Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
I have tested exactly as you described without terminating X $ sudo modprobe -r nvidia_uvm $lsmod | grep nvidia_uvm # empty result $ nvidia-modprobe -u#without sudo $ lsmod | grep nvidia_uvm # empty result! (not like it was expected) $ sudo nvidia-modprobe -u #this time with sudo -> modules are loaded nvidia_uvm765952 0 nvidia 13168640 1447 nvidia_modeset,nvidia_uvm 2018-02-06 11:45 GMT+01:00 Andreas Beckmann : > On 2018-02-05 22:35, Krzysztof Marczak wrote: >> Module loading mechanism seem to be working properly. >> I have tested it as you described: >> >> $ sudo modprobe -r nvidia >> $ lsmod | grep nvidia # there was no output >> $ sudo nvidia-modprobe -u >> $ lsmod | grep nvidia #returned folowing: >> >> nvidia_uvm765952 0 >> nvidia 13168640 1 nvidia_uvm >> >> It loaded nvidia_uvm as it should be. >> After that I started sddm and tested OpenCL. It worked properly (like after >> sudo clinfo). >> So the problem is somewhere else. > > There has been some progress made in #888952. > > Looks like a capability issue. > > Something new to try, no need to reboot or terminate X, we are just > playing around with the nvidia_uvm module: > > $ sudo setcap cap_sys_module+ep /usr/bin/nvidia-modprobe > $ sudo modprobe -r nvidia_uvm > $ lsmod | grep nvidia_uvm # expect empty > $ nvidia-modprobe -u# NO sudo! > $ lsmod | grep nvidia_uvm # expect result > > > Andreas
Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
On 2018-02-05 22:35, Krzysztof Marczak wrote: > Module loading mechanism seem to be working properly. > I have tested it as you described: > > $ sudo modprobe -r nvidia > $ lsmod | grep nvidia # there was no output > $ sudo nvidia-modprobe -u > $ lsmod | grep nvidia #returned folowing: > > nvidia_uvm765952 0 > nvidia 13168640 1 nvidia_uvm > > It loaded nvidia_uvm as it should be. > After that I started sddm and tested OpenCL. It worked properly (like after > sudo clinfo). > So the problem is somewhere else. There has been some progress made in #888952. Looks like a capability issue. Something new to try, no need to reboot or terminate X, we are just playing around with the nvidia_uvm module: $ sudo setcap cap_sys_module+ep /usr/bin/nvidia-modprobe $ sudo modprobe -r nvidia_uvm $ lsmod | grep nvidia_uvm # expect empty $ nvidia-modprobe -u# NO sudo! $ lsmod | grep nvidia_uvm # expect result Andreas
Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
Module loading mechanism seem to be working properly. I have tested it as you described: $ sudo modprobe -r nvidia $ lsmod | grep nvidia # there was no output $ sudo nvidia-modprobe -u $ lsmod | grep nvidia #returned folowing: nvidia_uvm765952 0 nvidia 13168640 1 nvidia_uvm It loaded nvidia_uvm as it should be. After that I started sddm and tested OpenCL. It worked properly (like after sudo clinfo). So the problem is somewhere else. 2018-02-05 6:01 GMT+01:00 Andreas Beckmann : > Control: tag -1 moreinfo > > On 2018-02-02 17:19, Krzysztof Marczak wrote: > > Thank you for quick reply. > > You were right. It's look like it's the same problem as reported in > #888952 > > When after reboot I don't run clinfo as a root, the NVidia OpenCL > platform > > is not visible. After running 'sudo clinfo' it starts to work properly. > > It's reproducible all the time. > I cannot reproduce the problem here. (Tried the 384.111 driver from > stretch-backports, using both ocl-icl-libopencl1 and nvidia-libopencl1. > Only possible difference is that I'm running a rather old non-distro > kernel for unrelated reasons.) > > But it should be easy for you to test the module loading mechanism, > assuming you don't have X running using the nvidia driver: > > $ sudo modprobe -r nvidia # unload all nvidia modules > $ lsmod | grep nvidia # expect no output > $ nvidia-modprobe -u# NVIDIA's setuid root helper > $ lsmod | grep nvidia # expect nvidia and nvidia-uvm > > If that doesn't work for you, the setuid helper nvidia-modprobe is not > working properly on your system > > $ ls -la /usr/bin/nvidia-modprobe > > should print > > -rwsr-xr-x 1 root root ... >^ > > > Andreas >
Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
