Re: [gentoo-user] Re: "No CUDA device found" with nvidia-drivers newer than nvidia-drivers-396.24-r1(

[Corentin “Nado” Pazdera]

August 15, 2018 5:45 PM, tuxic@××××××.de wrote:

> I put nvidia-uvm explictly into /etc/conf.d/modules - which was not
> necessary ever before....and it shows the same problems: No cuda
> devices.
>
> I think I will dream this night of no cuda devices... ;(
>
> On 08/15 05:11, tuxic@××××××.de wrote:
>
>> On 08/15 02:32, Corentin “Nado” Pazdera wrote:
>> August 15, 2018 2:59 PM, tuxic@××××××.de wrote:
>
> Yes I did reboot the sustem. In my initial mail I mentioned a tool
> called CUDA-Z and Blender, which both reports a missing CUDA device.
>> Ok, so you do not have a specific error which might have been thrown by the module?
>> Other ideas, check dev-util/nvidia-cuda-toolkit version and double check nvidia/nvidia_uvm with
>> modinfo to ensure they are installed and loaded correctly with the right version?
>> Could you also run /opt/cuda/extras/demo_suite/deviceQuery (from nvidia-cuda-toolkit) and show its
>> output?
>>
>> My installation works, so at least we know their version is not completely broken...
>> Driver version: 396.51
>> Cuda version: 9.2.88
>>
>> --
>> Corentin “Nado” Pazdera
>>
>> I compiled the new version of the driver again and rebooted the
>> system.
>>
>> # dmesg | grep -i nvidia:
>>
>> [ 11.375631] nvidia_drm: module license 'MIT' taints kernel.
>> [ 12.313260] nvidia-nvlink: Nvlink Core is being initialized, major device number 246
>> [ 12.313586] nvidia 0000:07:00.0: vgaarb: changed VGA decodes:
>> olddecodes=io+mem,decodes=none:owns=io+mem
>> [ 12.313691] nvidia 0000:02:00.0: enabling device (0000 -> 0003)
>> [ 12.313737] nvidia 0000:02:00.0: vgaarb: changed VGA decodes:
>> olddecodes=io+mem,decodes=none:owns=none
>> [ 12.313826] NVRM: loading NVIDIA UNIX x86_64 Kernel Module 396.51 Tue Jul 31 10:43:06 PDT 2018
>> (using threaded interrupts)
>> [ 12.491106] input: HDA NVidia HDMI as
>> /devices/pci0000:00/0000:00:0b.0/0000:02:00.1/sound/card2/input9
>> [ 12.492291] input: HDA NVidia HDMI as
>> /devices/pci0000:00/0000:00:0b.0/0000:02:00.1/sound/card2/input10
>> [ 12.493772] input: HDA NVidia HDMI as
>> /devices/pci0000:00/0000:00:02.0/0000:07:00.1/sound/card1/input11
>> [ 12.494605] input: HDA NVidia HDMI as
>> /devices/pci0000:00/0000:00:02.0/0000:07:00.1/sound/card1/input12
>> [ 13.963644] caller _nv001112rm+0xe3/0x1d0 [nvidia] mapping multiple BARs
>> [ 34.236553] caller _nv001112rm+0xe3/0x1d0 [nvidia] mapping multiple BARs
>> [ 34.516495] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 396.51
>> Tue Jul 31 14:52:09 PDT 2018
>>
>> # modprobe -a nvidia-uvm
>>
>> # dmesg | grep uvm
>>
>> [ 209.441956] nvidia-uvm: Loaded the UVM driver in 8 mode, major device number 245
>>
>> # /opt/cuda/extras/demo_suite/deviceQuery
>> /opt/cuda/extras/demo_suite/deviceQuery Starting...
>>
>> CUDA Device Query (Runtime API) version (CUDART static linking)
>>
>> cudaGetDeviceCount returned 30
>> -> unknown error
>> Result = FAIL
>> [1] 5086 exit 1 /opt/cuda/extras/demo_suite/deviceQuery
>>
>> CUDA-Z shows also "no CUDA device"
>>
>> # modinfo nvidia-uvm
>> filename: /lib/modules/4.18.0-RT/video/nvidia-uvm.ko
>> supported: external
>> license: MIT
>> depends: nvidia
>> name: nvidia_uvm
>> vermagic: 4.18.0-RT SMP preempt mod_unload
>> parm: uvm_perf_prefetch_enable:uint
>> parm: uvm_perf_prefetch_threshold:uint
>> parm: uvm_perf_prefetch_min_faults:uint
>> parm: uvm_perf_thrashing_enable:uint
>> parm: uvm_perf_thrashing_threshold:uint
>> parm: uvm_perf_thrashing_pin_threshold:uint
>> parm: uvm_perf_thrashing_lapse_usec:uint
>> parm: uvm_perf_thrashing_nap_usec:uint
>> parm: uvm_perf_thrashing_epoch_msec:uint
>> parm: uvm_perf_thrashing_max_resets:uint
>> parm: uvm_perf_thrashing_pin_msec:uint
>> parm: uvm_perf_map_remote_on_native_atomics_fault:uint
