UniformBuffer

Hi,
i would like to try running some Android app on Armbian. I see that Anbox is available on the repository, but i cannot get running. When starting i got
 
Failed to start as either binder or ashmem kernel drivers are not loaded
 
Obviously i'm not asking support for Anbox, i would like to know how to enable binder and ashmem kernel modules. When i try to modprobe them i got:
 
modprobe: FATAL: Module binder not found in directory /lib/modules/5.6.15-meson64
modprobe: FATAL: Module ashmem not found in directory /lib/modules/5.6.15-meson64
 
On the git there is a guide on how compile and install them, but since they are kernel modules they are probably used for some other programs and maybe they could be installed in some easy way, maybe using apt. So before trying to compile and install them by myself, i would like to know if is possible to install them more easily.
 
I'm using a AML-S905X-CC board and i'm on Debian Buster image upgraded to Sid.
 
Thanks for the attention,
Have a good day

Hi,
i'm using a Raspad tablet-like device with a aml-s905x-cc inside and i'm experimenting some mobile interface that i can use with the touchscreen, but also with mouse and keyboard. I'm simply curious about convergence desktop designs and mobile desktop environments. The obvious choice would be GNOME, but it is very heavy, both on memory and cpu usage and my board with only 2GB of ram cannot handle it if combined with other heavy programs. I knew that Librem was developing a desktop environment for their Librem 5 linux phone. Since they focus on opensource software i thought if i can install their desktop environment on my tablet to try it and.... yes, it is possible! And it is surprisingly easy to do.
So, if someone would like also to try their desktop environment, i make a "guide" on how to install it.
 
Like any wayland desktop environment, there is a compositor (called phoc) and a graphical shell (called phosh). On the git they state that to run phosh should run on any compositor that implement wlr-layer-shell, so maybe you could try other compositors!
 
First clone the git of phoc with:
git clone https://source.puri.sm/Librem5/phoc
 
Change the directory on the just downloaded folder:
cd phoc
 
Then, since they provide the file needed for debian packaging, i suggest to generate the deb package and install it instead of generating the binary directly with meson.
In this way you can uninstall the program simply with apt like any other program.
 
From the control file, i can see the needed dependencies, but i suggest to get directly from the build command below.
If some package is missing, the packaging tool will tell you which. Then you can simply do "sudo apt install [all the missing dependencies]"
 
To create the debian package run:
debuild -i -us -uc -b
 
When the command finish, it should create a debian package inside the upper folder called "phoc_0.4.0_[architecture].deb"
You can install it with "sudo apt install /path to the file/phoc_0.4.0_[architecture].deb"
 
Once phoc is installed, it's time to install phosh.
Go to the previous folder with:
cd ..
 
Clone the phosh git with:
git clone https://source.puri.sm/Librem5/phosh
 
Change the directory on the just downloaded folder:
cd phosh
 
Dependencies should be these, but like before, i suggest to get them directly from the build command and install them as the same way as phoc:
 
To create the debian package run:
debuild -i -us -uc -b
 
At the end of the compilation, you should find the package inside the upper folder and like before you can install it with "sudo apt install /path to the file/phosh_0.4.0_[architecture].deb"
 
That's it. Reboot and on you favourite login manager should appear the entry "Phosh" that can be started like any other desktop environment.
 
During the debian packaging compilation it will generate some other packages, like the debugging symbols, but they are not needed to make phosh work.
Hope someone could find this "guide" useful.
Have fun!
 

Thanks for the info, like i said i'm not very skilled with kernel things, but thanks to your guide i was able to change cma allocation from 256M to 512M.
I have also tried to set 1GB with 0x40000000, that should be aligned with 0x400000, but after setting it i got no monitor output (from leds pattern i can say it was working, sysreq magic keys also worked, so the kernel was on).
Anyway, that's not foundamental, i have also tried to set the cma with the same value of the patch you linked (0x38000000) and it worked, making the cma 896MB.
 
I have tested some h264 and vp9 videos and the performance it's still choppy (after increasing to 512MB it become a little better).
I have tried mpv with `mpv --hwdec=yes video.mkv` and ffplay with `ffplay -vcodec h264_v4l2m2m video_h264.mkv` and `ffplay -vcodec vp9_v4l2m2m video_vp9.webm` .
For now i got the best performance with vp9 format using ffplay.
A strange thing is that ffplay perform better than mpv, but mpv use ffmpeg like ffmplay, so they should have more or less the same performance.
Anyway the low performance seems to be related to a VERY single threaded behavior. After increasing cma to 512MB i got some free cma left (40-80MB) and increasing again to 896MB increase the free cma proportionally, so it seems that meson_vdec does not get advantage from memory after ~512MB.
 
