>>> The kernel
>>> TLB miss appears to be occuring when the TLBmod exception handler is
>>> trying to load the first-level page table entry.
>>>
>>>                pmap  = p->p_vmspace->vm_map.pmap;
>>>                if (!(pte = pmap_segmap(pmap, vaddr)))
>>
>> I threw in the printfs, they showed (1 out of 1 try) that pmap was not NULL, 
>> but pmap->pm_segtab was.
>
> It's weird.  'pm_segtab' entry should never be NULL except for kernel map.

One possiblity is pmap_release() was mistakenly called for a live
process (not a dying process), then ->pm_segtab was assigned with
value NULL.  But, this theory would not explain symptoms found with
MIPS1-only kernels.

My MAXINE still keeps running doing 'libc make' with load averages
near 5.0.  (I started wondering some of TLB related CP0 registers
have bogus values in some generation of R3000 processors...)

Tohru Nishimura