Hi Dan,
I think you have a bug in this function, the list_splice_init adds the
new slots in the head of the chain_node, but you get the
old_chain_tail (latest descriptor) from the tail of the chain!!

 > +static dma_cookie_t
 > +iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
 > +{
 > +
 > +       old_chain_tail = list_entry(iop_chan->chain.prev,
 > +               struct iop_adma_desc_slot, chain_node);
 > +       list_splice_init(&sw_desc->group_list, &old_chain_tail->chain_node);
 > +
 > +       /* fix up the hardware chain */
 > +       iop_desc_set_next_desc(old_chain_tail, grp_start->phys);
 > +