global_page_state() was renamed to global_zone_page_state() by commit
c41f012ade0b ("mm: rename global_page_state to global_zone_page_state").
However, some more changes are needed (and were in fact needed even with
kernels older than 4.14. In particular, several users reported failures
when starting VMs on hosts with 4.13 kernel with message "not enough
physical memory" due to wrong accounting of NR_SLAB_UNRECLAIMABLE pages.
Since commit 385386cff4c6 ("mm: vmstat: move slab statistics from zone to
node counters") in v4.13-rc1, NR_SLAB_UNRECLAIMABLE needs to be used with
global_node_page_state(), in-tree callers were fixed by commit d507e2ebd2c7
("mm: fix global NR_SLAB_.*CLAIMABLE counter reads").
Since commit 599d0c954f91 ("mm, vmscan: move LRU lists to node") in
v4.8-rc1, NR_UNEVICTABLE needs global_node_page_state() rather than
global_page_state().
Since commit 50658e2e04c1 ("mm: move page mapped accounting to the node")
in v4.8-rc1, NR_ANON_PAGES needs global_node_page_state() as well. This was
shortly before it was renamed to NR_ANON_MAPPED but as both got into
mainline in v4.8-rc1, we can do with one #ifdef.
To keep HostIF_EstimateLockedPageLimit() readable, extract the version
dependent calls into inline helpers.
Michal Kubecek
fadedd9c8a