8361752: Double free in CompileQueue::delete_all after JDK-8357473 #26294

shipilev · 2025-07-14T13:57:09Z

See the bug for more analysis.

The short summary is that CompileQueue::delete_all walks the entire compile queue and deletes the tasks. It normally goes smoothly, unless there are blocking tasks. Then, the actual waiters have to delete the task, lest we delete the task under waiter's feet. Full deletion and blocking waits coordinate with waiting_for_completion_count counter. This mechanism -- added by JDK-8343938 in JDK 25 to solve a similar problem -- almost works. Almost.

There is a subtle race window, where blocking waiter could have already unparked, dropped waiting_for_completion_count to 0 and proceeded to delete the task, see CompileBroker::wait_for_completion(). Then the queue deletion code could assume there are no actual waiters on the blocking task, and proceed to delete the task again. Before JDK-8357473 this race was fairly innocuous, as second attempt at insertion into the free list was benign. But now, CompileTask-s are delete-d, and the second attempt leads to double free.

I suspect we can fix that by complicating the coordination protocol even further, e.g. by tracking the counters more thoroughly. But, recognizing CompileQueue::delete_all() is basically only called from the compiler shutdown code (things are already bad), and it looks completely opportunistic (it does not delete the whole compiler threads, so skipping synchronous deletes on a few compile tasks are not a big deal), we should strive to simplify it.

This PR summarily delegates all blocking task deletes to waiters. I think it stands to reason (and can be seen in CompilerBroker code) that if a blocking task is in queue, then there is a waiter that would call CompileBroker::wait_for_completion() on it.

Additional testing:

Linux AArch64 server fastdebug, tier1
Linux x86_64 server fastdebug, all

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-8361752: Double free in CompileQueue::delete_all after JDK-8357473 (Bug - P3)

Reviewers

Vladimir Kozlov (@vnkozlov - Reviewer)
Vladimir Ivanov (@iwanowww - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/26294/head:pull/26294
$ git checkout pull/26294

Update a local copy of the PR:
$ git checkout pull/26294
$ git pull https://git.openjdk.org/jdk.git pull/26294/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 26294

View PR using the GUI difftool:
$ git pr show -t 26294

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/26294.diff

Using Webrev

Link to Webrev Comment

bridgekeeper · 2025-07-14T13:58:26Z

👋 Welcome back shade! 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 · 2025-07-14T13:59:30Z

@shipilev 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:

8361752: Double free in CompileQueue::delete_all after JDK-8357473

Reviewed-by: kvn, vlivanov

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 86 new commits pushed to the master branch:

03230f8: 8351073: [macos] jpackage produces invalid Java runtime DMG bundles
a3843e8: 8361751: Test sun/tools/jcmd/TestJcmdSanity.java timed out on Windows
60c29ff: 8362524: Fix confusing but harmless typos in x86 CPU Features
... and 83 more: https://git.openjdk.org/jdk/compare/14c79be1613c9d737a9536087ac48914ee4ba8d9...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 · 2025-07-14T14:00:18Z

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

hotspot-compiler

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 · 2025-07-14T15:27:51Z

Webrevs

01: Full - Incremental (76bfa8d1)
00: Full (13625998)

shipilev · 2025-07-14T15:33:25Z

I am pretty convinced this is it. But I still struggle to reproduce the failure locally. So I would appreciate if @TobiHartmann or @dholmes-ora could give it a spin through the CI where this reproduces. Probably after JDK-8360048 lands, if that one is not a test-only bug?

mhaessig · 2025-07-14T16:11:15Z

I kicked off a CI run. I'll keep you posted on the results.

vnkozlov · 2025-07-14T17:16:07Z

src/hotspot/share/compiler/compileBroker.cpp

+  // Wake up all blocking task waiters to delete all remaining blocking
+  // tasks. This is not a performance sensitive path, so we do this
+  // unconditionally to simplify coding.
+  {
+    MonitorLocker ml(Thread::current(), CompileTaskWait_lock);
+    ml.notify_all();
+  }
+


What about other compiler threads which still in process of compiling for blocking tasks? They still need it CompileTask object.
delete_all() is called by one compiler thread which finished compilation but other threads may not.

I don't see any compiler thread checks shut_down state to stop compilation.

AFAIU, that's the point of the existing protocol to force waiters to delete the task: the blocking waiter would wait for compiler thread to complete the task one way or the other. This PR makes that protocol even stronger: only blocking waiters are allowed to delete the blocking task.

Ah, your question is what happens if we notify here, and compilations are still running? Well, I think current protocol should nominally allow waiters to wait until compilation is over and then allow them to delete the task. But then I see wait_for_compilation can exit when compilation is shut down:

while (!task->is_complete() && !is_compilation_disabled_forever()) { ml.wait(); }

This will proceed to delete the task while compiler thread is running. Grrr. Looks to be another hole in this protocol.

Can compiler thread delete its own blocking task when it finished. And let Java thread resume execution when compilation disabled as it do now but do nothing about task in such case?

I don't think that works. There is no "own" blocking task, there are nearly always two threads involved: the compiler thread and the waiter (Java) thread. Waiter is checking the task status under the lock. Logically, the last user should delete the task, that is waiter.

