Subject: followup: make coredumping
To: None <port-sun3@NetBSD.ORG>
From: der Mouse <mouse@Collatz.McRCIM.McGill.EDU>
List: port-sun3
Date: 11/07/1995 08:13:40
A while ago I posted something saying that I'd been seeing make
coredumping, always in exactly the same place, during "make build"
working on libc.

I built a version of make with -g, and it coredumped too, though not at
the same place.  Upon looking at the core with gdb, it appears that it
simply ran out of memory.  In the source,

    PAlloc (nLNode, ListNode);
    nLNode->datum = d;
    nLNode->useCount = nLNode->flags = 0;

nLNode is a nil pointer just after the PAlloc call.

Examining the sbrk-maintained curbrk pointer reveals that it's
0x0081dffc.  The coredump is 8397408 bytes.  size on the make binary
says
	text    data    bss     dec     hex
	114688  8192    18476   141356  2282c
The text segment size in hex is 1c000.

This all adds up, to me, to make running out of VM at about 8
megabytes.  This would not be interesting, except that 8 megabytes is
exactly how much RAM there is in the machine.

Therefore, I speculate that the NetBSD/sun3 simply does not page
correctly (though my experience indicates it does know how to swap).

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu