Created attachment 791541 [details] example memleak.py output on the same tumbleweed install The same kind of effect regarding the lack of stack trace entries can be seen when using memleak.py from the bcc tools (as discussed in bug 1116448). Kernel side, this uses: \ bpf_get_stackid \ get_perf_callchain In comparison, I think ftrace uses \ __ftrace_trace_stack When I looked earlier, it seems that they both call into the same code, unwind_start, unwind_next_frame, ... I don't know why the difference. I've attached a sample memleak.py output which shows partial stack traces similar to the perf output in comment 0.