ppc64 kernel config changes heads up

[Al Dunsmuir]

On Friday, June 26, 2015, 6:44:15 PM, Peter Robinson wrote:
>> With all due respect, you do realize that several companies actually
>> still manufacture and sell 32-bit PowerPC processors, correct? Do you
>> think this move will be well-met by them?

This change only affects the Fedora builds, not the support for kernel
builds  for  32-bit  ppc in general. As such it is indeed a reasonable
step for Peter to make at this time.

> With all due respect do you realise we actually dropped support for
> "32-bit PowerPC processors" for all release post Fedora 20?

> Fedora 20 went EOL this week but it was known when Fedora 20 went GA
> in December of 2013, around 18 months ago, that ppc-32 was EOL. This
> isn't a shock or unknown, it was widely publicised that that if people
> were interested in 32 bit PPC in general that they needed to pick up
> the architecture maintainership, and in fact the kernel hasn't had the
> bits in place to build for ppc-32 for some time, in the ppc32 case
> it's merely cleaning up the left over bits of something, in terms of
> the Fedora kernel, that has long been dead.

Josh  wanted  to  remove  the PPC32 kernal build support last year. At
that time, I asked him to keep it in place, and I do still plan to get
a ppc32 remix running AFTER I get Mac G5 boot fixed in the main Fedora
ppc64  secondary  arch.

The  PPC64 effort is running late (largely due to limited work time, a
lot of things to learn and F20 (build env) having issues after updates
that prevented network installs. Peter resolved the latter in April.

The reason Josh wanted it removed was twofold:
- It was dead within Fedora AND any ACTIVE remix
- It was extremely complex, building both a SMP and non-SMP kernel.

Ubuntu  dropped  the  non-SMP  ppc32  kernel from their builds several
years ago. Non-SMP ppc32 machines simply use the SMP kernel with 1 CPU
at  a  tiny  performance  penalty.  It  saves  a lot of build time and
complexity!  Even  IBM  did  this  for AIX 5.1 (as witnessed on my B50
32-bit box).

We do not want to remove all ppc32 support, as a remix should be built
with  minimum  delta from the primary release.  In this case, it makes
sense.

When  I'm actively working on ppc32, I plan to add a new spec patch to
do  the  builds  like Ubuntu and will base it on Peter's updates. That
would  be  used  for  builds  of  the remix, and would be suitable for
inclusion  in  the main Fedora spec with their review/agreement purely
to minimize the deltas from primary release.

>>> On Jun 26, 2015, at 2:40 PM, Peter Robinson <pbrobinson at gmail.com> wrote:
>>>
>>> Hi All,
>>>
>>> If you don't care about powerpc you can stop reading :-)
>>>
>>> If you do.... I went to do what I thought was a basic fix up to the
>>> kernel builds on POWER to get rid of the last bits of the ppc32
>>> legacy, on my list for some time, and it has become more of a slightly
>>> epic slash and burn!
>>>
>>> The resulting output kernel configs are for all intents and purpose
>>> unchanged but the git diff is some what epic.
>>>
>>> The top level of the changes are:
>>>
>>> Core changes that apply too all kernels (ppc64/ppc64p7/ppc64le) reside
>>> in config-powerpc64-generic
>>>
>>> config-powerpc64 -> legacy platforms that are older than POWER7 and
>>> their dependencies (here lies the ghosts of PowerPC G5 and friends).
>>>
>>> config-powerpc64le and config-powerpc64p7 contain anything specific to
>>> POWER7 and newer that might affect older platforms, or LE/BE specific
>>> to p7 and later
>>>
>>> I've removed anything that pertains to PPC32 and Cell processors (the
>>> later were all disabled now anyway) but if the kernel docs were wrong
>>> and they do by chance work on some other 64 bit platform that still
>>> works do let me know.
>>>
>>> So the only thing of real note is the ppc64p7 and ppc64le configs no
>>> longer have any old IDE configs enabled or any of the legacy Apple
>>> bits pulled in, I don't believe that should cause any pain what so
>>> ever, I only bring this up because it's the only thing of note on
>>> diffs of before/after resulting config diffs.
>>>
>>> So ultimately please test kernel builds going forward. There's a
>>> scratch build here [1] until we get an official one to test. There's
>>> no perf packages, that appears broeken... even on vanilla pre commit
>>> :)
>>>
>>> For those that are morbidly interested in the diff (and don't have the
>>> Fedora kernel package check out) you can see that here [2].
>>>
>>> Ultimately the large git commit looks quite scary but the actual
>>> change to the 3 configs are minimal and I believe this should make
>>> things cleaner moving forward :-)
>>>
>>> I look forward to the testing feedback including any further suggested
>>> cleanups to the new configs, oh and please test :-)
>>>
>>> Peter
>>>
>>> [1] http://ppc.koji.fedoraproject.org/koji/taskinfo?taskID=2582705
>>> [2] http://pkgs.fedoraproject.org/cgit/kernel.git/commit/?id=c9229e16082bdc9b2f74a08bfdc77bcfed51df18