Early memory allocation

[Norman Feske]

Hello Sven,

>> from the 'Platform::_setup_mem_alloc' function. ;-) You will notice that
>> the startup delay gets much reduced. But you will certainly also notice
>> corruptions in Pistachio's mapping database.
> 
> No, I need a reliable kernel :) But I will even now investigate that,
> especially because we go for Genode on OKL4-Pistachio (on qemu,
> l4ka-pistachio was easier to set up for a start...). The OKL4 kernel was
> derived from l4ka, so it might show similar corruptions.

I took your mail as reason to try out Pistachio's "new mapping database"
feature, which is disabled by default. Apparently, this feature works
much better so you may give it a shot. Please note that the changed
Pistachio kernel config updates the kernel headers so you need to
rebuild not just the kernel but also the Genode userland.

BTW, OKL4 is not prone to the mentioned mapping-database corruption
problem because it does not support recursive mappings. On Pistachio,
we observed the problem when mapping a 4K page out of a locally mapped
4M page to another process. (In fact, the real problem appeared later,
when unmapping the said mapping) On OKL4, this never happens because
the map operation takes a physical memory descriptor as source argument
(the kernal does not maintain a hierarchical mapping database).

Regards
Norman