In rhbz#1752252 we tried to build innoextract-1.8. It failed on ppc64le: https://koji.fedoraproject.org/koji/taskinfo?taskID=38684075 with error:
collect2: fatal error: ld terminated with signal 11 [Segmentation fault], core dumped
https://kojipkgs.fedoraproject.org//work/tasks/4100/38684100/build.log
All other arches are fine. I re-ran the build, failed again. I built it against F31, the same problem. Then I built it in mock locally:
$ mock -r fedora-31-ppc64le --forcearch ppc64le --rebuild ./innoextract-1.8-1.fc32.src.rpm
and it worked fine. Then I ssh'd to a test machine ppc64le-test.fedorainfracloud.org, and I also built it there just fine (in mock, against F31).
ppc64le-test.fedorainfracloud.org
There seems to be some problem with specific ppc64le machine(s) in koji infra, but I don't know exactly what. Perhaps it's running out of memory?
System seems to be having a lot of errors
[1946607.467291] ld64.so.2[15758]: floating point exception (8) at 7fffaf3801c4 nip 7fffaf3801c4 lr 10001cb8 code 3 in libm.so[7fffaf370000+110000] [1946607.467302] ld64.so.2[15758]: code: 60000000 e90282e8 554a00ee 5529056a 7d295378 79290020 e9480058 7c080066 [1946607.467305] ld64.so.2[15758]: code: 714a0400 7d294378 7c090166 41820030 <fffe058e> 74692000 41820018 fc00048e
[1980091.125497] c-index-test[29862]: unhandled trap (5) at 7fff8946d040 nip 7fff8946d040 lr 7fff8946ca08 code 1 in libclangLex.so.7[7fff893b0000+100000] [1980091.125510] c-index-test[29862]: code: 80e90010 6ce67361 2e867263 41960014 38e00001 0c870000 419e0200 4bfffdd4 [1980091.125514] c-index-test[29862]: code: 88e90014 2e870068 4096ffe8 38e00000 <0c870000> 4bffffe4 2b890004 409efa24
I have updated and rebooted the server to see if that clears the issue.
I've done another scratch build and it still shows the same error: https://koji.fedoraproject.org/koji/taskinfo?taskID=38690348
OK looking at the system.. it is on a different guest on a different hardware server... so my worry about bad RAM on ppc09-01 goes away
[Thu Oct 31 15:30:09 2019] ld.gold[24354]: segfault (11) at 7645656d616e34 nip 1322927a8 lr 132292774 code 3 in ld.gold[132210000+3e0000] [Thu Oct 31 15:30:09 2019] ld.gold[24354]: code: e9230040 2c0a0000 41820060 e91f0260 e95f0268 7929e102 e8810028 7d485050 [Thu Oct 31 15:30:09 2019] ld.gold[24354]: code: 7d4a2674 7c295040 4080015c 792926e4 <7d28482e> 2c290000 4182012c e8640290
So all the ppc boxes on both 8 and 9 architectures have a long list of segfault 11's from various builds. Not sure if that helps any other than.. you aren't going to get this to build for a while.
I think we should ask @sharkcz about this. It is unlikely our hardware, but it could be firmware or something.
Metadata Update from @kevin: - Issue priority set to: Waiting on Assignee (was: Needs Review)
I'm on an OpenPOWER conference these days, but no Power HW at hand :-) I'll look on Monday when I'm back. But it sounds weird if a koji build segfaults and mock build succeeds.
OK, so I seem to be able to reproduce it locally, F-30 (bare-metal Power9) and ld.gold segfaults ... possible workaround - don't use ld.gold (AFAIK it's not being developed anymore), but regular ld
diff --git a/innoextract.spec b/innoextract.spec index 980bbb3..69efd3d 100644 --- a/innoextract.spec +++ b/innoextract.spec @@ -38,6 +38,7 @@ non-windows systems without running the actual installer using wine. -DCMAKE_INSTALL_DATAROOTDIR="%{_datadir}" \ -DCMAKE_INSTALL_MANDIR="%{_mandir}" \ -DCMAKE_INSTALL_BINDIR="%{_bindir}" \ + -DUSE_LDGOLD=FALSE \ . make %{?_smp_mflags}
is the fix
Hello, the build still fails: https://koji.fedoraproject.org/koji/taskinfo?taskID=39012403
Any update to this? Thanks.
-1 shouldn't fail as it excluded ppc64le, but I have pushed -2 that uses ld.bfd instead of ld.gold. The root cause in ld.gol hasn't been fixed yet.
Thanks, @sharkcz. I completely misread your previous comment that this can be fixed in the innoextract spec file.
the segfault has been reported as https://bugzilla.redhat.com/show_bug.cgi?id=1772935
Metadata Update from @kevin: - Issue close_status updated to: Upstream - Issue status updated to: Closed (was: Open)
Login to comment on this ticket.