review (S) for 6956931: assert(SafepointSynchronize::is_at_safepoint()) failed: must be executed at a safepoint

[Tom Rodriguez]

On Jun 2, 2010, at 11:21 AM, Daniel D. Daugherty wrote:

> src/share/vm/code/nmethod.cpp
>   Looks like the old code (1344-1358) had the possibility of a
>   double event post for both JVM/TI and dtrace. The newly
>   refactored code doesn't have that (possible) issue. Nice fix.
> 
>   I'm not aware of any open bugs complaining about double
>   CompiledMethodUnload events or dtrace probes, but if there are
>   I think you've fixed those too.
> 
> src/share/vm/prims/jvmtiExport.hpp
>   internal_post_compiled_method_unload() should be
>   post_compiled_method_unload_internal() if you want to follow
>   the existing style in jvmtiExport.
> 
> src/share/vm/prims/jvmtiExport.cpp
>   Same internal_post_compiled_method_unload() naming issue

Fixed.  Thanks!

tom

> 
> Dan
> 
> 
> On 6/2/2010 10:51 AM, Tom Rodriguez wrote:
>> http://cr.openjdk.java.net/~never/6956931
>> 
>> 6956931: assert(SafepointSynchronize::is_at_safepoint()) failed: must be executed at a safepoint
>> Reviewed-by:
>> 
>> SSince the sweeper no longer runs at a safepoint the compiled method
>> unload logic doesn't always run at a safepoint so we need a new
>> synchronous notification path that's equivalent to the
>> compiled_method_load logic.  I refactored the code a bit to share the
>> core notification logic and I refactored the notification in nmethod
>> to remove duplication.  Tested with NSK jvmti,jdb,jdi,hprof suites.