fredrum

Has anyone spotted a change in very recent kernel driver of pancfs turning into 'panthor'? It's mentioned here https://lwn.net/Articles/940957/ and is about Mali Mesa continous work. I guess I'll try to take a look at kernel repo's myself but I'm not used to doing that so thought I'd drop the info here as well. Cheers Update: Nope if I'm looking in the right place I can't see it in linux kernel 6.6-rc4 even. I think it should be living here 'drivers/gpu/drm/panthor' as per https://lore.kernel.org/linux-arm-kernel/20230809171017.64409d60@collabora.com/T/ for example. I guess its still way too early.

Hello! Has anyone tried to play straight unconverted hevc video from a GoPro 11? Iv'e tried ffplay and gst-play but I'm just getting a static image and a bunch of these errors, [swscaler @ 0x7f49d399c0] [swscaler @ 0x7f49d47a50] deprecated pixel format used, make sure you did set range correctly I have successfully played example jellyfish videos of hevc and hevc 10bit formats, ffmpeg reports the following details from the GoPro footage, Stream #0:0[0x1](eng): Video: hevc (Main) (hvc1 / 0x31637668), yuvj420p(pc, bt709), 3840x2160 [SAR 1:1 DAR 16:9], 119725 kb/s, 119.88 fps, 119.88 tbr, 120k tbn (default) Any suggestions? Cheers!

This Mekatronics box is using the rk3588 with the Mali-G610. (same gpu as Opi5) Presumably using the vendor blob drivers. glmark score: 3696 [ YouTube video showing test + result] I have been testing the Collabora new Mesa source and using Balbes test kernel builds with pancsf but I'm not getting anything near these numbers. IIRC I get around 1200-1300. I don't know what the Collabora plan is here as there doesn't seem to be much action as far as I can tell as an outside observer. Not sure how ready they think this Panfrost version ('v10') is? EDIT: Btw regarding video decoding the HW seems extremely capable. Here's a thread talking about 16 1080p streams simultaneously, https://forum.radxa.com/t/decoding-and-rendering-16-1080p-streams-with-gstreamer/16307

I am currently using your 20230331-current image. Should I upgrade to one of your later ones? I had the smae issue before with getting the GPU recognized and not sure why. Could it be because I install to /usr/lib ? I do edit some ld config files and do 'sudo ldconfig' and it does make the system pick up my own built mesa but always with llvm. I did get GPU with a mesa build you linked to and I just copied straight to the default repo system folder. Could it just be the folder where I install it?

Hello! Does anyone, and by anyone I probably really mean @balbes150 :D, know what I need to do when building official Mesa for it to be recognized in the experimental 6.x kernel environment? There's a panfrost focused dev/staging branch that the panfrost devs are using that I would love to try and be able to build for OPi5 but when I do it I always just get llvm. balbes, you are managing to get the GPU recognized would you be able to share how that's done in the building? I'm not asking for you to provide a build I would just love to know how you do it so I can as well and test the new versions as they happen. This is the branch and commits btw that I read on 'the internet' that its the bleeding edge for rk3588/Mali g610 mesa. https://gitlab.freedesktop.org/panfrost/mesa/-/commits/panfrost/v10-wip Iv'e successfully built that mesa branch and took the latest commits but the gpu is still not recognized so there must be a trick I'm missing. Does anyone have any suggestions? Cheers!

