Hi Ryan, On Mon, Jul 03, 2017 at 03:39:35PM -0700, Ryan Tandy wrote: > Hi debian-powerpc, > > Would a ppc64(el) porter be able to help me look at #866122? I have > requested a porterbox account but it's not gone through yet, and I am unable > to reproduce the issue at all in a qemu VM.
You can also request a VM on the minicloud at http://openpower.ic.unicamp.br/minicloud/ if you wish. They are usually quick on creating accounts. > The openldap test suite is failing on ppc64 and ppc64el in stretch and > unstable: the slapd-mtread helper program segfaults (exit 139) in a certain > test case. > > It looks like the builds tend to succeed on jessie's kernel and fail on > stretch's kernel: In fact, this problem seems to reproduce once in a while, thus, I would not trust that it might be related to the kernel/gcc combination at this very beginning. I am suspecting that it might be related to the amount of threads created and the order, but I do not have a conclusion yet. Still investigating. > apt-get build-dep openldap > apt-get source openldap > cd openldap-*/ > DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage -T build > cd debian/build/tests > ./run -b bdb test060-mt-hot Nice. I was able to reproduce it and debug it further. The problem seems to be related to a invalid branch/jump, the the next address is not memory mapped, thus the segfault. The new address is completely random, and definitely is wrong. The link register (LR), which is register that shows the return of the branch (similar to call() on amd64) is always constant when ALSR is disabled. Other than that, I also saw a stack corruption, which caused the backtrace to be completely bogus. Anyway, myself and a colleague are still investigating this problem. I will keep you informed of the progress of this issue. Thanks, Breno
