Re: [RFC PATCH] arm: dma-mapping: fallback allocation for cma failure

Gioh Kim <gioh.kim@xxxxxxx> · Wed, 21 May 2014 09:24:50 +0900

2014-05-21 오전 3:22, Michal Nazarewicz 쓴 글:
On Mon, May 19 2014, Joonsoo Kim wrote:
On Tue, May 20, 2014 at 02:57:47PM +0900, Gioh Kim wrote:

Thanks for your advise, Michal Nazarewicz.

Having discuss with Joonsoo, I'm adding fallback allocation after __alloc_from_contiguous().
The fallback allocation works if CMA kernel options is turned on but CMA size is zero.

Hello, Gioh.

I also mentioned the case where devices have their specific cma_area.
It means that this device needs memory with some contraint.
Although I'm not familiar with DMA infrastructure, I think that
we should handle this case.

How about below patch?

------------>8----------------

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 6b00be1..4023434 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -379,7 +379,7 @@ static int __init atomic_pool_init(void)
  	unsigned long *bitmap;
  	struct page *page;
  	struct page **pages;
-	void *ptr;
+	void *ptr = NULL;
  	int bitmap_size = BITS_TO_LONGS(nr_pages) * sizeof(long);

  	bitmap = kzalloc(bitmap_size, GFP_KERNEL);
@@ -393,7 +393,8 @@ static int __init atomic_pool_init(void)
  	if (IS_ENABLED(CONFIG_DMA_CMA))
  		ptr = __alloc_from_contiguous(NULL, pool->size, prot, &page,
  					      atomic_pool_init);
-	else
+
+	if (!ptr)
  		ptr = __alloc_remap_buffer(NULL, pool->size, gfp, prot, &page,
  					   atomic_pool_init);
  	if (ptr) {
@@ -701,10 +702,22 @@ static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle,
  		addr = __alloc_simple_buffer(dev, size, gfp, &page);
  	else if (!(gfp & __GFP_WAIT))
  		addr = __alloc_from_pool(size, &page);
-	else if (!IS_ENABLED(CONFIG_DMA_CMA))
-		addr = __alloc_remap_buffer(dev, size, gfp, prot, &page, caller);
-	else
-		addr = __alloc_from_contiguous(dev, size, prot, &page, caller);
+	else {
+		if (IS_ENABLED(CONFIG_DMA_CMA)) {
+			addr = __alloc_from_contiguous(dev, size, prot,
+							&page, caller);
+			/*
+			 * Device specific cma_area means that
+			 * this device needs memory with some contraint.
+			 * So, we can't fall through general remap allocation.
+			 */
+			if (!addr && dev && dev->cma_area)
+				return NULL;
+		}
+
+		addr = __alloc_remap_buffer(dev, size, gfp, prot,
+							&page, caller);
+	}

__arm_dma_free will have to be changed to handle the fallback as well.
But perhaps Marek is right and there should be no fallback for regular
allocations?  Than again, non-CMA allocation should be performed at
least in the case of cma=0.


I think in the cases of cma=0, CONFIG_CMA_SIZE_SEL_MBYTES=0 and CONFIG_CMA_SIZE_SEL_PERCENTAGE=0
kernel should be act as like CMA is not turned on.
non-CMA allocation should be performed in the cases.


I will send a patch soon with __arm_dma_free changes.




  	if (addr)
  		*handle = pfn_to_dma(dev, page_to_pfn(page));




--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>







