notes@ucbcad.UUCP (11/24/83)
#N:ucbesvax:1900001:000:361 ucbesvax!turner Nov 24 06:19:00 1983 We're running 4.2. The end(3) man page refers to brk(2), saying that the value of the system break is "reliably" returned by brk(0). But, looking at man page brk(2), it's say nothing about this special case. I'm not apprised of the v7 versions--is the documentation the same on this count? What can I assume here? --- Michael Turner (ucbvax!ucbesvax.turner)
mjs@rabbit.UUCP (11/28/83)
It is sbrk(0) that returns the current break (since V6). Brk's
argument is the absolute break to set; sbrk's argument is the number of
bytes to move the break up.
--
Marty Shannon
UUCP: {alice,rabbit,research}!mjs
Phone: 201-582-3199padpowell@wateng.UUCP (PAD Powell [Admin]) (11/29/83)
WARNING: I have just read the documentation for sbrk, and have come to
the conclusion that calling sbrk(0) will cause at least some memory to be
allocated, as the docuemtation states that this will be done.
I have just experimented, and SUNAVBITCH! it doesn't seem to do this on 4.1,
at least here at Waterloo, but the behaviours on 4.1a/Version 7/System 3/
may be slightly different.
Try the following little gem to see what you have:
#define BLKSIZE 1024
main()
{
static char *new_base, *old_base, *new_top, *old_top;
char *sbrk();
int i;
for(i=0; i<100; ++i ){
old_base = new_base;
old_top = new_top;
new_base = sbrk(0);
new_top = sbrk(BLKSIZE); /* read sbrk(2) about rounding up */
if( (int)new_base == -1 || (int)new_top == -1 ){
printf( "THUD! limit reached\n" );
return;
}
printf( "old_base %ld, old_top %ld, new_base %ld, new_top %ld\n",
(long int)old_base,(long int)old_top,
(long int)new_base,(long int)new_top);
if( old_top && old_top != new_base ){
printf( "Gotcha! funny alignment snuck in\n" );
return;
}
if( new_top != new_base ){
printf( "Gotcha! allocated something for sbrk(0)\n" );
return;
}
new_top += BLKSIZE;
}
}
I must admit that this area seems to be one of the biggest botches
in the kernel design. Break function, and C break? Sigh.
Patrick Powell
P.S.- Why don't we simply make this a new defintion of sbrk, modify the
man pages, and FORCE implementors to return the top address for
sbrk(0)???