Am 09.11.2010 um 11:02 schrieb Artyom Tarasenko:
On Mon, Nov 8, 2010 at 10:53 PM, Tarl Neustaedter tarl-b2@tarl.net wrote:
On 2010-11-6 1:46 PM, Tarl Neustaedter wrote:
On 2010-11-6 6:43 AM, Andreas Färber wrote:
[...] Hm. It's right for reg, with is one cell (0) only. But it logically doesn't fit the MMU's available property. So it's telling me we need to special-case that somehow for 5/5. Which in turn means that we'll need to pass the #address-cells and #size-cells via stack in 3/5 to cover this use case.
"The property values are as defined for the standard “reg” format, with single-cell virtual addresses. The regions of virtual address space denote the virtual address space that is currently unallocated by the Open Firmware and is available for use by client programs." (IEEE 1275 3.6.5)
Hmm. It's "recommended", and I believe that is incorrect for SPARC 64. I'll check on monday - I remember recently seeing that the / virtual-memory available property used two-cell virtual addresses. I'll check what it does with MMUs.
Duhh... Sun systems haven't had an MMU node for a _long_ time. I think the SS5 had one, but I literally can't find a Sun old enough to have such a node. Never mind...
I guess the first one which had it was Ultra-1. SS-5 didn't have it:
ok .version Release 2.15 Version 5 created 95/03/29 14:21:55 ok show-devs
[snip]
Just so that we don't misunderstand each other, this is not about a node named "MMU" that we might see in show-devs. It's about some node - anonymous or named - that gets referenced by the mmu property in the /chosen node.
On my PowerMac G3 it points to /cpus/PowerPC,750@0 (the CPU), for example.
Andreas