martin@media-lab.media.mit.edu (Martin Friedmann) (10/08/90)
It seems that SunOS4.1 off the tape crashes almost regularly when some certain applications are run. I have found two of these applications, one of which I wrote and use often getx11 from the Utah Raster Toolkit (without the -f flag) (-f means no fork()!) is an image display program for X11 which crashed the kernel about 10% of the times. The kernel crashes in a call to fork(). This I am certain of. Another application which will crash a SparcStation (EVERYTIME I've tried this!) is rcp. try the following on a SparcStation1 running vanilla SunOS4.1 a-machine% rcp a-file a-sparcstation:/dev/audio That crashes every Sparc everytime! Othere things about SunOS4.1 seem fine and dandy. I am just having problems with my getx11 program. Is there a fix to 4.1 available? Do you want a kernel traceback? a PANIC dump? Am I alone in my problems? Marty. p.s. People at Sun should know that I'm not flaming SUN. I like Sun, Suns and things that run on Suns! So don't feel like I hate you! But you gotta FIX THE BUGS!!!
jis@mtgzx.att.com (J Mukerji) (11/04/90)
> It seems that SunOS4.1 off the tape crashes almost regularly when some > certain applications are run. . . . . . . . > Is there a fix to 4.1 available? Do you want a kernel traceback? a > PANIC dump? Am I alone in my problems? > Marty. > p.s. People at Sun should know that I'm not flaming SUN. I like Sun, Suns and things that run on Suns! So don't feel like I hate you! But > you gotta FIX THE BUGS!!! I sent kernel dump to our friendly Sun Rep. They analyzed it and sent me a patch that fixed the kernel trap problem. The README file from the patch is excerpted below: > Keywords: 4.1, 4/110, flush_wind, trap, fault, data > Synopsis: data fault trap in flush_windows on 4/110 running 4.1 > Date: August 13, 1990 > SunOS release: 4.1 > BugId's fixed with this patch: 1029939 > Architectures for which this patch is available: sun4 sun4c > Obsoleted by: 4.1.1 > Problem Description: > On a fork, the child gets a copy of the parent's > process control block, which for the child will be invalid. > If it enters the kernel because of a register window overflow > or underflow, the code will fault in a user stack page. > When the process is eventually resumed, the bad pcb causes > the panic. Your problem sounds very similar to the one that I had. Applying this patch may fix your problem. Good Luck!