arm: Add option to skip buffer zeroing

The DMA framework currently zeros all buffers because it (righfully so)
assumes that drivers will soon need to pass the memory to a device.
Some devices/use case may not require zeroed memory and there can
be an increase in performance if we skip the zeroing. Add a DMA_ATTR
to allow skipping of DMA zeroing.

Change-Id: Ib78a19cb341c3c441f91d5b004c6375c80c10413
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>

arch/arm/mm/dma-mapping.c

@@ -291,7 +291,7 @@ static void __dma_free_buffer(struct page *page, size_t size)
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
 				     pgprot_t prot, struct page **ret_page,
 				     const void *caller,
-				     bool no_kernel_mapping);
+				     struct dma_attrs *attrs);
 
 static void *__alloc_remap_buffer(struct device *dev, size_t size, gfp_t gfp,
 				 pgprot_t prot, struct page **ret_page,
@@ -397,7 +397,7 @@ static int __init atomic_pool_init(void)
 
 	if (IS_ENABLED(CONFIG_CMA))
 		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
-					      atomic_pool_init, false);
+					      atomic_pool_init, NULL);
 	else
 		ptr = __alloc_remap_buffer(NULL, pool->size, gfp, prot, &page,
 					   atomic_pool_init);
@@ -611,13 +611,15 @@ static int __free_from_pool(void *start, size_t size)
 static void *__alloc_from_contiguous(struct device *dev, size_t size,
 				     pgprot_t prot, struct page **ret_page,
 				     const void *caller,
-				     bool no_kernel_mapping)
+				     struct dma_attrs *attrs)
 {
 	unsigned long order = get_order(size);
 	size_t count = size >> PAGE_SHIFT;
 	unsigned long pfn;
 	struct page *page;
 	void *ptr;
+	bool no_kernel_mapping = dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING,
+							attrs);
 
 	pfn = dma_alloc_from_contiguous(dev, count, order);
 	if (!pfn)
@@ -625,6 +627,7 @@ static void *__alloc_from_contiguous(struct device *dev, size_t size,
 
 	page = pfn_to_page(pfn);
 
+	if (!dma_get_attr(DMA_ATTR_SKIP_ZEROING, attrs))
 		__dma_clear_buffer(page, size);
 
 	if (PageHighMem(page)) {
@@ -706,7 +709,7 @@ static void *__alloc_simple_buffer(struct device *dev, size_t size, gfp_t gfp,
 
 static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 			 gfp_t gfp, pgprot_t prot, bool is_coherent,
-			 const void *caller, bool no_kernel_mapping)
+			 const void *caller, struct dma_attrs *attrs)
 {
 	u64 mask = get_coherent_dma_mask(dev);
 	struct page *page = NULL;
@@ -747,7 +750,7 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
 	else
 		addr = __alloc_from_contiguous(dev, size, prot, &page, caller,
-						no_kernel_mapping);
+						attrs);
 
 	if (addr)
 		*handle = pfn_to_dma(dev, page_to_pfn(page));
@@ -764,14 +767,12 @@ void *arm_dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
 {
 	pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel);
 	void *memory;
-	bool no_kernel_mapping = dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING,
-					attrs);
 
 	if (dma_alloc_from_coherent(dev, size, handle, &memory))
 		return memory;
 
 	return __dma_alloc(dev, size, handle, gfp, prot, false,
-			   __builtin_return_address(0), no_kernel_mapping);
+			   __builtin_return_address(0), attrs);
 }
 
 static void *arm_coherent_dma_alloc(struct device *dev, size_t size,
@@ -779,14 +780,12 @@ static void *arm_coherent_dma_alloc(struct device *dev, size_t size,
 {
 	pgprot_t prot = __get_dma_pgprot(attrs, pgprot_kernel);
 	void *memory;
-	bool no_kernel_mapping = dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING,
-					attrs);
 
 	if (dma_alloc_from_coherent(dev, size, handle, &memory))
 		return memory;
 
 	return __dma_alloc(dev, size, handle, gfp, prot, true,
-			   __builtin_return_address(0), no_kernel_mapping);
+			   __builtin_return_address(0), attrs);
 }
 
 /*

include/linux/dma-attrs.h

@@ -19,6 +19,7 @@ enum dma_attr {
 	DMA_ATTR_SKIP_CPU_SYNC,
 	DMA_ATTR_FORCE_CONTIGUOUS,
 	DMA_ATTR_STRONGLY_ORDERED,
+	DMA_ATTR_SKIP_ZEROING,
 	DMA_ATTR_MAX,
 };