But I think we can handle this hole by ignoring the blocking task deletion during compiler shutdown. For the same reason described in PR body: we already leave cruft behind in that case, and it costs us quite a bit of complexity to deal with every corner case during shutdown. So it seems simpler to just drop the tasks on the floor in that corner case.

I did a variant of this in new commit, seems to still work well under stress testing. More testing is running now...

…task

mhaessig · 2025-07-15T11:52:04Z

I kicked off a CI run.

FWIW, tier1-tier3, and 100 repeats of TestStressBailout.java on Linux x64 & aarch64, Windows x64, and Mac aarch64 all passed.

Let me know when I should kick off another round.

shipilev · 2025-07-15T12:42:34Z

FWIW, tier1-tier3, and 100 repeats of TestStressBailout.java on Linux x64 & aarch64, Windows x64, and Mac aarch64 all passed.

Let me know when I should kick off another round.

Thank you, that is good to know!

New version handles even more obscure corner case, that I doubt would show up easily :) My Linux x86_64 server fastdebug make test TEST=all run just completed without problems, so we can test this version more broadly as well.

mhaessig · 2025-07-16T08:42:24Z

[...] we can test this version more broadly as well.

tier1 - tier3 and 100 repeats of TestStressBailout.java on Linux x64 & aarch64, Windows x64, and Mac x64 & aarch64 all passed.

vnkozlov

Good.

shipilev · 2025-07-16T17:24:06Z

Thanks! I think I need another Review.

vnkozlov · 2025-07-18T17:56:26Z

@iwanowww please look?

iwanowww

Looks good.

shipilev · 2025-07-21T06:03:40Z

Thank you! I re-tested locally after local merge with current master, and it still works. Here goes.

/integrate

openjdk · 2025-07-21T06:04:21Z

Going to push as commit 9609f57.
Since your change was applied there have been 91 commits pushed to the master branch:

441dbde: 8362556: New test jdk/jfr/event/io/TestIOTopFrame.java is failing on all platforms
ee0bcc5: 8362379: Test serviceability/HeapDump/UnmountedVThreadNativeMethodAtTop.java should mark as /native
ceb51d4: 8362829: Exclude CDS test cases after JDK-8361725
... and 88 more: https://git.openjdk.org/jdk/compare/14c79be1613c9d737a9536087ac48914ee4ba8d9...master

Your commit was automatically rebased without conflicts.

openjdk · 2025-07-21T06:04:35Z

@shipilev Pushed as commit 9609f57.

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

shipilev added 4 commits July 14, 2025 14:02

Sample fix

0300eaa

Remove ProblemList entry

5ec4a3d

Comment touchup

47d1583

More comment touchups

1362599

openjdk bot added the hotspot-compiler hotspot-compiler-dev@openjdk.org label Jul 14, 2025

shipilev marked this pull request as ready for review July 14, 2025 15:23

openjdk bot added the rfr Pull request is ready for review label Jul 14, 2025

vnkozlov reviewed Jul 14, 2025

View reviewed changes

Also handle the corner case when compiler threads might be using the …

76bfa8d

…task

vnkozlov approved these changes Jul 16, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Jul 16, 2025

iwanowww approved these changes Jul 18, 2025

View reviewed changes

openjdk bot added the integrated Pull request has been integrated label Jul 21, 2025

openjdk bot closed this Jul 21, 2025

openjdk bot removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Jul 21, 2025

8361752: Double free in CompileQueue::delete_all after JDK-8357473 #26294

8361752: Double free in CompileQueue::delete_all after JDK-8357473 #26294

Uh oh!

Conversation

shipilev commented Jul 14, 2025 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Issue

Reviewers

Reviewing

Uh oh!

bridgekeeper bot commented Jul 14, 2025

Uh oh!

openjdk bot commented Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openjdk bot commented Jul 14, 2025

Uh oh!

mlbridge bot commented Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

shipilev commented Jul 14, 2025

Uh oh!

mhaessig commented Jul 14, 2025

Uh oh!

vnkozlov Jul 14, 2025

Choose a reason for hiding this comment

Uh oh!

shipilev Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shipilev Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vnkozlov Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shipilev Jul 15, 2025

Choose a reason for hiding this comment

Uh oh!

mhaessig commented Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

shipilev commented Jul 15, 2025

Uh oh!

mhaessig commented Jul 16, 2025

Uh oh!

vnkozlov left a comment

Choose a reason for hiding this comment

Uh oh!

shipilev commented Jul 16, 2025

Uh oh!

vnkozlov commented Jul 18, 2025

Uh oh!

iwanowww left a comment

Choose a reason for hiding this comment

Uh oh!

shipilev commented Jul 21, 2025

Uh oh!

openjdk bot commented Jul 21, 2025

Uh oh!

openjdk bot commented Jul 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

4 participants

shipilev commented Jul 14, 2025 •

edited by openjdk bot

Loading

openjdk bot commented Jul 14, 2025 •

edited

Loading

mlbridge bot commented Jul 14, 2025 •

edited

Loading

shipilev Jul 14, 2025 •

edited

Loading

shipilev Jul 14, 2025 •

edited

Loading

vnkozlov Jul 14, 2025 •

edited

Loading

mhaessig commented Jul 15, 2025 •

edited

Loading