mm: free compound page with correct order
authorYu Zhao <[email protected]>
Wed, 29 Oct 2014 21:50:26 +0000 (14:50 -0700)
committerLinus Torvalds <[email protected]>
Wed, 29 Oct 2014 23:33:14 +0000 (16:33 -0700)
commit5ddacbe92b806cd5b4f8f154e8e46ac267fff55c
tree45394bb991b7403f73ce540ce9add87fa874f350
parentf601de204465048bdf0d5537f630729622ebc3a6
mm: free compound page with correct order

Compound page should be freed by put_page() or free_pages() with correct
order.  Not doing so will cause tail pages leaked.

The compound order can be obtained by compound_order() or use
HPAGE_PMD_ORDER in our case.  Some people would argue the latter is
faster but I prefer the former which is more general.

This bug was observed not just on our servers (the worst case we saw is
11G leaked on a 48G machine) but also on our workstations running Ubuntu
based distro.

  $ cat /proc/vmstat  | grep thp_zero_page_alloc
  thp_zero_page_alloc 55
  thp_zero_page_alloc_failed 0

This means there is (thp_zero_page_alloc - 1) * (2M - 4K) memory leaked.

Fixes: 97ae17497e99 ("thp: implement refcounting for huge zero page")
Signed-off-by: Yu Zhao <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Bob Liu <[email protected]>
Cc: <[email protected]> [3.8+]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
mm/huge_memory.c