[PATCH] arm64/dma-mapping: Fix null-pointer check

On Mon, Jun 12, 2017 at 01:50:28PM -0700, Olav Haugan wrote:

On 17-06-12 13:29:04, Catalin Marinas wrote:

quoted

On Sat, Jun 10, 2017 at 12:41:10PM -0700, Olav Haugan wrote:

quoted

@@ -149,6 +140,11 @@ static void *__dma_alloc(struct device *dev, size_t size,
 	bool coherent = is_device_dma_coherent(dev);
 	pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, false);
 
+	if (!dev) {
+		WARN_ONCE(1, "Use an actual device structure for DMA allocation\n");
+		return NULL;
+	}
+
 	size = PAGE_ALIGN(size);
 
 	if (!coherent && !gfpflags_allow_blocking(flags)) {

@@ -192,8 +188,13 @@ static void __dma_free(struct device *dev, size_t size,
 		       void *vaddr, dma_addr_t dma_handle,
 		       unsigned long attrs)
 {
-	void *swiotlb_addr = phys_to_virt(dma_to_phys(dev, dma_handle));
+	void *swiotlb_addr;
 
+	if (!dev) {
+		WARN_ONCE(1, "Use an actual device structure for DMA free\n");
+		return;
+	}
+	swiotlb_addr = phys_to_virt(dma_to_phys(dev, dma_handle));

I don't think we need the checks anymore. With commit 1dccb598df54
("arm64: simplify dma_get_ops") __generic_dma_ops() returns
dummy_dma_ops when dev == NULL, so the above __dma_alloc/__dma_free
functions would not be called.

We don't need the check in is_device_dma_coherent() either then right?

Right. The only user of this function outside
arch/arm64/mm/dma-mapping.c is Xen (shared with arm32 under
arch/arm/xen/) but since arm32 doesn't do this NULL check either, we
should be fine.

-- 
Catalin