2.8 fixes/bugs

utzoo!decvax!cwruecmp!ordy utzoo!decvax!cwruecmp!ordy
Thu Dec 30 16:45:02 AEST 1982


	Some 2.8 BSD stuff.

	1) in the makefile for 'csh', there is a warning that you
had better link the executable separate I & D if you can because
the construction: set xxx = `date` dumps core when sourced. There is a claim
that the problem may be in the kernel. Well, it is not, it is just
that there are too many large buffers which reside on the stack,
and in that case they add up to a segmentation violation. The define
in the makefile of 'BUFSIZ' is 1024 by default, and I have reduced
it to 400 with no ill effects, and the above construction now works
fine on an 11/23+ (nonseparate I & D). Note that BUFSIZ controls the
size of some I/O buffering and the maximum command word size. The
only common large words I have found are termcap strings, and in fact
the use of tset in the .login file prompted me to explore this.
If you don't care about long words, you can reduce BUFSIZ even further.
Note: 512 was not small enough.


	2) I believe there is an error in the machdep.c kernel source
file if out of kernel address space buffers are used, and there are
greater than 64 buffers. There is a calculation of the number of
clicks to allocate from memory for buffer space. It goes something
like: btoc(NBUF*BSIZE). All arithmetic is done in 16 bits, and with
BSIZE = 1024 and NBUF greater than 64, boom. It is in there twice,
so be sure to get both. The obvious fix is to assume that BSIZE is
always an integer multiple of the click size, and perform the multiplication
after converting BSIZE to clicks. (NBUF*btoc(BSIZE))
	Kernel comments elsewhere indicate you can have up to 256 buffers.
Anybody ever get like 150 or so of them??? I have both an 11/44 and 11/23+
with about 1 meg each, and I am about to try.

	3) There have been rumors about ps and w not working on
nonseparate I & D machines, bombing out with the 'no namelist'
diagnostic. Any fixes??

	4) If anybody has an 11/23+ (22 bit addressing), and needs
the simple mods to make the RL02 disk controller talk 22 bits, I have them.

	5) Any words on 2.8.1 -- that is 2.8 with job control and
a few more things??? I have a prerelease, which boots, but craps out
in a few commands. Anybody got it working, or hints on what to look
for???

	6) I also have been experiencing a problem where running
overlay programs on an 11/23 (nonseparate I & D) corrupts the kernel,
and an eventual crash results. As this is not my machine, time and
access for debugging are limited. Any ideas on what to check for???

	The systems I am playing with 2.8 on are not very close
to me, and as a result I don't have the time or opportunity to
really get in and play. Any further 2.8 hints would be appreciated,
and if mailed to me I will compose and present in a week or two.

	Greg Ordy (decvax!cwruecmp!ordy)



More information about the Comp.bugs.2bsd mailing list