From: William C. <wc...@re...> - 2010-04-07 16:30:54
|
On 04/07/2010 11:38 AM, Maynard Johnson wrote: > William Cohen wrote: >> In Fedora 12 the upstream version of oprofile is not able to make use of the debuginfo files. When one is trying > > Will, do you mean the latest release -- 0.9.6 -- or a CVS pull? The problem also is reproduceable in the cvs pull version of oprofile. I have traced the problem down to the following change: PatchSet 2826 Date: 2008/08/07 09:01:15 Author: maynardj Branch: HEAD Tag: (none) Branches: Log: Fix a couple problems relating to overlay symbols for Cell SPE applications Members: ChangeLog:1.1828->1.1829 libutil++/bfd_support.cpp:1.10->1.11 libutil++/op_bfd.cpp:1.84->1.85 libutil++/op_bfd.h:1.52->1.53 I commented out the conditional in op_bfd::symbol_size() to avoid using op_bfd_symbol::symbol_endpos() and the code works. So there is something in that computation that doesn't work on Fedora12. > > to get more detailed information from an executable that has a separate debuginfo file with "opreport -l executable_name". Instead of getting a breakdown of samples for each function end up just getting the following error: >> >> opreport error: profile_t::samples_range(): start > end something wrong with kernel or module layout ? > > There was a regression in 0.9.5 where we completely failed to process separate debuginfo files, but that's been fixed in 0.9.6. Of course, the error you're seeing ("start > end") is a different problem. Nevertheless, I just did some testing on a distro that's shipping oprofile 0.9.6 and it handled separate debuginfo files OK. Are there any patches being applied to the oprofile rpm that's being built for F12? > > -Maynard the problem is observed in the pristine oprofile cvs check, so it doesn't appear to be due to patches added in the the fedora oprofile package. -Will |