Re: [PATCH v4 tip 3/7] tracing: allow BPF programs to call ktime_get_ns()

On Sat, Feb 28, 2015 at 2:20 AM, Peter Zijlstra [off-list ref] wrote:

quoted

+static u64 bpf_ktime_get_ns(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
+{
+     return ktime_get_ns();
+}

Please use ktime_get_mono_fast_ns() instead. If you ever want to allow
running BPF stuff from general tracepoints and the like you need to
be NMI safe.

good point. will do.
Currently in trace_call_bpf() I have:
if (in_nmi()) /* not supported yet */
  return 1;
to tackle nmi after all pieces are in place.
Likely nmi-safe programs will have access to reduced set
of helper functions.