8329528: G1 does not update TAMS correctly when dropping retained regions during Concurrent Start pause

[tschatzl]

Hi all,

please review this change to fix the problem that when g1 ages out regions from the retained region collection set candidate list during a concurrent start gc, the TAMSes of these regions are not set correctly.

The change achieves this by changing the order of some calculations in G1YoungCollector::pre_evacuate_collection_set: instead of setting TAMSes before collection set calculation, do it later, adjusting the predicate determining whether to update the TAMSes at concurrent start pause start to exclude regions in the collection set.

Several attempts for different fixes were tried, one of them is still viewable by only considering the changesets before the "alternate implementation" commit. However they quickly grew to fairly large size and complicated the code a lot.

Testing: tier1-3, test case not failing any more

Thanks,
Thomas

---

Progress

• Change must be properly reviewed (1 review required, with at least 1 Reviewer)
• Change must not contain extraneous whitespace
• Commit message must refer to an issue

Issue

• JDK-8329528: G1 does not update TAMS correctly when dropping retained regions during Concurrent Start pause (Bug - P1)

Reviewers

• Albert Mingkun Yang (@albertnetymk - Reviewer)
• Ivan Walulya (@walulyai - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/18692/head:pull/18692
$ git checkout pull/18692

Update a local copy of the PR:
$ git checkout pull/18692
$ git pull https://git.openjdk.org/jdk.git pull/18692/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 18692

View PR using the GUI difftool:
$ git pr show -t 18692

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/18692.diff

Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back tschatzl! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

@tschatzl This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8329528: G1 does not update TAMS correctly when dropping retained regions during Concurrent Start pause

Reviewed-by: ayang, iwalulya

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 45 new commits pushed to the master branch:

• 8817ba4: 8330000: ZGC: ZObjArrayAllocator may unnecessarily clear TypeArrays twice
• f778642: 8330024: [AIX] replace my_disclaim64 helper by direct call to disclaim64
• 7df4926: 8327741: JVM crash in hotspot/share/opto/compile.cpp - failed: missing inlining msg
• 2ceeb6c: 8328822: C2: "negative trip count?" assert failure in profile predicate code
• b04b304: 8329254: optimize integral reverse operations on x86 GFNI target.
• d9c84e7: 8329656: assertion failed in MAP_ARCHIVE_MMAP_FAILURE path: Invalid immediate -5 0
• 5e544f1: 8329491: GetThreadListStackTraces function should use JvmtiHandshake
• 643dd48: 8329432: PopFrame and ForceEarlyReturn functions should use JvmtiHandshake
• 70944ca: 8330033: com/sun/net/httpserver/bugs/B6431193.java fails in AssertionError after JDK-8326568
• 279ed0d: 8329961: Buffer overflow in os::Linux::kernel_version
• ... and 35 more: https://git.openjdk.org/jdk/compare/58e39c193211579316cfd09dc4117d510e317985...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

[openjdk]

@tschatzl The following label will be automatically applied to this pull request:

• hotspot-gc

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

[mlbridge]

Webrevs

• 00: Full (989da8eb)

[albertnetymk]

The fix looks good. However, in the long run, pulling out the logic around G1NumCollectionsKeepPinned from the cset construction seems cleaner. Also, NoteStartOfMarkHRClosure can keep the original simpler condition.

[walulyai]

LGTM!

[tschatzl]

The fix looks good. However, in the long run, pulling out the logic around G1NumCollectionsKeepPinned from the cset construction seems cleaner.

That is true at first glance, but when trying that out for this change, to me the drawbacks outweighed the cleanliness argument that resulted in much more code:

• extra loop over the collection set candidates, where that loop is at a different place in the code, called from pre_evacuate_collection_set. This creates an extra additional dependency when tracking dependencies there. One could imagine to add time tracking for this phase as well, resulting in more extra code.
• logging of this filtering will become disjointed from collection set selection. Dropping a region from the collection set candidates implies not selecting it, and reflecting this properly in the selection process takes additional effort unless one agrees with having separate log messages. (E.g. it is a bit surprising that when you know that there were pinned regions, the cset selection may actually tell you that there were none).
• this is very little extra code piggybacking it on the collection set selection

I think I considered this extra step for filtering when implementing this the first time, rejected it for all the extra effort (probably in the range of 100 LOC, see the diff for the 0c46d6a change - which is incomplete btw as it does not adjust the regression test and potentially add proper timing/logging) just to basically avoid this single if clause there.

Anyway, modifying this in this change at least it detracts from the problem at hand. I will file an RFE.

Also, NoteStartOfMarkHRClosure can keep the original simpler condition.

The additional clause is an arbitrary requirement/assumption by the existing code, asserting somewhere that cset regions do not have tams set initially.
All code paths for regions in the collection set update the TAMS, either clear again when freeing the region, or set it properly during evacuation failure, afair. I wanted to keep the initial assumption of cset regions having tams == bottom throughout.

Could you approve the change if it looks good to you?

[albertnetymk]

The fix is small and easy to backport; can probably further improve the logic around pinning&cset at later PRs.

[tschatzl]

Thanks @albertnetymk @walulyai for your reviews
/integrate

[openjdk]

Going to push as commit ff5c9a4.
Since your change was applied there have been 47 commits pushed to the master branch:

• 9acce7a: 8329774: Break long lines in jdk/src/jdk.hotspot.agent/doc /transported_core.html
• f0cd866: 8329704: Implement framework for proper handling of JDK_LIBS
• 8817ba4: 8330000: ZGC: ZObjArrayAllocator may unnecessarily clear TypeArrays twice
• f778642: 8330024: [AIX] replace my_disclaim64 helper by direct call to disclaim64
• 7df4926: 8327741: JVM crash in hotspot/share/opto/compile.cpp - failed: missing inlining msg
• 2ceeb6c: 8328822: C2: "negative trip count?" assert failure in profile predicate code
• b04b304: 8329254: optimize integral reverse operations on x86 GFNI target.
• d9c84e7: 8329656: assertion failed in MAP_ARCHIVE_MMAP_FAILURE path: Invalid immediate -5 0
• 5e544f1: 8329491: GetThreadListStackTraces function should use JvmtiHandshake
• 643dd48: 8329432: PopFrame and ForceEarlyReturn functions should use JvmtiHandshake
• ... and 37 more: https://git.openjdk.org/jdk/compare/58e39c193211579316cfd09dc4117d510e317985...master

Your commit was automatically rebased without conflicts.

[openjdk]

@tschatzl Pushed as commit ff5c9a4.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.