Hi Steven, I completely forgot about this issue until recently when I encountered it again. Instrumenting the ftrace_ops_assist_func symbol and some other symbol seems to be causing problems.
Placing kretprobes like in the following configuration crashes my kernel (4.16.0-rc5) on a Qemu/KVM virtual machine: config 1: echo "r:event_1 __fdget" >> kprobe_events echo "r:event_2 ftrace_ops_assist_func" >> kprobe_events config 2: echo "r:event_1 __fdget_pos" >> kprobe_events echo "r:event_2 ftrace_ops_assist_func" >> kprobe_events config 3: echo 'r:event_1 arch_dup_task_struct' >> kprobe_events echo 'r:event_2 ftrace_ops_assist_func' >> kprobe_events config 4: echo 'r:event_1 sys_open' >> kprobe_events echo 'r:event_2 ftrace_ops_assist_func' >> kprobe_events Here is my kernel config [1]: In a previous email [2], you mentioned that you would like to add the ftrace-related symbols to a section to un-blacklist them all at once on demand but wanted to discuss it at Linux Plumbers. Do you still think that it's the right approach? I can easily test any patch regarding this issue. [1] http://paste.ubuntu.com/p/BJWvgMnW8z/ [2] https://lkml.org/lkml/2017/7/14/568 Thank you, 2017-07-14 14:29 GMT-04:00 Steven Rostedt <rost...@goodmis.org>: > On Fri, 14 Jul 2017 10:58:35 -0400 > Francis Deslauriers <francis.deslauri...@efficios.com> wrote: > >> This function is called when a kprobe is hit. Thus it should be >> blacklisted to prevent kprobe to be triggered by kprobes. >> >> Signed-off-by: Francis Deslauriers <francis.deslauri...@efficios.com> >> --- >> kernel/trace/ftrace.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c >> index b308be3..c473d9b 100644 >> --- a/kernel/trace/ftrace.c >> +++ b/kernel/trace/ftrace.c >> @@ -36,6 +36,7 @@ >> >> #include <trace/events/sched.h> >> >> +#include <asm/kprobes.h> >> #include <asm/sections.h> >> #include <asm/setup.h> >> >> @@ -5739,6 +5740,7 @@ static void ftrace_ops_assist_func(unsigned long ip, >> unsigned long parent_ip, >> preempt_enable_notrace(); >> trace_clear_recursion(bit); >> } >> +NOKPROBE_SYMBOL(ftrace_ops_assist_func); > > Continuing from what I said in the other email, this is fixing a > symptom and not the problem. The real fix will be much more involved. I > have a good idea on how to accomplish it too. > > -- Steve > > >> >> /** >> * ftrace_ops_get_func - get the function a trampoline should call > -- Francis Deslauriers Software developer EfficiOS inc.
