In article <6.0.0.22.2.20031120150329.02fa7418@localhost>
matt@3am-software.com wrote:

> I have a possible fix which I'll test tonight on my IBM NWS1000.
> Rather than restoring ci_curpm, I'm moving the initial trapframe down
> by one callframe.  Then directly after the trapframe, I'm storing a pointer
> to the current process'es pmap.  This is nicer than ci_curpm since I don't
> have to swap it on context switch.  Since I know the page is mapped and
> accessible since I've restored the userstate from it, getting the pmap can
> not result in another DSI fault.
> 
> I've attached the diffs (they at least build) but haven't tested them yet.

Unfortunately, this doesn't seem to fix the problem on my Apus2000/200.
I rebuild a kernel with this patch but it stops after mountroot as before,
and I can't get ddb prompt even on serial console after the hang.
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp