Hi Markus,

On 20/02/2016 1:55 AM, Markus Gronlund wrote:
Greetings,

Please review this small change lowering the lock rankings  of some locks.

Have we actually verified the new ranking constraints (ie that special guarantees not to block) by code inspection?

This is done in order to reduce the risk for potential deadlocks and to
increase the surface area for event generation.

Bug: https://bugs.openjdk.java.net/browse/JDK-8149803

Patch of this tiny change is attached.

- def(JfrStream_lock , Mutex, nonleaf, true, Monitor::_safepoint_check_never); + def(JfrStream_lock , Mutex, special+1, true, Monitor::_safepoint_check_never);


Not clear what "special+1" is supposed to signify here - doesn't that make it the same rank as suspend_resume?

  enum lock_types {
       event,
       special,
       suspend_resume,
       leaf        = suspend_resume +   2,
       safepoint   = leaf           +  10,
       barrier     = safepoint      +   1,
       nonleaf     = barrier        +   1,
       max_nonleaf = nonleaf        + 900,
       native      = max_nonleaf    +   1
  };


Thanks,
David


Thanks in advance

Markus

Reply via email to