[Buildroot] [PATCH] glibc: add version 2.24

[Yann E. MORIN <yann.morin.1998@free.fr>]

Khem, All,

On 2016-08-09 12:42 -0700, Khem Raj spake thusly:
> 
> > On Aug 9, 2016, at 12:33 PM, Yann E. MORIN <yann.morin.1998@free.fr> wrote:
> > 
> > Thomas, All,
> > 
> > On 2016-08-09 10:49 +0200, Thomas Petazzoni spake thusly:
> >> On Mon, 8 Aug 2016 18:42:09 +0200, Yann E. MORIN wrote:
> >> 
> >>>> It is not written completely clearly: they first say that on
> >>>> x86/x86-64, 2.6.32 is sufficient, but they conclude that 3.2 is the
> >>>> minimum version on all architectures. It would be good to clarify this
> >>>> aspect.
> >>> 
> >>> What this means, at least what I understand it means, is that:
> >>> 
> >>>  - glibc 2.24 will not *run* on kernels more ancient than 3.2, except on
> >>>    x86/x86_64, when it will not run on kernels older than 2.6.32.
> >>> 
> >>>    This is a runtime dependency.
> >>> 
> >>>  - glibc needs kernel headers 3.2 (or later) for all architectures,
> >>>    even for x86/x86_64.
> >>> 
> >>>    This is a built-time dependency.
> >> 
> >> OK. From Buildroot's perspective, all what matters is the build-time
> >> dependency. If the 3.2 kernel headers are needed to build glibc 2.24,
> >> then glibc 2.24 should depend on headers >= 3.2.
> > 
> > Yes. That's what Vicente did in his v3 of his patch:
> >    https://patchwork.ozlabs.org/patch/657159/
> > 
> >>>> Regardless of this detail, this means we will have to encode this
> >>>> dependency somehow. Indeed, we still have people using kernels older
> >>>> than 3.2 I believe on various platforms.
> >>> 
> >>> On all but x86_x86_64, yes.
> >> 
> >> I don't get this. From my point of view, the x86/x86_64 "exception" is
> >> of no use to us: all we are interested in is the build-time dependency.
> > 
> > Well, I mixed it a bit in my head... What I was thinking was that we
> > should still allow building a glibc-based toolchain even if the running
> > kernel was older than 3.2 for x86/x86_64. But we have no way to enforce
> > that: we have no _AT_LEAST_X_Y symbols for the running kernel (not that
> > we want to have them).
> 
> you can in theory use --enable-kernel=VERSION to configure glibc in order
> to indicate minimum kernel it can run on, older than that will be flagged
> at runtime practically,

Well, yes indeed, but do we care to support that? There are cases where
we do not and can't know the version of the kernel the user will be
running on (because he builds the kernel outside of Buidlroot, for
whatever reason).

So we just let glibc configure itself for the oldest kernel it can
support on the targeted arch.

I even doubt it would make sense to have such a knob for size reasons.
glibc is already big, and I doubt removing fallbacks for older kernels
would gain much more than a few KiBs. At best.

> are there BR users who want to stick to such
> an old kernel on x86 and use latest BR with glibc 2.24 ?

Yes. There are BSPs out there that are stuck on 3.0. I even know of some
that are even stuck to some 2.6.18-ish... :-(

Regards,
Yann E. MORIN.

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'