Changeset 201929 in webkit

Timestamp:

Jun 10, 2016 11:01:54 AM (8 years ago)

Author:

Simon Fraser

Message:

[iOS WK2] Rare RELEASE_ASSERT under RemoteLayerTreeDrawingArea::flushLayers()
​https://bugs.webkit.org/show_bug.cgi?id=158622
rdar://problem/26609452

Reviewed by Tim Horton.

It's possible for a CADisplayLink to fire after being paused sometimes, possibly
when an app is running another CADisplayLink whose callback takes some time. When
this happens, RemoteLayerTreeDrawingAreaProxy could erroneously send a second
didUpdate() to the web process between commits, which would clear the m_waitingForBackingStoreSwap
flag too early, and allow a subsequent RemoteLayerTreeDrawingArea::flushLayers()
to proceed when the m_pendingBackingStoreFlusher was still flushing.

Fix by preventing two didUpdates from being sent from the UI process between
commits.

Not easily testable.

• UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.h:
• UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.mm:

(WebKit::RemoteLayerTreeDrawingAreaProxy::commitLayerTree):
(WebKit::RemoteLayerTreeDrawingAreaProxy::didRefreshDisplay):

Location:

trunk/Source/WebKit2

Files:

• ChangeLog (modified) (1 diff)
• UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.h (modified) (1 diff)
• UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.mm (modified) (2 diffs)

trunk/Source/WebKit2/ChangeLog

@@ +1 @@
+2016-06-10  Simon Fraser  <simon.fraser@apple.com>
+
+        [iOS WK2] Rare RELEASE_ASSERT under RemoteLayerTreeDrawingArea::flushLayers()
+        https://bugs.webkit.org/show_bug.cgi?id=158622
+        rdar://problem/26609452
+
+        Reviewed by Tim Horton.
+        
+        It's possible for a CADisplayLink to fire after being paused sometimes, possibly
+        when an app is running another CADisplayLink whose callback takes some time. When
+        this happens, RemoteLayerTreeDrawingAreaProxy could erroneously send a second
+        didUpdate() to the web process between commits, which would clear the m_waitingForBackingStoreSwap
+        flag too early, and allow a subsequent RemoteLayerTreeDrawingArea::flushLayers()
+        to proceed when the m_pendingBackingStoreFlusher was still flushing.
+
+        Fix by preventing two didUpdates from being sent from the UI process between
+        commits.
+        
+        Not easily testable.
+
+        * UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.h:
+        * UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.mm:
+        (WebKit::RemoteLayerTreeDrawingAreaProxy::commitLayerTree):
+        (WebKit::RemoteLayerTreeDrawingAreaProxy::didRefreshDisplay):
+
 2016-06-10  Carlos Garcia Campos  <cgarcia@igalia.com>
 

trunk/Source/WebKit2/UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.h

@@ -92 +92 @@
     RemoteLayerTreeHost m_remoteLayerTreeHost;
     bool m_isWaitingForDidUpdateGeometry { false };
+    bool m_haveSentDidUpdateSinceLastCommit { false };
 
     WebCore::IntSize m_lastSentSize;

trunk/Source/WebKit2/UIProcess/mac/RemoteLayerTreeDrawingAreaProxy.mm

@@ -220 +220 @@
     m_webPageProxy.layerTreeCommitComplete();
 
+    m_haveSentDidUpdateSinceLastCommit = false;
+
 #if PLATFORM(IOS)
     [m_displayLinkHandler schedule];
@@ -390 +392 @@
         return;
 
+    if (m_haveSentDidUpdateSinceLastCommit)
+        return;
+    
+    m_haveSentDidUpdateSinceLastCommit = true;
+
     // Waiting for CA to commit is insufficient, because the render server can still be
     // using our backing store. We can improve this by waiting for the render server to commit