Re: [PATCH v6 tip 2/8] tracing: attach BPF programs to kprobes

[Peter Zijlstra <peterz-wEGCiKHe2LqWVfeAwA7xHQ@public.gmane.org>]

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);