physical memory limit?
Andreas Tobler
andreast-list at fgznet.ch
Wed Oct 8 19:23:35 UTC 2008
Hi Peter,
Andreas Tobler wrote:
> Peter Grehan wrote:
>
>>> I remember back in 05 we had a similar issue when I tried
>>> 'buildworld' on a G4 alu book. Then you suggested to me
>>> to reduce hw.physmem to 512MB.
>>
>> Indeed, and I'm certain those problems were fixed in r1.115
>> of mmu_oea.c where the secondary PTEG hash was corrected.
>
> Ok, then I trust you :)
Hm, I just read the commit log:
---
Remove bogus increment of re-hashed PTEG index. This snuck in with r1.12
of pmap.c, and is potentially the cause of hangs reported on machines
with a small amount of memory. On machines with sufficient RAM, and
without a lot of processes running, this situation would probably never
occur.
---
Irritating issue is the part with 'hangs reported on machines with a
small amount of memory'. I do not have small amount of ram, neither did
I have hangs. I had 11 at that time. Now I have 11 again, but maybe due
to other issues. Well on OS-X the machine does not complain. Even
memtest doen't complain during the one hour period I let it run. Also,
it might be that this needs a looong run of memtest to uncover something.
Anyway.
>>> I'll reduce/replace the memory and see how it behaves.
>>
>> OK, let us know how it goes.
>
> First step passed, I reduced to one 512MB DIMM and a full buildworld
> with parallel qt build passed.
> I'll try overnight with the other DIMM to exclude single DIMM failures.
> But still, I have some more possibilities then, slot1/2 is defect, DIMM
> 1 does not like co'op with DIMM 2, ECC vs. non ECC (one is marked with
> ECC support, the other isn't).
> Will take some hours to find out.
Continued and found that both DIMM's seem to work in the first bank.
Running a buildworld with a DIMM in the second bank failed for both,
ECC/non ECC. So it seems it might be a problem with my HW. :(
I even bought a new, non ECC DIMM and I seem to have the same pattern.
Right now I try with 768MB (512MB/256MB ECC).
Let you know about findings.
Regards,
Andreas
More information about the freebsd-ppc
mailing list