Bug 624398

Summary:	x11-drivers/nvidia-drivers-384.47: "nvidia_uvm: Unknown symbol nvUvmInterfaceGetBigPageSize (err 0)"
Product:	Gentoo Linux	Reporter:	Will Simoneau <bugzilla>
Component:	Current packages	Assignee:	Jeroen Roovers (RETIRED) <jer>
Status:	RESOLVED INVALID
Severity:	normal
Priority:	Normal
Version:	unspecified
Hardware:	AMD64
OS:	Linux
Whiteboard:
Package list:		Runtime testing required:	---

Description Will Simoneau 2017-07-10 02:46:15 UTC

With x11-drivers/nvidia-drivers-384.47, the nvidia_uvm module fails to load on my system.

# modprobe nvidia_uvm
modprobe: ERROR: could not insert 'nvidia_uvm': Unknown symbol in module, or unknown parameter (see dmesg)

dmesg contains this:
nvidia_uvm: Unknown symbol nvUvmInterfaceGetBigPageSize (err 0)

The nVidia X11 driver itself seems to work (including OpenGL), but this apparently prevents things like ffmpeg's h264_nvenc encoder (and probably CUDA stuff in general) from working.

Just for completeness, ffmpeg dies with these messages:
[h264_nvenc @ 0x20d06a0] Cannot init CUDA
[h264_nvenc @ 0x20d06a0] cuCtxPushCurrent failed

My kernel is 4.4.38 plus a custom driver module I wrote for a PCIe card, which I unfortunately haven't had a chance to port to a newer kernel yet.

Comment 1 Jeroen Roovers (RETIRED) gentoo-dev

2017-07-10 05:19:19 UTC

384.47 does not use nvUvmInterfaceGetBigPageSize.
381.22 does use nvUvmInterfaceGetBigPageSize.

It's likely that your nvidia-uvm.ko belongs to the old 381.22 and that your nvidia.ko belongs to 384.47.

Needless to say, it works fine for me.

Comment 2 Will Simoneau 2017-07-10 13:10:31 UTC

Nice catch. As it turns out, the uvm USE-flag had somehow become unset on this system, leading to a stale nvidia-uvm.ko in /lib/modules. I don't know why the flag changed states, but turning it on fixes the problem.

Thanks and sorry for the noise...