Most "can it run DOOM?" projects are a punchline. Robin M.'s takes on the Spengler PSL50, a self-serve coffee kiosk, and it doubles as a short case study in how embedded Linux products get opened up. The weaknesses he found are common ones: a debug header nobody disabled, a bootloader with no password, and a firmware update scheme whose secret ships on the device it is supposed to protect. The DOOM port came at the end, once the machine was already open.
The project is written up on Robin M.'s Hackaday.io page. Hackaday covered it on 10 November 2026 and Hackster.io also reported on it. Hackaday credits the tip to robidev.
It started with a microphone
Curiosity about the hardware came before any interest in games. According to his project page, Robin M. was reading the Spengler PSL50 service manual and noticed that the screen module, listed as STUM_XXL, included a PIR sensor, a microphone, an ECO function and a speaker. A microphone in a coffee machine is an odd thing to find. When he asked, the supplier said it was meant for a future presence-detection option and was not in use. His project page says requests to investigate further were denied.
Instead of taking that answer on trust, he bought a display/controller module on eBay and started probing it.
The hardware
Hackaday describes the PSL50 as a smart self-serve, pay-to-drink coffee kiosk with a Linux touchscreen module. Both outlets name the processor as an NXP i.MX 6 with an Arm Cortex-A9 core. Neither source gives a specific i.MX 6 variant, core count, clock speed, RAM or flash size, and we won't guess at them here.
Hackster lists the module's interfaces as Ethernet, USB and microSD, plus connections for payment hardware, cup dispensers and energy-saving modules. The software is a Yocto-based embedded Linux. Hackaday calls it a 2013-era Linux build, and Hackster puts the environment at Linux 3.10.17 with glibc 2.18. That age matters later in the story.
Step one: an unlabelled header that turned out to be a console
Hackaday says the debug port was found by probing. Hackster adds detail: an unidentified connector on the board was a 3.3 V UART running at 115200 baud. With a Raspberry Pi connected as the serial adapter, the board printed its boot console.
None of this is unusual. Many production boards keep a UART populated because it is the cheapest diagnostic tool a factory or repair technician has. What matters is what the console gives access to, and on this board it gave access to nearly everything.
Step two: U-Boot with no lock
Both outlets report that the U-Boot bootloader had no password and no other protection. Hackaday notes that a setenv was enough to get a sh prompt. Hackster lays out the sequence: change the kernel command line, launch /bin/sh at boot instead of the normal init process, mount the filesystem, and change the root password. After that he had a persistent root login on the device.
This is a well-known weakness. If the bootloader environment can be edited from the serial console, then nothing that boots after it can be trusted to enforce its own security. Common mitigations include a locked or compiled-in environment, a disabled or password-protected autoboot interrupt, and a verified boot chain. None of the sources say why they were missing here. Hackaday's description of a 2013-era build points to a platform designed before such hardening was routine.
Step three: the update mechanism
The finding with the widest consequences is the USB update path. Hackster reports that the update routine looks for an autorun.sh script and an accompanying hash on attached USB storage. In principle a check like that stops anyone from running arbitrary scripts on the machine. In this case the verification depends on a secret stored in plaintext on the machine itself, so once Robin M. had read the filesystem, he could build an update package the device would accept as valid.
This matters more than the UART for one practical reason. Using the UART means opening the unit and reaching a header on the board. A USB update path, depending on how a particular installation exposes it, is designed for field use. Integrity checking only holds up when the key that validates updates cannot be recovered from the device being updated. Asymmetric signing, where the device holds only a public key, is the usual approach. A shared secret stored in the clear amounts to a checksum with extra steps.
Robin M.'s project page says he used this mechanism to run his own code. Neither the project page nor either news report says the attack works without physical access to the module, and none of them describe a remote vector.
Step four: DOOM on the framebuffer
With root and code execution, the DOOM port was mostly a packaging problem. Hackaday reports that he used fbDOOM, a port that renders directly to the Linux framebuffer and so needs no X server or compositor. Because the system libraries on the device are about 13 years old, he compiled it as a self-contained binary rather than linking against the libraries on the device. That is a sensible approach for any legacy embedded target where rebuilding the vendor's userland is not an option.
What It Means for Makers
This is not a product launch. Nothing here can be bought, pre-ordered or backed. The only acquisition detail in the sources is that Robin M. got his module on eBay. None of the sources give a price for the module or the kiosk, so we don't quote one.
The project is useful in three ways.
- For hardware hackers and right-to-repair tinkerers, surplus kiosk and appliance controllers are inexpensive, capable ARM Linux boards with displays attached. The method shown here is the standard one: read the service documentation, find the UART, interrupt the bootloader, get a shell. It still works on more commercial gear than it should.
- For embedded engineers shipping products, the project is a list of what to fix. Restrict or password-protect the bootloader console in production builds. Don't let the kernel command line be edited from an open serial port. Don't validate update packages with a secret the device stores in readable form. Use public-key signatures so the device can verify updates but can't create them. The secret on the device is the actual weakness. Robin M.'s Raspberry Pi only exposed it.
- For educators, the project is a compact, real-world example that covers serial consoles, bootloader environments, Linux init, update integrity and cross-compiling static binaries for an old userland, all on a single board.
The microphone question that started it all ended anticlimactically. The supplier said it was an unused provision for future presence detection. According to Hackster, Robin M. ran a custom diagnostic script that checked the hardware, networking, audio playback and recording, found no evidence that the machine was listening to anyone, and noted that most configurations apparently lack the necessary audio hardware altogether.