root@bu-cs.UUCP (Barry Shein) (08/26/85)
Ok, I am trying to get more info on this, but maybe someone with the
sources can help me with hypotheses? There seems to be some strange
interaction between wrastop() and floating point (!) which can cause
the machine to panic, usually with panic: init died, or sometimes
just the 'available memory' message like it's trying to reboot (but
it just sits there.) I have a program which does:
#include <math.h>
...
int i,j ;
double x ;
for(...)
{
i = 75*sin(x) ;
...
}
wrastop(...) ;
and it will work fine. Now the same exact program with just:
(same everything)
j = 75 ;
for(...)
{
i = j*sin(x) ;
....
}
wrastop(....) ;
panics the system, that is, the only difference is that we multiply
by j rather than constant 75 (I originally had j as a loop counter which
also panic'd so I narrowed it down to this, it just draws some ellipses.)
Doing a diff on the .s files for each reveals a lot of differences so
my questions are:
a) Is this machine memory protected? I assume it has to
be (I mean the kernel from user programs) and maybe it's
wrastop that's scribbling all over things in the kernel?
b) Is floating point simulated? I notice cc does not take
the -f flag but rather passes it on to ld, for what it's
worth.
That is, my current theory is that the floating point simulator
and wrastop() are somehow scribbling on the same memory locations
in the kernel area.
At any rate, I would be happy to speak directly with a developer if
any are listening.
-Barry Shein, Boston University
P.S. Yes, I like the machine, I didn't buy it, it was loaned to me,
I woulda gotten it out of here if I didn't like it, there's no room
in my office for junk, c'mon guys, a bug fix release and life'll be
perfect :-)