[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ARMv4-support in armel/squeeze?

+++ Christoph Biedl [2010-12-20 20:06 +0100]:
> Thanks for your detailled answer.
> Wookey wrote...
> > arm will stay around for
> > another 18 months or so as 'oldstable', and then v4 machines will be
> > pretty-much unsupported in Debian unless someone steps up to support
> > them.
> Do I hear a call for volunteers?

Without volunteers it's guaranteed that nothing will happen on the v4
front. But volunteers is not sufficient. A case needs to be made that
it is worthwhile. Obviously an unoffical port could be maintained
outside Debian by anyone who was sufficiently enthused. 

> > This is unfortunate for people still running v4 hardware (strongarm,
> > pretty much - do have something else which is v4?), but continuing to
> > support v4 hardware imposes significant overhead on people running
> > later hardware. There is plenty of pressure to move to v5-or-later
> > (i.e. drop v4t too). Ubuntu has already moved to v7 or later only.
> Do I understand correctly the main reason for dropping arm is lack of
> manpower to support several flavours of ARM, while the toolchain and
> things like that are still in good shape?

It's not just manpower - it's Debian archive space too, but yes
manpower for maintaining the port (or set of ports) remains surprisingly
limited given the number of people using it.

> > Debian has always tried hard to support as much hardware as possible,
> > but it does (often, and in this case) affect newer hardware so at some
> > point one has to let hardware drop off the bottom.
> FWIW, my box is a "Raidsonic NAS IB-4220-B" (machtype 2038, FA526 CPU,
> StromARM if I read /proc/cpuinfo and Wikipedia right), and support is
> just about to enter the kernel. These machines were sold at least
> until two years ago, this is a rather short time compared to some
> 15-year-old pentiums that are still supported in i386 and will be for
> a long time.

Yes, you are quite right. They are rather awkward from our point of
view. Before that hardware the last v4 CPU still in use was the
StrongARM (SA1100) available from 1998 to 2004. Dropping that, whilst
still disappointing a few people, was not deemed unreasonable. How many
FA526 Debian users are there?

> > It's very difficult to support people with reliable old Netwinders at
> > one end, and others who want to use all the shiny stuff on their new
> > arm netbooks/tablets at the other witout building 6 different arm
> > distros. What an appropriate compromise is, is currently a hot topic.
> Um, really six (the ARM world is new to me so I might miss irony)?

Something like that. A minimum set to support everybody to some
reasonable level of adequacy might be v4, v4t, v5, v7, v7+VFP. 

People could also make a case for v7+VFP+NEON, and maybe v6. This is
before considering hardfloat/softfloat options or core/SoC
optimisations e.g. for cortexA8 or A9, or thumb2 vs ARM. 

> Just to improve my understanding, how bad was the impact of a
> hypothetical "armv4" architecture on ARMv4t, i.e. without thumb
> support?

In the EABI port (armel) it siply wasn't supported (because it was
difficult, as clarified by Paul).

> What I'd like to see is continued support for my hardware. Not the
> full project of course, but those somewhat 300 packages that make up a
> typical installation for a NAS box. 

One of the issues with the current Debian infrstructure is that if
there is a port then it must build all 17,000 packages - there is no
facility to say 'only build these N that people actually care about'. 

If most of these arm variants were only mini-distros with a
much-reduced package set then supporting them would be a much smaller

'Subarches' is the term that has been coined to cover this idea of
rebuilding a subset of packages for a different flavour
(ISA/optimisation). I'm not aware of any concrete proposals for how
this should be implemnted. You are not the only person interested in
this, so it may happen as some point.

> Now that I see Debian will not
> provide it, I can - and probably will - do that for myself, but I'd
> take the opportuniy to switch to EABI. If there's broader interest for
> an "armv4eabi" architecture, this should be done in a way other people
> can use it, too (and I'd need a lot more knowledge about toolchains,
> buildds and the like, then).

That would be apprecated by FA526 and strongarm users. It may be
possible to use the Debian ports infrastructure, which removes a lot of the
pain. Ask the admin. 

All you need to do is build a toolchain with Paul's v4 patches,
defaulting to use one of the v4 options, then build for eabi/armel (and
fix the packages that break - there will be some). 

This is not a trivial job, but neither is it epic. Getting the initial
system will be the hardest bit, because existing armel packages will
not run on your hardware.

Principal hats:  Linaro, Emdebian, Wookware, Balloonboard, ARM

Reply to: