[Top] [All Lists]

Re: Syscalls in NetBSD?

To: (Per Fogelstrom)
Subject: Re: Syscalls in NetBSD?
From: Systemkennung Linux <>
Date: Mon, 16 Oct 1995 10:08:33 +0100 (MET)
In-reply-to: <> from "Per Fogelstrom" at Oct 16, 95 09:15:14 am
Hi Per,

I'm CC-ing this to linux-mips.

> > as you may already have read on the MIPS mailinglist I've changed
> > Linux/MIPS' system call interface to work like the one of Risc/OS
> > and IRIX.  This give a nice way to be more compatible than the
> > ABI demands ...
> > 
> Yes, i saw that. Seems resonable. Only one thought: Isn't irix and
> risc/os big endian systems? Considering that, would it not be nicer
> to have Linux and NetBSD to be compatible on the system call level
> so they could share binaries without compat interfaces?

The big deal that I want is that I can even use libraries like
from an SGI.  I think this is usefull to make running comercial
software useable on Linux/MIPS.

Implementing yet another syscall interface for the NetBSD syscalls would
be nice.  Unluckily the syscall numbers are not identical :-(  Even
worse - I cloned the behaviour of IRIX to return an error status from
a syscall in a3.  I think we now still have the time to redesign the
basic interfaces of our two OSes to make compatibility without kludges
possible.  I think of renumbering the syscalls the following way:

     0 - 999  SVR4 flavoured calls a la Risc/OS, IRIX
  1000 - 1999 SYSV flavoured calls a la Risc/OS, IRIX
  2000 - 2999 BSD flavoured calls a la Risc/OS, IRIX
  3000 - 3999 POSIX flavoured calls a la Risc/OS, IRIX
  4000 - 4999 Linux flavoured calls
  5000 - 5999 NetBSD flavoured calls

For compatibility you could use the number range from 0 - 999 to
mean the same thing as 5000 - 5999 for quite some time.  This might
even be a good solution as I found no more syscall in the 0 - 999
number range in the IRIX 5.3 header files (/usr/include/sys.s).

Note that I could move the Linux syscall numbers again with no 
problem; It's just that I try to avoid the known to be in use
range from 0 - 3999.  I'm also interested in Ultrix in that direction.
Oh, and didn't Sony have some BSD flavour for it's News systems?

About the byte order - I try to make the design decisions with
respect to both byte orders;  there is interest in running Linux/MIPS
in big endian mode on Magnum and Olivetti systems.  Oh, and there
are some wired ideas like bi-endian OSes.  Mips Computer Systems, Inc.
tried to write it but by that time Mips was bought by SGI and so
the project was cancelled.

> > What I'm now interested in is some info about how NetBSD does
> > system calls:
> > 
> >   - What are the calling conventions (parameters/return values/
> >     errors)?
> >   - Which number range is being used for the syscalls? (Background:
> >     Risc/OS and Irix reserve numbers 2000-2999 for BSD.  Would be
> >     nice if NetBSD would do the same.
> > 
> Calling convention is the same as in a function call, A0 = syscall
> number, A1 = arg 1 and so on. Return value in v0 and perhaps v1.
> Syscall numbers are in the low range from zero and upwards.
> Syscall emulation interfaces are implemented. For the Mips systems
> one exist for, ouch, ultrix. Well, nice in a way, but NetBSD is
> not bug compatible with ultrix so expect programs to work better :-)

Or to work at all ...  I've written some not so nice things about Risc/OS
in comp.sys.mips but what I've heared till now about Ultrix beats every-
thing by an order of magnitude ...


<Prev in Thread] Current Thread [Next in Thread>
  • Re: Syscalls in NetBSD?, Systemkennung Linux <=