Document the JIT bootstrap process

[dsouzai]

Closes #12506

[dsouzai]

@a7ehuo could you please review?

[a7ehuo]

Thank you very much Irwin for documenting the JIT bootstrap process! I just have a few very minor comments

[dsouzai]

@a7ehuo good for another review

[a7ehuo]

Thank you for the update! I just have two minor comments

[a7ehuo]

Thank you for the update! LGTM

[dsouzai]

@0xdaryl could you review/merge?

[dsouzai]

@0xdaryl review reminder. If you're not able to, perhaps @mpirvu do you mind reviewing/merging?

[0xdaryl]

Sorry, this slipped off my radar. I'll review, but I would still appreciate @mpirvu's eyes on this as well.

[mpirvu]

Looks good to me.
Personally I would also want to see a comment on when the compilation threads are stopped and also when does "LIBRARIES_ONUNLOAD" get executed. In all my tests I see only "JVM_EXIT_STAGE"; "LIBRARIES_ONUNLOAD" frees some extra data structure, so would be nice to know how to test that code path.

[dsouzai]

In all my tests I see only "JVM_EXIT_STAGE"; "LIBRARIES_ONUNLOAD" frees some extra data structure, so would be nice to know how to test that code path.

It's quite a roundabout mechanism for LIBRARIES_ONUNLOAD. It looks like LIBRARIES_ONUNLOAD is invoked via runShutdownStage, but this happens really early in protectedInitializeJavaVM. It only invokes libraries that have set the FORCE_UNLOAD flag. The JIT does set this under the following condition when it is hooked in the DLL_LOAD_TABLE_FINALIZED stage.

         static bool TR_DisableFullSpeedDebug = feGetEnv2("TR_DisableFullSpeedDebug", (void *)vm)?1:0;
         if (!fullSpeedDebugSet || TR_DisableFullSpeedDebug)
            {
            loadInfo->loadFlags |= FORCE_UNLOAD;
            break;
            }

It seems that if we use the env var to disable FSD, we tell the VM to just unload the JIT.

[dsouzai]

@mpirvu @0xdaryl comments addressed.

[0xdaryl]

Docs only change. No need for CI.