All,

I am using NetBSD 1.2.1/i386 and have recently upgraded a Dell Pentium 
system from 32MB of RAM to 128MB of RAM. 

When I rebooted, my kernel crashed, so I modified the kernel config file 
to set EXTMEM_SIZE=130048, and increased the value of the NKPDE 
constant in /usr/src/sys/arch/i386/include/pmap.h. For the kernel,
I then did a make clean, and proceeded to build and install a kernel.

However, even with these changes, my kernel still crashes. What I am 
doing wrong, or forgetting to do?

The FAQ that Dave Burgess maintains at http://cynjut.neonramp.com/
says that the value for EXTMEM_SIZE should be the number of bytes in the
extended memory range for the system. His example says that for a 128MB
system, set the value to (128-1) * 1024, or 130048. It also says that
the constant NKPDE needs to be increased. The FAQ recommends 16, so I
increased the default value of 12 to 16.

When I reboot, I see a message that says something like:

  Real memory = 133820416
  panic: ptdi 24b023
  Stopped at _Debugger +0x4: leave

The value for real memory doesn't seem to make sense... I am not sure
how it is calculated, but 128 * 1024 * 1000 = 131072000. This is the value
that I am expecting to see, but for whatever reason, the system thinks
there is an additional 2748416 bytes of memory.

As I am writing this, I am now wondering where the build process 
really gets the value for NKPDE from -- I assume it gets it from 
/usr/src/sys/arch/i386/pmap.h, but if it is really using a header file
from /usr/include, then I have to make change the NKPDE constant there.
I am writing this at my office, and will check this out when I go home 
today.

I assume others have seen this problem before, so please tell me what you
think I am doing wrong... Everytime I want to test the new memory in my 
system, I take out the 4 32MB SIMMS and replace it with 2 8MB SIMMS, and 
I am growing tired of that process...

>From watching the postings to current-users, it appears that having to
set the EXTMEM_SIZE option in NetBSD 1.3 will no longer be necessary
as the routines for detecting memory have been improved. Can anyone 
verify this?

Thanks in advance,


Brian

-------------------------------------------------------------------------
| Brian Stark                       | Internet : bstark@siemens-psc.com |
| Siemens Energy & Automation, Inc. |          : bstark@empros.com      |
| Power Transmission & Distribution |                                   |
| 7225 Northland Drive              | Voice    : +1 612 536-4697        |
| Brooklyn Park, MN 55428   USA     | Fax      : +1 612 536-4919        |
-------------------------------------------------------------------------