User:Rellla

Hardware

A10

• Cubieboard
• Mele A100
• Mele A2000

A20

• Cubietech Cubieboard2
• Cubietech Cubietruck
• LeMaker Banana Pi

H3

• Xunlong Orange Pi Plus

H5

• Xunlong Orange Pi PC 2

H6

• Xunlong Orange Pi One Plus
• Eachlink H6 Mini

Software

• Filesystem/ Distribution: Debian from scratch (User:Rellla/Debian as my own helpful installation record)
• Armbian
• Github
• Freedesktop Gitlab

Notices

Sunxi display driver bugs collection:

• https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_layer.c#L566
• https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_video.c#L77
• https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/de_fe.c#L1385
• File:Ve tile format v1.pdf
• Tile format conversion code

Useful links

• https://github.com/dvdhrm/docs/tree/master/drm-howto
• http://events.linuxfoundation.org/sites/events/files/slides/brezillon-drm-kms.pdf

[DISP] not supported image0 pixel sequence

Sometimes we have some issues with the 3.4 display driver. dmesg is flooded with messages like:

[DISP] not supported image0 pixel sequence:208 in img_sw_para_to_reg

or

[DISP] not supported yuv channel pixel sequence:0 in img_sw_para_to_reg

I think, this can be tracked down to https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_video.c#L388 Hal_Set_Frame executes https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_video.c#L357 and https://github.com/linux-sunxi/linux-sunxi/blob/sunxi-3.4/drivers/video/sunxi/disp/disp_video.c#L362 which could be both responsible for the dmesg output in case the parameters don't exist anymore in memory. This would explain the different and not-reasonable values. Hal_Set_Frame is executed during VBlanking, which would explain the massive flooding. This all can happen when ((g_video[sel][id].enable == TRUE) && (g_video[sel][id].have_got_frame == TRUE)), which is triggered by the DISP_CMD_VIDEO_START and DISP_CMD_VIDEO_SET_FB ioctl. Both are set false with DISP_CMD_VIDEO_STOP again.

So, in case we forget the DISP_CMD_VIDEO_STOP ioctl, we may run into some issues with faulty parameters in Hal_Set_Frame itself. To debug that kernel bug, we could set some logs into Hal_Set_Frame, to see what happens. Or simply try to consequently use DISP_CMD_VIDEO_STOP. I could see this issue sometimes, if i CTRL-C a running program (e.g. sth. with libvdpau-sunxi) and then do another CTRL-C. Maybe everything is killed but g_video[sel][id].enable isn't set to FALSE again, which keeps Hal_Set_Frame executed every VBlank.

I think it has nothing to do with the other ioctls which trigger the img_sw_para_to_reg function in the end because of the massive (most likely 50Hz) flood of dmesg. With this background, one maybe can reproduce the issue.