Blank screen on boot?

Hi - firstly, I need to apologize for posting here; I have tried in several places from different angles, and most recently I was told to go here, maybe because they got tired of me.

I’m building a net-boot image of Debian 11, which needs to start up without a graphical desktop (it is part of a compute cluster). Some of the systems will have an NVIDIA GPU from the cheaper end of the spectrum, ie. the kind that are often used for gaming PCs, which seem to have enough compute power for our purposes for now. The systems boot up well enough, but when I disable the nouveau driver, I lose the text consoles, which is a nuisance; although the systems can be accessed via ssh, there are situations when looking at the console is necessary.

As I understand it, what goes wrong is that the framebuffer driver depends on nouveau, which seems absurd, but there you are. So, does anybody here know how to fix this problem, so I can boot up with a GPU, without nouveau, and still have the text consoles?

The text console should not depend on the nouveau driver, it should initially be handled by efifb (or vga/vesa console in case of csm boot).

Hmm, from what I find in google it looks like efifb is for MacBooks(?) - does it work on Intel/AMD based computers too?

Eversince uefi is used on x86 hardware, >10 years, it’s the standard output method.
When booting, sudo dmesg |grep efifb should output something like
pci 0000:00:02.0: BAR 2: assigned to efifb
This tells about the boot gpu used.

What version of the kernel is your image using? There is a range of kernels that contain a bug where the boot console will be disabled whenever any driver calls the function to remove conflicting framebuffers, even if the boot console does not conflict with the driver making the call. So disabling nouveau and then loading any other DRM driver or the vfio-pci driver will disable the boot console.

That bug is fixed in newer kernels so upgrading might be worth a try.

That sounds strangely plausible - I’m on 5.10.0-22-amd64. I know there are ver 6 kernels, but I have to mount a lustre filesystem, and I’ve had a lot of trouble trying to build the client on ver 6.

I’ll look into what options I have here - do you have a pointer to that bug, like a name or so?

I’ll look into what options I have here - do you have a pointer to that bug, like a name or so?

It was this one: 216303 – Commit ee7a69aa38d87a3bbced7b8245c732c05ed0c6ec broke legacy frame buffer with NVIDIA

If you have the NVIDIA driver installed, this bug is what prompted adding the fbdev=1 feature to nvidia-drm. It’s still experimental so it can still be a bit rough around the edges but if upgrading the kernel will be tough, you could try enabling that.