arnold@gatech.UUCP (06/10/83)
There is something that I've always wondered about on
pdp11 unix systems. What exactly is the difference between the
-i and -n loader options? The manual page is about as clear
as the New York Skyline is even.
I know that they both separate your program into
separate I and D space, and that -i has to be used when a program
gets too big for -n. I don't understand how or why tho.
Thanks in advance,
Arnold Robbins
--
"The parchment and quill pen of Arnold Robbins"
Arnold @ GATech (CS Net)
Arnold.GATech @ UDel-Relay (ARPA) School of ICS
...!{sb1, allegra}!gatech!arnold (uucp) Georgia Tech
...!duke!mcnc!msdc!gatech!arnold Atlanta, GA 30332
...!decvax!cornell!allegra!gatech!arnoldlwa%mit-csr@sri-unix.UUCP (06/11/83)
The -i and -n options to the pdp-11 loader actually do different things. -n produces a load image with separate "text" and "data" segments, such that the text segment is read-only and can be shared. The data segment is relocated to begin at the next 4K word boundary above the end of the text segment (_etext) so that it will mapped by a different mapping register - this allows the text and data segments to be non-contiguous in physical memory. The -i option goes further - it produces a program linked to run with separated instruction and data (I and D) spaces. In this case both the text segment and the data segment are relocated to begin at virtual address 0; the processor determines from the type of reference (instruction fetch or not) whether the address is to be mapped using the I-space mapping registers or the D-space mapping registers. Programs linked with -n can be run on any pdp-11; programs linked with -i can be larger, but will only run on 11/44's, 45's, 50's, 55's, and 70's. -Larry Allen -------
edhall%rand-unix@sri-unix.UUCP (06/12/83)
Actually, the data segment in separate I/D programs begins at location 2. This is to confirm to K&R's dictum that ``it is guaranteed that assignment of the constant 0 to a pointer will produce a null pointer distinguishable from a pointer to any object''. A while back there was a drawn-out discussion of de-referencing null pointers, and I believe this fact was mentioned in passing. A look at the source to ld will confirm it. -Ed Hall edhall@rand-unix ucbvax!trw-unix!randvax!edhall
jcp%brl-bmd@sri-unix.UUCP (06/13/83)
From: J. C. Pistritto <jcp@brl-bmd> The -n option on the PDP-11 loader is often referred to as 'write-protected text' loading. It loads your program in to one (1) 16 bit address space, and results in a program that runs with both its instruction and data memory management registers, (on machines that have both, pointing to the same physical region of memory, just that the pages which refer to 'code' are write-protected. A small side effect of this is that since data follows code, the first data address in your program will be a multiple of 8192 (bytes), as the data is adjusted upward in memory to the beginning of the next memory managment register address space, (there are only 8 for the entire 64kbytes possible). This means you could actually have a program which will load and run correctly without -n that will NOT with -n. I have actually seen this happen, and it means your program is\ getting close the hairy edge. One note: a '-n' loaded program may be run on ANY PDP-11 with memory management, (ie. runs UNIX). This is NOT true of a -i loaded program. The -i option directs the loader to load code and data into two PHSYCIALLY DISTINCT regions of memory, which each have their own set of memory managment registers when the program is running. This also means that this will NOT work on any machine that does not support the Split I&D register sets. (Ex. PDP-11/23s, PDP-11/34s, PDP-11/60??). Also, it means you get an entire 64Kbytes for code and 64Kbytes, (minus the stack), for data. Also, there are certain functions which may not be used in a split I&D program, (on version 6, NARGS was an example of this). Of course, it is not possible to write self-modifying code when either -n or -i are specified, (in fact, this is often the purpose of specifing them...) -JCP-
guy@rlgvax.UUCP (06/13/83)
Actually, the PDP-11 "crt0" C startup routine puts in a shim at location
0 to insure that real data starts at location 2; if you don't link with
it, you can end up with real data at location 0. Anybody out there
planning an implementation of UNIX should ENSURE, using this or any
other trick, that no code (other than the start-up code) or data appear
at location 0 (one implementation did that, and it gave some of our
programs gastric distress; fortunately, they fixed that bug early on).
Guy Harris
RLG Corporation
{seismo,mcnc,we13,brl-bmd,allegra}!rlgvax!guytom@rlgvax.UUCP (06/14/83)
Another note should be made about the "ld" option "-n".
As discussed, if NEITHER "-n" or "-i" is specified, you get the
default, which is that each user runs his own copy of the complete
program, with no sharing of the text and text is not write-protected.
In short, there is no distinction between text and data, and they are
put in the same memory mapping segments.
A program may fit in 16-bit addressing in the default case, but may
not fit when link-edited with the "-n" option. This is because the text
must be set up in separate map segments to be protected and shared.
This generally requires an extra memory mapping register. If your
program is big and using all them already it will no longer fit.
I believe the error you get is a simple "too big" when you try to execute
it without the program ever getting to first base.
- Tom Beres
{mcnc, seismo, allegra, brl-bmd}!rlgvax!tom