On Tue, Mar 10, 2015 at 09:18:48PM -0700, Alexei Starovoitov wrote:

+unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
+{
+	unsigned int ret;
+	int cpu;
+
+	if (in_nmi()) /* not supported yet */
+		return 1;
+
+	preempt_disable_notrace();
+
+	cpu = raw_smp_processor_id();
+	if (unlikely(per_cpu(bpf_prog_active, cpu)++ != 0)) {
+		/* since some bpf program is already running on this cpu,
+		 * don't call into another bpf program (same or different)
+		 * and don't send kprobe event into ring-buffer,
+		 * so return zero here
+		 */
+		ret = 0;
+		goto out;
+	}
+
+	rcu_read_lock();

You've so far tried very hard to not get into tracing; and then you call
rcu_read_lock() :-)

So either document why this isn't a problem, provide
rcu_read_lock_notrace() or switch to RCU-sched and thereby avoid the
problem.

+	ret = BPF_PROG_RUN(prog, ctx);
+	rcu_read_unlock();
+
+ out:
+	per_cpu(bpf_prog_active, cpu)--;
+	preempt_enable_notrace();
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(trace_call_bpf);