CRIU restore with -Xdump using CRIUSupport.registerRestoreOptionsFile()

[JasonFengJ9]

Added an internal JVM checkpoint hook criuRestoreInitializeDump() invoking criuReloadXDumAgents() after CRIU restore;
Refactored configureDumpAgents() to handle both startup and CRIU restore scenarios;
Not disable dump agent hooks within JIT for CRIU builds;
Added a test to verify -Xdump:java:events=vmstop and -Xdump:java:events=throw,filter=java/lang/OutOfMemoryError,request=exclusive+prepwalk+serial+preempt.

Signed-off-by: Jason Feng fengj@ca.ibm.com

[JasonFengJ9]

@dsouzai could you please review JIT changes?
FYI @vijaysun-omr

[dsouzai]

The JIT changes look fine; however I don't think they're needed. The change in J9Options.cpp only executes once during startup, so it doesn't really impact anything post-restore. As for the change in VMJ9.cpp, the call to ((*vmHooks)->J9HookDisable(vmHooks, J9HOOK_VM_EXCEPTION_CATCH) != 0); essentially boils down to

		if (HOOK_FLAGS(commonInterface, eventNum) & (J9HOOK_FLAG_RESERVED | J9HOOK_FLAG_HOOKED)) {
			rc = -1;
		}

in hookable.cpp which is basically what you've added in the J9VM_OPT_CRIU_SUPPORT guarded sections.

I think the calls to J9HookDisable in the compiler to determine whether the hook is enabled should be cleaned up, but you don't need to bother with it in this PR; I'll open a PR to clean it all up.

Also, on a related note, is it possible for J9HOOK_FLAG_HOOKED to be set but not J9HOOK_FLAG_RESERVED?

[JasonFengJ9]

I think the calls to J9HookDisable in the compiler to determine whether the hook is enabled should be cleaned up, but you don't need to bother with it in this PR; I'll open a PR to clean it all up.

Assuming the cleanup is to replace J9HookDisable with J9_EVENT_IS_HOOKED() || J9_EVENT_IS_RESERVED(), -Xdump at restore won't work if the hook is disabled.
Will remove these JIT changes, and mark this PR depending on the JIT change.

is it possible for J9HOOK_FLAG_HOOKED to be set but not J9HOOK_FLAG_RESERVED?

J9HookRegisterWithCallSitePrivate always set J9HOOK_FLAG_HOOKED | J9HOOK_FLAG_RESERVED.
J9HookUnregister only clears J9HOOK_FLAG_HOOKED.
So it is possible for J9HOOK_FLAG_RESERVED set but not J9HOOK_FLAG_HOOKED.
On the other hand, it won't be the case that J9HOOK_FLAG_HOOKED is set but not J9HOOK_FLAG_RESERVED.

[dsouzai]

Spoke to Jason offline. Turns out the compiler changes are required. The reason being that if pre-checkpoint no -Xdump option is set then J9HookDisable will set the J9HOOK_FLAG_DISABLED when called (because J9HOOK_FLAG_RESERVED/J9HOOK_FLAG_HOOKED wont be set). However this means that post-restore, it won't be possible to enable (reserve) this event.

I'll open a PR to clean the JIT calls later, but for 0.38 the compiler changes ensure that we don't set the J9HOOK_FLAG_DISABLED flag.

[tajila]

@JasonFengJ9 Please rebase your changes

[JasonFengJ9]

@tajila just rebased, please have a look.

[tajila]

jenkins test sanity xlinux jdk17

[tajila]

jenkins compile win jdk8

[JasonFengJ9]

Earlier builds:
https://openj9-jenkins.osuosl.org/job/Build_JDK8_x86-64_windows_Personal/850/
https://openj9-jenkins.osuosl.org/job/Build_JDK17_x86-64_linux_Personal/442/
https://openj9-jenkins.osuosl.org/job/Test_openjdk17_j9_sanity.functional_x86-64_linux_Personal/278/

@tajila this is ready for another look.

[tajila]

jenkins test sanity xlinux jdk17

[tajila]

@JasonFengJ9 please rebase

[JasonFengJ9]

@tajila just rebased

[tajila]

jenkins test sanity xlinux jdk17

[tajila]

Only needed compile, will merge now

[tajila]

@JasonFengJ9 please make a PR for 0.38

[JasonFengJ9]

@tajila please review

• (v0.38.0-release) CRIU restore with -Xdump using CRIUSupport.registerRestoreOptionsFile()  #16950