Segfault running Dalli test suite

[nirvdrum]

While working on fixing a TruffleRuby compatibility issue with Dalli, a popular memcache client, I encountered a segfault while running its test suite. I haven't seen the segfault every time. I don't know what the frequency is.

I'm running the latest TruffleRuby release on macOS ARM, testing 4c4a5a2354707604456f6f1bf08d020f1909b49e of Dalli:

> ruby -v
truffleruby 23.1.2, like ruby 3.2.2, Oracle GraalVM Native [aarch64-darwin]

Segfault log

I'll try out dev builds and see if it's still a problem there. If not, at least we'll have a log with an open source project to refer to.

Fatal error: No exception handler registered for deopt target
encodedBci: 37 (bci 5 rethrowException)
Method info: org.truffleruby.language.control.WhileNode.execute(WhileNode.java:37)
Partial Deoptimized Stack
org.truffleruby.language.control.WhileNode.execute(WhileNode.java:37)

[nirvdrum]

I'm still playing with different versions and combinations to see which TruffleRuby builds are affected. So far I've managed to experience the segfault with three different configurations on x86_64 Linux (Ubuntu 23.10). The point at which it segfaults differs in each case, but they all happen with rethrowException.

• 23.1.2 GFTC Native log
• 24.1.0-dev GFTC Native log
• 24.1.0-dev CE Native log

[eregon]

Thanks for the report and trying on various versions.
Fatal error: No exception handler registered for deopt target certainly feels like Native Image bug.

Could you post detailed repro instructions so I can share them on the internal issue?
Did you just run bundle exec rake? With/without RUN_SASL_TESTS=1? On which branch/commit?
EDIT: from this log it's clear it's bundle exec rake.

[eregon]

I'll regroup the header of each log here for convenience:

truffleruby 23.1.2, like ruby 3.2.2, Oracle GraalVM Native [aarch64-darwin]
Fatal error: No exception handler registered for deopt target
encodedBci: 37 (bci 5 rethrowException)
Method info: org.truffleruby.language.control.WhileNode.execute(WhileNode.java:37)
Partial Deoptimized Stack
org.truffleruby.language.control.WhileNode.execute(WhileNode.java:37)

---

truffleruby 24.1.0-dev-eedda850, like ruby 3.2.2, Oracle GraalVM Native [x86_64-linux]
Fatal error: No exception handler registered for deopt target
encodedBci: 313 (bci 74 rethrowException)
Method info: org.truffleruby.core.array.ArrayEachIteratorNode.iterateMany(ArrayEachIteratorNode.java:66)
Partial Deoptimized Stack
org.truffleruby.core.array.ArrayEachIteratorNode.iterateMany(ArrayEachIteratorNode.java:66)

---

truffleruby 24.1.0-dev-677ac08b, like ruby 3.2.2, GraalVM CE Native [x86_64-linux]
Fatal error: No exception handler registered for deopt target
encodedBci: 113 (bci 24 rethrowException)
Method info: org.truffleruby.language.RubyProcRootNode.execute(RubyProcRootNode.java:77)
Partial Deoptimized Stack
org.truffleruby.language.RubyProcRootNode.execute(RubyProcRootNode.java:77)

---

truffleruby 23.1.2, like ruby 3.2.2, Oracle GraalVM Native [x86_64-linux]
Fatal error: No exception handler registered for deopt target
encodedBci: 313 (bci 74 rethrowException)
Method info: org.truffleruby.core.array.ArrayEachIteratorNode.iterateMany(ArrayEachIteratorNode.java:66)
Partial Deoptimized Stack
org.truffleruby.core.array.ArrayEachIteratorNode.iterateMany(ArrayEachIteratorNode.java:66)

So they are all Fatal error: No exception handler registered for deopt target but for different methods.

