Re: S390: Add support for vdso getcpu symbol.

[Stefan Liebler <stli at linux dot vnet dot ibm dot com>]

On 05/04/2016 06:33 PM, Carlos O'Donell wrote:
> Stefan,
>
> Looks good to me, but I have one question.
>
> Why are the versions of these new symbols LINUX_2.6.29?
>
> If the kernel really wants to follow the userspace conventions
> for symbol versioning, the version of a newly added symbol is
> that of the Linux version that released the symbol.
>
> This allows users and developers to know exactly which upstream
> public kernel version exported the symbol, and talk sensible about
> the exported ABI/API in terms of these versions. Right now having
> them all be LINUX_2.6.29 is useful only for adding compat symbols
> at newer versions.
>
> At this point it's too late, you have a release with these symbols
> at these versions, and you can't change it.
>
> Martin,
>
> Did you consider using LINUX_4.5 version for these symbols?

I had realised this version mismatch, too.
I had asked Martin about the version number before I posted the patch
and his first thought was, that it reflects the version, the symbol was 
introduced. Then he realised, that it does not in this case.
But he won't change that for getcpu.

As observation:
E.g. <kernel-src>/arch/powerpc/kernel/vdso64/vdso64.lds.S was extended 
with __kernel_getcpu, __kernel_time.
E.g. <kernel-src>/arch/tile/kernel/vdso/vdso.lds.S was extended with 
__vdso_clock_gettime.
These additions were made without a new version number.
The corresponding glibc commits logically does not introduce a new 
linux-version to find vdso-symbol in _libc_vdso_platform_setup(), too.


For documentation:
Link to discussion "Should Linux VDSO be using symbol version based on 
the released kernel?"
(https://www.sourceware.org/ml/libc-alpha/2016-05/msg00056.html)

I've pushed the patch.
Thanks.

Bye
Stefan