Amlogic-NO discussion

But it looks like Kodi do freeze after a longer suspend like > 15min.

I look for something a differention in case of CMA memory from mali driver, which was in kernel 4.9 and is missing now in kernel 5.15 (according to dmesg log). But maybe you know more about it or this information is too old. Only for your information:

From dmesg of system CE21NG with kernel 4.9 was information about CMA in log e.g.:

Apr 26 08:15:09.319445 CoreELEC kernel: Machine model: UGOOS AM6
Apr 26 08:15:09.319493 CoreELEC kernel: earlycon: aml-uart0 at MMIO 0x00000000ff803000 (options ‘’)
Apr 26 08:15:09.319552 CoreELEC kernel: bootconsole [aml-uart0] enabled
Apr 26 08:15:09.319601 CoreELEC kernel: 07400000 - 07500000, 1024 KB, ramoops@0x07400000
Apr 26 08:15:09.319653 CoreELEC kernel: __reserved_mem_alloc_size, start:0x0000000005000000, end:0x0000000005400000, len:4 MiB
Apr 26 08:15:09.319702 CoreELEC kernel: 05000000 - 05400000, 4096 KB, linux,secmon
Apr 26 08:15:09.319750 CoreELEC kernel: __reserved_mem_alloc_size, start:0x0000000077800000, end:0x0000000078000000, len:8 MiB
Apr 26 08:15:09.319799 CoreELEC kernel: 77800000 - 78000000, 8192 KB, linux,meson-fb
Apr 26 08:15:09.319846 CoreELEC kernel: ed000000 - f5000000, 131072 KB, linux,ion-dev
Apr 26 08:15:09.319895 CoreELEC kernel: eb400000 - ed000000, 28672 KB, linux,ion-fb
Apr 26 08:15:09.319960 CoreELEC kernel: Reserved memory: created DMA memory pool at 0x00000000eb400000, size 0 MiB
Apr 26 08:15:09.320008 CoreELEC kernel: eb400000 - eb400000, 0 KB, linux,ppmgr
Apr 26 08:15:09.320057 CoreELEC kernel: d0400000 - eb400000, 442368 KB, linux,codec_mm_cma
Apr 26 08:15:09.320105 CoreELEC kernel: d0400000 - d0400000, 0 KB, linux,codec_mm_reserved
Apr 26 08:15:09.320151 CoreELEC kernel: cf400000 - d0400000, 16384 KB, linux,galcore
Apr 26 08:15:09.320197 CoreELEC kernel: cma: Reserved 8 MiB at 0x00000000cec00000
Apr 26 08:15:09.320245 CoreELEC kernel: On node 0 totalpages: 1003520
Apr 26 08:15:09.320293 CoreELEC kernel: DMA zone: 15680 pages used for memmap
Apr 26 08:15:09.320339 CoreELEC kernel: DMA zone: 0 pages reserved
Apr 26 08:15:09.320385 CoreELEC kernel: DMA zone: 1003520 pages, LIFO batch:31
Apr 26 08:15:09.322184 CoreELEC kernel: Memory: 3268424K/4014080K available (13310K kernel code, 2370K rwdata, 5768K rodata, 5888K init, 1429K bss, 135352K reserved, 610304K cma-reserved)

According odroid forum was issue with mali driver due to missing of cma in newer kernels:

It is unlikely that a memory leak is causing the issue. On Amlogic platforms, the display controller does not have an IOMMU and therefore requires CMA memory (contiguous). When more than one allocation is made, and then freed, the memory becomes “fragmented”. This means that although enough memory may be available, it is not contiguous and the allocation fails. CMA memory only comes from a reserved pool to guarantee its contiguous. You can try adjusting this value as a kernel parameter.

