drm.ko missing for CUDA 6.5 / Ubuntu 14.04 / AWS EC2 GPU instance g2.2xlarge

Question

To install CUDA 6.5 on Ubuntu 14.04.1 LTS on AWS EC2 g2.2xlarge instance, whether I install via the .deb file or .run file

.sudo ./cuda_6.5.14_linux_64.run --kernel-source-path=/usr/src/linux-headers-3.13.0-34-generic

I always get the same error about a missing drm.ko. The code compilation seems successful. Below was the log. (I rebooted before installing)

Kernel module compilation complete.

Unable to determine if Secure Boot is enabled: No such file or directory

Kernel module load error: No such file or directory

Kernel messages:

[ 3.595939] type=1400 audit(1408809902.911:5): apparmor="STATUS"

operation="profile_replace" profile="unconfined"

name="/usr/lib/NetworkManager/nm-dhcp-client.action" pid=492

comm="apparmor_parser"

[ 3.595942] type=1400 audit(1408809902.911:6): apparmor="STATUS"

operation="profile_replace" profile="unconfined"

name="/usr/lib/connman/scripts/dhclient-script" pid=492

comm="apparmor_parser"

[ 3.596140] type=1400 audit(1408809902.915:7): apparmor="STATUS"

operation="profile_replace" profile="unconfined"

operation="profile_replace" profile="unconfined"

name="/usr/lib/connman/scripts/dhclient-script" pid=492

comm="apparmor_parser"

[ 4.696067] init: failsafe main process (833) killed by TERM signal

[ 4.793261] type=1400 audit(1408809904.107:8): apparmor="STATUS"

operation="profile_replace" profile="unconfined" name="/sbin/dhclient"

pid=952 comm="apparmor_parser"

[ 4.793267] type=1400 audit(1408809904.107:9): apparmor="STATUS"

operation="profile_replace" profile="unconfined"

name="/usr/lib/NetworkManager/nm-dhcp-client.action" pid=952

comm="apparmor_parser"

[ 5.036249] init: plymouth-upstart-bridge main process ended, respawning

[ 6.589233] init: udev-fallback-graphics main process (1203) terminated

with status 1

[ 136.367014] nvidia: module license 'NVIDIA' taints kernel.

[ 136.367019] Disabling lock debugging due to kernel taint

[ 136.370281] nvidia: module verification failed: signature and/or

required key missing - tainting kernel

[ 136.370383] nvidia: Unknown symbol drm_open (err 0)

[ 136.370393] nvidia: Unknown symbol drm_poll (err 0)

[ 136.370404] nvidia: Unknown symbol drm_pci_init (err 0)

[ 136.370449] nvidia: Unknown symbol drm_gem_prime_handle_to_fd (err 0)

[ 136.370462] nvidia: Unknown symbol drm_gem_private_object_init (err 0)

[ 136.370474] nvidia: Unknown symbol drm_gem_mmap (err 0)

[ 136.370478] nvidia: Unknown symbol drm_ioctl (err 0)

[ 136.370486] nvidia: Unknown symbol drm_gem_object_free (err 0)

[ 136.370496] nvidia: Unknown symbol drm_read (err 0)

[ 136.370509] nvidia: Unknown symbol drm_gem_handle_create (err 0)

[ 136.370515] nvidia: Unknown symbol drm_prime_pages_to_sg (err 0)

[ 136.370550] nvidia: Unknown symbol drm_pci_exit (err 0)

[ 136.370563] nvidia: Unknown symbol drm_release (err 0)

[ 136.370565] nvidia: Unknown symbol drm_gem_prime_export (err 0)

The driver installation is unable to locate the kernel source. Please make sure that the kernel source packages are installed and set up correctly.

Answer 1

The error was caused by missing drm module required by NVIDIA driver. By default, Ubuntu AMI installs minimal generic Linux kernel(linux-image-virtual), which doesn't include drm module. To fix it, install the complete generic kernel linux-image-generic. Installing linux-image-extra-virtual would work as it is merely a transitional package to linux-image-generic. I would suggest install linux-generic to include both headers and image. To summarize:

sudo apt-get install linux-generic

There is similar question asked on AWS forum

Answer 2

Actually right after the fresh launch of the GPU instance, apt-get upgrade wanted to keep back 4 packages as linux-virtual, linux-image-virtual. I still installed them so that I got strictly nothing more to upgrade. (The fresh setup doesn't have previous nvidia or any nouveau drivers.)

The thing is that linux-image-virtual is a lean build with no drm.ko. Just do

sudo apt-get install linux-image-extra-virtual

which contains drm.ko.

Then go on installing CUDA with either the .deb or .run file.