This page collects some technical information for the Philips (a.k.a. TPVision) Android TV; specifically, for my 2018 65OLED873 series (“TPM171E”).

NOTE: This document is work in progress!

Install telnetd from Google Play, enable the adb service.

Enable developer options by tapping on the build number in the Android settings five times. There, enable USB Debugging. Try adb connect your.tv.address from a computer. If no confirmation dialog pops up and adb shell doesn’t work either, try rebooting (e.g. with Ctrl+Alt+Del on a USB keyboard) and quickly connecting during boot. It may take a few tries, but should succeed eventually.

Something in this sequence has temporarily made “USB 1” (the upper port) a USB host, speaking the adb protocol (you’ll need a USB A-to-A cable). This seems to always be the case in fastboot mode, out of which you can (apparently only) get with fastboot continue. Don’t type reboot bootloader!

• mmcblk0boot0, mmcblk0p1 (uboot), p2 (ubootB): uboot.bin
• p3 (misc): ???
• p4 (recovery), p8 (recoveryB): recovery.img
• p5 (boot): boot.img (androidboot)
• p6 (uboot_env), p7 (ubootB_env): uenv
• p9 (bootB), p22 (tzbp): tz.bin.lzhs
• p10-p13: system, userdata, cache, dmverity
• p14 (linux_rootfs): rootfs
• p15 (factory): factory
• p16 (factory_setting): ???
• p17 (3rd_ro): 3rd file
• p18-p21: 3rd_rw, reserved, channelA, channelB
• p23 (adsp): adsp
• p24-p27: ciplus, dvbsDB, nvm, ambl
• p28 (pq): pq
• p29 (aq): aq
• p30-p33: SCHEDPVR, usersetting, basic

uboot.bin (mmcblk0p1 and 2) is comprised of mtk_1bl, the first-stage bootloader. It decompresses mtkloader, which decompresses uboot. uboot is an actual U-Boot bootloader (2011.12) and loads the ARM TrustZone kernel (lk/MP) after decompressing tz.bin (p9/p22). TZ seems to load the Linux kernel.

The TZ stuff seems entirely irrelevant in practice. An U-Boot shell can be accessed through a serial console (SERV.U port), boot args can be set and different partitions can be booted.

For example, to load a different Android boot image from recoveryB on /dev/mmcblk0p8 (which seems unused and its content otherwise matches recovery, so seems safe to overwrite):

# Run 'getenv bootargs' beforehand and modify root= only
setenv bootargs lpj=120000 console=ttyMT0,115200n1 root=/dev/mmcblk0p8 vmalloc=650mb mtdparts=mt53xx-emmc:2M(uboot),2M(ubootB),1M(misc)enc,32M(recovery)enc,20M(boot)enc,2M(uboot_env)enc,2M(ubootB_env)enc,32M(recoveryB)enc,20M(bootB)enc,2048M(system),10288M(userdata),1200M(cache),1M(dmverity),61M(linux_rootfs),20M(factory),10M(factory_setting),128M(3rd_ro)enc,64M(3rd_rw)enc,1M(reserved),1M(channelA),1M(channelB),1M(tzbp)enc,4M(adsp),1M(ciplus),1M(dvbsDB),2M(nvm),1M(ambl),20M(pq),5M(aq),1M(SCHEDPVR),720M(usersetting),1M(basic) usbportusing=1,1,0,1 usbpwrgpio=42:1,43:1,44:1,210:1 usbocgpio=404:0,404:0,405:0,405:0 usbhubrstgpio=-1:-1 msdcgpio=76,-1,-1,176,-1,-1 msdcautok=0x200000a,0x242000,0x0,0x403c0007,0xfffe00c9,0x84821803,0x2e889f2,0x1030204,0x3030303,0x0,0x0,0x1201d,0x0,0x0,0x0,0x0,0x0 tzsz=48m no_console_suspend minigzip=1 forbid_uart0=1 perfor_uart0=0 gpustart=924712960 gpusize=0 gpuionsize=0 androidboot.serialno=000070af2414fb93 adb_enable=0 androidboot.hardware=mt5891 bootreason=reboot modelgroup=PHTV17AMT0 device_name=65OLED873/12
eboot recoveryB

Make sure that the root= partition matches the one you are booting with eboot, or your kernel will use a different initrd and you might get surprising results.

make the loader target and adb push it to /data/local/tmp or so. The first run takes about two minutes, as the loader needs to acquire a thread dump to locate the running kernel module address. It relocates the bundled getroot.elf to the address of the _CmdVersion handler, writes it there, executes it via b.ver and execve()s a shell.

