[PATCH v2 3/8] spi: imx: add support for all SPI word width for DMA transfer

Robin Gong b38343 at freescale.com
Thu Oct 8 02:51:03 PDT 2015


On Thu, Oct 01, 2015 at 12:34:54AM +0000, Bondarenko, Anton wrote:
> On 30.09.2015 10:35, Robin Gong wrote:
> > On Fri, Sep 25, 2015 at 07:57:10PM +0200, Anton Bondarenko wrote:
> >> @@ -91,11 +91,15 @@ struct spi_imx_data {
> >>  
> >>  	struct completion xfer_done;
> >>  	void __iomem *base;
> >> +	unsigned long base_phys;
> >> +
> >>  	struct clk *clk_per;
> >>  	struct clk *clk_ipg;
> >>  	unsigned long spi_clk;
> >>  	unsigned int spi_bus_clk;
> >>  
> >> +	unsigned int bpw_w;
> >> +
> > It's better to change bytes_per_word for clear understanding,since bpw in spi means
> > bits_per_word...
> Agree. Fixed in V3
> >>  	unsigned int count;
> >>  	void (*tx)(struct spi_imx_data *);
> >>  	void (*rx)(struct spi_imx_data *);
> >> @@ -201,9 +205,13 @@ static bool spi_imx_can_dma(struct spi_master *master, struct spi_device *spi,
> >>  			 struct spi_transfer *transfer)
> >>  {
> >>  	struct spi_imx_data *spi_imx = spi_master_get_devdata(master);
> >> +	unsigned int bpw_w = transfer->bits_per_word;
> >>  
> >> -	if (spi_imx->dma_is_inited &&
> >> -	    (transfer->len > spi_imx->wml * sizeof(u32)))
> >> +	if (!bpw_w)
> >> +		bpw_w = spi->bits_per_word;
> >> +
> >> +	bpw_w = DIV_ROUND_UP(bpw_w, 8);
> >> +	if (spi_imx->dma_is_inited && (transfer->len > spi_imx->wml * bpw_w))
> > Please remove bpw_w here as I talked in the first patch.
> As explained in patch1 we need to use SPI word size in calculation to decide 
> if we want to go with DMA or PIO mode. Just a short example:
> We need to transfer 24 SPI words with BPW == 32. This will take ~ 24 PIO writes
> to FIFO (and same for reads). But transfer->len will be 24*4=96 bytes.
> WML is 32 SPI words. The decision will be incorrect if we do not take into account
> SPI bits per word. 
Yes, you are right, but I'm thinking so many 'DIV_ROUND_UP()'in your patch to
get bpw, can you centralize it or just use the 'bpw_w' in 'struct spi_imx_data'?
> >>  		return true;
> >>  	return false;
> >>  }
> >> @@ -1007,7 +1058,8 @@ static int spi_imx_dma_transfer(struct spi_imx_data *spi_imx,
> >>  					    DMA_FROM_DEVICE);
> >>  
> >>  			spi_imx->rx_buf = pio_buffer;
> >> -			spi_imx->txfifo = left;
> >> +			spi_imx->txfifo = DIV_ROUND_UP(left, spi_imx->bpw_w);
> >> +
> > Looks not enough here, you have to set 'spi_imx->rx' per the right bpw.
> What do you mean? We have had bpw_w == 1 before so 
> spi_imx->txfifo = DIV_ROUND_UP(left, spi_imx->bpw_w);
> is equivalent to
> spi_imx->txfifo = left;
> Now we could have bpw_w equal to 2 or 4 also. txfifo is number of SPI words
> and not number of bytes.
Sorry, please ignore this comment, since 'spi_imx->rx' will be correctly set
per the right bpw in spi_imx_setupxfer whatever in DMA or PIO mode.
> >>  			reinit_completion(&spi_imx->xfer_done);
> >>  
> >>  			spi_imx->devtype_data->intctrl(spi_imx, MXC_INT_TCEN);



More information about the linux-arm-kernel mailing list