Ohhhh with that one I now do get the GPU detected with glxinfo -B ! Nice one! [ I kept updating this maybe skip to the end for the success part? ] There are some very cool and strong graphical artifacts in some areas now and then and there's a error message from the glxinfo. Is there a device file I need to flip some permission on or something maybe? Using gnome+wayland. fred@orangepi5:~$ glxinfo -B name of display: :0 DRM_IOCTL_MODE_CREATE_DUMB failed: Permission denied Failed to create scanout resource display: :0 screen: 0 direct rendering: Yes Extended renderer info (GLX_MESA_query_renderer): Vendor: Mesa (0xffffffff) Device: Mali-G610 (Panfrost) (0xffffffff) Version: 23.0.0 Accelerated: yes Video memory: 7695MB Unified memory: yes Preferred profile: compat (0x2) Max core profile version: 0.0 Max compat profile version: 3.0 Max GLES1 profile version: 1.1 Max GLES[23] profile version: 3.1 OpenGL vendor string: Mesa OpenGL renderer string: Mali-G610 (Panfrost) OpenGL version string: 3.0 Mesa 23.0.0-devel (git-cbe80448cc) OpenGL shading language version string: 1.30 OpenGL context flags: (none) DRM_IOCTL_MODE_CREATE_DUMB failed: Permission denied Failed to create scanout resource OpenGL ES profile version string: OpenGL ES 3.1 Mesa 23.0.0-devel (git-cbe80448cc) OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.10 Trying this but it doesn't help, sudo chmod a+rw /dev/dri/* I'll google around a bit! Don't worry if this is eating up your time and you'd rather spend that on other things. If that's the case I'll just leave this until its more mature. If not I'm happy to keep trying stuff. UPDATE: trying glmark2-es2-wayland but it doesn't find the GPU it is running with llvm. Both these ways of running: > glmark2-es2-wayland and > LD_LIBRARY_PATH="/usr/lib/aarch64-linux-gnu" glmark2-es2-wayland libEGL warning: egl: failed to create dri2 screen libEGL warning: failed to get driver name for fd 0 libEGL warning: MESA-LOADER: failed to retrieve device information libEGL warning: failed to get driver name for fd 0 glxgears also not happy: DRM_IOCTL_MODE_CREATE_DUMB failed: Permission denied Failed to create scanout resource Segmentation fault Checking dmesg for drm: [just the tail of output] [ 2.449518] rockchip-vop2 fdd90000.vop: [drm:vop2_crtc_atomic_enable [rockchipdrm]] dclk_out0 div: 0 dclk_core0 div: 2 [ 2.449565] rockchip-vop2 fdd90000.vop: [drm:vop2_crtc_atomic_enable [rockchipdrm]] set dclk_vop0 to 148500000, get 148500000 [ 2.861015] rockchip-drm display-subsystem: [drm] fb0: rockchipdrmfb frame buffer device [ 2.878911] [drm] Initialized rockchip 3.0.0 20140818 for display-subsystem on minor 1 [ 3.117857] [drm:rockchip_drm_fault_handler [rockchipdrm]] *ERROR* iommu fault handler flags: 0x0 [ 3.243729] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.244607] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.245466] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.246323] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.247183] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.248039] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.248895] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 3.249757] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 [ 4.753891] systemd[1]: Starting modprobe@drm.service - Load Kernel Module drm... [ 4.802028] systemd[1]: modprobe@drm.service: Deactivated successfully. [ 4.803050] systemd[1]: Finished modprobe@drm.service - Load Kernel Module drm. [ 24.800483] [drm:rockchip_drm_fault_handler [rockchipdrm]] *ERROR* iommu fault handler flags: 0x0 [ 24.801159] rockchip-vop2 fdd90000.vop: [drm:vop2_isr [rockchipdrm]] *ERROR* POST_BUF_EMPTY irq err at vp0 UPDATE 2: Waaiiit a minute. It seems my custom build Mesa was interfering even after I had reset the ld conf files + sudo ldconfig. Now I was able to run, > LD_LIBRARY_PATH="/usr/lib/aarch64-linux-gnu" glmark2-es2-wayland and without changing the default governor I got: glmark2 Score: 2189 Success!! There were still some strong glitching in the following, [jellyfish] [terrain] [refract] but maybe something in my own Mesa install still interefered. I still also get the DRM_IOCTL_MODE_CREATE_DUMB failed: Permission denied from glxinfo. UPDATE 3: After sticking the governor to performance glmark2-es2-wayland returns: glmark2 Score: 3455 A new record for me on this board! The other issues mentioned above are still there but maybe I'll have to re-run on a clean flashed SD card and see what goes away. That'll be for later because now I'm itching to try Dolphin again on this setup!

Ok I think I'm still a bit fuzzy about what that means. Iv'e tried so many different patchy files from different threads now When I do this I 'brick' my system its not showing my anything and I have to start from scratch again and reflash, dpkg -i --force-overwrite mesa_23.0.0-dev-2_arm64.deb dpkg -i --force-overwrite mesa_23.0.0-devel-3_arm64.deb from: https://mega.nz/folder/j9QSDQSQ#6WpasOlbZYIInfw6yo4phQ/folder/LkgF1SoY I also tried again to build mesa myself but that also did still not find the GPU.

