Chrome OS for N2(+)

[AreaScout]

Hi all

I have successfully build and patched chromium browser to run Ozone-gbm abstraction layer with Aura window system with Panfrost GPU drivers, soon there will be an Debian package or Debian minimal image (bullseye) to test with, here are some screenshots running it

Here is the official explanation:

Ozone is a platform abstraction layer beneath the Aura window system that is used for low level input and graphics. Once complete, the abstraction will support underlying systems ranging from embedded SoC targets to new X11-alternative window systems on Linux such as Wayland or Mir to bring up Aura Chromium by providing an implementation of the platform interface.

https://chromium.googlesource.com/chrom ... verview.md

Screenshot 2021-09-03 9.11.27 PM.png (121.07 KiB) Viewed 24805 times

Screenshot 2021-09-03 9.16.02 PM.png (106.93 KiB) Viewed 24805 times

Screenshot 2021-09-03 9.19.21 PM.png (1.05 MiB) Viewed 24805 times

Screenshot 2021-09-03 9.21.54 PM.png (473.71 KiB) Viewed 24805 times

Screenshot 2021-09-03 9.23.42 PM.png (62.58 KiB) Viewed 24805 times

RG

[mad_ady]

This is going to be interesting to see...

Pardon my ignorance, but what exactly is chrome os (I know it's a browser-only OS for chromebooks)? In this case it's linux, but what gui is it? Google's? Can you run regular Linux apps? Can you run Android apps?

[AreaScout]

This is going to be interesting to see...

Pardon my ignorance, but what exactly is chrome os (I know it's a browser-only OS for chromebooks)? In this case it's linux, but what gui is it? Google's? Can you run regular Linux apps? Can you run Android apps?

ChromeOS is exactly what you said, it is made as a browser OS only, first apps could only be installed within the browser but today apps can be installed from the google web store and as I understand and google wrote, they want it to run on as much platforms as possible I don't think it can run regular Linux apps as it has it's own windowing system Aura, do not ask me how to run Android Apps on it, I don't know, I only have read it should be possible

So it has it's own windowing system and a browser, so the most important things anyone wants to do, you can do with it -> write emails, surf the Internet, Watch on Streaming Platforms like Disney, Netflix, install apps, have your own file manager a.s.o, the fun thing is that this is already compiled in your local chrome browser as of version m87, you only have to start chrome with this flags ./chrome --enable-features=UseOzonePlatform --ozone-platform={x11/wayland} x11 or wayland, when it comes for DRM/GBM backend like this build has (running from console only distributions) Ozone-GBM uses the GBM API to create buffers. However, it doesn’t use Mesa’s GBM implementation, but ships its own in the form of the minigbm library, you need this library and it has to be modified to run on the N2 !

This GBM version, is able to run without GUI like I said (X11, Wayland) on e.g. NAS systems, Gaming OS with Emulationstation to quickly start the Browser and surf the internet, write emails, watch photos on you NAS, watch Netflix and other streaming services on your Gamestation, install messengers a.s.o

I will look if I can upstream my source code changes too, so that the chrome package from your local Linux distribution will support your platform, but if you want to try it earlier you have to try my build

RG

[brad]

Yes very interesting, I am wondering if it is restricted to 32 bit linux installations or if it just runs ChromeOS (or an implementation of it) in 32 bit mode on 64 bit linux.

[AreaScout]

I have compiled chromium as arm64, so it's a native arm64 application, it runs ChromeOS with an compatibility layer called Ozone, dunno if there are some differences (e.g. application addons) to ChromeOS on a chromebook, because I don't have one

[odroid]

Great OS!
Time to consider building a home made Chromebook

[AreaScout]

Screenshot 2021-09-05 5.43.20 PM.png (798.75 KiB) Viewed 24660 times

Ok time for you to test it if you want

Download the Armbian Bullseye image from here https://www.armbian.com/odroid-n2/#kernels-archive-all and flash it

After booting you have to login and create a user

As root user add meverics backport repo to install latest mesa

Code: Select all

cd  /etc/apt/sources.list.d/
wget https://oph.mdrjr.net/meveric/sources.lists/meveric-bullseye-backports.list
wget -O- http://oph.mdrjr.net/meveric/meveric.asc | apt-key add -
apt update

Add your new user to groups needed, exchange testUser with your chosen user name

Code: Select all

usermod -aG sudo,adm,audio,operator,input,video,render,tty,staff,games,users,plugdev,netdev,disk testUser

