BUG: withOwnerThread closes MappedMemorySegment

[Ty Young]

Forgot to mention, I'm using default GC.


On 9/25/20 10:00 AM, Jorn Vernee wrote:
> Hi Ty,
>
> I'm trying to reproduce your crash locally, but no luck so far. From 
> the error message, it seems like it might be caused by a thread being 
> suspended while inside a native call, which triggers a slow path that 
> seems to have a bug in it; not setting the thread state to 
> _thread_in_java again.
>
> The following patch fixes that mistake [1], but since I'm not able so 
> far to reproduce the crash I can't say for sure that that will fix the 
> problem you are seeing.
>
> It would help if you could provide some steps to reproduce, and/or try 
> out the patch at [1] if you can (should be enough to put it in a file 
> and then use `git apply <file>`).
>
> I'm also curious to know if you are running with the default GC, or 
> maybe another one?
>
> Thanks,
> Jorn
>
> [1] :
>
> diff --git a/src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp 
> b/src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp
> index 51b9082cfae..887369fc5c8 100644
> --- a/src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp
> +++ b/src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp
> @@ -3810,11 +3810,12 @@ void NativeInvokerGenerator::generate() {
>    __ safepoint_poll(L_safepoint_poll_slow_path, r15_thread, rscratch1);
>    __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
>    __ jcc(Assembler::notEqual, L_safepoint_poll_slow_path);
> -  // change thread state
> -  __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 
> _thread_in_Java);
>
>    __ bind(L_after_safepoint_poll);
>
> +  // change thread state
> +  __ movl(Address(r15_thread, JavaThread::thread_state_offset()), 
> _thread_in_Java);
> +
>    __ block_comment("reguard stack check");
>    Label L_reguard;
>    Label L_after_reguard
>