cgutman
commented
10 months ago Please run Sunshine with the MESA_DEBUG=1 environment variable per https://docs.mesa3d.org/debugging.html and post the updated log.
Open SzczurekYT opened 10 months ago
cgutman
commented
10 months ago Please run Sunshine with the MESA_DEBUG=1 environment variable per https://docs.mesa3d.org/debugging.html and post the updated log.
SzczurekYT
commented
10 months ago Here you go sunshine.log
cgutman
commented
10 months ago Hmm, out of memory is pretty strange...
Mesa: error: GL_OUT_OF_MEMORY in glGetTexImage
[2024:01:20:16:49:19]: Error: GL: /run/build/sunshine/src/platform/linux/graphics.cpp:562: [00000505]Can you try with Sunshine's min log level set to Debug?
If you can give me steps to create a VKMS setup similar to yours, I can try it here. I have used AMDGPU's VKMS support and it worked fine for me after #2030.
SzczurekYT
commented
10 months ago I just do sudo modprobe vkms and use correct number in sunshine. I also set some resolution in the system settings I think.
Never heard of amdgpu vkms though.
For the log level I'll try when I will have some time, maybe tomorrow.
SzczurekYT
commented
10 months ago Can you try with Sunshine's min log level set to Debug?
Also: Where does flatpak sunshine store it's config? Sometimes I would like to edit something manually but I just can't find it. It's not in ~/.config/sunshine and ~/.var/app/dev.lizardbyte.sunshine/config/sunshine doesn't seem to be used.
cgutman
commented
10 months ago OK, I dug into this quite a while. This will not be easy to resolve properly.
The problem turns out to be that the dmabufs used as frame buffers on the vkms device (at least with Mutter) are imported from the rendering GPU. If you force the compositor to run using llvmpipe instead of a hardware accelerated GPU, I'll bet it works fine.
You can see this in debugfs:
cat /sys/kernel/debug/dri/0/state
plane[31]: plane-0
crtc=crtc-0
fb=48
allocated by = gnome-shell
refcount=3
format=XR24 little-endian (0x34325258)
modifier=0x0
size=1024x768
layers:
size[0]=1024x768
pitch[0]=4096
offset[0]=0
obj[0]:
name=0
refcount=2
start=00147000
size=3145728
imported=yes <-------- HERE
crtc-pos=1024x768+0+0
src-pos=1024.000000x768.000000+0.000000+0.000000
rotation=1
normalized-zpos=0
color-encoding=ITU-R BT.601 YCbCr
color-range=YCbCr limited rangeThis is a problem because the kernel explicitly forbids mapping of imported dmabufs
kms_swrast needs to map the buffer to read/copy it, which fails here:
(gdb) bt
#0 kms_sw_displaytarget_map (ws=..., dt=..., flags=...) at ../src/gallium/winsys/sw/kms-dri/kms_dri_sw_winsys.c:272
#1 0x00007fffae198379 in llvmpipe_resource_map (resource=..., level=..., layer=..., tex_usage=...) at ../src/gallium/drivers/llvmpipe/lp_texture.c:523
#2 0x00007fffae198710 in llvmpipe_transfer_map_ms (pipe=..., resource=..., level=..., usage=..., sample=sample@entry=..., box=..., transfer=...) at ../src/gallium/drivers/llvmpipe/lp_texture.c:777
#3 0x00007fffae198769 in llvmpipe_transfer_map (pipe=..., resource=..., level=..., usage=..., box=..., transfer=...) at ../src/gallium/drivers/llvmpipe/lp_texture.c:806
#4 0x00007fffadbf05fe in pipe_texture_map_3d (transfer=..., d=..., h=..., w=..., z=..., y=..., x=..., access=..., level=..., resource=..., context=...) at ../src/gallium/auxiliary/util/u_inlines.h:577
#5 st_texture_image_map (st=st@entry=..., stImage=stImage@entry=..., usage=usage@entry=..., x=x@entry=..., y=y@entry=..., z=z@entry=..., w=..., h=..., d=..., transfer=...) at ../src/mesa/state_tracker/st_texture.c:315
#6 0x00007fffadbc1dee in st_MapTextureImage (ctx=ctx@entry=..., texImage=texImage@entry=..., slice=..., x=x@entry=..., y=y@entry=..., w=w@entry=..., h=..., mode=..., mapOut=..., rowStrideOut=...)
at ../src/mesa/state_tracker/st_cb_texture.c:547
#7 0x00007fffade141bf in get_tex_rgba_uncompressed (transferOps=..., texImage=..., pixels=..., type=..., format=..., depth=..., height=..., width=..., zoffset=..., yoffset=..., xoffset=..., dimensions=..., ctx=...)
