> From: "Theo de Raadt" <[email protected]> > Date: Mon, 10 Nov 2025 08:27:12 -0700 > > Mark Kettenis <[email protected]> wrote: > > > I suspect that init(8) tries to open /dev/console and fails. > > Likely, and then it does not run the first rc script.
Worse. It will sleep for a few seconds and try to open /dev/console again. At least that is what happens if /dev/console doesn't exist on the filesystem. > > > A workaround is not the way to handle this problem. > > > > Do we need a "dummy" console device that we can fall back on as a last > > resort? > > I have a theory that back in the legacy days, the vga / pckbd on isa > always responded sufficiently for a console attachment to be created. The wscons console code doesn't really care about whether we have an input device or not. But it really wants an output device. And yes, for a long time the pcengines apu machines were pretty much the only x86 devices where the legacy vga stuff didn't respond. > But in a new EFI "legacy-free" circumstance, RAMDISK may be missing > some code (because is is large) and therefore not attach a console. With the GENERIC kernel amdgpu(4) attaches and will configure a framebuffer even if the KVM isn't pluggend in. > With a console of last resort that has no input method (does it block? > or does it always indicate ready and EOF?) worries me a little bit. > We could be trading one problem for another. It should behave as if you have PC without a keyboard plugged in. And yes, that means that if the upgrade script stops and asks for input, it will just hang. Not much you can do about that I fear. > I thought when these VM's are booted with GENERIC, they have a console. This is not a VM.
