Possible PPC kernel bug on builders

[Jerry James]

I posted about this 5 days ago on ppc list [1], but have had no
response.  I tried to get some attention on #fedora-ppc today, also
with no success.  I am failing miserably to get the attention of any
of the PPC folks, so I am trying email here to see if this will work.

GCL is failing to build sometimes on ppc64/ppc64le.  I have had both
successful [2] and unsuccessful [3] builds.  Both of those build logs
contain lots of extra debugging junk I threw in so I could gather
information on what is going on.  Here's the critical part.  The
"uname -a" invocation I added at the top of %build always shows a 3.14
kernel when the build succeeds and always shows a 3.16 kernel when the
build fails.

I believe there is a bug in the 3.16 ppc64/ppc64le kernels on (some
of?) the builders.  The bug, whatever it is, is causing spurious
mprotect() failures.  If somebody can either confirm or deny that, I
am very interested.  I have not had this problem on any other
architecture.  Thank you.

Footnotes:
[1] https://lists.fedoraproject.org/pipermail/ppc/2014-October/003024.html
[2] http://ppc.koji.fedoraproject.org/koji/taskinfo?taskID=2138082
[3] http://ppc.koji.fedoraproject.org/koji/taskinfo?taskID=2138060
-- 
Jerry James
http://www.jamezone.org/