diff mbox

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

Message ID f8844a7a52ab44f00e7b6a1c3ce55d72d7999514.1442923630.git.anton_bondarenko@mentor.com (mailing list archive)
State New, archived
Headers show

Commit Message

Anton Bondarenko Sept. 25, 2015, 5:57 p.m. UTC
DMA transfer for SPI was limited to up to 8 bits word size until now.
Sync in SPI burst size and DMA bus width is necessary to correctly
support other BPW supported by HW.

Signed-off-by: Anton Bondarenko <anton_bondarenko@mentor.com>
---
 drivers/spi/spi-imx.c | 121 ++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 87 insertions(+), 34 deletions(-)

Comments

Robin Gong Sept. 30, 2015, 8:35 a.m. UTC | #1
On Fri, Sep 25, 2015 at 07:57:10PM +0200, Anton Bondarenko wrote:
> DMA transfer for SPI was limited to up to 8 bits word size until now.
> Sync in SPI burst size and DMA bus width is necessary to correctly
> support other BPW supported by HW.
> 
> Signed-off-by: Anton Bondarenko <anton_bondarenko@mentor.com>
> ---
>  drivers/spi/spi-imx.c | 121 ++++++++++++++++++++++++++++++++++++--------------
>  1 file changed, 87 insertions(+), 34 deletions(-)
> 
> diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c
> index 6c98eda..d9b730d 100644
> --- a/drivers/spi/spi-imx.c
> +++ b/drivers/spi/spi-imx.c
> @@ -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...
>  	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.
>  		return true;
>  	return false;
>  }
> @@ -761,11 +769,62 @@ static irqreturn_t spi_imx_isr(int irq, void *dev_id)
>  	return IRQ_HANDLED;
>  }
>  
> +static int spi_imx_sdma_configure(struct spi_master *master)
> +{
> +	int ret;
> +	enum dma_slave_buswidth dsb_default = DMA_SLAVE_BUSWIDTH_1_BYTE;
> +	struct dma_slave_config slave_config = {};
> +	struct spi_imx_data *spi_imx = spi_master_get_devdata(master);
> +
> +	switch (spi_imx->bpw_w) {
> +	case 4:
> +		dsb_default = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		break;
> +	case 2:
> +		dsb_default = DMA_SLAVE_BUSWIDTH_2_BYTES;
> +		break;
> +	case 1:
> +		dsb_default = DMA_SLAVE_BUSWIDTH_1_BYTE;
> +		break;
> +	default:
> +		pr_err("Not supported word size %d\n", spi_imx->bpw_w);
> +		ret = -EINVAL;
> +		goto err;
> +	}
> +
> +	slave_config.direction = DMA_MEM_TO_DEV;
> +	slave_config.dst_addr = spi_imx->base_phys + MXC_CSPITXDATA;
> +	slave_config.dst_addr_width = dsb_default;
> +	slave_config.dst_maxburst = spi_imx_get_fifosize(spi_imx)
> +					- spi_imx->wml;
> +	ret = dmaengine_slave_config(master->dma_tx, &slave_config);
> +	if (ret) {
> +		pr_err("error in TX dma configuration.\n");
> +		goto err;
> +	}
> +
> +	memset(&slave_config, 0, sizeof(slave_config));
> +
> +	slave_config.direction = DMA_DEV_TO_MEM;
> +	slave_config.src_addr = spi_imx->base_phys + MXC_CSPIRXDATA;
> +	slave_config.src_addr_width = dsb_default;
