i had some very short conversation with hpa (thank you for that valuable input!) and he is telling, that grub2 bypasses standard linux kernel initialization code . he thinks this is not good at all and wonders, what makes grub2 think it`s smart enough to do so.
by piece of luck, grub2 "circumvents" the pae problem with cpu`s not correctly reporting it - and that made it possible to provide installation-quirks like being described on the ubuntu wiki.
from my limited point of view, skipping standard initialization code from the linux kernel does not appear to be smart to me too, because it made finding the real problem so extremly difficult.
i guess, if grub2 (and so the kernel) would just behave the standard way, maybe we already would have an appropriate quirk in the linux initialization code, or at least a hint that there are cpu`s which have pae but do not report about - and a kernel param to force skipping the pae check.
you can easily verify that hpa is correct - just edit the grub commandline and replace "linux /boot/vmlinz....and "initrd /boot/initrd..." with "linux16 /boot/vmlinuz..." and "initrd16 /boot/inititrd..." and - voila, the problem also happens on bootable systems.
i think this is stuff for being further discussed on linux-kernel and grub mailing lists...
so here we go:
i had some very short conversation with hpa (thank you for that valuable input!) and he is telling, that grub2 bypasses standard linux kernel initialization code . he thinks this is not good at all and wonders, what makes grub2 think it`s smart enough to do so.
by piece of luck, grub2 "circumvents" the pae problem with cpu`s not correctly reporting it - and that made it possible to provide installation-quirks like being described on the ubuntu wiki.
from my limited point of view, skipping standard initialization code from the linux kernel does not appear to be smart to me too, because it made finding the real problem so extremly difficult.
i guess, if grub2 (and so the kernel) would just behave the standard way, maybe we already would have an appropriate quirk in the linux initialization code, or at least a hint that there are cpu`s which have pae but do not report about - and a kernel param to force skipping the pae check.
you can easily verify that hpa is correct - just edit the grub commandline and replace "linux /boot/vmlinz....and "initrd /boot/initrd..." with "linux16 /boot/vmlinuz..." and "initrd16 /boot/inititrd..." and - voila, the problem also happens on bootable systems.
i think this is stuff for being further discussed on linux-kernel and grub mailing lists...