Control: tag -1 moreinfo On 2018-02-02 17:19, Krzysztof Marczak wrote: > Thank you for quick reply. > You were right. It's look like it's the same problem as reported in #888952 > When after reboot I don't run clinfo as a root, the NVidia OpenCL platform > is not visible. After running 'sudo clinfo' it starts to work properly. > It's reproducible all the time. I cannot reproduce the problem here. (Tried the 384.111 driver from stretch-backports, using both ocl-icl-libopencl1 and nvidia-libopencl1. Only possible difference is that I'm running a rather old non-distro kernel for unrelated reasons.) But it should be easy for you to test the module loading mechanism, assuming you don't have X running using the nvidia driver: $ sudo modprobe -r nvidia # unload all nvidia modules $ lsmod | grep nvidia # expect no output $ nvidia-modprobe -u# NVIDIA's setuid root helper $ lsmod | grep nvidia # expect nvidia and nvidia-uvm If that doesn't work for you, the setuid helper nvidia-modprobe is not working properly on your system $ ls -la /usr/bin/nvidia-modprobe should print -rwsr-xr-x 1 root root ... ^ Andreas
Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
Thank you for quick reply. You were right. It's look like it's the same problem as reported in #888952 When after reboot I don't run clinfo as a root, the NVidia OpenCL platform is not visible. After running 'sudo clinfo' it starts to work properly. It's reproducible all the time. It looks like nvidia_uvm kernel module is loaded when clinfo is used when I run folowing commands: $ lsmod >modules_before.txt $ sudo clinfo $ lsmod >modules_after.txt $ diff modules_before.txt modules_after.txt the output if diff is following: 1a2 > nvidia_uvm765952 0 79c80 < nvidia 13168640 1057 nvidia_modeset --- > nvidia 13168640 1055 nvidia_modeset,nvidia_uvm This is the output of clinfo run as a root: X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 Number of platforms 2 Platform Name Intel Gen OCL Driver Platform Vendor Intel Platform VersionOpenCL 2.0 beignet 1.3 Platform ProfileFULL_PROFILE Platform Extensions cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable _store cl_khr_3d_image_writes cl_khr_image2d_from_buffer cl_khr_depth_images cl_khr_spir cl_khr_icd cl_intel_accelerator cl_intel_subgroups cl_intel_subgroups_short cl_khr_gl_sharing Platform Extensions function suffix Intel X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 Platform Name NVIDIA CUDA Platform Vendor NVIDIA Corporation Platform VersionOpenCL 1.2 CUDA 9.0.282 Platform ProfileFULL_PROFILE Platform Extensions cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 cl_khr_byte _addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_copy_opts cl_nv_create_buffer Platform Extensions function suffix NV Platform Name Intel Gen OCL Driver Number of devices 1 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 X server found. dri2 connection failed! DRM_IOCTL_I915_GEM_APERTURE failed: Invalid argument Assuming 131072kB available aperture size. May lead to reduced performance or incorrect rendering. get chip id failed: -1 [22] param: 4, val: 0 Device Name Intel(R) HD Graphics Haswell GT2 Desktop Device Vendor Intel Device Vendor ID0x8086 Device Version OpenCL 1.2 beignet 1.3 Driver Version 1.3 Device OpenCL C Version OpenCL C 1.2 beignet 1.3 Device Type GPU Device AvailableYes Device Profile FULL_PROFILE Max compute units 20 Max clock frequency 1000MHz Device Partition(core) Max number of sub-devices 1 Sup
Bug#889097: nvidia-libopencl1: Missing NVidia OpenCL platform
On 2018-02-01 22:30, Krzysztof Marczak wrote: > After lastest update to version nvidia-graphics-drivers 384.111-3 the OpenCL > platform in no longer available in the system. clinfo command lists only other > platforms (e.g. Intel). > Just before this update everything worked properly. > > I have tried to unistall (purge) all opencl-icd packages and install again, > but > it did not help. > > In actual state there is not possible to run any software using OpenCL with > nvidia proprietary drivers. Looks like a duplicate of #888952 Running clinfo as root helped there - is some module not loaded? Which package version combination was working previously? Andreas