Re: MACHINE_NEW_NONCONTIG

[David Hopper <dhop%nwlink.com@localhost>]

Thanks, Michael:  -n1 and -n2 does properly post memory now, but it looks like 
there are bigger
problems now.  An ugly UVM fault occurs when I concatenate memory, after a few 
seconds of running
through the rc.  First run, it happened at inetd; now it occurs at fsck_ffs, so 
maybe it's happening
whenever it hits the phys mem boundary:

[...]
Automatic boot in progress, starting file system checks.
/dev/rsd1a: file system is clean; not checking
uvm_fault(0xcd980, 0x0, 0, 0x1) -> 0x1
  type 8, code [mmu,,ssw]: 505
trap type 8, code = 505, v = c
pid = 16, pc = 000B8568, ps = 2404, sfc = 1, dfc = 1
[... kernel stack ...]
panic: MMU fault
Stopped in fsck_ffs at 0x95356         unlk     a6
db>

As a sanity check I used 'loadbsd -a -n2 netbsd' on Ignatios' bona-fide kernel 
from netbsd.org, and
I get the same UVM dump.

--dhop

"Michael L. Hitch" wrote:

> David Hopper <dhop%nwlink.com@localhost> writes:
> > Perhaps I misunderstand NONCONTIG's function, then.  NetBSD stopped 
> > concatenating memory
> > segments for me around 1.3, and I assumed it was because of the UVM 
> > overhaul goings-on.
>
>   MACHINE_NEW_NONCONTIG replaced the older NONCONTIG option.  Since all ports
> can only use UVM, and UVM requires MACHINE_NEW_NONCONTIG, that is no longer
> an option.
>
> > I have 32 megs on a Warp Engine, and 16+2 on my A3K motherboard.  The 
> > relevant bits of the
> > dmesg output are here:
> >
> > NetBSD 1.4.1 (TIAMAT) #1: Wed Sep  8 04:05:31 PDT 1999
> >     bard@tiamat:/usr/src/sys/arch/amiga/compile/TIAMAT
> > Amiga 3000 (m68040 CPU/MMU/FPU)
> > real  mem = 33554432 (4096 pages)
> > avail mem = 30138368 (3679 pages)
> > using 217 buffers containing 1777664 bytes of memory
> > memory segment 0 at 8000000 size 2000000
> > memory segment 1 at 7000000 size 1000000
> > memory segment 2 at 0 size 200000
> > mainbus0 (root)
> > ...
> >
> > Userland insists that I only have 32 megs.  Booting into AmigaOS enables 
> > all 50 megs.  NetBSD
> > 1.2 and earlier would post all 50+ megs at the real mem post; 1.3 stopped 
> > doing this.  Strange
> > (is it?) that it posts all memory segments, but won't enable segments 1 or 
> > 2.
>
>   Are you using the "-n1" or "-n2" option when booting the kernel?  The Amiga
> only uses one memory segment by default, and requires the "-n" option to
> enable multiple segments.  The bootup message will always display all the
> memory segments found by the boot loader, even if the kernel doesn't use
> them all.
>
>   I haven't actually verified that multiple segments still work, so it is
> possible that something got broken somewhere along the way.
>
> Michael
> ---
> Michael L. Hitch                        mhitch%montana.edu@localhost
> Computer Consultant,  Information Technology Center
> Montana State University, Bozeman, MT     USA