On Tue, 2006-03-07 at 11:36 -0800, Andrew Morton wrote:
> Because there was a change which could have affected this.  But it was
> merged in late October and was present in 2.6.15.

It certainly is possible that my patch caused the bug.  However, my
patch only affects limit_regions(), which is only called when the user
specifies a mem= argument on the command-line.  When they do this, in
addition to the BIOS-e820 printout, they should also see a "user-defined
physical RAM map:", which I don't see in the diff.  Also, I'm pretty
sure that this e820 printout runs before parse_cmdline_early(), where
limit_regions() is called.

Martin, in any case, I have debugged things in that code recently, and
I'd be happy to help you fix your problem.  I've attached a patch that
does a ton of e820 debug printks.  If you could get me a full copy of
your dmesg with that applied, I should be able to locate the problem a
bit more quickly.

-- Dave