Handle Continuation scan during Concurrent Scavenger and Metronome #16157

LinHu2016 · 2022-10-20T21:10:54Z

For CS and Metronome, due to synchronous nature of their barriers (they read the value of the slot and act on it immediately, for example they may evacuate object for CS and mark object for Metronome), We fully scan java stacks of related continuation during the virtual thread mount.

depends: eclipse/omr#6781

Signed-off-by: Lin Hu linhu@ca.ibm.com

pshipton · 2022-10-20T21:22:17Z

Note we can run system testing in a PR build, i.e. ... test sanity.system <plats> <vers>

amicic · 2022-10-21T14:23:34Z

runtime/gc_base/ObjectAccessBarrier.hpp

@@ -267,6 +269,9 @@ class MM_ObjectAccessBarrier : public MM_BaseVirtual
 virtual bool postObjectRead(J9VMThread *vmThread, J9Object *srcObject, fj9object_t *srcAddress);
 virtual bool postObjectRead(J9VMThread *vmThread, J9Class *srcClass, J9Object **srcAddress);

+ virtual void preMountContinuation(J9VMThread *vmThread, j9object_t contObject) {}
+ virtual void postUnmountContinuation(J9VMThread *vmThread, j9object_t contObject) {}
+ virtual void registerScavenger(MM_Scavenger *scavenger) {}


I would not introduce subclass specific method in a base class. This method should be only in StandardAccessBarrier and not even virtual.

Then, of course, pointer to accessBarrier must be upcasted when invoking registerScavenger.

Prior to that we can assert 2 things are valid, to make sure that upcast is reasonable (none tried to register to wrong type of accessBarrier) :

isStandardGC() - this is exact guard that is used when we instantiate StandardAccessBarrier

scavengerEnabled

amicic · 2022-10-21T14:36:32Z

runtime/gc_modron_standard/StandardAccessBarrier.cpp

@@ -265,10 +265,12 @@ MM_StandardAccessBarrier::postObjectStoreImpl(J9VMThread *vmThread, J9Object *ds
 {
 /* If the source object is NULL, there is no need for a write barrier. */
 if(NULL != srcObject) {
+#if defined(OMR_GC_CONCURRENT_SCAVENGER)
 if (_extensions->isConcurrentScavengerEnabled() && !_extensions->isScavengerBackOutFlagRaised()) {
 Assert_MM_false(_extensions->scavenger->isObjectInEvacuateMemory(dstObject));


missed to use cached value

amicic · 2022-10-24T13:17:13Z

jenkins test sanity.system aix jdk8

amicic · 2022-10-24T13:17:51Z

jenkins test sanity win jdk19

amicic · 2022-10-24T16:59:44Z

Hm, sanity.system failure does not seem to be related:

TST stderr j> 2022/10/24 14:37:55.467 Mon Oct 24 14:37:55 UTC 2022 - main : Threads started, waiting for deadlock to establish
TST stderr j> 2022/10/24 14:37:55.467 Mon Oct 24 14:37:55 UTC 2022 - Deadlock Thread 1 : Acquired second lock
TST stderr j> 2022/10/24 14:37:55.518 Mon Oct 24 14:37:55 UTC 2022 - main : Deadlock established
TST stderr j> 2022/10/24 14:38:00.545 Mon Oct 24 14:38:00 UTC 2022 - Deadlock Thread 1 : Interrupted trying to get second lock
TST stderr j> 2022/10/24 14:38:00.546 Mon Oct 24 14:38:00 UTC 2022 - Deadlock Thread 2 : Interrupted trying to get second lock
TST stderr Exception in thread "main" java.lang.AssertionError: Reported Deadlocked Threads are Incorrect

pshipton · 2022-10-24T17:05:50Z

There was a bad change merged #15956

amicic · 2022-10-24T18:19:05Z

test/testng.xml

+ <class name=".TestGCPolicyNogc"/>
+ </classes>
+ </test> <!-- Test -->
+</suite> <!-- Suite -->


wrong change.
expected to touch one of module.xml to allow include from gc_base to gc_modron_standard

amicic · 2022-10-24T18:19:33Z

runtime/gc_base/GCExtensions.cpp

+MM_GCExtensions::registerScavenger(MM_Scavenger *scavenger)
+{
+ MM_GCExtensionsBase::registerScavenger(scavenger);
+ if (isStandardGC()) {


