Bug#105451: update
You have made a lot of good suggestions. I can only act myself on the ones
involving setserial. I will include some commentary on this in the setserial
documentation. This clearly should be documented better elsewhere too, and
to be honest I only worked it out by first believing that there was going to
be a limit and then hunting through the kernel source code looking for the
code that did it.
Glad to have been of help.
G.
> -----Original Message-----
> From: Andrew Young [mailto:aty@sciences.sdsu.edu]
> Sent: 31 July 2001 20:15
> To: G.Russell@napier.ac.uk
> Cc: 105451@bugs.debian.org; adam@onshore.com
> Subject: Re: Bug#105451: update
>
>
> I think this should solve the problem!
>
> ...
> > In any case mknod just creates files with minor and major
> numbers, and does
> > not ask the kernel if these are valid. The validation only
> happens when
> > the file is opened. The kernel only recognises major 4,
> with minor 64-67
> > as serial ports by default. Kernel variable CONFIG_SERIAL_MANY_PORTS
> > provides extra minor numbers (how many ports this adds I
> cannot say).
> > This may or may not be set for you in the
> > kernel build... this information is printed out at boot
> time as " MANY
> > PORTS"
> > and I gather that this is not printed when the module is
> loaded, and thus I
> > am guessing that it is not set. I am getting this from the
> 2.4.7 kernel
> > source
> > so things may be different in other kernel versions (but I
> doubt it).
>
> You are correct. It doesn't appear in any of the 2.2.19 flavors, so
> far as I can see from the kernel-config files.
>
> > The names of the ports and there numbers are really not
> that important.
> > I am hoping that you only have a few serial ports in total
> (less than 4), so
>
> Yes, there are just 3: the modem, the mouse, and a serial line my wife
> uses to log in from a glass tty in her office.
>
> > you can hopefully use ttyS3 or ttyS4 and use setserial to
> configure them to
> > what you want without having to recompile the kernel. I can
> think of no good
> > reason why MANY PORTS is not the default, though it
> probably adds a few 100
> > bytes to the kernel size. You probably chose S14 as this is
> > a SPARE tty (whatever this means) in serial.h when
> MANY_PORTS is defined.
>
> I was in fact just following the convention adopted in my old
> Slackware
> system with the ttyS14 (actually, cua14) in the rc.serial file.
>
> > Hope this helps.
>
> I think it should solve the problem! I tried the "vanilla" kernel and
> of course had the same problem with ttyS14 as before. Ordinarily I
> would be reluctant to trample on the debian port assignment,
> but if you
> say it is OK to redefine ttyS3 as my modem, I'll go ahead and try
> that.
>
> The MANY_PORTS business had actually been suggested yesterday by
> another correspondent, but without such definite information
> about what
> is going on; so I decided to try the "vanilla" kernel first as it
> seemed unlikely to me that MANY_PORTS was going to be needed, and the
> "vanilla" version is described as having lots of drivers -- with all
> its support for exotic hardware, who'd think it wouldn't support a
> nonstandard serial port? From the naming of the modules, I had
> supposed that MANY_PORTS would be required only if I
> literally had many
> ports, not just a high-numbered one; while NONSTANDARD sounded more
> likely to be required for a serial port with an odd irq and memory
> address.
>
> This whole fiasco could have been headed off if the necessary
> information had been available in places one would normally look for
> it. For example, if the MANY_PORTS business had been explained in the
> SERIAL HOWTO, or even if the information had been available in the
> Debian bug-report archives (where a search for "ttyS14" turns up
> practically nothing), I could have avoided the problem.
>
> It would also be useful if such things were explained in the
> Installation instructions -- or at least, if there were a
> pointer there
> to where the information could be found. I suppose this is too
> uncommon and subtle a problem to need coverage in the installation
> scripts themselves -- though it might be useful to say,
> either there or
> in the installation manual, that *only* serial ports 0-3 are usable
> without rebuilding the kernel. It would also be helpful if this were
> mentioned in section 5.2 of the installation manual, where "choosing a
> flavor" is discussed. The present manual makes it sound as if the
> "vanilla" flavor has just about eveything needed.
>
> Another place it would be useful to document this is in the
> /etc/serial.conf file, which is already full of useful
> comments. As it
> has dummy entries for ttyS14 and 15, I assumed I could just
> fill in the
> blanks and setserial would set up the serial port; there is no
> indication that special kernel support is needed to make ttyS14/15
> work. (Probably other entries in serial.conf should have comments
> added to indicate what kernel modules they need.)
>
> While I am suggesting clarifications, let me also suggest that the
> kernel-config files have at least a 1-line header comment to indicate
> which kernel and flavor they refer to, as I wasted a day by having
> inadvertently looked at the wrong one in chasing after a fix for my
> problem. There are a lot of very similar files and subdirectories in
> the Debian ftp tree, and it's easy to get lost (particularly if, like
> me, you are not familiar with this distribution).
>
> I think the fact that someone with 15 years' experience with
> UNIX and 5
> years' experience with earlier Linux systems (me), as well as several
> people with considerable familiarity with Debian (e.g., Adam) were
> unable to spot the problem and solve it right off shows that more
> information about this obscure point needs to be made available in the
> documentation where people would normally look, such as the SERIAL
> HOWTO and the Installation manual.
>
> Would it be reasonable to add a note about this to, say, the mknod and
> setserial man pages as well? It's all very well to say that all
> problems are ultimately resolvable by reading the source
> code; but when
> one is installing the system for the first time, it isn't obvious
> whether the problem is in the kernel, mknod or setserial, or what --
> not to mention that *most* new users don't have a clue about reading
> source code, and need a good human-readable document (preferably the
> Installation instructions; the HOWTOs are a bit obscure for non-Linux
> people).
>
> Thanks very much for all your help. I'll try this and let you know if
> just fudging ttyS3 in serial.conf does the trick.
>
> -- Andy Young
>
Reply to: