Re: [PATCH] serial: sh-sci: don't filter on DMA device, use only channel ID

[Guennadi Liakhovetski <g.liakhovetski@gmx.de>]

On Tue, 30 Aug 2011, Vinod Koul wrote:

> On Tue, 2011-08-30 at 09:54 +0200, Guennadi Liakhovetski wrote:
> > On Mon, 29 Aug 2011, Vinod Koul wrote:
> > > > >> You should not assign chan->private.
> > > > >> Please move this  to dma_slave_control API
> > > > >
> > > > > I haven't seen any reply to this comment and this patch seems to still be
> > > > > outstanding, is there an updated version of this patch that I've missed?
> > > I don't recall seeing any updated version fixing this 
> > 
> > As a matter of fact, when you say "use dma_slave_control API," you 
> > actually mean the dma_slave_config, right? Then, how is one supposed to 
> > use it in this case? Shall we be issuing a DMA_SLAVE_CONFIG call inside of 
> > the filter and check the return code? The problem is, that not all DMA 
> > controllers on sh-mobile SoCs can service the same slave devices. So, if 
> > we don't check in the filter we might well get an unsuitable DMA channel.
> Here you are assigning to chan->private your specific values, which
> should be moved to the dma_slave_config. 
> But here you are removing the checks, and accepting the first channel
> you got, so how do you find channel is suitable or not.

That's done in the driver's .device_alloc_chan_resources() method. It 
checkx the .private pointer, tries to find a suitable channel, if it fails 
- it returns -EINVAL. See 
drivers/dma/shdma.c::sh_dmae_alloc_chan_resources():

	if (param) {
		const struct sh_dmae_slave_config *cfg;

		cfg = sh_dmae_find_slave(sh_chan, param);
		if (!cfg) {
			ret = -EINVAL;
			goto efindslave;
		}

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/