[lsb-discuss] LSB and ARM

Dallman, John john.dallman at siemens.com
Thu May 19 06:15:41 PDT 2011


Mats wrote:

> 1. LSB is very much a model of defining a common base,
> and ARM deployments have had wide variability - you're
> left with a question of "which ARM".

Since I've just been finding this out, I might as well explain a
little more. ARM do not make chips. They define the architecture,
design implementations, and license them. Other companies actually
make them - any decent chip fabricator can do it.

Most ARMs, for a long time, have gone into various kinds of embedded
systems. Those tend to only run the software supplied with them, so
instruction set standardisation wasn't a priority. So a culture has
evolved which is quite alien to people working with Intel, AMD or
PowerPC chips, of lots and lots of slight variations of the ARM
instruction sets. Many embedded system makers have requested extra
instructions for specialised purposes, or felt that instructions
they did not need could be left out. They've also had their ARM
cores built into chips that also contained significant amounts of
other functionality, as is common for micro-controllers.

Since a key feature of LSB is that it defines binary compatibility,
these variations in instruction set complicate the issue. Recently,
as ARMs have been used in systems that can load considerable extra
software, such as smart-phones and tablets, standardisation has
become more important: Apple obviously want the successors of the
current iPad/iPhone models to be able to run the same instruction
set.

ARM have been defining various standardised instruction sets for
different target profiles. But that's still instruction sets, plural.

> It may be that the
> current armv7hl is that target, although that certainly won't
> run on the entire range of ARM devices out there - is it
> okay to leave out those that don't have harfp capability?



More information about the lsb-discuss mailing list