http://bugzilla.suse.com/show_bug.cgi?id=1118066 Bug ID: 1118066 Summary: perf outputs partial kernel backtraces Classification: openSUSE Product: openSUSE Tumbleweed Version: Current Hardware: Other OS: Other Status: NEW Severity: Normal Priority: P5 - None Component: Kernel Assignee: kernel-maintainers@forge.provo.novell.com Reporter: bpoirier@suse.com QA Contact: qa-bugs@suse.de Found By: --- Blocker: --- Created attachment 791540 --> http://bugzilla.suse.com/attachment.cgi?id=791540&action=edit example output on sle12-sp3 and tumbleweed Forked from bug 1116448 comment 15. As noted there:
Doesn't answer your question but I had a look earlier this week and the aborted stack traces seem related to the unwinder used by bpf/perf. When I was testing recording events with stacktraces, it seems that ftrace gets "fuller" stack traces than perf, but I don't know why since they both call into the same unwinding code, unwind_start(), unwind_next_frame(),... They use different glue code but I didn't have time to investigate more.
No clue why it would work better with ubuntu than opensuse on the same kernel.
Not to get sidetracked on this issue too much, but for example, compare the output of trace-cmd record -F -e kmem:kmalloc -T uname trace-cmd report and perf record -e kmem:kmalloc -g uname perf script I tested on tumbleweed with 4.18.15 and on SLE12-SP3 and, in both cases, perf lacks many entries at the bottom of the kernel stack trace. -- You are receiving this mail because: You are on the CC list for the bug.