RFR: 8347006: LoadRangeNode floats above array guard in arraycopy intrinsic
Aleksey Shipilev
shade at openjdk.org
Wed Jan 8 12:29:49 UTC 2025
On Wed, 8 Jan 2025 12:07:16 GMT, Tobias Hartmann <thartmann at openjdk.org> wrote:
> C2's arraycopy intrinsic adds guards that check that the source and destination objects are arrays:
> https://github.com/openjdk/jdk/blob/afe543414f58a04832d4f07dea88881d64954a0b/src/hotspot/share/opto/library_call.cpp#L5917-L5919
>
> If these guards pass, the array length is loaded:
> https://github.com/openjdk/jdk/blob/afe543414f58a04832d4f07dea88881d64954a0b/src/hotspot/share/opto/library_call.cpp#L5930-L5933
>
> But since the `LoadRangeNode` is not pinned, it might float above the array guard:
> https://github.com/openjdk/jdk/blob/afe543414f58a04832d4f07dea88881d64954a0b/src/hotspot/share/opto/graphKit.cpp#L1214
>
> If the object is not an array, we will read garbage. That's usually fine because the result will not be used (the array guard will trigger) but with `-XX:+UseCompactObjectHeaders` it can happen that the memory right after the header is not mapped and we crash.
>
> The fix is to add a `CheckCastPPNode` to propagate the information that the operand is an array and prevent the load from floating.
>
> Thanks to @shipilev for identifying the root cause!
>
> I was able to reliably reproduce the issue with `compiler/arraycopy/TestArrayCopyNoInit.java` and `-XX:-UseTLAB -XX:+UnlockExperimentalVMOptions -XX:-UseCompressedClassPointers` on Linux AArch64 and verified that the fix solves the problem.
>
> Best regards,
> Tobias
I have a question about the test :)
test/hotspot/jtreg/compiler/arraycopy/TestArrayCopyNoInit.java line 32:
> 30: * compiler.arraycopy.TestArrayCopyNoInit
> 31: * @run main/othervm -XX:-BackgroundCompilation -XX:-UseOnStackReplacement -XX:TypeProfileLevel=020
> 32: * -XX:+UnlockExperimentalVMOptions -XX:+UseCompactObjectHeaders -XX:-UseTLAB
You have been able to reproduce this with `-UseCompressedClassPointers`, right? If so, I'd suggest we do a run config with `-UseCCP` instead of `+UseCOH`, because this gives us a cleaner way for backports, if we need one later.
-------------
PR Review: https://git.openjdk.org/jdk/pull/22967#pullrequestreview-2537002862
PR Review Comment: https://git.openjdk.org/jdk/pull/22967#discussion_r1907099990
More information about the hotspot-compiler-dev
mailing list