Modular kernel for the Gemini PDA available from the gemian repository

Recently Adam Boardman and I have managed to integrate the modular kernel for the Gemini PDA into the gemian kernel repository. So, from now on, whenever the kernel gets improved, the modular kernel gets built and the update is available for Debian via apt.

Installing the modular kernel

From now on, the modular kernel for the Gemini PDA can be installed easily using apt:

sudo apt install gemian-modular-kernel

When using the new bootloader the kernel gets flashed to the boot partition automatically. This works because the new bootloader passes the current boot partition’s name to the kernel using the kernel cmdline. The cmdline can be examined with:

cat /proc/cmdline

When using the old bootloader with the Gemini PDA this information is not available to the kernel and consecutively to the operating system, thus one still has to flash the kernel image manually after installing or updating the kernel package. For this one has to carefully decide which boot partition the Linux system is being booted from. Using the wrong partition name can render other installed operating systems unbootable. To recover, flashing the wrongly overwritten boot partition using the flash tool might be necessary. When knowing the boot partition the new kernel can be flashed using dd (the X in bootX has to be replaced with the number of the boot partition) as shown below.

sudo dd if=/usr/share/kernel/linux-boot.img of=/dev/disk/by-partlabel/bootX

After flashing the kernel the either or the other way a reboot is necessary. The boot partition number can be determined from the scatter file that has been used initially to flash the Gemini. Alternatively it can be found out from the key combination that has been used to boot the Gemini. Detailed information on this can be found in the Gemini bootloader documentation.

Building out of tree modules

For building out of tree modules with the Gemini (in example for using USB devices that are not supported with the kernel), in addition to the kernel, the kernel-headers package has to be installed:

sudo apt install gemian-modular-kernel-headers

With the kernel headers and the appropriate build toolchain (gcc, etc.) additional kernel modules can be compiled on the Gemini. Instructions on how to do this can usually be found with the module source.

Some prebuilt modules (iptables mirror target, frandom, 88XXau) for the kernel can be downloaded below:

gemini-modules-extra-3.18.41.tar.gz (117 downloads)

To use the modules, the downloaded archive has be extracted to the root directory. Afterwards depmod has to be executed:

cd /; sudo tar -xzf /path/to/gemini-modules-extra-3.18.41.tar.gz; sudo depmod

With high probability these modules should still be usable after upgrading the modular kernel to a newer build.


1 Star2 Stars3 Stars4 Stars5 Stars (2 votes, average: 5.00 out of 5)

Building a kernel module for the awus1900 Wifi stick and the Gemini PDA

A few days ago I have been asked if it is possible to build a driver for the awus1900 Wifi stick for the Gemini PDA. To be honest, I did not know, so I gave it a try.

The awus1900 uses Realtek’s rtl8814au chipset. The Linux driver for this chipset is available at many locations around the net. Most ones, I have tried, have not been compilable against the Gemini’s kernel. The driver at has been compilable with some minor modifications against the kernel source used for the kernel in Modular Linux kernel for the Gemini PDA with lid close fix.

First of all some parameters in the Makefile had to be changed to match the Gemini:

  • CONFIG_PLATFORM_I386_PC = n (disable x86 build)
  • CONFIG_PLATFORM_ARM64_RPI = y (enable arm64 build)

Some more parameters have been enabled for features in the hope that these do not cause problems:

  • CONFIG_80211W = y

With all these changes the build fails complaining about STATION_INFO_SIGNAL and many more being undeclared. The module’s source expects these defines to be present in the kernel source for kernels below version 4.0. Most probably the Gemini kernel tree is different than other 3.x trees. So the line 23

