[syslinux] Memory upgrade breaks net boot (PXE + NFS Root)

[Alex Still]

On 2/8/07, H. Peter Anvin <hpa at zytor.com> wrote:
>
> Alex Still wrote:
> > Hi list !
> >
> > A memory upgrade (from 2 to 4Gb) on a bunch of IBM LS21 blades has
> triggered
> > a very weird behavior.
> > I really can't say if this could be PXE related or not - apologies if it
> > isn't.
> >
> > These machines start via PXE and then NFS-mount their root.
> >
> > Following the upgrade, none of them would boot. I've played with the
> "mem="
> > kernel option, and the result is :
> > - if mem < 3250m : everything's fine
> > - for some values of mem : Machine auto-reboots while loading the initrd
> > "Loading initrd...*bam*"
> > - some others : Initrd loads, but turns out it's corrupted ("can't find
> a
> > valid initrd", then panic)
> > - some others : Kernel starts booting, bnx2 driver errors all over the
> > place, machine hangs during the boot process
> >
> > Unfortunately I have no other hardware to try this on. We're using a 64
> bits
> > kernel, which should handle 4Gb fine.
> > Booting from a CD works fine. At this point I can't figure out if this
> is
> > related to PXE or the kernel (nfs-root setup)
> >
> > If anyone else has seen this, I'd love to hear about it !
> >
>
> I would start by running memtest86+ on this box.  My first guess would
> be a bad DIMM (or an overloaded memory bus), or a BIOS (SMM?) bug
> related to very large amounts of memory.
>
>         -hpa
>
>
I was unclear,  "no other hardware",  meant I only have LS21 blades. But all
40 of them fail in the exact same way ;)
memtest86+ didn't find anything on the 3 blades I tried it on.

Now, further tests show that pxe booting a fedora 6 install kernel/initrd
works just fine.
(using the same pxelinux.0 on the same TFTP server)
So I guess the problem is with our kernel.  The thing I don't understand is
why it sometimes reboots before it is even done downloading the initrd.

Back to testing !

Cheers,
-- 
Alex