8345296: AArch64: VM crashes with SIGILL when prctl is disallowed

[krk]

Ref. openjdk/jdk#22479

We have caught this in some prod environments, where prctl is forbidden by the sandboxing mechanism. This fails the JVM

Does not apply cleanly, the backport depends on FloatRegister changes from JDK-8339063 which are included in src/hotspot/cpu/aarch64/register_aarch64.hpp.

Additional testing:

• Verified the bug is fixed with the seccomp repro from JDK-8345296: Segfault without the patch, works with.

---

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
• JDK-8345296 needs maintainer approval

Issue

• JDK-8345296: AArch64: VM crashes with SIGILL when prctl is disallowed (Bug - P3 - Approved)

Reviewers

• Paul Hohensee (@phohensee - Reviewer) Review applies to 8ec87dce
• Aleksey Shipilev (@shipilev - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk21u-dev.git pull/1222/head:pull/1222
$ git checkout pull/1222

Update a local copy of the PR:
$ git checkout pull/1222
$ git pull https://git.openjdk.org/jdk21u-dev.git pull/1222/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 1222

View PR using the GUI difftool:
$ git pr show -t 1222

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk21u-dev/pull/1222.diff

Using Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back krk! 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]

@krk 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:

8345296: AArch64: VM crashes with SIGILL when prctl is disallowed

Reviewed-by: shade, phh
Backport-of: 3c60f0b2bb75150d49da9ab94d88b767275de5e2

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 3 new commits pushed to the master branch:

• 29c1188: 8314909: tools/jpackage/windows/Win8282351Test.java fails with java.lang.AssertionError: Expected [0]. Actual [1618]:
• a021113: 8320916: jdk/jfr/event/gc/stacktrace/TestParallelMarkSweepAllocationPendingStackTrace.java failed with "OutOfMemoryError: GC overhead limit exceeded"
• e7111b2: 8346108: [21u][BACKOUT] 8337994: [REDO] Native memory leak when not recording any events

Please see this link for an up-to-date comparison between the source branch of this pull request and the master branch.
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.

As you do not have Committer status in this project an existing Committer must agree to sponsor your change. Possible candidates are the reviewers of this PR (@phohensee, @shipilev) but any other Committer may sponsor as well.

➡️ To flag this PR as ready for integration with the above commit message, type /integrate in a new comment. (Afterwards, your sponsor types /sponsor in a new comment to perform the integration).

[openjdk]

This backport pull request has now been updated with issue from the original commit.

[mlbridge]

Webrevs

• 02: Full - Incremental (704459c2)
• 01: Full - Incremental (3fc28084)
• 00: Full (8ec87dce)

[phohensee]

/issue JDK-8339063

[openjdk]

@phohensee Only the author (@krk) is allowed to issue the /issue command.

[phohensee]

Please add "/issue JDK-8339063" (I'm unsure of the exact syntax, but it's something like that :)), since it's included. Otherwise lgtm.

[krk]

/issue JDK-8339063

[openjdk]

@krk
Adding additional issue to issue list: 8339063: [aarch64] Skip verify_sve_vector_length after native calls if SVE supports 128 bits VL only.

[openjdk]

⚠️ @krk This change is now ready for you to apply for maintainer approval. This can be done directly in each associated issue or by using the /approval command.

[shipilev]

Seeing the report for Corretto 21 that demonstrates this issue affects some MacOS and Alpine configs:
corretto/corretto-21#85

[shipilev]

I am not entirely convinced the modifications that JDK-8339063 does to this code are 100% safe. It tries to install the max vector size and exits the VM if not possible? That sounds like a behavioral change with unknown consequences. And unfortunately, that patch is only in JDK 24, so we do not yet know the impact well. Looks like we are mixing that backport because we need access to new FloatRegister constants?

If so, I think it would be safer to just take that particular hunk in register_aarch64.hpp only. Then if/when we do JDK-8339063 backport, it would slide right in. And meanwhile, we keep our reliability posture better.

[krk]

Thanks for the details. Yes, it was to access the new FloatRegister constants. Simplifying this now.

[openjdk]

@krk Please do not rebase or force-push to an active PR as it invalidates existing review comments. Note for future reference, the bots always squash all changes into a single commit automatically as part of the integration. See OpenJDK Developers’ Guide for more information.

