On Wed, 26 Mar 1997 11:17:21 -0800 (PST) 
 Bob Sutterfield <Bob@XC.Org> wrote:

 > Running NetBSD-1.2 on a Hexium-200, we're still seeing nightly panics.
 > 
 > Because for a while many of the panics said "vm_map_entry_create: out
 > of map entries", I replaced vm/vm_map.c with "1.26 1997/02/25
 > 23:27:08" from the NetBSD-current source tree on March 12.  I also
 > increased MAX_KMAP from 10 to 100 and MAX_KMAPENT from 500 to 5000.
 > That flavor of panic seems to have subsided.
 > 
 > Now we've started seeing "panic: ptdi %x".  I'm not familiar enough
 > with the changes of -1.2.1 and -current in arch/i386/i386/* to know
 > whether I'm safe to just replace pmap.c, or whether I need to worry
 > about its routines' external interfaces' compatibility with other
 > modules in that area.  Or even whether it's confined to pmap.c.

You're running out of kernel PDEs.  You've increased the number
of entries for the MI VM mappings, but not for the low-level
machine-dependent mappings.

In -current, the number of required kernel PDEs is computed at boot
time, but I'm not sure if this is in 1.2.1 (it required changes to
pmap.c, pmap.h, locore.s, and machdep.c).

As a work around, you should try bumping the "NKPDE" constant in
i386/include/pmap.h.  24 is the highest it should ever go.  I believe
the default is/was 12.

 > Or is now the time to schedule the 1.2.1 upgrade?  Has the pmap
 > mechanism been fixed in 1.2.1?  Any insight, advice, and suggestions
 > would be most welcome.

...there have been some changes to the i386 pmap in -current, but none
that really change how it works.  The most significant was the
dynamic computation of nkpde.

Hope this helps.

Jason R. Thorpe                                       thorpej@nas.nasa.gov
NASA Ames Research Center                               Home: 408.866.1912
NAS: M/S 258-6                                          Work: 415.604.0935
Moffett Field, CA 94035                                Pager: 415.428.6939