Description of problem:
We would like to update you with a bug fix we have identified on the
latest Btgeneric 5334 (ia32exec.bin) we have delivered you on Sep
15th â targeted for EL3 U4 â
Bug description: the translator maintains processor ordering
compatibility of the source IA-32 application by either using
affinity solution or using strong order version of the load/store IPF
instructions ( acquire-release versions of the load / store
instructions ). We have identified a bug in one of the optimizations
that in certain case the generated code did not use the acquire-
release versions of the instructions when it actually needed to.
The bug was discovered during a code review â we have not seen
failures in our test environment due to this bug â even on MP systems
running stress testing -
Bug fix: the fix was by eliminating the problematic optimization
Although this bug did not duplicate on real tests, we still highly
recommend the fix
Version-Release number of selected component (if applicable):
i tried to attach the binary with the fix, but its size exceeds the
limit. we sent the binary thru email to firstname.lastname@example.org and
email@example.com on Sep 28. Please let me know if it was not
We did receive the new binary; thanks. We are requesting that Intel
shoulder the testing effort with this new binary (as with previous
ones!). In particular, we request that you perform benchmark tests as
well as gcc, glibc, and mozilla-type tests.
Elena Zannoni has already spoken to Geoff Gustafson, Intel engineer
onsite at RH, about testing; Geoff can repeat some of the simpler
testing that we we (with gdb and a simple ia32 program (break main;
run) but can't go too far beyond that.
We've posted the latest rpms (ia32el-1.1-3 in dist-4E-lacd and
ia32el-1.1-2.EL3 in dist-3.0E-lacd-U4) at the following locations:
Please grab these rpms and test as rigorously as possible by next
Wednesday, 6 October. We're quickly approaching RHEL 4 Beta 2 package
freezee and need to ensure that testing is done ASAP. Many thanks.
Latest packages are now ia32el-1.1-6.EL3 (for RHEL3 U4 lacd) and
ia32el-1.1-7 (for RHEL4 lacd).
Can we close this bug?
Yes, I'm closing it.