hotspot JIT producing incorrect code -how should I report a bug?

[Mark Wielaard]

Hi,

On Wed, Jul 16, 2008 at 02:23:21PM -0700, Chuck Rasbold wrote:
> Ismael Juma wrote:
> > Chuck Rasbold <Chuck.Rasbold at ...> writes:
> > 
> >>Thanks for bringing 6707044 to our attention. Is was previously misfiled
> >>in the wrong subcategory. I'm looking into a diagnosis right now.
> > 
> > Thanks for looking into it Chuck. It looks like a scary bug and it would
> > be very nice to get a fix.
> 
> With a fix now in jdk7/hotspot-comp/hotspot, we're now looking into
> getting this fix backported into the 6u10 release.

This was indeed a scary bug. And easily triggered.
I backported it to the jdk6 hotspot as used in icedtea.
I also added the  test case for it (that obviously fails before and
succeeds after the patch). It would be nice if it could also be added
to the hotspot tests. There are unfortunately very little public tests
for hotspot and increasing their number would help with making sure
no regressings slip in.

I didn't take your followup patches (for #6726504 and #6730192) because
I wanted a minimal backport. But if you think it would be good to
backport those too then I will be happy to take a stab at it.

Cheers,

Mark

P.S. Note to Gary, I didn't try to fix it in the zero hotspot 7b24
     version. We should probably just refresh that one completely.
-------------- next part --------------
diff -r 4aebfff4f8a2 hotspot/src/share/vm/opto/parse2.cpp
--- openjdk.orig/hotspot/src/share/vm/opto/parse2.cpp	Mon Sep 15 11:38:34 2008 +0200
+++ openjdk/hotspot/src/share/vm/opto/parse2.cpp	Thu Sep 25 21:54:26 2008 +0200
@@ -868,6 +868,8 @@
   return prob < PROB_MIN;
 }
 
+//-------------------------------repush_if_args--------------------------------
+// Push arguments of an "if" bytecode back onto the stack by adjusting _sp.
 inline void Parse::repush_if_args() {
 #ifndef PRODUCT
   if (PrintOpto && WizardMode) {
@@ -896,7 +898,6 @@
     if (PrintOpto && Verbose)
       tty->print_cr("Never-taken backedge stops compilation at bci %d",bci());
 #endif
-    repush_if_args(); // to gather stats on loop
     uncommon_trap(Deoptimization::Reason_unreached,
                   Deoptimization::Action_reinterpret,
                   NULL, "cold");
diff -r 51798f0e554f test/compiler/6707044/LuceneCrash.java
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ openjdk/hotspot/test/compiler/6707044/LuceneCrash.java	Thu Sep 25 21:55:28 2008 +0200
@@ -0,0 +1,40 @@
+/*
+ * @test
+ * @bug 6707044
+ * @summary uncommon_trap of ifnull bytecode leaves garbage on expression stack
+ * @run main/othervm -Xbatch LuceneCrash
+ */
+public class LuceneCrash {
+
+  public static void main(String[] args) throws Throwable {
+    new LuceneCrash().crash();
+  }
+
+  private Object alwaysNull;
+
+  final void crash() throws Throwable {
+    for (int r = 0; r < 3; r++) {
+      for (int docNum = 0; docNum < 10000;) {
+        if (r < 2) {
+          for(int j=0;j<3000;j++)
+            docNum++;
+        } else {
+          docNum++;
+          doNothing(getNothing());
+          if (alwaysNull != null) {
+            throw new RuntimeException("BUG: checkAbort is always null: r=" + r + " of 3; docNum=" + docNum);
+          }
+        }
+      }
+    }
+  }
+
+  Object getNothing() {
+    return this;
+  }
+
+  int x;
+  void doNothing(Object o) {
+    x++;
+  }
+}