8325397: sun/java2d/Disposer/TestDisposerRace.java fails in linux-aarch64

Reviewed-by: alanb

Author: Viktor Klang

src/java.base/share/classes/java/util/concurrent/locks/AbstractQueuedLongSynchronizer.java

@@ -277,6 +277,40 @@ private static void signalNextIfShared(Node h) {
         }
     }
 
+    /**
+     * Repeatedly invokes acquire, if its execution throws an Error or a Runtime Exception,
+     * using an Unsafe.park-based backoff
+     * @param node which to reacquire
+     * @param arg the acquire argument
+     */
+    private final void reacquire(Node node, long arg) {
+        try {
+            acquire(node, arg, false, false, false, 0L);
+        } catch (Error | RuntimeException firstEx) {
+            // While we currently do not emit an JFR events in this situation, mainly
+            // because the conditions under which this happens are such that it
+            // cannot be presumed to be possible to actually allocate an event, and
+            // using a preconstructed one would have limited value in serviceability.
+            // Having said that, the following place would be the more appropriate
+            // place to put such logic:
+            //     emit JFR event
+
+            for (long nanos = 1L;;) {
+                U.park(false, nanos); // must use Unsafe park to sleep
+                if (nanos < 1L << 30)            // max about 1 second
+                    nanos <<= 1;
+
+                try {
+                    acquire(node, arg, false, false, false, 0L);
+                } catch (Error | RuntimeException ignored) {
+                    continue;
+                }
+
+                throw firstEx;
+            }
+        }
+    }
+
     /**
      * Main acquire method, invoked by all exported acquire methods.
      *
@@ -1299,7 +1333,7 @@ else if ((node.status & COND) != 0) {
             }
             LockSupport.setCurrentBlocker(null);
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (interrupted)
                 Thread.currentThread().interrupt();
         }
@@ -1346,7 +1380,7 @@ public final void await() throws InterruptedException {
             }
             LockSupport.setCurrentBlocker(null);
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (interrupted) {
                 if (cancelled) {
                     unlinkCancelledWaiters(node);
@@ -1389,7 +1423,7 @@ public final long awaitNanos(long nanosTimeout)
                     LockSupport.parkNanos(this, nanos);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)
@@ -1433,7 +1467,7 @@ public final boolean awaitUntil(Date deadline)
                     LockSupport.parkUntil(this, abstime);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)
@@ -1478,7 +1512,7 @@ public final boolean await(long time, TimeUnit unit)
                     LockSupport.parkNanos(this, nanos);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)

src/java.base/share/classes/java/util/concurrent/locks/AbstractQueuedSynchronizer.java

@@ -656,6 +656,40 @@ private static void signalNextIfShared(Node h) {
         }
     }
 
+    /**
+     * Repeatedly invokes acquire, if its execution throws an Error or a Runtime Exception,
+     * using an Unsafe.park-based backoff
+     * @param node which to reacquire
+     * @param arg the acquire argument
+     */
+    private final void reacquire(Node node, int arg) {
+        try {
+            acquire(node, arg, false, false, false, 0L);
+        } catch (Error | RuntimeException firstEx) {
+            // While we currently do not emit an JFR events in this situation, mainly
+            // because the conditions under which this happens are such that it
+            // cannot be presumed to be possible to actually allocate an event, and
+            // using a preconstructed one would have limited value in serviceability.
+            // Having said that, the following place would be the more appropriate
+            // place to put such logic:
+            //     emit JFR event
+
+            for (long nanos = 1L;;) {
+                U.park(false, nanos); // must use Unsafe park to sleep
+                if (nanos < 1L << 30)            // max about 1 second
+                    nanos <<= 1;
+
+                try {
+                    acquire(node, arg, false, false, false, 0L);
+                } catch (Error | RuntimeException ignored) {
+                    continue;
+                }
+
+                throw firstEx;
+            }
+        }
+    }
+
     /**
      * Main acquire method, invoked by all exported acquire methods.
      *
@@ -1678,7 +1712,7 @@ else if ((node.status & COND) != 0) {
             }
             LockSupport.setCurrentBlocker(null);
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (interrupted)
                 Thread.currentThread().interrupt();
         }
@@ -1725,7 +1759,7 @@ public final void await() throws InterruptedException {
             }
             LockSupport.setCurrentBlocker(null);
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (interrupted) {
                 if (cancelled) {
                     unlinkCancelledWaiters(node);
@@ -1768,7 +1802,7 @@ public final long awaitNanos(long nanosTimeout)
                     LockSupport.parkNanos(this, nanos);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)
@@ -1812,7 +1846,7 @@ public final boolean awaitUntil(Date deadline)
                     LockSupport.parkUntil(this, abstime);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)
@@ -1857,7 +1891,7 @@ public final boolean await(long time, TimeUnit unit)
                     LockSupport.parkNanos(this, nanos);
             }
             node.clearStatus();
-            acquire(node, savedState, false, false, false, 0L);
+            reacquire(node, savedState);
             if (cancelled) {
                 unlinkCancelledWaiters(node);
                 if (interrupted)

