On Thu, Aug 13, 2015 at 04:25:05PM +0100, Russell King - ARM Linux wrote:

On Thu, Aug 13, 2015 at 05:04:08PM +0200, Christoph Hellwig wrote:

quoted

diff --git a/arch/arm/common/dmabounce.c b/arch/arm/common/dmabounce.c
index 1143c4d..260f52a 100644
--- a/arch/arm/common/dmabounce.c
+++ b/arch/arm/common/dmabounce.c

@@ -440,14 +440,6 @@ static void dmabounce_sync_for_device(struct device *dev,
 	arm_dma_ops.sync_single_for_device(dev, handle, size, dir);
 }
 
-static int dmabounce_set_mask(struct device *dev, u64 dma_mask)
-{
-	if (dev->archdata.dmabounce)
-		return 0;
-
-	return arm_dma_ops.set_dma_mask(dev, dma_mask);

Are you sure about this?  A user of dmabounce gets to request any mask
with the original code (even though it was never written back... which
is a separate bug.)  After this, it seems that this will get limited
by the dma_supported() check.  As this old code is about bouncing any
buffer into DMA-able memory, it doesn't care about the DMA mask.

I think you're right.  With the default dma_supported implementation
it would be fine, but ARM uses a custom one.  I'll keep the arm
specific dma_set_mask implementation for the next round.