#if (CFG80211_API_LEVEL >= KERNEL_VERSION(4, 0, 0))

in os_dep/linux/ioctl_cfg80211.c has been replaced with


to get the module source build against the Gemini’s kernel. Afterwards it has been possible to cross compile the kernel module by running make:

make ARCH=arm64 KSRC=/path_to_lib_modules_dir/3.18.41+/build

After building the module it can be copied to /lib/modules/3.18.41+/extra/ (or any other proper directory) on the Gemini and used afterwards. For those who do not want to build the module themselves, the binary modules for the kernel shared in the article Modular Linux kernel for the Gemini PDA with lid close fix can be downloaded from here: (92 downloads)


1 Star2 Stars3 Stars4 Stars5 Stars (1 votes, average: 5.00 out of 5)

udev-182 needs CONFIG_ DEVTMPFS in kernel

After the latest upgrades on my gentoo vserver system running a 3.3.0 Linux  vserver-kernel (vserver-sources-, the system did not start up properly anymore. No kernel modules got loaded and even the network devices have not been available after a reboot. This is more or less the worst case, since then one has to be physically in front of the machine and can not repair the system via ssh remote login.

The kernel upgrade was not the reason for this,  but the upgrade to udev-182. This is what the log said:

Mar 21 17:20:05 mittelerde /etc/init.d/sshd[5563]: ERROR: cannot start sshd as net.eth0 would not start
Mar 21 17:20:09 mittelerde /etc/init.d/udev-mount[6075]: Udev uses a devtmpfs mounted on /dev to manage devices.
Mar 21 17:20:09 mittelerde /etc/init.d/udev-mount[6076]: This means that CONFIG_DEVTMPFS=y is required
Mar 21 17:20:09 mittelerde /etc/init.d/udev-mount[6077]: in the kernel configuration.
Mar 21 17:20:09 mittelerde /etc/init.d/udev-mount[6067]: ERROR: udev-mount failed to start
Mar 21 17:20:09 mittelerde /etc/init.d/udev[6066]: ERROR: cannot start udev as udev-mount would not start
Mar 21 17:21:06 mittelerde /etc/init.d/net.eth0[6463]: ERROR: interface eth0 does not exist

With the information “CONFIG_DEVTMPFS=y is required” the log contains the necessary hint to get things to work. The CONFIG_DEVTMPFS option had to be enabled in the kernel. Afterwards the kernel has to be recompiled. The option can be found in menuconfig under Device Drivers-> Generic Driver options and is called Maintain a devtmpfs filesystem to mount at /dev.  For getting the devfs automatically mounted at boot time it makes sense to also enable the option Automount devtmpfs at /dev, after the kernel mounted the rootfs (CONFIG_DEVTMPFS_MOUNT).

It is safe to enable these options with older udev versions. Doing so protects your system from not working any more when you get the udev update later.


1 Star2 Stars3 Stars4 Stars5 Stars (2 votes, average: 5.00 out of 5)

nvidia-drivers-295.17 solve black screen problem

Using nvidia linux drivers from version 270.X to 275.X with some graphics boards, in example the Quadro FX 350M which is built into the Dell Precision M65 notebook,  resulted in a black screen or window for OpenGL applications. Even glxgears did only output a black window. The problem has been discussed on A downgrade to a lower driver version, in example the version 260.X drivers is not applicable anymore, since these drivers do dot build against a recent linux-3 kernel. A upgrade to newer drivers also was not possible, since driver versions from 285.X to 295.10 did not work at all for this graphics board. Recently version 295.17 of nvidias beta driver has become available, which solves this issue. Download links are available on

For gentoo users I have modified the nvidia-drivers  ebuild for the 295.17 driver. You can download my modified overlay, [download#83] and extract it in /usr/local/portage. Be sure to include the following line in your /etc/make.conf:


Afterwards you may emerge nvdidia-drivers-295.17.




1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)

tuxonice problems with nvidia-drivers-270.41.06

A few days ago nvidia-drivers-270.41.06 appeared in the gentoo partage tree. During my regular updates it got installed. Afterwards turonice did not work anymore with the configuration I wrote about in zen-sources-2.6.38_p20110501 with tuxonice.

