stuart@hcr.UUCP (Stuart Thompson) (06/27/88)
The C-shell has some VAX machine dependent code which makes some very basic
assumptions about the address space of the VAX machine to control memory
allocation/dealloction. Due to the way the C-shell mixes pointers to data
from both the data segment (static data) and the break area (dynamic data),
a special function called xfree() is used to determine which data area on
the stack a pointer refers to before memory is deallocated (using free()).
In the figure below, the data segment, which includes both the "bss" and
"data" areas of the stack has a lower address space than that of the break
area (shown as "sbrk area" the figure). The xfree() function does the
following:
xfree(cp)
char *cp;
{
extern int end; /* end of data segment */
if (cp >= end && cp < (char *)&cp)
free();
}
The if statement in xfree() checks that:
1) cp points above the data segment, i.e. not in the data segment
2) the address for the actual data for cp is below the current
stack frame, i.e. is not local in the current stack frame
VAX STACK
hi addr _________________
| |
| | <- pointer to data (&cp)
|===============| current stack frame
| |
|---------------|
| |
| sbrk |
| area |
| | <- address of data (cp)
| |
|---------------| (end)
| bss |
|---------------|
| data |
lo addr -----------------
It should be noted that the if statement in xfree() will only work on machines
with a stack structure similar to that of the VAX. The onlyread() function
in the C-shell also contains the same sort of machine dependent assumptions.
For many other machine architectures whose address space does not correspond
to that of a VAX, porting the C-shell requires making interesting workarounds.
One such implementation is in the VX/VE UNIX emulation on CDC CYBER machines.
Segments on CYBERs are quite different than on a VAX, the break area resides
in an extensible segment which has no fixed relation to either the stack segment
or data segments. In this case, we must modify xfree() and onlyread() to
compare segment numbers between pointers rather than relative stack addresses.
A truly portable C-shell should use non-machine dependent mechanisms to
distinguish between pointers to dynamic data and pointers to static data.
_____
Stuart E. Thompson
Just another techie
{utzoo,utcsri}!hcr!stuartboyd@basser.oz (Boyd Roberts) (07/08/88)
Those coders who believe all the world is a VAX are just wrong-wrong-wrong. There's a machine in the UK called an Orion made by some people called High Level Hardware. It has TWO (count them) stacks and is word (32 bit) addressed. Miss-aligned byte references don't fault and the byte offset is in the HIGH order address bits. So you get a byte address space that looks something like this: 0x00000000 0x10000000 0x20000000 0x30000000 0x00000001 0x10000001 0x20000001 0x30000001 etc... But then the two stacks come in. A high order bit (in fact two) select which stack you're referencing. Either, the ``scalar'' or the ``vector'' stack. The address space is NOT contiguous. Stick to K&R and you're fine. Somehow High Level Hardware ported 4.2BSD to their box. They must have been crazy!! I used it & it worked. Hats off to HLH, but their hardware people should've been shot. Worse still I think I heard rumours about them releasing NFS on it. Now if XDR ain't VAX dependant, I don't know what is. SysV NFS 2.0 XDR has some real cuties for non VAX byte order machines. Puke! One Sunday afternoon & a lot of printf's later I found it. Assuming how unions/structures are packed & passed to function calls as their constituent elements is NO way to write a ``portable'' piece of code. Boyd Roberts boyd@basser.cs.su.oz boyd@necisa.necisa.oz ``When the going gets wierd, the weird turn pro...''
chris@softway.oz (Chris Maltby) (07/08/88)
Yet another reason not to use 'csh'... $ whatis sh csh whatis /bin/sh unset csh builtin whatis $ -- Chris Maltby - Softway Pty Ltd (chris@softway.oz) PHONE: +61-2-698-2322 UUCP: uunet!softway.oz!chris FAX: +61-2-699-9174 INTERNET: chris@softway.oz.au
drb@praxis.co.uk (David Brownbridge) (07/12/88)
In article <1296@basser.oz> boyd@basser.oz (Boyd Roberts) writes: >Those coders who believe all the world is a VAX are just >wrong-wrong-wrong. yes Yes YES! >Somehow High Level Hardware ported 4.2BSD to their box. ... >... I think I heard rumours about them releasing NFS on it. I worked on the NFS port when I was at HLH. The "portable" NFS sources were certainly written with VAXen in mind. Imagine my Joy(!) at finding a simple piece of pointer arithmetic broken by gratuitous casting: some_routine() /* located in an NFS kernel near you */ { char *a, *b, *c; /* ... code setting b,c to point to parts of an array ... */ a = (int) b - (int) c; } Because Orion integer arithmetic is different from pointer arithmetic the answer delivered in "a" is garbage. Remove the casts and all is ok. As a historical note, the Orion you mention is now superceded by the Clipper-based Orion 1/05 and HLH probably posses the most portable set of BSD/NFS sources in the known universe! David Brownbridge +44-225-444700 drb%praxis@ukc.ac.uk Praxis Systems plc 20 Manvers St Bath Avon UK BA1 1PX
maart@cs.vu.nl (Maarten Litmaath) (07/16/88)
In article <2629@newton.praxis.co.uk> drb%praxis@ukc.ac.uk (David Brownbridge) writes:
\ char *a, *b, *c;
\
\ a = (int) b - (int) c;
That should be:
int a;
char *b, *c;
Right?
--
I'd rather live in Russia |Maarten Litmaath @ Free U Amsterdam:
than in South-Africa... |maart@cs.vu.nl, mcvax!botter!ark!maart