mm: vmalloc search restart fix

Current vmalloc restart search for a free area in case we can't find one.
The reason is there are areas which are lazily freed, and could be
possibly freed now.  However, current implementation start searching the
tree from the last failing address, which is pretty much by definition at
the end of address space.  So, we fail.

The proposal of this patch is to restart the search from the beginning of
the requested vstart address.  This fixes the regression in running KVM
virtual machines for me, described in http://lkml.org/lkml/2008/10/28/349,
caused by commit db64fe0.

Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c

@@ -324,14 +324,14 @@ static struct vmap_area *alloc_vmap_area(unsigned long size,
 
 	BUG_ON(size & ~PAGE_MASK);
 
-	addr = ALIGN(vstart, align);
-
 	va = kmalloc_node(sizeof(struct vmap_area),
 			gfp_mask & GFP_RECLAIM_MASK, node);
 	if (unlikely(!va))
 		return ERR_PTR(-ENOMEM);
 
 retry:
+	addr = ALIGN(vstart, align);
+
 	spin_lock(&vmap_area_lock);
 	/* XXX: could have a last_hole cache */
 	n = vmap_area_root.rb_node;