And the stacktraces are all similar:

Stacktrace for the failing thread 0x00007f6a88004f80 (A=AOT compiled, J=JIT compiled, D=deoptimized, i=inlined):
  i  SP 0x00007f6a911fa830 IP 0x00007f6b58152550 size=48    [image code] com.oracle.svm.core.jdk.VMErrorSubstitutions.shutdown(VMErrorSubstitutions.java:148)
  A  SP 0x00007f6a911fa830 IP 0x00007f6b58152550 size=48    [image code] com.oracle.svm.core.jdk.VMErrorSubstitutions.shouldNotReachHere(VMErrorSubstitutions.java:141)
  A  SP 0x00007f6a911fa860 IP 0x00007f6b58226954 size=16    [image code] com.oracle.svm.core.util.VMError.shouldNotReachHere(VMError.java:90)
  A  SP 0x00007f6a911fa870 IP 0x00007f6b580cb305 size=80    [image code] com.oracle.svm.core.deopt.Deoptimizer.fatalDeoptimizationError0(Deoptimizer.java:1322)
  i  SP 0x00007f6a911fa8c0 IP 0x00007f6b580c553c size=48    [image code] com.oracle.svm.core.deopt.Deoptimizer.fatalDeoptimizationError(Deoptimizer.java:1301)
  A  SP 0x00007f6a911fa8c0 IP 0x00007f6b580c553c size=48    [image code] com.oracle.svm.core.deopt.DeoptimizedFrame.throwMissingExceptionHandler(DeoptimizedFrame.java:413)
  A  SP 0x00007f6a911fa8f0 IP 0x00007f6b580c5457 size=80    [image code] com.oracle.svm.core.deopt.DeoptimizedFrame.takeException(DeoptimizedFrame.java:402)
  i  SP 0x00007f6a911fa940 IP 0x00007f6b5820203f size=224   [image code] com.oracle.svm.core.snippets.ExceptionUnwind.deoptTakeExceptionInterruptible(ExceptionUnwind.java:266)
  A  SP 0x00007f6a911fa940 IP 0x00007f6b5820203f size=224   [image code] com.oracle.svm.core.snippets.ExceptionUnwind.defaultUnwindException(ExceptionUnwind.java:223)
  A  SP 0x00007f6a911faa20 IP 0x00007f6b58202e70 size=32    [image code] com.oracle.svm.core.snippets.ExceptionUnwind.unwindExceptionInterruptible(ExceptionUnwind.java:129)
  A  SP 0x00007f6a911faa40 IP 0x00007f6b582031ce size=1232  [image code] com.oracle.svm.core.snippets.ExceptionUnwind.unwindExceptionWithCalleeSavedRegisters(ExceptionUnwind.java:106)
  A  SP 0x00007f6a911faf10 IP 0x00007f6b5825a8f5 size=1280  [image code] com.oracle.svm.truffle.api.SubstrateThreadLocalHandshake.pollStub(SubstrateThreadLocalHandshake.java)
  D  SP 0x00007f6a911fb410 IP 0x00007f6b580ca190 size=64    [image code, deopt] org.truffleruby.language.RubyProcRootNode.execute**(RubyProcRootNode.java:77)
  D  SP 0x00007f6a911fb410 IP 0x00007f6b580ca190 size=64    [image code, deopt] com.oracle.truffle.runtime.OptimizedCallTarget.executeRootNode**(OptimizedCallTarget.java:745)

This looks like a TruffleSafepoint poll that ends up throwing and then triggers this error.

[eregon]

FWIW I tried to repro just raising an exception in another thread but that didn't trigger this issue (maybe because it doesn't deopt):

ruby -e 't=Thread.new { sleep 1; 100.times { Thread.pass until Thread.main.backtrace&.first&.include? "each"; Thread.main.raise "bye" } }; a=Array.new(100_000_000, 0); while true; begin; a.each { |e| }; rescue => e; p e.class; end; end'

I will try to repro on dalli.

[eregon]

Filed as GR-51913 internally.