The dmesg output during supend to disk was:

Freezing processes & syncing filesystems.
Stopping fuse filesystems.
Freezing user space processes … (elapsed 0.01 seconds) done.
Stopping normal filesystems.
Freezing remaining freezable tasks … (elapsed 0.01 seconds) done.
Preparing Image. Try 1.
Restarting normal filesystems.
Stopping fuse filesystems.
Freezing user space processes … (elapsed 0.00 seconds) done.
Stopping normal filesystems.
Freezing remaining freezable tasks …
Freezing of tasks failed after 20.00 seconds (1 tasks refusing to freeze, wq_busy=0):
khugepaged      R  running task        0   619      2 0x00800000
ffff88004ef24fa0 ffffffff81692eb9 ffff88004ef24fa0 ffff8800cf13e0c0
ffff8800c861e000 ffff8800c861e010 ffff8800c861e000 ffff8800cf13e330
ffff8800c861ffd8 ffff8800c861e010 ffff8800cf13e148 ffff8800cf13e330
Call Trace:
[<ffffffff81692eb9>] ? schedule+0x729/0xea0
[<ffffffff8169308b>] ? schedule+0x8fb/0xea0
[<ffffffff8110d4ab>] ? __page_check_address+0x10b/0x190
[<ffffffff8110d9f6>] ? page_referenced_one+0x96/0x220
[<ffffffff8110eb7a>] ? page_referenced+0x2da/0x370
[<ffffffff810f13fd>] ? shrink_page_list+0x21d/0x5d0
[<ffffffff810f1b9e>] ? shrink_inactive_list+0x15e/0x4d0
[<ffffffff812d74b8>] ? __next_cpu+0x18/0x30
[<ffffffff8103877b>] ? resched_best_mask+0x3b/0xd0
[<ffffffff810f22bb>] ? shrink_zone+0x3ab/0x500
[<ffffffff810f32bf>] ? do_try_to_free_pages+0xcf/0x470
[<ffffffff810f3965>] ? try_to_free_pages+0xa5/0x1c0
[<ffffffff810e7d5c>] ? __alloc_pages_nodemask+0x49c/0x900
[<ffffffff81050496>] ? try_to_del_timer_sync+0x76/0x110
[<ffffffff81050690>] ? process_timeout+0x0/0x10
[<ffffffff81126461>] ? khugepaged_alloc_hugepage+0x51/0xe0
[<ffffffff81061c70>] ? autoremove_wake_function+0x0/0x30
[<ffffffff811266bd>] ? khugepaged+0xad/0x10e0
[<ffffffff81061c70>] ? autoremove_wake_function+0x0/0x30
[<ffffffff81126610>] ? khugepaged+0x0/0x10e0
[<ffffffff81126610>] ? khugepaged+0x0/0x10e0
[<ffffffff81061706>] ? kthread+0x96/0xa0
[<ffffffff8103c6af>] ? schedule_tail+0x4f/0x110
[<ffffffff81003bd4>] ? kernel_thread_helper+0x4/0x10
[<ffffffff81061670>] ? kthread+0x0/0xa0
[<ffffffff81003bd0>] ? kernel_thread_helper+0x0/0x10

Cleaning up…
Restarting all filesystems …
Restarting tasks … done.
video LNXVIDEO:00: Restoring backlight state
TuxOnIce debugging info:
– TuxOnIce core  : 3.2
– Kernel Version :
– Compiler vers. : 4.4
– Attempt number : 1
– Parameters     : 17 700428 2 1 -2 4
– Overall expected compression percentage: 50.
– Compressor is ‘lzf’.
– Block I/O active.
Used 0 pages from swap on /dev/sda6.
– Max outstanding reads 1. Max writes 0.
Memory_needed: 1024 x (4096 + 360 + 112) = 4677632 bytes.
Free mem throttle point reached 0.
– Swap Allocator enabled.
Swap available for image: 2098481 pages.
– File Allocator active.
Storage available for image: 0 pages.
– No I/O speed stats available.
– Extra pages    : 0 used/500.
– Result         : Hibernation was aborted.
: Freezing filesystems and/or tasks failed.

