Re: lilo in i386 - fail for 64bit ?
As far as I know, mem=nopentium is no longer needed. You are probably
safe without it in both 32 and 64bit kernels.
On Tue, 2004-06-15 at 20:22, Alex Perry wrote:
> Purely for the record in the mailing list archive, should someone else
> observe the same symptom ...
>
> Alex Perry wrote:
>
> >My M6805 machine has suddenly lost the ability to boot a 64 bit kernel.
> >32 bit works fine, but any 64 bit or biarch kernel gets to a split
> >second after the decompression finishes and triggers a hard reboot. I
> >didn't think I'd made any changes to the system, other than upgrading to
> >the current Testing, so I'm wondering whether anybody else has observed
> >this odd phenomenon.
> >
>
> Somehow I managed to have "mem=nopentium" appended to all my kernels,
> not just the pure32 ones. This mistake had not been committed through
> lilo so, when the lilo upgrade was installed, the kernel parameters were
> modified. This append has the documented effect (and is needed for
> Athlon processors) in pure32. In pure64 and 32/64 kernels it seems to
> be interpreted as a request for a memory size of zero, which fails.
>
> I've noticed some installers ask the user whether they have an AMD
> processor and, for an affirmative answer, will automatically add
> "mem=nopentium" to the kernel parameters. It occurs to me that this
> might cause people who are installing a dual boot pure32/pure64 to have
> trouble booting their new biarch kernel.
>
> While I'm on the subject, the bootloaders decide whether a kernel is a
> "zimage" or a "bzimage" by inspection of the content and not by looking
> at the file name. If you get a "too big" error from the bootloader, it
> probably means you copied the "zimage" file from the kernel build tree
> when you wanted the "bzimage".
>
> Alex.
>
Reply to: