Version-Release number of selected component (if applicable): elfutils-0.137-3.el5 How reproducible: 100% Steps to Reproduce: 1. eu-unstrip -n --core=some_coredump Actual results: eu-unstrip: No modules recognized in core file Expected results: list of module, filenames and build_ids Additional info: Attaching the coredump on which it fails.
I found no attachment. Please try this case on RHEL6/Fedora elfutils.
Created attachment 439324 [details] coredump
Attached the coredump, the case fails even on the F13 using the attached core from RHEL5.
RHEL5 core dumps do not have partial segments for ELF text mappings. This leaves no information that is accessible from a cold start to see what the modules in the process were. To handle a core dump that has no part of the text segments, you need to use "eu-unstrip -n -e exe --core core" where exe is the original executable that was run to create this core. You can do "echo 0x33 > /proc/.../coredump_filter" to change the core dump behavior (to match the behavior of RHEL6 and Fedora>=8). Then you may have enough information for plain "eu-unstrip -n --core core" to at least list the modules from the link_map list. But even then, RHEL5 binaries do not have build IDs, so just names will be as good as it gets. RHEL5 eu-unstrip indeed has bugs in handling core files even with a partial-segments dump. It is unlikely we will have a business justification to do an elfutils update in RHEL5 just to backport those fixes. Fedora/RHEL6 eu-unstrip suboptimally suppresses the file names it finds in this kind of core file, so you only get the soname or basename as the module name. 0.149 will fix that to print those full file names too.
Thanks for the info, I'll try to experiment with it to see how it will work with ABRT. Btw, the bussiness justification might be the backporting of ABRT to RHEL5 which relies on elfutils. So I'm setting this as a blocker for backporting ABRT just to track it. Thank you, Jirka
Closing report of non-bug.