[krk]

Picked only the necessary hunks from JDK-8339063.

[shipilev]

Also need /issue remove JDK-8339063 now?

[krk]

I only included an independeny chunk, not implementing JDK-8339063. It could be independently reviewed.

[krk]

/issue remove JDK-8339063

[openjdk]

@krk
Removing additional issue from issue list: 8339063.

[shipilev]

This looks correct.

[shipilev]

Right, include as well.

[krk]

/approval request Fix VM crash on production workloads when prctl is disallowed, on AArch64 with SVE.

[openjdk]

@krk
8345296: The approval request has been created successfully.

[mrserb]

@GoeLin We should consider fixing this bug in the upcoming January 2025 release. Jvm crash on different platforms, tested Alpine, AL2, AL2023, macOS 15.2+docker.
@shipilev please confirm

[shipilev]

@GoeLin We should consider fixing this bug in the upcoming January 2025 release. Jvm crash on different platforms, tested Alpine, AL2, AL2023, macOS 15.2+docker. @shipilev please confirm

Yes, we are planning to pick this fix in Corretto release for January. I think this patch is beneficial to have in all January releases as well. This is ultimately 21u and 17u maintainers' call.

[jerboaa]

/approve yes

Since this is related to prctl returning -1 in some Docker environments on newer MacOS 15.2+. It does not seem reproducible on native Aarch64 Linux or MacOSX 15.2 and there is a work-around, -XX:UseSVE=0, lets keep this on the April schedule for now. It should be a safe change, but I don't see the urgency either.

[openjdk]

@jerboaa
8345296: The approval request has been approved.

[krk]

/sponsor

[openjdk]

@krk Only Committers are allowed to sponsor changes.

[krk]

/integrate

[openjdk]

@krk
Your change (at version 704459c) is now ready to be sponsored by a Committer.

[shipilev]

/sponsor

[openjdk]

Going to push as commit 21b76f3.
Since your change was applied there have been 3 commits pushed to the master branch:

• 29c1188: 8314909: tools/jpackage/windows/Win8282351Test.java fails with java.lang.AssertionError: Expected [0]. Actual [1618]:
• a021113: 8320916: jdk/jfr/event/gc/stacktrace/TestParallelMarkSweepAllocationPendingStackTrace.java failed with "OutOfMemoryError: GC overhead limit exceeded"
• e7111b2: 8346108: [21u][BACKOUT] 8337994: [REDO] Native memory leak when not recording any events

Your commit was automatically rebased without conflicts.

[openjdk]

@shipilev @krk Pushed as commit 21b76f3.

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

[lprimak]

Hi @shipilev @krk

Looking a bit further into this, I do not understand what the "root cause" of this issue is.
Is this PR a workaround?
For example, running on any Arm machines I have access to (Mac, Graviton Linux, etc.) I cannot get the
-XX:UseSVE=1 to actually work.

What architecture does this support currently, if any?
It doesn't work on Mac M4, Mac M1, and under VM in Mac M1 or M4, etc.
Is the root cause somewhere in MacOS? If it is, I can't find one.

Thank you

[shipilev]

Looking a bit further into this, I do not understand what the "root cause" of this issue is. Is this PR a workaround?

This fix makes sure VM does not crash if lower layer (whether native OS or whatever sandbox/virtualization layer) does not allow us to execute prctl(PR_SVE_GET_VL) to get SVE vector length. Whether this is a fix or a workaround depends on whether prohibiting prctl(PR_SVE_GET_VL) is a bug or not from the lower layer perspective :) I tend to think the innocuous prctl-s should be enabled, so I tend to think about this fix as the workaround.

What architecture does this support currently, if any? It doesn't work on Mac M4, Mac M1, and under VM in Mac M1 or M4, etc. Is the root cause somewhere in MacOS? If it is, I can't find one.

By "this" you mean SVE? SVE is a feature of modern AArch64 CPUs. AFAIU, this means Apple M4+, Graviton 3+, etc.

[lprimak]

Ok. I think I get it now.
the only machine with SVE I have access to is M4 Mac. My Graviton machines are G1 so they wouldn’t work.

So the root cause is that Sonoma 15.2 introduced a regression that breaks SVE.
it makes it unavailable natively and crashes JVM under a virtualization environments

thank you.