Anyway, even if there are some problems, i'm happy to be able to see progresses with my eyes!
Thanks again for the help and for your hard work

Hello uniformBuffer,
 
   The CMA memory is set by a device tree entry, so that would be the only way to change it to my knowledge.  The drivers are extremely unoptimized presently for the vdec/venc, and the recommended setting is over 800 MB.  This is an obvious problem for a general-purpose distribution, it renders the La Frite 512 MB boards unbootable, and leaves no real room for any working tasks on the 1 GB ones.  It may be possible to implement this as an overlay, but in the meantime modifying the device tree is the only route:
 
https://github.com/armbian/build/blob/cc7ab6a6b1d91977bd9e154245307e85f7f76519/patch/kernel/meson64-current/0302-arm64-dts-meson-set-dma-pool-to-896MB.patch
 
This patch gives you the handle and value to get the required 896 MB CMA pool.

Thankfully the kernel does not need recompiled for this, only the device tree.  See the post below, I b elieve this is still accurate as far as device tree decompile/recompile method, can be done on the device (use the correct device tree for your board )
 

I just did a quick test on Le Potato, increasing CMA to 512 MB was very easy following the above instructions with the /boot/dtb/amlogic/meson-gxl-s905x-libretech-cc.dtb device tree.  The property yo uwant to change is right at the top of the file as a handle under "reserved-memory"
 
Any value that is 0x400000 aligned should be valid, so you can experiment.
 
 

The two worst things that could happen are either the compile failes or the board does not boot after kernel replacement. So if possible make a backup of your sd card first, just in case...
 
Also take note that running a custom kernel voids support from Armbian side

Not possible via apt or any other way to sideload.
However making a working kernel package for your board is fairly easy using the build script: https://github.com/armbian/build
Clone, run, select your board and that you wish to adjust the kernel config. Done. The script will handle the rest.

if X selects the best resolution, but the image is bigger than screen, you need adjust display's overscan setting. but if there is an offset, maybe there is a setting in display but i don't know. it looks like you already had a fixed version.

you need to use alsamixer in console, select audio card with F6, and enable some switch and enlarge volume to have hdmi output.

What mean you can't login? It does not appear on the list of available desktop environment on the display manager you are using? If yes, probably the package miss to install the "x.desktop" file, where x is the desktop environment name, that is needed to the display manager to keep track of the available desktop environment in the system. If is that the case, you can start manually the desktop environment, at least to try if it work, then i will explain how to create and place them. You have to move on a different TTY (using ctrl+alt+*a number from 0 to 9*, login, and run the appropriate launch command:
-"sway" to run sway
-"weston-launch" to run weston
-"gnome-session" to run gnome (i'm not very sure, it does not work for me)
 
Like i said i have not made strange steps to run these desktop environment, i simply installed them from apt and they worked. If you give me some more information about errors and problems you get, i will help as i can

I needed to send it complete the first time, sorry.
 

 
I've run glxinfo, same as glmark2, mali450 mesa.

Sure, there are no strange steps. I tried a lot of desktop environment installed simply from apt. The only thing that i have changed on my system is that i get packages from Debian Sid, transforming my system from stable updates to rolling release, so i get more updated packages compared to stable debian . You can switch adding some lines on apt edit-sources, but this will make your system unstable and unsecure (Sid is also called Debian Unstable) and the procedure will be, mostly, irreversible without breaking the apt system. If you can use your system without update to unstable, don't do it. I suggest before to test some of the following desktop environment , then if you want, i can guide you (it is not complicated, but it is IRREVERSIBLE, so you have to know what you are doing).
The desktop environment that i have tested:
- Weston v8.0.0-1: around 200fps, it is the wayland reference compositor, so it is the slowest. You can install directly from apt using "sudo apt install weston"
-Gnome(Wayland) v1:3.30+2: around 300-350fps. You can install directly from apt using "sudo apt install gnome-core" (gnome-core is for minimal gnome installation, for full installation of gnome and applications use "gnome"). Be sure to run this on wayland since it have also xorg version. You can check using "weston-info" command. If it gives you some infos, you are on wayland.
-Sway v1.4-2: around 500-600fps. You can install directly from apt using "sudo apt install sway". I have noticed that Debian Stable does not have Sway package, only Debian Sid have it. You can get a precompiled from debian packages (https://packages.debian.org/sid/sway) or from their git (https://github.com/swaywm/sway/releases/tag/1.4).
Since differences on packages versions, sway could not be possible to install from debian package without upgrading to Sid, but you could try. Installing the package in this way should not mess up your system since dependancies are untouched: if dependancies are satisfied, apt will install it, otherwise not. Sway is a bit different from others. It is a tiling window manager, so windows behave differently from the others that are stacking window manager.
-Wayfire v0.4.0: around 600-700fps. I have compiled and installed this from the git repository (probably that the reason for that fps count, it is much optimized for my system). It is composed by 3 programs: Window Manager(wayfire),Shell (that include panel and background) (wf-shell) and Config Manager(wcm). They can be retrieved from the repository pool with all build instructions: https://github.com/WayfireWM.
You could also use wf-install that they provide. It is a script that directly download and compile the 3 component mentioned before.
If you want i can guide you to install wayfire, but i would suggest to try at least one among the other, just to see if you get better performance that xorg, the other desktop environment are simply plug and play.
 
