tmm3@quads.uchicago.edu (todd matthew morin) (03/05/91)
Hello, I've inherited an assembly language routine that was written a few years ago and writes directly to the mac VIA to enable millisecond accurate timing. Unfortunately, this code predates all but the MacPlus and SE (and the 512K, etc). The problem is that on newer machines, most notably the SE/30, unexpected crashes occur as a result of this code. Using MacsBug I've been able to figure that the program is crashing on an RTS instruction. The program itself (in which this timer is included), displays files on the screen synchronously with the vertical interrupt, then starts the timer. The SE/30 does not consistently crash at the same file being displayed, but only consistently on the RTS instruction. This leads me to believe that the SE/30 is doing something with VIA1 during the millitimer routine. I have a copy of the new mac hardware reference (Guide to the Macintosh Family Hard- ware, 2nd edition), and have gone through the excellent section on the VIAs, but since no sound is being played, no disk access is occurring, etc., I am at a loss to understand exactly what is happening. Anyone have any ideas. The source code is available upon request for those who think it'll help figure this out. This code BTW was modified for MPW by David Powell of NIH in 1986, and originally written under MDS by Scott Gillespie of Reed College, and has been floating around for at least 4 years...so many of you may have already seen it. Anyway, I'd appreciate any ideas any of you might have. If you need a copy of the source code to help, drop me a line. Finally, please e-mail responses and I will definately post a summary. Thanks, Todd M. Morin tmm3@midway.uchicago.edu
Lawson.English@p88.f15.n300.z1.fidonet.org (Lawson English) (03/06/91)
todd matthew morin writes in a message to All
TMM> I've inherited an assembly language routine that was written
TMM> a few years ago and writes directly to the mac VIA to enable
TMM> millisecond accurate timing. Unfortunately, this code predates
TMM> all but the MacPlus and SE (and the 512K, etc). The problem is
TMM> that on newer machines, most notably the SE/30, unexpected crashes
TMM> occur as a result of this code. Using MacsBug I've been able
TMM> to figure that the program is crashing on an RTS instruction.
TMM> The program itself (in which this timer is included), displays
TMM> files on the screen synchronously with the vertical interrupt,
TMM> then starts the timer. The SE/30 does not consistently crash
TMM> at the same file being displayed, but only consistently on the
TMM> RTS instruction. This leads me to believe that the SE/30 is doing
TMM> something with VIA1 during the millitimer routine
I too have some code that breaks with the newer systems. It doesn't seem to
break due to hardware problems, but because we were using a low-memory global
as a scratch address which is now overwritten. As far as I can tell, VIA1 should
work as it always does. It does for my code, I simply have to rework the scratch
address to work with post System 4.1 systems.
It is an interactive profiler that examines all address space and graphs memory
usage "on the fly." It uses the VIA to set an interupt every millisecond. When
it was working, we could zoom into any heavily used area of memory and map the
most used sections. It had accuracy from 1/256 of the total address space down
to 2 bytes, so we could see single-machine-instruction bottlenecks.
I suspect that your problem isn't with the VIA, it's System-dependent like mine
was(is). The other possibility is a stack problem. There may be differences
in the way the stack RTS is handled returning from an interupt.
Lawson
--
Uucp: ...{gatech,ames,rutgers}!ncar!asuvax!stjhmc!300!15.88!Lawson.English
Internet: Lawson.English@p88.f15.n300.z1.fidonet.org