IOAudio2Device

IOAudio2Device is a kernel-extension for playing (and recording?) audio. It's interacted with exclusively by CoreAudio, so it can't be used directly if CoreAudio is in the picture.

Methods

Selector	Action	Input	Output
0	start	-	-
1	end	-	-
2	setControlValue	uint32_t controlId, uint32_t value	-
4	setSampleRate	uint64_t sampleRate (fixed-point 32.32)	-
5	setStreamFormat	uint32_t streamId, struct AudioStreamBasicDescription (officially documented)	-
6	setStreamActive	uint32_t streamId, uint32_t active	-

Notifications

Notifications are registered via IOConnectSetNotificationPort, they are sent on any audio event.

struct IOAudio2DeviceNotification {
  <standard message header (ID is unknown, ignored by CoreAudio)>
  // note: no NDR

  uint32_t refCon; // passed to IOConnectSetNotificationPort
  uint32_t numEvents;
  IOAudio2DeviceEvent events[numEvents];
};

struct IOAudio2DeviceEvent {
  uint32_t id;
  uint32_t type; // 4cc, e.g. 'goin'
  uint32_t args[6];
};

Some known events are:

Type	Description	Id	Args
'goin'	Sent after start is called	0	-
'halt'	Sent after end is called	0	-
'prop'	Sent after property is changed	0 or stream id	property (?), scope (?), 0 (?)
'cval'	Sent after control value is changed	control id	new value

Mapped memory

The following memory can be mapped via IOConnectMapMemory:

Type	Size	Description
0	Single page?	struct IOAudio2DeviceStatus
0x10000001	from service property "io buffer frame size", multiplied by sample size	input sound buffer
0x10000002	from service property "io buffer frame size", multiplied by sample size	output sound buffer

struct IOAudio2DeviceStatus {
  uint64_t bufferSampleOffset;
  uint64_t bufferTime;
  uint64_t ???;
};

Operation

Once audio is started, the driver periodically reads some data from the output buffer and plays it. It zeroes the data after reading it (so that silence is played if the buffer is being neglected by user-land)

Once the driver finishes going over the entire output buffer, it adds the amount of samples in the buffer to bufferSampleOffset and updates bufferTime to the current time.

The output buffer sample format is typically 16-bit PCM stereo.

(input sound buffer operation is likely similar but with writing instead of reading).