RFR: 8253180: ZGC: Implementation of JEP 376: ZGC: Concurrent Thread-Stack Processing

[Stefan Karlsson]

On Tue, 22 Sep 2020 11:43:41 GMT, Erik Österlund <eosterlund at openjdk.org> wrote:

> This PR the implementation of "JEP 376: ZGC: Concurrent Thread-Stack Processing" (cf.
> https://openjdk.java.net/jeps/376).
> Basically, this patch modifies the epilog safepoint when returning from a frame (supporting interpreter frames, c1, c2,
> and native wrapper frames), to compare the stack pointer against a thread-local value. This turns return polls into
> more of a swiss army knife that can be used to poll for safepoints, handshakes, but also returns into not yet safe to
> expose frames, denoted by a "stack watermark".  ZGC will leave frames (and other thread oops) in a state of a mess in
> the GC checkpoint safepoints, rather than processing all threads and their stacks. Processing is initialized
> automagically when threads wake up for a safepoint, or get poked by a handshake or safepoint. Said initialization
> processes a few (3) frames and other thread oops. The rest - the bulk of the frame processing, is deferred until it is
> actually needed. It is needed when a frame is exposed to either 1) execution (returns or unwinding due to exception
> handling), or 2) stack walker APIs. A hook is then run to go and finish the lazy processing of frames.  Mutator and GC
> threads can compete for processing. The processing is therefore performed under a per-thread lock. Note that disarming
> of the poll word (that the returns are comparing against) is only performed by the thread itself. So sliding the
> watermark up will require one runtime call for a thread to note that nothing needs to be done, and then update the poll
> word accordingly. Downgrading the poll word concurrently by other threads was simply not worth the complexity it
> brought (and is only possible on TSO machines). So left that one out.

Next set of comments.

src/hotspot/share/gc/z/zStackWatermark.cpp line 78:

> 76:   ZThreadLocalData::do_invisible_root(_jt, ZBarrier::load_barrier_on_invisible_root_oop_field);
> 77:
> 78:   ZVerify::verify_thread_frames_bad(_jt);

Every time I read the name verify_thread_no_frames_bad I read it as "verify that no frames are bad", but that's not at
all what that function does.

Is there still a reason why verify_thread_no_frames_bad and verify_thread_frames_bad are split up into two functions?
Or could we fuse them into a ZVerify::verify_thread_bad?

src/hotspot/share/gc/z/zBarrier.cpp line 130:

> 128: uintptr_t ZBarrier::load_barrier_on_invisible_root_oop_slow_path(uintptr_t addr) {
> 129:   return during_relocate() ? relocate(addr) : mark<DontFollow, Strong, Publish>(addr);
> 130: }

There's a style skew between load_barrier_on_oop_slow_path and load_barrier_on_invisible_root_oop_slow_path. The former
calls wrapper function relocate_or_mark, which does the during_relocate() ... check. Maybe do the same for
load_barrier_on_invisible_root_oop_slow_path, and introduce a relocate_or_mark_no_follow?

src/hotspot/share/gc/z/zBarrierSet.cpp line 85:

> 83:   ZThreadLocalData::set_address_bad_mask(thread, ZAddressBadMask);
> 84:   if (thread->is_Java_thread()) {
> 85:     JavaThread* const jt = static_cast<JavaThread*>(thread);

Use as_Java_thread() here?

src/hotspot/share/gc/z/zCollectedHeap.cpp line 235:

> 233:   return true;
> 234: }
> 235:

Weird placement between store barrier functions. But even weirder that we still have those functions. I'll remove them
with JDK-8253516.

src/hotspot/share/gc/z/zDriver.cpp line 108:

> 106:     return false;
> 107:   }
> 108:

Group needs_inactive_gc_locker, skip_thread_oop_barriers, and allow_coalesced_vm_operations together?

Add a comment about why we chose to skip coalescing here.

src/hotspot/share/gc/z/zHeapIterator.cpp line 90:

> 88:   virtual void do_thread(Thread* thread) {
> 89:     if (thread->is_Java_thread()) {
> 90:       StackWatermarkSet::finish_processing(static_cast<JavaThread*>(thread), NULL /* context */,
> StackWatermarkKind::gc);

as_Java_thread() ?

src/hotspot/share/gc/z/zNMethod.cpp line 244:

> 242:   ZNMethodToOopsDoClosure(OopClosure* cl, bool should_disarm_nmethods) :
> 243:     _cl(cl),
> 244:     _should_disarm_nmethods(should_disarm_nmethods) {}

Indentation is off.

src/hotspot/share/gc/z/zObjectAllocator.cpp line 2:

> 1: /*
> 2:  * Copyright (c) 2015, 2020, Oracle and/or its affiliates. All rights reserved.

Revert.

src/hotspot/share/gc/z/zRootsIterator.cpp line 200:

> 198:   ZRootsIteratorClosure* const _cl;
> 199:   // The resource mark is needed because interpreter oop maps are not reused in concurrent mode.
> 200:   // Instead, they are temporary, and resource allocated.

temporary, and => temporary and ?

src/hotspot/share/gc/z/zStackWatermark.cpp line 96:

> 94: void ZStackWatermark::process(const frame& fr, RegisterMap& register_map, void* context) {
> 95:   ZVerify::verify_frame_bad(fr, register_map);
> 96:   frame(fr).oops_do(closure_from_context(context), &_cb_cl, &register_map);

With recent frame::oops_do cleanups, frame(fr) can be changed to simply fr.

src/hotspot/share/gc/z/zVerify.cpp line 377:

> 375: void ZVerify::verify_frame_bad(const frame& fr, RegisterMap& register_map) {
> 376:   ZVerifyBadOopClosure verify_cl;
> 377:   frame(fr).oops_do(&verify_cl, NULL, &register_map);

With recent frame::oops_do cleanups, frame(fr) can be changed to simply fr.

src/hotspot/share/gc/z/zStackWatermark.hpp line 46:

> 44:   ZOnStackCodeBlobClosure();
> 45:
> 46:   virtual void do_code_blob(CodeBlob* cb);

Could be private.

src/hotspot/share/gc/z/zStackWatermark.hpp line 57:

> 55:   OopClosure* closure_from_context(void* context);
> 56:
> 57: protected:

No need to be protected, so can be private. We don't have sub-classes that overrides ZStackWatermark.

-------------

PR: https://git.openjdk.java.net/jdk/pull/296