Skip to contentSkip to conversation

Discontiguous memory fun

DISCUSSION Jon Masters 3 participants 4 messages Oct 2, 2005 – Oct 3, 2005 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

Replying to my own post...

Quoted 5 lines · Jon Masters

I've got a hardware platform based on an ARM922, a bit like Excalibur,
which has a nice large hole in memory:

0x0000_0000 - 32MB (or whatever) of SDRAM.
0x8000_0000 - 64MB (or whatever) of SDRAM.

I fixed it using a bodge similar to PHYS_OFFSET and disabled the lower bank of memory if the kernel was booted at the higher address since it was extremely difficult to play with the half-support for discontig in old 2.4 ARM kernels (all of the existing boards are just playing a game of luck with their memory/node mappings) without a working BDI.

I'd love to know what the state of discontig memory is like on 2.6 series ARM kernels and highmem too for that matter, but I've not had chance to look at it (I'm usually a ppc guy).

Jon.

#2 lore raw
Quoted 7 lines · Jon Masters

Replying to my own post...

On 9/1/05, Jon Masters <[email protected]> wrote:

I'd love to know what the state of discontig memory is like on 2.6 series ARM kernels and highmem too for that matter, but I've not had chance to look at it (I'm usually a ppc guy).

It works fine.

The node mapping is performed by macros in asm-arm/arch-*/memory.h.

#3 lore raw
Quoted 5 lines · Marc Singer
Quoted 4 lines · Jon Masters

Replying to my own post...

On 9/1/05, Jon Masters <[email protected]> wrote:

Quoted 7 lines
Quoted 3 lines

I'd love to know what the state of discontig memory is like on 2.6 series ARM kernels and highmem too for that matter, but I've not had chance to look at it (I'm usually a ppc guy).

It works fine.

The node mapping is performed by macros in asm-arm/arch-*/memory.h.

Hmmm...as it was in 2.4. I'll take a look for my general interest.

Jon.

#4 lore raw

--Jon Masters <[email protected]> wrote (on Monday, October 03, 2005 03:10:28 +0100):

Quoted 11 lines · Marc Singer
Quoted 10 lines · Jon Masters
Quoted 7 lines

Replying to my own post...

On 9/1/05, Jon Masters <[email protected]> wrote:

I'd love to know what the state of discontig memory is like on 2.6 series ARM kernels and highmem too for that matter, but I've not had chance to look at it (I'm usually a ppc guy).

Quoted 5 lines
Quoted 3 lines

It works fine.

The node mapping is performed by macros in asm-arm/arch-*/memory.h.

Hmmm...as it was in 2.4. I'll take a look for my general interest.

If you're going to mess with it, please use sparsemem, not discontigmem. The intent is to deprecate and delete the latter.

M.