libcuda.so.1 is not a symbolic link

[sirredbeard]

This is cross-post of Ubuntu bug 1886532 which is pending investigation.

On Windows 10 build 20161 WSL2 running sudo apt upgrade results in warning:

"libcuda.so.1 is not a symbolic link"

It appears to be a result of a postinst hook in libc-bin.

https://bugs.launchpad.net/ubuntuwsl/+bug/1886532

[therealkenc]

Yeah, copies not symlinks.

[zaxioms0]

Is there a solution for this? My /usr/lib/wsl/lib looks exactly like this. Where do I need symlinks and from where to where?

[therealkenc]

It's in %SYSTEMROOT%\System32\lxss\lib. It might be possible to make it a symlink in that folder, though I haven't tried and mucking your System32 is not recommended for lots of good reasons. You could also probably make the warning go away by mount binding a symlink over the file on the WSL side. Haven't tried that either.

Thus far the warning hasn't proved blocking in any way. It isn't even that clear (to me) why ldconfig feels inclined to point it out, because there isn't anything technically "wrong" with the file not being a symlink, except for violating a convention. Absent some heretofore unstated blocking use-case, the smart play ("solution") is to ignore the warning.

[zaxioms0]

What would cause problems with symlinking from system32. Like I know generally altering system32 is bad, but would symlinks cause problems?

[wanfuse123]

Directory of C:\Windows\System32\lxss\lib

09/18/2020 03:53 PM

.
08/30/2020 09:51 AM 124,664 libcuda.so
09/12/2020 08:44 AM 832,936 libd3d12.so
09/12/2020 08:44 AM 5,073,944 libd3d12core.so
09/12/2020 08:44 AM 25,069,816 libdirectml.so
09/12/2020 08:44 AM 878,768 libdxcore.so
08/30/2020 09:51 AM 40,496,936 libnvwgf2umx.so
6 File(s) 72,477,064 bytes
1 Dir(s) 643,723,309,056 bytes free

C:\Windows\System32\lxss\lib>mklink libcuda.so.1 libcuda.so
symbolic link created for libcuda.so.1 <<===>> libcuda.so

[wanfuse123]

simlynk generates new error

sudo docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi
Failed to initialize NVML: Unknown Error

[wanfuse123]

docker run --gpus all nvcr.io/nvidia/k8s/cuda-sample:nbody nbody -privileged -gpu -benchmark
Run "nbody -benchmark [-numbodies=]" to measure performance.
-fullscreen (run n-body simulation in fullscreen mode)
-fp64 (use double precision floating point values for simulation)
-hostmem (stores simulation data in host memory)
-benchmark (run benchmark to measure performance)
-numbodies= (number of bodies (>= 1) to run in simulation)
-device= (where d=0,1,2.... for the CUDA device to use)
-numdevices= (where i=(number of CUDA devices > 0) to use for simulation)
-compare (compares simulation results running once on the default GPU and once on the CPU)
-cpu (run n-body simulation on the CPU)
-tipsy=<file.bin> (load a tipsy model file for simulation)

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

Error: only 0 Devices available, 1 requested. Exiting.

NOTE: the -privileged switch changes the error from above

[PeerXu]

$ sudo docker run --gpus all nvcr.io/nvidia/k8s/cuda-sample:nbody nbody -privileged -gpu -benchmark
CUDA error at bodysystemcuda_impl.h:159 code=46(cudaErrorDevicesUnavailable) "cudaEventCreate(&m_deviceData[0].event)" 
Run "nbody -benchmark [-numbodies=<numBodies>]" to measure performance.
	-fullscreen       (run n-body simulation in fullscreen mode)
	-fp64             (use double precision floating point values for simulation)
	-hostmem          (stores simulation data in host memory)
	-benchmark        (run benchmark to measure performance) 
	-numbodies=<N>    (number of bodies (>= 1) to run in simulation) 
	-device=<d>       (where d=0,1,2.... for the CUDA device to use)
	-numdevices=<i>   (where i=(number of CUDA devices > 0) to use for simulation)
	-compare          (compares simulation results running once on the default GPU and once on the CPU)
	-cpu              (run n-body simulation on the CPU)
	-tipsy=<file.bin> (load a tipsy model file for simulation)

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

> Windowed mode
> Simulation data stored in video memory
> Single precision floating point simulation
> 1 Devices used for simulation
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1

> Compute 6.1 CUDA device: [GeForce GTX 1080]

[wanfuse123]

What happens if tou ad -- compair after benchmark ? Forget it if one or two dashes

[Shohou]

/usr/lib/wsl/lib is included through /etc/ld.so.conf.d. You will find a file there with instructions how to prevent its generation. You have to delete that file and create /etc/wsl.conf with the fillowing lines:

[automount]
ldconfig = false

[thierryherrmann]

Same problem here using Win 10 build 20257.1 both with ubuntu 18.04 and ubuntu 20.04 (kernel 4.19.128).
Problem shows up either through sudo apt-get install -y nvidia-docker2 or sudo apt-get install -y cuda-toolkit-11-0, two steps mentioned in the official NVidia CUDA-wsl procedure.
Must have worked at some point but something must have been broken

[ligewu]

1. disable automount in /etc/wsl.conf

[automount]
ldconfig = false

2. copy /usr/lib/wsl/lib to /usr/lib/wsl2/lib (in wsl, writable)

3. edit /etc/ld.so.conf.d/ld.wsl.conf

/usr/lib/wsl/lib --> /usr/lib/wsl2/lib (new location)

4. rm /usr/lib/wsl2/lib/libcuda.so.* and sudo ldconfig

works for CUDA in WSL, but "Segmentation fault" in DirectML

[ImanHosseini]

Will this issue ever be fixed?