Difference between /dev/ttyS0 and /dev/cua0

You might want to try posting your question on comp.os.linux.embedded

got

A /dev/cua0 port can be opened by multiple applications at the same time, while a /dev/ttyS0 port cannot.

Meindert

That is simply not true.

Any number of applications that can open /dev/ttyS0. I just opened /dev/ttyS0 8 times. Four are "cat >/dev/ttyS0" and four are copies of kermit:

$ ls -l /dev/ttyS0 lr-xr-xr-x 1 root root 5 Oct 28 12:40 /dev/ttyS0 -> tts/0

$ lsof | grep /dev/tts cat 15176 grante 1w CHR 4,64 604 /dev/tts/0 cat 15177 grante 1w CHR 4,64 604 /dev/tts/0 cat 15178 grante 1w CHR 4,64 604 /dev/tts/0 cat 15179 grante 1w CHR 4,64 604 /dev/tts/0 kermit 15244 grante 4u CHR 4,64 604 /dev/tts/0 kermit 15292 grante 4u CHR 4,64 604 /dev/tts/0 kermit 15340 grante 4u CHR 4,64 604 /dev/tts/0 kermit 15388 grante 4u CHR 4,64 604 /dev/tts/0

I quite regularly have /dev/ttyS0 open simultaneously by more than one application. It's usually by accident, and things don't work right -- but that's another issue.

Historically, /dev/cua* was used to "dial out" and when you opened it, it did not wait for carrier detect. /dev/ttys* was used for dial-in, and it _did_ block waiting for carrier detect when you tried to open/read it. An open of one would block if the port was already opened via the _other_ device. That was many years ago, and I'm pretty sure it doesn't work that way any more.

I'm surprised you've even found a distro that still has /dev/cua0.

...so that you could have a dialin service configured on /dev/ttyS0 without having to build in any knowledge of other applications that want to share the port for dialout. Then you configure those dialout applications to use /dev/cuaa0 without having to teach them anything about the daemons.

If you plan to "own" the serial port you should use cuaa0 to avoid the carrier-detect issues with ttyS0.

Exactly. That's the way it worked years ago in kernel 2.0 and older.

Wrong.

Use of /dev/cuaN devices has been depreciated for years (since kernel 2.1), and a decently modern distro won't even have the cua devices. The /dev/cuaN devices (major number 5) are absent in 2.4 and later.

The carrier detect issues you're talking about are in the old tty devices. That stuff was all re-written a half-decade ago. See this thread from 1988 for an excellent explination by Ted Ts'o, the main man behind the serial stuff for years:

Uh, that's obviously a typo -- it was 1998.

Oh, I didn't catch the original post that mentioned Linux. Other unices have retained the cua/tty distinction.

Ok, my turn: Wrong. Slackware 10 (kernel 2.4 I believe) still has 'em.....

Meindert

Interesting. Are they a different major number (5, IIRC) than the /dev/ttySN devices (4), or are they just alternative names for the same devices?

The /dev/cuaN devices are major 5 and the /dev/ttySN are 4.

Meindert

Weird. Do they work?

I checked a few of my systems: Deban stable (2.4.18) and unstabled (2.4.25) and Mandrake 10.0 (2.6.3). None of them had /dev/cuaN devices. When I created them (e.g. with major 5, minor 64), and open of the cua device failed with an error of "no such device".

It appears so. I could 'cat' a file to both /dev/ttyS0 and /dev/cua0

Meindert