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 [details] 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.