Visual DRM plane artifacts on vertical edges.

dividuum commented 1 month ago

Describe the bug

While playing around in my software, I managed to come across a DRM plane layout that results in artifacts on vertical plane edges. Look at this screenshot:

screenshot

I'm playing three videos across two displays (separated by the white line in the screenshot). As a result my software splits the middle video into two planes (labeled '2' and '3'). To reduce overdraw, my software also removes the non-visible part of the plane '3' that's hidden behind the plane '4'.

The video is 560x360 pixel, DRM "YU12". The relevant output in /sys/kernel/debug/dri/1/state for plane '3' is:

    crtc-pos=360x848+0+124
    src-pos=65.000000x320.000000+280.000000+0.000000

For plane '4':

    crtc-pos=1560x1080+360+0
    src-pos=560.000000x320.000000+0.000000+0.000000

So plane '4' touches plane '3' at x=360. That edge has odd bleeding of pixels going on:

pixel

Steps to reproduce the behaviour

Not sure yet. That specific configuration doesn't work. Resizing the middle video (and thus plane '2' and '3') likely fixes the issue. Tested only on a Pi5 for now.

Device (s)

Raspberry Pi 5

System

My custom info-beamer OS. Using 6.6.33 kernel from https://github.com/raspberrypi/rpi-firmware:

Linux info-beamer-xxxxxxxxx 6.6.33-v8+ #1770 SMP PREEMPT Wed Jun 12 17:47:55 BST 2024 aarch64 GNU/Linux                                                                                                                                                                                                                                                                                                                                  
2024/06/05 16:41:49 
Copyright (c) 2012 Broadcom
version 6fe0b091 (release) (embedded)

Logs

No response

Additional context

No response

popcornmix commented 1 month ago

Might be best to provide minimal test code that shows the issue(*). Can you show the issue with a single display and two video planes?

(*) it may be possible to configure kmstest in a similar way

6by9 commented 1 month ago

(Sorry, read and then forgot about this issue).

Can you also provide the contents of /sys/kernel/debug/dri/1/hvs_dlists, so that we can analyse what the HVS has been asked to compose?

dividuum commented 1 month ago

Can you show the issue with a single display and two video planes?

Not sure. I've not observed it with just a single display so far, but that might just be a coincidence. The dual display layout was just something I randomly put together and other layouts even across both displays don't show this glitch.

HVS chan 0:
dlist: 98: 0x5b000008
dlist: 99: 0x007c0168
dlist: 100: 0x4205fff0
dlist: 101: 0x034f0617
dlist: 102: 0x013f0117
dlist: 103: 0x808f011c
dlist: 104: 0x00160c0a
dlist: 105: 0x7cf80000
dlist: 106: 0x00000240
dlist: 107: 0x001b100a
dlist: 108: 0x7cfad000
dlist: 109: 0x00000120
dlist: 110: 0x001e140a
dlist: 111: 0x7cfb8400
dlist: 112: 0x00000120
dlist: 113: 0x00000000
dlist: 114: 0x4016f970
dlist: 115: 0x40304d70
dlist: 116: 0xc00071e3
dlist: 117: 0x402df260
dlist: 118: 0x40609a60
dlist: 119: 0x0000e3c6
dlist: 120: 0x00000020
dlist: 121: 0x00000020
dlist: 122: 0x00000020
dlist: 123: 0x00000020
dlist: 124: 0x80000000
dlist: 125: 0x5b000008
dlist: 126: 0x01130034
dlist: 127: 0x4205fff0
dlist: 128: 0x024005e7
dlist: 129: 0x013f022f
dlist: 130: 0x809f013f
dlist: 131: 0x0042480a
dlist: 132: 0x7cf00000
dlist: 133: 0x00000240
dlist: 134: 0x00474c0a
dlist: 135: 0x7cf2d000
dlist: 136: 0x00000120
dlist: 137: 0x004a500a
dlist: 138: 0x7cf38400
dlist: 139: 0x00000120
dlist: 140: 0x00000012
dlist: 141: 0x402f6870
dlist: 142: 0x4046fc70
dlist: 143: 0x40007700
dlist: 144: 0x405ed060
dlist: 145: 0x408df860
dlist: 146: 0x0000ee00
dlist: 147: 0x00000020
dlist: 148: 0x00000020
dlist: 149: 0x00000020
dlist: 150: 0x00000020
dlist: 151: 0x80000000
dlist: 152: 0x80000000
HVS chan 1:
dlist: 208: 0x5b000008
dlist: 209: 0x007c0000
dlist: 210: 0x4205fff0
dlist: 211: 0x034f0167
dlist: 212: 0x013f0040
dlist: 213: 0x806900d1
dlist: 214: 0x0000180a
dlist: 215: 0x7cf80118
dlist: 216: 0x00000240
dlist: 217: 0x00051c0a
dlist: 218: 0x7cfad08c
dlist: 219: 0x00000120
dlist: 220: 0x0008200a
dlist: 221: 0x7cfb848c
dlist: 222: 0x00000120
dlist: 223: 0x00000035
dlist: 224: 0x40171c70
dlist: 225: 0x40304d70
dlist: 226: 0x800046a2
dlist: 227: 0x402e3860
dlist: 228: 0x40609a60
dlist: 229: 0x80008d44
dlist: 230: 0x00000020
dlist: 231: 0x00000020
dlist: 232: 0x00000020
dlist: 233: 0x00000020
dlist: 234: 0x80000000
dlist: 235: 0x5b000008
dlist: 236: 0x00000168
dlist: 237: 0x4205fff0
dlist: 238: 0x04370617
dlist: 239: 0x013f022f
dlist: 240: 0x806600ca
dlist: 241: 0x004d540a
dlist: 242: 0x7ce80000
dlist: 243: 0x00000240
dlist: 244: 0x0052580a
dlist: 245: 0x7cead000
dlist: 246: 0x00000120
dlist: 247: 0x00555c0a
dlist: 248: 0x7ceb8400
dlist: 249: 0x00000120
dlist: 250: 0x00000039
dlist: 251: 0x402df270
dlist: 252: 0x4025ed70
dlist: 253: 0x8000579b
dlist: 254: 0x405be461
dlist: 255: 0x404bda60
dlist: 256: 0x8000af36
dlist: 257: 0x00000020
dlist: 258: 0x00000020
dlist: 259: 0x00000020
dlist: 260: 0x00000020
dlist: 261: 0x80000000
dlist: 262: 0x80000000
HVS chan 2 disabled

dividuum commented 1 month ago

I won't be able to provide a condensed test case within the next two weeks, but I can try to put one together after that. Don't put too much energy into this in the mean time.

dividuum commented 1 month ago

Ok. Got something to easily reproduce this:

Latest Raspberry Pi OS (I've used the 32bit userland version for my test case)
Download this tgz containing the 32bit info-beamer binary and stripped down code to trigger this issue.
Might need libao4 (apt-get install libao4)
Stop any wayland or X11, so DRM/KMS is free to use (service lightdm stop).
Connect two FullHD capable displays.
From a console, run ./run-testcase

The issue also happens with 64bit. If you want to run that instead, download the 64bit binary of info-beamer from here and replace the included info-beamer binary with the one from that download.

Let me know if that's helpful or if there's anything else I might provide.

pi@raspberrypi:~ $ uname -a
Linux raspberrypi 6.6.31+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.31-1+rpt1 (2024-05-29) aarch64 GNU/Linux
pi@raspberrypi:~ $ vcgencmd version
2024/06/05 16:41:49 
Copyright (c) 2012 Broadcom
version 6fe0b091 (release) (embedded)

raspberrypi / linux