Merge

src/hotspot/share/gc/g1/g1ConcurrentMark.cpp

@@ -2008,13 +2008,21 @@ void G1ConcurrentMark::print_stats() {
 }
 
 void G1ConcurrentMark::concurrent_cycle_abort() {
-  if (!cm_thread()->in_progress() || _has_aborted) {
-    // We haven't started a concurrent cycle or we have already aborted it. No need to do anything.
+  // We haven't started a concurrent cycle no need to do anything; we might have
+  // aborted the marking because of shutting down though. In this case the marking
+  // might have already completed the abort (leading to in_progress() below to
+  // return false), however this still left marking state particularly in the
+  // shared marking bitmap that must be cleaned up.
+  // If there are multiple full gcs during shutdown we do this work repeatedly for
+  // nothing, but this situation should be extremely rare (a full gc after shutdown
+  // has been signalled is alredy rare), and this work should be negligible compared
+  // to actual full gc work.
+  if (!cm_thread()->in_progress() && !_g1h->concurrent_mark_is_terminating()) {
     return;
   }
 
-  // Clear all marks in the next bitmap for the next marking cycle. This will allow us to skip the next
-  // concurrent bitmap clearing.
+  // Clear all marks in the next bitmap for this full gc as it has been used by the
+  // marking that is interrupted by this full gc.
   {
     GCTraceTime(Debug, gc) debug("Clear Next Bitmap");
     clear_next_bitmap(_g1h->workers());
@@ -2028,9 +2036,8 @@ void G1ConcurrentMark::concurrent_cycle_abort() {
   for (uint i = 0; i < _max_num_tasks; ++i) {
     _tasks[i]->clear_region_fields();
   }
-  _first_overflow_barrier_sync.abort();
-  _second_overflow_barrier_sync.abort();
-  _has_aborted = true;
+
+  abort_marking_threads();
 
   SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
   satb_mq_set.abandon_partial_marking();
@@ -2041,6 +2048,13 @@ void G1ConcurrentMark::concurrent_cycle_abort() {
                                  satb_mq_set.is_active() /* expected_active */);
 }
 
+void G1ConcurrentMark::abort_marking_threads() {
+  assert(!_root_regions.scan_in_progress(), "still doing root region scan");
+  _has_aborted = true;
+  _first_overflow_barrier_sync.abort();
+  _second_overflow_barrier_sync.abort();
+}
+
 static void print_ms_time_info(const char* prefix, const char* name,
                                NumberSeq& ns) {
   log_trace(gc, marking)("%s%5d %12s: total time = %8.2f s (avg = %8.2f ms).",

src/hotspot/share/gc/g1/g1ConcurrentMark.hpp

@@ -496,6 +496,11 @@ public:
   void concurrent_cycle_abort();
   void concurrent_cycle_end();
 
+  // Notifies marking threads to abort. This is a best-effort notification. Does not
+  // guarantee or update any state after the call. Root region scan must not be
+  // running.
+  void abort_marking_threads();
+
   void update_accum_task_vtime(int i, double vtime) {
     _accum_task_vtime[i] += vtime;
   }

src/hotspot/share/gc/g1/g1ConcurrentMarkThread.cpp

@@ -139,6 +139,15 @@ void G1ConcurrentMarkThread::run_service() {
 }
 
 void G1ConcurrentMarkThread::stop_service() {
+  if (in_progress()) {
+    // We are not allowed to abort the marking threads during root region scan.
+    // Needs to be done separately.
+    _cm->root_regions()->abort();
+    _cm->root_regions()->wait_until_scan_finished();
+
+    _cm->abort_marking_threads();
+  }
+
   MutexLocker ml(CGC_lock, Mutex::_no_safepoint_check_flag);
   CGC_lock->notify_all();
 }

src/hotspot/share/opto/loopTransform.cpp

@@ -3437,6 +3437,11 @@ bool IdealLoopTree::do_one_iteration_loop(PhaseIdealLoop *phase) {
 //=============================================================================
 //------------------------------iteration_split_impl---------------------------
 bool IdealLoopTree::iteration_split_impl(PhaseIdealLoop *phase, Node_List &old_new) {
+  if (!_head->is_Loop()) {
+    // Head could be a region with a NeverBranch that was added in beautify loops but the region was not
+    // yet transformed into a LoopNode. Bail out and wait until beautify loops turns it into a Loop node.
+    return false;
+  }
   // Compute loop trip count if possible.
   compute_trip_count(phase);
 

src/jdk.jpackage/share/native/applauncher/JvmLauncher.cpp

@@ -23,6 +23,7 @@
  * questions.
  */
 
+#include <cstddef>
 #include <cstring>
 #include "tstrings.h"
 #include "JvmLauncher.h"

test/hotspot/jtreg/compiler/loopopts/TestIterationSplitWithRegionHead.java

@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) 2022, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/*
+ * @test
+ * @requires vm.compiler2.enabled
+ * @bug 8279837
+ * @summary Tests infinite loop with region head in iteration split.
+ * @run main/othervm -Xcomp -XX:-TieredCompilation
+ *      -XX:CompileCommand=compileonly,compiler.loopopts.TestIterationSplitWithRegionHead::test
+ *      -XX:CompileCommand=dontinline,compiler.loopopts.TestIterationSplitWithRegionHead::*
+ *      compiler.loopopts.TestIterationSplitWithRegionHead
+ */
+
+package compiler.loopopts;
+
+public class TestIterationSplitWithRegionHead {
+
+    static boolean flagFalse = false;
+
+    public static void main(String[] args) {
+        test();
+    }
+
+    public static void test() {
+        // 1) The loop tree is built. We find that nested loop N2 is an infinite loop and add a NeverBranch
+        // to the inner loop to make it reachable. But the current loop tree does not have N2, yet. The
+        // resulting loop tree is:
+        //
+        //   Loop: N0/N0  has_call has_sfpt
+        //     Loop: N77/N121  has_call // N1 outer
+        //       Loop: N77/N111  has_call sfpts={ 111 97 } // N1 inner
+        //
+        // 2) beautify_loops() finds that the outer loop head of N1 is shared and thus adds a new region
+        // in merge_many_backedges(). As a result, the loop tree is built again. This time, the NeverBranch
+        // in the inner loop of N2 allows that a loop tree can be built for it:
+        //
+        //   Loop: N0/N0  has_call has_sfpt
+        //     Loop: N216/N213  limit_check profile_predicated predicated has_call sfpts={ 111 97 } // N1 shared loop head
+        //     Loop: N196/N201  sfpts={ 201 } // N2 inner loop now discovered with the new NeverBranch
+        //
+        // However, a LoopNode is only added by beautify_loops() which won't be called until the next iteration of loop opts.
+        // This means that we have a Region node (N196) as head in the loop tree which cannot be handled by iteration_split_impl()
+        // resulting in an assertion failure.
+
+        // Nested loop N1
+        while (flagFalse) {
+            while (dontInlineFalse()) {
+            }
+        }
+        dontInlineFalse();
+
+        // Nested loop N2
+        while (flagFalse) {
+            while (true) ;  // Detected as infinite inner loop by C2 -> NeverBranch added
+        }
+    }
+
+    public static boolean dontInlineFalse() {
+        return false;
+    }
+}