RE: [PATCH v7 1/2] i2c: imx: add DMA support for freescale i2c driver

[Yao Yuan <yao.yuan@freescale.com>]

On Wednesday, September 17, 2014 2:17 AM, Marek Vasut wrote:
> On Wednesday, September 10, 2014 at 04:48:01 PM, Yao Yuan wrote:
> > On Friday, September 05, 2014 6:41 PM, Marek Vasut wrote:
> > > On Friday, September 05, 2014 at 12:32:40 PM, Yao Yuan wrote:
> > > [...]
> > >
> > > > > > +static int i2c_imx_dma_write(struct imx_i2c_struct *i2c_imx,
> > > > > > +					struct i2c_msg *msgs)
> > > > > > +{
> > > > > > +	int result;
> > > > > > +	unsigned int temp = 0;
> > > > > > +	unsigned long orig_jiffies = jiffies;
> > > > > > +	struct imx_i2c_dma *dma = i2c_imx->dma;
> > > > > > +	struct device *dev = &i2c_imx->adapter.dev;
> > > > > > +
> > > > > > +	dev_dbg(dev, "<%s> write slave address: addr=0x%x\n",
> > > > > > +		__func__, msgs->addr << 1);
> > > > > > +
> > > > > > +	reinit_completion(&i2c_imx->dma->cmd_complete);
> > > > > > +	dma->chan_using = dma->chan_tx;
> > > > > > +	dma->dma_transfer_dir = DMA_MEM_TO_DEV;
> > > > > > +	dma->dma_data_dir = DMA_TO_DEVICE;
> > > > > > +	dma->dma_len = msgs->len - 1;
> > > > > > +	result = i2c_imx_dma_xfer(i2c_imx, msgs);
> > > > > > +	if (result)
> > > > > > +		return result;
> > > > > > +
> > > > > > +	temp = imx_i2c_read_reg(i2c_imx, IMX_I2C_I2CR);
> > > > > > +	temp |= I2CR_DMAEN;
> > > > > > +	imx_i2c_write_reg(temp, i2c_imx, IMX_I2C_I2CR);
> > > > > > +
> > > > > > +	/*
> > > > > > +	 * Write slave address.
> > > > > > +	 * The first byte muse be transmitted by the CPU.
> > > > > > +	 */
> > > > > > +	imx_i2c_write_reg(msgs->addr << 1, i2c_imx, IMX_I2C_I2DR);
> > > > > > +	result = wait_for_completion_interruptible_timeout(
> > > > > > +				&i2c_imx->dma->cmd_complete,
> > > > > > +				msecs_to_jiffies(IMX_I2C_DMA_TIMEOUT));
> > > > > > +	if (result <= 0) {
> > > > > > +		dmaengine_terminate_all(dma->chan_using);
> > > > > > +		if (result)
> > > > > > +			return result;
> > > > > > +		else
> > > > > > +			return -ETIMEDOUT;
> > > > >
> > > > > Shouldn't you force-disable the DMA here somehow (like unsetting
> > > > > I2CR_DMAEN bit), if it failed or timed out?
> > > >
> > > > [Yuan Yao] Yes, I put the code for force-disable DMA in
> > > > i2c_imx_start(). In order to make sure any DMA error will not
> > > > effect the I2C.
> > > > It seems almost the same as put the code here, how about your think?
> > >
> > > Would that mean that the "crashed" DMA would be running until the
> > > next transmission is scheduled ?
> >
> > [Yuan Yao] No, In fact any DMA timeout will result the failure of I2C
> > transmission and then it will turn to report the exception and wait
> > for next transmission.
> 
> Can you tell when the next transmission will happen? What if I issue a
> single transmission and that one fails ? Will the DMA run until who knows
> when ?

[Yuan Yao] 
Sorry for my unclear description. In fact, During the DMA transmission  if
an error happened or time out, DMA will stop at once and be disabled.
I just continue to route the TX and RX request to signal the DMA controller.
Because the DMA is disabled, it will ignore those signals.

In a word, I just want to block the I2C TX, RX and interrupt signal when
DMA mode failed until the next I2C transmission start.

In fact, the bit "I2CR_DMAEN" is a switch which decide whether I2C route
the TX, RX and interrupt signal to DMA controller. 

> > The only thing I worried about is I2C may still receive some feedbacks
> > after DMA timeout. In this case the feedbacks may lead to abnormal
> > state in PIO mode.But it will be ignored in DMA model.
> > That's why I tend to delay force-disable DMA until the next
> > transmission begin. Could you please give me some suggestion?
> 
> No, this design just seems flawed to me. You should stop the DMA
> immediatelly if there is an error to avoid wasting resources and prevent
> possible other adverse effects.
>
[Yuan Yao] 
Yes, I have stopped the DMA immediately. However I keep the I2C DMA
single route.

I don't have the exact evidence to prove that my design is acceptable.
So if you are sure it's flawed, I will change it in the next version(V8).

Best regards,
Yuan Yao