mm: nobootmem: Correct alloc_bootmem semantics.
authorDavid Miller <davem@davemloft.net>
Wed, 25 Apr 2012 20:10:50 +0000 (16:10 -0400)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 26 Apr 2012 04:18:01 +0000 (21:18 -0700)
commit4e1c2b284461fd8aa8d7b295a1e911fc4390755b
treef51ab604d23af8ac0f6087516f877af5604c3477
parentaf3a3ab2966112c0d0a44df7eeb1e95fe32d4495
mm: nobootmem: Correct alloc_bootmem semantics.

The comments above __alloc_bootmem_node() claim that the code will
first try the allocation using 'goal' and if that fails it will
try again but with the 'goal' requirement dropped.

Unfortunately, this is not what the code does, so fix it to do so.

This is important for nobootmem conversions to architectures such
as sparc where MAX_DMA_ADDRESS is infinity.

On such architectures all of the allocations done by generic spots,
such as the sparse-vmemmap implementation, will pass in:

__pa(MAX_DMA_ADDRESS)

as the goal, and with the limit given as "-1" this will always fail
unless we add the appropriate fallback logic here.

Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/nobootmem.c