x86: Fix the initialization of physnode_map

With DISCONTIGMEM, the mapping between a pfn and its owning node is
initialized using data provided by the BIOS. However, the initialization
may fail if the extents are not aligned to section boundary (64M).

The symptom of this bug is an early boot failure in pfn_to_page(),
as it tries to access NODE_DATA(__nid) using index from an unitialized
element of the physnode_map[] array.

While the bug is always present, it is more likely to be hit in kdump
kernels on large machines, because:

1. The memory map for a kdump kernel is specified as exactmap, and
   exactmap is more likely to be unaligned.

2. Large reservations are more likely to span across a 64M boundary.

[ hpa: fixed incorrect use of "pfn" instead of "start" ]

Signed-off-by: Petr Tesarik <ptesarik@suse.cz>
Link: http://lkml.kernel.org/r/20140201133019.32e56f86@hananiah.suse.cz
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>

arch/x86/mm/numa_32.c

@@ -52,6 +52,8 @@ void memory_present(int nid, unsigned long start, unsigned long end)
 			nid, start, end);
 	printk(KERN_DEBUG "  Setting physnode_map array to node %d for pfns:\n", nid);
 	printk(KERN_DEBUG "  ");
+	start = round_down(start, PAGES_PER_SECTION);
+	end = round_up(end, PAGES_PER_SECTION);
 	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
 		physnode_map[pfn / PAGES_PER_SECTION] = nid;
 		printk(KERN_CONT "%lx ", pfn);