eglInitialize failed

[guten-tag]

Hi Community,
I am trying to run a testing code for EGL, but it failed in initialization. Here is my testing code.

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

int main()
{
    EGLDisplay display = eglGetDisplay( EGL_DEFAULT_DISPLAY );
    if (display == EGL_NO_DISPLAY)
    {
        std::cout << "unable to open connection to local windowing system." << std::endl;
    }

    EGLint major, minor;
    if (!eglInitialize(display, &major, &minor))
    {
        std::cout << "unable to initialize EGL." << std::endl;
    }

    std::cout << "egl error code : " << eglGetError() << std::endl;

    return 0;
}

And the output is :

unable to initialize EGL.
egl error code : 12289

My platform is CentOS 7.6 without X server. Besides, I am trying to run the testing code on DCU, which is the GPU of AMD.

Let me know if you want more information.
Thanks for your help.

[Dark_Photon]

@guten-tag

I am trying to run a testing code for EGL, but it failed in initialization. Here is my testing code.

    EGLDisplay display = eglGetDisplay( EGL_DEFAULT_DISPLAY );
    if (display == EGL_NO_DISPLAY)
    {
        std::cout << "unable to open connection to local windowing system." << std::endl;
    }

And the output is :

unable to initialize EGL.

Try this instead:

#include <stdio.h>
#include <stdlib.h>
#include <X11/Xlib.h>
#include <EGL/egl.h>
...

  const char *display_str = getenv( "DISPLAY" ) ? getenv( "DISPLAY" ) : ":0.0";

  Display *display = XOpenDisplay( display_str );
  if ( !display )
  {
    printf( "ERROR: XOpenDisplay( %s ) failed.\n", display_str );
    exit( 1 );
  }

  EGLDisplay egl_display = eglGetDisplay( display );
  if ( !egl_display )
  {
    printf( "ERROR: eglGetDisplay() failed.\n" );
    exit( 1 );
  }
...

[guten-tag]

the output changed into :

No protocol specified
ERROR: XOpenDisplay( :0.0 ) failed.

[Dark_Photon]

The user you’re running as can’t connect to the X server running on the machine. Fix that first. Then retry your EGL connect test.

Once apps like glxgears and glxinfo run properly, you’re ready to retry your EGL test.

Which driver and driver version are you running? I have no idea if AMD drivers support creating EGL contexts without access to the X server, like NVIDIA does. However, if they do, there may be some tricks to ensuring that the user that you’re running as can connect to the underlying graphics driver. You can investigate this using strace. See this post where I detailed a similar investigation I did to figure out what access NVIDIA’s EGL support needed in order to create an EGL context without an X server connection.

[guten-tag]

There is no X server running on my DCU server. When I try to run glxgears or glxinfo, the output is :
Error: unable to open display.
So I try the testing code on another linux server, which is also based on CentOS. There is no GPU or DCU on it, and also without X server. After testing, I got this:
ERROR: XOpenDisplay( :0.0 ) failed.
I am trying to run glxgears of glxinfo on this linux server. There is some questions when I using EGL:

1. When there is no GPU on the server, whether EGL would use CPU or not?
2. I also run the testing code on Ubuntu system, it performs well. So I wonder is it a system problem that caused the wrong result.
   Looking forward to your reply.

[Dark_Photon]

@guten-tag

There is no X server running on my DCU server.

When I try to run glxgears or glxinfo, the output is : Error: unable to open display.

That is expected. You can’t connect to an X server if it doesn’t exist. You need an X server running. This provides the graphical display capability. You should get a similar error when running xdpyinfo.

So I try the testing code on another linux server, which is also based on CentOS.
There is no GPU or DCU on it, and also without X server.

After testing, I got this: ERROR: XOpenDisplay( :0.0 ) failed.

Again, that’s the expected result. You can’t connect to an X server if it doesn’t exist.

There is some questions when I using EGL:

1. When there is no GPU on the server, whether EGL would use CPU or not?

So you want to initialize EGL on a Linux system that both:

1. Does not have an X server running (for window display capability) and
2. Does not contain a GPU.

Obviously, you won’t be connecting to an EGL implementation in the graphics driver provided by your GPU vendor, because there isn’t one. You’ll need to use a library or environment that provides EGl and either OpenGL and/or OpenGL ES for completely CPU-based off-screen rendering purposes. Mesa3D might do this. Check with them.

[guten-tag]

Thanks for your reply.
Yesterday, I have tried to run the testing code on a personal computer, with centos 8 on it, and the glxgears runs properly. Of course the X server is well. After testing, everything is ok.
As you mentioned before, EGL is support for off-screen rendering without X server. However, when I try it again on a server without X server, it failed as the same. It seems like the device could not find a display for EGL. I have no idea about the relationship between display and X server. I wonder whether the display initializes correctly or not without X server.
By the way, I have tried this . But it still initializes failed.

[Dark_Photon]

@guten-tag

As you mentioned before, EGL is support for off-screen rendering without X server.