Ok I got a little bit further but no cigar. I managed to get the ubuntu gnome version 20230331-current going. It seems that image didn't like my regular HD Samsung SyncMaster SA650 monitor (usb-c to displayport). Its a ~10 year old monitor that's been very reliable and works fine otherwise. It also worked on your 20230329-current image. Anyway I got visuals with a different Asus HD monitor I had lying around (HDMI). Great! I then tried with both the rk3588s-orangepi-5.dtb files you guys suggested to try. I copied them to Desktop then did Ctrl+Alt+F1 to switch to tty and ran the copy command to /boot/dtb/rockchip from there. I rebooted and tried with glxinfo -B but I still get only llvm. @balbes150 when you say 'installed the archive' do you just refer to that .dtb copy or something else? UPDATE: Ah I just did the old, dmesg | grep pancsf and that's looking better and outputs, [ 2.193441] pancsf fb000000.gpu: clock rate = 198000000 [ 2.193943] pancsf fb000000.gpu: bus_clock rate = 200000000 [ 2.194454] pancsf fb000000.gpu: [drm:pancsf_devfreq_init [pancsf]] More than 1 supply is not supported yet [ 2.195400] pancsf fb000000.gpu: Looking up mali-supply from device tree [ 2.195575] pancsf fb000000.gpu: Looking up sram-supply from device tree [ 2.196482] pancsf fb000000.gpu: EM: OPP:400000 is inefficient [ 2.196489] pancsf fb000000.gpu: EM: OPP:300000 is inefficient [ 2.196625] pancsf fb000000.gpu: EM: created perf domain [ 2.197427] pancsf fb000000.gpu: mali-g610 id 0xa867 major 0x0 minor 0x0 status 0x5 [ 2.198112] pancsf fb000000.gpu: Features: L2:0x07120306 Tiler:0x00000809 Mem:0x301 MMU:0x00002830 AS:0xff [ 2.198956] pancsf fb000000.gpu: shader_present=0x50005 l2_present=0x1 tiler_present=0x1 [ 2.200827] pancsf fb000000.gpu: Firmware protected mode entry not be supported, ignoring [ 2.201576] pancsf fb000000.gpu: Firmware protected mode entry not be supported, ignoring [ 2.202288] pancsf fb000000.gpu: Firmware protected mode entry not be supported, ignoring [ 2.203232] pancsf fb000000.gpu: CSF FW v1.1.0, Features 0 Instrumentation features 71 [ 2.204081] [drm] Initialized pancsf 1.0.0 20230120 for fb000000.gpu on minor 0 So to me that looks good at least now on the pancsf front? Something else now instead then...hmm Just checked /boot/extlinux/extlinux.conf and it looks ok I guess, label Armbian kernel /boot/Image initrd /boot/uInitrd fdt /boot/dtb/rockchip/rk3588s-orangepi-5.dtb append root=UUID=9667a537-98e8-470a-8819-5aa4dd9b0b9b console=ttyS02,1500000 console=tty0 rw no_console_suspend consoleblank=0 fsck.fix=yes fsck.repair=yes net.ifnames=0 splash plymouth.ignore-serial-consoles

I took the .dtb that balbes150 links a couple of posts up. So you took that file from a non-desktop image? UPDATE: Thank you I'll try with that one. UPDATE2: Ok so I instead copied in the .dtb that you just attached above here and rebooted. Something is different (I think filesize is different but haven't checked) because now it won't boot up. Incidentally the same thing happens as when I use the latest 20230331-current images, Power goes on with the red LED then after 10 seconds or so the green starts to repeatedly blink but nothing shows up on the monitor its completely black and I don't think it gets signal. I have tried both using the usb-c and the hdmi port.

