[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-serviceability-dev
Subject:    Re: RFR: 8261236: C2: ClhsdbJstackXcompStress test fails when StressGCM is enabled
From:       Vladimir Kozlov <kvn () openjdk ! java ! net>
Date:       2021-07-27 15:51:29
Message-ID: yRCLT3lLpp5zrWoRmHJUa75QJqQ1xJb4JGxt27Ev-6M=.b0438c16-9520-4eb8-a362-c0a67457dca1 () github ! com
[Download RAW message or body]

On Fri, 16 Jul 2021 10:24:56 GMT, Nick Gasson <ngasson@openjdk.org> wrote:

> The clhsdb jstack command crashes when the debugged program is run with `-Xcomp \
> -XX:+StressGCM -XX:StressSeed=2` on AArch64: 
> 
> sun.jvm.hotspot.utilities.AssertionFailure: sanity check
> at jdk.hotspot.agent/sun.jvm.hotspot.utilities.Assert.that(Assert.java:32)
> at jdk.hotspot.agent/sun.jvm.hotspot.runtime.RegisterMap.setLocation(RegisterMap.java:160)
>  at jdk.hotspot.agent/sun.jvm.hotspot.compiler.ImmutableOopMapSet.updateRegisterMap(ImmutableOopMapSet.java:303)
>  at jdk.hotspot.agent/sun.jvm.hotspot.runtime.aarch64.AARCH64Frame.senderForCompiledFrame(AARCH64Frame.java:407)
>  
> 
> The assertion failure here is:
> 
> 
> Assert.that(0 <= i && i < regCount, "sanity check");
> 
> 
> I.e. there's an invalid register in the oop map.
> 
> The problem seems to be caused by the changes in JDK-8231586 which changed \
> `OopMapValue::oop_types` from a bit mask to normal integer enum. However the \
> changes in the C++ code weren't mirrored in SA's OopMapStream which still treats \
> OopTypes as a bit mask. 
> The particular oop map this is crashing on looks like this:
> 
> 
> ImmutableOopMap {[24]=Oop [32]=Oop [40]=Derived_oop_[24] } pc offsets: 324
> 
> 
> The code is looking for callee saved values (type=2) by AND-ing with each oop value \
> type in the set, so it wrongly interprets the derived oop [40] (type=3) as a callee \
> saved register. 
> This patch just mirrors the changes to the C++ code into the corresponding SA \
> classes.  The C++ OopMapStream constructor no longer takes a type filter argument \
> and callers are expected filter themselves, so I've made the same change to the \
> Java code. 
> This bug can also be seen using the clhsdb "disassemble" command.  For example the \
> above oop map is currently printed incorrectly as: 
> 
> OopMap:
> NarrowOops:[40]
> Callee saved:[40] = [24]
> Derived oops:[40] = [24]
> 
> 
> With this patch it becomes:
> 
> 
> OopMap:
> Oops:[24]  [32]
> Derived oops:[40] = [24]
> 
> 
> This bug was reported on AArch64 but it seems to be just luck that we don't see it \
> on other platforms. 
> Tested tier1 and hotspot_serviceability on AArch64 and x86.

Good.

-------------

Marked as reviewed by kvn (Reviewer).

PR: https://git.openjdk.java.net/jdk/pull/4807


[prev in list] [next in list] [prev in thread] [next in thread]