EDIT: got it working, dont know why I had to put CONFIG_MEDIA_TUNER_R912=m in linux.aarch64.conf also
ah yes, the R912 driver was added by Your patch, so it was not enabled in the official LibreElec build. Somthing to remember when it lands into the upstream libreelec.
I think that if You test it and find no issues there, and chewitt test it on his boxes (I an already running it for couple of days without any issues) than the patches are ready to be included in the official LibreElec.
ashi can You test this kernel branch: https://github.com/LibreELEC/amlo…mczerski-claude ?
this is a claude rewrite of the vendor dvb demux driver. I aplied Your changes to it.
To build Libreelec with that kernel You need to additionally change two things in the LibreElec build:
- remove "dvb" from PKG_PATCH_DIRS of packages/linux/package.mk (to remove patches of the vendor dvb driver)
- add CONFIG_DVB_MESON_DEMUX=m to projects/Amlogic/linux/linux.aarch64.conf (to select new dvb driver)
If I install shader preset addon and select some other video filters (like scale 2x, scale 2x plus or super eagle) the screen is no longer black.
In case there is still some users of a95x f3 air then current nightly (and probably all) LE version will not boot on it even though there is meson-sm1-a95xf3-air.dtb device tree for this box available. At least from me, I have two of them. But the fix is simple, apply this patch to LE repository:
diff --git a/projects/Amlogic/bootloader/scripts/s905_autoscript.src b/projects/Amlogic/bootloader/scripts/s905_autoscript.src
index 8b83de1748..5f4f45be14 100644
--- a/projects/Amlogic/bootloader/scripts/s905_autoscript.src
+++ b/projects/Amlogic/bootloader/scripts/s905_autoscript.src
@@ -7,5 +7,7 @@ setenv try_boot_start 'if fatload ${devtype} ${devnum} ${loadaddr} KERNEL; then
setenv devtype mmc
setenv devnum 0
run try_boot_start
-setenv devtype usb
-for devnum in 0 1 2 3 ; do run try_boot_start ; done
+bootm start
+bootm loados
+bootm prep
+bootm gomake image amd prepare sd card with it. Now it should boot.
Or use attached compiled s905_autoscript script (replace s905_autoscript on your sd card with it).
background:
on a95x f3, command bootm ${loadaddr} - ${dtb_mem_addr} fails with some signature check failure. But this happens also on CoreElec build that I used from the beginning. So I looked how CoreElec is booting and it uses this procedure from the patch which make the board boot at the end.
I worked a little bit with claude and to its suggestions made some tests: I run kmscube which is working, and installed visualization.shadertoy addon which is also working.
https://paste.libreelec.tv/superb-louse.log
the log was taken on snes9x emulator. I tested also uae with same result. Will try some other emulators later.
just tested, same issue. I forgot to mention that audio and controller input are working, so the core itself seems to be working - just the display fails.
I have a similar issue. Black screen after launching the game using retroplayer. Tested on snes9x and uae emulator cores. No obvious errors in kodi log. No errors in dmesg. The issue is present on s905d (GXL), but on s905x3 (SM1) it is working.
LE 13 (2026-04-24, 30131d75e592a2ca6d72c642311ea6b0b3509f50)
attaching kodi.log, the game is launched at 2026-04-27 22:07:39.729
BTW. on LE12.2 on the same hardware I had no issues with retroplayer.
I would appreciate any tips on how to debug this problem.
HI taki ,
Regarding topic 1. I also think that the demod/tuner drivers should be distributed as addons. In my linux branch (https://github.com/mczerski/linux/tree/amlogic-dvb) I try to make the frontend attachment modular. Current version allows to plug in individual demod / tuner drivers with or monolitic driver implemntations with DTS. But current approach requires each plugin to "behave" in certain not documented way. I also want to try with slightly different approach and to use component framework. But there will be always some code required on the demod/driver side to be added - this is why addon approach will be the best.
