Re: [PATCH v2] staging: bcm2835-audio: Release resources on module_exit()

[Kirill Marinushkin]

On 03/13/18 22:23, Andy Shevchenko wrote:
> On Tue, Mar 13, 2018 at 9:34 PM, Kirill Marinushkin
> <k.marinushkin@xxxxxxxxx> wrote:
>> In the current implementation, `rmmod snd_bcm2835` does not release
>> resources properly. It causes an oops when trying to list sound devices.
>>
>> This commit fixes it.
> Nice catch!
>
> See my comments below.
>
>>  static void snd_devm_unregister_child(struct device *dev, void *res)
>>  {
>>         struct device *childdev = *(struct device **)res;
>> +       struct bcm2835_chip *chip = dev_get_drvdata(childdev);
>> +       struct snd_card *card = chip->card;
>> +
>> +       snd_card_free(card);
>> +       dev_set_drvdata(childdev, NULL);
> AFAIU this is done by device core.

Maybe you are right. But I don't know, which function in the device core does it.
It is safe to have this line. So, I suggest to keep it.

>
>>         device_unregister(childdev);
>>  }
>> +static void snd_devm_release(struct device *dev)
>> +{
>> +       struct bcm2835_chip *chip = dev_get_drvdata(dev);
>> +
>> +       kfree(chip);
>> +}
>
>>  /* chip-specific constructor
>> @@ -122,7 +136,7 @@ static int snd_bcm2835_create(struct snd_card *card,
>>
>>         err = snd_device_new(card, SNDRV_DEV_LOWLEVEL, chip, &ops);
>>         if (err) {
>> -               snd_bcm2835_free(chip);
>> +               kfree(chip);
> Do you call device_register() inside snd_device_new()?
> In this case you might need put_device() here instead of simple kfree().

No, from what I see, device_register() does not happen inside snd_device_new().

>>                 return err;
>>         }