Audio

Audio is set up as on stock, with Nokia's cards under Nokia's names, so Harmattan's PulseAudio and its policy run unchanged. The speaker, the earpiece and microphone paths, headphone detection and the headset button work. Not tried yet: listening to the DAC33's low-power FIFO mode, an ECI remote, and a real call (no SIM so far; see Modem).

The cards

ALSA card	Interface	Codec	Used for
0 dfl61twl4030	McBSP3 (I2S)	TWL5031 (mainline's twl4030 codec)	the speaker (its amplifier on TWL GPIO 7), the earpiece, the digital microphones, the headset microphone, call routing
1 dfl61dac33	McBSP2	TLV320DAC33, then the TPA6140A2 headphone amplifier	headphones
2 dfl61wl1273	McBSP4	Nokia's WL1273 codec	Bluetooth SCO and the FM radio (Bluetooth, FM)

Harmattan's PulseAudio opens the first two by name (hw:dfl61twl4030, hw:dfl61dac33), and its system.pa fails until both exist. The third is optional for PulseAudio; fmrxd expects it at index 2.

The machine driver

sound/soc/ti/dfl61.c is Nokia's dfl61-twl4030.c, dfl61-tlv320dac33.c and dfl61-wl1273.c as one DT driver (omap3-n9-audio.dtsi). It runs on mainline's codec drivers, whose control names match PulseAudio's profile sets in /usr/share/pulseaudio/alsa-mixer.

Commit
ac1f7c3f8 ASoC: ti: add Nokia's dfl61 machine driver	the two main cards, the jack and its plug classification; the TPA6140A2 as its own model (mainline matched it as a TPA6130A2 and gave it the wrong volume control)
bf07a897c ASoC: ti: dfl61: McBSP threshold mode and stock's sysfs paths	each McBSP in threshold mode, as Nokia's machine drivers set it; links at 2.6.32's paths: /sys/devices/platform/omap-mcbsp.{2,3,4}/max_{tx,rx}_thres and soc-audio.{0,1,2}/pmdown_time, which PulseAudio's pre-start script writes
24b20f193 ASoC: ti: dfl61: the AV jack's buttons and Nokia's ECI accessories	below
cf6c40eba ASoC: ti: dfl61: switch the AV jack to video for a TV-out cable	TVOUT_EN (GPIO 40) high for a video cable, low otherwise (TV-out)

Without the sysfs paths, PulseAudio's pre-start only logs a failure: the McBSPs stay in element mode (one DMA request per sample) and the power-down delay stays at 5 s instead of stock's values.

Headphones and the DAC33's low-power mode

Harmattan's /etc/pulse/pulseaudio_pre-start.sh selects the DAC33's 'FIFO Mode' 'Mode 7 Low Power'. That mode was Nokia's own (mainline has Bypass, Mode 1 and Mode 7) and is ported:

• 2cfd2979c ASoC: tlv320dac33: Nokia's Mode 7 Low Power FIFO mode: the DAC33 reads from the McBSP in bursts. An alarm interrupt ("ATHR time", 10 ms by default) starts the McBSP before the FIFO runs low, and a timer stops it after the burst. In between, the McBSP is force-idled with its functional clock gated, and a CPU latency request of the ATHR time is held while playing.
• 755ab739b, b778d7c84: Nokia's runtime power-level requests for the CPU DAI and the McBSP.
• 0c2950672 bypass the FIFO for short periods, as Nokia's: the control only selects the mode, and streams with periods of 40 ms or less run in Bypass whatever is selected. PulseAudio uses 10 ms periods normally and 140 ms when its policy asks for long buffering, so as on stock only long-buffered playback uses the FIFO.
• 43ef969c6, a5ea5ba25: the burst bit clock divider and the continuous bit clock from Nokia's board data.

Measured with 140 ms periods: 8-10 DAC33 interrupts a second, McBSP2's clock gated about 80% of the time, and the PER domain entering retention during playback (about 1000 times in 10 s, none in Bypass). Headphones played a test tone and the FM radio during the port; the low-power mode itself has not been listened to yet (music with the screen off).

The AV jack

The card has Nokia's jack input device, "dfl61-twl4030 Jack". It is fed by AvPlugDet (TWL GPIO 0, high when something is plugged in), and ohmd's accessories plugin follows its switches. Each plug is classified as Nokia's twl5031-aci did it, from the TWL5031's accessory comparators and the mic line measured through the MADC:

Result	Notes
wrong pinout	a CTIA headset such as Apple EarPods; the speaker stays on, as with stock's logic
headphones	mic line below 40 mV; reported as SND_JACK_HEADPHONE alone
video cable	below 500 mV; the jack switches to video (TV-out)
ECI accessory	answers Nokia's ECI bus reset; its buttons below
basic headset, carkit	one button, KEY_PHONE
open cable, unknown

The headphone amplifier is off before the first measurement, and passes repeat until two in a row agree, giving up with "unknown" after 24, as Nokia's driver did. The kernel log names each result ("AV jack: ...").

