Re: Dropping support for the smallest armel machines
On Wed, Jun 26, 2013 at 04:39:15PM +0100, Ben Hutchings wrote:
> On Wed, Jun 26, 2013 at 03:13:48PM +0200, Arnaud Patard wrote:
> > Ben Hutchings <ben@decadent.org.uk> writes:
> >
> > > We have a recurring problem with building kernels for armel: three
> > > flavours (iop32x, ixp4xx, orion5x) require the kernel image size to be
> > > less than 1.4-1.5 MB in order to fit into a fixed flash partition.
> > >
> > > As more features continue to be added to Linux and cannot always
> > > configurable as a module, it is necessary to override and disable them
> > > on these three configurations[1]. I don't think this is sustainable
> > > unless someone who particularly cares about these older platforms steps
> > > up to take on this task.
> > >
> > > The iop32x and ixp4xx hardware appears to been discontinued in 2008. If
> > > we remove these flavours now, they will still be supported in Debian 7
> > > until 2016. I think 8 years of support is pretty good.
> >
> > btw, iop32x is used on n2100 which are used on buildd/porter boxes. If
> > we stop supporting it, I'm not sure how the DSA people will react about
> > that.
> [...]
Currently only 3 out of 10 are N2100s; the remainder are Marvell
development boards running the mv78xx0 flavour.
> Bet they're slower than QEMU versatile emulation on an x86. And DSA
> loves virtual machines. :-)
This speed comparison might not be true, and I know people worry about
accuracy of emulation and being able to reproduce architecture-
specific bugs.
The other option that has been suggested repeatedly is to put armel
chroots on ARMv7 hardware. Unaligned accesses behave differently on
v7, but they weren't consistent between different v5 implementations
(http://www.heyrick.co.uk/armwiki/Unaligned_data_access) so I don't
think this is critical.
Ben.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Reply to: