[gentoo-alt] rap prefix and old kernel (<3.10)

[Francois Bissey]

So I went and re-installed my ppc64 rap prefix. After

some back and forth I finally reached the end successfully \o/



I have a few things to say but I decided to split this particular

issue because it has wider implication.



The OS is SuSE linux Enterprise Server 11 SP1 and it has kernel

2.6.32. SP3 and SP4 have 3.0 and would also be affected I believe.

2.6.32 is also the kernel of redhat/centos 6 which is still in wide

use.



So rap tried to install glibc 2.25 which was still the highest 

version available at the time. Compile went apparently fine until

the install stage. When glibc told me politely but firmly:

FATAL KERNEL TOO OLD

And looking back at the log, it was initially configured to use

2.6.32 as the oldest kernel but the configure script changed that

to 3.10 (If I remember correctly) later during configuration.



After masking 2.25 I did try 2.24, the message was not as clear, there was

a “FATAL” but I don’t think my kernel got insulted that time but it was

the same place. And configure again rewrote the required minimal kernel

version supported.



Masking 2.24 I got down to 2.23 which in the end worked (after some massaging

more about that later). 



So message #1 if you have an older kernel (<3.10) the latest glibc

available to you is 2.23. That probably should be in the wiki.



The compiling actually didn’t go without problems. Courtesy of binutils-2.29.1.

There was two issues:

* http://lists.busybox.net/pipermail/buildroot/2017-August/199812.html

glibc 2.23 and 2.24 were affected by this issue, I think 2.25 was OK.

glibc takes user patch from etc/portage/patches so adding this was easy.

* an assembly bug specific to ppc that plagued all versions and apparently

was spotted starting binutils-2.27.51. I can send a link to that one as well

but it potentially just affects me unlike the first one.



# message #2 if you are stuck with an older glibc, it may need patching

to deal with newer binutils and possibly other toolchain elements.



Cheers,

François