Regarding topic 2. Could you explain in more details what issues you have ? From my expirience there is no issues with i2c axccept one case, where two demods are on the same i2c bus - I had to modify m88rs6060 driver because it did not work well with i2c transactions interleaved with i2c transactions for the second demod.
Also I would like to test Your current driver version. Can You share it through some git repository ?
I think that TVHeadend is just requesting the whole TS stream that contains all of the channels to be able to provide all channels from one mux to multiple clients. And in such situation the demux driver does not have to do demuxing thus there is no dmx irqs. Does that make sense ?
taki it might be a bug that I found when porting vendor driver to mainline.
In aml_dmx.c You will find sunch code:
#define DEMUX_INT_MASK\
((0<<(AUDIO_SPLICING_POINT)) |\
(0<<(VIDEO_SPLICING_POINT)) |\
(1<<(OTHER_PES_READY)) |\
(1<<(SUB_PES_READY)) |\
(1<<(SECTION_BUFFER_READY)) |\
(0<<(OM_CMD_READ_PENDING)) |\
(1<<(TS_ERROR_PIN)) |\
(1<<(NEW_PDTS_READY)) |\
(0<<(DUPLICATED_PACKET)) |\
(0<<(DIS_CONTINUITY_PACKET)))Change 1<<(NEW_PDTS_READY) to 0<<(NEW_PDTS_READY) and check if it helps.
For some reason this NEW_PDTS_READY interrupt source triggers at a very high rate and the kernel disables it. Disabling this interrupt is only a workaround but I don't know how to fix that better. And by looking at the code this interrupt source is not relevant.
The other thing is that when using tvheadend I also have no dmx_irqs (only dvr_irqs).
But when using dvbv5-zap (dvbv5-zap -c dvb_channel.conf -r "Some channel") I get dmx_irqs:
48: 308 0 0 0 GICv2 55 Edge dmx irq
49: 0 0 0 0 GICv2 37 Edge dmx irq
50: 0 0 0 0 GICv2 53 Edge dmx irq
51: 638 0 0 0 GICv2 51 Edge dvr irq
52: 0 0 0 0 GICv2 57 Edge dvr irqI think we must decide how we approach this topic. The work I did was more like a proof of concept that it is even possible to have DVB working under mainline kernel. To make it real it must be merged into upstream.
taki You said:
I have developed this according to modern kernel architecture.
Is that mean that You wrote it from scratch or did You take the vendor driver and adapted it like I did ?
One thing is to write driver code according to mainline kernel infrastructure, other thing is to have this driver merged into upstream. I do not have expirience in pushing code to mainline kernel but I think that the code from my branch is far from this goal 
great to hear that there is simpler solution 
Well, the first thing is to access the serial console. You have to google how to do it as I don't have this G-box. Do You have the box from that link: Geektech - GBox Midnight 2.1 UART (Serial Connection) ? if jest You would need to solder wires to serial console pads. And then connect the serial to usb adapter to it.
First: I'm sure I've seen already somewhere the solution for that (but can't find it now). Try to find this answer.
Second: If You will not find the answer I could guide You how to do that. But You need access to serial console of the board, and You will need to change some u-boot environment variable. Basically I need You to paste here the output of the "printenv" command from the u-boot console. Do You know how to do that ?
So it is my understanding that when libreelec is installed, it installs the core components to the NAND flash of the device, and then upon first boot it sets the device storage to external flash (USB formatted with the proper filesystem) to use in the future.
Where You read this ? I can tell You that I put LibreELEC-Amlogic_G02REF.arm-9.0-devel-20190307225115-0e98e15.img.gz onto sd card and booted the box with it. Libreelec installed on the sd card and did not touch the nand. I can dual boot, when sd card is inserted the libreelec is booted and whithout sd card the stock android rom is booted from nand.
I have different request if I may I'm using device with dvb-t tuner si2168, and it is not detected in this build. It is however working on the stock android rom. Is it possible to add driver for this tuner ?
