[pups] Stray Interupts
Bill Gunshannon
bill at cs.scranton.edu
Mon Feb 26 12:45:53 AEST 2001
On Mon, 26 Feb 2001, Fred N. van Kempen wrote:
> > > It could mean that running Ultrix on a simulator isn't on
> > without some work on the emulators themselves.
> I have been running Ultrix-11 V3.1 on Ersatz-11 without any of that. It
Which I think says a lot about the quality of Ersatz-11.
> _could_ be linked to interrupt latency issues- Ultrix tells controller
> to do something (e.g., three commands to read a sector). Controller does
> as told, generating an interrupt for each of the requests saying its ready.
> However, because of latency, only ints 1 and 3 actually get delivered within
> the expected timeframe (can happen).
>
> Usually (from my experience with writing Unix kernel drivers), this is not
> a problem, because the "message" from (in this case) int2 will be picked up
> when we start to service int3, which we _did_ see. So, even though we didnt
> get int2, we were fine.
>
> Now... emulator wakes up again, goes "oi, i messed up, better go send that
> int now" and sends the int. The driver no longer _awaits_ an interrupt
> (because
> we cleared the AttentionNeeded flags when servicing int3), so... we get the
> "stray int" message.
Based on my experience over this weekend, I can definitely agree with
all of the above. It makes perfect sense and goes a long way toward
explaining my problems.
>
> If this logic is correct, it will get worse when loading the host system
> heavily, so latency will occur more often.
I can also vouch for this. I finally gave up on trying to do anything
I/O intensive on the emulated RP disk. Emulator was continuously
crashing.
> On a very fast box (like my quad CPU
> P3/850 Linux box) it should hardly occur.
While this is not a solution most people here are likely to be able to
apply :-) I also have doubts that it will solve the problem. I also
doubt that the problem is as easy as just throwing away the stray interrupt.
bill
--
Bill Gunshannon | de-moc-ra-cy (di mok' ra see) n. Three wolves
bill at cs.scranton.edu | and a sheep voting on what's for dinner.
University of Scranton |
Scranton, Pennsylvania | #include <std.disclaimer.h>
Received: (from major at localhost)
by minnie.cs.adfa.edu.au (8.9.3/8.9.3) id SAA49935
for pups-liszt; Mon, 26 Feb 2001 18:59:24 +1100 (EST)
(envelope-from owner-pups at minnie.cs.adfa.edu.au)
More information about the TUHS
mailing list