> > cache entries before unmapping the page.  This seems to do the trick,
> > an MP kernel on a dual hypersparc SS20 can get to start_init() now
> > (see my post to tech-smp)
> 
> Oooooh... so I should REALLY put the pieces of my two sparc10 (one
> with working memory, one with a dual-cpu board) together...? 

I wouldn't celebrate yet.  In order to get as far as start_init() I
need to have the semaphore operations do a printf() - otherwise we
typically get a watchdog reset (nested trap actually - thanks eeh)
long before then.  This seems to be because ctw_invalid is trying to
save an insane number of windows - presumably because we are not
initializing the secondary cpu's completely?

I _think_ the printf's help because the printf mutex forces
the cpu's into lockstep, or it could simply be a  timing issue.

Anyway, there's a lot of work yet to be done and I expect it needs
sparc clue - which I don't have -  I had a lot of help from Bill
Sommerfeld tonight ;-)

Which is why if no one strenously objects I thought it would be
usefult to commit kern_sem.c so that folk with more clue can have a
play too.

--sjg