RFR: 8177845: Need a mechanism to load Graal

[Doug Simon]

I've had to update the webrev again to support selection of a "null" compiler (i.e. one that raises an
exception upon a compilation request) and added -Djvmci.Compiler=null to a large number of JVMCI jtreg
tests to prevent Graal being selected and initialized by the JVMCI compiler auto-selection mechanism.
Initializing Graal will currently fail with errors (see stack trace below) until Graal is updated to
the version compatible with the JVMCI API changes.

In addition to resolving the compatibility issue, explicitly selecting the "null" compiler for these
tests better isolates them from parts of the runtime they are not aiming to test.

org.graalvm.compiler.debug.GraalError: java.lang.ClassCastException: java.base/java.util.ImmutableCollections$MapN cannot be cast to java.base/java.util.Properties
	at jdk.internal.vm.compiler/org.graalvm.compiler.hotspot.HotSpotGraalCompilerFactory.getSavedProperties(HotSpotGraalCompilerFactory.java:217)
	at jdk.internal.vm.compiler/org.graalvm.compiler.hotspot.HotSpotGraalCompilerFactory.initializeOptions(HotSpotGraalCompilerFactory.java:138)
	at jdk.internal.vm.compiler/org.graalvm.compiler.hotspot.HotSpotGraalCompilerFactory.onSelection(HotSpotGraalCompilerFactory.java:95)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCICompilerConfig.getCompilerFactory(HotSpotJVMCICompilerConfig.java:104)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCIRuntime.<init>(HotSpotJVMCIRuntime.java:290)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCIRuntime.<init>(HotSpotJVMCIRuntime.java:65)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCIRuntime$DelayedInit.<clinit>(HotSpotJVMCIRuntime.java:73)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCIRuntime.runtime(HotSpotJVMCIRuntime.java:83)
	at jdk.internal.vm.ci/jdk.vm.ci.runtime.JVMCI.initializeRuntime(Native Method)
	at jdk.internal.vm.ci/jdk.vm.ci.runtime.JVMCI.<clinit>(JVMCI.java:58)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotJVMCIRuntime.runtime(HotSpotJVMCIRuntime.java:82)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotVMConfig.config(HotSpotVMConfig.java:41)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotResolvedJavaMethodImpl.getHolder(HotSpotResolvedJavaMethodImpl.java:92)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.HotSpotResolvedJavaMethodImpl.fromMetaspace(HotSpotResolvedJavaMethodImpl.java:110)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.CompilerToVM.asResolvedJavaMethod(Native Method)
	at jdk.internal.vm.ci/jdk.vm.ci.hotspot.CompilerToVMHelper.asResolvedJavaMethod(CompilerToVMHelper.java:185)
	at compiler.jvmci.common.CTVMUtilities.getResolvedMethod(CTVMUtilities.java:59)
	at compiler.jvmci.common.CTVMUtilities.getResolvedMethod(CTVMUtilities.java:64)
	at compiler.jvmci.compilerToVM.AllocateCompileIdTest.runSanityCorrectTest(AllocateCompileIdTest.java:125)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1378)
	at compiler.jvmci.compilerToVM.AllocateCompileIdTest.main(AllocateCompileIdTest.java:71)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
	at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
	at java.base/java.lang.reflect.Method.invoke(Method.java:563)
	at com.sun.javatest.regtest.agent.MainWrapper$MainThread.run(MainWrapper.java:115)
	at java.base/java.lang.Thread.run(Thread.java:844)
Caused by: java.lang.ClassCastException: java.base/java.util.ImmutableCollections$MapN cannot be cast to java.base/java.util.Properties
	at jdk.internal.vm.compiler/org.graalvm.compiler.hotspot.HotSpotGraalCompilerFactory.getSavedProperties(HotSpotGraalCompilerFactory.java:215)
	... 26 more

-Doug

> On 19 Apr 2017, at 23:26, Doug Simon <doug.simon at oracle.com> wrote:
> 
> I've updated http://cr.openjdk.java.net/~dnsimon/8177845/hotspot/ with these changes:
> 
> 1. JVMCIServiceLocator.getProvider(Class<S>) is now protected
> 2. JVMCIServiceLocator.getProviders(Class<S>) now checks JVMCIPermission
> 3. Rename: jdk.vm.ci.services.internal.JDK9 -> jdk.vm.ci.services.internal.ReflectionAccessJDK
> 
> -Doug
> 
>> On 19 Apr 2017, at 23:12, Doug Simon <doug.simon at oracle.com> wrote:
>> 
>>> 
>>> On 19 Apr 2017, at 21:40, Christian Thalinger <cthalinger at twitter.com> wrote:
>>> 
>>>> 
>>>> On Apr 19, 2017, at 9:27 AM, Doug Simon <doug.simon at oracle.com> wrote:
>>>> 
>>>>> 
>>>>> On 19 Apr 2017, at 21:04, Mandy Chung <mandy.chung at oracle.com> wrote:
>>>>> 
>>>>> 
>>>>>> On Apr 19, 2017, at 11:55 AM, Christian Thalinger <cthalinger at twitter.com> wrote:
>>>>>> 
>>>>>> 
>>>>>>> On Apr 19, 2017, at 8:38 AM, Mandy Chung <mandy.chung at oracle.com> wrote:
>>>>>>> 
>>>>>>> Since jdk.internal.vm.compiler becomes an upgradeable module, it is not hashed with java.base to allow it to be upgraded and there is no integrity check.  Such qualified export will be granted to any module named jdk.internal.vm.compiler at runtime.  The goal is for upgradeable modules not to use any internal APIs and eliminate the qualified exports.
>>>>>>> 
>>>>>>> The main thing is that jdk.vm.ci.services API would need to be guarded if it’s used by non-Graal modules.
>>>>>> 
>>>>>> This all makes sense but where is the restriction that only jdk.internal.vm.compiler can use jdk.vm.ci.services?  
>>>>> 
>>>>> It's unqualified and no restriction in this change.
>>>> 
>>>> The public methods currently in jdk.vm.ci.services are:
>>>> 
>>>> 1. JVMCIServiceLocator.getProvider(Class<S>)
>>>> 2. JVMCIServiceLocator.getProviders(Class<S>)
>>>> 3. Services.initializeJVMCI()
>>>> 4. Services.getSavedProperties()
>>>> 5. Services.exportJVMCITo(Class<?>)
>>>> 6. Services.load(Class<S>)
>>>> 7. Services.loadSingle(Class<S>, boolean)
>>>> 
>>>> 1 should be made protected. I'll update the webrev with this change.
>>> 
>>> Good.
>>> 
>>>> 
>>>> 2 should check for JVMCIPermission. I'll update the webrev with this change.
>>> 
>>> Good.
>>> 
>>>> 
>>>> 3 is harmless from a security perspective in my opinion.
>>> 
>>> Would be good if one of Oracle’s security engineers could take a quick look just to be sure.
>> 
>> Vladimir, can you please bring this to the attention of the relevant engineer.
>> 
>>>> 
>>>> 4 checks for JVMCIPermission.
>>> 
>>> Ok.
>>> 
>>>> 
>>>> 5, 6 and 7 will be removed in a follow bug that updates Graal from upstream (and removes its usage of these methods).
>>> 
>>> About this, will this Graal update happen for JDK 9?
>> 
>> Yes.
>> 
>>> It’s awfully late in the cycle...
>> 
>> These are jigsaw related changes and I've been told jigsaw has an FC exception (although I don't exactly know what that is).
>> 
>> -Doug
>