I'm not having much luck not sure what I'm doing wrong. I'm working with 20230329-current lunar gnome desktop, fresh flash with nothing added or changed. Linux orangepi5 6.2.0-rc1-station-m3 #trunk SMP PREEMPT_DYNAMIC Sat Mar 25 17:10:20 MSK 2023 aarch64 aarch64 aarch64 GNU/Linux (the 20230331-current desktop images will not boot for me) I tried copy the rk3588s-orangepi-5.dtb to /boot/dtb/rockchip/ and reboot. Running the gnome wayland desktop. But no Mali GPU in glxinfo just llvm still. If I run, dmesg | grep pancsf i get, [ 2.206484] pancsf fb000000.gpu: clock rate = 198000000 [ 2.206963] pancsf fb000000.gpu: bus_clock rate = 200000000 [ 2.207490] pancsf fb000000.gpu: [drm:pancsf_devfreq_init [pancsf]] More than 1 supply is not supported yet [ 2.208395] pancsf fb000000.gpu: Looking up mali-supply from device tree [ 2.209459] pancsf fb000000.gpu: Looking up sram-supply from device tree [ 2.210974] pancsf fb000000.gpu: EM: OPP:400000 is inefficient [ 2.211270] pancsf fb000000.gpu: EM: OPP:300000 is inefficient [ 2.211372] pancsf fb000000.gpu: EM: created perf domain [ 2.212044] pancsf fb000000.gpu: mali-g610 id 0xa867 major 0x0 minor 0x0 status 0x5 [ 2.213481] pancsf fb000000.gpu: Features: L2:0x07120306 Tiler:0x00000809 Mem:0x301 MMU:0x00002830 AS:0xff [ 2.214961] pancsf fb000000.gpu: shader_present=0x50005 l2_present=0x1 tiler_present=0x1 [ 2.216816] pancsf fb000000.gpu: Direct firmware load for mali_csffw.bin failed with error -2 [ 2.218010] pancsf fb000000.gpu: Failed to load firmware image 'mali_csffw.bin' [ 2.219478] pancsf fb000000.gpu: Fatal error during GPU init [ 2.220192] pancsf: probe of fb000000.gpu failed with error -2 I actually also tried to install the mali_csffw.bin file following an older jeffyCN instruction from a different thread on the forum but it didn't help.

Sweet thank you so much for posting that .dtb I will try it out as soon as I get time which will probably be Sunday!

Hi I'm excited to try mesa Panfrost (not panfork) with this kernel and have managed to get a build. But I keep getting llvm instead of the GPU and was wondering if anyone know what I need to do? I'm using this config command: meson --prefix /usr --libdir lib -Dvulkan-drivers= -Dlibunwind=false -Dgallium-drivers=panfrost,kmsro,swrast -Dplatforms=x11,wayland -Dbuildtype=release -Degl=true -Dgles1=true -Dgles2=true build I'm also adding the /usr/lib and /usr/lib/dri paths to ld conf files and it seems they are found. But glxinfo still gives me just llvm. Cheers!

Awesome is this Panfrost-Panfrost, ie not Panfork? Id'e love to try it but do not have experience trying out different kernels. Ive seen the/a 'midstream' kernel and that it includes the wip/new pancsf. Could you make some 1-2-3 steps instruction for making an install that has this? I would really love to try it.

Ive made a new video with a few more games showing! [ UPDATED VIDEO SHOWING GAMES WITH ARMBIAN SYSTEM TWEAKS ] Its with the system tweaks for some extra performance as described here,

Awesome I installed those kernel patches and they do actually work!! I'll make another Dolphin video some time later today but it makes a significant difference to fps. glmark2-es2 now scores 1244 whereas before I think it was 8-900 something like that. (its insane that the vendor blob drivers had ~3700) Here's what you can do if you want to also try those kernel patches(?): Download the three .deb files from https://github.com/ChisBread/linux-orangepi/releases/tag/v1.06.002 cd to where they are. You need to install old python (python2) I'm sure there's a correct way to do the following but I did, sudo apt install python2 sudo ln -s /usr/bin/python2 /usr/bin/python sudo dpkg -i ./linux-dtb-legacy-rockchip-rk3588_1.0.6_arm64.deb sudo dpkg -i ./linux-headers-legacy-rockchip-rk3588_1.0.6_arm64.deb sudo dpkg -i ./linux-image-legacy-rockchip-rk3588_1.0.6_arm64.deb I then also did the following as it was mentioned. I haven't tested before-after so not sure if its actually neccessary, PAN_MESA_DEBUG=gofaster export PAN_MESA_DEBUG=gofaster glmark2 Score: 1244 UPDATE: I just redid the benchmark test but this time with that governor set to 'performance' and got even higher score! Also I used the wayland version of the benchmark app not sure if that makes much difference as I think the rendering is done off screen anyway. su - echo performance > /sys/devices/platform/fb000000.gpu/devfreq/fb000000.gpu/governor echo performance > /sys/devices/platform/dmc/devfreq/dmc/governor glmark2-es2-wayland glmark2 Score: 1730 I can't notice any big difference in Dolphin fps so maybe the governor thing mostly helps with the benchmark? And warning that the governor settings will set your opi to run these constantly at full throttle so it'll get hotter and draw more power. You can easily reverse this when you don't want them any longer. It would be amazing if someone was able to try out the official Panfrost driver with that new 'pancsf' thinghy. As they mention, "Glmark2 improves from ~500 to ~1700 with performance governor." (though I have no idea what hardware that is, I think it might be some rk3588 though)

