minnie.tuhs.org/unixen.tar/unixen/tools/CrossEnv/V6_Vax/README

Text File | 1998-02-25 | 7.4 KB | 199 lines

  A Package to Support VAX Compatability Mode on UNIX-32V

                      Arthur W. Wetzel
                        735 LIS Bldg
    Interdisciplinary Department of Information Science
                  University of Pittsburgh
                   Pittsburgh, Pa. 15260
                       (412)-624-5203


This is a brief description of a package to support the exe-
cution of PDP-11 programs on VAX UNIX-32V or Berkeley VMUNIX
in compatability mode.  The major functions are to

1)  allocate  a  block  of memory as the PDP-11 memory space
     (this must start at location 0),
2) read compatability mode program images  into  memory  and
     lay them out properly (with arguments etc),
3)  actually  handle  the  change  to and from compatability
     mode,
4) simulate system calls for what ever operating  system  is
     being simulated and
5) simulate floating point (FPU and FIS) instructions.

Unfortunately programs requiring separated I/D space can not
be run.  Loading of the package is  rather  slow  since  the
entire process is about 80K bytes (64K is the PDP-11 space).
Once execution begins however, the speed  is  similar  to  a
PDP-11/70.   There  is considerable overhead for each excep-
tion condition so that programs with a lot of  system  calls
or  especially  with  floating  point will be greatly slowed
down.  Note that the text segment must be writable since the
PDP-11 memory space is there.

Three  quick  changes  to  UNIX-32V and Berkeley VMUNIX were
made in the course of constructing this package.

First, it is necessary  to  patch  a  bug  in  the  original
     mchdep.c.   The  bug in the sendsig routine is that the
     condition codes are masked out of the psl before it  is
     stacked  when  catching signals.  This affects all pro-
     grams not just compatability mode ones although  is  is
     not usually a frequent problem execept in this applica-
     tion.  The mask which was 0xfff1 should be  changed  to
     0xffff.  If this is not done, the condition codes after
     a signal trap routine returns will  always  be  cleared
     which  can  result in many strange problems when condi-
     tion codes are being checked in loops or in  this  case
     after  an  "illegal  instruction"  trap.  This same bug
     remains in the Third Berkeley Software Tape version  of
     Virtual Memory UNIX.

Second,  although  it is easy to get into compatability mode
     one also needs a way to  get  back  when  an  exception









                             -2-


     condition arises.  This can be done by changing another
     mask in the last line of the same  routine.   The  0x1f
     mask  should be changed to 0x8000001f.  This clears the
     compatability mode bit so that all signals  are  neces-
     sarily  caught  in native mode where native code can do
     something about the situation.

Finally, if one wants compatibility mode  programs  to  have
     SETUID and SETGID status, there must be a way to change
     the effective uid or gid without  clobbering  the  real
     uid  or gid.  This is easily done by adding seteuid and
     setegid system calls to UNIX-32V.  My method  of  doing
     this  was  to modify setuid and getuid so that the high
     order 16 bits of the  argument  in  the  actual  system
     calls  is a flag (uids and gids are only 16 bits in the
     low order part of the word) to indicate either a  regu-
     lar  setuid or getuid function or alternately a seteuid
     and setegid function.  Appropriate functions  seteuid()
     and setegid() have been added to our libc.a which auto-
     matically set up the flags while setuid() and  setgid()
     insure that the flags are zeroed.

Most  of  the  programming was done in late August 1979 with
additions being made occasionally thru August 1980.   Compi-
lation  procedures are specified in Makefile.  An effort was
made to minimize the amount of assembly language  coding  so
that  only two small assembler routines are found here.  One
of these (memsiz.s) simply  specifies  how  much  memory  is
being  allocated  for  PDP-11  images and makes it available
through certain global variables.  The other assembler  file
(compat.s) handles the protocol for getting into compatabil-
ity mode at a certain pc and with a  certain  ps.   It  also
includes  a  getreg  function which copies machine registers
into known places.  The heart of the entire package is  run-
compat.c which is used for all RTSs (Run Time Systems).  The
function main here simply checks for the  existence  of  the
file  to be executed and sets the required uid and gid.  The
execute function actually copies the file to memory and sets
trap conditions.  Finally illtrap() catches illegal instruc-
tions and goes to the code appropriate for what is found  as
the  illegal  instruction.   The bulk of the lines of C code
are in unixtraps.c and dofloat.c which do UNIX system  calls
in  either version 6 or 7 format and simulate floating point
operations.  (Since PWB-UNIX is upward compatable with  ver-
sion  6,  the version 6 system support also includes PWB sys
calls.)  There are probably a number of bugs in the floating
point  simulation code just waiting to be found.  If you are
running programs which already include the  PDP-11  floating
point  interpretation  code, you may want to disable dofloat
as the illegal instructions can be caught and  simulated  in
the PDP-11 code.  To do this just make dofloat.o with "cc -c
-O -DNOFPSIM dofloat.c".

A shell which will automatically invoke  compatability  mode









                             -3-


programs is in the modshell directory as difference listings
from the original UNIX-32V shell.  Most of the new  code  is
in  a  new  function compat.c.  The automatic recognition of
PDP-11 UNIX version 6/7 programs relies on the fact that the
second  word  (16 bit) of a PDP-11 a.out file (text size) is
nonzero whereas it is 0 for 32V a.outs.  No easy distinction
can be made between version 6 and version 7 a.outs so that a
shell variable RTS sets up the name of a  default  Run  Time
System.  On our system version 6 a.outs have been patched so
that word 6 of the header which is  unused  is  a  1.   This
hoaky?  method  seems to work just fine.  A program v6flag.c
is in the modshell directory to do this.

One possible use of this package is  to  get  programs  like
INGRES running on the VAX without going through what appears
to be a nontrivial conversion effort.  There are two ways of
running  such  programs.  Firstly if the shell is patched to
automatically recognize and run  compatability  mode  a.outs
(as in modshell), the PDP-11 a.out files for the program can
be just put on the system with their normal names and run as
usual.   Note  however  that  you will be using the UNIX-32V
shell so that any shell files from  PDP-11  version  6  will
have  to  be  modified for this to work correctly with some-
thing like INGRES.  The second approach is to make a  direc-
tory  hierarchy somewhere which corresponds to what would be
on a PDP-11 including the appropriate PDP-11 shell.  In that
case  just execute that shell in compatability mode with the
root directory set to the top of the PDP-11 hierarchy.  This
is  the quickest way to get something going in a hurry since
no changes are required to existing  PDP-11  code  or  shell
files.

Emulation  of  RT-11 system calls provided by Dan Strick are
not being distributed at this time.

Please foreward any comments, bug fixes or  quick  questions
to the author at the above address.