[systemd-devel] Please help: timeout waiting for /dev/tty* console device

[Gabriel L. Somlo]

On Fri, Jan 06, 2023 at 07:15:14PM -0500, Gabriel L. Somlo wrote:
> Hi,
> 
> I'm failing to get a login prompt on the serial console of my system,
> because a few steps earlier serial-getty at .service fails due to a
> dependency on the actual tty device, which times out:
> 
> [ TIME ] Timed out waiting for device …ttyLXU0.device - /dev/ttyLXU0.
> [DEPEND] Dependency failed for seri…ice - Serial Getty on ttyLXU0.
> 
> This eventually results in "Failed to start systemd-logind.service",
> and no login prompt on the serial console.
> 
> I'm trying get a riscv64 port of fedora (systemd version 251) running
> on a system that can be considered "exotic" and rather on the slow
> side -- it's an FPGA soft-cpu system using the RocketChip running at
> 50MHz.
> 
> I got it working successfully on an older version of the fedora-riscv
> port, f33, using systemd 246.
> 
> I can also get it working with the current systemd on a 4-core Rocket
> chip deployed on a large enough FPGA.
> 
> It (udev?) times out on the single-core version of the design, and I'm
> wondering if there are any available options to get systemd and/or
> udev to be more "patient".
> 
> I tried booting with `udev.children_max=1` and `udev.event_timeout=800`
> on the kernel command line.
> 
> I also tried modifying systemd-udev-settle.service like so:
> 	TimeoutSec=720
> 	ExecStart=udevadm settle --timeout=680
> 
> None of the above seem to help on the single-core 50MHz rocket-chip
> system (and are not needed on the 4-core version).
> 
> Any other tricks I can use to force it to wait (or to otherwise
> encourage it to find, sooner) /dev/ttyLXU0?
> 
> I can't log into this machine to run any tests. It does manage connect
> to the network, start NetworkManager.service and sshd.service, but any
> attempt to ssh in over the network fails (after successful auth) with
> "conection closed by <ip address> port 22". Not sure if due to the
> user login service having failed, or some other unrelated timeout.

Turns out it was just really slow loading bash :) After a while, I did
manage to log in and obtain a root shell over ssh. From there I did:

	udevadm trigger --type=devices --action=add
	udevadm trigger --type=subsystems --action=add

which had no visible effect. After that, I did:

	systemctl restart serial-getty at ttyLXU0.service

which resulted in a login prompt showing up on the serial console!

I *can* run any tests y'all might suggest to further debug the state
of the system. But at this point I really do believe there is (or
should be :) a way to extend the timeout during initial boot to force
the system to wait for /dev/ttyLXU0 to become available (via udev?).

Thanks much,
--Gabriel