Inline native call to java/lang/Class.getStackClass when possible

[r30shah]

Native call to java/lang/Class.getStackClass from the methods that are
being inlined in another method can be folded to load of JavaLangClass
from J9Class pointer when the inlined index of the caller of the
getStackClass is higher than the known constant argument to
getStackClass.

Signed-off-by: Rahil Shah rahil@ca.ibm.com

[r30shah]

@joransiu / @vijaysun-omr / @hzongaro Can I get your review on this change. I have tested this change internally through OpenJ9 - Personal/job/Pipeline-Build-Test-Personal/21826/. I see some criu related failures on x86 (Not related to these changes).

For the background, on JDK18 and up, we have introduced call to getStackClass [1][2][3] at various critical path which triggered a stack class iteration and stack walk. If we are inlining to the point that we can know the caller of getStackClass at index at compile time, we can inline those calls. On ILOG-ODM, these extra calls to getStackClass resulted in 5% performance penalty in throughput. With this change, we recovered that 5% back.

[1].

openj9/jcl/src/java.base/share/classes/java/lang/Class.java

Line 406 in b3ab6b6

return forName(className, getStackClass(1));

[2].

openj9/jcl/src/java.base/share/classes/java/lang/Class.java

Line 532 in b3ab6b6

Class<?> caller = getStackClass(1);

[3].

openj9/jcl/src/java.base/share/classes/java/lang/Class.java

Line 634 in b3ab6b6

Class<?> caller = getStackClass(1);

[vijaysun-omr]

The change itself looks good to me. There appears to be some sort of trailing H char in the commit message, that it would be good to rectify. The change also includes a change for Class.forName as a recognized method, but this isn't mentioned in the commit message or title. After these questions are answered, we could run more testing.

[joransiu]

Logic looks fine to me, besides some wording nits of the performTransformation message.

[r30shah]

The change itself looks good to me. There appears to be some sort of trailing H char in the commit message, that it would be good to rectify. The change also includes a change for Class.forName as a recognized method, but this isn't mentioned in the commit message or title. After these questions are answered, we could run more testing.

I have removed the trailing H character from commit message.
I have also cleaned up the code for Class.forName(). I originally implemented this change to improve calls to getStackClass from Class.forName(), but while working on it, I realized it doesn't need to be only from Class.forName. So in further refinement of the changes, I covered other cases as well. So now we do not need Class.forName to be recognized anymore.

[vijaysun-omr]

jenkins test sanity all jdk21

[r30shah]

Checking out the failures on some of the builds.

1. JDK21 x86-64 : Build passes but job failed with following error when trying to upload the files

22:06:44  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:44  Attempting retry #1
22:06:47  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:47  Attempting retry #2
22:06:50  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:50  Attempting retry #3
22:06:52  [consumer_1] Deploying artifact: https://ub18p8art.casa.cs.unb.ca/artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/test-images.tar.gz
22:06:53  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:53  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:53  Attempting retry #1
22:06:56  Error occurred for request PUT /artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/OpenJ9-JDK21-x86-64_linux-20240429-204010.tar.gz;build.parentNumber=483;build.parentName=Pipeline_Build_Test_JDK21_x86-64_linux;build.buildIdentifier=eclipse-openj9%2Fopenj9%2319365;build.timestamp=1714437508323;build.name=Build_JDK21_x86-64_linux_Personal;build.number=128 HTTP/1.1: No route to host (Host unreachable).
22:06:56  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:56  Attempting retry #1
22:06:56  Attempting retry #2
22:06:59  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:06:59  Error occurred for request PUT /artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/OpenJ9-JDK21-x86-64_linux-20240429-204010.tar.gz;build.parentNumber=483;build.parentName=Pipeline_Build_Test_JDK21_x86-64_linux;build.buildIdentifier=eclipse-openj9%2Fopenj9%2319365;build.timestamp=1714437508323;build.name=Build_JDK21_x86-64_linux_Personal;build.number=128 HTTP/1.1: No route to host (Host unreachable).
22:06:59  Attempting retry #3
22:06:59  Attempting retry #2
22:07:02  Error occurred for request GET /artifactory/api/system/version HTTP/1.1: No route to host (Host unreachable).
22:07:02  Error occurred for request PUT /artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/OpenJ9-JDK21-x86-64_linux-20240429-204010.tar.gz;build.parentNumber=483;build.parentName=Pipeline_Build_Test_JDK21_x86-64_linux;build.buildIdentifier=eclipse-openj9%2Fopenj9%2319365;build.timestamp=1714437508323;build.name=Build_JDK21_x86-64_linux_Personal;build.number=128 HTTP/1.1: No route to host (Host unreachable).
22:07:02  Attempting retry #3
22:07:05  Error occurred for request PUT /artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/OpenJ9-JDK21-x86-64_linux-20240429-204010.tar.gz;build.parentNumber=483;build.parentName=Pipeline_Build_Test_JDK21_x86-64_linux;build.buildIdentifier=eclipse-openj9%2Fopenj9%2319365;build.timestamp=1714437508323;build.name=Build_JDK21_x86-64_linux_Personal;build.number=128 HTTP/1.1: No route to host (Host unreachable).
22:07:05  Error occurred for request PUT /artifactory/ci-openj9/Build_JDK21_x86-64_linux_Personal/128/test-images.tar.gz;build.parentNumber=483;build.parentName=Pipeline_Build_Test_JDK21_x86-64_linux;build.buildIdentifier=eclipse-openj9%2Fopenj9%2319365;build.timestamp=1714437508323;build.name=Build_JDK21_x86-64_linux_Personal;build.number=128 HTTP/1.1: No route to host (Host unreachable).
22:07:05  Attempting retry #1
22:07:05  [consumer_0] An exception occurred during execution:
22:07:05  java.lang.RuntimeException: java.net.NoRouteToHostException: No route to host (Host unreachable)
22:07:05  	at org.jfrog.build.extractor.clientConfiguration.util.spec.SpecDeploymentConsumer.consumerRun(SpecDeploymentConsumer.java:44)
22:07:05  	at org.jfrog.build.extractor.producerConsumer.ConsumerRunnableBase.run(ConsumerRunnableBase.java:11)
22:07:05  	at java.base/java.lang.Thread.run(Thread.java:840)

