Re: PIC question

[Johnny Billquist <bqt%softjar.se@localhost>]

John Klos wrote:
> > I'm starting to suspect it might be in connection to page faults and 
> > pages paged in. I should really test a few things on something else 
> > than a Unibus-machine though. It might be some problem related to the 
> > MSCP device driver...
>
> I can give you serial access to a VAXstation 4000/60, if that'd help. 
> Not sure how much I could help in other ways, but I'd certainly try.

Rbanks for the offer. I have a 4000/90 here myself though, so that won't 
be neccesary.
What I would appreciate if someone else could try to reproduce the 
problem itself, or even better, fix it... :-)

I'll see if I can post a small patch tomorrow that adds an event counter 
for missed clock interrupts, and then people could try to exercise their 
system and see if they can get that counter to move.
I don't have any problems getting missed interrupts on the 8650, but in 
todays terms, that is a very slow machine, although I have plenty of 
memory on it.

If anyone have any ideas of any situation where mi_switch might actually 
be called at IPL_SCHED would also be interesting to hear about.

I mean, it raises the IPL to IPL_SCHED inside mi_switch, which is 
obviously a good idea during the context switch, but sometimes the 
system ends up running the user process at IPL_SCHED as well, and it 
would seem to be connected to mi_switch being called already at 
IPL_SCHED. But I'm not entirely sure that's a correct conclusion yet.

Basically, I'm trying a lot of different debug prinouts, panics to get 
into ddb when I see that I've missed interrupts, and poke around in ddb, 
and other ways to try and figure out when/why the interrupts are missed. 
I mean, it's obvious why the interrupts are missed, we are running for 
long times at raised IPL levels, but why are we doing that, and how do 
we get there. That's what I'm trying to understand. And more pair of 
eyes would really be helpful, but that requires that others also can 
reproduce the problem...

I'm not even sure it is VAX specific. Maybe it's just that on faster 
machines, noone have really paid enough attention to notice that it can 
happen, and the effect isn't really a big problem on faster machines.

The 8650 don't really crash or anything from this either, it just means 
that the clock sometimes drift away seriously much, and as an effect, 
ntpd cannot be used either, since it cannot handle the randomness or the 
big deltas the drift can have.

        Johnny