Red Hat Bugzilla – Bug 813095
Firefox freezes on Fedora 17 Alpha for PPC64
Last modified: 2012-09-10 11:35:04 EDT
Created attachment 577842 [details]
strace -s 1024 -ff -o firefox /usr/lib64/firefox/firefox
Description of problem:
Firefox freezes after a few minutes of usage. I'm not sure what triggers the issue, though a few minutes browsing random pages has been enough to reproduce the issue.
I've attached an strace. Firefox seems to be stuck, there is a LOT of messages in the strace like this:
madvise(0xfff574ce000, 4096, MADV_DONTNEED) = -1 EINVAL (Invalid argument)
Version-Release number of selected component (if applicable):
A few minutes browsing random pages has been enough to reproduce the issue.
Steps to Reproduce:
1. Install Fedora 17 Alpha release for PPC64 from http://download.fedoraproject.org/pub/fedora-secondary/releases/test/17-Alpha
2. Install updates: yum update --nogpgcheck
3. Run firefox
4. Browse some random pages for a few minutes (1 or 2 minutes has been enough to me)
Firefox freezes (UI stops responding and/or updating)
Firefox UI should continue working.
I've tested only using vnc. I don't have a graphical terminal to this machine.
Created attachment 577846 [details]
gdb full backtrace
Attached gdb backtrace.
Moved upstream - https://bugzilla.mozilla.org/show_bug.cgi?id=746112
From upstream BZ:
Do you happen to be using JACK on that machine?
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=660960 is a similar issue triggered by JACK locking memory it doesn't own.
I'm unable to reproduce the hang with Firefox 10.0.4 for RHEL which contains the workaround patch. You may need to build and test the package on Fedora ppc64 too. And don't forget to disable all JIT there, like it's in the RHEL package.
The workaround works fine on RHEL6:
so this issue has to be something Fedora 17 specific.
A few comments on what I've done trying to narrow down this issue:
1) Since RHEL6.2 has a newer version of firefox (10.0.4 as opposed to 10.0.1 on f17), I built a firefox/xulrunner-10.0.4 on f17. It still freezes.
2) Installing RHEL6.2 binaries of xulrunner makes firefox to work fine on f17. Here is what I did to get RHEL binaries of xulrunner on f17:
rpm -Uvh --force --nodeps xulrunner-10.0.4-1.el6_2.ppc64.rpm (from RHEL repo)
rpm -Uvh firefox-10.0.4-1.fc17.ppc64.rpm (my custom firefox-10.0.4)
ln -s /usr/lib64/libhunspell-1.3.so.0 /usr/lib64/libhunspell-1.2.so.0 (otherwise firefox refuses to start)
3) As the previous configuration worked fine, I built the src.rpm of xulrunner from RHEL on f17. The following patches from fedora source were needed to get the RHEL src.rpm of xulrunner building on f17 (probably because of different gcc version):
The RHEL src.rpm built on f17 (as describe above) also freezes. Since the above patches are quite minimal, this leads us to believe that it must be something in the build environment.
4) Talking to Martin in the IRC:
(11:30:25 AM) stransky: gustavold, or take the rhel one and disable optimalization in the spec file - in MOZ_OPT_FLAGS. plus change --disable-optimize/--enable-debug in spec
(11:30:50 AM) stransky: gustavold, you need to remove the -O2 flag from MOZ_OPT_FLAGS
Only removing -O2 from MOZ_OPT_FLAGS doesn't help. It still freezes.
Adding --disable-optimize makes the build fail with:
/home/gustavold/rpmbuild/BUILD/xulrunner-10.0.4/mozilla-esr10/xulrunner/installer/../../toolkit/mozapps/installer/precompile_cache.js:73: TypeError: Cc is null
This issue is related to the gcc version used in f17. I installed gcc-4.6.3-2.fc16 with its dependencies in my f17 mock environment used to compile xulrunner. Firefox works fine on f17 using the xulrunner compiled with gcc-4.6.3-2.fc16.
This is how I installed gcc 4.6 on my f17 mock environment:
<mock-chroot>[root@coho gcc]# rpm -ivh --force cpp-4.6.3-2.fc16.ppc64.rpm gcc-c++-4.6.3-2.fc16.ppc64.rpm libgomp-4.6.3-2.fc16.ppc64.rpm libstdc++-devel-4.6.3-2.fc16.ppc64.rpm gcc-4.6.3-2.fc16.ppc64.rpm gmp-4.3.2-4.fc16.ppc64.rpm libstdc++-4.6.3-2.fc16.ppc64.rpm
Should I file a new bug for the --disable-optimize and --enable-debug build issues?
Created attachment 582135 [details]
the patch from mozilla bug 691898 backported to xulrunner 12
Created attachment 582136 [details]
fix build on secondary arches
Update: We have firefox and xulrunner working locally here but it requires two fixes.
1. RH bugz 824463 which is a gcc fix to fix a small regression.
2. the nodecommit patch in https://bugzilla.mozilla.org/show_bug.cgi?id=746112
The patch has been added to git.
(In reply to comment #11)
> The patch has been added to git.
Works on powerpc with this patch. Thanks folks - I can use ff12 now!
The patch we need is https://bug746112.bugzilla.mozilla.org/attachment.cgi?id=616080
The patch you added on your last commit (9ccd78f5a81ca7c84a8381e99b473a39e0f07914) is not needed now that we have fixed gcc.
The confusion comes from the fact that there are two different bugs being discussed on that bugz (probably my fault).
Okay, uploaded to CVS.
It should be already fixed on F17, please reopen if not.