Make the nodes other than node 0 use bdata->last_success for fast
search too.

We need to use __alloc_bootmem_core() for vmemmap allocation for other
nodes when numa and sparsemem/vmemmap are enabled.

Also, make fail_block path increase i with incr only after ALIGN
to avoid extra increase when size is larger than align.
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>