Hi, to those who are pushing the -fno-omit-frame-pointer change: Are you aware that neither that flag nor even -mno-omit-leaf-frame-pointer actually guarantee that every leaf function is going to carry a frame pointer, as required for your backtraces?
See for yourself: https://godbolt.org/z/TjzTsWoWT The only way to get GCC to generate a frame pointer for this function is to not use any optimization at all (-O0, or just leave the flags empty, which unfortunately defaults to -O0 in GCC). I have tried: * just -O2 * -O2 -fno-omit-frame-pointer * -O2 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer and all of them produce just imul+mov+ret, which means that an unwinder based purely on frame pointers will NOT be able to locate the caller of this function if the profiler snapshots it in the middle. Another issue is that even if the function does have a frame pointer, it takes time to set up the frame pointer (push rbp; mov rbp, rsp), and it also needs to be popped at the end (pop rbp) before returning (ret), so a randomly sampled snapshot can always happen to be taken in the short time window where the frame pointer is not ready, which will also lead to the caller being unable to be located. This inherently makes purely frame- pointer-based unwinding unreliable. Frame pointers sound like a simple solution to unwinding, but they are not. They are no complete replacement for unwinding information. Kevin Kofler _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue
