Re: [PATCH v6 2/4] iio: adc: Add Xilinx AMS driver

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

...

> >   
> > > +	if (IS_ERR(ams->base))
> > > +		return PTR_ERR(ams->base);
> > > +
> > > +	ams->clk = devm_clk_get(&pdev->dev, NULL);
> > > +	if (IS_ERR(ams->clk))
> > > +		return PTR_ERR(ams->clk);
> > > +	clk_prepare_enable(ams->clk);
> > > +	devm_add_action_or_reset(&pdev->dev, (void  
> > *)clk_disable_unprepare,  
> > > +				 ams->clk);
> > > +
> > > +	INIT_DELAYED_WORK(&ams->ams_unmask_work,  
> > ams_unmask_worker);  
> > > +	devm_add_action_or_reset(&pdev->dev, (void  
> > *)cancel_delayed_work,
> > 
> > I'm not keen on casting away the function pointer type.  Normally we'd just
> > wrap it in a local function, to make it clear it was deliberate and avoid
> > potential nasty problems if the signature of the function ever changes.
> > 
> > It's 3 lines of boilerplate, but will give me warm fuzzy feelings!
> > Same for the other case above.  The fact this isn't done in exising kernel code
> > make this particularly risky.  
> 
> Makes sense. I will revert the code back to its original and handle the cases using goto
> and inside remove()
Ah.  Not what I meant.  I was suggesting you add a little function locally
that has the right type and in turn calls cancel_delayed_work().

As that directly exposes the actual function calls, any signature change in future
will cause compile breakage (or be picked up any automated tools doing that refactor).

> 
> >   
> > > +				 &ams->ams_unmask_work);
> > > +
> > > +	ret = ams_init_device(ams);
> > > +	if (ret) {
> > > +		dev_err(&pdev->dev, "failed to initialize AMS\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	ret = ams_parse_dt(indio_dev, pdev);
> > > +	if (ret) {
> > > +		dev_err(&pdev->dev, "failure in parsing DT\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	ams_enable_channel_sequence(indio_dev);
> > > +
> > > +	ams->irq = platform_get_irq(pdev, 0);  
> > 
> > platform_get_irq () can return errors, in particular -EPROBE_DEFER so I'd
> > check that and return before you call devm_request_irq() I'm not sure
> > devm_request_irq() will not eat that error code.
> >   
> 
> Will fix this in next series.
> 
> >   
> > > +	ret = devm_request_irq(&pdev->dev, ams->irq, &ams_irq, 0, "ams-  
> > irq",  
> > > +			       indio_dev);
> > > +	if (ret < 0) {
> > > +		dev_err(&pdev->dev, "failed to register interrupt\n");
> > > +		return ret;
> > > +	}
> > > +
> > > +	platform_set_drvdata(pdev, indio_dev);
> > > +
> > > +	return iio_device_register(indio_dev); }
> > > +
> > > +static int ams_remove(struct platform_device *pdev) {
> > > +	struct iio_dev *indio_dev = platform_get_drvdata(pdev);
> > > +
> > > +	iio_device_unregister(indio_dev);  
> > 
> > If this is all you have in remove, then you can use devm_iio_device_register()
> > in probe() and not need an remove() callback at all.  
> 
> I think remove will have more functions since I am getting rid of devm_add_action_or_reset()

See above.

J

> 
> >   
> > > +
> > > +	return 0;
> > > +}
> > > +  
> > 
> > ...