Login with your new created user

Code: Select all

su - testUser

Download my pre-compiled chromium package and extract it

Code: Select all

wget https://www.areascout.at/chromium-OzoneChromeOS.tar.gz
tar xvf chromium-OzoneChromeOS.tar.gz
cd chromium-OzoneChromeOS

Install some package dependencies

Code: Select all

sudo apt install libnss3 libnspr4 libdbus-1-3 libexpat1 libxcb1 libxkbcommon0 libasound2 fontconfig-config libgbm1 libgl1 libegl1-mesa libgles2-mesa

Now start ChomeOS for the first time

Code: Select all

./chrome --ozone-platform=drm --force-system-compositor-mode --login-profile=user --user-data-dir=/home/area/.config/google-chrome-gbm --use-gl=egl --in-process-gpu --crosh-command=/bin/bash

--force-system-compositor-mode is mandatory, --login-profile and --user-data-dir is just for testing to do not overwrite your default profile

More command line switches to experiment you can find here https://peter.sh/experiments/chromium-c ... -switches/

My source code changes are here https://pastebin.com/J3JS9QRN libminigbm didn't had to be changed, I was mistaken, so this is the only change you have to make if you want to build for your own

If you want to build it for your own you have to build it on a i386 (x86_64) machine, the build system only supports this arch, it will download cross compilers, my args.gn looks like this:

Code: Select all

use_ozone = true
target_os = "chromeos"
target_cpu = "arm64"
use_system_minigbm = false
use_glib = false
use_gtk = false
ozone_platform = "drm"
ozone_platform_drm = true
is_debug = false
is_official_build = false
use_xkbcommon = true
dcheck_always_on = true

Also interesting for you to read if you want to compile it https://chromium.googlesource.com/chrom ... one_drm.md

And lastly here is the biggest downside what this thing has now, your Meson kernel does not support a cursor plane but only a universal plane https://github.com/torvalds/linux/blob/ ... ane.c#L550 which is needed to draw a cursor, without a cursor you are moving blind with your mouse and the whole thing is only usable if you have a touch screen, touch screens can be purchased at

https://www.hardkernel.com/shop/odroid- ... apability/
https://www.hardkernel.com/shop/odroid- ... apability/

They work out of the box with this kernel

Have fun

[hatsu]

Hi @AreaScout,

And lastly here is the biggest downside what this thing has now, your Meson kernel does not support a cursor plane but only a universal plane https://github.com/torvalds/linux/blob/ ... ane.c#L550 which is needed to draw a cursor, without a cursor you are moving blind with your mouse and the whole thing is only usable if you have a touch screen, touch screens can be purchased at

Wonder whether @meveric new upstream kernel 5.13.11-arm64 supports a cursor plane?

[AreaScout]

I don't think so, the only chance for this to use with a mouse on meson kernel is that the cursor will be drawn in the universal plane, if support is already there for this then I didn't get it to work, if not someone has to make support for it

Another thing is that the GPU governor is simply_ondemand, but we need performance for it to render smooth, performance governor is not select able, so you have to stress out the GPU to scale up to full performance until the UI becomes smooth

RG

[mad_ady]

Another thing is that the GPU governor is simply_ondemand, but we need performance for it to render smooth, performance governor is not select able, so you have to stress out the GPU to scale up to full performance

Isn't this related to armbian kernel config? You could presumably compile a kernel with performance governor for gpu, right?

[AreaScout]

Isn't this related to armbian kernel config? You could presumably compile a kernel with performance governor for gpu, right?

Sure you are right, it's not a tragedy, just want it to point out, not that you wonder if the UI stutters sometimes, the reason is the governor, just start youtube or webgl content and the UI runs flowless

[AreaScout]

Screenshot 2021-09-11 5.42.25 PM.png (446.63 KiB) Viewed 24496 times

Adding --arc-availability=officially-supported command line switch will enable google app store, but I am still not certain if this really works ...

RG

[odroid]

I really hope the Google play store will work.

[AreaScout]

I really hope the Google play store will work.

Me too ! I get an error because a file path is missing, they are maintaining there own kernel, maybe this kernel is needed for it to work, also a device ID seems to be missing, so ChromeOS can be only used as a Demo by now, but having a full functioning browser on a non X11/Wayland system is very very good, i am testing if i can get this to run on the N1 and also on the XU4, if I have it running on XU4 (I hope) then I will add it to my CloudShell2 NAS Image, imagine how cool that is if you can connect for NAS to TV and have a browser to config your OpenMediaVault NAS, you do not need a PC anymore to do this, or just receive emails a.s.o, for the HC4 this would also be nice !

