Re: Fresh 7.10 netinstall problems...
On Fri, Mar 14, 2008 at 12:03:40PM -0400, Douglas A. Tutty wrote:
> On Thu, Mar 13, 2008 at 10:19:09PM +0000, digbyt@skaro.afraid.org wrote:
> > On Sat, Oct 27, 2007 at 05:30:49PM +0100, Digby Tarvin wrote:
>
> > I have now had time to get back to investigating this, and after a
> > systematic test of all the differences between the 486 config and
> > the unstable 686 config, I have discovered that the critical item
> > was the setting for High Memory Support. In the 486 Kernel this
> > is set to 'off', and for the 686 kernel it is set to '4GB'.
> >
> > Having identified that this setting determines if my kernel will be
> > stable or not, I then tried booting the default (686) kernel with
> > the boot option 'mem=896M' (foregoing the top 128MB or ram)
> >
> > Sure enough, the full pentium Debian kernel is now stable!
> >
> > Still trying to find a way to determine if this indicates a fault
> > in this machine, or some subtle compatability issue with it.
> >
> > If anyone has any idea what might be causing this particular fault I
> > would be most intetested to hear. I would have thought some bad ram
> > would have been detected by the BIOS or the install disk memory tests.
> > And I am told that the Windows system which it came with runs without
> > any problem. I wonder if there is anything about the way HIGHMEM is
> > used that would show up some subtle flaw in this machine?
>
> What happens if you swap memory around in the box. If it causes errors
> on your previously-stable kernel, then that's a sign. What happens if
> you install memtest+ which puts a line in grub's menulist so that you
> can boot into pure memtest and let it cook for a couple of days?
I did try the memory test from the install disc back when I fist started,
although can't remember if I used the debian or the ubuntu disc. Anyway,
it failed to find anything wrong. I assume that would be the same as
memtest+ wouldn't it?
Good idea on swapping the memory around. I'll give it a try.
I may have spoken too soon about the mem= argument though. It certainly
allowed HIGHMEM enable kernel to boot all the way to X (which it never does
otherwise), but I still had my kernel build fall over after a couple of
hours with a bus error. I have just re-built the standard 686 kernel
with HIGHMEM off to confirm that that gets me back to complete
stability.
I am tempted to adjust my memory boundaries so that I can use the full 1GB
of ram without having HIGHMEM enabled (ie change the 3G/1G split to a
2.75G/1.25G split). That would give me access to all the resources - but
it would be more satisfying to know exactly what the issue is..
Regards,
DigbyT
--
Digby R. S. Tarvin digbyt(at)digbyt.com
http://www.digbyt.com
Reply to: