Remove swapped-out usage in --site-per-process.

chrome/renderer/content_settings_observer.cc

Index: chrome/renderer/content_settings_observer.cc
diff --git a/chrome/renderer/content_settings_observer.cc b/chrome/renderer/content_settings_observer.cc
index 3769a85cce7d56ff343fd4b4d6e9e4a73460401f..8cc8acc07f4cc65f8a51d5fdf696dde9ae698c39 100644
--- a/chrome/renderer/content_settings_observer.cc
+++ b/chrome/renderer/content_settings_observer.cc
@@ -170,11 +170,12 @@ ContentSettingsObserver::ContentSettingsObserver(
   ClearBlockedContentSettings();
   render_frame->GetWebFrame()->setContentSettingsClient(this);
 
-  if (render_frame->GetRenderView()->GetMainRenderFrame() != render_frame) {
+  content::RenderFrame* main_frame =
+      render_frame->GetRenderView()->GetMainRenderFrame();
+  if (main_frame && main_frame != render_frame) {
     // Copy all the settings from the main render frame to avoid race conditions
     // when initializing this data. See http://crbug.com/333308.

[Charlie Reis, 2015/06/04 00:02:11]

Unrelated: Will this bug regress for OOPIFs, since they won't find the main
frame?  I wonder if we should file a OOPIF bug for it.

[nasko, 2015/06/04 14:57:12]

I think it deserves a bug and attention, as with OOPIF we aren't guaranteed a
main frame in the process. Filed crbug.com/496670 to ensure we don't lose track
of this.

-    ContentSettingsObserver* parent = ContentSettingsObserver::Get(
-        render_frame->GetRenderView()->GetMainRenderFrame());
+    ContentSettingsObserver* parent = ContentSettingsObserver::Get(main_frame);
     allow_displaying_insecure_content_ =
         parent->allow_displaying_insecure_content_;
     allow_running_insecure_content_ = parent->allow_running_insecure_content_;