Somehow there seems to be a problem with nvidia-drivers-270.41.06. After downgrading to nvidia-drivers-260.19.44 tuxonice worked as usual.


This solution was partly wrong. As I noticed later, the problem still persists occasionally. With the older nvidia-driver the probability that the problem occurs just was lower. However, the main reason  for the tuxonice abort is some incompatibility between tuxonice and the new transparent hugepage feature of the kernel. As a workaround one may disable the transparent hugepage support before suspension and reenable it afterwards. I.e. one can include something like the following in his hibernate configuration:

OnSuspend 90 echo ‘never’ > /sys/kernel/mm/transparent_hugepage/enabled

OnResume 90 echo ‘always’ > /sys/kernel/mm/transparent_hugepage/enabled


OnResume 90 echo ‘madvise’ > /sys/kernel/mm/transparent_hugepage/enabled


1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)

Flightgear with VR920 headtracking

Recently I basically got Flightgear to work with quad buffered stereo. The only thing that was still missing for having the Vusix VR920 head mounted display fully supported in the flight simulator was headtracking.

However, with my new headtracking driver, VR920 headtracking in Flightgear is possible at last. A good part of the work has been done by Anders Gidenstam who provided the original Nasal module, the headtracking protocol description and usage instructions for his webcam based headtracking solution for Flightgear.

Download and copy the protocol description [download#59] to $FG_ROOT/Protocol. For me (gentoo system) this location is /usr/share/games/FlightGear/Protocol/, probably for many others it is /usr/share/FlightGear/Protocol/

Afterwards download unzip the modified Nasal module [download#58] to ~/.fgfs/Nasal. It is important to use your home directory and NOT i.e. /usr/share/games/FlightGear/Nasal/.

Then make sure that the vr920 headtracking driver runs in UDP mode. If running Flightgear on the same machine as the headtracking driver, which should be the usual case, just use as destionation IP for the driver and use 4242 as destination port. These are the default settings of the driver.

Finally run Flightgear with these options: –generic=socket,in,<hz>,,<port>,udp,headtrack –prop:/sim/headtracking/enabled=1

If you also want to have quad buffered stereo with it (you need an nvidia quadro board, with assumably a pre G80 Chip or probably an ATI FireGL, never tried that, and a stereo enabled xserver) use the patch from FlightGear with quad buffered stereo. For instructions on how to get the xserver to work in stereoscopic mode see: Vuzix VR920 with Linux and active 3D stereo

For the described configuration you can use the following little startup script:

fgfs –generic=socket,in,25,,4242,udp,headtrack –prop:/sim/headtracking/enabled=1

Now have much fun and enjoy a new experience with your VR920 and Flightgear in stereo with headtracking.

best regards


1 Star2 Stars3 Stars4 Stars5 Stars (1 votes, average: 5.00 out of 5)

VRTrack 1.0 – headtracking driver for the vr920 HMD

As I promised in New version of the vr920 headtracking driver coming soon here is the new version of my headtracking driver for the Vuzix VR920 iwear for Linux. It calculates yaw, pitch and roll from the accelerometer and magnetometer data (The device has got three of each). This makes a 3DOF tracking possible and allows you to look around in a 3D Scene.  In example you can use the driver with my stereoscopic image viewer SIV. The driver averages the sensor readings with an improved algorithm, which gives a far smoother experience than with the initial driver version. The driver package consists of a daemon which can be run in the background and for convenience a basic control application that enables one to easily tweak the various driver settings and to callibrate the device. For general Information on how to use the device with Linux see: Vuzix VR920 with Linux and active 3D stereo.

