Fix exec instrumentation: only for kernels 3.4.0 and higher

[lttng-modules.git] / lttng-context-vppid.c

diff --git a/lttng-context-vppid.c b/lttng-context-vppid.c
index 3d55fc5d6a09750f23bb08561e9f8d9725ffb75e..06377a3098ff7f72b992f0c9848f693fbc10f7dc 100644 (file)
--- a/lttng-context-vppid.c
+++ b/lttng-context-vppid.c
@@ -48,11 +48,21 @@ void vppid_record(struct lttng_ctx_field *field,
        pid_t vppid;
 
        /*
-        * nsproxy can be NULL when scheduled out of exit.
+        * current nsproxy can be NULL when scheduled out of exit. pid_vnr uses
+        * the current thread nsproxy to perform the lookup.
         */
+
+       /*
+        * TODO: when we eventually add RCU subsystem instrumentation,
+        * taking the rcu read lock here will trigger RCU tracing
+        * recursively. We should modify the kernel synchronization so
+        * it synchronizes both for RCU and RCU sched, and rely on
+        * rcu_read_lock_sched_notrace.
+        */
+
        rcu_read_lock();
        parent = rcu_dereference(current->real_parent);
-       if (!parent->nsproxy)
+       if (!current->nsproxy)
                vppid = 0;
        else
                vppid = task_tgid_vnr(parent);