> I wonder if this is the cause of the break, it seems a bit too old.
> Anyone precisely noted when the serial port stopped working?

Replying to myself :o)

Ok, the problem was here. When the driver has been moved to the arm32
directory, there has been a small modification to sacom_filltx().
Reverting to the old version fixes the panic. I tried the serial port,
it works well (again), I can use cu to connect to another machine. 

===================================================================
RCS file: /cvsroot/syssrc/sys/arch/arm/sa11x0/sa11x0_com.c,v
retrieving revision 1.5
diff -r1.5 sa11x0_com.c
1178c1178
<       while(bus_space_read_4(sacomconstag, sacomconsioh, SACOM_SR1)
---
>       while(bus_space_read_4(iot, ioh, SACOM_SR1)

Now, I'd be happy to commit this change, but I must admit I don't really
understand what I changed here. Could anyone with a deeper understanding
of this code tell me if it is okay to commit the above patch? Why I/O
tag and handle became static variables instead of getting them from
softc?

-- 
Emmanuel Dreyfus
UNIX *is* user friendly. It is just a bit selective about his friends
manu@netbsd.org