[RFC] Memory management

[Vladimir 'phcoder' Serbinenko]

Hello. As appeared multiple times grub2's memory management can't
handle some problems.
1) First of all it manages grub_malloc and os_area separately which
forces initialisation code to split memory into two parts. Let's take
a case of initrd for linux and solaris both having size of 270 MiB.
And also a machine with 512 MiB memory. Linux loader puts initrd in
os_area but multiboot allocates memory with grub_malloc. So in this
configuration only one will work. One solution is to make multiboot
loader use os_area but then non-loaders may need big amounts of memory
too (e.g. graphics)
2) Cross-firmware. Currently every loader uses firmware functions to
claim memory ranges. These functions need to be abstracted and work
for systems where either grub (e.g. on BIOS) or firmware (e.g. on EFI)
is owner of memory
3) Hibernation. At least one loader (xnu) needs to supply a
hibernation image to the kernel. This is particularly nasty because it
has to be in one chunk and can be up to 100-epsilon percent of RAM.
So I propose the following adjustments:
1) When doing grub_memalign (); caller can optionally specify an
allocation policy. On BIOS there would be 3 allocation policies:
DEFAULT, KERNEL and LOWMEM.
DEFAULT will allocate starting at the end of memory, KERNEL starting
from the start of memory and ignoring everything upper than 4 GiB.
This decision is to make big allocation containing kernel and modules
needed for loader to be easily expandable when adding modules.
2) grub_request_cells will allow asking for a specified range to be
unmanaged by memory system. This would allow e.g. to load kernel at
1MiB mark
grub_request_cells can only request ranges aligned on cell boundaries.
A separate function grub_request available from a module will allow
sub-cell requests by maintaining a linked list of partialy allocated
cells.
You can see my current progress on
http://repo.or.cz/w/grub2/phcoder.git branch mm. Currently
grub_request is mapped to grub_request_cells. Everything compiles only
on bios and I haven't launched it at all yet. In other words you
shouldn't boot with it yet unless in vm you don't care about
-- 
Regards
Vladimir 'phcoder' Serbinenko