2. JDK21 PPC64LE : Build aborted with following error,

All nodes of label ‘[ci.role.build&&hw.arch.ppc64le&&sw.os.cent.7](https://openj9-jenkins.osuosl.org/label/ci.role.build&&hw.arch.ppc64le&&sw.os.cent.7/)’ are offline

Also checking out OpenJ9 slack : Seems like bunch of UNB machine went down on Friday, (https://openj9.slack.com/archives/CDS7QE9HB/p1714157270356889 ](https://openj9.slack.com/archives/C8PQL5N65/p1714397285116169) which was not fixed yesterday. Looking at https://openj9-jenkins.osuosl.org/label/ci.geo.unb/, seems like bunch of there are still down

[r30shah]

Jenkins test sanity xlinux jdk21

[r30shah]

Failure seen in JDK21 x86 build was due tio #19431, Now as it is merged, relaunching the JDK21 x86 build

[r30shah]

Jenkins test sanity xlinux jdk21

[r30shah]

Jenkins test sanity xlinux jdk21

[r30shah]

Seems like last launched build finished successfully but test job could not download the build to start the tests.

I am going to rebase my changes and run the tests internally again to cover some of the missing targets.

[r30shah]

jenkins test sanity all jdk21

[r30shah]

Looking into the build results (Both from internal and open)

1. JDK21 AARCH64 Linux : Failed in test testJitserverArguments_0 with following error, issue is already seen : testJitserverArguments_0_FAILED FAILURE: JITServer::StreamFailure: Connect failed: Connection refused #19151

19:06:04   [ERR] JITServer is ready to accept incoming requests
19:06:04   [ERR] can't bind server address: Address already in use
19:06:04   [ERR] Failed to open server socket on port 43896

I do have this target passing on hyc jenkins : build

2. PPC64LE linux : Build aborted due to machine unavailability, corresponding build from internal jenkins, both sanity.functional and sanity.openjdk passes.

3. x86-64 Linux : Builds can not be downloaded to test, corresponding build from internal hyc jenkins: build

4. x86-64 Mac : Same issue where test job can not reach to artifactory to download builds, I also ran this target internally in build where sanity.functional passes (Sanity.openjdk was aborted due to infrastructure issue).

@vijaysun-omr I am not sure when the external infrastructure issues will be resolved, but I did try to cover most of the platforms combining internally and externally. Here is the JDK11 and JDK21 internal build. One failure seen internally on s390 is same as #19151 which is passing externally.

This one can be merged if you are OK with the testing.

[vijaysun-omr]

Thanks @r30shah

I am going to merge this on the basis of the justification that you provided.