Heres linus's take on it - "Paul (no@thanks.com) on 8/24/10 wrote:
>
>'ARM Ltd. announced extensions for virtualization and 40-bit
>addressing to the ARMv7a architecture, two of the key new
>aspects of ARM's upcoming Eagle core.'
Oh gods. Please don't say that it's another PAE mess. It's
a major pain, and having to specially map all the physical
pages that you can't fit directly in the virtual address
space leads to some nasty issues.
I hope they have a plan to move it towards a real 64-bit
architecture. But it sounds like ARM is doing the whole
"do all the mistakes x86 ever did" thing.
And the funny thing is, ARM people think that x86 is the
ugly architecture. But with x86 cleaning its stuff up,
and ARM apparently convinced it needs to do all the mistakes
x86 did, we'll soon see ARM as the real cesspool.
Yeah, PAE wasn't a good idea when Intel did it in 1996 and it's not a good idea now for ARM, but Linux already has highmem support so I'm sure ARM highmem patches are imminent and people who need more than 4GB RAM will use them.
Still there's a good reason not to use 64 bits: interconnections, transistor count, die size. You can't have a CPU that sips milliwats of power and make an architecture of a 50 Watts beast.
But to the extent they want to compete with Intel and AMD in server space they're going to have to go all in to a full 64 bit design. What is the Eagle design for? Is there any other target market that needs a PAE half-measure like this?
I hope they have a plan to move it towards a real 64-bit architecture. But it sounds like ARM is doing the whole "do all the mistakes x86 ever did" thing.
And the funny thing is, ARM people think that x86 is the ugly architecture. But with x86 cleaning its stuff up, and ARM apparently convinced it needs to do all the mistakes x86 did, we'll soon see ARM as the real cesspool.
from http://realworldtech.com/forums/index.cfm?action=detail&...