RG

[mctom]

Hah, to think that I bought my mom a Chromebook as a christmas gift, just replaced the ChromeOS with Gallium. Now ChromeOS is actually on demand?

[AreaScout]

Hah, to think that I bought my mom a Chromebook as a christmas gift, just replaced the ChromeOS with Gallium. Now ChromeOS is actually on demand?

I was just searching for a solution to get chromium running on GBM, didn't expect to get an OS with it although this is not mandatory for me it is something cool to see, even on the Go Super it could be fun to have browser on all that gaming images

RG

[AreaScout]

And AreaScout says 'there should be a mouse cursor' and there was a mouse cursor !

Sorry for the shaky mouse movement



RG

[mad_ady]

And on N1, nonetheless!

[AreaScout]

And on N1, nonetheless!

Yes the N1 has the luxury to have a cursor plane, but I have to say that the rest of the DRM implementation on the N2 is pretty good when it comes to Chrome compatibility, on the N1 it was harder to get it running

I will try if I can use the overlay plane on the N2 as cursor plane on the weekend, this thing turns out very interesting for me and I hope for others too

[nzone]

Hi AreaScout, I downloaded your chrome app last week, chrooted into an armbian folder. Had a bit of trouble with render group on the old armbian so temporally added --no-sandbox to the chrome command line you provided and to my amazement it worked :O

I even run my script to extract only the needed file to make a small chrome only chroot image, and it still worked. I'm amazed, a working desktop OS for 440 megabytes (can be smaller with some effort).

Getting the cursor working would be cool because it seems to work really well.

One annoying thing that will probably stop me from using this more often is it does not obey virtual terminal switching, so when I change to another virtual terminal, CTL ALT F1 for example, it keeps hogging the video keyboard and mouse. Sadly there is probably no easy solution for this.

Anyway, thanks for your efforts.

[AreaScout]

@nzone

20210915_200327_resize_21.png (1002.05 KiB) Viewed 24325 times

A lot has been improved since the last build, the login manager was crashing, but I managed it to fix it, also if you add a google API key + secrets bound to your google account, it unlocks a whole new features to use, cloud storage, voice recognition a.s.o some of this features are not free you have to pay for it but others are not
I have to rebuild it without my API keys, I can't give them into public you can then add your own to unlock OS features

@odroid, BayLibre or any, I think drawing the cursor in the overlay plane will only work as a hack but not fix, because it is used as video plane, it would be better to add real cursor plane, I have found a code example where cursor is used, if you look to this code sniped, the fourth parameter of this function is used for the cursor plane

https://gist.github.com/carlocaione/81c ... drv-c-L667

While on the mainline kernel the fourth parameter of the function is NULL

https://github.com/torvalds/linux/blob/ ... rtc.c#L687

Could we initialize a cursor plane with meson_plane_create here to add one ->

https://github.com/torvalds/linux/blob/ ... ane.c#L531

Why was that not added/removed did it made problems ? It would good to have one !

RG

[mad_ady]

@AreaScout - is it possible to combine several desktop environments into one OS and choose (e.g. via NCURSES, at boot) if you want to boot into ChromeOS or a regular lightdm? Is it anything preventing coexistence? Might make sense if you have power users that need a standard desktop and light users that can live with ChromeOS.

[AreaScout]

Hello mad_ady, yes of course that is possible, here is what it can do

• Run under X11 (build for platform x11)
• Run under Wayland (build for platform wayland)
• Run from console DRM and GBM buffers (build for platform DRM)
• Run with own Login Manager (add --login-manager to command line)
• Run without Login Manager
• Run in Kiosk mode, Browser opens on start directly if you want with pre-defined website (e.g cool for a NAS with OMV admin page on start)

RG

[mad_ady]

How very convenient! Thanks!

[chewitt]

@AreaScout ping Neil Armstrong in #linux-amlogic on Libera. I have fuzzy recall there's a reason there is no cursor plane in the upstream codebase.

[AreaScout]

Will do tomorrow, thx for info !

[nzone]

# AreaScout

There is a bit of imfo at

https://github.com/superna9999/linux/wi ... res-Status

Cursor Plane support
More complex, would need refactor so support updating the cursor plane manually when in interlaced mode

Neil Armstrong has done so much for all of us odroid users over the years, it feels wrong to ask for more....... but wouldn't it be cool to have it working.