> +	slave_config.src_maxburst = spi_imx_get_fifosize(spi_imx)
> +					- spi_imx->wml;
> +	ret = dmaengine_slave_config(master->dma_rx, &slave_config);
> +	if (ret)
> +		pr_err("error in RX dma configuration.\n");
> +
> +err:
> +	return ret;
> +}
> +
>  static int spi_imx_setupxfer(struct spi_device *spi,
>  				 struct spi_transfer *t)
>  {
>  	struct spi_imx_data *spi_imx = spi_master_get_devdata(spi->master);
>  	struct spi_imx_config config;
> +	unsigned int bpw_w_new;
> +	int ret = 0;
>  
>  	config.bpw = t ? t->bits_per_word : spi->bits_per_word;
>  	config.speed_hz  = t ? t->speed_hz : spi->max_speed_hz;
> @@ -789,9 +848,18 @@ static int spi_imx_setupxfer(struct spi_device *spi,
>  		spi_imx->tx = spi_imx_buf_tx_u32;
>  	}
>  
> -	spi_imx->devtype_data->config(spi_imx, &config);
> +	bpw_w_new = DIV_ROUND_UP(config.bpw, 8);
> +	if (spi_imx->dma_is_inited && spi_imx->bpw_w != bpw_w_new) {
> +		spi_imx->bpw_w = bpw_w_new;
> +		ret = spi_imx_sdma_configure(spi->master);
> +		if (ret != 0)
> +			pr_err("Can't configure SDMA, error %d\n", ret);
> +	}
>  
> -	return 0;
> +	if (!ret)
> +		ret = spi_imx->devtype_data->config(spi_imx, &config);
> +
> +	return ret;
>  }
>  
>  static void spi_imx_sdma_exit(struct spi_imx_data *spi_imx)
> @@ -812,10 +880,8 @@ static void spi_imx_sdma_exit(struct spi_imx_data *spi_imx)
>  }
>  
>  static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
> -			     struct spi_master *master,
> -			     const struct resource *res)
> +			     struct spi_master *master)
>  {
> -	struct dma_slave_config slave_config = {};
>  	int ret;
>  
>  	/* use pio mode for i.mx6dl chip TKT238285 */
> @@ -832,17 +898,6 @@ static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
>  		goto err;
>  	}
>  
> -	slave_config.direction = DMA_MEM_TO_DEV;
> -	slave_config.dst_addr = res->start + MXC_CSPITXDATA;
> -	slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
> -	slave_config.dst_maxburst = spi_imx_get_fifosize(spi_imx)
> -					- spi_imx->wml;
> -	ret = dmaengine_slave_config(master->dma_tx, &slave_config);
> -	if (ret) {
> -		dev_err(dev, "error in TX dma configuration.\n");
> -		goto err;
> -	}
> -
>  	/* Prepare for RX : */
>  	master->dma_rx = dma_request_slave_channel(dev, "rx");
>  	if (!master->dma_rx) {
> @@ -851,23 +906,19 @@ static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
>  		goto err;
>  	}
>  
> -	slave_config.direction = DMA_DEV_TO_MEM;
> -	slave_config.src_addr = res->start + MXC_CSPIRXDATA;
> -	slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
> -	slave_config.src_maxburst = spi_imx_get_fifosize(spi_imx)
> -					- spi_imx->wml;
> -	ret = dmaengine_slave_config(master->dma_rx, &slave_config);
> -	if (ret) {
> -		dev_err(dev, "error in RX dma configuration.\n");
> -		goto err;
> -	}
> -
>  	init_completion(&spi_imx->dma_rx_completion);
>  	init_completion(&spi_imx->dma_tx_completion);
