Good news!  It turns out that my use of the new MIPS_DYNAMIC_STATUS_MASK
feature wasn't working properly.  In some cases, this macro will not work
properly in locore.S, etc., if e.g. the interrupt enable bit was in effect
on a previous line.  You'll see what I mean if you look at the code.
So I ended up sticking an ifdef in arch/mips/include/cpuregs.h to set
MIPS_INT_MASK to mask off INT_MASK_4, because I don't know any other way
to make it shut up!  So gladly, things are looking promising.  Net booting
appears to start up properly (though I still need to unpack the binary sets
onto an nfs server).  What's odd is that the sonic driver reports 
"no heartbeat" twice, even if I set the SQE jumper on my transceiver.  But
the messages would appear harmless.  Also, I tried enabling scsi, and the
"waiting 2 seconds for devices to settle" seems to wait much longer, maybe
20 seconds.  I don't know if this is because there AREN'T currently any
drives connected.  I'll try attaching a drive to see if it's any different.
And of course, I'll send my current patches to the list as soon as I've 
verified that net booting and nfs mounting completely works, probably later
tonight.

Cheers,
-Mark