This will turn a text-terminal (or emulator PC) into a "serial console". Many messages from the system are normally only sent to the console (the PC monitor). Some of the messages sent to the console at boot-time may also be seen on any terminal after the boot succeeds by typing the command: dmesg. If the boot failed this will not be of any use since the terminal can't work without an operating system. It's possible to modify the Linux kernel so as to make a terminal serve as the console and receive all the messages from Linux intended for the console. Unfortunately, the messages from the BIOS (which display on the monitor when a PC is first started) will not display on this terminal (but still display on the monitor).
There's a Remote-Serial-Console-HOWTO on this topic. Some people use a serial console when they run a PC without a monitor or keyboard. Normally, a PC will not start up without a keyboard and video card but some BIOSs permit it. If you are lucky enough to have such a BIOS that supports "console re-direct" you will likely then need to setup the BIOS using the CMOS menus when you start your PC.
The method of creating a "serial console" depends on your kernel version. In any case serial support needs to be compiled into the kernel and not supplied as a module.
The instructions for creating a serial-console are included with the kernel documentation in: Documentation/serial-console.txt. Kernel 2.4+ has better documentation and mentions the need to run getty on the serial port. Normally, the device /dev/console is linked to tty0 (the PC console). For a serial-console you create a new /dev/console by
mknod -m 622 console c 5 1
Here is an example /etc/lilo.conf file contents (for ttyS0):
prompt
timeout=50
boot = /dev/sda
vga = normal # force sane state
install = /boot/boot.b
message = /boot/message
image = /vmlinuz
root = /dev/sda1
label = console
serial = 0,9600n8
append = "console=ttyS0"
The same PC may have more than one serial console but the last console mentioned in the "append" statement becomes the main console that you interact with ?? See the kernel docs (but it's not too clear).
The Linux Journal in April 1997 had an article on patching the Linux kernel. You add a couple of #defines at the start of src/linux/drivers/char/console.c:
#define CONFIG_SERIAL_ECHO
#define SERIAL_ECHO_PORT 0x2f8 /* Serial port address */
The following was not in the Linux Journal article.
In kernel 2.+ (and earlier ??) you need to also set the baud
rate (unless 9600 is OK). Find these 2 lines:
serial_echo_outb(0x00, UART_DLM); /* 9600 baud */
serial_echo_outb(0x0c, UART_DLL);
Change 0x0c in the line above (depending on the baud rate you want):
115200 baud: 0x01 19200 baud: 0x06 2400 baud: 0x30
57600 baud: 0x02 9600 baud: 0x0c 1200 baud: 0x60
38400 baud: 0x03 4800 baud: 0x18
If you currently use the console to select which operating system to boot (using LILO), but would like to do this from a terminal, then you need to add a line to the /etc/lilo.conf file. See the manual page for lilo.conf and search for "serial=".
One way to do this is to just use a terminal (serial console) for the monitor See Make a Serial Terminal the Console. You may still need a video card since many BIOSs require one to get the PC started. Your BIOS may also require a keyboard to get started or it may have an option where you can set the BIOS not to require a keyboard.
If you boot without a monitor, make sure that the boot loader (such as LILO or GRUB) doesn't try to present any image on the screen. A text-terminal can't display it and the booting may hang.
If you have a keyboardless terminal, it can also be used as a monitor by use of the ttysnoop program. As of yet, it doesn't work like a console since it doesn't get all the initial boot-time messages. See Use a Keyboardless Terminal as the Monitor
While you might think that a terminal without a keyboard is
useless it is possible to use it as the monitor and type on the PC's
own keyboard. This may be done by using the spy program
ttysnoop. This program permits a person at one terminal to spy
on (or snoop) what someone else is typing at another terminal (or the
PC console-monitor).
Now suppose you are typing away at the monitor tty1. Imagine that someone with a terminal on ttyS2 is spying on you (with ttysnoop) and has a screen that looks just like your screen. Everything you type at tty1 also displays on ttyS2. Now move the spy terminal (on ttyS2) so it is side-by-side with your monitor (on tty1). Everything you type on the PC keyboard will now display on the two screens in front of you: the monitor and the spy terminal. Now just look only at the spy terminal as you type. Disconnect both the monitor and the terminal keyboard since you're not using either. Thus you are now using a keyboardless terminal as a monitor. What a simple but clever solution!
One possible problem (which turns out to be no problem at all) is that normally, the type of spy terminal should be the same as the type of terminal being spied upon. Since the monitor is normally declared as a terminal of type "linux" (which is close to vt100), you might think that the real terminal should also be (or at least emulate) a vt100. Not necessarily so! For example, if you have a wyse60 terminal you simply falsely declare that you have a wyse terminal on tty1 (tell the getty program for tty1 that it's a wyse60).
Here's why you can get away with this. Go back the the scenario where you have both the monitor and the wyse60 terminal in front of you, both displaying the same thing (or trying to). Ttysnoop will be sending the wyse60 the same bytes as are going to the monitor. If you've falsely claimed that the monitor is a wyse60, then you'll have wyse60 escape sequences going to both the monitor and the wyse60 terminal (via ttysnoops). The display on the wyse60 is fine but the display on the monitor is corrupted since it's getting wyse60 escape sequences. Since you will not use the monitor (and are about to disconnect it) this corruption is never a problem (you simply don't see it).
This is not the ideal setup since ttysnoop runs so late that the login prompt doesn't appear. This example is for a wyse60 terminal on ttyS2 and the missing monitor/PC-keyboard on tty1. It uses the agetty program for getty as supplied by the Debian distribution. Your getty may require a different format.
In /etc/inittab:
1:2345:respawn:/sbin/getty 38400 tty1 wyse60 -ln /usr/sbin/ttysnoops
Note that ttysnoop/ttysnoops is a client-server combo so the
"s" is not a typo. If you don't use -n you may not see a login prompt
on the terminal. With no -n, agetty issues the prompt before the
ttysnoop is activated. With -n, agetty doesn't issue the prompt (but
login does instead). If you use -n, agetty will no longer
automatically detect parity but if you don't use parity all is OK.
In /etc/snooptab:
# tty snoopdev type execpgm
tty1 /dev/ttyS3 init /usr/local/bin/sterm
/usr/local/bin/sterm:
#!/bin/sh
stty rows 24
/bin/login $@