Changeset 284811 in webkit

Timestamp:

Oct 25, 2021, 12:10:10 PM (5 years ago)

Author:

Alan Coon

Message:

Cherry-pick r284644. rdar://problem/84153991

Rare crash under DisplayLink::displayLinkCallback()
​https://bugs.webkit.org/show_bug.cgi?id=232101
<rdar://84153991>

Reviewed by Tim Horton.

The crash was a divide by zero under m_currentUpdate.nextUpdate(), indicating that
m_currentUpdate.updatesPerSecond was zero. Previous assumptions that this was caused
by weird display configs were wrong. The actual issue is a race condition where
the callback can fire while we're still inside CVDisplayLinkStart(), or at least
before we've updated m_currentUpdate.

The fix is to initialize m_currentUpdate before we call CVDisplayLinkStart().

• UIProcess/mac/DisplayLink.cpp: (WebKit::DisplayLink::addObserver):

git-svn-id: ​https://svn.webkit.org/repository/webkit/trunk@284644 268f45cc-cd09-0410-ab3c-d52691b4dbfc

Location:

branches/safari-612-branch/Source/WebKit

Files:

• ChangeLog (modified) (1 diff)
• UIProcess/mac/DisplayLink.cpp (modified) (1 diff)

branches/safari-612-branch/Source/WebKit/ChangeLog

@@ +1 @@
+2021-10-25  Null  <null@apple.com>
+
+        Cherry-pick r284644. rdar://problem/84153991
+
+    Rare crash under DisplayLink::displayLinkCallback()
+    https://bugs.webkit.org/show_bug.cgi?id=232101
+    <rdar://84153991>
+    
+    Reviewed by Tim Horton.
+    
+    The crash was a divide by zero under m_currentUpdate.nextUpdate(), indicating that
+    m_currentUpdate.updatesPerSecond was zero. Previous assumptions that this was caused
+    by weird display configs were wrong. The actual issue is a race condition where
+    the callback can fire while we're still inside CVDisplayLinkStart(), or at least
+    before we've updated m_currentUpdate.
+    
+    The fix is to initialize m_currentUpdate before we call CVDisplayLinkStart().
+    
+    * UIProcess/mac/DisplayLink.cpp:
+    (WebKit::DisplayLink::addObserver):
+    
+    
+    git-svn-id: https://svn.webkit.org/repository/webkit/trunk@284644 268f45cc-cd09-0410-ab3c-d52691b4dbfc
+
+    2021-10-21  Simon Fraser  <simon.fraser@apple.com>
+
+            Rare crash under DisplayLink::displayLinkCallback()
+            https://bugs.webkit.org/show_bug.cgi?id=232101
+            <rdar://84153991>
+
+            Reviewed by Tim Horton.
+
+            The crash was a divide by zero under m_currentUpdate.nextUpdate(), indicating that
+            m_currentUpdate.updatesPerSecond was zero. Previous assumptions that this was caused
+            by weird display configs were wrong. The actual issue is a race condition where
+            the callback can fire while we're still inside CVDisplayLinkStart(), or at least
+            before we've updated m_currentUpdate.
+
+            The fix is to initialize m_currentUpdate before we call CVDisplayLinkStart().
+
+            * UIProcess/mac/DisplayLink.cpp:
+            (WebKit::DisplayLink::addObserver):
+
 2021-10-19  Russell Epstein  <repstein@apple.com>
 

branches/safari-612-branch/Source/WebKit/UIProcess/mac/DisplayLink.cpp

@@ -102 +102 @@
     if (!CVDisplayLinkIsRunning(m_displayLink)) {
         LOG_WITH_STREAM(DisplayLink, stream << "[UI ] DisplayLink for display " << m_displayID << " starting CVDisplayLink with fps " << m_displayNominalFramesPerSecond);
+
+        m_currentUpdate = { 0, m_displayNominalFramesPerSecond };
+
         CVReturn error = CVDisplayLinkStart(m_displayLink);
         if (error)
             RELEASE_LOG_FAULT(DisplayLink, "DisplayLink: Could not start the display link: %d", error);
-
-        if (!m_displayNominalFramesPerSecond) {
-            RELEASE_LOG_FAULT(DisplayLink, "DisplayLink: displayNominalFramesPerSecond is 0, using %d", WebCore::FullSpeedFramesPerSecond);
-            m_displayNominalFramesPerSecond = WebCore::FullSpeedFramesPerSecond;
-        };
-
-        m_currentUpdate = { 0, m_displayNominalFramesPerSecond };
     }
 }