Higher Half Kernel
| Kernel Designs |
|---|
| Models |
| Other Concepts |
It is traditional and generally good to have your kernel mapped in every user process. Linux, for instance (and many other Unices) reside at the virtual addresses 0xC0000000 - 0xFFFFFFFF of every address space, leaving the range 0x00000000 - 0xBFFFFFFF for user code, data, stacks, libraries, etc. Kernels that have such design are said to be "in the higher half" by opposition to kernels that use lowest virtual addresses for themselves, and leave higher addresses for the applications.
Advantages of a higher half kernel are:
- It's easier to set up VM86 processes since the region below 1MB is userspace.
- More generically, user applications are not dependent on how many memory is kernel space (Your application can be linked to 0x400000 regardless of whether kernel is at 0xC0000000, 0x80000000 or 0xE0000000 ...), which makes ABI's nicer.
- If your OS is 64-bits, then 32-bit applications will be able to use the full 32-bit address space.
- 'mnemonic' invalid pointers such as 0xCAFEBABE, 0xDEADBEEF, 0xDEADC0DE, etc. can be used.
Initialization
To setup a higher half kernel, you have to map your kernel to the appropriate virtual address. How to do this basically depends on when you'd like your kernel to believe it's in the higher end, and when you set up paging.
Custom Bootloader
The easiest way is to load your kernel to any physical location you wish (for instance in the lowest 1MB) and prepare page tables that will perform the appropriate translation. Let's say you loaded your kernel at 0x00010000 to 0x0009FFFF and want it to appear at 0xC0010000, you could do the following:
- Pick 3 page-aligned (0x1000-aligned) addresses where you'll put your page directory and system tables. Make sure they are zeroed (memclr them or memset them to 0).
- Fill the lowest 256 entries of one table to set up Identity Paging for at least the BIOS and your bootloader (it's probably best to use 1:1 mapping for the entire lowest 1MB).
- In the other table, fill entry #0x10 (#16) with 0x00010003, entry #0x11 (#17) with 0x00011003, and so on (do this for every page your kernel has or needs).
- Fill entry #0x0 (#0) of the directory with the address of the first table (and make sure it's set to present).
- Fill entry #0x300 (#768) of the directory with the address of the second table (and make sure it's set to present).
When switching to Protected Mode, use this assembly example:
mov eax, physical_address_of_the_directory ; Get the physical address of the page directory...
mov cr3, eax ; ... and store it in CR3.
mov eax, cr0 ; Get what's in CR0...
or eax, 0x80000001 ; ... enable protected mode and paging ...
mov cr0, eax ; ... and put the new value back in CR0.
The GDT Trick
If you don't want to enable paging right from the start, it is still possible to have your kernel appearing in the higher half. Tim Robinson's GDT Trick works by using segmentation to select an appropriate base for the code and data segments. Say you've loaded your kernel at 0x10000 and we want it to appear at 0xC0000000, then all we need to do is find a base _X_, such as _X_ + 0xC0000000 = 0x10000. The bootloader will then initialize the GDT with cs.base = 0x40010000 = ds.base. This also means that special care must be taken for VRAM (video RAM) access, as 0xB8000 is now somewhere above 1GB. Either use a special 0-based additional data-segment or use
#define logical_to_physical(x) (((void*) x) + 0x40010000)
#define physical_to_logical(x) (((void*) x) - 0x40010000)
short *vram = physical_to_logical(0xB8000);
When you eventually enable paging, create the page tables as mentioned above (keeping in mind that you need to undo the address conversion again, e.g.
pgentry *pagedirectory = physical_to_logical(0x9D000);
pgentry *lowesttable = physical_to_logical(0x9C000);
pgentry *kerneltable = physical_to_logical(0x9B000);
/* prepare the lowest table and the kernel table... */
pagedirectory[0] = mkpgentry(0x9C000, PG_PRESENT | any_other_flag_you_might_need);
pagedirectory[0xC0000000 >> 12] = mkpgentry(0x9B000, PG_PRESENT | PG_SYSTEM | any_other_flag_you_might_need);
set_cr3_and_update_gdt(0x9D000);
Along with setting CR3's value, you'll have to clear the base of all the previous segments and reload segment registers so that, when set_cr3_and_update_gdt returns, any memory references now use the 0-based code and data segment and that the page tables perform the translation required to make 0x10000 appear at 0xC0000000.
With GRUB From the Start
GRUB will load your kernel at the desired physical target address, but it will leave segments in the GDT with a 0 base and it will not enable paging.
With GRUB Later On
Even with GRUB it's still possible to use TimRobinson's GDT trick: just set up a "fake" GDT then jump to your C code and enable paging. See Higher Half With GDT