>  	master->can_dma = spi_imx_can_dma;
>  	master->max_dma_len = MAX_SDMA_BD_BYTES;
>  	spi_imx->bitbang.master->flags = SPI_MASTER_MUST_RX |
>  					 SPI_MASTER_MUST_TX;
> +	spi_imx->bpw_w = 1;
> +	ret = spi_imx_sdma_configure(master);
> +	if (ret) {
> +		dev_info(dev, "cannot get setup DMA.\n");
> +		goto err;
> +	}
> +
>  	spi_imx->dma_is_inited = 1;
>  
>  	return 0;
> @@ -924,7 +975,7 @@ static int spi_imx_dma_transfer(struct spi_imx_data *spi_imx,
>  	int ret;
>  	unsigned long timeout;
>  	unsigned long transfer_timeout;
> -	const int left = transfer->len % spi_imx->wml;
> +	const int left = transfer->len % (spi_imx->wml * spi_imx->bpw_w);
>  	struct spi_master *master = spi_imx->bitbang.master;
>  	struct sg_table *tx = &transfer->tx_sg, *rx = &transfer->rx_sg;
>  
> @@ -989,7 +1040,7 @@ static int spi_imx_dma_transfer(struct spi_imx_data *spi_imx,
>  		dmaengine_terminate_all(master->dma_rx);
>  	} else {
>  		transfer_timeout = spi_imx_calculate_timeout(spi_imx,
> -							     spi_imx->wml * 2);
> +					spi_imx->bpw_w * spi_imx->wml * 2);
>  		timeout = wait_for_completion_timeout(
>  				&spi_imx->dma_rx_completion, transfer_timeout);
>  		if (!timeout) {
> @@ -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.
>  			reinit_completion(&spi_imx->xfer_done);
>  
>  			spi_imx->devtype_data->intctrl(spi_imx, MXC_INT_TCEN);
> @@ -1207,6 +1259,7 @@ static int spi_imx_probe(struct platform_device *pdev)
>  		ret = PTR_ERR(spi_imx->base);
>  		goto out_master_put;
>  	}
> +	spi_imx->base_phys = res->start;
>  
>  	irq = platform_get_irq(pdev, 0);
>  	if (irq < 0) {
> @@ -1246,8 +1299,8 @@ static int spi_imx_probe(struct platform_device *pdev)
>  	 * Only validated on i.mx6 now, can remove the constrain if validated on
>  	 * other chips.
>  	 */
> -	if (spi_imx->devtype_data == &imx51_ecspi_devtype_data
> -	    && spi_imx_sdma_init(&pdev->dev, spi_imx, master, res))
> +	if (spi_imx->devtype_data == &imx51_ecspi_devtype_data &&
> +	    spi_imx_sdma_init(&pdev->dev, spi_imx, master))
>  		dev_err(&pdev->dev, "dma setup error,use pio instead\n");
>  
>  	spi_imx->devtype_data->reset(spi_imx);
> -- 
> 2.5.2
>
Anton Bondarenko Oct. 1, 2015, 12:34 a.m. UTC | #2
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.  
>>  		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.
>>  			reinit_completion(&spi_imx->xfer_done);
>>  
>>  			spi_imx->devtype_data->intctrl(spi_imx, MXC_INT_TCEN);
Robin Gong Oct. 8, 2015, 9:51 a.m. UTC | #3
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);
Anton Bondarenko Oct. 20, 2015, 11:28 p.m. UTC | #4
On 08.10.2015 11:51, Robin Gong wrote:
> 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'?
We can't use bpw_w from spi_imx_data because this is external callback 
for SPI core and driver does not know the purpose of this call. We also 
can't update current bpw_w for the same reason.