Two gotchas from the port:

• Report headphones as SND_JACK_HEADPHONE alone. With SND_JACK_MECHANICAL set too, ohmd reads the plug as an unknown accessory and keeps the speaker.
• A widget forced on with snd_soc_dapm_enable_pin() (for the measurement) stays forced. DAPM then never re-evaluates what follows it, and the TPA's mute stayed on during playback. Widgets are now handed back by disabling and then enabling them.

Headset buttons and ECI accessories

Stock read the jack's buttons with Nokia's twl5031-aci and eci drivers. They are ported into dfl61.c (24b20f193), and Harmattan sees what it did on stock:

• One input device, "ECI Accessory". udev links it as /dev/input/eci and /dev/input/accessory. qmkeyd2 reads it, X takes it as a keyboard, and ohmd finds it by name.
• A basic headset's (or carkit's) button: KEY_PHONE (169), from the ACI's comparator 1 with Nokia's timing (a change reported 200 ms after its sample, sampled every 220 ms while held), with autorepeat (period 400 ms).
• ECI accessories (Nokia remotes and carkits with a chip on the mic line, reached over a one-wire bus through the TWL5031's ACI block): classification tries Nokia's bus reset between the video-cable and open-cable steps, then reads the accessory's memory, its "enhancement features" block and the interrupt set-up, and maps the buttons with Nokia's table (volume, play/pause, stop, forward/rewind, send/end, mute...). ohmd reads the memory from /sys/devices/platform/ECI_accessory.0/memory. The accessory's microphone follows the "Headset Mic" widget.
• The ACI interrupt is the TWL's PIH line 6 (interrupts-extended = <&twl 6>), and VAUX1 (stock's "v28") is held while a plug is in.

Differences from stock, each needed on 6.18: bus transactions run from process context (6.18 nests all TWL interrupts in one thread, where waiting would deadlock), the mic switch runs from a work (Nokia's ran inside the DAPM event, which needs DAPM's lock here), and an ECI accessory whose init fails is still reported as a headset. Every held key is released on unplug, and the memory walk is bounded. TWL5031 ES1.0/ES1.1 (an address-decode bug that stock's twl-core worked around) get no buttons. The probe prints the revision ("TWL5031 ACI v.N (IDCODE ... rev ...): jack buttons"); the tested phone has 0x2b.

Tested with Nokia's one-button headset: it is classified as a headset; KEY_PHONE arrives on press and release, repeats while held, and a double press comes through as two presses; the music player pauses and resumes on it. Unplugged while held, the key is released, and plugged in again it works. No CPU latency request is left behind between transactions. Not tested: an ECI remote, and CORE retention with a headset in and USB out.

Call audio

On the N9, call audio is digital all the way. The modem and the SoC exchange 20 ms PCM frames over the SSI link, through Nokia's cmt_speech (/dev/cmt_speech) and PulseAudio's module-meego-cmtspeech, which holds the device open from boot. No TWL5031 voice port is involved. PulseAudio converts between the frames and 48 kHz on hw:dfl61twl4030, and its routes use controls that exist on card 0: the earpiece (DAC2, Earpiece Mixer), the loudspeaker (PredriveR), the headset (PredriveL and the DAC33's analog bypass into the TPA6140A2), the digital microphone or the headset's, and sidetone through the digital loopback. Without a call, setting these paths by hand plays to the earpiece and records the digital microphones. The modem side, the kernel fixes and the SIM-less tests are on the Modem page.

Checking it

cat /proc/asound/cards                              # 0 dfl61twl4030, 1 dfl61dac33, 2 dfl61wl1273
amixer -c dfl61dac33 sget 'FIFO Mode'               # Mode 7 Low Power once PulseAudio has started
dmesg | grep "AV jack\|jack buttons"                # each plug's classification
grep -A4 "ECI Accessory" /proc/bus/input/devices    # the buttons' input device
grep twl5031_aci /proc/interrupts                   # one or two per button press
od -An -tx1 /sys/devices/platform/ECI_accessory.0/memory   # "No such device or address" without an ECI accessory

PulseAudio's job reboots the phone if PulseAudio dies too often: stop it with initctl stop pulseaudio (and start again), never with kill.