#mad_ady

All the options mentioned by AreaScout are good but sadly ....

# rant start

Most linux users are forced into booting the kernel, then handing over the amazing potential offered by the kernel to an /sbin/init which instantly locks the user into a never ending exponential bloatware explosion.

This is why I'm exited by AreaScout's work, the chance we can have a fully operational desktop with the ability to run "killall chrome" and be instantly back to bare bone console to run what ever the user wants..... e.g barebone X

Xorg $DISPLAY -listen tcp vt1; init_windowmanager
Starts faster than most gnome apps but apparently XServers are evil

# rant end

Thank you for your time

[chewitt]

Neil is awesome but ultimately paid to do a day-job which currently doesn't involve much work on Amlogc SoCs, so it's hard for him (and others on the Baylibre payroll) to absorb some of the tasks on the to-do list, and there are many higher priority tasks on the list than cursor plane support when they do have a moment to do pro-bono things.

[AreaScout]

@chewitt

I will just ask him if it's possible to get something to test, every thing else must be clarified with odroid after Chuseok holiday

[AreaScout]

@all

viewtopic.php?p=334946#p334946

I have uploaded a new tar archive with updated ChromeOS build wget https://www.areascout.at/chromium-OzoneChromeOS.tar.gz, this one also runs on RockChip SoC's. login manager works, I hope I have removed my API keys because I only rebuild the binaries not the libraries (they can be build into the binary or add as environment variables), it could be that they removed the need of API's keys with this commit https://chromium-review.googlesource.co ... /+/3162199 or I have not really an idea what this commit is for, if it complains on login manger that API keys are missing, please follow this howto https://www.chromium.org/developers/how-tos/api-keys you then have to use the google cloud console to generate it https://cloud.google.com/console it's kind of Microsoft Azure, when it says "In the 'Application typ' section check the 'Other' option" there is no Other option, I have chosen 'Desktop app' and it worked, then add the keys as environment variables, adding it to /etc/chrome_dev.conf did not work for me

Code: Select all

GOOGLE_API_KEY=your_api_key
GOOGLE_DEFAULT_CLIENT_ID=your_client_id
GOOGLE_DEFAULT_CLIENT_SECRET=your_client_secret

After login and a small setup process you are logged in with your real google account, your settings are stored permanently, more apps can be installed a.s.o, after re-login you must have a different login screen

Please be careful to not add unintentionally payed services from your google cloud console, it tells you if it's free or not

Regarding the cursor plane if have spoken with Neil and it should be possible to add, the problems with interlaced resolutions should "not a big deal" he said and for the problems "it would only affect CVBS 480i/576i output" which is not very often used

P.S.: If you type with your keyboard, the output is also written to tty/console which is odd, to prevent this stop tty service sudo systemctl stop getty@tty1.service

RG

[nzone]

Regarding the cursor plane if have spoken with Neil and it should be possible to add, the problems with interlaced resolutions should "not a big deal" he said and for the problems "it would only affect CVBS 480i/576i output" which is not very often used

Excellent!!!

from https://oftc.irclog.whitequark.org/panfrost/2021-09-17

08:36 <icecream95> Evidently the bug where regions of windows randomly disappear on rk3288 is still present (kernel 5.14.0)
<robmur01> the long-outstanding kernel fixes have only just landed in mainline this week

11:52 <alyssa> tomeu: PAN_MESA_DEBUG=nocrc for u
11:53 <alyssa> PAN_MESA_DEBUG=nocrc,noafbc for good measure

Hoping the panfrost instability that I alway get at https://glslsandbox.com/ for instants which include system lockups will be helped...... Kernel
5.15 rc2 out tomorrow.

P.S.: If you type with your keyboard, the output is also written to tty/console which is odd, to prevent this stop tty service sudo systemctl stop getty@tty1.service

I noticed the same thing when changing virtual teminals. I cound see what I had typed until I typed a character or moved the mouse.

I thought a bunch of the service start/stop on virtual terminal changes could be a better short term workaround than "killall chrome" but can't really do much until I get a cursor as I have not got a touch screen option.

Anyway thanks again.

