In <mtuvgvflrdn.fsf@contents-vnder-pressvre.mit.edu>
nathanw@MIT.EDU wrote:

>>>I'd rather avoid trying to upgrade to 1.5_ALPHA2 just for the sake of checking,
>>>especially since a buildworld will most likely result in the same "kmem_map"
>> >related panic that got me into this mess in the first place.
>> 
>> Yeah, can someone else speak to this? Nathan?
> 
> I get this on 1.5_ALPHA2 on my 3/80, but not my 3/60. Haven't had time
> to do a deeper analysis.

I've tested a kernel with "options KMEMSTATS".
The kernel does not panic, but all userland
processes hang up during make build.

dump_kmemstats() says "VM pmap" wastes kernel memory:

---
[ entries are sorted ]
 :
34 VM pmap              2516608
82 UVM amap             74144
62 ttys                 63232
 2 devbuf               29472
28 UFS mount            27664
49 Export Host          8192
 :
----

malloc() for M_VMPMAP is only in sun3x/pmap.c:pmap_create()
and it should be free()'ed in sun3x/pmap.c:pmap_destroy().
I don't know about pmap very much, but what causes this problem?

Some simple_locks are needed like sun3/pmap.c ?
pmap_remove() should be called in sun3x/dvma.c:dvma_mapout() ?
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp