Try not to let the AUHAL driver get stuck sending silence.

media/audio/mac/audio_auhal_mac.cc

Index: media/audio/mac/audio_auhal_mac.cc
diff --git a/media/audio/mac/audio_auhal_mac.cc b/media/audio/mac/audio_auhal_mac.cc
index 4d0d61430f15bd7f13e3ca0905c7065163d68286..ed3b193416712945ff0a3f56e8291f7718845435 100644
--- a/media/audio/mac/audio_auhal_mac.cc
+++ b/media/audio/mac/audio_auhal_mac.cc
@@ -72,6 +72,7 @@ AUHALStream::AUHALStream(
       volume_(1),
       hardware_latency_frames_(0),
       stopped_(false),
+      notified_for_possible_device_change_(false),
       input_buffer_list_(NULL) {
   // We must have a manager.
   DCHECK(manager_);
@@ -159,6 +160,7 @@ void AUHALStream::Start(AudioSourceCallback* callback) {
   }
 
   stopped_ = false;
+  notified_for_possible_device_change_ = false;
   {
     base::AutoLock auto_lock(source_lock_);
     source_ = callback;
@@ -209,6 +211,16 @@ OSStatus AUHALStream::Render(
     // In this case either audio input or audio output will be broken,
     // so just output silence.
     ZeroBufferList(io_data);
+
+    // In case we missed a device notification, notify AudioManager
+    if (!notified_for_possible_device_change_) {
+      // We should really be using TryPostTask here since we don't want to lock,

[Chris Rogers, 2013/06/20 19:13:19]

This codepath is never happening in normal steady-state rendering, right?  It's
just at the end?
If so, then the tryLock() is probably not that important.  If this codepath is
part of normal steady-state rendering then it's very important.

[DaleCurtis, 2013/06/20 19:25:30]

Correct, it's a one time thing too, so you're right TryLock is probably not
necessary here.  I'll remove the comment.

+      // but MessageLoopProxy doesn't expose this, only MessageLoop.
+      notified_for_possible_device_change_ = true;
+      manager_->GetMessageLoop()->PostTask(FROM_HERE, base::Bind(
+          &AudioManagerMac::HandleDeviceChanges, base::Unretained(manager_)));
+    }
+
     return noErr;
   }