On Mon, 26 Dec 2022 12:27:17 +0800 haibo.chen@xxxxxxx wrote: > From: Haibo Chen <haibo.chen@xxxxxxx> > > The ADC in i.mx93 is a total new ADC IP, add a driver to support > this ADC. > > Currently, only support one shot normal conversion triggered by > software. For other mode, will add in future. > > Signed-off-by: Haibo Chen <haibo.chen@xxxxxxx> Hi Haibo, I think there are still improvements to be made in ordering in probe()/remove() and also you aren't calling pm_runtime_dont_use_autosuspend() which is a requirement if manually handling runtime pm disabling on remove() Jonathan > --- > > diff --git a/drivers/iio/adc/imx93_adc.c b/drivers/iio/adc/imx93_adc.c > new file mode 100644 > index 000000000000..677f13a040f8 > --- /dev/null > +++ b/drivers/iio/adc/imx93_adc.c > @@ -0,0 +1,477 @@ > +static int imx93_adc_probe(struct platform_device *pdev) > +{ > + struct imx93_adc *adc; > + struct iio_dev *indio_dev; > + struct device *dev = &pdev->dev; > + int ret; > + > + indio_dev = devm_iio_device_alloc(dev, sizeof(*adc)); > + if (!indio_dev) { > + dev_err(dev, "Failed allocating iio device\n"); > + return -ENOMEM; > + } > + > + adc = iio_priv(indio_dev); > + adc->dev = dev; > + > + mutex_init(&adc->lock); > + adc->regs = devm_platform_ioremap_resource(pdev, 0); > + if (IS_ERR(adc->regs)) > + return PTR_ERR(adc->regs); > + > + /* The third irq is for ADC conversion usage */ > + adc->irq = platform_get_irq(pdev, 2); > + if (adc->irq < 0) > + return adc->irq; > + > + adc->ipg_clk = devm_clk_get(dev, "ipg"); > + if (IS_ERR(adc->ipg_clk)) > + return dev_err_probe(dev, PTR_ERR(adc->ipg_clk), > + "Failed getting clock.\n"); > + > + adc->vref = devm_regulator_get(dev, "vref"); > + if (IS_ERR(adc->vref)) > + return dev_err_probe(dev, PTR_ERR(adc->vref), > + "Failed getting reference voltage.\n"); > + > + ret = regulator_enable(adc->vref); > + if (ret) { > + dev_err(dev, "Can't enable adc reference top voltage.\n"); You can use dev_err_probe() for all such handling in probe() whether or not it can defer. That tends to simplify things and avoids the need for reviewers to consider if a function can defer of not. > + return ret; > + } > + > + platform_set_drvdata(pdev, indio_dev); > + > + init_completion(&adc->completion); > + > + indio_dev->name = "imx93-adc"; > + indio_dev->info = &imx93_adc_iio_info; > + indio_dev->modes = INDIO_DIRECT_MODE; > + indio_dev->channels = imx93_adc_iio_channels; > + indio_dev->num_channels = ARRAY_SIZE(imx93_adc_iio_channels); > + > + ret = clk_prepare_enable(adc->ipg_clk); > + if (ret) { > + dev_err(&pdev->dev, "Could not prepare or enable the clock.\n"); > + goto error_regulator_disable; > + } > + > + ret = request_irq(adc->irq, imx93_adc_isr, 0, IMX93_ADC_DRIVER_NAME, adc); > + if (ret < 0) { > + dev_err(dev, "Failed requesting irq, irq = %d\n", adc->irq); > + goto error_ipg_clk_disable; > + } > + > + ret = imx93_adc_calibration(adc); > + if (ret < 0) > + goto error_free_adc_irq; > + > + imx93_adc_config_ad_clk(adc); > + > + ret = iio_device_register(indio_dev); > + if (ret) { > + dev_err(dev, "Couldn't register the device.\n"); > + goto error_free_adc_irq; > + } > + > + pm_runtime_set_active(dev); > + pm_runtime_set_autosuspend_delay(dev, 50); > + pm_runtime_use_autosuspend(dev); > + pm_runtime_enable(dev); > + > + return 0; > + > +error_free_adc_irq: > + free_irq(adc->irq, adc); > +error_ipg_clk_disable: > + clk_disable_unprepare(adc->ipg_clk); > +error_regulator_disable: > + regulator_disable(adc->vref); > + > + return ret; > +} > + > +static int imx93_adc_remove(struct platform_device *pdev) > +{ > + struct iio_dev *indio_dev = platform_get_drvdata(pdev); > + struct imx93_adc *adc = iio_priv(indio_dev); > + struct device *dev = adc->dev; > + > + /* adc power down need clock on */ > + pm_runtime_get_sync(dev); > + > + iio_device_unregister(indio_dev); > + imx93_adc_power_down(adc); Why is there no similar power down in the error path in probe for iio_device_register() returning an error? > + free_irq(adc->irq, adc); > + clk_disable_unprepare(adc->ipg_clk); > + regulator_disable(adc->vref); > + > + pm_runtime_disable(dev); > + pm_runtime_put_noidle(dev); I think I caused confusion a bit here by pointing out the device unregister needed to be first. That's now fine, but the rest would benefit from a rethink. To my mind, the ideal situation is that the remove() is a reverse of the probe() function, so I'd expect to see these pm_runtime_disable(), pm_runtime_put_noidle() at the start of this function. Note that you also need to call pm_runtime_dont_use_autosuspend() somewhere in here - or take all the probe/remove devm_ managed and use devm_pm_runtime_enable() which tidies that up for you as needed. (see docs in pm_runtime.h) > + > + return 0; > +}