Systemd, Mobile Linux, and Containers

I’ve had some problems running apps I want on my Furilabs FLX1s [1], so I decided to install some container environments to test various versions. I started with Debian/Testing so I can test the build process for some packages I’m about to upload to Unstable.

Systemd Issues

When running debootstrap testing testing to setup the chroot the process aborted with errors including the following from the systemd postinst:

Failed to enable units: Protocol driver not attached.
Cannot open '/etc/machine-id': Protocol driver not attached

This turned out to be from trying to run systemctl in the postinst, I just removed the “set -e” line from /chroot/testing/var/lib/dpkg/info/systemd.postinst and kept on going (I’m not planning to actually use systemd so it’s failure to setup wasn’t a problem).

Then I installed a bunch of -dev packages needed to build my package which had a dependency chain that included udev leading to the following error:

Setting up udev (260.1-1) ...
Failed to chase and open directory '/etc/udev/hwdb.d', ignoring: Protocol driver not attached
Failed to chase and open directory '/usr/lib/udev/hwdb.d', ignoring: Protocol driver not attached

Udev is also a part of systemd.

Googling for this turned up a closed systemd bug about this indicating that it has a minimum kernel version of 5.10 [2]. The Furiphone has kernel 4.19.325-furiphone-radon due to being based on Android.

Checking the kernel version isn’t that hard to do, if the systemd programs in question checked the version and reported “can’t run on kernels prior to 5.10 then it would avoid a lot of confusion – and also bug reports that the systemd developers don’t want.

Some Debian package dependencies can probably do with revision. Installing the packages “libkdb3-dev libkf5archive-dev qtdeclarative5-dev qtpositioning5-dev qttools5-dev” ideally wouldn’t have a dependency chain leading to udev.

The Furilabs people appear to have patched the latest Debian version of systemd to work with the older kernels, the version is currently 260.1-1+furios0+git20260425023744.8401044.forky.production.

Compile Times

I got this working by just editing every postinst script and either removing the “set -e” or adding an “exit 0” at the top, I don’t need things to be configured properly for a running OS I just need the files in the right locations for a container.

One issue I discovered when I started compiling is that it was only running on 1 core and the “nprocs” program was returning “1”. The “lscpu” program showed that only 1 of the 8 cores was online, it was a single Cortex-A78 core. Some combination of putting it in “caffeine mode” and having the screen on enabled all 6*Cortex-A55 and 2*Cortex-A78 cores.

The below table compares compiling Harbour-Amazfish on the Furiphone with all 8 CPU cores active, my E5-2696 v4 workstation (almost the fastest socket 2011-3 CPU ever made), running ARM64 software emulation on a system with two E5-2699A v4 CPUs, and a Radxa 8 core ARM SBC (which I will review in a future blog post).

Given that the source apparently limits the parallelism to less than 7 cores on average it’s pretty impressive for the elapsed time to be only 2.5* longer on the phone. Emulating the ARM64 build at about 4* the system CPU time is impressive too, as the system has 4.5* as many CPU cores it could theoretically compile ARM code faster than the native ARM hardware I own for any project that uses enough cores.

• [1] https://etbe.coker.com.au/2026/04/14/furilabs-flx1s-finally-working/
• [2] https://github.com/systemd/systemd/issues/41250