I'll add a function like this:
int spi_imx_get_bytes_per_word(int bpw)
{
     return DIV_ROUND_UP(bpw, BITS_PER_BYTE);
}

>>>>   		return true;
>>>>   	return false;
>>>>   }

Regards,
Anton
diff mbox

Patch

diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c
index 6c98eda..d9b730d 100644
--- a/drivers/spi/spi-imx.c
+++ b/drivers/spi/spi-imx.c
@@ -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;
+
 	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))
 		return true;
 	return false;
 }
@@ -761,11 +769,62 @@  static irqreturn_t spi_imx_isr(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
+static int spi_imx_sdma_configure(struct spi_master *master)
+{
+	int ret;
+	enum dma_slave_buswidth dsb_default = DMA_SLAVE_BUSWIDTH_1_BYTE;
+	struct dma_slave_config slave_config = {};
+	struct spi_imx_data *spi_imx = spi_master_get_devdata(master);
+
+	switch (spi_imx->bpw_w) {
+	case 4:
+		dsb_default = DMA_SLAVE_BUSWIDTH_4_BYTES;
+		break;
+	case 2:
+		dsb_default = DMA_SLAVE_BUSWIDTH_2_BYTES;
+		break;
+	case 1:
+		dsb_default = DMA_SLAVE_BUSWIDTH_1_BYTE;
+		break;
+	default:
+		pr_err("Not supported word size %d\n", spi_imx->bpw_w);
+		ret = -EINVAL;
+		goto err;
+	}
+
+	slave_config.direction = DMA_MEM_TO_DEV;
+	slave_config.dst_addr = spi_imx->base_phys + MXC_CSPITXDATA;
+	slave_config.dst_addr_width = dsb_default;
+	slave_config.dst_maxburst = spi_imx_get_fifosize(spi_imx)
+					- spi_imx->wml;
+	ret = dmaengine_slave_config(master->dma_tx, &slave_config);
+	if (ret) {
+		pr_err("error in TX dma configuration.\n");
+		goto err;
+	}
+
+	memset(&slave_config, 0, sizeof(slave_config));
+
+	slave_config.direction = DMA_DEV_TO_MEM;
+	slave_config.src_addr = spi_imx->base_phys + MXC_CSPIRXDATA;
+	slave_config.src_addr_width = dsb_default;
+	slave_config.src_maxburst = spi_imx_get_fifosize(spi_imx)
+					- spi_imx->wml;
+	ret = dmaengine_slave_config(master->dma_rx, &slave_config);
+	if (ret)
+		pr_err("error in RX dma configuration.\n");
+
+err:
+	return ret;
+}
+
 static int spi_imx_setupxfer(struct spi_device *spi,
 				 struct spi_transfer *t)
 {
 	struct spi_imx_data *spi_imx = spi_master_get_devdata(spi->master);
 	struct spi_imx_config config;
+	unsigned int bpw_w_new;
+	int ret = 0;
 
 	config.bpw = t ? t->bits_per_word : spi->bits_per_word;
 	config.speed_hz  = t ? t->speed_hz : spi->max_speed_hz;
@@ -789,9 +848,18 @@  static int spi_imx_setupxfer(struct spi_device *spi,
 		spi_imx->tx = spi_imx_buf_tx_u32;
 	}
 
-	spi_imx->devtype_data->config(spi_imx, &config);
+	bpw_w_new = DIV_ROUND_UP(config.bpw, 8);
+	if (spi_imx->dma_is_inited && spi_imx->bpw_w != bpw_w_new) {
+		spi_imx->bpw_w = bpw_w_new;
+		ret = spi_imx_sdma_configure(spi->master);
+		if (ret != 0)
+			pr_err("Can't configure SDMA, error %d\n", ret);
+	}
 
-	return 0;
+	if (!ret)
+		ret = spi_imx->devtype_data->config(spi_imx, &config);
+
+	return ret;
 }
 
 static void spi_imx_sdma_exit(struct spi_imx_data *spi_imx)
@@ -812,10 +880,8 @@  static void spi_imx_sdma_exit(struct spi_imx_data *spi_imx)
 }
 
 static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
-			     struct spi_master *master,
-			     const struct resource *res)
+			     struct spi_master *master)
 {
-	struct dma_slave_config slave_config = {};
 	int ret;
 
 	/* use pio mode for i.mx6dl chip TKT238285 */
@@ -832,17 +898,6 @@  static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
 		goto err;
 	}
 
