eglGetDisplay(EGL_DEFAULT_DISPLAY) returns EGL_NO_DISPLAY

[Thomas Foster]

I am using EGL to obtain offscreen rendering contexts, but upgrading this package today broke my application. I have written a small test case, that I expect to print "PASS":

```
#include <EGL/egl.h>
#include <iostream>

int main(int argc, char **argv) {
    EGLDisplay eglDpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);
    std::cout << ((eglDpy == EGL_NO_DISPLAY) ? "FAIL" : "PASS") << std::endl;
}
```

Build with e.g. `g++ egl_test.cpp -lEGL`. Using nvidia-375 (375.39-0ubuntu0.16.04.1), this test program will print "FAIL", because `eglGetDisplay(EGL_DEFAULT_DISPLAY)` returns `EGL_NO_DISPLAY`. If I install the 375.39 driver using Nvidia's installer, the test passes, so I think it is not an upstream issue.

ProblemType: Bug
DistroRelease: Ubuntu 16.04
Package: nvidia-375 375.39-0ubuntu0.16.04.1
ProcVersionSignature: Ubuntu 4.4.0-67.88-generic 4.4.49
Uname: Linux 4.4.0-67-generic x86_64
NonfreeKernelModules: nvidia_uvm nvidia_drm nvidia_modeset nvidia
ApportVersion: 2.20.1-0ubuntu2.5
Architecture: amd64
CurrentDesktop: MATE
Date: Tue Mar 21 12:41:31 2017
EcryptfsInUse: Yes
InstallationDate: Installed on 2016-10-11 (160 days ago)
InstallationMedia: Xubuntu 16.04.1 LTS "Xenial Xerus" - Release amd64 (20160719)
SourcePackage: nvidia-graphics-drivers-375
UpgradeStatus: No upgrade log present (probably fresh install)

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[helltone (gafunchal)]

EGL seems to be completely broken in both the current version of the nvidia-367 package (which installs 375.39) on Ubuntu 16.04, and also the nvidia-375 package on Ubuntu 16.10. I can't get past the initialisation using eglGetDisplay, same as show in the testcase in the bug description.

[Dario Limongi (limee)]

Seems the EGL linking is broken, libEGL.so.1 doesn't link to anything (in both /usr/lib/nvidia-375 and /usr/lib32/nvidia-375).

These fixed my machine and could open an EGL display:

cd /usr/lib/nvidia-375
sudo ln -sf libEGL.so.1 libEGL.so.375.39

cd /usr/lib32/nvidia-375
sudo ln -sf libEGL.so.1 libEGL.so.375.39

[Adrien Beau (adrienbeau)]

I came to this issue from another route: ever since installing package nvidia-375, I have the following warnings every time APT runs the libc-bin trigger:

Processing triggers for libc-bin (2.23-0ubuntu7) ...
/sbin/ldconfig.real: /usr/lib/nvidia-375/libEGL.so.1 is not a symbolic link
/sbin/ldconfig.real: /usr/lib32/nvidia-375/libEGL.so.1 is not a symbolic link

The warning is correct:

$ ls -l /usr/lib/nvidia-375/libEGL.so*
lrwxrwxrwx 1 root root 11 mars 17 11:31 /usr/lib/nvidia-375/libEGL.so -> libEGL.so.1
-rw-r--r-- 1 root root 77264 févr. 1 04:35 /usr/lib/nvidia-375/libEGL.so.1
-rw-r--r-- 1 root root 20608 févr. 1 04:35 /usr/lib/nvidia-375/libEGL.so.375.39

$ ls -l /usr/lib32/nvidia-375/libEGL.so*
lrwxrwxrwx 1 root root 11 mars 17 11:30 /usr/lib32/nvidia-375/libEGL.so -> libEGL.so.1
-rw-r--r-- 1 root root 71688 févr. 1 04:15 /usr/lib32/nvidia-375/libEGL.so.1
-rw-r--r-- 1 root root 15940 févr. 1 04:15 /usr/lib32/nvidia-375/libEGL.so.375.39

All the files appear to be valid ELF binaries, but with significant differences in size. I have no idea which one to remove and replace by a symlink.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Cody Planteen (planteen)]

I have the same problem. Attempting to link against /usr/lib/nvidia-375/libEGL.so.375.39, /usr/lib/x86_64-linux-gnu/mesa-egl/libEGL.so.1.0.0, and /usr/lib/nvidia-375/libEGL_nvidia.so.375.39 instead of -lEGL does not fix it. It does not fix it as in eglGetDisplay() still returns EGL_NO_DISPLAY.

/usr/lib/nvidia-375$ ls -l *EGL*
lrwxrwxrwx 1 root root 23 Mar 17 04:31 libEGL_nvidia.so.0 -> libEGL_nvidia.so.375.39
-rw-r--r-- 1 root root 1079216 Jan 31 22:06 libEGL_nvidia.so.375.39
lrwxrwxrwx 1 root root 11 Mar 17 04:31 libEGL.so -> libEGL.so.1
-rw-r--r-- 1 root root 77264 Jan 31 20:35 libEGL.so.1
-rw-r--r-- 1 root root 20608 Jan 31 20:35 libEGL.so.375.39

It appears that libEGL.so.1 itself is an ELF:
/usr/lib/nvidia-375$ file libEGL.so.1
libEGL.so.1: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, stripped

[Thomas Foster (thomasvfoster)]

My understanding is that libEGL.so.1 is GLVND-based, where libEGL.375.39 is
non-GLVND. I think both of them are at some level wrappers around
libEGL_nvidia.so, which is much larger than either of the libEGL.so and I guess
contains the meat of Nvidia's EGL implementation.

The GLVND one looks up the EGL implementations present by inspecting the files
in /usr/share/glvnd/egl_vendor.d/. A similar mechanism declares Vulkan
implementations with JSON files in /usr/share/vulkan/icd.d/, and, sure enough,
the nvidia-375 package places nvidia_icd.json in /usr/share/vulkan/icd.d/.

However, nvidia-375 doesn't put anything in /usr/share/glvnd/egl_vendor.d/, so
libEGL.so.1 doesn't find out about libEGL_nvidia.so and can't forward to it.

If I create /usr/share/glvnd/egl_vendor.d/ and copy in 10_nvidia.json from
Nvidia's runfile installer, my test case passes.

[Thomas Foster (thomasvfoster)]

This suggests a change to the source package to copy 10_nvidia.json from the upstream package to /usr/share/glvnd/egl_vendor.d/. I think that libEGL.so.375.39 is vestigial and could be omitted, but that's optional; it doesn't seem to be doing any harm.

[Thomas Foster (thomasvfoster)]

Pull request here: https://github.com/tseliot/nvidia-graphics-drivers/pull/9

[Dario Limongi (limee)]

Beware my fix comment had a mistake, the linking is the other way round
(you'll have to reinstall nvidia if you already did the wrong linking: sudo apt-get reinstall nvidia-375)

Fixed linking:

cd /usr/lib/nvidia-375
sudo ln -sf libEGL.so.375.39 libEGL.so.1

cd /usr/lib32/nvidia-375
sudo ln -sf libEGL.so.375.39 libEGL.so.1

[Launchpad Janitor (janitor)]

This bug was fixed in the package nvidia-graphics-drivers-375 - 375.39-0ubuntu3

---------------
nvidia-graphics-drivers-375 (375.39-0ubuntu3) zesty; urgency=medium

  [ Thomas Foster ]
  * debian/nvidia-375.install:
    - install glvnd EGL vendor configuration file (LP: #1674677)

 -- Alberto Milone <email address hidden> Wed, 05 Apr 2017 10:14:48 +0200