No. While EGL can in some some circumstances be used for off-screen rendering via OpenGL or OpenGL ES without a window system (e.g. w/o an X server running in the Linux world), its primary purpose is as an interface between OpenGL or OpenGL ES and the native window system.

EGL is very commonly used on Android mobile platforms, but can be found on PC as well.

I have no idea about the relationship between display and X server.

Think of a “display” as a connection to the windowing system. And in the X server world, a display can have multiple “screens”. These screens usually correspond to the monitors on your system.

However, some graphics drivers have the ability to provide access to a virtual display (in cases where there is no underlying windowing system) for off-screen rendering purposes only. NVIDIA’s graphics drivers for example support this.

By the way, I have tried this [NVIDIA OpenGL rendering without an X server]. But it still initializes failed.

1. Buy an NVIDIA GPU.
2. Install it and NVIDIA Graphics Drivers on your system.

And then you’ll be able to perform off-screen rendering via EGL and OpenGL or OpenGL ES, without an X server running on your system, when running directly on the machine (not in a VM).

This EGL + OpenGL / OpenGL ES rendering support without an X server is a feature of the NVIDIA graphics drivers specifically, not a feature of Linux in general.

[guten-tag]

Hi,
I have changed into a NVIDIA GPU Cluster, and the version of NVIDIA Graphics Drivers is 440.33.01. Also I have tried to run the testing code on it. When I was running the code on login node, it was well. However once I submit it into the computing node for running, it went wrong. The output is :

ERROR: XOpenDisplay( localhost:14.0 ) failed.

I have no idea what’s wrong with it. Looking forward to your reply.

[Dark_Photon]

ERROR: XOpenDisplay( localhost:14.0 ) failed suggests two things:

1. You are trying to use the EGL connect method that requires a connection to the X server, and
2. You are not directly logged into the X server / display manager on that system, but rather are running in an SSH connection to or a VM session on that system.

If you want to create an EGL connection without a connection to the X server, you don’t want to use the X server connection method:

    Display *display = XOpenDisplay();
    ...
    EGLDisplay egl_display = eglDisplay( display )

You want to use the:

    EGLDisplay egl_display = eglGetDisplay( EGL_DEFAULT_DISPLAY );

method described in this article:

• EGL Eye: OpenGL Visualization without an X Server

[Zhao-yunfei]

Hi, I’m in the same situation. Have you solved it yet?
my platform is ubuntu16.04 without X, and nvidia driver is 440.33.01 too.
i also use eglGetDisplay(EGL_DEFAULT_DISPLAY) to get a display.
i have a bad result that eglInitialize failed.
looking forward to your reply.

[Dark_Photon]

@Zhao-yunfei

Hi, I’m in the same situation. Have you solved it yet?

my platform is ubuntu16.04 without X, and nvidia driver is 440.33.01 too.

i also use eglGetDisplay(EGL_DEFAULT_DISPLAY) to get a display.

i have a bad result that eglInitialize failed.

See the thread linked below, and in particular that post.

In order to:

• create a GPU-accelerated EGL context
• via the NVIDIA graphics driver on Linux
• using the display eglGetDisplay( EGL_DEFAULT_DISPLAY ) (instead of XOpenDisplay())

the Linux usermode driver/libs need to be able to open the NVIDIA Linux driver’s kernel special files. Namely:

• /dev/nvidia0
• /dev/nvidiactl
• /dev/nvidia-modeset
• /dev/dri/card0 , and
• /dev/dri/renderD128

If your system is setup similarly to the one here:

> ls -l /dev/nvidia0 /dev/nvidiactl /dev/nvidia-modeset /dev/dri/card0 /dev/dri/renderD128
crw-rw----+ 1 root video 226,   0 Aug  5 17:34 /dev/dri/card0
crw-rw----+ 1 root video 226, 128 Aug  5 17:34 /dev/dri/renderD128
crw-rw----+ 1 root video 195, 254 Aug  5 17:34 /dev/nvidia-modeset
crw-rw----+ 1 root video 195,   0 Aug  5 17:34 /dev/nvidia0
crw-rw----+ 1 root video 195, 255 Aug  5 17:34 /dev/nvidiactl

that just involves adding your username to the video group in the /etc/group file. You may then need to log out and back in to activate this change.

OpenGL version is forced to 1.4, but card can support 4.6.0

@azuresilent, I figured out what the issue was with trying to render via an NVidia GPU without X server access, and it works great now: Basically, add the user you want to be able to have this access to the “video” group (in /etc/group). The reason is, in order for the EGL+GL init to succeed for the NVidia path, the app must be able to open kernel special files associated with the NVidia driver which are owned by group “video”. Special files such as: /dev/nvidia0, /dev/nvidiactl, /dev/nvidia-…

[Zhao-yunfei]

Thank you very much. there is no /dev/dri. that’s the reason why i failed.

[Dark_Photon]

Ok, sounds good.

If you haven’t already, run your app under strace (strace <appname> <args>), direct the output to a file, and look for refs to those devices. If you see the driver trying and failing to open one of them, that could be it.