memblock: Fix alloc failure due to dumb underflow protection in membl…

…ock_find_in_range_node()

7bd0b0f ("memblock: Reimplement memblock allocation using
reverse free area iterator") implemented a simple top-down
allocator using a reverse memblock iterator.  To avoid underflow
in the allocator loop, it simply raised the lower boundary to
the requested size under the assumption that requested size
would be far smaller than available memblocks.

This causes early page table allocation failure under certain
configurations in Xen.  Fix it by checking for underflow directly
instead of bumping up lower bound.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: rjw@sisk.pl
Cc: xen-devel@lists.xensource.com
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20120113181412.GA11112@google.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>

mm/memblock.c

@@ -106,14 +106,17 @@ phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
 		end = memblock.current_limit;
 
-	/* adjust @start to avoid underflow and allocating the first page */
-	start = max3(start, size, (phys_addr_t)PAGE_SIZE);
+	/* avoid allocating the first page */
+	start = max_t(phys_addr_t, start, PAGE_SIZE);
 	end = max(start, end);
 
 	for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
 		this_start = clamp(this_start, start, end);
 		this_end = clamp(this_end, start, end);
 
+		if (this_end < size)
+			continue;
+
 		cand = round_down(this_end - size, align);
 		if (cand >= this_start)
 			return cand;