The driver provides the trackingdata in different formats to the application using it. It always writes the data to /dev/headtracking. A line read from /dev/vrtrack consists of six floats that correspond a sensor reading in this format:

yaw pitch roll x y z

Yaw, pitch and roll are angles from 0 to 360 degrees. X, y and z are always zero for the vr920, since it only supports three degrees of freedom. These values are reserved for future devices which may support six degrees of freedom, in the hope to propose a standard for tracking devices.

The driver can scale the readings and invert the axes independantly to get the needed value range for the used application and a pleasant experience.

For maximum compatibility with existing applications there are four other modes of operation available that can be enabled separately:

  • Joystick emulation
    The driver emulates a joystick device /dev/input/jsX. The readings for yaw, pitch and roll are the X,Y and Z axis of the emulated joystick. This may be used to enable basic headtracking support in games that do not natively support headtracking.
  • Mouse emulation
    The driver emulates a joystick device /dev/input/mouseX. The readings for yaw and pitch are being translated to X and Y of the mouse device, so when you look right the mouse pointer moves to the right and when you look up the pointer moves upwards and vice versa.  This may also be used to enable basic headtracking support in games that do not natively support headtracking. It can also be used to just control the mouse pointer of the window system. Controlling the viewport of the window system can also be a resonable purpose. With the new MPX extension in xorg this may be possible.
  • UDP – network
    In UDP mode the driver sends the tracking data via network as UDP unicast. The approach to send the data out via network makes the language used for writing the application independant from the language used for developing the driver. The packet sent to the clients contains the three angles, yaw, pitch and roll and x,y and z as 32 bit fixed point in Q16.16 format. This mode may i.e. used to control flightgear.
  • Multicast – network
    In multicast mode the driver sends the tracking data via network as UDP multicast, thus many clients may read the data, which makes parallelization more possible, i.e. one could use one machine for rendering and another machine for calculations. In addition to this, the approach to send the data out via network makes the language used for writing the application independant from the language used for developing the driver. The tracking data sent to the clients contains the three angles, yaw, pitch and roll and for easy usage a viewmatrix, one can directly use with scenegraph libraries. If you intend to develop an application using the headtracking of the VR920 see the file democlient.cpp included in the download for details on how to get the data into your application. This mode is used by the stereoscopic image viewer SIV.
Below is a screenshot of the control application during callibration of a vr920 device:
control_appvrtrack driver during calibration ( screenshot)

Important note: During calibration make sure that the display of the device is displaying something. Since the displays not only showing a blue screen influences the sensor data (at least with my device) you’ll end with wrong calibration else. You may use i.e. nvidia-settings to ensure this. For detailed usage instructions see the readme included in the download.


I decided to publish the driver under the creative common noncommercial license. You may download the full source from here: [download#55], an x86_64 binary from here: [download#56] , or an i686 binary from here: [download#57]. An Archlinux PKGBUILD provided by Feilen is available here: More binary/distribution specific formats may be available in the future. The x86_64 binary has been build on an up to date gentoo system, the i686 binary on ubuntu hardy. For the i686 binary you may install libconfig++ i.e. libconfig++8_1.3.2-2 from here: libconfig++ If none of the binaries works for you, you may have to build from source…

You need to have libusb, libconfig++, libfuse and libcurses installed on your system. For ubuntu users I included the small shell script that installs the dependencies. Maybe it works also for for other Debian-based distributions. Gentoo users just have to make sure that  libusb, ncurses, fuse, and libconfig have been emerged. Your kernel version has to be at least 2.6.31 and you must have cuse enabled in your kernel.


If you like the driver, feel free to link to If you developed an application using the tracking data provided by the driver please leave a comment, because then I can review the application and eventually write about it. To request commercial licenses contact us at info(at) Well, if you just want to support our work on use the donate button 😉

best regards


1 Star2 Stars3 Stars4 Stars5 Stars (1 votes, average: 5.00 out of 5)