NetBSD/sys/arch/amiga/AMIGA-MMU

26 lines
1.4 KiB
Plaintext

The HP300 version uses a kernel which is loaded into the range 0xfffxxx... but
is linked for range 0x000.. This causes the problem that with switching the
MMU to the mapped kernel, the PC becomes invalid. The HP version solves the
problem by mapping 1:1 the last page of physical memory into virtual memory.
On the Amiga, there *is* memory at PA 0x0, so we might as well use it, causes
much less grief and weirdness in locore.s. However, since memory down there
is CHIPMEM, inherently slower than FASTMEM, and badly needed for framebuffer
storage space, we'll copy the kernel over to the first bank of FASTRAM, and
when enabling the MMU, the kernel will execute in FASTRAM, although the PC
stays the same.
This strategy has the big advantage (compared to the solution in Amiga MACH)
that we can link the kernel absolutely to VA 0, for all memory models, since
we'll never execute the kernel in the FASTMEM space while the MMU is turned
off.
This strategy also means, that we don't have to relocate any addresses while
bootstrapping the mmu!!
Initialization of the MMU happens in amiga_init.c. This file is quite a mess,
I have generated it trying to understand what's happening in the hp300 locore.s
file. I think, it should be able to handle MMU initialization much cleaner
now that we don't have to think about relocation until the MMU is enabled.
If you need your daily bit of horror, take a look at amiga_init.c...