-	slave_config.direction = DMA_MEM_TO_DEV;
-	slave_config.dst_addr = res->start + MXC_CSPITXDATA;
-	slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
-	slave_config.dst_maxburst = spi_imx_get_fifosize(spi_imx)
-					- spi_imx->wml;
-	ret = dmaengine_slave_config(master->dma_tx, &slave_config);
-	if (ret) {
-		dev_err(dev, "error in TX dma configuration.\n");
-		goto err;
-	}
-
 	/* Prepare for RX : */
 	master->dma_rx = dma_request_slave_channel(dev, "rx");
 	if (!master->dma_rx) {
@@ -851,23 +906,19 @@  static int spi_imx_sdma_init(struct device *dev, struct spi_imx_data *spi_imx,
 		goto err;
 	}
 
-	slave_config.direction = DMA_DEV_TO_MEM;
-	slave_config.src_addr = res->start + MXC_CSPIRXDATA;
-	slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
-	slave_config.src_maxburst = spi_imx_get_fifosize(spi_imx)
-					- spi_imx->wml;
-	ret = dmaengine_slave_config(master->dma_rx, &slave_config);
-	if (ret) {
-		dev_err(dev, "error in RX dma configuration.\n");
-		goto err;
-	}
-
 	init_completion(&spi_imx->dma_rx_completion);
 	init_completion(&spi_imx->dma_tx_completion);
 	master->can_dma = spi_imx_can_dma;
 	master->max_dma_len = MAX_SDMA_BD_BYTES;
 	spi_imx->bitbang.master->flags = SPI_MASTER_MUST_RX |
 					 SPI_MASTER_MUST_TX;
+	spi_imx->bpw_w = 1;
+	ret = spi_imx_sdma_configure(master);
+	if (ret) {
+		dev_info(dev, "cannot get setup DMA.\n");
+		goto err;
+	}
+
 	spi_imx->dma_is_inited = 1;
 
 	return 0;
@@ -924,7 +975,7 @@  static int spi_imx_dma_transfer(struct spi_imx_data *spi_imx,
 	int ret;
 	unsigned long timeout;
 	unsigned long transfer_timeout;
-	const int left = transfer->len % spi_imx->wml;
+	const int left = transfer->len % (spi_imx->wml * spi_imx->bpw_w);
 	struct spi_master *master = spi_imx->bitbang.master;
 	struct sg_table *tx = &transfer->tx_sg, *rx = &transfer->rx_sg;
 
@@ -989,7 +1040,7 @@  static int spi_imx_dma_transfer(struct spi_imx_data *spi_imx,
 		dmaengine_terminate_all(master->dma_rx);
 	} else {
 		transfer_timeout = spi_imx_calculate_timeout(spi_imx,
-							     spi_imx->wml * 2);
+					spi_imx->bpw_w * spi_imx->wml * 2);
 		timeout = wait_for_completion_timeout(
 				&spi_imx->dma_rx_completion, transfer_timeout);
 		if (!timeout) {
@@ -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);
+
 			reinit_completion(&spi_imx->xfer_done);
 
 			spi_imx->devtype_data->intctrl(spi_imx, MXC_INT_TCEN);
@@ -1207,6 +1259,7 @@  static int spi_imx_probe(struct platform_device *pdev)
 		ret = PTR_ERR(spi_imx->base);
 		goto out_master_put;
 	}
+	spi_imx->base_phys = res->start;
 
 	irq = platform_get_irq(pdev, 0);
 	if (irq < 0) {
@@ -1246,8 +1299,8 @@  static int spi_imx_probe(struct platform_device *pdev)
 	 * Only validated on i.mx6 now, can remove the constrain if validated on
 	 * other chips.
 	 */
-	if (spi_imx->devtype_data == &imx51_ecspi_devtype_data
-	    && spi_imx_sdma_init(&pdev->dev, spi_imx, master, res))
+	if (spi_imx->devtype_data == &imx51_ecspi_devtype_data &&
+	    spi_imx_sdma_init(&pdev->dev, spi_imx, master))
 		dev_err(&pdev->dev, "dma setup error,use pio instead\n");
 
 	spi_imx->devtype_data->reset(spi_imx);