percpu: fix how @gfp is interpreted by the percpu allocator
authorTejun Heo <tj@kernel.org>
Wed, 8 Oct 2014 16:01:52 +0000 (12:01 -0400)
committerTejun Heo <tj@kernel.org>
Wed, 8 Oct 2014 16:01:52 +0000 (12:01 -0400)
commit6ae833c7fe0c6ef1f0ab13cc775da230d6f4c256
tree23fc4c7e1ef8a8d97fa039e0da7162c48d08bae7
parent17497acbdce9506fd6a75115dee4ab80c3cc5ee5
percpu: fix how @gfp is interpreted by the percpu allocator

When @gfp is specified, the percpu allocator is interested in whether
it contains all of GFP_KERNEL or not.  If it does, the normal
allocation path is taken; otherwise, the atomic allocation path.
Unfortunately, pcpu_alloc() was incorrectly testing for whether @gfp
contains any part of GFP_KERNEL.

Fix it by testing "(gfp & GFP_KERNEL) != GFP_KERNEL" instead of
"!(gfp & GFP_KERNEL)" to decide whether the allocation should be
atomic or not.

Signed-off-by: Tejun Heo <tj@kernel.org>
mm/percpu.c