test/jdk/sun/java2d/Disposer/TestDisposerRace.java

@@ -23,6 +23,7 @@
 import java.util.LinkedList;
 import java.util.List;
 import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.Supplier;
 import javax.swing.SwingUtilities;
 
 import sun.java2d.Disposer;
@@ -39,31 +40,38 @@ public final class TestDisposerRace {
     private static final AtomicInteger recordsCount = new AtomicInteger();
     private static volatile boolean disposerDone = false;
 
+    private static final String KO_OVERFLOW = "Some records have not been disposed!";
+    private static final String KO_UNDERFLOW = "Disposed more records than were added!";
+
     public static void main(String[] args) throws Exception {
-        TestDisposerRace test = new TestDisposerRace();
-        test.run();
+        new TestDisposerRace().run();
 
         checkRecordsCountIsSane();
         if (recordsCount.get() > 0) {
+            System.err.println(KO_OVERFLOW); // In case the next line fails to allocate due to OOME
             throw new RuntimeException("Some records (" + recordsCount + ") have not been disposed");
         }
     }
 
+    interface ThrowingRunnable<E extends Exception> {
+        void run() throws E;
+    }
+
     TestDisposerRace() {
         addRecordsToDisposer(30_000);
     }
 
     void run() throws Exception {
         generateOOME();
         for (int i = 0; i < 1000; ++i) {
-            SwingUtilities.invokeAndWait(Disposer::pollRemove);
-            if (i % 10 == 0) {
-                // Adding records will race with the diposer trying to remove them
+            retryOnOOME(() -> SwingUtilities.invokeAndWait(Disposer::pollRemove));
+
+            // Adding records will race with the diposer trying to remove them
+            if (i % 10 == 0)
                 addRecordsToDisposer(1000);
-            }
         }
 
-        Disposer.addObjectRecord(new Object(), new FinalDisposerRecord());
+        retryOnOOME(() -> Disposer.addObjectRecord(new Object(), new FinalDisposerRecord()));
 
         while (!disposerDone) {
              generateOOME();
@@ -72,18 +80,45 @@ void run() throws Exception {
 
     private static void checkRecordsCountIsSane() {
         if (recordsCount.get() < 0) {
-            throw new RuntimeException("Disposed more records than were added");
+            throw new RuntimeException(KO_UNDERFLOW);
+        }
+    }
+
+    private static <T> T retryOnOOME(Supplier<T> allocator) {
+        for(;;) {
+            try {
+                return allocator.get();
+            } catch (OutOfMemoryError ignored1) {
+                try {
+                    Thread.sleep(1); // Give GC a little chance to run
+                } catch (InterruptedException ignored2) {}
+            }
+        }
+    }
+
+    private static <E extends Exception> void retryOnOOME(ThrowingRunnable<E> tr) throws E {
+        for(;;) {
+            try {
+                tr.run();
+                break;
+            } catch (OutOfMemoryError ignored1) {
+                try {
+                    Thread.sleep(1); // Give GC a little chance to run
+                } catch (InterruptedException ignored2) {}
+            }
         }
     }
 
     private void addRecordsToDisposer(int count) {
         checkRecordsCountIsSane();
 
-        recordsCount.addAndGet(count);
+        MyDisposerRecord disposerRecord = retryOnOOME(MyDisposerRecord::new);
 
-        MyDisposerRecord disposerRecord = new MyDisposerRecord();
-        for (int i = 0; i < count; i++) {
-            Disposer.addObjectRecord(new Object(), disposerRecord);
+        while(count > 0) {
+            recordsCount.incrementAndGet(); // pre-add to make sure it doesn't go negative
+            var o = retryOnOOME(Object::new);
+            retryOnOOME(() -> Disposer.addObjectRecord(o, disposerRecord));
+            --count;
         }
     }
 
@@ -106,8 +141,8 @@ private static void giveGCAChance() {
     }
 
     private static void generateOOME() throws Exception {
-        final List<Object> leak = new LinkedList<>();
         try {
+            final List<Object> leak = new LinkedList<>();
             while (true) {
                 leak.add(new byte[1024 * 1024]);
             }