Rockchip kernels with RGA/MPP VPU hw acceleration

sk4g · Friday at 12:32 PM

Can someone explain the purpose of the different kernels available in the Armbian repos and Armbian Github?

linux-image-legacy-media/bookworm 24.5.1 arm64
   Armbian Linux legacy kernel image 5.10.110-legacy-media

linux-image-current-media/bookworm 24.5.1 arm64
   Armbian Linux current kernel image 6.1.92-current-media

linux-image-current-rockchip64/bookworm 25.2.3 arm64
   Armbian Linux current kernel image 6.12.19-current-rockchip64

The linux-image-legacy-media and linux-image-current-media kernels are on BSP kernel versions 5.10 and 6.1. But aren't suffixed with the vendor name, which is typically done for BSP kernels.

The linux-image-current-rockchip64 kernel is on a mainline version of 6.12.19. Why is it suffixed with rockchip64?

There are also kernels such as rk-6.1-rkr5 that are not in the repos, but can be built from https://github.com/armbian/linux-rockchip

These are on BSP kernel versions such as 5.10 and 6.1.

What is the purpose of the armbian-bsp packages such as armbian-bsp-cli-rock64-current and armbian-bsp-desktop-rock64-current? What are their differences?

Do any of these kernels support RGA/MPP Rockchip VPU hw acceleration?
The Jellyfin docs state,

Quote

Make sure dma_heap, dri, mpp_service and rga exist in /dev. Otherwise upgrade your BSP kernel to 5.10 LTS and newer.

6.12.19-current-rockchip64 is missing rga and mpp

root@rock64:~# uname -r
6.12.19-current-rockchip64

root@rock64:~# ls -l /dev | grep -E "mpp|rga|dri|dma_heap"
drwxr-xr-x 2 root root          80 Jan  1  1970 dma_heap
drwxr-xr-x 3 root root         120 Mar 21 10:59 dri

root@rock64:~# lsmod | grep rockchip
rockchip_vdec          77824  0
v4l2_vp9               20480  2 rockchip_vdec,hantro_vpu
v4l2_h264              16384  2 rockchip_vdec,hantro_vpu
rockchip_rga           20480  0
videobuf2_dma_contig    20480  2 rockchip_vdec,hantro_vpu
videobuf2_dma_sg       16384  1 rockchip_rga
v4l2_mem2mem           24576  3 rockchip_vdec,hantro_vpu,rockchip_rga
videobuf2_v4l2         20480  4 rockchip_vdec,hantro_vpu,rockchip_rga,v4l2_mem2mem
videodev              237568  5 rockchip_vdec,videobuf2_v4l2,hantro_vpu,rockchip_rga,v4l2_mem2mem
snd_soc_rockchip_spdif    12288  0
snd_soc_rockchip_i2s    24576  4
videobuf2_common       53248  8 rockchip_vdec,videobuf2_dma_contig,videobuf2_v4l2,hantro_vpu,rockchip_rga,videobuf2_dma_sg,v4l2_mem2mem,videobuf2_memops
mc                     53248  6 rockchip_vdec,videodev,videobuf2_v4l2,hantro_vpu,videobuf2_common,v4l2_mem2mem
snd_soc_core          204800  7 snd_soc_spdif_tx,snd_soc_hdmi_codec,snd_soc_rockchip_spdif,snd_soc_rk3328,snd_soc_simple_card_utils,snd_soc_rockchip_i2s,snd_soc_simple_card
nvmem_rockchip_efuse    12288  0
rockchipdrm           180224  0
dw_hdmi_qp             20480  1 rockchipdrm
dw_mipi_dsi            16384  1 rockchipdrm
dw_hdmi                45056  2 dw_hdmi_i2s_audio,rockchipdrm
analogix_dp            40960  1 rockchipdrm
drm_display_helper    131072  4 rockchipdrm,dw_hdmi,analogix_dp,dw_hdmi_qp

I did get ffmpeg hw acceleration working using v4l2request from https://github.com/jernejsk/FFmpeg/tree/v4l2-request-n7.1 using the 6.12.19-current-rockchip64 kernel. But wanted to see if jellyfin-ffmpeg7 or https://github.com/nyanmisaka/ffmpeg-rockchip performed better.

v4l2request does 49 fps, software decoding does 60 fps, but uses 2.6x more cpu.

root@rock64:~/github/FFmpeg# ffmpeg -hwaccel v4l2request -i 1080p.264 -f null - -benchmark
ffmpeg version 904a851 Copyright (c) 2000-2024 the FFmpeg developers
  built with gcc 12 (Debian 12.2.0-14)
  configuration: --enable-v4l2-request --enable-libudev --enable-libdrm --enable-gnutls --prefix="$HOME/ffmpeg_build"
  libavutil      59. 39.100 / 59. 39.100
  libavcodec     61. 19.100 / 61. 19.100
  libavformat    61.  7.100 / 61.  7.100
  libavdevice    61.  3.100 / 61.  3.100
  libavfilter    10.  4.100 / 10.  4.100
  libswscale      8.  3.100 /  8.  3.100
  libswresample   5.  3.100 /  5.  3.100
Input #0, h264, from '/root/1080p.264':
  Duration: N/A, bitrate: N/A
  Stream #0:0: Video: h264 (Main), yuv420p(progressive), 1920x1080 [SAR 1:1 DAR 16:9], 25 fps, 59.94 tbr, 1200k tbn
Stream mapping:
  Stream #0:0 -> #0:0 (h264 (native) -> wrapped_avframe (native))
Press [q] to stop, [?] for help
[h264 @ 0xaaaaf4be97b0] Using V4L2 media driver rkvdec (6.12.19) for S264
Output #0, null, to 'pipe:':
  Metadata:
    encoder         : Lavf61.7.100
  Stream #0:0: Video: wrapped_avframe, nv12(progressive), 1920x1080 [SAR 1:1 DAR 16:9], q=2-31, 200 kb/s, 29.97 fps, 29.97 tbn
      Metadata:
        encoder         : Lavc61.19.100 wrapped_avframe
[out#0/null @ 0xaaaaf4be7320] video:485KiB audio:0KiB subtitle:0KiB other streams:0KiB global headers:0KiB muxing overhead: unknown
frame= 1129 fps= 49 q=-0.0 Lsize=N/A time=00:00:37.67 bitrate=N/A speed=1.64x    
bench: utime=21.094s stime=2.279s rtime=22.936s
bench: maxrss=72164KiB

Werner · Friday at 12:49 PM

moved

17 minutes ago, sk4g said:
media

Where did you get those? media branch has been abandoned for over a year now.

sk4g · Friday at 12:56 PM

They are in the debian bookworm repos. I ran apt search linux-image-*

Werner · Friday at 01:01 PM

Alright, didn't know they are still there. Well as mentioned media won't receive any further updates since all functionally is merged into standard branches.

Anyway, Rock64 is rk3399 I think and you should be good to go with mainline kernel, either current which is 6.12.y or edge which is 6.14-rc7 atm.

The branches at https://github.com/armbian/linux-rockchip are made from Rockchip BSP code and mostly for rk35xx socs. I don't think there is much work there for rk3399.