Re: [PATCH v3 5/6] i2c: Add Actions Semiconductor Owl family S900 I2C driver

[Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>]

On Sun, Jul 01, 2018 at 01:43:33PM +0530, Manivannan Sadhasivam wrote:
> Hi Andy,
> 
> On Sun, Jul 01, 2018 at 12:11:00AM +0300, Andy Shevchenko wrote:
> > On Sat, Jun 30, 2018 at 4:33 PM, Manivannan Sadhasivam
> > <manivannan.sadhasivam@linaro.org> wrote:
> > > Add Actions Semiconductor Owl family S900 I2C driver.
> > 
> > Thanks for an update. Few left comments and it would LGTM.
> > 
> 
> Thanks :)
> 
> > > +static int owl_i2c_reset(struct owl_i2c_dev *i2c_dev)
> > > +{
> > 
> > > +       mdelay(1);
> > 
> > But now, since it's not used in atomic context, we may switch to
> > usleep_range() / msleep() instead.
> > 
> 
> okay, will use msleep()
>

Just realised, I have to use spinlock for the entire owl_i2c_master_xfer
function, so can't use sleep* for delay.

> > > +       owl_i2c_update_reg(i2c_dev->base + OWL_I2C_REG_CTL,
> > > +                               OWL_I2C_CTL_EN, true);
> > > +
> > 
> > > +       /* Wait 50ms for FIFO reset complete */
> > > +       do {
> > 
> > > +               mdelay(1);
> > 
> > Especially in this case it's very important.
> > 
> 
> Okay.

Same here, but I'm not sure about the latency. What is your suggestion?

Thanks,
Mani

> 
> > > +       } while (timeout++ < OWL_I2C_MAX_RETRIES);
> > 
> > > +}
> > 
> > > +       val = (i2c_dev->clk_rate + i2c_dev->bus_freq * 16 - 1) /
> > > +                               (i2c_dev->bus_freq * 16);
> > 
> > This is effectively DIV_ROUND_UP(->clk_rate, ->bus_freq * 16).
> > 
> 
> Ack.
> 
> > > +       /*
> > > +        * By default, 0 will be returned if interrupt occurred but no
> > > +        * read or write happened. Else if msg_ptr equals to message length,
> > > +        * message count will be returned.
> > > +        */
> > 
> > > +       if (i2c_dev->msg_ptr == msg->len)
> > > +               ret = num;
> > 
> > I dunno if
> > 
> > ret = ->msg_ptr == len ? num : 0;
> > 
> > would be slightly more explicit (yes, I aware about ret == 0).
> > 
> > Up to you to choose.
> > 
> 
> As per Peter's comment, returning 0 will get changed to an error value.
> Will use this pattern once we settle with a proper error value.
> 
> > > +       /* We support only frequencies of 100k and 400k for now */
> > > +       if (i2c_dev->bus_freq != OWL_I2C_DEF_SPEED_HZ &&
> > > +                       i2c_dev->bus_freq > OWL_I2C_MAX_SPEED_HZ) {
> > 
> > I think it should be != in the second case as well.
> > 
> 
> yeah, agree. We don't support any other frequencies now.
> 
> Thanks,
> Mani
> 
> > > +               dev_err(dev, "invalid clock-frequency %d\n", i2c_dev->bus_freq);
> > > +               return -EINVAL;
> > > +       }
> > 
> > -- 
> > With Best Regards,
> > Andy Shevchenko