Clarify comment on FrameTree::AddFrame.

This behavior was updated in https://crrev.com/299939, but the comment
didn't explain why the method might return null.

BUG=none
TEST=No behavior change.

Review URL: https://codereview.chromium.org/938073002

Cr-Commit-Position: refs/heads/master@{#317084}

content/browser/frame_host/frame_tree.cc

@@ -150,6 +150,8 @@ RenderFrameHostImpl* FrameTree::AddFrame(FrameTreeNode* parent,
   // it is in the same SiteInstance as the parent frame. Ensure that the process
   // which requested a child frame to be added is the same as the process of the
   // parent node.
+  // We return nullptr if this is not the case, which can happen in a race if an
+  // old RFH sends a CreateChildFrame message as we're swapping to a new RFH.
   if (parent->current_frame_host()->GetProcess()->GetID() != process_id)
     return nullptr;
 

content/browser/frame_host/frame_tree.h

@@ -70,6 +70,10 @@ class CONTENT_EXPORT FrameTree {
   void ForEach(const base::Callback<bool(FrameTreeNode*)>& on_node) const;
 
   // Frame tree manipulation routines.
+  // |process_id| is required to disambiguate |new_routing_id|, and it must
+  // match the process of the |parent| node.  Otherwise this method returns
+  // nullptr.  Passing MSG_ROUTING_NONE for |new_routing_id| will allocate a new
+  // routing ID for the new frame.
   RenderFrameHostImpl* AddFrame(FrameTreeNode* parent,
                                 int process_id,
                                 int new_routing_id,