Hi Pedro. From a compiler point of view we take the 'do no harm' approach. The compiler is intended for the Cortex-A series and all new work is focused on that area, but the ARMv5 support should be no worse than the upstream version. Give the toolchain a try as the bug fixes and performance improvements could be worthwhile.

I'll let others reply on the test head area.

-- Michael

On Wed, Aug 25, 2010 at 7:24 AM, Pedro I. Sanchez psanchez@fosstel.com wrote:

Hello,

I'm looking for some advice on the feasibility of using Linaro on one of my projects based on an AT91SAM9260 Atmel processor, which has an ARM926EJ-S ARMv5 processor core.

I see that the Linaro builds do not support v5 and I wonder, what would be involved in "recompiling" Linaro with v5 support? Is there anyone else looking at this possibility? Any suggestions?

Thanks,

-- Pedro

---

linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev

On Thu, Aug 26, 2010 at 07:01:50PM -0300, Christian Robottom Reis wrote:

Note that Debian does maintain a v5 build at the moment; if they transition to using the Linaro toolchain for their armel build then it may be the best possible scenario for v5 validation. I'm copying Matthias who may have more information on the current status of that project.

Debian is currently targeting v4t rather than v5. There was a recent discussion on debian-arm about bumping the compat baseline to v5, and the consensus seemed to be that there wasn't enough performance improvement to justify the effort and the loss of compatibility with some widely-deployed v4t developer kit.

But as long as the goal is to have something that can /run/ on v5 hw (as opposed to something that runs /optimally/ on v5 hw), Debian fits the bill.

On Mon, Aug 30, 2010, Pedro I. Sanchez wrote:

Well, I have several v5 boards I could use for testing. Bootstrap code and U-Boot issues are sorted out already and I could concentrate on Linaro's toolchain and rootfs components initially. I can also allocate a 64-bit Ubuntu 10.04 machine to do builds. Would this help?

I'd need help getting started with the whole build process but I'll be glad to put some effort into this. I'm really interested in getting the v5 port up and running soon.

So there are two ways we currently do builds: * automated native builds * manual cross-builds

The former is the classical Debian/Ubuntu archive setup, and we currently provide only an ARMv7A Thumb2 VFPv3D16 reference archive based on the Ubuntu armel port. Debian provides a base armel archive built for ARMv4T. If you want to go that route it's the one which will allow you to build the largest amount of packages for v5, but you'll need to rebootstrap the port starting from Debian because the Ubuntu binaries don't run on your hardware. The process is basically: - copy Debian archive into Linaro armv5 bootstrap_archive - setup chroot from bootstrap_archive - grab Linaro toolchain sources and change the defaults to armv5 instead of armv7, build by hand, put in target_archive - for all source packages - pull source package from source_archive - build using bootstrap_archive + target_archive binaries - upload to target_archive Eventually, you'll have more and more packages in the target_archive, and after some rebuild cycles, a bootstraped armv5 archive from sources out of source_archive.

The latter approach is using cross-compilation, which is technically possible with Debian/Ubuntu tools, but very fragile and almost not automated at all. There are multiple tools to help you cross-compile from existing binaries, so that you don't have to break all the build-deps loops, but it's still a very manual process. Peter Pearse (ppearse on IRC) is working his way doing an armv5 and an armv6 cross-rebuild; you should ping him to hear how he is doing it exactly! :-) Cc:ing him.