Cool thanks I think Iv'e missed that message from ChrisBread so will definitely see if that kernel patch works with the current Armbian images. Or if I can find a setup that works. I'm not so so sharp with kernel hacking patching but I can probably figure it out. UPDATE: Just found this on the panfork issues page relating to the same kernel patch, https://gitlab.com/panfork/mesa/-/issues/12 I have tried the 'governor' trick as iv'e come across it in a few threads. There's a few discussions on the *cough* radxa *cough* Rock 5 forums. For just the governor setting I saw no noticable fps increase. Maybe because once an app start doing gpu work it should request 100% effort by the gpu? Or it might be because the drivers are not efficient enough the way they work so there's already a different type of performance ceiling? UPDATE2: There's also some currently ongoing work by collabora with some 'pancsf' driver that *might* help the sitation I'm not 100% clear but I belive it might improve gl task scheduling amongs things. Something which if I understand correctly is one of the things that are not great with the current panfork driver, even according to 'icecream95' himself. If someone is good with kernel hacking you might try this which seems to have an early version of that new collabora driver, NOTE though that this work is done for the main 'Panfrost' driver. (Not the rockchip panfork one) https://github.com/Googulator/linux-rk3588-midstream https://github.com/Googulator/linux-rk3588-midstream/commit/f563c09394db72ff99e18fd038c78abd46228a77

I have been looking into this as well over the last two weeks or so to try to get a claerer picture. But don't have that yet. I'm no kernel guy just a hobbyist programmer but I have used drm prime in a software I've modified for Raspberry, Chaiki for Raspberry I'm currently trying to port that over to OPi5. There's various places that can give you some hints but I'm going to assume that you guys are all aware of those already so not going to post. But did you see this one? I just spotted it but haven't had time to try yet https://github.com/rockchip-linux/mpp/pull/122/commits I didn't have luck running the current default 'mpi_dec_test' but there might be some drm prime hints in that PR?

Someone asked in a different thread how I got Dolphin to run so thought I'd start a thread so not to pollute the other one. Basically it wasn't that hard especially if youve done some git'hubbing before. I think I followed the 'Ubuntu' instructions here. Make sure you get the dependencies and do all the steps correctly. It should build without problems if you do. (remember you have a lot of cores on Opi5 so can use 'make -j6' or even higher) https://github.com/dolphin-emu/dolphin/wiki/Building-for-Linux You then want to set a config flag to make Dolphin use GLES instead of big brother OpenGL. (you probably want to run it at least once to generate the config files) nano ~/.config/dolphin-emu/GFX.ini add a line: PreferGLES = True But when running one of the more common games I got a shader compilation error. I hacked around that and also reported it on the Dolpin bug report site and it should be fixed soon when the 'PR' gets accepted. In the meantime you can read here what to do, https://bugs.dolphin-emu.org/issues/13198# Note that in my youtube video I get a good fps but later on in the game when you get to the first village the transparencies in the central square bonfire completely kills fps. The same happens in Metroid Prime when you have the large bluish holo walls in view. Mario Kart runs pretty well most of the time. Mario Sunshine as well at least in the plaza area I have't tried further. Cheers and Good Luck!

Hey if you start a new thread about Dolphin I'll reply to you there! It would be good a have a special Dolphin thread anyway! UPDATE: I made a thread with descriptions here for you,

Might be worth pointing out even if maybe many people are aware that there's two completely differernt uses of the term 'DRM'. DRM = Digital Rights Management. Ie support for enforcing copy right protection of media. also DRM = Direct Render Manager. A render and display path in Linux that can show images when not under X11. https://en.wikipedia.org/wiki/Direct_Rendering_Manager

Do you have any pointers to how to go about doing that? I searched around and found a few messages here and there about OPP tables but they were for CPU only. I couldn't find an example for GPU.

Hello!! I got Dolphin running under Armbian and it seems its mainly GPU bound for the games that Iv'e tested. I am aware that the Panfrost drivers are not optimal but I thought I'd like to try and squeeze a few extra fps out of this setup by overclocking the GPU if that's possible. Is it? And does anyone know how I can do that? Cheers!

Ok I see. Thank you for the estimate!

+1 When might we see a switch to a 6.x mainline kernel!?

thanks for the suggestion its running now but its very very slow and says its going to take more than an hour! (I usually get ~30megabytes (not megabit) download speeds when all's well) UPDATE: Nice! That worked the keyboard is now working fine! Thank You!