Find a strange problem about using project-panama on macos

[David Holmes]

cc mailing list

On 14/08/2023 6:51 pm, 刘希晨 wrote:
> Sorry I am not an export in native debugging, it took me several hours 
> finding the actual line causing jconsole and jmap not working properly 
> on macos, and except for that, the application works perfectly well in 
> my computer, that's the reason why I haven't found the case before, so I 
> guess it's not a problem about safepoint, because if this single line 
> would cause the application not able to reach a safepoint, why would it 
> work on windows, and the application could GC well (I tested add 
> System.gc() in the above main method, and it could be done well, I think 
> that should suggest the safepoint could be reached, perhaps I am wrong, 
> I don't know much about safepoint but it seems GC needs to reach a 
> safepoint to run), it runs completely normally except can't be accessed 
> by jmap and jconsole (By the way, the application could be debugged by 
> idea properly, which is quite suprising, I am using Idea version 
> 2023.2), so I guess the openjdk21-ea implementation on macos of 
> SymbolLookup should have some problems internally.

When we are dealing with native code different platforms have different 
native code and so they can behave differently. The symptoms you 
describe suggest an inability of the VM to reach a safepoint.

But without getting a native stack dump from a debugger there is no way 
to investigate this further.

Also please respond to the mailing list rather than directly to myself, 
so others can see what is happening.

David

> David Holmes <david.holmes at oracle.com <mailto:david.holmes at oracle.com>> 
> 于2023年8月14日周一 12:51写道：
> 
>     On 14/08/2023 2:45 pm, David Holmes wrote:
>      >
>      >
>      > On 14/08/2023 12:42 pm, 刘希晨 wrote:
>      >>
>      >> I have found a really strange problem when using project-panama on
>      >> macos, and it might be a bug.
>      >>
>      >> *Environment*: JDK-21 ea build 34, cpu : M1 Pro, operating system :
>      >> Ventura 13.5
>      >>
>      >> Here are the minimal reproducible codes:
>      >>
>      >> public class SimpleExample {
>      >>      public static void main(String[] args)throws Throwable {
>      >>          SymbolLookup crypto =
>      >>
>     SymbolLookup.libraryLookup("/Users/liuxichen/workspace/tenet-lib/lib/libcrypto.dylib", Arena.global());
>      >>          Thread.sleep(Long.MAX_VALUE);
>      >>      }
>      >> }
>      >>
>      >> The libcrypto.dylib file is the OPENSSL dynamic library file, I
>     think
>      >> other libraries probably could cause the same problem.
>      >>
>      >> *Problem description*:
>      >> A normal java program would work fine with *jconsole* and
>     *jmap*, if
>      >> running the above code in macos, jconsole won't be able to
>     connect to
>      >> the target process, and jmap will give the following error:
>      >>
>      >> *~/Dow/jdk-21/C/H/bin  jps*
>      >>
>      >> 1665 RemoteMavenServer36
>      >> 2673 Jps
>      >> 532
>      >> 1671 Launcher
>      >> 1672 SimpleExample
>      >> 1211
>      >>
>      >>
>      >> * ~/Dow/jdk-21/C/H/bin  ./jmap -histo:live,file=histo5.data 1672*
>      >>
>      >> Exception in thread"main"
>      >> com.sun.tools.attach.AttachNotSupportedException: Unable to open
>      >> socket file
>      >> /var/folders/k4/j5wzcbnd3m96s5lmqw_8lh8r0000gn/T/.java_pid1672:
>     target
>      >> process1672 doesn't respond within 10500ms or HotSpot VM not loaded
>      >> at
>      >>
>     jdk.attach/sun.tools.attach.VirtualMachineImpl.<init>(VirtualMachineImpl.java:95)
>      >>      at
>      >>
>     jdk.attach/sun.tools.attach.AttachProviderImpl.attachVirtualMachine(AttachProviderImpl.java:58)
>      >>      at
>      >>
>     jdk.attach/com.sun.tools.attach.VirtualMachine.attach(VirtualMachine.java:207)
>      >>      at
>     jdk.jcmd/sun.tools.jmap.JMap.executeCommandForPid(JMap.java:127)
>      >>      at jdk.jcmd/sun.tools.jmap.JMap.histo(JMap.java:199)
>      >>      at jdk.jcmd/sun.tools.jmap.JMap.main(JMap.java:112)
>      >
>      > These problems, and the workaround, suggest that a thread is
>     preventing
>      > the VM from going to safepoint, but I can't see why that should
>     be the
>      > case. If you apply the StartAttachListener workaround what does
>     jstack
>      > show you is happening?
> 
>     Actually if the VM can't reach a safepoint then jstack won't be able to
>     run either. You need to run a native debugger against the process and
>     get the stack traces.
> 
>     David
>     ------
> 
>      > David
>      > -----
>      >
>      >>
>      >> Several tests:
>      >> 1. The above code runs fine on windows and macos, and the process
>      >> could be accessed by jconsole and jmap on windows, only fail on
>     macos
>      >> 2. The problem could be solved by adding VM options, for jconsole
>      >> usable, add -Dcom.sun.management.jmxremote for jmap usable add
>      >> -XX:+StartAttachListener , if there is no
>      >> *SymbolLookup.libraryLookup*(), then jconsole and jmap would
>     work well
>      >> without these two arguments
>      >>
>      >> So, I guess there must be some internal logic that blocks the
>     jconsole
>      >> and jmap from entering java program inside
>      >> *SymbolLookup.libraryLookup*(), please correct me if I am wrong
>      >>
>      >>
>      >>
>