The more relevant answer is that unless your intent to actually zero-copy the display output, a CMA buffer is not required. This is the case in X11 and Wayland without DMABUF protocol support. The use of DRM_IOCTL_MODE_CREATE_DUMB explicitly asks the display driver to allocate and supply a buffer. As noted before, this will always be CMA on current Amlogic platforms. For general purpose applications, the buffer allocation should be done with “gbm” instead. This requests a buffer from the GPU (Mali) that does have an IOMMU and can utilize any available free memory for the allocation (DMA scatter/gather). For maximum 3D performance, allocations should also come from “gbm” for a different reason. Like many GPUs, Mali is a tile renderer. Its internal memory layout is in blocks rather than linear (as used by the display). Having to convert to/from tiled memory to linear memory will decrease performance since this should only be done once a frame is complete (swapped).

With regard to rotation, console (text) rotation can be accomplished with a kernel parameter. Wayland supports rotation in ‘weston.ini’. For X11, if the ‘modeset’ driver is in use, then GPU based rotation should also be available.

[edit]
I should probably add that (at least on ARM platforms), the use of DRM_IOCTL_MODE_CREATE_DUMB is highly discouraged. Drivers typically provide a vendor specific IOCTL that provides the same functionally, but enables write-back caching on the buffer. The libgbm attempts to consolidate this with vendor specific packaging.

https://forum.odroid.com/viewtopic.php?t=47788

System Log file
log-2025-04-26-07.43.01.zip (97.1 KB)

Interesting. On my device ugoos am6plus with s922xj at CE22NO is not information about cma:
https://paste.coreelec.org/RefuelSettled

reboot device to get start of dmesg

yes now i see

https://paste.coreelec.org/CalmlyOutcome

Update of the 27th, we have sleep mode working again on the s905x4, congratulations to the team. However, the system continues to be slow on the first boot, if I try to play something it starts to stutter (reading and writing from the SD card is slow), however, when entering sleep mode and turning the TV back on, everything is normal.

Edit 29th Abril: It continues the same thing above in tox3, the performance only works normally without bottleneck after suspending the system once.

Yes at am6plus is it same. Samples from sd card but also from other disk are played after longer waiting time. After wakeup from suspend mode is time of start playing again ok.

https://paste.coreelec.org/MunchObituary

Is it normal to the database to migrate to a “new” version when updating?
It was on “video133” which is right one for kodi 22 but I now have “video134” and “video135”
(Running the nightly on 2 boxes, they both were using the same SQL database 134 so I deleted 133 and now it created 135)

Version is bumped from time to time and 135 is the last one.

On TOX3 CE 22 NO MariaDB not working properly.it connects to the DB fine, but no files shown at all using current nightly. the last good 100% working nightly is from 20250406.
Is it working to anyone using MariaDB or MySQL ?

That is very normal in the pre-alpha and alpha stages, don’t update nightles if you don’t want to upgrade all your clients when you use a central database.

That was my point, both are using the same NO version and kodi docs state database version to be 133.

I’m using MariaDB and is working fine. Were you using a previous version like ce21 and it hasn’t migrated properly?

You missed my post above with latest version 135 ?

I can not help much when no sample with this issue is available.
It looks your DVB source.
You must record the stream directly in Kodi to reproduce the issue.

It sound the same issue someone have with france streams.
But without a source to test there is no chance to find the commit what it does break it.

Fair enough but the wiki is not the official documentation, @vpeter pointed to the source.

Tried with other DVB sources same problem. I recorded the stream directly to Kodi.
Simultaneous recording of two TV channels on the same TP.
I applied simultaneous recording for comparison.
https://s6.dosya.tc/server22/r8aabq/Stream_Record__Kodi.rar.html
Note:The recording named paratoner is a recording of the problematic channel.
Other channel works fine.

Does playback work for the recording?
If yes, it’s not usefull.

I can watch the recordings I made.
I can also watch it when I turn on timeshift on the channel
Tested with TBS5520SE, TBS5530 and TBS5930 DVB tuners and on SK1, AM8 and Gt King II.
I can open terminal sharing.