[AreaScout]

Hi all

I did another video of ChromeOS with screen capture this time to show how much CPU power is consumed while capturing and playing YouTube video



Have fun

[narmstrong]

Hi guys,

I found a few hours to implement HW cursor support with OSD2, here is the patch on 5.15-rc2 (should apply on older kernels):
https://github.com/superna9999/linux/co ... 326399d231

Notice I only tested on A311D (same as S922X), under weston, sway & modetest.
Should work fine on S905X4.
I didn't test on earlier GXBB, GXL & GXM SoCs.

Under Xorg the cursor doesn't move, I'll need to check why.

Please tell me if you hit a bug and how to reproduce it.

Neil

[mad_ady]

Bless you!

[AreaScout]

@narmstrong

Wow that was fast, without having time to test it, does the cursor plane only support atomic DRM or also legacy DRM ? Thanks for the patch, I can test it at earliest tomorrow

edit: A good test would be testgles2 from SDL2 test suite build for KMSDRM or of course ChromeOS

RG

[narmstrong]

@narmstrong

Wow that was fast, without having time to test it, does the cursor plane only support atomic DRM or also legacy DRM ? Thanks for the patch, I can test it at earliest tomorrow

edit: A good test would be testgles2 from SDL2 test suite build for KMSDRM or of course ChromeOS

RG

It should support the legacy and the atomic way

Good idea, let me test SDL2 then maybe ChromeOS

[narmstrong]

Tested with SDL2 wich uses the legacy cursoe API and it works just fine.

[narmstrong]

And with your prebuilt ChromeOS, and it works just fine.

[AreaScout]

@narmstrong

Good job ! thx again for this, I for myself can only test it on late afternoon, but having this now is so cool !

[AreaScout]

@narmstrong

Building it on remote server while on breakfast , it fails on 5.10 kernel with missing include file <drm/drm_gem_atomic_helper.h>, building on 5.14 now and it looks ok so far

RG

[AreaScout]

@all

Just to confirm, tested it with Atomic and Legacy DRM and both works fine, it's so much better with cursor ! I will build an Image with preinstalled ChromiumOS/ChromeOS (as this two seems to get closer and closer)

Prove video



Screen size is only 1024x600 (VU7A) but it looks great and even better on 1080p !

Merci

RG

[istanbulls]

I watch the works with curiosity and interest. There have been rapid developments in a short time.
For novices like me, I can't wait to test it when a burnable image is created

[nzone]

@narmstrong

I replaced the folder drivers/gpu/drm/meson in kernel 5.15rc3 that I built the day before your post with the same folder from superna9999/linux and in 30 seconds my n2+ rebuilt 5.15rc3 and low and behold it all just miraculously worked.

To my amazement, not only does AreaScout's chromeOS have a cursor but my entire system seems to be flying. I've been testing all sorts of things from webkit browsers like surf (really rips now, out premforms chome ????), tinyX (software cursor ??? all cursor blinks gone????). I been testing html5 canvas web pages with tinyX (Xserver 1.3 with fbdev, circa 2010) and it's fast with no cursor blinking :O
I'm just astounded by it all. Maybe it's just the new kernel but either way thank you very much.

Anyway I should calm down and report that the cursor works for about 10 seconds on my n2+ hardware accelerated Xserver before it stops. I can see it trying to move and the cursor images do change on mouse over events. It will be interesting to see how well it all works when working properly given all the other performance gains I've mentioned above.

Update:

After some playing around, it appears the cursor works fine on accelerated Xserver when an app that uses the hardware such as glmark2-es2 or the chromium web browser is running . As soon as you close these apps and go to a non accelerated app, the cursor stop moving. Restarting a hardware using app starts the cursor moving again. I hope this can help someone determine whats going on.

So I've been leaving chromium running.
As for Xserver running with the hardware accelerated cursor everything is buttery smooth and fluent.

P.S maybe this xserver stuff should be moved to it's own thread? Maybe "N2 hardware debugging" ????


@AreaScout

I apologize for once again polluting your thread. Congrats on you amazing achievement.

[AreaScout]

@AreaScout

I apologize for once again polluting your thread. Congrats on you amazing achievement.

That's not a problem for this post as it's somehow related

Update:

After some playing around, it appears the cursor works fine on accelerated Xserver when an app that uses the hardware such as glmark2-es2 or the chromium web browser is running . As soon as you close these apps and go to a non accelerated app, the cursor stop moving. Restarting a hardware using app starts the cursor moving again. I hope this can help someone determine whats going on.

So I've been leaving chromium running.
As for Xserver running with the hardware accelerated cursor everything is buttery smooth and fluent.