See the install-magisk script to install Magisk for a more permanent way of acquiring root. It does not seem to affect OTA updates in any way, except that it needs reinstallation after an update.

/dev/cli is the debug interface to dtv_driver.ko. It supports some ioctls (cf. cli_ioctl()).

• ioctl(fd, 0, &char): Places a character into the UART input buffer.

• ioctl(fd, 1, &str): Passes a (0-terminated) string to CLI_Parser().

• ioctl(fd, 0x100): Acquires a semaphore.

• ioctl(fd, 0x101): Releases this semaphore.

• ioctl(fd, 0x102, void*): Calls DMX_SetCapture() with the user-supplied argument.

• ioctl(fd, 0x103, void*): Calls DMX_GetCaptureInfo() with the user-supplied argument.

I haven’t played around with any other ioctl than 0 or 1.

/dev/cli is writable by the shell user, but some (obscure) commands return EPERM. That said, most don’t. cli_shell is a pre-installed binary that does (essentially) the following:

int fd = open("/dev/cli", O_RDWR);
system("echo 0 > /proc/sys/kernel/printk");
char *command = append(join(" ", argc, argv), "\r\n");
char c;
while (c = *command++)
  ioctl(fd, 0, &c);
system("echo 7 > /proc/sys/kernel/printk");
usleep(200000);

It is probably meant for a user on the serial console, to which the kernel log would get printed. With adb access, you can simply run logcat -s MTK_KL|tr \\r \\n.

The cli in this repository simplifies the access via adb. It only prints the actual command output before exiting, by placing a sentinel alias before and after the command, all of which are sent with ioctl 1 instead of 0. I couldn’t see a difference. It sometimes fails; if there is no output after a couple of seconds, and the CLI isn’t blocked (e.g. by a running b.da), try hitting ^C and try again.

ls gets you a list of (most) commands. See cmdlist-guest.txt and cmdlist-sv.txt for mine. The CLI knows a default “guest” mode and a higher-privileged “supervisor” mode. The b.sv command is apparently meant to detect this based on some external trigger, but b.scm 0 (set cli mode?) lets us handily switch to supervisor mode and b.scm 2 back. See _arDefaultCmdTbl in dtv_driver.ko for all commands - I couldn’t find any interesting hidden ones (// for line comment is one).

Generally, the something.q commands seem to query only and are probably harmless. .d_on, .d_off and .d_l set/show debug/log levels.

Some random notes:

• cust or customer refers to Philips/TPVision. It’s likely that dtv_driver.ko was mostly written by Mediatek.

• The kernel and the modules are always loaded in the same order at the same addresses.

• MID refers to a “Memory Intrustion Detection” and probably is some HDCP Nonsense. It does nothing about the kernel, however, which starts at 0xc0008000.

• linuxmode puzzles me a bit, it seems to move the base offsets for the memory read/write commands only. My current guess is that some other CPU (which is the “Linux” one, as opposed to the “Android” one) has its memory mapped to that offset and one can poke around on the other CPU.

• The code has little error checking, to put it mildly. E.g. the do command for repeating a CLI command locks up with do 0 ls.

• b.da gives a thread dump of the kernel portion of all tasks.

• dtv_driver calls its own CLI_Parser (which has a global mutex) every once in a while.

• The printk()s are full of carriage-returns that you may want to strip off when viewing logcat output.

I haven’t really played around with anything else than the memory read/write operations to get some code running. But see the Makefile for how to link and run code in the kernel.