JDK-8249880 : JVMCI calling register_nmethod without CodeCache lock
  • Type: Bug
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 11,15,16
  • Priority: P2
  • Status: Closed
  • Resolution: Fixed
  • Submitted: 2020-07-22
  • Updated: 2021-01-13
  • Resolved: 2020-07-22
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
JDK 11 JDK 15 JDK 16
11.0.10-oracleFixed 15Fixed 16 b07Fixed
Description
When running a CompileTheWorld test using Graal and libGraal, we are hitting a guarantee caused by a missing lock on the CodeCache when registerring an nmethod.

Here is a proposed fix:

diff --git a/src/hotspot/share/jvmci/jvmciRuntime.cpp b/src/hotspot/share/jvmci/jvmciRuntime.cpp
--- a/src/hotspot/share/jvmci/jvmciRuntime.cpp
+++ b/src/hotspot/share/jvmci/jvmciRuntime.cpp
@@ -668,6 +668,7 @@
 
   // Since we've patched some oops in the nmethod,
   // (re)register it with the heap.
+  MutexLocker ml(CodeCache_lock, Mutex::_no_safepoint_check_flag);
   Universe::heap()->register_nmethod(nm);
 }

Here is the crash output:

#
# A fatal error has been detected by the Java Runtime Environment:
#
# Internal Error (heapRegionRemSet.cpp:459), pid=36332, tid=36346
# guarantee(!_code_roots.contains(nm)) failed: duplicate entry found
#
# JRE version: Java(TM) SE Runtime Environment Server VM Graal:compiler_9692d20aac4f22c736dded0b94aab16fc4e9ed50 (15.0) (build 15-internal+0-2020-07-21-1834055.bvandett...)
# Java VM: Java HotSpot(TM) 64-Bit Server VM Server VM Graal:compiler_9692d20aac4f22c736dded0b94aab16fc4e9ed50 (15-internal+0-2020-07-21-1834055.bvandett..., mixed mode, sharing, tiered, jvmci, jvmci compiler, compressed oops, g1 gc, linux-amd64)
# Problematic frame:
# V [libjvm.so+0x729054] HeapRegionRemSet::remove_strong_code_root(nmethod*)+0xa4
#
# Core dump will be written. Default location: Core dumps may be processed with ���/usr/share/apport/apport %p %s %c %d %P %E��� (or dumping to /export/users/bobv/bitbucket/graal/vm/core.36332)
#
# An error report file with more information is saved as:
# /export/users/bobv/bitbucket/graal/vm/hs_err_pid36332.log
Compiled method (JVMCI)  26076 45372    4    sun.net.www.MessageHeader::canonicalID (99 bytes)
 total in heap [0x00007f4ec5fe4990,0x00007f4ec5fe5068] = 1752
 relocation   [0x00007f4ec5fe4ae8,0x00007f4ec5fe4b40] = 88
 main code   [0x00007f4ec5fe4b40,0x00007f4ec5fe4e60] = 800
 stub code   [0x00007f4ec5fe4e60,0x00007f4ec5fe4ec0] = 96
 oops      [0x00007f4ec5fe4ec0,0x00007f4ec5fe4ec8] = 8
 metadata    [0x00007f4ec5fe4ec8,0x00007f4ec5fe4ed0] = 8
 scopes data  [0x00007f4ec5fe4ed0,0x00007f4ec5fe4f40] = 112
 scopes pcs   [0x00007f4ec5fe4f40,0x00007f4ec5fe4fc0] = 128
 dependencies  [0x00007f4ec5fe4fc0,0x00007f4ec5fe4fc8] = 8
 handler table [0x00007f4ec5fe4fc8,0x00007f4ec5fe5058] = 144
 JVMCI data   [0x00007f4ec5fe5058,0x00007f4ec5fe5068] = 16
Comments
The issue was mainly observed when running CompileTheWorld with Graal. Right now there is no easy way to verify the issue because Graal was disabled in jdk16.
13-01-2021
Fix request for JDK 15 approved.
22-07-2020
Fix Request: This fix is required for stability running libgraal on JDK15 in the Metropolis EA release. The VM will crash under heavy compilation load without this fix. This fix has already been integrated into the labsjdk and JDK 16. The fix was reviewed by Tom Rodriguez, Erik Osterlund, Vladimir Kozlov and Doug Simon.
22-07-2020
Changed to P2 because affects Metropolis EA release for JDK 15.
22-07-2020
URL: https://hg.openjdk.java.net/jdk/jdk/rev/c3a4a7ea7c30 User: bobv Date: 2020-07-22 17:59:48 +0000
22-07-2020