Re: Thread names in gdb7.7(using info threads command)
Pedro Alves <[email protected]>
| Newsgroups | gmane.comp.gdb.devel |
|---|---|
| Message-ID | <[email protected]> |
On 08/20/2014 07:08 AM, Sreejith M M wrote: > Thanks Pedro for the reply. > > To make things clearer, I get the thread names when I do live > debugging( Run the image from gdb and crash) > I am not getting the information only when I independently run the > image and take the core dump to be analyzed in gdb. > > What my question is , whether coredump will have the required > information or not. Unfortunately, the Linux core dump format is underdocumented. You'll have to web search and read sources to get more familiar with it, sorry. Start by looking at /usr/include/linux/elfcore.h, and looking at bfd's and gdb's sources for the structures defined there. Look around for info on core notes (NT_PRPSINFO, etc.). The elfutils sources would also be another good source. As to whether the info is there or not, I'm not sure; you'll have to experiment. Try using "eu-readelf -n" on the core, and see if the thread name is there somewhere. I think it might end up in PRPSINFO/fname, but I'm not 100% sure, and my time budget doesn't allow for investigating further at this point, sorry. > Why I am asking this again is because I did not > understand how a gdb patch (as you have mentioned) can read the right > data from core dump. Could you please explain a bit more? > > PS: I am newbie to gdb and dont understand the code quite well. If you > can give some pointers, I will try to implement the patch if it is > useful. See the target_thread_name calls in gdb/thread.c. When debugging against a live target, that ends up calling into linux-nat.c:linux_nat_thread_name, which extracts the thread name out of /proc/PID/task/TID/comm. When debugging a core dump, that target_thread_name call ends up in the corelow.c target_ops implementation instead. You'll need to add a to_thread_name hook to core_ops, that then calls info bfd to get the thread name somehow, assuming the info is indeed there. >> Or maybe it is, in NT_PRPSINFO / prpsinfo->pr_fname ? In that case, >> all we'd need is a gdb patch to make use of it. See the >> target_thread_name target method. We'd need an implementation in >> corelow.c. Thanks, Pedro Alves