at ../src/mesa/main/texgetimage.c:478
#8 get_tex_rgba (texImage=..., pixels=..., type=..., format=..., depth=..., height=..., width=..., zoffset=..., yoffset=..., xoffset=..., dimensions=..., ctx=...) at ../src/mesa/main/texgetimage.c:605
#9 _mesa_GetTexSubImage_sw (ctx=ctx@entry=..., xoffset=xoffset@entry=..., yoffset=..., yoffset@entry=..., zoffset=..., zoffset@entry=..., width=width@entry=..., height=...,
height@entry=..., depth=..., format=..., type=..., pixels=..., texImage=...) at ../src/mesa/main/texgetimage.c:760
#10 0x00007fffadbc4b85 in st_GetTexSubImage
(ctx=ctx@entry=..., xoffset=xoffset@entry=..., yoffset=yoffset@entry=..., zoffset=zoffset@entry=..., width=width@entry=..., height=height@entry=..., depth=..., format=..., type=..., pixels=..., texImage=...)
at ../src/mesa/state_tracker/st_cb_texture.c:2733
#11 0x00007fffade13539 in get_texture_image
(ctx=ctx@entry=..., texObj=texObj@entry=..., target=..., level=level@entry=..., xoffset=xoffset@entry=..., yoffset=yoffset@entry=..., zoffset=..., width=..., height=..., depth=..., format=..., type=..., pixels=..., caller=...) at ../src/mesa/main/texgetimage.c:1441
#12 0x00007fffade1515a in _mesa_GetTextureSubImage (texture=..., level=..., xoffset=..., yoffset=..., zoffset=..., width=..., height=..., depth=..., format=..., type=..., bufSize=..., pixels=...)
at ../src/mesa/main/texgetimage.c:1633
#13 0x00000000006620d8 in platf::kms::display_ram_t::snapshot(std::function<bool (std::shared_ptr<platf::img_t>&)> const&, std::shared_ptr<platf::img_t>&, std::chrono::duration<long, std::ratio<1l, 1000l> >, bool)
(this=..., pull_free_image_cb=..., img_out=..., timeout=..., cursor=...) at /home/cgutman/moonlight-src/LB_Sunshine/src/platform/linux/kmsgrab.cpp:1267
#14 0x000000000066178f in platf::kms::display_ram_t::capture(std::function<bool (std::shared_ptr<platf::img_t>&&, bool)> const&, std::function<bool (std::shared_ptr<platf::img_t>&)> const&, bool*)
(this=..., push_captured_image_cb=..., pull_free_image_cb=..., cursor=...) at /home/cgutman/moonlight-src/LB_Sunshine/src/platform/linux/kmsgrab.cpp:1162
#15 0x00000000005dd928 in video::captureThread (capture_ctx_queue=..., display_wp=..., reinit_event=..., encoder=...) at /home/cgutman/moonlight-src/LB_Sunshine/src/video.cpp:1299Fortunately this only applies to the swrast codepath that is used when performing CPU encoding or readback VAAPI encoding. It is possible to avoid this issue when performing GPU encoding assuming the primary GPU can import the FB from VKMS.
LizardByte-bot
commented
7 months ago It seems this issue hasn't had any activity in the past 90 days. If it's still something you'd like addressed, please let us know by leaving a comment. Otherwise, to help keep our backlog tidy, we'll be closing this issue in 10 days. Thanks!
d-air1
commented
7 months ago Keep alive.
LizardByte-bot
commented
4 months ago It seems this issue hasn't had any activity in the past 90 days. If it's still something you'd like addressed, please let us know by leaving a comment. Otherwise, to help keep our backlog tidy, we'll be closing this issue in 10 days. Thanks!
SzczurekYT
commented
4 months ago Nope, still not resolved, don't mark it as stale. @cgutman Do you still plan on finishing your pr #2053 ?
LizardByte-bot
commented
1 month ago It seems this issue hasn't had any activity in the past 90 days. If it's still something you'd like addressed, please let us know by leaving a comment. Otherwise, to help keep our backlog tidy, we'll be closing this issue in 10 days. Thanks!
paccerdk
commented
1 month ago Unresolved, not stale, this is also needed if wanting to use a tablet / computer as an additional screen
Is there an existing issue for this?
Is your issue described in the documentation?
Is your issue present in the nightly release?
Describe the Bug
Hello I'm creating a second virtual monitor with vkms and trying to stream it with sunshine. But when Moonlight client connects you see a black screen and one error line is spammed in the logs. [2024:01:20:14:04:07]: Error: GL: /run/build/sunshine/src/platform/linux/graphics.cpp:562: [00000505]
Expected Behavior
Working stream displaying a virtual monitor.
Additional Context
I'm using latest nightly to incorporate latest fix without which the virtual monitor wasn't even detected properly. https://github.com/LizardByte/Sunshine/pull/2030 I'm on wayland.
Side question: The Capture Method (Linux Only) fields lists PipeWire as a capture method but I don't see any mention of it in the docs. Does sunshine have or has plans on supporting pipewire capture? I think that could be useful and a nice way to get another capture method on wayland.
Host Operating System
Linux
Operating System Version
Nobara Linux 39 (fedora fork)
Architecture
64 bit
Sunshine commit or version
a10ec3a98ac6a3f5618dc53469f3c5b48dab36ee (Latest nightly as of now)
Package
Linux - flatpak
GPU Type
AMD
GPU Model
AMD Radeon RX 5600 XT
GPU Driver/Mesa Version
Mesa 23.3.3
Capture Method (Linux Only)
KMS
Config
Apps
Relevant log output