Re: Runtime failure running sh:qemu in -next due to 'sh: fix copy_from_user()'

From: Rob Landley
Date: Wed Sep 28 2016 - 22:36:41 EST


On 09/18/2016 10:17 AM, Rich Felker wrote:
> On Sat, Sep 17, 2016 at 11:40:28PM -0500, Rob Landley wrote:
>>
>>
>> On 09/16/2016 09:23 PM, Guenter Roeck wrote:
>>> On 09/16/2016 04:32 PM, Rich Felker wrote:
>>>>> 4.6.3 from kernel.org.
>>>>
>>>> That is utterly ancient and probaby very buggy. I would recommend 5.x+
>>>> or at the very least 4.7 or 4.8.
>>>>
>>> Unfortunately that is the latest one available from kernel.org :-(.
>>> I'll try to build one myself.
>>
>> Rich, you really, really need to get an actual release version of
>> https://github.com/richfelker/musl-cross-make posted.
>
> What do you mean? Binaries? There are release tags, though it would
> probably be a good time to make another one.
>
> But this project (musl-cross-make) is not needed for building kernels
> -- stock gcc, any modern-ish version, should work fine. The canonical
> way (from prior to my involvement) to build sh* kernels is to use a
> gcc that supports any ISA level, and this can be done without multilib
> libgcc since the kernel provides its own libgcc replacement functions.

The above was an example of somebody using a broken toolchain because
there isn't a known-good reference toolchain for the architecture, which
the kernel maintainer is known to regression test against. Having such a
thing might help people distinguish "bug in kernel" from "bug in gcc".

> Rich

Rob