Possible bug in ftrace_profile_enable_event

[Paul Mackerras]

I was looking through kernel/trace/trace_event_profile.c and I saw
this code:

static int ftrace_profile_enable_event(struct ftrace_event_call *event)
{
	char *buf;
	int ret = -ENOMEM;

	if (atomic_inc_return(&event->profile_count))
		return 0;

	if (!total_profile_count++) {
		buf = (char *)alloc_percpu(profile_buf_t);
		if (!buf)
			goto fail_buf;

		rcu_assign_pointer(trace_profile_buf, buf);

		buf = (char *)alloc_percpu(profile_buf_t);
		if (!buf)
			goto fail_buf_nmi;

		rcu_assign_pointer(trace_profile_buf_nmi, buf);
	}

	ret = event->profile_enable();
	if (!ret)
		return 0;

	kfree(trace_profile_buf_nmi);
fail_buf_nmi:
	kfree(trace_profile_buf);
fail_buf:
	total_profile_count--;

...

So we only allocate trace_profile_buf and trace_profile_buf_nmi if
total_profile_count was zero on entry, but if we get an error returned
from event->profile_enable(), we free them both unconditionally,
regardless of the value of total_profile_count.  That seems wrong.  Is
there a subtle reason why that is the right thing to do?

(Also, is kfree the appropriate counterpart to alloc_percpu?)

Paul.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/