tanglu or debian

[Vincent Cheng]

On Sat, Aug 22, 2015 at 3:33 AM, Joshua N Pritikin <jpritikin at pobox.com> wrote:
> On Fri, Aug 21, 2015 at 07:09:50PM -0700, Vincent Cheng wrote:
>> On Fri, Aug 21, 2015 at 6:13 PM, Joshua N Pritikin <jpritikin at pobox.com> wrote:
>> > I reported this problem on the tanglu forum,
>> >
>> >   http://tangluusers.org/thread/opencl-with-the-nvidia-proprietary-driver-78/
>> >
>> > Shall I also file a bug with debian? I expect tanglu is using the
>> > debian packages unmodified so my issues are probably relevant to
>> > debian.
>>
>> Have you tried installing nvidia-opencl-dev?
>
> No, I tried just now:
>
> joshua at droplet:~$ dpkg --get-selections |grep nvid
> glx-alternative-nvidia                          install
> libegl1-nvidia:amd64                            install
> libgl1-nvidia-glx:amd64                         install
> libgl1-nvidia-glx:i386                          install
> libgl1-nvidia-glx-i386:i386                     install
> libgles1-nvidia:amd64                           install
> libgles2-nvidia:amd64                           install
> libnvidia-compiler:amd64                        install
> libnvidia-eglcore:amd64                         install
> libnvidia-ml1:amd64                             install
> libnvidia-ml1:i386                              install
> nvidia-alternative                              install
> nvidia-detect                                   install
> nvidia-driver                                   install
> nvidia-driver-bin                               install
> nvidia-installer-cleanup                        install
> nvidia-kernel-common                            install
> nvidia-kernel-dkms                              install
> nvidia-modprobe                                 install
> nvidia-opencl-common                            install
> nvidia-opencl-dev:amd64                         install
> nvidia-opencl-icd:amd64                         install
> nvidia-settings                                 install
> nvidia-smi:i386                                 install
> nvidia-support                                  install
> nvidia-vdpau-driver:amd64                       install
> xserver-xorg-video-nvidia                       install
>
> I also installed nvidia-opencl-icd as suggested on the tanglu forum.
> However, it still doesn't work:
>
> joshua at droplet:~$ clinfo
> modprobe: ERROR: could not insert 'nvidia_uvm': Invalid argument
> I: ICD loader reports no usable platforms
>
> I checked journalctl and found:
>
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceChannelDestroy
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceChannelDestroy (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceQueryCaps
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceQueryCaps (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceMemoryAllocSys
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceMemoryAllocSys (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceMemoryCpuMap
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceMemoryCpuMap (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceKillChannel
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceKillChannel (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceMemoryCpuUnMap
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceMemoryCpuUnMap (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceAddressSpaceCreateMirrored
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceAddressSpaceCreateMirrored (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceServiceDeviceInterruptsRM
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceServiceDeviceInterruptsRM (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceDeRegisterUvmOps
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceDeRegisterUvmOps (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceMemoryFree
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceMemoryFree (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceGetUvmPrivRegion
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceGetUvmPrivRegion (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceGetAttachedUuids
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceGetAttachedUuids (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceSessionDestroy
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceSessionDestroy (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceCheckEccErrorSlowpath
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceCheckEccErrorSlowpath (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceAddressSpaceCreate
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceAddressSpaceCreate (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceCopyEngineAllocate
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceCopyEngineAllocate (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceAddressSpaceDestroy
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceAddressSpaceDestroy (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceRegisterUvmCallbacks
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceRegisterUvmCallbacks (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceGetGpuArch
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceGetGpuArch (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceChannelAllocate
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceChannelAllocate (err -22)
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: no symbol version for nvUvmInterfaceSessionCreate
> Aug 22 06:32:33 droplet kernel: nvidia_uvm: Unknown symbol nvUvmInterfaceSessionCreate (err -22)

This looks like a regression I introduced in 346.72-1 (reported as
Debian bug #789483 [1]). It's fixed in later versions; please install
352.30-1 from Debian experimental (or ask Tanglu's maintainers to pull
in 352.30-1 into their repositories).

Regards,
Vincent

[1] https://bugs.debian.org/789483