let's actually make this an assert

amicic · 2022-10-24T18:42:42Z

jenkins test sanity.system xLinux jdk8

amicic · 2022-10-24T18:43:08Z

jenkins test sanity aix jdk19

amicic · 2022-10-24T18:47:39Z

runtime/gc_base/module.xml

@@ -53,6 +53,7 @@ SPDX-License-Identifier: EPL-2.0 OR Apache-2.0 OR GPL-2.0 WITH Classpath-excepti
 <include path="j9gcstats"/>
 <include path="$(OMR_DIR)/gc/stats" type="relativepath"/>
 <include path="j9gcgluejava"/>
+ <include path="j9modronstandard"/>


need copyright update

from internal builds I see, we need -I../omr/gc/base/standard too

In file included from �[01m�[K../omr/gc/base/Configuration.hpp:33:0�[m�[K, from �[01m�[K../gc_modron_standard/StandardAccessBarrier.hpp:37�[m�[K, from �[01m�[KGCExtensions.cpp:40�[m�[K: �[01m�[K../gc_glue_java/ConfigurationDelegate.hpp:34:10:�[m�[K �[01;31m�[Kfatal error: �[m�[KConcurrentGC.hpp: No such file or directory

For CS and Metronome, due to synchronous nature of their barriers (they read the value of the slot and act on it immediately, for example they may evacuate object for CS and mark object for Metronome), We fully scan java stacks of related continuation during the virtual thread mount. Signed-off-by: Lin Hu <linhu@ca.ibm.com>

amicic · 2022-10-24T23:18:52Z

The only commit since jenkins tests were launched was touching only a module.xml file that is relevant only for IBM internal builds, so I did not relaunch them, however results got hidden:

aix sanity.functional jdk19 all green:
https://openj9-jenkins.osuosl.org/job/PullRequest-OpenJ9/2907/

xLinux sanity.system jdk8, all green except known, but unrelated failures in JLM:
https://openj9-jenkins.osuosl.org/job/PullRequest-OpenJ9/2906/

LinHu2016 force-pushed the supportConcurrentScavenger2 branch 3 times, most recently from 87d1aa8 to a8e639b Compare October 21, 2022 12:52

amicic reviewed Oct 21, 2022

View reviewed changes

LinHu2016 force-pushed the supportConcurrentScavenger2 branch 2 times, most recently from be9e78b to 0c18a0d Compare October 21, 2022 15:50

LinHu2016 force-pushed the supportConcurrentScavenger2 branch from 0c18a0d to 0f878c0 Compare October 24, 2022 18:15

amicic reviewed Oct 24, 2022

View reviewed changes

LinHu2016 force-pushed the supportConcurrentScavenger2 branch 2 times, most recently from 3ce7ca7 to bcdcf87 Compare October 24, 2022 18:41

amicic reviewed Oct 24, 2022

View reviewed changes

LinHu2016 force-pushed the supportConcurrentScavenger2 branch from bcdcf87 to c7a0338 Compare October 24, 2022 19:22

LinHu2016 mentioned this pull request Oct 24, 2022

Handle continuation mounting case during Metronome GC #16183

Merged

amicic approved these changes Oct 24, 2022

View reviewed changes

amicic merged commit 1a8e36d into eclipse-openj9:master Oct 24, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Handle Continuation scan during Concurrent Scavenger and Metronome #16157

Handle Continuation scan during Concurrent Scavenger and Metronome #16157

LinHu2016 commented Oct 20, 2022

pshipton commented Oct 20, 2022

amicic Oct 21, 2022

amicic Oct 21, 2022

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022 •

edited

Loading

pshipton commented Oct 24, 2022

amicic Oct 24, 2022

amicic Oct 24, 2022

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022

amicic Oct 24, 2022

amicic Oct 24, 2022

amicic commented Oct 24, 2022

Handle Continuation scan during Concurrent Scavenger and Metronome #16157

Handle Continuation scan during Concurrent Scavenger and Metronome #16157

Conversation

LinHu2016 commented Oct 20, 2022

pshipton commented Oct 20, 2022

amicic Oct 21, 2022

Choose a reason for hiding this comment

amicic Oct 21, 2022

Choose a reason for hiding this comment

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022 • edited Loading

pshipton commented Oct 24, 2022

amicic Oct 24, 2022

Choose a reason for hiding this comment

amicic Oct 24, 2022

Choose a reason for hiding this comment

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022

amicic Oct 24, 2022

Choose a reason for hiding this comment

amicic Oct 24, 2022

Choose a reason for hiding this comment

amicic commented Oct 24, 2022

amicic commented Oct 24, 2022 •

edited

Loading