[PATCH v5 2/2] DMAEngine: Add DMAEngine driver based on old MSM ADM DMA APIs

[Vinod Koul]

On Mon, 2012-03-26 at 09:17 +0200, Linus Walleij wrote:
> On Thu, Mar 22, 2012 at 3:23 PM, Ravi Kumar V <kumarrav at codeaurora.org> wrote:
> 
> This stuff:
> 
> > +enum adm_flush_type {
> > +       ADM_FORCED_FLUSH,
> > +       ADM_GRACEFUL_FLUSH,
> > +};
> 
> > +/*
> > + *  Controlling the hardware channel like stopping, flushing.
> > + */
> > +static int msm_dma_chan_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
> > +                                                       unsigned long arg)
> > +{
> > +       enum adm_flush_type cmd_type = arg;
> > +
> > +       if (cmd == DMA_TERMINATE_ALL) {
> > +               switch (cmd_type) {
> > +               case ADM_GRACEFUL_FLUSH:
> > +                               msm_dmov_stop_cmd(chan->chan_id, NULL, 1);
> > +                       break;
> > +               case ADM_FORCED_FLUSH:
> > +                       /*
> > +                        * We treate default as forced flush
> > +                        * so we fall through
> > +                        */
> > +               default:
> > +                               msm_dmov_stop_cmd(chan->chan_id, NULL, 0);
> > +                       break;
> > +               }
> > +       }
> > +       return 0;
> > +}
> 
> This is an interesting way of using the opaque arg. Can you describe
> the different semantics of graceful vs forced flush so we understand
> the usecases for them?
> 
> DMA_TERMINATE_ALL is usually not very graceful.
> 
> DMA_PAUSE might be the semantic you're looking for as the
> graceful path, and then you can do away with this strange
> enum altogether.
I am not a big fan of opaque objects. Btw client should use
dmaengine_terminate_all() only


-- 
~Vinod