Re: [PATCH] dmaengine: imx-sdma: Add device to device support

[Shengjiu Wang <shengjiu.wang@freescale.com>]

On Tue, Jul 07, 2015 at 01:24:20PM +0800, Shengjiu Wang wrote:
> Hi vinod
> 
> On Tue, Jul 07, 2015 at 09:50:57AM +0530, Vinod Koul wrote:
> > On Tue, Jun 23, 2015 at 04:42:54PM +0800, Shengjiu Wang wrote:
> > > +static void sdma_set_watermarklevel_for_p2p(struct sdma_channel *sdmac)
> > > +{
> > > +	struct sdma_engine *sdma = sdmac->sdma;
> > > +
> > > +	int lwml = sdmac->watermark_level & 0xff;
> > > +	int hwml = (sdmac->watermark_level >> 16) & 0xff;
> > > +
> > > +	if (sdmac->event_id0 > 31) {
> > > +		sdmac->event_mask[0] |= 0;
> > > +		__set_bit(28, &sdmac->watermark_level);
> > why not use set_bit(), you are modifying driver memory
> Original driver all use the __set_bit. do you think we need to change
> all the __set_bit to set_bit? And from the header file "arch/arm/include/asm
> /bitops.h", the set_bit is same as __set_bit.
> > 
> > 
> > > +		sdmac->event_mask[1] |=
> > > +				BIT(sdmac->event_id0 % 32);
> > and then why not use set_bit() here too?
> > 
> > > +	} else {
> > > +		sdmac->event_mask[0] |= 0;
> > > +		sdmac->event_mask[1] |=
> > > +				BIT(sdmac->event_id0 % 32);
> > > +	}
> > > +	if (sdmac->event_id1 > 31) {
> > > +		sdmac->event_mask[1] |= 0;
> > > +		__set_bit(29, &sdmac->watermark_level);
> > > +		sdmac->event_mask[0] |=
> > > +			BIT(sdmac->event_id1 % 32);
> > > +	} else {
> > > +		sdmac->event_mask[1] |= 0;
> > > +		sdmac->event_mask[0] |=
> > > +			BIT(sdmac->event_id1 % 32);
> > > +	}
> > pattern for eventidX is repeated, also in that we can make generic macro to
> > handle and reduce code size
> I will change this.
> > 
> > > +
> > > +	/*
> > > +	 * If LWML(src_maxburst) > HWML(dst_maxburst), we need
> > > +	 * swap LWML and HWML of INFO(A.3.2.5.1), also need swap
> > > +	 * r0(event_mask[1]) and r1(event_mask[0]).
> > > +	 */
> > > +	if (lwml > hwml) {
> > > +		sdmac->watermark_level &= ~0xff00ff;
> > Magic number?
> > 
> > >  static int sdma_config_channel(struct dma_chan *chan)
> > >  {
> > >  	struct sdma_channel *sdmac = to_sdma_chan(chan);
> > > @@ -869,6 +945,12 @@ static int sdma_config_channel(struct dma_chan *chan)
> > >  		sdma_event_enable(sdmac, sdmac->event_id0);
> > >  	}
> > >  
> > > +	if (sdmac->event_id1) {
> > > +		if (sdmac->event_id1 >= sdmac->sdma->drvdata->num_events)
> > > +			return -EINVAL;
> > > +		sdma_event_enable(sdmac, sdmac->event_id1);
> > > +	}
> > > +
> > >  	switch (sdmac->peripheral_type) {
> > >  	case IMX_DMATYPE_DSP:
> > >  		sdma_config_ownership(sdmac, false, true, true);
> > > @@ -887,19 +969,21 @@ static int sdma_config_channel(struct dma_chan *chan)
> > >  			(sdmac->peripheral_type != IMX_DMATYPE_DSP)) {
> > >  		/* Handle multiple event channels differently */
> > >  		if (sdmac->event_id1) {
> > > -			sdmac->event_mask[1] = BIT(sdmac->event_id1 % 32);
> > > -			if (sdmac->event_id1 > 31)
> > > -				__set_bit(31, &sdmac->watermark_level);
> > > -			sdmac->event_mask[0] = BIT(sdmac->event_id0 % 32);
> > > -			if (sdmac->event_id0 > 31)
> > > -				__set_bit(30, &sdmac->watermark_level);
> > > -		} else {
> > > +			if (sdmac->peripheral_type == IMX_DMATYPE_ASRC_SP ||
> > > +			    sdmac->peripheral_type == IMX_DMATYPE_ASRC)
> > > +				sdma_set_watermarklevel_for_p2p(sdmac);
> > > +		} else
> > >  			__set_bit(sdmac->event_id0, sdmac->event_mask);
> > > -		}
> > > +
> > >  		/* Watermark Level */
> > >  		sdmac->watermark_level |= sdmac->watermark_level;
> > >  		/* Address */
> > > -		sdmac->shp_addr = sdmac->per_address;
> > > +		if (sdmac->direction == DMA_DEV_TO_DEV) {
> > Okay the direction is depreciated, so can you store both source and
> > destination and use them based on direction in prepare()
> > 
> > Also I see driver is not doing this, so while at it, can you fix this is
> > current code as well
> > 
> which prepare() do you mean? sdma_prep_dma_cyclic, sdma_prep_slave_sg?
> 
I exchange the meaning of per_address and per_address2 for p2p. So can remove
the direction checking here.
> > -- 
> > ~Vinod
> > 
> 
> 
I have sent a V2 patch for review. Thanks.

> best regards
> wang shengjiu