P.S maybe this xserver stuff should be moved to it's own thread? Maybe "N2 hardware debugging" ????

I don't know a lot of X11 and Meson, but that sounds like xserver is using fbdev and chromium and other apps make use of the hardware cursor ? Are you using a xf86-video-fbdev driver ?, if yes they are not using the DRM path, I know that armsoc driver (xf86-video-armsoc) makes use of DRM which holds a cursor plane, with this the cursor should work, we have that running on XU4 and other Exynos boards.

And yeah for a in-depth discussion it would be better to open a separate thread

[AreaScout]

I watch the works with curiosity and interest. There have been rapid developments in a short time.
For novices like me, I can't wait to test it when a burnable image is created

Aha ! Those guys calling themselves a novices are normally the professionals And of course an image will be there

RG

[mad_ady]

@AreaScout if you're putting the effort in making new images, will you consider also one for XU4? Perhaps you can use something like ansible to orchestrate building images - adding your packages on top of existing images.

[nzone]

I don't know a lot of X11 and Meson, but that sounds like xserver is using fbdev and chromium and other apps make use of the hardware cursor ? Are you using a xf86-video-fbdev driver ?, if yes they are not using the DRM path, I know that armsoc driver (xf86-video-armsoc) makes use of DRM which holds a cursor plane, with this the cursor should work, we have that running on XU4 and other Exynos boards.

And yeah for a in-depth discussion it would be better to open a separate thread

I'm using the modesetting driver.
I can confirm, I am seeing a big improvement in performance with this cursor change not only in the apps that use the hardware acceleration but other apps as well.
It could be a very simple fix. You are probably aware of the following link but it's a good read for anyone wanting to try to understand the complexity involved.

https://en.wikipedia.org/wiki/Direct_Rendering_Manager

I think it is very related to what you have achieved here because the windows in your chromeos feel very "floaty/laggy" when moving them, however the complexity of this problem means it needs it's own thread.

[AreaScout]

You are probably aware of the following link but it's a good read for anyone wanting to try to understand the complexity involved.

https://en.wikipedia.org/wiki/Direct_Rendering_Manager

No that's fine, the forum should transport as much knowledge as possible, however the knowledge scale from peoples here is enormous large, from beginner to extra ordinary expert, some people will get attracted and some get bored "Man ! I know this for years !" by the information you post, that's how it is ..

I think it is very related to what you have achieved here because the windows in your chromeos feel very "floaty/laggy" when moving them, however the complexity of this problem means it needs it's own thread.

Didn't feel that window movement is laggy but I was testing most of the time with 1024x600 Display, maybe that's why, the command line switch I added --in-process-gpu does already own threading on GPU tasks, there are a loooot of command line switches to tune and speed up ChomeOS or lets better say ChromiumOS, try 1280x720 and be sure you use it on a 4GB model

@mad_ady

Let's see if I can find time for XU4 also, I think it will be harder to get it working without mesa drivers ...

RG

[nzone]

Didn't feel that window movement is laggy but I was testing most of the time with 1024x600 Display, maybe that's why, the command line switch I added --in-process-gpu does already own threading on GPU tasks, there are a loooot of command line switches to tune and speed up ChomeOS or lets better say ChromiumOS, try 1280x720 and be sure you use it on a 4GB model

I need to do a full chromeOS install at some stage. I have made a small X chroot with only hardware accelerated apps and a tiny windows manager and everything is running quick so disregard that comment for now.


--in-process-gpu

Thanks, that fixed --enable-native-gpu-memory-buffers command switch which always left chrome tab and location bar blank.

Let's see if I can find time for XU4 also, I think it will be harder to get it working without mesa drivers ...

RG

I haven't tested panfrost on armhf for a while but can confirm it's straight forward to build and has worked well in the past. Kernel can still be 64bit aarch64.
I see you have netflix icon in your video. I thought widevine was still needed for that which is only available armhf???

[AreaScout]

I haven't tested panfrost on armhf for a while but can confirm it's straight forward to build and has worked well in the past. Kernel can still be 64bit aarch64.
I see you have netflix icon in your video. I thought widevine was still needed for that which is only available armhf???

The Mali T-628 GPU used in XU4 is not supported by Panfrost, that's the problem and the way a Surface is acquired in Chrome does not work with the vendor provided libraries if I am not mistaken, so it needs some hacks to work

Edit: That's just a Netflix resolution helper App not for video streaming itself

RG