Postpone registering for exception events

[keithc-ca]

In Java 25+, jdk.internal.misc.Unsafe.INVALID_FIELD_OFFSET has type long, not int, so a NoSuchFieldError was thrown by GetStaticFieldID() called within sun.misc.Unsafe.registerNatives() (prior to #22949). Normally that error is just ignored, but if a user supplies an -Xdump option which requires checking the call stack, that check fails because the offset of Throwable.walkback has not been captured for use by J9VMJAVALANGTHROWABLE_WALKBACK(), etc. The result is that the lock word (at offset zero after the header) is read and interpreted as a heap reference. On must platforms, the initial lock word is zero which behaves like null which iterateStackTraceImpl() handles as an unavailable walkback object. On POWER, however, the initial lock is normally non-zero (but not a valid heap reference), resulting in a crash.

This change avoids problems like this early in the lifetime of a JVM by not registering dump hooks for exception events until after all the known classes, fields and methods have been resolved (including the offset of Throwable.walkback).

The underlying issue motivated these related changes:

1. the removal of useless/problematic code from sun.misc.Unsafe.registerNatives() (Remove useless/problematic code from sun.misc.Unsafe.registerNatives() #22949)
2. a general cleanup of trigger.c, with an improvement to error checking (Reserve hooks that are postponed so they can be registered later #22947)

Related to RTC 153110.

[pshipton]

jenkins test sanity plinux,amac jdk25

[pshipton]

Pls check out the internal RAS failures http://vmfarm.rtp.raleigh.ibm.com/jobs_by_status.php?build_id=105877&status=FAILED

Some of these may be expected, but there are many more failures than usually occur. You can compare against the ibuild.

[keithc-ca]

I expect the failures are related to https://github.com/eclipse-openj9/openj9/blob/master/runtime/rasdump/dmpsup.c#L544-L547 which seems to be negatively affected by the delay installing the exception-related hooks.

[keithc-ca]

I do note that the failures in rasDumpEnableHooks() are silently ignored, except the last attempt. That's probably something that should be fixed as well.

[keithc-ca]

I think I'll improve the cleanup part of this and deliver it separately. In draft mode until after that.

[keithc-ca]

Update: The third commit has already been merged via #22949. The first commit is superseded by #22947. The result, then, will be a single commit (a rebased copy of 0ace927).

[pshipton]

jenkins test sanity amac jdk11