Hi,
On Mon, Aug 08, 2011 at 10:32:38AM +0200, Per Forlin wrote:
[snip]
> -/* Work function invoked from DMA callback to handle rx transfers. */
> -static void ux500_rx_work(struct work_struct *data)
> +void ux500_dma_callback(void *private_data)
> {
> - struct ux500_dma_channel *ux500_chann
Skip the use of work queue and call musb_dma_completion() directly from
DMA callback context.
Here follows measurements on a Snowball board with ondemand governor active.
Performance using work queue:
(105 MB) copied, 6.23758 s, 16.8 MB/s
(105 MB) copied, 5.7151 s, 18.3 MB/s
(105 MB) copied, 5.83
