Re: [PATCH V3] i2c: designware: fix wrong tx/rx fifo for ACPI

[Andy Shevchenko]

Thanks for an update! My comments below.

On Mon, 2016-12-12 at 15:36 +0700, Tin Huynh wrote:
> ACPI always sets txfifo and rxfifo to 32. This configuration will
> cause problem if the IP core supports a fifo size of less than 32.
> The driver should read the fifo size from the IP and select theÂ
> smaller one of the two.

I would use FIFO in capital to be consistent with what you refer to
(apparently not a variable name), so

Tx FIFO, Rx FIFO, FIFO, and so on.


> 
> Signed-off-by: Tin Huynh <tnhuynh@xxxxxxx>
> 
> ---
> Âdrivers/i2c/busses/i2c-designware-platdrv.c |ÂÂÂ26
> ++++++++++++++++++++------
> Â1 files changed, 20 insertions(+), 6 deletions(-)
> 
> Change from V2:
> Â-Add a helper function to set fifo size.
> 
> Change from V1:
> Â-Revert the default 32 for fifo, read parameter from IP core
> Âand pick the smaller one of the two.
> Â-Correct the title to describe new approach.
> 
> diff --git a/drivers/i2c/busses/i2c-designware-platdrv.c
> b/drivers/i2c/busses/i2c-designware-platdrv.c
> index 0b42a12..665f491 100644
> --- a/drivers/i2c/busses/i2c-designware-platdrv.c
> +++ b/drivers/i2c/busses/i2c-designware-platdrv.c
> @@ -150,6 +150,24 @@ static int i2c_dw_plat_prepare_clk(struct
> dw_i2c_dev *i_dev, bool prepare)
> Â	return 0;
> Â}
> Â
> +static void dw_i2c_set_fifo_size(struct dw_i2c_dev *dev)
> +{
> +	u32 param1, tx_fifo_depth, rx_fifo_depth;
> +
> +	param1 = i2c_dw_read_comp_param(dev);

You name it as param1 because you read *_PARAM1? For me it's not clear
from the name of helper function.

u32 param would work otherwise.

> +	tx_fifo_depth = ((param1 >> 16) & 0xff) + 1;
> +	rx_fifo_depth = ((param1 >> 8)ÂÂ& 0xff) + 1;
> +	if (!dev->tx_fifo_depth) {
> +		dev->tx_fifo_depth = tx_fifo_depth;
> +		dev->rx_fifo_depth = rx_fifo_depth;
> +	} else if (tx_fifo_depth) {
> +		dev->tx_fifo_depth = min_t(u32, dev->tx_fifo_depth,
> +				tx_fifo_depth);
> +		dev->rx_fifo_depth = min_t(u32, dev->rx_fifo_depth,
> +				rx_fifo_depth);
> +	}

So, let's clarify here:
Is it possible to have an IP without parameter block enabled? I mean to
read something arbitrary (or zeroes, or all-ones) from param1.

If not, just remove second condition at all.

> +}
> +
> Âstatic int dw_i2c_plat_probe(struct platform_device *pdev)
> Â{
> Â	struct dw_i2c_platform_data *pdata = dev_get_platdata(&pdev-
> >dev);
> @@ -246,13 +264,9 @@ static int dw_i2c_plat_probe(struct
> platform_device *pdev)
> Â				1000000);
> Â	}
> Â
> -	if (!dev->tx_fifo_depth) {
> -		u32 param1 = i2c_dw_read_comp_param(dev);
> -
> -		dev->tx_fifo_depth = ((param1 >> 16) & 0xff) + 1;
> -		dev->rx_fifo_depth = ((param1 >> 8)ÂÂ& 0xff) + 1;
> 

> +	if (!dev->tx_fifo_depth)
> Â		dev->adapter.nr = pdev->id;

Now you spread condition to two locations and it's hard to remember
ordering without looking closer to the code.

That's why I suggested to pass an ID parameter in the first place.

> -	}
> +	dw_i2c_set_fifo_size(dev);
> Â
> Â	adap = &dev->adapter;
> Â	adap->owner = THIS_MODULE;

-- 
Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
Intel Finland Oy