On 02/07/2011 12:59 PM, Mark Brown wrote:
On Mon, Feb 07, 2011 at 12:04:22AM +0100, Lars-Peter Clausen wrote:
This patch fixes two issues with the bluethooth DAI registration:
- The DAI has to be registered only after the sound device has been added.
This isn't a problem, components in the system can be registered in any order.
By device I mean the soc-core platform device, which has to exist before a dai can be registered with that device.
- Currently the DAI is registered with snd_soc_register_dai which causes it to be named after the device it was registered for. Which is in this case the "soc-audio" device. (Ab)use snd_soc_register_dais which causes the DAI to be named after the DAI driver, which is what we want in this case.
This doesn't tie up with your change which...
- /* register bluetooth DAI here */
- ret = snd_soc_register_dai(&neo1973_gta02_snd_device->dev, &bt_dai);
- if (ret)
goto err_put_device;
- /* register bluetooth DAI here */
- ret = snd_soc_register_dais(&neo1973_gta02_snd_device->dev, &bt_dai, 1); if (ret)
goto err_unregister_dai;
goto err_unregister_device;...uses the same device in both cases and looking briefly at the code both register functions just use the dev that was passed in.
The difference between the two is that snd_soc_register_dais calls fmt_multiple_name, but snd_soc_register_dai uses fmt_single_name. The later uses the name of device for which the dai is registered to generate the dai_device name, while the former uses the name of the dai_driver. So by calling snd_soc_register_dai the bluetooth dai_device will be name "soc-core" instead of "bluetooth-dai". So currently there is no match between the dai_link and the dai_device, as a result the sound card is not instantiated.
In any case, we clearly shouldn't be applying patches which bodge around the core.
Well the alternative would be a patch looking like this:
- .cpu_dai_name = "bluetooth-dai", + .cpu_dai_name = "soc-core",
Which isn't really nice either.
- Lars