On Fri, Jan 29, 2016 at 10:22:41AM -0500, Steven Rostedt wrote:
> > No, that doesn't fix it (current Linus' master):
> >
> > # uname -a
> > Linux p2345007 4.5.0-rc1-00032-g26cd83670f2f #26 SMP Fri Jan 29 15:39:47
> > CET 2016 s390x s390x s390x GNU/Linux
> >
> > # cat stack_max_size
> > 4496
> >
> > # cat stack_trace
> > Depth Size Location (0 entries)
> > ----- ---- --------
>
> Ah this stack trace. This is different than what Pratyush Anand is
> fixing. This is the stack tracer which uses function tracing. The other
> fix is to deal with stack traces from events, specifically kprobes. As
> he stated, you test with:
>
> # echo stacktrace > /sys/kernel/debug/tracing/trace_options
> # echo "p kfree" >> /sys/kernel/debug/tracing/kprobe_events
> # echo 1 > /sys/kernel/debug/tracing/events/kprobes/enable
Yes, I was aware of that, however the output of the trace file looked like
this:
bash-1516 [003] d.s. 44.002233: p_kfree_0: (kfree+0x0/0x1a8)
bash-1516 [003] d.s. 44.002233: <stack trace>
bash-1540 [000] d... 44.002335: p_kfree_0: (kfree+0x0/0x1a8)
bash-1540 [000] d... 44.002336: <stack trace>
bash-1540 [000] d... 44.002338: p_kfree_0: (kfree+0x0/0x1a8)
bash-1540 [000] d... 44.002338: <stack trace>
Which made me check the stack tracer which did not work anymore.
> But anyway, I'm curious, does this fix the issue for you?
>
> -- Steve
>
> diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
> index dda9e6742950..db1c26c385c7 100644
> --- a/kernel/trace/trace_stack.c
> +++ b/kernel/trace/trace_stack.c
> @@ -125,6 +125,9 @@ check_stack(unsigned long ip, unsigned long *stack)
> break;
> }
>
> + if (i == stack_trace_max.nr_entries)
> + i = 0;
> +
With this patch the trace file now contains:
bash-1543 [001] d... 31.273255: p_kfree_0: (kfree+0x0/0x1c0)
bash-1543 [001] d... 31.273256: <stack trace>
=> pipe_release
=> __fput
=> task_work_run
=> do_notify_resume
=> system_call
bash-1543 [001] d... 31.273400: p_kfree_0: (kfree+0x0/0x1c0)
bash-1543 [001] d... 31.273400: <stack trace>
=> load_elf_binary
=> search_binary_handler
=> do_execveat_common.isra.14
=> SyS_execve
=> system_call
Which looks much better. The stack tracer also works again!
If the above is supposed to be the final fix, please feel free to add
Tested-by: Heiko Carstens <heiko.carst...@de.ibm.com>
Thanks a lot, Steven!
