TV-out

Composite video, PAL or NTSC, over the AV jack works as on stock: plug a TV-out cable and Harmattan clones the screen to the TV, with the standard and aspect ratio from Settings. Tested with a TV: six minutes of picture without a sync loss, a video from the Videos app, PAL/NTSC and the aspect ratio switched from Settings, and the unplug. After a sync loss the TV stays dark until the cable is replugged.

How it works

The kernel only switches the jack to video and reports the cable; Harmattan turns the TV on. The chain is stock's:

1. The jack. At each plug the TWL5031 card of Nokia's dfl61 machine driver sets TVOUT_EN (GPIO 40) low and classifies the plug. A video cable (mic line below 500 mV; the tested cable read about 99 mV) gets TVOUT_EN high, then the jack reports SND_JACK_AVOUT (SW_LINEOUT_INSERT and SW_VIDEOOUT_INSERT) on the input device dfl61-twl4030 Jack. The kernel log says AV jack: video cable (...). An unplug leaves TVOUT_EN as it is.
2. ohmd (the accessories plugin) takes SW_VIDEOOUT_INSERT without SW_JACK_PHYSICAL_INSERT as "tvout". Its policy routes video to the screen and the TV, with the standard and aspect from GConf (/system/tvout/encoding, pal by default; /system/tvout/aspectratio, wide). The Settings applet only writes those keys: there is no on/off switch, the cable is the switch.
3. ohmd's videoep asks X through XRandR: output TV, CRTC 1, mode 854x480, properties SignalProperties and TVAspectRatio.
4. Nokia's X driver sets omapdss up: a video plane (OMAPFB_SETUP_PLANE on /dev/fb2) showing fb0's memory, moved to the tv manager, then display1/timings (pal or ntsc), display1/enabled 1 and display1/wss (widescreen signalling: 8, 7, 1, 11 for PAL 4:3, PAL 16:9, NTSC 4:3, NTSC 16:9). At an unplug or DPMS off: the plane off, display1/enabled 0.

On the TV: PAL 720x574 at a 13.5 MHz pixel clock, the 854x480 screen scaled to 645x516 in the frame, WSS 16:9. Views in portrait show sideways, as on stock.

X needs display0 to be lcd and display1 to be tv, with manager N on display N; the probe order gives that.

The kernel side

The tv display is omapfb's VENC with the analog TV connector (composite-video-connector and the VDAC, 1.8 V, in omap3-n9.dts). Its register tables are the same as Nokia's DSS2 VENC. What it took:

Commit	What
9d625f491	VENC starts with PAL timings. A display1/wss write before the first enable hit a BUG() with VENC's lock held (a mainline bug; stock set PAL at probe)
e455ff905	L3 held at OPP100 from VENC's power on to its power off, as stock's request for any started display: the TV fetches from SDRAM all the time
f8a8d6628	the 3630's VENC takes the 96 MHz TV DAC clock (dss_96m_fck) too. omapfb took it on the 3430 and AM35xx only, so the clock framework could stop DPLL4's 96 MHz output under VENC
cf6c40eba	the TWL5031 card drives TVOUT_EN as stock. The DT had hogged it low, so the jack stayed in audio mode
95e3de665	the DISPC on DSS1_ALWON_FCLK, as stock. On the DSI PLL it lost its clock at every ULPS entry of the panel: "SYNC_LOST on channel tv" at each update after a pause
5fbca8420	after the sync losses, a stalled half of a panel update gave the DSI bus lock back twice: unlocked updates, a warning flood, then a panic. Now once

Clocks, as stock: dss_tv_fck 54 MHz (DPLL4's M3 divider 16, as NOLO leaves it; stock relied on NOLO too) and dss_96m_fck 96 MHz.

Headphones still classify as headphones, and TVOUT_EN stays low for them.

Checking it

As root, with debugfs at /sys/kernel/debug:

D=/sys/devices/platform/omapdss
cat $D/display1/name $D/display1/enabled $D/display1/timings  # tv; 1 with a TV cable in
dmesg | grep -i "AV jack\|SYNC_LOST"
grep tvout /sys/kernel/debug/gpio                    # gpio-40: hi with a TV cable in
cat /sys/kernel/debug/clk/dss_tv_fck/clk_rate        # 54000000
cat /sys/kernel/debug/clk/dss_96m_fck/clk_rate       # 96000000
cat /sys/kernel/debug/omap3_l3                       # the dss hold while the TV is on

At an unplug display1/enabled goes back to 0, the L3 hold goes and CORE retains again in standby.

Limitations

• After a sync loss ("SYNC_LOST on channel tv") omapdss's recovery leaves the TV's plane off, and X doesn't set it up again; stock's error handler turned it back on. Replug the cable. Since the DISPC clock fix nothing has caused a sync loss in testing; a fix (turning the planes back on after the restart) was written but isn't merged, since without a sync loss it can't be tested.
• NTSC was switched from Settings during the test but not confirmed in the logs.
• Writing display1/timings turns the TV on (6.18 disables, sets and re-enables); stock applied them only to a running TV. X writes them right before enabling the display, so Harmattan isn't affected.
• Stock's debugfs avswitch isn't there; only Nokia's factory test servers used it.

The LCD side of the DSS: Display. The jack's audio side: Audio.