa node,
set nodemask 2, which indicates all memory allocated in
node1, but when a numa system don't have node1, these cases
will failed, such like the box, available nodes: (0,2,4,7)
[root@hp-magnycours-01 oom]# ./oom02
oom02 0 TINFO : set overcommit_memory to 1
oom02 0 TINFO : process mempolicy.
oom02 0 TINFO : start normal OOM testing.
oom02 0 TINFO : expected victim is 9112.
oom02 1 TBROK : set_mempolicy: errno=EINVAL(22): Invalid argument
oom02 2 TBROK : Remaining cases broken
oom02 0 TINFO : set overcommit_memory to 0
oom02 1 TFAIL : the victim unexpectedly failed: 512
oom02 0 TINFO : start OOM testing for mlocked pages.
...
the patch fixed the above issue.
Also inside cpuset01 case, depend on the reproducer, mem hog thread
allocated memory on *all* nodes, but 'nmask += exp2f(i);' can't
do that, and the patch fixed it.
Signed-off-by: Zhouping Liu <zliu@redhat.com>
Reviewed-by: Caspar Zhang <caspar@casparzhang.com>
Signed-off-by: Wanlong Gao <gaowanlong@cn.fujitsu.com>
4 files changed