Re: [Ksummit-discuss] Fwd: Rough notes from testing unconference

[Olof Johansson <olof@lixom.net>]

On Sun, Aug 24, 2014 at 11:37 AM, Guenter Roeck <linux@roeck-us.net> wrote:
> On 08/24/2014 11:15 AM, Geert Uytterhoeven wrote:
>>
>> On Sun, Aug 24, 2014 at 7:12 PM, Grant Likely <grant.likely@secretlab.ca>
>> wrote:
>>>
>>> Cross building the kernel isn't the problem. That's easy. Getting a
>>> rootfs and the tests into the rootfs can be hard, depending on the
>>> platform (a problem for some embedded targets).
>>
>>
>> In the past, I have used OpenWRT to build the rootfs for the test
>> target.
>>
> I very much prefer Grant's direct method - it ensures that the available
> test code is automatically integrated into the root file system, and that
> there are no architecture / target dependencies.
>
> For my part I have used various means to create root file systems for my
> own tests, depending on the architecture. I see it as significant
> improvement
> to have it all together in a single place, in our control, and to have it
> as simple as possible.

Yep. Unfortunately, having a cross-toolchain that includes libc is a
significant complexity improvement over just having one that can build
a kernel. I wouldn't mind having pre-built rootfs:es available just
like we have toolchains today, but that won't help me when it comes to
building the kselftest binaries.

I have traditionally been using Segher's buildall
(http://git.infradead.org/users/segher/buildall.git) to build and
populate toolchains locally, and that will no longer be enough. Unless
someone convinces Segher to improve his scripts, of course. :)


-Olof