C2 crash in PhaseIdealLoop::build_loop_late_post_work

[Doerr, Martin]

Hi Christian,

JDK-8251994 doesn't resolve this issue. The crash has occurred after it was fixed.

Seems like replay doesn't help, because we'd need a generated Scala class:
java.lang.NoClassDefFoundError: org/apache/spark/sql/catalyst/expressions/GeneratedClass$SpecificUnsafeProjection
Error while parsing line 6957: Can't find holder klass

Debugging with gdb and core file is not easy, either.

Best regards,
Martin


> -----Original Message-----
> From: Christian Hagedorn <christian.hagedorn at oracle.com>
> Sent: Dienstag, 27. Oktober 2020 16:31
> To: Doerr, Martin <martin.doerr at sap.com>; 'hotspot-compiler-
> dev at openjdk.java.net' <hotspot-compiler-dev at openjdk.java.net>
> Subject: Re: C2 crash in PhaseIdealLoop::build_loop_late_post_work
> 
> Hi Martin
> 
> Thanks for the pointer! I will have a look at that.
> 
> Best regards,
> Christian
> 
> On 27.10.20 16:28, Doerr, Martin wrote:
> > Hi Christian,
> >
> > thanks for your reply.
> >
> > Interesting. Another issue related to JDK-8249749 was recently fixed:
> > 8251994: VM crashed running TestComplexAddrExpr.java test with -
> XX:UseAVX=X
> > https://bugs.openjdk.java.net/browse/JDK-8251994
> > Maybe it helps for this, too.
> >
> > I don’t have a fast reproducer atm.
> >
> > Best regards,
> > Martin
> >
> >
> >> -----Original Message-----
> >> From: Christian Hagedorn <christian.hagedorn at oracle.com>
> >> Sent: Dienstag, 27. Oktober 2020 13:03
> >> To: Doerr, Martin <martin.doerr at sap.com>; 'hotspot-compiler-
> >> dev at openjdk.java.net' <hotspot-compiler-dev at openjdk.java.net>
> >> Subject: Re: C2 crash in PhaseIdealLoop::build_loop_late_post_work
> >>
> >> Hi Martin
> >>
> >> Yes, this is a known problem (see JDK-8251925). We also see it failing
> >> with only this benchmark, often in log-regression. I played around with
> >> different settings by just running log-regression, for example, which
> >> did not seem to trigger it, though. Do you have a fast reproducer?
> >>
> >> It started to fail after JDK-8249749 which, however, just seem to have
> >> revealed the problem.
> >>
> >> Best regards,
> >> Christian
> >>
> >> On 27.10.20 12:07, Doerr, Martin wrote:
> >>> Hi,
> >>>
> >>> we observe C2 crashes on all x86_64 platforms when running
> Renaissance
> >> Benchmark "log-regression" since 2020-08-10.
> >>> (hg was at "8250848: [aarch64] nativeGotJump_at() missing call to
> verify()."
> >> when we've seen it first.)
> >>>
> >>>
> >>> #  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at
> >> pc=0x00007ffd594e9322, pid=1439804, tid=1516348
> >>>
> >>> #
> >>>
> >>> # JRE version: OpenJDK Runtime Environment (16.0.0.1) (build 16.0.0.1-
> >> internal+0-adhoc.openjdk.jdk-dev)
> >>>
> >>> # Java VM: OpenJDK 64-Bit Server VM (16.0.0.1-internal+0-
> >> adhoc.openjdk.jdk-dev, mixed mode, sharing, tiered, compressed oops,
> g1
> >> gc, windows-amd64)
> >>>
> >>> # Problematic frame:
> >>>
> >>> # V  [jvm.dll+0x559322]
> >> PhaseIdealLoop::build_loop_late_post_work+0x212
> >>>
> >>>
> >>>
> >>> Host: Intel(R) Xeon(R) CPU E7-8880 v4 @ 2.20GHz, 8 cores, 15G,
> Windows
> >> Server 2016 , 64 bit Build 14393 (10.0.14393.3630)
> >>>
> >>> Time: Sat Oct 24 19:40:57 2020 W. Europe Daylight Time elapsed time:
> >> 16.016253 seconds (0d 0h 0m 16s)
> >>>
> >>>
> >>> V  [jvm.dll+0x559322]
> PhaseIdealLoop::build_loop_late_post_work+0x212
> >> (loopnode.cpp:5312)
> >>>
> >>> V  [jvm.dll+0x5590b4]  PhaseIdealLoop::build_loop_late+0x214
> >> (loopnode.cpp:5159)
> >>>
> >>> V  [jvm.dll+0x558418]  PhaseIdealLoop::build_and_optimize+0x7b8
> >> (loopnode.cpp:3874)
> >>>
> >>> V  [jvm.dll+0x21410a]  Compile::optimize_loops+0x15a
> (compile.cpp:1960)
> >>>
> >>> V  [jvm.dll+0x20d900]  Compile::Optimize+0xf90  (compile.cpp:2184)
> >>>
> >>> V  [jvm.dll+0x20b105]  Compile::Compile+0xba5  (compile.cpp:735)
> >>>
> >>> V  [jvm.dll+0x19de6b]  C2Compiler::compile_method+0xab
> >> (c2compiler.cpp:104)
> >>>
> >>> Is this a known problem?
> >>>
> >>> Best regards,
> >>> Martin
> >>>