On Mon, 2011-05-09 at 11:33 +0530, viresh kumar wrote:

On 05/09/2011 10:50 AM, Koul, Vinod wrote:

quoted

quoted

quoted

@@ -695,9 +695,15 @@ dwc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		reg = dws->tx_reg;
 		for_each_sg(sgl, sg, sg_len, i) {
 			struct dw_desc	*desc;
-			u32		len;
-			u32		mem;
+			u32		len, dlen, mem;
 
+			mem = sg_phys(sg);
+			len = sg_dma_len(sg);
+			mem_width = 2;

hardcoding mem_width doesn't make sense, you should take this from input
params

Firstly, this change is not introduced in this patch, i have just rearranged this.
So, will send separate patch if this change is required.

Secondly, peripheral width is taken from chan->private. And by 2 for mem_width,
we meant word-by-word here. Shouldn't we always try word-by-word here?
How should we pass width for memory?

Ah, then there is another todo for this driver. Move from chan->private
to DMA_SLAVE_CONFIG.

struct dma_slave_config has dedicated fields for both source and
destination widths, src_addr_width & dst_addr_width please use them...
You should pass this structure to device control API as:
dmaengine_device_control(
	chan, DMA_SLAVE_CONFIG, (unsigned long)&slave_config);

-- 
~Vinod