mm: huge_memory: use GFP_TRANSHUGE when charging huge pages
authorJohannes Weiner <hannes@cmpxchg.org>
Wed, 6 Aug 2014 23:05:49 +0000 (16:05 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 7 Aug 2014 01:01:17 +0000 (18:01 -0700)
Transparent huge page charges prefer falling back to regular pages
rather than spending a lot of time in direct reclaim.

Desired reclaim behavior is usually declared in the gfp mask, but THP
charges use GFP_KERNEL and then rely on the fact that OOM is disabled
for THP charges, and that OOM-disabled charges don't retry reclaim.
Needless to say, this is anything but obvious and quite error prone.

Convert THP charges to use GFP_TRANSHUGE instead, which implies
__GFP_NORETRY, to indicate the low-latency requirement.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/huge_memory.c

index 4b95ff4120f5ae49cf01a29b2b921c6501f4c002..24e354c2b59e5b8e166fb51a6e7bc61753a35df8 100644 (file)
@@ -827,7 +827,7 @@ int do_huge_pmd_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
                count_vm_event(THP_FAULT_FALLBACK);
                return VM_FAULT_FALLBACK;
        }
-       if (unlikely(mem_cgroup_charge_anon(page, mm, GFP_KERNEL))) {
+       if (unlikely(mem_cgroup_charge_anon(page, mm, GFP_TRANSHUGE))) {
                put_page(page);
                count_vm_event(THP_FAULT_FALLBACK);
                return VM_FAULT_FALLBACK;
@@ -1132,7 +1132,7 @@ alloc:
                goto out;
        }
 
-       if (unlikely(mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL))) {
+       if (unlikely(mem_cgroup_charge_anon(new_page, mm, GFP_TRANSHUGE))) {
                put_page(new_page);
                if (page) {
                        split_huge_page(page);
@@ -2399,7 +2399,7 @@ static void collapse_huge_page(struct mm_struct *mm,
        if (!new_page)
                return;
 
-       if (unlikely(mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)))
+       if (unlikely(mem_cgroup_charge_anon(new_page, mm, GFP_TRANSHUGE)))
                return;
 
        /*