Cubox? Linux 3.7 released, bringing generic ARM support

Scott Sullivan scott-lxSQFCZeNF4 at public.gmane.org
Fri Dec 14 17:13:12 UTC 2012


On 12/14/2012 11:54 AM, D. Hugh Redelmeier wrote:
> | From: Thomas Milne <thomas.bruce.milne-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org>
[...]
>
> What is it slow doing?  Why?
>
> As far as I know, the major thing that *might* get sped up in another
> distro is floating point.  Most programs don't care about FP
> performance, but a few care a lot -- do you run any of those?

The Cubox SoC is a ARMv7 instruction set, which means it has a mandatory 
hardware floating-point unit. It's also why Ubuntu works on it as 
Ubuntu's developers have choose only to support ARMv7 architecture and 
above.

> My perception is that these little devices are sometimes slow because:
[...]
> - not enough RAM (new desktops start at 4G; most ARM systems stop at
>    1G)

The Cubox is fortunately equipped with 1G of ram, but core CPU speed is 
much lower then it's peers at a partly 800Mhz.

http://archlinuxarm.org/platforms/armv7

> | Will this 3.7 kernel mean that Linux will now run on Cubox with normal
> | install? I know Debian won't have the 3.7 kernel for a while.
>
> Who knows.  It is a future direction so it may not be embraced by old
> products.  Too many embedded systems are "fire and forget" by their
> producers.  I don't have any knowledge of the Cubox folks intention.
> You could ask them.

The ease of installation we see on x86 dervied systems has to do with 
the consistent boot environment provided by the BIOS. There is no 
equivalent in the ARM SoC world and this is complicated by the genuine 
competition in the market as their are literally dozen of CPU design 
companies, and hundreds of board design companies. To contrast, there is 
only really 3 x86 CPU designers (Intel, AMD and Via) and only much 
smaller cabal of board designers.

The closest that comes a BIOS is uboot, but that also does the job of a 
boot loader (so it's part BIOS part Grub). But there are plenty of board 
designers that don't use uboot, or fork versions of uboot with limited 
and inconsistent feature sets that they don't merge back. Then there are 
devices like the Raspberry that does something completely different.

So, the community is still coming to grips with all of this. The 3.7 
kernel work is the first major foundation effort to bring make the 
kernel agile in the face of all of this and allow the maintainers and 
board designers to build on a leverage more common code.

-- 
Scott Sullivan
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists





More information about the Legacy mailing list