jhsdb jstack cannot work correctly

[Yasumasa Suenaga]

Hi Jini,

Thank you for your advice!
I could get thread dump with this change:

-----------------
diff -r 5acd2b561936 src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/runtime/x86/X86Frame.java
--- a/src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/runtime/x86/X86Frame.java     Tue Aug 02 20:55:27 2016 -0700
+++ b/src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/runtime/x86/X86Frame.java     Wed Aug 03 23:42:36 2016 +0900
@@ -48,7 +48,7 @@
    private static final int SENDER_SP_OFFSET           =  2;

    // Interpreter frames
-  private static final int INTERPRETER_FRAME_MIRROR_OFFSET    =  2; // for native calls only
+  private static int INTERPRETER_FRAME_MIRROR_OFFSET;
    private static final int INTERPRETER_FRAME_SENDER_SP_OFFSET = -1;
    private static final int INTERPRETER_FRAME_LAST_SP_OFFSET   = INTERPRETER_FRAME_SENDER_SP_OFFSET - 1;
    private static final int INTERPRETER_FRAME_METHOD_OFFSET    = INTERPRETER_FRAME_LAST_SP_OFFSET - 1;
@@ -74,7 +74,8 @@
    }

    private static synchronized void initialize(TypeDataBase db) {
-    INTERPRETER_FRAME_MDX_OFFSET                  = INTERPRETER_FRAME_METHOD_OFFSET - 1;
+    INTERPRETER_FRAME_MIRROR_OFFSET               = INTERPRETER_FRAME_METHOD_OFFSET - 1;
+    INTERPRETER_FRAME_MDX_OFFSET                  = INTERPRETER_FRAME_MIRROR_OFFSET - 1;
      INTERPRETER_FRAME_CACHE_OFFSET                = INTERPRETER_FRAME_MDX_OFFSET - 1;
      INTERPRETER_FRAME_LOCALS_OFFSET               = INTERPRETER_FRAME_CACHE_OFFSET - 1;
      INTERPRETER_FRAME_BCX_OFFSET                  = INTERPRETER_FRAME_LOCALS_OFFSET - 1;
-----------------

I hope to resolve this issue ASAP.


Yasumasa


On 2016/08/03 22:23, Jini Susan George wrote:
> Hi Yasumasa,
>
> Thanks for noticing this. Regarding the illegal bci, you seem to be right. From a quick look, this looks like this came in with the introduction of interpreter_frame_mirror_offset. (https://bugs.openjdk.java.net/browse/JDK-8154580).
> From frame_x86.hpp, we have:
>
> ...
> 71     interpreter_frame_last_sp_offset                 = interpreter_frame_sender_sp_offset - 1,
> 72     interpreter_frame_method_offset                  = interpreter_frame_last_sp_offset - 1,
> 73     interpreter_frame_mirror_offset                  = interpreter_frame_method_offset - 1,
> 74     interpreter_frame_mdp_offset                     = interpreter_frame_mirror_offset - 1,
> 75     interpreter_frame_cache_offset                   = interpreter_frame_mdp_offset - 1,
> 76     interpreter_frame_locals_offset                  = interpreter_frame_cache_offset - 1,
> 77     interpreter_frame_bcp_offset                     = interpreter_frame_locals_offset - 1,
> ...
>
> This does not match with what we have in share/classes/sun/jvm/hotspot/runtime/x86/X86Frame.java.
>
> 77     INTERPRETER_FRAME_MDX_OFFSET                  = INTERPRETER_FRAME_METHOD_OFFSET - 1;
> 78     INTERPRETER_FRAME_CACHE_OFFSET                = INTERPRETER_FRAME_MDX_OFFSET - 1;
> 79     INTERPRETER_FRAME_LOCALS_OFFSET               = INTERPRETER_FRAME_CACHE_OFFSET - 1;
> 80     INTERPRETER_FRAME_BCX_OFFSET                  = INTERPRETER_FRAME_LOCALS_OFFSET - 1;
>
> We need to have the mirror offset reflected in the SA code also. I will file a bug for this and fix this.
>
> Thank you,
> Jini.
>
>
>> -----Original Message-----
>> From: Yasumasa Suenaga [mailto:yasuenag at gmail.com]
>> Sent: Wednesday, August 03, 2016 5:51 PM
>> To: serviceability-dev at openjdk.java.net
>> Subject: jhsdb jstack cannot work correctly
>>
>> Hi all,
>>
>> I use "jhsdb jstack" command in JDK 9 EA b129.
>> But it does not seem to work fine.
>>
>> I encountered two problems:
>>
>>    1. AssertionFailure: illegal bci
>>    2. Normal mode cannot work
>>
>>
>> 1. AssertionFailure: illegal bci
>>     I saw error stack as below with JDK 9 EA b129 Linux x64:
>> -------------------
>> 0x00007fe06bd270c2      * java.lang.Object.wait(long) bci:-520544688
>> (Interpreted frame)
>> 0x00007fe06bd1a443      sun.jvm.hotspot.utilities.AssertionFailure: illegal bci
>>          at sun.jvm.hotspot.utilities.Assert.that(jdk.hotspot.agent at 9-
>> ea/Assert.java:32)
>>          at
>> sun.jvm.hotspot.oops.ConstMethod.getLineNumberFromBCI(jdk.hotspot.age
>> nt at 9-ea/ConstMethod.java:297)
>>          at
>> sun.jvm.hotspot.oops.Method.getLineNumberFromBCI(jdk.hotspot.agent at 9-
>> ea/Method.java:282)
>>          at sun.jvm.hotspot.tools.PStack.getJavaNames(jdk.hotspot.agent at 9-
>> ea/PStack.java:239)
>>          at sun.jvm.hotspot.tools.PStack.run(jdk.hotspot.agent at 9-
>> ea/PStack.java:112)
>> -------------------
>>
>>     jhsdb can parse compiled frame correctly. It seems to be a problem for
>> interpreter frame only.
>>     What is the cause of this problem?
>>     interpretedVFrame in HotSpot has changed the location to store BCP?
>>
>>
>> 2. Normal mode cannot work
>>     "jhsdb jstack" should work as normal mode without being added --mixed
>> option.
>>     However, this command always works as mixed mode.
>>     I think we can fix as below:
>> -------------------
>> diff -r 5acd2b561936
>> src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/tools/JStack.java
>> --- a/src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/tools/JStack.java
>> Tue Aug 02 20:55:27 2016 -0700
>> +++
>> b/src/jdk.hotspot.agent/share/classes/sun/jvm/hotspot/tools/JStack.java
>> Wed Aug 03 21:18:57 2016 +0900
>> @@ -68,6 +68,10 @@
>>       }
>>
>>       public void runWithArgs(String... args) {
>> +        // We should recover these mode to default value.
>> +        mixedMode = false;
>> +        concurrentLocks = false;
>> +
>>           int used = 0;
>>           for (int i = 0; i < args.length; i++) {
>>               if (args[i].equals("-m")) {
>> -------------------
>>
>>
>> Has someone worked for them?
>> If not so, and patch of 2. is correct, I file it to JBS and upload webrev.
>>
>>
>> Thanks,
>>
>> Yasumasa
>>
>>