[OpenAFS-devel] Re: Compiling openafs on arm64/aarch64 architecture

Micheal Waltz mwaltz@qualcomm.com
Wed, 3 Sep 2014 15:03:25 -0700


On 09/03/14 14:45, Andrew Deason wrote:
> The next release would be 1.6.10, currently in progress. This might be a
> bit late for that, since I'm not sure exactly what we should do. The
> patches I gave you would not go in the release as-is, since presumably
> they would break other arm systems.

Sorry, I was looking at 1.6.7 which ships with Ubuntu 14.04 and meant 
1.6.10 or a future release. We don't have an immediate need for it now, 
just looking towards the future and I completely understand about fully 
going through regression testing.

> It seems like maybe aarch64 should be its own sysname, though, from what
> I've seen here? Since it doesn't define __arm__ when compiling,
> apparently, and it has its own gnu toolchain triplet component. Do you
> (or others) agree? Should it be arm64_linux26, aarch64_linux26,
> armv8a_linux26, or something else?
I would like to see aarch64/arm64 at least as it's own sysname. From 
what I've experienced with using armv6 (raspberrypi) and armv7 
(ifc6410), they both have the arm_linux26 sysname, but binaries can be 
incompatible. If for whatever reason we need to use armv6, armv7, and 
armv8 binaries in the same AFS cell we would have all sorts of @sys issues.

This is further confusing since ARM uses aarch64, but 
Debian/Ubuntu/Linux Kernel uses arm64.

My vote is for arm64_linux26 since it's more recognizable, is the same 
name as the arch in the Linux kernel, and stays with the AFS convention 
of having arm_linux26 for 32-bit and arm64_linux26 for 64-bit.

-- 
Micheal Waltz
Unix Configuration Management (go/ucm)
Qualcomm Inc.
Phone: 858-845-6083
Cell: 858-882-7079