On Wed, Jul 31, 2013 at 08:17:39AM +0200, Jean-Francois Moine wrote:

The function kirkwood_i2s_dev_remove() may be used when probe fails.

Looking at this deeper, I'm not happy with this.

+static int kirkwood_i2s_dev_remove(struct platform_device *pdev) +{

• struct kirkwood_dma_data *priv = dev_get_drvdata(&pdev->dev);
• snd_soc_unregister_component(&pdev->dev);

...

@@ -519,30 +532,17 @@ static int kirkwood_i2s_dev_probe(struct platform_device *pdev)

err = snd_soc_register_component(&pdev->dev, &kirkwood_i2s_component, soc_dai, 1);

• if (err) {

• dev_err(&pdev->dev, "snd_soc_register_component failed\n");
  

• goto fail;
  

• }
• return 0;

+fail:

• kirkwood_i2s_dev_remove(pdev);

What this means is that if snd_soc_register_component() fails, we end up calling snd_soc_unregister_component(). This may be fine with the way snd_soc_unregister_component() is currently implemented, but you're making the assumption that it's fine to call snd_soc_unregister_component() for a device which hasn't been registered. Technically, this is a layering violation, which makes this change fragile if the behaviour of snd_soc_unregister_component() changes in the future.

For the sake of two calls in the error path, I don't think the benefits of this patch outweigh the risk.