Re: NetBSD 6.0 fails to boot on large server

[Chuck Silvers <chuq%chuq.com@localhost>]

On Fri, Jan 18, 2013 at 12:08:46AM +1100, Darren Reed wrote:
> On 17/01/2013 7:12 PM, David Laight wrote:
> > On Thu, Jan 17, 2013 at 12:47:50AM +0100, Darren Reed wrote:
> >> I tried to boot 6.0 (could not find the -current ISO) on one of the
> >> larger servers that I have access to...
> >>
> >> It didn't go very well.
> >>
> >> Once the boot loader started to happen, the updates through the ILO
> >> screen were very slow...
> >>
> >> http://imgur.com/a/y3zel
> >>
> >> The last message about "cpu0 at mainbus0 apid 0" has been there for ~10
> >> minutes
> >>
> >> The "total memory" is out by around 522405MB.
> > 
> > How much memory and how many cpus should there be?
> > Any idea how many different memory regions?
> > (I remember something about the system not like lots of memory regions.)
> 
> 512G, 8 CPUs in one of these:
> http://www.supermicro.com/products/system/5u/5086/sys-5086b-trf.cfm

right now we only support memory with physical addresses less than 512 GB,
and this system apparently assigns some of its memory to physical addresses
higher than that (eg. to leave room for PCI memory space below 4 GB).
this is because the "direct map" of all of physical memory only uses one
L4 page table page, which only maps 512 GB of virtual space.

for right now you can build a kernel that will ignore the memory with
physical addresses higher than 512 GB with:

options PHYSMEM_MAX_ADDR="0x8000000000L"

I'll look at adding support for more memory this weekend
(unless someone else would like to do it).

-Chuck