= Proposal topic =
rebuild required: any gcc-compiled[*] multi-threaded tool is suspect
[*] if gcc-compiled using glibc-headers-2.14.90-12.999 or newer
= Overview =
Starting with glibc-2.14.90-12.999, any program compiled with gcc used headers declaring many functions with the "leaf" attribute. That attribute tells gcc -O2 that it can perform code motion across affected function calls. When that code motion moves guarded code out of a critical section, it exposes a multi-threaded program to a race-condition bug. With git, we saw that it would sometimes be easy to reproduce, and others, not. In one case, I was seeing a failure only 10% of the time.
As Jakub mentioned in http://sourceware.org/bugzilla/show_bug.cgi?id=13344#c2 this may even affect any printf-using code that registers a hook function that is defined in the same compilation unit where such a hook-using printf invocation appears.
Other basic tools may be affected. I've seen xz compression fail recently, though have no proof that this was the cause. Even sort is multi-threaded.
For many more details, see: http://bugzilla.redhat.com/747377 and http://thread.gmane.org/gmane.comp.version-control.git/184184 and http://sourceware.org/bugzilla/show_bug.cgi?id=13344
= Problem space =
Release F16 without this particular bug.
= Solution Overview =
Fix glibc and rebuild affected packages.
Adding glibc-owner to this ticket.
a) Is there a fix?
b) How can we identify things that need rebuild when there is?
we're handling this in the release management loop. we do have tools in place and can turn this around reasonably fast. jim meyering is working on a fix for glibc, and dennis will rebuild affected packages once the update is done.
we have a list of ~90 packages that were built against the affected glibc that are already in f16 'stable', and will rebuild those first. then we will rebuild any packages that are pending to be pushed through the freeze to fix blocker or NTH issues. then we'll rebuild updates-candidate builds, which will likely be a bigger number.
I've submitted a build for glibc (jim doesn't have provenpackager privs, so I just applied his changes and submitted):
http://koji.fedoraproject.org/koji/taskinfo?taskID=3462105
Once it's done I'll submit it as an update, and we can start the rebuilds.
update is submitted:
https://admin.fedoraproject.org/updates/glibc-2.14.90-14
build has been tagged into the buildroot and dgilmore is working on the mass rebuild. all builds from now on should be unaffected.
I'll put -14 into rawhide too.
This got taken care of, AFAIK.
Login to comment on this ticket.