Shared deletion list is not naturally freed and triggers too many resets in V8 with JIT optimization

From byron.c....@gmail.com on September 17, 2014 20:32:07

In a run of Octane on V8 with inference-based JIT optimization (not yet committed), all shared deletion list entries remain pending until the application exits. There are ~6,000 shared deletion list entries accumulated at that point. This is not a problem in Ion. It looks like this is caused by idle worker threads in V8, which never enter DR and therefore never clear their reference count on pending deletions.

This is probably also an issue in base DR (without JIT optimization), but it is much less of a problem because only a few shared fragments are individually deleted (most deletions occur via flush, which performs a synch).

Original issue: http://code.google.com/p/dynamorio/issues/detail?id=1543