>> parm: uvm_hmm:Enable (1) or disable (0) HMM mode. Default: 0. Ignored if CONFIG_HMM is not set, or
>> if NEXT settings conflict with HMM. (int)
>> parm: uvm_global_oversubscription:Enable (1) or disable (0) global oversubscription support. (int)
>> parm: uvm_leak_checker:Enable uvm memory leak checking. 0 = disabled, 1 = count total bytes
>> allocated and freed, 2 = per-allocation origin tracking. (int)
>> parm: uvm_force_prefetch_fault_support:uint
>> parm: uvm_debug_enable_push_desc:Enable push description tracking (int)
>> parm: uvm_page_table_location:Set the location for UVM-allocated page tables. Choices are: vid,
>> sys. (charp)
>> parm: uvm_perf_access_counter_mimc_migration_enable:Whether MIMC access counters will trigger
>> migrations.Valid values: <= -1 (default policy), 0 (off), >= 1 (on) (int)
>> parm: uvm_perf_access_counter_momc_migration_enable:Whether MOMC access counters will trigger
>> migrations.Valid values: <= -1 (default policy), 0 (off), >= 1 (on) (int)
>> parm: uvm_perf_access_counter_batch_count:uint
>> parm: uvm_perf_access_counter_granularity:Size of the physical memory region tracked by each
>> counter. Valid values asof Volta: 64k, 2m, 16m, 16g (charp)
>> parm: uvm_perf_access_counter_threshold:Number of remote accesses on a region required to trigger a
>> notification.Valid values: [1, 65535] (uint)
>> parm: uvm_perf_reenable_prefetch_faults_lapse_msec:uint
>> parm: uvm_perf_fault_batch_count:uint
>> parm: uvm_perf_fault_replay_policy:uint
>> parm: uvm_perf_fault_replay_update_put_ratio:uint
>> parm: uvm_perf_fault_max_batches_per_service:uint
>> parm: uvm_perf_fault_max_throttle_per_service:uint
>> parm: uvm_perf_fault_coalesce:uint
>> parm: uvm_fault_force_sysmem:Force (1) using sysmem storage for pages that faulted. Default: 0.
>> (int)
>> parm: uvm_perf_map_remote_on_eviction:int
>> parm: uvm_channel_num_gpfifo_entries:uint
>> parm: uvm_channel_gpfifo_loc:charp
>> parm: uvm_channel_gpput_loc:charp
>> parm: uvm_channel_pushbuffer_loc:charp
>> parm: uvm_enable_debug_procfs:Enable debug procfs entries in /proc/driver/nvidia-uvm (int)
>> parm: uvm8_ats_mode:Override the default ATS (Address Translation Services) UVM mode by disabling
>> (0) or enabling (1) (int)
>> parm: uvm_driver_mode:Set the uvm kernel driver mode. Choices include: 8 (charp)
>> parm: uvm_debug_prints:Enable uvm debug prints. (int)
>> parm: uvm_enable_builtin_tests:Enable the UVM built-in tests. (This is a security risk) (int)
>>
>> # ls -l /lib/modules/4.18.0-RT/video/nvidia-uvm.ko
>> -rw-r--r-- 1 root root 1405808 Aug 15 16:49 /lib/modules/4.18.0-RT/video/nvidia-uvm.ko
>> (just installed minytes before)
>>
>> # uname -a
>> Linux solfire 4.18.0-RT #1 SMP PREEMPT Mon Aug 13 05:15:26 CEST 2018 x86_64 AMD Phenom(tm) II X6
>> 1090T Processor AuthenticAMD GNU/Linux
>> (the kernel version matches)
>>
>> # eix nvidia-cuda-toolkit
>>
>> [I] dev-util/nvidia-cuda-toolkit
>> Available versions: [M](~)6.5.14(0/6.5.14) [M](~)6.5.19-r1(0/6.5.19) [M](~)7.5.18-r2(0/7.5.18)
>> [M](~)8.0.44(0/8.0.44) [M](~)8.0.61(0/8.0.61) (~)9.0.176(0/9.0.176) (~)9.1.85(0/9.1.85)
>> (~)9.2.88(0/9.2.88) {debugger doc eclipse profiler}
>> Installed versions: 9.2.88(0/9.2.88)(06:31:32 PM 08/14/2018)(-debugger -doc -eclipse -profiler)
>> Homepage: https://developer.nvidia.com/cuda-zone
>> Description: NVIDIA CUDA Toolkit (compiler and friends)
>>
>> It becomes even more weird...

It is weird indeed... Im running on kernel 4.15.16 and I needed to disable MSI in
/etc/modprobe.d/nvidia.conf with " NVreg_EnableMSI=0" appended to the line "options nvidia ...".
Thats the main differences I see with you from the software side.

This kind of error is usually due to a failed reload (not rebooting) or because of a version
mismatch according to google, but I can't find any mismatch in the info you gave us.

Good luck

--
Corentin “Nado” Pazdera