Have fun on testing

So you suggest to change the command to "command" right? I tried it but nothing happen. Maybe your suggestion works, but the command itself doesn't. As i know, all current kernel boot parameter can be found on /proc/cmdline file, but there isn't.
 
Solution found:
I tried to use some of the texts you suggest, specifically the extraargs=video=composite-1:d (without quotes) works.
Seems that the script look for the extraargs= prefix and add everything next to kernel parameters.
It appear on /proc/cmdline and the command seems to work, now that connector is listed ad "disconnected".
Thanks a lot for the help and the dedicated time!

I would try "video=composite-1:d".. this goes in over u-boot variables, and they tend to mess up stuff.. cause it's then more like extraargs=video=composite-1:d whereas it should be extraargs="video=composite-1:d"..
 
 

Thanks for the info, now i can check if some patches are applied.
Anyway after some troubleshooting, i think that the problem with the drm has something related with the monitor (that is initializated by the lima driver? i don't know). Even with randr if i try to get some info, it report:
Searching a bit deeper i tried to get some info directly from the edid of the monitor using the program "edid-decode"(maybe the monitor is not recognized correctly).
In the /sys/class/drm/ folder i have:
card0/             card0-HDMI-A-1/    renderD128/        
card0-Composite-1/ card1/             version
 
If i check the edid in card0-HDMI-A-1, it report:
It sounds correct, but I could be wrong.
 
Instead if i try to read the edid of card0-Composite-1 i got:
EDID extract of '/sys/class/drm/card0-Composite-1/edid' failed
 
In the previous posts i have reported the "modetest" result and it says that the composite connector have 2 resolution modes, 720x576i and 720x480i, that i think they are wrong. The HDMI connector instead report a lot of resolutions, including as first 1280x800 that is the native resolution of the monitor.
 
I think that drm programs crash because they select the first connector they get from the drm resources enumeration (so the "corrupted" one).
There is any way to blacklist that connector?
Also, for curiosity, there are any other people that have the same problem (if not, probably has something to do with my strange monitor). To check i think is sufficient to be able to run weston or kmscube successfully.
 
Thanks for the attention and the dedicated effort,
Have a nice day

Here are the patches currently used in the v20.02 kernel

https://github.com/armbian/build/tree/v20.02/patch/kernel/meson64-current

After some research i discovered some error in the dmesg log regarding the lima driver initialization. Following an extract of dmesg:
 
I'm refering to the errors like: IRQ ppmmu3 not found. Maybe have something to do with the drm problem.
I found a report on the lima git where the author of the post have the exactly same error on the dmesg: https://gitlab.freedesktop.org/lima/linux/issues/26
He s905 with kernel 5.4. These should be similar to my configuration.
On the same post a patch is suggested to fix the problem and seems, from the author reaction, that worked. I don't know if this patch is already merged or not and i have not tried myself. Anyway i thought it could be useful for solving the problem, so i posted it.
Thanks again for the hard work and dedicated effort.

Echoing what Lanefu said, thanks for the info, and sorry about the rough edges!  A new set of sound patches have been submitted upstream, I will take a look at them and see if they can be used as a means of cleaning up our current implementation (targeted originally for 4.19).  
 
The memory limit seems to be a potential u-boot issue.  For the display, the Amlogic clock framework has been an ongoing project for mainline, with a lot of changes.  I will spend some quality time with everyone's favorite S905X board and get back to you.  🙂

best thing you can do is probably roll back
 
sudo apt install linux-image-current-meson64=19.11.3
 
 

Here's lepotato holding together and using lots of ram. with 2 instances of armbianmonitor -z running
 

Okay i've got patched 5.4..... sound works.... was able to use more than a gig of ram when i ran armbianmonitor -z and watched htop from the desktop
 
https://github.com/armbian/build/pull/1805

Just downloaded the new kernel with your fixes. Like you said, audio and memory bug is now fixed. Also the meminfo now report correctly:
CmaTotal: 262144 kB instead of ~1GB
Thanks again for the effort and the dedicated time!

Wow, you managed to solve it, can't wait to see your fix merged. Also, your suggestion to rollback the kernel works, now i have sound again while waiting your patch.
Thanks a lot for your blazing fast support and the effort employed to solve these problems. I would also like to contribute to development but i have not yet reached the necessary skill level.... for now :-)