ARM: Allocate dma pages via the page allocator and not via the slab allocator

Slab allocations are *not* guaranteed to be page aligned and slab allocators
may use the page structs for their own purposes. Using the page allocator
yields a properly aligned page and also makes the page flushing logic work right.

Passing a kmalloced "page" to a flushing function will not work reliably.

This will hopefully address the issue with SLUB on ARM. SLUB uses the
page->mapping field which is also checked by the flushing logic. The
flushing logic expect a normal page and not a slab page.

Signed-off-by: Christoph Lameter <clameter@sgi.com>

---
 arch/arm/mm/consistent.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: linux-2.6/arch/arm/mm/consistent.c
===================================================================
--- linux-2.6.orig/arch/arm/mm/consistent.c	2007-06-21 18:18:15.000000000 -0700
+++ linux-2.6/arch/arm/mm/consistent.c	2007-06-21 18:29:16.000000000 -0700
@@ -277,7 +277,7 @@ dma_alloc_coherent(struct device *dev, s
 	if (arch_is_coherent()) {
 		void *virt;
 
-		virt = kmalloc(size, gfp);
+		virt = get_free_pages(gfp, get_order(size));
 		if (!virt)
 			return NULL;
 		*handle =  virt_to_dma(dev, virt);
@@ -364,7 +364,7 @@ void dma_free_coherent(struct device *de
 	WARN_ON(irqs_disabled());
 
 	if (arch_is_coherent()) {
-		kfree(cpu_addr);
+		free_pages((unsigned long)cpu_addr, get_order(size));
 		return;
 	}